Creating a suite of source code benchmarks to provide a consistent, comparative measure of OpenCL, OpenACC, OpenMP and compute-intensive performance is not simple. The challenges are numerous, for example:
And in the end, you hope you have something that is fair, relevant, and understandable, that will address technology for a period of years.
Such is the case with SPEC ACCEL, which is a new benchmark. The new benchmark suite:
All of this is due to the efforts of a large team of people from around the world, that SPEC would like to thank (with apologies and thanks to those who may not be listed).
SPEC ACCEL benchmarks are based on real-world applications and tests. We therefore thank the authors of those codes.
All the benchmarks in the final OpenCL suite come from PARBOIL (Prof Wen-Mei Hwu's research group at UIUC) and RODINIA (Prof Kevin Skadron's research group at UVA). To these research teams we are very thankful.
For more information regarding the SPEC benchmarks, including information about other contributors, see the documentation in the individual benchmarks directories.
We would also like to thank:
SPEC management, who provided us with resources, and kept the SPEC HPG Subcommittee on their toes:
The SPEC office, who were the ones who really provided us with the resources:
The long time SPEC HPG leaders, who initiated this benchmark and successfully carried it to version 1.0 and 1.1.
Most importantly, thank you to the people who sweated through technical details and porting issues, debated their view points vigorously, and were willing to drive issues to conclusion. Without these people this benchmark suite would truly not exist.
A benchmark "Project Leader" serves as the focal point for porting a program to all the tested platforms, improves the stability of the code, removes code that SPEC does not want to test (such as platform-specific extensions), writes documentation, consults with both the author and the users, and generally tries to give each benchmark its best chance at making it through SPEC's testing process. Thank you to these project leaders:
Thank you also to the following people who worked very closely with the project leaders:
In addition to the above, many other people from the member companies assisted with benchmark development, including but not limited to Alexander Bobyr, Bill Brantley, Huiyu Feng, Oscar Hernandez, Wayne Joubert, Huian Li, Junjie Li, Daniel Neilson, Maxim Perminov, Dave Raddatz, Pavel Shelepugin, Alexy Titov, Veronica Vergara Larrea, Matthis van Waveren, Sandra Wienke, Rengan Xu.
Analysis of the candidate benchmarks was a substantial task. Contributors to profiling and other forms of analysis included: John Stratton, Ke Wang, Brian Whilney, Sandra Wienke.
Cloyce Spradling is the primary author of the SPEC CPU/HPG toolset.
Finally, we thank all of the people behind the scenes in the compiler groups, architecture groups and performance groups who supported their work for SPEC at their respective companies.
Brian Whitney, Release Manager, SPEC HPG Committee
Junjie Li, Secretary, SPEC HPG Committee
Guido Juckeland, Vice-Chair, SPEC HPG Committee
Robert Henschel, Chair, SPEC HPG Committee
Copyright © 2014-2017 Standard Performance Evaluation Corporation
All Rights Reserved