The worst thing is however that they do not even cite the practically relevant memory efficient implementation of strassen (https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.39.6887 ). One can argue that all matmul algorithms with better complexity than Strassen are irrelevant due to their constants, but not even comparing to the best memory implementation is odd-especially as they don't show improvement in asymptotic complexity.