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
|
#include <algorithm>
#include <cfloat>
#include <vector>
#include "thrust/device_vector.h"
#include "caffe/layer.hpp"
#include "caffe/util/math_functions.hpp"
#include "caffe/vision_layers.hpp"
namespace caffe {
template <typename Dtype>
__global__ void kernel_channel_max(const int num, const int channels,
const int spatial_dim, const Dtype* data, Dtype* out) {
CUDA_KERNEL_LOOP(index, num * spatial_dim) {
int n = index / spatial_dim;
int s = index % spatial_dim;
Dtype maxval = -FLT_MAX;
for (int c = 0; c < channels; ++c) {
maxval = max(data[(n * channels + c) * spatial_dim + s], maxval);
}
out[index] = maxval;
}
}
template <typename Dtype>
__global__ void kernel_channel_subtract(const int count,
const int num, const int channels,
const int spatial_dim, const Dtype* channel_max, Dtype* data) {
CUDA_KERNEL_LOOP(index, count) {
int n = index / channels / spatial_dim;
int s = index % spatial_dim;
data[index] -= channel_max[n * spatial_dim + s];
}
}
template <typename Dtype>
__global__ void kernel_exp(const int count, const Dtype* data, Dtype* out) {
CUDA_KERNEL_LOOP(index, count) {
out[index] = exp(data[index]);
}
}
template <typename Dtype>
__global__ void kernel_channel_sum(const int num, const int channels,
const int spatial_dim, const Dtype* data, Dtype* channel_sum) {
CUDA_KERNEL_LOOP(index, num * spatial_dim) {
int n = index / spatial_dim;
int s = index % spatial_dim;
Dtype sum = 0;
for (int c = 0; c < channels; ++c) {
sum += data[(n * channels + c) * spatial_dim + s];
}
channel_sum[index] = sum;
}
}
template <typename Dtype>
__global__ void kernel_channel_div(const int count,
const int num, const int channels,
const int spatial_dim, const Dtype* channel_sum, Dtype* data) {
CUDA_KERNEL_LOOP(index, count) {
int n = index / channels / spatial_dim;
int s = index % spatial_dim;
data[index] /= channel_sum[n * spatial_dim + s];
}
}
template <typename Dtype>
__global__ void kernel_channel_dot(const int num, const int channels,
const int spatial_dim, const Dtype* data_1, const Dtype* data_2,
Dtype* channel_dot) {
CUDA_KERNEL_LOOP(index, num * spatial_dim) {
int n = index / spatial_dim;
int s = index % spatial_dim;
Dtype dot = 0;
for (int c = 0; c < channels; ++c) {
dot += (data_1[(n * channels + c) * spatial_dim + s]
* data_2[(n * channels + c) * spatial_dim + s]);
}
channel_dot[index] = dot;
}
}
template <typename Dtype>
void SoftmaxLayer<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom,
const vector<Blob<Dtype>*>& top) {
const Dtype* bottom_data = bottom[0]->gpu_data();
Dtype* top_data = top[0]->mutable_gpu_data();
Dtype* scale_data = scale_.mutable_gpu_data();
int count = bottom[0]->count();
int channels = top[0]->shape(softmax_axis_);
caffe_copy(count, bottom_data, top_data);
// We need to subtract the max to avoid numerical issues, compute the exp,
// and then normalize.
// compute max
// NOLINT_NEXT_LINE(whitespace/operators)
kernel_channel_max<Dtype><<<CAFFE_GET_BLOCKS(outer_num_ * inner_num_),
CAFFE_CUDA_NUM_THREADS>>>(outer_num_, channels, inner_num_, top_data,
scale_data);
// subtract
// NOLINT_NEXT_LINE(whitespace/operators)
kernel_channel_subtract<Dtype><<<CAFFE_GET_BLOCKS(count),
CAFFE_CUDA_NUM_THREADS>>>(count, outer_num_, channels, inner_num_,
scale_data, top_data);
// exponentiate
// NOLINT_NEXT_LINE(whitespace/operators)
kernel_exp<Dtype><<<CAFFE_GET_BLOCKS(count), CAFFE_CUDA_NUM_THREADS>>>(
count, top_data, top_data);
// sum after exp
// NOLINT_NEXT_LINE(whitespace/operators)
kernel_channel_sum<Dtype><<<CAFFE_GET_BLOCKS(outer_num_ * inner_num_),
CAFFE_CUDA_NUM_THREADS>>>(outer_num_, channels, inner_num_, top_data,
scale_data);
// divide
// NOLINT_NEXT_LINE(whitespace/operators)
kernel_channel_div<Dtype><<<CAFFE_GET_BLOCKS(count),
CAFFE_CUDA_NUM_THREADS>>>(count, outer_num_, channels, inner_num_,
scale_data, top_data);
}
template <typename Dtype>
void SoftmaxLayer<Dtype>::Backward_gpu(const vector<Blob<Dtype>*>& top,
const vector<bool>& propagate_down, const vector<Blob<Dtype>*>& bottom) {
const Dtype* top_diff = top[0]->gpu_diff();
const Dtype* top_data = top[0]->gpu_data();
Dtype* bottom_diff = bottom[0]->mutable_gpu_diff();
Dtype* scale_data = scale_.mutable_gpu_data();
int count = top[0]->count();
int channels = top[0]->shape(softmax_axis_);
caffe_copy(count, top_diff, bottom_diff);
// Compute inner1d(top_diff, top_data) and subtract them from the bottom diff.
// NOLINT_NEXT_LINE(whitespace/operators)
kernel_channel_dot<Dtype><<<CAFFE_GET_BLOCKS(outer_num_ * inner_num_),
CAFFE_CUDA_NUM_THREADS>>>(outer_num_, channels, inner_num_,
top_diff, top_data, scale_data);
// NOLINT_NEXT_LINE(whitespace/operators)
kernel_channel_subtract<Dtype><<<CAFFE_GET_BLOCKS(count),
CAFFE_CUDA_NUM_THREADS>>>(count, outer_num_, channels, inner_num_,
scale_data, bottom_diff);
// elementwise multiplication
caffe_gpu_mul<Dtype>(top[0]->count(), bottom_diff, top_data, bottom_diff);
}
INSTANTIATE_LAYER_GPU_FUNCS(SoftmaxLayer);
} // namespace caffe
|
