A parameterisable and scalable Smith-Waterman algorithm implementation on CUDA-compatible GPUs
スポンサーリンク
概要
- 論文の詳細を見る
2009 IEEE 7th Symposium on Application Specific Processors (SASP) : San Francisco, CA, USA, 2009.07.27-2009.07.28This paper describes a multi-threaded parallel design and implementation of the Smith-Waterman (SM) algorithm on compute unified device architecture (CUDA)-compatible graphic processing units (GPUs). A novel technique has been put forward to solve the restriction on the length of the query sequence in previous GPU implementations of the Smith-Waterman algorithm. The main reasons behind this limitation in previous GPU implementations were the finite size of local memory and number of threads per block. Our solution to this problem uses a divide and conquer approach to compute the alignment matrix involved in each pairwise sequence alignment, as it divides the entire matrix computation into multiple sub-matrices and allocates the available amount of threads and memory resources to each submatrix iteratively. Intermediate data is stored in shared and global memory on the fly depending on the length of sequences in hand. The proposed technique resulted in up to 4.2 GCUPS (Giga Cell Updates per Second) performance when tested against the SWISS-PROT protein database, which is up to 15 times faster than a equivalent optimised CPU-only implementation running on a Pentium4 3.4GHz desktop computer. Moreover, our implementation can cope with any query or subject sequence size, unlike previously reported GPU implementations of the Smith-Waterman algorithm which makes it fully deployable in real world bioinformatics applications.
論文 | ランダム
- F-0516 純チタンの切削加工 : 表面あらさの向上について(S40-2 難削材の加工(2))(S40 難削材の加工)
- 新時代における地域金融機関のリテール営業戦略
- 10.糖尿病治療における運動指導の実際 : 第61回日本体力医学会関東地方会 : 招待講演
- Real-Time Gesture Recognition Under Dynamic Viewpoints(Gestures)
- 労働運動最前線 女性いきいき--全労連女性部