On the relation between functional and data parallel programming languages

Abstract

Data parallel programming is becoming an increasingly important tool for exploiting parallelism in data-intensive applications, especially on SIMD and vector computers. Many algorithms appearing in such applications are very succinctly expressed in data parallel languages: this indicates that data parallel programming can be a powerful abstract programming paradigm rather than just a syntax for explicit programming of SIMD computers. The data parallel languages in practical use today are, however, exponents of exactly the latter point of view: even though they incorporate some elements of abstraction, their semantics are all to some extent based on a SIMD execution model. Therefore it is hard to use these languages to express algorithms in the problem domain in an abstract, machine-independent way. This is likely to make programming in these languages more errorprone and programs less portable than if they had been designed with a more clean-cut abstract semantics.