Performance Impact of Using Denormalized Numbers in Basic Floating-point Operations

Manipulation of denormalized numbers in circuits for floating-point operations comes at a reasonable cost. IEEE standard 754 defines the use of denormalized values. They are incorporated into high-performance CPUs and they are gaining space in commercial products. This paper describes the extra processing steps that are required for the manipulation of denormalized numbers in floating-point addition and floating-point multiplication. It also provides the results of several experiments that show the average additional cost that designers face when creating these basic operations using cell libraries, for some CMOS technologies.