Parallel Functional Computation
Final rept. 15 Jun 1986-14 Aug 1989
COLORADO STATE UNIV FORT COLLINS
Pagination or Media Count:
The magnitude of problems that people need to solve using automatic computation often exceeds the processing power of available sequential computers. However, most algorithms contain components that can execute concurrently instead of serially, and vendors have provided a variety of machines capable of parallel processing. But expected reductions in execution times did not always occur owing to ineffective utilization of available parallel hardware or excessive interprocessor communication. Programmers could sometimes manually optimize program structures to solve these problems, but this work is extraordinarily difficult, and not practical for large programs. Research into automatic parallelization follows two paths. First, because of the large investment in extant software, attempts continue toward compilers for conventional languages that map ordinary programs onto parallel architectures efficiently. Systems of this kind work best when programmers use certain easily recognizable forms in source programs and understand underlying parallel hardware. Second, research continues in this and other projects to design, implement, and evaluate new languages for which automatic parallel execution is more easily attainable.
- Computer Programming and Software