The Parameterized Synchronous Dataflow (PSDF)

The Parameterized Synchronous Dataflow (PSDF) domain is an extension of the Synchronous Dataflow (SDF) domain that allows for more extensive reconfiguration of models. The SDF domain uses static analysis of variable dependence to detect cases where rate parameters may change. By default, SDF disallows reconfiguration of all rate parameters. If rate parameters are allowed to change, then SDF checks that rate parameters do not change during execution of the schedule, and declares that inferred rate parameters for external ports change as often as the internal rate parameters.

This domain offers two key extensions:

  1. Dependence analysis is used to determine if the modification to the rate parameters occurs during execution of the SDF schedule. If this test passes, then a parameterized schedule is constructed.
  2. The generated schedule is checked for local synchrony [1], to determine if external rate parameters may change. The correct dependency information for external ports is then declared.

Note that the resulting behavior is identical to the SDF scheduler, with rate parameter changes allowed, except much more efficient, since scheduling on the fly is not necessary during every reconfiguration. Additionally, code can be generated for the PSDF domain that allows for efficient reconfiguration. The added precision of dependency analysis for external rate parameters also means that some hierarchical models that would be ruled out by the SDF checks are allowed.

PSDF uses $PTII/lib/mapss.jar, which is a preliminary and experimental version of the MAPSS (MAryland Package for System Synthesis) package, developed at the University of Maryland. Publications related to the MAPSS project can be found at the link below. An overview of some of the core features in MAPSS is given in [2]. The mapss.jar file includes a basic subset of MAPSS features used to support scheduling in the Ptolemy II PSDF domain. A more complete distribution of MAPSS, including source code and documentation, is planned for the near future.

References

  1. B. Bhattacharya and S. S. Bhattacharyya. Parameterized dataflow modeling for DSP systems. IEEE Transactions on Signal Processing, 49(10):2408-2421, October 2001.
  2. C. Hsu, F. Keceli, M. Ko, S. Shahparnia, and S. S. Bhattacharyya. DIF: An interchange format for dataflow-based design tools. In Proceedings of the International Workshop on Systems, Architectures, Modeling, and Simulation, Samos, Greece, July 2004.

References [1] and [2] are available from the DSP-CAD Group publications page.