Calcul embarqué haute performance, Technologies du futur, Méthodes et outils
LLVM support for Variable and eXtended Precision (VXP) floating-point computations
To meet the growing needs of high-performance numerical computation, our laboratory is engaged in the development of a hardware and software stack for computations with Variable and eXtended Precision (VXP) floating-point numbers.
The hardware part consists of a RISC-V host processor whose instruction set architecture (ISA) was extended with new instructions for a VXP accelerator. The software part consists of an extension to the C and possibly C++ languages that allows programming with variable precision, which can varies dynamically at runtime. The targeted applications are linear algebra kernels and other High Performance Computing (HPC) calculations.
Based on a first prototype resulting from a PhD [1], the laboratory is looking for a candidate to carry out the support of C and C++ extensions for variable precision programming in the LLVM compiler chain. The objective is to maximize the expressive power of the language extensions while maintaining the benefits of the multiple optimization passes already present in LLVM, and developing VXP-specific passes and runtime.
This work will be in close collaboration with the hardware development team of the RISC-V accelerator, as well as with the team that is porting applications and HPC kernels to this new architecture.
The candidate will join the CEA’s project team of research engineers in Grenoble (France) and may be in contact with industrial or academic partners, in France or in Europe. Travel and occasional missions may be necessary for trade shows or visits to our partners.
Pour candidater sur le site du CEA: https://www.emploi.cea.fr/offre-de-emploi/emploi-developpeur-logiciel-pour-compilateurs-llvm-et-gcc-h-f_17413.aspx
Fichier PDF
Télécharger le PDF