namespace cpp {

OpView::OpView(OpSpec op)
    : op_(op),
      input_args_(op_.Inputs().begin(), op_.Inputs().end()),
      output_args_(op_.Outputs().begin(), op_.Outputs().end()),
      argument_attrs_(op_.Attributes().begin(), op_.Attributes().end()) {
  // Initialize function arguments
  all_arguments_.push_back(OpArgumentView("AbstractContext*", "ctx"));
  for (const auto& arg : op_.Inputs()) {

namespace internal {
Status Reset(AbstractOperation* op_, const char* op,
             const char* raw_device_name, ForwardOperation* forward_op_) {
  forward_op_->op_name = op;
  forward_op_->attrs.Reset(op);
  return op_->Reset(op, raw_device_name);
}
Status AddInput(AbstractOperation* op_, AbstractTensorHandle* input,
                ForwardOperation* forward_op_) {
  TF_RETURN_IF_ERROR(op_->AddInput(input));

LogicalResult Tf2XlaRewriter::LegalizeOp() {
  for (Type ty : op_->getOperandTypes()) {
    auto ranked_ty = mlir::dyn_cast<ShapedType>(ty);
    // Only bounded operands are supported in the XLA builders.
    if (!IsBounded(ranked_ty)) {
      return op_->emitRemark()
             << "lowering requires bounded tensor operands " << ranked_ty;
    }
  }

  if (HasSymbolRefAttr(op_)) {

    tensorflow::StringPiece s(data, length);
    op_->node_builder.Attr(attr_name, s);
    return absl::OkStatus();
  }
  Status SetAttrInt(const char* attr_name, int64_t value) override {
    op_->node_builder.Attr(attr_name, static_cast<int64_t>(value));
    return absl::OkStatus();
  }
  Status SetAttrFloat(const char* attr_name, float value) override {
    op_->node_builder.Attr(attr_name, value);

class Output {
 public:
  Output() = default;
  explicit Output(Node* n) : op_(n) {}
  Output(Node* n, int32_t index) : op_(n), index_(index) {}
  Output(const Operation& op, int32_t index) : op_(op), index_(index) {}

  Operation op() const { return op_; }
  Node* node() const { return op().node(); }
  int32 index() const { return index_; }
  DataType type() const { return op_.output_type(index_); }
  std::string name() const {

  void set_op(string op) {
    DCHECK(IsEmpty());
    op_ = std::move(op);
  }

  void set_assigned_device(string assigned_device) {
    DCHECK(IsEmpty());
    assigned_device_ = std::move(assigned_device);
  }

  void set_constant_value(Tensor constant_value) {
    DCHECK(IsEmpty());
    constant_value_ = std::move(constant_value);
    op_ = "Const";
  }

  void set_inputs(InputSeqMatcher inputs) {

// by cloning the root op to replicate that value. The old op will get folded.
def CloningOpResult : NativeCodeCall<
  "$_builder.clone(*op0)->getOpResult(0)">;

// Same as CloningOpResult but is used for ops with multiple results.
class CloningOpResults<int returns> : NativeCodeCall<
  "$_builder.clone(*op0)->getOpResults()", returns>;

// Creates an 1D array const with float values.
class Create1DConst<string values> : NativeCodeCall<

namespace generator {
namespace cpp {

class OpCommentRenderer : public Renderer {
 public:
  explicit OpCommentRenderer(RendererContext context, OpView op);

  void Render();

 private:
  OpView op_;
};

}  // namespace cpp
}  // namespace generator
}  // namespace tensorflow

  string FunctionName() const;
  string VariableName() const;
  string OpNameString() const;
  string Summary() const;
  std::vector<string> Description() const;
  bool IsListOp() const;

 private:
  OpSpec op_;
  std::vector<ArgView> input_args_;
  std::vector<ArgView> output_args_;
  std::vector<AttrView> argument_attrs_;
  std::vector<OpArgumentView> all_arguments_;
};

}  // namespace cpp
}  // namespace generator

			opxy = op_NY
		case AOR:
			opxy = op_OG
		case AORW:
			opx = op_O
			opxy = op_OY
		case AXOR:
			opxy = op_XG
		case AXORW:
			opx = op_X
			opxy = op_XY
		}
		if opx != 0 && 0 <= d2 && d2 < DISP12 {
			zRX(opx, uint32(r1), uint32(x2), uint32(b2), uint32(d2), asm)
		} else {
			zRXY(opxy, uint32(r1), uint32(x2), uint32(b2), uint32(d2), asm)
		}