BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20200129T163600Z
LOCATION:501
DTSTART;TZID=America/Denver:20191118T115200
DTEND;TZID=America/Denver:20191118T121400
UID:submissions.supercomputing.org_SC19_sess117_ws_mchpc111@linklings.com
SUMMARY:Generalized Sparse Matrix-Matrix Multiplication for Vector Engines
and Graph Applications
DESCRIPTION:Workshop\n\nGeneralized Sparse Matrix-Matrix Multiplication fo
r Vector Engines and Graph Applications\n\nLi, Wang, Araki, Qiu\n\nGeneral
ized sparse matrix-matrix multiplication (SpGEMM) is a key primitive kerne
l for many high-performance graph algorithms as well as for machine learni
ng, and data analysis algorithms. Although many SpGEMM algorithms have bee
n proposed, such as ESC and SPA, there is currently no SpGEMM kernel optim
ized for vector engines (VEs). NEC SX-Aurora is the new vector computing s
ystem that can achieve high performance by leveraging high bandwidth memor
y of 1.2TB/s and long vector of VEs, where the execution of scientific app
lications is limited by memory bandwidth. In this paper, we demonstrate si
gnificant initial work of SpGEMM kernel for a vector engine and implement
it to vectorize several essential graph analysis algorithms: Butterfly cou
nting and Triangle counting. We propose a SpGEMM algorithm with a novel hy
brid method based on sparse vectors and loop raking to maximize the length
of vectorizable code for vector machine architectures. The experimental r
esults show that the vector engine has advantages on more massive data set
s. This work contributes to high performance and portability of the SpGEMM
kernel to a new family of heterogeneous computing systems, which is Vecto
r Host (VH) equipped with different accelerators or VEs.\n\nTag: Workshop
Reg Pass, HPC, Memory, OS and Runtime Systems, Runtime Systems\n\nRegistra
tion Category: Workshop Reg Pass, HPC, Memory, OS and Runtime Systems, Run
time Systems
URL:https://sc19.supercomputing.org/presentation/?id=ws_mchpc111&sess=sess
117
END:VEVENT
END:VCALENDAR