Recovering numerical reproducibility in hydrodynamic simulations

Philippe Langlois 1 Rafife Nheili 1 Christophe Denis 2
1 DALI - Digits, Architectures et Logiciels Informatiques
LIRMM - Laboratoire d'Informatique de Robotique et de Microélectronique de Montpellier, UPVD - Université de Perpignan Via Domitia
Abstract : HPC simulations suffer from failures of numerical reproducibility because of floating-point arithmetic peculiarities. Different computing distributions of a parallel computation may yield different numerical results. We are interested in a finite element computation of hydrodynamic simulations within the openTelemac software where parallelism is provided by domain decomposition. One main task in a finite element simulation consists in building one large linear system and to solve it. Here the building step relies on element-by-element storage mode and the solving step applies the conjugated gradient algorithm. The subdomain parallelism is merged within these steps. We study why reproducibility fails in this process and which operations have to be corrected. We detail how to use compensation techniques to compute a numerically reproducible resolution. We illustrate this approach presenting the reproducible version of hydrodynamic simulations for one test cases provided with the openTelemac software suite.
Complete list of metadatas

Cited literature [12 references]  Display  Hide  Download

https://hal-lirmm.ccsd.cnrs.fr/lirmm-01274671
Contributor : Rafife Nheili <>
Submitted on : Tuesday, February 16, 2016 - 10:07:22 AM
Last modification on : Tuesday, July 16, 2019 - 10:06:02 AM
Long-term archiving on : Tuesday, May 17, 2016 - 5:23:25 PM

File

LaNhDe_2016 (1).pdf
Files produced by the author(s)

Identifiers

Citation

Philippe Langlois, Rafife Nheili, Christophe Denis. Recovering numerical reproducibility in hydrodynamic simulations. ARITH: Computer Arithmetic, Jul 2016, Silicon Valley, Santa Clara, CA, United States. pp.63-70, ⟨10.1109/ARITH.2016.27⟩. ⟨lirmm-01274671⟩

Share

Metrics

Record views

328

Files downloads

623