A Comprehensive Environment for Automatic Packaging and Scheduling of Parallelism

Abstract

The state of the art in vectorizing and parallelizing compilers has overcome the language barrier and offers a viable (and not necessarily exclusive) alternative to parallel programming. The evolution of parallel programming languages has not kept up with the evolution of parallel architectures [KDLS86], [GGKM83], [Chen83], [MiUc84]. Even though this may change in the future, currently we are confined to using enhanced versions of existing languages which do not provide either the power or the flexibility for programming modern parallel supercomputers [ANSI86], [GuPL88], [MeRo85]. Under this reality parallel programming remains still a highly complex and empirical art, even for the skillful programmer. Parallelizing compilers offer a solution to both of these problems. A serial program can be automatically vectorized and/or parallelized, and in many cases automatic concurrentization can proceed to the maximum possible degree.