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:20210916T132456Z
LOCATION:
DTSTART;TZID=Europe/Stockholm:20210706T173000
DTEND;TZID=Europe/Stockholm:20210706T190000
UID:submissions.pasc-conference.org_PASC21_sess182_post131@linklings.com
SUMMARY:P16 - MARS: Mesh Adaptive Refinement for Supercomputing
DESCRIPTION:Poster\n\nP16 - MARS: Mesh Adaptive Refinement for Supercomput
 ing\n\nGanellari, Zulian, Fink, Fadel, Cumming...\n\nMARS is an open-sourc
 e mesh management library designed to handle N-dimensional elements (N <
 ;= 4). MARS is developed in C++ and makes use of template meta-progra
 mming to have compile time dimensions of elements and vectors, thus allowi
 ng for both compile time performance optimizations and concise and reusabl
 e code.<br /><br />The main features of MARS consist of:<br /><br />1.mesh
  generation<br />2.adaptive mesh refinement using bisection algorithms<br 
 />3.conforming mesh data-structure<br />4.mesh quality estimators to study
  the output of different mesh-refinement strategies<br />5.performance por
 table algorithms and data-structures targetting different accelerators.<br
  /><br />MARS targets multi-core CPUs and GPUs using the C++ Kokkos progra
 mming model. The mesh is entirely constructed and stored on the devic
 e (GPUs). This enables libraries using MARS to perform further o
 perations directly on the device, avoiding going through the host. Current
 ly, MARS supports as its performance portable, parallel, adaptive refineme
 nt based algorithm the LEPP (Longest edge propagation path) from Rivara. M
 esh generation is fully supported in parallel. As an example, 143 million 
 Hex8 elements can be generated on a single node in just 0.86 sec.<br /><br
  />A distributed memory, parallel implementation based on MPI is ongoing w
 ork, and forest of octrees and space filling curves algorithms for efficie
 nt mesh partitioning are being planned.
END:VEVENT
END:VCALENDAR