https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122635
--- Comment #1 from Robin Dapp <rdapp at gcc dot gnu.org> --- The length is not 0 when vectorizing, the 0 length is introduced by unrolling. We don't unroll a VLA loop which is why we see the issue with VLS and not with VLA. This part works as designed. We are not deleting the dead stores, though, which is an issue. I'll have a look but it might still take some days.
