1. Heterochronous Dataflow in Ptolemy II Brian K. Vogel vogel@eecs.berkeley.edu EE249 Project Presentation, Dec. 4, 1999

2. Outline Project goals Project goals Ptolemy II background Ptolemy II background Heterochronous dataflow (HDF) semantics Heterochronous dataflow (HDF) semantics Current results Current results Future work Future work Demo Demo References References

3. Project Goals Implement the heterochronous dataflow (HDF) MOC in Ptolemy II Implement the heterochronous dataflow (HDF) MOC in Ptolemy II –Extend the FSM domain to allow interaction with SDF and HDF domains –Implement *charts interaction semantics for SDF with hierarchical FSMs –Implement *charts interaction semantics for HDF with hierarchical FSMs Create an interesting Ptolemy demo using SDF/HDF with hierarchical FSMs Create an interesting Ptolemy demo using SDF/HDF with hierarchical FSMs

4. Ptolemy II Background Software supporting heterogeneous, concurrent modeling and design. Software supporting heterogeneous, concurrent modeling and design. Component view of design Component view of design Models of computation are implemented in Ptolemy II domains Models of computation are implemented in Ptolemy II domains Written in Java Written in Java

5. Heterochronous dataflow (HDF) semantics HDF [1] is a generalization of cyclo-static dataflow (CSDF) and also of static dataflow (SDF) HDF [1] is a generalization of cyclo-static dataflow (CSDF) and also of static dataflow (SDF) –An HDF actor has a finite number of type signatures –The order in which type signatures are used is, in general, not cyclic

6. Heterochronous dataflow (HDF) semantics (cont.) An HDF actor has an initial type signature when execution starts An HDF actor has an initial type signature when execution starts Can solve balance equations => find an iteration Can solve balance equations => find an iteration The type signature is only allowed to change after the last firing of an iteration The type signature is only allowed to change after the last firing of an iteration

7. Heterochronous dataflow (HDF) semantics (cont.) It may be interesting to combine HDF with FSMs. An interaction semantics of HDF with FSMs is specified in *charts [1] It may be interesting to combine HDF with FSMs. An interaction semantics of HDF with FSMs is specified in *charts [1] *charts is a family of MOCs describing the semantics of hierarchical FSMs combined with multiple concurrency models *charts is a family of MOCs describing the semantics of hierarchical FSMs combined with multiple concurrency models In *charts, an HDF actor can refine to an FSM In *charts, an HDF actor can refine to an FSM –The type signature is determined by the current state –The type signature may change when a state transition occurs –A state may refine to an SDF diagram, HDF diagram, or another FSM (no restriction on number of levels in the hierarchy)

8. Heterochronous dataflow (HDF) properties HDF is more expressive than SDF but retains desirable decidability properties HDF is more expressive than SDF but retains desirable decidability properties –Static schedules are possible –Deadlock, bounded memory are decidable A potential drawback is that a global solution to the balance equations determines when a state transition can occur (could be difficult/non-intuitive to use) A potential drawback is that a global solution to the balance equations determines when a state transition can occur (could be difficult/non-intuitive to use)

9. HDF example b$0 && b$1 b$0 a(2) b(5,2) c(1,3)d(1) b(5)c(1) b(2)c(3) HDF FSM SDF

10. Current results Completed a preliminary implementation in Ptolemy II Completed a preliminary implementation in Ptolemy II –SDF with hierarchical FSMs –HDF with hierarchical FSMs Appears to be fully functional, but more testing is needed to verify this Appears to be fully functional, but more testing is needed to verify this –Follows the semantics of [1] –HDF, SDF, FSM can be combined, with arbitrary level of hierarchy –Ptolemy II expression language is used to specify state transition guard expressions

11. Future work Optimizations Optimizations –Investigate precomputing all schedules before execution begins (currently construct schedules dynamically) –Optimize token transfers Allow HDF actors to have 0-rate ports Allow HDF actors to have 0-rate ports Construct more test cases to verify that the implementation is correct Construct more test cases to verify that the implementation is correct Create interesting demo applications that are a good match for HDF Create interesting demo applications that are a good match for HDF

12. Demo Hierarchical FSMs with SDF: Hysteresis Hierarchical FSMs with SDF: Hysteresis

13. References [1] A. Girault, B. Lee, and E. A. Lee, "Hierarchical Finite State Machines with Multiple Concurrency Models," IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, Vol. 18, No. 6, pp.742-760, 1999.