v4.11.0:

HIGHLIGHTS:
- NRD: added "GetMaxAccumulatedFrameNum" helper, emphasizing the importance of time-based accumulation, also added "DEFAULT_ACCUMULATION_TIME" constants for all denoisers
- NRD: added "cameraAttachedReflectionMaterialID" to "CommonSettings" (currently used by REBLUR only), helping to deal with mirror reflections of objects attached to the camera. It's a suboptimal solution, future releases will address this problem better
- REBLUR/SIGMA: "stabilizationStrength" replaced with "maxStabilizedFrameNum" for consistency with other accumulation related settings and "GetMaxAccumulatedFrameNum" helper. The progression "0, 1, 2, 3... 9" is more understandable than "1.00, 0.50, 0.33, 0.25... 0.10"

BREAKING CHANGES:
- settings related (please, read UPDATE.md)
- "NRD_FrontEnd_UnpackNormalAndRoughness" returns "materialID" in the same range, it was passed to "NRD_FrontEnd_PackNormalAndRoughness"

DETAILS:
- NRD: added "GetMaxAccumulatedFrameNum" helper to the main header
- NRD: added "cameraAttachedReflectionMaterialID" to "CommonSettings" (currently used by REBLUR only, WIP longer term)
- NRD: "NRD_FrontEnd_UnpackNormalAndRoughness" returns "materialID" in the same range, it was passed to "NRD_FrontEnd_PackNormalAndRoughness"
- NRD: improved code reuse between REBLUR and RELAX, reduced divergence
- NRD: improved per frame rotators (better noise properties)
- NRD: interface polishing (no functional changes)
- RELAX: added clamping to user provided "max accumulated frame num" values
- RELAX: reduced memory usage of DIFFUSE denoiser (had a dummy allocation)
- RELAX: diff-spec lobe fraction replaced with one shared fraction (simplified settings)
- RELAX: partial refactoring to match overall NRD style
- REBLUR/RELAX: hot/cold branch sorting
- REBLUR/RELAX: improved parallax computations, affecting many aspects of specular tracking
- REBLUR/RELAX: improved curvature computations
- REBLUR/RELAX: added forgotten initialization of "gStrandThickness" constant (oops!)
- REBLUR: minor performance opts
- REBLUR: fixed minor regressions related to migration to stochastic bilinear filter for RGBA1010102 oct-packed normals (since prev data repacking is gone)
- REBLUR: stochastic bilinear gets replaced with native bilinear if normals are interpolate-able
- REBLUR: slightly decreased hit distance sensitivity in case of 0 roughness
- REBLUR: slightly reduced "lobeAngleFraction" default value to squeeze more normal details (using the old default is OK)
- REBLUR: "lobeHalfAngle" uses code from "GetNormalWeightParam" (reduced code entropy)
- REBLUR: adjusted virtual parallax based confidence to be a bit more aggressive
- REBLUR: improved screen space sampling (now used in more shaders without IQ loss, improves perf, WIP)
- REBLUR: "GetGeometryWeightParams" should not become more strict if history is long (the rest is handled by normal weights)
- REBLUR: minimized deprecated "GetBilateralWeight" usage
- REBLUR: relaxed maximum value of slope scaling for disocclusion threshold
- REBLUR: improved antilag (which didn't do much last year), also made almost FPS independent, tuned to "not hurt"
- REBLUR: finally fixed ancient and ugly issue leading to "bright lighting crawling", caused by unbiased (bright) fast history selectively accelerating the biased (fireflies suppressed) main history
- REBLUR/SIGMA: "stabilization strength" settings replaced with "max stabilized frame num"
- SIGMA: removed unnecessary checks in spatial filters
- SIGMA: "closest viewZ" replaced with "longest MV" logic in the temporal stabilization pass
- SIGMA: history copying decoupled into a separate pass
- SIGMA: added disocclusion tracking to get rid of ghosting
- SIGMA: fixed ugly white outlines (suboptimal weighting scheme between dense and sparse passes)
- SIGMA: tuned defaults settings
- SIGMA: introduced screen space sampling to better respect low-discrepancy based per frame rotators
- updated docs
- updated dependencies
- refactoring

Include/NRD.h

@@ -28,14 +28,12 @@ license agreement from NVIDIA CORPORATION is strictly prohibited.
 #include <cstddef>
 
 #define NRD_VERSION_MAJOR 4
-#define NRD_VERSION_MINOR 10
+#define NRD_VERSION_MINOR 11
 #define NRD_VERSION_BUILD 0
-#define NRD_VERSION_DATE "9 October 2024"
+#define NRD_VERSION_DATE "5 November 2024"
 
-#if defined(_MSC_VER)
-    #define NRD_CALL __fastcall
-#elif !defined(__aarch64__) && !defined(__x86_64) && (defined(__GNUC__)  || defined (__clang__))
-    #define NRD_CALL __attribute__((fastcall))
+#if defined(_WIN32)
+    #define NRD_CALL __stdcall
 #else
     #define NRD_CALL
 #endif

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 10
+#define NRD_DESCS_VERSION_MINOR 11
 
 static_assert(NRD_VERSION_MAJOR == NRD_DESCS_VERSION_MAJOR && NRD_VERSION_MINOR == NRD_DESCS_VERSION_MINOR, "Please, update all NRD SDK files");
 
@@ -167,9 +167,9 @@ namespace nrd
           - Optional inputs are in ()
         */
 
-        // =============================================================================================================================
+        //=============================================================================================================================
         // REBLUR
-        // =============================================================================================================================
+        //=============================================================================================================================
 
         // INPUTS - IN_DIFF_RADIANCE_HITDIST (IN_DIFF_CONFIDENCE, IN_DISOCCLUSION_THRESHOLD_MIX)
         // OUTPUTS - OUT_DIFF_RADIANCE_HITDIST
@@ -211,9 +211,9 @@ namespace nrd
         // OUTPUTS - OUT_DIFF_DIRECTION_HITDIST
         REBLUR_DIFFUSE_DIRECTIONAL_OCCLUSION,
 
-        // =============================================================================================================================
+        //=============================================================================================================================
         // RELAX
-        // =============================================================================================================================
+        //=============================================================================================================================
 
         // INPUTS - IN_DIFF_RADIANCE_HITDIST (IN_DIFF_CONFIDENCE, IN_DISOCCLUSION_THRESHOLD_MIX)
         // OUTPUTS - OUT_DIFF_RADIANCE_HITDIST
@@ -239,9 +239,9 @@ namespace nrd
         // OUTPUTS - OUT_DIFF_SH0, OUT_DIFF_SH1, OUT_SPEC_SH0, OUT_SPEC_SH1
         RELAX_DIFFUSE_SPECULAR_SH,
 
-        // =============================================================================================================================
+        //=============================================================================================================================
         // SIGMA
-        // =============================================================================================================================
+        //=============================================================================================================================
 
         // INPUTS - IN_PENUMBRA, OUT_SHADOW_TRANSLUCENCY
         // OUTPUTS - OUT_SHADOW_TRANSLUCENCY
@@ -251,9 +251,9 @@ namespace nrd
         // OUTPUTS - OUT_SHADOW_TRANSLUCENCY
         SIGMA_SHADOW_TRANSLUCENCY,
 
-        // =============================================================================================================================
+        //=============================================================================================================================
         // REFERENCE
-        // =============================================================================================================================
+        //=============================================================================================================================
 
         // INPUTS - IN_SIGNAL
         // OUTPUTS - OUT_SIGNAL

Integration/NRDIntegration.hpp

@@ -567,6 +567,7 @@ void Integration::Dispatch(nri::CommandBuffer& commandBuffer, nri::DescriptorPoo
         {
             const ResourceDesc& nrdResource = dispatchDesc.resources[n];
 
+            // Get texture
             nri::TextureBarrierDesc* nrdTexture = nullptr;
             if (nrdResource.type == ResourceType::TRANSIENT_POOL)
                 nrdTexture = &m_TexturePool[nrdResource.indexInPool + instanceDesc.permanentPoolSize];
@@ -578,13 +579,19 @@ void Integration::Dispatch(nri::CommandBuffer& commandBuffer, nri::DescriptorPoo
                 NRD_INTEGRATION_ASSERT(nrdTexture && nrdTexture->texture, "'userPool' entry can't be NULL if it's in use!");
             }
 
-            const nri::AccessBits nextAccess = nrdResource.descriptorType == DescriptorType::TEXTURE ? nri::AccessBits::SHADER_RESOURCE : nri::AccessBits::SHADER_RESOURCE_STORAGE;
-            const nri::Layout nextLayout =  nrdResource.descriptorType == DescriptorType::TEXTURE ? nri::Layout::SHADER_RESOURCE : nri::Layout::SHADER_RESOURCE_STORAGE;
-            bool isStateChanged = nextAccess != nrdTexture->after.access || nextLayout != nrdTexture->after.layout;
-            bool isStorageBarrier = nextAccess == nri::AccessBits::SHADER_RESOURCE_STORAGE && nrdTexture->after.access == nri::AccessBits::SHADER_RESOURCE_STORAGE;
+            // Prepare barrier
+            nri::AccessLayoutStage next = {};
+            if (nrdResource.descriptorType == DescriptorType::TEXTURE)
+                next = {nri::AccessBits::SHADER_RESOURCE, nri::Layout::SHADER_RESOURCE, nri::StageBits::COMPUTE_SHADER};
+            else
+                next = {nri::AccessBits::SHADER_RESOURCE_STORAGE, nri::Layout::SHADER_RESOURCE_STORAGE, nri::StageBits::COMPUTE_SHADER};
+
+            bool isStateChanged = next.access != nrdTexture->after.access || next.layout != nrdTexture->after.layout;
+            bool isStorageBarrier = next.access == nri::AccessBits::SHADER_RESOURCE_STORAGE && nrdTexture->after.access == nri::AccessBits::SHADER_RESOURCE_STORAGE;
             if (isStateChanged || isStorageBarrier)
-                transitions[transitionBarriers.textureNum++] = nri::TextureBarrierFromState(*nrdTexture, {nextAccess, nextLayout}, 0, 1);
+                transitions[transitionBarriers.textureNum++] = nri::TextureBarrierFromState(*nrdTexture, next);
 
+            // Create descriptor
             uint64_t resource = m_NRI->GetTextureNativeObject(*nrdTexture->texture);
             uint64_t key = CreateDescriptorKey(resource, isStorage);
             const auto& entry = m_CachedDescriptors.find(key);
@@ -607,6 +614,9 @@ void Integration::Dispatch(nri::CommandBuffer& commandBuffer, nri::DescriptorPoo
         }
     }
 
+    // Barriers
+    m_NRI->CmdBarrier(commandBuffer, transitionBarriers);
+
     // Allocating descriptor sets
     uint32_t descriptorSetSamplersIndex = instanceDesc.constantBufferSpaceIndex == instanceDesc.samplersSpaceIndex ? 0 : 1;
     uint32_t descriptorSetResourcesIndex = instanceDesc.resourcesSpaceIndex == instanceDesc.constantBufferSpaceIndex ? 0 : (instanceDesc.resourcesSpaceIndex == instanceDesc.samplersSpaceIndex ? descriptorSetSamplersIndex : descriptorSetSamplersIndex + 1);
@@ -664,7 +674,6 @@ void Integration::Dispatch(nri::CommandBuffer& commandBuffer, nri::DescriptorPoo
     m_NRI->UpdateDescriptorRanges(*descriptorSets[descriptorSetResourcesIndex], instanceDesc.samplersSpaceIndex == instanceDesc.resourcesSpaceIndex ? 1 : 0, pipelineDesc.resourceRangesNum, resourceRanges);
 
     // Rendering
-    m_NRI->CmdBarrier(commandBuffer, transitionBarriers);
     m_NRI->CmdSetPipelineLayout(commandBuffer, *pipelineLayout);
 
     nri::Pipeline* pipeline = m_Pipelines[dispatchDesc.pipelineIndex];

README.md

@@ -1,4 +1,4 @@
-# NVIDIA REAL-TIME DENOISERS v4.10.0 (NRD)
+# NVIDIA REAL-TIME DENOISERS v4.11.0 (NRD)
 
 [![Build NRD SDK](https://github.com/NVIDIAGameWorks/RayTracingDenoiser/actions/workflows/build.yml/badge.svg)](https://github.com/NVIDIAGameWorks/RayTracingDenoiser/actions/workflows/build.yml)
 
@@ -15,11 +15,11 @@ For quick starting see *[NRD sample](https://github.com/NVIDIAGameWorks/NRDSampl
 - *RELAX* - A-trous based denoiser, has been designed for *[RTXDI (RTX Direct Illumination)](https://developer.nvidia.com/rtxdi)*
 - *SIGMA* - shadow-only denoiser
 
-Performance on RTX 4080 @ 1440p (native resolution, default denoiser settings):
-- `REBLUR_DIFFUSE_SPECULAR` - 2.45 ms
-- `RELAX_DIFFUSE_SPECULAR` - 2.90 ms
-- `SIGMA_SHADOW` - 0.35 ms (0.25 ms if temporal stabilization is off)
-- `SIGMA_SHADOW_TRANSLUCENCY` - 0.45 ms (0.32 ms if temporal stabilization is off)
+Performance on RTX 4080 @ 1440p (native resolution, default denoiser settings, `NormalEncoding::R10_G10_B10_A2_UNORM`):
+- `REBLUR_DIFFUSE_SPECULAR` - 2.40 ms (2.15 in performance mode)
+- `RELAX_DIFFUSE_SPECULAR` - 2.95 ms
+- `SIGMA_SHADOW` - 0.40 ms
+- `SIGMA_SHADOW_TRANSLUCENCY` - 0.50 ms
 
 Supported signal types:
 - *RELAX*:
@@ -322,14 +322,14 @@ The *Persistent* column (matches *NRD Permanent pool*) indicates how much of the
 |            |    REBLUR_DIFFUSE_SPECULAR_OCCLUSION |            67.94 |            38.12 |            29.81 |
 |            |           REBLUR_DIFFUSE_SPECULAR_SH |           266.12 |           105.62 |           160.50 |
 |            | REBLUR_DIFFUSE_DIRECTIONAL_OCCLUSION |            84.56 |            42.25 |            42.31 |
-|            |                        RELAX_DIFFUSE |            99.25 |            63.31 |            35.94 |
+|            |                        RELAX_DIFFUSE |            90.81 |            54.88 |            35.94 |
 |            |                     RELAX_DIFFUSE_SH |           158.31 |            88.62 |            69.69 |
 |            |                       RELAX_SPECULAR |           101.44 |            63.38 |            38.06 |
 |            |                    RELAX_SPECULAR_SH |           168.94 |            97.12 |            71.81 |
 |            |               RELAX_DIFFUSE_SPECULAR |           168.94 |            97.12 |            71.81 |
 |            |            RELAX_DIFFUSE_SPECULAR_SH |           303.94 |           164.62 |           139.31 |
-|            |                         SIGMA_SHADOW |            15.00 |             0.00 |            15.00 |
-|            |            SIGMA_SHADOW_TRANSLUCENCY |            33.94 |             0.00 |            33.94 |
+|            |                         SIGMA_SHADOW |            31.88 |             8.44 |            23.44 |
+|            |            SIGMA_SHADOW_TRANSLUCENCY |            50.81 |             8.44 |            42.38 |
 |            |                            REFERENCE |            33.75 |            33.75 |             0.00 |
 |            |                                      |                  |                  |                  |
 |      1440p |                       REBLUR_DIFFUSE |           150.06 |            75.00 |            75.06 |
@@ -342,14 +342,14 @@ The *Persistent* column (matches *NRD Permanent pool*) indicates how much of the
 |            |    REBLUR_DIFFUSE_SPECULAR_OCCLUSION |           120.06 |            67.50 |            52.56 |
 |            |           REBLUR_DIFFUSE_SPECULAR_SH |           472.56 |           187.50 |           285.06 |
 |            | REBLUR_DIFFUSE_DIRECTIONAL_OCCLUSION |           150.06 |            75.00 |            75.06 |
-|            |                        RELAX_DIFFUSE |           176.31 |           112.50 |            63.81 |
+|            |                        RELAX_DIFFUSE |           161.31 |            97.50 |            63.81 |
 |            |                     RELAX_DIFFUSE_SH |           281.31 |           157.50 |           123.81 |
 |            |                       RELAX_SPECULAR |           180.06 |           112.50 |            67.56 |
 |            |                    RELAX_SPECULAR_SH |           300.06 |           172.50 |           127.56 |
 |            |               RELAX_DIFFUSE_SPECULAR |           300.06 |           172.50 |           127.56 |
 |            |            RELAX_DIFFUSE_SPECULAR_SH |           540.06 |           292.50 |           247.56 |
-|            |                         SIGMA_SHADOW |            26.38 |             0.00 |            26.38 |
-|            |            SIGMA_SHADOW_TRANSLUCENCY |            60.12 |             0.00 |            60.12 |
+|            |                         SIGMA_SHADOW |            56.38 |            15.00 |            41.38 |
+|            |            SIGMA_SHADOW_TRANSLUCENCY |            90.12 |            15.00 |            75.12 |
 |            |                            REFERENCE |            60.00 |            60.00 |             0.00 |
 |            |                                      |                  |                  |                  |
 |      2160p |                       REBLUR_DIFFUSE |           318.88 |           159.38 |           159.50 |
@@ -362,14 +362,14 @@ The *Persistent* column (matches *NRD Permanent pool*) indicates how much of the
 |            |    REBLUR_DIFFUSE_SPECULAR_OCCLUSION |           255.06 |           143.44 |           111.62 |
 |            |           REBLUR_DIFFUSE_SPECULAR_SH |          1004.12 |           398.44 |           605.69 |
 |            | REBLUR_DIFFUSE_DIRECTIONAL_OCCLUSION |           318.88 |           159.38 |           159.50 |
-|            |                        RELAX_DIFFUSE |           374.69 |           239.12 |           135.56 |
+|            |                        RELAX_DIFFUSE |           342.81 |           207.25 |           135.56 |
 |            |                     RELAX_DIFFUSE_SH |           597.81 |           334.75 |           263.06 |
 |            |                       RELAX_SPECULAR |           382.69 |           239.12 |           143.56 |
 |            |                    RELAX_SPECULAR_SH |           637.69 |           366.62 |           271.06 |
 |            |               RELAX_DIFFUSE_SPECULAR |           637.69 |           366.62 |           271.06 |
 |            |            RELAX_DIFFUSE_SPECULAR_SH |          1147.69 |           621.62 |           526.06 |
-|            |                         SIGMA_SHADOW |            56.19 |             0.00 |            56.19 |
-|            |            SIGMA_SHADOW_TRANSLUCENCY |           127.81 |             0.00 |           127.81 |
+|            |                         SIGMA_SHADOW |           119.94 |            31.88 |            88.06 |
+|            |            SIGMA_SHADOW_TRANSLUCENCY |           191.56 |            31.88 |           159.69 |
 |            |                            REFERENCE |           127.50 |           127.50 |             0.00 |
 
 # INTEGRATION VARIANTS
@@ -636,7 +636,7 @@ When denoising reflections in pure mirrors, some advantages can be reached if *N
 Notes, requirements and restrictions:
 - the primary hit (0th bounce) gets replaced with the first "non-pure mirror" hit in the bounce chain - this hit becomes *PSR*
 - all associated data in the g-buffer gets replaced by *PSR* data
-- the camera "sees" PSR like the mirror surface in-between don't exist. This space is called virtual world space
+- the camera "sees" PSR like the mirror surface in-between doesn't exist. This space is called virtual world space
   - virtual space position lies on the same view vector as the primary hit position, but the position is elongated. Elongation depends on `hitT` and curvature at hits, starting from the primary hit
   - virtual space normal is the normal at *PSR* hit mirrored several times  in the reversed order until the primary hit is reached
 - *PSR* data is NOT always data at the *PSR* hit!
@@ -674,7 +674,7 @@ IN_MV = GetMotionAt( B );
 
 ## INTERACTION WITH FRAME GENERATION TECHNIQUES
 
-Frame generation (FG) techniques boost FPS by interpolating between 2 last available frames. *NRD* works better when framerate increases, because it gets more data per second. It's not the case for FG, because all rendering pipeline underlying passes (like, denoising) continue to work on the original non-boosted framerate.
+Frame generation (FG) techniques boost FPS by interpolating between 2 last available frames. *NRD* works better when frame rate increases, because it gets more data per second. It's not the case for FG, because all rendering pipeline underlying passes (like, denoising) continue to work on the original non-boosted framerate. `GetMaxAccumulatedFrameNum` helper should get a real FPS, not a fake one.
 
 ## HAIR DENOISING TIPS
 
@@ -704,7 +704,7 @@ Hair strands tangent vectors *can't* be used as "normals guide" for *NRD* due to
 
 **[NRD]** *NRD* requires non-jittered matrices.
 
-**[NRD]** Most of denoisers do not write into output pixels outside of `CommonSettings::denoisingRange`.
+**[NRD]** Most denoisers do not write into output pixels outside of `CommonSettings::denoisingRange`.
 
 **[NRD]** When upgrading to the latest version keep an eye on `ResourceType` enumeration. The order of the input slots can be changed or something can be added, you need to adjust the inputs accordingly to match the mapping. Or use *NRD integration* to simplify the process.
 
@@ -771,7 +771,7 @@ maxAccumulatedFrameNum > maxFastAccumulatedFrameNum > historyFixFrameNum
 
 **[SIGMA]** Using "blue" noise helps to minimize shadow shimmering and flickering. It works best if the pattern has limited number of animated frames (4-8) or it is static on the screen.
 
-**[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:
+**[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 following benefits:
  - light count independent memory usage
  - no need to manage history buffers for lights
 

Resources/Version.h

@@ -22,7 +22,7 @@ Versioning rules:
 */
 
 #define VERSION_MAJOR                   4
-#define VERSION_MINOR                   10
+#define VERSION_MINOR                   11
 #define VERSION_BUILD                   0
 
 #define VERSION_STRING STR(VERSION_MAJOR.VERSION_MINOR.VERSION_BUILD encoding=NRD_NORMAL_ENCODING.NRD_ROUGHNESS_ENCODING)

Shaders.cfg

@@ -230,14 +230,15 @@ RELAX_DiffuseSpecularSh_PrePass.cs.hlsl -T cs
 RELAX_DiffuseSpecularSh_SplitScreen.cs.hlsl -T cs
 RELAX_DiffuseSpecularSh_TemporalAccumulation.cs.hlsl -T cs
 RELAX_Validation.cs.hlsl -T cs
-SIGMA_ShadowTranslucency_Blur.cs.hlsl -T cs
-SIGMA_ShadowTranslucency_ClassifyTiles.cs.hlsl -T cs
-SIGMA_ShadowTranslucency_PostBlur.cs.hlsl -T cs
-SIGMA_ShadowTranslucency_SplitScreen.cs.hlsl -T cs
-SIGMA_ShadowTranslucency_TemporalStabilization.cs.hlsl -T cs
+SIGMA_Copy.cs.hlsl -T cs
+SIGMA_SmoothTiles.cs.hlsl -T cs
 SIGMA_Shadow_Blur.cs.hlsl -T cs
 SIGMA_Shadow_ClassifyTiles.cs.hlsl -T cs
 SIGMA_Shadow_PostBlur.cs.hlsl -T cs
-SIGMA_Shadow_SmoothTiles.cs.hlsl -T cs
 SIGMA_Shadow_SplitScreen.cs.hlsl -T cs
 SIGMA_Shadow_TemporalStabilization.cs.hlsl -T cs
+SIGMA_ShadowTranslucency_Blur.cs.hlsl -T cs
+SIGMA_ShadowTranslucency_ClassifyTiles.cs.hlsl -T cs
+SIGMA_ShadowTranslucency_PostBlur.cs.hlsl -T cs
+SIGMA_ShadowTranslucency_SplitScreen.cs.hlsl -T cs
+SIGMA_ShadowTranslucency_TemporalStabilization.cs.hlsl -T cs