submission 655098
wbh · python · License unknown
Use it
Vendorable · source mirrored · license unknownView source →
No package. Vendor the mirrored source: 42 lines, June 9 Researcher Reciprocity License v1.0.
submission.py
curl "https://kernelindex.com/api/v1/implementations/kernelbot-amd-mxfp4-mm-655098?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:ea557370f7e540302d36a24a5dad84778a9ca0443aa350b6add1a6ebf39b0e75
license declaredunknown
license concludedunknown
authorswbh
imported2026-08-26
Kernel source
submission.py42 lines
from task import input_t, output_t
import aiter
from aiter import QuantType, dtypes
from aiter.ops.triton.quant import dynamic_mxfp4_quant
from aiter.utility.fp4_utils import e8m0_shuffle
def _quant_mxfp4(x, shuffle=True):
x_fp4, bs_e8m0 = dynamic_mxfp4_quant(x)
if shuffle:
bs_e8m0 = e8m0_shuffle(bs_e8m0)
return x_fp4.view(dtypes.fp4x2), bs_e8m0.view(dtypes.fp8_e8m0)
def _kernel_gemm_a4w4(A_q, B_shuffle, A_scale_sh, B_scale_sh):
out_gemm = aiter.gemm_a4w4(
A_q,
B_shuffle,
A_scale_sh,
B_scale_sh,
dtype=dtypes.bf16,
bpreshuffle=True,
)
return out_gemm
def custom_kernel(data: input_t) -> output_t:
A, B, B_q, B_shuffle, B_scale_sh = data
A = A.contiguous()
B = B.contiguous()
m, k = A.shape
n, _ = B.shape
A_q, A_scale_sh = _quant_mxfp4(A, shuffle=True)
out_gemm = _kernel_gemm_a4w4(
A_q,
B_shuffle,
A_scale_sh,
B_scale_sh,
)
return out_gemm
scrolls · 42 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