Exploiting parallelism across program execution: a unification technique and its analysis

A new technique for source-to-source transformation of sequential programs is described. It is shown that the transformed programs so generated provide significant speedups over the original program on vector processors and vector multiprocessors. The parallelism that arises when multiple instances of a program are executed on simultaneously available data sets is exploited. This is in contrast to the existing approaches that aim at detecting parallelism within a program. The analytic model is used to prove the optimality of the complete first policy for block selection for a class of program graphs known as nonregressive graphs. Analytic and simulation models of the technique clearly indicate the speedups that could be achieved when several data sets are available simultaneously, as is the case in many fields of interest