for (int64_t i = 0; i < elements_in.size(); ++i) {
      elements_out(i) = elements_in(i);
    }
  } else {
    AllocatorAttributes attr;
    attr.set_gpu_compatible(true);
    attr.set_nic_compatible(true);
    TF_RETURN_IF_ERROR(context->allocate_temp(
        var->tensor()->dtype(), var->tensor()->shape(), &tmp, attr));
    tensorflow::Status s;
    TF_Tensor* tf_tmp = TF_TensorFromTensor(tmp, &s);