Description du sujet de thèse

After the great wave of the object paradigm in programming, which led to data organization of the AoS (for 'Array of Structures') type, the computing community is gradually returning to the SoA (for 'Structure of Arrays'), which lends itself more effectively to the parallelization of SIMD (CPU) and SIMT (GPU) calculations, necessary for optimum performance today. Going one step further, we can build AoSoAs (for 'Array of SoA'), which can adapt to processor cache hierarchies. Curiously, although widely used to optimize supercomputing, these approaches have been little researched and published in the computer science research community [1].
In particle physics, researchers and engineers are studying similar optimizations [2], adapting them to the specific problems of High-Throughput Computing (HTC), notably by frequently resorting to 'jagged' [3] or 'awkward' [4] arrays, a dense but irregular hyper-matrix.
This thesis project brings together players from LISN (computer science research laboratory), IJCLab and CERN (physics laboratories), who already contribute for the development of the software library KIWAKU [5] which lets you manipulate vectors, matrices, tensors, etc.
The PhD candidate will be tasked with identifying the latest theories and practices in AoSoA and seeing if they can be adapted to the irregular data structures of particle physics. Finally, he/she will validate his/her proposals on KIWAKU and on full-scale use cases for the international LHCb collaboration, in particular for the reconstruction of data from the Velo sub-detector. The latter aims to unify the data structures used in CPU and GPU variants, with the use of SoA and multi-dimensional irregular arrays.
[1] Data layout and SIMD abstraction layers: decoupling interfaces from implementations. Sylvain Jubertie, Ian Masliah, Joel Falcou. https://hal.science/hal-01915529v1
[2] Fast and flexible data structures for the LHCb Run 3 software trigger. Sevda Esen, Arthur Marius Hennequin, Michel De Cian. https://doi.org/10.48550/arXiv.2307.03689
[3] Evolution of the ATLAS event data model for the HL-LHC. Attila Krasznahorkay, Paul Gessinger, Scott Snyder, Nicholas Swatman. https://indico.cern.ch/event/1338689/contributions/6010097/
[4] The Awkward World of Python and C++. Manasvi Goyal, Ianna Osborne, Jim Pivarski. https://arxiv.org/pdf/2303.02205
[5] Kiwaku, a C++20 library for multidimensional arrays Application to ACTS tracking. Sylvain Joube, Hadrien Grasland, David Chamont, Joël Falcou. https://hal.science/hal-04401240v1

Contexte de travail

The thesis will be attached to the STIC doctoral school of the Université Paris-Saclay. The PhD candidate will work both at IJCLab and LISN.
The Irène Joliot-Curie Physics Laboratory of 2 Infinities (IJCLab) is a physics laboratory under the supervision of the CNRS (IN2P3), the University of Paris-Saclay and the University of Paris. It is located on the campus of the University of Paris-Saclay in Orsay, 20 km south of Paris and easily accessible by RER in 35 minutes. The laboratory has extensive technical capabilities (around 280 engineers and technicians) in all the major fields required to design, develop and implement the experimental devices needed for its scientific activity. The PhD candidate will be attached to the Software Development Service of the IT Department.
IJCLAB bases its recruitment policy on the promotion of equality, diversity and inclusion. Essential values, they allow the professional development of agents, who are real actors in a collective success, but also the development of the laboratory itself.

The Laboratoire Interdisciplinaire des Sciences du Numérique (LISN), based on the Université Paris-Saclay campus, is a multidisciplinary research laboratory created by bringing together the teams of two leading computer science laboratories at the Université Paris-Saclay: the Laboratoire d'Informatique pour la Mécanique et les Sciences de l'Ingénieur (LIMSI Laboratoire d'Informatique pour la Mécanique et les Sciences de l'Ingénieur, founded in 1972, whose teams joined those of the LRI in 2021 to form the LISN.) and the Laboratoire de Recherche en Informatique (LRI).

Administratively, LISN is a Unité Mixte de Recherche (UMR9015), with CNRS and Université Paris-Saclay as its main supervisors, and INRIA and CentraleSupélec as secondary supervisors.

Le poste se situe dans un secteur relevant de la protection du potentiel scientifique et technique (PPST), et nécessite donc, conformément à la réglementation, que votre arrivée soit autorisée par l'autorité compétente du MESR.

Contraintes et risques

The PhD candidate will be required to work alternately at IJCLab and LISN (Orsay), in different ways over time, depending on the needs of his/her research. They may also be asked to travel to CERN (Genève) for periods ranging from a few days to a few weeks.

Informations complémentaires

The candidate must hold an engineering degree and/or a master's degree in scientific computing at the time of taking up the post.
The position requires solid knowledge of C++, Linux and Git, together with good oral and written communication skills (French and English required) to present at conferences and write articles for scientific journals. In addition, the candidate must be able to work as part of a team and report frequently on results, within large international collaborations involving foreign physicists.
Applications should include an unabridged CV, a letter of motivation and at least two references (persons likely to be contacted).