Clang UB sanitizer CI test: increase coverage

[lucafedeli88]

In CI test based on clang UB sanitizers, most of the time (~ 1h30) is spent in compiling the code, while just a few minutes are spent actually running some simulations. This means that we can increase the coverage of the test by adding some more simulations to the tests with a negligible increase of the total runtime.
This PR does just that: now most of the cases in Examples/Physics_applications are tested with the UB sanitizer.

Note that some cases cannot run in double precision (see below). For this reason, the PR also splits the UB sanitizer test into single precision and double precision (in double precision only the cases that cannot run in single precision are tested).

Updates:

1) Issue found while running inputs_test_3d_beam_beam_collision --> We need to run this case in double precision

The tool has found an issue while running mpirun -n 2 ./build/bin/warpx.3d Examples/Physics_applications/beam_beam_collision/inputs_test_3d_beam_beam_collision :

STEP 1 starts ...
/home/runner/work/WarpX/WarpX/build/_deps/fetchedpicsar-src/multi_physics/QED/include/picsar_qed/containers/picsar_tables.hpp:310:17: runtime error: -nan is outside the range of representable values of type 'int'
/home/runner/work/WarpX/WarpX/build/_deps/fetchedpicsar-src/multi_physics/QED/include/picsar_qed/containers/picsar_tables.hpp:310:17: runtime error: -nan is outside the range of representable values of type 'int'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior /home/runner/work/WarpX/WarpX/build/_deps/fetchedpicsar-src/multi_physics/QED/include/picsar_qed/containers/picsar_tables.hpp:310:17 in 
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior /home/runner/work/WarpX/WarpX/build/_deps/fetchedpicsar-src/multi_physics/QED/include/picsar_qed/containers/picsar_tables.hpp:310:17 in 

I've temporarily commented out this case while I investigate the cause. For the moment, I am not able to reproduce the issue on my local machine. The issue is using single precision for this specific test case! Specifically, momenta end up being NaN and the sanitizer detects the attempt to convert a floating point NaN into an integer.
I have added an ABLASTR_ALWAYS_ASSERT_WITH_MESSAGE in PoissonSolver.H so that a more readable error message is provided to the user.

2) Issue found while running inputs_test_2d_background_mcc --> We need to run this case in double precision

MLMG does not converge in single precision for this simulation case. We need to run it in double precision.

3) Issue found while running free_electron_laser --> We need to run this case in double precision

I have observed this issue:

 STEP 444 starts ...
0::1::Assertion `m_current_z_lab[i_buffer] >= m_buffer_domain_lab[i_buffer].lo(m_moving_window_dir) and m_current_z_lab[i_buffer] <= m_buffer_domain_lab[i_buffer].hi(m_moving_window_dir)' failed, file "/home/runner/work/WarpX/WarpX/Source/Diagnostics/BTDiagnostics.cpp", line 870, Msg: 
 ### ERROR   : z-slice in lab-frame (0.299976) is outside the buffer domain
#            physical extent (0.299976 to 0.299988).
 !!!

which seems to be related to using single precision instead of double precision. Therefore, we need to run this case in double precision.

4) Issue found while running inputs_test_2d_laser_ion_acc --> bugfix in WarpX

inputs_test_2d_laser_ion_acc case has the following issue in single precision:

--- INFO    : Writing openPMD file diags/openPMDbw000000
terminate called after throwing an instance of 'std::runtime_error'
  what():  Datatypes of chunk data (FLOAT) and record component (DOUBLE) do not match.
SIGABRT
See Backtrace.0.0 file for details

This comes from the fact that the datatype of this dataset in ParticleHistogram2D.cpp is hard-coded as double:

    auto dataset = io::Dataset(
            io::determineDatatype<double>(),
            {static_cast<unsigned long>(m_bin_num_ord), static_cast<unsigned long>(m_bin_num_abs)});

this PR modifies these lines as follows:

    auto dataset = io::Dataset(
            io::determineDatatype<amrex::Real>(),
            {static_cast<unsigned long>(m_bin_num_ord), static_cast<unsigned long>(m_bin_num_abs)});

5) Perform all the simulations in double precision

See #5936

6) Use ctest instead of running the tests directly

7) In IntervalsParser prevent the use of an uninitialized object

This was discovered by adding a new test case.
temp_slice.getStart() > m_slices[i_slice].getStart() && i_slice < static_cast<int>(m_slices.size()) must be reversed: i_slice < static_cast<int>(m_slices.size()) && temp_slice.getStart() > m_slices[i_slice].getStart() otherwise we call a method of a non-existent object.

[lucafedeli88]

ping, @EZoni !

[dpgrote]

This looks good, thanks!

I will comment that NaNs always make me nervous that there is something bad lurking in the code. Do you know why they appear in that case with single precision?

[lucafedeli88]

This looks good, thanks!

I will comment that NaNs always make me nervous that there is something bad lurking in the code. Do you know why they appear in that case with single precision?

Thanks, @dpgrote . The NaN comes from this line in PoissonSolver.H. beta_solver[2] is 1.00059783 here due to numerical errors, so the square root is negative. We could indeed try to fix that.

geom[lev].CellSize(2)/std::sqrt(1._rt-beta_solver[2]*beta_solver[2]))};

[lucafedeli88]

This looks good, thanks!
I will comment that NaNs always make me nervous that there is something bad lurking in the code. Do you know why they appear in that case with single precision?

Thanks, @dpgrote . The NaN comes from this line in PoissonSolver.H. beta_solver[2] is 1.00059783 here due to numerical errors, so the square root is negative. We could indeed try to fix that.

geom[lev].CellSize(2)/std::sqrt(1._rt-beta_solver[2]*beta_solver[2]))};

Actually, in this case, we have a 250 GeV electron beam. So we can't represent beta reliably in single precision for such high energies. I think that the best we can do is to add an assert to ensure that beta < 1 .

[lucafedeli88]

@dpgrote , I've added this check inside PoissonSolver.H :

    ABLASTR_ALWAYS_ASSERT_WITH_MESSAGE (
        std::all_of(beta_solver.begin(), beta_solver.end(),
            [](const auto b){return (b<1.0_rt);}),
        "Components of beta_solver must be < 1.");

so that at least the user has a more readable error message.

[EZoni]

Looks good to me.

[EZoni]

Just one last comment, in case it helps make this more compact and robust.

I left one more comment to address.