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:20241120T082409Z
LOCATION:HG F 3
DTSTART;TZID=Europe/Stockholm:20240605T113000
DTEND;TZID=Europe/Stockholm:20240605T120000
UID:submissions.pasc-conference.org_PASC24_sess154_msa219@linklings.com
SUMMARY:A Shared Compilation Stack for HPC Stencil DSLs
DESCRIPTION:Minisymposium\n\nGeorge Bisbas (Imperial College London); Anto
 n Lydike, Emilien Bauer, Nick Brown, and Mathieu Fehr (University of Edinb
 urgh); Paul H.J. Kelly (Imperial College London); and Tobias Grosser (Univ
 ersity of Cambridge)\n\nDomain Specific Languages can massively improve co
 mputational science\nproductivity and also provide high performance. High-
 level DSLs capture\nabstractions that the DSL compiler can exploit\nto tar
 get current- and next-generation supercomputers.\nConsequently, there have
  been many DSL projects, notably in finite-difference stencil\ncomputation
 s - but implementations fail to share code and fail to harness combined de
 veloper effort.\nA large portion of their code base is dedicated to reason
 ing about generic HPC concepts, such as generation of directives for share
 d-memory parallelism, message-passing communications for distributed-memor
 y parallelism, vectorization, arithmetic (factorization, sub-expression el
 imination), and loop optimizations (blocking, fusion, fission). These gene
 ral-purpose optimizations are often combined with domain-specific ones to 
 maximize performance.\nThis talk presents joint work spanning three stenci
 l DSL projects, which aims to realize this lost potential - in Devito, Psy
 clone, and the Open Earth Compiler. We present a) how we tailor the widely
  adopted MLIR compiler framework to support optimizations suitable for FD-
 stencil computations and generate HPC-ready code running on multi-node CPU
 s and single-node GPUs, and b) how, for example, in Devito, we can leverag
 e these HPC-enabling contributed abstractions in MLIR and achieve better o
 r on-par performance by building on MLIR dialects and transformations that
  are shared across all three projects.\n\nDomain: Computational Methods an
 d Applied Mathematics\n\nSession Chair: Phillip Colella (Lawrence Berkeley
  National Laboratory, University of California)
END:VEVENT
END:VCALENDAR