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:20210916T132452Z
LOCATION:Ernesto Bertarelli
DTSTART;TZID=Europe/Stockholm:20210707T150000
DTEND;TZID=Europe/Stockholm:20210707T153000
UID:submissions.pasc-conference.org_PASC21_sess144_msa297@linklings.com
SUMMARY:Automatic Differentiation of C++ Codes on Emerging Manycore Archit
ectures with Sacado
DESCRIPTION:Minisymposium\n\nAutomatic Differentiation of C++ Codes on Eme
rging Manycore Architectures with Sacado\n\nPhipps, Pawlowski, Trott\n\nAu
tomatic differentiation (AD) is a well-known technique for evaluating anal
ytic derivatives of calculations implemented on a computer, with numerous
software tools available for incorporating AD technology into complex appl
ications. However, a growing challenge for AD is the efficient differenti
ation of parallel computations implemented on emerging manycore computing
architectures such as multicore CPUs, GPUs, and accelerators as these devi
ces become more pervasive. In this work, we explore forward-mode, operato
r overloading-based differentiation of C++ codes on these architectures us
ing the widely available Sacado AD software package. In particular, we le
verage Kokkos, a C++ tool providing APIs for implementing parallel computa
tions that is portable to a wide variety of emerging architectures. We de
scribe the challenges that arise when differentiating code for these archi
tectures using Kokkos, and two approaches for overcoming them that ensure
optimal memory access patterns as well as expose additional dimensions of
fine-grained parallelism in the derivative calculation. We describe the r
esults of several computational experiments that demonstrate the performan
ce of the approach on several contemporary CPU and GPU architectures. We
then conclude with applications of these techniques to the simulation of d
iscretized systems of partial differential equations.\n\nDomain: CS and Ma
th, Emerging Applications, Physics, Engineering
END:VEVENT
END:VCALENDAR