Dataflow-based Design and Implementation of Image Processing Applications

Abstract

Dataflow is a well known computational model and is widely used for expressing the functionality of digital signal processing (DSP) applications, such as audio and video data stream processing, digital communications, and image processing. These applications usually require real-time processing capabilities and have critical performance constraints. Dataflow provides a formal mechanism for describing specifications of DSP applications, imposes minimal data-dependency constraints in specifications, and is effective in exposing and exploiting task or data level parallelism for achieving high performance implementations.

To demonstrate dataflow-based design methods in a manner that is concrete and easily adapted to different platforms and back-end design tools, we present in this report a number of case studies based on the lightweight dataflow (LWDF) programming methodology. LWDF is designed as a "minimalistic" approach for integrating coarse grain dataflow programming structures into arbitrary simulation- or platform-oriented languages, such as C, C++, CUDA, MATLAB, SystemC, Verilog, and VHDL. In particular, LWDF requires minimal dependence on specialized tools or libraries. This feature --- together with the rigorous adherence to dataflow principles throughout the LWDF design framework --- allows designers to integrate and experiment with dataflow modeling approaches relatively quickly and flexibly into existing design methodologies and processes.