Refactored connected to use blas

Joseph Redmon

2014-10-14 7756cccb793bb4950c241f2804195ea859d1b407

Refactored connected to use blas

 Makefile

@@ -1,5 +1,5 @@
CC=gcc
GPU=1
GPU=0
COMMON=-Wall -Wfatal-errors `pkg-config --cflags opencv` -I/usr/local/cuda/include/
ifeq ($(GPU), 1) 
COMMON+=-DGPU
@@ -7,6 +7,7 @@
endif
UNAME = $(shell uname)
OPTS=-Ofast -flto
OPTS=-Ofast -flto
ifeq ($(UNAME), Darwin)
COMMON+= -isystem /usr/local/Cellar/opencv/2.4.6.1/include/opencv -isystem /usr/local/Cellar/opencv/2.4.6.1/include
ifeq ($(GPU), 1)

 src/cnn.c

@@ -916,8 +916,8 @@
    //test_ensemble();
    //test_nist_single();
    //test_nist();
    //train_nist();
    test_convolutional_layer();
    train_nist();
    //test_convolutional_layer();
    //test_col2im();
    //test_cifar10();
    //train_cifar10();

 src/connected_layer.c

@@ -26,7 +26,6 @@

    layer->weight_updates = calloc(inputs*outputs, sizeof(float));
    //layer->weight_adapt = calloc(inputs*outputs, sizeof(float));
    layer->weight_momentum = calloc(inputs*outputs, sizeof(float));
    layer->weights = calloc(inputs*outputs, sizeof(float));
    float scale = 1./inputs;
    scale = .05;
@@ -35,7 +34,6 @@

    layer->bias_updates = calloc(outputs, sizeof(float));
    //layer->bias_adapt = calloc(outputs, sizeof(float));
    layer->bias_momentum = calloc(outputs, sizeof(float));
    layer->biases = calloc(outputs, sizeof(float));
    for(i = 0; i < outputs; ++i){
        //layer->biases[i] = rand_normal()*scale + scale;
@@ -50,24 +48,19 @@

void update_connected_layer(connected_layer layer)
{
    int i;
    for(i = 0; i < layer.outputs; ++i){
        layer.bias_momentum[i] = layer.learning_rate*(layer.bias_updates[i]) + layer.momentum*layer.bias_momentum[i];
        layer.biases[i] += layer.bias_momentum[i];
    }
    for(i = 0; i < layer.outputs*layer.inputs; ++i){
        layer.weight_momentum[i] = layer.learning_rate*(layer.weight_updates[i] - layer.decay*layer.weights[i]) + layer.momentum*layer.weight_momentum[i];
        layer.weights[i] += layer.weight_momentum[i];
    }
    memset(layer.bias_updates, 0, layer.outputs*sizeof(float));
    memset(layer.weight_updates, 0, layer.outputs*layer.inputs*sizeof(float));
    axpy_cpu(layer.outputs, layer.learning_rate, layer.bias_updates, 1, layer.biases, 1);
    scal_cpu(layer.outputs, layer.momentum, layer.bias_updates, 1);

    scal_cpu(layer.inputs*layer.outputs, 1.-layer.learning_rate*layer.decay, layer.weights, 1);
    axpy_cpu(layer.inputs*layer.outputs, layer.learning_rate, layer.weight_updates, 1, layer.weights, 1);
    scal_cpu(layer.inputs*layer.outputs, layer.momentum, layer.weight_updates, 1);
}

void forward_connected_layer(connected_layer layer, float *input)
{
    int i;
    for(i = 0; i < layer.batch; ++i){
        memcpy(layer.output+i*layer.outputs, layer.biases, layer.outputs*sizeof(float));
        copy_cpu(layer.outputs, layer.biases, 1, layer.output + i*layer.outputs, 1);
    }
    int m = layer.batch;
    int k = layer.inputs;
@@ -82,8 +75,8 @@
void backward_connected_layer(connected_layer layer, float *input, float *delta)
{
    int i;
    gradient_array(layer.output, layer.outputs*layer.batch, layer.activation, layer.delta);
    for(i = 0; i < layer.outputs*layer.batch; ++i){
        layer.delta[i] *= gradient(layer.output[i], layer.activation);
        layer.bias_updates[i%layer.outputs] += layer.delta[i];
    }
    int m = layer.inputs;

 src/connected_layer.h

@@ -21,9 +21,6 @@
    float *weight_adapt;
    float *bias_adapt;

    float *weight_momentum;
    float *bias_momentum;

    float *output;
    float *delta;
    

 src/network.c

@@ -229,6 +229,8 @@
        return layer.output;
    } else if(net.types[i] == DROPOUT){
        return get_network_output_layer(net, i-1);
    } else if(net.types[i] == FREEWEIGHT){
        return get_network_output_layer(net, i-1);
    } else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.output;
@@ -258,6 +260,8 @@
        return layer.delta;
    } else if(net.types[i] == DROPOUT){
        return get_network_delta_layer(net, i-1);
    } else if(net.types[i] == FREEWEIGHT){
        return get_network_delta_layer(net, i-1);
    } else if(net.types[i] == CONNECTED){
        connected_layer layer = *(connected_layer *)net.layers[i];
        return layer.delta;
@@ -424,6 +428,10 @@
        dropout_layer layer = *(dropout_layer *) net.layers[i];
        return layer.inputs;
    }
    else if(net.types[i] == FREEWEIGHT){
        freeweight_layer layer = *(freeweight_layer *) net.layers[i];
        return layer.inputs;
    }
    else if(net.types[i] == SOFTMAX){
        softmax_layer layer = *(softmax_layer *)net.layers[i];
        return layer.inputs;
@@ -451,6 +459,10 @@
        dropout_layer layer = *(dropout_layer *) net.layers[i];
        return layer.inputs;
    }
    else if(net.types[i] == FREEWEIGHT){
        freeweight_layer layer = *(freeweight_layer *) net.layers[i];
        return layer.inputs;
    }
    else if(net.types[i] == SOFTMAX){
        softmax_layer layer = *(softmax_layer *)net.layers[i];
        return layer.inputs;