result = session->Run(run_options_proto, input_pairs, output_tensor_names,
                          target_oper_names, &outputs, &run_metadata_proto);

    // Serialize back to upstream client, who now owns the new buffer
    if (run_metadata != nullptr) {
      status->status = MessageToBuffer(run_metadata_proto, run_metadata);
      if (!status->status.ok()) return;
    }
  } else {

  TF_DeleteAbstractOp(matmul_op);

  TF_AbstractFunction* func =
      TF_FinalizeFunction(graph_ctx, mm_outputs, status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

  /* Now that the graph is built, test graph implementation on matmul example:
    [[1,1] ,   *   [[1,1] ,   =  [[2,2],
     [1,1]]         [1,1]]        [2,2]]
  */

  // Build eager context.

  void Run(const std::vector<std::pair<TF_Operation*, TF_Tensor*>>& inputs,
           TF_Operation* output, int32_t expected_result) {
    Run(inputs, {{output, 0}}, {expected_result});
  }

  // Run the host graph, which now contains a function and check that
  // outputs are as expected.
  // 'T' stands for 'tensor' since the outputs are tensors, not scalars.
  void RunT(const std::vector<std::pair<TF_Operation*, TF_Tensor*>>& inputs,

  EXPECT_EQ(attr_value.type(), tensorflow::DT_INT32);
  ASSERT_TRUE(GetAttrValue(add, "N", &attr_value, s)) << TF_Message(s);
  EXPECT_EQ(attr_value.i(), 2);

  // Placeholder oper now has a consumer.
  ASSERT_EQ(1, TF_OperationOutputNumConsumers(TF_Output{feed, 0}));
  TF_Input feed_port;
  EXPECT_EQ(1, TF_OperationOutputConsumers(TF_Output{feed, 0}, &feed_port, 1));
  EXPECT_EQ(add, feed_port.oper);