auto input_type = operands[0].getType().dyn_cast<ShapedType>();
  if (!input_type || !input_type.hasRank()) {
    // If input is unranked, then so is output.
    inferredReturnTypes.assign(
        num_value, UnrankedTensorType::get(input_type.getElementType()));
    return success();
  }

  if (input_type.hasStaticShape() && input_type.getNumElements() <= 0) {

    const std::vector<std::string>& input_dtypes,
    const std::vector<std::optional<std::vector<int>>>& input_shapes,
    const std::vector<std::string>& output_arrays,
    const std::vector<std::string>& control_output_arrays,
    const GraphdefToMlirOptions& import_options, mlir::MLIRContext* context) {
  auto module_or = GraphdefToMlirImport(
      input, input_arrays, input_dtypes, input_shapes, output_arrays,

absl::StatusOr<mlir::OwningOpRef<mlir::ModuleOp>>
GraphdefToMlirTranslateFunction(
    llvm::StringRef input, const std::vector<std::string>& input_arrays,
    const std::vector<std::string>& input_dtypes,
    const std::vector<std::optional<std::vector<int>>>& input_shapes,
    const std::vector<std::string>& output_arrays,
    const std::vector<std::string>& control_output_arrays,

  // * 3rd output is the outer offset.
  auto input_type = GetInputType(func, 0);
  if (!input_type || !mlir::isa<StringType>(input_type.getElementType()) ||
      !input_type.hasRank()) {
    return func.emitError() << "Input should be a string tensor";
  }

  const std::vector<int> kValidNumOfOutput = {1, 2, 3};
  if (input_type.getRank() >= kValidNumOfOutput.size()) {
    return func.emitError()

  auto input_type = input.getType().cast<ShapedType>();
  if (permutation_array.size() != input_type.getRank()) {
    return nullptr;
  }
  llvm::SmallVector<int64_t> transposed_shape(permutation_array.size());
  for (int64_t i = 0; i < permutation_array.size(); ++i) {
    transposed_shape[i] = input_type.getDimSize(permutation_array[i]);
  }
  auto transposed_type =

        file("inputs/input").text = "initial"
    }

    def "can override @Internal with @#inputType.simpleName"() {
        buildFile << """
            class InternalBaseTask extends BaseTask {
                @Internal def input
            }
            class CustomTask extends InternalBaseTask {
                @${inputType.name} def input
            }
            custom {
                def layout = project.layout

  llvm::SmallVector<int64_t, 4> input_shape(4, ShapedType::kDynamic);
  auto input_type = mlir::cast<TensorType>(op.getInput().getType());
  if (input_type.hasRank()) {
    if (input_type.getRank() != 4)
      return op.emitOpError()
             << "requires input to be a 4D tensor, but got " << input_type;

    int64_t input_batch = input_type.getDimSize(0);
    if (input_batch != ShapedType::kDynamic &&

    llvm::ArrayRef<int64_t> permutation_array, ShapedType input_type,
    ConversionPatternRewriter& rewriter) {
  assert(permutation_array.size() == input_type.getRank());
  llvm::SmallVector<int64_t> transposed_shape(permutation_array.size());
  for (int64_t i = 0; i < permutation_array.size(); ++i) {
    transposed_shape[i] = input_type.getDimSize(permutation_array[i]);
  }
  auto transposed_type =

  DataType input_type;
  string method;
  TF_RETURN_IF_ERROR(GetNodeAttr(op.node()->attrs(), "method", &method));
  TF_RETURN_IF_ERROR(GetNodeAttr(op.node()->attrs(), "T", &input_type));
  auto image_shape = Shape(scope, op.input(0));
  grad_outputs->push_back(CropAndResizeGradImage(
      scope, grad_inputs[0], op.input(1), op.input(2), image_shape, input_type,
      CropAndResizeGradImage::Method(method)));

    // can have smaller memory usage.
    auto input_type =
        mlir::dyn_cast<RankedTensorType>(dequantize_op.getOutput().getType());
    auto output_type = mlir::dyn_cast<RankedTensorType>(
        passthrough_op->getResult(0).getType());
    if (!input_type || !output_type ||
        get_num_elements(input_type) <= get_num_elements(output_type)) {
      return failure();
    }