https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83800
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |RESOLVED
Resolution|--- |WONTFIX
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I think only the double routines in glibc guarantee 0.5ulp precise results,
libquadmath is based on the quad long double code in glibc generally and
certainly doesn't have such guarantees. I think the sqrtq isn't even based on
that, and just does an approximation using sqrt or sqrtl and 1-2 Newton
iterations. From what you've said, it isn't even off by whole 1ulp, but
something in between 0.5ulp and 1ulp.
