https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127055
--- Comment #13 from Matthias Kretz (Vir) <mkretz at gcc dot gnu.org> --- OK, I still don't understand md files (though I really would like to). FWIW, the patch I posted resolves my issue (except for spurious failures to inline the function with the single vround* instruction).
