Date
1 - 4 of 4
Review: smarter runflags for lazy eval (issue194117)
larry...@...
Reviewers: ,
Description: If you called context->execute() with a "sparse" set of runflags (such as if you were picking the subset of a set of rays that hit objects with the same shaders), it would use those runflags for any unconditionally-run layers, but lazy layer evaluation was done with all points on, which is very wasteful for sparse run sets. This change records the top-level original runflags passed in, and uses those runflags when running upstream layers lazily. Please review this at http://codereview.appspot.com/194117/show Affected files: src/liboslexec/context.cpp src/liboslexec/exec.cpp src/liboslexec/oslexec_pvt.h |
|
aco...@...
|
|
cku...@...
LGTM2
Can't wait to see the difference on those benchmark cases. http://codereview.appspot.com/194117/show |
|
larry...@...
On 2010/01/27 23:22:32, ckulla wrote:
Can't wait to see the difference on those benchmark cases.OK, I committed it, so give it a try. This should throw into even more stark relief the waste associated with sparse runflags. I'll tackle that after "rebind" is done. http://codereview.appspot.com/194117/show |
|