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:20210402T160102Z
LOCATION:Track 4
DTSTART;TZID=America/New_York:20201118T160000
DTEND;TZID=America/New_York:20201118T163000
UID:submissions.supercomputing.org_SC20_sess177_pap341@linklings.com
SUMMARY:GE-SpMM: General-Purpose Sparse Matrix-Matrix Multiplication on GP
 Us for Graph Neural Networks
DESCRIPTION:Paper\n\nGE-SpMM: General-Purpose Sparse Matrix-Matrix Multipl
 ication on GPUs for Graph Neural Networks\n\nHuang, Dai, Wang, Yang\n\nThe
  acceleration of Graph Neural Networks (GNNs) requires efficient and frame
 work-compatible Sparse-Dense Matrix-Matrix Multiplication (SpMM). From the
  compatibility perspective, the sophisticated sparse matrix representation
 s in state-of-the-art SpMM designs cause heavy preprocessing overhead for 
 the framework. From the efficiency perspective, optimizations for Sparse M
 atrix-Vector (SpMV) do not apply well to SpMM, leading to redundant and un
 coalesced global memory access. We propose GE-SpMM, which takes the CSR fo
 rmat consistent with GNN frameworks to enable integration without the form
 at transformation overhead. We use coalesced row caching to ensure coalesc
 ed access to both sparse and dense data in the global memory. We use coars
 e-grained warp merging to reduce redundant data loading among GPU warps. E
 xperiments on a real-world graph dataset demonstrate up to 1.41× speedup o
 ver Nvidia cuSPARSE and up to 1.81× over GraphBLAST. We embed GE-SpMM in G
 NN frameworks and get up to 3.67× speedup on popular GNN models like GCN a
 nd GraphSAGE.\n\nTag: Accelerators, FPGA, and GPUs, Linear Algebra, Machin
 e Learning, Deep Learning and Artificial Intelligence\n\nRegistration Cate
 gory: Tech Program Reg Pass
END:VEVENT
END:VCALENDAR