I guess we were discussing "does this fix #xxx" and that triggered gitea to close the report.
@Sergey I was suggesting to add something like alpha2 = max(alpha2, 5e-8f); inside bsdf_D().
I can confirm the banding artefacts with Beckmann. Not sure if we can improve the precision in this case, otherwise it's fine to clamp for Beckmann when evaluating, Beckmann's tail is not obvious…
Should the value be per-sample or total?
Total value makes more sense to me.
What unit should we use for time values?
I think we can have a post-processing step to normalize the…
I have tested it myself on Mac and it looks good, it would be nice if other reviewers who has access to more hardwares could test what threshold is good.
I should say ctx->max_t was not yet computed.
Because we have not looped through the array at this point, so we don't have ctx->max_t available. I just initialzed it to ray->tmax to ensure…
0 or any small value is fine, is just an initialization before running the loop, I can change to 0 if that seems more understandable.
Somehow fmod() is quite slow on my CPU. Rendering the attached scene gives
Is clicking Render -> Bake -> Bake from the attached scene enough to reproduce the crash? In that case I can reproduce in 4.2 with the same crash log, but not with 4.3 f6e6f2b2d222, not even…
This part of cleanup can be separated too.