SystemVerilog pre_randomize & post_randomize

Variables declared rand or randc inside a class aren't randomized automatically — you always call the built-in randomize() method to do it, and it returns 1 on success or 0 on failure. Two callback functions, pre_randomize() and post_randomize(), let you hook into that process to run your own code immediately before and after the solver assigns new random values.

6 min read | Intermediate Level

What You'll Learn

• How to call the built-in randomize() method and check its return value
• When pre_randomize() and post_randomize() run relative to randomize()
• How overriding these two callbacks changes an object's randomization behavior
• What happens to post_randomize() when randomize() fails

Calling randomize()

Class objects are not randomized automatically — you always have to call the built-in randomize() method yourself. It can fail for a variety of reasons, such as conflicting constraints or a solver that simply can't find a value satisfying every constraint at once, so the method returns a status you can check.

virtual function int randomize ();

Here's a simple example showing how randomize() is called:

class Beverage;
  rand bit [7:0]	beer_id;
  
  constraint c_beer_id { beer_id >= 10;
                        beer_id <= 50; };
  
endclass

module tb;
   Beverage b;
  
    initial begin
      b = new ();
      $display ("Initial beerId = %0d", b.beer_id);
      if (b.randomize ())
      	$display ("Randomization successful !");
      $display ("After randomization beerId = %0d", b.beer_id);
    end
endmodule

Output

run -all;
# KERNEL: Initial beerId = 0
# KERNEL: Randomization successful !
# KERNEL: After randomization beerId = 25
# KERNEL: Simulation has finished. There are no more test vectors to simulate.

beer_id starts out at its default value of 0, then randomize() assigns it a fresh value between 10 and 50 as required by c_beer_id, and the if check confirms the call succeeded before printing the result. There are also two callback functions that randomize() automatically calls for you — one right before it computes random values, and one right after.

pre_randomize(): Running Code Before Randomization

This function is defined within the same class whose object will be randomized, and is called before the solver assigns random values.

function void pre_randomize();

We'll add a pre_randomize() function to the example discussed earlier, using the same testbench:

class Beverage;
  rand bit [7:0]	beer_id;
  
  constraint c_beer_id { beer_id >= 10;
                        beer_id <= 50; };
  
  function void pre_randomize ();
  	$display ("This will be called just before randomization");
  endfunction
  
endclass

Output

run -all;
# KERNEL: Initial beerId = 0
# KERNEL: This will be called just before randomization
# KERNEL: Randomization successful !
# KERNEL: After randomization beerId = 25
# KERNEL: Simulation has finished. There are no more test vectors to simulate.

post_randomize(): Running Code After Randomization

This function is also defined within the same class, and is called after the solver has finished assigning random values.

function void post_randomize();

We'll add a post_randomize() function to the same example:

class Beverage;
  rand bit [7:0]	beer_id;
  
  constraint c_beer_id { beer_id >= 10;
                        beer_id <= 50; };
  
  function void pre_randomize ();
  	$display ("This will be called just before randomization");
  endfunction
  
  function void post_randomize ();
  	$display ("This will be called just after randomization");
  endfunction
  
endclass

Output

run -all;
# KERNEL: Initial beerId = 0
# KERNEL: This will be called just before randomization
# KERNEL: This will be called just after randomization
# KERNEL: Randomization successful !
# KERNEL: After randomization beerId = 25
# KERNEL: Simulation has finished. There are no more test vectors to simulate.

The order in the output tells the whole story: pre_randomize() runs first, then the solver assigns beer_id, then post_randomize() runs, and only after all of that does b.randomize() actually return — which is why "Randomization successful !" prints last, even though it's the first line of code inside the if.

Overriding the Callbacks

What we did above is override the existing, empty pre_randomize() and post_randomize() methods with our own definitions — a clean way to change an object's randomization behavior without touching randomize() itself. If a class is derived from a base class and doesn't define its own version of either method, both automatically call their super version instead.

pre_randomize() and post_randomize() are not declared virtual, but they behave as if they were. If you try to make them virtual yourself, you'll hit a compiler error:

class Beverage;
  rand bit [7:0]	beer_id;
  
  virtual function void pre_randomize ();
  	$display ("This will be called just before randomization");
  endfunction
  
endclass

Output

virtual function void pre_randomize ();
                                    |
ncvlog: *E,CLSMNV (testbench.sv,7|36): The pre_randomize() method cannot be declared virtual.

• If randomize() fails, post_randomize() is not called
• The randomize() method itself is built-in and cannot be overridden
• If randomization fails, the variables retain their original values and are not modified

Common Beginner Mistakes

class Beverage;
  rand bit [7:0]	beer_id;
       bit [7:0]	checksum;
  
  constraint c_beer_id { beer_id >= 10;
                        beer_id <= 50; };
  
  function void post_randomize ();
  	checksum = ~beer_id;
  	$display ("beer_id = %0d checksum = %0d", beer_id, checksum);
  endfunction
  
endclass