That is an interesting paper! While, yes, it does have some similarities to macro-regions it looks a lot more similar to an old algorithm I read once (see link below). That particular algorithm dynamically increased and decreased the size of each of the grid's voxels depending on the amount of rays entering it and the amount of primitives to be intersected inside the voxel (i.e. the load). It worked on an old style MPP architecture. The difference with the irregular grid paper above is the SAH is pre-computed and the voxels are resized as a pre-processing step instead of being dynamically re-computed during rendering like in the old paper (that I link below):http://graphicsinterface.org/wp-content ... 1986-9.pdf
It is a shame Nemoto & Omachi's (1986) paper was not linked in that article since it seems a lot more relevant than most of the things it references. There are also more recent references with macro-regions e.g. "Macro 64-regions for uniform grids on GPU" (2014), Eugene M. Taranta II, Sumanta N. Pattanaik. The proximity clouds references are kinda pointless. Although it is also a space skipping technique it's different. Too many pointless BVH references too.
Their grid construction could still probably be faster though. Looking at their 'emit_new_refs', for example, it uses a per primitive loop to generate the prim and cell ids, which will have poor workload distribution in scenes with dissimilar sized primitives. We solve that in our paper:https://www.academia.edu/13928983/Effic ... hitectures
It is nice to see people working on grids. They have been IMHO under explored in the present literature so this paper is very welcome!