loc, revised_output_type, original_input, shape);

    // Replace the original strided_slice.
    uint64_t revised_begin_mask = strided_slice_op.getBeginMask();
    uint64_t revised_end_mask = strided_slice_op.getEndMask();
    // Since we expand the dims, we need to apply them to the begin_mask &
    // end_mask.
    revised_begin_mask |= strided_slice_op.getNewAxisMask();
    revised_end_mask |= strided_slice_op.getNewAxisMask();

  // All masks are `0` except `shrink_axis_mask` which is equal to `1` (slicing
  // scalar value from input vector).
  if (slice_op.getBeginMask() != 0 || slice_op.getEllipsisMask() != 0 ||
      slice_op.getEndMask() != 0 || slice_op.getNewAxisMask() != 0 ||
      slice_op.getShrinkAxisMask() != 1)
    return {};

  // Returns a value if the `value` is defined by a ConstOp with a single

        strided_slice_op.getEnd().getLoc(), new_added_value);

    if (strided_slice_op.getBeginMask() != 0) return failure();
    if (strided_slice_op.getEndMask() != 0) return failure();
    if (strided_slice_op.getEllipsisMask() != 0) return failure();
    mlir::TFL::StridedSliceOp new_strided_slice_op =
        rewriter.create<TFL::StridedSliceOp>(

  if (num_input_dims + num_added_axis > 5) return failure();
  return success();
}

OpFoldResult StridedSliceOp::fold(FoldAdaptor) {
  // Currently only support all masks being 0.
  if (getBeginMask() != 0 || getEndMask() != 0 || getEllipsisMask() != 0 ||
      getNewAxisMask() != 0 || getShrinkAxisMask() != 0)
    return {};

  auto input_type = getInput().getType().dyn_cast_or_null<RankedTensorType>();

    // verifier.
    int64_t slicing_dim_size =
        mlir::cast<RankedTensorType>(op.getBegin().getType()).getDimSize(0);
    uint64_t begin_mask = op.getBeginMask();
    uint64_t end_mask = op.getEndMask();
    const int input_rank = input_shape.size();
    for (int d = 0; d < input_rank; ++d) {
      // Each dimension is either sliced fully or has size of one.