https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87812
--- Comment #4 from Iain Sandoe <iains at gcc dot gnu.org> --- (In reply to H.J. Lu from comment #3) > Without AVX, v32qi should be passed the same way before AVX was > added to compiler. For example, > > If there were AVX1024, GCC would pass y in vector register. When AVX1024 > were disabled, GCC would pass y in memory and issue an ABI change warning. > So LLVM is wrong. ---- As an ABI co-author, perhaps you would be kind enough to file a bug there then? --- For current affected systems (at least freeBSD 10+ and Darwin 11+ I'd say) we will have to figure some work-around.
