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_post122@linklings.com
SUMMARY:P09 - Distributed Linear Algebra with (HPX) Futures
DESCRIPTION:Poster\n\nP09 - Distributed Linear Algebra with (HPX) Futures\
 n\nInvernizzi, Nikolov, Querciagrossa, Solcà\n\nThe de-facto standard libr
 ary for distributed linear algebra is ScaLAPACK, a library that has been d
 eveloped in 1995, when supercomputers were based on nodes which had a sing
 le CPU core. Since then, the node architecture has evolved; nowadays, supe
 rcomputers are built upon multi-socked nodes, multi-core CPUs, and acceler
 ators. The parallelism approach of the ScaLAPACK library does not perform 
 well on these new architectures and the corresponding algorithms have not 
 evolved to keep up with developments in hardware or modern solving strateg
 ies.  We chose a task-based approach aiming to reduce the amount of s
 ynchronization necessary between distinct routines. We introduced fine-gra
 ined task dependencies, which allow a second algorithm to start even if th
 e previous calculation is not completely finished. Among different task-ba
 sed runtimes, our choice naturally fell on the HPX library, since it is a 
 C++ library for Concurrency and Parallelism, which implements all of the c
 orresponding facilities as defined by the C++ Standard and functionalities
  proposed as part of the ongoing C++ standardization process. We started i
 mplementing the Cholesky decoposition, which shows very good performance r
 esults compared to state-of-the-art solvers. Inspired by an inner-product 
 kernel found in electronic structure codes such as SIRIUS, we also impleme
 nted a tall-skinny matrix-matrix multiplication.
END:VEVENT
END:VCALENDAR