(2014-11-10)
CMake does a great job in auto-detecting your system. But what to do if you want to be in control of the choice of compilers, flags etc ? Use toolchain files.
The start of my cmake thing is here.
Please consider the discussion on cross compiling on the cmake home page. Using this approach I was able to cross-compile code for Windows (via mingw on Linux, including GUI (fltk) and OpenGL.
The basic approach is to split your CMake system into two parts:
one generic (wrt. system) part which describes your project
one system specific part which describes the compiler etc.
For an example, we set up a second version of our project in tinyproject2.zip by just adding one file to it, let us call it intel.cmake.
It contains
set(CMAKE_SYSTEM_NAME Linux)
set(CMAKE_COMPILER_VENDOR "intel")
set(CMAKE_C_COMPILER icc)
set(CMAKE_C_FLAGS "-O3 " CACHE STRING "" FORCE)Now, we do the standard out-of-source build thing with one modification:
$ mkdir build $ cd build $ cmake -DCMAKE_TOOLCHAIN_FILE=../intel.cmake .. $ cd .. $ make -C build $ .build/hello Hello world, the answer is 42!
During the cmake run you should see that cmake detects the intel compiler and then uses it.
Please note, that due to the out-of-source work flow it is very easy to work in parallel on different versions, e.g.
gnu-dbg gnu-opt intel-dbg intel-opt
etc. Each has its own toolchain file and build directory.