cjolivier01 commented on a change in pull request #9862: Fix a race condition 
in converting data layouts in MKLDNN.
URL: https://github.com/apache/incubator-mxnet/pull/9862#discussion_r170116831
 
 

 ##########
 File path: src/operator/nn/mkldnn/mkldnn_base.cc
 ##########
 @@ -270,9 +270,24 @@ void FallBackCompute(FCompute fn, const nnvm::NodeAttrs 
&attrs,
                      const std::vector<OpReqType> &req,
                      const std::vector<NDArray> &outputs) {
   std::vector<TBlob> in_blobs(inputs.size());
+  std::vector<NDArray> in_bufs;
   for (size_t i = 0; i < in_blobs.size(); i++) {
+    // If the input data isn't stored in the default format, we shouldn't
+    // call data() directly, which will change the layout of the NDArray.
+    // Instead, we should save the converted data in another NDArray.
+    // TODO(zhengda) we should use temp space to save the converted data.
+    if (inputs[i].IsDefaultData()) {
       in_blobs[i] = inputs[i].data();
+    } else {
+      in_bufs.emplace_back(inputs[i].shape(), inputs[i].ctx(),
 
 Review comment:
   nit: it would be nice to call in_bufs.reserve(in_blobs.size()) on this first 
pass (ie if in_bufs.empty())
   If there is often more than one in_bufs to be inserted

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


With regards,
Apache Git Services

Reply via email to