Performance Portability and Programmability for Heterogeneous Many-core Architectures (PEPPHER)
Forskningsprojekt, 2010
– 2012
To develop a framework for programming and optimizing applications for heterogeneous many-core processors. It will allow applications developers to extract optimal performance with fast time-to-market
The emergence of heterogene-ous many-core processors, in a large spectrum of systems from embedded and general-purpose to high-end computing systems, poses major challenges to the European software industry. In general, there is no guarantee that software developed for a particular architecture will be executable on another, related architecture. For instance, H.264 codecs are essentially rewritten for each generation of DSP proc-essor in heterogeneous architec-tures, a practice that is clearly wasteful in resources and should be avoided as far as possible. A related issue is the coding of multiple versions of H.264 codecs to play back or encode different resolutions of video (QVGA, VGA, WVGA, D1, 720p, 1080i, or 1080p) using differing numbers of cores and memory in order to meet the proc-essing requirements for each resolution. The PEPPHER project is de-veloping a unified framework for programming and optimizing appli-cations for heterogeneous many-core processors to ensure func-tional and performance portability with reasonable programming effort. PEPPHER envisions two kinds of programmers: (1) “mainstream programmers” who build applications using the PEPPHER framework and specify performance-expectations, (2) “expert programmers” who imple-ment and extend the PEPPHER framework. Once an application is devel-oped using the PEPPHER methodology and framework, no further porting effort of the mainstream programmer is needed for efficient execution of application on any of the target PEPPHER platforms. For a set of target applications and platforms, PEPPHER will reduce the porting effort of the mainstream programmer to zero.
Deltagare
Philippas Tsigas (kontakt)
Chalmers, Data- och informationsteknik, Nätverk och system
Samarbetspartners
Codeplay Software Ltd
London, United Kingdom
Institut National de Recherche en Informatique et en Automatique (INRIA)
Le Chesnay Cedex, France
Intel Deutschland GmbH
Feldkirchen, Germany
Karlsruher Institut für Technologie (KIT)
Karlsruhe, Germany
Linköpings universitet
Linköping, Sweden
Movidius Ltd
Dublin, Ireland
Technische Universität Wien
Wien, Austria
Universität Wien
Wien, Austria
Université de Bordeaux
Bordeaux, France
Finansiering
Europeiska kommissionen (EU)
Projekt-id: EC/FP7/248481
Finansierar Chalmers deltagande under 2010–2012