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:20210916T132446Z
LOCATION:Mère Royaume
DTSTART;TZID=Europe/Stockholm:20210705T140000
DTEND;TZID=Europe/Stockholm:20210705T143000
UID:submissions.pasc-conference.org_PASC21_sess213_pap_dec120@linklings.co
 m
SUMMARY:Eventify: Event-Based Task Parallelism for Strong Scaling
DESCRIPTION:Paper\n\nEventify: Event-Based Task Parallelism for Strong Sca
 ling\n\nHaensel, Morgenstern, Beckmann, Kabadshow, Dachsel\n\nToday's proc
 essors become fatter, not faster. However, the exploitation of these massi
 vely parallel compute resources remains a challenge for many traditional H
 PC applications regarding scalability, portability and programmability. To
  tackle this challenge, several parallel programming approaches such as lo
 op parallelism and task parallelism are researched in form of languages, l
 ibraries and frameworks. Task parallelism as provided by OpenMP, HPX, Star
 PU, Charm++ and Kokkos is the most promising approach to overcome the chal
 lenges of ever increasing parallelism.<br />The aforementioned parallel pr
 ogramming technologies enable scalability for a broad range of algorithms 
 with coarse-grained tasks, e.g. in linear algebra and classical N-body sim
 ulation. However, they do not fully address the performance bottlenecks of
  algorithms with fine-grained tasks and the resultant large task graphs. A
 dditionally, we experienced the description of large task graphs to be cum
 bersome with the common approach of providing in-, out- and inout-dependen
 cies. We introduce event-based task parallelism to solve the performance a
 nd programmability issues for algorithms that exhibit fine-grained task pa
 rallelism and contain repetitive task patterns. With user-defined event li
 sts, the approach provides a more convenient and compact way to describe l
 arge task graphs. Furthermore, we show how these event lists are processed
  by a task engine that reuses user-defined, algorithmic data structures. A
 s use case, we describe the implementation of a fast multipole method for 
 molecular dynamics with event-based task parallelism.<br />The performance
  analysis reveals that the event-based implementation is 52 % faster than 
 a loop-parallel implementation with OpenMP.\n\nDomain: CS and Math
END:VEVENT
END:VCALENDAR