|In reply to Comment 33 (Johan Rönnblom):|
>I don't think Don wanted to compare the JIT's. I think he wanted to
know which system would be faster for him, personally. If he's using
certain ImageEngineer filters a lot, then that's a good test.
If ImageEngineer is not native software, he will be comparing JIT, if he wants that or not. If it is native software, he will compare programs compiled by different compilers or at least different compiler backends. That is questionable.
>Compiling is a good test if you're compiling a lot of big sources. Etc
etc. :-) [...] compiler speed is not that interesting to me.
You don't seem to get my point. It doesn't matter what comes out of it (a new highscore or a compiled program). It doesn't matter one bit if you compile or not. The idea is about finding a meaningful test for a component like Trance. What matters is that you use a test that stresses the component you want to test in many different yet reasonable ways rather than with a few arbitrary lines of code. The result will then be meaningful to many people, whether they play games or do something else with the computer. The test you have suggested will be meaningful exactly to one sort of people: those running a blurr filter. A tiny variation of parametsr (other filter) would invalidate the result.