[Bug target/120083] Use vcvtneps2bf16 AVX512BF instruction to convert float to __bf16 on -O3 without needing -Ofast

[mcccs at gmx dot com via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120083

--- Comment #5 from mcccs at gmx dot com ---
Ah I see then there's the problem of denormals. I guess I'll report this to
LLVM so that they may cripple naive AI libraries using O3.