[Bug target/88892] [8/9 Regression] Double-to-float conversion uses wrong rounding mode when followed by memcpy

[segher at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88892

--- Comment #11 from Segher Boessenkool <segher at gcc dot gnu.org> ---
Good point.  So we cannot use stfs (etc.) as float_truncate+store ever, not
even with full fast math options.