mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
c0ed38e644
See https://github.com/pytorch/pytorch/pull/129751#issue-2380881501. Most changes are auto-generated by linter. You can review these PRs via: ```bash git diff --ignore-all-space --ignore-blank-lines HEAD~1 ``` Pull Request resolved: https://github.com/pytorch/pytorch/pull/129754 Approved by: https://github.com/ezyang
37 lines
920 B
Python
37 lines
920 B
Python
import operator_benchmark as op_bench
|
|
|
|
import torch
|
|
import torch.nn.functional as F
|
|
|
|
|
|
"""Microbenchmarks for instancenorm operator."""
|
|
|
|
instancenorm_configs_short = op_bench.cross_product_configs(
|
|
dims=(
|
|
(32, 8, 16),
|
|
(32, 8, 56, 56),
|
|
),
|
|
tags=["short"],
|
|
)
|
|
|
|
|
|
class InstanceNormBenchmark(op_bench.TorchBenchmarkBase):
|
|
def init(self, dims):
|
|
num_channels = dims[1]
|
|
self.inputs = {
|
|
"input": (torch.rand(*dims) - 0.5) * 256,
|
|
"weight": torch.rand(num_channels, dtype=torch.float),
|
|
"bias": torch.rand(num_channels, dtype=torch.float),
|
|
"eps": 1e-5,
|
|
}
|
|
|
|
def forward(self, input, weight, bias, eps: float):
|
|
return F.instance_norm(input, weight=weight, bias=bias, eps=eps)
|
|
|
|
|
|
op_bench.generate_pt_test(instancenorm_configs_short, InstanceNormBenchmark)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
op_bench.benchmark_runner.main()
|