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
|
import operator_benchmark as op_bench
import torch
import torch.ao.nn.quantized as nnq
from typing import List
"""Microbenchmarks for quantized Cat operator"""
# Configs for PT Cat operator
qcat_configs_short = op_bench.config_list(
attr_names=['M', 'N', 'K', 'L', 'dim'],
attrs=[
[256, 512, 1, 2, 0],
[512, 512, 2, 1, 1],
],
cross_product_configs={
'contig': ('all', 'one', 'none'),
'dtype': (torch.quint8, torch.qint8, torch.qint32),
},
tags=['short'],
)
qcat_configs_long = op_bench.cross_product_configs(
M=[128, 1024],
N=[128, 1024],
K=[1, 2],
L=[5, 7],
dim=[0, 1, 2],
contig=['all', 'one', 'none'],
dtype=[torch.quint8],
tags=['long']
)
class QCatBenchmark(op_bench.TorchBenchmarkBase):
def init(self, M, N, K, L, dim, contig, dtype):
f_input = (torch.rand(M, N, K) - 0.5) * 256
self.qf = nnq.QFunctional()
scale = 1.0
zero_point = 0
self.qf.scale = scale
self.qf.zero_point = zero_point
assert(contig in ('none', 'one', 'all'))
q_input = torch.quantize_per_tensor(f_input, scale, zero_point, dtype)
permute_dims = tuple(range(q_input.ndim - 1, -1, -1))
q_input_non_contig = q_input.permute(permute_dims).contiguous()
q_input_non_contig = q_input_non_contig.permute(permute_dims)
if contig == 'all':
self.input = (q_input, q_input)
elif contig == 'one':
self.input = (q_input, q_input_non_contig)
elif contig == 'none':
self.input = (q_input_non_contig, q_input_non_contig)
self.inputs = {
"input": self.input,
"dim": dim
}
self.set_module_name('qcat')
def forward(self, input: List[torch.Tensor], dim: int):
return self.qf.cat(input, dim=dim)
op_bench.generate_pt_test(qcat_configs_short + qcat_configs_long,
QCatBenchmark)
if __name__ == "__main__":
op_bench.benchmark_runner.main()
|
