You should prefer using the standardized likely/unlikely attributes instead.