%0:6 = tf_executor.graph {
      %outputs, %control = tf_executor.island wraps "tf.ReadVariableOp"(%arg2) {device = ""} : (tensor<*x!tf_type.resource>) -> tensor<*xf32>
      %outputs_0, %control_1 = tf_executor.island wraps "tf.ReadVariableOp"(%arg1) {device = ""} : (tensor<*x!tf_type.resource>) -> tensor<*xf32>

//     tf_device.return %5 : tensor<i1>
//   }) {device = "/DEVICE:3"} : () -> tensor<i1>
//   tf_executor.yield %a1, %b1 : tensor<i1>, tensor<i1>
// }
LogicalResult CreateIslandsFromReplicate(const Dialect* tf_dialect,
                                         tf_executor::GraphOp graph_op,
                                         tf_executor::IslandOp island_op,
                                         tf_device::ReplicateOp replicate_op,

#include "tensorflow/compiler/mlir/tensorflow/ir/tf_executor.h"
#include "tensorflow/compiler/mlir/tensorflow/ir/tf_ops.h"
#include "tensorflow/compiler/mlir/tensorflow/ir/tf_saved_model.h"
#include "tensorflow/compiler/mlir/tensorflow/translate/import_model.h"

namespace mlir {
namespace quant {
namespace {

using ::mlir::tf_executor::FetchOp;
using ::mlir::tf_executor::GraphOp;
using ::mlir::tf_executor::IslandOp;

    %0 = tf_executor.graph {
      %outputs, %control = tf_executor.island wraps "tf.Const"() {device = "", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>
      %outputs_0, %control_1 = tf_executor.island wraps "tf.Const"() {device = "", dtype = f32, value = dense<[]> : tensor<0xf32>} : () -> tensor<0xf32>

// tf_executor dialect.
absl::StatusOr<mlir::OwningOpRef<mlir::ModuleOp>> ConvertGraphdefToMlir(
    const GraphDef& graphdef, const GraphDebugInfo& debug_info,
    const GraphImportConfig& specs, mlir::MLIRContext* context);

// Given a Graph, returns a MLIR module containing the graph, expressed with
// tf_executor dialect.

// to the island. Otherwise a NoOp will be added to the island.
tf_executor::IslandOp CreateIsland(TypeRange result_types,
                                   ValueRange control_inputs,
                                   const tf_executor::ControlType& control_type,
                                   const Location& loc, Operation* sub_op,
                                   tf_executor::IslandOp original_island) {
  OpBuilder builder(original_island);

    %0 = tf_executor.graph {
      %outputs, %control = tf_executor.island wraps "tf.VariableV2"() {container = "", device = "/job:localhost/replica:0/task:0/device:CPU:0", shape = #tf_type.shape<>, shared_name = "v_load_44"} : () -> tensor<!tf_type.f32ref>

      }
    } else if (auto graph = dyn_cast<tf_executor::GraphOp>(op)) {
      for (auto &arg : graph.GetFetch()->getOpOperands()) {
        if (arg.getOperandNumber() < graph.getNumResults()) {
          auto result = graph.getResult(arg.getOperandNumber());
          this->join(getLatticeElement(result), *getLatticeElement(arg.get()));
        }
      }
    } else if (auto island = dyn_cast<tf_executor::IslandOp>(op)) {

  }
  return absl::OkStatus();
}

// Collects control ret Nodes based on tf_executor.graph's associated
// tf_executor.fetch control inputs.
Status Exporter::GetControlRetNodes(
    mlir::tf_executor::FetchOp fetch,
    absl::flat_hash_set<Node*>* control_ret_nodes) {
  for (Value fetch_operand : fetch.getOperands()) {
    if (mlir::isa<mlir::tf_executor::ControlType>(fetch_operand.getType())) {
      Operation* defining_op =

  }
  return absl::OkStatus();
}

// Collects control ret Nodes based on tf_executor.graph's associated
// tf_executor.fetch control inputs.
Status Exporter::GetControlRetNodes(
    mlir::tf_executor::FetchOp fetch,
    absl::flat_hash_set<Node*>* control_ret_nodes) {
  for (Value fetch_operand : fetch.getOperands()) {
    if (mlir::isa<mlir::tf_executor::ControlType>(fetch_operand.getType())) {
      Operation* defining_op =