From 08d6910521d9aa05a11d0db154ae70874e47d0e1 Mon Sep 17 00:00:00 2001
From: Alexey <AlexeyAB@users.noreply.github.com>
Date: Wed, 09 Aug 2017 21:06:24 +0000
Subject: [PATCH] Update Readme.md
---
src/convolutional_kernels.cu | 8 ++++++++
1 files changed, 8 insertions(+), 0 deletions(-)
diff --git a/src/convolutional_kernels.cu b/src/convolutional_kernels.cu
index 709f77f..03c9ab7 100644
--- a/src/convolutional_kernels.cu
+++ b/src/convolutional_kernels.cu
@@ -2,6 +2,10 @@
#include "curand.h"
#include "cublas_v2.h"
+#ifdef CUDNN
+#pragma comment(lib, "cudnn.lib")
+#endif
+
extern "C" {
#include "convolutional_layer.h"
#include "batchnorm_layer.h"
@@ -123,6 +127,7 @@
activate_array_ongpu(l.output_gpu, l.outputs*l.batch, l.activation);
//if(l.dot > 0) dot_error_gpu(l);
if(l.binary || l.xnor) swap_binary(&l);
+ //cudaDeviceSynchronize(); // for correct profiling of performance
}
void backward_convolutional_layer_gpu(convolutional_layer l, network_state state)
@@ -133,6 +138,9 @@
if(l.batch_normalize){
backward_batchnorm_layer_gpu(l, state);
+ //axpy_ongpu(l.outputs*l.batch, -state.net.decay, l.x_gpu, 1, l.delta_gpu, 1);
+ } else {
+ //axpy_ongpu(l.outputs*l.batch, -state.net.decay, l.output_gpu, 1, l.delta_gpu, 1);
}
float *original_input = state.input;
--
Gitblit v1.10.0