when do you need floating point units?

when will you need floating point units in your application?

In what application, only integer units is good?

thanks

Let's say your salary is 49.5% of $1,000,000.00 per year. Would you be happy receiving nothing?

Have you read about the history of fractions?

Can you refine your question a little more?

John

It depends what you mean by "need". It is possible to do all computation with integers, but sometimes we use floating point representation for convenience.

Software development will often make wide use of floating point numbers, since these are often easier for the programmer and modern microprocessors have excellent capabilities for dealing with floats.

However, floats are often completely avoided in high-speed hardware development. Basic arithmetic blocks (such as adders and multipliers) are much less complex for integers than floats. Since hardware resources are often very limited (or expensive), it is often worth going to great lengths to solve all your sums using integer/fixed-point arithmetic wherever possible.

It is often avoidable in many embedded systems, but sometimes, it is quite difficult to avoid and it comes in very handy to have a floating point unit on board and make use of it. Once such example is in industrial motion control systems. The user may want to control the speed of the machine in terms of his/her "user units", which could be "coffins-per-minute" or packets per minute, or feet-per-second, or chasses-per-hour, etc... Your system needs to be able to handle fractional user units, and convert them into the appropriate encoder revolutions and through whatever gearbox ratios to motor revolutions. While it could be done by limiting the user to specific increments of user units he can choose, this isn't always acceptable. It is alot easier and more straightforward to be able to simply calculate, for example, the 32.347 feet per second the user wants in floating point into the nearest integer number of motor revolutions and encoder counts. But, if one is always working in lower level machine units, then using all integer math or scaled integer math common in fixed point DSPs is still common.

when dealing with high speed data acquisition and digital signal processing hardware floating point units can give one the real number processing speed required
**broken link removed**
https://www.ti.com/lsds/ti/dsp/overview.page