An analysis of floating-point addition

The major problem in designing a floating-point system is the addition operation. Knowledge of the relative importance of each of the suboperations can assist the designer in materially improving performance. Of special import are the provisions for shifting in pre-addition alignment of the operand radix points and in post-addition normalization of the result. Since these shifts vary over a wide range, provision must be made for the worst cases, without requiring the time of these extremes for all cases.