Computation of numerical algorithms with parametric floating point operators

• Published in: Journal of systems architecture Vol. 43; no. 8; pp. 587 - 599
• Main Authors: Bartolucci, Massimo, Guazzoni, Paolo, Manfredi, Giorgio, Sechi, Giacomo, Zetta, Luisa
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.05.1997; Elsevier Sequoia S.A
• Subjects: Algorithms; Arithmetic and conversion operators; Computer architecture; Nuclear electronics; Real time; Real-time processing; Studies; Real-time processing; Nuclear electronics; Arithmetic and conversion operators
• ISSN: 1383-7621; 1873-6165
• DOI: 10.1016/S1383-7621(96)00117-8

The computation complexity of a large number of applications, makes useful the development of new design strategies both in the direction of correctness and reduction of computational time and size of the used devices. Since a large part of numerical algorithms involves the computation of polynomial expansions in terms of sum and products of floating point numbers, a way for reducing both dimension and execution time of the devices, is to adapt the elementary operators and the number of terms of the polynomial expansion to the very computational needs. In this work a design method consisting in using an ad hoc designed set of Elementary Floating Point Operators parametrically defined is presented. Moreover, we discuss the mantissa size influence on real-time computation of numerical algorithms, in order to state the truncation limit for which the correctness constraints are anyhow saved for different applications in order to allow a more reliable choice of the hardware, and to limit also the general costs.