Resources

• Try out EDA Playground for online verilog simulation.
• OMG and try DigitalJS Online to VISUALISE THE SYNTHESIS RESULTS!!
• See
  • FPGAs For Dummies
  • Verilog Pro
  • Nand Land Verilog Tutorials
  • FPGA Architectures: An Overview
  • Advanced Verilog - ECS 270 v10/23/0
  • IEEE Standard for Verilog(R) Hardware Description Language
  • Learn FPGA
  • FPGATutorial.com

Quick Looks At FPGAs

Field programmable Gate Arrays (FPGAs) are pre-fabricated silicon devices that can be electrically programmed in the field to become almost any kind of digital circuit or system. For low to medium volume productions, FPGAs provide cheaper solution and faster time to market as compared to Application Specific Integrated Circuits (ASIC) which normally require a lot of resources in terms of time and money to obtain first device

Normally FPGAs comprise of:

• Programmable logic blocks which implement logic functions.
• Programmable routing that connects these logic functions.
• I/O blocks that are connected to logic blocks through routing interconnect and that make off-chip connections.

Modern FPGAs consist of mixes of configurable static random access memory (SRAM or Flash), high-speed input/output pins (I/Os), logic blocks, and routing. More specifically, an FPGA contains programmable logic elements called, naturally, logic elements (LEs), as well as a hierarchy of reconfigurable interconnects that allow the LEs to be physically connected to one another. You can configure LEs to do complex functions or simply perform basic logic gates, such as AND and OR. Most FPGAs also contain memory blocks ...

... The core of an FPGA is simply an array of these logic gates and wires etched into an integrated circuit in a way that allows you to reconfigure them ...

Major FPGA companies are people like Xilinx (now AMD) and Altera (now Intel). Also Cypress. They tend to produce higher end devices in terms of density, speed etc.

Peope like Lattice produce devices that are more mid-range, and Efinix devices are at the lower end and priced accordingly.

Small Example Of Parallel Execution

• Unlike "normal" software programming languages like "C", which is sequention, HDL statements are executed in parallel and continually execute (always) for a specified number of times or times.

//        +-----+
// A -----| NOR |----+
// B --+--|     |    |   D +-----+
//     |  +-----+    +-----| OR  |----- Q
//     |             +-----|     |
//     |  +-----+    |   E +-----+
//     +--| AND |----+
// C -----|     |
//        +-----+

`timescale 1 ns/1 ps
module smallExample (A,B,C,Q);

    input A,B,C; // Inputs
    output Q;    // Outputs
    wire D,E;    // Internmediate signals internal to FPGA

    always @(A,B,C,D,E)     // always @(<sensitivity-list>). Could also be written always@(*)
                            // where the @(*) syntax is a shorthand for specifying sensitivity
                            // to all signals in this block.
                            // "," means AND. To or replace "," with " or ".
    begin
        D <= A nor B;       // "<=" means PARALLEL EXECUTION
        E <= B and C;
        Q <= D or E;
    end

endmodule

The above is an example of combinatorial logic: logic that does not require a clock to operate.

In the above, the execution of the logic is done in parallel, as noted by the use of the <= operator.

Where = to be used, then the execution would be sequential. In sequential logic, where the output depends on the current state and inputs, you might use something like always @(posedge clk or posedge rst) to trigger the block on positive clock edges or resets.

The always block is continuously executed, triggered on changes to any of the signals in the sensitivity list.

Note that the following assignments in the always block shown are identical to continuous assignments outside of a block:

always @(A,B,C,D,E)
begin
    D <= A nor B;
    E <= B and C;
    Q <= D or E;
end

Is identical to:

assign D = A nor B;
assign E = B and C;
assign Q = D or E;

Because there is no clock being used.

At the start of the example we see timescale 1 ns/1 ps`, which is used to specify the time unit and precision for the simulation:timescale time_unit / time_precision`.

Levels Of Descriptions

1. Behavioural model: Design based on truth table.
2. Data flow model: Design based on boolean equations.
3. Gate level model: Design based on gates.

E.g. 2x1 MUX, inputs D0 and D1, select pin S and output Y:

Behavioural

module mux(
    input D0,
    input D1,
    input S,
    output Y
);

    always @(D0 or D1 or S)
    begin
        if (S)
            Y = D1;
        else
            Y = D0;            
    end

endmodule

Dataflow

module mux(
    input D0,
    input D1,
    input S,
    output Y
);

    assign Y = S ? D1: D0;

endmodule

Gate Level

module mux(
    input D0,
    input D1,
    input S,
    output Y
);

    wire T1, T2, Sbar;

    and (T1, D1, S);
    and (T2, D0, Sbar);
    not (SBar, S);
    or(Y, T1, T2);

endmodule

Verilog Constructs and Datatypes

• Comments: Just Like C/C++ comments
• Number specification: (Size)'(Radix)(Value). E.g. 4'b100 is a 4 bit binary value, 4'd9 is a 4 bit decimal value, 4'hF is a 4 bit hexadecimal value.
• Data types:
  • Logic values: 1, 0, x, z. x means "don't care" and z means "high impedance".
  • wire (nets):
    • Wires are used for connecting different elements. They can be treated as physical wires - they are a type of net.
    • They can be read or assigned, but only by continuous assignment can they be assigned. I.e. use of assign statement outside of an always block or driven directly from an output port.
    • No values get stored in them: teh value is derived from what is being driven from its drivers.
    • They need to be driven by either continuous assign statement or from a port of a module. [Ref].
    • By default they are 1-but wide, i.e, they are scalar. A vector is required to hold values other than 0 or 1 (see later).
  • reg (registers):
    • Use when you want to represent a piece of storage.
    • Can only be assigned using proceedural statements, I.e., drive from an always/initial block.
    • Contrary to their name, regs don't necessarily correspond to physical registers. They represent data storage elements in Verilog/SystemVerilog. They retain their value until the next value is assigned to them (not through an assign statement). They can be synthesized to FF, latch or combinatorial circuits. [Ref].
    • Put another way, "a register remembers a piece of information until it is told to remember something else" [Ref].
  • reg v.s. wire
    • See also
      • Verilog reg, Verilog wire, SystemVerilog logic. What’s the difference? by Jason Yu
    • For wire vs reg, Jason Yu (see link above) has some good rule's of thumb:
      1. Use Verilog reg when you want to represent a piece of storage, and use Verilog wire when you want to represent a physical connection.
      2. Drive a Verilog wire with assign statement or port output
      3. Drive a Verilog reg from an always block. If you want to drive a physical connection with combinatorial logic inside an always@(*) block, then you have to declare the physical connection as Verilog reg.
  • integer
  • real
  • string
  • time (current time is $time)
  • parameter: same as const in C
  • vectors
  • arrays
• Operators
  • Conacatenation {} [Ref].
    • Can concatenate two more more types/signals (not necessarily the same).
    • Examples:
      • wire [7:0] C = {A, B}, where A = 4'b0111 and B=4'b1100, results in C=8'01111100 or 0x7C.
      • wire [15:0] C = {A, B}, where A = 4'b0111 and B=4'b1100, results in C=8'0000000001111100 or 0x007C. This demonstrates padding.
      • r_SHIFT_REG[7:0] <= {r_SHIFT_REG[6:0], r_SHIFT_REG[7]}; demonstrates a circular rotation of a buffer.
  • Replication {}.
    • <repition count>{<thing-to-replicate>}
    • The repetition multiplier must be a constant.
    • Example, if reg [3:0] V = 4'b0111;:
      • 3{V} results in 12b011101110111.
  • Reduction
    • Reduces a vector of bits to a single boolean value.
    • E.g. &4'b1101 is 0 as bits are all AND'ed together. I.e. &4'b1101 == 1 AND 1 AND 0 AND 1 == 0.
    • E.g. |4'b1101 is 1 as bits are all OR'ed together.
    • Can be used in assignments, e.g. reg a = |4'b1101 is equivalent to reg a = 1.

Assignment

Proceedural Assignment

Procedural assignments update the value of variables under the control of the procedural flow constructs that surround them.

...procedural assignments put values in variables. The assignment does not have duration; instead, the variable holds the value of the assignment until the next procedural assignment to that variable.

Procedural assignments occur within procedures such as always, initial (see 9.9), task, and function (see Clause 10) and can be thought of as "triggered" assignments...

A brief note on assignment operators [Ref]:

• <= is non-blocking and is performed on every positive edge of clock. These are evaluated in parallel so no guarantee of order. An example of this would be a register.
• = is blocking assignment, which should only be used inside always statements and enforces sequential order.

Both <= and = must be used within an initial or always block.

Be warned: Nonblocking Assignments in Verilog Synthesis, Coding Styles That Kill!.

Blocking Proceedural Assignment

A blocking procedural assignment statement shall be executed before the execution of the statements that follow it in a sequential block ... A blocking procedural assignment statement shall not prevent the execution of statements that follow it in a parallel block ... [Ref]

In other words:

A blocking assignment gets its name because a blocking assignment must evaluate the RHS arguments and complete the assignment without interruption from any other Verilog statement. The assignment is said to "block" other assignments until the current assignment has completed [Ref]

My colleage has warned me off using blocking assignment in proceedural blocks:

= should not be used in always blocks, <= should only be used in always blocks.

...

Because, for synthesizable logic, all the things that happen inside an always block must happen simultaneously. They can't happen in sequence, as would be the case with a blocking assignment.

...

The thing with VHDL and Verilog is they were both originally designed for modelling digital systems and then kind of borrowed for doing FPGA development, so it's completely possible to write code that is perfectly valid, in terms of syntax and function but completely impossible to synthesize into something that'll run in an FPGA. So, it is valid to put blocking assignments in an always block, but most tools won't synthesize that because there is no implicit way for an FPGA to sequence the assignments.

...

Synthesizability, for some tools, may also depend on whether blocking assignments, inside an always block, have interdependencies. For example:

always @(posedge clk) begin
  x = b;
  y = c;
end

may synthesize fine, because it's irrelevant whether assignment was blocking or non-blocking, in that case. But:

always @(posedge clk) begin
  x = b;
  y = x;
end

may not, because y depends on x having been modified first.

I guess, in short, what I'm saying is; blocking assignments, inside always blocks (at least, podedge/negedge always blocks) cause headaches, non-blocking assignments don't.

FPGAs are designed to 'run' synchronous register transfer logic - That is what they're optimized for. They are based on the concept of clock domains where everything transitions on a clock edge and persists between clock edges. If you write your VHDL/Verilog with this principal in mind and describe, explicitly, what happens on a clock edge, you maximize your chances of the code being synthesizable.

Non-Blocking Proceedural Assignment

The nonblocking procedural assignment allows assignment scheduling without blocking the procedural flow. The nonblocking procedural assignment statements can be used whenever several variable assignments within the same time step can be made without regard to order or dependence upon each other. [Ref]

If there is no timing control in the block in which the <= is being used, then the order of evalulation of the expressions is not specified - i.e., no particular ordering is guaranteed.

However, if timing control is specified, then evaluation occurs in two steps:

1. At the trigger the RHS of the nonblocking assignments are evaluated, but the updates are secheduled for "later" in a second phase, the "nonblocking assign update events".
2. In the nonblock assign update event the RHS values that were evaluated and remembered in step one are assigned to the LHS.

The specification [Ref] gives this example:

init begin
    a = 0;
    b = 1;
end

always @(posedge c) begin
    a <= b;
    b <= a;
end

As <= does not block, the only way these two statements could execute concurrently and not involve a race condition is by using the above strategy. That is why the output of the above is deterministic and after the first clock edge a will be 1 and b will be 0.

As another example, the following creates a race condition because the two always blocks execute simultaneously:

always @(posedge clk)
    a = b;

always @(posedge clk)
    b = a;

This is a race condition because both equations are scheduled to execute in the same simulation time step, and when blocking assignments are scheduled to execute in the same time step, the order execution is unknown [Ref].

A Verilog race condition occurs when two or more statements that are scheduled to execute in the same simulation time-step, would give different results when the order of statement execution is changed, as permitted by the IEEE Verilog Standard. [Ref]

However, if non-blocking assignment is used the race condition is removed.

always @(posedge clk)
    a <= b; // No race condition any more

always @(posedge clk)
    b <= a; // No race condition any more

This is because for non-blocking assignment, as we saw, on the rising edge the LHS of all of the <= expressions will be evaluated and the assigned in the next "phase" (but obs on the same rising edge).

A slightly more involved example showing the "why" in terms of how the hardware works:

Blocking v.s. Non-Blocking Example

Compare the following two code blocks, one using blocking and the other using non-blocking assignment to create different connections between two D-type flip flops. [Ref]. It demonstrates the difference between blocking and non blocking assignment nicely...

module block (
    input D, 
    input clk, 
    output Q1, 
    output Q2
);
    always @(posedge clk)
    begin
        Q1 = D;    // BLOCKING
        Q2 = Q1;   // STEP-BY-STEP EXECUTION
    end
endmodule
        

module block (
    input D, 
    input clk, 
    output Q1, 
    output Q2
);
    always @(posedge clk)
    begin
        Q1 <= D;    // NON-BLOCKING
        Q2 <= Q1;   // PARALLEL EXECUTION
    end
endmodule
        

This produces the following output:

                                    ____      ____      ____
                           CLK ____|    |____|    |____|    |_
                                   :         :         :
                                 _______     :  _________
                           D   _|  :    |______|       : |____
                                   :         :         :
                                   :         :         :
BLOCKING ASSIGNMENT                :                                     
                                   :         :         :                                _______
                                   :_________           ______           } D ------+---|D     Q|----Q1
                           Q1  ____|         |_________|                 }         |   |       |
                                   :         :         :                 }         |   |       |
                                    _________           ______           }         |   |__CLK__|
                           Q1  ____|         |_________|                 } CLK-+---|-------^     
                                   :         :         :                       |   |    _______
                                   :         :         :                       |   +---|D     Q|----Q2
                                   :         :         :                       |       |       |
                                   :         :         :                       |       |       |
                                   :         :         :                       |       |__CLK__|          
NON BLOCKING ASSIGNMENT            :         :         :                       +-----------^     
                                   :         :         :                             
                                   :_________           ______           }        _______   Q1  _______      
                           Q1  ____|         |_________|                 } D-----|D     Q|-----|D     Q|---- Q2
                                   :         :         :                 }       |       |     |       |
                                              _________                  }       |       |     |       |
                           Q1  ______________|         |______           }       |__CLK__|     |__CLK__|
                                                                           C---------^-------------^         

WARNING: The first example uses blocking assignment in an always proceedural block which may not be synthesiable - DigitalJS actually warns about this!

Seen on DigiJS:

Blocking v.s. Non-Blocking Guidelines

Taken verbatim from [Ref].

1. When modeling sequential logic, use nonblocking assignments.
2. When modeling latches, use nonblocking assignments.
3. When modeling combinational logic with an always block, use blocking assignments.
4. When modeling both sequential and combinational logic within the same always block, use nonblocking assignments.
5. Do not mix blocking and nonblocking assignments in the same always block.
6. Do not make assignments to the same variable from more than one always block.
7. Use $strobe to display values that have been assigned using nonblocking assignments.
8. Do not make assignments using #0 delays.

Continual Assignment

The keyword assign is used for continual assignment to a wire outside an always statement. The value of LHS is updated when RHS changes.

Continuous assignments drive nets and are evaluated and updated whenever an input operand changes value.

Assignments on nets shall be continuous and automatic. In other words, whenever an operand in the righthand expression changes value, the whole right-hand side shall be evaluated. If the new value is different from the previous value, then the new value shall be assigned to the left-hand side.

Continual assignment is done using the assign keyword and must be done outside a block. It is limited to basic boolean and ternary-if (bool ? do-if-true : do-if-false) operators.

Vectors

Wires and registers are, by default, scalar. I.e., they are 1-bit wide and can hold the numbers 0 or 1. To hold larger numbers a vector is required. Note, they can also hold z - high impedance and x - don't care.

Vectors are declared as type [upper:lower] name;, so for example:

wire [7:0] an_8_bit_wire;
reg  [7:0] an_8_bit_reg;

In verilog it is easy to mismatch vectors and silenty discard bits! So use a linter!, e.g. Verilator.

Arrays

Arrays are similar to C arrays - they are a repetition of a type. So,

reg my_reg1[0:11];      // is an array of 12 1-bit register variables
reg [7:0]my_reg2[0:11]; // is an array of 12 8-bit register variables - an array of vectors, known as a MEMORY ARRAY.

Verilog arrays can only be reference one element at a time.

Vectors and Arrays Layout

In Verilog, you don't usually have very much control over how things get laid out. The tools will usually infer whether, for example, an FPGA's block RAM will be used to implement something or logic units will be used instead. This a memory array (see above) may or may not get allocated to BRAM.

Always Blocks

See Tutorial – Sequential Code on your FPGA -- Using Process (in VHDL) or Always Block (in Verilog) with Clocks.

NOTE only registers (reg) can be used in always blocks and registers are a precious resource.

Not usually used for combinational logic as the same effect can be achieved with assign, outside of a block.

The NANDLAND author gives the following two, almost equivalent pieces of code:

always @ (*)
  begin
    and_gate <= input_1 & input_2;
  end

... and ...

// Taken from https://nandland.com/tutorial-sequential-code-on-your-fpga/
always @ (posedge i_clock)
  begin
    and_gate <= input_1 & input_2;
  end

They look almost identical, except that the first uses combintational logic and the latter sequential logic. Why? In the first and_gate is immediately updated whenever any of its inputs change. In the latter, and_gate is not immediately updated whenever an input changes... it has to wait for the next rising clock edge!

In traditional Verilog, procedural always blocks are used to model combinational, latch, and sequential logic. Synthesis and simulations tools have no way to know what type of logic an engineer intended to represent. Instead, these tools can only interpret the code within the procedural block, and then "infer", which just a nice way to say "guess", the engineer's intention. [Ref]

The author also gives the following example:

// Taken from https://nandland.com/tutorial-sequential-code-on-your-fpga/
reg test1 = 1'b1;
reg test2 = 1'b0;
reg test3 = 1'b0;
reg test4 = 1'b0;

always @ (posedge i_clock)
  begin
    test2 <= test1;
    test3 <= test2;
    test4 <= test3;
  end

On each rising clock edge all of the lines are executed in parallel -- at the same time.

The main point of this example is that although test4 may go high on the 3rd clock edge it is not valid until the fourth clock cycle due to propogation time. This is the case for all the clock cycles: testX will go high, but its not until the next clock cycle that it is valid and can be clocked by the next flip-flop as it propogates somewhere between the current and before the next clock edge.

This chimes well with an interesting post I found on reddit: When to use continuous vs procedural assignment?.

When using continuous asignments one has to think about propogation delay because it can restrict the clock frequency.

Its replicated here:

The guys code was:

...
assign next_test = ((a + b) > 4'hF) ? 4'h0 : (a + b);

always @(posedge clk)
begin
    if(next_test == 4'hF) begin
        test <= 4'h0;
    end 
    else begin
        test <= next_test
    end
end
...

The question is should next_test be assigned in the always block rather than continuously assigned.

One commenators has the following to say:

You can stack any number of continuous assignments on top of each other. But it means that the entire pipeline would have to fit the timing constraints. As it is coded, your software will try to get from the values of 'a' and 'b' to the value of 'test' in one clock. Along the way, it needs to calculate a+b, compare it with 0, do conditional assignment, etc., etc. That's a long path with multiple chained operations, each of which incurs some propagation delay, which brings down the maximum clock rate at which the design can run.

If you calculate 'next_test' procedurally, you need one clock to get from 'a' and 'b' to 'next_test' and a second clock to get from 'next_test' to 'test'. You use more registers and you have higher latency, but the maximum achievable clock rate is higher.

Incidentally, your code seems wrong. If a, b, and next_test are all 4-bit, I think that the result of ((a+b)>4'hF) is always false, because (a+b) is going to be 4-bit too. You'd need something like

wire [4:0] temp;
assign temp = a+b;
assign next_test=temp>5'hF ? 4'h0 : temp[3:0];

NANDLAND has an excellent article on the effects of propogation delay, which is well worth a read!

The longer the propagation delay, the slower your clock is able to run.

The reason for this is that both Flip-Flops use the same clock. The first Flip-Flop drives its output at clock edge 1. The second Flip-Flop does not see the change on the output of the first Flip-Flop until clock edge 2, at which point it drives its output. If the signal can safely travel from Flip-Flop 1 to Flip-Flop 2 in one clock period, your design is good! If not, you will run into problems. [Ref]

Also read about how setup and hold time relate to propogation delay:

Setup time is the amount of time required for the input to a Flip-Flop to be stable before a clock edge. Hold time is similar to setup time, but it deals with events after a clock edge occurs. Hold time is the minimum amount of time required for the input to a Flip-Flop to be stable after a clock edge. [Ref]

Another thing to be careful about is NOT CREATING LATCHES in always blocks due to incomplete assignment. See this NANDLAND article on how to avoid creatign latches.

Scope

The always block does not have scope like one might expect in an actual software (i.e., hot hardware) language like C. And because all of the always blocks in a module (program?) execute CONCURRENTLY, the following should throw an error because var can't get two values concurrently!

always @(posedge clk)
    ...
    var1 = ....;
    ...
end

...

always @(posedge clk)
    ...
    var1 = ....;
    ...
end

Compiler Directives.

Directives are compile time macros that execute during compilation.

All compiler directives begin with a ` and take the format ` ``<keyword> ....

• ` ``define <macro-name> <value>: Defines a text macro that is substitutes the text of the macro when ever it enounters the macro name. This is analogous to the C #define.
• ` ``include <filename>: Just like C's #include.
• ` ``timescale <time_unit> / <time_precision>:
• ` ``ifdef <define_name>:
• ` ``else:

Behavioural Modeling

Keywords include if-else, case, which are synthesizable statements, and while, for, forever, repeat, fork-join, which are not syntheziable and are only really used in the test bench.

Assignment statements such as = and <= should only be written within a procedural block: initial or always. Writing assignment statements outside of a precedural block is an error.

The initial block is only executed once and starts at time 0. Useful, for example, initialising variables. It may be syntehsizeable. As my colleage said reply to my question:

Q: The initial block - is that simulation only?

A: Not necessarily (some synthesis tools may use initial conditions to set the state of the FPGA when it leaves it's configuration phase - personally, I set that explicitly and only use initial statements for simulation).

The always block is repeatedly executed in a never ending loop and starts after the intial block.

module module_name(...);

    initial
    begin
        ...
    end

    always
    begin
        ...
    end

endmodule

Parameterized Verilog Modules

Instantiation parameters can be added to modules to make them analogous to C++ templates. This is done with the pound (#) symnbol.

module my_module #(paremeter WIDTH) (
    input  [WIDTH-1:0] A,
    output [WIDTH-1:0] A,
);
    ... do stuff ...
endmodule

The above decalares what is like a module template... when instantiated with a concrete value for WIDTH, a module instance with WIDTH replaced by its value is created:

my_module #( .WIDTH(8) ) instance_name (
    .input(...),
    .output(...)
);

Testbench

See ...

• ECE 128 – Verilog Tutorial: Practical Coding Style for Writing Testbenches
• Assert statement in Verilog, SO.
• Using Verilog for Testbenches, S Capkun et al.
• Writing Efficient Testbenches, M Hamid

These are used to check Verilog designs are working as expected and consist on non-synthesizable verilog code which generates inputs to the design and checks that the outputs are correct [Ref].

Simulation tools show the output waveforms. Open source alternatives exist like Icarus Verilog with GTKWave.

To install on Ubuntu (not latest version):

sudo apt update
sudo apt install iverilog
sudo apt install gtkwave

To get latest release, download from GitHub and untar source.

Then

sudo apt update
sudo apt install -y autoconf gperf make gcc g++ bison flex
tar xzvf iverilog-12_0.tar.gz
cd iverilog-12_0/
sh autoconf.sh
./configure
make
make check
sudo make install  ##< Installs to /usr/local

Time

Testbench code is not synthesized: so need "special" constructs to create delays: the pound character followed by a number of time units to model delays. The unit of time is set by the timescale` directive which we have seen before:timescale / `:

`timescale 1 ns/1 ps
module smallTest ();
    ...    
    #10.5 a = 1'b1; //< `a` is set to 1 after 10.5 units of time...
                    //< ...so in this case, after 10.5 nanoseconds 

The resolution determines the smallest step we can make in a delay. The delay is also relative to the moment the line is encountered... it is not an absolute time. Therefore:

`timescale 1 ns/1 ps
...
#10 a = 1'b1; // At 10ns
#10 a = 1'b0; // At 20ns
#10 a = 1'b1; // At 30ns
#10 a = 1'b0; // At 40ns

This is also equivalent to:

`timescale 1 ns/1 ps
...
#10
a = 1'b1; // At 10ns
#10
a = 1'b0; // At 20ns
#10
a = 1'b1; // At 30ns
#10
a = 1'b0; // At 40ns

Note the following:

The most fundamental non-synthesizable piece of code is a delay statement. The FPGA has no concept of time, so it is impossible to tell the FPGA to wait for 10 nanoseconds. [Ref]

Generating Clocks

// Declare a clock period constant.
Parameter ClockPeriod = 10;

// Clock Generation method 1:
initial begin
    forever Clock = #(ClockPeriod / 2) ~ Clock;
end

// Clock Generation method 2:
initial begin
    always #(ClockPeriod / 2) Clock = ~Clock;
end

Forever Loops

Is an infinite loop. Useful for generating clock signals in test benches. The following generates a clock with a period of 2 nanoseconds:

// See https://fpgatutorial.com/how-to-write-a-basic-verilog-testbench/
`timescale 1 ns/1 ps
initial begin
   clk = 1'b0;
   forever begin
     #1 clk = ~clk;
   end
end

Loops must be written inside proceedural or generate blocks.

Repeating Yourself

To repeat something a number of times use repeat(<n>) statement; as a one liner, or repeat(<n>) begin .... end over multiple lines.

The code below toggles the clock 4 times, with a delay of 10 time units between each toggle, before lowering rst and generating a clock that runs forever.

// Taken from https://alchitry.com/writing-test-benches-verilog
initial begin
  clk = 1'b0;
  rst = 1'b1;
  repeat(4) #10 clk = ~clk;
  rst = 1'b0;
  forever #10 clk = ~clk; // generate a clock
end

Wait For A Clock Edge

The testbench can wait for an edge of a signal using @(negedge signal-name) or @(posedge signal-name). This can be combined with repeat(<n>) to wait for n clock edges, for example repeat(10) @(posedge clk) waits for 10 rising clock edges before continuing.

Reading In Test Vector Files

Great for automation of test benches. Use $readmemb() (binary), $readmemh() (hexadecimal) etc.

The idea is that rather than use a bag of simulation variables, a vector is used instead and you can compare conditions to those vectors step by step... or just use them as plain old vectors... either way if you need a list of them, they can be read in from a file.

Lets say we have a test vector file, test.tv with the following contents, and we are testing a simple AND gate:

00_0
01_0
10_0
11_1

This can be read in an an initial block like so:

reg [3:0] testvectors[3:0];

initial
    begin
        $readmemb("example.tv", testvectors);
        ...
    end

The we could change the inputs to the gate under test just after every clock edge and check the results on the falling clock edge.

always @(posedge clk)
    begin
        // Apply inputs with some delay (1ns) AFTER clock
        // WARNING: This is important as inputs should not change at the same time with clock
        #1; {a, b, c, yexpected} = testvectors[vectornum];

always @(negedge clk)
    begin
    if (y !== yexpected)
    begin
        $display("Error: inputs = %b", {a, b, c});
        $display(" outputs = %b (%b exp)",y,yexpected);
        errors = errors + 1;
    end
    ...

    vectornum = vectornum + 1;
    if (testvectors[vectornum] === 4'bx)
    begin
        $display("%d tests completed with %d errors", vectornum, errors);
        $finish;
    end
end

This is pretty good, but not brute force testing for components when the number of possible input and output combinations is large will not be possible - will take too much time. At that point it's down to selective testing, formal verification etc.

Verilog System Tasks/Functions

Always begin with a $ and are built-in, ready to be used. Again, not synthesizable, only for testbench use.

// General syntax
$display(, 
            

// General syntax
$monitor(, );

// Example - monitor the values of the in_a and in_b signals
$monitor("in_a=%b, in_b=%b\n", in_a, in_b);

$display("Current simulation time = %t", $time);

Structure

module my_testbench(); // Create a verilog module with no inputs or outputs, which will
                       // be used to instantiate the DUT so that signals can be
                       // connected to it to provide stimuli to test it with.

    // Declare wires and registers
    wire a;
    wire b; // etc etc...

    // Instantiate the device under test (DUT)
    my_device dut (....);

    // First initial block should start setting the initial conditions for your test
    // bench, setup $monitors etc
    initial begin
       $monitor ("TIME = %d, test_var1= %b, test_var2= %b", $time, test_var1, test_var2);
        a = 1'b0;
        b = 1'b1; // etc etc...
    end

    // Second initial block can be used to do limit the run time of the simulation
    initial begin
        #finishtime // everything below will printout after "finishtime" expires
        $display ("Finishing simulation due to simulation constraint.");
        $display ("Time is - %d",$time);
        $finish;
    end

    // Third initial block can do your testing
    initial begin
        ...
    end

    // Last initial block can open files to save results to a Value Change Dump (VCD) File
    initial begin
        $dumpfile("path/to/dumpfile.vcd");
        $dumpvars;
    end

To make an assert-style statement something like the following can be used [Ref]:

`define assert(signal, value) \
        if (signal !== value) begin \
            $display("ASSERTION FAILED in %m: signal != value"); \
            $finish; \
        end

...
...

initial begin // assertions
    #32 `assert(q, 16'hF0CB)
end

NOTE, in Icarus the define must be done on one line, without using \, otherwise line numbers will be incorrect!