.. _device-support:

Device Support
==============


Data Parallel C++ supports several types of devices:


- Host device: Performs computations directly on the current CPU
- CPU device: Performs computations on a CPU using OpenCL™
- GPU device: Performs computations on a GPU


All Data Parallel C++ routines of |O-MKL| RNG support at least the Host and CPU devices. GPU devices are supported for the following engines:


- ``oneapi::mkl::rng::mcg31m1``
- ``oneapi::mkl::rng::mcg59``
- ``oneapi::mkl::rng::mrg32k3a``
- ``oneapi::mkl::rng::mt19937``
- ``oneapi::mkl::rng::mt2203``
- ``oneapi::mkl::rng::philox4x32x10``
- ``oneapi::mkl::rng::sobol``


GPU devices are supported for all distributions.


.. note::
   Some of the distributions use double precision inside the implementations, so not all GPUs are supported for them.

Refer to :ref:`Engines<engines-basic-random-number-generators>` and :ref:`distributions` for more detailed descriptions of each routine.