Task-Based Parallelization of the Fast Multipole Method on NVIDIA GPUs and Multicore Processors

Emmanuel Agullo (1) , Bérenger Bramas (1) , Olivier Coulaud (1, 2) , Eric Darve (3, 4) , Matthias Messner (1) , Toru Takahashi (5)

1 HiePACS - High-End Parallel Algorithms for Challenging Numerical Simulations
2 LaBRI - Laboratoire Bordelais de Recherche en Informatique
3 Department of Mechanical Engineering [Stanford]
4 ICME - Institute for Computational and Mathematical Engineering [Stanford]
5 Department of Mechanical Science and Engineering

Fast Multipole Methods are a fundamental operation for the simulation of many physical problems. In this talk, we present a new approach for implementing these methods that achieves high performance across many different computer architectures. Our method consists of expressing the FMM algorithm as a task flow and employing a state-of-the-art runtime system, StarPU, to process the tasks on the different computing units.

Mots clés: Runtime; GPU; HPC; FMM

Domaines: Calcul parallèle, distribué et partagé [cs.DC]

Emmanuel Agullo, Bérenger Bramas, Olivier Coulaud, Eric Darve, Matthias Messner, et al.. Task-Based Parallelization of the Fast Multipole Method on NVIDIA GPUs and Multicore Processors. SIAM Conference on Computational Science and Engineering (SIAM CSE 2015), Mar 2015, Salt Lake City, United States. ⟨hal-01135258⟩