I have built XFree86 at least 3 times in the past week, all with varying
levels of optimization, from -O to -O3 and ALWAYS with -march=pentium3.
All of the builds succeeded, although I had stability problems with -O2
and above. Are you _certain_ this is a compiler bug?
On Tue, 24 Sep 2002, Vallo Kallaste wrote:
> On Mon, Sep 23, 2002 at 03:13:15PM -0700, Kris Kennaway <[EMAIL PROTECTED]> wrote:
> > On Mon, Sep 23, 2002 at 01:55:18PM +0300, Vallo Kallaste wrote:
> > > This isn't a yesterdays problem, I've had this for a month or so.
> > > The problem is explicit declaration of -march=p, use
> > > CPUTYPE=i686 in /etc/make.conf and you get further. The second way
> > > is to not set CPUTYPE at all, logic in /usr/share/mk/bsd.cpu.mk will
> > > set -mcpu to what appears to fit.
> > That's not true; it adds -mcpu=pentiumpro by default to optimize
> > instruction scheduling for 686-class CPUs (without breaking binary
> > compatibility down to i386s), but it doesnt "autodetect" anything, and
> > you'll get better performance on a 686-class CPU by specifying it in
> > CPUTYPE (since you'll then also get pentium pro instructions).
> > -mcpu != -march
> Yes I know and it was what I meant to say. By setting CPUTYPE=p
> excplicitly in /etc/make.conf you'll get the build failure in
> XFree86-4-Server because bsd.cpu.mk will set excplicit -march, not
> -mcpu. But the build error happens _only_ if one sets CPUTYPE=p
> and not in case CPUTYPE=i686, the latter sets -march=pentiumpro, not
> -march=pentium. Hope this is all clear now and sorry about
> confusion, english isn't my native tongue after all.
To Unsubscribe: send mail to [EMAIL PROTECTED]
with "unsubscribe freebsd-current" in the body of the message