// an array of length at least `max_inputs` (ideally set to
// TF_OperationNumInputs(oper)).  Beware that a concurrent
// modification of the graph can increase the number of inputs of
// an operation.
TF_CAPI_EXPORT extern void TF_OperationAllInputs(TF_Operation* oper,
                                                 TF_Output* inputs,
                                                 int max_inputs);

  absl::optional<std::vector<std::unique_ptr<ParallelTensor>>> Execute(
      TFE_Context* context, const std::vector<ParallelTensor*>& inputs,
      const char* operation_name, const TFE_OpAttrs* attributes,
      int expected_max_outputs, TF_Status* status) const;

  // A non-blocking version of `Execute`. After each call, `Join` must be called
  // before `StartExecute` is called again. Using `StartExecute` with `Join`