Classification and transformation of dynamic dataflow programs

Dataflow programming has been used to describe signal processing applications for many years, traditionally with cyclostatic dataflow (CSDF) or synchronous dataflow (SDF) models that restrict expressive power in favor of compile-time analysis and predictability. Dynamic dataflow is not restricted with respect to expressive power, but it does require runtime scheduling in the general case. Fortunately, most signal processing applications are far from being entirely dynamic, and parts with static behavior need not be dynamically scheduled. This paper presents a method to automatically analyze and classify blocks of a dynamic dataflow program within more restrictive dataflow models when possible, and to transform the blocks classified as static to improve execution speed by reducing the number of FIFO accesses. We used this method on actors of two dynamic dataflow descriptions of an MPEG-4 part 2 decoder, and study how classification and transformation increases decoding speed.