RE: [sv-ec] In-line random variable control

Section 13.10 describe the behavior of in-line variable control as:
 
"This mechanism controls the set of active random variables for the
duration of the call to randomize, which is conceptually equivalent to
making a set of calls to the rand_mode() method to disable or enable the
corresponding random variables."
 
The LRM does NOT make any mention of changing the set of active
constraints.
 
When a class is randomized, the randomization includes all the active
constraints in the class. So a
contraint on state variables that is not met will cause the
randomization to fail.
 
Consider the following:
 
module top;
 
  class example;
    rand int xval, yval;
    int aval, bval;
    constraint c1 { aval < bval; };
    constraint c2 { aval < xval; };
  endclass
 
  example E1 = new;
 
  initial begin
        E1.aval = 10;
        E1.bval = 10;
        assert ( E1.randomize() ); // fails constraint c1
        E1.aval = 9;
        assert ( E1.randomize() );   // succeeds
        E1.xval = E1.aval;
        E1.xval.rand_mode(0);
        assert ( E1.randomize() );    // fails constraint c2
        E1.xval.rand_mode(1);
        assert ( E1.randomize(yval) );    // fails constraint c2
  end
endmodule
 
The first randomize fails due to constraint c1. The values of the state
variables (aval, bval)
do not meet the constraint.
The second randomize succeeds because the state variable now have good
values.
 
The third randomize fails because 'xval' is now a state variable. The
values of state
variables (xval, aval) cause constraint c2 to fail.
 
The fourth randomize is similar to the third. The constraint 'c2' will
fail because the
values of state variable (xval, aval).
 
 
- Ray


________________________________

	From: owner-sv-ec@server.eda.org
[mailto:owner-sv-ec@server.eda.org] On Behalf Of Symons, Tom H
	Sent: Friday, April 14, 2006 9:33 AM
	To: sv-ec@server.eda.org
	Subject: [sv-ec] In-line random variable control
	
	

	When calling randomize with arguments, should the constraint
solver fail if random variables not specified as arguments do not
satisfy their constraints at the time of the call?

	 

	Or should the solver just focus on the random variables given as
arguments, and only consider constraints that directly or indirectly
involve them?

	 

	The two simulators we are working with have come to opposite
conclusions on this issue.

	 

	I would think in-line random variable control would be much more
useful if only related constraints were involved.

	 

	Thanks

	Tom Symons