[CMake] CUDA language support with host compiler flags

[Robert Maynard]

Currently to get proper propagation of architecture flags such as
-arch=sm_50,  -compute=compute_X  you need to place these into the
CMAKE_CUDA_FLAGS. This is a known issue, as flags specified by
`target_compile_options` are not propagate to the device linking step,
which needs the correct architecture flags.


On Thu, Feb 28, 2019 at 8:13 AM Máté Ferenc Nagy-Egri
<csiga.biga at aol.com> wrote:
>
> Hi Robert!
>
>
>
> Thank you for the help. I don’t know if I could’ve find that solution myself. I guess selecting CUDA as the COMPILE_LANGUAGE I can controll .cu compiler options to select architecture and what not.
>
>
>
> Anyhow, if someone needed the minimal template, here it is:
>
>
>
> https://gist.github.com/MathiasMagnus/0edacac888a758fe233cb69f3e291d62
>
>
>
> Cheers,
>
> Máté
>
>
>
> Feladó: Robert Maynard
> Elküldve: 2019. február 27., szerda 15:00
> Címzett: Nagy-Egri MĂĄtĂŠ Ferenc
> Másolatot kap: CMake MailingList
> Tárgy: Re: [CMake] CUDA language support with host compiler flags
>
>
>
> You need to guard the flags with `$<COMPILE_LANGUAGE:CXX>` the
>
> evaluation on a given compiler id is done for all sources of a target,
>
> and not on each target source file.
>
>
>
> So you will need something like:
>
>
>
> set(cxx_flags "$<$<OR:$<CXX_COMPILER_ID:GNU>,$<CXX_COMPILER_ID:Clang>>:-Wall
>
> -Wextra -pedantic>
>
>                                                $<$<CXX_COMPILER_ID:MSVC>:/W4")
>
> target_compile_options(${PROJECT_NAME} PRIVATE
>
> $<$<COMPILE_LANGUAGE:CXX>:${cxx_flags}>)
>
>
>
> On Fri, Feb 22, 2019 at 4:11 AM Máté Ferenc Nagy-Egri via CMake
>
> <cmake at cmake.org> wrote:
>
> >
>
> > Hi All!
>
> >
>
> > I am trying to compile CUDA code with controlling both host and device compiler flags.
>
> >
>
> > Currently my CMakeLists.txt looks like:
>
> >
>
> > ```
>
> > cmake_minimum_required(VERSION 3.8) # CUDA language support
>
> >
>
> > project(CUDA_test LANGUAGES CXX CUDA)
>
> >
>
> > if (MSVC)
>
> >   string(REGEX REPLACE "/W[0-9]" "" CMAKE_CXX_FLAGS ${CMAKE_CXX_FLAGS})
>
> > endif (MSVC)
>
> >
>
> > set(Hdrs)
>
> >
>
> > set(Srcs Main.cu)
>
> >
>
> > add_executable(${PROJECT_NAME} ${Hdrs} ${Srcs})
>
> >
>
> > target_include_directories(${PROJECT_NAME} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
>
> >
>
> > target_compile_options(${PROJECT_NAME} PRIVATE $<$<OR:$<CXX_COMPILER_ID:GNU>,$<CXX_COMPILER_ID:Clang>>:-Wall -Wextra -pedantic>
>
> >                                                $<$<CXX_COMPILER_ID:MSVC>:/W4>)
>
> >
>
> > set_target_properties(${PROJECT_NAME} PROPERTIES CUDA_SEPARABLE_COMPILATION ON
>
> >                                                  CUDA_STANDARD 14
>
> >                                                  CUDA_STANDARD_REQUIRED ON
>
> >                                                  CUDA_EXTENSIONS OFF
>
> >                                                  CXX_STANDARD 14
>
> >                                                  CXX_STANDARD_REQUIRED ON
>
> >                                                  CXX_EXTENSIONS OFF)
>
> >
>
> > source_group ("Headers" FILES ${Hdrs})
>
> > source_group ("Sources" FILES ${Srcs})
>
> > ```
>
> >
>
> > However, when I compile the code I get the following error:
>
> >
>
> > [1/3] Building CUDA object CMakeFiles/CUDA_test.dir/Main.cu.o
>
> > FAILED: CMakeFiles/CUDA_test.dir/Main.cu.o
>
> > /usr/bin/nvcc   -I/var/tmp/src/e75971a9-7e91-6137-abfa-df34048cc171/GCC-Debug-WSL -g   -Wall -Wextra -pedantic -std=c++14 -x cu -dc /var/tmp/src/e75971a9-7e91-6137-abfa-df34048cc171/GCC-Debug-WSL/Main.cu -o CMakeFiles/CUDA_test.dir/Main.cu.o && /usr/bin/nvcc   -I/var/tmp/src/e75971a9-7e91-6137-abfa-df34048cc171/GCC-Debug-WSL -g   -Wall -Wextra -pedantic -std=c++14 -x cu -M /var/tmp/src/e75971a9-7e91-6137-abfa-df34048cc171/GCC-Debug-WSL/Main.cu -MT CMakeFiles/CUDA_test.dir/Main.cu.o -o CMakeFiles/CUDA_test.dir/Main.cu.o.d
>
> > nvcc fatal   : Unknown option 'Wall'
>
> > ninja: build stopped: subcommand failed.
>
> >
>
> > CMake seems to pass -Wall -Wextra -pedantic to the device compiler as well, even though it is neither GNU nor Clang, but CUDA. How can I specify warning and similar flags separatly for the host and device compilers?
>
> > --
>
> >
>
> > Powered by www.kitware.com
>
> >
>
> > Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ
>
> >
>
> > Kitware offers various services to support the CMake community. For more information on each offering, please visit:
>
> >
>
> > CMake Support: http://cmake.org/cmake/help/support.html
>
> > CMake Consulting: http://cmake.org/cmake/help/consulting.html
>
> > CMake Training Courses: http://cmake.org/cmake/help/training.html
>
> >
>
> > Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html
>
> >
>
> > Follow this link to subscribe/unsubscribe:
>
> > https://cmake.org/mailman/listinfo/cmake
>
>