avgpool_layer_kernels.cu
1.59 KB
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
#include "cuda_runtime.h"
#include "curand.h"
#include "cublas_v2.h"
extern "C" {
#include "avgpool_layer.h"
#include "cuda.h"
}
__global__ void forward_avgpool_layer_kernel(int n, int w, int h, int c, float *input, float *output)
{
int id = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
if(id >= n) return;
int k = id % c;
id /= c;
int b = id;
int i;
int out_index = (k + c*b);
output[out_index] = 0;
for(i = 0; i < w*h; ++i){
int in_index = i + h*w*(k + b*c);
output[out_index] += input[in_index];
}
output[out_index] /= w*h;
}
__global__ void backward_avgpool_layer_kernel(int n, int w, int h, int c, float *in_delta, float *out_delta)
{
int id = (blockIdx.x + blockIdx.y*gridDim.x) * blockDim.x + threadIdx.x;
if(id >= n) return;
int k = id % c;
id /= c;
int b = id;
int i;
int out_index = (k + c*b);
for(i = 0; i < w*h; ++i){
int in_index = i + h*w*(k + b*c);
in_delta[in_index] += out_delta[out_index] / (w*h);
}
}
extern "C" void forward_avgpool_layer_gpu(avgpool_layer layer, network net)
{
size_t n = layer.c*layer.batch;
forward_avgpool_layer_kernel<<<cuda_gridsize(n), BLOCK>>>(n, layer.w, layer.h, layer.c, net.input_gpu, layer.output_gpu);
check_error(cudaPeekAtLastError());
}
extern "C" void backward_avgpool_layer_gpu(avgpool_layer layer, network net)
{
size_t n = layer.c*layer.batch;
backward_avgpool_layer_kernel<<<cuda_gridsize(n), BLOCK>>>(n, layer.w, layer.h, layer.c, net.delta_gpu, layer.delta_gpu);
check_error(cudaPeekAtLastError());
}