Efficient and Accurate Whole-Program Alias Analysis for ARM Executable Code

• Main Authors: Yi-Fang Chung, 鍾宜芳
• Other Authors: Peng-Sheng Chen
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/82536630740263457263

碩士 === 國立中正大學 === 資訊工程研究所 === 101 === Alias analysis is used to determine whether two memory references access the same memory location. Most researches have focused on high-level analysis, but rarely on low-level analysis. When doing low-level alias analysis, we have most program information (e.g., system libraries, and 3rd-party libraries) that cannot be obtained in high-level alias analysis. In addition, low-level alias analysis is important for many compiler optimizations and post-link optimizations. However, large amount of memory usage cause analysis extremely slowly when processing an inter-procedural, low-level alias analysis. According to the observations, we found the configuration of abstracting memory address, and strategies of memory information table and data-flow analysis significantly affect the memory usage and performance of the alias analyzer and analytical precision. In this thesis, we study how to appropriately configure three factors in an inter-procedural, low-level alias analysis. An inter-procedural, low-level alias analysis is implemented based on Diablo, a retargetable link-time binary rewriting framework. The target architecture is ARM processors. Different configurations are used to evaluate the impacts on analysis performance, memory usage, and accuracy. For the tested benchmark programs, the experimental results show that appropriately setting in abstracting memory address, not propagating memory information table among basic blocks, and appropriate meet operation can have the analytical result whose accuracy is 2.95 times compared with Debray's approach. The analysis can be finished in 7 minutes with 133MB memory usage.