"whose value is non-negative">;

def TFL_BoolTrue : AttrConstraint<
    CPred<"$_self.isa<BoolAttr>() && $_self.cast<BoolAttr>().getValue()">,
    "whose value is true">;

def TFL_BoolFalse : AttrConstraint<
    CPred<"$_self.isa<BoolAttr>() && !$_self.cast<BoolAttr>().getValue()">,
    "whose value is false">;

class TFL_StringEqualsTo<string value> : AttrConstraint<

                                      ImplicitLocOpBuilder& builder,
                                      bool is_lhs) {
  auto operand_type = mlir::cast<ShapedType>(operand.getType());
  BoolAttr true_attr = builder.getBoolAttr(true);
  auto operand_shape = builder.create<TF::ShapeOp>(operand, true_attr);
  const int64_t operand_rank = operand_type.getRank();

    // Set to `nullptr` because this attribute only matters when the input is
    // dynamic-range quantized.
    const BoolAttr asymmetric_quantize_inputs = nullptr;

    const int lhs_rank = lhs_value.getType().cast<TensorType>().getRank();
    const BoolAttr adj_x =
        (lhs_contracting_dims[0] == lhs_rank - 2 ? rewriter.getBoolAttr(true)

                    Attribute value) {
  ShapedType type;
  if (auto elem_attr = mlir::dyn_cast<ElementsAttr>(value)) {
    return ConstOp::build(builder, result, elem_attr);
  } else if (mlir::isa<BoolAttr, FloatAttr, IntegerAttr>(value)) {
    // All TensorFlow types must be tensor types. In the build() method,
    // we want to provide more flexibility by allowing attributes of scalar

  auto array = attr.cast<mlir::ArrayAttr>().getValue();

  for (int i = 0; i < array.size(); i++) {
    if (llvm::isa<mlir::BoolAttr>(array[i])) {
      flex_builder->Bool(name.c_str(),
                         array[i].cast<mlir::BoolAttr>().getValue());
    } else if (llvm::isa<mlir::StringAttr>(attr)) {
      flex_builder->String(name.c_str(),

                        Value filter, Value bias,
                        StringAttr fused_activation_function,
                        StringAttr weights_format, BoolAttr keep_num_dims,
                        BoolAttr asymmetric_quantize_inputs) {
  auto fc_op = builder->create<FullyConnectedOp>(
      result[0].getLoc(), result.getTypes(), input, filter, bias,

  return mlir::OpTrait::impl::verifyCompatibleOperandBroadcast(
      op.getOperation());
}

void NotEqualOp::build(OpBuilder &builder, OperationState &result, Value x,
                       Value y, BoolAttr incompatible_shape_error) {
  auto result_type = DeduceEqualCmpOpType(&builder, result.location, x, y,
                                          incompatible_shape_error);

  }

  // Default to non-time_major.
  std::optional<mlir::NamedAttribute> time_major_attr =
      attr.getNamed("time_major");
  bool time_majored =
      time_major_attr ? time_major_attr->getValue().cast<BoolAttr>().getValue()
                      : false;

  int64_t batch =
      time_majored ? input_type.getDimSize(1) : input_type.getDimSize(0);
  int64_t time =

  auto composite_control_flow_op = [&](const std::string& name) {
    result.name = mlir::OperationName(get_full_op_name(name), context_);
    bool stateless = absl::StartsWith(node_type_name, "Stateless");
    mlir::BoolAttr val = builder_.getBoolAttr(stateless);
    result.attributes.push_back(builder_.getNamedAttr("is_stateless", val));
  };

  // Map Case/If/While and StatelessCase/If/While op in TensorFlow to the common