CPred<"$_op.getOperand(0).getType().cast<ShapedType>().getRank() <= 1">,
          CPred<"$_op.getOperand(0).getType().cast<ShapedType>().getRank() == 2 && !$_op.getOperand(0).getType().cast<ShapedType>().hasStaticShape()">,
          CPred<"$_op.getOperand(0).getType().cast<ShapedType>().getRank() == 2 && $_op.getOperand(0).getType().cast<ShapedType>().getShape()[1] <= 4">]>>]> {

          .isExactlyValue(0.0f));

  EXPECT_EQ(fused_lstm_func_.getFunctionType().getNumResults(), 1);
  auto output_types = fused_lstm_func_.getFunctionType().getResults();
  SmallVector<int64_t, 2> output_shape{1, mlir::ShapedType::kDynamic};
  EXPECT_EQ(mlir::cast<RankedTensorType>(output_types[0]).getShape().size(),
            output_shape.size());
  for (int i = 0; i < output_shape.size(); i++) {

      return failure();
    }

    // Remove identity reshape with both static result and input shape.
    auto result_type = mlir::cast<ShapedType>(op.getType());
    auto input_type = mlir::cast<ShapedType>(op.getInput().getType());

    // Constant folding
    // If the result type isn't static, tries to derive the result type from
    // the #2 operand.

// not.
const char kLegalizeTflVariables[] = "tfl._legalize_tfl_variables";

bool HasSupportedElementType(Operation* op) {
  return utils::IsSupportedVariableType(op);
}

bool IsSupportedElementType(ShapedType type) {
  return utils::IsSupportedVariableType(type);
}

#include "tensorflow/compiler/mlir/lite/transforms/generated_legalize_variables.inc"

// Pass which legalizes TF variables which are already passed as bounded

    const xla::ConvolutionDimensionNumbers &dnums, PatternRewriter &rewriter) {
  StringAttr conv_padding = op.getPaddingAttr();
  SmallVector<int64_t> padding_nums;
  ShapedType lhs_shape = mlir::cast<ShapedType>(op.getLhs().getType());
  ShapedType rhs_shape = mlir::cast<ShapedType>(op.getRhs().getType());

  // Handle only static shape cases.
  // TODO(b/260284866): Handle dynamic shape cases.
  if (!lhs_shape.hasStaticShape()) {

// returns the return type of the tfl.average_pool_2d emitted. Note that the
// aten.avg_pool2d works with the NCHW layout while tfl.average_pool_2d assumes
// NHWC.
ShapedType GetNhwcReturnTypeFromNchw(Operation* old_op);

}  // namespace odml

}  // namespace mlir

    if (index < 0 || index >= static_cast<int>(op->getNumResults()))
      return false;
    Value res = op->getResult(index);
    return res.getType().isa<ShapedType>() &&
           res.getType().cast<ShapedType>().getElementType().isa<FloatType>();
  }

  // A method to retrieve the name for the given op.
  OperationToName op_to_name_;

// or if 'op' is null.
bool NotTFLQuantDequantizeOp(Operation* op);

// Returns true if it is a shaped type of f32 elements.
inline bool IsF32ShapedType(Type t) {
  if (auto shaped_type = mlir::dyn_cast_or_null<ShapedType>(t)) {
    return shaped_type.getElementType().isF32();
  }
  return false;
}

// Return true when the given element_type is QI8.
inline bool IsQI8Type(Type t) {

                 std::back_inserter(real_values), [&](APFloat value) -> float {
                   return value.convertToFloat();
                 });
  const ShapedType new_dense_type = dyn_cast_or_null<ShapedType>(
      q_type.castExpressedToStorageType(real_values_attr.getType()));
  const int width = dyn_cast<IntegerType>(q_type.getStorageType()).getWidth();

    // Slice each row of `i` and `v` to perform a separate dynamic-update-slice
    // on the contents of `x`.
    auto input_type = mlir::cast<ShapedType>(input.getType());
    auto updates_type = mlir::cast<ShapedType>(updates.getType());
    auto indices_type = mlir::cast<ShapedType>(indices.getType());
    if (!input_type.hasRank()) return failure();
    if (!updates_type.hasRank() || updates_type.isDynamicDim(0))