Part of the article (and the whole article it references) are about how algorithm A that is guaranteed to perform fewer swaps than B is nevertheless slower in p

Part of the article (and the whole article it references) are about how algorithm A that is guaranteed to perform fewer swaps than B is nevertheless slower in practice on modern machines because of other issues (predictability, data flow). > Most contemporary implementations of quicksort use Hoare partition, for obvious reasons: it does as many comparisons as the Lomuto partition and fewer swaps. > Given that Hoare partition clearly does less work than Lomuto partition, the question would be why ever teach or use the latter at all. […] implemented in a branch-free manner, Lomuto partition is a lot faster than Hoare partition on random data.