From 537d135feba179636e9bbfe296e078d51f59914c Mon Sep 17 00:00:00 2001
From: AlexeyAB <alexeyab84@gmail.com>
Date: Mon, 19 Mar 2018 23:16:51 +0000
Subject: [PATCH] Improve training performance - batch-norm using cuDNN.
---
src/convolutional_kernels.cu | 10 ++++++----
1 files changed, 6 insertions(+), 4 deletions(-)
diff --git a/src/convolutional_kernels.cu b/src/convolutional_kernels.cu
index 44b9a0f..603d531 100644
--- a/src/convolutional_kernels.cu
+++ b/src/convolutional_kernels.cu
@@ -37,7 +37,7 @@
int i = 0;
float mean = 0;
for(i = 0; i < n; ++i){
- mean += abs(input[i*size + s]);
+ mean += fabs(input[i*size + s]);
}
mean = mean / n;
for(i = 0; i < n; ++i){
@@ -59,7 +59,7 @@
int i = 0;
float mean = 0;
for(i = 0; i < size; ++i){
- mean += abs(weights[f*size + i]);
+ mean += fabs(weights[f*size + i]);
}
mean = mean / size;
for(i = 0; i < size; ++i){
@@ -205,8 +205,10 @@
if (l.batch_normalize) {
forward_batchnorm_layer_gpu(l, state);
- }
- add_bias_gpu(l.output_gpu, l.biases_gpu, l.batch, l.n, l.out_w*l.out_h);
+ }
+ else {
+ add_bias_gpu(l.output_gpu, l.biases_gpu, l.batch, l.n, l.out_w*l.out_h);
+ }
activate_array_ongpu(l.output_gpu, l.outputs*l.batch, l.activation);
//if(l.dot > 0) dot_error_gpu(l);
--
Gitblit v1.10.0