Improved Area Light Support for Hair #6157

johnpars · 2021-10-27T20:05:13Z

Purpose of this PR

This PR adds an improved method for supporting rectangular area lights for hair.

Due to lack of support for LTC with anisotropic models, we have up until now fallen back to the GGX LTC for hair. This PR tries to improve on this by instead deferring to a "Representative Point" as a stop-gap method for both the Approximate (Kajiya) and Physical (Marschner) based model until further research is given for LTC.

The result of this work is that we should now be much closer to our references.

Below is some results for our approximation for Physical.

And here is some results for our approximation for Approximate.
[TODO IMAGE]

Performance

Performance-wise this seems like a slight win over LTC for Marschner model and a pretty significant win for Kajiya. Additionally, I measured the MRP method with and without the brute force solid angle computation and did find a measurable difference, but in both cases these still resulted in lower times than the LTC, so I think this is good justification to stick with the brute force solid angle route.

Performance metrics were captured from a standalone Windows built with an NVIDIA Quadro RTX 8000, at a 3840x2160 resolution, 8x MSAA.

Model	Area Light Method	Forward Opaque Time Per-Groom (ms)
Marschner (+Multiple Scattering)	MRP, Brute Force Solid Angle	1.31
Kajiya-Kay	MRP, Brute Force Solid Angle	0.97
Marschner (+Multiple Scattering)	MRP, Optimized Solid Angle	1.27
Kajiya-Kay	MRP, Optimized Solid Angle	0.89
Marschner (GGX)	LTC	1.57
Kajiya-Kay (GGX)	LTC	1.43

Test scenario scene. Left Kajiya, right Marschner.

Testing status

I manually verified the following behaviors:

Varying area light shapes and orientations are in good agreement with the reference.
Cookie support.
Barn door support.
Shadow support.

Updated the hair area light graphics test to reflect the improved area light support, as well as added more coverage for the marschner model. Local tests are passing + launched yamato.

Comments to reviewers

I ended up finding a reasonable way to approximate Kajiya with this approach. It doesn't exactly use MRP, but instead just treats the area light as a normal analytic light and modulates the roughness automatically based on the solid angle. It required some tuning to make sure we are in good agreement with the reference.

…sphere scattering.

…mals.

…ipleScattering/HairMultipleScattering.hlsl

github-actions · 2021-10-27T20:05:30Z

Hi! This comment will help you figure out which jobs to run before merging your PR. The suggestions are dynamic based on what files you have changed.
Link to Yamato: https://unity-ci.cds.internal.unity3d.com/project/902/
Search for your PR branch using the search bar at the top, then add the following segment(s) to the end of the URL (you may need multiple tabs depending on how many packages you change)

HDRP
/jobDefinition/.yamato%2Fall-hdrp.yml%23PR_HDRP_trunk
With changes to HDRP packages, you should also run
/jobDefinition/.yamato%2Fall-lightmapping.yml%23PR_Lightmapping_trunk

Depending on the scope of your PR, you may need to run more jobs than what has been suggested. Please speak to your lead or a Graphics SDET (#devs-graphics-automation) if you are unsure.

johnpars · 2021-10-27T20:56:17Z

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.cs

@@ -161,6 +161,9 @@ public struct BSDFData

            public float perceptualRoughnessRadial;

+            // Normalization factor for area lights.
+            public Vector3 distributionNormalizationFactor;


Added this new term that area lights compute to normalize the distribution. This is necessary for energy conservation since we are moving the position of the light.

johnpars · 2021-10-27T20:57:06Z

com.unity.render-pipelines.high-definition/Runtime/Lighting/LightEvaluation.hlsl

@@ -318,7 +327,7 @@ SHADOW_TYPE EvaluateShadow_Directional( LightLoopContext lightLoopContext, Posit
 #include "Packages/com.unity.render-pipelines.high-definition/Runtime/Lighting/PunctualLightCommon.hlsl"

 float4 EvaluateCookie_Punctual(LightLoopContext lightLoopContext, LightData light,
-                               float3 lightToSample)
+                               float3 lightToSample, float lod = 0)


Slight adjustment to punctual cookie sampling to take an optional mip level. MRP lights will take advantage of this to select the appropriate level (instead of defaulting to 0 like normal analytic lights).

johnpars · 2021-10-27T20:58:38Z

com.unity.render-pipelines.high-definition/Runtime/Lighting/LightEvaluation.hlsl

-    shadow = min(shadow, NdotL > 0.0 ? GetContactShadow(lightLoopContext, light.contactShadowMask, light.isRayTracedContactShadow) : 1.0);
+{
+    // In certain cases (like hair) we allow to force the contact shadow sample.
+    #ifdef LIGHT_EVALUATION_CONTACT_SHADOW_DISABLE_NDOTL


For non-shadow mapped lights we add this new light evaluation flag that allows hair to override the NdotL which cancels to contact shadow sample. This is important for Marschner since it's not exactly cosine weighted by NdotL due to the transmission lobe

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl

johnpars · 2021-10-27T21:01:35Z

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl

+
+        float3 L;
+
+        if (HasFlag(bsdfData.materialFeatures, MATERIALFEATUREFLAGS_HAIR_MARSCHNER))


MRP for Marschner

johnpars · 2021-10-27T21:02:00Z

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl

+
+            bsdfData.distributionNormalizationFactor = sqrt(alpha / alphaPrime);
+        }
+        else


Light Center + Solid Angle modulated roughness for Kajiya

johnpars · 2021-10-27T21:03:03Z

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl

@@ -1063,7 +1291,11 @@ DirectLighting EvaluateBSDF_Area(LightLoopContext lightLoopContext,
    }
    else
    {
+#if 0


Should we add a toggle to the shader graph that allows a user to choose between the default LTC/GGX or improved method? Or just keep LTC always disabled for now.

I think this is probably quite robust and strictly closer to reference in terms of actual shading but also overall brightness.

My only worry is with elongated and skewed area lights (one part of the elongated area light being much closer to the shaded surface) and with lights with cookies. These might result in brightness that's quite far off the reference, and be jarring if the hair is e.g. contrasting with forehead that's correctly lit with LTC. If in those cases GGX/LTC on hair gives a less incorrect result, I'd see that as the only reason to keep it as an option. But maybe it's sufficiently worse overall, that it wouldn't be a practical choice anyway?

We can't afford both in the code for perf reasons anyway (vgpr). So totally fine with disabling LTC

johnpars · 2021-10-27T21:03:41Z

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/HairPathTracing.hlsl

@@ -22,7 +22,7 @@ void ProcessBSDFData(PathIntersection pathIntersection, BuiltinData builtinData,
 {
    // NOTE: Currently we don't support ray-aligned ribbons in the acceleration structure, so our only H-calculation routines
    // are either stochastic or derived from a tube intersection.
-#if 1
+#if 0


Decided to disable near field path traced model for now, just in case users provide hair geometry with bad strand normals- they will still get correct results for far field.

johnpars · 2021-10-27T21:04:02Z

...pelines.high-definition/Runtime/Material/Hair/MultipleScattering/HairMultipleScattering.hlsl

@@ -27,7 +27,7 @@ float DecodeHairStrandCount(float3 L, float4 strandCountProbe)
        HALF_SQRT_3_DIV_PI * L.x
    );

-    return abs(dot(strandCountProbe, Ylm));
+    return max(dot(strandCountProbe, Ylm), 0);


Fixing a mistake I made with SH anti-ring.

...pelines.high-definition/Runtime/Material/Hair/MultipleScattering/HairMultipleScattering.hlsl

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl

robcupisz · 2021-10-28T13:13:33Z

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl

@@ -1063,7 +1291,11 @@ DirectLighting EvaluateBSDF_Area(LightLoopContext lightLoopContext,
    }
    else
    {
+#if 0


I think this is probably quite robust and strictly closer to reference in terms of actual shading but also overall brightness.

My only worry is with elongated and skewed area lights (one part of the elongated area light being much closer to the shaded surface) and with lights with cookies. These might result in brightness that's quite far off the reference, and be jarring if the hair is e.g. contrasting with forehead that's correctly lit with LTC. If in those cases GGX/LTC on hair gives a less incorrect result, I'd see that as the only reason to keep it as an option. But maybe it's sufficiently worse overall, that it wouldn't be a practical choice anyway?

...pelines.high-definition/Runtime/Material/Hair/MultipleScattering/HairMultipleScattering.hlsl

robcupisz · 2021-10-28T13:20:27Z

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl

+            // Construct the most representative direction.
+            // We must consider multiple specular lobes, on the backward (R, TRT) and forward (TT) scattering hemisphere.
+            // For the backward hemisphere we handle R and TRT similarly, and use the MRP result (based on the "fake" normal just like how we use it for IBL).
+            // For the forward hemisphere we need to approximate harsher. We can get away with falling back to the light center and modifying the roughness.


My concern here is that elongated lights with one end closer to the shaded point will appear too dark. I suppose I should try it out in action :)

What was the reasoning behind going with the solid-angle-based roughness tweak as opposed to MRP? Were the results worse with MRP?

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl

…arily add back rect ref

kecho

Run yamato, and add a comment on the removal of that division by PI :) ensure yamato is green!

...pelines.high-definition/Runtime/Material/Hair/MultipleScattering/HairMultipleScattering.hlsl

…of the divide by PI

…tudinal distribution

johnpars · 2021-11-02T03:04:28Z

Re-running Yamato. All of these failures seem like instabilities (including the smoke test one, still persisting after I merged master). I also provided QA a test project to validate this PR.

johnpars · 2021-11-02T15:44:21Z

Looks like we are 'green' test-wise, the failures are known instabilities.

sebastienlagarde · 2021-11-04T08:51:39Z

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl

 #endif

 #include "Packages/com.unity.render-pipelines.high-definition/Runtime/Lighting/LightEvaluation.hlsl"
 #include "Packages/com.unity.render-pipelines.high-definition/Runtime/Material/MaterialEvaluation.hlsl"
 #include "Packages/com.unity.render-pipelines.high-definition/Runtime/Lighting/SurfaceShading.hlsl"

+float3 RayPlaneIntersect(in float3 rayOrigin, in float3 rayDirection, in float3 planeOrigin, in float3 planeNormal)


Those set of function either alredy exist or should be moved into our geometry.hlsl common file

sebastienlagarde · 2021-11-04T09:19:24Z

@qa: Merging to 22.1, the coverage by John is already great.

* Initial commit MRP implementation * Barn door application * Add dominant marschner lobe direction skeleton * Finalize rect light MRP for marschner, handle forward + backward hemisphere scattering. * Force pathtracer to far-field in case users use geometry with bad normals. * com.unity.render-pipelines.high-definition/Runtime/Material/Hair/MultipleScattering/HairMultipleScattering.hlsl * Fix SH ringing for multiple scattering * Add trivial support for cookies * Add Kajiya support for area light approximation (non-LTC) * remove Line area light MRP for now * Remove area rect reference * Update area light hair test scene and reference images * Select cookie mip based on solid angle, rename some functions, temporarily add back rect ref * Remove the useless div-by-1 and add a comment to explain the removal of the divide by PI * Add a comment to further explain the normalization we do on the longitudinal distribution * Remove the ref once again * Update test images for metal and vulkan * Add a zero-div guard to silence compiler warning

JMargevics

Tested thoroughly this on Windows platforms and found no visual issues both with strands and hair cards. ✔👍

* Initial commit MRP implementation * Barn door application * Add dominant marschner lobe direction skeleton * Finalize rect light MRP for marschner, handle forward + backward hemisphere scattering. * Force pathtracer to far-field in case users use geometry with bad normals. * com.unity.render-pipelines.high-definition/Runtime/Material/Hair/MultipleScattering/HairMultipleScattering.hlsl * Fix SH ringing for multiple scattering * Add trivial support for cookies * Add Kajiya support for area light approximation (non-LTC) * remove Line area light MRP for now * Remove area rect reference * Update area light hair test scene and reference images * Select cookie mip based on solid angle, rename some functions, temporarily add back rect ref * Remove the useless div-by-1 and add a comment to explain the removal of the divide by PI * Add a comment to further explain the normalization we do on the longitudinal distribution * Remove the ref once again * Update test images for metal and vulkan * Add a zero-div guard to silence compiler warning

* Increasing HDRP fined pruned light tile count to 63. #5771 * Fix Upscaled Film Grain & Dither in HDRP (#6111) When using particular types of upscaling, (CAS, FSR, Catmull) the film grain quality produced by HDRP was lower than expected. This was caused by errors in the UV scale calculations for the grain texture. This change fixes the issue by modifying the scale calculations to use the final viewport size rather than the size of the current scaled/unscaled viewport. This works because film grain is always handled by "FinalPass" which always renders at final viewport size anyways. This change also fixes a few cases where the current resolution group would have been incorrect when CAS or FSR were being used as the upscaling method. * Update CHANGELOG.md * Fix regression that was introduced in a previous PR on the diffuse denoiser #6119 * Fix Light Loop Variant Warnings (1372256) #6135 * Bump the HDRP Template IET Framework Package Version #6134 * Update ray tracing and path tracing docs regarding custom interpolator limitations (#6126) * [HDRP] Disable DoF for orthographic cameras #6124 * update occlusion radius for directional light sample (#6154) Co-authored-by: Sean Puller <sean.puller@unity3d.com> * Clean up Public API Documentation #6151 * Fix Sky override not taken into account #6127 * [Fogbugz # 1372245] Hdrp/drs fixing pyramid blur #6136 * Improved Area Light Support for Hair (#6157) * Initial commit MRP implementation * Barn door application * Add dominant marschner lobe direction skeleton * Finalize rect light MRP for marschner, handle forward + backward hemisphere scattering. * Force pathtracer to far-field in case users use geometry with bad normals. * com.unity.render-pipelines.high-definition/Runtime/Material/Hair/MultipleScattering/HairMultipleScattering.hlsl * Fix SH ringing for multiple scattering * Add trivial support for cookies * Add Kajiya support for area light approximation (non-LTC) * remove Line area light MRP for now * Remove area rect reference * Update area light hair test scene and reference images * Select cookie mip based on solid angle, rename some functions, temporarily add back rect ref * Remove the useless div-by-1 and add a comment to explain the removal of the divide by PI * Add a comment to further explain the normalization we do on the longitudinal distribution * Remove the ref once again * Update test images for metal and vulkan * Add a zero-div guard to silence compiler warning * Update CHANGELOG.md * [HDRP][Path Tracing] Exposed 3 methods related to path tracing and accumulation. #6197 * [HDRP] Fix infitnite material import loop materials #6185 * fixed typo: users -> uses (#6162) * Don't blur if nothing to blur (saves memory due to RG) (#6183) * Fix local fog volumes z axis calculation #6068 * fixed passive voice (#6092) * Fixed grammar errors * fixed passive voice in decal doc * Update Decal.md * Update Decal.md * Update Decal.md * AxF Raytracing: fix performance mode distillation that was broken and removed in #9d6db83478c213bc066e5f728d94b7faa6662543 (#6104) * Add scale parameter in eye shader #6132 * updated Override Fog page (#6200) * updated fog override screenshot * changed Depth Extent to Volumetric Fog Distance * Fix missing pragma (#6204) * [HDRP] Fix custom pass motion vector texture access #6205 * Fix typo in MipBias tooltip in HDRenderPipelineUI.Skin (#6210) * Replace gizmos FindObjectsOfType with list registrations to avoid constant slowdown in large scenes. Merged editor guards blocks. (#6168) * [HDRP] Fix APV tooltip for Geometry Distance Offset (#6150) * Update tooltip * Update ProbeVolumeUI.Skin.cs * Fixed the point distribution for the diffuse denoiser sometimes not being properly intialized. #6194 * Fixed the bad blending between the sun and the clouds (case 1373282). #6193 * Fix shadow mask fade and optimize it at same time (#6083) * [HDRP][Path Tracing] Minor fix for SSS when combined with fog (#6215) * Minor fix for SSS + fog. * Comments. * Update 1401_HairGraph_Area_Light.png * update screenshots Co-authored-by: Kleber Garcia <kleber.garcia@unity3d.com> Co-authored-by: gmitrano-unity <89797527+gmitrano-unity@users.noreply.github.com> Co-authored-by: anisunity <42026998+anisunity@users.noreply.github.com> Co-authored-by: John Parsaie <johnpa@unity3d.com> Co-authored-by: Pavlos Mavridis <pavlos.mavridis@unity3d.com> Co-authored-by: Sean Puller <43151199+SeanPuller@users.noreply.github.com> Co-authored-by: Sean Puller <sean.puller@unity3d.com> Co-authored-by: Adrien de Tocqueville <adrien.tocqueville@unity3d.com> Co-authored-by: Emmanuel Turquin <emmanuel@turquin.org> Co-authored-by: Antoine Lelievre <antoinel@unity3d.com> Co-authored-by: emilybrown1 <88374601+emilybrown1@users.noreply.github.com> Co-authored-by: FrancescoC-unity <43168857+FrancescoC-unity@users.noreply.github.com> Co-authored-by: RemyUnity <32760367+RemyUnity@users.noreply.github.com> Co-authored-by: slunity <37302815+slunity@users.noreply.github.com> Co-authored-by: James-Burr-Unity <61244495+James-Burr-Unity@users.noreply.github.com> Co-authored-by: Torbjorn Laedre <271210+tlaedre@users.noreply.github.com>

johnpars added 10 commits October 14, 2021 23:51

Initial commit MRP implementation

c1c1b2b

Barn door application

94cc3d6

Add dominant marschner lobe direction skeleton

24f1138

Finalize rect light MRP for marschner, handle forward + backward hemi…

9f7fac6

…sphere scattering.

Force pathtracer to far-field in case users use geometry with bad nor…

fca4625

…mals.

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Mult…

f91ce44

…ipleScattering/HairMultipleScattering.hlsl

Fix SH ringing for multiple scattering

3de6891

Merge branch 'master' into HDRP/hair-mrp-area-lights

7f1b311

Add trivial support for cookies

1d4625c

Add Kajiya support for area light approximation (non-LTC)

f0c4b2d

johnpars added the HDRP label Oct 27, 2021

johnpars added 3 commits October 27, 2021 16:28

remove Line area light MRP for now

62d588c

Remove area rect reference

800ec98

Update area light hair test scene and reference images

b3f142a

johnpars requested review from a team, sebastienlagarde, kecho and robcupisz October 27, 2021 20:46

johnpars commented Oct 27, 2021

View reviewed changes

com.unity.render-pipelines.high-definition/Runtime/Material/Hair/Hair.hlsl Show resolved Hide resolved

johnpars commented Oct 27, 2021

View reviewed changes

...pelines.high-definition/Runtime/Material/Hair/MultipleScattering/HairMultipleScattering.hlsl Outdated Show resolved Hide resolved

robcupisz reviewed Oct 28, 2021

View reviewed changes

Select cookie mip based on solid angle, rename some functions, tempor…

993388f

…arily add back rect ref

kecho suggested changes Nov 1, 2021

View reviewed changes

...pelines.high-definition/Runtime/Material/Hair/MultipleScattering/HairMultipleScattering.hlsl Outdated Show resolved Hide resolved

johnpars added 6 commits November 1, 2021 12:37

Remove the useless div-by-1 and add a comment to explain the removal …

94454d8

…of the divide by PI

Add a comment to further explain the normalization we do on the longi…

f44de0d

…tudinal distribution

Remove the ref once again

93096ec

Update test images for metal and vulkan

8283223

Add a zero-div guard to silence compiler warning

fc8b128

Merge branch 'master' into HDRP/hair-mrp-area-lights

4d9ea1d

robcupisz approved these changes Nov 2, 2021

View reviewed changes

kecho approved these changes Nov 2, 2021

View reviewed changes

johnpars marked this pull request as ready for review November 2, 2021 14:22

github-actions bot added the testing label Nov 2, 2021

sebastienlagarde reviewed Nov 4, 2021

View reviewed changes

sebastienlagarde approved these changes Nov 4, 2021

View reviewed changes

sebastienlagarde added the needs-backport-2021.2 label Nov 4, 2021

sebastienlagarde merged commit d5bdfdb into master Nov 4, 2021

sebastienlagarde deleted the HDRP/hair-mrp-area-lights branch November 4, 2021 09:18

JMargevics reviewed Nov 5, 2021

View reviewed changes

sebastienlagarde added backported-2021.2 and removed needs-backport-2021.2 labels Nov 7, 2021

sebastienlagarde mentioned this pull request Nov 7, 2021

[HDRP][Backport 21.2] Backport various PRs #6231

Merged


		float3 L;

		if (HasFlag(bsdfData.materialFeatures, MATERIALFEATUREFLAGS_HAIR_MARSCHNER))

Improved Area Light Support for Hair #6157

Improved Area Light Support for Hair #6157

Uh oh!

Conversation

johnpars commented Oct 27, 2021 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose of this PR

Performance

Testing status

Comments to reviewers

Uh oh!

github-actions bot commented Oct 27, 2021

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

kecho left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

johnpars commented Nov 2, 2021

Uh oh!

johnpars commented Nov 2, 2021

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sebastienlagarde commented Nov 4, 2021

Uh oh!

JMargevics left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

johnpars commented Oct 27, 2021 •

edited

Loading