libcu++: The NVIDIA C++ Standard Library

libcu++ is the NVIDIA C++ Standard Library, bringing you familiar features from the C++ Standard Library that you can seamlessly use in CUDA C++ in both host and device code.

If you know how to use your C++ Standard Library, then you know how to use libcu++. It's included in your NVIDIA HPC SDK or CUDA Toolkit installation. No additional installation or compiler flags are needed.

All you have to do is add cuda/ to the start of your includes and cuda:: before any uses of std:::

#include <cuda/std/atomic>
cuda::std::atomic<int> x;

Examples are a great way to get familiar with libcu++; the documentation is also pretty helpful.

cuda:: and cuda::std::

When used with NVCC, NVIDIA C++ Standard Library facilities live in their own header hierarchy and namespace with the same structure as, but distinct from, the host compiler's Standard Library:

• std::/<*>: Your host compiler's Standard Library that works in __host__ code only. When using NVCC, the NVIDIA Standard Library does not replace or interfere with host compiler's Standard Library.
• cuda::std::/<cuda/std/*>: Strictly conforming implementations of facilities from the Standard Library that work in __host__ __device__ code.
• cuda::/<cuda/*>: Conforming extensions to the Standard Library that work in __host__ __device__ code.
• cuda::device/<cuda/device/*>: Conforming extensions to the Standard Library that work only in __device__ code.

// Standard C++, __host__ only.
#include <atomic>
std::atomic<int> x;

// CUDA C++, __host__ __device__.
// Strictly conforming to the C++ Standard.
#include <cuda/std/atomic>
cuda::std::atomic<int> x;

// CUDA C++, __host__ __device__.
// Conforming extensions to the C++ Standard.
#include <cuda/std/atomic>
cuda::atomic<int, cuda::thread_scope_block> x;

cuda:: is Heterogeneous

libcu++ facilities work across your entire codebase, in both host and device code. Everything in cuda:: is __host__ __device__.

Unless otherwise noted, any libcu++ object which is copyable or movable can be copied or moved between host and device code.

Synchronization objects work across host and device code, and can be used to synchronize between host and device threads. However, there are some restrictions to be aware of; please see the synchronization library section for more details.

cuda::device::

A small number of libcu++ facilities only work in device code, usually because there is no sensible implementation in host code.

Such facilities live in cuda::device::.

cuda:: is Incremental

libcu++ delivers a high-priority subset of the C++ Standard Library today, and each release increases the feature set. But it is a subset; not everything is available today. The API section lists the facilities available and the releases they were first introduced in.

libcu++ is Open Source

libcu++ is an open source project developed on GitHub. It is NVIDIA's variant of LLVM's libc++. libcu++ is distributed under the Apache License v2.0 with LLVM Exceptions.