Run tests in parallel on CI

[Nanoseb]

closes #20

[Nanoseb]

I had to use --oversubscribe because the CI runner doesn't seem to have more than 1 core. Though it shouldn't matter for the logic of the tests themselves.

[semi-h]

Allocator and reorder tests are the ones that doesn't require a parallel run. Both of them don't do any communication and even in a full simulation with multiple ranks they're always local to their own rank. Running these on multiple ranks is still a valid test but if we can avoid this easily it is better I guess.

[Nanoseb]

that's a good point, though I can't think of a clean way of doing it yet.

[pbartholomew08]

You could use a function to build the test definition, I'll see if I can dig out an example where I've done this

[pbartholomew08]

Something like the following should work

function(define_test test_name np)
  add_test(NAME ${test_name} COMMAND sh -c "mpirun --oversubscribe -np ${np} ...")
endfunction()

# Make a single process test (assumes there is a test_foo target)
define_test(test_foo 1)

# Make a parallel test
define_test(test_bar ${CMAKE_CTEST_NPROCS})

[pbartholomew08]

Minor comment, otherwise looks good to me.

[semi-h]

Can we specify number of ranks we want to test with somehow when building with a command line option? Or do we have to edit the relevant line in the CMakeLists.txt?

[Nanoseb]

Can we specify number of ranks we want to test with somehow when building with a command line option? Or do we have to edit the relevant line in the CMakeLists.txt?

Yes, you don't have to edit the CMakeLists.txt file. It is set to 4 by default, but is accessible via the CMAKE_CTEST_NPROCS variable so cmake -DCMAKE_CTEST_NPROCS=42 ... will work too. Or editing it later on via ccmake.