return_operands.push_back(cloned_output);
  }
  function_builder.create<mlir::func::ReturnOp>(new_func.getLoc(),
                                                return_operands);
  ops_added.func_op = new_func;
  module.push_back(new_func);
  return new_func;
}

void ExtractSubgraphToFunc(const Subgraph& subgraph, OpBuilder& builder,
                           ModuleOp& module, OpsAdded& ops_added) {

    : tfr_module_(tfr_module), pm_(tfr_module_.getContext()) {
  mlir::OpPassManager& func_pm = pm_.nest<mlir::func::FuncOp>();

  // Prepare the imported graph.
  func_pm.addPass(mlir::CreateExecutorDialectToFunctionalConversionPass());

  // Run TFR lowering, inlining and raising to tf.
  func_pm.addPass(mlir::TFR::CreateDecomposeTFOpsPass(tfr_module_));
  func_pm.addPass(mlir::TFR::CreateRaiseToTFOpsPass(

  // Insert CalibrationStatisticsSaverOp to the end of each region.
  for (auto func_op : module_op.getOps<func::FuncOp>()) {
    int32_t output_file_idx = 0;
    StringRef func_name = func_op.getSymName();

    func_op.walk([&output_file_idx, &ctx, &func_name, &aggregator_ops_to_ignore,
                  this](Operation* op) {
      for (Region& region : op->getRegions()) {

 public:
  explicit ConvertMaxUnpoolingFunc(func::FuncOp func, mlir::TF::FuncAttr attr)
      : func_(func), attr_(attr) {}

  LogicalResult RewriteFunc();

  LogicalResult VerifySignature();

 private:
  LogicalResult CreateCustomOptions(std::string& custom_option_buffer);

  func::FuncOp func_;
  mlir::TF::FuncAttr attr_;
};

// Fuse DenseImageWarp ops annotated by tf.function to a TFLite custom op.

  return op->getName().getStringRef().str() == kClusterFuncOpName;
}

void VerifyInputDialectToExecutorPass::runOnOperation() {
  Operation* func_op = getOperation();

  auto walk_result = func_op->walk([&](Operation* op) {
    if (!tensorflow::tf2xla::internal::IsInBridgeAcceptableDialects(op)) {
      std::string error = "op is in dialect " +

  get_alg_op->erase();
  return success();
}

void TFXLADeviceSpecificTransforms::runOnOperation() {
  if (!device_type_.hasValue()) return;
  auto func_op = getOperation();

  auto walk_result = func_op->walk([&](TF::StatelessRandomGetAlgOp op) {
    if (failed(ConvertGetAlgOp(op))) {
      op->emitOpError(
          "Could not convert and remove Device specific information");

    : public impl::VerifyClusteringPassBase<VerifyClusteringPass> {
 public:
  void runOnOperation() override;
};

void VerifyClusteringPass::runOnOperation() {
  Operation* func_op = getOperation();

  auto walk_result = func_op->walk([&](Operation* op) {
    if (!tensorflow::tf2xla::internal::IsInBridgeAcceptableDialects(op)) {
      std::string error = "op is in dialect " +

 public:
  void runOnOperation() override;
};

void InputMetricsLoweringPass::runOnOperation() {
  bool has_dynamic_op = false;
  Operation* func_op = getOperation();

  func_op->walk([&](Operation* op) {
    auto abstractOp = op->getRegisteredInfo();
    if (!abstractOp) return WalkResult::advance();

    if (mlir::mhlo::IsDynamicPadderOp(abstractOp->getTypeID())) {

  module_op->walk([&](func::FuncOp func_op) {
    FunctionName func_name = func_op.getSymName().str();
    existing_func_names.insert(func_name);
    // We may retrieve the original function's name from the attribute.
    // Functions without this attribute are ignored.
    auto original_func_name =
        func_op->getAttrOfType<StringAttr>("tf._original_func_name");
    if (original_func_name) {

// Returns a list of "tf.entry_function" attribute's "inputs" comma-split
// values.
//
// Example: if `func_op` has attribute `tf.entry_function = {inputs =
// "arg0:0,arg1:0"}`, then this function returns `{"arg0:0", "arg1:0"}`.
SmallVector<StringRef> GetEntryFunctionInputs(func::FuncOp func_op) {
  auto entry_function_attr =
      func_op->getAttrOfType<DictionaryAttr>("tf.entry_function");

  SmallVector<StringRef> inputs;