Multi-operand floating-point addition utilizing operand sorting
F. Hartwig, Arild Lacroix · 2003
In many signal processing applications repeated additions have to be calculated with a considerable number of operands. It is known that proper sorting of operands has a positive influence on the precision of the result of multioperand floating-point additions. This is investigated for different operand statistics and for proper sorting strategies. In addition to successive addition techniques, adder trees are included in the investigation. The densities of the addition errors for different floating-point addition methods are compared.>