In the first post I introduced the Uranium project, the motivations around it and some basic explanations around compilers. So far so good, no radiation in the air.

As a first milestone, I set the generation of some basic IR code that I can compile using LLVM. For both generating the IR code and compiling the code we are going to use LLVM. It’s not really needed for the generation phase, but LLVM has some C++ classes that assists you in the process. Instead of writing the IR code by myself, I can use the LLVM libraries to generate a collection of objects that I can easily dump in IR code.

1. Compiling and installing LLVM and Clang

Obviously, before using LLVM and Clang I needed to install them. If you don’t want to build your own binaries, you can use a pre-build distribution if your platform is among those available. I’m using Linux Mint that is based on Ubuntu, so I suppose that there are pre-build binaries for my Linux distribution. Anyway, I didn’t try to install them but I know it’s an option. I’m trying to build a compiler, so I prefer to compile the compiler that is going to help me to compile my compiler ^.^

Compiling LLVM and Clang was easier than I expected. The most difficult part was waiting until the end of the compilation. You only need patience (and CMake and a modern C++ compiler :).

The documentation specifies the requirements and the build and installation process quite well. Using CMake for building the libraries is the easier way. I know it is well described there, but anyways, I’m going to post here the process with the exact commands I ran.

This is the LLVM repository. You can use git clone or download a package with an specific LLVM version. For example, for installing the version 10.0.0 using git your should execute:

The version 10.0.0 was the last stable version at the time of writing. You can explore the existing tags in the repository with git tag -l. Choose the tag that correspond to the version you want to install.

Inside the llvm-project folder run:

And now you need to generate the Makefiles using CMake. We can pass several variables to CMake to change the generated build files. Some of the options I used were:

• CMAKE_INSTALL_PREFIX: Where the libraries are going to be installed. By default they are installed in /usr/local but thinking about the future I preferred to create a separate directory for each version I had compile. I set this variable to /usr/local/llvm10.
• As I said in the previous post, clang is the C++ compiler we are going to use. Apart from compile our compiler, clang is going to be very useful for comparing the generated IR code of our language to the one generated by an equivalent program written in C/C++. We’ll use that as proof that we’re generating proper code. To indicate that we want to build clang we need to add the project name to the variable LLVM_ENABLE_PROJECTS.

The line I ran was:

You can also compile clang later running again the cmake command with DLLVM_ENABLE_PROJECTS. You don’t even need the repeat the CMAKE_INSTALL_PREFIX variable if you did it before:

Compile the code just with a simple make command. The compilation process for LLVM + clang takes around 4 hours in my computer.

Once compiled, install the binaries. If your install directory does not exists, you need to create it first.

1.1. Build documentation

If you want to build the documentation you can re-execute the cmake with the LLVM_ENABLE_DOXYGEN variable active. I tried to build the documentation with Sphynx but I was getting an “Warning treated as error” that didn’t allow me to finish the docs generation. Doxygen is working well for me:

Your documentation is ready in build/docs/doxygen/html/index.html. If you install it, you can also find it in your install directory, <install-dir>/docs/html/html/index.html. For example, in my case it is in /usr/local/llvm10/docs/html/html/index.html. Keep in mind that this documentation is only for the LLVM libraries and does not include clang documentation or any other extra tutorials.

I tried to install the clang documentation after installing the LLVM documentation and it turns out that the clang documentation overwrites the LLVM documentation. Maybe I did something wrong, but I didn’t invest time on solving the problem. Why? Read the next paragraph.

If you do not want to generate the documentation you can use the online version. Just be sure that you are using the documentation of the installed version of LLVM to avoid confusions. Search for the appropriate version of your documentation here. For example, for the version 10.0.0 that I installed I need to use the given link.

2. Installing Flex and Bison

This step in Ubuntu is pretty easy:

If you want to compile from source or install them in a different system you will need to do a web search.

3. Check the installation

If you want to use all the LLVM tools easily, add the bin directory under your install folder to the PATH environment variable. Add the following line to your ~/.bashrc file:

Using this approach you can easily select the version of LLVM you can use in your projects. Just change the value of the PATH variable.

Restart your Shell and execute clang --version and later lli --version. The system should find the existing executable files and show the next outputs.

clang --version:

lli --version:

Check also the Flex and Bison installation.

flex --version:

bison --version:

4. Next step

After compiling, installing and checking that the binaries are in the correct path, we are going to compile our first program using the LLVM libraries. For doing that we need a proper way to build our code. Like LLVM, I’m going to use CMake for generating a Makefile. See you in the next post!!

5. References

Apart from the given links, no more references are been used for this post. Here are all the links used in this page:

To make it easier, I’ve created an index page with all the Uranium posts.