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

Add fix for slow GPU aware MPI to JSC booster script #1133

Conversation

AlexanderSinn
Copy link
Member

@AlexanderSinn AlexanderSinn commented Jul 3, 2024

In my testing so far, setting this environment variable greatly improves performance of GPU-aware MPI on the booster. 

export UCX_CUDA_COPY_REG_WHOLE_ALLOC=on
  • Small enough (< few 100s of lines), otherwise it should probably be split into smaller PRs
  • Tested (describe the tests in the PR description)
  • Runs on GPU (basic: the code compiles and run well with the new module)
  • Contains an automated test (checksum and/or comparison with theory)
  • Documented: all elements (classes and their members, functions, namespaces, etc.) are documented
  • Constified (All that can be const is const)
  • Code is clean (no unwanted comments, )
  • Style and code conventions are respected at the bottom of https://github.com/Hi-PACE/hipace
  • Proper label and GitHub project, if applicable

@AlexanderSinn AlexanderSinn added documentation Improvements or additions to documentation GPU Related to GPU acceleration performance optimization, benchmark, profiling, etc. Parallelization Longitudinal and transverse MPI decomposition labels Jul 3, 2024
Copy link
Member

@MaxThevenet MaxThevenet left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh wow, thanks for fixing this!!

@MaxThevenet MaxThevenet merged commit ac24074 into Hi-PACE:development Jul 16, 2024
10 checks passed
@MaxThevenet
Copy link
Member

Did you exchange with JSC on this? Any chance this could be useful feedback to them?

@AlexanderSinn
Copy link
Member Author

Yes I talked to JSC support a lot about this. However, we agreed that the setting might not improve performance (and could worsen it) for most other codes.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
documentation Improvements or additions to documentation GPU Related to GPU acceleration Parallelization Longitudinal and transverse MPI decomposition performance optimization, benchmark, profiling, etc.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants