ABSTRACT
The floating-point processor has an important role in high-precision calculation, image acceleration and digital signal processing.The floating-point division is the most time-consuming among four floating-point arithmetic operations, and it is an important part of the floating-point processor. In the floating-point division design, functional iteration is widely used (Flynn 1970). An important factor influencing functional iteration algorithm is the precision of initial reciprocal approximation (Fowler & Smith 1989). The precision of the initial reciprocal approximation decides the number of iteration for the final result, which influence the performance of division operation (Tang 1991, Oberman & Flynn1997). In order to generate more accurate reciprocal approximation with little table size, this paper presents different methods and their error analysis in detail. And an optimized symmetric table with redundant Booth encoding is discussed particularly, combining getting initial approximation and Booth encoding together, which can reduce the logic circuit area of the division based on functional iteration.
