lgtm This seems like exactly the kind of optimization tweaking options that we ignore for gcc compatibility. If we implemented this, I guess it would control shrink-wrapping optimizations in the backend. http://reviews.llvm.org/D4357