[Bug tree-optimization/93271] [11/12/13/14 regression] SRA producing wrong code on denormals

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=93271

--- Comment #15 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
It is even worse with `int128_t` and `long double` due to `long double` having
padding bits (and a padding byte). So it is not only denormals but even bits in
the padding bits/byte that might get lost.