3 ms·

Good call, thank you. Updated version has been republished.

• mturmon 12y ago
  Since the row has "n" entries, and errors can accumulate in each add, something like (n+2)*eps may be more appropriate. (I wrote a similar library in Matlab.)