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:20190719T085744Z
LOCATION:HG EO Nord
DTSTART;TZID=Europe/Stockholm:20190613T195000
DTEND;TZID=Europe/Stockholm:20190613T215000
UID:submissions.pasc-conference.org_PASC19_sess179_post113@linklings.com
SUMMARY:CSM01 - Abstracting Scientific Array-Based Computations in High-Pe
 rformance Computing Environments
DESCRIPTION:Poster\n\n\nCSM01 - Abstracting Scientific Array-Based Computa
 tions in High-Performance Computing Environments\n\nMacià, Haveraaen\n\nMa
 nipulating multi-dimensional arrays is at the core of numerical methods an
 d algorithms for solving equations describing physical phenomena, a signif
 icant domain for high-performance computing. The layout and the access pat
 terns of array data are crucial to the optimal performance of array-based 
 computations. As we move towards exascale computing, and ever more varied 
 hardware architectures, an abstraction for an effective data parallel comp
 utation is needed to overcome portability challenges. Here we present a co
 mplete library-based environment to abstract the continuous high-level app
 lication layer from the final detailed parallel low-level code. The applic
 ation domain’s concepts are transparently mapped and described at an
  intermediate layer using the mathematics of arrays (MoA) formalism and th
 e ψ-calculus — an algebra of multi-dimensional arrays and an ind
 ex calculus, specifying the data, its decomposition and its mapping to the
  underlying architecture. Programs written towards such an environment are
  independent of underlying layout and access pattern modifications, thus e
 asily ported between architectures. Numerical and parallel implementations
  become application-independent and can be designed for reuse and tuned to
 wards the efficient exploitation of different computer architectures. The 
 approach thus gives a clean and natural separation between the application
  code, the solver algorithm, and the underlying hardware architectures.
END:VEVENT
END:VCALENDAR