These are the PRs that fixed CUDA support in LLVM and CppInterOp by @anutosh491.
[clang-repl] : Fix clang-repl crash with --cuda flag - llvm
Add error propagation to interpreter creation logic - CppInterOp

Support for CUDA in xeus-cpp (Taking inspiration from xeus-clang-repl)

• Fixing cuda support in clang-repl.
• -cuda flag is already supported by CppInterOp's createClangInterpreter.
• We need to pass the cuda flag while calling Cpp::CreateInterpreter to trigger the above.
  

  xeus-cpp/src/xinterpreter.cpp

  Line 51 in 3c5328d

  return Cpp::CreateInterpreter(ClangArgs/*, {"-cuda"}*/);

• Need to add the following kernel(It should resemble something like this).

{
  "display_name": "CppInterOp (C++17)",
  "env": {
      "CPLUS_INCLUDE_PATH":"@CMAKE_CPLUS_INCLUDE_PATH@",
      "PATH":"@CMAKE_PATH@",
      "LD_LIBRARY_PATH":"@CMAKE_LD_LIBRARY_PATH@",
      "PYTHONPATH":"@CMAKE_PYTHONPATH@"
  },
  "argv": [
      "@CMAKE_INSTALL_PREFIX@/@CMAKE_INSTALL_BINDIR@/xcpp",
      "-f",
      "{connection_file}",
      "-cuda",
      "-std=c++17"@CMAKE_OMP@
  ],
  "language": "CUDA"
}

• Demo files like kalman_CUDA_demo and image_CUDA_demo.
• Having tests for CUDA in test_interpreter.cpp.

I hope these will bring CUDA support in xeus-cpp. I will be updating if any other changes are required.