BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Intelligent Thread-Level Speculation - Jeremy Singer (School of Co
 mputer Science\, University of Manchester)
DTSTART:20090717T141500Z
DTEND:20090717T151500Z
UID:TALK19052@talks.cam.ac.uk
CONTACT:Dominic Orchard
DESCRIPTION:Thread-level speculation (TLS) is a mechanism for improving th
 e\nexecution of sequential programs on multi-core processors. TLS relies\n
 on the insertion of potential thread-spawn points into sequential\nprogram
 s\, either statically or dynamically. Spawned threads generate\nparallelis
 m\, but they are only effective if (a) the parallel execution\nsaving outw
 eighs the thread management overhead\, and (b) there are few\ndata depende
 nce vioations with other concurrent threads.\n\nUntil now\, spawn points h
 ave been identified via (a) exhaustive\nprofiling of programs\, and/or (b)
  heuristics devised by human\nexperts. We have recently begun a project th
 at uses machine learning\ntechniques to generate spawning heuristics autom
 atically\, with less\nprogram profiling. This presentation reports on the 
 use of Java\nprograms' static attributes and dynamic behaviour as features
  for\nlearning effective TLS spawning policies.
LOCATION:GS15\, Computer Laboratory
END:VEVENT
END:VCALENDAR