Thank you to @nakakoutv for the amazing original project! (https://scratch.mit.edu/projects/1272522593/) This uses the single stamp 4x4 grid renderer first done by @nakakoutv and improved by @yukku. My configuration is now as optimised as possible and runs on mobile without errors and is 23% faster than the original. (50% faster if focusing only on the draw loop) This uses 32 64x64 costumes, each pair of 2 costumes stores the 2^12 combinations of bits excluding corners. Then this is duplicated 16 times for each corner combination. On mobile the gpu shader precision is much lower, so I've bounded the whirl range to just +-350. Achieving this required improving my simulated annealing search, taking advantage of very small whirl increments, and a lot of fisheye values. Again thank you @nakakoutv for the exceptional original project.❤️ I'd have remixed it if this project didn't use devious methods to even be uploaded in the first place.
