Add GEMM-based standard SDPA benchmark and fix custom_sdpa_out signature#18715
Open
kimishpatel wants to merge 1 commit intogh/kimishpatel/235/basefrom
Open
Add GEMM-based standard SDPA benchmark and fix custom_sdpa_out signature#18715kimishpatel wants to merge 1 commit intogh/kimishpatel/235/basefrom
kimishpatel wants to merge 1 commit intogh/kimishpatel/235/basefrom
Conversation
Adds a standalone GEMM-based (non-tiled) attention benchmark alongside the existing ET flash attention benchmark, allowing direct comparison of the two algorithms. The standard SDPA uses cblas_sgemm for Q@K^T and scores@V with 3-pass softmax, matching the approach used by ONNX Runtime's GQA operator. Both transposed [B,H,S,D] and standard [B,S,H,D] cache layouts are supported via BLAS leading dimension parameter. Validation tests run before benchmarks to ensure correctness against ET's custom_sdpa_out. Also fixes broken custom_sdpa_out calls to match the new 3-bool signature (is_seq_dim_2, is_k_seq_dim_2, is_v_seq_dim_2). Authored with Claude. Differential Revision: [D99677686](https://our.internmc.facebook.com/intern/diff/D99677686/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/18715
Note: Links to docs will display an error until the docs builds have been completed. ❌ 128 New FailuresAs of commit e968c4a with merge base fb1618e ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
This was referenced Apr 6, 2026
Open
This PR needs a
|
This was referenced Apr 6, 2026
meta-cla
bot
added
the
CLA Signed
Contributor
Author
|
submitted by accident, not meant to land immedidately |
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.
Stack from ghstack (oldest at bottom):
Adds a standalone GEMM-based (non-tiled) attention benchmark alongside
the existing ET flash attention benchmark, allowing direct comparison
of the two algorithms. The standard SDPA uses cblas_sgemm for Q@K^T
and scores@V with 3-pass softmax, matching the approach used by
ONNX Runtime's GQA operator.
Both transposed [B,H,S,D] and standard [B,S,H,D] cache layouts are
supported via BLAS leading dimension parameter. Validation tests run
before benchmarks to ensure correctness against ET's custom_sdpa_out.
Also fixes broken custom_sdpa_out calls to match the new 3-bool
signature (is_seq_dim_2, is_k_seq_dim_2, is_v_seq_dim_2).
Authored with Claude.
Differential Revision: D99677686