Decomposing functional programs for parallel computation

摘要

An approach is presented for automatically decomposing functional programs for execution on a multiprocessor system. A program is translated into combinator code and it is compiled into fixed program code for evaluation by one or more stack machines. The decomposition of the resulting target code is handled dynamically at runtime. This capability is introduced through the use of a set of combinators whose defining properties play a direct role in the allocation, synchronization and deallocation of processors. The authors have developed a fixed program multiprocessor simulator and are conducting performance evaluation studies to determine the effectiveness of this approach to parallel computation. Results from these studies are given.