https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111751
--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> --- (In reply to JuzheZhong from comment #5) > (In reply to Andrew Pinski from comment #4) > > The issue for aarch64 with SVE is that MASK_LOAD is not optimized: > > > > ic = "\x00\x03\x06\t\f\x0f\x12\x15\x18\x1b\x1e!$\'*-"; > > ib = "\x00\x03\x06\t\f\x0f\x12\x15\x18\x1b\x1e!$\'*-"; > > vect__1.7_9 = .MASK_LOAD (&ib, 8B, { -1, -1, -1, -1, -1, -1, -1, -1, -1, > > -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, > > ... }); > > vect__2.10_35 = .MASK_LOAD (&ic, 8B, { -1, -1, -1, -1, -1, -1, -1, -1, -1, > > -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, > > ... }); > > I don't ARM SVE has issues ... It does as I mentioned if you use -fno-vect-cost-model, you get the above issue which should be optimized really to a constant vector ...