Problematic handling of LLVM arithmetic with overflow intrinsics

[shaobo-he]

Current integer-overflow checking assertions are inserted whenever an LLVM arithmetic with overflow intrinsics occur. This is incorrect because such intrinsics are not necessarily used for integer-overflow checking. For example, __builtin_add_overflow is compiled to one of the intrinsics. We need to decouple the handling of these intrinsics with overflow checking.

[zvonimir]

I don't quite understand this issue. Are you saying we are injecting checks even when overflow checking is not enabled in the command line?