Merged
Conversation
Contributor
There was a problem hiding this comment.
Pull request overview
This PR adds Gluon kernel implementations for GEMM_AFP4WFP4 (GEMM with FP4 activations and FP4 weights) operations, specifically targeting AMD CDNA4 (gfx950) devices. The implementation includes a new Gluon-based kernel, test infrastructure updates to support both Triton and Gluon implementations, and benchmark support.
Changes:
- Adds new Gluon kernel implementation for AFP4WFP4 GEMM operations with split-K support
- Updates tests to parameterize implementation choice between Triton and Gluon variants
- Adds benchmark support for performance comparison between implementations
Reviewed changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 5 comments.
| File | Description |
|---|---|
| aiter/ops/triton/gluon/gemm_afp4wfp4.py | New Gluon kernel implementation with main GEMM kernel, reduce kernel, and configuration loading |
| aiter/ops/triton/configs/gemm/gluon/gfx950-GEMM-AFP4WFP4.json | Configuration file with tuning parameters for gfx950 architecture |
| op_tests/triton_tests/gemm/basic/test_gemm_afp4wfp4.py | Updated tests to support both Triton and Gluon implementations via parameterization |
| op_tests/op_benchmarks/triton/bench_gemm_afp4wfp4.py | Added Gluon implementation option to benchmark suite |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
ahmed-bsod
force-pushed
the
ahmed-bsod/gluon_gemm_afp4wfp4
branch
from
3770da4 to
51aafa6
Compare
cagrikymk
previously approved these changes
Jan 13, 2026
Contributor
|
Besides the nitpicking, LGTM! |
ahmed-bsod
force-pushed
the
ahmed-bsod/gluon_gemm_afp4wfp4
branch
from
51aafa6 to
6538132
Compare
ahmed-bsod
force-pushed
the
ahmed-bsod/gluon_gemm_afp4wfp4
branch
2 times, most recently
from
ac77116 to
798cbc9
Compare
also changed waves_per_eu to 0 Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
ahmed-bsod
force-pushed
the
ahmed-bsod/gluon_gemm_afp4wfp4
branch
from
798cbc9 to
389d342
Compare
lburzawa
approved these changes
Jan 21, 2026
cagrikymk
approved these changes
Jan 21, 2026
gyohuangxin
pushed a commit
that referenced
this pull request
Jan 22, 2026
* added gluon kernel for gemm_afp4wfp4 * updated the gemm_afp4wfp4 test and bench scripts to support running the kernel
yzhou103
pushed a commit
that referenced
this pull request
Jan 28, 2026
* added gluon kernel for gemm_afp4wfp4 * updated the gemm_afp4wfp4 test and bench scripts to support running the kernel
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
4 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
Added gluon kernels for GEMM_AFP4FP4
Perf
Tuned for M=N=K=8192
Running the bench script
python bench_gemm_afp4wfp4.py --shape 8192 8192 8192 --gluongives a performance of 2500ish tflops on avg
Testing
Passes all the shapes in
test_gemm_afp4wfp4.pyAlso tested the
_gemm_afp4wfp4_reduce_kernelby setting k_split to 2, 4, 8 in config file