Add cutlass decode kernel to TritonBench #4853
Open
Conversation
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
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
|
This pull request was exported from Phabricator. Differential Revision: D80041532 |
added 2 commits
September 22, 2025 12:54
Summary: D80992628 introduced SWA FWD kernel changes which did not support decode kernels (i.e., supporting sm100_fmha_fwd but not sm100_fmha_gen). Similarly, softmax_scale introduced in D82788784 did not support decode kernels either. In blackwell_fmha_test, the these parameters are dropped during decode kernel selection (https://www.internalfb.com/code/fbsource/[cd7066706035]/fbcode/deeplearning/fbgemm/fbgemm_gpu/experimental/gen_ai/test/attention/blackwell_fmha_test.py?lines=182) To avoid confusion, do not test test_decode with ignored parameters. Differential Revision: D82991496
Summary: 1. Reduce pipeline stages to avoid exceeding smem limit 2. Add static_assert to make sure smem capacity violation is raised during compilation rather than runtime 3. Select the TMEM intrinsics based on sizeof(Element). 4. Update unittest to include bf16 5. Also label decode kernel test name with their corresponding test parameters. Differential Revision: D82991495
|
@Aya-ZIbra has exported this pull request. If you are a Meta employee, you can view the originating diff in D80041532. |
Aya-ZIbra
force-pushed
the
export-D80041532
branch
from
8bead29 to
3b2e442
Compare
Aya-ZIbra
added a commit
to Aya-ZIbra/FBGEMM
that referenced
this pull request
Sep 22, 2025
Summary: Pull Request resolved: pytorch#4853 X-link: facebookresearch/FBGEMM#1875 Add cutlass blackwell FMHA decode kernel implementation to TritonBench benchmarking suite . Reviewed By: sryap Differential Revision: D80041532
|
@Aya-ZIbra has exported this pull request. If you are a Meta employee, you can view the originating diff in D80041532. |
Aya-ZIbra
added a commit
to Aya-ZIbra/FBGEMM
that referenced
this pull request
Sep 22, 2025
Summary: Pull Request resolved: pytorch#4853 X-link: facebookresearch/FBGEMM#1875 Add cutlass blackwell FMHA decode kernel implementation to TritonBench benchmarking suite . Reviewed By: sryap Differential Revision: D80041532
Aya-ZIbra
force-pushed
the
export-D80041532
branch
from
3b2e442 to
68081f5
Compare
Aya-ZIbra
added a commit
to Aya-ZIbra/tritonbench
that referenced
this pull request
Sep 23, 2025
Summary: X-link: pytorch/FBGEMM#4853 X-link: facebookresearch/FBGEMM#1875 Add cutlass blackwell FMHA decode kernel implementation to TritonBench benchmarking suite . Reviewed By: sryap Differential Revision: D80041532
Summary: X-link: meta-pytorch/tritonbench#376 Pull Request resolved: pytorch#4853 X-link: facebookresearch/FBGEMM#1875 Add cutlass blackwell FMHA decode kernel implementation to TritonBench benchmarking suite . Reviewed By: sryap Differential Revision: D80041532
|
@Aya-ZIbra has exported this pull request. If you are a Meta employee, you can view the originating diff in D80041532. |
Aya-ZIbra
force-pushed
the
export-D80041532
branch
from
68081f5 to
5589cbb
Compare
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.
Summary:
X-link: https://github.com/facebookresearch/FBGEMM/pull/1875
Add cutlass blackwell FMHA decode kernel implementation to TritonBench benchmarking suite .
Reviewed By: sryap
Differential Revision: D80041532