-
Notifications
You must be signed in to change notification settings - Fork 30
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
[CI] Add new shape and with/without ukernel for benchmark tracking #909
Conversation
build_tools/ci/cpu_comparison/run.py
Outdated
512, | ||
512, | ||
4096, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It should be M == 512, N == 4096, K == 512 to correspond to the problem size we're tracking.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Please tell me all the sizes we're tracking (sorry I lost the link to the doc again). I'd like to make this the official source of the tracker.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
M == 512, N == 4096, K == 512
M == 4096, N == 512, K == 512
d3387a2
to
7805b30
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Any idea of the windows failure?
Yes I'm 90% sure it'll be fixed after rebasing on #911 (windows + ukernel = doesn't work) |
7805b30
to
d2c93ea
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Great to have these shapes benchmarked on every commit!
Now prints performance numbers for shapes
(M, N, K) = (512, 512, 4096) and (512, 4096, 512) and (4096, 512, 512)
with and without ukernels. Numbers with ukernels are interesting, take a look in CI, different gain with ukernel for the different shapes.
But looks like there's some improvement we can make on the vectorization of matmul front.