This article explains how you can build Fastcomp’s sources using a fully manual process.
Fastcomp is delivered pre-built as part of the SDK on most platforms, but can also be built from source from within the SDK.
If you are building a large project, you will need to create a 64-bit build of LLVM and Clang. Compiling and optimizing can take more memory than is available to the 32-bit build.
First verify you have the compiler toolchain for your platform, download and install it as necessary.
Then follow the instructions for your platform showing how to manually build Emscripten from source. This page helps you with the specific task of building fastcomp (LLVM + clang).
To build the Fastcomp code from source:
Create a directory to store the build. It doesn’t matter where, because Emscripten gets the information from the compiler configuration file (~/.emscripten). We show how to update this file later in these instructions:
mkdir myfastcomp cd myfastcomp
Clone the fastcomp LLVM repository (https://github.com/kripken/emscripten-fastcomp):
git clone https://github.com/kripken/emscripten-fastcomp
Clone the kripken/emscripten-fastcomp-clang repository into emscripten-fastcomp/tools/clang:
cd emscripten-fastcomp git clone https://github.com/kripken/emscripten-fastcomp-clang tools/clang
You must clone it into a directory named clang as shown, so that Clang is present in tools/clang!
Create a build directory (inside the emscripten-fastcomp directory) and then navigate into it:
mkdir build cd build
Configure the build using cmake:
cmake .. -DCMAKE_BUILD_TYPE=Release -DLLVM_TARGETS_TO_BUILD="host;JSBackend" -DLLVM_INCLUDE_EXAMPLES=OFF -DLLVM_INCLUDE_TESTS=OFF -DCLANG_INCLUDE_TESTS=OFF
On Windows you will need Visual Studio 2015 or newer to build.
Determine the number of available cores on your system (Emscripten can run many operations in parallel, so using more cores may have a significant impact on compilation time):
- On Mac OS X you can get the number of cores using: Apple menu | About this mac | More info | System report. The Hardware overview on the resulting dialog includes a Total number of cores entry.
- On Linux you can find the number of cores by entering the following command on the terminal:
cat /proc/cpuinfo | grep "^cpu cores" | uniq.
- On Windows the number of cores is listed on the Task Manager | Performance Tab. You can open the Task Manager by entering Ctrl + Shift + Esc from the Desktop.
Call make to build the sources, specifying the number of available cores:
If the build completes successfully, clang, clang++, and a number of other files will be created in the release directory (<LLVM root>/build/Release/bin).
The final step is to update the ~/.emscripten file, specifying the location of fastcomp in the
If you’re building the whole of Emscripten from source, following the platform-specific instructions in Building Emscripten from Source, you won’t yet have Emscripten installed. In this case, skip this step and return to those instructions.
If you already have an Emscripten environment (for example if you’re building Fastcomp using the SDK), then set
LLVM_ROOT to the location of the clang binary under the build directory. This will be something like <LLVM root>/build/Release/bin or <LLVM root>/build/bin:
You should use the same branch (incoming, or master) for building all three repositories:
Mixing incoming and master branches may result in errors when building the three repositories.
emcc -v to check if the branches are synchronized.
emcc -v checks the code in the repositories, not the builds. Before building make sure that you fetch the latest changes to LLVM and Clang.
Bisecting across multiple git trees can be hard. We use version numbers to help synchronize points between them:
Version numbers are typically
Xis a major number (changes very rarely).
Yis a release number (changes each time we merge incoming to master, so these numbers indicate points where all tests passed).
Zis minor update that is just a sync point between the repos, or is needed when libc changes in emscripten (version changes clear the cache).