-
Notifications
You must be signed in to change notification settings - Fork 645
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
[LLVMGPU] Disable scf.forall distribution for matmulSimt #19302
Conversation
We are moving towards tile&fuse pipeline. `matmulSimt` pipeline is to be deprecated. Meanwhile, all existing tests that depend on matmulsimt pipeline shouldn't break and hence disabling the forall distribution for the pipeline.
@pashu123 seems like there may be issues with SDXL perf after merging this PR |
cc: @qedawkins for visibility :^ |
That's surprising because this is just reverting to the behavior we've had for a long time, are we sure it's this PR? |
It started failing just after merging this one, but I can't say for sure because I haven't tried reproducing this locally |
@kuhar It might be because of 2602a2a (this is just reverting to the old behaviour) that @qedawkins mentioned. |
I see a similar huge regression right before this PR too: https://github.com/iree-org/iree/actions/runs/12023092109/job/33516722915 maybe an unstable runner? |
cc: @saienduri |
It started here: 77ff99c |
) We are moving towards tile&fuse pipeline. `matmulSimt` pipeline is to be deprecated. Meanwhile, all existing tests that depend on matmulsimt pipeline shouldn't break and hence disabling the forall distribution for the pipeline. Signed-off-by: Giacomo Serafini <[email protected]>
We are moving towards tile&fuse pipeline.
matmulSimt
pipeline is to be deprecated. Meanwhile, all existing tests that depend on matmulsimt pipeline shouldn't break and hence disabling the forall distribution for the pipeline.