https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96600
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |amodra at gcc dot gnu.org
--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Even the j0 > 103 case is wrong. First of all, it is unclear where that 103
comes from, e.g. for the quad version it uses j0 > 111, i.e. two smaller than
the 113 bit precision, but 103 is 3 smaller than 106.
But more importantly, it assumes that such values have no fractional part,
which is not given.
E.g. the upper double could be 0x1.0p+110 and lower 0x1.0f0f0f0f0f0fp+30.
