Energy and power constrains are conditioning the next-generation supercomputers toward highly parallel architectures, based either on hybrid CPU-GPU design or on homogeneous many cores System on a Chip. Applications need to be restructured to cope with massive parallelism that is present at all levels of the architecture (vector core, node, system wide), and new programming paradigms must be implemented to exploit them.
Moreover, in many application domains, high level code gluing (e.g. python) are being used to couple applications or to link domain specific libraries as a way to leverage the already existing massively parallel codes.
This school focus on innovative software development techniques to address the implementation of new HPC applications and the re-factory of existing ones, in the era of energy efficient, massively parallel architectures.
Software engineering techniques and high productivity languages will complement more traditional lectures on parallel programming, to allow the implementation of applications that can be maintained across a complex and fast evolving HPC architectures.
The school, in the hands-on session, will give the possibility to exercise using both Nvidia and Intel massively parallel chips.
* Characteristics of Hybrid and Homogeneous architectures
* Elements of software engineering
* Parallel programming techniques for hybrid CPU-GPU architectures
* Parallel programming techniques for homogeneous Many core architectures
* New OpenMP stadard and task based programming
* Introduction to Python for high performance computing
* Integration of different programming paradigm: MPI, OpenMP, python and device specific
The school is aimed at PRACE users, final year master students, PhD students, and young researchers in computational sciences and engineering, with different backgrounds, interested in applying the emerging technologies on high performance computing to their research.
Basic knowledge of parallel programming with MPI and/or OpenMP, knowledge of FORTRAN and C languages.
Basic knowledge of parallel computer architectures.
Attendance is free. A grant of 250 EUR (for students working abroad) and 150 EUR (for students working in Italy) will be available for participants not funded by their institution and not working or living in the Bologna area. Documentation will be required. Lunch will be provided by Cineca. Each student will be given a two month access to the Cineca's supercomputing resources.
January 16th, 2016
The number of participants for each edition is limited to 24 students. Applicants will be selected according to their experience, qualification and scientific interest.
EDITION FOR 2016
15 - 19 of February, 2016 - Bologna
This edition will be held in English and non-Italian speaking students are welcome!