http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45505
--- Comment #15 from Martin Jambor <jamborm at gcc dot gnu.org> 2011-01-06 16:41:34 UTC --- I've played around with this a bit more and came to the conclusion that we could refine SRA heuristics some more to not scalarize this if we added two more attributes to struct access, one meaning "read as a scalar" and another for "written as a scalar." (I'm quite confident this would work, I have a different patch that works too but it uses a rather ad-hoc approach). However, I'm not sure whether we should be adding more attributes when we have already quite a few just in order to be able to make slightly better judgments about single-field structures like this one. (Maybe we really could have a location for return instead?). In either case, it is nothing for stage4. BTW, is this even a 4.6 regression?