Skip to content

[EXTERNAL] Fix v_mov_b16_t16 index in folding pass #2011

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
Merged
merged 2 commits into from
Oct 4, 2025
Merged

[EXTERNAL] Fix v_mov_b16_t16 index in folding pass #2011

merged 2 commits into from
Oct 4, 2025
+95 −1

Conversation

justinrosner
Copy link
Contributor

@justinrosner justinrosner commented Oct 3, 2025 •
edited
Loading

Motivation

This is a pulldown from an upstream change (llvm/llvm-project#161764) that provides a fix for functional failures in the MIGraphX CI on Navi3X.

Technical Details

With true16 mode, v_mov_b16_t16 is added as a new foldable copy inst, but the src operand is in a different index. This change fixes a bug in the folding pass so that we are using the correct index.

Test Plan

  • Run failing MIGraphX CI (test_group_query_attention_decode_small)
  • Add a new rocMLIR unit test to catch this

Test Result

  • Failing MIGraphX CI (test_group_query_attention_decode_small)

Submission Checklist

@justinrosner
Copy link
Contributor Author

Once the changes on upstream are finalized, I'll create two separate commits that I'll merge in: 1 for the external changes and 1 for rocMLIR changes

@justinrosner justinrosner marked this pull request as ready for review October 3, 2025 13:48
@justinrosner justinrosner requested a review from causten as a code owner October 3, 2025 13:48
@justinrosner justinrosner force-pushed the justinr-const-fold-patch branch from 2c94c5c to c11b4e0 Compare October 3, 2025 16:02
@broxigarchen broxigarchen mentioned this pull request Oct 3, 2025
Merged
@justinrosner
Copy link
Contributor Author

The Tune selected rocMLIR configs stage of the CI is continuously timing out for Navi3x. E.g., https://ml-ci-internal.amd.com/job/MLIR/job/mlir/job/PR-2011/8/pipeline-overview/. I've kicked off a local run of the same tuningRunner commands to ensure that this passes.

@justinrosner
Copy link
Contributor Author

justinrosner commented Oct 4, 2025 •
edited
Loading

The Tune selected rocMLIR configs stage of the CI is continuously timing out for Navi3x. E.g., https://ml-ci-internal.amd.com/job/MLIR/job/mlir/job/PR-2011/8/pipeline-overview/. I've kicked off a local run of the same tuningRunner commands to ensure that this passes.

I tried running the 'problematic' tuningRunner using both develop and this branch, and they both take multiple hours (i.e., not going to finish in the time allotted by the CI). All other CI jobs pass: https://ml-ci-internal.amd.com/job/MLIR/job/mlir/job/PR-2011/10/pipeline-overview/. Given that the tuningRunner jobs pass locally on Navi3x, this seems safe to merge.

@justinrosner justinrosner merged commit d3a00e0 into develop Oct 4, 2025
16 checks passed
@justinrosner justinrosner deleted the justinr-const-fold-patch branch October 4, 2025 03:05
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@dhernandez0 dhernandez0 dhernandez0 approved these changes

@umangyadav umangyadav umangyadav approved these changes

@causten causten Awaiting requested review from causten causten is a code owner

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@justinrosner @dhernandez0 @umangyadav