Do not define -mlong-double-128 on ppc when --without-fp specified?
Joseph S. Myers
joseph@codesourcery.com
Thu Mar 16 22:12:00 GMT 2006
On Wed, 15 Mar 2006, Roland McGrath wrote:
> Sorry, it's going to be broken until someone volunteers to fill in the
> missing pieces. We do not want to have different glibc ABIs for different
> powerpc configurations.
Note that the function-call ABI for soft-float and hard-float PowerPC is
in any case incompatible, you can't mix soft-float binaries with
hard-float libc or vice versa.
> The glibc ABI for powerpc will use IBM extended
> format for long double from now on, regardless of the particular hardware
> constraints you are configuring for. To restore working support for
> --without-fpu powerpc configurations, GCC needs to be made to support it.
> This will require writing the soft-fp support for IBM extended format that
> GCC will generate calls to.
Rather than trying to fit support for a fundamentally non-IEEE
floating-point format into an IEEE software floating point library, I
think it would be simpler and more compatible with the hard-float case to
use the existing IBM long double support in libgcc (darwin-ldouble.c)
(falling back to soft-fp operations on double) for +-*/, plus simple
functions for the cases the .md file can expand directly (conversions,
negation, comparison). Then, the only soft-fp support you need to add is
for fused multiply-add.
--
Joseph S. Myers
joseph@codesourcery.com
More information about the Libc-alpha
mailing list