=== r128multiply_noopt cycle breakdown (90 calls) ===
  convert x2              :     219.8 cyc/call
  multiply                :      27.5 cyc/call
  divide /10^16 (rescale) :      72.9 cyc/call
  round + overflow check  :      29.5 cyc/call
  divide /10^16 (split)   :     682.0 cyc/call
  ----------------------------------------
  total (timed)           :    1031.7 cyc/call
