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:20210916T132448Z
LOCATION:Mère Royaume
DTSTART;TZID=Europe/Stockholm:20210706T110000
DTEND;TZID=Europe/Stockholm:20210706T113000
UID:submissions.pasc-conference.org_PASC21_sess147_msa325@linklings.com
SUMMARY:Futhark: Painless HPC on GPUs
DESCRIPTION:Minisymposium\n\nFuthark: Painless HPC on GPUs\n\nMalaspinas, 
 El-Kharroubi\n\nThe advent of new HPC architectures such as GPUs often cha
 llenges the long known CPU-based best practices. They usually offer a grea
 t improvement in performance at the cost of requiring a deep understanding
  of the hardware from the developer who needs to perform very precise opti
 mizations. In the same spirit that many man-made optimizations and fine tu
 ning for specific CPU architectures where beaten by having compilers produ
 cing more efficient code. The development of more advanced compilers for G
 PUs can greatly enhance the development speed on such architectures withou
 t loss of performance. With this in mind the University of Copenhagen is c
 urrently developing the Futhark programming language: a simple array-based
  functional programming language that allows its compiler to heavily optim
 ize the code for any platform CPU or GPU (Cuda or OpenCL) without requirin
 g any knowledge on the specificities of the platform from the developer. T
 he Futhark language is built with the philosophy to allow the creation of 
 very efficient libraries which can be called from end user applications th
 at can be written in Python or C.In this talk we will describe  the langua
 ge and discuss the performance of a fluid flow solver based on the lattice
  Boltzmann method. We will show that without any hardware-specific optimiz
 ations from the developer and no knowledge on the underlying hardware comp
 arable performance with respect to state-of-the-art optimized by hand code
  is achieved.\n\nDomain: CS and Math, Chemistry and Materials, Physics, En
 gineering
END:VEVENT
END:VCALENDAR