Compiler Optimization Verification and Maintenance

Due to the complexity of a compiler, it is difficult for compiler developers to prove the correctness of output of intermediate representations or profiling information manually after applying optimizations. Moreover, it is important to choose good test cases to test the optimization code’s changes, however, it is not simple to build separated test cases to cover every possible optimizations. This proposal proposes automatic techniques to prove the correctness of an IR, the correctness of its annotations for correct profiling information, and to generate test cases according to the kinds of optimizations. Moreover, the test case size can be reduced by using bug isolation techniques. We focus our techniques on loop nest optimizations in the GNU C compiler with SPEC-CPU2006 benchmark test suites.