Co-optimizing Dataflow Graphs and Actors with MLIR - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Co-optimizing Dataflow Graphs and Actors with MLIR

Résumé

Dataflow programming is considered a good solution for the implementation of parallel signal processing applications. However, the strict separation between kernel and coordination codes limits the variety of possible optimizations and the compatibility with state-of-the-art compiler frameworks. We present a prototype static dataflow compiler, built with the LLVM MLIR framework, that overcomes these limitations and enables a previously impossible combination of optimization strategies that leverages information from the dataflow topology. Initial results show 30% wall time improvement and 53% memory usage improvement on a video processing workload.
Fichier principal
Vignette du fichier
dfmlir___SiPS22.pdf (281.64 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03845902 , version 1 (09-11-2022)

Identifiants

  • HAL Id : hal-03845902 , version 1

Citer

Pedro Ciambra, Mickaël Dardaillon, Maxime Pelcat, Hervé Yviquel. Co-optimizing Dataflow Graphs and Actors with MLIR. 2022 IEEE Workshop on Signal Processing Systems (SiPS), Nov 2022, Rennes, France. ⟨hal-03845902⟩
39 Consultations
351 Téléchargements

Partager

Gmail Facebook X LinkedIn More