Research
Research topics
- Computer arithmetic (fixed-point and floating-point arithmetic).
- Software implementation of floating-point operators and mathematical functions.
- Automated code generation and validation.
- Automation of debugging scientific floating-point programs.
- Precision tuning and performance improvement of numerical programs.
Projects
- PADOC (Performance- and accuracy-aware data format optimization in numerical codes), ANR JCJC, 2019-2025.
- PERPREVAL-AUTO (Performance, précision, validation automatiques), grant from Région Occitanie and UPVD, 2018-2021.
- CTAOptSim (Optimization of performance and numerical accuracy of CTA software), CNRS PEPS Astro-Informatique, 2018-2020.
- REPRISES (Portability, sustainability, and reproducibility of numerical programs in physic), IN2P3 R&D project, 2017-.
- MetaLibm (Code generation for mathematical functions and filters), ANR INS, 2013-2018.
- DEFIS (Design of fixed-point embedded systems), ANR INS, 2011-2015.
Students
Ph.D. students
- Youssef Fakhreddine, with Philippe Langlois (2022-, ANR JCJC PADOC).
- Harris Ludwig Boudi, with Philippe Langlois (2018-2021, PERPREVAL-AUTO).
- Hugues de Lassus Saint-Geniès, with David Defour (2014-2018, MetaLibm).
- Amine Najahi, with Matthieu Martel (2011-2014, DEFIS).
MS students
- Youssef Fakhreddine (2021, M2 CHPS, UPVD).
- Mouad Ettalbi (2016, M1 CHPS, UPVD), with Philippe Langlois.
Software
- FLIP: Software support for binary32 floating-point arithmetic on integer processors.
- CGPE: Generation of fast and certified codes for the evaluation of polynomials in fixed-point arithmetic.
- FPLA: Generation of linear algebra basic blocks in fixed-point arithmetic.
- PPT: Generation of error-free tables for the implementation of trigonometric and hyperbolic functions based on Pythagorean triples.
- MetaLog: Meta-implementation of faithfully rounded logarithm function.
- GenericImplementPoly: Automated adaptation of precision in polynomial evaluation in floating-point arithmetic.
- vectoptim: Precision reduction of instructions appearing in vectorizable blocks.
- fp2mp: Analysis of the impact on accuracy of the transformation of floating-point formats.
- loop-splitter: LLVM loop splitter.
- dd-fp2mp: Automated adaptation of instruction floating-point format in iterative routines.