The Affinity Analysis Of Program Reference Data Based On Locality Theory

The rapid growth of CPU and the slow growth of the memory make the speed gap between CPU and memory bigger and bigger in recent years, which leads to the memory system becoming a performance bottleneck. Cache is widely used in modern computer architecture to reduce its effect, but the cache miss will cause the extra time of memory access, In fact, whether cache can be fully utilized depends on program locality and data reuse mode. Generally array regrouping and structure splitting in program locality optimization techniques can improve program locality, but it needs to change the internal organization or adjacent relation of the program memory data, The affinity analysis of program reference data can guide the change of the internal organization or adjacent relation of the program memory data. The affinity analysis of program reference data can be the critical part of program locality optimization.This paper mainly studies the affinity analysis algorithm of program reference data, additionly, before applying to the affinity analysis algorithm we usually need to gather program performance data through program analysis, such as program reference data. The main points we are about to discuss in this paper are as follows: Firstly, in order to obtain the program reference data, this paper proposes a program analysis framework based on the Pin. The main principle of program analysis is carried out at different levels of code instrumentation based on dynamic instrumentation system Pin to obtain program analysis results. The main modules of this framework include program reference data acquisition, instruction statistics and extraction of call graph. Secondly, when analysing acquired program reference data, the affinity analysis algorithm of program reference data based on the reuse distance sequence similarity is proposed. The algorithm firstly calculates reuse distance sequences of program reference data and then the reuse distance sequences similarity are calculated, finally the affinity of program reference data can be obtained.And at last according to the module of the program reference data acquisition in program analysis framework based on the Pin and data affinity analysis algorithm proposed above, an affinity analysis tool of program reference data is completed and the validity of the affinity analysis tool is verified by using the test case.