SmallVector<Value, 4> operands;
  operands.reserve(operands_size);
  operands.append(while_op.getOperands().begin(), while_op.getOperands().end());
  operands.append(extra_operands.begin(), extra_operands.end());
  SmallVector<Type, 4> new_types;
  new_types.reserve(operands_size);
  new_types.append(while_op.getResultTypes().begin(),

    return result;
  }
  std::vector<ParallelTensor*> parallel_inputs;
  std::vector<std::unique_ptr<ParallelTensor>> implicitly_broadcast_tensors;
  parallel_inputs.reserve(inputs.size());
  implicitly_broadcast_tensors.reserve(inputs.size());  // not tight
  for (const auto& input : inputs) {
    if (absl::holds_alternative<TFE_TensorHandle*>(input)) {
      if (operation_name == std::string("_EagerConst")) {

  if (type.getElementType().isF32()) {
    tflite::internal::sparsity::FormatConverter<float> format_converter(
        shape, traversal_order, format, b_size, b_map);
    std::vector<float> data;
    data.reserve(type.getNumElements());
    for (const auto val : attr.getValues<float>()) data.push_back(val);
    format_converter.DenseToSparse(data.data());
    sparsity =

  if (output_names) {
    output_names_vec.reserve(noutputs);
    for (int i = 0; i < noutputs; ++i) {
      output_names_vec.push_back(string(output_names[i]));
    }
  }

  // Process control output names.
  std::vector<string> control_output_names_vec;
  if (control_output_names) {
    control_output_names_vec.reserve(ncontrol_outputs);
    for (int i = 0; i < ncontrol_outputs; ++i) {

  std::vector<std::unique_ptr<ParallelTensor>> per_device_outputs;
  per_device_outputs.reserve(first_op_output_count);
  for (int i = 0; i < first_op_output_count; ++i) {
    std::vector<TensorHandlePtr> components;
    components.reserve(underlying_devices_.size());
    for (int j = 0; j < underlying_devices_.size(); ++j) {

  // directed edge (i->j) into the graph.
  std::vector<std::vector<uint64_t>> dependency_graph;
  dependency_graph.reserve(n);

  Operation &yield_op = body.back();
  auto results = yield_op.getOperands();
  for (auto i = 0; i < n; ++i) {
    std::vector<uint64_t> neighbors;
    neighbors.reserve(n);
    FindProducers(results[i], neighbors);
    dependency_graph.push_back(neighbors);
  }

  SmallVector<int64_t> new_start;
  SmallVector<int64_t> new_limit;
  SmallVector<int64_t> new_slice_shape;
  new_start.reserve(first.getStrides().size());
  new_limit.reserve(first.getStrides().size());
  new_slice_shape.reserve(first.getStrides().size());

  for (int i = 0; i < first.getStrides().size(); ++i) {
    const int64_t first_start =

std::optional<ConstBytesAttr> CustomOptions(MLIRContext* context,
                                            mlir::Operation* op) {
  if (auto reserve =
          llvm::dyn_cast_or_null<mlir::TF::TensorListReserveOp>(op)) {
    tflite::TensorType tflite_type =
        tflite::ConvertTypeToTensorType(reserve.getElementDtype());

    return CreateListReserveOptions(context, tflite_type);
  }

  llvm::SmallVector<mlir::APInt> ret;
  size_t read_size = sizeof(T);
  int bytes_len = bytes.size();
  assert(bytes_len % read_size == 0);

  int elem_count = bytes_len / read_size;
  ret.reserve(elem_count);

  const char* data_ptr = reinterpret_cast<const char*>(bytes.data());
  for (int i = 0; i < elem_count; i++) {
    T val = llvm::support::endian::readNext<T, llvm::endianness::native,

                                              device_->parsed_name().id));

    std::vector<xla::PjRtBuffer*> executable_args;
    executable_args.reserve(result->input_mapping.size());
    absl::flat_hash_set<int> non_donatable_input_indices;
    TF_EXPECT_OK(PreparePjRtExecutableArguments(
        /*num_missing_prefix_ctx_inputs=*/0, result->input_mapping, inputs,