submission 740652
abhinava-sai · 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.
submission.py
curl "https://kernelindex.com/api/v1/implementations/kernelbot-amd-mxfp4-mm-740652?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:abde732721740468fb2a20bb7bac9a14237100b9a5d15890087d5864097dfdcc
license declaredunknown
license concludedunknown
authorsabhinava-sai
imported2026-08-26
Kernel source
submission.py36 lines
#!POPCORN leaderboard amd-mxfp4-mm
#!POPCORN gpu MI355X
from task import input_t, output_t
import aiter
from aiter import dtypes
from aiter.ops.triton.quant import dynamic_mxfp4_quant
from aiter.utility.fp4_utils import e8m0_shuffle
_quant = dynamic_mxfp4_quant
_shuf = e8m0_shuffle
_gemm = aiter.gemm_a4w4
_fp4 = dtypes.fp4x2
_fp8 = dtypes.fp8_e8m0
_bf16 = dtypes.bf16
def custom_kernel(data: input_t) -> output_t:
A = data[0]
B_shuffle = data[3]
B_scale_sh = data[4]
A = A.contiguous()
x_fp4, bs_e8m0 = _quant(A)
bs_e8m0 = _shuf(bs_e8m0)
return _gemm(
x_fp4.view(_fp4),
B_shuffle,
bs_e8m0.view(_fp8),
B_scale_sh,
dtype=_bf16,
bpreshuffle=True
)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