Automatic Hierarchical Discovery of Quasi-Static Schedules of RVC-CAL Dataflow Programs

Jani Boutellier (1) , Mickaël Raulet (2) , Olli Silvén (1)

1 MVG - Machine Vision Group
2 IETR - Institut d'Electronique et de Télécommunications de Rennes

RVC-CAL is an actor-based dataflow language that enables concurrent, modular and portable description of signal processing algorithms. RVC-CAL programs can be compiled to implementation languages such as C/C++ and VHDL for producing software or hardware implementations. This paper presents a methodology for automatic discovery of piecewise-deterministic (quasi-static) execution schedules for RVC-CAL program software implementations. Quasi-static scheduling moves computational burden from the implementable run-time system to design-time compilation and thus enables making signal processing systems more efficient. The presented methodology divides the RVC-CAL program into segments and hierarchically detects quasi-static behavior from each segment: first at the level of actors and later at the level of the whole segment. Finally, a code generator creates a quasi-statically scheduled version of the program. The impact of segment based quasi-static scheduling is demonstrated by applying the methodology to several RVC-CAL programs that execute up to 58 % faster after applying the presented methodology.

Domaines: Systèmes embarqués

Jani Boutellier, Mickaël Raulet, Olli Silvén. Automatic Hierarchical Discovery of Quasi-Static Schedules of RVC-CAL Dataflow Programs. Journal of Signal Processing Systems, 2013, 71 (1), pp.35-40. ⟨10.1007/s11265-012-0676-4⟩. ⟨hal-00717218⟩