Skip to content

7.2 cherry-pick PR#3221. #3372

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
Open
illsilin wants to merge 1 commit into
base: release/rocm-rel-7.2
Choose a base branch
from
Open

7.2 cherry-pick PR#3221. #3372

illsilin wants to merge 1 commit into from

Conversation

@illsilin
Copy link
Collaborator

@illsilin illsilin commented Dec 8, 2025

Proposed changes

This will include the workaround for fmha synchronization issue in rocm7.2

Checklist

Please put an x into the boxes that apply. You can also fill these out after creating the PR. If you're not sure, please don't hesitate to ask.

  • I have added tests relevant to the introduced functionality, and the unit tests are passing locally
  • I have added the test to REGRESSION_TESTS list defined at the top of CMakeLists.txt in tests/CMakeLists.txt, IF the test takes more than 30 seconds to run.
  • I have added inline documentation which enables the maintainers with understanding the motivation
  • I have removed the stale documentation which is no longer relevant after this pull request
  • (If this change is user-facing) I have added release notes which provide the end users with a brief summary of the improvement from this pull request
  • I have run clang-format on all changed files
  • Any dependent changes have been merged

@ex-rzr @illsilin
[CK_TILE] FMHA Reduce register spilling in fwd with dropout (workarou…
754b7c6 
…nd for CI failures with clang-22) (#3221)

* Use vectorized stores for dropout randvals

With no kPadSeqLenK the kernel uses 2 buffer_store_dwordx2 instead of
16 buffer_store_byte. This requires less registers and reduces spilling.

* Calculate dropout randvals for storing and applying only once

Even though it may add a small overhead when storing is not required,
it uses significantly less registers and hence no spilling.
@afagaj afagaj changed the title Cherry-pick PR#3221. 7.2 cherry-pick PR#3221. Dec 8, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@afagaj afagaj afagaj approved these changes

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

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

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

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

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

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

@andriy-ca andriy-ca Awaiting requested review from andriy-ca andriy-ca is a code owner

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

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

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

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

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

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

@vidyasagar-amd vidyasagar-amd Awaiting requested review from vidyasagar-amd vidyasagar-amd 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.

4 participants

@illsilin @afagaj @ex-rzr