Dead Block Replacement and Bypass with a Sampling Predictor

Abstract : We present a cache replacement and bypass policy driven by dead block prediction. A block is considered dead is it will be replaced before it will be used again. If dead blocks can be identified, then they can be replaced early. If a block is predicted to be “dead on arrival,” i.e., it will not be accessed again after it is placed in the cache, then it can bypass the cache. The predictor is based on one simple observation: if a block becomes dead after being touched by a particular instruction, then other blocks touched by that instruction are also likely to become dead. Ideally, we would track the tendency of blocks to become dead for every instruction accessing the cache. However, to fit within a realistic hardware budget, we sample only a few sets from the cache. This paper describes our sampling dead block predictor and the techniques used to make it fit within the allow hardware budget for the cache replacement contest.
Type de document :
Communication dans un congrès
Joel Emer. JWAC 2010 - 1st JILP Worshop on Computer Architecture Competitions: cache replacement Championship, Jun 2010, Saint Malo, France. 2010
Liste complète des métadonnées

Littérature citée [9 références]  Voir  Masquer  Télécharger

https://hal.inria.fr/inria-00492930
Contributeur : Ist Rennes <>
Soumis le : jeudi 17 juin 2010 - 14:23:27
Dernière modification le : lundi 20 juin 2016 - 14:10:32
Document(s) archivé(s) le : lundi 20 septembre 2010 - 17:25:49

Fichier

001_jiminez.pdf.pdf
Fichiers produits par l'(les) auteur(s)

Identifiants

  • HAL Id : inria-00492930, version 1

Collections

Citation

Daniel A. Jimenez. Dead Block Replacement and Bypass with a Sampling Predictor. Joel Emer. JWAC 2010 - 1st JILP Worshop on Computer Architecture Competitions: cache replacement Championship, Jun 2010, Saint Malo, France. 2010. 〈inria-00492930〉

Partager

Métriques

Consultations de la notice

119

Téléchargements de fichiers

255