https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112325
liuhongt at gcc dot gnu.org changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |liuhongt at gcc dot gnu.org
--- Comment #3 from liuhongt at gcc dot gnu.org ---
BB vectorizer relies on the backend support of .REDUC_PLUS for reduction, but
loop vectorizer can manually do reduction. That's why it's not vectorized after
cunrolli.
After adding reduc_plus_scal_v4si, it's vectorized.
Looks like we need to support
reduc_plus_scal_{v4si,v8si,v16si,v8hi,v16hi,v32hi}
Similar for reduc_{and,ior,xor}_scal_m.