if you had a displacement map texture with a full mip chain. and each mip was the maximum heights (not the average height) scaled down of the mip before it, wouldnt that be a way to reduce texture sampling - marching the ray across?
How come this has never come up anywhere on the net? (as far as ive checked) Does it suck compared or not work?
It would require sampling a desired level, so youd have to make sure you stopped at the right from eye distance.
Try looking into this presentation - http://www.google.cz/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&ved=0CCAQFjAA&url=http%3A%2F%2Fwww.drobot.org%2Fpub%2FM_Drobot_Programming_Quadtree%2520Displacement%2520Mapping.pdf&ei=sUU7UJbtB8724QSdh4C4Dw&usg=AFQjCNE0Nc86XGUZYO_1S2wu2SZGzkCoIw
It shows something called quadtree displacement mapping and thats exactly what you're looking for. Note that it probably will be outperformed by SAH KD-tree in real world scenarios (SAH KD-tree will just be a little better than quadtree on terrain that's not too plannar)
- terrain is static (you can pre-compute the KD-tree, no need to do it at processing time).