adding two fixed point numbers

Hi, this is probably a newbie question using fixed point designer. I am trying to add two fixed point numbers, one of which is extracted partially from another fixed point number. I don't need to do any hardware code generation later on so I only care about the math being correct. Please see below:

a=fi(1,0,3,0);
b=fi(1,0,3,0);
out=a.bin(1:2)+b;

a.bin(1:2)+b causes an error. I suppose that is because a.bin(1:2) is not a fixed point object.

I know the way to do this is to define a new fi object with the proper wordlength and then do the summation:

c=fi(0,0,2,0);
c.bin=a.bin(1:2);
out=c+b

Is there a different way to do this so I don't need to define an intermediate fi object, c?

Thanks you very much.

Answer 1

[Andy Bartlett]

Bin method returns a string

The bin method of a fi object

a=fi(1,0,3,0);
w = a.bin(1:2)
class(w)

returns a string

w =
    '00'
ans =
    'char'

and cannot be added to another fi object.

fi is both a constructor and a cast operator

When fi is called, it is both a constructor and a run-time cast operation.

Your goal of adding b to the two most significant bits of a without creating an intermediate variable can be achieved using fi as a cast.

wl = 3;
iMax = 2^wl - 1;
nPts = 8;
vA = randi([0 iMax],1,nPts);
vB = randi([0 iMax],1,nPts);
a=fi(vA,0,3,0);
b=fi(vB,0,3,0);
outFi = fi( a, 0, 2, -1, 'RoundMode', 'floor') + b
%
% check math to show that this is correct.
% Luxury doubles math answer and the fixed-point answer should be
% off by 0 or 1 depending on whether bit was dropped from a
%
outDbl = double(a) + double(b);
err = double(outFi) - outDbl;
fi_dbl_err = [
    double(outFi)
    outDbl
    err]

Running this script with this random number generator seed

rng(12345);

produced the following output

outFi = 
    11     7     5     7     4     4     8     9
      numerictype(0,4,0)

        RoundingMethod: Floor
        OverflowAction: Saturate
           ProductMode: FullPrecision
               SumMode: FullPrecision
fi_dbl_err =
    11     7     5     7     4     4     8     9
    12     7     6     8     4     4     9    10
    -1     0    -1    -1     0     0    -1    -1

[scc28x]

Thank you for your reply. I still don't quite understand the "cast" feature. Using the idea above. Here is an example:

a=fi(2,0,3,0);
k=fi(a,0,3,1)+1;

In this case, k becomes "3" and k.bin='0110'. Maybe I missed something but this is not achieving my goal. It is merely shifting the decimal location. The underlying value of k is retained at 2.0+1=3.0.

What I want to see is:

a.bin='0010' is bit shifted to left by 1, so we get '0100'.

Now when adding 1 to "a", we get '0101'. In other words, the underlying value actually changes.

I was thinking about using bitsliceget to accomplish this. Is it the best way?

Thanks again.