Parallel compilers analyze sequential programs, in particular their loops, to detect hidden parallelism for automatic reconstructing of sequential programs into parallel subtasks. Up to now, many researchers have concentrated on how to find efficient and precise parallel compilers for shared-memory multiprocessor systems to achieve high speedup. Parafrase-2, a famous parallel compiler developed at Illinois University, aims at exploring program transformation for parallel compilers. In this paper, Parafrase-2 is first traced; the data structures and the logical views of scalar analysis, data dependence analysis, and parallelization analysis passes along with their relevant relations between passes are then presented. To develop a prototype parallel compiler and evaluate its performance, Parafrase-2 was ported on an Acer Altos 10000 system under Mach OS. Furthermore, some improvable issues are also presented. Finally, concluding remarks and future work are also given.
Proceedings of the National Science Council, Republic of China, Part A: Physical Science and Engineering 18(5):450-462