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

Cycles: Reduce amount of malloc() calls from the kernel

Browse Source 
This commit makes it so malloc() is only happening once per volume and
once per transparent shadow query (per thread), improving scalability of
the code to multiple CPU cores.

Hard to measure this with a low-bottom i7 here currently, but from quick
tests seems volume sampling gave about 3-5% speedup.

The idea is to store allocated memory in kernel globals, which are per
thread on CPU already.

Reviewers: dingto, juicyfruit, lukasstockner97, maiself, brecht

Reviewed By: brecht

Subscribers: Blendify, nutel

Differential Revision: https://developer.blender.org/D1996
This commit is contained in:
Sergey Sharybin
2016-05-17 12:30:46 +02:00
parent 2433a537fa
commit 7b356a8565
4 changed files with 83 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
intern/cycles/kernel/kernel_shadow.h
View File

@@ -59,14 +59,20 @@ ccl_device_inline bool shadow_blocked(KernelGlobals *kg, PathState *state, Ray *
/* intersect to find an opaque surface, or record all transparent surface hits */
Intersection hits_stack[STACK_MAX_HITS];
Intersection *hits = hits_stack;
uint max_hits = kernel_data.integrator.transparent_max_bounce - state->transparent_bounce - 1;
const int transparent_max_bounce = kernel_data.integrator.transparent_max_bounce;
uint max_hits = transparent_max_bounce - state->transparent_bounce - 1;
/* prefer to use stack but use dynamic allocation if too deep max hits
* we need max_hits + 1 storage space due to the logic in
* scene_intersect_shadow_all which will first store and then check if
* the limit is exceeded */
if(max_hits + 1 > STACK_MAX_HITS)
hits = (Intersection*)malloc(sizeof(Intersection)*(max_hits + 1));
if(max_hits + 1 > STACK_MAX_HITS) {
if(kg->transparent_shadow_intersections == NULL) {
kg->transparent_shadow_intersections =
(Intersection*)malloc(sizeof(Intersection)*(transparent_max_bounce + 1));
}
hits = kg->transparent_shadow_intersections;
}
uint num_hits;
blocked = scene_intersect_shadow_all(kg, ray, hits, max_hits, &num_hits);
@@ -147,14 +153,8 @@ ccl_device_inline bool shadow_blocked(KernelGlobals *kg, PathState *state, Ray *
*shadow = throughput;
if(hits != hits_stack)
free(hits);
return is_zero(throughput);
}
/* free dynamic storage */
if(hits != hits_stack)
free(hits);
}
else {
Intersection isect;