mkl-service - Python package for run-time control of Intel® oneAPI Math Kernel Library (oneMKL).

To install conda package, use conda install -c https://software.repos.intel.com/python/conda/ mkl-service, or conda install -c conda-forge mkl-service.

To install PyPI package, use python -m pip install mkl-service.

Intel® oneAPI Math Kernel Library (oneMKL) supports functions are subdivided into the following groups according to their purpose:

• Version Information
• Threading Control
• Timing
• Memory Management
• Conditional Numerical Reproducibility Control
• Miscellaneous

A short example, illustrating its use:

>>> import mkl
>>> mkl.domain_set_num_threads(1, domain="fft") # oneMKL FFT functions to run sequentially
# 'success'

For more information about the usage of support functions see Developer Reference for Intel® oneAPI Math Kernel Library for C.

Building

A C compiler and Intel(R) oneAPI Math Kernel Library (oneMKL) are required to build mkl-service from source. The compiler must support C11 atomics (i.e., for Windows, Visual Studio 2022 17.5 or newer).

Executing

python -m pip install .

will pull in the required build dependencies, including mkl, and build mkl-service.

If you already have mkl installed (from your system or a Conda environment) and want to reuse it instead of pulling a fresh copy into an isolated build, first install the build dependencies:

python -m pip install mkl-devel meson-python cmake ninja "cython>=3.1.0"

then build against the existing installation with:

python -m pip install --no-build-isolation --no-deps .

Build options

Options are passed using -Csetup-args:

python -m pip install . -Csetup-args=-Dilp64=true

By default mkl-service is built against the LP64 interface, in which oneMKL's integer type MKL_INT is 32-bit. Enabling ilp64 makes MKL_INT 64-bit and requests the ILP64 interface layer from libmkl_rt on import. Use it when the oneMKL libraries you link against provide the ILP64 interface.

The Python API is identical in both configurations — no function signature or return value changes.

Warning: the oneMKL interface layer is process-global and can only be selected before the first oneMKL call. mkl-service requests it on import, so an ILP64 build is order-dependent and unsafe to mix with LP64 consumers:

• If imported before another oneMKL consumer initializes oneMKL, that consumer's LP64 calls get reinterpreted as ILP64. LAPACK routines in particular may crash.
• If the other consumer initializes oneMKL first, the ILP64 request is ignored and the process stays LP64.

Only enable ilp64 when every consumer uses the ILP64 interface or mkl-service is the sole oneMKL consumer in the process.