DPNP: NumPy Drop-In Replacement for Intel(R) XPU

API coverage summary

Full documentation

DPNP C++ backend documentation

The project contains:

• Python interface with NumPy-like API
• C++ library with SYCL based kernels

How to run

By default main CPU SYCL queue is used. To use Intel GPU please use:

DPNP_QUEUE_GPU=1 python examples/example1.py

Build from source:

git clone https://github.com/IntelPython/dpnp
cd dpnp
./0.build.sh

Install Wheel Package from Pypi

Install DPNP

python -m pip install --index-url https://pypi.anaconda.org/intel/simple -extra-index-url https://pypi.org/simple dpnp

Note: DPNP wheel package is placed on Pypi, but some of its dependencies (like Intel numpy) are in Anaconda Cloud. That is why install command requires additional intel Pypi channel from Anaconda Cloud.

Set path to Performance Libraries in case of using venv or system Python:

export LD_LIBRARY_PATH=<path_to_your_env>/lib

It is also required to set following environment variables:

export OCL_ICD_FILENAMES_RESET=1
export OCL_ICD_FILENAMES=libintelocl.so

Run test

. ./0.env.sh
pytest
# or
pytest tests/test_matmul.py -s -v
# or
python -m unittest tests/test_mixins.py

Run numpy external test

. ./0.env.sh
python -m tests.third_party.numpy_ext
# or
python -m tests.third_party.numpy_ext core/tests/test_umath.py
# or
python -m tests.third_party.numpy_ext core/tests/test_umath.py::TestHypot::test_simple

Building documentation:

Prerequisites:
$ conda install sphinx sphinx_rtd_theme
Building:
1. Install dpnp into your python environment
2. $ cd doc && make html
3. The documentation will be in doc/_build/html

Packaging:

. ./0.env.sh
conda-build conda-recipe/

Run benchmark:

cd benchmarks/

asv run --python=python --bench <filename without .py>
# example:
asv run --python=python --bench bench_elementwise

# or

asv run --python=python --bench <class>.<bench>
# example:
asv run --python=python --bench Elementwise.time_square

# add --quick option to run every case once but looks like first execution has additional overheads and takes a lot of time (need to be investigated)

Tests matrix: