https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109085
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|RESOLVED |NEW
Resolution|FIXED |---
--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
The only difference with -fexcess-precision=standard
Is after:
fdivrs 32(%esp)
We do this:
fstps 12(%esp)
flds 12(%esp)
Which is exactly what you expect, that is a truncation to 32bit float from long
double.
The reason why it "worked" before was there is an implict truncation when
passing to a function call ...