Some implementations of floating-point use support code to handle denormalized numbers. The performance of such systems, in calculations involving denormalized numbers, is much less than it is in normal calculations.

Flush-to-zero mode in floating-point always preserves the sign bit.