When your operations are statistical in nature, as Approximate Computing is, so then these operations themselves are different sampling methods.
If the for loop is calculating a sum, but that value only needs to be over a threshold, there is no reason to add the whole thing, you need a proper sample size and distribution and that it what Loop Perforation is giving you in this case.
