return TensorHandlePtr(TFE_NewTensorHandle(tensor.get(), status));
}

// Helper to un-pack `num_replicas` TFE_TensorHandles from one parallel handle.
template <std::size_t num_replicas>
void ExtractPerDeviceValues(
    TFE_Context* context, TFE_TensorHandle* input,
    std::array<TensorHandlePtr, num_replicas>* components, TF_Status* status) {
  std::unique_ptr<TFE_Op, decltype(&TFE_DeleteOp)> op(

  // successful. The resulting structure should be:
  //   second_device{first_device{1. * 3., 2. * 3.}, 3. * 3.}.
  std::array<TensorHandlePtr, 2> second_components;
  ExtractPerDeviceValues(context.get(), multiply_result.get(),
                         &second_components, status.get());
  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());

                 status.get()));
    ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
    std::array<TensorHandlePtr, 2> out_components;
    ExtractPerDeviceValues(context.get(), multiply_result.get(),
                           &out_components, status.get());
    ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());

                                        TF_Status* status);

// Helper to un-pack `num_replicas` TFE_TensorHandles from one parallel handle.
template <std::size_t num_replicas>
void ExtractPerDeviceValues(
    TFE_Context* context, TFE_TensorHandle* input,
    std::array<TensorHandlePtr, num_replicas>* components, TF_Status* status);

// Helper to pack `num_replicas` TFE_TensorHandles into one parallel handle.