BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20190719T085744Z
LOCATION:HG EO Nord
DTSTART;TZID=Europe/Stockholm:20190613T195000
DTEND;TZID=Europe/Stockholm:20190613T215000
UID:submissions.pasc-conference.org_PASC19_sess179_post108@linklings.com
SUMMARY:CSM06 - Auto-Tuning for Short-Range Particle Simulation
DESCRIPTION:Poster\n\n\nCSM06 - Auto-Tuning for Short-Range Particle Simul
 ation\n\nGratl, Seckler, Tchipev, Bungartz, Neumann\n\nIn N-Body simulatio
 ns the time-to-solution can change drastically depending on simulation par
 ameters and the configuration of the scenario. Different algorithms or par
 allelization patterns can be applied to improve the performance for a spec
 ific scenario. However, no combination is optimal for every situation. Hig
 hly impactful characteristics like the particle distribution or local dens
 ities can change significantly over the course of simulations. This means 
 that the optimal algorithm combination can change anytime during the simul
 ation. To tackle this problem of specialization we created the C++ library
  AutoPas. It is designed to act as a base layer to create arbitrary N-Body
  simulations while obtaining optimal node-level performance through auto-t
 uning. The library determines the best performing combination of algorithm
 s for finding particle pairs, data structures, and OpenMP parallelization 
 patterns. Should the optimum change, the library can alter the combination
  at runtime. We integrated AutoPas into the software ls1 mardyn, a molecul
 ar dynamics simulator for large numbers of small rigid molecules. This pos
 ter showcases the concepts behind AutoPas and how to integrate it into exi
 sting code bases. Furthermore, we show run time results for different expe
 riments in which we achieve and sustain optimal node-level performance by 
 this auto-tuning approach.
END:VEVENT
END:VCALENDAR