I'm wondering how loops of various sizes perform on recent x86 processors, as a function of number of uops.

Here's a quote from Peter Cordes who raised the issue of non-multiple-of-4 counts in ano...