Abstract: The “write-once-run-everywhere” programming models are still seen as a marketing trick in computer science. For example, OpenCL, the newest such model, is no exception: proposed in 2009 as an instrument to address the portability of parallel programming over multiple multi-/many-core architectures, it has been quickly criticized for its lack of “performance portability”. This talk is intended as a discussion on performance portability. Therefore, I would like to present answers to three relevant questions: (1) what is performance portability? (2) how can we evaluate performance portability? (3) can a programming model tackle performance portability? Along the talk, I am using OpenCL and multi-/many-core architectures (CPUs and GPUs) as running examples.
Bio: Ana has recently joined the faculty at TU Delft, after a spell in Amsterdam. Her PhD (which was at Delft) was on parallel programming models and tools. She has extensive experience of achieving high performance on exotic parallel hardware – multicore, GPUs, FPGAs, the Cell processor, etc. She has worked on applications in radio astronomy, video indexing, and radiation transport. She has also spent time at IBM Research and the Barcelona Supercomputing Centre. She is a graduate of the POLITEHNICA University of Bucharest. Further details of her work are here:
http://scholar.google.co.uk/citations?user=UypyD2wAAAAJ&hl=en