[CMake] Using CMake to link C++ and CUDA Object file

[Robert Maynard]

To be clear the FindCUDA module currently does two things it finds the
location of the CUDA SDK and it also adds the 'cuda_' commands.

In your case you might still need to use FindCUDA to find the location of
the thrust include directory, or you can find the location yourself by
using find_path like:

```
find_path(CUDA_TOOLKIT_INCLUDE
  device_functions.h # Header included in toolkit
  PATHS ${CMAKE_CUDA_IMPLICIT_LINK_DIRECTORIES}
  PATH_SUFFIXES include ../include
  NO_DEFAULT_PATH
  )
```


On Fri, Jul 6, 2018 at 11:30 AM Quang Ha <quang.t.ha.20 at gmail.com> wrote:

> Thanks for your response, Robert. If I don't use find_package(CUDA), I run
> into the following problem:
>
>  fatal error: thrust/host_vector.h: No such file or directory
>
> I guess this is needed if I am using Thrust/CUDA?
>
> Thanks,
> Quang
>
> On Fri, 6 Jul 2018 at 06:38, Robert Maynard <robert.maynard at kitware.com>
> wrote:
>
>> if you are using the new first class cuda language, as seen by `project(MyTest
>> LANGUAGES CXX CUDA)` you shouldn't use the old find cuda package or the
>> associated cuda_add_library calls. When using the first class cuda
>> language simply use add_library.
>>
>> The other difference I notice is that you makefile example is not
>> building shared libraries but object files, so to the same in CMake you
>> need to use 'add_library' with the OBJECT keyword not the SHARED keyword.
>>
>>
>>
>> On Fri, Jul 6, 2018 at 12:01 AM Quang Ha <quang.t.ha.20 at gmail.com> wrote:
>>
>>> Hi all,
>>>
>>> Following upon how to perform code linking between CUDA and C++ (here at
>>> https://devblogs.nvidia.com/separate-compilation-linking-cuda-device-code/),
>>> I have successfully come up with a Makefile that is currently working for
>>> my test case:
>>>
>>> ##################################Makefile###############
>>> ##################################
>>> main_objects = main.o
>>> host_objects = host.o
>>> device_objects = device.o
>>>
>>> all: $(main_objects) $(host_objects) $(device_objects)
>>>         g++ -O3 -fopenmp $(main_objects) $(host_objects)
>>> $(device_objects) -L/usr/local/cuda/lib64 -lcudart -o main
>>>
>>> $(main_objects): $(main_objects:.o=.cpp)
>>>         g++ -O3 -I/usr/local/cuda/include -c $< -o $@
>>>
>>> $(host_objects): $(host_objects:.o=.cpp)
>>>         g++ -O3 -DTHRUST_HOST_SYSTEM=THRUST_HOST_SYSTEM_OMP -fopenmp -c
>>> $< -o $@
>>>
>>> $(device_objects): $(device_objects:.o=.cu)
>>>         nvcc -O3 -DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_CUDA -c $<
>>> -o $@
>>>
>>> clean:
>>>         rm -f *.o main
>>> ####################################################################
>>> ##################################
>>>
>>> Now, I am trying to write a CMake for this, my current attempt is as
>>> followed:
>>>
>>>
>>> ##############################################CMakeLists.txt############################################
>>> cmake_minimum_required(VERSION 3.8)
>>> project(MyTest LANGUAGES CXX CUDA)
>>>
>>> # check requirements
>>> find_package(CUDA REQUIRED)
>>> find_package(OpenMP)
>>>
>>> if (CUDA_FOUND AND OPENMP_FOUND)
>>>     # OpenMP
>>>     set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
>>>     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
>>>     set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS}
>>> ${OpenMP_EXE_LINKER_FLAGS}")
>>>
>>>     # CUDA
>>>     include_directories(${CUDA_INCLUDE_DIRS})
>>>     set(ALL_CUDA_LIBS ${CUDA_LIBRARIES} ${CUDA_cusparse_LIBRARY}
>>> ${CUDA_cublas_LIBRARY})
>>>     set(LIBS ${LIBS} ${ALL_CUDA_LIBS})
>>>     set(CUDA_SEPARABLE_COMPILATION OFF)
>>>     set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS};--maxrregcount 32)
>>>
>>>     set(OMP_SOURCES host.cpp)
>>>     add_library(Host_OMP SHARED ${OMP_SOURCES})
>>>     set_target_properties(Host_OMP PROPERTIES COMPILE_FLAGS
>>> "-DTHRUST_HOST_SYSTEM=THRUST_HOST_SYSTEM_OMP")
>>>     set(LIBS ${LIBS} ${Host_OMP})
>>>
>>>     set(CUDA_SOURCES device.cu)
>>>     cuda_add_library(Device_CUDA SHARED ${CUDA_SOURCES})
>>>     set_target_properties(Device_CUDA PROPERTIES COMPILE_FLAGS
>>> "-DTHRUST_DEVICE_SYSTEM=THRUST_DEVICE_SYSTEM_CUDA")
>>>     set_target_properties(Device_CUDA PROPERTIES
>>> CUDA_SEPERABLE_COMPILATION ON)
>>>     set(LIBS ${LIBS} ${Device_CUDA})
>>>
>>>     add_executable(main main.cpp)
>>>     target_link_libraries(main ${LIBS})
>>>
>>> endif(CUDA_FOUND AND OPENMP_FOUND)
>>> ####################################################################
>>> ##################################
>>>
>>> It is still, unfortunately, not working. If I commented out the
>>> add_executable line, the object code for Host_OMP and Device_CUDA is
>>> compiled. I can then use the terminal and link them with main.cpp using gcc
>>> and generate the executable - i.e. using the final command in the Makefile
>>> above.
>>>
>>> What am I doing wrong here? It seems to be very fundamental, but I can't
>>> seem to get it right...
>>>
>>> Any suggestions would be greatly appreciated!!
>>>
>>> Thanks,
>>> Quang
>>> --
>>>
>>> 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
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://cmake.org/pipermail/cmake/attachments/20180706/531fd1f4/attachment.html>