Isn’t the usual reason for sampling isn’t resolution, but the overhead of collecting the statistics? But if the overhead problem can be reduced, high resolution would be nice; I suspect it would have shown that more time was spent in is/2 when I was trying to speed up the “countdown” problem.
I’m curious how you’d reduce the overhead of collecting the statistics? (I’ve written a few profilers, with sampling in a separate thread; and even when writing in assembler, there’s a non-trivial overhead and resulting statistics skew.)