Instructions for Ubuntu
Since it took quite some time to figure out which tools are necessary to build NumPy from source such that Intel MKL is enabled, I’m documenting the approach which worked for me here.
Step 1: Prepare the environment
The NumPy build process requires python3-dev.
sudo apt-get install python3-dev
In addition, it makes sense to install NumPy into a new Python virtual environment.
Python virtual environments can for example be created using venv.
sudo apt-get install python3-venv
We can now create a virtual environment as follows
python3 -m venv numpymkl
Activate it prior to building NumPy
source numpymkl/bin/activate
Step 2: Get Intel MKL
The Intel MKL can for example be obtained by installing the Intel oneAPI Base Toolkit. There are various ways of doing this which Intel documents on their site. I chose the apt variant. It is probably best to look up Intel’s instructions instead of following the following steps:
First, some basic tools are required
sudo apt update
sudo apt -y install cmake pkg-config build-essential
Second, allow the apt client to use the Intel repository
wget https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
sudo apt-key add GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
rm GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB
sudo add-apt-repository "deb https://apt.repos.intel.com/oneapi all main"
Third, install the basic compilers
sudo apt install intel-basekit
The recommendation is to put
source /opt/intel/oneapi/setvars.sh
into the ~/.bashrc.
Alternatively, just make sure to run this command before you compile or use NumPy.
Step 3: Install NumPy
Clone the NumPy git repository. It is tempting to download the zip instead of cloning the repository. However, you will be missing submodules which might lead to cryptic compilation errors. So, after cloning the repository, also pull the submodules
git submodule update --init
Now, checkout the NumPy version which you would like to install.
It makes sense to choose a particular tag.
For example, I installed version 1.26.0.
git fetch --all --tags
git checkout tags/v1.26.0
With the Python virtual environment active as well as with the Intel oneAPI variables set (source /opt/intel/oneapi/setvars.sh), navigate into the NumPy repository and install the Python packages that are required for the build process
python3 -m pip install -r build_requirements.txt
Then, install NumPy
python3 setup.py install
If it worked, python3 -m pip list will list numpy and the correct version number.
Navigate into a different folder and check if Intel MKL is enabled
python3
>>> import numpy
>>> numpy.show_config()
Some information about the BLAS and LAPACK libraries should be printed which tells you that Intel MKL is used.
Note: Do not import numpy while you are still in the numpy repository folder.
Note: The Intel oneAPI variables need to be set before you import numpy.