抄録
This paper proposes a hierarchical macro-dataflow computation scheme in a Fortran program. In addition to usually used the parallelism among loop iterations, the hierarchical macro-dataflow computation scheme allows us to hierarchically exploit the near fine grain parallelism among statements inside basic block, and the coarse grain parallelism among basic blocks, loops, or subroutines inside a main routine, subroutines, and loops. This paper describes the hierarchical definition of macrotasks, the parallelism extraction scheme among macrotasks defined inside a higher level layer macrotask and a scheduling scheme that assigns hierarchical macrotasks on hierarchical processor clusters. A result is also described of the performance of hierarchical macro-dataflow computation evaluated on a distributed shared memory and centralized memory multiprocessor system OSCAR.
本文言語 | English |
---|---|
ページ | 44-49 |
ページ数 | 6 |
出版ステータス | Published - 1995 1月 1 |
イベント | Proceedings of the 1995 IEEE Pacific RIM Conference on Communications, Computers, and Signal Processing - Victoria, BC, Can 継続期間: 1995 5月 17 → 1995 5月 19 |
Other
Other | Proceedings of the 1995 IEEE Pacific RIM Conference on Communications, Computers, and Signal Processing |
---|---|
City | Victoria, BC, Can |
Period | 95/5/17 → 95/5/19 |
ASJC Scopus subject areas
- 信号処理
- コンピュータ ネットワークおよび通信