Palm: an integrated parallelism enhancement environment with static-dynamic scheduling

While everyone agrees that algorithm, compiler, and architecture should operate hand in hand to produce the most efficient parallel code, a unified research effort leading to an environment that a user can quickly use to map a 'dusty deck' as well as software in newer languages to an efficient code for a variety of commercial compilers and architectures, has been lacking. The authors have undertaken an integrated approach leading to an environment that may be used for both important classes of architectures: shared memory, and private memory MIMD machines in a language independent manner. Various utilities permit measurement of potential parallelism in an algorithmic step, perform source code modification to assist the compiler in utilizing the embedded parallelism, and tune the code to a specific architecture. The Static Dynamic Scheduler, which is a part of the environment estimates the processor requirements of basic blocks of given program and allocates the processors partially at compile time and partially at run time, to obtain a good tradeoff between speedup and utilization.<<ETX>>