return operands_;
  }
  absl::Span<Predicate* const> operands() const { return operands_; }

 private:
  std::vector<Predicate*> operands_;
};

// Represents a logical negation of a set of predicates.
class NotPredicate : public Predicate {
 public:
  explicit NotPredicate(int64_t id, Predicate* operand)
      : Predicate(id), operands_({operand}) {}

			var i Expr
			if p.tok != _Colon {
				var comma bool
				if p.tok == _Rbrack {
					// invalid empty instance, slice or index expression; accept but complain
					p.syntaxError("expected operand")
					i = p.badExpr()
				} else {
					i, comma = p.typeList(false)
				}
				if comma || p.tok == _Rbrack {
					p.want(_Rbrack)
					// x[], x[i,] or x[i, j, ...]
					t := new(IndexExpr)

  for (auto operand : forward_caller->getOperands()) {
    if (llvm::isa<BlockArgument>(operand)) {
      // Pull this from the original operands to the original while op.
      auto arg = llvm::cast<BlockArgument>(operand);
      f_operands.push_back(loop_operands[arg.getArgNumber()]);
      continue;
    }
    auto src = operand.getDefiningOp();

// If an error occurs, x.mode is set to invalid.
func (check *Checker) matchTypes(x, y *operand) {
	// mayConvert reports whether the operands x and y may
	// possibly have matching types after converting one
	// untyped operand to the type of the other.
	// If mayConvert returns true, we try to convert the
	// operands to each other's types, and if that fails
	// we report a conversion failure.

    llvm::SmallVector<Value, 4> operands(addn_op.getInputs().begin(),
                                         addn_op.getInputs().end());

    int64_t n = operands.size();
    // Keep doing tree-based reduction when there are more than one operand.
    while (n > 1) {
      for (int64_t i = 0; i < n; i += 2) {
        // Add two adjacent operands if applicable.
        operands[i / 2] =

// we will use the original operand's type, otherwise update it with the
// unranked tensor type.
template <typename R>
void UpdateTensorListTypes(
    const llvm::SmallSet<int, 4> &tensor_list_index,
    const llvm::SmallSet<int, 4> &resized_tensor_list_index,
    ArrayRef<Type> types, R &&range, ValueRange operands,
    llvm::SmallVectorImpl<Type> *updated_types) {
  int i = 0;

  auto replace_operand_usage = [&](OpOperand& operand) {
    if (mlir::TF::CanBeRefined(operand.get().getType()) ||
        HasDynamicOutputs(operand.getOwner())) {
      return insertion_point->getParentRegion()->isAncestor(
          operand.getOwner()->getParentRegion());
    }
    return insertion_point->getParentRegion()->isAncestor(
               operand.getOwner()->getParentRegion()) ||

    (BinaryOp:$result (TFL_TileOp $input, (Arith_ConstantOp $tile)),
     $operand, $act_func),
    (BinaryOp $input, $operand, $act_func),
  [(OperandsBroadcastToOutputType $input, $operand, $result),
   (HasRankAtMost<4> $input),
   (HasRankAtMost<4> $operand)]>;

  def FuseTileBroadcastToBinaryOp2#BinaryOp : Pat<
    (BinaryOp:$result $operand,
      (TFL_TileOp $input, (Arith_ConstantOp $tile)), $act_func),

// CHECK-NOT:  "tf.TPUReplicatedOutput"


// Test replication with replicated operands and replicated results. The cluster
// will be wrapped in a `tf_device.cluster` first and then by a replicate.
// TPUReplicatedInput and TPUReplicatedOutput nodes will be replaced by the
// replicate operands and results.
// CHECK-LABEL: func @replication

          new_return.setOperand(resource_arg_to_new_output[index], value);
        });
  }

  // Recreate the op without resource operands.
  OpBuilder builder(op);
  // Now use the filtered original operands, which will be replaced by
  // AddLoadsStoresOutsideControlFlowOp().
  auto new_operands =
      FilterRange<Value, OperandRange>(op.getInput(), resource_arg_uses);