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:20210916T132454Z
LOCATION:Mère Royaume
DTSTART;TZID=Europe/Stockholm:20210709T143000
DTEND;TZID=Europe/Stockholm:20210709T150000
UID:submissions.pasc-conference.org_PASC21_sess115_msa157@linklings.com
SUMMARY:Abstracting Heterogeneous Memory Systems for High Performance Appl
 ications
DESCRIPTION:Minisymposium\n\nAbstracting Heterogeneous Memory Systems for 
 High Performance Applications\n\nDykes, Marsden\n\nHeterogeneous memory sy
 stems are present in almost all modern high-performance computing systems,
  from traditional cache hierarchies and NUMA domains in main memory, to hi
 gh bandwidth memories in accelerator and coprocessor devices, off-node fla
 sh storage, and more recently on-node NVDIMMs. In this landscape, memory o
 ptimisation is crucial for performance, however addressing such memory sys
 tems is an ongoing challenge for application programmers both in terms of 
 technical implementation and effective utilisation in scientific codes. EP
 iGRAM-HS is an EU project focused on enabling extreme-scale applications o
 n heterogeneous hardware. We will present our efforts to support applicati
 on programmers in identifying, utilising, and transporting data in modern 
 heterogeneous memory environments, and preliminary application of those ef
 forts in existing scientific codes. In particular, we will discuss the Mam
 ba library for abstracting memory which is based on an array-like object a
 nd accompanying API. We will also discuss the challenges of utilizing hete
 rogeneous memory in weather and climate codes, and present an example of M
 amba usage in a mini-app, demonstrating GPU compute on Mamba-managed memor
 y, in a manner that could facilitate incremental GPU-porting of large code
  bases.\n\nDomain: CS and Math, Climate and Weather, Engineering
END:VEVENT
END:VCALENDAR