Performance estimation of linear algebra numerical libraries - Archive ouverte HAL Access content directly
Journal Articles Journal of Numerical Mathematics Year : 2015

Performance estimation of linear algebra numerical libraries

Abstract

In this work, numerical algebraic operations are performed by using several libraries whose algorithm are optimized to drain resources from hardware architecture. In particular, dot product of two vectors and the matrix-matrix product of two dense matrices are computed. In addition, the Cholesky decomposition on a real, symmetric, and positive definite matrix is performed through routines for band and sparse matrix storage. The involved CPU time is used as an indicator of the performance of the employed numerical tool. Results are compared to naive implementations of the same numerical algorithm, highlighting the speed-up due to the usage of optimized routines.
Fichier principal
Vignette du fichier
DEROSIS_JNM_2015.pdf (359.76 Ko) Télécharger le fichier
Origin : Publisher files allowed on an open archive
Loading...

Dates and versions

hal-01298333 , version 1 (12-04-2016)

Identifiers

Cite

Alessandro de Rosis. Performance estimation of linear algebra numerical libraries. Journal of Numerical Mathematics, 2015, 23, pp.13-19. ⟨10.1515/jnma-2015-0002⟩. ⟨hal-01298333⟩
60 View
482 Download

Altmetric

Share

Gmail Facebook X LinkedIn More