PADOC : Performance- and Accuracy-aware Data format Optimization in numerical Codes
ANR JCJC, 2018.
Partners: DALI/LIRMM.
Project leader: Guillaume Revy (LIRMM, Perpignan)
Project description
Project members
Philippe Langlois (LIRMM, Perpignan), David Parello (LIRMM, Perpignan), Guillaume Revy (LIRMM, Perpignan)
CTAOptSim
CNRS PEPS Astro-Info, 2018.
Partners:
- Laboratoire Univers et Particules de Montpellier (CNRS/IN2P3/UMR5299/LUPM), Université de Montpellier, équipe Expérience et Modélisation en Astroparticules (EMA)
- Laboratoire d'Informatique, de Robotique et de Microélectronique de Montpellier (CNRS/INS2I/UMR5506/LIRMM), Université de Perpignan via Domitia (UPVD), équipe Digits, architectures et logiciels informatiques (DALI)
Project leader: Johan Bregeon (CNRS, LUPM, Montpellier)
Project description
The Cherenkov Telescope Array (CTA), currently under construction, is the next-generation instrument in the field of very high energy gamma-ray astronomy. The first data are expected by the end of 2018, while scientific operations shall start around 2022 and last around 30 years. In order to characterise the instrument response to the Cherenkov light emitted by atmospheric cosmic ray showers, detailed Monte Carlo simulations will be regularly performed in parallel to CTA operations. The estimated CPU time associated to these simulations is very high, of the order of 200 millions HS06 hours per year. Reducing the CPU time devoted to a single simulation would allow either to reduce infrastructure cost or to better cover the large phase space.
For CTA, the simulation of the atmospheric cascade induced by the primary particle, and the successive generation and propagation of Cherenkov photons through the atmosphere to the ground is handled by the corsikaprogram and a dedicated C module. Our initial studies have shown that up to 70% of the computing time is indeed spent in the Cherenkov photon propagation module, which optimization has become the first target of our project. We're willing to study different options of code optimization, with a particular focus on vectorization facilities (SIMD instructions) together with computing precision tuning, and mathematical functions evaluation. This work is also used as a real life test bench to develop automatic tools able to optimise code performance while keeping a fair balance with the required precision and stability, a question that is of high importance for the computer science community in HPC.Project members
- Johan BREGEON, chargé de recherche, équipe EMA au LUPM (P.I.)
- Luisa ARRABITO, ingénieure de recherche, service informatique du LUPM
- Philippe LANGLOIS, professeur de l'UPVD, équipe DALI du LIRMM
- David PARELLO, maître de conférence à l'UPVD, équipe DALI du LIRMM
- Guillaume REVY, maître de conférence à l'UPVD, équipe DALI du LIRMM
QUARENUM: QUAlity and REproducibility in NUMerical simulations
PEPS INS2I, 2013.
Partners: LRI, LIP6, LIP, LIRMM and ONERA.
Project leader: Marc Baboulin (LRI).
Project description
The QUARENUM project aims at designing and implementing efficient algorithms for numerical validation and reproducibility in high-performance computing (HPC) applications. This project focuses on various methods to analyze the sensitivity of HPC applications to errors (e.g., simulation of rounding error propagation using stochastic or interval arithmetics, estimation of condition number and backward errors). These aspects and the objective of reproducibility are studied in the framework of massively parallel heterogeneous architectures.
Project members
Marc Baboulin (LRI, Orsay), David Defour (LIRMM, Perpignan), Joel Falcou (LRI, Orsay), Stef Graillat (LIP6, Paris), Fabienne Jézéquel (LIP6), Philippe Langlois (LIRMM, Perpignan), Thien-Hiep Lê (ONERA, Châtillon), Nicolas Louvet (LIP/ENS Lyon), Nathalie Revol (LIP/ENS Lyon).
DEFIS : Design of fixed-point embedded systems
ANR 2011-2014, Programme "Ingénierie Numérique et Sécurité"
Partners: IRISA, LIP6, CEA, Thales, and Inpixal.
Project leader: Olivier Sentieys (IRISA).
Float to Fix Conversion
Context: Embedded systems are usually coming with stringent constraints. Fixed-point arithmeticis mainstream in many embedded systems. However, fixed-point conversion is a tedious, time-consuming and error prone task. Efficient tools are required to automate the fixed-point conversion: reduce time-to-marke, validate numerical behavior, optimize implementation (energy, cost, performance).
Objectives: Aim of the DEFIS project is threefold
–To provide new analytical methods for fixed-point refinement of complex applications
–To develop a complete software infrastructure for fixed-point conversion automation
–To demonstrate the quality of DEFIS flow on two industrial applications