[ ] number of samples (blocker 25 and filter 64) seems to yield satisfying results
[ ] is it correct to only do texture2DCompare in PCSSPCFFilter (not sure doing lerp or any kernel pcf helping since we’re already “sampling around” within the golden vogel disk; performance is dreadful if we should to do that and we'd need to use less filter sample than 64).
[ ] PCSS texture cube for point lights doesn’t have biased z
[ ] sampling pattern very visible in specular reflection when light intensity high. Higher pixel ratio and shadow map size can help reduce a bit, at a perf cost.
Also needing pex-renderer tweaks:
[ ] it is still dependent on shadowmap size
[ ] sportlight penumbra should be relative to light space z (.bulbRadiusLs / zLight;)
[ ] if bulbRadius is too big, we see the edge of the light depth map because we’re sampling outside its bbox (it is normal, unless we find a heuristic to make lightbox view size dependent on that bulbRadius?)
Also needing pex-renderer tweaks:
.bulbRadiusLs / zLight;)