From 08e21e799bf381a25f139f133b37687dfe3b2c2d Mon Sep 17 00:00:00 2001
From: Jonathan L Long <jonlong@cs.berkeley.edu>
Date: Sat, 13 Sep 2014 17:41:59 -0700
Subject: [PATCH] accumulate gradients in cudnn conv layer

---
 src/caffe/layers/cudnn_conv_layer.cu | 2 --
 1 file changed, 2 deletions(-)

diff --git a/src/caffe/layers/cudnn_conv_layer.cu b/src/caffe/layers/cudnn_conv_layer.cu
index 071014e1b48..b5bfdb098e0 100644
--- a/src/caffe/layers/cudnn_conv_layer.cu
+++ b/src/caffe/layers/cudnn_conv_layer.cu
@@ -54,12 +54,10 @@ void CuDNNConvolutionLayer<Dtype>::Backward_gpu(const vector<Blob<Dtype>*>& top,
   if (this->param_propagate_down_[0]) {
     weight = this->blobs_[0]->gpu_data();
     weight_diff = this->blobs_[0]->mutable_gpu_diff();
-    caffe_gpu_set(this->blobs_[0]->count(), Dtype(0), weight_diff);
   }
   Dtype* bias_diff = NULL;
   if (this->bias_term_ && this->param_propagate_down_[1]) {
     bias_diff = this->blobs_[1]->mutable_gpu_diff();
-    caffe_gpu_set(this->blobs_[1]->count(), Dtype(0), bias_diff);
   }
   for (int i = 0; i < top.size(); ++i) {
     const Dtype* top_diff = top[i]->gpu_diff();