Formal approach to guarantee the correctness of compiler optimization using temporal logic Department of Mathematical and Computing Sciences

Optimization is a very important phase of compilation. Depending on the circumstances, it can improve the performance of programs by double-digit percentages. It is essential that the compiler optimizer is implemented so that it does not change the semantics of programs it is applied to. However, many optimizations are too complex to o er such correctness. Guaranteeing the correctness of an optimization of a real program is still an unresolved issue. In this thesis, We propose two techniques to address this problem. One technique involves a Java-compiler optimizer that uses temporal logic and the second involves a compiler-optimizer veri er for the existing COINS (COmpiler INfraStructure) compiler, again using temporal logic. The Javacompiler optimizer analyzes a Java program using a type of temporal logic and optimizes the program according to the results of this analysis. The compiler-optimizer veri er checks whether optimizations implemented for normal program languages such as C or Java have been executed correctly by using temporal logic. we use a temporal logic called CTL-FV, a type of CTL (Computational Tree Logic) that uses free variables (FV) in the formulas and can handle past and future temporal operators. Several research works have analyzed and optimized programs using temporal logic. However, no evaluation of the optimization time or execution time of these implementations has been undertaken for any real programming language until now. In our Java-compiler optimizer, We present a system that generates a Java optimizer, based on CTL-FV temporal logic, from our original domain-speci cation language. This language includes conditions for optimization and rewrite rules that will be applied when the conditions are satis ed. The condition part can express typical optimization rules very naturally and concisely, and the optimizers that are generated run e ciently. By adding rewriting conditions to allow for temporary variables and