Only Ubuntu and MacOS are known to work, but these instructions are likely to work other platforms. Please let us know if you successfully build on a different platform, so we can improve these instructions.
Attention: Please only run these instructions on your own machine. If you are building on Grace 2, please skip forward to Download the Source Code.
Starting with a fresh install of Ubuntu 16.04 is recommended.
sudo apt update && sudo apt upgrade
sudo apt install cmake git
Install Ninja (optional)
It is recommended to build LLVM using Ninja to avoid running out of memory during linking. Using Ninja should also result in faster builds.
wget -q https://github.com/ninja-build/ninja/releases/download/v1.9.0/ninja-linux.zip && unzip -q ninja-linux.zip && sudo cp ninja /usr/bin && rm ninja ninja-linux.zip
apt install ninja
Proceed to Download the Source Code.
Open Terminal and run
xcode-select --install
and select either Install or Get Xcode, if you want to install Xcode and have not already.
Install Homebrew (optional)
Homebrew is a package manager for MacOS - it provides a simple way to install and manage software on your Mac.
To install homebrew, open Terminal and run
/usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
Install CMake
brew install cmake
Visit the CMake downloads page, and download the file ending in Darwin-x86_64.dmg.
Open the downloaded image, and drag the CMake application into the Applications folder.
Install Ninja (optional)
It is recommended to build LLVM using Ninja to avoid running out of memory during linking. Using Ninja should also result in faster builds.
brew install ninja
Download and install Ninja 1.9:
wget -q https://github.com/ninja-build/ninja/releases/download/v1.9.0/ninja-mac.zip && unzip -q ninja-mac.zip && sudo cp ninja /usr/bin && rm ninja ninja-mac.zip
Install Xcode (optional)
If you would like to use Xcode to build LLVM, and do not already have it installed, go to the Mac App Store,
search for Xcode, and click Get.
In this method, LLVM is built separately, with its location passed to OptSched. This can help reduce rebuilds by restricting it to rebuilding OptSched only, rather than all of LLVM. Additionally, it allows you to do things such as build OptSched with debug info, but leave off LLVM's debug info.
See the Manual Build on downloading LLVM and building clang and llvm. Follow those steps, except do not clone OptSched inside of LLVM. Simply build LLVM just with the spilling info patch.
Additionally, add a -DCMAKE_INSTALL_PREFIX=/path/to/install/llvm-release to the cmake ... command,
and also run ninja install or make install after building.
Remember that CMAKE_INSTALL_PREFIX; it will be passed to CMake when building OptSched itself.
Clone OptSched to a fresh directory, and make a build directory for OptSched:
git clone https://github.com/CSUS-LLVM/OptSched
cd OptSched
mkdir build && cd build
Run CMake:
cmake .. -GNinja -DCMAKE_BUILD_TYPE=Debug -DLLVM_DIR=</path/to/install/llvm-release>/lib/cmake/llvm -DOPTSCHED_ENABLE_AMDGPU=OFF
Note: if you use -DCMAKE_BUILD_TYPE=Debug, make sure that you built LLVM with assertions in the previous step.
If you did not build LLVM with assertions, use -DCMAKE_BUILD_TYPE=Release.
Build:
ninja
# Or if you left off -GNinja, use Make:
make -j6 # or however many CPUs you have
To let a script manage cloning and installing all dependencies, placing OptSched inside llvm for you.
1. Create a build directory.
mkdir build && cd build
2. Configure
We want to configure against the OptSched/cmake/superbuild directory.
Use the generator you want, be it Ninja with -GNinja, explicitly specified makefiles with -G'Unix Makefilex',
or something else.
If you have ccache installed, consider adding -DCMAKE_CXX_COMPILER_LAUNCHER=ccache.
This will speed up subsequent builds. If you do so, be sure to disable the ccache hash_dir setting.
To build OptSched inside LLVM:
cmake -GNinja -DCMAKE_BUILD_TYPE=Debug -DLLVM_PARALLEL_LINK_JOBS=1 ../cmake/superbuild
Note: In debug builds, linking uses a lot of memory. Set LLVM_PARALLEL_LINK_JOBS=2 if you have >= 32G memory, otherwise use LLVM_PARALLEL_LINK_JOBS=1.
If you also wish to build flang, add -DOPTSCHEDSUPER_FLANG=ON.
The flang compiler cannot be built with ninja, so if you are using Ninja, add -DOPTSCHEDSUPER_FLANG_COMPILER_CMAKE_GENERATOR='Unix Makefiles'
Complete command for building with flang:
cmake -GNinja -DCMAKE_BUILD_TYPE=Debug -DLLVM_PARALLEL_LINK_JOBS=1 -DOPTSCHEDSUPER_FLANG_COMPILER_CMAKE_GENERATOR='Unix Makefiles' -DOPTSCHEDSUPER_FLANG=ON ../cmake/superbuild
3. Run the Super-build
Invoke the generator you chose.
Ninja:
ninja
Make:
make # Consider adding -jN where N is the number of parallel compile processes you want
Generic:
cmake --build .
The CMake super-build script will clone, configure, and build llvm-project along with flang if specified.
The main build directory for LLVM, where unit tests can be run, is llvm-prefix/src/llvm-build.
The flang binaries and libraries will be installed to flang-install inside your build directory.
The llvm binaries and libraries, including OptSched.so, will be installed to llvm-install inside your build directory.
These directories may be changed at the configure step by specifying -DOPTSCHEDSUPER_<Type>_INSTALL_PREFIX=/path/to/install/dir,
where <Type> is either LLVM or FLANG.
To manually build this, such as if you want to place OptSched inside an existing clone of llvm.
1. Clone the LLVM source code from GitHub.
git clone https://github.com/llvm/llvm-project
2. Checkout LLVM release 6.
cd llvm-project && git checkout release/6.x
3. Clone OptSched into the projects directory.
cd llvm/projects && git clone https://github.com/CSUS-LLVM/OptSched
4. Create a build directory.
mkdir build && cd build
5. Apply this patch to print spilling info.
git am ../OptSched/patches/llvm6.0/llvm6-print-spilling-info.patch
These instructions follow after Download the Source Code, and so assume that you are in the llvm-project/llvm/projects/build directory.
Using Ninja (recommended)
Note: In debug builds, linking uses a lot of memory. Set LLVM_PARALLEL_LINK_JOBS=2 if you have >= 32G memory, otherwise use LLVM_PARALLEL_LINK_JOBS=1.
cmake -GNinja -DLLVM_PARALLEL_LINK_JOBS=1 -DLLVM_ENABLE_PROJECTS='clang' -DCMAKE_BUILD_TYPE=Debug '-DLLVM_TARGETS_TO_BUILD=X86' -DLLVM_BUILD_TOOLS=ON -DLLVM_INCLUDE_TESTS=ON -DLLVM_OPTIMIZED_TABLEGEN=ON ../.. -DOPTSCHED_ENABLE_AMDGPU=OFF
ninja
Using Make
Note: Debug builds use a lot of memory. The build will fail if you do not have enough. If this happens, try using Ninja to build.
cmake -DLLVM_ENABLE_PROJECTS='clang' -DCMAKE_BUILD_TYPE=Debug '-DLLVM_TARGETS_TO_BUILD=X86' -DLLVM_BUILD_TOOLS=ON -DLLVM_INCLUDE_TESTS=ON -DLLVM_OPTIMIZED_TABLEGEN=ON ../.. -DOPTSCHED_ENABLE_AMDGPU=OFF
make
A Debug build of LLVM on a single thread will take a long time.
See Building with CMake for more build options.
These instructions follow after Download the Source Code, and so assume that you are in the llvm-project/llvm/projects directory.
1. Build an Xcode project
cmake -G Xcode -DLLVM_ENABLE_PROJECTS='clang' -DCMAKE_BUILD_TYPE=Debug '-DLLVM_TARGETS_TO_BUILD=X86' -DLLVM_BUILD_TOOLS=ON -DLLVM_INCLUDE_TESTS=ON -DLLVM_OPTIMIZED_TABLEGEN=ON ../.. -DOPTSCHED_ENABLE_AMDGPU=OFF
This will create an Xcode project in llvm-project/llvm/projects/build.
2. Open the project in Xcode
Open Xcode, and go to File > Open... (or press Cmd+O).
Navigate to llvm-projects/llvm/projects/build and click Open.
3. Create clang and OptSched schemes
Upon opening the project, Xcode will prompt you to create schemes. We want to create them manually - this will prevent you from having to sort through tons of build targets later.
Press the + button and select clang from the long Targets dropdown list.
Repeat this step, selecting OptSched from the list.
You should now have two schemes - clang and OptSched. Press Close.
4. Change the language standard to C++14
In the file navigation window on the left, scroll to the very top and select the LLVM
project file, under which all other files should be listed. It has a blue icon.
Just to the right of the file navigator there will be a list of all build targets.
Select OptSched from this list.
Click on Build Settings at the top of the page, and scroll down to the
Apple Clang - Custom Compiler Flags section.
Under Other C++ Flags, double-click on the long list of compiler flags to the right of Debug.
From the text box that pops up, double-click on -std=c++11 and change it to -std=c++14.
Note: If you want to compile in Release mode, or any of the others, you will have to make the same change for that mode.
5. Build clang and OptSched
At the top left, there is a "Run" (►) button and a "Stop" (■) button. To the right of those is where you can chose your scheme.
Select the clang scheme and then select Product > Build (or press Cmd + B).
Wait for the build to complete, and repeat this with the OptSched scheme.
Super-build
Run ctest with make test (or ninja test) or by running ctest directly with ctest . .
This will run the unit tests and it will do a test run of the compiler with OptSched enabled.
Unit Tests
Invoke the check-optsched-unit target of the build system generated by CMake.
For make, that is make check-optsched-unit. In general, that is cmake --build . --target check-optsched-unit
If you used the superbuild, you need to be in the llvm-prefix/src/llvm-build directory first.
Command Line Build
From the llvm-project/llvm/projects/build directory, run:
echo 'int main(){};' | ./bin/clang -xc - -O3 -fplugin=lib/OptSched.so -mllvm -misched=optsched -mllvm -enable-misched -mllvm -optsched-cfg=../OptSched/example/optsched-cfg -mllvm -debug-only=optsched
MacOS Xcode Build
From the llvm-project/llvm/projects/build directory, run:
echo 'int main(){};' | Debug/bin/clang -xc - -O3 -fplugin=lib/OptSched.so -mllvm -misched=optsched -mllvm -enable-misched -mllvm -optsched-cfg=../OptSched/example/optsched-cfg -mllvm -debug-only=optsched
CCache
Are your compile times slow because you keep finding you have to rebuild files (e.g. all of LLVM)?
Use ccache. To do this, install ccache (sudo apt-get install ccache) and add the following arguments to your cmake command:
-DCMAKE_CXX_COMPILER_LAUNCHER=ccache -DCMAKE_C_COMPILER_LAUNCHER=ccache
Also, you probably want to configure ccache to not hash the directory: ccache -o hash_dir=false.
With CCache enabled, if the compiler would be building a file with no changes from before, it will instead use the cached file.