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 E 1.2
DTSTART;TZID=Europe/Stockholm:20190613T144500
DTEND;TZID=Europe/Stockholm:20190613T151500
UID:submissions.pasc-conference.org_PASC19_sess135_msa262@linklings.com
SUMMARY:Generating High Performance Tensor Operations for Earthquake Simul
 ations
DESCRIPTION:Minisymposium\nComputer Science and Applied Mathematics, Physi
 cs, Solid Earth Dynamics\n\nGenerating High Performance Tensor Operations 
 for Earthquake Simulations\n\nUphoff, Bader\n\nModern HPC systems enable e
 arthquake simulations with unprecedented accuracy and complexity, allowing
  the resolution of frequencies up to 10 Hz and dynamic simulation of the r
 upture process. But, in order to exploit the capabilities of today's super
 computers, all levels of parallelism need to be leveraged, starting from i
 nstruction level parallelism and ending at asynchronous communication and 
 I/O. So, on the one hand expert knowledge and long development times are r
 equired. On the other hand, moving to higher frequency regimes requires to
  account for advanced rheological models, such as viscoelasticity or plast
 icity, which requires a large codebase to be maintained and adapted to new
  hardware. Moreover, new hardware pushes the machine balance, such that on
 e is going to need to come up with new ideas, in order to make the most ef
 ficient use of invested energy. Here, we present a possible ansatz to deal
  with software complexity, which we have implemented in the software packa
 ge SeisSol. Our approach is based on a domain specific language for tensor
  operations, which we think to be a suitable abstraction. With our DSL, te
 nsor operations get automatically tuned and mapped to highly tuned matrix 
 multiplication routines, leading to high performance and reduced developme
 nt time.
END:VEVENT
END:VCALENDAR