基本信息
浏览量:511
职业迁徙
个人简介
The coming decades of high-performance computational mathematics will be increasingly dominated by heterogeneous computing based on hybrid multiprocessors with of a mix of conventional CPU cores and high-throughput GPU cores. This trend is driven by power constraints and the need for ever-increasing performance. Although these new heterogeneous systems can deliver high performance with low energy requirements, new algorithms must be developed to exploit them. The most challenging problems for these systems require a mix of regular and irregular computation. My research into direct methods for sparse linear systems lies within this problem domain and is also a critical component for many applications in computational mathematics.
My experience makes me well-poised for future research in GPU-based algorithms. As a world leader in algorithmic research for sparse matrix computations, my work combines graph-theoretic methods and numerical techniques to create algorithms for solving problems in computational science that arise across a wide range of applications. I incorporate my novel theory and algorithms into robust library-quality open-source software, which is widely used in industry, academia, and government labs. In the past decade, I have published more software in the ACM Transactions on Mathematical Software than any other author (9% of the algorithmic output of that journal).
A primary thrust for my current and future work focuses on creating parallel algorithms for sparse multifrontal LU, QR, and Cholesky factorization for hybrid multicore CPU/GPU multiprocessors. The workflow in these algorithms is an irregular tree, where the nodes are the bulk of the work (regular operations on dense frontal matrices of varying sizes), and the edges are the irregular assembly of data from child to parent. The GPUs operate on many frontal matrices at a time, and data is assembled from child to parent frontal matrix without moving it to the CPU. No other sparse direct method employs these strategies for GPU computing. By creating widely-used high-performance algorithms and software that encapsulate these ideas, this research will extend the capability of heterogeneous computing to a wide domain of applications in computational science and mathematics.
In recognition of my research efforts in high-performance
Awards & Honors
Fellow of the Society for Industrial and Applied Mathematics (citation: for contributions to sparse matrix algorithms and software, including the University of Florida Sparse Matrix Collection).
Fellow of the Association for Computing Machinery (ACM), class of 2015.
Fellow of the Institute of Electrical and Electronics Engineers (IEEE), class of 2016.
computing via GPU-based algorithms, NVIDIA has designated Texas A&M as a CUDA Research Center.
研究兴趣
论文共 310 篇作者统计合作学者相似作者
按年份排序按引用量排序主题筛选期刊级别筛选合作者筛选合作机构筛选
时间
引用量
主题
期刊级别
合作者
合作机构
2023 IEEE HIGH PERFORMANCE EXTREME COMPUTING CONFERENCE, HPEC (2023): 1-9
IPDPSpp.322-332, (2023)
引用0浏览0EIWOS引用
0
0
ACM Transactions on Mathematical Softwareno. 3 (2023): 1-30
引用0浏览0EI引用
0
0
2022 IEEE High Performance Extreme Computing Conference (HPEC)pp.1-8, (2022)
加载更多
作者统计
合作学者
合作机构
D-Core
- 合作者
- 学生
- 导师
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn