GeezerVison

The World Wide Web, restored to 800 pixels of glory.

GeezerVison is showing intelpython.github.io — Quick Start Guide - Data Parallel Extension for NumPy (dpnp) 0.22.0dev1 documentation

Quick Start Guide¶

Device Drivers¶

To start programming data parallel devices beyond CPU, you will need an appropriate hardware. The Data Parallel Extension for NumPy* works fine on Intel © laptops with integrated graphics. In majority of cases, your Windows*-based laptop already has all necessary device drivers installed. But if you want the most up-to-date driver, you can always update it to the latest one. Follow device driver installation instructions to complete the step.

Python Interpreter¶

You will need Python 3.11, 3.12, 3.13, 3.14 or 3.15 installed on your system. If you do not have one yet the easiest way to do that is to install Intel Distribution for Python*. It installs all essential Python numerical and machine learning packages optimized for the Intel hardware, including Data Parallel Extension for NumPy*. If you have Python installation from another vendor, it is fine too. All you need is to install Data Parallel Extension for NumPy* manually as shown in the next installation section.

Installation¶

Install Package from Intel(R) channel¶

You will need one of the commands below:

Conda: conda install dpnp -c https://software.repos.intel.com/python/conda/ -c conda-forge --override-channels
Pip: python -m pip install --index-url https://software.repos.intel.com/python/pypi dpnp

These commands install dpnp package along with its dependencies, including dpctl package with Data Parallel Control Library and all required compiler runtimes and OneMKL.

Warning

Packages from the Intel channel are meant to be used together with dependencies from the conda-forge channel, and might not work correctly when used in an environment where packages from the anaconda default channel have been installed. It is advisable to use the miniforge installer for conda/mamba, as it comes with conda-forge as the only default channel.

Note

Before installing with conda or pip it is strongly advised to update conda and pip to latest versions

Build and Install Conda Package¶

Alternatively you can create and activate a local conda build environment:

conda create -n build-env conda-build
conda activate build-env

And to build dpnp package from the sources:

conda-build conda-recipe -c https://software.repos.intel.com/python/conda/ -c conda-forge --override-channels

Finally, to install the result package:

conda install dpnp -c local

Build and Install with scikit-build¶

Another way to build and install dpnp package from the source is to use Python setuptools and scikit-build. You will need to create a local conda build environment by command below depending on hosting OS.

On Linux:

conda create -n build-env dpctl cython dpcpp_linux-64 mkl-devel-dpcpp tbb-devel        \
      onedpl-devel cmake scikit-build ninja versioneer pytest ocl-icd-system           \
      -c dppy/label/dev -c https://software.repos.intel.com/python/conda/ -c conda-forge --override-channels
conda activate build-env

On Windows:

conda create -n build-env dpctl cython dpcpp_win-64 mkl-devel-dpcpp tbb-devel          \
      onedpl-devel cmake scikit-build ninja versioneer pytest khronos-opencl-icd-loader\
      -c dppy/label/dev -c https://software.repos.intel.com/python/conda/ -c conda-forge --override-channels
conda activate build-env

To build and install the package on Linux OS, run:

python setup.py build_ext --inplace -- -G Ninja -DCMAKE_C_COMPILER:PATH=icx -DCMAKE_CXX_COMPILER:PATH=icpx
python -m pip install -e . --no-build-isolation

To build and install the package on Windows OS, run:

python setup.py build_ext --inplace -- -G Ninja -DCMAKE_C_COMPILER:PATH=icx -DCMAKE_CXX_COMPILER:PATH=icx
python -m pip install -e . --no-build-isolation

Alternatively, to develop on Linux or Windows OS, you can use the driver script:

python scripts/build_locally.py

Building for custom SYCL targets¶

Project dpnp is written using generic SYCL and supports building for multiple SYCL targets, subject to limitations of CodePlay plugins implementing SYCL programming model for classes of devices.

Building dpnp for these targets requires that these CodePlay plugins be installed into DPC++ installation layout of compatible version. The following plugins from CodePlay are supported:

oneAPI for NVIDIA(R) GPUs

oneAPI for AMD GPUs

Building dpnp also requires building Data Parallel Control Library for custom SYCL targets.

Builds for CUDA and AMD devices internally use SYCL alias targets that are passed to the compiler. A full list of available SYCL alias targets is available in the DPC++ Compiler User Manual.

CUDA build¶

To build for CUDA devices, use the --target-cuda argument.

To target a specific architecture (e.g., sm_80):

python scripts/build_locally.py --target-cuda=sm_80

To use the default architecture (sm_50), run:

python scripts/build_locally.py --target-cuda

Note that kernels are built for the default architecture (sm_50), allowing them to work on a wider range of architectures, but limiting the usage of more recent CUDA features.

For reference, compute architecture strings like sm_80 correspond to specific CUDA Compute Capabilities (e.g., Compute Capability 8.0 corresponds to sm_80). A complete mapping between NVIDIA GPU models and their respective Compute Capabilities can be found in the official CUDA GPU Compute Capability documentation.

AMD build¶

To build for AMD devices, use the --target-hip=<arch> argument:

python scripts/build_locally.py --target-hip=<arch>

Note that the oneAPI for AMD GPUs plugin requires the architecture be specified and only one architecture can be specified at a time.

To determine the architecture code (<arch>) for your AMD GPU, run:

rocminfo | grep 'Name: *gfx.*'

This will print names like gfx90a, gfx1030, etc. You can then use one of them as the argument to --target-hip.

For example:

python scripts/build_locally.py --target-hip=gfx90a

Multi-target build¶

The default dpnp build from the source enables support of Intel devices only. Extending the build with a custom SYCL target additionally enables support of CUDA or AMD device in dpnp. Besides, the support can be also extended to enable both CUDA and AMD devices at the same time:

python scripts/build_locally.py --target-cuda --target-hip=gfx90a

Command-Line Interface¶

The python -m dpnp command provides options to query the include paths needed when building C++ extensions with dpnp:

python -m dpnp --includes            # print -I flag for dpnp include directory
python -m dpnp --include-dir         # print path to dpnp include directory
python -m dpnp --tensor-includes     # print -I flag for libtensor include directory
python -m dpnp --tensor-include-dir  # print path to libtensor include directory

These options are useful when building pybind11 extensions that use dpnp4pybind11.hpp or libtensor kernel headers.

Testing¶

To run the test suite from the source tree, after building the package as described above, run:

python -m pytest dpnp/tests

To run the tests of an installed dpnp package, run:

python -m pytest --pyargs dpnp.tests

Examples¶

The examples below demonstrates a simple usage of the Data Parallel Extension for NumPy*

How to create an array and to sum the elements¶

 1"""
 2
 3import dpnp as np
 4
 5x = np.asarray([1, 2, 3])
 6print("Array x allocated on the device:", x.device)
 7
 8y = np.sum(x)
 9
10print("Result y is located on the device:", y.device)  # The same device as x
11print("Shape of y is:", y.shape)  # 0-dimensional array
12print("y =", y)  # Expect 6

How to create an array on the specific device type and how the next computations follow it¶

 1"""
 2
 3import dpnp as np
 4
 5x = np.empty(3)
 6try:
 7    x = np.asarray([1, 2, 3], device="gpu")
 8except Exception:
 9    print("GPU device is not available")
10
11print("Array x allocated on the device:", x.device)
12
13y = np.sum(x)
14
15print("Result y is located on the device:", y.device)  # The same device as x
16print("Shape of y is:", y.shape)  # 0-dimensional array
17print("y=", y)  # Expect 6

More examples on how to use dpnp can be found in dpnp/examples.

You are visitor 01402503 — GeezerVison does not store the pages you look up.
Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.