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:20210402T160544Z
LOCATION:Track 6
DTSTART;TZID=America/New_York:20201112T113000
DTEND;TZID=America/New_York:20201112T121000
UID:submissions.supercomputing.org_SC20_sess212_ws_llvmf101@linklings.com
SUMMARY:Static Neural Compiler Optimization via Deep Reinforcement Learnin
 g
DESCRIPTION:Workshop\n\nStatic Neural Compiler Optimization via Deep Reinf
 orcement Learning\n\nMammadli, Jannesari, Wolf\n\nThe phase-ordering probl
 em of modern compilers has received a lot of attention from the research c
 ommunity over the years, yet remains largely unsolved. Various optimizatio
 n sequences exposed to the user are manually designed by compiler develope
 rs. In designing such a sequence developers have to choose the set of opti
 mization passes, their parameters and ordering within a sequence. Resultin
 g sequences usually fall short of achieving optimal runtime for a given so
 urce code and may sometimes even degrade the performance when compared to 
 unoptimized version. In this paper, we employ a deep reinforcement learnin
 g approach to the phase-ordering problem. Provided with sub-sequences cons
 tituting LLVM's O3 sequence, our agent learns to outperform the O3 sequenc
 e on the set of source codes used for training and achieves competitive pe
 rformance on the validation set, gaining up to 1.32x speedup on previously
 -unseen programs. Notably, our approach differs from autotuning methods by
  not depending on one or more test runs of the program for making successf
 ul optimization decisions. It has no dependence on any dynamic feature, bu
 t only on the statically-attainable intermediate representation of the sou
 rce code. We believe that the models trained using our approach can be int
 egrated into modern compilers as neural optimization agents, at first to c
 omplement, and eventually replace the hand-crafted optimization sequences.
 \n\nRegistration Category: Workshop Reg Pass
END:VEVENT
END:VCALENDAR