I recently did a graphics experiment with volume rendering where I wanted to partition a grid into groups of active voxels. Imagining a Morton or Hilbert curve

I recently did a graphics experiment with volume rendering where I wanted to partition a grid into groups of active voxels. Imagining a Morton or Hilbert curve would be best, I compared it to row-major/scanline order and Boutrophedonic, as maybe a couple different baselines. I was very surprised to find that Boustrophedonic was the best of these in terms of minimizing volume and overlap of the groups. I was packing naively and not being careful to use powers of two, which definitely compromised the usage of Morton and Hilbert, so my result isn’t necessarily the best or generally applicable, but it was still surprising and interesting to me and some others people I discussed with, that doing it the dumb way lead to back-and-forth ordering being better than space filling curves with what seem to be more apparent locality properties. (Well, it wasn’t as surprising to the guy who suggested I should try a Boustrophedonic ordering. Maybe he experienced the same surprise earlier.)