Shadow computation with BFloat16 to estimate thenumerical accuracy of summations - Archive ouverte HAL Access content directly
Conference Papers Year : 2021

Shadow computation with BFloat16 to estimate thenumerical accuracy of summations

Pablo de Oliveira Castro
Matei Istoan
Eric Petit

Abstract

In this article, we propose to exploit the new computational capability offered by the Bfloat16 representation format to perform shadow computations and compute estimations of the relative error. We demonstrate and evaluate the assumptions under which shadow computation is valid for the summation problem.
Fichier principal
Vignette du fichier
Pivoting_strategies (22).pdf (799.22 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03159965 , version 1 (04-03-2021)
hal-03159965 , version 2 (26-05-2021)

Identifiers

  • HAL Id : hal-03159965 , version 2

Cite

David Defour, Pablo de Oliveira Castro, Matei Istoan, Eric Petit. Shadow computation with BFloat16 to estimate thenumerical accuracy of summations. IEEE 28th Symposium on Computer Arithmetic (ARITH), Jun 2021, Virtual Conference, France. ⟨hal-03159965v2⟩
144 View
344 Download

Share

Gmail Facebook X LinkedIn More