An Efficient Data Distribution Technique for Distributed Memory Parallel Computers

書誌事項

タイトル別名
  • Efficient Data Distribution Technique f
  • 並列化コンパイラ

この論文をさがす

説明

In this paper we propose a linear data distribution technique which extends the traditional BLOCK or CYCLIC distribution for intra-dimension as in HPF to permit partitioning the array elements along slant lines.The array distribution patterns are determined by analyzing the array subscript refrences in loop nests.If the data are distributed along the slant lines then we show the conversion algorithm between global address and local address and the conversion algorithm from global iteration space to local iteration space.Some experimental results on a distributed memory parallel computer CP-PACS show the effciency of our subscript analysis and also show that the proposed data distribution technique can achieve better performance than the traditional data distribution for some scientific applications.

In this paper,we propose a linear data distribution technique,which extends the traditional BLOCK or CYCLIC distribution for intra-dimension as in HPF,to permit partitioning the array elements along slant lines.The array distribution patterns are determined by analyzing the array subscript refrences in loop nests.If the data are distributed along the slant lines,then we show the conversion algorithm between global address and local address,and the conversion algorithm from global iteration space to local iteration space.Some experimental results on a distributed memory parallel computer,CP-PACS,show the effciency of our subscript analysis,and also show that the proposed data distribution technique can achieve better performance than the traditional data distribution for some scientific applications.

収録刊行物

参考文献 (17)*注記

もっと見る

キーワード

詳細情報 詳細情報について

問題の指摘

ページトップへ