-
Notifications
You must be signed in to change notification settings - Fork 254
/
pooling.cpp
231 lines (206 loc) · 8.83 KB
/
pooling.cpp
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
#include <algorithm>
#include <limits>
#include <cstring>
#include <cstdio>
template <typename T>
void max_pooling_fwd(const T* global_input, T *global_output, size_t *global_mask,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
int input_offset = width*height;
int output_offset = pooled_width*pooled_height;
#pragma omp parallel for
for (int n = 0; n < num; ++n) {
for (int c = 0; c < channels; ++c) {
int offset = (n * channels + c);
const T *input = global_input + input_offset * offset;
T *output = global_output + output_offset * offset;
size_t *mask = global_mask + output_offset * offset;
for (int ph = 0; ph < pooled_height; ++ph) {
for (int pw = 0; pw < pooled_width; ++pw) {
int hstart = ph*stride_h - pad_h;
int wstart = pw*stride_w - pad_w;
int hend = std::min(hstart + kernel_h, height);
int wend = std::min(wstart + kernel_w, width);
hstart = std::max(hstart, 0);
wstart = std::max(wstart, 0);
int pool_index = ph * pooled_width + pw;
T maxval = -std::numeric_limits<T>::max();
size_t maxidx = 0;
for (int h = hstart; h < hend; ++h) {
for (int w = wstart; w < wend; ++w) {
int index = h * width + w;
if (input[index] > maxval) {
maxval = input[index];
maxidx = index;
}
}
}
output[pool_index] = maxval;
mask[pool_index] = maxidx;
}
}
}
}
}
template <typename T>
void max_pooling_bwd(T* global_input, const T *global_output, const size_t *global_mask,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
int input_offset = width*height;
int output_offset = pooled_width*pooled_height;
memset(global_input, 0, input_offset*channels*num*sizeof(T));
#pragma omp parallel for
for (int n = 0; n < num; ++n) {
for (int c = 0; c < channels; ++c) {
int offset = (n * channels + c);
T *input = global_input + input_offset * offset;
const T *output = global_output + output_offset * offset;
const size_t *mask = global_mask + output_offset * offset;
for (int ph = 0; ph < pooled_height; ++ph) {
for (int pw = 0; pw < pooled_width; ++pw) {
int pool_index = ph * pooled_width + pw;
input[mask[pool_index]] += output[pool_index];
}
}
}
}
}
template <typename T>
void mean_pooling_fwd(const T* global_input, T *global_output,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
int input_offset = width*height;
int output_offset = pooled_width*pooled_height;
int kernel_size = kernel_w * kernel_h;
#pragma omp parallel for
for (int n = 0; n < num; ++n) {
for (int c = 0; c < channels; ++c) {
int offset = (n * channels + c);
const T *input = global_input + input_offset * offset;
T *output = global_output + output_offset * offset;
for (int ph = 0; ph < pooled_height; ++ph) {
for (int pw = 0; pw < pooled_width; ++pw) {
int hstart = ph*stride_h - pad_h;
int wstart = pw*stride_w - pad_w;
int hend = std::min(hstart + kernel_h, height);
int wend = std::min(wstart + kernel_w, width);
hstart = std::max(hstart, 0);
wstart = std::max(wstart, 0);
int pool_index = ph * pooled_width + pw;
T meanval = 0;
for (int h = hstart; h < hend; ++h) {
for (int w = wstart; w < wend; ++w) {
meanval += input[h * width + w];
}
}
output[pool_index] = meanval / kernel_size;
}
}
}
}
}
template <typename T>
void mean_pooling_bwd(T* global_input, const T *global_output,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
int input_offset = width*height;
int output_offset = pooled_width*pooled_height;
int kernel_size = kernel_w * kernel_h;
memset(global_input, 0, input_offset*channels*num*sizeof(T));
#pragma omp parallel for
for (int n = 0; n < num; ++n) {
for (int c = 0; c < channels; ++c) {
int offset = (n * channels + c);
T *input = global_input + input_offset * offset;
const T *output = global_output + output_offset * offset;
for (int ph = 0; ph < pooled_height; ++ph) {
for (int pw = 0; pw < pooled_width; ++pw) {
int hstart = ph*stride_h - pad_h;
int wstart = pw*stride_w - pad_w;
int hend = std::min(hstart + kernel_h, height);
int wend = std::min(wstart + kernel_w, width);
hstart = std::max(hstart, 0);
wstart = std::max(wstart, 0);
int pool_index = ph * pooled_width + pw;
for (int h = hstart; h < hend; ++h) {
for (int w = wstart; w < wend; ++w) {
input[h * width + w] += output[pool_index] / kernel_size;
}
}
}
}
}
}
}
extern "C" {
void max_pooling_fwd_float(const float* global_input, float *global_output, size_t *global_mask,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
max_pooling_fwd(global_input, global_output, global_mask,
width, height, channels, num,
pooled_width, pooled_height,
kernel_w, kernel_h, pad_w, pad_h, stride_w, stride_h);
}
void max_pooling_fwd_double(const double* global_input, double *global_output, size_t *global_mask,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
max_pooling_fwd(global_input, global_output, global_mask,
width, height, channels, num,
pooled_width, pooled_height,
kernel_w, kernel_h, pad_w, pad_h, stride_w, stride_h);
}
void max_pooling_bwd_float(float* global_input, const float *global_output, const size_t *global_mask,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
max_pooling_bwd(global_input, global_output, global_mask,
width, height, channels, num, pooled_width, pooled_height,
kernel_w, kernel_h, pad_w, pad_h, stride_w, stride_h);
}
void max_pooling_bwd_double(double* global_input, const double *global_output, const size_t *global_mask,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
max_pooling_bwd(global_input, global_output, global_mask,
width, height, channels, num, pooled_width, pooled_height,
kernel_w, kernel_h, pad_w, pad_h, stride_w, stride_h);
}
void mean_pooling_fwd_float(const float* global_input, float *global_output,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
mean_pooling_fwd(global_input, global_output,
width, height, channels, num, pooled_width, pooled_height,
kernel_w, kernel_h, pad_w, pad_h, stride_w, stride_h);
}
void mean_pooling_fwd_double(const double* global_input, double *global_output,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
mean_pooling_fwd(global_input, global_output,
width, height, channels, num, pooled_width, pooled_height,
kernel_w, kernel_h, pad_w, pad_h, stride_w, stride_h);
}
void mean_pooling_bwd_float(float* global_input, const float *global_output,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
mean_pooling_bwd(global_input, global_output,
width, height, channels, num, pooled_width, pooled_height,
kernel_w, kernel_h, pad_w, pad_h, stride_w, stride_h);
}
void mean_pooling_bwd_double(double* global_input, const double *global_output,
int width, int height, int channels, int num,
int pooled_width, int pooled_height,
int kernel_w, int kernel_h, int pad_w, int pad_h, int stride_w, int stride_h) {
mean_pooling_bwd(global_input, global_output,
width, height, channels, num, pooled_width, pooled_height,
kernel_w, kernel_h, pad_w, pad_h, stride_w, stride_h);
}
} // extern "C"