BiELL: A bisection ELLPACK-based storage format for optimizing SpMV on GPUs
详细信息 查看全文
- 作者：Cong Zhenga ; zhengcong55@126.com" class="auth_mailAuthor Vitae ; Shuo GubAuthor Vitae ; Tong-Xiang Guc ; txgu@iapcm.ac.cn" class="auth_mail ; gu_tx@sina.com" class="auth_mailAuthor Vitae ; Bing YangaAuthor Vitae ; Xing-Ping LiucAuthor Vitae
- 关键词：Sparse matrix&ndash ; vector multiplication ; GPU ; BiELL ; BiJAD
- 刊名：Journal of Parallel and Distributed Computing
- 出版年：July 2014
- 年：2014
- 卷：74
- 期：7
- 页码：2639-2647
- 全文大小：1904 K
文摘
- •
SpMV is one of the most important high level operations for basic linear algebra.
- •
We will show how to achieve high performance for general SpMV on GPUs.
- •
BiELL format and its improvement BiJAD format were proposed.
- •
They have better load balance, access the data coalescing and thus higher performance.
- •
They perform better especially when the number of non-zero elements per row varies a lot.