https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119014
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
That is the normal behavior of extended precision.
Because without -fexcess-precision=16 _Float16 is evaluated in float precision,
the first 42.42f16 is actually 42.42f returned in _Float16 and thus converted
to the _Float16 precision, while in the x != 42.42f16 case there is no explicit
cast or assignment and so you're actually comparing against 42.42f.
You can use x != (_Float16) 42.42f16 if the constant is supposed to be in
_Float16 precision.
