Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Performance degradation with HIP memory kinds enabled in OFI conduit #28

Open
elliottslaughter opened this issue Mar 27, 2024 · 1 comment

Comments

@elliottslaughter
Copy link
Contributor

elliottslaughter commented Mar 27, 2024

We've had some performance issues that were fixed by rolling back that change. On a HIP system running on a single GPU the performance of GPU kernels was about a factor of 30 slower when we used tracing in the code. Interestingly the code ran with its normal performance as soon as the Legion profiler was switched on. I wonder if it has any side effect that could cause this.

Originally posted by @tukss in #27 (comment)

@elliottslaughter
Copy link
Contributor Author

@tukss I moved this to a new issue so that we can discuss here.

I believe that enabling Legion profiling with HIP will cause Legion to synchronize the event stream so that we can determine when the kernels have finished. @eddy16112 or @seemamirch might have more information.

Why that would then make performance recover, I cannot say.

@bonachea or @PHHargrove might have other ideas to try.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant