Classification-Based Optimization of Dynamic Dataflow Programs

Hervé Yviquel (1) , Emmanuel Casseau (2) , Matthieu Wipliez (3) , Jérôme Gorin (4) , Mickaël Raulet (1)

1 IETR - Institut d'Electronique et de Télécommunications de Rennes
2 CAIRN - Energy Efficient Computing ArchItectures with Embedded Reconfigurable Resources
3 Synflow [Rennes]
4 Télécom ParisTech

This chapter reviews dataflow programming as a whole and presents a classification-based methodology to bridge the gap between predictable and dynamic dataflow modeling in order to achieve expressiveness of the programming language as well as efficiency of the implementation. The authors conduct experiments across three MPEG video decoders including one based on the new High Efficiency Video Coding standard. Those dataflow-based video decoders are executed onto two different platforms: a desktop processor and an embedded platform composed of interconnected and tiny Very Long Instruction Word-style processors. The authors show that the fully automated transformations presented can result in a 80% gain in speed compared to runtime scheduling in the more favorable case.

Domaines: Systèmes embarqués

Hervé Yviquel, Emmanuel Casseau, Matthieu Wipliez, Jérôme Gorin, Mickaël Raulet. Classification-Based Optimization of Dynamic Dataflow Programs. Advancing Embedded Systems and Real-Time Communications with Emerging Technologies, IGI Global, pp.282-301, 2014, 9781466660342. ⟨10.4018/978-1-4666-6034-2.ch012⟩. ⟨hal-01068648⟩