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 D 3.2
DTSTART;TZID=Europe/Stockholm:20190613T181500
DTEND;TZID=Europe/Stockholm:20190613T184500
UID:submissions.pasc-conference.org_PASC19_sess152_msa132@linklings.com
SUMMARY:MPI+CUDA/OpenACC in Plasma Simulation Codes
DESCRIPTION:Minisymposium\nComputer Science and Applied Mathematics, Emerg
 ing Application Domains, Climate and Weather, Physics, Life Sciences, Engi
 neering\n\nMPI+CUDA/OpenACC in Plasma Simulation Codes\n\nAsahi, Yasushiro
 \n\nThe emergence of GPU computing has been changing the high-performance 
 computing in science drastically. Indeed, GPUs are considered to be key in
 gredients to satisfy enormous computational demands with reasonable energy
  consumption. Unfortunately, it is not easy to extract their high computat
 ional power, when porting massively parallelized legacy simulation codes t
 o GPU platforms.As a case study, we have ported massively parallelized pla
 sma turbulence simulation codes GKV [Watanabe, NF, 2006] and GT5D [Idomura
 , CPC, 2008] to GPU platforms. GKV (resp. GT5D) code was ported with CUDA 
 (resp. OpenACC). The characteristics of these plasma simulation codes may 
 be their high-dimensionality (3D space and 2D velocity space), which is di
 fferent from usual 3D fluid simulations.In the talk, we discuss the advant
 ages and disadvantages of CUDA and OpenACC approaches through our porting 
 experience. Focusing on the MPI communications, we present the pipeline-ba
 sed communication and computation overlapping technique, which can be appl
 icable as long as both the computations and communications along a specifi
 c direction are independent. This can be said as an effective usage of hig
 h dimensional feature. The overlapping is applied not only to the MPI-comm
 unications and computations, but also to the data transfer between CPUs an
 d GPUs.
END:VEVENT
END:VCALENDAR