blob: 83fba1477e3dea459b2565192a5e64752dc8c020 (
plain)
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
|
// Copyright 2013 Yangqing Jia
#include <ctime>
#include <string>
#include <vector>
#include "cuda_runtime.h"
#include "fcntl.h"
#include "google/protobuf/text_format.h"
#include "caffe/blob.hpp"
#include "caffe/common.hpp"
#include "caffe/net.hpp"
#include "caffe/filler.hpp"
#include "caffe/proto/caffe.pb.h"
#include "caffe/util/io.hpp"
#include "caffe/solver.hpp"
using boost::shared_ptr;
using namespace caffe; // NOLINT(build/namespaces)
int main(int argc, char** argv) {
int total_iter = 50;
if (argc < 2) {
LOG(ERROR) << "net_speed_benchmark net_proto [iterations=50] [CPU/GPU] "
<< "[Device_id=0]";
return 0;
}
if (argc >=3) {
total_iter = atoi(argv[2]);
}
LOG(ERROR) << "Testing for " << total_iter << "Iterations.";
if (argc >= 4 && strcmp(argv[3], "GPU") == 0) {
LOG(ERROR) << "Using GPU";
uint device_id = 0;
if (argc >= 5 && strcmp(argv[3], "GPU") == 0) {
device_id = atoi(argv[4]);
}
LOG(ERROR) << "Using Device_id=" << device_id;
Caffe::SetDevice(device_id);
Caffe::set_mode(Caffe::GPU);
} else {
LOG(ERROR) << "Using CPU";
Caffe::set_mode(Caffe::CPU);
}
Caffe::set_phase(Caffe::TRAIN);
NetParameter net_param;
ReadProtoFromTextFile(argv[1],
&net_param);
Net<float> caffe_net(net_param);
// Run the network without training.
LOG(ERROR) << "Performing Forward";
// Note that for the speed benchmark, we will assume that the network does
// not take any input blobs.
caffe_net.Forward(vector<Blob<float>*>());
LOG(ERROR) << "Performing Backward";
LOG(ERROR) << "Initial loss: " << caffe_net.Backward();
const vector<shared_ptr<Layer<float> > >& layers = caffe_net.layers();
vector<vector<Blob<float>*> >& bottom_vecs = caffe_net.bottom_vecs();
vector<vector<Blob<float>*> >& top_vecs = caffe_net.top_vecs();
LOG(ERROR) << "*** Benchmark begins ***";
if (Caffe::mode() == Caffe::GPU) {
cudaDeviceSynchronize();
}
clock_t forward_start = clock();
for (int i = 0; i < layers.size(); ++i) {
const string& layername = layers[i]->layer_param().name();
if (Caffe::mode() == Caffe::GPU) {
cudaDeviceSynchronize();
}
clock_t start = clock();
for (int j = 0; j < total_iter; ++j) {
layers[i]->Forward(bottom_vecs[i], &top_vecs[i]);
}
if (Caffe::mode() == Caffe::GPU) {
cudaDeviceSynchronize();
}
LOG(ERROR) << layername << "\tforward: "
<< static_cast<float>(clock() - start) / CLOCKS_PER_SEC
<< " seconds.";
}
LOG(ERROR) << "Forward pass: "
<< static_cast<float>(clock() - forward_start) / CLOCKS_PER_SEC
<< " seconds.";
clock_t backward_start = clock();
for (int i = layers.size() - 1; i >= 0; --i) {
const string& layername = layers[i]->layer_param().name();
if (Caffe::mode() == Caffe::GPU) {
cudaDeviceSynchronize();
}
clock_t start = clock();
for (int j = 0; j < total_iter; ++j) {
layers[i]->Backward(top_vecs[i], true, &bottom_vecs[i]);
}
if (Caffe::mode() == Caffe::GPU) {
cudaDeviceSynchronize();
}
LOG(ERROR) << layername << "\tbackward: "
<< static_cast<float>(clock() - start) / CLOCKS_PER_SEC
<< " seconds.";
}
LOG(ERROR) << "Backward pass: "
<< static_cast<float>(clock() - backward_start) / CLOCKS_PER_SEC
<< " seconds.";
LOG(ERROR) << "Total Time: "
<< static_cast<float>(clock() - forward_start) / CLOCKS_PER_SEC
<< " seconds.";
LOG(ERROR) << "*** Benchmark ends ***";
return 0;
}
|
