On Wed, Mar 07, 2001 at 09:09:36AM +0000, Antony T Curtis wrote:
> Kris Kennaway wrote:
> >
> > On Wed, Mar 07, 2001 at 09:30:33AM +0100, Hartmann, O. wrote:
> > > Dear Sirs.
> > >
> > > I'm really confused and surprised by the fact, that with the upcoming new
> > > FreeBSD 4.3 in its make.conf we can choose architectural parameters for the
> > > CPU architecture. And I will tell you why.
>
> <snip>
>
> > I've been running NIS on FreeBSD with world and kernel compiled with
> > -march=pentiumpro for about 9 months with no ill effects. In fact
> > I've never heard of reports of bad code generation using the -march
> > settings -- it appears to be very stable. Are you sure you didn't
> > have a nonstandard optimization setting (-O2, etc) as well?
>
> I have world compiled with -O2 with no ill effect (that I have noticed)
> but if I compile the kernel with -O2, it looks okay but has problems
> with talking TCP/IP - seems to generate faulty checksums and some other
> quirks.... To put it simply, stable kernels are only with -O (in my
> experience)
>
> (however, I have compiled KLDs with -O2 with no ill effect)
Yes, this is well known (and documented in /etc/make.conf). I've
*never* heard failure reports associated with using an -march option,
nor seen any myself (all my systems run with -march=<whatever> and
have never shown failures; contrasted with -O2 which inevitably causes
problems each time I'm tempted to try it)
I think Jordan's commit to /etc/make.conf warning people off of using
CPUTYPE was premature.
Kris
PGP signature
