Method InsertReorderToUserMem

tensorflow/core/util/mkl_util.h:1654–1661

Function to actually insert reorder primitive in the net This function completes remaining part of output reordering. It inserts a reordering primitive from the temporary buffer that holds the output to the user-specified output buffer. @input: net - net to which to add reorder primitive @input: net_args - net to which user and reorder memories are added if needed. Each entry is a key-value pair

  ///                    needed. Each entry is a key-value pair of the form
  ///                    <argument-type, dnnl::memory>.
  inline void InsertReorderToUserMem(std::vector<primitive>& net,
                                     std::vector<MemoryArgsMap>& net_args) {
    DCHECK(user_memory_);
    DCHECK(reorder_memory_);
    net.push_back(CreateReorder(reorder_memory_, user_memory_));
    net_args.push_back(MemoryArgsMap{{DNNL_ARG_FROM, *reorder_memory_},
                                     {DNNL_ARG_TO, *user_memory_}});
  }

  /// TODO: this is a faster path with reorder primitive cache compared with
  ///       InsertReorderToUserMem(net, net_args), will remove