"https://github.com/tensorflow/tensorflow/issues if this feature is "
      "important to you");
#else
  std::vector<tensorflow::Output> y_arg = OutputsFromTFOutputs(y, ny);
  std::vector<tensorflow::Output> x_arg = OutputsFromTFOutputs(x, nx);
  std::vector<tensorflow::Output> dy_arg;

  {
    // We need to hold on to the lock while we have a scope that uses TF_Graph.
    mutex_lock graph_lock(g->mu);