Blender Git Commit Log

Git Commits -> Revision fd0dea5

October 19, 2016, 12:26 (GMT)
Fix Cycles CUDA performance on CUDA 8.0.

Mostly this is making inlining match CUDA 7.5 in a few performance critical
places. The end result is that performance is now better than before, possibly
due to less register spilling or other CUDA 8.0 compiler improvements.

On benchmarks scenes, there are 3% to 35% render time reductions. Stack memory
usage is reduced a little too.

Reviewed By: sergey

Differential Revision: https://developer.blender.org/D2269

Commit Details:

Full Hash: fd0dea585cfcb50a8b056f0d1101505f12b508bc
Parent Commit: 4d14bd1
Committed By: Sergey Sharybin
Lines Changed: +82, -94

20 Modified Paths:

/intern/cycles/kernel/bvh/bvh.h (+9, -8) (Diff)
/intern/cycles/kernel/bvh/bvh_nodes.h (+15, -15) (Diff)
/intern/cycles/kernel/bvh/bvh_traversal.h (+8, -13) (Diff)
/intern/cycles/kernel/bvh/bvh_types.h (+1, -1) (Diff)
/intern/cycles/kernel/closure/bsdf.h (+2, -2) (Diff)
/intern/cycles/kernel/closure/bsdf_ashikhmin_shirley.h (+1, -1) (Diff)
/intern/cycles/kernel/closure/bsdf_microfacet.h (+1, -1) (Diff)
/intern/cycles/kernel/closure/bsdf_microfacet_multi.h (+20, -20) (Diff)
/intern/cycles/kernel/closure/bsdf_microfacet_multi_impl.h (+2, -2) (Diff)
/intern/cycles/kernel/closure/bssrdf.h (+3, -3) (Diff)
/intern/cycles/kernel/geom/geom_curve.h (+3, -3) (Diff)
/intern/cycles/kernel/kernel_accumulate.h (+1, -7) (Diff)
/intern/cycles/kernel/kernel_compat_cuda.h (+1, -0) (Diff)
/intern/cycles/kernel/kernel_compat_opencl.h (+1, -0) (Diff)
/intern/cycles/kernel/kernel_path.h (+3, -3) (Diff)
/intern/cycles/kernel/kernel_path_branched.h (+2, -2) (Diff)
/intern/cycles/kernel/kernel_random.h (+3, -9) (Diff)
/intern/cycles/kernel/kernel_shadow.h (+3, -3) (Diff)
/intern/cycles/kernel/osl/osl_services.cpp (+1, -1) (Diff)
/intern/cycles/util/util_types.h (+2, -0) (Diff)
By: Miika HämäläinenLast update: Nov-07-2014 14:18MiikaHweb | 2003-2021