[CK Tile] Enable LSE output for fp8bf16 V3 FMHA kernels#7723
Merged
Conversation
Contributor
|
LGTM — the LSE formula fix for StandardAttention is correct (matches the async pipeline reference) and the plumbing changes are straightforward. One ask: please make sure the FMHA test is enabled in CI for this PR so the |
poyenc
approved these changes
May 26, 2026
Contributor
poyenc
left a comment
poyenc
left a comment
There was a problem hiding this comment.
LGTM — LSE formula fix is correct, CI all green including FMHA tests on gfx950/gfx942/gfx90a/gfx1201.
assistant-librarian Bot
pushed a commit
to ROCm/composable_kernel
that referenced
this pull request
May 28, 2026
[CK Tile] Enable LSE output for fp8bf16 V3 FMHA kernels
(#7723)
### Motivation
The V3 pipeline (qr_async_trload_v3) for fp8bf16 FMHA kernels did not
support LSE (Log-Sum-Exp) output. This PR enables LSE output support for
fp8bf16 V3 FMHA kernels, allowing users to retrieve attention statistics
alongside attention outputs.
### Technical Details
- StandardAttention: lse = softmax_scale * m + log(l)
- LogitsSoftCap: lse = (m / log2(e)) + log(l)
### Test Plan
Run FMHA forward example with fp8bf16 precision and LSE output enabled:
- Test 1: Basic LSE functionality
./build/bin/tile_example_fmha_fwd -v=1 -b=1 -h=8 -s=1024 -d=128
-prec=fp8bf16 -init=3 -qscale=1 -lse=1
- Test 2: LSE with LogitsSoftCap (CMakeList should remove Logits filter)
./build/bin/tile_example_fmha_fwd -v=1 -b=1 -h=8 -s=1024 -d=128
-prec=fp8bf16 -init=3 -qscale=1 -lse=1 -logits_soft_cap=30.0
yenong-amd
pushed a commit
that referenced
this pull request
May 28, 2026
### Motivation
The V3 pipeline (qr_async_trload_v3) for fp8bf16 FMHA kernels did not
support LSE (Log-Sum-Exp) output. This PR enables LSE output support for
fp8bf16 V3 FMHA kernels, allowing users to retrieve attention statistics
alongside attention outputs.
### Technical Details
- StandardAttention: lse = softmax_scale * m + log(l)
- LogitsSoftCap: lse = (m / log2(e)) + log(l)
### Test Plan
Run FMHA forward example with fp8bf16 precision and LSE output enabled:
- Test 1: Basic LSE functionality
./build/bin/tile_example_fmha_fwd -v=1 -b=1 -h=8 -s=1024 -d=128
-prec=fp8bf16 -init=3 -qscale=1 -lse=1
- Test 2: LSE with LogitsSoftCap (CMakeList should remove Logits filter)
./build/bin/tile_example_fmha_fwd -v=1 -b=1 -h=8 -s=1024 -d=128
-prec=fp8bf16 -init=3 -qscale=1 -lse=1 -logits_soft_cap=30.0
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
2 participants
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.
Motivation
The V3 pipeline (qr_async_trload_v3) for fp8bf16 FMHA kernels did not support LSE (Log-Sum-Exp) output. This PR enables LSE output support for fp8bf16 V3 FMHA kernels, allowing users to retrieve attention statistics alongside attention outputs.
Technical Details
Test Plan
Run FMHA forward example with fp8bf16 precision and LSE output enabled:
Test 1: Basic LSE functionality
./build/bin/tile_example_fmha_fwd -v=1 -b=1 -h=8 -s=1024 -d=128 -prec=fp8bf16 -init=3 -qscale=1 -lse=1
Test 2: LSE with LogitsSoftCap (CMakeList should remove Logits filter)
./build/bin/tile_example_fmha_fwd -v=1 -b=1 -h=8 -s=1024 -d=128 -prec=fp8bf16 -init=3 -qscale=1 -lse=1 -logits_soft_cap=30.0