If you have pip installed, you should be
able to install the latest stable release of
scikit-cuda by running the
pip install scikit-cuda
All dependencies should be automatically downloaded and installed if they are not already on your system.
Obtaining the Latest Software¶
The latest stable and development versions of
scikit-cuda can be downloaded
Online documentation is available at https://scikit-cuda.readthedocs.org
scikit-cuda requires that the following software packages be
- Python 2.7 or 3.4.
- Setuptools 0.6c10 or later.
- Mako 1.0.1 or later.
- NumPy 1.2.0 or later.
- PyCUDA 2014.1 or later (some
scikit-cudamight not work properly with earlier versions).
- NVIDIA CUDA Toolkit 5.0 or later.
Note that both Python and the CUDA Toolkit must be built for the same architecture, i.e., Python compiled for a 32-bit architecture will not find the libraries provided by a 64-bit CUDA installation. CUDA versions from 7.0 onwards are 64-bit.
To run the unit tests, the following packages are also required:
Some of the linear algebra functionality relies on the CULA toolkit; as of 2017, it is available to premium tier users of E.M. Photonics’ HPC site Celerity Tools:
- CULA R16a or later.
To build the documentation, the following packages are also required:
The software has been developed and tested on Linux; it should also work on other Unix-like platforms supported by the above packages. Parts of the package may work on Windows as well, but remain untested.
Building and Installation¶
scikit-cuda searches for CUDA libraries in the system library
search path when imported. You may have to modify this path (e.g., by adding the
path to the CUDA libraries to
/etc/ld.so.conf and running
root or to the
LD_LIBRARY_PATH environmental variable on Linux, or by adding the CUDA
library path to the
DYLD_LIBRARY_PATH on MacOSX) if the libraries are
not being found.
To build and install the toolbox, download and unpack the source release and run:
python setup.py install
from within the main directory in the release. To rebuild the documentation, run:
python setup.py build_sphinx
Running the Unit Tests¶
To run all of the package unit tests, download and unpack the package source tarball and run:
python setup.py test
from within the main directory in the archive. Tests for individual
modules (found in the
tests/ subdirectory) can also be run
The functions provided by
scikit-cuda are grouped into several submodules in
skcuda namespace package. Sample code demonstrating how to use
different parts of the toolbox is located in the
demos/ subdirectory of the
source release. Many of the high-level functions also contain doctests that
describe their usage.