Digital Signal Processing Reference
In-Depth Information
selects both a function
from the function repertoire and the corresponding input and output ports. Note that
a function that is selected from the function repertoire may read arguments and write
results from non-unique input ports and to non-unique output ports, respectively, as
is also the case with polyhedral process networks [ 73 ] .
This allows to separate function selection and binding, so that reading, executing,
and writing can proceed in a pipelined fashion. Although standard blocking read
and blocking write synchronization is possible, the SBF actor allows for a more
general deterministic dynamic approach. In this approach, the actor behavior is
divided into a channel checking part and a scheduling part. See also Sect. 5 . Inthe
channel checking part, channels are visited without empty or full channel blocking .
A visited input channel C in returns a C in .
In the SBF actor, the controller's selection function
μ
1 signal when the channel is not empty, and
a C in .
0 signal when the channel is empty. And similarly for output channels. These
signals indicate whether or not a particular function from the function repertoire
can fire. In the scheduling part, a function can only be invoked when the channel
checking signals allow it to fire. If not, then the function will not be invoked. As
a consequence, the actor is blocked. Clearly channel checking and scheduling can
proceed in parallel.
9
Summary
This chapter, has reviewed several DSP-oriented dataflow models of computa-
tion that are oriented towards representing dynamic dataflow behavior. As signal
processing systems are developed and deployed for more complex applications,
exploration of such generalized dataflow modeling techniques is of increasing
importance. This chapter has complemented the discussion in [ 31 ] , which focuses
on the relatively mature class of decidable dataflow modeling techniques, and builds
on the dynamic dataflow principles introduced in certain specific forms [ 15 , 21 ] .
Acknowledgements In this work, Bhattacharyya has been supported in part by the US Air Force
Office of Scientific Research. The authors also thank Marc Geilen (m.c.w.geilen@tue.nl) and
Sander Stuijk (s.stuijk@tue.nl), both from the Eindhoven University of Technology, for their
contributiontoSect. 6 .
References
1. Annevelink, J.: HIFI: A design method for implementing signal processing algorithms on
VLSI processor arrays. Ph.D. thesis, Delft University of Technology, Department of Electical
Engineering, Delft, The Netherlands (1988)
2. Backus, J.: Can programming be liberated from the von Neumann style? A functional style and
its algebra of programs. Communications of the ACM 21 (8), 613-641 (1978)
 
 
Search WWH ::




Custom Search