.. _libcudacxx-setup-requirements:

Requirements
============

All requirements are applicable to the ``main`` branch on GitHub. For
details on specific releases, please see the
:ref:`changelog <libcudacxx-releases-changelog>`.

Usage Requirements
------------------

To use the NVIDIA C++ Standard Library, you must meet the following
requirements.

System Software
~~~~~~~~~~~~~~~

The NVIDIA C++ Standard Library requires either the `NVIDIA HPC
SDK <https://developer.nvidia.com/hpc-sdk>`_ or the `CUDA
Toolkit <https://developer.nvidia.com/cuda-toolkit>`_.

libcu++ was first released in NVHPC 20.3 and CUDA 10.2. Some features
are only available in newer releases. Please see the :ref:`Standard API
section <libcudacxx-standard-api>`, :ref:`Extended API
section <libcudacxx-extended-api>`, and :ref:`release section <libcudacxx-releases>`
to find which features require newer releases.

Releases of libcu++ are only tested against the latest releases of NVHPC
and CUDA. It may be possible to use a newer version of libcu++ with an
older NVHPC or CUDA installation by using a libcu++ release from GitHub,
but please be aware this is not officially supported.

C++ Dialects
~~~~~~~~~~~~

The NVIDIA C++ Standard Library supports the following C++ dialects:

-  C++11
-  C++14
-  C++17

A number of features have been backported to earlier standards. Please
see the [API section] for more details.

NVCC Host Compilers
~~~~~~~~~~~~~~~~~~~

When used with NVCC, the NVIDIA C++ Standard Library supports the
following host compilers:

-  MSVC 2019 and 2022.
-  GCC 6, until 13.
-  Clang 14 until 18.
-  ICPC latest.
-  NVHPC 20.9 and 20.11.

Device Architectures
~~~~~~~~~~~~~~~~~~~~

The NVIDIA C++ Standard Library fully supports the following NVIDIA
device architectures:

-  Volta: SM 70 and 72.
-  Turing: SM 75.
-  Ampere: SM 80.

The following NVIDIA device architectures are partially supported:

-  Maxwell: SM 50, 52 and 53.

   -  Synchronization facilities are supported.

-  Pascall: SM 60, 61 and 62.

   -  Blocking synchronization facilities (e.g. most of the
      synchronization primitives) are not supported. Please see the
      :ref:`synchronization primitives section <libcudacxx-extended-api-synchronization>` for details.

Host Architectures
~~~~~~~~~~~~~~~~~~

The NVIDIA C++ Standard Library supports the following host
architectures:

-  aarch64.
-  x86-64.

Host Operating Systems
~~~~~~~~~~~~~~~~~~~~~~

The NVIDIA C++ Standard Library supports the following host operating
systems:

-  Linux.
-  Windows.
-  Android.
-  QNX.

Build and Test Requirements
---------------------------

To build and test libcu++ yourself, you will need the following in
addition to the usage requirements:

-  `CMake <https://cmake.org>`_.
-  `LLVM <https://github.com/llvm>`_.

   -  You do not have to build LLVM; we only need its CMake modules.

-  `lit <https://pypi.org/project/lit/>`_, the LLVM Integrated Tester.

   -  We recommend installing lit using Python's pip package manager.