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:20250822T115805Z
LOCATION:Campussaal - Plenary Room
DTSTART;TZID=Europe/Stockholm:20250616T102000
DTEND;TZID=Europe/Stockholm:20250616T105000
UID:submissions.pasc-conference.org_PASC25_sess149_pos102@linklings.com
SUMMARY:P01 - Achieving Performance Portability on ECMWF’s Open-Source Ope
 rational Wave Model ecWAM Using Source-To-Source Translation and GPU-Aware
  Data-Structures
DESCRIPTION:Michael Staneker and Ahmad Nawab (ECMWF)\n\nIt can be quite ch
 allenging to adapt production numerical weather prediction (NWP) codes for
  GPU execution. Those codes have typically been developed and optimised fo
 r multi-core CPUs and are continually being updated by domain scientists. 
 Additional complexity arises from the vast size of these codebases, the in
 creased diversity in available platform architectures with native and deri
 ved programming models as well as the necessity of vendor-specific modific
 ations to achieve optimal performance. At ECMWF, we manage this complexity
  using Loki, our source-to-source translation toolchain, and FIELD API, a 
 GPU-aware data-structures library. In this poster we present how these two
  tools have been used to achieve performance portability on ECMWF’s operat
 ional wave-model, ecWAM. Starting from the original CPU optimised Fortran 
 code, we present different GPU-capable variants that can be generated via 
 Loki. The variants presented are diverse in terms of optimisation strategi
 es and employed programming models. As one of the highlights, Loki is capa
 ble of translating the original Fortran kernels to C-style kernels like CU
 DA for NVIDIA GPUs and HIP for AMD GPUs.  With this, we present not only p
 erformance across multiple architectures but also showing potential perfor
 mance benefits resulting from translation to native kernel languages.\n\nS
 ession Chair: Chris Cantwell (Imperial College London)\n\n
END:VEVENT
END:VCALENDAR