Typecasting int->float

[Kevin Mayer]

Consider the following (on a PIC16F877):

float f;
int i = 23;
f = (float) i;
printf("Int is \%D, float is \%3.3f",i,f);

The output is:
Int is 23, float is 22.9

Basically, whatever int I cast to a float, the float is always 0.1 less than the int - this isn't due to the lack of accuracy in the floating point number implementation, so why could this be happening?
___________________________
This message was ported from CCS's old forum
Original Post ID: 1576

[Dale Botkin]

:= float f;
:= int i = 23;
:= f = (float) i;
:= printf("Int is \%D, float is \%3.3f",i,f);
:=
:=The output is:
:= Int is 23, float is 22.9

Is this on a PIC or a Pentium?

8-) Sorry, but /somebody/ had to say it...
___________________________
This message was ported from CCS's old forum
Original Post ID: 1583

[Kevin Mayer]

Actually - no-one really had to say it....the first line of my post is:

"Consider the following (on a PIC16F877):"

so it's on a PIC, not a Pentium...
___________________________
This message was ported from CCS's old forum
Original Post ID: 1584

[Neutone]

Their are rounding errors in the floating number format but they are not causing the 0.1 error. You can use the "Pconvert.exe" program distributed with PCW to look at this. The error you are seeing is probably in the printf function. \%3.3f should have displayed something like 22.999
___________________________
This message was ported from CCS's old forum
Original Post ID: 1591

[Kevin Mayer]

I only have the PCM compiler, not the whole PCW suite and hence I don't have pconvert.exe :(

Sorry - it was actually a printf \%3.1f - but shouldn't printf round it? (or is that asking too much?)
___________________________
This message was ported from CCS's old forum
Original Post ID: 1602

[Neutone]

Try adding .0000001 to the float prior to printing the number. It should look right to at least 5 decimal places (\%3.5f).


:=I only have the PCM compiler, not the whole PCW suite and hence I don't have pconvert.exe <img src="http://www.ccsinfo.com/pix/forum/sad.gif" border="0">
:=
:=Sorry - it was actually a printf \%3.1f - but shouldn't printf round it? (or is that asking too much?)
___________________________
This message was ported from CCS's old forum
Original Post ID: 1605

[Guest]

I have the same problem, but is not a casting problem, the proble is in the printf function.
Anybody have a solution to this problem???

Thnaks in advance

[PCM programmer]

To fix the printf rounding problem, see this thread.
http://www.ccsinfo.com/forum/viewtopic.php?t=23290

So, if your precision is ".1", then you add .05 to the value.

if the precision is ".2", then you add .005

if it's ".3", you add .0005

This should fix the CCS round-off problem.

[Douglas Kennedy]

There is no reason why this should happen from a math point of view.
23 is an integer clearly within the range of the mantissa.
Assuming the 4 byte in float representation is accurate of the decimal value 23 then printf and its possible divide by ten logic to tease out the digits for printing is causing an error.
A necessary for accuracy move on a float before CCS attempts a printf conversion would be to first see if the value can be expressed within the mantissa with a zero exponent. This is certainly true of all integers less than 2^23. If the exponent can be balanced down to zero ( it would actually be 127 due to the bias needed to allow for negative exponents)and at the same time adjusting the mantissa for the exponent changes then printf should treat the float now as a 23bit integer. Printf can print integers correctly so it should also print float correctly. Again this is a bug probably caused by starting with the floats normalized 0.1...... binary mantissa representation rather than up shifting when they see a zero in the right most (23 bit of the mantissa) and balancing with a decrement in the exponent.
It's a bug from my point of view.