Skip to content

Commit

Permalink
note cuDNN v2 convolutional TODOs
Browse files Browse the repository at this point in the history
  • Loading branch information
@shelhamer
shelhamer committed Mar 24, 2015
1 parent 6d1444f commit 3cc9b73
Showing 1 changed file with 4 additions and 1 deletion.
5 changes: 4 additions & 1 deletion src/caffe/layers/cudnn_conv_layer.cu
View file
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,10 @@ void CuDNNConvolutionLayer<Dtype>::Forward_gpu(
for (int g = 0; g < this->group_; g++) {
cudnnConvolutionFwdAlgo_t algo;

// get the desired convolution algorithm
// pick the convolution algorithm
// TODO(shelhamer) this should be done during reshape
// TODO(shelhamer) the choice of automatic or manual algorithm picking
// should be exposed in proto
CUDNN_CHECK(cudnnGetConvolutionForwardAlgorithm(handle_[g],
bottom_descs_[i],
filter_desc_,
Expand Down

3 comments on commit 3cc9b73

@xiangyangli-cn
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I just downdload the latest caffe-master and cudnn v2,but this occour:
cudnn_conv_layer.cu:65] Check failed: status == CUDNN_STATUS_SUCCESS (8 vs. 0) CUDNN_STATUS_EXECUTION_FAILED

How can i fix this problem

@jmozah
Copy link

@jmozah jmozah commented on 3cc9b73 Mar 25, 2015

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I too get the same error...

@shelhamer
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry about that -- there should be a fix shortly for the automatic selection. This is happening because cuDNN is selecting the fastest but memory consuming algorithm so that the GPU runs out of memory.

Please sign in to comment.