if (!HasTfSavedModelSemantics(module)) {
      return;
    }

    auto init_op = GetSessionInitializerOp(module);
    if (init_op == nullptr || init_op.getInitializers().empty()) {
      return;
    }

    SymbolTable symbol_table(module);
    for (auto initializer : init_op.getInitializers()) {
      auto func = symbol_table.lookup<func::FuncOp>(
          mlir::cast<FlatSymbolRefAttr>(initializer).getValue());

  bool is_ok = arg.ok();
  auto graph_analysis_failure = WasGraphAnalysisFailure(arg);
  return testing::ExplainMatchResult(
      testing::IsTrue(), is_ok || graph_analysis_failure, result_listener);
}

MATCHER_P2(IncrementedOrFiltered, metric, value,
           "Metric was incremented by value or Status equal to the Graph "
           "Analysis failure") {
  auto graph_analysis_failure = WasGraphAnalysisFailure(arg);

BlockArgument InsertFilePrefixArgument(func::FuncOp func_op,
                                       OpBuilder& builder) {
  const auto filename_op_type = RankedTensorType::get(
      /*shape=*/{}, /*elementType=*/builder.getType<TF::StringType>());
  const auto file_prefix_attr = builder.getStringAttr(kTfFilePrefix);
  const auto arg_attrs = builder.getDictionaryAttr({builder.getNamedAttr(

  if (element_type.isF16()) {
    auto floatType = mlir::FloatType::getF16(element_type.getContext());
    auto floatAttr = mlir::FloatAttr::get(floatType, static_cast<float>(value));
    std::vector<Attribute> floatValues({floatAttr});
    return DenseElementsAttr::get(shaped_type, floatValues);
  } else if (element_type.isBF16()) {
    auto floatType = mlir::FloatType::getBF16(element_type.getContext());

    return mlir::failure();
  }
  // Cluster outermost partitioned calls with _xla_compile_device_type
  // attribute.
  for (auto &pcall_op : outermost_pcall_ops) {
    auto call = llvm::cast<CallOpInterface>(pcall_op.getOperation());
    CallInterfaceCallable callable = call.getCallableForCallee();
    auto sym = callable.get<mlir::SymbolRefAttr>();
    EncapsulatePartitionedCall(pcall_op, sym.getRootReference());
  }

    flatbuffers::FlatBufferBuilder* builder,
    std::map<std::string, uint8_t>* hardware_names) {
  uint8_t index = 0;
  for (auto& func : *funcs) {
    func.walk([&hardware_names, &index](mlir::Operation* op) {
      auto device_name = GetDeviceName(op);
      if (!device_name.has_value()) return;

      auto iter = hardware_names->find(*device_name);
      if (iter == hardware_names->end()) {

                                     mlir::TF::RuntimeDevices* devices) {
  auto devices_attr = op->getAttr(kDevicesAttr);
  if (!devices_attr) return mlir::success();

  if (auto array_attr = mlir::dyn_cast<mlir::ArrayAttr>(devices_attr)) {
    return GetDevicesFromOp(op, array_attr, devices);

  } else if (auto dict_attr =
                 mlir::dyn_cast<mlir::DictionaryAttr>(devices_attr)) {

          init_table_from_text_file_ops_to_erase;
      auto asset = LookupBoundInputOfType<AssetOp>(func, i, symbol_table);

      if (!asset) continue;

      auto arg = func.getArgument(i);
      bool arg_is_deletable = true;
      for (auto user : arg.getUsers()) {
        if (auto read_op =
                llvm::dyn_cast<TF::InitializeTableFromTextFileV2Op>(user)) {

    DenseElementsAttr constant;
    if (auto rhs = bmm_op.getY(); !matchPattern(rhs, m_Constant(&constant))) {
      // The constant may be preceded by QDQs in models with QDQ format, so we
      // should set it to the real constant.
      auto dq = dyn_cast_or_null<DequantizeOp>(rhs.getDefiningOp());
      if (!dq) return failure();
      auto q = dyn_cast_or_null<QuantizeOp>(dq.getInput().getDefiningOp());

  DeviceSet device_set;
  for (auto& device : devices) device_set.AddDevice(device.get());
  AddDevicesToOp(*module_ref, &device_set);

  auto devices_attr =
      (*module_ref)->getAttrOfType<mlir::DictionaryAttr>("tf.devices");
  ASSERT_NE(devices_attr, nullptr);
  ASSERT_EQ(devices_attr.size(), 3);

  // CPU device added with an empty metadata.