v4.8.0:

HIGHLIGHTS: - NRD: improved FP16 viewZ support - SIGMA: made more friendly to per light shadow denoising (multi-light scenario) - removed unused (non-denoisers) "SPECULAR_REFLECTION_MV" & "SPECULAR_DELTA_MV" DETAILS: - NRD: exposed "CommonSettings::viewZScale", which should help to improve FP16 support - SIGMA: "IN_SHADOWDATA" => "IN_PENUMBRA", "IN_SHADOW_TRANSLUCENCY" => "IN_TRANSLUCENCY" - SIGMA: massaged front-end functions - SIGMA: switched to use common "IN_VIEWZ" input instead of a custom one (it eliminates a need to clone ViewZ in case of multiple lights) - REBLUR: minor tweaks - SPECULAR_REFLECTION_MV: deprecated and removed - SPECULAR_DELTA_MV: deprecated and removed - updated docs
NVIDIAGameWorks · May 9, 2024 · 8d3c245 · 8d3c245
1 parent f20158d
commit 8d3c245
Show file tree

Hide file tree

Showing 58 changed files with 330 additions and 970 deletions.
diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -195,7 +195,6 @@ if (NOT NRD_DISABLE_SHADER_COMPILATION)
 
  if (NRD_EMBEDS_SPIRV_SHADERS)
  set (SHADERMAKE_COMMANDS ${SHADERMAKE_COMMANDS} COMMAND ShaderMake -p SPIRV --compiler "${DXC_SPIRV_PATH}" ${SHADERMAKE_GENERAL_ARGS}
- -D VULKAN
  --sRegShift 100
  --tRegShift 200
  --bRegShift 300

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 7
+#define NRD_VERSION_MINOR 8
 #define NRD_VERSION_BUILD 0
-#define NRD_VERSION_DATE "17 April 2024"
+#define NRD_VERSION_DATE "9 May 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 7
+#define NRD_DESCS_VERSION_MINOR 8
 
 static_assert(NRD_VERSION_MAJOR == NRD_DESCS_VERSION_MAJOR && NRD_VERSION_MINOR == NRD_DESCS_VERSION_MINOR, "Please, update all NRD SDK files");
 
@@ -92,10 +92,11 @@ namespace nrd
  IN_SPEC_SH0,
  IN_SPEC_SH1,
 
- // Shadow data and optional translucency (RG16f+ and RGBA8+ for translucency)
- // SIGMA: use "SIGMA_FrontEnd_PackShadow" for encoding
- IN_SHADOWDATA,
- IN_SHADOW_TRANSLUCENCY,
+ // Penumbra and optional translucency (R16f+ and RGBA8+ for translucency)
+ // SIGMA: use "SIGMA_FrontEnd_PackPenumbra" for penumbra properties encoding
+ // SIGMA: use "SIGMA_FrontEnd_PackTranslucency" for translucency encoding
+ IN_PENUMBRA,
+ IN_TRANSLUCENCY,
 
  // Some signal (R8+)
  IN_SIGNAL,
@@ -108,44 +109,37 @@ namespace nrd
  // OUTPUTS
  //=============================================================================================================================
 
- // IMPORTANT: These textures can be potentially used as history buffers!
  // IMPORTANT: Most of denoisers do not write into output pixels outside of "CommonSettings::denoisingRange"!
 
  // Radiance and hit distance
  // REBLUR: use "REBLUR_BackEnd_UnpackRadianceAndNormHitDist" for decoding (RGBA16f+)
  // RELAX: use "RELAX_BackEnd_UnpackRadiance" for decoding (R11G11B10f+)
- OUT_DIFF_RADIANCE_HITDIST,
- OUT_SPEC_RADIANCE_HITDIST,
+ OUT_DIFF_RADIANCE_HITDIST, // IMPORTANT: used as history if "stabilizationStrength != 0"
+ OUT_SPEC_RADIANCE_HITDIST, // IMPORTANT: used as history if "stabilizationStrength != 0"
 
  // SH data
  // REBLUR: use "REBLUR_BackEnd_UnpackSh" for decoding (2x RGBA16f+)
  // RELAX: use "RELAX_BackEnd_UnpackSh" for decoding (2x RGBA16f+)
- OUT_DIFF_SH0,
- OUT_DIFF_SH1,
- OUT_SPEC_SH0,
- OUT_SPEC_SH1,
+ OUT_DIFF_SH0, // IMPORTANT: used as history if "stabilizationStrength != 0"
+ OUT_DIFF_SH1, // IMPORTANT: used as history if "stabilizationStrength != 0"
+ OUT_SPEC_SH0, // IMPORTANT: used as history if "stabilizationStrength != 0"
+ OUT_SPEC_SH1, // IMPORTANT: used as history if "stabilizationStrength != 0"
 
  // Normalized hit distance (R8+)
  OUT_DIFF_HITDIST,
  OUT_SPEC_HITDIST,
 
  // Bent normal and normalized hit distance (RGBA8+)
  // REBLUR: use "REBLUR_BackEnd_UnpackDirectionalOcclusion" for decoding
- OUT_DIFF_DIRECTION_HITDIST,
+ OUT_DIFF_DIRECTION_HITDIST, // IMPORTANT: used as history if "stabilizationStrength != 0"
 
  // Shadow and optional transcluceny (R8+ or RGBA8+)
  // SIGMA: use "SIGMA_BackEnd_UnpackShadow" for decoding
- OUT_SHADOW_TRANSLUCENCY,
+ OUT_SHADOW_TRANSLUCENCY, // IMPORTANT: used as history if "stabilizationStrength != 0"
 
  // Denoised signal (R8+)
  OUT_SIGNAL,
 
- // 2D screen-space specular motion (RG16f+), MV = previous - current
- OUT_REFLECTION_MV,
-
- // 2D screen-space refraction motion (RG16f+), MV = previous - current
- OUT_DELTA_MV,
-
  // (Optional) Debug output (RGBA8+), .w = transparency
  // Used if "CommonSettings::enableValidation = true"
  OUT_VALIDATION,
@@ -168,10 +162,8 @@ namespace nrd
  /*
  IMPORTANT: IN_MV, IN_NORMAL_ROUGHNESS, IN_VIEWZ are used by any denoiser
  These denoisers DON'T use:
- SIGMA_SHADOW - IN_VIEWZ
- SIGMA_SHADOW_TRANSLUCENCY - IN_VIEWZ
+ SIGMA_SHADOW & SIGMA_SHADOW_TRANSLUCENCY - IN_MV, if "stabilizationStrength = 0"
  REFERENCE - IN_MV, IN_NORMAL_ROUGHNESS, IN_VIEWZ
- SPECULAR_DELTA_MV - IN_NORMAL_ROUGHNESS, IN_VIEWZ
  */
 
  // =============================================================================================================================
@@ -225,18 +217,6 @@ namespace nrd
  // OUTPUTS - OUT_DIFF_DIRECTION_HITDIST
  REBLUR_DIFFUSE_DIRECTIONAL_OCCLUSION,
 
- // =============================================================================================================================
- // SIGMA
- // =============================================================================================================================
-
- // INPUTS - IN_SHADOWDATA, OUT_SHADOW_TRANSLUCENCY (used as history)
- // OUTPUTS - OUT_SHADOW_TRANSLUCENCY
- SIGMA_SHADOW,
-
- // INPUTS - IN_SHADOWDATA, IN_SHADOW_TRANSLUCENCY, OUT_SHADOW_TRANSLUCENCY (used as history)
- // OUTPUTS - OUT_SHADOW_TRANSLUCENCY
- SIGMA_SHADOW_TRANSLUCENCY,
-
  // =============================================================================================================================
  // RELAX
  // =============================================================================================================================
@@ -272,24 +252,24 @@ namespace nrd
  RELAX_DIFFUSE_SPECULAR_SH,
 
  // =============================================================================================================================
- // REFERENCE
+ // SIGMA
  // =============================================================================================================================
 
- // INPUTS - IN_SIGNAL
- // OUTPUTS - OUT_SIGNAL
- REFERENCE,
+ // INPUTS - IN_PENUMBRA, OUT_SHADOW_TRANSLUCENCY
+ // OUTPUTS - OUT_SHADOW_TRANSLUCENCY
+ SIGMA_SHADOW,
+
+ // INPUTS - IN_PENUMBRA, IN_TRANSLUCENCY, OUT_SHADOW_TRANSLUCENCY
+ // OUTPUTS - OUT_SHADOW_TRANSLUCENCY
+ SIGMA_SHADOW_TRANSLUCENCY,
 
  // =============================================================================================================================
- // MOTION VECTORS
+ // REFERENCE
  // =============================================================================================================================
 
- // INPUTS - IN_SPEC_HITDIST
- // OUTPUTS - OUT_REFLECTION_MV
- SPECULAR_REFLECTION_MV,
-
- // INPUTS - IN_DELTA_PRIMARY_POS, IN_DELTA_SECONDARY_POS
- // OUTPUT - OUT_DELTA_MV
- SPECULAR_DELTA_MV,
+ // INPUTS - IN_SIGNAL
+ // OUTPUTS - OUT_SIGNAL
+ REFERENCE,
 
  MAX_NUM
  };
@@ -479,7 +459,7 @@ namespace nrd
  ComputeShaderDesc computeShaderSPIRV;
  const char* shaderFileName;
  const char* shaderEntryPointName;
- const ResourceRangeDesc* resourceRanges; // up to 2 ranges: "TEXTURE" inputs (optional) and "TEXTURE_STORAGE" outputs 
+ const ResourceRangeDesc* resourceRanges; // up to 2 ranges: "TEXTURE" inputs (optional) and "TEXTURE_STORAGE" outputs
  uint32_t resourceRangesNum;
 
  // Hint that pipeline has a constant buffer with shared parameters from "InstanceDesc"

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 7
+#define NRD_SETTINGS_VERSION_MINOR 8
 
 static_assert(NRD_VERSION_MAJOR == NRD_SETTINGS_VERSION_MAJOR && NRD_VERSION_MINOR == NRD_SETTINGS_VERSION_MINOR, "Please, update all NRD SDK files");
 
@@ -109,6 +109,9 @@ namespace nrd
  uint16_t rectSize[2] = {};
  uint16_t rectSizePrev[2] = {};
 
+ // (>0) - viewZ = IN_VIEWZ * viewZScale (mostly for FP16 viewZ)
+ float viewZScale = 1.0f;
+
  // (ms) - user provided if > 0, otherwise - tracked internally
  float timeDeltaBetweenFrames = 0.0f;
 
@@ -257,22 +260,6 @@ namespace nrd
  bool usePrepassOnlyForSpecularMotionEstimation = false;
  };
 
- // SIGMA
-
- 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;
-
- // (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
 
  const uint32_t RELAX_MAX_HISTORY_FRAME_NUM = 255;
@@ -379,25 +366,27 @@ namespace nrd
  bool enableMaterialTestForSpecular = false;
  };
 
- // REFERENCE
+ // SIGMA
 
- struct ReferenceSettings
+ struct SigmaSettings
  {
- // (>= 0) - maximum number of linearly accumulated frames ( = FPS * "time of accumulation")
- uint32_t maxAccumulatedFrameNum = 1024;
- };
+ // Direction to the light source
+ // IMPORTANT: it is needed only for directional light sources (sun)
+  float lightDirection[3] = {0.0f, 0.0f, 0.0f};
 
- // SPECULAR_REFLECTION_MV
+ // (normalized %) - represents maximum allowed deviation from local tangent plane
+ float planeDistanceSensitivity = 0.005f;
 
- struct SpecularReflectionMvSettings
- {
- float unused;
+  // (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;
  };
 
- // SPECULAR_DELTA_MV
+ // REFERENCE
 
- struct SpecularDeltaMvSettings
+ struct ReferenceSettings
  {
- float unused;
+ // (>= 0) - maximum number of linearly accumulated frames ( = FPS * "time of accumulation")
+ uint32_t maxAccumulatedFrameNum = 1024;
  };
 }
diff --git a/README.md b/README.md
@@ -1,4 +1,4 @@
-# NVIDIA REAL-TIME DENOISERS v4.7.0 (NRD)
+# NVIDIA REAL-TIME DENOISERS v4.8.0 (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 +18,7 @@ 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 (0.24 mns if temporal stabilization is off)
+- `SIGMA_SHADOW` - 0.30 ms (0.24 ms if temporal stabilization is off)
 - `SIGMA_SHADOW_TRANSLUCENCY` - 0.40 ms (0.30 ms if temporal stabilization is off)
 
 Supported signal types:
@@ -309,8 +309,8 @@ The *Persistent* column (matches *NRD Permanent pool*) indicates how much of the
 | | REBLUR_DIFFUSE_SPECULAR_OCCLUSION | 72.12 | 38.12 | 34.00 |
 | | REBLUR_DIFFUSE_SPECULAR_SH | 270.31 | 105.62 | 164.69 |
 | | REBLUR_DIFFUSE_DIRECTIONAL_OCCLUSION | 86.69 | 42.25 | 44.44 |
-| | SIGMA_SHADOW | 23.38 | 0.00 | 23.38 |
-| | SIGMA_SHADOW_TRANSLUCENCY | 42.31 | 0.00 | 42.31 |
+| | SIGMA_SHADOW | 15.00 | 0.00 | 15.00 |
+| | SIGMA_SHADOW_TRANSLUCENCY | 33.94 | 0.00 | 33.94 |
 | | RELAX_DIFFUSE | 99.25 | 63.31 | 35.94 |
 | | RELAX_DIFFUSE_SH | 158.31 | 88.62 | 69.69 |
 | | RELAX_SPECULAR | 101.44 | 63.38 | 38.06 |
@@ -329,8 +329,8 @@ The *Persistent* column (matches *NRD Permanent pool*) indicates how much of the
 | | REBLUR_DIFFUSE_SPECULAR_OCCLUSION | 127.56 | 67.50 | 60.06 |
 | | REBLUR_DIFFUSE_SPECULAR_SH | 480.06 | 187.50 | 292.56 |
 | | REBLUR_DIFFUSE_DIRECTIONAL_OCCLUSION | 153.81 | 75.00 | 78.81 |
-| | SIGMA_SHADOW | 41.38 | 0.00 | 41.38 |
-| | SIGMA_SHADOW_TRANSLUCENCY | 75.12 | 0.00 | 75.12 |
+| | SIGMA_SHADOW | 26.38 | 0.00 | 26.38 |
+| | SIGMA_SHADOW_TRANSLUCENCY | 60.12 | 0.00 | 60.12 |
 | | RELAX_DIFFUSE | 176.31 | 112.50 | 63.81 |
 | | RELAX_DIFFUSE_SH | 281.31 | 157.50 | 123.81 |
 | | RELAX_SPECULAR | 180.06 | 112.50 | 67.56 |
@@ -349,8 +349,8 @@ The *Persistent* column (matches *NRD Permanent pool*) indicates how much of the
 | | REBLUR_DIFFUSE_SPECULAR_OCCLUSION | 271.00 | 143.44 | 127.56 |
 | | REBLUR_DIFFUSE_SPECULAR_SH | 1020.06 | 398.44 | 621.62 |
 | | REBLUR_DIFFUSE_DIRECTIONAL_OCCLUSION | 326.81 | 159.38 | 167.44 |
-| | SIGMA_SHADOW | 88.06 | 0.00 | 88.06 |
-| | SIGMA_SHADOW_TRANSLUCENCY | 159.69 | 0.00 | 159.69 |
+| | SIGMA_SHADOW | 56.19 | 0.00 | 56.19 |
+| | SIGMA_SHADOW_TRANSLUCENCY | 127.81 | 0.00 | 127.81 |
 | | RELAX_DIFFUSE | 374.69 | 239.12 | 135.56 |
 | | RELAX_DIFFUSE_SH | 597.81 | 334.75 | 263.06 |
 | | RELAX_SPECULAR | 382.69 | 239.12 | 143.56 |
@@ -749,7 +749,11 @@ maxAccumulatedFrameNum > maxFastAccumulatedFrameNum > historyFixFrameNum
 
 **[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:
+**[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
+
+**[SIGMA]** In theory *SIGMA_TRANSLUCENT_SHADOW* can be used as a "single-pass" shadow denoiser for shadows from multiple light sources:
 
 *L[i]* - unshadowed analytical lighting from a single light source (**not noisy**)<br/>
 *S[i]* - stochastically sampled light visibility for *L[i]* (**noisy**)<br/>
@@ -766,7 +770,9 @@ Or:<br/>
 Input data preparation example:
 ```cpp
 float3 Lsum = 0;
-float2x3 multiLightShadowData = SIGMA_FrontEnd_MultiLightStart( );
+float3 LSsum = 0.0;
+float Wsum = 0.0;
+float Psum = 0.0;
 
 for( uint i = 0; i < N; i++ )
 {
@@ -775,15 +781,20 @@ for( uint i = 0; i < N; i++ )
 
  // "distanceToOccluder" should respect rules described in NRD.hlsli in "INPUT PARAMETERS" section
  float distanceToOccluder = SampleShadow( i );
+ float shadow = !IsOccluded( distanceToOccluder );
+ LSsum += L * shadow;
 
  // The weight should be zero if a pixel is not in the penumbra, but it is not trivial to compute...
  float weight = ...;
+ weight *= Luminance( L );
+ Wsum += weight;
 
- SIGMA_FrontEnd_MultiLightUpdate( L, distanceToOccluder, tanOfLightAngularRadius, weight, multiLightShadowData );
+ float penumbraRadius = SIGMA_FrontEnd_PackPenumbra( ... ).x;
+ Psum += penumbraRadius * weight;
 }
 
-float4 shadowTranslucency;
-float2 shadowData = SIGMA_FrontEnd_MultiLightEnd( viewZ, multiLightShadowData, Lsum, shadowTranslucency );
+float3 translucency = LSsum / max( Lsum, NRD_EPS );
+float penumbraRadius = Psum / max( Wsum, NRD_EPS );
 ```
 
 After denoising the final result can be computed as:
@@ -797,7 +808,3 @@ 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 7
+#define VERSION_MINOR 8
 #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.cfg b/Shaders.cfg
@@ -241,5 +241,3 @@ 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
-SpecularDeltaMv_Compute.cs.hlsl -T cs
-SpecularReflectionMv_Compute.cs.hlsl -T cs
diff --git a/Shaders/Include/Common.hlsli b/Shaders/Include/Common.hlsli
@@ -122,7 +122,7 @@ Usage:
  PrintfAt( "a = %f, b = %f, c = %u", a, b, c );
  #endif
 */
-#if( defined( __hlsl_dx_compiler ) && defined( VULKAN ) )
+#if( defined( __hlsl_dx_compiler ) && defined( __spirv__ ) )
  #define PRINTF_AVAILABLE
  #define PrintfAt(...) \
  if( uint( pixelPos.x ) == gPrintfAt.x && uint( pixelPos.y ) == gPrintfAt.y ) \
@@ -198,6 +198,11 @@ static const float3 g_Special8[ 8 ] =
  #define CompareMaterials( m0, m, mask ) 1.0
 #endif
 
+float UnpackViewZ( float z )
+{
+ return abs( z * gViewZScale );
+}
+
 float PixelRadiusToWorld( float unproject, float orthoMode, float pixelRadius, float viewZ )
 {
  return pixelRadius * unproject * lerp( viewZ, 1.0, abs( orthoMode ) );