BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/New_York
X-LIC-LOCATION:America/New_York
BEGIN:DAYLIGHT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
TZNAME:EDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
TZNAME:EST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250626T234543Z
LOCATION:B311
DTSTART;TZID=America/New_York:20241120T110000
DTEND;TZID=America/New_York:20241120T113000
UID:submissions.supercomputing.org_SC24_sess369_pap294@linklings.com
SUMMARY:A Sparsity-aware Distributed-memory Algorithm for Sparse-sparse Ma
 trix Multiplication
DESCRIPTION:Yuxi Hong and Aydin Buluc (Lawrence Berkeley National Laborato
 ry (LBNL))\n\nMultiplying two sparse matrices (SpGEMM) is a common computa
 tional primitive used in many areas including graph algorithms, bioinforma
 tics, algebraic multigrid solvers, and randomized sketching. Distributed-m
 emory parallel algorithms for SpGEMM have mainly focused on sparsity-obliv
 ious approaches that use 2D and 3D partitioning. Sparsity-aware 1D algorit
 hms can theoretically reduce communication by not fetching nonzero of the 
 sparse matrices that do not participate in the multiplication. \nHere, we 
 present a distributed-memory 1D SpGEMM algorithm and implementation.  It u
 ses MPI RDMA operations to mitigate the cost of packing/unpacking submatri
 ces for communication, and it uses a block fetching strategy to avoid exce
 ssive fine-grained messaging. Our results show that our 1D implementation 
 outperforms state-of-the-art 2D and 3D implementations within CombBLAS for
  many configurations, inputs, and use cases, while remaining conceptually 
 simpler.\n\nTag: Algorithms, Artificial Intelligence/Machine Learning, Gra
 ph Algorithms, Linear Algebra\n\nRegistration Category: Tech Program Reg P
 ass\n\nSession Chair: Jiajia Li (North Carolina State University)\n\n
END:VEVENT
END:VCALENDAR