A Combined Data and Program Partitioning Algorithm for Distributed Memory Multiprocessors

概要

論文の詳細を見る
In this paper we propose an algorithm to perform data partitioning and program partitioning simultaneously on the Data Partitioning Graph, an intermediate representation for parallelizing compilers. Conventional and, therefore, conservative parallelizing compilers usually activate program partitioning prior to data partitioning. However, on distributed memory multiprocessors, since communication costs change depending on a data partitioning and distribution decision, it is quite difficult to partition a program effectively with consideration of data partitioning. The proposed algorithm resolves this confliction by handling these inseparable partitioning problems simultaneously with a branch-and-bound based scheme.
一般社団法人情報処理学会の論文
1998-11-26