Integer math vs floating point math

[MiniMe]

Hi !

As I'm dealing with my very first bigger project, I'm up to make a decision: to choose integer math over floating point variables

Why I assume it is good:
- less rom and ram usage for math operations
- faster math operations
- good to use as I'm dealing with low range values ( temperature, many values)
- no need to convert for further action with int variables ( for example to change a number in menu system)

My main reason to think that it is a good idea is: to avoid full ROM in a finished program (if to choose different devise with less program memory later, because atm. it seems like to be a problem).

Is it worth doing ? Because so far nothing is certain... at the moment. Or just to do it using floating point variables and if full rom problem occur later... just to re - do whole thing.

I'm asking for Your help and advice because I have very little c language experience and do not much about embedded system design... because it is my first project. Your answer and advise could save a day of low result work for me.


Thank you in advance !

[SherpaDoug]

I only use floats if I really absolutely have to. Integers give you control over rounding and precision. They generate faster and smaller code. It may take a little more thought at the beginning of the project, but it forces you to really understand the data you are working with, and that is usually good. That extra thought can avoid bugs that otherwise show up late in the project just before delivery.

My first really big PIC project was gear to precisely measure ship hulls to find warping and dents. I could measure any ship afloat with 24 bit integer millimeters with lots of room to spare.

Tell us more about the nature of the project and its data and we can tell you more.
_________________
The search for better is endless. Instead simply find very good and get the job done.

[Ttelmah]

Yes.
Remember also, that CCS contains the ability to automatically display a scaled integer 'as if' it is a decimal value, using the %LW printf format.
Most people here who have experience with PIC's, will treat FP, as something to avoid if possible.

Best Wishes

[bkamen]

And PIC's aren't the only MPU's without FP capability.

On ANY integer only MPU, one always wants to consider FP Math.

Some MPU's are fast enough so that maybe it's no big deal.. depending on the application. But that's always the catch. Carefully consider what's being done and see if FP is really needed.

I use FP on PICs when I know speed isn't an issue.

-Ben
_________________
Dazed and confused? I don't think so. Just "plain lost" will do. :D

[SherpaDoug]

It is not just a speed issue. With floating point it is often difficult to tell just when rounding errors will occur. With fixed point the rounding errors are very predictable, and if the integers are properly scaled the system resolution is better than from any equal data length floating point math.
_________________
The search for better is endless. Instead simply find very good and get the job done.

[Jeff King]

Any suggestions on integer based math libs? I.e. sin cos, etc etc.?

[Ttelmah]

'Integer arithmetic', implies 'no decimals', implies no possible results for sin/cos etc.....

Only with _scaled_ integers, do results for things like sin/cos exist.

For most applications, if you are only using (say) 2 digits), and coding for speed, then the fastest solution, is a lookup table. Just code your own. There have been several posted examples here for some specific applications.
For three digits, a 2 digit lookup, and linear interpolation is usually the simplest . There have been interpolation examples posted here as well.
You need to decide what scaling you want to use, and what accuracy you need, then people may be able to help.

Best Wishes

[Jeff King]

Ok, thanks.

About 7 places to the right of the decimal point Range of -180 to +180. GPS data calculating distances. Already using the Haversine forumla. Greater speed would be a plus but biggest concern is getting rid of rounding errors and extending precision.

Just looking for some references to some code libraries/techniques if you might know any. Or some seed words for the search engines to get me started on google/here.

Thanks again.

[languer]

Don't know if you've seen this... CORDIC 32bit Functions

Have not personally used them, but have seen many GPS projects using CORDIC arithmetic.