// Otherwise, output type has same rank as the input.
    return RankedTensorType::get(
        SmallVector<int64_t, 4>(rank, ShapedType::kDynamic), element_ty);
  }

  int64_t num_reduce_dim = 0;
  llvm::SmallVector<bool, 4> is_reduce_dim(rank, false);
  for (const APInt &index : indices.getValues<APInt>()) {
    int64_t dim = GetDimForAxis(index.getSExtValue(), rank);
    // Invalid input.

std::optional<std::pair<SmallVector<int64_t>, SmallVector<int64_t>>>
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;

    auto can_convert_to_bilinear = [](bool align_corners, int64_t dilation,
                                      int64_t padding, int64_t stride,
                                      int64_t input_spatial,
                                      int64_t output_spatial) {
      int64_t input_spatial_size =
          align_corners ? input_spatial - 1 : input_spatial;
      int64_t output_spatial_size =

    // (spatial_filter_shape[i]-1) * dilation_rate[i]) / strides[i])
    auto get_output_dim_for_valid_padding =
        [](int64_t input_dim, int64_t dilation_dim, int64_t kernel_dim,
           int64_t stride_dim) -> int64_t {
      return std::ceil((input_dim - (kernel_dim - 1) * dilation_dim) /
                       stride_dim);
    };
    return output_height ==

// reached convergence, false otherwise.
FailureOr<bool> InferShapeForFunction(func::FuncOp func,
                                      ArrayRef<ArrayRef<int64_t>> arg_shapes,
                                      int64_t graph_version,
                                      int64_t max_iterations = 10,
                                      ArrayRef<TypeID> ops_to_skip = {});

}  // namespace TF

}  // namespace mlir

  int64_t input_rank = input_ty.getRank();
  ArrayRef<int64_t> input_shape = input_ty.getShape();
  SmallVector<int64_t, 4> normalized_sizes;

  for (int64_t i = 0; i < input_rank; ++i) {
    int64_t input_size = input_shape[i];
    int64_t start_index =
        constant_start_indices.getValues<IntegerAttr>()[i].getInt();
    int64_t slice_size = slice_sizes.getValues<IntegerAttr>()[i].getInt();

#include "mlir/IR/BuiltinTypes.h"  // from @llvm-project

namespace tensorflow {

llvm::SmallVector<int64_t> ConvertTFShapeToMlir(llvm::ArrayRef<int64_t> shapes);

llvm::SmallVector<int64_t> ConvertMlirShapeToTF(llvm::ArrayRef<int64_t> shape);

static constexpr int64_t kTFDynamicSize = -1;
mlir::RankedTensorType GetTypeFromTFTensorShape(llvm::ArrayRef<int64_t> shape,
                                                mlir::Type elementType,

      std::vector<int64_t> shape;
      shape.push_back(static_cast<int64_t>(op->window_strides.size()));
      attributes.emplace_back(builder.getNamedAttr(
          "window_strides",
          BuildVhloTensorV1Attr(shape, op->window_strides, builder)));
    } else {
      std::vector<int64_t> data(op->input_spatial_dimensions.size(), 1);
      std::vector<int64_t> shape;

  // Extract B, Y1, C from %lhs.
  ArrayRef<int64_t> shape_lhs =
      reshape_before.getOperand().getType().getShape();
  ArrayRef<int64_t> shape_b = shape_lhs.take_front(batch_dims_count);
  ArrayRef<int64_t> shape_c = shape_lhs.take_back(contracting_dims_count);
  ArrayRef<int64_t> shape_y1 =
      shape_lhs.drop_front(shape_b.size()).drop_back(shape_c.size());

#include "tensorflow/compiler/mlir/tensorflow/ir/tf_op_interfaces.h"

namespace mlir {

class MLIRContext;

namespace TF {

SmallVector<int64_t, 4> ReversePermutation(ArrayRef<int64_t> permutation);

SmallVector<int64_t, 4> GetDataFormatPermutation(StringRef from, StringRef to);

// Shuffle elements in the `attr` according to the permutation. Optional