https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119244
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Note, initially libquadmath was used always by libgfortran on targets with some
__float128 support.
Later on libgfortran has been conditionalized, so that on targets where the
sufficient support in libc is present it uses just that (and for backwards
compatibility links --as-needed with libquadmath), and otherwise link with
libquadmath.
Important question is whether the COBOL compiler emits directly into compiled
code calls to the various libc *f128 functions, or calls always libgcobol APIs
and only those APIs then call *f128 functions. In the latter case, this would
be simpler, libgcobol would decide at configure time whether to use libc or
libquadmath APIs.
As for adding _Float128 support to targets which don't currently have _Float128
support, that is mostly a matter of ABI, somebody needs to codify an ABI how to
pass/return those, add needed backend support and port libquadmath.
