Skip to content
KernelIndex
Search⌘K
Sign in

submission 630031

Hamza · python · License unknown
Kernel source · 59 lines ↓

Use it

Vendorable · source mirrored · license unknownView source →

No package. Vendor the mirrored source: 59 lines, June 9 Researcher Reciprocity License v1.0.

submission.py
curl "https://kernelindex.com/api/v1/implementations/kernelbot-amd-moe-mxfp4-630031?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
172.1µs
#335 of 782
2026-03-25
Run →

Reported · How evidence levels are derived →

Source and license

sourceavailable
revision digestsha256:e391368f767ad10ffcd0c251530665b39f24cf7a4f0219345c867535d2df7660
license declaredunknown
license concludedunknown
authorsHamza
imported2026-08-15

Kernel source

submission.py59 lines
#!POPCORN leaderboard amd-moe-mxfp4
#!POPCORN gpu MI355X

import os
import torch
from task import input_t, output_t

from aiter import ActivationType, QuantType
from aiter.fused_moe import fused_moe

_PAD_CACHE = {}


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

    M = topk_ids.shape[0]

    # For small batch sizes, ksplit=2 triggers the cktile backend
    # which is ~30% faster for 32-expert shapes with M<=128.
    # For large M, the default ck backend is better.
    # get_ksplit() is LRU-cached per shape, so the env var only
    # matters on the first (warmup) call for each unique shape.
    if M <= 128:
        os.environ["AITER_KSPLIT"] = "2"
    else:
        os.environ["AITER_KSPLIT"] = "0"

    cfg_id = id(config)
    cached = _PAD_CACHE.get(cfg_id)
    if cached is not None:
        hidden_pad, intermediate_pad = cached
    else:
        hidden_pad = config["d_hidden_pad"] - config["d_hidden"]
        intermediate_pad = config["d_expert_pad"] - config["d_expert"]
        _PAD_CACHE[cfg_id] = (hidden_pad, intermediate_pad)

    return fused_moe(
        hidden_states, gate_up_weight_shuffled, down_weight_shuffled,
        topk_weights, topk_ids,
        activation=ActivationType.Silu, quant_type=QuantType.per_1x32,
        w1_scale=gate_up_weight_scale_shuffled,
        w2_scale=down_weight_scale_shuffled,
        hidden_pad=hidden_pad, intermediate_pad=intermediate_pad,
    )
scrolls · 59 lines total

Source code from GPU Mode and the KernelBot dataset · June 9 Researcher Reciprocity License v1.0

Changes from previous submission

Against this author's previous submission submission 625722.

  #!POPCORN leaderboard amd-moe-mxfp4
  #!POPCORN gpu MI355X
   
+ import os
  import torch
  from task import input_t, output_t
   
  from aiter import ActivationType, QuantType
  from aiter.fused_moe import fused_moe
   
- # Cache pad values per config id (they don't depend on M)
  _PAD_CACHE = {}
   
   
  ⋯ 13 unchanged lines
          config,
      ) = data
   
+     M = topk_ids.shape[0]
+  
+     # For small batch sizes, ksplit=2 triggers the cktile backend
+     # which is ~30% faster for 32-expert shapes with M<=128.
+     # For large M, the default ck backend is better.
+     # get_ksplit() is LRU-cached per shape, so the env var only
+     # matters on the first (warmup) call for each unique shape.
+     if M <= 128:
+         os.environ["AITER_KSPLIT"] = "2"
+     else:
+         os.environ["AITER_KSPLIT"] = "0"
+  
      cfg_id = id(config)
      cached = _PAD_CACHE.get(cfg_id)
-     if cached is None:
+     if cached is not None:
+         hidden_pad, intermediate_pad = cached
+     else:
          hidden_pad = config["d_hidden_pad"] - config["d_hidden"]
          intermediate_pad = config["d_expert_pad"] - config["d_expert"]
-         cached = (hidden_pad, intermediate_pad)
-         _PAD_CACHE[cfg_id] = cached
-     hidden_pad, intermediate_pad = cached
+         _PAD_CACHE[cfg_id] = (hidden_pad, intermediate_pad)
   
      return fused_moe(
-         hidden_states,
-         gate_up_weight_shuffled,
-         down_weight_shuffled,
-         topk_weights,
-         topk_ids,
-         activation=ActivationType.Silu,
-         quant_type=QuantType.per_1x32,
+         hidden_states, gate_up_weight_shuffled, down_weight_shuffled,
+         topk_weights, topk_ids,
+         activation=ActivationType.Silu, quant_type=QuantType.per_1x32,
          w1_scale=gate_up_weight_scale_shuffled,
          w2_scale=down_weight_scale_shuffled,
-         hidden_pad=hidden_pad,
-         intermediate_pad=intermediate_pad,
+         hidden_pad=hidden_pad, intermediate_pad=intermediate_pad,
      )
scrolls · 60 diff lines total

Best evidence level for this revision: reported

JSON