[SOLVED] checking saturation logic

[rakeshk.r]

Hi, Lets say, I have a 6 bit rounded result (4 Int,2 Frac) from a signed multiplication. But i need a 4 bit output (2 int and 2 frac). So now I will have to check if the result overflows when i remove some integer bits. From this point could you check if i am doing the following logic correctly.

Assume 'x' is the 6 bit rounded result. and 'res' is the 4 bit output needed.

if x(5 downto 4) /= (x(3)&x(3)) then 

    if x(5) = '1' then
         res <= (3=>'0', others=>'1');
    else
        res <= (3=>'1', others=>'0');
    end if;

else
    res <= x (3 downto 0);     
end if;

 

[TrickyDicky]

why not simply write this:

if x > POS_SAT_VALUE then
  res <= MAX;
elsif x < NEG_SAT_VALUE then
  res <= MIN;
else
  res <= x(res'range);
end if;

It is much clearer.

 

[rakeshk.r]

considering my example above, POS_SAT_VALUE = "0111" and NEG_SAT_VALUE = "1000". MAX ="0111" and MIN = "1000" right ?

 

[TrickyDicky]

Yes.

If you tried out the fixed point packages, it would easy to do something like:

if x > to_sfixed(1.75, 4, -2) then
  res <= to_sfixed(1.75, 1, -2);
elsif x < to_sfixed(-2.0, 4, -2) then
  res <= to_sfixed(-2.0, 1, -2);
else
  res <= x(1 downto -2);
end if;

 

[rakeshk.r]

I think I made a mistake in the code mentioned in the question.

if x(5) = '1' then
         res <= (3=>'1', others=>'0'); -- max neg
    else
        res <= (3=>'0', others=>'1'); -- max pos 
    end if;

Isn't it ?

- - - Updated - - -

why not simply write this:

if x > POS_SAT_VALUE then
  res <= MAX;
elsif x < NEG_SAT_VALUE then
  res <= MIN;
else
  res <= x(res'range);
end if;

It is much clearer.

Does this logic work for the case where guard bits are used ?