On the Machine-independent Target Language for Parallelizing Compilers

Although shared memory machines provide one of the easier models for parallel programming, the lack of standardization for expressing parallelism on these machines makes it difficult to write efficient portable code. The Guide TM Programming System is one solution to this problem. In this paper, we discuss a back-end to the Polaris parallelizing compiler that generates Guide TM directives. We then compare the performance of parallel programs expressed in this way to programs automatically parallelized by a machine's native compiler, and by code expressing parallelism with native directives. The resulting performance is presented and the feasibility of this directive set as a portable parallel language is discussed. 1 Introduction Shared memory machines provide one of the easier conceptual models for parallel programming. Although this makes programming these machines relatively straight forward, there has been a lack of standardization of expressing parallelism. In moving from one..