Instrumentation-driven model detection and actor partitioning for dataflow graphs

Dataflow modeling offers a myriad of tools to improve optimization and analysis of signal processing applications, and is often used by designers to help design, implement, and maintain systems on chip for signal processing. However, maintaining and upgrading legacy systems that were not originally designed using dataflow methods can be challenging. Designers often convert legacy code to dataflow graphs by hand, a process that can be difficult and time consuming. In this paper, the authors developed a method to facilitate this conversion process by automatically detecting the dataflow models of the core functions from bodies of legacy code. They focus first on detecting static dataflow models, such as homogeneous and synchronous dataflow, and then present an extension that can also detect dynamic dataflow models. Building on the authors' algorithms for dataflow model detection, they present an iterative actor partitioning process that can be used to partition complex actors into simpler sub-functions that are more prone to analysis techniques.