https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89798
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> --- Note that in theory we can handle vector sizes up to 1<<(1<<10) given we store TYPE_VECTOR_SUBPARTS logarithmic and have 10 bits for the exponent... If we want to limit it we should document the limit if it is not so already.
