Parallel processing scheme for a FORTRAN program on a multiprocessor system OSCAR

A parallel processing scheme is proposed for a FORTRAN program on a multiprocessor system named OSCAR (optimally scheduled advanced multiprocessor). The scheme combines parallel processing of fine grain tasks (each of which consists of an assignment statement), macro-data-flow computation (which uses parallelism among coarse grain tasks), and the traditional loop concurrentization. A parallelizing compiler that accomplishes the proposed scheme has been implemented on OSCAR. OSCAR has been designed to effectively support task-scheduling, which has been an obstacle to the realization of efficient parallel processing. A performance evaluation of the scheme on OSCAR is also described.<
>