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:20210916T132449Z
LOCATION:Lise Girardin
DTSTART;TZID=Europe/Stockholm:20210706T120000
DTEND;TZID=Europe/Stockholm:20210706T123000
UID:submissions.pasc-conference.org_PASC21_sess124_msa181@linklings.com
SUMMARY:Development of Exascale full-f Gyrokinetic Simulation on Summit an
 d FUGAKU
DESCRIPTION:Minisymposium\n\nDevelopment of Exascale full-f Gyrokinetic Si
 mulation on Summit and FUGAKU\n\nIdomura\n\nThe Gyrokinetic Toroidal 5D fu
 ll-f Eulerian code GT5D [Idomura, CPC2008] is based on a semi-implicit fin
 ite difference scheme, in which a stiff linear 4D convection operator is s
 ubject to implicit time integration, and the implicit finite difference so
 lver for fast kinetic electrons occupies more than 80% of the total comput
 ing cost. The implicit solver was originally developed using a Krylov subs
 pace method (GCR), in which global collective communications and halo data
  communications were becoming bottlenecks on the latest accelerator based 
 platforms. This issue was partly resolved by introducing a communication-a
 voiding Krylov subspace method (CA-GMRES) [Idomura, ScalA’17]. Howev
 er, the remaining halo data communications in SpMV still occupy significan
 t costs. To resolve this issue, the number of SpMVs and thus, halo data co
 mmunications is reduced by improving the convergence property using a new 
 FP16 preconditioner [Idomura, SC20]. The CA-GMRES solver with the FP16 pre
 conditioner is designed for the smooth linear operator by fully utilizing 
 the new support for FP16 SIMD operations on FUGAKU (A64FX), and achieved a
 n order of magnitude smaller number of iterations, leading to significant 
 speedup compared to the GCR and CA-GMRES solvers without preconditioning. 
 We discuss the performance portability of GT5D with the new implicit solve
 r on FUGAKU and SUMMIT.\n\nDomain: CS and Math, Physics
END:VEVENT
END:VCALENDAR