fixed point representation vs floating point

Question

(I don't know if this is the right place to post this, if not sorry and can it be moved to the right site please)

Under what circumstances would fixed point representation be used rather than floating point?

Answer 1

It really depends on CPU architecture, but it would generally be used for either:

• Improvements in performance

• Cases where the target platform does not have a floating point unit (FPU)

Many embedded devices do not have an FPU and would make use of fixed point representation. On some architectures, using fixed point would also provide noticeable performance improvements.

Answer 2

If your target architecture has no (or a slow) floating point unit (for example, many DSPs) then fixed-point will be potentially much faster to compute.

Some algorithms behave very badly when floating point is used. For example, if you end up trying to add very small numbers to very large numbers, the small numbers can look like zeros and the update doesn't make any difference. This can have major knock-on effects.

This is well worth a read: What Every Computer Scientist Should Know About Floating-Point Arithmetic