ISSN 1000-1239 CN 11-1777/TP

Journal of Computer Research and Development ›› 2021, Vol. 58 ›› Issue (3): 668-680.doi: 10.7544/issn1000-1239.2021.20190728

Previous Articles    

An Ideal Performance Oriented Approach for Cross-Framework Compiler Analysis

Lai Qingkuan1,3, Lü Fang1,2, He Chunlin4, He Xianbo4, Feng Xiaobing1,2,3   

  1. 1(State Key Laboratory of Computer Architecture(Institute of Computing Technology, Chinese Academy of Sciences), Beijing 100190);2(Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190);3(University of Chinese Academy of Sciences, Beijing 100049);4(School of Computer Science, China West Normal University, Nanchong, Sichuan 637009) 
  • Online:2021-03-01
  • Supported by: 
    This work was supported by the National Key Research and Development Program of China(2016YFB0200803), the CCF Research Fund Project of Tencent, the National Natural Science Foundation of China(61802368, 61521092, 61432016, 61432018, 61332009, 61702485, 61872043), the Talent Project of China West National University(17YC149), Nanchong City Major Scientific and Technological Achievement Conversion Project(18SXHZ0386), and the Ministry of Education Industry-University Cooperation Collaborative Education Project(201702049002).

Abstract: Compiler performance is the embodiment that computer system architecture makes full use of its advantages. Compiler optimization is influenced by the machine platform and compiler characteristics. Compiler analysis is carried out between the target compiler and the multiple reference compiler,  and the target platform and the multiple reference platform,  that is to say,  the combination of the compiler and the platform is the foundation of the analysis. Only in the circumstance of multiple combinations,  we can provide the maximum possibility of performance room for improvement and prioritization schemes in detail for the target compiler optimization. However,  adding to the combination of compiler and platform will add to the analysis amount of work which is unable to measure in most cases. For this purpose,  one kind of analytical technique in face of cross-platform and cross-compiler on the strength of the peak value architecture is put forward. On the strength of the peak value architecture set,  we structure the ideal performance space for the target compiler. In the combination of the advantage of fine grit,  we optimize the location technique,  and provide the optimization options with advantages and optimized direction for the target compiler,  so as to implement compiler optimization. In the end,  by means of experiments that we have carried out,  we verify the practical nature and pervasive nature of this analytical technique,  what is more,  we provide the optimized direction for the target compiler (gcc) on Intel platform.

Key words: compiler optimization, peak value architecture, ideal performance space, optimization options with advantages, optimized direction

CLC Number: