Expand configs with undefined dtypes to allowed WMMA ones for attention

[dorde-antic]

Motivation

Resolves known problems we have in MITuna until we fix them on MITuna so that our CI doesn't get blocked for dtypes problems...

Technical Details

Expand configs with undefined dtypes to allowed WMMA ones for attention

Test Plan

Tuning Attention

(Commit: 3d9f673 - with embedded py script right before tuning attn)
✅ https://ml-ci-internal.amd.com/blue/organizations/jenkins/MLIR%2Fmlir/detail/PR-2123/5/pipeline/667 (fails for other reasons)

(Commit: f3322d2 - after moving filtering logic in groovy function)
⌛https://ml-ci-internal.amd.com/job/MLIR/job/mlir/job/PR-2123/

Test Result

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[dorde-antic]

@umangyadav @dhernandez0
https://ml-ci-internal.amd.com/blue/organizations/jenkins/MLIR%2Fmlir/detail/PR-2123/5/pipeline/698
by this way of handling configs without datatype defined, there are no f32 tried for attn on navi in CI. (the present errors in run I've pasted here are not related to tuning f32)

We can move this expanding logic (which i tried directly in Jenkinsfile) for example to handleNewConfigs script (or other script used in CI) and call that function in CI before tuning attn on navi? Then we would reuse defined dtypes from perfRunner in handleNewConfigs, and it won't be hardcoded as in case I did in Jenkinsfile direclty in this PR (did it like that for testing purposes)

[dhernandez0]

@umangyadav @dhernandez0 https://ml-ci-internal.amd.com/blue/organizations/jenkins/MLIR%2Fmlir/detail/PR-2123/5/pipeline/698 by this way of handling configs without datatype defined, there are no f32 tried for attn on navi in CI. (the present errors in run I've pasted here are not related to tuning f32)

We can move this expanding logic (which i tried directly in Jenkinsfile) for example to handleNewConfigs script (or other script used in CI) and call that function in CI before tuning attn on navi? Then we would reuse defined dtypes from perfRunner in handleNewConfigs, and it won't be hardcoded as in case I did in Jenkinsfile direclty in this PR (did it like that for testing purposes)

I'm happy with this solution as a temporary fix until we find how to properly fix it.

[Copilot]

Pull request overview

This PR modifies the Jenkins pipeline to handle attention configurations for gfx1 chips by expanding entries without explicit dtypes to include all allowed WMMA data types (i8, f16, bf16), while filtering out f32 configurations.

Key Changes:

• Replaces simple grep filtering with a Python script that processes attention configuration files
• Expands configuration entries lacking explicit -t dtype flags to multiple entries with each allowed dtype
• Preserves comments and empty lines in the configuration file

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[umangyadav]

i am okay with keeping logic here in jenkinsfile but drawback is that we have to keep same logic in multiple different places now.

[dorde-antic]

i am okay with keeping logic here in jenkinsfile but drawback is that we have to keep same logic in multiple different places now.

Once I finish resolving it on MITuna we can remove redundant code for this
@umangyadav