Optimization of a discontinuous finite element solver with OpenCL and StarPU

Bérenger Bramas 1 Philippe Helluy 2, 3 Laura Mendoza 2, 3 Bruno Weber 4
1 CAMUS - Compilation pour les Architectures MUlti-coeurS
Inria Nancy - Grand Est, ICube - Laboratoire des sciences de l'ingénieur, de l'informatique et de l'imagerie
3 TONUS - TOkamaks and NUmerical Simulations
IRMA - Institut de Recherche Mathématique Avancée, Inria Nancy - Grand Est
Abstract : schnaps is a finite element solver designed to simulate various physical phenomena. It is designed to run on hybrid computers made of several CPUs and GPUs. In order to address the hybrid architectures we rely on the StarPU runtime. StarPU allows to optimize in an incre- mental way a sequential algorithm in order to migrate to multicore parallelism and then to hybrid computing with OpenCL accelerators. StarPU proposes several task scheduling strategies in or- der to efficiently exploit the available computing power. We present the design of schnaps and the performance gain that we have obtained in electromagnetic simulations thanks to OpenCL codelets.
Document type :
Preprints, Working Papers, ...
Complete list of metadatas

Cited literature [13 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01942863
Contributor : Philippe Helluy <>
Submitted on : Monday, July 8, 2019 - 11:38:14 PM
Last modification on : Wednesday, July 10, 2019 - 1:27:23 AM

File

schnaps_opencl_helluy.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01942863, version 2

Citation

Bérenger Bramas, Philippe Helluy, Laura Mendoza, Bruno Weber. Optimization of a discontinuous finite element solver with OpenCL and StarPU. 2019. ⟨hal-01942863v2⟩

Share

Metrics

Record views

44

Files downloads

34