Blazingly fast voxel data generation

My voxel engine (Rust+Bevy), is now 25 TIMES FASTER. I mainly used 4 techniques, 1 of which I've never heard anyone do. 1. Extremity bound checking (automated) 2. Noise up-sampling 3. Noise Cashing 4. RLE based runtime voxel data baseline: 20 million voxels / sec. optimizations: 500 million voxels / sec. I can fly 320 blocks per second, and the engine keeps up. RLE allows for 32x256x32 chunk sizes, which has some surprising benefits. Low RAM, instant (file/network) no compression needed, faster chunk management/scanner, fast y range memory writes, fast surface detection, data fits in L1 cpu cache. the cons: slower random memory access+modification. ^this con has not affected me yet, because I do not have surface feature implemented yet. Q: What about ambient occlusion + RLE? To binary greedy mesh, we always transform chunk data into bit-arrays. So when sampling for AO, I don't random memory access the RLE. it's fairly cache-friendly to unpack the RLE into the bit-arrays. Fun fact: I sample AO in a 5x5 range, and it's still this fast :o Data gen also traverse in such a way that we NEVER have to randomly memory access the RLE. That's how I can use RLE, and still be very fast! Q: How does the binary greedy mesher work now? Considering my greedy mesher works upon u32 bit-arrays, i can't fit 256Y inside the bit-array data. Hence, I segment up the world into 8 seperate (32*32*32) volumes. If my chunk data contains faces spanning y:0 to y:64, My mesher will only generate 2 seperate meshes. #bevy #rustlang #voxel