[Clfs-dev] Embedded clfs problems and workarounds

[Andrew Bradford]

On Thu, Feb 10, 2011 at 8:38 PM, Joe Ciccone <jciccone at gmail.com> wrote:
> All of the gcc builds in embedded should have --disable-multilib if they
> do not already.
>
> If the target is i?86-*-*-* with or without --disable-multilib should
> produce a compiler for 32bit only. If the target is x86_64-*-*-* with
> --disable-multilib will produce a compiler for 64bit only. This is the
> desired behavior in the embedded builds.
>
> AFAIK GCC automatically will build a multilib compiler if the target
> triplet is for a 64bit architecture that has a 32bit subset.
>
> We really should have --disable-multilib everywhere in the embedded
> builds. Since there is no multilib, there should be no reason to have to
> specify -m32/-m64 via BUILD. The compiler should be built to produce the
> appropriate output rather then being told on the command line at every
> instance. Easily accomplished in non-multilib builds.
>
> --
> Joe Ciccone

Joe,

For x86, I'd agree.  I'm adding --disable-multilib to both gcc-static
and gcc-final in my repo.

But for ARM (and maybe MIPS with --with-divide?), using
--disable-multilib without specifying hard/soft floating point or
arm/thumb instructions could be an issue.  I'm not sure what GCC would
do there.  Do you know?

I'd like to add --with-float and --with-mode to ARM first, before
adding --disable-multilib, but I'm being conservative because I don't
know the full impacts.

I'm reading the GCC configure docs at: http://gcc.gnu.org/install/configure.html

-Andrew