Note on FastTwoSum with Directed Rounding
Résumé
In [1], Graillat and Jézéquel prove a bound on the maximal error for the FastTwoSum algorithm with directed roundings. We improve that bound by a factor 2, and show that FastTwoSum is exact when the exponent difference of the inputs does not exceed the current precision.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|