[weinzierl]

Why does the number of Total Cycles increase for the faster version?

[jhpratt]

I'm not sure! It's certainly counterintuitive, particularly given that it is faster.

[mmastrac]

There is a preliminary version with more -- but the final version has fewer cycles

[zokier]

It is still bit odd considering tha in the actual benchmarks the preliminary version is significantly faster than the old one. So there seems to be fairly large discrepancy between the cycles estimation and actual runtime