// TPUReplicatedInput and TPUReplicatedOutput respectively. For example, with
// two component devices, running `x = TPUReplicatedInput(inputs=[a, b])` on the
// parallel device creates a parallel tensor `x` with `a` on the first of
// `underlying_devices` and `b` on the second. Running `a_unpacked, b_unpacked =
// TPUReplicatedOutput(input=x, num_replicas=2)` un-packs the parallel tensor
// into its components.
//

  // Adds `value` to the existing value of the variable.
  void AssignAdd(TFE_Context* context, TFE_TensorHandle* value,
                 TF_Status* status);

 private:
  // Helper for running any single-argument assignment ops (Assign, AssignAdd,
  // AssignSub, ...).
  void GeneralAssignment(const char* op_name, TFE_Context* context,
                         TFE_TensorHandle* value, TF_Status* status);

TF_CAPI_EXPORT extern uint64_t TF_NowSeconds(void);

// Populates a TF_ThreadOptions struct with system-default values.
TF_CAPI_EXPORT extern void TF_DefaultThreadOptions(TF_ThreadOptions* options);

// Returns a new thread that is running work_func and is identified
// (for debugging/performance-analysis) by thread_name.
//
// The given param (which may be null) is passed to work_func when the thread

    if (!released_) f_();
    released_ = src.released_;
    f_ = src.release();
    return *this;
  }

  ~Cleanup() {
    if (!released_) f_();
  }

  // Releases the cleanup function instead of running it.
  // Hint: use c.release()() to run early.
  F release() {
    released_ = true;
    return std::move(f_);
  }

  bool is_released() const { return released_; }

 private:

// If heavy_load_on_streaming_rpc is true, send some rpc requests before the one
// which creates a remote input, to simulate a scenario that the remote input
// is not ready when we start running an op or a function.
void TestRemoteExecuteSilentCopies(bool async, bool remote, bool func,
                                   bool heavy_load_on_streaming_rpc,
                                   bool remote_func_outputs = false,

  TF_Graph* const graph;

  tensorflow::mutex mu TF_ACQUIRED_AFTER(TF_Graph::mu);
  int last_num_graph_nodes;

  // If true, TF_SessionRun and similar methods will call
  // ExtendSessionGraphHelper before running the graph (this is the default
  // public behavior). Can be set to false if the caller needs to call
  // ExtendSessionGraphHelper manually.
  std::atomic<bool> extend_before_run;
};