This is exactly my thought, check the video I posted on twitter @serialdev , I was thinking of the usual loop unrolling , caching , vectorize , if with a managed Lang compile, check intrinsics and then even explore SIMD. I'm honestly impressed by ChatGPT