Is it possible to use CUDA C for object oriented programming?

Question

Is it possible to use CUDA C for object oriented programming? Are there any classes in CUDA C like in C++?

Answer 1

CUDA does support classes, however not all C++ features are available. CUDA C++ support has significantly evolved over the time. Here is a non-exhaustive list of supported features:

• C++ standards supported
  • CUDA 11 introduced C++17 support (see release notes)
  • CUDA 12 now supports C++20 (see release notes)
• Standard library support
  • libcu++ has many standard library functions available (non-exhaustive list here)
  • NVCC option --expt-relaxed-constexpr allows to use any constexpr host code in device code (powerful tool to reuse existing C++ code)
• Alternative compilers
  • NVIDIA offers nvc++, a compiler that allows to write parallel C++ code using C++17 execution policies (more information on this page)

Answer 2

nvcc since version 3.0 uses g++ (in Linux) for compiling host code, so you can use the whole set of C++ features for the host part.

For the device code it supports restricted set of C++. But I think in the future releases of CUDA it will be more complete.