if (*requested_placement == '\0') {
    TF_SetStatus(
        status, TF_INTERNAL,
        "Ops must be placed on the parallel device explicitly, or their inputs "
        "first un-packed. Got an un-placed op with an input placed on the "
        "parallel device.");
    return;
  }
  TFE_Context* context = TFE_OpGetContext(original_op, status);
  if (TF_GetCode(status) != TF_OK) return;

      Multiply(context.get(), second_combined_value.get(),
               second_negative_one.get(), status.get()));
  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());

  // Un-pack the parallel tensor to verify that the operation was
  // successful. The resulting structure should be:
  //   second_device{first_device{1. * 3., 2. * 3.}, 3. * 3.}.
  std::array<TensorHandlePtr, 2> second_components;

                   &FloatDeallocator, nullptr),
      TF_DeleteTensor);
  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,