CP2K v2024.1

[laraPPr]

• link to support ticket: #2024022760001074
• website: https://www.cp2k.org/
• installation docs: https://github.com/cp2k/cp2k/blob/master/INSTALL.md
• toolchain: ?
• easyblock to use: ``
• required dependencies:
  • SIRIUS
    • {lib}[foss/2023a] SIRIUS v7.5.2, Umpire v2024.02.1, COSTA v2.2.2, ... easybuilders/easybuild-easyconfigs#20989
  • Libint
  • libxc
  • libxsmm
  • libvori
  • FFTW
  • PLUMED
• optional dependencies:
  • ...
• notes:
  • ...
• effort: (TBD)
• other install methods
  • conda:
  • container image:
  • pre-built binaries (RHEL8 Linux x86_64):
  • easyconfig outside EasyBuild:

[pavelToman]

• {chem}[foss/2023a] CP2K v2024.1 easybuilders/easybuild-easyconfigs#20798

[pavelToman]

Easyblock update:

• update CP2K easyblock to use do_regtest.py script for recent versions (WIP) easybuilders/easybuild-easyblocks#2984

[boegel]

Initial attempt by @pavelToman results in lots of failing tests with 2024.1 (after unsetting OMP_PROC_BIND), see https://github.com/vscentrum/vsc-software-stack/blob/wip/297_CP2K/test_error_1.txt:

------------------------------- Summary --------------------------------
Number of FAILED  tests 56
Number of WRONG   tests 0
Number of CORRECT tests 31
Total number of   tests 87

That may be because too many threads/ranks are used by default, resulting in very limited memory being available.

Next step would be to try with tests/do_regtest.py --mpiranks 2 --ompthreads 2 and see if that improves the situation.

@smoors Do you have any ideas/feedback here?

[boegel]

Instructions for running CP2K test suite are available at https://www.cp2k.org/dev:regtesting

[boegel]

Something I overlooked initially: they also wanted to include https://github.com/electronic-structure/SIRIUS as a dependency for CP2K.

@pavelToman Can you take a look at that one?

[pavelToman]

Initial attempt by @pavelToman results in lots of failing tests with 2024.1 (after unsetting OMP_PROC_BIND), see https://github.com/vscentrum/vsc-software-stack/blob/wip/297_CP2K/test_error_1.txt:

------------------------------- Summary --------------------------------
Number of FAILED  tests 56
Number of WRONG   tests 0
Number of CORRECT tests 31
Total number of   tests 87

That may be because too many threads/ranks are used by default, resulting in very limited memory being available.

Next step would be to try with tests/do_regtest.py --mpiranks 2 --ompthreads 2 and see if that improves the situation.

@smoors Do you have any ideas/feedback here?

I let it run whole night but it stops (freeze) on test step with a command:
python /tmp/vsc47063/easybuild/build/CP2K/2024.1/foss-2023a/cp2k-2024.1/tests/do_regtest.py --maxtasks 1 --mpiranks 2 --ompthreads 2 Linux-x86-64-foss psmp

• the SLURM envs unseted
• the OMP_PROC_BIND unseted

[hajgato]

I have installed CP2K from easybuilders/easybuild-easyconfigs#20798 with --skip-test-step and it runch with my CP2K test, from an interactive session (2 nodes, 8 ppn) as well as from a script job (with vsc-mympirun)

[hajgato]

there are lots of failed jobs with

cp2k.psmp: /tmp/vsc43020/easybuild/CP2K/2024.1/foss-2023a/cp2k-2024.1/src/dbm/dbm_distribution.c:42: dbm_dist_1d_new: Assertion `0 <= coords[i] && coords[i] < dist->nranks' failed.
cp2k.psmp: /tmp/vsc43020/easybuild/CP2K/2024.1/foss-2023a/cp2k-2024.1/src/dbm/dbm_distribution.c:42: dbm_dist_1d_new: Assertion `0 <= coords[i] && coords[i] < dist->nranks' failed

[hajgato]

when regtest is done, --maxtasks a --mpiranks b --ompthreads c, then a>=b*c
By default, a is taken form /proc/cpuinfo. Therefore if not a full node is requested for a job, the regtest would run overcommitted. In general a should be the same than the number of parallel builds (the number after make -j)