art, cudnn

Joseph Redmon

2016-05-13 13209df7bb53de19aa3f82e870db11eb5b7587f1

 src/network_kernels.cu

@@ -11,17 +11,21 @@
#include "image.h"
#include "data.h"
#include "utils.h"
#include "params.h"
#include "parser.h"

#include "crop_layer.h"
#include "connected_layer.h"
#include "rnn_layer.h"
#include "gru_layer.h"
#include "crnn_layer.h"
#include "detection_layer.h"
#include "convolutional_layer.h"
#include "activation_layer.h"
#include "deconvolutional_layer.h"
#include "maxpool_layer.h"
#include "avgpool_layer.h"
#include "normalization_layer.h"
#include "batchnorm_layer.h"
#include "cost_layer.h"
#include "local_layer.h"
#include "softmax_layer.h"
@@ -37,6 +41,7 @@

void forward_network_gpu(network net, network_state state)
{
    state.workspace = net.workspace;
    int i;
    for(i = 0; i < net.n; ++i){
        state.index = i;
@@ -48,12 +53,20 @@
            forward_convolutional_layer_gpu(l, state);
        } else if(l.type == DECONVOLUTIONAL){
            forward_deconvolutional_layer_gpu(l, state);
        } else if(l.type == ACTIVE){
            forward_activation_layer_gpu(l, state);
        } else if(l.type == LOCAL){
            forward_local_layer_gpu(l, state);
        } else if(l.type == DETECTION){
            forward_detection_layer_gpu(l, state);
        } else if(l.type == CONNECTED){
            forward_connected_layer_gpu(l, state);
        } else if(l.type == RNN){
            forward_rnn_layer_gpu(l, state);
        } else if(l.type == GRU){
            forward_gru_layer_gpu(l, state);
        } else if(l.type == CRNN){
            forward_crnn_layer_gpu(l, state);
        } else if(l.type == CROP){
            forward_crop_layer_gpu(l, state);
        } else if(l.type == COST){
@@ -62,6 +75,8 @@
            forward_softmax_layer_gpu(l, state);
        } else if(l.type == NORMALIZATION){
            forward_normalization_layer_gpu(l, state);
        } else if(l.type == BATCHNORM){
            forward_batchnorm_layer_gpu(l, state);
        } else if(l.type == MAXPOOL){
            forward_maxpool_layer_gpu(l, state);
        } else if(l.type == AVGPOOL){
@@ -79,6 +94,7 @@

void backward_network_gpu(network net, network_state state)
{
    state.workspace = net.workspace;
    int i;
    float * original_input = state.input;
    float * original_delta = state.delta;
@@ -97,6 +113,8 @@
            backward_convolutional_layer_gpu(l, state);
        } else if(l.type == DECONVOLUTIONAL){
            backward_deconvolutional_layer_gpu(l, state);
        } else if(l.type == ACTIVE){
            backward_activation_layer_gpu(l, state);
        } else if(l.type == LOCAL){
            backward_local_layer_gpu(l, state);
        } else if(l.type == MAXPOOL){
@@ -109,10 +127,18 @@
            backward_detection_layer_gpu(l, state);
        } else if(l.type == NORMALIZATION){
            backward_normalization_layer_gpu(l, state);
        } else if(l.type == BATCHNORM){
            backward_batchnorm_layer_gpu(l, state);
        } else if(l.type == SOFTMAX){
            if(i != 0) backward_softmax_layer_gpu(l, state);
        } else if(l.type == CONNECTED){
            backward_connected_layer_gpu(l, state);
        } else if(l.type == RNN){
            backward_rnn_layer_gpu(l, state);
        } else if(l.type == GRU){
            backward_gru_layer_gpu(l, state);
        } else if(l.type == CRNN){
            backward_crnn_layer_gpu(l, state);
        } else if(l.type == COST){
            backward_cost_layer_gpu(l, state);
        } else if(l.type == ROUTE){
@@ -136,6 +162,12 @@
            update_deconvolutional_layer_gpu(l, rate, net.momentum, net.decay);
        } else if(l.type == CONNECTED){
            update_connected_layer_gpu(l, update_batch, rate, net.momentum, net.decay);
        } else if(l.type == GRU){
            update_gru_layer_gpu(l, update_batch, rate, net.momentum, net.decay);
        } else if(l.type == RNN){
            update_rnn_layer_gpu(l, update_batch, rate, net.momentum, net.decay);
        } else if(l.type == CRNN){
            update_crnn_layer_gpu(l, update_batch, rate, net.momentum, net.decay);
        } else if(l.type == LOCAL){
            update_local_layer_gpu(l, update_batch, rate, net.momentum, net.decay);
        }