1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
|
import operator_benchmark as op_bench
import torch
"""Microbenchmarks for channel_shuffle operator."""
# Configs for PT channel_shuffle operator
channel_shuffle_long_configs = op_bench.cross_product_configs(
batch_size=[4, 8],
channels_per_group=[32, 64],
height=[32, 64],
width=[32, 64],
groups=[4, 8],
channel_last=[True, False],
tags=["long"]
)
channel_shuffle_short_configs = op_bench.config_list(
attr_names=["batch_size", "channels_per_group", "height", "width", "groups"],
attrs=[
[2, 16, 16, 16, 2],
[2, 32, 32, 32, 2],
[4, 32, 32, 32, 4],
[4, 64, 64, 64, 4],
[8, 64, 64, 64, 8],
[16, 64, 64, 64, 16],
],
cross_product_configs={
"channel_last": [True, False],
},
tags=["short"]
)
class ChannelSHuffleBenchmark(op_bench.TorchBenchmarkBase):
def init(self, batch_size, channels_per_group, height, width, groups, channel_last):
channels = channels_per_group * groups
data_shape = (batch_size, channels, height, width)
input_data = torch.rand(data_shape)
if channel_last:
input_data = input_data.contiguous(memory_format=torch.channels_last)
self.inputs = {
"input_data": input_data,
"groups": groups
}
self.set_module_name('channel_shuffle')
def forward(self, input_data, groups: int):
return torch.channel_shuffle(input_data, groups)
op_bench.generate_pt_test(channel_shuffle_short_configs + channel_shuffle_long_configs,
ChannelSHuffleBenchmark)
if __name__ == "__main__":
op_bench.benchmark_runner.main()
|
