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:20260422T000711Z
LOCATION:301-302-303
DTSTART;TZID=America/Denver:20231114T103000
DTEND;TZID=America/Denver:20231114T110000
UID:submissions.supercomputing.org_SC23_sess172_pap119@linklings.com
SUMMARY:Legate Sparse: Distributed Sparse Computing in Python
DESCRIPTION:Rohan Yadav (Stanford University); Wonchan Lee, Melih Elibol, 
 Manolis Papadakis, Taylor Lee-Patti, and Michael Garland (NVIDIA Corporati
 on); Alex Aiken and Fredrik Kjolstad (Stanford University); and Michael Ba
 uer (NVIDIA Corporation)\n\nThe sparse module of the popular SciPy Python 
 library is widely used across applications in scientific computing, data a
 nalysis, and machine learning. The standard implementation of SciPy is res
 tricted to a single CPU and cannot take advantage of modern distributed an
 d accelerated computing resources. We introduce Legate Sparse, a system th
 at transparently distributes and accelerates unmodified sparse matrix-base
 d SciPy programs across clusters of CPUs and GPUs, and composes with cuNum
 eric, a distributed NumPy library. Legate Sparse uses a combination of sta
 tic and dynamic techniques to performantly compose independently written s
 parse and dense array programming libraries, providing a unified Python in
 terface for distributed sparse and dense array computations. We show that 
 Legate Sparse is competitive with single-GPU libraries like CuPy and the i
 ndustry-standard PETSc library on up to 1280 CPU cores and 192 GPUs of the
  Summit supercomputer, while offering the productivity benefits of idiomat
 ic SciPy and NumPy.\n\nTag: Heterogeneous Computing, Programming Framework
 s and System Software, Task Parallelism\n\nRegistration Category: Tech Pro
 gram Reg Pass\n\nReproducibility Badges: Artifact Available, Artifact Func
 tional, Results Reproduced\n\nSession Chair: Claudia Fohry (University of 
 Kassel, Germany)\n\n
END:VEVENT
END:VCALENDAR