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:20190614T090000
DTEND;TZID=Europe/Stockholm:20190614T093000
UID:submissions.pasc-conference.org_PASC19_sess182_pap_jan128@linklings.co
m
SUMMARY:Porting the COSMO Weather Model to Manycore CPUs
DESCRIPTION:Paper\nClimate and Weather\n\nPorting the COSMO Weather Model
to Manycore CPUs\n\nThaler, Moosbrugger, Osuna, Bianco, Vogt...\n\nWeather
and climate simulations are a major application driver in high-performanc
e computing (HPC). With the end of Dennard scaling and Moore's law, the HP
C industry increasingly employs specialized computation accelerators to in
crease computational throughput. Manycore architectures, such as Intel's K
nights Landing (KNL), are a representative example of future processing de
vices. However, software has to be modified to use these devices efficient
ly. In this work, we demonstrate how an existing domain-specific language
that has been designed for CPUs and GPUs can be extended to Manycore archi
tectures such as KNL. We achieve comparable performance to the NVIDIA Tesl
a P100 GPU architecture on hand-tuned representative stencils of the dynam
ical core of the COSMO weather model and its radiation code. Further, we p
resent performance within a factor of two of the P100 of the full DSL-base
d GPU-optimized COSMO dycore code. We find that optimizing code to full pe
rformance on modern manycore architectures requires similar effort and har
dware knowledge as for GPUs. Further, we show limitations of the present a
pproaches, and outline our lessons learned and possible principles for des
ign of future DSLs for accelerators in the weather and climate domain.
Full paper: https://doi.org/10.1145/3324989.3325723
END:VEVENT
END:VCALENDAR