Re: [AD] Pentium Pro optimizations under *nix |
[ Thread Index |
Date Index
| More lists.liballeg.org/allegro-developers Archives
]
- To: "Allegro Developers list" <conductors@xxxxxxxxxx>
- Subject: Re: [AD] Pentium Pro optimizations under *nix
- From: Evert Glebbeek <eglebbk@xxxxxxxxxx>
- Date: Sun, 6 Oct 2002 12:10:10 +0200
- Invalid-uucp-from: From eglebbk@xxxxxxxxxx Sun Oct 6 21:20:39 2002
- Organization: University of Amsterdam
> How about passing the relevant options directly to gcc then ? I just
re-read
> the manual and there are plenty of variants in the new versions: i386,
i486,
> i586 (=pentium), i686 (=pentiumpro), k6, k6-2, k6-3, pentium-2,
pentium-3,
> pentium-4, athlon, athlon-xp, athlon-mp.
>
> We could have --enable-optimizations=xxx for the -mcpu switch
> and --enable-exclusive-optimizations=xxx for the -march switch, the
default
> being --enable-optimizations=pentium.
That would be even easier to maintain, of course.
The only pitfalls I see involve non-intel targets, where -mcpu=pentium
doesn't make much sense. I guess the solution would be to set a flag when
a --enable*-optimizations option is given and let the script pick the
default compiler switch if the flag is not set.
The other pitfall is that -mcpu and -march don't appear to be
cross-platform switches. For instance, corresponding switches for Sparc
targets would seem to be -mtune and -mcpu (!) respectively. At least that
is the impression I get from the gcc docs.
Still, Intel targets (and in particular Linux) are the most immediate, so
I'll focus on those first. I'll play around with things some more and post
back when I get it to work.