Still feeling oddly tired, all the time.

On top of that, I think that the initial results showing an 80% win seemed to have been a bug related to a bogus write-back in a prototype resulting the algorithm exiting earlier without being complete. So far, I have put forward about 3 optimizations, each of which has been something like 2-5% improvement. In concrete terms, this means that a micro-bench went from 28.1 seconds down to 26.3 seconds.

I have another few ideas but we will see what happens since none of them are showing huge wins, or anything. I could further specialize this case so that it could take some short-cuts but that increases the breadth of the codebase so I am trying to find more generalized solutions.

We will see, but this is the kind of thing I was worried would delay OctoberPeaks 1.0 completion.