作者: Peter-Michael Seidel
DOI: 10.1016/S0167-9260(99)00008-5
关键词: Double-precision floating-point format 、 Booth's multiplication algorithm 、 Algorithm 、 Single-precision floating-point format 、 Normalized number 、 Rounding 、 Mathematics 、 Approximation error 、 Floating point 、 Reciprocal
摘要: Abstract This paper presents a fast implementation for reciprocal approximation, that can compute redundant of normalized number with precision 2 −28 in roughly 16–17 logic levels. Moreover, less accurate, but much cheaper is proposed. The representation the directly be fed into common Booth multiplier. allows to implement IEEE floating-point division correct rounding all modes latency 7 clock cycles double and 4 single precision. We also consider compressions from carry–save representations Booth-digit representations.