[feat]: fix splitK ASM kernel wrong output when splitK=0
Open
@junxiaguo is already working on this.
Since Aug 20, 2025.
- Dominant language
- Python
- Stars
- 565
- Forks
- 585
- Avg merge
- 3d 4h
- Merged PRs (30d)
- 366
Description
Problem Description
eg: f4gemm_bf16_per1x32Fp4_BpreShuffle_KSplit_128x512.co
With splitk=0 parameter, the kernel should not use zero to reset out（means kernel should not use atomic output operation）:
out_reset = torch.zeros(
(out.shape[0] + 255) // 256 * 256, out.shape[1], dtype=dtype
)
out = out_reset
Operating System
Ubuntu22.04
CPU
9800X3d
GPU
AMD MI355
ROCm Version
Rocm6.3
ROCm Component
No response
Steps to Reproduce
d, avg_d = run_gemm_asm(
x,
wshuffle,
x_scales_shuffle,
w_scales_shuffle,
out3,
"_ZN5aiter49f4gemm_bf16_per1x32Fp4_BpreShuffle_KSplit_128x512E",
bias_f32,
bpreshuffle=True,
log2_k_split=0,
)
(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support
No response
Additional Information
No response
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Assessment
This issue has not been assessed yet.