A Loop-level Analysis and Optimal Sub-sequence Mining

Compiler pipeline is divided into three phases frontend, optimization and backend. Each and every phase has its own importance and research scope. This thesis work is based on the optimization phase of compiler pipeline. Inside the optimization phase, we focus on loop level optimizations, as most of the execution time of the program is taken by loops. It is well understood that the search space of possible optimizations is large. So, building the precise optimizations that give good performance involves a lot of manual effort and expert advice. In my work, the focus is to find the patterns in loops, analyze their behaviour, and observe their interactions with optimization(s). If it is possible to understand the patterns in loops then it could be helpful to design optimization heuristics with Machine-Learning approaches. In the second chapter, we make a detailed study of unoptimized loops from the Fotonik benchmark, where we analyze the access patterns of these loops. The third chapter of this report briefly explains the interactions of loops with different optimization sequences. In the final chapter, we find the similarity patterns in loops