Parallel Executions of Loops in a Multiprocessor UNIX(R) System
01 January 1989
Different iterations of a program loop can be executed concurrently by several processors to speed up the overall computation. One of the major constraints on the parallelization of loop execution is the data dependency among different iterations. A synchronization framework is presented in this paper, and a set of primitives are defined and placed based on data flow analyses. in addition to traditional live variable analysis, we present other types of analysis for placement of synchronization primitives and release of semaphores.