Efficient parallelization of large-scale hard real-time applications

Parallelisation efficace de larges applications temps-reel

Keryan Didier (1) , Dumitru Potop-Butucaru (1) , Guillaume Iooss (2) , Albert Cohen (2) , Jean Souyris (3) , Philippe Baufreton (4) , Amaury Graillat (5)

1 AOSTE2 - Models and methods of analysis and optimization for systems with real-time and embedded contraints
2 Parkas - Parallélisme de Kahn Synchrone
3 Airbus [France]
4 SAFRAN Group
5 Kalray

We present a parallel compilation method for embedded control applications. The method is fully automatic and scales up, being based on low-complexity heuristics. Unlike classical compilation, it also takes as input non-functional requirements, e.g. real-time or resource limits. The main objective is not optimization per se, but the respect of requirements. To this end, static resource allocation and code generation algorithms perform a safe accounting of non-functional properties. Accounting starts from per-component time and memory footprint worst-case bounds, automatically obtained through calls to state-of-the-art static analysis tools. Experiments show that our method produces efficient code for large-scale, real-life avionics applications.

Mots clés: Offlne scheduling; Parallelization; Compilation; Hard real-time; Many-core; Code generation; Timing analysis; Lustre; Scade; Shared memory

Domaines: Systèmes embarqués

Keryan Didier, Dumitru Potop-Butucaru, Guillaume Iooss, Albert Cohen, Jean Souyris, et al.. Efficient parallelization of large-scale hard real-time applications. [Research Report] RR-9180, INRIA Paris. 2018. ⟨hal-01810176v2⟩