copysignl.c

[Bernhard Reutner-Fischer]

On Tue, Mar 16, 2010 at 09:10:40PM +0100, Peter Korsgaard wrote:
>>>>>> "Nuno" == Nuno Pereira <fxss at iol.pt> writes:
>
>Hi,
> Nuno> This fix allows to compile BR with software floating point support
> Nuno> with shared libgcc.
>
> Nuno> If you keep software floating point support enabled and disable shared
> Nuno> libgcc, it compiles ok but when you enable shared libgcc, compilation
> Nuno> blows due to an invalid reference to copysignl.c.
>
>Thanks for the description. uClibc people, is this a known issue?

See gcc thread cited below, yes.
And no,
emitting libcalls to math functions and not linking against libm is just
a gcc bug and not a call to make libc a superset of libm

I've locally changed that for good (and better, size-wise) for shared
setups since they (end up) want(ing) IEEE m anyway in usual setups.
>
> >>  Nuno> Hi,
> >>  Nuno> In the latest release of Buildroot, I think that fix should be included:
> >> 
> >>  Nuno> http://lists.busybox.net/pipermail/buildroot/2010-February/031987.html
> >> 
> >> What exactly does this fix? I'm using BR on ppc daily, and never seen
> >> this problem.

default sanity is imposed by 405 soft-fp mmu C99, to keep your eyebrows
down.

thanks,
Bernhard