if (dyn_cast_or_null<quantfork::QuantizeCastOp>(next_op)) return failure();

    auto input_type = mlir::cast<ShapedType>(transpose_op.getInput().getType());
    auto perm_type = mlir::cast<ShapedType>(transpose_op.getPerm().getType());
    if (input_type.hasStaticShape() && perm_type.hasStaticShape()) {
      if (perm_type.getNumElements() != input_type.getRank()) {
        return transpose_op.emitOpError(

  BoolAttr narrow_range = builder.getBoolAttr(false);

  auto add_quantize_op = [&](Location loc, Type input_type, Block* block,
                             Block::iterator insertion_point, Value arg,
                             int i) {
    if (auto shaped = mlir::dyn_cast<ShapedType>(input_type)) {
      if (mlir::isa<FloatType>(shaped.getElementType())) {

  auto output_type = RankedTensorType::get(output_shape, builder->getF32Type());
  SmallVector<mlir::Type, 4> input_types{input_type, weight_type, bias_type,
                                         projection_type,
                                         layer_norm_scale_type};
  auto func_type = builder->getFunctionType(input_types, output_type);

  auto func = func::FuncOp::create(

  BoolAttr narrow_range = builder.getBoolAttr(false);

  auto add_quantize_op = [&](Location loc, Type input_type, Block* block,
                             Block::iterator insertion_point, Value arg,
                             int i) {
    if (auto shaped = mlir::dyn_cast<ShapedType>(input_type)) {
      if (mlir::isa<FloatType>(shaped.getElementType())) {

GetBroadcastShapesForBatchMatmul(ShapedType input_type,
                                 ShapedType weight_type) {
  ArrayRef<int64_t> input_shape = input_type.getShape();
  ArrayRef<int64_t> weight_shape = weight_type.getShape();

  const int64_t num_matmul_dim = 2;
  const int64_t num_input_batch_dim = input_type.getRank() - num_matmul_dim;

  mlir::OpBuilder builder(func.getBody());
  auto& bb = func.front();
  auto loc = func.getLoc();

  int num_args = bb.getNumArguments();
  llvm::SmallVector<Type, 4> input_types;
  input_types.reserve(num_args);
  // Edit the block arguments and create the new input ops in place to replace
  // the old input ops and quantize ops.
  for (int i = 0; i != num_args; ++i) {

class Operation {
 public:
  Operation() : node_(nullptr) {}
  explicit Operation(Node* n);

  int32 num_inputs() const { return node_->num_inputs(); }
  DataType input_type(int32_t o) const { return node_->input_type(o); }
  Output input(int32_t i) const;

  int32 num_outputs() const { return node_->num_outputs(); }
  DataType output_type(int32_t o) const { return node_->output_type(o); }

  return attr_enforced_quantizable || trait_enforced_quantizable;
}

// Returns the quantized type for the
// input_type/min/max/storag_type_width/narrow_range.
// This is entry point to the Quant dialect and used for both quantizing
// activations and weights.
Type GetQuantizedType(Builder builder, const Type input_type,
                      const ArrayRef<double> min, const ArrayRef<double> max,

  const tflite::TensorType input_type =
      FromTocoDataTypeToTflitToTensorType(input_data_type);
  const tflite::TensorType output_type =
      FromTocoDataTypeToTflitToTensorType(output_data_type);

  std::string output_model;
  const absl::string_view input_model_buffer(buf, length);
  auto status = mlir::lite::QuantizeModel(
      input_model_buffer, input_type, output_type, inference_tensor_type,

  // Create a tfl.transpose op that performs ZX transpose on `input`.
  auto create_z_x_transpose_op = [&](Value input) -> Value {
    RankedTensorType input_type = mlir::cast<RankedTensorType>(input.getType());
    const int input_rank = input_type.getRank();

    // Create a 1D I32 tensor for representing the dimension permutation.
    auto permuation_tensor_type =