https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126000
--- Comment #2 from vekumar at gcc dot gnu.org --- EVEX is shorter only when the displacement D satisfies: * D is a multiple of the instruction's CD8 scale N, and * (D / N) fits in a signed byte [-128, 127], and * D itself does NOT fit in a signed byte (otherwise VEX already uses disp8).
