Menu

Youenn LEBRAS

MONTPELLIER

En résumé

Mes compétences :
implementation of used cases and examples
static and dynamic analysis
Performance Optimization
Compilation
High Performance Computing

Entreprises

  • Université De Versailles Saint-quentin-en-yvelines - Doctorant

    2015 - 2019 Modern high performance processor architectures tackle performance issues by heavily relying on increased vector lengths and advanced memory hierarchies to deliver high performance.
    Manual optimization of production HPC codes became a difficult task when having to manage multiple architecture dependent transformations.
    Developers usually trust compilers to automatically address these performance issues, but they deploy static performance models and heuristics which, must remain conservative or even fail in the worst case.
    Compiler optimization choices can be refines by using feedback data from dynamic profiling.
    But, it does not always consider some metrics and is rarely aggressive enough regarding metric data collection.
    On the other hand, performance analysis tools are pretty good at detecting specific performance issues, but only return observations on the quality and on the execution of the code.
    Our goal is to develop a framework will allow to perform of source code transformations based on performance analysis tools metrics.
    This framework will be integrated into the MAQAO tool suite.
    We present an FDO tool with a set of source-to-source transformations guided by metrics coming from the various MAQAO tools and open to user advices.
    This framework can also be used to simplify the development by automatically performing some simple, but time-consuming and error-prone transformations (e.g. loop/function specialization).

    tldr; Code Optimization Based on Source-to-Source transformations guided by static and dynamic profilers analysis.

    Ph.D. advisor: Prof. William Jalby

Formations

Réseau

Pas de contact professionnel

Annuaire des membres :