* Expand comments.
* Swap variable names in the sum_exact code so that x and y
  are consistently chosen as the larger and smaller magnitude
  values respectively.