An optimizing compiler for FP*-a data-parallel dialect of FP

The authors present an overview of a compiler for the functional language FP* that generates programs suitable for execution on massively parallel architectures. Data-parallel programs are naturally expressed using FP* functionals, which map efficiently onto massively parallel architectures. However, the creation of large and unnecessary intermediate results is a general source of inefficiencies in functional programs. The FP* compiler uses an optimization technique that significantly reduces such unnecessary copying of data structures, thereby minimizing inter-processor communication in massively parallel computers. The technique uses a set of axioms and inference rules to formally analyze FP* programs. The compiler is specified by a formal compilation scheme which produces efficient data-parallel programs.<
>