Function CalculateTFStrides

tensorflow/core/util/mkl_util.h:1181–1190

Function to calculate strides given tensor shape in Tensorflow order E.g., if dims_tf_order is {1, 2, 3, 4}, then as per Tensorflow convention, dimension with size 1 is outermost dimension; while dimension with size 4 is innermost dimension. So strides for this tensor would be {4 * 3 * 2, 4 * 3, 4, 1}, i.e., {24, 12, 4, 1}. @input Tensorflow shape in memory::dims type @return memory::dims contain

/// @input Tensorflow shape in memory::dims type
/// @return memory::dims containing strides for the tensor.
inline memory::dims CalculateTFStrides(const memory::dims& dims_tf_order) {
  CHECK_GT(dims_tf_order.size(), 0);
  memory::dims strides(dims_tf_order.size());
  int last_dim_idx = dims_tf_order.size() - 1;
  strides[last_dim_idx] = 1;
  for (int d = last_dim_idx - 1; d >= 0; d--) {
    strides[d] = strides[d + 1] * dims_tf_order[d + 1];
  }
  return strides;
}

/// Helper function to create memory descriptor in Blocked format
///