submission 686253
namecd_95693 · python · License unknown
Use it
Vendorable · source mirrored · license unknownView source →
No package. Vendor the mirrored source: 36 lines, June 9 Researcher Reciprocity License v1.0.
sub.py
curl "https://kernelindex.com/api/v1/implementations/kernelbot-amd-mxfp4-mm-686253?include=source"interfacepython
Compatibility
measured onAMD Instinct MI355X
declared hardwareAMD Instinct MI355X
architecturesgfx950
dtypesbf16, mxfp4
Benchmark evidence
1 measurement across 1 GPU, fastest first.
Operation / workload
Hardware
Latency
Rank
Observed
Reported · How evidence levels are derived →
Source and license
sourceavailable
revision digestsha256:ab0435d5b010d0097ee285590e47b7e2159f7ead601405ac6e176d6dd46d2703
license declaredunknown
license concludedunknown
authorsnamecd_95693
imported2026-08-26
Techniques
Extracted from the mirrored source by pattern, never inferred. Each row cites its line.
fp4
FP4 quant + FP4 GEMM optimized for AMD MI355XKernel source
sub.py36 lines
"""
FP4 quant + FP4 GEMM optimized for AMD MI355X
bf16 A, MXFP4 B -> MXFP4 per-1x32 quant A -> gemm_a4w4 -> bf16 C
"""
import torch
from task import input_t, output_t
def custom_kernel(data: input_t) -> output_t:
"""
MXFP4 GEMM kernel optimized for MI355X
"""
import aiter
from aiter import dtypes
from aiter.ops.triton.quant import dynamic_mxfp4_quant
from aiter.utility.fp4_utils import e8m0_shuffle
A, B, B_q, B_shuffle, B_scale_sh = data
A = A.contiguous()
B = B.contiguous()
A_fp4, A_scale_e8m0 = dynamic_mxfp4_quant(A)
A_scale_sh = e8m0_shuffle(A_scale_e8m0)
A_q = A_fp4.view(dtypes.fp4x2)
A_scale = A_scale_sh.view(dtypes.fp8_e8m0)
out = aiter.gemm_a4w4(
A_q,
B_shuffle,
A_scale,
B_scale_sh,
dtype=dtypes.bf16,
bpreshuffle=True,
)
return out
scrolls · 36 lines total
Source code from GPU Mode and the KernelBot dataset · June 9 Researcher Reciprocity License v1.0
Best evidence level for this revision: reported
JSON