Design and performance analysis of multiprocessor streaming applications

The processing of regularly structured data streams frequently occurs in the context of real-time audio, video, and digital signal processing in consumer electronics devices, communication systems, and medical systems, to name a few. Software programs developed for this purpose are commonly referred to as streaming applications. The steadily growing computational demand of streaming applications increasingly leads to implementations on heterogeneous single-chip multiprocessors. This trend calls for suitable design automation techniques to deal with the complexity of the resulting systems and raises the questions of how (real-time) streaming applications can be efficiently implemented on multiprocessors and how the timing behavior of these applications can be analyzed. In this thesis, these questions are tackled by proposing the distributed operation layer (Dol) design flow that automates the implementation and performance analysis of multiprocessor streaming applications. To this end, a streaming application is specified as a dataflow process network, that is, as a network of actors that explicitly communicate over first-in first-out channels. Then, fundamental properties of dataflow process networks are leveraged to automatically transform this high-level specification into a platform-specific implementation and an analytic performance analysis model.