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:20260202T201228Z
LOCATION:Second Floor Atrium
DTSTART;TZID=America/Chicago:20251118T080000
DTEND;TZID=America/Chicago:20251118T170000
UID:submissions.supercomputing.org_SC25_sess537_drs114@linklings.com
SUMMARY:Accelerating Sparse Tensor Contractions
DESCRIPTION:Saurabh Raje (University of Utah)\n\nSparse tensor contraction
 s (SpTC) are a bottleneck for several algorithms in scientific computing, 
 data science, artificial intelligence and graphics. The SpTC operation is 
 any expression of the form R(l0,l1, r0) = X(l0, l1, c0) * Y(r0, c0) where 
 two tensors are multiplied along several dimensions to form a multidimensi
 onal result. Sparse tensor networks are an extension of this problem such 
 that there are more than two inputs. This thesis aims to accelerate both—t
 he SpTC primitive and sparse tensor networks with multiple SpTC terms. We 
 develop kernels and IR optimizations to improve code-generation for sparse
  tensor networks.\n\nTo generate efficient code for a sparse tensor networ
 k, several inter-dependent optimizations must be made on the intermediate 
 representation (IR). This includes sparse tensor mode order, and loop fusi
 on to reduce intermediate tensors. Correctness requirements impose constra
 ints on these variables.\n\nWe develop CoNST, a code-generator that co-opt
 imizes these variables. An integer constraint system is solved by the Z3 S
 MT solver and the result lowers to a unique fused loop structure and tenso
 r mode layouts for the entire contraction tree. CoNST outperforms state-of
 -the-art compilers by orders of magnitude in run-time.\n\nTo accelerate th
 e SpTC operation, we perform the first analysis of data-access costs and m
 emory requirements for loop orders. We develop FaSTCC, a hash-based parall
 el implementation of the SpTC operation that uses the fastest loop order w
 ith minimal memory overhead. FaSTCC introduces a new 2D tiled contraction-
 index-outer scheme and a corresponding tile-aware design. It outperforms p
 revious state-of-the-art by 2-5x on up to 64 CPU threads.\n\nTag: Research
  & ACM SRC Posters\n\nRecording: Not Livestreamed, Not Recorded\n\nRegistr
 ation Category: Technical Program Reg Pass\n\nSession Chairs: Kento Sato (
 RIKEN Center for Computational Science (R-CCS)); Chris Schlipalius (Pawsey
  Supercomputing Research Centre; Commonwealth Scientific and Industrial Re
 search Organisation (CSIRO), Australia); and Anja Gerbes (Georg-August-Uni
 versität Göttingen)\n\n
END:VEVENT
END:VCALENDAR