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 a cycles.csv target which recomputes only if the inputs or the xdsl commit change #61

Merged
merged 9 commits into from
Nov 3, 2023

Conversation

superlopuh
Copy link
Collaborator

@superlopuh superlopuh commented Nov 3, 2023

This is a step towards caching more of the traces being computed, I think it's a shame for us to wait for a long time to recompute things that don't change, hopefully by transitioning to a make-based system we'll spend a lot less time waiting for results.

This PR doesn't commit the csv files generated for me locally, but I think we should consider adding them, and changin the CI to only check that there is no work to be done when running make results/combined.csv on the server.

@superlopuh
Copy link
Collaborator Author

Not sure why the CI is failing, isn't xdsl a submodule located at /src/xdsl for the CI also?

@compor
Copy link
Contributor

compor commented Nov 3, 2023

having a look

snitch/Makefile.rules Outdated Show resolved Hide resolved
@superlopuh
Copy link
Collaborator Author

superlopuh commented Nov 3, 2023

On second thought, here are some csvs generated, I think it's worth discussing committing them now

Copy link

github-actions bot commented Nov 3, 2023

kernel size version cycles
relu 16x16xf64 baseline.x 1339
relu 16x16xf64 ssr.x 846
relu 16x16xf64 ssr_frep_unroll.x 334
relu 16x16xf64 snitch_stream.x 322
relu 16x16xf64 linalg.x 1337
relu 16x16xf64 ssr_frep.x 327
dsum 8x16xf32 baseline.x 1202
dsum 8x16xf32 ssr2d.x 273
dsum 8x16xf32 ssr1d_frep1d.x 187
dsum 8x16xf32 scf.x 1227
dsum 8x16xf32 ssr1d.x 253
dsum 8x16xf32 linalg.x 1089
dsum 8x16xf32 noalias.x 1202
matmul 8x8xf64 baseline.x 4230
matmul 8x8xf64 linalg.x 6214

@compor compor self-requested a review November 3, 2023 18:12
@compor
Copy link
Contributor

compor commented Nov 3, 2023

Maybe I misunderstood, but do you mean committing these specific CSVs per kernel directory?
If yes, is this in order to detect changes?

Nvm, I'm tired, you discuss this in the main PR message; I agree and we had discussed yesterday in person.

Moreover, I can simplify the results collection if we keep them.

@compor
Copy link
Contributor

compor commented Nov 3, 2023

I'll start work on this while waiting for others to chip in.

@superlopuh
Copy link
Collaborator Author

ah nice if you feel like doing this that would be great. I think it's less of a priority than getting the other kernels in but I think it will let us iterate more quickly once we do have it

@superlopuh
Copy link
Collaborator Author

I think meanwhile I'll merge it it's additive at the moment and lets me test the other branches more easily

@superlopuh superlopuh merged commit 2f43517 into main Nov 3, 2023
2 checks passed
@superlopuh superlopuh deleted the sasha/make-csv branch November 3, 2023 23:29
Copy link

github-actions bot commented Nov 3, 2023

kernel size version cycles
relu 16x16xf64 baseline.x 1339
relu 16x16xf64 ssr.x 846
relu 16x16xf64 ssr_frep_unroll.x 334
relu 16x16xf64 snitch_stream.x 322
relu 16x16xf64 linalg.x 1337
relu 16x16xf64 ssr_frep.x 327
dsum 8x16xf32 baseline.x 1202
dsum 8x16xf32 ssr2d.x 273
dsum 8x16xf32 ssr1d_frep1d.x 187
dsum 8x16xf32 scf.x 1227
dsum 8x16xf32 ssr1d.x 253
dsum 8x16xf32 linalg.x 1089
dsum 8x16xf32 noalias.x 1202
matmul 8x8xf64 baseline.x 4230
matmul 8x8xf64 linalg.x 6214

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

Successfully merging this pull request may close these issues.

2 participants