Embedding data mappers with distributed memory machine compilers
R. Ponnusamy, Joel Haskin Saltz, Raja Das, Charles H. Koelbel, Alok Choudhary · ACM SIGPLAN Notices · 1993
In scalable multiprocessor systems, high performance demands that computational load be balanced evenly among processors and that interprocessor communication be limited as much as possible. Compilation techniques for achieving these goals have been explored extensively in recent years [3, 9, 11, 13, 17, 18]. This research has produced a variety of useful techniques, but most of it has assumed that the programmer specifies the distribution of large data structures among processor memories. A few projects have attempted to automatically derive data distributions for regular problems [12, 10, 8, 1]. In this paper, we study the more challenging problem of automatically choosing data distributions for irregular problems.