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
|
#include <stdio.h>
// Replace this with the saxpy from cuBlas or whatever?
// I doubt it matters, but it's definitely weird to have this
void __device__
saxpy(float* X, const float* Y, float scale, int n)
{
for (int i=0; i < n; ++i)
X[i] += Y[i] * scale;
}
void __global__
maxout(float* best__bo, int* which__bo,
const float* cands__bop, int B, int O, int P)
{
int b = blockIdx.x * blockDim.x + threadIdx.x;
if (b >= B) return;
for (int o=0; o < O; ++o)
{
which__bo[0] = 0;
best__bo[0] = cands__bop[0];
cands__bop += 1;
for (int p=1; p < P; ++p)
{
if (cands__bop[0] > best__bo[0])
{
which__bo[0] = p;
best__bo[0] = cands__bop[0];
}
cands__bop += 1;
}
best__bo += 1;
which__bo += 1;
}
}
void __global__
mean_pool(float* means__bo,
const float* X__to, const int* lengths__b, int B, int T, int O)
{
// Compute means of a batch of concatenated sequences, using the lengths.'''
int b = blockIdx.x; // Batch-item we're averaging
if (b >= B) return;
// Go to the regions we're working on
for (int i=0; i < b; ++i) {
means__bo += O;
X__to += lengths__b[i] * O;
}
int length = lengths__b[b];
// Each invocation of the kernel averages one batch.
float scale = 1. / length;
for (int _=0; _ < length; ++_) // Iterate over rows
{
saxpy(means__bo, X__to, scale, O);
X__to += O;
}
}
void __global__
sum_pool(float* sums__bo,
const float* X__to, const int* lengths__b, int B, int T, int O)
{
// Compute sums of a batch of concatenated sequences, using the lengths.'''
int b = blockIdx.x; // Batch-item we're summing
if (b >= B) return;
// Go to the regions we're working on
for (int i=0; i < b; ++i) {
sums__bo += O;
X__to += lengths__b[i] * O;
}
int length = lengths__b[b];
// Each invocation of the kernel sums one batch.
for (int _=0; _ < length; ++_) // Iterate over rows
{
saxpy(sums__bo, X__to, 1.0, O);
X__to += O;
}
}
void __global__
max_pool(float* maxes__bo, int* which__bo,
const float* X__to, const int* lengths__b, int B, int T, int O)
{
// Compute means of a batch of concatenated sequences, using the lengths.'''
int b = blockIdx.x; // Batch-item we're averaging
if (b >= B) return;
// Go to the regions we're working on
for (int i=0; i < b; ++i) {
maxes__bo += O;
which__bo += O;
X__to += lengths__b[i] * O;
}
// Each invocation of the kernel maxes one batch.
// Start by assuming maxes are at i=0
for (int j=0; j < O; ++j) {
maxes__bo[j] = X__to[j];
which__bo[j] = 0;
}
X__to += O;
int length = lengths__b[b];
for (int i=1; i < length; ++i) // Iterate over rows
{
for (int j=0; j < O; ++j)
{
if (X__to[j] > maxes__bo[j])
{
maxes__bo[j] = X__to[j];
which__bo[j] = i;
}
}
X__to += O;
}
}
void __global__
backprop_mean_pool(float* dX__to, const float* d_means__bo, const int* lengths__b,
int B, int T, int O)
{
int b = blockIdx.x; // Batch-item we're averaging
if (b >= B) return;
// Go to the regions we're working on
for (int i=0; i < b; ++i) {
d_means__bo += O;
dX__to += lengths__b[i] * O;
}
int length = lengths__b[b];
float scale = 1./ length;
for (int _=0; _ < length; _++)
{
saxpy(dX__to, d_means__bo, scale, O);
dX__to += O;
}
}
void __global__
backprop_sum_pool(float* dX__to, const float* d_sum__bo, const int* lengths__b,
int B, int T, int O)
{
int b = blockIdx.x; // Batch-item we're averaging
if (b >= B) return;
// Go to the regions we're working on
for (int i=0; i < b; ++i) {
d_sum__bo += O;
dX__to += lengths__b[i] * O;
}
int length = lengths__b[b];
for (int _=0; _ < length; _++)
{
saxpy(dX__to, d_sum__bo, 1.0, O);
dX__to += O;
}
}
void __global__
backprop_max_pool(float* dX__to,
const float* d_maxes__bo, const int* which__bo, const int* lengths__b, int B, int T, int O)
{
int b = blockIdx.x; // Batch-item we're averaging
if (b >= B) return;
// Go to the regions we're working on
for (int i=0; i < b; ++i) {
d_maxes__bo += O;
which__bo += O;
dX__to += lengths__b[i] * O;
}
int length = lengths__b[b];
for (int i=0; i < length; ++i)
{
for (int j=0; j < O; ++j)
{
if (which__bo[j] == i)
dX__to[j] += d_maxes__bo[j];
}
dX__to += O;
}
}
|
