Dispatches / a short note
Ordered access
Ordered access of data and minimizing working set reduces CPU cache misses
Okay, radix sort is pretty cool. I had the same concern with bucket allocation, although it happens only 4 times for a 32bit wide .group id (32/8 (chunk bits used at a time) = 4). And it’s a single pass O(n) groupby algorithm.
I feel the author actually meant .group and not .index in his approach following groupby. Author features MSD radix sort, and out there in the wild we have 3-way radix quicksort, LSD radix (yummy). gotta check them.