Plusieurs experts internationaux dans les domaines de l’embarqué et du calcul hautes performances participent à cet évènement.
L’agenda prévisionnel du workshop est le suivant :
- 10:00 - 11:00 Tutorial : programming large scale heterogeneous systems (LSHS) using mathematical extensions to standard programming models
- 11:00 - 12:00 invited talks : problems and approaches to program LSHS, chair : Lutz Schubert
- 12:00 - 13:00 panel discussion : how can we overcome the problems with programming future LSHS
L’évènement sera suivi, l’après midi, par une réunion interne du projet avec les membres de son comité de conseil.
Ci dessous le résumé du workshop anglais :
As scale and heterogeneity keeps increasing in next-generation computing systems on all levels, ranging from embedded to high performance computing systems, traditional programming models fail to a growing extent. With modern multi- & many-core SoC architecture, reconfigurable chips and specialised co-processors, the traditional von Neumann based concepts for programming need to be urgently revised. Hybrid parallel programming approaches such as MPI+OpenMP suffer from increasingly higher programming effort, but hardly offer any benefit in terms of exploitation of heterogeneity. Other approaches such as OpenCL, OpenACC etc. all offer increased heterogeneity, but at the cost of performance and scalability. More and more initiatives start investigating alternative programming models such as data flow based programming or functional programming paradigms as an alternative to deal with the communication problems and the higher-level representation of the algorithm’s intention. The workshop focuses on new and extended approaches to programming large scale systems with a strong heterogeneous setup, respectively programming portable software that still manages to exploit the system specifics and parallelism efficiently. Presentations shall focus on key challenges that arise from modern systems and suggestions to tackle them.