if (transpose_a) {
      TF_SetAttrBool(desc, "transpose_a", 1);
    }
    if (transpose_b) {
      TF_SetAttrBool(desc, "transpose_b", 1);
    }
    TF_AddInput(desc, {l, 0});
    TF_AddInput(desc, {r, 0});
    TF_Operation* op = TF_FinishOperation(desc, s);
    EXPECT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);
    return op;
  }

  void UseHelper(const std::vector<TF_Output>& inputs, TF_Operation** op) {
    TF_OperationDescription* desc =
        TF_NewOperation(host_graph_, func_name_, func_node_name_);
    for (auto input : inputs) {
      TF_AddInput(desc, input);
    }
    // Set device to CPU because some ops inside the function might not be
    // available on GPU.
    TF_SetDevice(desc, "/cpu:0");
    *op = TF_FinishOperation(desc, s_);