tra accepted this revision. tra added a comment. This revision is now accepted and ready to land. OK. If someone attempts to rely on this feature in CUDA it will be obviously broken due to the missing C-style mangled name. We can deal with it then. http://reviews.llvm.org/D16501