Parallel Fortran AnalyZER

点击翻译按钮获取中文摘要

Research into the area of automatically parallelizing programs for distributed memory architectures is inadequate. Current models of parallelization ignore the problems associated with the distribution of data structures and the corresponding communication overhead. The objective of this research is to establish mechanisms for the automatic distribution of FORTRAN programs for a loosely-coupled, message passing parallel processing system. The Parallel Fortran AnalyZER (PFAZER) will use three novel approaches to reduce the overhead associated with interprocessor communication. First, PFAZER will utilize an enhanced cost model which will more accurately estimate the overhead of interprocessor commuication in a distributed processing environment. Second, PFAZER will utilize a software cache management system which will provide access to ambiguously referenced variables. Last, PFAZER will incorporate a new data dependence decision algorithm which will reduce the number of data dependencies in the source program which will further reduce interprocessor communication.