https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119244
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Guess the bitint lowering pass could handle lowering of INTEGER_TYPE with TYPE_PRECISION > 64 && TYPE_PRECISION <= 128 on !targetm.scalar_mode_supported_p (TImode) targets fairly easily at least in cases where it doesn't need libgcc support. What needs library support is multiplication, division/modulo and conversion from and to floating point. But then each backend needs to decide on the ABI how to pass that as argument, return it, what alignment it has.
