学术活动

Sparse Matrix Computations on Modern High-Performance Processors

首页 - 学术活动

报告人：

Weifeng Liu, Professor, Super Scientific Software Laboratory, China University of Petroleum-Beijing

邀请人：

Bastien Vieublé, Doctor & Xin Liu, Professor

题目：

Sparse Matrix Computations on Modern High-Performance Processors

时间地点：

15:00-16:00 September 3(Tuesday), Z311

摘要：

Sparse matrix computations are one of the most important common patterns in numerical linear algebra, and play a fundamental role in scientific and engineering computing. On high-performance processors with modern architectures, e.g., tensor core units and asymmetric multi-cores, sparse matrix computations often face more challenges such as diverse sparse formats, load imbalance, low efficiency in random memory accesses, poor data locality, complex task dependencies, and difficulties in utilizing acceleration units. As the scale of the matrix order being processed continues to expand, addressing these challenges becomes increasingly urgent. This presentation will start by analyzing the computational patterns of sparse basic linear algebra subprograms, discuss how to reduce the spatial cost of storing sparse data and achieve load-balanced computation, and then introduce how to utilize sparse block data structures to reduce random memory accesses and improve data locality. Furthermore, the talk will introduce synchronization-free algorithms to avoid barrier overhead caused by task dependencies and further discuss how to accelerate sparse computation using dense tensor units.

Short Bio: Weifeng Liu is currently a Full Professor at the Super Scientific Software Laboratory of China University of Petroleum-Beijing. Formerly, he was a Marie Curie Fellow at the Norwegian University of Science and Technology. He received his PhD in 2016 from the Niels Bohr Institute of the University of Copenhagen. He has been shortly working as a Research Associate at the STFC Rutherford Appleton Laboratory in 2016. He also has been working as a Senior Researcher in high performance computing technology at the SINOPEC Exploration and Production Research Institute for about six years (2006-2012). He received his BE and ME degrees in computer science, both from the China University of Petroleum-Beijing, in 2002 and 2006, respectively. His research on large-scale sparse direct solver PanguLU received the Best Paper Award of SC ’23. He is a Senior Member of the IEEE and a Member of the ACM and the SIAM. His research interests are in high performance numerical linear algebra, in particular include domain specific architectures, data structures, parallel and distributed algorithms, and mathematical software for sparse matrix computations.