Direct Parallelization of Call Statements: A Review

摘要

This paper gives a brief review on the problem of handling call statements in a multiprocessor environment. The approach is to try first the direct parallelization of each call, and then expansion only if needed. Direct parallelization of a call statement means treating the call as a single entity that is represented by a single node on the dependence graph. It requires a careful analysis of the effects of procedures and translation of the knowledge about a called procedure into information on variables of the calling procedure; a general dependence testing method is also needed. We describe direct parallelization, compare it with expansion, and compare this dependence test with the dependence tests currently in use. 14 refs. (ERA citation 13:013838)