Forcing one-sided results in Goldschmidt algorithm

This paper presents a method to obtain one-sided error results from Goldschmidt (GLD) algorithm. In some applications is useful to obtain a one-sided error result. This is done introducing error bias in the intermediate iterations. An error analysis permits to obtain expressions to estimate the loss of precision and to compensate it. In this way, the one-sided error results are obtained without significant additional hardware requirements.

keywords: Analytical models, Application software, Approximation algorithms, Computer errors, Concurrent computing, Error analysis, Hardware, Roundoff errors, System performance