The Community for Technology Leaders
Green Image
Multiprocessor systems have become common place, but little software takes advantage of their capabilities. Automatic parallelization is particularly attractive as it enables sequential code to exploit parallel hardware and realize improved performance, without additional programmer effort. This article demonstrates that automatic parallelization techniques are now mature enough to parallelize many numeric programs written in both Fortran and C. Using these techniques, the SPEC92fp and SPEC95fp benchmarks were successfully parallelized and run on an 8-processor Digital AlphaServer 8400 machine to obtain the highest recorded SPEC92fp and SPEC95fp ratios. The capabilities of state-of-the-art parallelizing compilers should be taken into account in future processor design. A multiprocessor in combination with a parallelizing compiler may outperform approaches to processor design which attempt to exploit increasing levels of instruction-level parallelism.
Multiprocessing, parallelization, numeric programs, compilers

C. S. Wilson et al., "Multiprocessors from a Software Perspective," in IEEE Micro, vol. 16, no. , pp. 52-61, 1996.
92 ms
(Ver 3.3 (11022016))