DeepMind has made a novel discovery in the superoptimization space that could turbocharge the abstractions produced by C libraries. By applying their deep learning knowledge gained from building AlphaGo to the discipline of superoptimization, DeepMind found tinier kernels for sorting algorithms. While the company’s contribution to LLVM libcxx earned well-deserved attention, it was noted that their published paper could have been better explained. DeepMind’s contribution aims to improve the way compilers optimize sorting codes because cleverer ways exist to code this type of problem. The example demonstrated produces about 2x speed improvement for sorting ascending ‘longs.’
https://justine.lol/sorting/