...Float multiplication/division not working

PICman · Joined: 02 Nov 2007 Posts: 26

Hi....

My code is a controller for a DDS chip (AD9850).

I want to enter the RF frequency directly ("frequence" variable) and then, do a simple calculation (one multiplication and a division) involving the reference frequency, to get the right frequency word to be sent (serially) in the DDS chip.

My problem is that the PIC does not perform the calculations.

If i enter the frequency word directly, everything works perfectly ! The data is sent and the DDS generates the required signal..

I've done a lot of C for computers (PC) without any problem, but for the PIC, i don't know why, it doesn't work !

The variables are "float"..

In the "calcul_bits" subroutine, i do perform a substraction with success, but something prevents the multiplication/division to be performed.

Compiler CCS 4.120, µC: PIC12F1840

asmboy · Joined: 20 Nov 2007 Posts: 2128 Location: albany ny

PICman · Joined: 02 Nov 2007 Posts: 26

1 - 0 for your HP35 !!!!

In debugging, i tried different values for my DDS.

And in editing / translating (from french to english) my code to send it here, i erased the wrong value. 34359738 is for a 1 MHz output signal...

(1000000*4294967296/125000000)

I never tried casting.... Is casting used to "translate" a variable between float and int ? Because my variables are all float.

Hmmm.. back to the CCS manual !!!
_________________
The ideal electronic world: A place where the words VERSION and REVISION do NOT exist !

PCM programmer · Joined: 06 Sep 2003 Posts: 21708

RF_Developer · Joined: 07 Feb 2011 Posts: 839

Take a look at http://www.ccsinfo.com/forum/viewtopic.php?t=24066 which deals with a very similar problem. maybe even the same problem.

For someone who has "done a lot of C for computers (PC) without any problem" you seem unaware of basic C, such as casting and constants which are not specific to CCS. Surprised

Float does not have the precision to accurately represent the constants you are using. It may work well enough, however. The poster of the other thread noticed this and was hopign to use doubles instead, but CCS doesn't provide a double length floating point type.

What both you and they have failed to notice is that the constants include powers of two, such as 4294967296 and 2147403648, which are 2^31 and 2^32. This says that fixed point, i.e. scaled integer, arithmetic is most appropriate for this and should, if done carefully, retain all the precision, not loose a lot as floats will.

Also

PICman · Joined: 02 Nov 2007 Posts: 26

Hi RF_developer...

In calculations, i always try to privilege the use of ints and double precision floats. I rarely use single precision floats.

And i wanted to use INTs for this project... but can't !!!! The first multiplication already implies 2^32 (The 32bit DDS freq accumulator), so no multiplication possible, since it would require way more than 32 bits. I cannot invert the operations either (divide freq by reference first) since the result woult be zero. Finally i cannot divide 2^32 by the reference since an INT result would be in the magnitude of 34, thus mediocre precision.

I know that the final solution would be to use a power of 2 crystal (e.g. 134.217718 MHz), but i'd prefer by far to have the code do the work instead of having to order a custom crystal. Also, working with code would give me the option to "fine tune" the DDS by slightly warping the "reference" value instead of mechanically tune the quartz.

I didn't try casting since the variables are ALREADY formatted as float... But i'll try it !
_________________
The ideal electronic world: A place where the words VERSION and REVISION do NOT exist !

PICman · Joined: 02 Nov 2007 Posts: 26

Thanks PCM_programmer !!!!!

I had to add ".0" to all my values to "tell" the compiler that i need a float value.

But what i don't understand, is that i have ALREADY configured the variables as floats !!! It looks like an autocast function. You configure a variable as a float in your code, but if you enter a number that looks like an INT (no DP), the compiler switches (or casts) it to an INT.... Is it possible to disable this function ?

Thanks for the help !
_________________
The ideal electronic world: A place where the words VERSION and REVISION do NOT exist !

Douglas Kennedy · Joined: 07 Sep 2003 Posts: 755 Location: Florida

The compiler won't think like you..it will try to be a mathematician. The mathematical notation on the left side of an equal sign doesn't take on the notation of the left hand side until the calculation is done and ready to be assigned. Integers are notated without a decimal point and come in 8 bit 16 bit and 32 bit format. The compiler will minimize the calculation and use the smallest integer when possible. This means if you want 3 to be a 32bit integer you need to cast it to be so otherwise it will get 8 bits. The notation for float (since so far for CCS there is only one float size) is 3.0 this is how notation works. 3.0 notation is equivalent to (float)3 notation.
The fact that you ask for the result of the right hand-side to be placed in a float doesn't dictate float will be used during the intermediate calculations required to produce the result. Remember the internal notation of a number in a pic register is binary..the compiler will switch from decimal to binary for you...decimal and binary notations are only equivalent for integers and decimal fractions that are a power series of 2. Conversion to and from decimal notation to binary notation can never be exact except for the condition in the prior sentence.

PICman · Joined: 02 Nov 2007 Posts: 26

Thanks Douglas Kennedy...

That confirms what i said on my last post... CCS has an "autocast" function that automatically casts the variables to the smallest memory needs possible... Is that makes useless the variable declaration at the code's beginning (simply declare everything as "int" and let the compiler do the rest) ?
_________________
The ideal electronic world: A place where the words VERSION and REVISION do NOT exist !

Ttelmah · Joined: 11 Mar 2010 Posts: 19346

No.
This is not CCS, but C, and the variable sizes are distinct from the maths sizes.

In C, when you evaluate any mathematical function, the compiler looks at the value(s) used. It takes the _highest_ type of these, and this is the arithmetic type used for the operation. So:

2*3

Both values are integer, and both fit into 8bit (in CCS), so int8 arithmetic.

2L*3

Both values are integer, and the first is a 'long' (int16 in CCS), so int16 maths.

(int32)2*3

Both integer, but first value is int32, so int32 arithmetic used.

2.0*3

First value is now a float, so float arithmetic used.

Now, this applies for each section of an arithmetic formula, so:

3.0*(54*56)

Here the 54*56 is evaluated first, and since both values are integer, and will fit into an int8, int8 arithmetic is used. The result is then converted to a float, and float arithmetic used for the second multiplication by 3. Since the first value has overflowed, the result will be 208*3.0 = 624.0....

The type _is_ derived from the number, if a constant value is used, but is the variable type, if a variable is used. So: