The part of that paper that says how expensive Trap on Overflow is:
> At the `-02` optimization level, our compiler prototype showed only a 5.58% slowdown when running the SPECINT2006 macro-benchmark. Although that percentage represents the worst-case performance for AIR integers (because no optimizations were performed), it is still low enough for typical applications to enable this feature in deployed systems.
> At the `-02` optimization level, our compiler prototype showed only a 5.58% slowdown when running the SPECINT2006 macro-benchmark. Although that percentage represents the worst-case performance for AIR integers (because no optimizations were performed), it is still low enough for typical applications to enable this feature in deployed systems.