Error (no time sorry!)

• Normalization (-1;0] ULPs

• Approximation

  • Simulation i.e. N=12 [-7;5]
  • The normalization error gets (-3;0]

• Multiplication

  • Constant truncation
  • Result truncation (-1;0]
  • The constant is <1 and the errors from previous blocks are reduced.

• The error range is unbalanced (more negative) can be compensated when we add j