High performance pseudo-analytical simulation of multi-object adaptive optics over multi-GPU systems

Ahmad Abdelfattah, Eric Gendron, Damien Gratadour, David Keyes, Hatem Ltaief, Arnaud Sevin, Fabrice Vidal

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Multi-object adaptive optics (MOAO) is a novel adaptive optics (AO) technique dedicated to the special case of wide-field multiobject spectrographs (MOS). It applies dedicated wavefront corrections to numerous independent tiny patches spread over a large field of view (FOV). The control of each deformable mirror (DM) is done individually using a tomographic reconstruction of the phase based on measurements from a number of wavefront sensors (WFS) pointing at natural and artificial guide stars in the field. The output of this study helps the design of a new instrument called MOSAIC, a multi-object spectrograph proposed for the European Extremely Large Telescope (E-ELT)1. We have developed a novel hybrid pseudo-analytical simulation scheme that allows us to accurately simulate in detail the tomographic problem. The main challenge resides in the computation of the tomographic reconstructor, which involves pseudo-inversion of a large dense symmetric matrix. The pseudo-inverse is computed using an eigenvalue decomposition, based on the divide and conquer algorithm, on multicore systems with multi-GPUs. Thanks to a new symmetric matrix-vector product (SYMV) multi-GPU kernel, our overall implementation scores significant speedups over standard numerical libraries on multicore, like Intel MKL, and up to 60% speedups over the standard MAGMA implementation on 8 Kepler K20c GPUs. At 40,000 unknowns, this appears to be the largest-scale tomographic AO matrix solver submitted to computation, to date, to our knowledge and opens new research directions for extreme scale AO simulations.

Original languageEnglish (US)
Title of host publicationEuro-Par 2014
Subtitle of host publicationParallel Processing - 20th International Conference, Proceedings
PublisherSpringer Verlag
Pages704-715
Number of pages12
Volume8632
ISBN (Electronic)9783319098722
ISBN (Print)9783319098722
StatePublished - 2014
Event20th International Conference on Parallel Processing, Euro-Par 2014 - Porto, Portugal
Duration: Aug 25 2014Aug 29 2014

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume8632
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other20th International Conference on Parallel Processing, Euro-Par 2014
CountryPortugal
CityPorto
Period08/25/1408/29/14

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint Dive into the research topics of 'High performance pseudo-analytical simulation of multi-object adaptive optics over multi-GPU systems'. Together they form a unique fingerprint.

Cite this