https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97678
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
Ah, the vectorized epilogue is wrong since we do not represent the initial
value there correctly. We still have the {0, 0} invariant in the corresponding
SLP tree.
Hmm.
So previously we're relying on the actual PHI node(s) preheader value plus
eventually honoring LOOP_VINFO_MASK_SKIP_NITERS.
