Comment by ashpil

Another alternative that the author omitted is just casting everything to doubles and summing naively (or vectorized) in double precision, then casting the final result back to a float. Would be curious to see how this compares to the other methods and whether it’s on the Pareto frontier.