Top Up Prev Next Bottom Contents Index Search

1.5 Dataflow Models of Computation

One of the most mature domains included in the current system is the synchronous dataflow (SDF) domain [Lee87a,b], which is similar to that used in Gabriel. This domain is used for signal processing and communications algorithm development, and has particularly good support for multirate algorithms [Buc91]. It has been used at Berkeley for instruction, at both the graduate and undergraduate level [Lee92]. A dynamic dataflow (DDF) domain extends SDF by allowing data-dependent flow of control, as in Blosim. Boolean dataflow (BDF) [Buc93a,b,c] has a compile-time scheduler for dynamic dataflow graphs [Lee91a].

Several code-generation domains use dataflow semantics [Pin92][Mur93]. These domains are capable of synthesis of C code, assembly code for certain programmable DSPs [Won92], VHDL, and Silage [Kal93]. A significant part of the research that led to the development of these domains has been concerned with synthesizing code that is efficient enough for embedded systems [Bha93a,b,c][Bha94a,b][Buc93b,c]. A large amount of effort has also been put into the automatic parallelization of the code [Ha91][Ha92][Sih93a,b], and on parallel architectures that take advantage of it [Lee91b][Sri93].

A generalization of dataflow, called Kahn process networks [Kah74], has been realized by Tom Parks in the PN domain [Par95].



Top Up Prev Next Bottom Contents Index Search

Copyright © 1990-1997, University of California. All rights reserved.