count: 4
    }
  }
  unclustered_op_histogram {
    op: "NoOp"
    count: 2
  }
  unclustered_op_histogram {
    op: "_Arg"
    count: 1
  }
  unclustered_op_histogram {
    op: "_Retval"
    count: 1
  }
}
)",
      &expected_auto_clustering_activity);
  EXPECT_EQ(listener()->auto_clustering_activity().DebugString(),
            expected_auto_clustering_activity.DebugString());

  auto a = ops::_Arg(scope.WithOpName("A"), DT_INT32, 0);
  auto b = ops::_Arg(scope.WithOpName("B"), DT_INT32, 1);
  auto c = ops::Add(scope.WithOpName("C"), a, b);
  auto d = ops::_Retval(scope.WithOpName("D"), c, 0);
  TF_RETURN_IF_ERROR(scope.ToGraph(graph.get()));
  return graph;
}

StatusOr<FunctionDef> SampleFuntionAddXY(const std::string& name) {
  TF_ASSIGN_OR_RETURN(auto graph, SampleGraphAddXY());

      StrEq("input_tensor:0"));

  // Match the `_Retval` node that corresponds to the return value of @main.
  const auto retval_node_itr =
      llvm::find_if(exported_model->graph_def().node(),
                    [](const NodeDef& node) { return node.op() == "_Retval"; });
  ASSERT_NE(retval_node_itr, exported_model->graph_def().node().end());

// following steps:
//
// 1. Add a XLA computation key placeholder node (it will be used as input for
//    XlaRecvAtHost and XlaSendFromHost);
// 2. Replace all _Arg nodes with one single XlaRecvAtHost node;
// 3. Replace all _Retval nodes with one single XlaSendFromHost node;
// 4. Mark all nodes except key placeholder with attr `xla_cluster_attr_name`
//    and `outside_compilation_attr_name`;

    return false;
  }

  // _Arg nodes in a top-level function represent feeds and _Retval nodes in a
  // top-level function represent fetches.
  if (stack_depth == 1 &&
      (node.type_string() == "_Arg" || node.type_string() == "_Retval")) {
    absl::string_view uncompilable_reason = "top level _Arg or _Retval";
    MaybeMarkUncompilableNode(uncompilable_reason, *stack_trace,

    if (mul) {
      auto c = ops::Mul(scope.WithOpName("C"), a, b);
      auto d = ops::_Retval(scope.WithOpName("D"), c, 0);
      TF_RETURN_IF_ERROR(scope.ToGraph(graph.get()));
    } else {
      auto c = ops::Add(scope.WithOpName("C"), a, b);
      auto d = ops::_Retval(scope.WithOpName("D"), c, 0);
      TF_RETURN_IF_ERROR(scope.ToGraph(graph.get()));
    }

  auto a = ops::_Arg(scope.WithOpName("A"), DT_INT32, 0);
  auto b = ops::_Arg(scope.WithOpName("B"), DT_INT32, 1);
  auto c = ops::Add(scope.WithOpName("C"), a, b);
  auto d = ops::_Retval(scope.WithOpName("D"), c, 0);
  TF_RETURN_IF_ERROR(scope.ToGraph(graph.get()));
  return graph;
}

absl::StatusOr<FunctionDef> SampleFuntionAddXY(const std::string& name) {

static const char kImportModelDefaultGraphFuncName[] = "main";

// Please refer to the TFG dialect description for the list of used attributes.
// Belows are the attributes in TFE.
// TFE Arguments and Results (Got from "_Arg",
// "_Retval", .etc)
//  NodeDef.device <-> "tf.device"
//  NodeDef.attr <-> "tf."
//
// TFE general operations
//  NodeDef.device <-> "device"
//
// The following two functions are only used for mapping/excluding attributes

                                         .WithAttr("dtype", DT_FLOAT)
                                         .WithAttr("value", Tensor()));
    Node* b = ops::UnaryOp("Relu", a, builder.opts().WithName("B"));
    ops::UnaryOp("_Retval", b,
                 builder.opts()
                     .WithName("R")
                     .WithAttr("T", DT_FLOAT)
                     .WithAttr("index", 0));

    mlir::OperationName name = inner_op->getName();
    if (!name.isRegistered() &&
        // Skip unmodelled ops that are handled differently.
        (node_type_name != "_Arg" && node_type_name != "_Retval") &&
        !unmodelled_op_names_.count(name.getIdentifier())) {
      if (node.op_def().is_stateful()) {
        VLOG(1) << "[potentially conservative] Op type `" << node.type_string()