Summary
r.VolumetricFog.Filtering.Quality 2 (tricubic) samples the wrong depth because h0.z is overwritten.
What Type of Bug are you experiencing?
Rendering (Graphics / Niagara)
Steps to Reproduce
Add an exponential height fog with volumetric fog and a spot light with volumetric scattering.
Set r.VolumetricFog.Jitter 0 so the edge is static.
Compare r.VolumetricFog.Filtering.Quality 0, 1 and 2 at the cone edge near the light.
Expected Result
File: Engine/Shaders/Private/HeightFogCommon.ush, SampleVolumetricFogFiltered(), line 492.
Description: The B-spline setup shared by bicubic and tricubic ends with:
hlsl
h1.xy = min(h1.xy, View.VolumetricFogUVMax);
h0.z = VolumeUV.z;
h0.z = VolumeUV.z is correct for bicubic, which filters only in XY and samples Z linearly at the original depth. The tricubic branch also uses h0.z and h1.z with the g0.z weight: lerp(tex001, tex000, g0.z). So in Z it blends the unfiltered depth with the B-spline offset h1.z, and the Z filter is wrong. In practice, quality 2 looks about the same as trilinear (froxel steps along spot light edges stay visible) or worse, while quality 1 smooths them correctly.
Observed Result
Suggested fix: move h0.z = VolumeUV.z; inside the #elif PERMUTATION_VOLUMETRIC_FOG_FILTERING_QUALITY == 1 branch, so tricubic keeps the computed h0.z.
Affects Versions
5.8
Platform(s)
Windows
Pull Request: https://github.com/EpicGames/UnrealEngine/pull/15392