On Monday 26 May 2003 9:10 pm, Jesse Keating wrote:
> On Monday 26 May 2003 18:18, Steve Snyder wrote:
> > Also: shouldn't you have a "-O2" (or something) in amongst those
> > compile switches?
>
> Thats automatic in the rpmbuilding.
>
> Another thing you might be seeing, is that it may take longer to compile
> athlon binaries than i686 binaries. The funny thing is that you will
> gain very very small amounts of speed increase, if any at all. You've
> got to ask yourself, why are you recompiling all the packages? The ones
> that really do benifit from athlon/i686 like opts. are already provided
> by Red Hat compiled for this.
So the only packages that will benefit from CPU-approriate optimizations are
the kernel, GLibC and OpenSSL? Just those out of the 1344 packages ("rpm
-qa | wc -l") installed on my system? I assume that is a correct
paraphrase of your position since those are the only files in the
~/en/os/i686 directories on RH's ftp site.
The GCC developers are going to be really upset when they hear how little
fruit their efforts have produced.
I too rebuilt all my RHL v9 SRPMS files. Why? Because I want software that
will more fully utilize the advances made on microprocessors over the past
twenty years. I understand that Red Hat must build 80386-compatible
binaries for distribution because they don't know what hardware the code
will actually be run on.
I, on the other hand, know that the code will be running on Xeon processors.
It seems that the -march=pentium4 switch generates stable code (though the
-msse2 code gen is iffy in GCC v3.2.x), so why not better match the
software to my hardware?
