Software for Doubled-Precision Floating-Point Computations

A method for exact multiplication of two maxLmum-preclsion numbers, so that the result is represented usmg two such numbers, was proposed by T.J Dekker a decade ago. A modffication of Dekker's method Is presented and is proved to be valid in most existing arithmetics, while the original method is valid only in a qmte restricted class of anthmetms. The new method is computer independent, can be written using a high-level language, and still makes it possible to double quite econommally the "maximum available" precismn of a computer during crltwal phases of computatmns. Corresponding methods for doubling the preclsmn are presented and also analyzed for addition, subtraction, and diwsmn