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

[DistConv] MIOpen path through convolution algorithm selection #101

Open
benson31 opened this issue Sep 1, 2023 · 2 comments
Open

[DistConv] MIOpen path through convolution algorithm selection #101

benson31 opened this issue Sep 1, 2023 · 2 comments
Assignees
@benson31
Labels
HIP/ROCM Related to HIP/ROCm

Comments

@benson31
Copy link
Collaborator

benson31 commented Sep 1, 2023

Currently, any value for the "name" passed to get_{fwd,bwd_{data,filter}}_algorithm() (e.g., here) that is not "AUTOTUNE" will probably fail for one reason or another.

"HEURISTIC" will fail because this isn't supported in MIOpen. Other options will fail because MIOpen requires that their Find APIs be called, and we don't currently incorporate that when the algorithm is not "AUTOTUNE".
@benson31 benson31 self-assigned this Sep 1, 2023
@benson31 benson31 added the HIP/ROCM Related to HIP/ROCm label Sep 1, 2023
@benson31
Copy link
Collaborator Author

benson31 commented Sep 1, 2023

I don't know that a solution to this will be back-ported into DistConv (it's not impossible, but not likely), but I wanted to document this here to ensure ongoing refactors do not hit the same pitfall.

@benson31
Copy link
Collaborator Author

benson31 commented Sep 1, 2023

We should also more robustly support things like "autotune but pick a deterministic algorithm", when possible.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@benson31 benson31

Labels
HIP/ROCM Related to HIP/ROCm
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@benson31