plan.txt

coarse
// todo
// - pretransform triangles
//   - no need to write processed
// split and duplicate with subsets of tiles regions for more parallel?
// parallel prefix sum
// parallel init

fine
// todo 
// * run bounds by bb, remove bb test in intersect
// / precompute axis test per axis pair (done for plane test, not edge test. done for final axis, not middle axis)
// - pick output format for subvoxel count
// - iterate axes by dominance (inner iterator = long axis)
// - once all tests are at axis instead of voxel, improve write 

fine_blend
// todo
// - parallel over workgroups with i division, take up to end of your working tile when you reach upper limit

jfa
// todo
// - storage barrier for non-apple, avoid multiple draws?
// * only jump within update segment bounds (reduces 64,32,16,8 to up+down only)
// - push constant for jump size?