Profiling optimized code: a profiling system for an HPF compiler

High Performance Fortran (HPF), a portable data-parallel language, is based on a high-level model which abstracts programming details away from the user. To achieve high performance, the HPF compiler must optimize the code, which may result in a significant change to the original code structure. Because the performances of the optimized and non-optimized codes differ, profiling HPF programs with compiler optimizations must be supported. This requires additional support in the compiler to report the profile data with respect to the original code as opposed to the transformed code. We developed a profiling system which maintains a history of code transformations, maps the original code to the optimized code, and reports profile data based on the original code. A version of the profiling system was successfully implemented for the Fujitsu HPF compiler and tested for several benchmarks on the Fujitsu AP1000.