return OP(0, 6)
	case ASRA:
		return OP(0, 7)
	case AROTR:
		return OP(8, 6)
	case ASLLV:
		return OP(2, 4)
	case ASRLV:
		return OP(2, 6)
	case ASRAV:
		return OP(2, 7)
	case AROTRV:
		return OP(10, 6)
	case AADDV:
		return OP(5, 4)
	case AADDVU:
		return OP(5, 5)
	case ASUBV:
		return OP(5, 6)
	case ASUBVU, ANEGV:
		return OP(5, 7)
	case AREM,

        op.getLoc(), /*resultType0=*/
        mlir::cast<TensorType>(op.getResult().getType())
            .clone(output_uniform_quantized_type),
        /*lhs=*/op.getLhs(), /*rhs=*/op.getRhs(),
        /*dot_dimension_numbers=*/op.getDotDimensionNumbers(),
        /*precision_config=*/op.getPrecisionConfigAttr());

    rewriter.replaceAllUsesWith(op.getResult(), new_dot_general_op.getResult());

		// then clear remaining sizes as available
		for rem > 0 {
			op, size := ppc64.AMOVB, int64(1)
			switch {
			case rem >= 8:
				op, size = ppc64.AMOVD, 8
			case rem >= 4:
				op, size = ppc64.AMOVW, 4
			case rem >= 2:
				op, size = ppc64.AMOVH, 2
			}
			p := s.Prog(op)
			p.From.Type = obj.TYPE_REG
			p.From.Reg = ppc64.REG_R0
			p.To.Type = obj.TYPE_MEM

  TFE_Op* op = TFE_NewOp(ctx, "ReadVariableOp", status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_OpSetAttrType(op, "dtype", TF_FLOAT);
  TFE_OpAddInput(op, var_handle, status);
  ASSERT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  int num_retvals = 1;
  TFE_TensorHandle* value_handle = nullptr;
  TFE_Execute(op, &value_handle, &num_retvals, status);

       SingleBlockImplicitTerminator<"YieldOp">]> {
  let summary = "while operation";
  let description = [{
  The tf.WhileRegion op represents a while loop using 2 regions and a set of
  iteration variables. The iteration variables maintained by this Op have the
  same types as the inputs. The Op executes a while loop described by the
  following pseudo code:

  ```
     func WhileRegionOp(inputs) {

		m = min(m, r)
	}
	return m
}

// op pushes a regexp with the given op onto the stack
// and returns that regexp.
func (p *parser) op(op Op) *Regexp {
	re := p.newRegexp(op)
	re.Flags = p.flags
	return p.push(re)
}

// repeat replaces the top stack element with itself repeated according to op, min, max.
// before is the regexp suffix starting at the repetition operator.

  defm : FuseActFnIntoConvOpPat<!cast<Op>(actFnPair[0]), !cast<ConstantStrAttr>(actFnPair[1])>;
  defm : FuseActFnIntoPoolOpPat<!cast<Op>(actFnPair[0]), !cast<ConstantStrAttr>(actFnPair[1])>;
}

class CanFuseConvOrDepthwiseConv<string is_depthwise> : Constraint<
  CPred<"TFL::CanFuseConvOrDepthwiseConv($0, $1, " # is_depthwise # ")">>;

// If we see a binary op (add, sub) op adding a constant value to a convolution

		}
		x.val = constant.UnaryOp(op, x.val, prec)
		x.expr = e
		check.overflow(x, x.Pos())
		return
	}

	x.mode = value
	// x.typ remains unchanged
}

func isShift(op token.Token) bool {
	return op == token.SHL || op == token.SHR
}

func isComparison(op token.Token) bool {
	// Note: tokens are not ordered well to make this much easier
	switch op {

		}

		switch nxt.Op {
		case OpAdd8:
			nxt.Op = OpSub8
		case OpAdd16:
			nxt.Op = OpSub16
		case OpAdd32:
			nxt.Op = OpSub32
		case OpAdd64:
			nxt.Op = OpSub64
		case OpSub8:
			nxt.Op = OpAdd8
		case OpSub16:
			nxt.Op = OpAdd16
		case OpSub32:
			nxt.Op = OpAdd32
		case OpSub64:
			nxt.Op = OpAdd64
		default:
			panic("unreachable")
		}

	// then they can't overlap.
	switch p1.Op {
	case OpAddr, OpLocalAddr:
		if p2.Op == OpAddr || p2.Op == OpLocalAddr || p2.Op == OpSP {
			return true
		}
		return (p2.Op == OpArg || p2.Op == OpArgIntReg) && p1.Args[0].Op == OpSP
	case OpArg, OpArgIntReg:
		if p2.Op == OpSP || p2.Op == OpLocalAddr {
			return true
		}
	case OpSP: