Imperial College London
Portrait Picture

ProfessorAlastairDonaldson

Faculty of Engineering, Department of Computing

Professor of Programming Languages
 
 
 
//

Contact

 

+44 (0)20 7594 8266alastair.donaldson Website

 
 
//

Location

 

422Huxley BuildingSouth Kensington Campus

//

Summary

 

Publications

Citation

BibTex format

@inproceedings{Sorensen:2017:10.1145/3106237.3106265,
author = {Sorensen, T and Evrard, H and Donaldson, AF},
doi = {10.1145/3106237.3106265},
publisher = {ACM},
title = {Cooperative kernels: GPU multitasking for blocking algorithms},
url = {http://dx.doi.org/10.1145/3106237.3106265},
year = {2017}
}
Download

RIS format (EndNote, RefMan)

TY  - CPAPER
AB  - There is growing interest in accelerating irregular data-parallelalgorithms on GPUs. These algorithms are typicallyblocking, sothey require fair scheduling. But GPU programming models (e.g.OpenCL) do not mandate fair scheduling, and GPU schedulers areunfair in practice. Current approaches avoid this issue by exploit-ing scheduling quirks of today’s GPUs in a manner that does notallow the GPU to be shared with other workloads (such as graphicsrendering tasks). We proposecooperative kernels, an extension tothe traditional GPU programming model geared towards writingblocking algorithms. Workgroups of a cooperative kernelarefairlyscheduled, and multitasking is supported via a small set of languageextensions through which the kernel and scheduler cooperate. Wedescribe a prototype implementation of a cooperative kernel frame-work implemented in OpenCL 2.0 and evaluate our approach byporting a set of blocking GPU applications to cooperative kernelsand examining their performance under multitasking.
AU  - Sorensen,T
AU  - Evrard,H
AU  - Donaldson,AF
DO  - 10.1145/3106237.3106265
PB  - ACM
PY  - 2017///
TI  - Cooperative kernels: GPU multitasking for blocking algorithms
UR  - http://dx.doi.org/10.1145/3106237.3106265
UR  - http://hdl.handle.net/10044/1/49875
ER  -
Download