https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88892
--- Comment #10 from joseph at codesourcery dot com <joseph at codesourcery dot com> --- Use of stfs on a double-precision value without frsp first is worse than simply using the wrong rounding mode; in the case of overflow it does a bitwise defined operation which is pretty useless in floating-point terms (other than to make load/store of a single-precision NaN, even an sNaN, keep the bit-pattern unchanged) (and in the case of an exponent below -149 the results are undefined).
