Missions

The NumPex PEPR was launched in 2023 for a duration of six years. NumPEx contributes to the design and the development of numerical methods, software components and tools that support future productive European exascale and post-exascale systems.
NumPEx aims to support the Computational Scientific and Engineering (CSE) community in leveraging the capabilities and potential of these new architectures through expanded and reusable exascale software components (libraries, frameworks, tools) and through improved exascale software development practices. CSE application domains include, among others: Earth system, climate, and environment; astronomy and astrophysics; high-energy physics; biology and health sciences; fusion energy production and management; material sciences; aeronautics, automotive.
The NumPEx Exa-SofT project aims to develop an expanded coherent and exascale-ready software stack featuring breakthrough research advances that meets the needs of complex parallel applications and the requirements of heterogeneous exascale architectures while improving productivity, performance portability, scalability and resilience, performance and energy efficiency.
The NumPEx Exa-DI project aims to improve and accelerate the development of exascale CSE applications through proxy applications representative of cross-cutting computation and communication motifs, and Software Development Kits (SDKs) seen as logical collections of well-integrated and interoperable motif-based software components (libraries, frameworks, tools) that can be easily deployed and instantiated in the various CSE application environments. This is addressed iteratively through an agile co-design and co-development process across NumPEx and with the CSE application community.

The Exa-DI and the Exa-SofT projects are looking for a highly motivated candidate with a strong background in high performance scientific computing, multi-linear algebra and tensor contractions for heterogeneous exascale architectures. The successful candidate will join the NumPEx PEPR to reinforce collaboration between the Exa-SofT and the Exa-DI projects and better support multi-linear algebra and tensor contractions in exascale CSE applications and Machine Learning.

Activités

As part of the collaborative process between the Exa-DI and Exa-SofT projects, the mission activity will imply a split of ~50% working time on multi-linear algebra and tensor contractions topics in collaboration with the Exa-SofT research and software development teams, and ~50% working time in the Exa-DI CDT for better supporting and integrating tensor algebra and tensor contractions libraries in Motif-based proxy apps and SDKs.

As part of the Exa-DI CDT, the recruited person will:
- contribute, according to your expertise, to the Exa-DI Motif-based Working Group (WG1) on efficient high-order FEM discretisation for PDEs @exascale, to co-identify and specify new versions of the proxy-GEOS with explicit tensors trains for performance improvement and portability;
- develop in the CDT new versions of the Exa-DI proxy-GEOS for the integration of multi-linear algebra and tensor contractions software components (libraries, frameworks) leveraging parallel programming and execution models, together with abstraction layers
- design high-level algebraic interface that promotes the broadest and most flexible reuse of multi-linear algebra and tensor contractions software components and supports run-time selection of implementations tuned for a variety of computational devices (e.g. CPUs and GPUs) while ensuring overall consistency and performance.
- contribute to identify new CSE applications domains, such as condensed matter systems, quantum chemistry, computational biology with well specified Application Demonstrators that would need to better support tensor contractions and could possibly lead to new Exa-DI Motif-based activities by the end of 2026.
- provide technical guidance to less experienced CDT engineers and ensure knowledge transfer by providing support and training.
- play an active role in the Exa-DI Agile approach by contributing to process improvements, project planning and tracking, progress reviews and demonstrations, and coordination with other teams.

Exa-SofT mission
Task T4.4 in WP4 of the Exa-SofT project deals with the development of efficient algorithms and codes for multilinear algebra, with a particular focus on the use of innovative parallel programming models and tools. In the context of this task and as part of the Exa-SofT project, the recruited person will:
- design novel algorithm for large-scale block-sparse tensor computations that are suited for massively parallel accelerated supercomputers.
- investigate the use of task-based parallelism for the implementation of such algorithms. This task will involve the use of the StarPU runtime and will be carried in close collaboration with WP3 of Exa-SofT. This can potentially lead to the development of novel features in the StarPU runtime which can be potentially beneficial to other algorithms and applications.
- explore techniques to improve performance and efficiency on accelerators, such as GPUs. This might include the use of batching techniques to handle the case of tensor contractions with small variable-size blocks, as well as the use of advanced methods for load balancing in a strongly heterogeneous setting (e.g., CPUs+GPUs).

Additionally, this work can benefit from the developments of WP1 and WP2 of Exa-SofT around the development of optimization techniques for portability layers such as Kokkos, and those of WP5 and WP6 around the profiling and analysis of performance and energy consumption.

Compétences

- PhD in computer science or another field related to scientific computing
- A solid background in multi-linear abler and tensor contractions software development
- Proficiency in multiple programming languages (Python, C/C++, or Fortran), ideally with in-depth knowledge of parallel programming (GPU, multi-threading, etc.).
- Familiarity with standard collaborative development tools: Git, GitHub, CMake, Guix-HPC, Spack, GTest, CTest, etc.
- Pragmatism, analytical skills and ability to take initiative to confidently tackle complex problems with multiple constraints (deadlines, major technical challenges).
- Aptitude to teamwork and a strong interest in multi-disciplinary collaborations involving multiple stakeholders at the intersection of applied mathematics, computer science, and CSE applications.
- Excellent written and verbal communication skills, both in French and English.

Contexte de travail

This post-doc will be carried within the APO (Parallel Algorithms and Optimization) team of the IRIT (Institut de Recherche en Informatique de Toulouse) laboratory in the facilities of the ENSEEIHT school in Toulouse. It will be carried in the context of the national PEPR NumPEx (https://numpex.org/) Exa-DI and Exa-SofT projects.
The position requires regular travel (approximately 1 week per month) to the Paris-Saclay Campus for collaboration with the Exa-DI Computational Data Team (CDT), and possibly to the Bordeaux University Campus for collaboration with the Exa-SoFT research and software development teams.

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.