Compositionality in synchronous dataflow: modular code generation from hierarchical SDF graphs

Hierarchical SDF models are not compositional: a composite SDF actor cannot be represented as an atomic SDF actor without loss of information that can lead to rate inconsistency or deadlock. Motivated by the need for incremental and modular code generation from hierarchical SDF models, we introduce in this paper DSSF profiles. DSSF (Deterministic SDF with Shared FIFOs) forms a compositional abstraction of composite actors that can be used for modular compilation. We provide algorithms for automatic synthesis of non-monolithic DSSF profiles of composite actors given DSSF profiles of their sub-actors. We show how different trade-offs can be explored when synthesizing such profiles, in terms of compactness (keeping the size of the generated DSSF profile small) versus reusability (maintaining necessary information to preserve rate consistency and deadlock-absence) as well as algorithmic complexity. We show that our method guarantees maximal reusability and report on a prototype implementation