for (auto idx : val1_indices) {
    if (idx < 0) idx = idx + val1_shape.getRank();
    if (idx >= val1_shape.getRank() || val1_shape.isDynamicDim(idx)) {
      return false;
    }
    val1_result *= val1_shape.getDimSize(idx);
  }

  for (auto idx : val2_indices) {
    if (idx < 0) idx = idx + val2_shape.getRank();
    if (idx >= val2_shape.getRank() || val2_shape.isDynamicDim(idx)) {
      return false;
    }

  const std::vector<int> kValidNumOfOutput = {1, 2, 3};
  if (input_type.getRank() >= kValidNumOfOutput.size()) {
    return func.emitError()
           << "Unrecognized input rank: " << input_type.getRank();
  }
  if (func.getNumResults() != kValidNumOfOutput[input_type.getRank()]) {
    return func.emitError()
           << "Expect " << kValidNumOfOutput[input_type.getRank()]
           << "output(s) when input has rank " << input_type.getRank();

    data_rank = then_tensor.getRank();
    if (then_tensor.getRank() > 0)
      data_first_dim = then_tensor.getShape().front();
    if (else_tensor.getRank() > 0)
      data_first_dim = std::max(else_tensor.getShape().front(), data_first_dim);
  } else if (then_has_rank) {
    data_rank = then_tensor.getRank();
    if (then_tensor.getRank() > 0)
      data_first_dim = then_tensor.getShape().front();

  return type && type.getRank() == rank &&
         mlir::isa<FloatType>(type.getElementType());
}

// Returns true if the given `value` has the specified rank or has unranked
// type.
inline bool IsOfRankOrUnranked(Value value, int64_t rank) {
  RankedTensorType type = GetRankedTensorTypeForOperand(value);
  return !type || type.getRank() == rank;
}

    ShapedType& updates_type, ConversionPatternRewriter& rewriter) {
  auto canonical_update_window_dims = llvm::to_vector(
      llvm::seq<int64_t>(indices_type.getRank() - 1, updates_type.getRank()));

  if (canonical_update_window_dims == update_window_dims) return success();

  // Permute updates if `update_window_dims` are leading indices.

  if (!output_ty) return success();

  int64_t expected_output_rank = std::max(x_ty.getRank(), y_ty.getRank());
  if (output_ty.getRank() != expected_output_rank) {
    return op.emitOpError()
           << "found invalid output rank, expected " << expected_output_rank
           << " but got " << output_ty.getRank();
  }

  // Check output batch dim with potential broadcasting.

    auto reshape_input_ty = mlir::dyn_cast<ShapedType>(reshape_input.getType());
    if (!reshape_input_ty.hasStaticShape() || input_ty.getRank() == 0 ||
        reshape_input_ty.getRank() == 0 ||
        input_ty.getDimSize(input_ty.getRank() - 1) !=
            reshape_input_ty.getDimSize(reshape_input_ty.getRank() - 1)) {
      return failure();
    }

    // Connect the input to the one of reshape.

  auto input_shape = mlir::cast<ShapedType>(input.getType());
  auto filter_shape = mlir::cast<ShapedType>(filter.getType());
  if (!input_shape.hasRank() || input_shape.getRank() != 4 ||
      !filter_shape.hasRank() || filter_shape.getRank() != 4) {
    emitError(loc, "input and filter are expected to be 4D tensors");
    return {};
  }

  const int feature_group_cnt =

  auto operand_shape = builder.create<TFL::ShapeOp>(
      RankedTensorType::get(static_cast<int32_t>(operand_type.getRank()),
                            builder.getIntegerType(32)),
      operand);
  const int64_t operand_rank = operand_type.getRank();
  // Compute flattened out dimension and contracting dimension using
  // TFL::UnsortedSegmentProdOp.

  if (!output_type) return failure();

  // bias should be a vector sized of the last output dim.
  int64_t num_units = output_type.getDimSize(output_type.getRank() - 1);
  auto bias_type =
      mlir::RankedTensorType::get({num_units}, output_type.getElementType());

  mlir::DenseElementsAttr bias_attr;
  if (output_type.getElementType().isF32()) {
    float val = 0.0;