Float Precision

[Kimi]

Hello

First of all, thanks very much to everybody... this forum is great...

Now let's go to the problem... I'not getting enough precision in floating points maths...

x1 = -34.61084;
f1 = 12.44 * x1;
after this we expected f1 = -430.5588496, but we watch f1 = -430.558830

this is just the begining...as we continue doing maths with this floats, errors become biggers...

f1 = f1 + 110574; //just adding a number

Expected f1 = 110143.4411504 -- calculated by Pic f1 = 110143.438726

Is there any way to double the Floating point precision? or at least increase it... Is this a compiler limitation, or a MCU limitation?

Thanks very very much
Kimi

[Kimi]

I found that the printf is wrong...

dist1 = 110143.4411504;
fprintf (PC," %f ",dist1);

this should return 110143.4411504 but I see 110143.432021 in the PC.
anybody could solve this problem?

Thanks
Kimi

[Neutone]

The floating point format used by CCS has 6.5 significant digits. It's a 32 bit float not as much percision as used in most PC software.

[Kimi]

Thanks

Is there any usefull way to display the 4 bytes that compose the float into screen... first to check the operation, second to be able to send a float though the serial or spi port.

Thanks

Kimi

[SherpaDoug]

If you make a pointer to the float you can print the bytes in hex.
My advice is not to use floats unless you really REALLY have to. I measure temperature all the time in integer millidegrees. A while back I was measuring warping in ship hulls in 24bit integer 1/10 millimeters. I could measure everything from the thickness of the paint to the length of a supertanker. I used string functions to convert the results to meters for printing.
_________________
The search for better is endless. Instead simply find very good and get the job done.