Repository navigation
Run EP permute tests on ROCm gfx950 - #4987
Open
indianspeedster wants to merge 1 commit into
Open
indianspeedster wants to merge 1 commit into
indianspeedster wants to merge 1 commit into
Conversation
test/prototype/moe_training/ep/test_permute.py and ep/test_kernels.py were gated to CUDA 12.8+ with SM >= 100, so they never ran on ROCm. The kernels they test (generate_permute_indices, _triton_permute_bwd, and the MXFP8 permute) are Triton and also work on gfx950. Allow MI350-series (gfx950) as well, using the existing torchao.utils.is_MI350().
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4987
Note: Links to docs will display an error until the docs builds have been completed. This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
|
The following ciflow label(s) have been added but CI has not been triggered yet because the workflows are awaiting approval:
Once a maintainer approves the workflows (scroll to the bottom of the PR page), the corresponding CI jobs will be triggered automatically. Please ping one of the reviewers if you do not have access to approve and run workflows. |
This was referenced Oct 8, 2026
indianspeedster
marked this pull request as ready for review
October 8, 2026 20:03
indianspeedster
requested review from
andrewor14,
jerryzh168 and
vkuzo
as code owners
October 8, 2026 20:03
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
test/prototype/moe_training/ep/test_permute.pyandep/test_kernels.pyaregated to CUDA 12.8+ with SM >= 100, so they never run on ROCm. The code they test
(
generate_permute_indices,_triton_permute_bwd, and the MXFP8 permute) isTriton and also works on gfx950. This allows MI350-series (gfx950) as well,
using the existing
torchao.utils.is_MI350(). Test-only change.Split out of #4623 as requested in review.
Test plan
pytest test/prototype/moe_training/ep/test_permute.py test/prototype/moe_training/ep/test_kernels.py:mainThe MI355X runs used a local environment where the
flydslpackage didn't matchthe repo, so
flydslwas hidden and torchao took its non-flydslpath.CI
No upstream CI runner has SM100, so on
mainthese files run in no CI job. Withthis PR they run in the ROCm regression job (
regression_test_rocm.yml, whichruns
pytest teston the MI350 runner when triggered byciflow/rocmor a pushto
main).MI355X: torch 2.15.0.dev+rocm7.14, triton 3.8.0. B200: torch 2.14.1+cu130,
triton 3.8.0.
Authored with assistance from an AI coding agent.