This reduces considerably the time spent for the integration
phase:
Volumetric pass only on Intel HD 4600: 54ms -> 14ms
Differential D5740
EEVEE: Volume: Speedup: Use Image load store extension Authored by Clément Foucault (fclem) on Sep 10 2019, 2:18 PM. Tags None Subscribers None Tokens
Details This reduces considerably the time spent for the integration Volumetric pass only on Intel HD 4600: 54ms -> 14ms
Diff Detail
Event TimelineComment Actions Sounds great. I am getting some test failures (with Quadro RTX 5000 on Ubuntu). There are a dozen or so tests already failing in master, I'm not sure if those are platform specific differences or if you see the same thing. But ignoring those, with this patch volumes look quite a bit denser too. The result looks more accurate, for example in the "principled blackbody" test the volume is now properly appearing in front of the sphere as dense as Cycles. I'm not sure if that indicates a bug in the non-optimized code, or if it's intentional the optimized code is more accurate as well? Comment Actions OMG that's spot on! This was an off by one error in the previous code. Fix will be committed separately. |