Computing a correct and tight rounding error bound using rounding-to-nearest - Inria - Institut national de recherche en sciences et technologies du numérique
Communication Dans Un Congrès Année : 2016

Computing a correct and tight rounding error bound using rounding-to-nearest

Résumé

When a floating-point computation is done, it is most of the time incorrect. The rounding error can be bounded by folklore formulas, such as ε|x| or ε| o (x)|. This gets more complicated when underflow is taken into account as an absolute term must be considered. Now, let us compute this error bound in practice. A common method is to use a directed rounding in order to be sure to get an over-approximation of this error bound. This article describes an algorithm that computes a correct bound using only rounding to nearest, therefore without requiring a costly change of the rounding mode. This is formally proved using the Coq formal proof assistant to increase the trust in this algorithm.
Fichier principal
Vignette du fichier
article.pdf (261.08 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Loading...

Dates et versions

hal-01377152 , version 1 (06-10-2016)

Identifiants

  • HAL Id : hal-01377152 , version 1

Citer

Sylvie Boldo. Computing a correct and tight rounding error bound using rounding-to-nearest. 9th International Workshop on Numerical Software Verification, Jul 2016, Toronto, Canada. ⟨hal-01377152⟩
256 Consultations
496 Téléchargements

Partager

More