LogicalResult XlaVariadicSortOp::verify() {
  XlaVariadicSortOp op = *this;
  const auto &inputs_ty = op.getInputs().getType();
  int n_inputs = inputs_ty.size();
  auto input_ty_0 = inputs_ty[0].cast<ShapedType>();
  if (input_ty_0.hasStaticShape()) {
    for (int i = 1; i < n_inputs; ++i) {
      auto input_ty_i = inputs_ty[i].cast<ShapedType>();
      if (input_ty_i.hasStaticShape() &&

  DataFormatVecPermuteOp op = *this;
  auto input_ty = mlir::dyn_cast<RankedTensorType>(op.getX().getType());
  if (!input_ty) return success();

  int rank = input_ty.getRank();
  if (rank != 1 && rank != 2)
    return op.emitOpError("requires input of rank 1 or 2");

  if (rank == 1) {
    int64_t dim0 = input_ty.getDimSize(0);
    if (dim0 != ShapedType::kDynamic && dim0 != 4 && dim0 != 2)

    // as its an input requirement.
    if (!input_ty.hasRank() || input_ty.getRank() != 4) {
      return failure();
    }

    int64_t batch_cst = input_ty.getShape()[0];
    int64_t channels_cst = input_ty.getShape()[3];

    int64_t in_y_cst = input_ty.getShape()[1];
    int64_t in_x_cst = input_ty.getShape()[2];
    int64_t in_spatial_cst =

        readingList.add(toKatakana(text));
        return readingList;
    }

    protected String toKatakana(final String inputStr) throws IOException {
        final StringBuilder kanaBuf = new StringBuilder();

        final Reader rd = new StringReader(inputStr);
        try (TokenStream stream = createTokenStream(rd)) {
            if (stream == null) {
                throw new IOException("Invalid tokenizer.");

    for (auto const& x : out) {
      inputs_.push_back(x);
    }
  }

  typename std::vector<Input>::iterator begin() { return inputs_.begin(); }
  typename std::vector<Input>::iterator end() { return inputs_.end(); }
  typename std::vector<Input>::const_iterator begin() const {
    return inputs_.begin();
  }
  typename std::vector<Input>::const_iterator end() const {
    return inputs_.end();
  }

 private:

}

Type InferReductionOpType(Value input, Value reduction_indices,
                          BoolAttr keep_dims) {
  Type input_ty = input.getType();
  Type element_ty = getElementTypeOrSelf(input_ty);

  // Output type is unranked if input type is not ranked.
  auto ranked_ty = mlir::dyn_cast<RankedTensorType>(input_ty);
  if (!ranked_ty) return UnrankedTensorType::get(element_ty);
  int64_t rank = ranked_ty.getRank();

    }
    int axis = val.getSExtValue();

    auto input_ty = mlir::dyn_cast<RankedTensorType>(op.getInput().getType());
    if (!input_ty || !input_ty.hasStaticShape()) {
      return rewriter.notifyMatchFailure(
          op, "require the type of input to have static shapes");
    }
    ArrayRef<int64_t> input_shape = input_ty.getShape();
    int input_rank = input_ty.getRank();
    if (axis < 0) axis += input_rank;

  std::unordered_set<int> stop_backprop_nodes =
      GetStopBackpropNodes(reachable_nodes, output_nodes);

  // Populate `input_nodes_` from Outputs in `inputs_`.
  input_nodes_.reserve(inputs_.size());
  for (size_t i = 0; i < inputs_.size(); ++i) {
    input_nodes_.insert({inputs_[i], i});
  }

  // TODO(andydavis) Consider a more efficient data structure for `pending_` to

           const std::vector<T>& expected_output_values) {
    ClientSession session(scope_);

    DCHECK_EQ(input_values.size(), inputs_.size());
    ClientSession::FeedType feeds;
    for (int i = 0; i < inputs_.size(); ++i) {
      feeds.emplace(inputs_[i], input_values[i]);
    }

    std::vector<Tensor> out_tensors;
    TF_ASSERT_OK(session.Run(feeds, outputs_, &out_tensors));

  }

  void Run(const std::vector<TF_Output>& run_outputs,
           std::initializer_list<int> input_values) {
    DCHECK_EQ(inputs_.size(), input_values.size());
    std::vector<std::pair<TF_Operation*, TF_Tensor*>> inputs(inputs_.size());
    int i = 0;
    for (int v : input_values) {
      inputs[i] = {inputs_[i].oper, Int32Tensor(v)};
      ++i;
    }
    // TODO(skyewm): use std::make_unique or absl::make_unique when possible.