RadForce test failure

[haakoan]

The test RadForce, run with make test, fails soon after starting. All other tests run without issue.

The test gives the following output and error:

MPI initialized with 1 MPI processes
MPI initialized with thread support level 0
AMReX (21.09-16-gd187e374f11c) initialized

Coarse STEP 1 at t = 0 (0%) starts ...
terminate called after throwing an instance of 'std::runtime_error'
  what():  Hydro states invalid (..../quokka/src/RadhydroSimulation.hpp:363)
Aborted (core dumped)

Is there a reason why this particular test would fail, while all others pass?

Thanks in advance for any help.

[BenWibking]

All tests pass when I run them. Can you copy the output from cmake so I can see exactly what build options and compiler options you are using?

[BenWibking]

I noticed the AMReX version is not the current version used in the repository. It might be useful to do git pull and git submodule update. The version string for the current version of AMReX included with Quokka should read AMReX (21.10-24-gb60a773d6c87).

[haakoan]

After many tries to build the code on my local HPC system I got it working, but I lost the output of cmake for the version of my attempts that produced this issue. I think you are right and that it had to do with AMReX. I think the problem was due to the set up of the packages and environmental variables of the HPC system.

I redid everything, with GCC, and got it running.

I had to manually set some of the CMAKE options. I think HPC users might run into a similar issues, the cmake does not always find the right compilers/python installations if the system has a complicated set up.

[BenWibking]

Were you using the Intel compiler by chance? Issue #5 notes that the `classic' Intel compiler (icc or icpc) generates incorrect machine code. If that was the issue, I will put a warning in the installation and readme about this.

[haakoan]

Yes I was. I saw this issue and it is why, together with a rather annoying setup of modules on my HPC system, that I changed to GCC. I find it amazing that the intel compiler actually does this. On that note, have you compared the performance of GCC and Intel?

[BenWibking]

The 'new' intel compilers (icx, icpx) are about 5% faster than gcc in my tests.