When

Nov 23, 2020 to Nov 27, 2020
(Europe/Berlin / UTC100)

Add event to calendar

iCal

The diversity and rapid change of the many supercomputing architectures used to run weather and climate codes and of the programming models required seriously affects the development productivity and the ability to retain a single source code running efficiently everywhere. Domain-specific languages (DSLs) provide a solution to the productivity and portability of these codes. This training course provides insights into the DSLs considered in ESiWACE2 (PSyclone, and the Dawn DSL ecosystem) and demonstrates how to apply them to weather and climate models

Topics:

  • Insights into the PSyclone and Dawn DSLs.
  • Demonstration of the DSLs on different types of global models, such as LFric, ICON and NEMO.
  • Exercises to learn step by step how to develop models using a DSL.
  • Insights into the performance portability provided by DSLs for multiple architectures.

 

Audience: Participants interested in theoretically and practically learning how to use the DSL languages to implement weather and climate models targeting different computing architectures. During a hands-on session, participants will be encouraged to implement some of the model benchmarks studied during the training course and to build their own toy models, followed by an in-depth evaluation of the generated optimised implementations and their performance benefits. This will be an online training course and places will be limited.

Final agenda and presentation slides

The video recordings of the training are available on the ESiWACE YouTube channel.



Read more about the training in the Results section.