Skip to content

[AIMIGRAPHX-1050] Enable MLIR Attention by Default on MI350#4901

Merged
kahmed10 merged 4 commits into
developfrom
enable_attention_gfx950
May 22, 2026
Merged

[AIMIGRAPHX-1050] Enable MLIR Attention by Default on MI350#4901
kahmed10 merged 4 commits into
developfrom
enable_attention_gfx950

Conversation

@eddieliao
Copy link
Copy Markdown
Contributor

Motivation

rocMLIR attention is currently only default for gfx94x (MI300). We can probably enable this for MI350 as well.

Technical Details

Adds a list of gfx prefixes that will use MLIR attention by default.

Changelog Category

Add a CHANGELOG.md entry for any option other than Not Applicable

    • Added: New functionality.
    • Changed: Changes to existing functionality.
    • Removed: Functionality or support that has been removed. (Compared to a previous release)
    • Optimized: Component performance that has been optimized or improved.
    • Resolved Issues: Known issues from a previous version that have been resolved.
    • Not Applicable: This PR is not to be included in the changelog.

@eddieliao eddieliao requested review from kahmed10 and pfultz2 May 20, 2026 22:34
@eddieliao eddieliao self-assigned this May 20, 2026
@eddieliao eddieliao requested a review from causten as a code owner May 20, 2026 22:34
Copilot AI review requested due to automatic review settings May 20, 2026 22:34
@eddieliao eddieliao added simple small or simple changes Perf Improve rocMLIR labels May 20, 2026
Copy link
Copy Markdown
Contributor

Copilot AI left a comment

Choose a reason for hiding this comment

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

Pull request overview

Enables rocMLIR attention by default on additional GPU architectures (MI350-class) by expanding the set of supported gfx prefixes used to auto-enable MLIR attention.

Changes:

  • Extend the default-enable check for MLIR attention from gfx94* to both gfx94* and gfx95* devices.
  • Refactor the device-prefix logic to use a small prefix list and std::any_of.

Comment thread src/targets/gpu/fuse_mlir.cpp Outdated
@kahmed10 kahmed10 merged commit f221f4a into develop May 22, 2026
38 checks passed
@kahmed10 kahmed10 deleted the enable_attention_gfx950 branch May 22, 2026 18:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Perf Improve rocMLIR simple small or simple changes

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants