Build Instructions

libnomp

Clone libnomp repo and change directory to libnomp:

git clone https://github.com/nomp-org/libnomp.git
cd libnomp

Install dependencies using conda to build and run libnomp:

conda env create -f environment.yml
conda activate libnomp

You can use mamba to install the dependencies faster.

mamba env create -f environment.yml
mamba activate libnomp

If you are planning on contributing to libnomp, install the dev dependencies instead:

conda env create -f environment-dev.yml
conda activate libnomp-dev

Similarly, you can install the dev dependencies with mamba as well:

mamba env create -f environment-dev.yml
mamba activate libnomp-dev

Use lncfg to configure CMake build and lnbuild to build/install libnomp. To see the available options, check out lncfg –help. libnomp currently supports the following backends:

OpenCL
CUDA
HIP

For example, to build libnomp with OpenCL backend enabled, use the following commands:

cd libnomp
./lncfg --enable-opencl --install-prefix ${HOME}/.nomp
./lnbuild

NOMP_INSTALL_DIR environment variable must be set to libnomp install directory or it should be passed as a command line argument using –nomp-install-dir during initialization of libnomp. It is recommended to update the shell configuration file (e.g., .bashrc) to set environment variable NOMP_INSTALL_DIR and append PATH variable with NOMP_INSTALL_DIR/bin.

export NOMP_INSTALL_DIR=${HOME}/.nomp
export PATH=${NOMP_INSTALL_DIR}/bin:${PATH}

Setting the environment variable and adding NOMP_INSTALL_DIR/bin to PATH will enable you to use lnrun script without using its full path to open documentation, run tests, debug tests, etc. You can update the shell configuration file automatically by passing –update-shell to lnbuild script. You have to open a new shell in order for changes to take affect in case you update the shell configuration file.

cd libnomp
./lncfg --enable-opencl --install-prefix ${HOME}/.nomp
./lnbuild --update-shell

Use lnbuild –help to see all the available options supported by lnbuild script.

You might additionally want to specify OpenCL library path as below if CMake can’t find OpenCL:

./lncfg --enable-opencl --opencl-lib /lib/x86_64-linux-gnu/libOpenCL.so

If you used conda to install OpenCL (for example pocl), do the following:

./lncfg --enable-opencl --opencl-lib ${CONDA_PREFIX}/lib/libOpenCL.so --opencl-headers ${CONDA_PREFIX}/include

Run libnomp tests

You can run libnomp tests by executing lnrun test command. See below for a few examples on how to use the script:

lnrun test
lnrun test --backend opencl

Use lnrun help to see all supported options.

nompcc

Clone nompcc repo first and change directory to llvm-project:

git clone https://github.com/nomp-org/llvm-project.git
cd llvm-project

If you are using Linux, build the llvm-project as follows:

nprocs=$(grep -c ^processor /proc/cpuinfo)
mkdir build; cd build
cmake -G "Unix Makefiles" ../llvm                    \
                   -DLLVM_ENABLE_PROJECTS="clang"    \
                   -DLLVM_TARGETS_TO_BUILD="X86"     \
                   -DLLVM_OPTIMIZED_TABLEGEN=ON      \
                   -DCMAKE_BUILD_TYPE=RelWithDebInfo \
                   -DCMAKE_C_COMPILER=`which gcc`    \
                   -DCMAKE_CXX_COMPILER=`which g++`  \
                   -DBUILD_SHARED_LIBS=on
make -j${nprocs}

If you are using OSX with Apple silicon, build the llvm-project as follows:

nprocs=$(sysctl -n hw.ncpu)
mkdir build; cd build
cmake -G "Unix Makefiles" ../llvm                                \
                    -DLLVM_ENABLE_PROJECTS="clang"               \
                    -DLLVM_TARGETS_TO_BUILD="AArch64"            \
                    -DLLVM_OPTIMIZED_TABLEGEN=ON                 \
                    -DCMAKE_BUILD_TYPE=RelWithDebInfo            \
                    -DCMAKE_C_COMPILER=`which clang`             \
                    -DCMAKE_CXX_COMPILER=`which clang++`         \
                    -DCMAKE_OSX_ARCHITECTURES='arm64'            \
                    -DDEFAULT_SYSROOT="$(xcrun --show-sdk-path)" \
                    -DBUILD_SHARED_LIBS=on
make -j${nprocs}

This will build the clang compiler in bin/clang. Set environment variable NOMP_CLANG_DIR to point to this clang binary directory:

export NOMP_CLANG_DIR=`pwd`/bin

Documentation

We use Doxygen for in source documentations and render those with Sphinx and Breathe. These packages must be available if you install the dev dependencies using conda. You can enable docs by passing either -docs or –enable-docs option to lncfg script.

./lncfg --enable-docs
./lnbuild

Use lnrun to open the user documentation locally. You can specify the browser with option –browser. For example, to open the documentation in firefox:

lnrun docs --browser firefox

If you do not specify the browser, it opens the documentation in chrome by default.