Dynamic and Multidimensional Dataflow Graphs

Much of the work to date on data ow models for signal processing system design has focused decidable data ow models that are best suited for onedimensional signal processing. In this chapter, we review more general data ow modeling techniques that are targeted to applications that include multidimensional signal processing and dynamic data ow behavior. As data ow techniques are applied to signal processing systems that are more complex, and demand increasing degrees of agility and exibility, these classes of more general data ow models are of correspondingly increasing interest. We begin with a discussion of two data ow modeling techniques — multi-dimensional synchronous data ow and windowed data ow — that are targeted towards multidimensional signal processing applications. We then provide a motivation for dynamic data ow models of computation, and review a number of speci c methods that have emerged in this class of models. Our coverage of dynamic data ow models in this chapter includes Boolean data ow, the stream-based function model, CAL, parameterized data ow, and enable-invoke data ow. 1 Multidimensional synchronous data owgraphs (MDSDF) In many signal processing applications, the tokens in a stream of tokens have a dimension higher than one. For example, the tokens in a video stream represent images so that a video application is actually three-dimensional. Static multidimensional Shuvra S. Bhattacharyya University of Maryland, USA e-mail: [email protected] Ed F. Deprettere Leiden University, The Netherlands e-mail: [email protected] Joachim Keinert Fraunhofer Institute for Integrated Circuits, Germany e-mail: joachim.keinert@iis. fraunhofer.de 899 In S. S. Bhattacharyya, E. F. Deprettere, R. Leupers, and J. Takala, editors, Handbook of Signal Processing Systems, pages 899-930. Springer, 2010. 900 Shuvra S. Bhattacharyya, Ed F. Deprettere, and Joachim Keinert (MD) streaming applications can be modeled using unidimensional data ow graphs (see Chapter 30), but these are at best cyclo-static data ow graphs, often with many phases in the actor’s vector valued token production and consumption patterns. These models incur a high control overhead that can be avoided when the application is modeled in terms of a multidimensional data ow graph that may turn out to be just a multidimensional version of the unidimensional synchronous data ow graph. Such graphs are called multidimensional synchronous data ow graphs (MDSDF).