Repository navigation
FlashAttention: use subgroup size 32 on Intel Xe2 and Xe3 - #33808
Merged
Merged
Conversation
|
Azure Pipelines: There may be pipelines that require an authorized user to comment /azp run to run. |
Contributor
Author
|
/run bots |
Contributor
Author
Contributor
There was a problem hiding this comment.
🟢 Approval recommended
No active review issues remain for the change requesting a subgroup size of 32 on supported Intel adapters.
No open issues 🎉
What changed in this PR
This PR requests a subgroup size of 32 for WebGPU FlashAttention dense prefill on supported Intel adapters, allowing Tint to use a constant shuffle lane bound and helping avoid indirect register addressing.
Changes:
- Sets subgroup size to 32 when Intel hardware supports subgroup size control and the requested size.
- Preserves existing behavior when capability or tile-size compatibility checks fail.
| File | Description |
|---|---|
onnxruntime/contrib_ops/webgpu/bert/flash_attention.cc |
Adds guarded Intel subgroup-size selection for dense prefill. |
Review effort: Balanced
Contributor
|
The request was received. Approve workflow runs. Thanks. |
Contributor
|
LGTM! |
Contributor
Author
|
Jiajia Qin (@qjia7) PTAL |
Jie Chen (jchen10)
force-pushed
the
fa_simd32
branch
from
November 4, 2026 03:20
ad053b0 to
0a05eb5
Compare
Jie Chen (jchen10)
changed the title
Jiajia Qin (qjia7)
previously approved these changes
Nov 4, 2026
Jie Chen (jchen10)
force-pushed
the
fa_simd32
branch
from
November 5, 2026 01:27
0a05eb5 to
ed92d9f
Compare
Contributor
Author
|
/run bots |
Contributor
|
The request was received. Approve workflow runs. Thanks. |
An explicit subgroup size lets Tint use a constant lane clamp bound for subgroupShuffle, so IGC keeps direct register addressing.
Jie Chen (jchen10)
force-pushed
the
fa_simd32
branch
from
November 5, 2026 03:28
ed92d9f to
ba9e271
Compare
Jie Chen (jchen10)
changed the title
Contributor
Author
|
/run bots |
Contributor
|
The request was received. Approve workflow runs. Thanks. |
Jiajia Qin (qjia7)
approved these changes
Nov 6, 2026
Jiajia Qin (qjia7)
enabled auto-merge (squash)
November 6, 2026 07:49
Contributor
Author
|
/run failed bots |
Contributor
|
The request was received. Failed workflow jobs were queued for rerun. Thanks. |
Contributor
Author
|
/run bots |
Contributor
|
The request was received. Approve workflow runs. Thanks. |
Jiajia Qin (qjia7)
approved these changes
Nov 11, 2026
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
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.
An explicit subgroup size lets Tint use a constant lane clamp bound for subgroupShuffle, so IGC keeps direct register addressing.