On 13 Oct 2007, Dave Chapman stated:
> I think I've solved the problem by adding the following lines to 
> gcc-4.0.3/gcc/config/arm/t-arm-elf and rebuilding gcc:
>
> MULTILIB_OPTIONS    += mcpu=arm9e/mcpu=arm7tdmi/mcpu=arm1136jf-s
> MULTILIB_DIRNAMES   += armv5te armv4t armv6
>
> This is the same file that is patched by rockboxdev.sh to enable building a 
> thumb version of libgcc.a, so we could incorporate this
> change by modifying that patch and rename it to something generic like 
> "rockbox-arm-multilib-4.0.3.diff" (as it specifies all of the
> ARM multilib options we need for Rockbox) instead of the current 
> "thumb-interwork-4.0.3.diff".
>
> But I don't completely understand the multilib feature, so would welcome 
> confirmation from someone who does that what I've done is
> sensible, and if it isn't sensible, how should we do it?

That is indeed how you turn on multilib support.

(FWIW I'm using GCC 4.2-branch-tip to build ipod rockboxes and have no
trouble, at least using -O. The last time I used it, -O2 had periodic
trouble that I haven't got around to tracking down... I should try with
GCC HEAD and see how it flies...)

-- 
`Some people don't think performance issues are "real bugs", and I think 
such people shouldn't be allowed to program.' --- Linus Torvalds

Reply via email to