A New Prospect in Finding Optimization Opportunities for Multicore Architectures

• Main Authors: Ching-Chi Lin, 林敬棋
• Other Authors: 劉邦鋒
• Format: Others
• Language: en_US
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/90131566715680542588

碩士 === 臺灣大學 === 資訊工程學研究所 === 98 === There have two groups of works in optimizing program execution in the literature – static and dynamic program optimization. To our best knowledge, neither of these optimizations, while looking for optimization opportunities, considers interactions among threads in multi-core architecture. Therefore we would like to develop techniques that can identify the presence of thread interactions and use it to guide possible optimization. We observe that interaction among threads, like competition for shared cache, can lead to “unstable” execution performance. That is, the same part of program will have very different performance characteristics, therefore we identify those parts of program as dynamic optimization opportunity, so that they can be optimized for better performance. We propose a simple and efficient sampling method that analyzes performance variance among basic blocks, so as to differentiate “unstable” and “stable” basic blocks. The results from the analysis can be used as a reference to determine which parts of the program on which dynamic optimizer should make extra efforts during execution. By mapping EIP of each sample back to its basic block, we are able to choose representative basic block for each interval during execution, and compare the performance of each thread, so as to calculate the performance variance of each　basic block. This sampling technique can also be applied to single-threaded programs.