NVPL Python Usage#

NVPL does not provide Python modules or extensions. However, other Python libraries or applications may use NVPL libraries through runtime dynamic loading. Headers may be required for libraries that use JIT compilation. For these use cases, NVPL provides redistributable binary packages for each component library, installable with standard Python packaging tools.

Python Wheels#

NVPL Python wheels are hosted on PyPI.

Wheels are available for individual libraries or via the nvpl meta-package which will install all component libraries.

Installed locations:

  • Headers: <site-packages>/nvpl/include

  • Shared libraries: <site-packages>/nvpl/lib

Conda Packages#

NVPL is available as a BLAS/LAPACK backend in the conda-forge ecosystem, allowing users to easily install NumPy, SciPy, and other scientific computing packages that use NVPL for optimized linear algebra operations.

NVPL conda packages are also available on the conda-forge channel as standalone libraries for development use. Package names follow the pattern:

  • libnvpl-<lib>0 – Runtime libraries

  • libnvpl-<lib>-dev – Development packages with headers and CMake files

  • libnvpl-dev – Meta-package with all NVPL development packages

For detailed information on using conda with NVPL, including environment setup and BLAS backend configuration, see Using NVPL with Conda.

Building NumPy and SciPy with NVPL#

For users who want to build custom NumPy and SciPy wheels with NVPL from source, see Building NumPy and SciPy with NVPL for step-by-step instructions using pkg-config.

Python Documentation