Re: [gentoo-dev] [RFC] Adding CPUFLAGS USE_EXPAND variable to the profiles

[Martin Schlemmer]

On Fri, 2006-07-07 at 17:46 +0200, Kevin F. Quinn wrote:
> On Fri, 7 Jul 2006 16:20:08 +0200
> Danny van Dyk <kugelfang@g.o> wrote:

> Diego's proposal essentially generates CPU_SUBMODEL automatically from
> CFLAGS - which could be the default behaviour if CPU_SUBMODEL is not
> set.  That way we have the best of both worlds; people who are happy to
> let the system determine the configure options from the compiler
> architecture can do so, those who want to control things in more detail
> can do that as well.
>

I snipped your proposal, which I will reread better later on, but sounds
not too bad if the glimpse was true.

The big issue with Diego's proposal though that most of us for x86 have
issues, is that you tie configure time optimisations that in theory
should be good with most compilers, with gcc's potshots that may or may
not be good.  Sure, you might get away with it these days, because
either bad stuff are filtered, or patched away, but it really is
essentially not the same thing.

I might for example with gcc-4.1.1 rather want gcc to do all
optimization, as it does a better job than the coders do, but with 3.4.6
gcc that sucks at sse2 (ok, apparently this should be fixed with patches
Mike backported, but still), I want what the developer coded mmx/sse
code.

The other side of it as well, is for new cpu's you might have to disable
custom configure enabled mmx/sse/etc in general, as they break with the
code (think when p4 was released).

Sure, maybe adding auto detecting for USE="mmx sse sse2 etc" if they are
not -mmx/-sse/etc can be a cool feature, but that is totally different.

Hopefully that was clear - if not, point out what I should try to
elaborate on.


-- 
Martin Schlemmer