Supercomputers, like the ones made available to scientists within PRACE (Partnership for Advanced Computing in Europe), which provide computing power in the petaflops range, are very expensive research resources. Therefore, the available processing time and energy should be used efficiently. This requires a continuous search for optimization potential in programming parallel architectures based on either multi-core processors, many-core processors or even general purpose graphics processing units (GPGPUs). While performance analysis tools exist to help the developers detect bottlenecks, these tools do not give any recommendations on how to subsequently tune the code. A solution for this problem will become available through the AutoTune project. AutoTune will extend the performance analysis tool “Periscope”, an automatic online and distributed performance analysis tool developed by Technische Universität München (TUM), with automatic online tuning plugins for performance and energy efficiency tuning.
AutoTune is coordinated by TUM and has a budget of approximately three million Euros. It will be funded by the European Commission with about two million Euros. The project started mid of October 2011 and will run for three years. In addition to TUM, the Leibniz Supercomputing Centre (LRZ) of the Bavarian Academy of Sciences, CAPS entreprise, the Universitat Autònoma de Barcelona (UAB), the Centre for High-End Computing (ICHEC) at the University of Galway as well as the University of Vienna are partners in the project. IBM is an associated partner.
The expertise of this consortium covers all relevant areas. TUM developed the automatic performance analysis tool Periscope. The University of Vienna brings in their expertise in language, compiler and runtime techniques for efficient programming of GPGPU-accelerated architectures. CAPS entreprise is developing HMPP (Hybrid Multicore Parallel Programming), an easy-to-use directive-based programming interface for GPGPU. UAB developed the MATE (Monitoring, Analysis and Tuning Environment) framework for tuning pattern-based parallel applications. The LRZ delivers extensive experience with application tuning and will work together with IBM on energy efficiency tuning. IBM will provide extensions to its Load Leveler for LRZ‘s next petascale system named SuperMUC. ICHEC is especially experienced in the area of GPGPU programming and will bring in their application knowledge.
Univ.-Prof. Dipl.-Ing. Dr. Siegfried Benkner