Skip to content
KernelIndex
Search⌘K

submission 608437

N-45div · python · License unknown

Use it

Vendorable · source mirrored · license unknownView source →

No package. Vendor the mirrored source: 43 lines, June 9 Researcher Reciprocity License v1.0.

submission.py
curl "https://kernelindex.com/api/v1/implementations/kernelbot-amd-moe-mxfp4-608437?include=source"
interfacepython
Compatibility
measured onAMD Instinct MI355X
declared hardwareAMD Instinct MI355X
architecturesgfx950
dtypesbf16, fp32, fp8_e8m0, int32, mxfp4

Benchmark evidence

1 measurement across 1 GPU, fastest first.

Operation / workload
Hardware
Latency
Rank
Observed
AMD MXFP4 MoEsuite of 7 cases
AMD Instinct MI355X
185.5µs
#640 of 782
2026-03-22

Reported · How evidence levels are derived →

Source and license

sourceavailable
revision digestsha256:61d834c934c0e692c4e26b8b1d4cc57ec00210c45b3a6b7b7ab69efb8b1d9a95
license declaredunknown
license concludedunknown
authorsN-45div
imported2026-08-26

Techniques

Extracted from the mirrored source by pattern, never inferred. Each row cites its line.

fp4Optimized MoE MXFP4 fused kernel for MI355X.

Kernel source

submission.py43 lines
"""
Optimized MoE MXFP4 fused kernel for MI355X.
Directly calls aiter fused_moe with minimal Python overhead.
"""
from task import input_t, output_t
from aiter import ActivationType, QuantType
from aiter.fused_moe import fused_moe


def custom_kernel(data: input_t) -> output_t:
    (
        hidden_states,
        gate_up_weight,
        down_weight,
        gate_up_weight_scale,
        down_weight_scale,
        gate_up_weight_shuffled,
        down_weight_shuffled,
        gate_up_weight_scale_shuffled,
        down_weight_scale_shuffled,
        topk_weights,
        topk_ids,
        config,
    ) = data

    return fused_moe(
        hidden_states,
        gate_up_weight_shuffled,
        down_weight_shuffled,
        topk_weights,
        topk_ids,
        expert_mask=None,
        activation=ActivationType.Silu,
        quant_type=QuantType.per_1x32,
        doweight_stage1=False,
        w1_scale=gate_up_weight_scale_shuffled,
        w2_scale=down_weight_scale_shuffled,
        a1_scale=None,
        a2_scale=None,
        hidden_pad=config["d_hidden_pad"] - config["d_hidden"],
        intermediate_pad=config["d_expert_pad"] - config["d_expert"],
    )
scrolls · 43 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