Chapter 6, Randomization Topics

Introduction to randomization
What to randomize?
A simple class with random variables
Weighted distributions
Conditional constraints
Solution probabilities
Controlling constraints
Constraint tips and techniques
Common Randomization Problems
Iterative and Array Constraints
Atomic Stimulus Gen vs Scenario Gen
Random Device Configuration

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

Chapter 6: Randomization
Directed testing:
Checks only anticipated bugs
Scales poorly as requirements change
Little upfront work
Linear progress
Random testing:
Checks unanticipated bugs
Scales well as requirements change
More upfront work
Better than linear progress
100%
Random
Testing
Test Plan
Completion

Directed
Testing

Chapter
Time 6 Copyright 2011 G. Tumbush, C. Spear v1.4

6.1 Introduction
A testbench based on randomization is a shotgun
The features you are trying to test is the target

Features

How to cover untested areas?

More random testing with tighter constraints
Directed testing
When is testing done?
Functional coverage
Code coverage
Shotgun Verification or The Homer Simpson Guide to Verification, Peet James
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

6.2 What to randomize?

Much more than data
1. Device configuration
2. Environment configuration
3. Primary input data
4. Encapsulated input data
5. Protocol exceptions
6. Errors and violations
7. Delays
8. Test order
9. Seed for the random test

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

6.3 Randomization in SystemVerilog

Specified within a class along with constraints
Variable declared with rand keyword distributes values uniformly
rand bit [1:0] y;
3, 2, 0, 0, 3, 1, .....
Variable declared with randc keyword distributes values cyclically
No repeats within an iteration
randc bit [1:0] y;

initial permutation 0,3,2,1

next permutation 0,3,2,1
next permutation 2,0,1,3

Constraints specified with constraint keyword

constraint y_c {y >=1;y<3;}

New values selected when randomize() function called

Returns 1 if constraints can be solved, 0 otherwise
<handle>.randomize();
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

6.3 Randomization in SystemVerilog

Constraint syntax:
constraint constraint_name { contraint_expressions }
where:
constraint_name is the name of the constraint block. This name can be
used to enable or disable a constraint using the system task
$constraint_mode().
constraint_expressions is a list of expression statements that restrict the
range of a variable or define relations between variables. A constraint
expression is any SystemVerilog expression, or one of the constraintspecific operators: =>, inside and dist.

The declarative nature of constraints imposes the following

restrictions on constraint expressions:

Calling tasks or functions is not allowed

Operators with side effects, such as ++ and -- are not allowed.
randc variables cannot be specified in ordering constraints
dist expressions cannot appear in other expressions (unlike inside); they
can only be top level expressions
Source: SystemVerilog 3.1Random Constraints Proposal Synopsys
6

6.3.1 Simple class with Random Variables

class Packet;
rand bit [31:0] src, dst, data[8];
randc bit [7:0] kind;
constraint c {src > 10; src < 15;}
endclass
Packet p;
Constraint expressions
initial begin
p=new();
if (!p.randomize()) // Returns a 0 if problem
$finish;
transmit(p);
end

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

6.3.2 Checking the result from randomization

Always check the result of a call to randomize()
randomize()will not be called if using an assertion to check
result and assertions turned off
initial begin
packet p = new();
$assertoff;
assert(p.randomize());
p.display();
end

# src = 0, dst =

0, ...

$rand_mode() method can be used to enable or disable any

random variable.
When a random variable is disabled, it behaves in exactly the
same way as other non-random variables.
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

6.3.2 Checking the result from... (cont.)

Text uses a macro to check the results from randomization
`define SV_RAND_CHECK(r) \
Line continuation symbol
do begin \
if (!(r)) begin \
$display("%s:%0d: Randomization failed \"%s\"", \
`__FILE__, `__LINE__, `"r`"); \
$finish; \
end \
Character escape symbol
end while (0)
`SV_RAND_CHECK(p.randomize());
# test.sv:13: Randomization failed p.randomize()"
`__FILE__ is a built-in macro that will display the name of the file the macro is called from
`__LINE__ is a built-in macro that will display the line of the file the macro is called from
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

6.3.3 The constraint solver

Solves constraint expressions
Same seed results in the same random values
Use different seeds in each nightly regression run.
Constraints may take a long time to solve if constraints are complex
Solver is specific to each simulator vendor/release.

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

10

6.3.4 What can be randomized?

2-state variables
4-state variables, only generates 2 state values ( no X/Z created)
Integers
Bit vectors
Arrays
Time
Real, string
Handle in constraint

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

11

Constraint Exercise 1
Write the SystemVerilog code for the following items:
1) Create a class Exercise1 containing two variables, 8-bit data
and 4-bit address. Create a constraint block that keeps address to
3 or 4.
2) In an initial block, construct an Exercise1 object and
randomize it. Check the status from randomization.

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

12

Constraint Exercise 1
Write the SystemVerilog code for the following items:
1) Create a class Exercise1 containing two variables, 8-bit data and 4-bit address.
Create a constraint block that keeps address to 3 or 4.
2) In an initial block, construct an Exercise1 object and randomize it. Check the status
from randomization.
class Exercise1;
rand bit [7:0] data;
rand bit [3:0] address;
constraint address_c {
address > 2;
address < 5;
}
endclass

// or
((address==3) || (address==4));
// or
address inside {[3:4]};

initial begin
Exercise1 MyExercise1;
MyExercise1 = new;
`SV_RAND_CHECK(MyExercise1.randomize());
`SV_RAND_CHECK(MyExercise1.randomize());
$display("data = %0d, address = %0d", MyExercise1.data,
`SV_RAND_CHECK(MyExercise1.randomize());
$display("data = %0d, address = %0d", MyExercise1.data,
`SV_RAND_CHECK(MyExercise1.randomize());
$display("data = %0d, address = %0d", MyExercise1.data,
`SV_RAND_CHECK(MyExercise1.randomize());
$display("data = %0d, address = %0d", MyExercise1.data,
end

# data = 136, address = 3

# data = 59, address = 3
# data = 80, address = 3
# data = 125, address = 4
MyExercise1.address);
MyExercise1.address);

MyExercise1.address);
MyExercise1.address);

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

13

6.4.2 Simple Expressions

Each constraint expression should contain only 1 relational
operator
<, <=, ==, >,=>
class Order_bad;
rand bit [7:0] lo, med, hi;
constraint bad {lo < med < hi;}
endclass

lo=20,
lo=114,
lo=186,
lo=214,

med=224,
med=39,
med=148,
med=223,

hi=164
hi=189
hi=161
hi=201

Constraint bad is broken down into multiple binary relational

expressions from left to right.
lo < med is evaluated. Results in 0 or 1
0 or 1 > hi is evaluated.
constraint good{lo < med;
med < hi;}
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

14

6.4.3 Equivalence Expressions

Suppose you want to constrain a value to be equal to an expression
class order;
rand bit [7:0] addr_mode, size, len;
constraint order_c {len == addr_mode*4 + size;}
endclass

len must be declared as random

Using = is a syntax error

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

15

6.4.4 Weighted Distributions

Weighted distributions cause a non-uniform distribution
Weights do not have to add up to 100% and can be variables
Cannot be used with randc
What would this be used for?
For a CPU want less or more of a particular opcode
For a datapath want max neg of 0, and max pos more often
constraint <constraint name> {<variable name> dist {<distribution>}};

constraint c_dist {
src dist {0:=40, [1:3]:=60};
dst dist {0:/40, [1:3]:/60};
}
:= operator indicates the weight is the same for all values
:/ operator indicates the weight is distributed across all values
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

16

6.4.4 Weighted Distributions := operator

:= operator indicates the weight is the same for all values
constraint src_dist { src dist {0:=40, [1:3]:=60} ;}

40 + 60 + 60 +60 = 220
src = 0, weight = 40/220 =18%
src = 1, weight = 60/220 =27%
src = 2, weight = 60/220 =27%
src = 3, weight = 60/220 =27%

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

17

6.4.4 Weighted Distributions :/ operator

:/ operator indicates the weight is distributed across all values

constraint dst_dist {dst dist {0:/40,

[1:3]:/60} ;}

dst = 0, weight = 40/100 =40%

dst = 1, weight = 20/100 =20%
dst = 2, weight = 20/100 =20%
dst = 3, weight = 20/100 =20%

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

18

6.4.4 Weighted Distributions (cont.)

Weights can be constants, ranges, or variables.
Using variables allows the weights to be adjusted on the fly
module sample6p9();

class BusOp;
0
1
2
typedef enum {BYTE, WORD, LWRD} length_e;
rand length_e len;
bit [31:0] w_byte=1, w_word=3, w_lwrd=5;
constraint c_len {
len dist {BYTE := w_byte,
WORD := w_word,
LWRD := w_lwrd};
}
endclass

classBusOp;
typedefenum{BYTE,WORD,LWRD}length_e;
randlength_elen;
bit[31:0]w_byte=1,w_word=3,w_lwrd=5;
constraintc_len{
lendist{BYTE:=w_byte,
WORD:=w_word,
LWRD:=w_lwrd};
}
endclass//BusOp

BusOp myBusOp;

# len = 0
# len = 1
# len = 2
# len = 2
# len = 0

initial begin
myBusOp = new();
$display("len = %0b", myBusOp.len);
void'(myBusOp.randomize());
$display("len = %0b", myBusOp.len);
void'(myBusOp.randomize());
$display("len = %0b", myBusOp.len);
void'(myBusOp.randomize());
$display("len = %0b", myBusOp.len);
void'(myBusOp.randomize());
$display("len = %0b", myBusOp.len);
end

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

19

Constraint Exercise 2
Modify the solution for Exercise 1 to create a new class
Exercise2 so that:
1. data is always equal to 5
2. Probability of address = 4d0 is 10%
3. Probability of address being between [1:14] is 80%
4. Probability of address = 4d15 is 10%
Demonstrate its usage by generating 20 new data and address
values and check for error.

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

20

Constraint Exercise 2
package my_package;
class Exercise2;
rand bit [7:0] data;
rand bit [3:0] address;
constraint data_c{data == 5;}
constraint address_dist {
address dist{0:=10, [1:14]:/80, 15:=10};
}
function void print_all;
$display("data = %d, address = %d", data, address);
endfunction
endclass // Exercise2
endpackage

program automatic test;

import my_package::*;

# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =
# data =

initial begin
Exercise2 MyExercise2;
repeat (20) begin
MyExercise2 = new;
`SV_RAND_CHECK(MyExercise2.randomize());
MyExercise2.print_all();
end
end
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4
endprogram

5, address = 2
5, address = 12
5, address = 7
5, address = 4
5, address = 14
5, address = 15
5, address = 7
5, address = 10
5, address = 8
5, address = 1
5, address = 6
5, address = 15
5, address = 15
5, address = 1
5, address = 12
5, address = 3
5, address = 15
5, address = 8
5, address = 10
5, address = 7

21

6.4.5 Set membership and the inside op

Alternative to {var>value1 ; var<value2} is the inside keyword
constraint address_c {address > 2; address < 5;}
constraint address_range{address inside{[3:4]};}

equivalent
Using the ! operator can exclude ranges
constraint c_range {
!(c inside{[lo:hi]});
}

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

22

6.4.6 Using an array in a set

Suppose you want to create multiple equivalence constraints
For example: f can only equal 1, 2, 3, 5, 8
rand bit [7:0] f;
constraint c_fibonacci {
(f == 1) || (f==2) || (f==3) || (f==5) || (f==8));}

Alternate solution is to store the values in an array

rand bit [7:0] f;
bit [31:0] vals[]= {1,2,3,5,8};
constraint c_fibonacci {f inside {vals;}}

Can specify that values in the array are NOT to be chosen

rand bit [7:0] notf;
bit [31:0] vals[]= {1,2,3,5,8};
constraint c_fibonacci {!(notf inside vals);}
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

23

6.4.6 Using an array in a set

rand bit [7:0] f;
bit [31:0] vals[]= {1,2,3,5,8};
constraint c_fibonacci {f inside {vals;}}
#
#
#
#
#

count[1]=
count[2]=
count[3]=
count[5]=
count[8]=

3943
4054
3922
4094
3987

**************************************
***************************************
**************************************
****************************************
**************************************

Values listed multiple times in array still occur with same

as value listed once in array:
Not in book but needed for
rand bit [7:0] f;
bit [31:0] vals[]= {1,8,8,8,8};
constraint c_fibonacci {f inside {vals;}}

Questasim to compile

# count[1]=10075 ****************************************
# count[8]= 9925 ***************************************

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

24

6.4.6 Using an array in a set

class Days;
typedef enum {SUN, MON, TUE, WED, THU, FRI, SAT} days_e;
days_e choices[$];

rand days_e choice;

constraint cday {choice inside {choices};};
endclass // Days
initial begin
Days days;
days = new;
// Nothing in queue yet
$display("choices queue = %p", days.choices);

#
#
#
#
#
#

choices queue = '{}

choices queue = '{SUN, SAT}
Random weekend day SUN

Values
listed
timesSAT};
in array still occur with same
days.choices
= { multiple
Days:: SUN, Days::
Now SUN listed
and SAT in
choices
choices queue = '{MON, TUE,
as// value
once
in queue
array:
$display("choices queue = %p", days.choices);

WED, THU, FRI}

Random week day TUE

`SV_RAND_CHECK(days.randomize());
$display("Random weekend day %s\n", days.choice.name());
days.choices = {Days::MON, Days::TUE, Days::WED, Days::THU, Days::FRI};
$display("choices queue = %p", days.choices);
`SV_RAND_CHECK(days.randomize());
$display("Random week day %s\n", days.choice.name());
end

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

25

6.4.7 Bidirectional Constraints

Constraint blocks are not procedural but declarative.
All constraints are active at the same time.
rand bit [15:0] r,s,t;
constraint c_bidir {
r < t;
s == r;
t < 10;
s > 5;}

Solution

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

26

6.4.8 Implication Constraints

Suppose you want to impose different constraints depending on a var
Remember that the constraint is bidirectional
Solution 1:

constraint mode_c {
if (mode == small)
len < 10;
else if (mode == large)
len > 100;
}
{(a==1)->(b==0)};

equivalent
{if (a==1) b==0;}
equivalent

Solution

a
(mode==small)

b
(mode==large)

Comment

10>= len >= 100

len < 10 or len > 100

mode == small or
mode == large

{!(a==1) || (b==0);}

equivalent
{(a==0) || (b==0);}

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

27

6.4.8 Implication Constraints

Suppose you want to impose different constraints depending on a var
Solution 2:
Syntax is expression -> constraint set
If expression is true
Constraint solver should satisfy the constraint set
If expression is false
The random numbers generated are unconstrained
by the constraint set.
Boolean equivalent is:
Solution
a
b
(!a || b)
constraint mode_c {
(mode == small) -> len < 10;
(mode == large) -> len > 100;
}

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

28

6.4.9 Equivalence operator

The equivalence operator <-> is bidirectional.
Relationship is if and only if.
A<->B is defined as ((A->B) && (B->A))
rand bit d, e;
constraint c { d==1 <-> e==1; }

when d == 1, the variable e must be 1

when e==1, d can be 0 or 1

(d==1) -> (e==1)

(e==1) =false

(e==1)=true

(d==1)=false

true

true

(d==1)=true

false

true

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

29

6.4.1 Constraint Introduction - Exercise

class Stim;
const bit [31:0] CONGEST_ADDR = 42;
typedef enum {READ, WRITE, CONTROL} stim_e;
randc stim_e kind;
rand bit [31:0] len, src, dst;
bit congestion_test;

What are the constraints on len,

dst, and src for this code?

constraint c_stim {
len < 1000;
len > 0;
if (congestion_test) {
dst inside {[CONGEST_ADDR-10:CONGEST_ADDR+10]};
src == CONGEST_ADDR;
}
else
src inside {0, [2:10], [100:107]};
}
endclass
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

30

6.4.1 Constraint Introduction - Exercise

class Stim;
const bit [31:0] CONGEST_ADDR = 42;
typedef enum {READ, WRITE, CONTROL} stim_e;
randc stim_e kind;
rand bit [31:0] len, src, dst;
bit congestion_test;
constraint c_stim {
len < 1000;
len > 0;
if (congestion_test) {
dst inside {[CONGEST_ADDR-10:CONGEST_ADDR+10]};
src == CONGEST_ADDR;
}
else
src inside {0, [2:10], [100:107]};
}
endclass
len must be between 1 and 999 inclusive

if congestion_test == 1, dst must be inside 42-10

= 32 to 42+10=52 and src=42
else src can take values of 0, 2 to 10, and 100 to
107. dst is unconstrained
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

31

6.5 Solution Probabilities

Its important to understand how constraints affect the probability of
the solution.
Unconstrained:
class Unconstrained;
rand bit x;
rand bit [1:0] y;
endclass

As there are no constraints,

Each has the same probability

Solution

Probability

1/8

1/8

1/8

1/8

1/8

1/8

1/8

1/8

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

32

6.5.2 Implication
class Imp1;
rand bit x;
rand bit [1:0] y;
constraint c_xy {
(x==0) -> y==0;
}
endclass

Solution

Probability

1/2

1/8

1/8

1/8

1/8

(x==0) -> (y==0)

(y==0)=false

(y==0)=true

(x==0)=false

true

true

(x==0)=true

false

true

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

33

6.5.3 Implication and bidirectional

constraints
class Imp2;
rand bit x;
rand bit [1:0] y;
constraint c_xy {
y>0;
(x==0)->y==0;
}
endclass

Solution

Probability

1/3

1/3

1/3

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

34

6.5.4 Guiding Distribution with solve/before

Solve before tells the solver to solve for 1 variable before
another.
The possible solutions does not change, just the probability.
When x = 0, y must be 0, when x =1, y can be 0, 1, 2, or 3.
class SolveBefore;
rand bit x;
rand bit [1:0] y;
constraint c_xy {
(x==0)->y==0;
solve x before y;
}
endclass

Solution

Probability

1/2

1/8

1/8

1/8

1/8

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

35

solve y before x;
When y = 1, 2, or 3, X must be 1. When y = 0, x can be 0 or 1.
class Imp1;
rand bit x;
rand bit [1:0] y;
constraint c_xy {
(x==0)->y==0;
solve y before x;
}
endclass

Solution

Probability

1/8

1/8

1/4

1/4

1/4

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

36

Solution Probabilities Exercise

Complete the table for the following constraints
class MemTrans;
rand bit x;
rand bit [1:0] y;
constraint c_xy {
y inside{[x:3]};
solve x before y;
}
endclass

Solution

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

Probability

37

Solution Probabilities Exercise

Complete the table for the following constraints
Probability of x = 0 is
Probability of y = 0, 1, 2, 3 is
Total = * = 1/8

class MemTrans;
rand bit x;
rand bit [1:0] y;
constraint c_xy {
y inside{[x:3]};
solve x before y;
}
endclass

Solution

Probability

1/8

1/8

1/8

1/8

1/6

1/6

1/6

Probability of x = 1 is
Probability of y = 0 as not inside 1:3
Probability of y = 1, 2, 3 is 1/3
Total = 1/2 * 1/3 = 1/6
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

38

6.6 Controlling Multiple Constraint Blocks

Use the constraint_mode() function to turn constraints on/off
<handle>.constraint_mode(<0/1>);
<handle>.<constraint>.constraint_mode(<0/1>);

class Packet
rand bit [31:0] length;
constraint c_short {length inside {[1:32]};}
constraint c_long {length inside {[1000:1023]};}
endclass
Packet p:
initial begin
p=new();
p.c_short.constraint_mode(0);
`SV_RAND_CHECK(p.randomize());
transmit(p);
....

...
p.constraint_mode(0);
p.c_short.constraint_mode(1);
`SV_RAND_CHECK(p.randomize());
transmit(p);
end // initial

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

39

6.7 Valid Constraints

A suggested technique to creating valid stimulus is to create valid
constraints
Turn the constraint off to test the systems response to invalid
stimulus.
For example, suppose a read-modify-write command is only valid if
the length is a long word.
class Transaction;
typedef enum {BYTE, WORD, LWRD, QWRD} length_e;
typedef enum {READ, WRITE, RMW, INTR} access_e;
rand length_e length;
rand access_e access;
constraint valid_RMW_LWRD {
(access == RMW) -> (length == LWRD);
}
endclass
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

40

6.8 In-line Constraints

In-line constraints create constraints outside of the class.
Add to existing constraints if they are enabled.
For example, a single test needs to be written with tighter than
usual address constraints
class Transaction;
rand bit [31:0] addr, data;
constraint c1 {addr inside{[0:100], [1000:2000]};}
endclass
intitial begin
Transaction t;
t=new();
`SV_RAND_CHECK(t.randomize() with {addr >=50; addr <=1500;
data <10;});
driveBus(t);
`SV_RAND_CHECK(t.randomize() with {addr ==2000; data >10;});
driveBus(t);
end
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

41

Valid and In-line Constraint Exercise

For the following class create:
1. A constraint that limits read transactions addresses to the range
0 to 7, inclusive
2. Write behavioral code to turn off the above constraint. Construct
and randomize a MemTrans object with an in-line constraint that
limits read transaction addresses to the range 0 to 8, inclusive.
Test that the in-line constraint is working.
class MemTrans;
rand bit rw; // read if rw = 0, write if rw = 1
rand bit [7:0] data_in;
rand bit [3:0] address;
endclass

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

42

Valid and In-line Constraint Exercise (cont.)

For the following class create:
1. A constraint that limits read transactions addresses to the range 0 to 7, inclusive
2. Write behavioral code to turn off the above constraint. Construct and randomize a MemTrans
object with an in-line constraint that limits read transaction addresses to the range 0 to 8,
inclusive. Test that the in-line constraint is working.
class MemTrans;
rand bit rw; // read if rw = 0, write if rw = 1
rand bit [7:0] data_in;
rand bit [3:0] address;
constraint valid_rw_addr {(rw == 0)->(address inside {[0:7]});}
endclass // MemTrans
MemTrans MyMemTrans;
initial begin
MyMemTrans = new();
MyMemTrans.valid_rw_address.constraint_mode(0);
`SV_RAND_CHECK(MyMemTrans.randomize() with {(rw == 0)->(address inside {[0:8]});});
end

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

43

6.9 pre_randomize/post_randomize
Implicitly called before/after every call to randomize()
void function
Cannot consume time.
Can only call other functions.
Does not return a value
Overload to add your functionality
post_randomize() is good for cleaning up
$dist_exponential (seed_expression, mean_expression)
mean_expression parameter causes the average value of the return value to
approach the mean. The mean_expression must be an integer greater than
zero (0).

Other distributions available:

Integer and real
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

44

6.9 pre_randomize/post_randomize
module sample6p35_tb();
class Bathtub;
int value;
int WIDTH = 50, DEPTH = 6, seed = 1;
function void pre_randomize();
value = $dist_exponential(seed, DEPTH);
if (value > WIDTH) value = WIDTH;
if ($urandom_range(1))
value = WIDTH - value;
$display("value = %0d", value);
endfunction // pre_randomize
endclass // Bathtub

Bathtub bathtub;
initial begin
bathtub = new();

80
70
60
50
40
30
20
10
0
0

10

20

30

40

50

60

repeat(250)
bathtub.pre_randomize;
end
endmodule // sample6p35_tb

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

45

6.11 Constraint Tips and Techniques

Instead of hardcoding constraints, use variables with defaults
Allows the constraint to be modified without modifying the class
Allows invalid stimulus to be generated
class Packet;
rand bit [31:0] length;
constraint c_length {
length inside {[1:100]};
}
endclass

class Packet;
rand bit [31:0] length;
int max_length= 100;
constraint c_length {
length inside {[1:max_length]};
}
endclass

initial begin
Packet p1 = new();
p1.max_length = 200;
p1.randomize();
end
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

46

6.11.2 Using Nonrandom Values

constraint_mode() turns on/off constraints
rand_mode() makes a variable or every variable in an object non-

random
class Packet;
rand bit [7:0] length;
constraint c_length{length > 0;}
..... // Other constraints depending on length
endclass
initial begin
Packet p = new();
`SV_RAND_CHECK(p.randomize());
Make length nonrandom
p.length.rand_mode(0);
Create an invalid length
p.length = 0;
`SV_RAND_CHECK(p.randomize());
Value for length will be
p.rand_mode(0);
included in constraint solution
end
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

47

6.11.3 Checking Values using Constraints

If you change the value of random variables how do you know all your random variables
are still valid?
Use a call to <handle>.randomize(null) to check.
class Transaction;
rand bit [31:0] addr, data;
constraint c1 {addr inside{[0:100], [1000:2000]};}
endclass
Transaction t;
initial begin
t=new();
`SV_RAND_CHECK(t.randomize());
t.addr = 200;
`SV_RAND_CHECK(t.randomize(null));
end

...Randomization failed

`define SV_RAND_CHECK(r) \
do begin \
if (!(r)) begin \
$display("%s:%0d: Randomization failed \"%s\"", \
`__FILE__, `__LINE__, `"r`"); \
$finish; \
end \
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4
end while (0)

48

6.11.4 Randomizing Individual Variables

Can pass variables to randomize() to randomize only a subset of
variables
class Rising;
bit [7:0] low;
rand bit [7:0] med, hi;
constraint up { low < med; med < hi; }
endclass
initial begin
Rising r;
r = new();
`SV_RAND_CHECK(r.randomize());
`SV_RAND_CHECK(r.randomize(med));
`SV_RAND_CHECK(r.randomize(low));
`SV_RAND_CHECK(r.randomize(low, med));
end
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

49

6.11.5 Turn Constraints Off and On

Use many simple constraints instead of 1 complex constraint
Turn on the constraints needed
class Instruction;
typedef enum {NOP, HALT, CLR, NOT} opcode_e;
rand opcode_e opcode;
bit [1:0] n_operands;
constraint c_operands{
if (n_operands == 0)
(opcode == NOP) || (opcode == HALT);
else if (n_operands == 1)
(opcode == CLR) || (opcode == NOT);
.....
}
endclass

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

50

6.11.5 Turn Constraints Off and On(cont.)

class Instruction;
typedef enum {NOP, HALT, CLR, NOT} opcode_e;
rand opcode_e opcode;
constraint c_no_operands{
(opcode == NOP) || (opcode == HALT);}
constraint c_one_operand{
(opcode == CLR) || (opcode == NOT);}
.....
}
endclass
initial begin
Instruction instr = new();
instr.constraint_mode(0);
instr.c_no_operands.constraint_mode(1);
`SV_RAND_CHECK(instr.randomize());
end
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

51

6.12 Common Randomization Problems

Using a signed variable isnt an issue if you control the values
for (int i=0;i<=5;i++)

However, a randomized signed variable will produce negative values

class SignedVars;
rand byte pkt1_len, pk2_len;
constraint total_len {pkt1_len + pk2_len == 64;}
endclass

Some valid solutions of {pkt1_len, pkt2_len} are:

(32,32)
(2,62)
(-63, 127)

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

52

6.12.1 Use Signed Values with care

Might be temped to declare pkt1_len, pk2_len as large unsigned
class Vars32;
rand bit [31:0] pkt1_len, pk2_len;
constraint total_len {pkt1_len + pk2_len == 64;}
endclass

A valid solution of {pkt1_len, pkt2_len} is

(32h8000_0040, 32h8000_0000) = 33h140 = 32h40 == 32d64

One solution is to constrain the max values of pkt1_len and pk2_len

Best solution is to only use values as wide as required
class Vars8;
rand bit [7:0] pkt1_len, pkt2_len;
constraint total_len {pkt1_len + pkt2_len == 9d64;}
endclass
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

53

6.12.2 Solver performance Tips

Operators like /, *, % will slow down the constraint solver
Use left shifts for multiply by 2
Use right shifts for divide by 2
Use bit-wise AND and mask for modulo by 2
x % y (x & (y 1))
9%4 = 1001
& 0011
0001

8%4 = 1000
& 0011
0000

Use $urandom or $urandom_range instead of the solver.

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

54

6.13 Iterative and Array Constraints

Constraining the size of an array
class dyn_size;
rand bit [31:0] d[];
constraint d_size {d.size() inside {[1:10]};}
endclass

Constrains the array element

Constraining the elements of an array contents between 1 and 255
class good_sum5;
rand uint len[];
constraint c_len {foreach (len[i])
Constrains the sum
len[i] inside {[1:255]};
of the array contents to len.sum < 1024;
len.size() inside {[1:8]};
less than 1024
}
endclass
Constrains the # of array

elements from 1 to 8

The solver can handle

hundreds of constraints.

sum =
sum =
sum =
see

1011, val = 83 249 197 187 152

1012, val = 213 252 213 44 196
370, val = 118 76 176
Sample 6.59/6.60

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

95
20

40
20

8
54

55

`define SV_RAND_CHECK(r) \
do begin \
if (!(r)) begin \
$display("%s:%0d: Randomization failed \"%s\"", \
`__FILE__, `__LINE__, `"r`"); \
$finish; \
# sum= 137, val= 137
end \
# sum= 519, val=
1
end while (0)
# sum= 81, val= 81
# sum= 408, val= 158
typedef int unsigned uint;
module slide55_tb();
test t1();

endmodule // slide55_tb
program automatic test;
class good_sum5;
rand uint len[];
constraint c_len {foreach (len[i])
len[i] inside {[1:255]};
len.sum < 1024;
len.size() inside {[1:8]};}

#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#

sum= 624,
sum=1003,
sum= 142,
sum= 520,
sum= 135,
sum= 719,
sum= 798,
sum= 100,
sum= 529,
sum= 251,
sum= 879,
sum=1001,
sum= 763,
sum= 160,
sum= 702,
sum= 516,

function void display();

$write("sum=%4d, val=", len.sum());
foreach(len[i]) $write("%4d ", len[i]); $display;
endfunction
endclass

val=
val=
val=
val=
val=
val=
val=
val=
val=
val=
val=
val=
val=
val=
val=
val=

48
184
142
93
135
115
23
26
214
251
132
13
209
88
184
25

146

199

16

157

18
245
51

7
141
110

225
190
216

175

22

60

181

164

64
60
74
179

140
134

220
132

113
228
2
61
175
53

217
218
166
11
70
158

188

32

47

180
4

241

201

252
244
117

165
205
156

93
66

47

24
32

17
37

6
74

151
18

136

75
119

good_sum5 c;
initial begin
c = new();
repeat (20) begin
`SV_RAND_CHECK(c.randomize());
c.display();
end
end
endprogram

56

6.13.6 Randomizing an array of handles

Need to allocate all the elements before randomization as

the random solver never constructs objects.

For a dynamic array, allocate the maximum number of

elements needed and then use a constraint to resize array.
A dynamic array of handles can remain the same size or
shrink during randomization but it can never increase in
size.

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

57

6.13.6 Randomizing an array of handles

`define SV_RAND_CHECK(r) \
do begin \
if (!(r)) begin \
$display("%s:%0d: Randomization failed \"%s\"", \
`__FILE__, `__LINE__, `"r`"); \
$finish; \
end \
end while (0)
module sample6p67_tb();
parameter MAX_SIZE = 25;
class RandStuff;
rand bit [31:0] value;
endclass // RandStuff
class RandArray;
rand RandStuff array[];
constraint c {array.size() inside {[1:MAX_SIZE]};}
function new();
array = new[MAX_SIZE];
foreach (array[i])
array[i] = new();
endfunction; // new
endclass // RandArray

#
#
#
#
#
#
#
#
#
#
#
#
#
#
#
#

RandArray ra;
initial begin
ra = new();
`SV_RAND_CHECK(ra.randomize());
foreach (ra.array[i])
$display("ra.array[%2d]: ",i, ra.array[i].value);
end
endmodule // sample6p67

ra.array[ 0]:
ra.array[ 1]:
ra.array[ 2]:
ra.array[ 3]:
ra.array[ 4]:
ra.array[ 5]:
ra.array[ 6]:
ra.array[ 7]:
ra.array[ 8]:
ra.array[ 9]:
ra.array[10]:
ra.array[11]:
ra.array[12]:
ra.array[13]:
ra.array[14]:
ra.array[15]:

3507107891
4013053989
2773609279
528996684
3476166007
3634918746
1796308729
1730985904
1222782346
1298121202
1119090603
4220722116
1074911065
1269454440
2256144349
806154056

max decimal value for 32 bits:

4294967295
58

6.13.6 Randomizing an array of handles

For the homework in Chap6 you might create a fixed array of handles
localparam TESTS = 10;
Transaction Transaction_array[TESTS];

Can create a random array of handles

class TransactionArray;
rand Transaction Transaction_array[];
constraint c {Transaction_array.size() inside
{[1:TESTS]}; }
function new();
Transaction_array = new[TESTS];
foreach (Transaction_array[i])
Transaction_array[i] = new();
endfunction;
endclass
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

59

Array Constraint Exercise

Create a class for a graphics image that is 10x10 pixels. The value for
each pixel can be randomized to black or white. Randomly generate
an image that is, on average, 20% white.

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

60

package my_package;
parameter int HEIGHT = 10;
parameter int WIDTH = 10;
parameter int PERCENT_WHITE = 20;
typedef enum bit {BLACK, WHITE} colors_t;
class Screen;
rand colors_t pixels [HEIGHT] [WIDTH];
constraint colors_c {foreach (pixels[i,j]) pixels[i][j] dist { BLACK :=100-PERCENT_WHITE,
WHITE := PERCENT_WHITE};
// or
// constraint colors_c {(pixels.sum() with (item.sum() with (int'(item))) == PERCENT_WHITE);}
// This function is not necessary for the in-class solution
function void print_screen;
int
count = 0;
foreach (pixels[i])
$display("%p", pixels[i]);
foreach (pixels[i,j])
count += pixels[i][j];
$display("Num of pixels=WHITE = %0d", count);
endfunction //
endclass
endpackage
`default_nettype none
program automatic test;
import my_package::*;
Screen my_screen;
initial begin
my_screen=new();
`SV_RAND_CHECK(my_screen.randomize());
my_screen.print_screen;
end
endprogram
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

61

6.14 Atomic Stimulus Gen vs Scenario Gen

Up to this point weve covered creating atomic random transactions
Real-world scenarios have long sequences of transactions
Methods to create scenarios
An atomic generator with history
Randsequence
Random array of objects

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

62

6.14.1 An Atomic generator with History

Suppose we want to create a sequence of transactions where back to
back writes are not allowed
package my_package;
typedef enum {READ, WRITE} rw_t;
class Transaction;
rw_t old_rw;
rand rw_t rw;
rand bit [31:0] addr, data;
constraint rw_c{if (old_rw == WRITE) rw != WRITE;};
function void post_randomize;
old_rw = rw;
endfunction
function void print_all;
$display("addr = %d, data = %d, rw = %s", addr,
data, rw);
endfunction
endclass
endpackage
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4
63

module slide63_tb();
import my_package::*;
Transaction t;
initial begin
for (int i=0;i<=50;i++) begin
t = new();
`SV_RAND_CHECK (t.randomize());
t.print_all;
end
end
endmodule // slide63_tb
package my_package;
typedef enum {READ, WRITE} rw_t;

class Transaction;
rw_t old_rw;
rand rw_t rw;
rand bit [31:0] addr, data;
constraint rw_c{if (old_rw == WRITE) rw != WRITE;};
function void post_randomize;
old_rw = rw;
endfunction //
function void print_all;
$display("addr = %d, data = %d, rw = %s", addr,data, rw);
endfunction
endclass
endpackage
64

6.14.1 Usage of atomic generator w/ history

initial begin
for (int i=0;i<=50;i++) begin
t=new();
`SV_RAND_CHECK (t.randomize());
t.print_all;
end
Back to back writes
end

initial begin
t=new();
for (int i=0;i<=50;i++) begin
`SV_RAND_CHECK (t.randomize());
t.print_all;
end
end

#
#
#
#
#
#
#
#
#

rw
rw
rw
rw
rw
rw
rw
rw
rw

=
=
=
=
=
=
=
=
=

READ
WRITE
READ
READ
READ
WRITE
WRITE
WRITE
WRITE

#
#
#
#
#
#
#
#

rw
rw
rw
rw
rw
rw
rw
rw

=
=
=
=
=
=
=
=

READ
WRITE
READ
WRITE
READ
WRITE
READ
READ

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

65

6.14.2 Random Array of Objects

Create as many transaction objects as needed

Constrain the current object by previous objects

class Transaction;
rand rw_t rw;
rand bit [31:0] addr, data;
endclass
class RandTransaction;
rand Transaction trans_array[];
constraint rw_c {foreach (trans_array[i])
if ((i>0) && (trans_array[i-1].rw == WRITE))
trans_array[i].rw != WRITE;}
function new();
trans_array = new[TESTS];
foreach (trans_array[i])
trans_array[i] = new();
endfunction;
endclass
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

66

module slide66_tb();
parameter TESTS = 5;
typedef enum {READ, WRITE} rw_t;
class Transaction;
rand rw_t rw;
rand bit [31:0] addr, data;
endclass

#
#
#
#
#
#
#
#
#
#

i
i
i
i
i
i
i
i
i
i

=
=
=
=
=
=
=
=
=
=

0,
1,
2,
3,
4,
0,
1,
2,
3,
4,

class RandTransaction;
rand Transaction trans_array[];
constraint rw_c {foreach (trans_array[i])
if ((i>0) && (trans_array[i-1].rw == WRITE))
trans_array[i].rw != WRITE;}

rw
rw
rw
rw
rw
rw
rw
rw
rw
rw

=
=
=
=
=
=
=
=
=
=

'{rw:READ,
'{rw:READ,
'{rw:READ,
'{rw:READ,
'{rw:READ,
'{rw:READ,
'{rw:READ,
'{rw:READ,
'{rw:READ,
'{rw:READ,

addr:0, data:0}
addr:0, data:0}
addr:0, data:0}
addr:0, data:0}
addr:0, data:0}
addr:806154056, data:1796290893}
addr:3362437678, data:3276263896}
addr:2421260686, data:815502112}
addr:3293998593, data:1926967213}
addr:3933255193, data:2450738056}

function new();
trans_array = new[TESTS];
foreach (trans_array[i]) begin
trans_array[i] = new();
end
endfunction;
endclass
RandTransaction t;
initial begin
t = new();
foreach(t.trans_array[i])
$display("i = %0d, rw = %p",i, t.trans_array[i]);
`SV_RAND_CHECK(t.randomize());
foreach(t.trans_array[i])
$display("i = %0d, rw = %p",i, t.trans_array[i]);
end
endmodule
Chapter 6 Copyright 2011 G. Tumbush, C.
Spear v1.4

67

6.14.4 Randsequence
SystemVerilog introduces the randsequence construct
Uses a Backus-Naur Form (BNF) like syntax to describe the grammar
of the transaction
initial begin
for (int i=0; i<15; i++) begin
randsequence (stream)
stream : cfg_read := 1 | weights
io_read := 2 |
mem_read := 5 ;
cfg_read : { cfg_read_task; } |
{ cfg_read_task; } cfg_read;
mem_read : { mem_read_task; } |
{ mem_read_task; } mem_read;

cfg_read can either be a single

call to cfg_read_taskor a call
to the task followed by
another cfg_read

io_read : { io_read_task; } |
{ io_read_task; } io_read;
endsequence
end
end

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

68

Scenario Generation Exercise

Expand the Transaction class in slide 6.14.1 An Atomic generator
with History or slide 6.14.3 Random Array of Objects so that back
to back transactions of the same type do not have the same address.

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

69

Scenario Generation Exercise

Expand the Transaction class in slide 6.14.1 An Atomic generator
with History or slide 6.14.3 Random Array of Objects so that back
to back transactions of the same type do not have the same address.
If you modified the An Atomic generator with History (6.14.1) add to the
class:
bit [31:0] old_addr;
constraint addr_c{if (old_rw == READ)
old_addr != addr;}
to the post_randomize function add old_addr = addr;

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

70

6.17 Random Device Configuration

Randomizing data is not enough, randomize the environment
Using an example of a 4-port Ethernet switch
class EthCfg;
rand bit [3:0] in_use;
rand bit [47:0] mac_addr[4];
rand bit [3:0] is_100;
rand uint run_for_n_frames;

//
//
//
//

Ports used in test: 3, 2, 1, 0

MAC addresses
100mb mode for ports 3, 2, 1, 0
# frames in test

// Force some addr bits when running in unicast mode

constraint local_unicast {
foreach (mac_addr[i])
mac_addr[i][41:40] == 2'b00;
}
constraint reasonable {
// Limit test length
run_for_n_frames inside {[1:100]};}
endclass : eth_cfg

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

71

6.17 Random Device Configuration (cont.)

class Environment;
EthCfg cfg; EthGen gen[4]; EthMii drv[4];
function new(); cfg = new(); endfunction
// Use random config. to build the environment
function void build();
foreach (gen[i]) begin
gen[i] = new(); drv[i] = new();
if (cfg.is_100[i]) drv[i].set_speed(100); end
endfunction
function void gen_cfg;
`SV_RAND_CHECK(cfg.randomize());
endfunction
task run();
foreach (gen[i])
if (cfg.in_use[i]) begin
// Only start the testbench transactors that are in-use
gen[i].run();
...
end
endtask
endclass : Environment
Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

72

6.17 Random Device Configuration (cont.)

program test;
Environment env;
initial begin
env = new();
env.gen_cfg;
env.build();
env.run();
env.wrap_up();
end
endprogram

Turn on/off constraints here

Calls randomize here

Override generated values here

Chapter 6 Copyright 2011 G. Tumbush, C. Spear v1.4

73