File: softmax_test.py

package info (click to toggle)
pytorch-cuda 2.6.0%2Bdfsg-7
  • links: PTS, VCS
  • area: contrib
  • in suites: forky, sid, trixie
  • size: 161,620 kB
  • sloc: python: 1,278,832; cpp: 900,322; ansic: 82,710; asm: 7,754; java: 3,363; sh: 2,811; javascript: 2,443; makefile: 597; ruby: 195; xml: 84; objc: 68
file content (108 lines) | stat: -rw-r--r-- 2,346 bytes parent folder | download | duplicates (3)
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
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
import operator_benchmark as op_bench

import torch
import torch.nn as nn


"""
Microbenchmarks for the softmax operators.
"""


# Configs for softmax ops
softmax_configs_short = op_bench.config_list(
    attr_names=["N", "C", "H", "W"],
    attrs=[
        [1, 3, 256, 256],
        [4, 3, 256, 256],
    ],
    cross_product_configs={
        "device": ["cpu", "cuda"],
    },
    tags=["short"],
)


softmax_configs_long = op_bench.cross_product_configs(
    N=[8, 16],
    C=[3],
    H=[256, 512],
    W=[256, 512],
    device=["cpu", "cuda"],
    tags=["long"],
)


softmax_ops_list = op_bench.op_list(
    attr_names=["op_name", "op_func"],
    attrs=[
        ["Softmax", nn.Softmax],
        ["Softmax2d", nn.Softmax2d],
        ["LogSoftmax", nn.LogSoftmax],
    ],
)

softmax_two_dims_ops_list = op_bench.op_list(
    attr_names=["op_name", "op_func"],
    attrs=[
        ["Softmax", nn.Softmax],
        ["LogSoftmax", nn.LogSoftmax],
    ],
)


softmax_two_dims_configs = op_bench.config_list(
    attr_names=["M", "N", "dim"],
    attrs=[
        [700, 23258, 0],
        [700, 23258, 1],
        [1024, 23258, 1],
        [128, 128, 1],
        [48, 128, 1],
        [16, 1024, 1],
        [32, 1024, 1],
        [48, 1024, 1],
        [16, 512, 1],
        [32, 512, 1],
        [48, 512, 1],
        [16, 256, 1],
        [32, 256, 1],
        [48, 256, 1],
    ],
    cross_product_configs={
        "device": ["cpu", "cuda"],
    },
    tags=["long"],
)


class SoftmaxBenchmark(op_bench.TorchBenchmarkBase):
    def init(self, N, C, H, W, device, op_func):
        self.inputs = {"input": torch.rand(N, C, H, W, device=device)}
        self.op_func = op_func()

    def forward(self, input):
        return self.op_func(input)


class Softmax2DimsBenchmark(op_bench.TorchBenchmarkBase):
    def init(self, M, N, dim, device, op_func):
        self.inputs = {"input": torch.rand(M, N, device=device)}
        self.op_func = op_func(dim=dim)

    def forward(self, input):
        return self.op_func(input)


op_bench.generate_pt_tests_from_op_list(
    softmax_ops_list, softmax_configs_short + softmax_configs_long, SoftmaxBenchmark
)


op_bench.generate_pt_tests_from_op_list(
    softmax_two_dims_ops_list, softmax_two_dims_configs, Softmax2DimsBenchmark
)


if __name__ == "__main__":
    op_bench.benchmark_runner.main()