Data-parallel Extension for Numba* (numba-dpex) is an open-source standalone extension for the Numba Python JIT compiler. Numba-dpex provides a SYCL*-like API for kernel programming Python. SYCL* is an open standard developed by the Unified Acceleration Foundation as a vendor-agnostic way of programming different types of data-parallel hardware such as multi-core CPUs, GPUs, and FPGAs. Numba-dpex's kernel-programming API brings the same programming model and a similar API to Python. The API allows expressing portable data-parallel kernels in Python and then JIT compiling them for different hardware targets. JIT compilation is supported for hardware that use the SPIR-V intermediate representation format that includes OpenCL CPU (Intel, AMD) devices, OpenCL GPU (Intel integrated and discrete GPUs) devices, and oneAPI Level Zero GPU (Intel integrated and discrete GPUs) devices.
The kernel programming API does not yet support every SYCL* feature. Refer to the SYCL* and numba-dpex feature comparison page to get a summary of supported features. Numba-dpex only implements SYCL*'s kernel programming API, all SYCL runtime Python bindings are provided by the dpctl package.
Along with the kernel programming API, numba-dpex extends Numba's
auto-parallelizer to bring device offload capabilities to prange loops and
NumPy-like vector expressions. The offload functionality is supported via the
NumPy drop-in replacement library: dpnp.
Note that dpnp and NumPy-based expressions can be used together in the same
function, with dpnp expressions getting offloaded by numba-dpex and NumPy
expressions getting parallelized by Numba.
Refer the documentation and examples to learn more.
Numba-dpex is part of the Intel® Distribution of Python (IDP) and Intel® oneAPI AIKit, and can be installed along with oneAPI. Additionally, we support installing it from Anaconda cloud. Please refer the instructions on our documentation page for more details.
Once the package is installed, a good starting point is to run the examples in
the numba_dpex/examples directory. The test suite may also be invoked as
follows:
python -m pytest --pyargs numba_dpex.testsTo install numba_dpex from the Intel(R) channel on Anaconda
cloud, use the following command:
conda install numba-dpex -c intel -c conda-forgeThe numba_dpex can be installed using pip obtaining wheel packages either from PyPi or from Intel(R) channel on Anaconda.
To install numba_dpex wheel package from Intel(R) channel on Anaconda, run the following command:
python -m pip install --index-url https://pypi.anaconda.org/intel/simple numba-dpexPlease create an issue for feature requests and bug reports. You can also use the GitHub Discussions feature for general questions.
If you want to chat with the developers, join the #Data-Parallel-Python_community room on Gitter.im.
Also refer our CONTRIBUTING page.