Software Environments and Scalable High Level Data Structure Extensions of Fortran for Irregular Loosely Synchronous Problems on SIMD and MIMD Parallel Computers

点击翻译按钮获取中文摘要

The work to be supported will be a collaboration between Rice University, Syracuse University and the University of Maryland. It will investigate and develop concepts that will extend Fortran D to irregular, loosely-synchronous problems. Fortran D, extensions of Fortran 77 and Fortran 90 that permit the programmer to specify the distribution of data, a key step in parallelization on distributed memory machines, has already been developed at Rice and Syracuse. The research includes four major components: (1) Extending Fortran D to accommodate a wider class of irregular problems and to make it possible to use automatic data mapping strategies. (2) Investigation of high level data structures, including data trees, synchronous problems. (3) Development of irregular mapping algorithms and their implementation in the PARTI library. (4) Integration of the irregular mapping strategies into the Fortran D compiler under development at Rice and Syracuse