Extending the Task Dataflow Model with Speculative Data Accesses
Résumé
In this paper, we describe a data-centric version-based approach to extending the task dataflow model with new access types that provide speculation capabilities to applications. We further describe the performance benefits of such a model using two scenarios that model realistic applications ; the parallelization of a finite-state machine and Monte-Carlo simulations.
Origine | Fichiers produits par l'(les) auteur(s) |
---|