A comparison of synchronous and cycle-static dataflow

Abstract

We compare synchronous dataflow (SDF) and cyclo-static dataflow (CSDF), which are each special cases of a model of computation we call dataflow process networks. In SDF actors have static firing rules: they consume and produce a fixed number of data tokens in each firing. This model is well suited to multirate signal processing applications and lends itself to efficient static scheduling, avoiding the run-time scheduling overhead incurred by general implementations of process networks. In CSDF which is a generalization of SDF actors have cyclically changing firing rules. In some situations, the added generality of CSDF can unnecessarily complicate the scheduling. We show how higher-order functions can be used to transform a CSDF graph into a SDF graph, simplifying the scheduling problem. In other situations, CSDF has a genuine advantage over SDF: simpler precedence constraints. We show how this makes it possible to eliminate unnecessary computations and expose additional parallelism. We use digital sample rate conversion as an example to illustrate these advantages of CSDF.