Hi, > Performance is such a reason in my book, as I said, in just about any test > I've done clang-generated code performs consistently (and sometimes probably > significantly) worse than gcc-generated code does. I don't know to what > extent clang does auto-vectorisation, but I've been impressed by the progress > made in that domain by gcc (I've already seen an auto-vectorised version of a > "scalar" video format conversion function from Perian outperform the > hand-coded SSE2 version by almost a factor 10). Fortunately we're no longer > living in times where a few percents in performance gain or loss were > actually noticeable in common use scenarios! > I also find gcc's error messages to be much more readable (though less so in > the more recent versions).
What clang and gcc versions have you compared ? Without these, any comparison is meaningless. Fwiw, in my tests gcc 4.8 and clang 3.4 are very similar. Cannot agree with you on the compiler messages though. Clang's are just more useful, as far as i am concerned.. The bottom line though is this is all irrelevant. You simply cannot reliably use gcc (or to be more precise libstdc++), on os x 10.9 onwards, without most likely running into problems. This ship has sailed and no amount of arguments as to why gcc is better is going to change anything. Chris > _______________________________________________ > macports-users mailing list > macports-users@lists.macosforge.org > https://lists.macosforge.org/mailman/listinfo/macports-users _______________________________________________ macports-users mailing list macports-users@lists.macosforge.org https://lists.macosforge.org/mailman/listinfo/macports-users
