mirror of https://github.com/pjreddie/darknet.git
83 lines
2.5 KiB
C
83 lines
2.5 KiB
C
#include "softmax_layer.h"
|
|
#include "blas.h"
|
|
#include "cuda.h"
|
|
#include <float.h>
|
|
#include <math.h>
|
|
#include <stdlib.h>
|
|
#include <stdio.h>
|
|
#include <assert.h>
|
|
|
|
softmax_layer make_softmax_layer(int batch, int inputs, int groups)
|
|
{
|
|
assert(inputs%groups == 0);
|
|
fprintf(stderr, "softmax %4d\n", inputs);
|
|
softmax_layer l = {0};
|
|
l.type = SOFTMAX;
|
|
l.batch = batch;
|
|
l.groups = groups;
|
|
l.inputs = inputs;
|
|
l.outputs = inputs;
|
|
l.output = calloc(inputs*batch, sizeof(float));
|
|
l.delta = calloc(inputs*batch, sizeof(float));
|
|
|
|
l.forward = forward_softmax_layer;
|
|
l.backward = backward_softmax_layer;
|
|
#ifdef GPU
|
|
l.forward_gpu = forward_softmax_layer_gpu;
|
|
l.backward_gpu = backward_softmax_layer_gpu;
|
|
|
|
l.output_gpu = cuda_make_array(l.output, inputs*batch);
|
|
l.delta_gpu = cuda_make_array(l.delta, inputs*batch);
|
|
#endif
|
|
return l;
|
|
}
|
|
|
|
void forward_softmax_layer(const softmax_layer l, network_state state)
|
|
{
|
|
if(l.softmax_tree){
|
|
int i;
|
|
int count = 0;
|
|
for (i = 0; i < l.softmax_tree->groups; ++i) {
|
|
int group_size = l.softmax_tree->group_size[i];
|
|
softmax_cpu(state.input + count, group_size, l.batch, l.inputs, 1, 0, 1, l.temperature, l.output + count);
|
|
count += group_size;
|
|
}
|
|
} else {
|
|
softmax_cpu(state.input, l.inputs/l.groups, l.batch, l.inputs, l.groups, l.inputs/l.groups, 1, l.temperature, l.output);
|
|
}
|
|
}
|
|
|
|
void backward_softmax_layer(const softmax_layer l, network_state state)
|
|
{
|
|
axpy_cpu(l.inputs*l.batch, 1, l.delta, 1, state.delta, 1);
|
|
}
|
|
|
|
#ifdef GPU
|
|
|
|
void pull_softmax_layer_output(const softmax_layer layer)
|
|
{
|
|
cuda_pull_array(layer.output_gpu, layer.output, layer.inputs*layer.batch);
|
|
}
|
|
|
|
void forward_softmax_layer_gpu(const softmax_layer l, network_state state)
|
|
{
|
|
if(l.softmax_tree){
|
|
int i;
|
|
int count = 0;
|
|
for (i = 0; i < l.softmax_tree->groups; ++i) {
|
|
int group_size = l.softmax_tree->group_size[i];
|
|
softmax_gpu(state.input + count, group_size, l.batch, l.inputs, 1, 0, 1, l.temperature, l.output_gpu + count);
|
|
count += group_size;
|
|
}
|
|
} else {
|
|
softmax_gpu(state.input, l.inputs/l.groups, l.batch, l.inputs, l.groups, l.inputs/l.groups, 1, l.temperature, l.output_gpu);
|
|
}
|
|
}
|
|
|
|
void backward_softmax_layer_gpu(const softmax_layer layer, network_state state)
|
|
{
|
|
axpy_ongpu(layer.batch*layer.inputs, 1, layer.delta_gpu, 1, state.delta, 1);
|
|
}
|
|
|
|
#endif
|