A Compilation Flow for Parametric Dataflow: Programming Model, Scheduling, and Application to Heterogeneous MPSoC

Efficient programming of signal processing applications on embed-ded systems is a complex problem. High level models such as Syn-chronous dataflow (SDF) have been privileged candidates for deal-ing with this complexity. These models permit to express inherent application parallelism, as well as analysis for both verification and optimization. Parametric dataflow models aim at providing suffi-cient dynamicity to model new applications, while at the same time maintaining the high level of analyzability needed for efficient real life implementations. This paper presents a new compilation flow that targets para-metric dataflows. Built on the LLVM compiler infrastructure, it of-fers an actor based C++ programming model to describe paramet-ric graphs, a compilation front-end providing graph analysis fea-tures, and a retargetable back-end to map the application on real hardware. This paper gives an overview of this flow, with a spe-cific focus on scheduling. The crucial gap between dataflow mod-els and real hardware on which actor firing is not atomic, as well as the consequences on FIFOs sizing and execution pipelining are taken into account.The experimental results illustrate our compila-tion flow applied to compilation of 3GPP LTE-Advanced demodu-lation on a heterogeneous MPSoC with distributed scheduling fea-tures. This achieves performances similar to time-consuming hand made optimizations