Optimization strategies for implementing elastic wave propagators in Intel Xeon phi coprocessors

PreviousNext You have accessInternational Conference and Exhibition, Barcelona, Spain, 3-6 April 2016Optimization strategies for implementing elastic wave propagators in Intel Xeon phi coprocessorsAuthors: Albert FarrésMauricio HanzichDiego CaballeroSantiago FernándezAlejandro DuránAlbert FarrésBarcelona Supercomputing CenterSearch for more papers by this author, Mauricio HanzichBarcelona Supercomputing CenterSearch for more papers by this author, Diego CaballeroBarcelona Supercomputing CenterSearch for more papers by this author, Santiago FernándezRepsolSearch for more papers by this author, and Alejandro DuránIntel Corporation IberiaSearch for more papers by this authorhttps://doi.org/10.1190/ice2016-6520801.1 SectionsAbout ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked InRedditEmail Abstract Acoustic wave propagation has been the preferred engine for geophysicalexploration applications for the last few years due to the large cost involvedin using better approximations, especially for 3D full-wave fieldmodelling-based applications. Hence, simplified approaches have been used togenerate images of the subsurface so that data processing can be finished in areasonable time. The current trend in seismic imaging aims at using an improvedphysical model, considering that the Earth is not rigid but an elastic body.This new model takes simulations closer to the real physics of the problem, atthe cost of raising the needed computational resources. Moreover, to take thesimulation representation a step closer to the real physics, some kind ofanisotropy in the propagation medium should be considered. However, this againmay rice the computational cost of the simulation. On the hardware front, recently developed high-performing devices, calledaccelerators or co- processors, have shown that can outperform their generalpurpose counterparts by orders of magnitude in terms of performance per watt.These new alternatives may then provide the necessary resources for makingpossible to represent complex wave physics in a reasonable time. There mightbe, however, a penalty associated to the usage of such devices, as some portionof the simulation code might need some re-writing or new optimizationstrategies explored and applied. We have designed and implemented a complete suite of wave propagators, whichincludes an elastic one for fully anisotropic scenarios using a FiniteDifferences (FD) method over a Fully Staggered Grid (FSG). In this work we willshow that regarding the computational cost and complexity of this applicationit is possible to outperform contemporary regular processors. We will exposesome optimization strategies evaluated and applied to an elastic propagatorbased on a Fully Staggered Grid, running on the Intel Xeon Phi coprocessor.The evaluated set of optimizations ranges from memory to compute optimizations.Our results show that it is possible to obtain more than an order of magnitudeof improvement when comparing the fully optimized code with a naïve version,while up to a 7x of improvement is possible with little investment on codeoptimization. It is important to remark, that the propagator is able toreproduce elastic wave propagation, even for an arbitrary anisotropy. Keywords: finite difference, 3D, parallel, programming, elasticPermalink: https://doi.org/10.1190/ice2016-6520801.1FiguresReferencesRelatedDetails International Conference and Exhibition, Barcelona, Spain, 3-6 April 2016ISSN (online):2159-6832Copyright: 2016 Pages: 356 publication data© 2016 Published in electronic format with permission by the Society of Exploration Geophysicists and the American Association of Petroleum GeologistsPublisher:Society of Exploration Geophysicists HistoryPublished Online: 03 Apr 2016 CITATION INFORMATION Albert Farrés, Mauricio Hanzich, Diego Caballero, Santiago Fernández, and Alejandro Durán, (2016), "Optimization strategies for implementing elastic wave propagators in Intel Xeon phi coprocessors," SEG Global Meeting Abstracts : 144-144. https://doi.org/10.1190/ice2016-6520801.1 Plain-Language Summary Keywordsfinite difference3DparallelprogrammingelasticLoading ...