Visit ShanghaiTech University | 中文 | How to find us
HOME > News and Events > Events
Optimization of Control Divergences on GPGPU Platforms
Date: 2018/5/20             Browse: 91

Speaker:     Prof. Li Shen. NUDT

Time:          10:00—11:00, May 20 

Location:    Room 1A-200, SIST Building

Host:          Prof. Shu Yin

Abstract:

General purpose graphics processing units (GPGPUs) which work in a single-instruction, single-data (SIMD) mode have been widely used in current high-performance computing systems. Control divergences caused by branches is a very commonly occurring performance problem in GPGPU in which the execution of diverging branches is serialized to execute only one control flow path at a time. To reduce the performance loss caused by control divergence, some dynamic warp formation (DWF) techniques have been proposed to combine threads having same control flow path in a new warp. This report will introduce two difference DWF techniques, Dynamic Ray Shuffling (DRS) which is proposed to eliminate control divergences in Ray Tracing, and partial DWF which can avoid the unnecessary warp formation.

Bio:

Li Shen received the BS, Master and PhD degrees in computer science and technology from the National University of Defense Technology (NUDT), Changsha, China, in 1997, 2000, and 2003, respectively. Currently, he is a Professor of the School of Computer, NUDT. His research interests include programming model and compiler design, high performance processor architecture, virtualization technologies, and performance evaluation and workload characterization. He is a member of CCF and ACM.

SIST-Seminar 18035