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:20211207T055347Z
LOCATION:Online
DTSTART;TZID=America/Chicago:20211115T153000
DTEND;TZID=America/Chicago:20211115T160000
UID:submissions.supercomputing.org_SC21_sess422_ws_pmbsf101@linklings.com
SUMMARY:Customized Monte Carlo Tree Search for LLVM/Polly's Composable Loo
 p Optimization Transformations
DESCRIPTION:Workshop\n\nCustomized Monte Carlo Tree Search for LLVM/Polly'
 s Composable Loop Optimization Transformations\n\nKoo, Balaprakash, Kruse,
  Wu, Hovland...\n\nPolly is the LLVM project's polyhedral loop optimizer. 
 Recent user-directed loop transformation pragmas were proposed based on LL
 VM/Clang and Polly. The search space exposed by the transformation pragmas
  is a tree, wherein each node represents a specific combination of loop tr
 ansformations. To find the best combination of these loop transformations,
  we developed a search algorithm based on Monte Carlo tree search (MCTS). 
 The algorithm consists of two phases: exploring loop transformations at di
 fferent depths of the tree to identify promising regions and exploiting th
 ose regions.  A restart mechanism is used to avoid the MCTS trapped in a l
 ocal solution. The best and worst solutions are transferred from the previ
 ous restarts to leverage the search history. We compare our approach with 
 breadth-first, beam, global greedy and random search methods using PolyBen
 ch benchmarks and ECP proxy applications. Our MCTS algorithm achieves a sp
 eedup of 2.3x over Polly's heuristic optimizations on average.\n\nTag: Onl
 ine Only, Accelerator-based Architectures, Applications, Computational Sci
 ence, Emerging Technologies, Extreme Scale Computing, File Systems and I/O
 , Heterogeneous Systems, Parallel Programming Languages and Models, Perfor
 mance, Scientific Computing, Software Engineering\n\nRegistration Category
 : Workshop Reg Pass
END:VEVENT
END:VCALENDAR