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.

Sign up for GitHub

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

Jump to bottom

ROCM v4.1 introduces performance regression #319

Closed
stgeke opened this issue Jun 7, 2021 · 1 comment
Closed

ROCM v4.1 introduces performance regression #319

stgeke opened this issue Jun 7, 2021 · 1 comment

Comments

@stgeke
Copy link
Collaborator

stgeke commented Jun 7, 2021 •
edited
Loading

In the ROCm v4.1 release, the max blocks per thread is changed from 256 to 1024.
This introduced a serious performance regression.

The ROCm release notes recommend the following workaround: add --gpu-max-threads-per-block=256 to the compilation options for HIP programs.

This issue will be fixed as soon as we specify launch bounds for every kernel (requires #89).
@stgeke
Copy link
Collaborator Author

stgeke commented Sep 22, 2021

Fixed in 7a55f45

@stgeke stgeke closed this as completed Sep 22, 2021
stgeke pushed a commit that referenced this issue Nov 6, 2021
@MalachiTimothyPhillips
Introduce createMeshMG + fix allocation of o_lambdaPfloat (#319)
e9120f8 
* Remove erroneous allocation of o_lambdaPfloat -- these are set up in MGLevel
* Change `ellipticUpdateLambda` -> `ellipticMultiGridUpdateLambda` to improve readability
* Move copyDfloatToPfloat and copyPfloatToDfloat to platform
* Use `createMeshMG` to create the mesh object associated with the elliptic level.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@stgeke