Closed RobertPincus closed 6 months ago
Note that the standalone solver tests fail for nvfortran
24.1 and the default kernels. This is surprising because the same compiler/kernel combination passes similar tests in the end-to-end problem (tests/check_equivalence
). @alexeedm has promised to take a look in the coming days.
@skosukhin What would you think about merging this onto develop
despite the failing CI for nvidia+default-kernels? I'll chase that down after the fact.
Of course @alexeedm If you have any bandwidth to look at the failing nvfortran 24.1
/default kernels failures I'm grateful. You had thought there might be a compiler bug.
This PR adds some unit tests for the solvers alone (in
tests/rte[sl]w_unit_tests.F90
). Longwave solvers are tested for correctness in radiative equilibrium and for invariance to choices (vertical orientation, problem size...). Shortwave solvers are tested only for invariance.