A program partitioning method for parallelizing compiler on workstation clusters

In this paper, we propose a program partitioning method for Me parallelizing compiler. Our method is designed for workstation clusters. The most important issue for making parallel processing successful in workstation clusters is decreasing the inter-process communication frequency and overhead. In our method, we adopt the function as a unit of process generation, which makes the granularity of generated process coarser. In order to illustrate the parallel executability among functions, we introduce Me dependency value. in our program partitioning algorithm, the process generation depends on this dependency value. Namely, in our algorithm, the decreased processing time is estimated by the dependency value, and transforms a function into or process only when the processing time of whole program is decreased.