[See previous devlog entry: Global illumination pre-req visualizations]

I've implemented parallel reduction of voxel world space coordinates in a compute shader and unfortunately performance isn't anywhere near acceptable.  *sigh*  It was an interesting experiment none the less and may still be viable if I can find a way to speed up the algorithm significantly.  I may have to try an entirely different approach for per voxel world space coordinate reduction (lighting pre-pass)...

Here's a visualization animation (much much slower than actual speed of about 0.005 seconds total) of parallel reduction of voxel world space coordinates (color coded) generated from compute shader output passes:

P.S: Here are the timing results for compute passes on a 256x256 vbuffer:

span-size   time
        2   0.000293
        4   0.000388
        8   0.000317
       16   0.000249
       32   0.000171
       64   0.000262
      128   0.000198
      256   0.000323
      512   0.000181
     1024   0.000165
     2048   0.000165
     4096   0.000177
     8192   0.000326
    16384   0.000480
    32768   0.000916
    65536   0.000721
total:      0.005332