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 performance reference for important matmul kernels #642

Open
wants to merge 5 commits into
base: main_perf
Choose a base branch
from

Conversation

zhanglx13
Copy link

This PR adds a database of matmul configs that we want to keep track of their performance.
This PR also adds a reference performance collected on smc300x-ccs-aus-GPUF292 (MI300X).
The database and reference are serving two purposes:

  1. Performance tracking. We will update the reference on a regular basis as a performance tracking record.
  2. Regression detection. We will run benchmark on the database on a daily basis to catch any regression.

@xiaohuguo2023
Copy link
Member

should we add the other perf-kernels as well, eg, FA fwd kernel ?

Copy link
Member

Choose a reason for hiding this comment

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

what are the purposes of this database file ? are they used to benchmark against the ref.csv ? if that's the case, what if the parameters changed, e.g GROUP_SIZE_M change from 4 to 8 ?

Copy link
Member

Choose a reason for hiding this comment

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

we may need other database file format if this is a daily/per commit tasks ?

Copy link
Author

Choose a reason for hiding this comment

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

This database.yaml is used as the current best perf config. If you have an optimization that can improve the best perf number and requires a different config, we should update the database.
The main purpose is to catch regression.

@zhanglx13
Copy link
Author

should we add the other perf-kernels as well, eg, FA fwd kernel ?

Yes we should. I'll add FA related kernels later.

@zhanglx13 zhanglx13 force-pushed the perf_ref branch 4 times, most recently from db1424e to 089b464 Compare October 1, 2024 03:33
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