I have a design were I'm writing/reading to/from a RAM and perform some computation on the read values. In some cases, I read values from RAM locations where I haven't written anything to yet. This is intentional because in the cases where this happens, the uninitialized values don't affect the computation: in these cases, the uninitialized values are multiplied with 0.
However, multiplying an unsigned
/signed
type which contains 'U'
bits results in a "don't care" output (i.e. all bits of the multiplication output are 'X'
) even if the other operand is 0. Therefore, I can't check the final computation output in my testbench because it becomes "don't care" (it seems like "don't care" outputs are interpreted as 0).
To avoid this problem, I wrote a function that resolves any 'U'
or 'X'
bits in a std_logic_vector
to '0'
. The functions looks as follows
function f(x : std_logic_vector) return std_logic_vector is
variable y : std_logic_vector (x'range);
begin
y := x;
-- pragma synthesis off
for i in 0 to x'length-1 loop
case x(i) is
when 'U' | 'X' => y(i) := '0';
when others => y(i) := x(i);
end case;
end loop; -- i
-- pragma synthesis on
return y;
end;
Now I'd like to expand the function by not only setting 'X'
and 'U'
bits to '0'
but to randomly set them to either '0'
or '1'
. I've tried using the uniform
function within f
. The problem is that when I define the two seeds within the function, that each time the function f
is called it returns the same std_logic_vector
(when it is given the same std_logic_vector
). As I take it from the uniform
function description, I should pass the two seeds from outside the function f
because they are modified by the uniform
function for the next call to uniform
.
Is there a possibility how this can be achieved using a function?
There's a very good random library as part of the Open Source VHDL Verification Methodology here. There is a description and download link here.
http://www.synthworks.com/blog/osvvm/
It allows you to randomise much more than just a simple uniform distribution of floating point numbers. As well as isolating you from the state storage problem you have noted.
Regarding your specifics:
As I take it from the uniform function description, I should pass the two seeds from outside the function f because they are modified by the uniform function for the next call to uniform.
Yes, you should. Like this:
PROCESS
VARIABLE seed1, seed2: positive; -- Seed and state values for random generator
VARIABLE rand: real; -- Random real-number value in range 0 to 1.0
BEGIN
UNIFORM(seed1, seed2, rand);
So in your case, you'll have to pass those "state" variables into (and out of) your function too - which in practice means it has to be a procedure.
Or use the OSVVM library linked above which allows you to have a shared variable of protected type, which you can use from a variety of places. This keeps its own state "inside" the protected type.
Can I achieve this using a function or do I have to use a procedure?
Functions do not allow parameters to be inout, pointers, or protected types. This limits your choices. OSVVM's randompkg uses a protected type to hide the seed and uses an impure function to access it. It is pretty easy to use. Just download the package from http://www.synthworks.com/downloads and look at the RandomPkg_user_guide.pdf.
You can probably get what you are trying to do to work, however, it is going to be a challenge. You can define a signal or pair of signals in your package and use an impure function (YMMV, I have only used impure functions inside of protected types). You can initialize the seeds in the signal declarations.
Even with OSVVM, you will need to use an impure function and declare the randomization object as a shared variable.
Jim
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With