Automatic privatization for parallel execution of loops

Abstract

Privatization of data is an important technique that has been used by compilers to parallelize loops by eliminating storage-related dependences. The code can be executed on multi-processors machines in reduced period of time. In this paper, we present an approach to automatic privatization of variables involved in data dependences that permits for extracting loop parallelism. The input of the algorithm is a set of relation dependences, the output is a parallel loop when appropriate. The scope of the applicability of the approach is illustrated by means of the NAS Parallel Benchmark suite. Received results are compared with those produced by the tool Pluto. Future work is outlined. © 2012 Springer-Verlag Berlin Heidelberg.