https://gcc.gnu.org/bugzilla/show_bug.cgi?id=56160
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Keywords| |missed-optimization
Target| |x86_64-linux-gnu
Component|middle-end |target
--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
This is just IV-OPTs going wrong.
One thing which I will note does improve the code is doing:
__m128 n = _mm_cmpneq_ps(res, res);
int need = _mm_movemask_ps(n);
if (UNLIKELY(need)) {
comp_mult(r, a, b, i);
} else
_mm_store_ps((float*)&r[i], res);