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:20250626T233526Z
LOCATION:B302-B305
DTSTART;TZID=America/New_York:20241120T100000
DTEND;TZID=America/New_York:20241120T170000
UID:submissions.supercomputing.org_SC24_sess533_post165@linklings.com
SUMMARY:Improving Polyhedral-Based Optimizations with Dynamic Coordinate D
 escent
DESCRIPTION:Gaurav Verma (Stony Brook University; Stony Brook University, 
 Institute for Advanced Computational Science (IACS)); Michael Canesche (Un
 iversidade Federal de Minas Gerais, Brazil); Barbara Chapman (Stony Brook 
 University; Stony Brook University, Institute for Advanced Computational S
 cience (IACS)); and Fernando Magno Quintão Pereira (Universidade Federal d
 e Minas Gerais, Brazil)\n\nPolyhedral optimizations have been a cornerston
 e of kernel optimization for many years. These techniques use a geometric 
 model of loop iterations to enable transformations like tiling, fusion, an
 d fission. The elegance of this approach lies in its ability to produce hi
 ghly efficient code through fully static optimizations. However, modern ke
 rnel schedulers typically avoid the polyhedral model, opting instead for d
 ynamic sampling techniques, such as evolutionary searches, to generate eff
 icient code. The polyhedral model is often bypassed because, being entirel
 y static, it struggles to adapt to the fine details of hardware. In this w
 ork, we demonstrate that it is possible to overcome this limitation by com
 bining the polyhedral model with a post-optimization phase based on dynami
 c coordinate descent, which uses minimal sampling while still achieving ex
 cellent performance.\n\nRegistration Category: Tech Program Reg Pass, Exhi
 bits Reg Pass\n\nSession Chairs: Ayesha Afzal (Friedrich-Alexander Univers
 ity, Erlangen-Nuremberg; Erlangen National High Performance Computing Cent
 er); Sally Ellingson (University of Kentucky); and Alan Sussman (Universit
 y of Maryland)\n\n
END:VEVENT
END:VCALENDAR