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:20260202T201806Z
LOCATION:263-264
DTSTART;TZID=America/Chicago:20251118T163700
DTEND;TZID=America/Chicago:20251118T170000
UID:submissions.supercomputing.org_SC25_sess299_pap588@linklings.com
SUMMARY:Bridging the Gap Between Unstructured SpMM and Structured Sparse T
 ensor Cores
DESCRIPTION:Yukang Dong, Ziyuan Shen, Wenbin Jiang, Zhenghang Liu, Ye Xu, 
 Bingyi He, Ran Zheng, and Hai Jin (Huazhong University of Science and Tech
 nology)\n\nThe acceleration of Sparse-dense Matrix Multiplication (SpMM) u
 sing Tensor Cores (TCs) in GPUs has recently garnered significant attentio
 n. TCs are designed for block-wise matrix multiplication; however, block p
 artitioning of general unstructured sparse matrices often results in low-l
 evel density, causing a substantial waste of computational resources. Spar
 se Tensor Cores (SpTCs) can mitigate this issue by skipping 50% of zero va
 lues; however, SpTCs are limited to strict 2:4 or 1:2 structured sparsity.
  To bridge this gap, we propose MP-SpMM, a novel matching and padding appr
 oach that transforms general sparse matrices into structured sparsity, dra
 wing inspiration from the maximum matching problem in graph theory. Moreov
 er, we introduce a novel storage format and a highly optimized GPU kernel 
 that fully exploits the capabilities of SpTCs. Extensive experiments on mo
 dern GPUs demonstrate that MP-SpMM outperforms state-of-the-art SpMM libra
 ries, DTC-SpMM and RoDe, with an average speedup of 2.42x (up to 7.65x) an
 d 1.92x (up to 8.60x).\n\nTag: HPC for Machine Learning, Performance Measu
 rement, Modeling, & Tools, Programming Frameworks\n\nRecording: Livestream
 ed, Recorded\n\nRegistration Category: Technical Program Reg Pass\n\nSessi
 on Chair: Abdel-Hameed A. Badawy (New Mexico State University, Los Alamos 
 National Laboratory (LANL))\n\n
END:VEVENT
END:VCALENDAR