Distributed training

Joseph Redmon

2014-12-07 28e21152728cbea617948671df064ec75c7953e5

 src/convolutional_layer.c

@@ -59,11 +59,9 @@

    layer->filters = calloc(c*n*size*size, sizeof(float));
    layer->filter_updates = calloc(c*n*size*size, sizeof(float));
    layer->filter_momentum = calloc(c*n*size*size, sizeof(float));

    layer->biases = calloc(n, sizeof(float));
    layer->bias_updates = calloc(n, sizeof(float));
    layer->bias_momentum = calloc(n, sizeof(float));
    float scale = 1./(size*size*c);
    scale = .01;
    for(i = 0; i < c*n*size*size; ++i) layer->filters[i] = scale*2*(rand_uniform()-.5);
@@ -77,14 +75,13 @@
    layer->col_image = calloc(out_h*out_w*size*size*c, sizeof(float));
    layer->output = calloc(layer->batch*out_h * out_w * n, sizeof(float));
    layer->delta  = calloc(layer->batch*out_h * out_w * n, sizeof(float));

    #ifdef GPU
    layer->filters_cl = cl_make_array(layer->filters, c*n*size*size);
    layer->filter_updates_cl = cl_make_array(layer->filter_updates, c*n*size*size);
    layer->filter_momentum_cl = cl_make_array(layer->filter_momentum, c*n*size*size);

    layer->biases_cl = cl_make_array(layer->biases, n);
    layer->bias_updates_cl = cl_make_array(layer->bias_updates, n);
    layer->bias_momentum_cl = cl_make_array(layer->bias_momentum, n);

    layer->col_image_cl = cl_make_array(layer->col_image, out_h*out_w*size*size*c);
    layer->delta_cl = cl_make_array(layer->delta, layer->batch*out_h*out_w*n);
@@ -394,12 +391,16 @@
{
    cl_read_array(layer.filters_cl, layer.filters, layer.c*layer.n*layer.size*layer.size);
    cl_read_array(layer.biases_cl, layer.biases, layer.n);
    cl_read_array(layer.filter_updates_cl, layer.filter_updates, layer.c*layer.n*layer.size*layer.size);
    cl_read_array(layer.bias_updates_cl, layer.bias_updates, layer.n);
}

void push_convolutional_layer(convolutional_layer layer)
{
    cl_write_array(layer.filters_cl, layer.filters, layer.c*layer.n*layer.size*layer.size);
    cl_write_array(layer.biases_cl, layer.biases, layer.n);
    cl_write_array(layer.filter_updates_cl, layer.filter_updates, layer.c*layer.n*layer.size*layer.size);
    cl_write_array(layer.bias_updates_cl, layer.bias_updates, layer.n);
}

void update_convolutional_layer_gpu(convolutional_layer layer)