https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119979
--- Comment #18 from Oleg Endo <olegendo at gcc dot gnu.org> --- (In reply to Jeffrey A. Law from comment #17) > At this point I don't have a high degree of confidence that we even know the > breadth of breakage -- we know some targets were affected because my tester > flagged them, but no in-depth evaluation of the fallout from HJ's patch. > > That's really unfortunate, particularly since HJ has known about these > problems for nearly a year and he's know that the problems touch the ABI > space. Part of me wants to revert the change, but that's not a great option > at this stage. I've only briefly tried to look at what's going in SH. SH's TARGET_PROMOTE_FUNCTION_MODE uses parts of its own impl of TARGET_PROMOTE_PROTOTYPES. The fix for this issue is to not override TARGET_PROMOTE_PROTOTYPES in SH, but keep everything else. This means, effectively it just disables the newly added call to TARGET_PROMOTE_PROTOTYPES in calls.cc in this commit https://gcc.gnu.org/cgit/gcc/commit/?id=78db4753c9646a372512e6a951fced12f74de0bc So it looks like the original semantics of TARGET_PROMOTE_PROTOTYPES is changed. Perhaps it'd been better to make a new target callback function for this instead.
