Parallelism-High Performance Computing-Grid

Parallelism-High Performance Computing-Grid
Domain - extra
hardware architecture
october 2010
Hardware Architecture Generation from High-Level Algorithmic Specifications
Thesis advisor
Lionel Lacassagne, IEF, University Paris XI
Dominique Ginhac, LE2I, University of Burgundy
Collaboration with LE2I in the framework of ANR PARADIS project
This thesis is a joint work in the framework of the PARADIS project which aims at providing high-level tools for hardware architecture generation. In collaboration with LE2I and LASMEA for the hardware part, the PhD student will have to perform generic analysis of hardware component, build formal rules to maps application structure on a FPGA like platform and incorporate it into an existing algorithmic skeletons based tools – QUAFF – to build a software/hardware tools for parallel computing.
Generic and Generative programming proved themselves to be valuable techniques for developing high performance, customizable software system. Various applications to parallel programming have been proposed but most tackle the problem of generating an optimal program for a given architecture. This PhD thesis, in the framework of the PARADIS project, aims at investigating Generic programming applied to the dual problem: for a given application, find the optimal hardware to run it. This exploration is mainly doable by the use of constrained parallel constructs –like algorithmic skeletons- that reduces the exploration space size.
The thesis will focus on building the software environment needed to specify, simulate, debug and generate, for a given application, the architecture specification (in terms of reusable components from Task 3) and the application code itself. This integrated flow will leverage the classic difficulties of such development by using Quaff, a C++ meta-programming based skeleton library, as a base for the tools. Basically, we’ll focus on extending Quaff handling for architectural specifications by providing new back-end and code generators. The main challenge is to be able to define a comprehensive set of skeletons and to adapt their formal semantic and production rules to the hardware layer. This is basically the dual process of what’s Quaff is doing at the moment: instead of generating the optimal code for a fixed architecture, we have to find the proper architecture for a given application.
Work program
The work is subdivised in four parts:
  • Analyzing and proposing a set of skeletons suited for image and vision applications
  • Define the kind of hardware components to use for such applications
  • Define rules and program them as meta-programs inside QUAFF
  • Develop applications within this framework

Extra information
  • Une bibliothèque métaprogrammée pour la programmation parallèle, Joel Falcou et Jocelyn Sérot, in Technique et Science Informatique, Volume 28(5), pp 645 – 675
  • Formal semantics applied to the implementation of a skeleton-based parallel programming library.", J. Falcou and J. Serot, In Parallel Computing: Architectures, Algorithms and Applications (Proc. of PARCO 2007)
The student will ideally have experience in software development, mainly modern C++ idioms and generative techniques, or experience in hardware design.
Expected funding
Research contract
Status of funding
Mardi 23 février 2010 22:22:15 CET
dernière modif.
Mercredi 24 février 2010 11:04:15 CET

Fichiers joints

Aucun fichier joint à cette fiche

Ecole Doctorale Informatique Paris-Sud

Nicole Bidoit
Stéphanie Druetta
Conseiller aux thèses
Dominique Gouyou-Beauchamps

ED 427 - Université Paris-Sud
UFR Sciences Orsay
Bat 650 - aile nord - 417
Tel : 01 69 15 63 19
Fax : 01 69 15 63 87
courriel: ed-info à