Response here because I'm not on Twitter: https://github.com/google/highway/issues/736 https://github.com/google/highway/issues/736 In short, what is being com

Response here because I'm not on Twitter: https://github.com/google/highway/issues/736 https://github.com/google/highway/issues/736 In short, what is being compared is O(1) djbsort sorting network, vs our full quicksort with pivot sampling, partitioning, then sorting network. This is because our sorting network size is 16 * elements_per_vector i.e. 128 in this configuration.