Benny Amorsen wrote:
>>>>>> "MK" == Mark Kirkwood <markir@paradise.net.nz> writes:
>
> MK> Here are the results after building gcc 4.1.2 (repeating results
> MK> for gcc 3.4.6 for comparison). I suspect that performance is
> MK> probably impacted because gcc 4.1.2 (and also the rest of the
> MK> tool-chain) is built with gcc 3.4.6 - but it certainly suggests
> MK> that the newer gcc versions don't like the slice-8 algorithm for
> MK> some reason.
>
> They don't seem to like the old CRC algorithms either. It is quite
> strange, such dramatic performance regressions from 3.x to 4.x are
> rare.
>
Right - I think the regression is caused by libc and kernel being built
with gcc 3.4.6 and the test program being built with gcc 4.1.2.
Rebuilding *everything* with 4.1.2 (which I'm not sure is possible for
FreeBSD at the moment) would probably get us back to numbers that looked
more like my Gentoo ones [1].
Cheers
Mark
[1] Note that the upgrade process for switching Gentoo from gcc 3.4 to
4.1 involves precisely this - build 4.1, then rebuild everything using
4.1 (including 4.1 itself!)