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 fno-offload-uniform-block flag for rocm5.7 and up #838

Merged
merged 2 commits into from
Aug 9, 2023
Merged

Conversation

illsilin
Copy link
Collaborator

@illsilin illsilin commented Aug 8, 2023

This should take care of the changes in the default compiler settings starting from rocm version 5.7.

@junliume
Copy link
Collaborator

junliume commented Aug 8, 2023

@illsilin this PR is to fix #837

Meanwhile, could you add comment in the code such as:

# No assumption that HIP kernels are launched with uniform block size for backward compatibility
# SWDEV-413293 and https://reviews.llvm.org/D155213

@illsilin
Copy link
Collaborator Author

illsilin commented Aug 8, 2023

OK, sure.

@junliume junliume added the urgency_blocker blocking feature deliverables label Aug 9, 2023
@junliume junliume merged commit 6802611 into develop Aug 9, 2023
@atamazov
Copy link

atamazov commented Aug 9, 2023

@junliume @illsilin I do not recommend adding the new option unconditionally, see ROCm/MIOpen#2307 (comment)

@illsilin
Copy link
Collaborator Author

illsilin commented Aug 9, 2023

I can run the full QA suite with rocm5.7 RC1 and this flag and see if there is any performance degradation.

@atamazov
Copy link

atamazov commented Aug 9, 2023

@illsilin That would be very good, thanks! Please update us with the results.

@illsilin
Copy link
Collaborator Author

illsilin commented Aug 9, 2023

Apparently, the rocm5.7 RC1 is the threshold version so the flag does not get set for it:
hip_version_flat=500723302
We can wait for the RC2 to do this, or I can try to run the tests on a development node with the latest rocm build.

@atamazov
Copy link

atamazov commented Mar 1, 2024

@illsilin FYI this is how we use check_cxx_compiler_flag() in MIOpen (more future-proof solution): ROCm/MIOpen#2741

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

Successfully merging this pull request may close these issues.

3 participants