forked from BVLC/caffe
-
Notifications
You must be signed in to change notification settings - Fork 263
/
tanh_layer.cu
60 lines (51 loc) · 1.96 KB
/
tanh_layer.cu
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
// TanH neuron activation function layer.
// Adapted from ReLU layer code written by Yangqing Jia
#include <vector>
#include "caffe/layers/tanh_layer.hpp"
namespace caffe {
template <typename Dtype>
__global__ void TanHForward(const int n, const Dtype* in, Dtype* out) {
CUDA_KERNEL_LOOP(index, n) {
out[index] = tanh(in[index]);
}
}
template <typename Ftype, typename Btype>
void TanHLayer<Ftype, Btype>::Forward_gpu(const vector<Blob*>& bottom,
const vector<Blob*>& top) {
const Ftype* bottom_data = bottom[0]->gpu_data<Ftype>();
Ftype* top_data = top[0]->mutable_gpu_data<Ftype>();
const int count = bottom[0]->count();
cudaStream_t stream = Caffe::thread_stream();
// NOLINT_NEXT_LINE(whitespace/operators)
TanHForward<<<CAFFE_GET_BLOCKS(count), CAFFE_CUDA_NUM_THREADS, 0, stream>>>(
count, bottom_data, top_data);
CUDA_POST_KERNEL_CHECK;
CUDA_CHECK(cudaStreamSynchronize(stream));
}
template <typename Dtype>
__global__ void TanHBackward(const int n, const Dtype* in_diff,
const Dtype* out_data, Dtype* out_diff) {
CUDA_KERNEL_LOOP(index, n) {
float tanhx = out_data[index];
out_diff[index] = in_diff[index] * (Dtype(1.) - tanhx * tanhx);
}
}
template <typename Ftype, typename Btype>
void TanHLayer<Ftype, Btype>::Backward_gpu(const vector<Blob*>& top,
const vector<bool>& propagate_down,
const vector<Blob*>& bottom) {
if (propagate_down[0]) {
const Btype* top_data = top[0]->gpu_data<Btype>();
const Btype* top_diff = top[0]->gpu_diff<Btype>();
Btype* bottom_diff = bottom[0]->mutable_gpu_diff<Btype>();
const int count = bottom[0]->count();
cudaStream_t stream = Caffe::thread_stream();
// NOLINT_NEXT_LINE(whitespace/operators)
TanHBackward<<<CAFFE_GET_BLOCKS(count), CAFFE_CUDA_NUM_THREADS, 0, stream>>>(
count, top_diff, top_data, bottom_diff);
CUDA_POST_KERNEL_CHECK;
CUDA_CHECK(cudaStreamSynchronize(stream));
}
}
INSTANTIATE_LAYER_GPU_FUNCS_FB(TanHLayer);
} // namespace caffe