An Automated System for Deriving Efficient Parallel Programs

点击翻译按钮获取中文摘要

Under this project a prototype derivation system for scientific programs, designed to facilitate the production of efficient implementations on parallel processors will be developed. The system will be based upon a "wide-spectrum" language that supports all phases of the software development process including specification, design, prototyping, optimization, testing, debugging and maintenance. Building on their previous work on the scientific programming systems - the R programming environment, the R validation system, the Russell compiler and the PFC system for automatic detection of parallelsism - the project team will implement a sophisticated programming environment to provide automatic or semi-automatic support for program transformation and refinement at all stages of the derivation process. It will be possible to augment the environment's standard transformation support by producing libraries of design rules written in the wide-spectrum language itself and mechanisms for validation of refinement steps will be built in at every level. To take advantage of the existing base of scientific software, the system will support the integration of FORTRAN subprograms into programs written in the wide-spectrum language. Furthermore, it will be possible to translate a low-level dialect of the language to FORTRAN 77, extended by the inclusion of primitives for the explicit specification of parallelism. The resulting FORTRAN program, typically consisting of many small procedures, can be optimized to high efficiency using interprocedural analysis and optimization techniques developed for the R project.