Superloop Scheduling: Loop Optimization via Direct Statement Instance Reordering - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Communication Dans Un Congrès Année : 2023

Superloop Scheduling: Loop Optimization via Direct Statement Instance Reordering

Résumé

Loop optimization in the polyhedral model is supported by the expressiveness of affine scheduling functions to model statement iteration ordering. Discovering the best scheduling remains a grand challenge reinforced by the need to build affine functions. Automatic techniques based on solving systems of affine constraints and/or composing affine scheduling primitives are limited either by the affine modeling or by their primitive set. In this paper we propose a radically different approach to affine scheduling construction called superloop scheduling. We leverage basic-block-level instruction reordering techniques and the capacity to agglomerate statement instances into "super" loops offered by modern trace analysis tools. This approach enables deepest possible reasoning about instruction ordering and a global approach to loop optimization, e.g., where tiling and intra-tile optimization is considered along with other transformations.
Fichier principal
Vignette du fichier
paper5.pdf (373.33 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04393522 , version 1 (14-01-2024)

Licence

Paternité

Identifiants

  • HAL Id : hal-04393522 , version 1

Citer

Cédric Bastoul, Alain Ketterlin, Vincent Loechner. Superloop Scheduling: Loop Optimization via Direct Statement Instance Reordering. IMPACT 2023, Jan 2023, Toulouse, France. ⟨hal-04393522⟩
19 Consultations
2 Téléchargements

Partager

Gmail Facebook X LinkedIn More