rtems-cvs toolchain

[Ralf Corsepius]

On Mon, 2008-04-21 at 22:48 -0700, Till Straumann wrote:
> Nigel Spon wrote:
> > On 22/04/2008, at 11:00 AM, Joel Sherrill wrote:
> >   
> >> It would be nice to get a 405 BSP back in the tree but that
> >> one was using (God help me) .. the other exception model
> >> and was the last one.  So it was removed in the interest
> >> of forward motion.
> >>     
> >
> >
> >
> > Hmm. Forward motion is good, but I seem to have gone too far forward  
> > and wound up in a bog.
> >
> > After a lot of updating and unpacking, I find I can't compile gcc  
> > 4.3.0 because it needs mpfr and gmp. I have the latest versions of  
> > these (2.3.1 and 4.2.2) and they build, but they default to 64-bit (- 
> > m64 -mtune=k8). The gcc build doesn't like this:
> >   
> I unpacked the mpfr and gmp sources and then created
> symlinks from the gcc-4.3.0 topdir
> 
>    ln  ../mpfr-2.3.1 ./mpfr
>    ln   ../gmp-4.2.2 ./gmp
> 
> and then it built fine (linux-x86_64 host, powerpc-rtems target)
Correct. You can build them one-tree style (untar them into the GCC source tree).
GCC's configure then picks them up automatically.

Another alternative is building and __installing__ them separately in
advance to building GCC. 

AFAIS, Nigel either seems to have missed to install them, rsp. missed to
make these libraries available to the run-time environment (Probably
using a non-standard install-directory, LD_LIBRARY_PATH, ldsoconfig or
similar).

Also, so far, all platforms I've built gcc-4.3.0 for, ship a
sufficiently new gmp, but some of them ship an insufficient mpfr.

For them, I am building mpfr one-tree style.

Ralf