Hey :) awesome work! I used this for my own bot and it worked an absolute treat.

I managed to improve the performance significantly by adding -O3 optimisations and wrapping the main search for loop with #pragma omp parallel for and compiling with openmp. On my 2014 MacBook Pro Retina laptop it can now solve depth = 3 in ~0.1s and depth = 4 in ~2.7s.

Happy to throw up a patch when I have some time, but figured I'd document here for anyone reading :)

Thanks again for the great open source contribution!