Issue Description:

We are in need of tests and/or benchmarks to ensure that arithmetic operations in Swift compile down to single instructions (when a single instruction on the target captures the necessary semantics).

These could plausibly take the form of either benchmarks or regression tests; regression tests are good because we can directly validate that we get single IR or assembly operations where appropriate, benchmarks are also good in some cases, where we don't necessarily have a single instruction replacement for an operation (integer smart shifts, for example).