On Sun, Oct 3, 2010 at 10:00 AM, Andreas Färber <andreas.faerber(a)web.de> wrote:
Am 03.10.2010 um 08:54 schrieb Blue Swirl:
> Thanks, applied as r876.
> On Sun, Sep 19, 2010 at 9:39 AM, Andreas Färber <andreas.faerber(a)web.de>
>> Otherwise it chokes on ducell in include/kernel/stack.h.
>> Signed-off-by: Andreas Färber <andreas.faerber(a)web.de>
>> It seems I only mentioned this on the list but never sent a proper
>> Not sure about Linux/ppc or Darwin/x86, so I've limited it to Darwin/ppc
>> for now.
Darwin/x86 was already covered by the pre-existing "if" obviously.
Debian/ppc still doesn't want to install over my SATA controller.
Last night I ran into another related issue though. If I try to compile a
ppc64 version using the regular include/arch/ppc/types.h cell sizes as
indicated on the list, then after this patch and some ppc64 preparations in
config/scripts/switch-arch, include/kernel/stack.h will try to access .lo
and .hi fields although ppc doesn't define ducell to be of 128-bit width.
My thinking was that we may need to split this up into HOST_NEEDS_FAKE_128_T
and TARGET_DCELL_128_BITS? Or should the ppc definition be changed? For now
I've just removed -DNEED_FAKE_128_T from my config.mak.
I'm not sure what the real solution is. So far the fake stuff was only
limited to host forthstrap.
Perhaps we should add a library with arbitrary precision arithmetic
and use that explicitly for ducell math, maybe even for other cases to
avoid any libgcc use.