https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107920
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> --- It is specifically svld1rq_s8 folding into: # VUSE <.MEM> _5 = MEM <vector(16) signed char> [(signed char * {ref-all})x_3(D)]; Which is done by svld1rq_impl::fold (aarch64-sve-builtins-base.cc:1214).