@kephale I noticed, while doing this work, that https://fileadmin.cs.lth.se/cs/Personal/Tomas_Akenine-Moller/code/ has three different versions of the code, tribox.txt, tribox2.txt (faster), and tribox3.txt (faster and less buggy). The page unambiguously recommends using the third version, but your comment in the code references tribox.txt. Have you had a look at the differences between these, and compared to your implementation? Should we take the lessons learned from the 2nd and 3rd versions of tribox and apply them here?

Definitely up for learning from the other versions as well. I would have expected that I checked out multiple versions at the time, but that was a long time ago now. The current changes have certainly diverged from the previous version. I haven't revisited the originals recently. For these updates I just tried out some strategies that seemed reasonable.