Atomic operation about torch::PackedTensorAccessor

Is add operation of torch::PackedTensorAccessor atomic?