Skip to content
KernelIndex
Search⌘K
Sign in

submission 705543

stashuk-olek · python · License unknown
Kernel source · 20 lines ↓

Use it

Vendorable · source mirrored · license unknownView source →

No package. Vendor the mirrored source: 20 lines, June 9 Researcher Reciprocity License v1.0.

submission.py
curl "https://kernelindex.com/api/v1/implementations/kernelbot-conv2d-v2-705543?include=source"
interfacepython
Compatibility
measured onNVIDIA H100
declared hardwareNVIDIA H100
architecturessm_90
dtypesfp32

Benchmark evidence

1 measurement across 1 GPU, fastest first.

Operation / workload
Hardware
Latency
Rank
Observed
2D convolutionsuite of 5 cases
NVIDIA H100
47.7ms
#5 of 35
2026-04-03
Run →

Reported · How evidence levels are derived →

Source and license

sourceavailable
revision digestsha256:660d10d1217dcdedb10958640dbea5d6cd4be0d715c3d1cd99267e269443ad8c
license declaredunknown
license concludedunknown
authorsstashuk-olek
imported2026-08-15

Kernel source

submission.py20 lines
#!POPCORN leaderboard conv2d_v2
#!POPCORN gpu H100

import torch
import torch.nn.functional as F

from task import input_t, output_t

# Disable TF32: must match reference cuDNN computation exactly.
torch.backends.cudnn.allow_tf32 = False
torch.backends.cuda.matmul.allow_tf32 = False
# Enable cuDNN benchmarking to select the fastest algorithm.
torch.backends.cudnn.benchmark = True


def custom_kernel(data: input_t) -> output_t:
    input_tensor, kernel, output = data
    output[...] = F.conv2d(input_tensor, kernel, stride=1, padding=0)
    return output

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 704454.

  ⋯ 2 unchanged lines
   
  import torch
  import torch.nn.functional as F
+  
  from task import input_t, output_t
   
- # Disable TF32: H100 enables it by default, but eval.py reference uses exact float32
+ # Disable TF32: must match reference cuDNN computation exactly.
  torch.backends.cudnn.allow_tf32 = False
  torch.backends.cuda.matmul.allow_tf32 = False
+ # Enable cuDNN benchmarking to select the fastest algorithm.
+ torch.backends.cudnn.benchmark = True
   
   
  def custom_kernel(data: input_t) -> output_t:

Best evidence level for this revision: reported

JSON