Proposal to improve data format conversions for a hybrid number system processor
Lucian Jurca, Aurel Stefan Gontean, Florin Alexa, Daniel-Ioan Curiac · 2007
Abstract:- In this paper we show how the multiplication or division of two single-precision floating-point operands can be solved faster and more accurately than in previous works that used logarithmic arithmetic. The logarithm calculation and ALU operation are fused in order to perform one single non-redundant addition in the critical path for finding the logarithm of the result. A second non-redundant addition is used to produce the result in floating-point format. Using Matlab analysis, the conversion error was also diminished by using correction values in the look-up tables content.