Alfio Lazzaro

> One softer design question there would be if the CI could be run more freely in this project's case I think so, CI here is github with very low...

> This is unrelated, sorry... > > > "there are no plans to make a new release of DBCSR beyond the current RC in CP2K" > > It would be...

/packit build

I'm checking this output: https://download.copr.fedorainfracloud.org/results/lecris/dbcsr/fedora-rawhide-x86_64/07625725-dbcsr/builder-live.log.gz I see that the test which takes most of the time is: * Serial case ``` Start 11: dbcsr_unittest1 11/17 Test #11: dbcsr_unittest1 ....................................... Passed...

> Making the workload run with OpenMPI using `--oversubscribe` is wrong. It usually means the launcher does not know how many processing elements (PE) in OpenMPI slang are available on...

Note that the most expensive tests we have are really serial, so likely I will go for a single rank...

@LecrisUT Thanks for this PR. I like the idea to default the number of ranks to 2. Most of the tests are really intended for max 4 ranks in general......

> Small note about `TEST_MPI_RANKS`. It should be fixed to `2` instead of `auto` in order to better take advantage of `ctest` parallelization I've changed the default to be 2,...

So I understand you want to mix LLVM (clang) for C/C++ and GNU (gfortran) for Fortran compilers. As you pointed out, in this way you mix the two OpenMP runtimes...

Well, @dev-zero suggested a way to avoid the OpenMP C and C++ if we don't use ACCEL and C_API... I will try to implement it. However, there are two orthogonal...