A parallelization framework for recursive tree programs

Abstract

The automatic parallelization of "regular" programs has encountered a fair amount of success due to the use of the polytope model. However, since most programs are not regular, or are regular only in parts, there is a need for a parallelization theory for other kinds of programs. This paper explore the suggestion that some "irregular" programs are in fact regular on other data and control structures. An example of this phenomenon is the set of recursive tree programs, which have a well defined parallelization model and dependence test.