https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120398
--- Comment #2 from Alexander Monakov <amonakov at gcc dot gnu.org> ---
Right, aarch64 vectorizes at -O3 to the desired form, but not -O2:
.L3:
ldr d1, [x0, x2, lsl 3]
add x2, x2, 1
fmla v31.2s, v1.2s, v1.2s
cmp x1, x2
bne .L3
