https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66086
--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> --- > note that on x86-64 the used values are 48-bit only and a double provides > enough > accuracy to store them correctly. These kind of assumptions are bad and very unportable. I can think of targets were pa and va are going to be bigger than 48bits.
