[Bug target/68211] Free __m128d subreg of double

["glisse at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68211

--- Comment #6 from Marc Glisse <glisse at gcc dot gnu.org> ---
(In reply to Jakub Jelinek from comment #4)
> If the upper bits of the register can contain arbitrary garbage, then
> keeping it there might result in e.g. floating point exceptions being raised
> (it could be even a sNAN).  Of course a different thing is if we can prove
> what is in those upper bits and be sure it doesn't do any harm, or if the
> operations on it later on are masked.

For
  double f(double x){return __builtin_sqrt(x);}
we generate
        vsqrtsd %xmm0, %xmm0, %xmm0
with -O -fno-math-errno.
I don't see what makes the rounded version different from the non-rounded
version.