On the calling behaviour of procedures

Abstract

If a procedure P is relatively nonrecursive with respect to a procedure Q, i.e., P is always preceded by Q in the calling graph of the program, then P does not need an activation record of its own — it can be integrated into that of Q. Thus, the efficiency of a call to P is between that of a strictly recursive and a nonrecursive one.

An algorithm is presented which classifies each procedure (or function) in a program as being either nonrecursive, relatively nonrecursive or strictly recursive (i.e., recursive but not relatively nonrecursive) and associates each relatively nonrecursive with a strictly recursive one.

The method is based on finding dominators in the (transformed) calling graph.