Limerick(6) - SpinalHDL Tutorial: the Simulation Flow of Hello World
Hello, everyone. Welcome to the sixth episode of FPGA
limerick. In this episode, we will take a close look at the simulation flow of
SpinalHDL. And we will use the Hello World as an example for demonstration purpose.
Generally speaking, when it comes to digital circuits, there are two approaches for verification.
1. Formal Verification
The formal verification is more like a math approach, because basically every digital circuit is nothing but a mapping from input to output, which mathematically fits the definition of Function.
In that sense, the DUT is treated as a math function, whose property is described by SVA (System Verilog Assertion), as illustrated below:
As it is like a math approach, there is no need to prepare test vectors for input and output. And the formal verification tool will use analytic method to check the DUT against those SVAs, and to make sure those properties are satisfied in all conditions.
SpinalHDL also supports Formal Verification, which can be found in
Because Formal Verification is an analytic method, no simulation is needed. We will just leave the rest of it to inquisitive viewers.
2. Functional Simulation
This is the main topic we will discuss today. And for functional simulation, there are also two approaches:
a) static approach, for which we will prepare the expected input/output before hand and make them as test vectors. After the simulation run, the simulation output is matched against the test vectors to verify the DUT, as illustrated below
b) dynamic approach, for which the input is randomized. And the simulation output is compared against a functional model to verify the DUT, as illustrated below:
In the figure above, the driver will generate the random inputs, and feed them to both the simulator and the functional model. And the monitor will capture the simulation outputs, and feed them into the checker to match against the functional model.
Simulation Flow for SpinalHDL
For SpinalHDL, it can support multiple simulators as its backend. And the fastest one will be verilator.
As illustrated in the figure above, the DUT will be compiled into verilog and passed onto the verilator. The verilator will turn them into C++ code and use GCC to make a shared object. The test bench in scala, together with shared object, will finally produce the waveform dump.
Currently the waveform dump can be in either VCD format or FST format. Comparing to VCD (a text format), the FST is a binary format and is a lot smaller than the VCD. So we would recommend using the FST format. Both VCD and FST can be viewed by GTKWave (https://gtkwave.sourceforge.net/)
Now let’s take a look at the test bench in Hello World example
First of all, please note that although SpinalHDL is built on top of Scala language, the test bench is 100% in vanilla Scala. In other words, the test bench is not in SpinalHDL at all. So when we try to get and compare the value from DUT, every data type originally in SpinalHDL has to be converted into Scala type. For example, the IO port could be “bits” in SpinalHDL. But when comparing it to the test vector, we need to use .toInt or .toLong to convert it to a data type understandable by vanilla Scala. Another example is the Boolean type. In SpinalHDL, the Boolean type is actually spelled as Bool, and the True/False literature is spelled with the first level as capital letter. But for vanilla scala language, the Boolean type is spelled as Boolean, with the true/false literature spelled in all small letter. So when we need to use .toBoolean to convert the Boolean type in SpinalHDL to the one in vanilla Scala.
Also, when we drive the DUT’s input ports, we also need to use data type from vanilla Scala. For the Boolean type, it means we should use “true/false” instead of “True/False”, and the syntax would be like
dut.io.srst #= false
Boilerplate Code for Test Bench
As we can see from the Hello World example, the Scala testbench is pretty much in boilerplate form. And we need the following method call:
For the validation of the simulation, this SpinalConfig should match the one used by the DUT generation.
.withWave / .withFstWave / .withVcdWave
If waveform dump is needed, please use those methods to specify the dump format. As we will use verilator as our backend simulator, using .withFstWave is recommended.
Instantiate the DUT in this method
Now this is the main body of the simulation, which we will elaborate next
The Hello World example we use here demonstrates the static approach (test vector matching) for Functional Simulation. (And we will demonstrate the dynamic in later episodes when we talk about cocotb)
The main body of the simulation is contained in the .doSim method in the testbench. It basically has three threads:
And they are forked when they are created, and then joined afterwards. For more complicated designs, more than one compare thread could be launched to match multiple test vectors.
The measure thread is not typical in the verification. It is just included in the Hello World as a way to sanity check the verification.
The clkThread is used to initialized the test, and keep track of the start/end of the test with accurate clock count.
To View the Wave
After the simulation, if the waveform is dumped, we can use GTKWave to view the wave. On Windows, the GTKWave can be installed in MSYS2. Please open MSYS2 MINGW64, and type in
pacman -S mingw-w64-x86_64-gtkwave
And then open a command prompt and type in “gtkwave” to launch the waveviewer. For the Hello World example, the wave dump (.fst or .vcd) can be found as .simWorkspace/NcoCounter/test.fst
Limerick(5) - SpinalHDL Tutorial: the Design Flow of Hello World
Hello, everyone! Welcome to the fifth episode of FPGA
limerick. In this episode, we will take a close look at the “Hello World”
example for SpinalHDL. And based on that, we will also demonstrate the basic
design flow for SpinalHDL.
As mentioned in previous episodes, the Hello World can be
checked out like
And for a fresh checkout, it has the following files:
*) Files at the top level
configuration file for SBG
configuration file for scalafix (linter)
configuration file for scalastyle (linter)
configuration file for scalafmt (formatter)
project source for SpinalHDL
configuration file for log4j2
test/spinal: test bench in
test bench using cocotb
additional configuration files for SBT
And the HelloWorld can also serve as a template for other
projects. You can find a script called “make_new_prj.sh” under the scripts
folder. And it can be used to create a new project folder using HelloWorld as
the template. (Please run it under WSL.)
The design flow of SpinalHDL
As illustrated above, the general design flow for SpinalHDL is to: 1. Setup the Parameters of the SpinalHDL top module 2. Feed the source code into SpinalConfig 3. Generate the output in Verilog, System Verilog or VHDL
And now let's take a close look to the HelloWorld example to see how the flow works.
The background of the HelloWorld Example: NCO Counter
HelloWorld is actually an example of NCO Counter. NCO stands for
Numerically Controlled Oscillator. It tries to output a new clock (an
enable pulse train actually) from on a base clock. The frequency of the
output clock and the frequency of the base clock are predefined by
And mathematically, we need to do the following: 1. Find the GCD of the base clock and the output clock 2. Use the GCD to simplify the ratio between output clock and the base clock, i.e:
3. Once we get the simplified ratio of n over m, we can set the NCO counter like the following: a) Set the counter value to be zero (The initial phase) b) For every base clock cycle, counter = (counter + n) mod m c) Generate a pulse (output clock) every time the counter rolls over m
The code structure of the SpinalHDL module
Now let’s take a look at the code structure of the HelloWorld Example:
following the naming convention of JAVA package, the package name is
often the reverse form of a URL. But you are also free to choose other
name formats as you see fit. For example, in our FpgaLimerick repo, all
the common code is placed in a package called “common”
This is like the include section for Verilog or C++. And you can use scalafix to find and remove the unused imports.
3. case class extends Component
is similar to the “module” in Verilog. The SpinalHDL coding convention
recommends using “case class” instead of “class” in scala.
The parameters of the module. Here we use the convention for parameters
as all capital letter with G_ prefix. (G stands for generics, a
reminiscent from VHDL)
5. A bundle for port list
6. The main body of the design (combinatorial logic and sequential logic)
And a companion object that contains the main function, which will
generate the verilog / system verilog / VHDL based on the configuration.
comparing to verilog, it is a cinch for SpinalHDL to get GCD value, as
SpinalHDL is based on a high level programming language scala. If we
want to do the same implementation in verilog, we have to jump through a
lot of hoops to get the GCD value, as verilog’s system functions are
very primitive and do not cover GCD directly. Even if you can find a way
to calculate the GCD with verilog’s system functions, I bet the code
would be bulky and messy.
Configuration and Code Generation
Hello World uses the MainConfig to generate the code. The MainConfig
can be found in common/MainConfig.scala, which is derived from
SpinalConfig (in Spinal.scala from spinalhdl-core)
And in the MainConfig, the targetDirectory to be “gen”, and it sets the async reset to be active low.
From the Hello World example, it shows that the SpinalHDL is much more powerful and flexible for design parameterization.
Limerick(4) - Linter, Formatter, Logging and Editor
Hello, everyone. Welcome to the fourth episode of FPGA
limerick. In this episode, we will talk about the linter, formatter, logging
and editor for SpinalHDL. However, as we all know, the concepts like linter,
formatter etc. are universal. They are not only confined to SpinalHDL. So we
will expand the scope to other languages when it is necessary.
Linter for C/C++
Maybe we can start with a more popular language like C/C++.
An often-used linter for C/C++ is clang-tidy. On Ubuntu, (which is the Linux
distro we choose for WSL), it can be installed like
sudo apt install clang-tidy
And you can use the following to check a C/C++ source file:
clang-tidy -checks=* file_name.cpp
And you can use the following inline comment to ignore a
certain warning, like