mkl_umath

mkl_umath._ufuncs exposes Intel(R) Math Kernel Library powered version of loops used in the patched version of NumPy, that used to be included in Intel(R) Distribution for Python*.

Patches were factored out per community feedback (NEP-36).

mkl_umath started as a part of Intel (R) Distribution for Python* optimizations to NumPy, and is now being released as a stand-alone package. It can be installed into conda environment using

   conda install -c https://software.repos.intel.com/python/conda mkl_umath

---

To install mkl_umath Pypi package please use following command:

   python -m pip install --i https://software.repos.intel.com/python/pypi -extra-index-url https://pypi.org/simple mkl_umath

If command above installs NumPy package from the Pypi, please use following command to install Intel optimized NumPy wheel package from Intel Pypi Cloud:

   python -m pip install --i https://software.repos.intel.com/python/pypi -extra-index-url https://pypi.org/simple mkl_umath numpy==<numpy_version>

Where <numpy_version> should be the latest version from https://software.repos.intel.com/python/conda/

---

Building

Intel(R) C compiler and Intel(R) OneAPI Math Kernel Library (OneMKL) are required to build mkl_umath from source.

If these are installed as part of a oneAPI installation, the following packages must also be installed into the environment

• cmake
• ninja
• cython
• scikit-build
• numpy

If build dependencies are to be installed with Conda, the following packages must be installed from the Intel(R) channel

• mkl-devel
• dpcpp_linux-64 (or dpcpp_win-64 for Windows)
• numpy-base

then the remaining dependencies

• cmake
• ninja
• cython
• scikit-build

and for mkl-devel and dpcpp_linux-64 in a Conda environment, MKLROOT environment variable must be set On Linux

export MKLROOT=$CONDA_PREFIX

On Windows

set MKLROOT=%CONDA_PREFIX%

If using oneAPI, it must be activated in the environment

On Linux

source ${ONEAPI_ROOT}/setvars.sh

On Windows

call "%ONEAPI_ROOT%\setvars.bat"

finally, execute

CC=icx pip install --no-build-isolation --no-deps .