I am trying to apply openmp and mpi techniques to an open source C program which requires "cmake . && make" to be built. I already found at How to set linker flags for OpenMP in CMake's try_compile function how to include the -fopenmp flags but still confused about mpi. What can I do about that?
可以将文章内容翻译成中文,广告屏蔽插件可能会导致该功能失效(如失效,请关闭广告屏蔽插件后再试):
问题:
回答1:
OpenMP
Is this a question about OpenMP? Then all you have to do is compile with -fopenmp
which you can do by appending it to CMAKE_C_FLAGS
, for example:
SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fopenmp)
MPI
For MPI, you have to find mpi first
find_package(MPI) #make it REQUIRED, if you want
then add it's header files to your search path
include_directories(SYSTEM ${MPI_INCLUDE_PATH})
and finally link your program(s) (which is my_mpi_target
in my case)
target_link_libraries(my_mpi_target ${MPI_C_LIBRARIES})
回答2:
In modern CMake 3.X which is target based, the CMakeLists.txt should look like this:
cmake_minimum_required(VERSION 3.0)
project(main)
find_package(MPI REQUIRED)
# add this line only when you are using openmpi which has a different c++ bindings
add_definitions(-DOMPI_SKIP_MPICXX)
# Use imported targets would make things much eazier. Thanks Levi for pointing it out.
add_executable(main main.cpp)
target_link_libraries(main
PRIVATE
MPI_C)
# Old way.
#target_link_libraries(main
# PRIVATE
# ${MPI_C_LIBRARIES})
#target_include_directories(main
# PRIVATE
# ${MPI_C_INCLUDE_PATH})