Method CreateBlockedMemDesc

tensorflow/core/util/mkl_util.h:1324–1330

Helper function to create memory descriptor in Blocked format @input: Tensor dimensions @input: strides corresponding to dimensions. One can use utility function such as CalculateTFStrides to compute strides for given dimensions. @return: memory::desc object corresponding to blocked memory format for given dimensions and strides.

  /// @return: memory::desc object corresponding to blocked memory format
  ///          for given dimensions and strides.
  static inline memory::desc CreateBlockedMemDesc(const memory::dims& dim,
                                                  const memory::dims& strides) {
    dnnl_memory_desc_t blocked_md;
    TF_CHECK_OK(
        CreateBlockedMemDescHelper(dim, strides, MklDnnType<T>(), &blocked_md));
    return memory::desc(blocked_md);
  }

  /// A version of SetUsrMem call that allows user to create memory in blocked
  /// format. So in addition to accepting dimensions, it also accepts strides.