This resolves some issues with correlation artifacts at higher
sample counts. See T101356
Details
Details
Diff Detail
Diff Detail
- Repository
- rB Blender
Event Timeline
Comment Actions
I'm not totally sure I've done everything here correctly, particularly in terms of handling alloc/free and passing the pattern size around.
Comment Actions
Sorry for nit-picking.
In intern/cycles/kernel/sample/jitter.h on line 19 you have:
/* NUM_TAB_SOBOL_SAMPLES should be a power of two... */
Shouldn't it be NUM_PMJ_SAMPLES since we haven't switched over to tabulated Sobol yet?
Comment Actions
Seems to work well in my tests, including converging to consistent result with test file from T101356.