C++ (Cpp) forward_convolutional_layer_gpuの例

コード例 #1

ファイル: crnn_layer.c プロジェクト: imaami/darknet

void forward_crnn_layer_gpu(layer_t l, network_state state)
{
    NETWORK_STATE(s);
    s.train = state.train;
    int i;
    layer_t input_layer = *(l.input_layer);
    layer_t self_layer = *(l.self_layer);
    layer_t output_layer = *(l.output_layer);

    fill_ongpu(l.outputs * l.batch * l.steps, 0, output_layer.delta_gpu, 1);
    fill_ongpu(l.hidden * l.batch * l.steps, 0, self_layer.delta_gpu, 1);
    fill_ongpu(l.hidden * l.batch * l.steps, 0, input_layer.delta_gpu, 1);
    if(state.train) fill_ongpu(l.hidden * l.batch, 0, l.state_gpu, 1);

    for (i = 0; i < l.steps; ++i) {
        s.input = state.input;
        forward_convolutional_layer_gpu(input_layer, s);

        s.input = l.state_gpu;
        forward_convolutional_layer_gpu(self_layer, s);

        float *old_state = l.state_gpu;
        if(state.train) l.state_gpu += l.hidden*l.batch;
        if(l.shortcut){
            copy_ongpu(l.hidden * l.batch, old_state, 1, l.state_gpu, 1);
        }else{
            fill_ongpu(l.hidden * l.batch, 0, l.state_gpu, 1);
        }
        axpy_ongpu(l.hidden * l.batch, 1, input_layer.output_gpu, 1, l.state_gpu, 1);
        axpy_ongpu(l.hidden * l.batch, 1, self_layer.output_gpu, 1, l.state_gpu, 1);

        s.input = l.state_gpu;
        forward_convolutional_layer_gpu(output_layer, s);

        state.input += l.inputs*l.batch;
        increment_layer(&input_layer, 1);
        increment_layer(&self_layer, 1);
        increment_layer(&output_layer, 1);
    }
}

コード例 #2

ファイル: crnn_layer.c プロジェクト: KaiqiZhang/darknet

void forward_crnn_layer_gpu(layer l, network net)
{
    network s = net;
    int i;
    layer input_layer = *(l.input_layer);
    layer self_layer = *(l.self_layer);
    layer output_layer = *(l.output_layer);

    fill_gpu(l.outputs * l.batch * l.steps, 0, output_layer.delta_gpu, 1);
    fill_gpu(l.hidden * l.batch * l.steps, 0, self_layer.delta_gpu, 1);
    fill_gpu(l.hidden * l.batch * l.steps, 0, input_layer.delta_gpu, 1);
    if(net.train) fill_gpu(l.hidden * l.batch, 0, l.state_gpu, 1);

    for (i = 0; i < l.steps; ++i) {
        s.input_gpu = net.input_gpu;
        forward_convolutional_layer_gpu(input_layer, s);

        s.input_gpu = l.state_gpu;
        forward_convolutional_layer_gpu(self_layer, s);

        float *old_state = l.state_gpu;
        if(net.train) l.state_gpu += l.hidden*l.batch;
        if(l.shortcut){
            copy_gpu(l.hidden * l.batch, old_state, 1, l.state_gpu, 1);
        }else{
            fill_gpu(l.hidden * l.batch, 0, l.state_gpu, 1);
        }
        axpy_gpu(l.hidden * l.batch, 1, input_layer.output_gpu, 1, l.state_gpu, 1);
        axpy_gpu(l.hidden * l.batch, 1, self_layer.output_gpu, 1, l.state_gpu, 1);

        s.input_gpu = l.state_gpu;
        forward_convolutional_layer_gpu(output_layer, s);

        net.input_gpu += l.inputs*l.batch;
        increment_layer(&input_layer, 1);
        increment_layer(&self_layer, 1);
        increment_layer(&output_layer, 1);
    }
}