From d0b9326a352ed2fbc3ae66fdef40b4533a2f211d Mon Sep 17 00:00:00 2001
From: Joseph Redmon <pjreddie@gmail.com>
Date: Tue, 11 Aug 2015 06:22:27 +0000
Subject: [PATCH] Hacks to get nightmare to not break gridsizing
---
src/network_kernels.cu | 33 ++++++++++++++++++---------------
1 files changed, 18 insertions(+), 15 deletions(-)
diff --git a/src/network_kernels.cu b/src/network_kernels.cu
index 5e353ae..3340afa 100644
--- a/src/network_kernels.cu
+++ b/src/network_kernels.cu
@@ -15,10 +15,13 @@
#include "convolutional_layer.h"
#include "deconvolutional_layer.h"
#include "maxpool_layer.h"
+#include "avgpool_layer.h"
+#include "normalization_layer.h"
#include "cost_layer.h"
#include "softmax_layer.h"
#include "dropout_layer.h"
#include "route_layer.h"
+#include "blas.h"
}
float * get_network_output_gpu_layer(network net, int i);
@@ -30,6 +33,9 @@
int i;
for(i = 0; i < net.n; ++i){
layer l = net.layers[i];
+ if(l.delta_gpu){
+ scal_ongpu(l.outputs * l.batch, 0, l.delta_gpu, 1);
+ }
if(l.type == CONVOLUTIONAL){
forward_convolutional_layer_gpu(l, state);
} else if(l.type == DECONVOLUTIONAL){
@@ -44,8 +50,12 @@
forward_cost_layer_gpu(l, state);
} else if(l.type == SOFTMAX){
forward_softmax_layer_gpu(l, state);
+ } else if(l.type == NORMALIZATION){
+ forward_normalization_layer_gpu(l, state);
} else if(l.type == MAXPOOL){
forward_maxpool_layer_gpu(l, state);
+ } else if(l.type == AVGPOOL){
+ forward_avgpool_layer_gpu(l, state);
} else if(l.type == DROPOUT){
forward_dropout_layer_gpu(l, state);
} else if(l.type == ROUTE){
@@ -59,11 +69,12 @@
{
int i;
float * original_input = state.input;
+ float * original_delta = state.delta;
for(i = net.n-1; i >= 0; --i){
layer l = net.layers[i];
if(i == 0){
state.input = original_input;
- state.delta = 0;
+ state.delta = original_delta;
}else{
layer prev = net.layers[i-1];
state.input = prev.output_gpu;
@@ -75,10 +86,14 @@
backward_deconvolutional_layer_gpu(l, state);
} else if(l.type == MAXPOOL){
if(i != 0) backward_maxpool_layer_gpu(l, state);
+ } else if(l.type == AVGPOOL){
+ if(i != 0) backward_avgpool_layer_gpu(l, state);
} else if(l.type == DROPOUT){
backward_dropout_layer_gpu(l, state);
} else if(l.type == DETECTION){
backward_detection_layer_gpu(l, state);
+ } else if(l.type == NORMALIZATION){
+ backward_normalization_layer_gpu(l, state);
} else if(l.type == SOFTMAX){
if(i != 0) backward_softmax_layer_gpu(l, state);
} else if(l.type == CONNECTED){
@@ -120,6 +135,7 @@
cuda_push_array(*net.truth_gpu, y, y_size);
}
state.input = *net.input_gpu;
+ state.delta = 0;
state.truth = *net.truth_gpu;
state.train = 1;
forward_network_gpu(net, state);
@@ -134,20 +150,7 @@
{
layer l = net.layers[i];
cuda_pull_array(l.output_gpu, l.output, l.outputs*l.batch);
- if(l.type == CONVOLUTIONAL){
- return l.output;
- } else if(l.type == DECONVOLUTIONAL){
- return l.output;
- } else if(l.type == CONNECTED){
- return l.output;
- } else if(l.type == DETECTION){
- return l.output;
- } else if(l.type == MAXPOOL){
- return l.output;
- } else if(l.type == SOFTMAX){
- return l.output;
- }
- return 0;
+ return l.output;
}
float *get_network_output_gpu(network net)
--
Gitblit v1.10.0