Public SourceMage Discussion List

[Sukneet Basuta <sukneet AT gmail.com>]

On Sun, Jan 20, 2013 at 4:55 PM, Javier Vasquez <j.e.vasquez.v AT gmail.com> 
wrote:
> Hi,
>
> I wanted to change architecture from x86_64 to corei7_avx (core i7
> with sandy bridge), so that new casted spells became further
> optimized.
>
> However doing that without any further actions proved wrong.
>
> When getting binutils re-casted (new version showing up on test), it
> actually re-compiled and installed OK, but then gcc got broken unable
> to generate executables at the "configure"stage of compilation...
> Specially gettext no longer was able to compile...

Does setting your architecture to native result in the same problem? I
suggest you just stick to native either way. It will choose the best
optimizations for your architecture. Be aware, however, that
native-x86_64-64 does not force -fpic and -dpic flags (while every
other x64 arch in Source Mage does), so you may run into a few
programs complaining about it. I've fixed it in every spell I've
found, but I didn't test every spell. It's mainly only old programs
that have issues though. If you don't mind using pic, just specify
those additional flags when you select your architecture.

> I remember before doing "sorcery -r" after recent installation, and
> after changing optimization flags, it's recommended to:
>
> 1.-  re-cast linux.
> 2.-  re-cast gcc, g++, perl, gettext, glibc.
> 3.-  run sorcery -r
>
> I'm wondering if the order is OK...  Also if binutils shouldn't be
> re-casted, in some particular order as well.

I always follow the order listed at
http://www.sourcemage.org/projects/source-mage/wiki/InstallGuide#Post-Install

> Any ways.  I wanted to do the same for a pentium-m machine, which was
> originally set as i686, and I moved it recently to pentium-m, but I'll
> move it back to i686 so I can update it...
>
> Let me know id those steps are adequate, or if they require changes.
> Is "sorcery -r" totally required?  Or are there a smaller list of
> spells required to be re-casted and that would be it?

It shouldn't be. AFAIK you can compile pretty much any program with
different optimizations and they should work. binutils is a pretty
crucial package though and everything relies on it, so maybe its the
exception. That said, the wiki recommends casting binutils and gcc
first after selecting your architecture. I have done that quite a few
times and I've never broke gcc. I suspect the culprit is incorrect
optimization flags.