(Depth-based) Cache Culling

[Wumpf]

Currently there are loads of caches that are never used.

2 possible solutions:

• mark needed caches in a screenspace-pass before voxelization
  • mark special bit voxel volume - no need to compute hashes etc.
  • if "toggling" a voxel as solid does not destroy this bit, it can be read independently of the actual voxelization!
  • No atomics anywhere if implemented right!
• lookup depth values in a depth max map
  • do not try to use the depth buffer directly as this gives only depth for a single point and not the area of the voxel in screenspace!

Note: Solution 1 sounds simplifies a lot but takes a bit per voxel and another screenpass.