[test] Add new Amber NVE library test and refactor the CSCS Amber check

[hurricane642]

This is a continuation of #2084, devoted exclusively to the Amber test. A new file nve.py has been added, introducing the main class for the Amber test, as well as the CSCS tests inherited from it.

[jenkins-cscs]

Can I test this patch?

[teojgo]

Ok to test

[pep8speaks]

Hello @hurricane642, Thank you for updating!

Cheers! There are no PEP8 issues in this Pull Request!Do see the ReFrame Coding Style Guide

Comment last updated at 2021-10-05 21:45:31 UTC

[victorusu]

@jenkins-cscs retry all

[vkarak]

I also believe that the CSCS version of the test can be simplified significantly. I will work a bit on that.

[vkarak]

I have a proposal for both the library file and the test. I will push tomorrow or do a PR to your branch.

[hurricane642]

I have a proposal for both the library file and the test. I will push tomorrow or do a PR to your branch.

Ok, cool, I'll wait :))

[codecov-commenter]

Codecov Report

Merging #2172 (8cda9fe) into master (c610815) will not change coverage.
The diff coverage is n/a.

❗ Current head 8cda9fe differs from pull request most recent head 0e27602. Consider uploading reports for the commit 0e27602 to get more accurate results

@@           Coverage Diff           @@
##           master    #2172   +/-   ##
=======================================
  Coverage   85.54%   85.54%           
=======================================
  Files          55       55           
  Lines        9746     9746           
=======================================
  Hits         8337     8337           
  Misses       1409     1409           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 43042a7...0e27602. Read the comment docs.

[vkarak]

@hurricane642 I've force pushed to your branch, since I made some substantial changes especially in the CSCS version of the test. I think that both the library and the CSCS version of the test -which is a bit weird to say the least- are much better now. @victorusu @jjotero comments?

[vkarak]

ok to test

[vkarak]

The simplest tests I could write and pass were the following:

@rfm.simple_test
class my_amber_check(Amber_NVE):
    modules = ['Amber']
    valid_prog_environs = ['builtin']

    @run_after('init')
    def scope_systems(self):
        if self.platform == 'gpu':
            self.valid_systems = ['daint:gpu']
        else:
            self.valid_systems = ['daint:mc']
            self.num_tasks = 2

or these two:

@rfm.simple_test
class my_amber_check(Amber_NVE):
    modules = ['Amber']
    valid_systems = ['daint:mc']
    valid_prog_environs = ['builtin']
    num_tasks = 2
    platform = parameter(['cpu'])


@rfm.simple_test
class my_amber_check(Amber_NVE):
    modules = ['Amber']
    valid_systems = ['daint:gpu']
    valid_prog_environs = ['builtin']
    platform = parameter(['gpu'])

Actually, the Cellulose_production_NVE was the only one not to pass, because it hit the default time limit.

I also noticed that num_tasks must be >=2 for the cpu version of the test, which I think it should be reflected in the library.

[vkarak]

Actually, the simplest functional test is this!

import reframe as rfm
from hpctestlib.apps.amber.nve import amber_nve_check


@rfm.simple_test
class my_amber_check(amber_nve_check):
    valid_prog_environs = ['*']
    valid_systems = ['*']

Which you can scope and run from the command line:

./bin/reframe <options> --system=dom:gpu -n '.*cuda$' -p builtin -S modules=Amber -S num_tasks=1 -r

or

./bin/reframe <options> --system=dom:mc -n '.*mpi$' -p builtin -S modules=Amber -S num_tasks=2 -r

[vkarak]

@jenkins-cscs retry daint