中国邮电高校学报(英文) ›› 2009, Vol. 16 ›› Issue (5): 107-113.doi: 10.1016/S1005-8885(08)60276-9

• Others • 上一篇    下一篇

Structure splitting framework for data-layout optimization

付雄;王汝传   

  1. College of Computer, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
  • 收稿日期:2008-12-08 修回日期:1900-01-01 出版日期:2009-10-30
  • 通讯作者: 付雄

Structure splitting framework for data-layout optimization

FU Xiong (), WANG Ru-chuan   

  1. College of Computer, Nanjing University of Posts and Telecommunications, Nanjing 210003, China
  • Received:2008-12-08 Revised:1900-01-01 Online:2009-10-30
  • Contact: FU Xiong

摘要:

To improve data cache performance, optimizing program data layout by data reorganization has become an important method of decreasing the impact of increasing gap of speed between processor and memory. In this article, a structure splitting framework with an analysis model named structure field relation graph (SFRG) is presented to optimize program data layout. The SFRG can be used to quantify relationship between fields. It helps to find an optimal layout for structure as well as the optimal program data layout. And the data cache performance is improved through SFRG-based structure splitting. Experiments show that this framework is effective in optimizing program data layout and improving the performance of data cache and whole program.

关键词:

program;optimization,;structure;splitting,;data;layout,;cache;performance

Abstract:

To improve data cache performance, optimizing program data layout by data reorganization has become an important method of decreasing the impact of increasing gap of speed between processor and memory. In this article, a structure splitting framework with an analysis model named structure field relation graph (SFRG) is presented to optimize program data layout. The SFRG can be used to quantify relationship between fields. It helps to find an optimal layout for structure as well as the optimal program data layout. And the data cache performance is improved through SFRG-based structure splitting. Experiments show that this framework is effective in optimizing program data layout and improving the performance of data cache and whole program.

Key words:

program optimization;structure splitting;data layout;cache performance