https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63734
--- Comment #1 from Marc Glisse <glisse at gcc dot gnu.org> --- My first impression is that this is a target issue, we should be able to call TARGET_VECTORIZE_VEC_PERM_CONST_OK from gimple. Either the target should avoid calling gen_reg_rtx when we are only calling the predicate, or it should do what is necessary to make gen_reg_rtx not fail. I am surprised because I would expect it to be reproducible (with a different testcase) without my patch, since vector lowering calls can_vec_perm_p. Does anyone have a different interpretation of what is happening?
