HAL : hal-00622654, version 1

Fiche détaillée

Récupérer au format BibTeX - EndNote - TEI - RefWorks -

Stack-less SIMT reconvergence at low cost

Sylvain Collange (  Auteur correspondant ) 1

(12/09/2011)

Parallel architectures following the SIMT model such as GPUs benefit from application regularity by issuing concurrent threads running in lockstep on SIMD units. As threads take different paths across the control-flow graph, lockstep execution is partially lost, and must be regained whenever possible in order to maximize the occupancy of SIMD units. In this paper, we propose a technique to handle SIMT control divergence that operates in constant space and handles indirect jumps and recursion. We describe a possible implementation which leverage the existing memory divergence management unit, ensuring a low hardware cost. In terms of performance, this solution is at least as efficient as existing techniques.

1 :	ARENAIRE (Inria Grenoble Rhône-Alpes / LIP Laboratoire de l'Informatique du Parallélisme)
	INRIA – CNRS : UMR5668 – Université Claude Bernard - Lyon I – École Normale Supérieure - Lyon

Équipe de recherche : ARENAIRE

Domaine

:

Informatique/Architecture

Mots Clés : Control-flow reconvergence – SIMT – SIMD – GPU

hal-00622654, version 1

http://hal.archives-ouvertes.fr/hal-00622654

oai:hal.archives-ouvertes.fr:hal-00622654

Contributeur : Sylvain Collange <>

Soumis le : Lundi 12 Septembre 2011, 16:14:16

Dernière modification le : Vendredi 25 Novembre 2011, 15:12:10