Cr8-xyz/blender
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Fix Cycles CUDA performance on CUDA 8.0.

Browse Source 
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
This commit is contained in:
Brecht Van Lommel
2016-10-02 14:48:39 +02:00
parent 49ad4215ba
commit a3abb020e3
20 changed files with 82 additions and 94 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
intern/cycles/kernel/kernel_shadow.h
View File

@@ -155,7 +155,7 @@ ccl_device_inline bool shadow_blocked(KernelGlobals *kg, ShaderData *shadow_sd,
}
else {
Intersection isect;
blocked = scene_intersect(kg, ray, PATH_RAY_SHADOW_OPAQUE, &isect, NULL, 0.0f, 0.0f);
blocked = scene_intersect(kg, *ray, PATH_RAY_SHADOW_OPAQUE, &isect, NULL, 0.0f, 0.0f);
}
#ifdef __VOLUME__
@@ -205,7 +205,7 @@ ccl_device_noinline bool shadow_blocked(KernelGlobals *kg,
Intersection *isect = &isect_object;
#endif
bool blocked = scene_intersect(kg, ray, PATH_RAY_SHADOW_OPAQUE, isect, NULL, 0.0f, 0.0f);
bool blocked = scene_intersect(kg, *ray, PATH_RAY_SHADOW_OPAQUE, isect, NULL, 0.0f, 0.0f);
#ifdef __TRANSPARENT_SHADOWS__
if(blocked && kernel_data.integrator.transparent_shadows) {
@@ -221,7 +221,7 @@ ccl_device_noinline bool shadow_blocked(KernelGlobals *kg,
if(bounce >= kernel_data.integrator.transparent_max_bounce)
return true;
if(!scene_intersect(kg, ray, PATH_RAY_SHADOW_TRANSPARENT, isect, NULL, 0.0f, 0.0f))
if(!scene_intersect(kg, *ray, PATH_RAY_SHADOW_TRANSPARENT, isect, NULL, 0.0f, 0.0f))
{
#ifdef __VOLUME__
/* attenuation for last line segment towards light */