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
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
|
import operator_benchmark as op_bench
import torch
import torch.ao.quantization.observer as obs
qobserver_short_configs_dict = {
"attr_names": ("C", "M", "N", "dtype", "device"),
"attrs": (
(3, 512, 512, torch.quint8, "cpu"),
(3, 512, 512, torch.quint8, "cuda"),
),
"tags": ("short",),
}
q_hist_observer_short_configs_dict = {
"attr_names": ("C", "M", "N", "dtype", "device"),
"attrs": ((3, 512, 512, torch.quint8, "cpu"),),
"tags": ("short",),
}
qobserver_long_configs_dict = {
"C": (32, 64),
"M": (256, 1024),
"N": (256, 1024),
"device": ("cpu", "cuda"),
"dtype": (torch.quint8,), # dtype doesn't change the timing, keep the same
"tags": ("long",),
}
q_hist_observer_long_configs_dict = {
"C": (1, 3, 8),
"M": (256, 1024),
"N": (256, 1024),
"device": ("cpu",),
"dtype": (torch.quint8,), # dtype doesn't change the timing, keep the same
"tags": ("long",),
}
qobserver_per_tensor_configs_short = op_bench.config_list(
cross_product_configs={
"qscheme": (torch.per_tensor_affine, torch.per_tensor_symmetric)
},
**qobserver_short_configs_dict,
)
qobserver_per_tensor_configs_long = op_bench.cross_product_configs(
qscheme=(torch.per_tensor_affine, torch.per_tensor_symmetric),
**qobserver_long_configs_dict,
)
qobserver_per_channel_configs_short = op_bench.config_list(
cross_product_configs={
"qscheme": (torch.per_channel_affine, torch.per_channel_symmetric)
},
**qobserver_short_configs_dict,
)
qobserver_per_channel_configs_long = op_bench.cross_product_configs(
qscheme=(torch.per_channel_affine, torch.per_channel_symmetric),
**qobserver_long_configs_dict,
)
q_hist_observer_per_tensor_configs_short = op_bench.config_list(
cross_product_configs={
"qscheme": (torch.per_tensor_affine, torch.per_tensor_symmetric)
},
**q_hist_observer_short_configs_dict,
)
q_hist_observer_per_tensor_configs_long = op_bench.cross_product_configs(
qscheme=(torch.per_tensor_affine, torch.per_tensor_symmetric),
**q_hist_observer_long_configs_dict,
)
qobserver_per_tensor_list = op_bench.op_list(
attr_names=["op_name", "op_func"],
attrs=[
["MinMaxObserver", obs.MinMaxObserver],
["MovingAverageMinMaxObserver", obs.MovingAverageMinMaxObserver],
],
)
qobserver_per_channel_list = op_bench.op_list(
attr_names=["op_name", "op_func"],
attrs=[
["PerChannelMinMaxObserver", obs.PerChannelMinMaxObserver],
[
"MovingAveragePerChannelMinMaxObserver",
obs.MovingAveragePerChannelMinMaxObserver,
],
],
)
q_hist_observer_list = op_bench.op_list(
attr_names=["op_name", "op_func"],
attrs=[
["HistogramObserver", obs.HistogramObserver],
["HistogramObserverCalculateQparams", obs.HistogramObserver],
],
)
class QObserverBenchmark(op_bench.TorchBenchmarkBase):
def init(self, C, M, N, dtype, qscheme, op_func, device):
self.inputs = {"f_input": torch.rand(C, M, N, device=device)}
self.op_func = op_func(dtype=dtype, qscheme=qscheme).to(device)
def forward(self, f_input):
self.op_func(f_input)
return self.op_func.calculate_qparams()
class QObserverBenchmarkCalculateQparams(op_bench.TorchBenchmarkBase):
def init(self, C, M, N, dtype, qscheme, op_func, device):
self.f_input = torch.rand(C, M, N, device=device)
self.q_observer = op_func(dtype=dtype, qscheme=qscheme).to(device)
self.q_observer(self.f_input)
self.inputs = {}
def forward(self):
return self.q_observer.calculate_qparams()
op_bench.generate_pt_tests_from_op_list(
qobserver_per_tensor_list,
qobserver_per_tensor_configs_short + qobserver_per_tensor_configs_long,
QObserverBenchmark,
)
op_bench.generate_pt_tests_from_op_list(
qobserver_per_channel_list,
qobserver_per_channel_configs_short + qobserver_per_channel_configs_long,
QObserverBenchmark,
)
op_bench.generate_pt_tests_from_op_list(
q_hist_observer_list,
q_hist_observer_per_tensor_configs_short + q_hist_observer_per_tensor_configs_long,
QObserverBenchmarkCalculateQparams,
)
if __name__ == "__main__":
op_bench.benchmark_runner.main()
|
