An issue for the SC: horrible documentation quality of GCC

[Richard Kenner]

    As there is so often, there is an answer that is perfectly obvious, but  
    not helpful in practice.

    Have a cost function that determines how good the optimization of a  
    program is.

    Run all optimizations in some predetermined sequence.

    Compare the cost function of the output to the one of the input.

    As long as there was some improvement, do it again.

    Then use the best result.

    Compile times? What do you mean, compile times?

Even aside from compile times, since some optimizations are performed
before register allocation and some after, there's no such iteration possible.