Re: compiling atlas2 on athlon
This hasn't come back to me from the lists in the 16 hours since I sent it, so
let's try again:
Adam C Powell IV wrote:
> Daniel Jacobowitz wrote:
>
> > On Tue, Oct 17, 2000 at 07:09:27PM -0400, Camm Maguire wrote:
> > > > This reminds me, atlas doesn't build on PPC. debian/config.expect needs
> > > > the attached patch, and then it fails while trying to compile
> > > > tune/sysinfo/Linux/muladd.c: the compiler eats up all available memory
> > > > (96 MB) and swap (135 MB), then X crashes and boots me back to gdm!
> > > >
> > >
> > > An update here. This is definitely a compiler bug on this platform.
> > > Removing the default -funroll-loops default optimization option delays
> > > compilation failure somewhat further, but still winds up with an
> > > "Illegal instruction" (!) produced in the assembly. Even with just
> > > -g, the xfindCE binary segfaults. I'll try to find time to file a bug
> > > against the compiler, but it looks hopeless for atlas on PPC for now.
> > >
> > > Perhaps I could try gcc 2.7.3, if it existed on PPC. Anyone have any
> > > experience which might indicate that this would be worth a try? Or
> > > any PPC users care to share an insight as to what compiler weakness is
> > > most likely being touched on here?
> >
> > I'll let you know - there are probably going to be new compiler
> > packages in a week or two, and they might address this. GCC 2.7.3 does
> > not exist on powerpc, and never has (that we admit to anymore).
> >
> > A small testcase would be appreciated.
>
> Here's the compile command (and result) for muladd.c which is pretty small:
>
> /usr/bin/gcc -DL2SIZE=262144 -I/usr/src/packages/atlas-3.0/include
> -I/usr/src/packages/atlas-3.0/include/Linux -DAdd__ -DStringSunStyle
> -fomit-frame-pointer -O3 -funroll-all-loops -c muladd.c
> VM: killing process cc1
> gcc: Internal compiler error: program cc1 got fatal signal 9
>
> I've put the cpp output in http://lyre.mit.edu/~powell/debs/muladd.E (it should
> show up there within a few seconds).
>
> Zeen,
>
> -Adam P.
>
> Welcome to the best software in the world today cafe!
Reply to: