BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20191206T142457Z
LOCATION:Analog 1\, 2
DTSTART;TZID=Europe/Stockholm:20180627T134500
DTEND;TZID=Europe/Stockholm:20180627T141500
UID:isc_hpc_ISC High Performance 2018_sess152_pap157@linklings.com
SUMMARY:TaskGenX: A Hardware-Software Proposal for Accelerating Task Paral
 lelism
DESCRIPTION:Research Paper\n\nTaskGenX: A Hardware-Software Proposal for A
 ccelerating Task Parallelism\n\nChronaki, Casas, Moreto, Bosch, Badia\n\nA
 s chip multi-processors (CMPs) are becoming\nmore and more complex, softwa
 re solutions such as parallel\nprogramming models are attracting a lot of 
 attention. Task-based\nparallel programming models offer an appealing appr
 oach to\nutilize complex CMPs. However, the increasing number of cores\non
  modern CMPs is pushing research towards the use of fine\ngrained parallel
 ism. Task-based programming models need to\nbe able to handle such workloa
 ds and offer performance and\nscalability. Using specialized hardware for 
 boosting performance\nof task-based programming models is a common practic
 e in the\nresearch community.\nOur paper makes the observation that task c
 reation becomes\na bottleneck when we execute fine grained parallel applic
 ations\nwith task-based programming models. As the number of cores\nincrea
 ses the time spent generating the tasks of the application\nis becoming mo
 re critical to the entire execution. To overcome\nthis issue, we propose T
 askGenX. TaskGenX offers a solution\nfor minimizing task creation overhead
 s and relies both on the\nruntime system and a dedicated hardware. On the 
 runtime\nsystem side, TaskGenX decouples the task creation from the\nother
  runtime activities. It then transfers this part of the\nruntime to a spec
 ialized hardware. We draw the requirements\nfor this hardware in order to 
 boost execution of highly parallel\napplications. From our evaluation usin
 g 11 parallel workloads on\nboth symmetric and asymmetric systems, we obta
 in performance\nimprovements up to 15, averaging to 3.1 over the baselin
 e.\n\nPasses: Conference Pass, Heterogeneous Systems, Scheduling, System S
 oftware & Runtime Systems\n\nTag: Conference Pass, Heterogeneous Systems, 
 Scheduling, System Software & Runtime Systems
URL:https://2018.isc-program.com/presentation/?id=pap157&sess=sess152
END:VEVENT
END:VCALENDAR