-
Notifications
You must be signed in to change notification settings - Fork 52
/
flops_benchmark.py
261 lines (164 loc) · 7.48 KB
/
flops_benchmark.py
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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
#### https://github.com/warmspringwinds/pytorch-segmentation-detection/blob/master/pytorch_segmentation_detection/utils/flops_benchmark.py
import torch
# ---- Public functions
def add_flops_counting_methods(net_main_module):
"""Adds flops counting functions to an existing model. After that
the flops count should be activated and the model should be run on an input
image.
Example:
fcn = add_flops_counting_methods(fcn)
fcn = fcn.cuda().train()
fcn.start_flops_count()
_ = fcn(batch)
fcn.compute_average_flops_cost() / 1e9 / 2 # Result in GFLOPs per image in batch
Important: dividing by 2 only works for resnet models -- see below for the details
of flops computation.
Attention: we are counting multiply-add as two flops in this work, because in
most resnet models convolutions are bias-free (BN layers act as bias there)
and it makes sense to count muliply and add as separate flops therefore.
This is why in the above example we divide by 2 in order to be consistent with
most modern benchmarks. For example in "Spatially Adaptive Computatin Time for Residual
Networks" by Figurnov et al multiply-add was counted as two flops.
This module computes the average flops which is necessary for dynamic networks which
have different number of executed layers. For static networks it is enough to run the network
once and get statistics (above example).
Implementation:
The module works by adding batch_count to the main module which tracks the sum
of all batch sizes that were run through the network.
Also each convolutional layer of the network tracks the overall number of flops
performed.
The parameters are updated with the help of registered hook-functions which
are being called each time the respective layer is executed.
Parameters
----------
net_main_module : torch.nn.Module
Main module containing network
Returns
-------
net_main_module : torch.nn.Module
Updated main module with new methods/attributes that are used
to compute flops.
"""
# adding additional methods to the existing module object,
# this is done this way so that each function has access to self object
net_main_module.start_flops_count = start_flops_count.__get__(net_main_module)
net_main_module.stop_flops_count = stop_flops_count.__get__(net_main_module)
net_main_module.reset_flops_count = reset_flops_count.__get__(net_main_module)
net_main_module.compute_average_flops_cost = compute_average_flops_cost.__get__(net_main_module)
net_main_module.reset_flops_count()
# Adding varialbles necessary for masked flops computation
net_main_module.apply(add_flops_mask_variable_or_reset)
return net_main_module
def compute_average_flops_cost(self):
"""
A method that will be available after add_flops_counting_methods() is called
on a desired net object.
Returns current mean flops consumption per image.
"""
batches_count = self.__batch_counter__
flops_sum = 0
for module in self.modules():
if isinstance(module, torch.nn.Conv2d):
flops_sum += module.__flops__
return flops_sum / batches_count
def start_flops_count(self):
"""
A method that will be available after add_flops_counting_methods() is called
on a desired net object.
Activates the computation of mean flops consumption per image.
Call it before you run the network.
"""
add_batch_counter_hook_function(self)
self.apply(add_flops_counter_hook_function)
def stop_flops_count(self):
"""
A method that will be available after add_flops_counting_methods() is called
on a desired net object.
Stops computing the mean flops consumption per image.
Call whenever you want to pause the computation.
"""
remove_batch_counter_hook_function(self)
self.apply(remove_flops_counter_hook_function)
def reset_flops_count(self):
"""
A method that will be available after add_flops_counting_methods() is called
on a desired net object.
Resets statistics computed so far.
"""
add_batch_counter_variables_or_reset(self)
self.apply(add_flops_counter_variable_or_reset)
def add_flops_mask(module, mask):
def add_flops_mask_func(module):
if isinstance(module, torch.nn.Conv2d):
module.__mask__ = mask
module.apply(add_flops_mask_func)
def remove_flops_mask(module):
module.apply(add_flops_mask_variable_or_reset)
# ---- Internal functions
def conv_flops_counter_hook(conv_module, input, output):
# Can have multiple inputs, getting the first one
input = input[0]
batch_size = input.shape[0]
output_height, output_width = output.shape[2:]
kernel_height, kernel_width = conv_module.kernel_size
in_channels = conv_module.in_channels
out_channels = conv_module.out_channels
groups = conv_module.groups
# We count multiply-add as 2 flops
conv_per_position_flops = 2 * kernel_height * kernel_width * in_channels * out_channels / groups
active_elements_count = batch_size * output_height * output_width
if conv_module.__mask__ is not None:
# (b, 1, h, w)
flops_mask = conv_module.__mask__.expand(batch_size, 1, output_height, output_width)
active_elements_count = flops_mask.sum()
overall_conv_flops = conv_per_position_flops * active_elements_count
bias_flops = 0
if conv_module.bias is not None:
bias_flops = out_channels * active_elements_count
overall_flops = overall_conv_flops + bias_flops
conv_module.__flops__ += overall_flops
def batch_counter_hook(module, input, output):
# Can have multiple inputs, getting the first one
input = input[0]
batch_size = input.shape[0]
module.__batch_counter__ += batch_size
def add_batch_counter_variables_or_reset(module):
module.__batch_counter__ = 0
def add_batch_counter_hook_function(module):
if hasattr(module, '__batch_counter_handle__'):
return
handle = module.register_forward_hook(batch_counter_hook)
module.__batch_counter_handle__ = handle
def remove_batch_counter_hook_function(module):
if hasattr(module, '__batch_counter_handle__'):
module.__batch_counter_handle__.remove()
del module.__batch_counter_handle__
def add_flops_counter_variable_or_reset(module):
if isinstance(module, torch.nn.Conv2d):
module.__flops__ = 0
def add_flops_counter_hook_function(module):
if isinstance(module, torch.nn.Conv2d):
if hasattr(module, '__flops_handle__'):
return
handle = module.register_forward_hook(conv_flops_counter_hook)
module.__flops_handle__ = handle
def remove_flops_counter_hook_function(module):
if isinstance(module, torch.nn.Conv2d):
if hasattr(module, '__flops_handle__'):
module.__flops_handle__.remove()
del module.__flops_handle__
# --- Masked flops counting
# Also being run in the initialization
def add_flops_mask_variable_or_reset(module):
if isinstance(module, torch.nn.Conv2d):
module.__mask__ = None
def count_flops(model, batch_size, device, dtype, input_size, in_channels, *params):
net = model(*params)
# print(net)
net = add_flops_counting_methods(net)
net.to(device=device, dtype=dtype)
net = net.train()
batch = torch.randn(batch_size, in_channels, input_size, input_size).to(device=device, dtype=dtype)
net.start_flops_count()
_ = net(batch)
return net.compute_average_flops_cost() / 2 # Result in FLOPs