Ray benchmark miniapp - #238
Conversation
Waqar-ukaea
added
the
Miniapps/Tooling
pshriwise
left a comment
pshriwise
left a comment
There was a problem hiding this comment.
Thanks @Waqar-ukaea! It will be nice to have a single tool to do some of our benchmarking work.
|
@pshriwise I have gone ahead and made the changes you suggested and fleshed out some of them with some extra functionality. The only thing is with switching to using |
pshriwise
left a comment
pshriwise
left a comment
There was a problem hiding this comment.
Just one final suggestion, but regardless left but regardless I think this is good to go!
|
@pshriwise I think we are good on this one now. I may put up an additional PR that adds a python script to drive multiple runs of the benchmark too once I am at the point where I want to start doing proper comparison runs with cuBQL etc. I'll also close #194 in favour of this one now too. |
pshriwise
left a comment
pshriwise
left a comment
There was a problem hiding this comment.
Looks good! Thank you @Waqar-ukaea!
2 participants
This PR adds a simple miniapp to compare the total ray throughput between backends within XDG.
I previously had a PR up for something similar in #194 centred around GPRT. The difference here in this new PR is that the benchmark miniapp is being added independent of any GPU backend (GPRT or cuBQL) and only works with embree along with some minor changes/cleanup to get it to a state where it is much easier to extend to also work with cuBQL.
One immediate question is whether or not an explicit check for correctness is required in this benchmark. Initially I didn't want to do anything that checks for correct rays as it would slow down the benchmark but I wonder if that would be useful.