ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

School of Electrical and Computer Engineering
Cornell University
revision: 2015-09-22-23-01

Introduction

Verilog Modeling: Synthesizable vs. Non-Synthesizable RTL

Verilog Basics: Data Types, Operators, and Conditionals

3.1

Hello World . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.2

Logic Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.3

Shift Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.4

Arithmetic Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

10

3.5

Relational Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

11

3.6

Concatenation Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

13

3.7

Enum Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

14

3.8

Struct Data Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

16

3.9

Ternary Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

18

3.10 If Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

19

3.11 Case Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

20

3.12 Casez Statements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

21

Registered Incrementer

21

4.1

RTL Model of Registered Incrementer . . . . . . . . . . . . . . . . . . . . . . . . . . . .

22

4.2

Simulating a Model using iverilog . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

24

4.3

Verifying a Model with Unit Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26

4.4

Reusing a Model with Structural Composition . . . . . . . . . . . . . . . . . . . . . . .

28

4.5

Parameterizing a Model with Static Elaboration . . . . . . . . . . . . . . . . . . . . . .

30

Sort Unit

32

5.1

Flat Sorter Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

32

5.2

Using Verilog Line Traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

36

5.3

Structural Sorter Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

37

5.4

Evaluating the Sorter Using a Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . .

39

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

Greatest Common Divisor: Verilog Design Example

39

6.1

Control/Datapath Split Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . .

40

6.2

Evaluating GCD Unit using a Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . .

45

TravisCI for Continuous Integration

46

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

1. Introduction
In the lab assignments for this course, we will be using the PyMTL hardware modeling framework
for functional-level modeling, verification, and simulator harnesses. Students can choose to use
either PyMTL or Verilog to do their register-transfer-level (RTL) qmodeling. If you are planning to
use PyMTL, then you do not need to complete this tutorial. If you are planning to use Verilog, you
should still complete the PyMTL tutorial since we will always be using PyMTL for some aspects of
the lab assignment.
This tutorial briefly reviews the basics of the Verilog hardware description language, but primarily focuses on how we can integrate Verilog RTL modeling into our PyMTL framework. Although
we will be using the open-source tool Icarus Verilog (iverilog) for compiling some simple Verilog
models into simulators, we will primarily be using the open-source tool Verilator (verilator) as our
Verilog simulator. PyMTL has built-in support for testing Verilog simulators created using Verilator.
As in the PyMTL tutorial, we will also be using GTKWave (gtkwave) for viewing waveforms. All
tools are installed and available on amdpool. This tutorial assumes that students have completed the
Linux and Git tutorials.
To follow along with the tutorial, access the ECE computing resources, and type the commands
without the % character. In addition to working through the commands in the tutorial, you should
also try the more open-ended tasks marked with the H symbol.
Before you begin, make sure that you have sourced the ece4750-setup script or that you have added
it to your .bashrc script, which will then source the script every time you login. Sourcing the setup
script sets up the environment required for this class.
You should start by forking the tutorial repository on GitHub. Start by going to the GitHub page for
the tutorial repository located here:
https://github.com/cornell-ece4750/ece4750-tut4-verilog
Click on Fork in the upper right-hand corner. If asked where to fork this repository, choose your
personal GitHub account. After a few seconds, you should have a new repository in your account:
https://github.com/<githubid>/ece4750-tut4-verilog
Where <githubid> is your GitHub ID, not your NetID. Now access amdpool and clone your copy of
the tutorial repository as follows:
%
%
%
%
%
%

source setup-ece4750.sh
mkdir -p ${HOME}/ece4750
cd ${HOME}/ece4750
git clone https://github.com/<githubid>/ece4750-tut4-verilog.git tut4
cd tut4/sim
TUTROOT=${PWD}

2. Verilog Modeling: Synthesizable vs. Non-Synthesizable RTL

Verilog is a powerful language that was originally intended for building simulators of hardware as
opposed to models that could automatically be transformed into hardware (e.g., synthesized to an
FPGA or ASIC). Given this, it is very easy to write Verilog code that does not actually model any kind
of realistic hardware. Indeed, we actually need this feature to be able to write clean and productive
assertions and line tracing. Non-synthesizable Verilog modeling is also critical when implementing
3

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

test harnesses. So students must be very diligent in actively deciding whether or not they are
writing synthesizable register-transfer-level models or non-synthesizable code. Students must
always keep in mind what hardware they are modeling and how they are modeling it!
Students design work will almost exclusively use synthesizable register-transfer-level (RTL) models.
It is acceptable to include a limited amount of non-synthesizable code in students designs for the
sole purpose of debugging, assertions, or line tracing. If the student includes non-synthesizable code
in their actual design (i.e., not the test harness), they must explicitly demarcate this code by wrapping
it in ifndef SYNTHESIS and endif. This explicitly documents the code as non-synthesizable and
aids automated tools in removing this code before synthesizing the design. If at any time students
are unclear about whether a specific construct is allowed in a synthesizable concurrent block, they
should ask the instructors.
Appendix A includes a table that outlines which Verilog constructs are allowed in synthesizable
RTL, which constructs are allowed in synthesizable RTL with limitations, and which constructs are
explicitly not allowed in synthesizable RTL. There are no limits on using the Verilog preprocessor,
since the preprocessing step happens at compile time.

3. Verilog Basics: Data Types, Operators, and Conditionals

We will begin by writing some very basic code to explore Verilog data types, operators, and conditionals. We will not be modeling actual hardware yet; we are just experimenting with the language.
Start by creating a build directory to work in.
% mkdir ${TUTROOT}/build
% cd ${TUTROOT}/build
3.1. Hello World
Create a new Verilog source file named hello-world.v with the contents shown in Figure 1 using
your favorite text editor. A module is the fundamental hardware building block in Verilog, but for
now we are simply using it to encapsulate an initial block. The initial block specifies code which
should be executed at the beginning of time when the simulator starts. Since real hardware cannot
do anything at the beginning of time initial blocks are not synthesizable and you should not
use them in the synthesizable portion of your designs. However, initial blocks can be useful for
test harnesses and when experimenting with the Verilog language. The initial block in Figure 1
contains a single call to the display system task which will output the given string to the console.
We will be using iverilog to compile Verilog models into simulators in the beginning of this tutorial before we turn our attention to using Verilator. You can run iverilog as follows to compile
hello-world.v.

1
2
3
4
5

module top;
initial begin
$display( "Hello World!" );
end
endmodule

Code at https://github.com/cbatten/x/blob/master/ex-basics-hello-world.v

Figure 1: Verilog Basics: Display Statement The obligatory Hello, World! program to compiling
a basic Verilog program.

ECE 4750 Computer Architecture, Fall 2015

Logical Operators
&
|
^
^~
~

bitwise AND
bitwise OR
bitwise XOR
bitwise XNOR
bitwise NOT

&&
||
!

boolean AND
boolean OR
boolean NOT

Arithmetic Operators
+
-

addition
subtraction

Tutorial 4: Verilog Hardware Description Language

Reduction Operators
&
~&
|
~|
^
^~

reduce via AND

reduce via NAND
reduce via OR
reduce via NOR
reduce via XOR
reduce via XNOR
Shift Operators

>>
<<
>>>

shift right
shift left
arithmetic shift right

Relational Operators
==
!=
>
>=
<
<=

equal
not equal
greater than
greater than or equals
less than
less than or equals
Other Operators

{}
concatenate
{N{}} replicate N times

Table 1: Table of Verilog Operators Not all Verilog operators are shown, just those operators that
are acceptable for use in the synthesizable RTL portion of students designs.
% cd ${TUTROOT}/build
% iverilog -g2012 -o hello-world hello-world.v
The -g2012 option tells iverilog to accept newer SystemVerilog syntax, and the -o specifies the
name of the simulator that iverilog will create. You can run this simulator as follows.
% cd ${TUTROOT}/build
% ./hello-world
As discussed in the Linux tutorial you can compile the Verilog and run the simulator in a single step.
% cd ${TUTROOT}/build
% iverilog -g2012 -o hello-world hello-world.v && ./hello-world
This makes it easy to edit the Verilog source file, quickly recompile, and test your changes by switching to your terminal, pressing the up-arrow key, and then pressing enter.
H

To-Do On Your Own: Edit the string that is displayed in this simple program, recompile, and rerun
the simulator.
3.2. Logic Data Types
To understand any new modeling language we usually start by exploring the primitive data types
for representing values in a model. Verilog uses a combination of the wire and reg keywords which
interact in subtle and confusing ways. SystemVerilog has simplified modeling by introducing the
logic data type. We will be exclusively using logic as the general-purpose, hardware-centric data
type for modeling a single bit or multiple bits. Each bit can take on one of four values: 0, 1, X, Z.
X is used to represent unknown values (e.g., the state of a register on reset). Z is used to represent
high-impedance values. Although we will use variables with X values in this course, we will not use
variables with Z values (although you may see Z values if you forget to connect an input port of a
module).

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

module top;

2
3

// Declare single-bit logic variables.

4
5
6
7

logic a;
logic b;
logic c;

8
9

initial begin

10
11

// Single-bit literals

12
13
14
15
16

a
a
a
a

=
=
=
=

1'b0;
1'b1;
1'bx;
1'bz;

$display(
$display(
$display(
$display(

"1'b0
"1'b1
"1'bx
"1'bz

=
=
=
=

%x
%x
%x
%x

",
",
",
",

a
a
a
a

);
);
);
);

17
18

// Bitwise logical operators for doing AND, OR, XOR, and NOT

19
20
21

a = 1'b0;
b = 1'b1;

22
23
24
25
26

c
c
c
c

=
=
=
=

a & b;
a | b;
a ^ b;
~b;

$display(
$display(
$display(
$display(

"0 & 1
"0 | 1
"0 ^ 1
"~1

=
=
=
=

%x
%x
%x
%x

",
",
",
",

c
c
c
c

);
);
);
);

27
28

// Bitwise logical operators for doing AND, OR, XOR, and NOT with X

29
30
31

a = 1'b0;
b = 1'bx;

32
33
34
35
36

c
c
c
c

=
=
=
=

a & b;
a | b;
a ^ b;
~b;

$display(
$display(
$display(
$display(

"0 & x
"0 | x
"0 ^ x
"~x

=
=
=
=

%x
%x
%x
%x

",
",
",
",

c
c
c
c

);
);
);
);

37
38

// Boolean logical operators

39
40
41

a = 1'b0;
b = 1'b1;

42
43
44
45

c = a && b;
c = a || b;
c = !b;

$display( "0 && 1 = %x ", c );

$display( "0 || 1 = %x ", c );
$display( "!1
= %x ", c );

46
47

end

48
49

endmodule

Code at https://github.com/cbatten/x/blob/master/ex-basics-logic-sbit.v

Figure 2: Verilog Basics: Single-Bit Logic and Logical Operators Experimenting with single-bit
logic variables and literals, logical bitwise operators, and logical boolean operators.
Table 1 shows the operators that we will be primarily using in this course. Note that Verilog and
SystemVerilog support additional operators including * for multiplication, / for division, % for modulus, ** for exponent, and ===/!=== for special equality checks. There may occasionally be reasons
to use one of these operators in your assertion or line tracing logic. However, these operators are
either not synthesizable or synthesize poorly, so students are not allowed to use these operators in
the synthesizable portion of their designs.

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

Figure 2 shows an example program that illustrates single-bit logic types. Create a new Verilog
source file named logic-sbit.v and copy some or all of this code. Compile this source file and run
the resulting simulator.
Lines 1316 illustrate how to write single-bit literals to express constant values. Lines 2326 illustrate
basic bitwise logical operators (&, |, , ~). Whenever we consider an expression in Verilog, we should
always ask ourselves, What will happen if one of the inputs is an X? Lines 3336 illustrate what
happens if the second operand is an X for bitwise logical operators. Recall that X means unknown.
If we OR the value 0 with an unknown value we cannot know the result. If the unknown value is
0, then the result should be 0, but if the unknown value is 1, then the result should be 1. So Verilog
specifies that in this case the value of the expression is X. Notice what happens if we AND the value
0 with an unknown value. In this case, we can guarantee that for any value for the second operand
the result will always be 0, so Verilog specifies the value of the expression is 0.
In addition to the basic bitwise logical operators, Verilog also defines three boolean logical operators
(&&, ||, !). These operators are effectively the same as the basic logical operators (&, |, ~) when
operating on single-bit logic values. The difference is really in the designers intent. Using &&, ||, !
suggests that the designer is implementing a boolean logic expression as opposed to doing low-level
bit manipulation.
H

To-Do On Your Own: Experiment with more complicated multi-stage logic expressions by writing
the boolean logic equations for a one-bit full-adder. Use the display system task to output the
result to the console. Experiment with using X input values as inputs to these logic expressions.
Multi-bit logic types are used for modeling bit vectors. Figure 3 shows an example program that
illustrates multi-bit logic types. Create a new Verilog source file named logic-mbit.v and copy
some or all of this code. Compile this source file and run the resulting simulator.
Lines 58 declares multi-bit logic variables. The square brackets contain the index of the mostsignificant and the least-significant bit. In this course, you should always use zero as the index
of the least significant bit. Note that to declare a four-bit logic value, we use [3:0] not [4:0].
Lines 1417 illustrate multi-bit literals that can be used to declare constant values. These literals have
the following general syntax: <bitwidth><base><number> where <base> can be b for binary, h for
hexadecimal, or d for decimal. It is legal to include underscores in the literal, which can be helpful
for improving the readability of long literals.
Lines 2428 illustrate multi-bit versions of the basic bitwise logic operators. As before, we should
always ask ourselves, What will happen if one of the inputs is an X? Lines 3539 illustrate what
happens if two bits in the second value are Xs. Note that some bits in the result are X and some can
be guaranteed to be either a 0 or 1.
Lines 4550 illustrate the reduction operators. These operators take a multi-bit logic value and combine all of the bits into a single-bit value. For example, the OR reduction operator (|) will OR all of
the bits together.

To-Do On Your Own: We will use relational operators (e.g., ==) to compare two multi-bit logic
values, but see if you can achieve the same effect with the bitwise XOR/XNOR operators and
OR/NOR reduction operators.

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

module top;

2
3

// Declare multi-bit logic variables

4
5
6
7
8

logic
logic
logic
logic

[ 3:0]
[ 3:0]
[ 3:0]
[11:0]

A;
B;
C;
D;

//
//
//
//

4-bit
4-bit
4-bit
12-bit

logic
logic
logic
logic

variable
variable
variable
variable

9
10

initial begin

11
12

// Multi-bit literals

13
14
15
16
17

A
D
D
D

=
=
=
=

4'b0101;
12'b1100_1010_0101;
12'hca5;
12'd1058;

$display(
$display(
$display(
$display(

"4'b0101
"12'b1100_1010_0101
"12'hca5
"12'd1058

=
=
=
=

%x",
%x",
%x",
%x",

A
D
D
D

);
);
);
);

18
19

// Bitwise logical operators for doing AND, OR, XOR, and NOT

20
21
22

A = 4'b0101;
B = 4'b0011;

23
24
25
26
27
28

C
C
C
C
C

=
=
=
=
=

A & B;
A | B;
A ^ B;
A ^~ B;
~B;

$display(
$display(
$display(
$display(
$display(

"4'b0101 &
"4'b0101 |
"4'b0101 ^
"4'b0101 ^~
"~4'b0011

4'b0011 = %b", C );
4'b0011 = %b", C );
4'b0011 = %b", C );
4'b0011 = %b", C );
= %b", C );

29
30

// Bitwise logical operators when some bits are X

31
32
33

A = 4'b0101;
B = 4'b00xx;

34
35
36
37
38
39

C
C
C
C
C

=
=
=
=
=

A & B;
A | B;
A ^ B;
A ^~ B;
~B;

$display(
$display(
$display(
$display(
$display(

"4'b0101 &
"4'b0101 |
"4'b0101 ^
"4'b0101 ^~
"~4'b00xx

4'b00xx
4'b00xx
4'b00xx
4'b00xx

" &
"~&
" |
"~|
"^
"^~

=
=
=
=
=
=

=
=
=
=
=

%b",
%b",
%b",
%b",
%b",

C
C
C
C
C

);
);
);
);
);

40
41

// Reduction operators

42
43

A = 4'b0101;

44
45
46
47
48
49
50

C
C
C
C
C
C

=
=
=
=
=
=

&A;
~&A;
|A;
~|A;
^A;
^~A;

$display(
$display(
$display(
$display(
$display(
$display(

4'b0101
4'b0101
4'b0101
4'b0101
4'b0101
4'b0101

%b",
%b",
%b",
%b",
%b",
%b",

C
C
C
C
C
C

);
);
);
);
);
);

51
52

end

53
54

endmodule

Code at https://github.com/cbatten/x/blob/master/ex-basics-logic-mbit.v

Figure 3: Verilog Basics: Multi-Bit Logic and Logical Operators Experimenting with multi-bit
logic variables and literals, bitwise logical operators, and reduction operators.

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.3. Shift Operators

Figure 4 illustrates three shift operators on multi-bit logic values. Create a new Verilog source file
named logic-shift.v and copy some or all of this code. Compile this source file and run the resulting simulator.
Notice how the logical shift operators (<<, >>) always shift in zeros, but the arithmetic right shift
operator (>>>) replicates the most-significant bit. Verilog requires that the left-hand operand to the
arithmetic shift operator be explicitly denoted as signed, which we have done using the signed
system task. We recommend this approach and avoiding the use of signed data types.
To-Do On Your Own: Experiment different multi-bit logic values and shift amounts.

module top;

2
3
4
5

logic [7:0] A;
logic [7:0] B;
logic [7:0] C;

6
7

initial begin

8
9

// Fixed shift amount for logical shifts

10
11

A = 8'b1110_0101;

12
13
14
15

C = A << 1;
C = A << 2;
C = A << 3;

$display( "8'b1110_0101 << 1 = %b", C );

$display( "8'b1110_0101 << 2 = %b", C );
$display( "8'b1110_0101 << 3 = %b", C );

C = A >> 1;
C = A >> 2;
C = A >> 3;

$display( "8'b1110_0101 >> 1 = %b", C );

$display( "8'b1110_0101 >> 2 = %b", C );
$display( "8'b1110_0101 >> 3 = %b", C );

16
17
18
19
20
21

// Fixed shift amount for arithmetic shifts

22
23

A = 8'b0110_0100;

24
25
26
27

C = $signed(A) >>> 1;
C = $signed(A) >>> 2;
C = $signed(A) >>> 3;

$display( "8'b0110_0100 >>> 1 = %b", C );

$display( "8'b0110_0100 >>> 2 = %b", C );
$display( "8'b0110_0100 >>> 3 = %b", C );

28
29

A = 8'b1110_0101;

30
31
32
33

C = $signed(A) >>> 1;
C = $signed(A) >>> 2;
C = $signed(A) >>> 3;

$display( "8'b1110_0101 >>> 1 = %b", C );

$display( "8'b1110_0101 >>> 2 = %b", C );
$display( "8'b1110_0101 >>> 3 = %b", C );

34
35

// Variable shift amount for logical shifts

36
37
38

A = 8'b1110_0101;
B = 8'd2;

39
40
41

C = A << B;
C = A >> B;

$display( "8'b1110_0101 << 2 = %b", C );

$display( "8'b1110_0101 >> 2 = %b", C );

42
43

end

44
45

endmodule

Code at https://github.com/cbatten/x/blob/master/ex-basics-logic-shift.v

Figure 4: Verilog Basics: Shift Operators Experimenting with logical and arithmetic shift operators
and fixed/variable shift amounts.
9

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.4. Arithmetic Operators

Figure 5 illustrates the addition and subtraction operators for multi-bit logic values. Create a new
Verilog source file named logic-arith.v and copy some or all of this code. Compile this source file
and run the resulting simulator.
These operators treat the multi-bit logic values as unsigned integers. Although Verilog does include
support for signed arithmetic, these constructs may not be synthesizable so you are required to use
only unsigned arithmetic. Also recall that *, /, %, ** are not allowed in the synthesizable portion of
your design.
Note that carefully considering the bitwidths of the input and output variables is important. Lines 22
23 illustrate overflow and underflow. You can see that if you overflow the bitwidth of the output
variable then the result will simply wrap around. Similarly, since we are using unsigned arithmetic
negative numbers wrap around. This is also called modular arithmetic.
To-Do On Your Own: Try writing some code which does a sequence of additions resulting in
overflow and then a sequence of subtractions that essentially undo the overflow. For example,
try 200 + 400 + 400 - 400 - 400. Does this expression produce the expected answer even
though the intermediate values overflowed?

module top;

2
3
4
5

logic [7:0] A;
logic [7:0] B;
logic [7:0] C;

6
7

initial begin

8
9

// Basic arithmetic with no overflow or underflow

10
11
12

A = 8'd28;
B = 8'd15;

13
14
15

C = A + B;
C = A - B;

$display( "8'd28 + 8'd15 = %d", C );

$display( "8'd28 - 8'd15 = %d", C );

16
17

// Basic arithmetic with overflow and underflow

18
19
20

A = 8'd250;
B = 8'd15;

21
22
23

C = A + B;
C = B - A;

$display( "8'd250 + 8'd15 = %d", C );

$display( "8'd15 - 8'd250 = %d", C );

24
25

end

26
27

endmodule

Code at https://github.com/cbatten/x/blob/master/ex-basics-logic-arith.v

Figure 5: Verilog Basics: Arithmetic Operators Experimenting with arithmetic operators for addition and subtraction.

10

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.5. Relational Operators

Figure 6 illustrates the relational operators used for comparing two multi-bit logic values. Create a
new Verilog source file named logic-relop.v and copy some or all of this code. Compile this source
file and run the resulting simulator.
Lines 2833 illustrate what happens if some of the bits are Xs for these relational operators. Notice
that we can still determine two values are not equal even if some bits are unknown. If the bits we do
know are different then the unknown bits do not matter; we can guarantee that the full bit vectors
are not equal. So in this example, since we know that the top-two bits in 4b1100 and 4b10xx then
we can guarantee that the two values are not equal even though the bottom two bits of one operand
are unknown.
The <, >, <=, >= operators behave slightly differently than the == and != operators when considering
values with Xs. In this example, we should be able to guarantee that 4b1100 is always greater than
4b10xx (assuming these are unsigned values), since no matter what the bottom two bits are in the
second operand it cannot be greater than the first operand. However, if you run this simulation, then
you will see that the result is still X. This is not a bug and is correct according to the Verilog language
specification. This is a great example of how Verilog has relatively complicated and sometimes inconsistent language semantics. Originally, there was no Verilog standard. The most common Verilog
simulator was the de-factor language standard. I imagine the reason there is this difference between
how == and < handle X values is simply because in the very first Verilog simulators it was the most
efficient solution. These kind of simulator implementation issues can be found throughout the
Verilog standard.
Lines 4043 illustrates signed comparisons using the signed system task to to interpret the unsigned
input operands as signed values. To simplify our designs, we do not allow students to use signed
types. We should explicitly use the signed system task whenever we need to ensure signed comparisons.
H

To-Do On Your Own: Try composing relational operators with the boolean logic operators we
learned about earlier in this section to create more complicated expressions.

11

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

module top;

2
3

// Declare multi-bit logic variables

4
5
6
7

logic
a; // 1-bit logic variable
logic [ 3:0] A; // 4-bit logic variable
logic [ 3:0] B; // 4-bit logic variable

8
9

initial begin

10
11

// Relational operators

12
13

A = 4'd15; B = 4'd09;

14
15
16
17
18
19
20

a
a
a
a
a
a

=
=
=
=
=
=

(
(
(
(
(
(

A
A
A
A
A
A

==
!=
>
>=
<
<=

B
B
B
B
B
B

);
);
);
);
);
);

$display(
$display(
$display(
$display(
$display(
$display(

"(15
"(15
"(15
"(15
"(15
"(15

==
!=
>
>=
<
<=

9)
9)
9)
9)
9)
9)

=
=
=
=
=
=

%x",
%x",
%x",
%x",
%x",
%x",

a
a
a
a
a
a

);
);
);
);
);
);

21
22

// Relational operators when some bits are X

23
24

A = 4'b1100; B = 4'b10xx;

25
26
27
28
29

a
a
a
a

=
=
=
=

(
(
(
(

A
A
A
A

==
!=
>
<

B
B
B
B

);
);
);
);

$display(
$display(
$display(
$display(

"(4'b1100
"(4'b1100
"(4'b1100
"(4'b1100

==
!=
>
<

4'b10xx)
4'b10xx)
4'b10xx)
4'b10xx)

=
=
=
=

%x",
%x",
%x",
%x",

a
a
a
a

);
);
);
);

%x",
%x",
%x",
%x",

a
a
a
a

30
31

// Signed relational operators

32
33
34

A = 4'b1111; // -1 in twos complement

B = 4'd0001; // 1 in twos complement

35
36
37
38
39

a
a
a
a

=
=
=
=

(
A >
B );
(
A <
B );
( $signed(A) > $signed(B) );
( $signed(A) < $signed(B) );

$display(
$display(
$display(
$display(

"(-1
"(-1
"(-1
"(-1

>
<
>
<

1)
1)
1)
1)

=
=
=
=

);
);
);
);

40
41

end

42
43

endmodule

Code at https://github.com/cbatten/x/blob/master/ex-basics-logic-relop.v

Figure 6: Verilog Basics: Relational Operators Experimenting with relational operators.

12

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.6. Concatenation Operators

Figure 7 illustrates the concatenation operators used for creating larger bit vectors from multiple
smaller bit vectors. Create a new Verilog source file named logic-concat.v and copy some or all of
this code. Compile this source file and run the resulting simulator.
Lines 1820 illustrate concatenating three four-bit logic variables and then assigning the result to a
12-bit logic variable. Lines 2526 illustrate concatenating a four-bit logic variable with an eight-bit
logic variable. The repeat operator can be used to duplicate a given logic variable multiple times
when creating larger bit vectors. On Line 33, we replicate a four-bit logic value three times to create
a 12-bit logic value.
To-Do On Your Own: Experiment with different variations of concatenation and the repeat operator to create interesting bit patterns.

module top;

2
3
4
5
6
7

logic
logic
logic
logic
logic

[ 3:0]
[ 3:0]
[ 3:0]
[ 7:0]
[11:0]

A;
B;
C;
D;
E;

//
//
//
//
//

4-bit
4-bit
4-bit
18-bit
12-bit

logic
logic
logic
logic
logic

variable
variable
variable
variable
variable

8
9

initial begin

10
11

// Basic concatenation

12
13
14
15
16

A
B
C
D

=
=
=
=

4'ha;
4'hb;
4'hc;
8'hde;

17
18
19
20

E = { A, B, C };
E = { C, A, B };
E = { B, C, A };

$display( "{ 4'ha, 4'hb, 4'hc } = %x", E );

$display( "{ 4'hc, 4'ha, 4'hb } = %x", E );
$display( "{ 4'hb, 4'hc, 4'ha } = %x", E );

E = { A, D };
E = { D, A };

$display( "{ 4'ha, 8'hde } = %x", E );

$display( "{ 8'hde, 4'ha } = %x", E );

E = { A, 8'hf0 };
E = { 8'hf0, A };

$display( "{ 4'ha, 8'hf0 } = %x", E );

$display( "{ 8'hf0, 4'ha } = %x", E );

21
22
23
24
25
26
27
28

// Repeat operator

29
30
31

A = 4'ha;
B = 4'hb;

32
33
34

E = { 3{A} };
$display( "{ 4'ha, 4'ha, 4'ha } = %x", E );
E = { A, {2{B}} }; $display( "{ 4'ha, 4'hb, 4'hb } = %x", E );

35
36

end

37
38

endmodule

Code at https://github.com/cbatten/x/blob/master/ex-basics-logic-concat.v

Figure 7: Verilog Basics: Concatenation Operators Experimenting with the basic concatenation
operator and the repeat operator.

13

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.7. Enum Data Types

The logic data type will be the most common data type we use in our synthesizable RTL since a
logic variable has a direct one-to-one correspondence with a bit vector in hardware. However, there
are certain cases where using a logic variable can be quite tedious and error prone. SystemVerilog
has introduced two new kinds of user-defined types that can greatly simplify some portions of our
designs. In this subsection, we introduce the enum type which enables declaring variables that can
only take on a predefined list of labels.
Figure 8 illustrates creating and using an enum type for holding a state variable which can take on
one of four labels. Create a new Verilog source file named enum.v and copy all of this code. Compile
this source file and run the resulting simulator.
Lines 38 declare a new enum type named state_t. Note that state_t is not a new variable but is
instead a new type. We will use the _t suffix to distinguish type names from variable names. Note
that after the enum keyword we have included a base type of logic [$clog2(4)-1:0]. This base type
specifies how we wish variables of this new type to be stored. In this case, we are specifying that
state_t variables should be encoded as a two-bit logic value. The clog2 system task calculates the
number of bits in the given argument; it is very useful when writing more parameterized code. So
in this situation we just need to pass in the number of labels in the enum to clog2. SystemVerilog
actually provides many different ways to create enum types including anonymous types, types where
we do not specify the base type, or types where we explicitly define the value for each label. In this
course, you should limit yourself to the exact syntax shown in this example.
Line 14 declares a new variable of type state_t. This is the first time we have seen a variable which
has a type other than logic. The ability to introduce new user-defined types is one of the more
powerful features of SystemVerilog. Lines 2124 sets the state variable using the labels declared as
part of the new state_t type. Lines 2840 compare the value of the state variable with these same
labels, and these comparisons can be used to take different action based on the current value.
There are several advantages to using an enum type compared to the basic logic type to represent
a variable that can hold one of several labels including: (1) more directly capturing the designers
intent to improve code quality; (2) preventing mistakes by eliminating the possibility of defining
labels with the same value or defining label values that are too large to fit in the underlying storage;
and (3) preventing mistakes when assigning variables of a different type to an enum variable.
H

To-Do On Your Own: Create your own new enum type for the state variable we will use in the GCD
example later in this tutorial. The new enum type should be called state_t and it should support
three different labels: STATE_IDLE, STATE_CALC, STATE_DONE. Write some code to set and compare
the value of a corresponding state variable.

14

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

// Declare enum type

2
3
4
5
6
7
8

typedef enum logic [$clog2(4)-1:0] {

STATE_A,
STATE_B,
STATE_C,
STATE_D
} state_t;

9
10

module top;

11
12

// Declare variables

13
14
15

state_t state;
logic
result;

16
17

initial begin

18
19

// Enum lable literals

20
21
22
23
24

state
state
state
state

=
=
=
=

STATE_A;
STATE_B;
STATE_C;
STATE_D;

$display(
$display(
$display(
$display(

"STATE_A
"STATE_B
"STATE_C
"STATE_D

=
=
=
=

%d",
%d",
%d",
%d",

state
state
state
state

);
);
);
);

25
26

// Comparisons

27
28

state = STATE_A;

29
30
31

result = ( state == STATE_A );

$display( "( STATE_A == STATE_A ) = %x", result );

32
33
34

result = ( state == STATE_B );

$display( "( STATE_A == STATE_B ) = %x", result );

35
36
37

result = ( state != STATE_A );

$display( "( STATE_A != STATE_A ) = %x", result );

38
39
40

result = ( state != STATE_B );

$display( "( STATE_A != STATE_B ) = %x", result );

41
42

end

43
44

endmodule

Code at https://github.com/cbatten/x/blob/master/ex-basics-enum.v

Figure 8: Verilog Basics: Enum Data Types Experimenting with enum data types including setting
the value of an enum using a label and using the equality operator.

15

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.8. Struct Data Types

User-defined structures are now supported in SystemVerilog. Figure 9 illustrates creating and using a
struct type for holding a variable with predefined named bit fields. Create a new Verilog source file
named struct.v and copy all of this code. Compile this source file and run the resulting simulator.
Lines 37 declare a new struct type named point_t. Again note that point_t is not a new variable
but is instead a new type. As before we use the _t suffix to distinguish type names from variable
names. Note that after the struct keyword we have included the packed keyword which specifies
that variables of this type should have an equivalent underlying logic storage. SystemVerilog also
includes support for unpacked structs, but in this course, you should limit yourself to the exact
syntax shown in this example. In addition to declaring the name of the new struct type, we also
declare the named bit fields within the new struct type. The order of these bit fields is important;
the first field will go in the most significant position of the underlying logic storage, the second field
will go in the next position, and so on.
Lines 1314 declare two new variables of type point_t. Line 18 declares a new logic variable with a
bitwidth large enough to hold a variable of type point_t. We can use the bits system task to easily
determine the total number of bits required to store a struct type. Lines 2426 set the three fields of
the point variable and Lines 2830 read these three fields in order to display them. Line 34 copies
one point variable into another point variable. Line 42 and 49 illustrate how to convert a point
variable to/from a basic logic variable.
There are several advantages to using a struct type compared to the basic logic type to represent
a variable with a predefined set of named bit fields including: (1) more directly capturing the designers intent to improve code quality; (2) reducing the syntactic overhead of managing bit fields;
and (3) preventing mistakes in modifying bit fields and in accessing bit fields.
H

To-Do On Your Own: Create a new struct type for holding the an RGB color pixel. The struct
should include three fields named red, green, and blue. Each field should be eight bits. Experiment with reading and writing these named fields.

16

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

// Declare struct type

2
3
4
5
6
7

typedef struct packed {

logic [3:0] x;
logic [3:0] y;
logic [3:0] z;
} point_t;

// Packed format:
//
11 8 7 4 3 0
// +----+----+----+
// | x | y | z |
// +----+----+----+

8
9

module top;

10
11

// Declare variables

12
13
14

point_t point_a;
point_t point_b;

15
16

// Declare other variables using $bits()

17
18

logic [$bits(point_t)-1:0] point_bits;

19
20

initial begin

21
22

// Reading and writing fields

23
24
25
26

point_a.x = 4'h3;
point_a.y = 4'h4;
point_a.z = 4'h5;

27
28
29
30

$display( "point_a.x = %x", point_a.x );

$display( "point_a.y = %x", point_a.y );
$display( "point_a.z = %x", point_a.z );

31
32

// Assign structs

33
34

point_b = point_a;

35
36
37
38

$display( "point_b.x = %x", point_b.x );

$display( "point_b.y = %x", point_b.y );
$display( "point_b.z = %x", point_b.z );

39
40

// Assign structs to bit vector

41
42

point_bits = point_a;

43
44

$display( "point_bits = %x", point_bits );

45
46

// Assign bit vector to struct

47
48
49

point_bits = { 4'd13, 4'd9, 4'd3 };

point_a = point_bits;

50
51
52
53

$display( "point_a.x = %x", point_a.x );

$display( "point_a.y = %x", point_a.y );
$display( "point_a.z = %x", point_a.z );

54
55

end

56
57

endmodule

Code at https://github.com/cbatten/x/blob/master/ex-basics-struct.v

Figure 9: Verilog Basics: Struct Data Types Experimenting with struct data types including
read/writing fields and converting to/from logic bit vectors.

17

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.9. Ternary Operator

Figure 10 illustrates using the ternary operator for
conditional execution. Create a new Verilog source
file named ternary.v and copy some or all of this
code. Compile this source file and run the resulting
simulator.
Lines 1219 illustrate using the ternary operator to
choose what value to assign to the logic variable
c. We can nest multiple ternary operators to compactly create expressions with multiple conditions.
Lines 2331 illustrate using four levels of nesting to
choose among four different values for assigning c.
Lines 3553 illustrate how the ternary operator
functions if the conditional is unknown.
In
lines 3543, all bits of the conditional are unknown,
while in lines 4553 only one bit of the conditional
is unknown. If you examine the output from this
simulator, you will see that Verilog semantics require any bits which can be guaranteed to be either
0 or 1 to be set as such, while the remaining bits are
set to X. Regardless of the condition, the upper five
bits of c are guaranteed to be 00001.
Note that the four ternary operators cover all possible combinations of the two-bit input, so the final value (i.e., 8h0e) will never be used. In other
words, if the conditionals contain unknowns this
does not mean the condition evaluates to false. This
is very different from the if statements described
in the next subsection.

3
4
5
6

10

a;
b;
c;
sel;

initial begin
// ternary statement

11
12
13

a = 8'd30;
b = 8'd16;

14
15
16

c = ( a < b ) ? 15 : 14;
$display( "c = %d", c );

17
18
19

c = ( b < a ) ? 15 : 14;
$display( "c = %d", c );

20
21

// nested ternary statement

22
23

sel = 2'b01;

24
25
26
27
28
29

c =
:
:
:
:

(
(
(
(

sel
sel
sel
sel

==
==
==
==

2'b00
2'b01
2'b10
2'b11

)
)
)
)

?
?
?
?

8'h0a
8'h0b
8'h0c
8'h0d
8'h0e;

30
31

$display( "sel = 1, c = %b", c );

32
33

// nested ternary statement w/ X

34
35

sel = 2'bxx;

36
37
38
39
40

c =
:
:
:
:

(
(
(
(

sel
sel
sel
sel

==
==
==
==

2'b00
2'b01
2'b10
2'b11

)
)
)
)

?
?
?
?

8'h0a
8'h0b
8'h0c
8'h0d
8'h0e;

42
43

$display( "sel = x, c = %b", c );

44

sel = 2'bx0;

46
47
48

This obfuscates the code and is not necessary. We

are using a ternary operator to simply choose between 0 or 1. You should just assign the result of
the conditional expression to a and b like this:

49
50
51

c =
:
:
:
:

(
(
(
(

sel
sel
sel
sel

==
==
==
==

2'b00
2'b01
2'b10
2'b11

)
)
)
)

?
?
?
?

8'h0a
8'h0b
8'h0c
8'h0d
8'h0e;

52
53

$display( "sel = x, c = %b", c );

54
55

a = ( cond_a );
b = !( cond_b );
H

[7:0]
[7:0]
[7:0]
[1:0]

45

a = ( cond_a ) ? 1b1 : 1b0;

b = ( cond_b ) ? 1b0 : 1b1;

logic
logic
logic
logic

41

Aside: For some reason, many students insist on

writing code like this:

module top;

end

56
57

To-Do On Your Own: Experiment with different

uses of the ternary operator.

18

endmodule

Code at https://github.com/cbatten/x/blob/
master/ex-basics-ternary.v

Figure 10: Verilog Basics: Ternary Operator

Experimenting with the ternary operator including nested statements and what happens
if the conditional includes an unknown.

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.10. If Statements
Figure 11 illustrates using if statements. Create a
new Verilog source file named if.v and copy some
or all of this code. Compile this source file and run
the resulting simulator.

The if statement resembles similar constructs in

many other programming languages. Lines 1120
illustrate basic if statements and lines 2433 illustrate if/else statements.

There are some subtle issues involved in how an

if statement handles X values in the conditional.
Lines 3746 illustrate this issue. The sel value in
this example is a single-bit X. What would we expect the value of a to be after this if statement?
Since the conditional is unknown, we might expect
any variables that are written from within the if
statement to also be unknown. In other words, we
might expect the value of a to be X after this if
statement. If you run this example code, you will
see that the value of a is actually 8h0b. This means
that an X value in the conditional for an if statement is not treated as unknown but is instead simply treated as if the conditional evaluated to false!
This issue is called X optimism since unknowns are
essentially optimistically turned into known values. X optimism can cause subtle simulation/synthesis mismatch issues. If you are interested, there
are several resources on the public course webpage
that go into much more detail. For this course, we
dont need to worry too much about X optimism
since we are not actually synthesizing our designs.

3
4
5

To-Do On Your Own: Experiment with different

if statements including nested if statements.
Experiment with what happens when the conditional is unknown.

logic [7:0] a;
logic [7:0] b;
logic
sel;

initial begin

8
9

// if statement

10
11
12

a = 8'd30;
b = 8'd16;

13
14
15
16

if ( a == b ) begin
$display( "30 == 16" );
end

17
18
19
20

if ( a != b ) begin
$display( "30 != 16" );
end

21
22

// if else statement

23
24

sel = 1'b1;

25
26
27
28
29
30
31

if ( sel == 1'b0 ) begin

a = 8'h0a;
end
else begin
a = 8'h0b;
end

32
33

$display( "sel = 1, a = %x ", a );

34
35

// if else statement w/ X

36
37

sel = 1'bx;

38
39
40
41
42
43
44

module top;

if ( sel == 1'b0 ) begin

a = 8'h0a;
end
else begin
a = 8'h0b;
end

45
46

$display( "sel = x, a = %x ", a );

47
48

end

49
50

endmodule

Code at https://github.com/cbatten/x/blob/
master/ex-basics-if.v

Figure 11: Verilog Basics: If Statements

Experimenting with if statements including
what happens if the conditional includes an
unknown.

19

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.11. Case Statements

Figure 12 illustrates using case statements. Create
a new Verilog source file named case.v and copy
some or all of this code. Compile this source file
and run the resulting simulator.

The case statement resembles similar constructs in

many other programming languages. Lines 12
22 illustrate a basic case statement where a twobit sel variable is used to choose one of four case
items.

There are similar issues as with the if statement in

terms of how case statements handle X values in
the conditional. In lines 2636, the sel variable is
set to all Xs. We might expect since the input to
the case statement is unknown the output should
also be unknown. However, if we look at the value
of a after executing this case statement it will be
8h0e. In other words, if there is an X in the input to the case statement, then the case statement
will fall through to the default case. In order to
avoid X optimism, we recommend students always
include a default case that sets all of the output
variables to Xs.
Notice that it is valid syntax to use X values in the
case items, as shown on lines 4849. These will actually match Xs in the input condition, which is almost certainly not what you want. This does not
model any kind of real hardware; we cannot check
for Xs in hardware since in real hardware an unknown must be known (i.e., all Xs will either be a
0 or a 1 in real hardware). Given this, you should
never use Xs in the case items for a case statement.

5
6

logic [1:0] sel;

logic [7:0] a;
initial begin

9
10

// case statement

11
12

sel = 2'b01;

13
14
15
16
17
18
19
20

case ( sel )
2'b00
: a
2'b01
: a
2'b10
: a
2'b11
: a
default : a
endcase

=
=
=
=
=

8'h0a;
8'h0b;
8'h0c;
8'h0d;
8'h0e;

21
22

$display( "sel = 01, a = %x", a );

23
24

// case statement w/ X

25
26

sel = 2'bxx;

27
28
29
30
31
32
33
34

case ( sel )
2'b00
: a
2'b01
: a
2'b10
: a
2'b11
: a
default : a
endcase

=
=
=
=
=

8'h0a;
8'h0b;
8'h0c;
8'h0d;
8'h0e;

35
36

$display( "sel = xx, a = %x", a );

37
38
39

// Do not use x's in the case

// selection items

40
41

sel = 2'bx0;

42
43
44

46

To-Do On Your Own: Experiment with a larger

case statement for a sel variable with three instead of two bits.

// Declaring Variables

45

module top;

47
48
49
50
51

case ( sel )
2'b00
: a
2'b01
: a
2'b10
: a
2'b11
: a
2'bx0
: a
2'bxx
: a
default : a
endcase

=
=
=
=
=
=
=

8'h0a;
8'h0b;
8'h0c;
8'h0d;
8'h0e;
8'h0f;
8'h00;

52
53

$display( "sel = x0, a = %x", a );

54
55

end

56
57

endmodule

Code at https://github.com/cbatten/x/blob/
master/ex-basics-case.v

Figure 12: Verilog Basics: Case Statements

Experimenting with case statements including what happens if the selection expression
and/or the case expressions includes an unknown.
20

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

3.12. Casez Statements

Figure 13 illustrates using casez statements. Create
a new Verilog source file named casez.v and copy
some or all of this code. Compile this source file
and run the resulting simulator.

The casez statement is very different from what

you might find in other programming languages.
The casez statement is a powerful construct that
can enable very concise hardware models, but must
be used carefully. A casez statement enables a designer to do wildcard matching on the input variable. Lines 1023 illustrate using a casez statement to implement a leading-one detector. This
kind of logic outputs the bit position of the leastsignificant one in the input variable. We can use
? characters in the case items as wildcards that
will match either a 0 or 1 in the input variable. So
both 4b0100 and 4b1100 will match the fourth
case item. Implementing similar functionality using a case statement would require 16 items. Besides being more verbose, using a case statement
also opens up additional opportunities for errors.

A casez statement behaves similarly to a case

statement when there are Xs in the input. Lines 27
40 illustrate a situation where two of the bits in the
input variable are unknown. This will match the
default case and the output will be Xs.
Aside: Verilog includes a casex statement which
you should never use. The reasoning is rather subtle, but to be safe stick to using casez statement if
you need wildcard matching (and only if you need
wildcard matching).
H

To-Do On Your Own: Experiment with a larger

casez statement to implement a leading-one
detector for an input variable with eight instead
of four bits. How many case items would we
need if we used a case statement to implement
the same functionality?

module top;

2
3
4

logic [3:0] a;
logic [7:0] b;

5
6

initial begin
// casez statement

9
10

a = 4'b0100;

11
12

casez ( a )

13

4'b0000
4'b???1
4'b??10
4'b?100
4'b1000

14
15
16
17
18

:
:
:
:
:

b
b
b
b
b

=
=
=
=
=

8'd0;
8'd1;
8'd2;
8'd3;
8'd4;

19
20
21

default : b = 8'hxx;
endcase

22
23

$display( "a = 4'b0100, b = %x", b );

24
25

// casez statement w/ Xs

26
27

a = 4'b01xx;

28
29

casez ( a )

30

4'b0000
4'b???1
4'b??10
4'b?100
4'b1000

31
32
33
34
35

:
:
:
:
:

b
b
b
b
b

=
=
=
=
=

8'd0;
8'd1;
8'd2;
8'd3;
8'd4;

36
37
38

default : b = 8'hxx;
endcase

39
40

$display( "a = 4'b01xx, b = %x", b );

41
42

end

43
44

endmodule

Code at https://github.com/cbatten/x/blob/
master/ex-basics-casez.v

Figure 13: Verilog Basics: Casez Statements

Experimenting with casez statements to illustrate their use as priority selectors with wildcards.

4. Registered Incrementer
In this section, we will create our very first Verilog hardware model and learn how to test this module
using waveforms, ad-hoc testing, and a simple unit testing framework. As with PyMTL RTL design,
it is good design practice to usually draw some kind of diagram of the hardware we wish to model
before starting to develop the corresponding Verilog model. This diagram might be a block-level
21

ECE 4750 Computer Architecture, Fall 2015

in

8b

+1

8b

out

Tutorial 4: Verilog Hardware Description Language

Figure 14: Block Diagram for Registered Incrementer An eight-bit registered incrementer with an
eight-bit input port, an eight-bit output port, and an
(implicit) clock input.

diagram, a datapath diagram, a finite-state-machine diagram, or even a control signal table; the
more we can structure our Verilog code to match this diagram the more confident we can be that
our model actually models what we think it does. In this section, we wish to model the eight-bit
registered incrementer shown in Figure 14.
4.1. RTL Model of Registered Incrementer
Figure 16 shows the Verilog code which corresponds to the diagram in Figure 14. Every Verilog
file should begin with a header comment as shown on lines 19 in Figure 16. The header comment
should identify the primary module in the file, and include a brief description of what the module
does. Reserve discussion of the actual implementation for later in the file. In general, you should
attempt to keep lines in your Verilog source code to less than 74 characters. This will make your code
easier to read, enable printing on standard sized paper, and facilitate viewing two source files sideby-side on a single monitor. Note that the code in Figure 16 is artificially narrow so we can display
two code listings side-by-side. Lines 1112 create an include guard using the Verilog pre-processor.
An include guard ensures that even if we include this Verilog file multiple times the modules within
the file will only be declared once. Without include guards, the Verilog compiler will likely complain
that the same module has been declared multiple times. Make sure that you have the corresponding
end of the include guard at the bottom of your Verilog source file as shown on line 43.
Unlike Python, Verilog does not have a clean way to manage namespaces for macros and module
names. This means that if you use the same macro or module name in two different files it will create
a namespace collision which can potentially be very difficult to debug. We will follow very specific
naming conventions to eliminate any possibility of a namespace collision. Our convention will to
use the subdirectory path as a prefix for all Verilog macro and module names. Since the registered
incrementer is in the directory tut4_verilog/regincr, we will use TUT4_VERILOG_REGINCR_ as a
prefix for all macro names and tut4_verilog_regincr_ as a prefix for all module names. You can
see this prefix being used for the macros on lines 1112 and for the module name on line 14. To
reiterate, Verilog macro and module name must use the subdirectory path as a prefix. While a bit tedious,
this is essential to avoiding namespace collisions.
As always, we begin by identifying the modules interface which in this case will include an eightbit input port, eight-bit output port, and a clock input. Lines 1520 in Figure 16 illustrate how we
represent this interface using Verilog. A common mistake is to forget the semicolon (;) on line 20. A
couple of comments about the coding conventions that we will be using in this course. All module
names should always include the subproject name as a prefix (e.g., ex_regincr_). The portion of
the name after this prefix should usually use CamelCaseNaming; each word begins with a capital
letter without any underscores (e.g., RegIncr). Port names (as well as variable and module instance
names) should use underscore_naming; all lowercase with underscores to separate words. As shown
on lines 1619, ports should be listed one per line with a two space initial indentation. The bitwidth
specifiers and port names should all line up vertically. As shown on lines 15 and 20, the opening and
closing parenthesis should be on their own separate lines. Carefully group ports to help the reader
understand how these ports are related. Use port names (as well as variable and module instance

22

ECE 4750 Computer Architecture, Fall 2015

1
2
3
4
5
6
7
8
9

Tutorial 4: Verilog Hardware Description Language

//======================================
// Registered Incrementer
//======================================
// This is a simple example of a module
// for a registered incrementer which
// combines a positive edge triggered
// register with a combinational +1
// incrementer. We use flat register// transfer-level modeling.

1
2
3
4
5
6
7
8
9

10
11
12

10

`ifndef TUT4_VERILOG_REGINCR_REG_INCR_V
`define TUT4_VERILOG_REGINCR_REG_INCR_V

11
12

13
14
15
16
17
18
19
20

14
15
16
17
18
19
20

25
26
27
28
29
30

22

logic [7:0] reg_out;

always_ff @( posedge clk ) begin
if ( reset )
reg_out <= '0;
else
reg_out <= in;
end

24
25
26
27
28
29
30

35
36
37

32

logic [7:0] temp_wire;

always_comb begin
temp_wire = reg_out + 1;
end

34
35
36
37

39

assign out = temp_wire;

40

endmodule

41

42
43

reg [7:0] temp_wire;

always @(*) begin
temp_wire = reg_out + 1;
end

38

assign out = temp_wire;

40
41

// Combinational logic

33

38
39

reg [7:0] reg_out;

always @( posedge clk ) begin
if ( reset )
reg_out <= 0;
else
reg_out <= in;
end

31

// Combinational logic

33
34

// Sequential logic

23

31
32

module tut4_verilog_regincr_RegIncr
(
input
clk,
input
reset,
input [7:0] in,
output [7:0] out
);

21

// Sequential logic

23
24

`ifndef TUT4_VERILOG_REGINCR_REG_INCR_V
`define TUT4_VERILOG_REGINCR_REG_INCR_V

13

module tut4_verilog_regincr_RegIncr
(
input logic
clk,
input logic
reset,
input logic [7:0] in,
output logic [7:0] out
);

21
22

//======================================
// Registered Incrementer
//======================================
// This is a simple example of a module
// for a registered incrementer which
// combines a positive edge triggered
// register with a combinational +1
// incrementer. We use flat register// transfer-level modeling.

endmodule

42

`endif /* TUT4_VERILOG_REGINCR_REG_INCR_V */

43

Figure 15: Registered Incrementer An eightbit registered +1 incrementer corresponding to

the diagram in Figure 14.

`endif /* TUT4_VERILOG_REGINCR_REG_INCR_V */

Figure 16: Registered Incrementer An eightbit registered +1 incrementer using Verilog-2001

constructs.

names) that are descriptive; prefer longer descriptive names (e.g., write_en) over shorter confusing
names (e.g., wen).
Lines 2239 model the internal behavior of the module. We usually prefer using two spaces for each
level of indentation; larger indentation can quickly result in significantly wasted horizontal space.
You should always use spaces and never insert any real tab characters into your source code. You must limit
yourself to synthesizable RTL for modeling your design. We will exclusively use two kinds of always
blocks: always_ff concurrent blocks to model sequential logic (analogous to PyMTL @s.tick_rtl
concurrent blocks) and always_comb concurrent blocks to model combinational logic (analogous to
PyMTL @s.combinational concurrent blocks). We require students to clearly distinguishing between the portions of your code that are meant to model sequential logic from those portions meant
to model combinational logic. This simple guideline can save significant frustration by making it

23

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

easy to see subtle errors. For example, by convention we should only use non-blocking assignments
in sequential logic (e.g., the <= operator on line 27) and we should only use blocking assignments
in combinational logic (e.g., the = operator on line 36). We use the variable reg_out to hold the intermediate value between the register and the incrementer, and we use the variable temp_wire to
hold the intermediate value between the incrementer and the output port. reg_out is modeling a
register while temp_wire is modeling a wire. Notice that both of these variables use the logic data
type; what makes one model a register while the other models a wire is how these variables are used.
The sequential concurrent block update to reg_out means it models a register. The combinational
concurrent block update to temp_wire means it models a wire.
The register incrementer illustrates the two fundamental ways to model combinational logic. We
have used an always_comb concurrent block to model the actual incrementer logic and a continuous
assignment statement (i.e., assign) to model connecting the temporary wire to the output port. We
could just have easily written logic as part of the assign statement. For example, we could have used
assign out = reg_out + 1 and skipped the always_comb concurrent block. In general, we prefer continuous assignment statements over always @(*) concurrent blocks to model combinational
logic, since it is easier to model less-realistic hardware using always_comb concurrent blocks. There is
usually a more direct one-to-one mapping from continuous assignment statements to real hardware.
However, there are many cases where it is significantly more convenient to use always_comb concurrent blocks or just not possible to use continuous assignment statements. Students will need to
use their judgment to determine the most elegant way to represent the hardware they are modeling
while still ensuring there is a clear mapping from the model to the target hardware.
Figure 15 illustrates a few new SystemVerilog constructs. Figure 16 illustrates the exact same registered incrementer implemented using the older Verilog-2001 hardware description language. Verilog2001 uses reg and wire to specify variables instead of logic. All ports are of type wire by default.
Determining when to use reg and wire is subtle and error prone. Note that reg is a misnomer; it
does not model a register! On line 34, we must declare temp_wire to be of type reg even though it is
modeling a wire. Verilog-2001 requires using reg for any variable written by an always concurrent
block. Verilog-2001 uses a generic always block for both sequential and combinational concurrent
blocks. While the always @(*) syntax is an improvement over the need in Verilog-1995 to explicitly
define sensitivity lists, always_ff and always_comb more directly capture designer intent and allow
automated tools to catch common errors. For example, a Verilog simulator can catch errors where a
designer accidentally uses a non-blocking assignment in an always_comb concurrent block, or where
a designer accidentally writes the same logic variable from two different always_comb concurrent
blocks. SystemVerilog is growing in popularity and increasingly becoming the de facto replacement
for Verilog-2001, so it is worthwhile to carefully adopt new SystemVerilog features that can improve
designer productivity.
Edit the Verilog source file named RegIncr.v in the tut4_verilog/regincr subdirectory using your
favorite text editor. Add the combinational logic shown on lines 3439 in Figure 16 which models the
incrementer logic. We will be using iverilog to simulate this registered incrementer module, and
iverilog does not currently support always_ff and always_comb, which is why we are using the
Verilog-2001 construct for now.
4.2. Simulating a Model using iverilog
Now that we have developed a new hardware module, we can test its functionality using a simulation harness. Figure 17 shows an ad-hoc test using non-synthesizable Verilog. Note that we must
explicitly include any Verilog files which contain modules that we want to use; Line 5 includes the
Verilog source file that contains the registered incrementer. Lines 1112 setup a clock with a period

24

ECE 4750 Computer Architecture, Fall 2015

1
2
3

Tutorial 4: Verilog Hardware Description Language

//========================================================================
// RegIncr Ad-Hoc Testing
//========================================================================

4
5

`include "../tut4_verilog/regincr/RegIncr.v"

6
7

module top;

8
9

// Clocking

10
11
12

logic clk = 1;
always #5 clk = ~clk;

13
14

// Instaniate the design under test

15
16
17
18

logic
reset = 1;
logic [7:0] in;
logic [7:0] out;

19
20
21
22
23
24
25
26

tut4_verilog_regincr_RegIncr reg_incr
(
.clk
(clk),
.reset (reset),
.in
(in),
.out
(out)
);

27
28

// Verify functionality

29
30

initial begin

31
32

// Dump waveforms

33
34
35

$dumpfile("regincr-iverilog-sim.vcd");
$dumpvars;

36
37

// Reset

38
39
40

#11;
reset = 1'b0;

41
42

// Test cases

43
44
45
46
47
48
49

in = 8'h00;
#10;
if ( out != 8'h01 ) begin
$display( "ERROR: out, expected = %x, actual = %x", 8'h01, out );
$finish;
end

50
51
52
53
54
55
56

in = 8'h13;
#10;
if ( out != 8'h14 ) begin
$display( "ERROR: out, expected = %x, actual = %x", 8'h14, out );
$finish;
end

57
58
59
60
61
62
63

in = 8'h27;
#10;
if ( out != 8'h28 ) begin
$display( "ERROR: out, expected = %x, actual = %x", 8'h28, out );
$finish;
end

64
65
66
67

$display( "*** PASSED ***" );

$finish;
end

68
69

endmodule

25
Figure 17: Simulator for Registered Incrementer A Verilog simulator for the eight-bit registered
incrementer in Figure 16.

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

of 10 time steps. Notice that we are assigning an initial value to the clk net on line 11 and then
modifying this net every five timesteps; setting initial values such as this is not synthesizable and
should only be used in test harnesses. If you need to set an initial value in your design, you should
use properly constructed reset logic.
Lines 1924 instantiate the device under test. Notice that we use underscore_naming for the module
instance name (e.g., reg_incr). You should almost always use named port binding (as opposed to
positional port binding) to connect nets to the ports in a module instance. Lines 2123 illustrate the
correct coding convention with one port binding per line and the ports/nets vertically aligned. As
shown on lines 20 and 24 the opening and closing parenthesis should be on their own separate lines.
Although this may seem verbose, this coding style can significantly reduce errors by making it much
easier to quickly visualize how ports are connected.
Lines 2862 illustrate an initial block which executes at the very beginning of the simulation.
initial blocks are not synthesizable and should only be used in test harnesses. Lines 3233 instruct the simulator to dump waveforms for all nets. Line 35 is a delay statement that essentially
waits for one timestep. Delay statements are not synthesizable and should only be used in test harnesses. Lines 3944 implement a test by setting the inputs of the device under test, waiting for 10
time steps, and then checking that the output is as expected. If there is an error, we display an error
message and stop the simulation. We include two more tests, and if we make it to the bottom of the
initial block then the test has passed.
Edit the Verilog simulation harness named regincr-iverilog-sim.v in the tut4_verilog/regincr
subdirectory using your favorite text editor. Add the code on lines 2026 in Figure 17 to instantiate the registered incrementer model. Then use iverilog to compile this simulator and run the
simulation as follows:
% cd ${TUTROOT}/build
% iverilog -g2012 -o regincr-iverilog-sim ../tut4_verilog/regincr/regincr-iverilog-sim.v
% ./regincr-iverilog-sim
If everything goes as expected, then the ad-hoc test should display *** PASSED ***.
H

To-Do On Your Own: Edit the register incrementer so that it now increments by +2 instead of +1.
Use an assign statement instead of the always @(*) concurrent block to do the incrementer logic.
Recompile, rerun the ad-hoc test, and verify that the tests no longer pass. Modify the ad-hoc test
so that it includes the correct reference outputs for a +2 incrementer, recompile, rerun the ad-hoc
test, and verify that the test now pass. When you are finished, edit the register incrementer so
that it again increments by +1.
4.3. Verifying a Model with Unit Testing
Writing test and simulation harnesses in Verilog is very tedious. There are some industry standard
verification frameworks based on SystemVerilog, such as the Open Verification Methodology (OVM)
and the Universal Verification Methodology (UVM), but these frameworks are very heavyweight and
are not supported by open-source tools. In this course, we will be using PyMTL for FL modeling, but
also to productively write test and simulation harnesses for our Verilog RTL models. PyMTL includes
support for Verilog import by writing a special PyMTL wrapper model. Once we have created this
wrapper model, we can use all of the sophisticated techniques we learned in the PyMTL tutorial for
writing test and simulation harnesses.

26

ECE 4750 Computer Architecture, Fall 2015

1
2
3

Tutorial 4: Verilog Hardware Description Language

#=========================================================================
# RegIncr
#=========================================================================

4
5

from pymtl import *

6
7

class RegIncr( VerilogModel ):

8
9

# Verilog module setup

10
11

vprefix = "tut4_verilog_regincr"

12
13

# Constructor

14
15

def __init__( s ):

16
17

# Port-based interface

18
19
20

s.in_ = InPort ( Bits(8) )

s.out = OutPort ( Bits(8) )

21
22

# Verilog ports

23
24
25
26
27
28
29

s.set_ports({
'clk'
: s.clk,
'reset' : s.reset,
'in'
: s.in_,
'out'
: s.out,
})

30
31

# Line Tracing

32
33
34

def line_trace( s ):
return "{} () {}".format( s.in_, s.out )

Figure 18: Registered Incrementer Wrapper PyMTL wrapper for the Verilog module shown in
Figure 14.
Figure 18 illustrates such a PyMTL wrapper. Note that on line 7, we inherit from VerilogModel not
Model. This is how we tell PyMTL that this is a special wrapper model. One line 11, we specify
the vprefix which is the prefixed discussed above for avoiding namespace collisions. The vprefix
must match the prefix used for the name of the Verilog model we are wrapping. The interface on
lines 1920 should be identical to what we would use with a standard PyMTL implementation. On
lines 2429, we must specify how the Verilog ports correspond to the PyMTL ports. The Verilog ports
are listed on the left, and the PyMTL ports are listed on the right. Notice that you must explicitly
include the clk and reset signals. Finally, you can include a line tracing function in the wrapper,
but keep in mind that you can only access the ports of the wrapper for line tracing. We will see later
in this tutorial how we can use line tracing within our Verilog modules.
Once we have a PyMTL wrapper, we can use the exact same test harness we used in the PyMTL
tutorial. The following commands will run the RegIncr_test.py test script.
% cd ${TUTROOT}/build
% py.test ../tut4_verilog/regincr/RegIncr_test.py -sv
% py.test ../tut4_verilog/regincr/RegIncr_extra_test.py -sv
Keep in mind that PyMTL uses the open-source verilator simulator instead of iverilog for simulating Verilog modules. The framework will output the specific verilog command being used to

27

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

create the simulator. You can us all of the py.test features we learned in the PyMTL tutorial including using the -s option to output line tracing and the --dump-vcd option to output VCD.
% cd ${TUTROOT}/build
% py.test ../tut4_verilog/regincr/RegIncr_test.py -s --dump-vcd
% gtkwave tut4_verilog.regincr.RegIncr_test.test_basic.verilator1.vcd
Note that the actual name of the VCD file will be a little different compared to when we use PyMTL
RTL models. In the lab assignments, we will mostly provide you with the appropriate top-level
PyMTL wrappers, but you may need to write your own wrappers for new child models you want to
test.
As we learned in the PyMTL tutorial, when testing an entire directory, we can use an iterative process
to zoom in on a failing test case. We will start by running all tests in the directory to see an overview
of which tests are passing and which tests are failing. We then explicitly run a single test script with
the -v command line option to see which specific test cases are failing. Finally, we will use the -k or
-x command line options with --tb, -s, and/or --dump-vcd command line option to generate error
output, line traces, and/or waveforms for the failing test case. Here is an example of this three-step
process to zoom in on a failing test case:
%
%
%
%

cd ${TUTROOT}/build
py.test ../tut4_verilog/regincr
py.test ../tut4_verilog/regincr/RegIncr2stage_test.py -v
py.test ../tut4_verilog/regincr/RegIncr2stage_test.py -v -x --tb=short

To-Do On Your Own: Edit the register incrementer so that it now increments by +2 instead of +1.
Recompile, rerun the unit test, and verify that the tests no longer pass. Modify the unit test so
that it includes the correct reference outputs for a +2 incrementer, recompile, rerun the unit test,
and verify that the test now pass. When you are finished, edit the register incrementer so that it
again increments by +1.
4.4. Reusing a Model with Structural Composition
As in PyMTL, we can use modularity and hierarchy to structurally compose small, simple models
into large, complex models. Figure 19 shows a two-stage registered incrementer that uses structural
composition to instantiate and connect two instances of a single-stage registered incrementer. Figure 20 shows the corresponding Verilog module. Line 11 uses a include to include the child model
that we will be reusing. Notice how we must use the full path (from the root of the project) to the
Verilog file we want to include.
Lines 2531 instantiate the first registered incrementer and lines 3541 instantiate the second registered incrementer. As mentioned above, we should almost always used named port binding to

in

8b

8b
RegIncr

RegIncr

out

Figure 19: Block Diagram for Two-Stage Registered Incrementer An eight-bit two-stage
registered incrementer that reuses the registered incrementer in Figure 14 through structural composition.

28

ECE 4750 Computer Architecture, Fall 2015

1
2
3
4
5
6

Tutorial 4: Verilog Hardware Description Language

//========================================================================
// RegIncr2stage
//========================================================================
// Two-stage registered incrementer that uses structural composition to
// instantitate and connect two instances of the single-stage registered
// incrementer.

7
8
9

`ifndef TUT4_VERILOG_REGINCR_REG_INCR_2STAGE_V
`define TUT4_VERILOG_REGINCR_REG_INCR_2STAGE_V

10
11

`include "tut4_verilog/regincr/RegIncr.v"

12
13
14
15
16
17
18
19

module tut4_verilog_regincr_RegIncr2stage
(
input logic
clk,
input logic
reset,
input logic [7:0] in,
output logic [7:0] out
);

20
21

// First stage

22
23

logic [7:0] reg_incr_0_out;

24
25
26
27
28
29
30
31

tut4_verilog_regincr_RegIncr reg_incr_0
(
.clk
(clk),
.reset (reset),
.in
(in),
.out
(reg_incr_0_out)
);

32
33

// Second stage

34
35
36
37
38
39
40
41

tut4_verilog_regincr_RegIncr reg_incr_1
(
.clk
(clk),
.reset (reset),
.in
(reg_incr_0_out),
.out
(out)
);

42
43

endmodule

44
45

`endif /* TUT4_VERILOG_REGINCR_REG_INCR_2STAGE_V */

Figure 20: Two-Stage Registered Incrementer An eight-bit two-stage registered incrementer corresponding to Figure 19. This model is implemented using structural composition to instantiate and
connect two instances of the single-stage register incrementer.
connect nets to the ports in a module instance. Lines 2730 illustrate the correct coding convention
with one port binding per line and the ports/nets vertically aligned. As shown on lines 26 and 31 the
opening and closing parenthesis should be on their own separate lines. We usually declare signals
that will be connected to output ports immediately before instantiating the module.
We need to write a new PyMTL wrapper for our two-stage registered incrementer, although it will
be essentially the same as the wrapper shown in Figure 18 except with a different class name. This
illustrates a key point: the PyMTL wrapper simply captures the Verilog interface and is largely unconcerned with the implementation.

29

ECE 4750 Computer Architecture, Fall 2015

1
2
3
4
5
6
7
8

module vc_Incrementer
#(
parameter p_nbits
= 1,
parameter p_inc_value = 1
)(
input logic [p_nbits-1:0] in,
output logic [p_nbits-1:0] out
);

Tutorial 4: Verilog Hardware Description Language

Figure 21: Parameterized Incrementer from

vc A combinational incrementer from
vc that is parameterized by both the port
bitwidth and the incrementer amount.

9
10

assign out = in + p_inc_value;

11
12

endmodule

Edit the Verilog source file named RegIncr2stage.v. Add lines 33-41 from Figure 20 to instantiate
and connect the second stage of the two-stage registered incrementer. Then reuse the test harness
from the PyMTL tutorial as follows:
% cd ${TUTROOT}/build
% py.test ../tut4_verilog/regincr/RegIncr2stage_test.py -sv

To-Do On Your Own: Create a three-stage registered incrementer similar in spirit to the two-stage
registered incrementer in Figure 19. Verify your design by writing a test script that uses test
vectors.
4.5. Parameterizing a Model with Static Elaboration
As we learned in the PyMTL tutorial, To facilitate model reuse and productive design-space exploration, we often want to implement parameterized models. A common example is to parameterize
models by the bitwidth of various input and output ports. The registered incrementer in Figure 16 is
designed for only for only eight-bit input values, but we may want to reuse this model in a different
context with four-bit input values or 16-bit input values. We can use Verilog parameters to parameterize the port bitwidth for the registered incrementer shown in Figure 16; we would replace references
to constant 7 with a reference to nbits-1. Now we can specify the port bitwidth for our register
incrementer when we construct the model. We have included a library of parameterized Verilog RTL
models in the vc subdirectory. Figure 21 shows a combinational incrementer from vc that is parameterized by both the port bitwidth and the incrementer amount. The parameters are specified using
the special syntax shown on lines 25. By convention, we use a p_ prefix when naming parameters.
Unfortunately, writing highly parameterized models in Verilog can be very tedious or even impossible, which is one key motivation for the PyMTL framework. Having said this, Verilog-2001 does
provide generate statements which are meant for static elaboration. Recall that static elaboration
happens at compile time, not runtime. We can use static elaboration to generate hardware which is
fundamentally different from modeling hardware. Figure 22 illustrates using generate statements
to create a multi-stage registered incrementer that is parameterized by the number of stages. The
number of stages is specified using the the p_nstages parameter shown on line 13. We create a array
of signals to hold the intermediate values between stages (line 25), and then we use a generate for
loop to instantiate and connect the stages. Using generate statements is one of the more advanced
parts of Verilog, so we will not go into more detail within this tutorial. We can reuse the test harness
from the PyMTL tutorial as follows:
% cd ${TUTROOT}/build
30

ECE 4750 Computer Architecture, Fall 2015

1
2
3
4

Tutorial 4: Verilog Hardware Description Language

//========================================================================
// RegIncrNstage
//========================================================================
// Registered incrementer that is parameterized by the number of stages.

5
6
7

`ifndef TUT4_VERILOG_REGINCR_REG_INCR_NSTAGE_V
`define TUT4_VERILOG_REGINCR_REG_INCR_NSTAGE_V

8
9

`include "tut4_verilog/regincr/RegIncr.v"

10
11
12
13
14
15
16
17
18
19

module tut4_verilog_regincr_RegIncrNstage
#(
parameter p_nstages = 2
)(
input logic
clk,
input logic
reset,
input logic [7:0] in,
output logic [7:0] out
);

20
21
22
23

// This defines an _array_ of signals. There are p_nstages+1 signals

// and each signal is 8 bits wide. We will use this array of signals to
// hold the output of each registered incrementer stage.

24
25

logic [7:0] reg_incr_out [p_nstages+1];

26
27
28

// Connect the input port of the module to the first signal in the
// reg_incr_out signal array.

29
30

assign reg_incr_out[0] = in;

31
32
33

// Instantiate the registered incrementers and make the connections

// between them using a generate block.

34
35
36
37

genvar i;
generate
for ( i = 0; i < p_nstages; i = i + 1 ) begin: gen

38
39
40
41
42
43
44
45

tut4_verilog_regincr_RegIncr reg_incr
(
.clk
(clk),
.reset (reset),
.in
(reg_incr_out[i]),
.out
(reg_incr_out[i+1])
);

46
47
48

end
endgenerate

49
50
51

// Connect the last signal in the reg_incr_out signal array to the

// output port of the module.

52
53

assign out = reg_incr_out[p_nstages];

54
55

endmodule

56
57

`endif /* TUT4_VERILOG_REGINCR_REG_INCR_NSTAGE_V */

Figure 22: N-Stage Registered Incrementer A parameterized registered incrementer where the
number of stages is specified using a Verilog parameter.

31

ECE 4750 Computer Architecture, Fall 2015

in_val
in0
in1
in2
in3

Stage S1

Tutorial 4: Verilog Hardware Description Language

Stage S2

Stage S3

n
n

min
max

min
max
min
max

n
n

min
max

min
max

n
n

out_val
out0
out1
out2
out3

Figure 23: Block Diagram for Four-Element Sorter An n-bit pipelined four-element sorter which
arranges the four elements in ascending order.
% py.test ../tut4_verilog/regincr/RegIncrNstage_test.py -sv

5. Sort Unit
The previous section introduces the key Verilog concepts and primitives that we will use to implement more complex RTL models including: declaring a port-based module interface; declaring internal state and wires using logic variables; declaring always @( posedge clk ) concurrent blocks
to model logic that executes on every rising clock edge; declaring always @(*) concurrent blocks to
model combinational logic that executes one or more times within a clock cycle; and creating PyMTL
wrappers.
In this section, we will explore how we can implement the same sorting unit previously seen in the
PyMTL tutorial except using the Verilog hardware description language. As a reminder, the simple
pipelined four-element sorter is shown in Figure 23. Each min/max unit compares its inputs and
sends the smaller value to the top output port and the larger value to the bottom output. This specific
implementation is pipelined into three stages, such that the critical path should be through a single
min/max unit. Input and output valid signals indicate when the input and output elements are
valid. Most of the code for this section is provided for you in the tut4_verilog/sort subdirectory.
5.1. Flat Sorter Implementation
We will begin by exploring a flat implementation of the sorter that does not instantiate any additional
child modules. This implementation is provided for you in the file named SortUnitFlatRTL.v. Figure 24 illustrates the Verilog code that describes the interface for the sorter. Notice how we have
parameterized the interface by the bitwidth of each element. Lines 24 declare a parameter named
p_nbits and give it a default value of one bit. We use this parameter when declaring the bitwidth of
the input and output ports, and we will also use this parameter in the implementation.
Figure 25 shows the first pipeline stage of the flat implementation of the sorter. Notice how we
use the parameter p_nbits to declare various internal registers and wires. We cleanly separate the
sequential logic from the combinational logic. We use comments and explicit suffixes to make it clear
what pipeline stage we are modeling.
The corresponding wrapper is in SortUnitFlatRTL.py and is shown in Figure 26. There are two
important differences from the PyMTL wrapper for the registered incrementer. On line 12, we set
32

ECE 4750 Computer Architecture, Fall 2015

1
2
3
4
5
6

Tutorial 4: Verilog Hardware Description Language

Figure 24: Interface for the Four-Element

Sorter The interface corresponds to the diagram in Figure 23 and is parameterized by the
bitwidth of each element.

module tut4_verilog_sort_SorterFlat
#(
parameter p_nbits = 1
)(
input logic
clk,
input logic
reset,

7
8
9
10
11
12

input
input
input
input
input

logic
logic
logic
logic
logic

[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]

in_val,
in0,
in1,
in2,
in3,

output
output
output
output
output

logic
logic
logic
logic
logic

[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]

out_val,
out0,
out1,
out2,
out3

13
14
15
16
17
18
19

);

vlinetrace to be True. This tells the PyMTL framework that the Verilog module includes its own
line tracing code. On lines 2830, we can specify how the PyMTL parameters (e.g., nbits) correspond
to the Verilog parameters (e.g., p_nbits). We can run the unit tests for this module as follows:
% cd ${TUTROOT}/build
% py.test ../tut4_verilog/sort/SortUnitFlatRTL_test.py
The design should pass all of the tests.
H

To-Do On Your Own: The sorter currently sorts the four input numbers from smallest to largest.
Change to the sorter implementation so it sorts the numbers from largest to smallest. Recompile
and rerun the unit test and verify that the tests are no longer passing. Modify the tests so that
they correctly capture the new expected behavior. Make a copy of the sorter implementation file
so you can put things back to the way they were when you are finished.

33

ECE 4750 Computer Architecture, Fall 2015

1
2
3

Tutorial 4: Verilog Hardware Description Language

//---------------------------------------------------------------------// Stage S0->S1 pipeline registers

//----------------------------------------------------------------------

4
5
6
7
8
9

logic
logic
logic
logic
logic

[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]

val_S1;
elm0_S1;
elm1_S1;
elm2_S1;
elm3_S1;

10
11
12
13
14
15
16
17

always_ff
val_S1
elm0_S1
elm1_S1
elm2_S1
elm3_S1
end

@(
<=
<=
<=
<=
<=

posedge clk ) begin

(reset) ? 0 : in_val;
in0;
in1;
in2;
in3;

18
19
20
21
22
23
24
25

//---------------------------------------------------------------------// Stage S1 combinational logic

//---------------------------------------------------------------------// Note that we explicitly catch the case where the elements contain
// X's and propagate X's appropriately. We would not need to do this if
// we used a continuous assignment statement with a ternary conditional
// operator.

26
27
28
29
30

logic
logic
logic
logic

[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]

elm0_next_S1;
elm1_next_S1;
elm2_next_S1;
elm3_next_S1;

31
32

always_comb begin

33
34

// Sort elms 0 and 1

35
36
37
38
39
40
41
42
43
44
45
46
47

if ( elm0_S1 <= elm1_S1 ) begin

elm0_next_S1 = elm0_S1;
elm1_next_S1 = elm1_S1;
end
else if ( elm0_S1 > elm1_S1 ) begin
elm0_next_S1 = elm1_S1;
elm1_next_S1 = elm0_S1;
end
else begin
elm0_next_S1 = 'x;
elm1_next_S1 = 'x;
end

48
49

// Sort elms 2 and 3

50
51
52
53
54
55
56
57
58
59
60
61
62

if ( elm2_S1 <= elm3_S1 ) begin

elm2_next_S1 = elm2_S1;
elm3_next_S1 = elm3_S1;
end
else if ( elm2_S1 > elm3_S1 ) begin
elm2_next_S1 = elm3_S1;
elm3_next_S1 = elm2_S1;
end
else begin
elm2_next_S1 = 'x;
elm3_next_S1 = 'x;
end

63
64

end

Figure 25: First Stage of the Flat Sorter Implementation First pipeline stage of the sorter using a
flat implementation corresponding to the diagram in Figure 23.
34

ECE 4750 Computer Architecture, Fall 2015

1
2
3

Tutorial 4: Verilog Hardware Description Language

#=========================================================================
# SortUnitFlatRTL
#=========================================================================

4
5

from pymtl import *

6
7

class SortUnitFlatRTL( VerilogModel ):

8
9

# Verilog module setup

10
11
12

vprefix = "tut4_verilog_sort"
vlinetrace = True

13
14

# Constructor

15
16

def __init__( s, nbits=8 ):

17
18

# Interface

19
20
21

s.in_val
s.in_

= InPort (1)
= [ InPort (nbits) for _ in range(4) ]

22
23
24

s.out_val = OutPort(1)
s.out
= [ OutPort (nbits) for _ in range(4) ]

25
26

# Verilog parameters

27
28
29
30

s.set_params({
'p_nbits' : nbits,
})

31
32

# Verilog ports

33
34
35
36

s.set_ports({
'clk'
: s.clk,
'reset'
: s.reset,

37
38
39
40
41
42

'in_val'
'in0'
'in1'
'in2'
'in3'

:
:
:
:
:

s.in_val,
s.in_[0],
s.in_[1],
s.in_[2],
s.in_[3],

'out_val'
'out0'
'out1'
'out2'
'out3'

:
:
:
:
:

s.out_val,
s.out[0],
s.out[1],
s.out[2],
s.out[3],

43
44
45
46
47
48
49

})

Figure 26: Sort Unit Wrapper PyMTL wrapper for the Verilog RTL implementation of the flat sort
unit.

35

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

5.2. Using Verilog Line Traces

We learned about line tracing in the PyMTL tutorial. We have implemented a small library that
enables implementing line tracing code directly within your Verilog modules. While not as elegant
as PyMTL line tracing, Verilog line tracing will still be an essentially way to visualize our designs.
You can use the the -s command-line option to see a line trace of the four-element sorter.
% cd ${TUTROOT}/build
% py.test ../tut4_verilog/sort/SortUnitFlatRTL_test.py -k test_basic -s
Figure 27 shows a portion of a representative line trace with some additional annotation. The first
column indicates the current cycle. There are fixed-width columns showing the inputs and outputs
of the module along with the state in the S1, S2, and S3 pipeline registers at the beginning of the
cycle. Notice that we use spaces when the data is invalid which improves readability and makes it
easy to see when the hardware is actually doing useful work. If you compare this line trace to the line
trace generated in the PyMTL tutorial, they should look identical. Both PyMTL and Verilog enable
the same kind of cycle-accurate modeling. Regardless of whether we are using PyMTL or Verilog,
line traces are a powerful way to visualize your design and debug both correctness and performance
issues.
cycle input ports
stage S1
stage S2
stage S3
output ports
--------------------------------------------------------------------------2:
|
|
|
|
3: {04,02,03,01}|
|
|
|
4:
|{04,02,03,01}|
|
|
5:
|
|{02,04,01,03}|
|
6:
|
|
|{01,03,02,04}|{01,02,03,04}
7:
|
|
|
|

1
2
3
4
5
6
7
8

Figure 27: Line Trace Output for Sort Unit RTL Model This line trace is for the test_basic test
case and is annotated to show what each column corresponds to in the model. Each line corresponds
to one (and only one!) cycle, and the fixed-width columns correspond to either the state at the
beginning of the corresponding cycle or the output of combinational logic during that cycle. If the
valid bit is not set, then the corresponding list of values is not shown.
1

`ifndef SYNTHESIS

2
3
4
5

logic [`VC_TRACE_NBITS-1:0] str;

`VC_TRACE_BEGIN
begin

6
7

// Inputs

8
9
10
11

$sformat( str, "{%x,%x,%x,%x}", in0, in1, in2, in3 );

vc_trace.append_val_str( trace_str, in_val, str );
vc_trace.append_str( trace_str, "|" );

12
13

...

14
15
16

end
`VC_TRACE_END

17
18

`endif /* SYNTHESIS */

Figure 28: Example of Line Tracing Code This code generates the first fixed-width column for the
line trace shown in Figure 27

36

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

As mentioned above, we provide a verilog component (VC) library with various useful Verilog modules, including one module in vc/trace.v which makes it easier to create line traces in your own
Verilog modules. Figure 28 shows a small snippet of code that is used in the sorter implementation
to trace the input ports. Notice how we have wrapped the line tracing code in ifndef SYNTHESIS
and endif to clearly indicate that this code is not synthesizable even though it is included in out
design. Line 3 declares a temporary string variable that we will use when converting nets into
strings. Lines 416 use helper macros to declare a task called trace which takes a special argument called trace_str that holds the current line trace string. The job of this task is to append
trace information to the trace_str that describes the current state and operation of this module.
The vc_trace.append_str and vc_trace.append_val_str helper tasks add strings to the line trace.
You can also build line traces hierarchically by explicitly calling the trace task on a child module.
Although we will provide significant line tracing code in each lab harness, you are also strongly
encouraged to augment this code with your own line tracing.
H

To-Do On Your Own: Modify the line tracing code to show the pipeline stage label (in, S1, S2, S3,
out) before each stage. After your modifications, the line trace might look something like this:
24: in:{05,07,06,08}|S1:{a5,a3,a2,a7}|S2:{03,04,01,02}|S3:

|out:

5.3. Structural Sorter Implementation

The flat implementation in SortUnitFlatRTL.v is complex and monolithic and it fails to really exploit the structure inherent in the sorter. Just as in the PyMTL tutorial, we can use modularity and
hierarchy to divide complicated designs into smaller more manageable units; these smaller units
are easier to design and can be tested independently before integrating them into larger, more complicated designs. We have started a structural implementation in SortUnitStructRTL.v; the structural implementation will have an identical interface and behavior as the flat implementation in
SortUnitFlatRTL.v.
Figure 29 shows the first pipeline stage of the structural implementation of the sorter. Our design instantiates three kinds of modules: vc_ResetReg, vc_Reg, and tut4_verilog_sort_MinMaxUnit. The
register modules are provided in the VC library. Notice how we still use the parameter p_nbits
to declare various internal variables, but in addition, we use this parameter when instantiating parameterized sub-modules. For example, the vc_Reg module is parameterized, and this allows us to
easily create pipeline registers of any bitwidth. Even though we are using a structural implementation strategy, we still cleanly separate the sequential logic from the combinational logic. We still use
comments and explicit suffixes to make it clear what pipeline stage we are modeling.

37

ECE 4750 Computer Architecture, Fall 2015

1
2
3

Tutorial 4: Verilog Hardware Description Language

//---------------------------------------------------------------------// Stage S0->S1 pipeline registers

//----------------------------------------------------------------------

4
5

logic val_S1;

6
7
8
9
10
11
12
13

vc_ResetReg#(1) val_S0S1
(
.clk
(clk),
.reset (reset),
.d
(in_val),
.q
(val_S1)
);

14
15
16
17

// This is probably the only place where it might be acceptable to use

// positional port binding since (a) it is so common and (b) there are
// very few ports to bind.

18
19
20
21
22

logic
logic
logic
logic

[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]
[p_nbits-1:0]

elm0_S1;
elm1_S1;
elm2_S1;
elm3_S1;

23
24
25
26
27

vc_Reg#(p_nbits)
vc_Reg#(p_nbits)
vc_Reg#(p_nbits)
vc_Reg#(p_nbits)

elm0_S0S1(
elm1_S0S1(
elm2_S0S1(
elm3_S0S1(

clk,
clk,
clk,
clk,

elm0_S1,
elm1_S1,
elm2_S1,
elm3_S1,

in0
in1
in2
in3

);
);
);
);

28
29
30
31

//---------------------------------------------------------------------// Stage S1 combinational logic

//----------------------------------------------------------------------

32
33
34

logic [p_nbits-1:0] mmuA_out_min_S1;

logic [p_nbits-1:0] mmuA_out_max_S1;

35
36
37
38
39
40
41
42

tut4_verilog_sort_MinMaxUnit#(p_nbits) mmuA_S1
(
.in0
(elm0_S1),
.in1
(elm1_S1),
.out_min (mmuA_out_min_S1),
.out_max (mmuA_out_max_S1)
);

43
44
45

logic [p_nbits-1:0] mmuB_out_min_S1;

logic [p_nbits-1:0] mmuB_out_max_S1;

46
47
48
49
50
51
52
53

tut4_verilog_sort_MinMaxUnit#(p_nbits) mmuB_S1
(
.in0
(elm2_S1),
.in1
(elm3_S1),
.out_min (mmuB_out_min_S1),
.out_max (mmuB_out_max_S1)
);

Figure 29: First Stage of the Structural Sorter Implementation First pipeline stage of the sorter
using a structural implementation corresponding to the diagram in Figure 23.

38

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

To-Do On Your Own: The structural implementation is incomplete because the actual implementation of the min/max unit in MinMaxUnit.v is not finished. You should go ahead and implement
the min/max unit, and then as always you should write a unit test to verify the functionality of your
MinMax unit! Add some line tracing for the min/max unit. You should have enough experience based on the previous sections to be able to create a unit test from scratch and run it using
py.test. You should name the new test script MinMaxUnit_test.py. You can use the registered
incrementer model as an example for both implementing the min/max unit and for writing the
corresponding test script. Once your min/max unit is complete and tested, then test the structural sorter implementation like this:
% cd ${TUTROOT}/build
% py.test ../tut4_verilog/sort/SortUnitStructRTL_test.py -v
% py.test ../tut4_verilog/sort/SortUnitStructRTL_test.py -k test_basic -s
The line trace for the sort unit structural RTL model should be the same as in Figure 27, since
these are really just two different implementations of the sort unit RTL.
5.4. Evaluating the Sorter Using a Simulator
Just like we can use PyMTL test harnesses to verify Verilog modules, we can also use PyMTL simulation harnesses for evaluation. Since our PyMTL wrapper has the exact same interface as a PyMTL
RTL implementation, the exact same simulator from the PyMTL tutorial will work without change.
% cd ${TUTROOT}/build
% ../tut4_verilog/sort/sort-sim --stats --impl rtl-flat
% ../tut4_verilog/sort/sort-sim --stats --impl rtl-struct

6. Greatest Common Divisor: Verilog Design Example

In this section, we will apply what we have learned in the previous section to study a more complicated hardware unit that calculates the greatest common divisor (GCD) of two input operands. Our
design will be exactly the same as what we experimented with in the PyMTL tutorial, except now
we will be using the Verilog hardware description language. The code for this section is provided
for you in the tut4_verilog/gcd subdirectory.
As a reminder, Figure 30 illustrates the interface for our module. The GCD unit will take two n-bits
operands and produce an n-bit result. For flow-control we use the same latency-insensitive val/rdy
interface that we learned about in the PyMTL tutorial. Our implementation will use Euclids algorithm to iteratively calculate the GCD. Figure 31 illustrates this algorithm as an executable Python
function.
H

To-Do On Your Own: Experiment with the algorithm using a Python interpreter. Try calculating
the GCD for different input values. Add additional debugging output to track what the algorithm
is doing each iteration.

39

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

6.1. Control/Datapath Split Implementation

As discussed in the PyMTL tutorial, we will usually divide more complicated designs into two parts:
the datapath and the control unit. The datapath contains the arithmetic operators, muxes, and registers that work on the data, while the control unit is responsible for controlling these components to
achieve the desired functionality. The control unit sends control signals to the datapath and the datapath sends status signals back to the control unit. Figure 32 illustrates the datapath for the GCD unit
and Figure 33 illustrates the corresponding finite-state-machine (FSM) control unit. The Verilog code
for the datapath, control unit, and the top-level module which composes the datapath and control
unit is in GcdUnitRTL.v.
Take a look at the datapath interface which is also shown in Figure 35. We cleanly identify the data
signals, control signals, and status signals. Figure 35 also shows the first two datapath components,
but take a look in GcdUnitRTL.v to see the entire datapath. Notice how we use a very structural
implementation that exactly matches the datapath diagram in Figure 32. We leverage several modules
1
2
3

req.msg

32

16

GCD

req.val
req.rdy

req_q

resp_q

resp.msg

resp.val

resp.rdy

7
8

Figure 30: Functional-Level Implementation of

GCD Unit Input and output use latencyinsensitive val/rdy interfaces. The input message
includes two 16-bit operands; output message is an
16-bit result. Clock and reset signals are not shown.

req_msg.b

!req_val

is_b_zero

IDLE
zero?

req_val

b_reg
sub

req_msg.a

req_msg

less
than?

a_mux_sel

Figure 31: Euclids GCD Algorithm Iteratively subtract the smaller value from the
larger value until one of them is zero, at
which time the GCD is the non-zero value.
This is executable Python code.

resp_msg

is_a_lt_b
b_mux b_reg
_sel
_en

def gcd( a, b ):
while True:
if a < b:
a,b = b,a
elif b != 0:
a = a - b
else:
return a

a_reg

a<b
/swap

Figure 32: Datapath Diagram for GCD Datapath includes two state registers and required muxing and arithmetic units to iteratively implement Euclids algorithm.

40

b != 0
/sub

b == 0
!resp_rdy

a_reg_en

CALC

DONE

resp_rdy

Figure 33: FSM Diagram for GCD A

hybrid Moore/Mealy FSM for controlling the datapath in Figure 32. Mealy
transitions in the calc state determine
whether to swap or subtract.

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

from the VC library (e.g., vc_Mux2, vc_ZeroComparator, vc_Subtractor). You should use a similar
structural approach when building your own datapaths for this course. For a net that moves data
from left to right in the datapath diagram, we usually declare a dedicated wire right before the
module instance (e.g., a_mux_out and a_reg_out). For a net that moves data from right to left in
the datapath diagram, we need to declare a dedicated wire right before it is used as an input (e.g.,
b_reg_out and b_sub_out).
Take a look at the control unit and notice the stylized way we write FSMs. An FSM-based control
unit should have three parts: a sequential concurrent block for the state, a combinational concurrent
block for the state transitions, and a combinational concurrent block for the state outputs. We often
use case statements to compactly represent the state transitions and outputs. Figure 36 shows the
portion of the FSM responsible for setting the output signals. We use a task to set all of the control
signals in a single line; as long as the task does not include non-synthesizable constructs (e.g., delay
statements or system tasks) the task itself should be synthesizable. Essentially, we have created a
control signal table in our Verilog code which exactly matches what we might draw on a piece of
paper. There is one row for each state or Mealy transition and one column for each control signal.
These compact control signal tables simplify coding complicated FSMs (or indeed other kinds of
control logic) and can enable a designer to quickly catch bugs (e.g., are the enable signals always set
to either zero or one?).
Figure 37 shows a portion of the top-level module that connects the datapath and control unit together. Lines X and Y use the new implicit connection operator (.*) provided by SystemVerilog.
Using the implicit connection operator during module instantiation means to connect every port to a
signal with the same name. So these two lines take care of connecting all of the control and status signals. This is a powerful way to write more compact code which avoids connectivity bugs, especially
when connecting the datapath and control unit.
We can use the exact same source/sink testing that we used in the PyMTL tutorial to test our Verilog
implementation of the GCD unit. As a reminder, Figure 34 illustrates the overall connectivity in the
test harness. The test source includes the ability to randomly delay messages going into the DUT
and the test sink includes the ability to randomly apply back-pressure to the DUT. By using various
combinations of these random delays we can more robustly ensure that our flow-control logic is
working correctly. We can reuse the same test harnesses from the PyMTL tutorial as follows:
% cd ${TUTROOT}/build
% py.test ../tut4_verilog/gcd/GcdUnitRTL_test.py -v
% py.test ../tut4_verilog/gcd/GcdUnitRTL_test.py -sv -k basic_0x0
Figure 38 illustrates a portion of the line trace for the randomized testing. We use the line trace to
show the state of the A and B registers at the beginning of each cycle and use specific characters to
req_msg

Test Source

req_val
req_rdy

resp_msg

GCD Unit

resp_val
resp_rdy

Test Sink

Figure 34: Verifying GCD Using Test Sources and Sinks Parameterized test sources send a stream
of messages over a val/rdy interface, and parameterized test sinks receive a stream of messages over
a val/rdy interface and compare each message to a previously specified reference message. Clock
and reset signals are not shown.

41

ECE 4750 Computer Architecture, Fall 2015

1
2
3
4

Tutorial 4: Verilog Hardware Description Language

module tut4_verilog_gcd_GcdUnitDpath
(
input logic
clk,
input logic
reset,

// Data signals

6
7

input logic [31:0] req_msg,

output logic [15:0] resp_msg,

8
9
10

// Control signals

11
12

input
input
input
input

13
14
15
16

logic
logic
logic [1:0]
logic

a_reg_en,
b_reg_en,
a_mux_sel,
b_mux_sel,

//
//
//
//

Enable for A register

Enable for B register
Sel for mux in front of A reg
sel for mux in front of B reg

is_b_zero,
is_a_lt_b

// Output of zero comparator

// Output of less-than comparator

17

// Status signals

18
19
20
21
22

);

output logic
output logic

23
24

localparam c_nbits = 16;

25
26

// Split out the a and b operands

27
28
29

logic [c_nbits-1:0] req_msg_a = req_msg[31:16];

logic [c_nbits-1:0] req_msg_b = req_msg[15:0 ];

30
31

// A Mux

32
33
34
35

logic [c_nbits-1:0] b_reg_out;

logic [c_nbits-1:0] sub_out;
logic [c_nbits-1:0] a_mux_out;

36
37
38
39
40
41
42
43
44

vc_Mux3#(c_nbits) a_mux
(
.sel
(a_mux_sel),
.in0
(req_msg_a),
.in1
(b_reg_out),
.in2
(sub_out),
.out
(a_mux_out)
);

45
46

// A register

47
48

logic [c_nbits-1:0] a_reg_out;

49
50
51
52
53
54
55
56
57

vc_EnReg#(c_nbits) a_reg
(
.clk
(clk),
.reset (reset),
.en
(a_reg_en),
.d
(a_mux_out),
.q
(a_reg_out)
);

Figure 35: Portion of GCD Datapath Unit We use struct types to encapsulate both control and
status signals and we use a preprocessor macro from the GCD message struct to determine how to
size the datapath components.

42

ECE 4750 Computer Architecture, Fall 2015

1
2
3

Tutorial 4: Verilog Hardware Description Language

//---------------------------------------------------------------------// State Outputs

//----------------------------------------------------------------------

4
5
6
7
8

localparam
localparam
localparam
localparam

a_x
a_ld
a_b
a_sub

=
=
=
=

2'dx;
2'd0;
2'd1;
2'd2;

9
10
11
12

localparam b_x
localparam b_ld
localparam b_a

= 1'dx;
= 1'd0;
= 1'd1;

13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

task cs
(
input logic
input logic
input logic
input logic
input logic
input logic
);
begin
req_rdy
=
resp_val =
a_reg_en =
b_reg_en =
a_mux_sel =
b_mux_sel =
end
endtask

cs_req_rdy,
cs_resp_val,
[1:0] cs_a_mux_sel,
cs_a_reg_en,
cs_b_mux_sel,
cs_b_reg_en
cs_req_rdy;
cs_resp_val;
cs_a_reg_en;
cs_b_reg_en;
cs_a_mux_sel;
cs_b_mux_sel;

32
33

// Labels for Mealy transistions

34
35
36

logic do_swap;
logic do_sub;

37
38
39

assign do_swap = is_a_lt_b;

assign do_sub = !is_b_zero;

40
41

// Set outputs using a control signal "table"

42
43

always_comb begin

44
45
46
47
48
49
50
51
52
53

set_cs( 0, 0, a_x, 0, b_x, 0 );

case ( state_reg )
//
req resp
//
rdy val
STATE_IDLE:
cs( 1, 0,
STATE_CALC: if ( do_swap ) cs( 0, 0,
else if ( do_sub ) cs( 0, 0,
STATE_DONE:
cs( 0, 1,
default
cs('x, 'x,

a mux a
sel
en
a_ld, 1,
a_b,
1,
a_sub, 1,
a_x,
0,
a_x, 'x,

b mux b
sel
en
b_ld, 1 );
b_a, 1 );
b_x, 0 );
b_x, 0 );
b_x, 'x );

54
55

endcase

56
57

end

Figure 36: Portion of GCD FSM-based Control Unit for State Outputs We can use a task to create
a control signal table with one row per state or Mealy transition and one column per control signal.
Local parameters can help compactly encode various control signal values.

43

ECE 4750 Computer Architecture, Fall 2015

1
2
3
4

Tutorial 4: Verilog Hardware Description Language

module tut4_verilog_gcd_GcdUnitRTL
(
input logic
clk,
input logic
reset,

5
6
7
8

input logic
output logic
input logic [31:0]

req_val,
req_rdy,
req_msg,

output logic
input logic
output logic [15:0]

resp_val,
resp_rdy,
resp_msg

9
10
11
12
13

);

14
15
16
17

//---------------------------------------------------------------------// Connect Control Unit and Datapath

//----------------------------------------------------------------------

18
19

// Control signals

20
21
22
23
24

logic
logic
logic [1:0]
logic

a_reg_en;
b_reg_en;
a_mux_sel;
b_mux_sel;

25
26

// Data signals

27
28
29

logic
logic

is_b_zero;
is_a_lt_b;

30
31

// Control unit

32
33
34
35
36

tut4_verilog_gcd_GcdUnitCtrl ctrl
(
.*
);

37
38

// Datapath

39
40
41
42
43

tut4_verilog_gcd_GcdUnitDpath dpath
(
.*
);

44
45

endmodule

Figure 37: Portion of GCD Top-Level Module We use the new implicit connection operator (.*)
to automatically connect all of the control and status signals to both the control unit and datapath.
indicate which state we are in (i.e., I = idle, Cs = calc with swap, C- = calc with subtract, D = done).
We can see that the test source sends a new message into the GCD unit on cycle 296. The GCD unit
is in the idle state and transitions into the calc state. It does five subtractions and a final swap before
transitioning into the done state on cycle 304. The result is valid but the test sink is not ready, so the
GCD unit waits in the done state until cycle 310 when it is able to send the result to the test sink. On
cycle 311 the GCD unit accepts a new input message to work on. This is a great example of how an
effective line trace can enable you to quickly visualize how a design is actually working.

44

ECE 4750 Computer Architecture, Fall 2015

cycle src

Tutorial 4: Verilog Hardware Description Language

Areg Breg ST out

sink

2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

296:
297:
298:
299:
300:
301:
302:
303:
304:
305:
306:
307:
308:
309:
310:
311:
312:
313:
314:

002d00e1
#
#
#
#
#
#
#
#
#
#
#
#
#
#
002200cc
#
#
#

>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>

00e1:002d(0002
#
(00e1
#
(00b4
#
(0087
#
(005a
#
(002d
#
(0000
#
(002d
#
(002d
#
(002d
#
(002d
#
(002d
#
(002d
#
(002d
#
(002d
00cc:0022(002d
#
(00cc
#
(00aa
#
(0088

0000
002d
002d
002d
002d
002d
002d
0000
0000
0000
0000
0000
0000
0000
0000
0000
0022
0022
0022

I )
C-)
C-)
C-)
C-)
C-).
Cs)
C )
D )#
D )#
D )#
D )#
D )#
D )#
D )002d
I )
C-)
C-)
C-)

>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>

Figure 38: Line Trace for RTL Implementation of GCD State of A and B registers at the beginning of the cycle is shown,
along with the current state of the FSM.
I = idle, Cs = calc with swap, C- = calc with
subtract, D = done.

#
#
#
#
#
#
002d

To-Do On Your Own: Optimize the GCD implementation to improve the performance on the given
input datasets.
A first optimization would be to transition into the done state if either a or b are zero. If a is zero
and b is greater than zero, we will swap a and b and then end the calculation on the next cycle
anyways. You will need to carefully modify the datapath and control so that the response can
come from either the a or b register.
A second optimization would be to avoid the bubbles caused by the IDLE and DONE states.
First, add an edge from the CALC state directly back to the IDLE state when the calculation is
complete and the response interface is ready. You will need to carefully manage the response
valid bit. Second, add an edge from the CALC state back to the CALC state when the calculation
is complete, the response interface is ready, and the request interface is valid. These optimizations
should eliminate any bubbles and improve the performance of back-to-back GCD transactions.
A third optimization would be to perform a swap and subtraction in the same cycle. This will
require modifying both the datapath and the control unit, but should have a significant impact
on the overall performance.
6.2. Evaluating GCD Unit using a Simulator
As with the previous section, we have provided a simulator for evaluating the performance of the
GCD implementation. In this case, we are focusing on a single implementation with two different
input datasets. You can run the simulators and look at the average number of cycles to compute a
GCD for each input dataset like this:
% cd ${TUTROOT}/build
% ../tut4_verilog/gcd/gcd-sim --stats --impl cl --input random
% ../tut4_verilog/gcd/gcd-sim --stats --impl rtl --input random

45

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

Figure 39: TravisCI Settings Page

7. TravisCI for Continuous Integration

As discussed in the Git tutorial, TravisCI is an online continuous integration service that is tightly
coupled to GitHub. TravisCI will automatically run all tests for a students lab assignment every
time the students push their code to GitHub. We will be using the results reported by TravisCI to
evaluate the code functionality of the lab assignments. In this section, we do a small experiment to
illustrate how TravisCI works for PyMTL projects.
The first step is to enable TravisCI for the remote repository in GitHub. Log into TravisCI using your
GitHub ID and password:
https://travis-ci.org/profile
Once you have signed in, you should go to your TravisCI profile and find the list of your public
GitHub repositories. You may need to click Sync to ensure that TravisCI has the most recent view of
your public repositories on GitHub. Turn on TravisCI with the little switch next to the repository
we have been using in this tutorial (<githubid>/ece4750-tut4-verilog). Figure 39 shows what the
TravisCI settings page should look like and the corresponding switch. After enabling TravisCI for
the <githubid>/ece4750-tut4-verilog repository, you should be able to go to the TravisCI page
for this repository:
https://travis-ci.org/<githubid>/ece4750-tut4-verilog
TravisCI will report that there are no builds for this repository yet. Go ahead and commit all of
the work you have done in this tutorial, then push your local commits to the remote repository on
GitHub. If you revisit the TravisCI page for this repository, you should see TravisCI starting to build
and run all of your tests. Figure ?? shows what the TravisCI build log will look like for a brand new
fork of the tutorial repository. Study the TravisCI log output to verify that TravisCI is: (1) installing

46

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

Figure 40: TravisCI Log

Verilator; (2) installing PyMTL; (3) creating a build directory; and (4) running all of your unit tests.
Confirm that if all of the tests pass on amdpool then they also pass on TravisCI.

Acknowledgments
This tutorial was developed for ECE 4750 Computer Architecture course at Cornell University by
Christopher Batten.

47

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

Appendix A: Constructs Allowed in Synthesizable RTL

Always Allowed in
Synthesizable RTL
logic
logic [N-1:0]
& | ^ ^~ ~ (bitwise)
&& || !
& ~& | ~| ^ ^~ (reduction)
+ >> << >>>
== != > <= < <=
{}
{N{}} (repeat)
?:
always_ff, always_comb
if else
case, endcase
begin, end
module, endmodule
input, output
assign
parameter
localparam
genvar
generate, endgenerate
generate for
generate if else
generate case
$clog2()
$bits()
named port connections
named parameter passing

Allowed in
Synthesizable RTL
With Limitations

Explicitly
Not Allowed in
Synthesizable RTL

always1
enum2
struct3
casez, endcase4
task, endtask5
function, endfunction5
= (blocking assignment)6
<= (non-blocking assignment)7
typedef8
packed9
$signed()10
read-modify-write signal11

wire, reg12
integer, real, time, realtime
signed13
===, !==
* / % ** 14
#N (delay statements)
inout15
initial
variable initialization16
negedge17
casex, endcase
for, while, repeat, forever18
fork, join
deassign, force, release
specify, endspecify
nmos, pmos, cmos
rnmos, rpmos, rcmos
tran, tranif0, tranif1
rtran, rtranif0, rtranif1
supply0, supply1
strong0, strong1
weak0, weak1
primitive, endprimitive
defparam
unnamed port connections19
unnamed parameter passing20
all other keywords
all other system tasks

1 Students should prefer using always_ff and always_comb instead of always. If students insist on using
always, then it can only be used in one of the following two constructs: always @(posedge clk) for sequential logic, and always @(*) for combinational logic. Students are not allowed to trigger sequential
blocks off of the negative edge of the clock or create asynchronous resets, nor use explicit sensitivity lists.
2 enum can only be used with an explicit base type of logic and explicitly setting the bitwidth using the following syntax: typedef enum logic [$clog2(N)-1:0] { ... } type_t; where N is the number of labels
in the enum. Anonymous enums are not allowed.
3 struct can only be used with the packed qualifier (i.e., unpacked structs are not allowed) using the following syntax: typedef struct packed { ... } type_t; Anonymous structs are not allowed.
4 casez can only be used in very specific situations to compactly implement priority encoder style hardware
structures.
5 task and function blocks must themselves contain only synthesizable RTL.

48

ECE 4750 Computer Architecture, Fall 2015

Tutorial 4: Verilog Hardware Description Language

Blocking assignments should only be used in always_comb blocks and are explicitly not allowed in
always_ff blocks.

Non-blocking assignments should only be used in always_ff blocks and are explicitly not allowed in
always_comb blocks.

typedef should only be used in conjunction with enum and struct.

packed should only be used in conjunction with struct.

10 Reading a signal, performing some arithmetic on the corresponding value, and then writing this value back
to the same signal (i.e., read-modify-write) is not allowed within an always_comb concurrent block. This is
a combinational loop and does not model valid hardware. Read-modify-write is allowed in an always_ff
concurrent block with a non-blocking assignment, although we urge students to consider separating the
sequential and combinational logic. Students can use an always_comb concurrent block to read the signal, perform some arithmetic on the corresponding value, and then write a temporary wire; and use an
always_ff concurrent block to flop the temporary wire into the destination signal.
11 $signed() can only be used around the operands to >>>, >, >=, <, <= to ensure that these operators perform
the signed equivalents.
12 wire and reg are perfectly valid, synthesizable constructs, but logic is much cleaner. So we would like
students to avoid using wire and reg.
13 signed types can sometimes be synthesized, but we do not allow this construct in the course.
14 Multipliers can be synthesized and small multipliers can even be synthesized efficiently, but we do not
allow this construct in the course. If you need to multiply or divide by a power of two, then you can use
the left and right shift operators.
15 Ports with inout can be used to create tri-state buses, but tools often have trouble synthesizing hardware
from these kinds of models.
16 Variable initialization means assigning an initial value to a logic variable when you declare the variable.
This is not synthesizable; it is not modeling real hardware. If you need to set some state to an initial
condition, you must explicitly use the reset signal.
17 Triggering a sequential block off of the negedge of a signal is certainly synthesizable, but we will be exclusively using a positive-edge-triggered flip-flop-based design style.
18 If you would like to generate hardware using loops, then you should use generate blocks.
19 In very specific, rare cases unnamed port connections might make sense, usually when there are just one or
two ports and there purpose is obvious from the context.
20 In very specific, rare cases unnamed parameter passing might make sense, usually when there are just one
or two parameters and their purpose is obvious from the context.

49