Runtime autotuning of tile LU factorization for CPU/GPU hybrid environments
Project/Area Number |
26400197
|
Research Category |
Grant-in-Aid for Scientific Research (C)
|
Allocation Type | Multi-year Fund |
Section | 一般 |
Research Field |
Foundations of mathematics/Applied mathematics
|
Research Institution | University of Yamanashi |
Principal Investigator |
|
Project Period (FY) |
2014-04-01 – 2017-03-31
|
Project Status |
Completed (Fiscal Year 2016)
|
Budget Amount *help |
¥4,420,000 (Direct Cost: ¥3,400,000、Indirect Cost: ¥1,020,000)
Fiscal Year 2016: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
Fiscal Year 2015: ¥650,000 (Direct Cost: ¥500,000、Indirect Cost: ¥150,000)
Fiscal Year 2014: ¥2,600,000 (Direct Cost: ¥2,000,000、Indirect Cost: ¥600,000)
|
Keywords | タイルアルゴリズム / タイルサイズチューニング / CPU-GPU実装 / LU分解 / ピボット選択 / ハイパフォーマンスコンピューティング |
Outline of Final Research Achievements |
The purpose of this research is to implement the tile algorithms for matrix decomposition efficiently on a CPU/GPU computing environment. This implementation makes it possible to speed up the LU decomposition for a large-scale dense matrix. For this purpose, data structures for adaptive tile size tuning, efficient task scheduling method, construction of the performance model and new pivoting strategy were examined. Among them, satisfactory results were obtained regarding the tile size tuning and the performance model. We also implemented the tile algorithm for matrix decomposition on the GPU supercomputer TUBAME 2.5. About this research, 13 oral presentations and two international conference papers with peer review were given.
|
Report
(4 results)
Research Products
(18 results)