Skip to content

[TRITON] Instruction shape fix for Gluon gemm_a8w8_blockscale kernel #1261

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
eky-amd merged 10 commits into from
Nov 7, 2025
Merged

[TRITON] Instruction shape fix for Gluon gemm_a8w8_blockscale kernel #1261

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
0e4300e
fixed instr_shape error for mfma layout
eky-amd Oct 23, 2025
3a0c2e9
completing merge
eky-amd Oct 24, 2025
7235e83
removed test skips for uneven K
eky-amd Oct 24, 2025
2017516
completing merge
eky-amd Oct 27, 2025
a0774b5
completing merge
eky-amd Oct 29, 2025
c3ddd09
completing merge
eky-amd Oct 31, 2025
6d9d5f6
Merge branch 'main' into eky/fix-gluon-gemm-a8w8-blockscale
gyohuangxin Nov 4, 2025
f2ab847
Merge branch 'main' into eky/fix-gluon-gemm-a8w8-blockscale
eky-amd Nov 6, 2025
2723a68
Merge branch 'main' into eky/fix-gluon-gemm-a8w8-blockscale
eky-amd Nov 6, 2025
49cdfb1
Merge branch 'main' into eky/fix-gluon-gemm-a8w8-blockscale
eky-amd Nov 7, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion aiter/ops/triton/gluon/gemm_a8w8_blockscale.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -120,7 +120,7 @@ def _gemm_a8w8_blockscale_kernel(
)
mfma_layout: gl.constexpr = gl.amd.AMDMFMALayout(
version=4,
instr_shape=[16, 16],
instr_shape=[16, 16, 32], # V_MFMA_F32_16X16X32_FP8_FP8 instruction
transposed=True,
warps_per_cta=[NUM_WARPS // 2, 2],
)
Expand Down
5 changes: 0 additions & 5 deletions op_tests/triton_tests/test_gemm_a8w8_blockscale.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -183,11 +183,6 @@ def test_gemm(dtype, M, N, K, layout, output, impl: str):
torch.cuda.synchronize()

block_shape_n, block_shape_k = block_shape
if K % block_shape_k != 0:
pytest.skip(
"Latest upstream compiler as of Aug 22 (necessary for Gluon) causes"
" infinite hang when EVEN_K is false. Try seeing if it's fixed if it's been a while."
)

if impl == "gluon" and int(DEVICE_ARCH.split("MI")[1].replace("X", "")) < 350:
pytest.skip(
Expand Down