Multigrain Affinity for Heterogeneous Work Stealing

Jean-Yves Vet 1 Patrick Carribault 1 Albert Cohen 2
2 Parkas - Parallélisme de Kahn Synchrone
DI-ENS - Département d'informatique de l'École normale supérieure, ENS Paris - École normale supérieure - Paris, Inria Paris-Rocquencourt, CNRS - Centre National de la Recherche Scientifique : UMR 8548
Abstract : In a parallel computing context, peak performance is hard to reach with irregular applications such as sparse linear algebra operations. It requires dynamic adjustments to automatically balance the workload between several processors. The problem becomes even more complicated when an architecture contains processing units with radically different computing capabilities. We present a hierarchical scheduling scheme designed to harness several CPUs and a GPU. It is built on a two-level work stealing mechanism tightly coupled to a software-managed cache. We show that our approach is well suited to dynamically control heterogeneous architectures, while taking advantage of a reduction of data transfers.
Type de document :
Communication dans un congrès
Programmability Issues for Heterogeneous Multicores, Jan 2012, France
Liste complète des métadonnées


https://hal.archives-ouvertes.fr/hal-00875338
Contributeur : Jean-Yves Vet <>
Soumis le : lundi 21 octobre 2013 - 17:15:42
Dernière modification le : mercredi 28 septembre 2016 - 16:11:03
Document(s) archivé(s) le : vendredi 7 avril 2017 - 14:22:00

Fichier

multigrain.pdf
Fichiers éditeurs autorisés sur une archive ouverte

Identifiants

  • HAL Id : hal-00875338, version 1

Collections

INRIA | CEA | PSL

Citation

Jean-Yves Vet, Patrick Carribault, Albert Cohen. Multigrain Affinity for Heterogeneous Work Stealing. Programmability Issues for Heterogeneous Multicores, Jan 2012, France. <hal-00875338>

Partager

Métriques

Consultations de
la notice

543

Téléchargements du document

166