https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119430
--- Comment #9 from Alexandre Oliva <aoliva at gcc dot gnu.org> --- I'm not sure I've run into the same problem, but the one I'm looking at is in the training stage, in ali.o, one of many that raise an assertion failure from SInput.Get_Source_File_Index.Assertions. Interesting situation: the caller is inlined into expand_sloc (cross-language inlinining, wheee :-), it sets up the static chain record correctly, stores it in r12, and then calls the Assertions nested function. But the call seems to go through an lto-introduced (?) mi thunk that clobbers r12 (used as a scratch register). Very very "ugh!" moment. Not sure yet how to go about this.
