Retargetable ModelBased Code Generation in Ptolemy II

1
Retargetable Model-Based Code Generation in
Ptolemy II
Man-Kit Leung, Tony Huang, Christopher Brooks,
Prof. Edward A. Lee
Eighth Biennial Ptolemy Miniconference 2009
Starmac Quadrotor

• Domain Generality
• The code generator provides an interface for
  various models of specification. These
  specification languages are often tailored for
  the application domains which relies on
  properties and analyzability of the language. In
  Ptolemy II, we called these MoCs. The code
  generator is designed to interface with different
  MoCs and nicely reuse the software components for
  handling heterogeneous models.
• The Generation Process
• Front-end - interface with Ptolemy model
  components (directors, actors, and etc.).
• Middle-end - analyses or partial evaluations that
  can be performed to optimize the generated code.
• Back-end - interface with different targets (HW
  platforms SW API's).
• Various techniques are employed, including
  composition of partial-evaluated results. We use
  an adaptor pattern that associates meta code and
  a target code template with model component. We
  use type polymorphism to specialize the generated
  function code while retaining software resue.

Mission This research investigates the
infrastructure of a retargetable and
domain-general C code generator for actor models
and use the code generator to program different
platforms. The main mission is to carefully
transform and compile models specified in formal
models of computation (MoCs). Retargetability One
very desireable property of the code generator
is its ability to generate code for multiple
targets. C code generation differs from
traditional compiler backend in that software
libraries and APIs are also considered
targetable.
CT SR
Software synthesis is the key component in the
correct-by-construction theory. It bridges the
gap between the simulation environment and
implementation code. It allows rapid
experimentation and shortens the development
cycle.
PN SDF FSM
This is an example heterogeneous model. The
top-level is modeled as a Kahn Process Network
(PN) that contains two hierarchical components
(Actor1 and Actor2) which in turn contain
refinement models.
Since actors directed by PN are executed in
separate threads of execution, we can gain
speedup from executing them on parallel
architectures.
MPI (Nersc Supercomputer)
Acknowledgements This work was supported in
part by the Center for Hybrid and Embedded
Software Systems (CHESS) at UC Berkeley, which
receives support from the National Science
Foundation (NSF awards 0720882 (CSR-EHS PRET)
and 0720841 (CSR-CPS)), the U. S. Army Research
Office (ARO W911NF-07-2-0019), the U. S. Air
Force Office of Scientific Research (MURI
FA9550-06-0312), the Air Force Research Lab
(AFRL), the State of California Micro Program,
and the following companies Agilent, Bosch,
Lockheed-Martin, National Instruments, and Toyota.
The heterogeneity of MoCs is accomplished through
hierarchical composition.
Luminary
i-Robot Create
Composing PN with other MoCs give us good tools
in managing and reasoning concurrency. For
example, By composing PN, SDF and FSM, we
effectively control concurrency while retaining
determinacy and understandability in the
subcomponents..

• Current Work
• Target code generation for the Synchronous/Reactiv
  e (SR) domain for general platforms.
• Target code generation for Giotto, a time
  triggered MoC, with the OpenRTOS and PRET
  platforms.
• Research into generalizing the relationship and
  descriptions of different targets.

The model specifies the control logic for the
iRobot Roomba. The controller combines together
the SDF and FSM domains.