On Wednesday September 28 2011, Professsor Michael O’Boyle (University of Edinburgh) will give a talk on Machine Learning based Compiler Optimisation and Parallelisation from 11:00 to 12:00 at the ExaScience Lab in imec, room 3.1A (building 3, 1st floor, meeting room A).. External visitors should register at imec’s Reception. Here is information on how to reach imec by various means of transportation.
The event is free of charge, but you have to register by sending an e-mail to Roel Wuyts (roelwuytsimecbe) (roelwuytsimecbe) , indicating the number of persons attending the event.
Machine Learning based Compiler Optimisation and Parallelisation
Professsor Michael O’Boyle (University of Edinburgh)
We live in exciting times. Platforms are multi-core, increasingly heterogeneous and rapidly evolving from one generation to the next. Compiler writers therefore face challenges on 3 fronts – how to find and exploit parallelism, how to utilise GPUs and other devices and how to do this before it all changes again. This talk describes how machine learning (ml) can be a useful tool in tackling each of these challenges. It describes how ml can dramatically reduce the amount of auto-tuning needed for improving code. It then shows how such an approach can be coupled with dynamic dependence analysis to auto-parallelise sequential programs for multi-cores. This can be then extended for heterogeneous systems where it can determine when a program can usefully exploit GPU capability.
About the Speaker
Michael O’Boyle graduated with a PhD in Computer Science from the University of Manchester in 1992. Prior to moving to a lectureship at Edinburgh in 1997, he was an SERC postdoctoral research fellow, a visiting scientist at INRIA and a visiting research fellow at the University of Vienna. In 2001 he was awarded an EPSRC Advanced Research Fellowship and later that year, a readership in the School of Informatics. Since then he has been a visiting scholar at Stanford University and a visiting professor at UPC Barcelona. In 2006, he was awarded a Personal Chair in Computer Science. His main research interests are in auto-parallelisation, machine-learning based compilation, mapping for heterogeneous systems, and compiler/architecture co-design. He is currently the Director of the Institute for Computing Systems Architecture and the ARM research centre of excellence.
- MulticoreBSP software updated
- Talk on “Semi-automatic SIMD-efficient data layouts for object-oriented programs” at CHEP2013
- Seminar “Computational methods for modelling protein folding and aggregation” by Joost Schymkowitz (VIB SWITCH Laboratory) on 26 June, 2013
- Best paper award at ISPASS for Multi-Threaded Sampling paper
- PRACE Preparatory Access Grant Obtained