How to compile C++ as CUDA using CMake

Question

I'm writing some code that can be compiled as C++ or as CUDA. In the latter case, it makes use of CUDA kernels, in the former it just runs conventional code.

Having created a file named test.cpp, I can compile it manually thus:

g++ test.cpp          # build as C++ with GCC
nvcc -x cu test.cpp   # build as CUDA with NVCC

where -x cu tells nvcc that although it's a .cpp extension, I'd like it to treat it as CUDA. So far, so good.

However, when I migrate to using CMake, I don't know how to do the same thing. That is: how to ask CMake to compile the .cpp file with NVCC, rather than GCC.

cmake_minimum_required(VERSION 3.9)
project(cuda_test LANGUAGES CUDA CXX)
add_executable(cuda_test test.cpp)     # builds with GCC

If I create a symlink to the original file:

ln -s test.cpp test.cu

then change CMakeLists.txt:

add_executable(cuda_test test.cu)     # builds with NVCC

But I'd like to be able to specify the equivalent of NVCC's -x switch within CMake, rather than playing games with extensions. Something like:

set_target_properties(cuda_test PROPERTIES FORCE_LANGUAGE CUDA)

or even

set_target_properties(test.cpp PROPERTIES FORCE_LANGUAGE CUDA)

Does such an incantation exist?

Answer 1

By default, CMake chooses compiler for a source file according to the file's extension. But you may force CMake to use the compiler you want by setting LANGUAGE property for a file:

set_source_files_properties(test.cpp PROPERTIES LANGUAGE CUDA)

(This just calls CUDA compiler for a file using normal compiler options. You still need to pass additional options for that compiler, so it could work with the unusual file's extension.)