absl::StatusOr<core::RefCountPtr<FunctionRecord>> GetFunctionRecord()
      override {
    return func_record_.GetNewRef();
  }

  // For LLVM style RTTI.
  static bool classof(const AbstractFunction* ptr) {
    return ptr->getKind() == kGraph;
  }

 private:
  core::RefCountPtr<FunctionRecord> func_record_;
};
}  // namespace graph
}  // namespace tracing
}  // namespace tensorflow

// Set the config in TF_SessionOptions.options.
// config should be a serialized tensorflow.ConfigProto proto.
// If config was not parsed successfully as a ConfigProto, record the
// error information in *status.
TF_CAPI_EXPORT extern void TF_SetConfig(TF_SessionOptions* options,
                                        const void* proto, size_t proto_len,

TF_CAPI_EXPORT extern TFE_ContextOptions* TFE_NewContextOptions(void);

// Set the config in TF_ContextOptions.options.
// config should be a serialized tensorflow.ConfigProto proto.
// If config was not parsed successfully as a ConfigProto, record the
// error information in *status.
TF_CAPI_EXPORT extern void TFE_ContextOptionsSetConfig(
    TFE_ContextOptions* options, const void* proto, size_t proto_len,
    TF_Status* status);

    TF_Status* status);

// Create and modify a variable placed on a parallel device which composes
// `first_device` and `second_device`.
void BasicTestsForTwoDevices(TFE_Context* context, const char* first_device,
                             const char* second_device);

// Implementations of templated functions ******************************

template <std::size_t num_replicas>

// 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.
//
// The filled `device` struct and the allocated `device_info` struct may be

  std::vector<tensorflow::string> missing_unused_key_names_data;
};

struct TF_DeviceList {
  std::vector<tensorflow::DeviceAttributes> response;
};

struct TF_Function {
  tensorflow::FunctionRecord* record;
};

struct TF_ApiDefMap {
  explicit TF_ApiDefMap(const tensorflow::OpList& op_list)
      :
#if !defined(IS_MOBILE_PLATFORM) && !defined(IS_SLIM_BUILD)
        api_def_map(op_list),

    if (grad_it == gradients.end()) {
      result[i] = nullptr;
    } else {
      if (grad_it->second.size() > 1) {
        Gradient* grad = vspace.AggregateGradients(grad_it->second);
        grad_it->second.clear();
        grad_it->second.push_back(grad);
      }
      result[i] = grad_it->second[0];
      used_gradient_ids.insert(tensor_id);
    }
  }
  VLOG(1) << "Final gradients size: "

    if (it == cache_.end()) {
      return false;
    }
    lru_list_.erase(it->second.lru_iterator);
    if (timer_seconds_() - it->second.timestamp > max_age_) {
      cache_.erase(it);
      return false;
    }
    *value = it->second.value;
    lru_list_.push_front(it->first);
    it->second.lru_iterator = lru_list_.begin();
    return true;
  }

  // [Optional]
  // Create function for optimizer.
  void* (*create_func)();

  // Optimizer function for optimizer. The first param is an optimizer created
  // by create_func. The second param is input graph. The third param is
  // GrapplerItem. The fourth param is output graph.
  void (*optimize_func)(void*, const TF_Buffer*, const TF_GrapplerItem*,
                        TF_Buffer*, TF_Status*);

  //
  // Arguments provide enough information to reconstruct the original `TFE_Op`,
  // or construct a transformed version, by inspecting the passed `op`.
  //
  // TFE_OpGetDevice(op) records the original placement of the operation. It may
  // be an empty string if no device was explicitly requested, but will
  // otherwise be the name of this custom device. Ops are placed onto a custom