EYRAUD-DUBOIS, Lionel; LAMBERT, Thomas

doi:10.1145/3225058.3225066

Métadonnées

Licence d’utilisation du document

EYRAUD-DUBOIS, Lionel
Reformulations based algorithms for Combinatorial Optimization [Realopt]
Laboratoire Bordelais de Recherche en Informatique [LaBRI]

LAMBERT, Thomas
Reformulations based algorithms for Combinatorial Optimization [Realopt]
School of Computer Science [Manchester]

Langue

Communication dans un congrès

Ce document a été publié dans

ICPP 2018 - 47th International Conference on Parallel Processing, 2018-08-13, Eugene, OR.

Résumé en anglais

We consider the problem of data allocation when performing matrix multiplication on a heterogeneous node, with multicores and GPUs. Classical (cyclic) allocations designed for homogeneous settings are not appropriate, but the advent of task-based runtime systems makes it possible to use more general allocations. Previous theoretical work has proposed square and cube partitioning algorithms aimed at minimizing data movement for matrix multiplication. We propose techniques to adapt these continuous square partitionings to allocating discrete tiles of a matrix, and strategies to adapt the static allocation at run-time. We use these techniques in an implementation of Matrix Multiplication based on the StarPU runtime system, and we show through extensive experiments that this implementation allows to consistently obtain a lower communication volume while improving slightly the execution time, compared to standard state-of-the-art dynamic strategies.< Réduire

Mots clés en anglais

Mathematics of computing → Solvers

Computing methodologies → Linear algebra algorithms

Parallel algorithms

General and reference → Performance

URI

https://oskar-bordeaux.fr/handle/20.500.12278/193293

DOI

http://dx.doi.org/10.1145/3225058.3225066

Project ANR

Solveurs pour architectures hétérogènes utilisant des supports d'exécution - ANR-13-MONU-0007

Origine

Importé de hal

Unités de recherche

Institut de Mathématiques de Bordeaux (IMB) - UMR 5251