BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20230124T171526Z
LOCATION:C141-143-149
DTSTART;TZID=America/Chicago:20221115T140000
DTEND;TZID=America/Chicago:20221115T143000
UID:submissions.supercomputing.org_SC22_sess155_pap287@linklings.com
SUMMARY:Addressing Irregular Patterns of Matrix Computations on GPUs and T
heir Impact on Applications Powered by Sparse Direct Solvers
DESCRIPTION:Paper\n\nAddressing Irregular Patterns of Matrix Computations
on GPUs and Their Impact on Applications Powered by Sparse Direct Solvers\
n\nAbdelfattah, Ghysels, Boukaram, Tomov, Li...\n\nSeveral scientific appl
ications rely on sparse direct solvers for their numerical robustness. How
ever, performance optimization for these solvers remains a challenging tas
k, especially on GPUs. This is due to workloads of small dense matrices th
at are different in size. Matrix decompositions on such irregular workload
s are rarely addressed on GPUs.\n\nThis paper addresses irregular workload
s of matrix computations on GPUs and shows their impact on a sparse LU sol
ver. We designed an interface for the basic matrix operations supporting p
roblems of different sizes. The interface enables us to develop irrLU-GPU,
an LU decomposition on matrices of different sizes. We demonstrate the im
pact of irrLU-GPU on sparse LU solvers using NVIDIA and AMD GPUs. Experime
ntal results are shown for a sparse direct solver based on multifrontal sp
arse LU decomposition applied to linear systems arising from the simulatio
n, using finite element discretization on unstructured meshes, of a high f
requency indefinite Maxwell problem.\n\nSession Format: Recorded\n\nTag: A
pplications, Numerical Algorithms, Security\n\nRegistration Category: Tech
Program Reg Pass\n\nReproducibility Badges: Artifact Available, Artifact
Functional, Results Reproduced
END:VEVENT
END:VCALENDAR