跳到主要导航 跳到搜索 跳到主要内容

Optimizing GCC with ICC as the Performance Goal for SPEC CPU 2006 and 2017

  • Hanke Zhang
  • , Puzhou Hou
  • , Xiaohua Shi*
  • *此作品的通讯作者
  • Beihang University

科研成果: 书/报告/会议事项章节会议稿件同行评审

摘要

SPEC CPU 2006 and 2017 are widely recognized CPU benchmark suites designed to evaluate the performance of system processors, memories, and compilers across various platforms. This paper tries to explore why ICC has significantly higher runtime efficiency than GCC on some benchmarks by analyzing the specific optimizations performed by ICC through various methods, aiming to figure out the reasons behind its high scores as well as the generality and portability of these optimizations, like loop fusion, SCEV and parallelization, structure peeling, vectorization, loop tiling, etc. The optimized GCC is more than 5.8% and 47% faster than the original GCC 12.3 for the rate metric on SPEC CPU 2017 and 2006, respectively. The performance gap between the optimized GCC and ICC on SPEC CPU 2006 has been reduced from 36.4% to 4.2% (speed metric) and from 51.8% to 3.0% (rate metric). It also surpassed ICC for both rate and speed metrics on SPEC CPU 2017, on an Intel Core i9-12900KF CPU.

源语言英语
主期刊名Computational Science and Computational Intelligence - 11th International Conference, CSCI 2024, Proceedings
编辑Hamid R. Arabnia, Leonidas Deligiannidis, Farzan Shenavarmasouleh, Soheyla Amirian, Farid Ghareh Mohammadi
出版商Springer Science and Business Media Deutschland GmbH
167-180
页数14
ISBN(印刷版)9783031951299
DOI
出版状态已出版 - 2025
活动11th International Conference on Computational Science and Computational Intelligence, CSCI 2024 - Las Vegas, 美国
期限: 11 12月 202413 12月 2024

出版系列

姓名Communications in Computer and Information Science
2506 CCIS
ISSN(印刷版)1865-0929
ISSN(电子版)1865-0937

会议

会议11th International Conference on Computational Science and Computational Intelligence, CSCI 2024
国家/地区美国
Las Vegas
时期11/12/2413/12/24

指纹

探究 'Optimizing GCC with ICC as the Performance Goal for SPEC CPU 2006 and 2017' 的科研主题。它们共同构成独一无二的指纹。

引用此