ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

update details about unittests

Browse Source
This commit is contained in:
Axel Kohlmeyer
2023-12-13 00:49:41 -05:00
parent 1cff1deedc
commit f53e2c3398
1 changed files with 18 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
doc/src/Developer_unittest.rst
View File

@ -274,9 +274,7 @@ Tests for using the Fortran module are in the ``unittest/fortran``
folder. Since they are also using the GoogleTest library, they require
to also implement test wrappers in C++ that will call fortran functions
which provide a C function interface through ISO_C_BINDINGS that will in
turn call the functions in the LAMMPS Fortran module. This part of the
unit tests is incomplete since the Fortran module it is based on is
incomplete as well.
turn call the functions in the LAMMPS Fortran module.
Tests for the C++-style library interface
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
@ -397,10 +395,10 @@ compare with the reference and also start from the data file. A final
check will use multi-cutoff r-RESPA (if supported by the pair style) at
a 1:1 split and compare to the Verlet results. These sets of tests are
run with multiple test fixtures for accelerated styles (OPT, OPENMP,
INTEL) and for the latter two with 4 OpenMP threads enabled. For
these tests the relative error (epsilon) is lowered by a common factor
due to the additional numerical noise, but the tests are still comparing
to the same reference data.
INTEL, KOKKOS (OpenMP only)) and for the latter three with 4 OpenMP
threads enabled. For these tests the relative error (epsilon) is lowered
by a common factor due to the additional numerical noise, but the tests
are still comparing to the same reference data.
Additional tests will check whether all listed extract keywords are
supported and have the correct dimensionality and the final set of tests
@ -434,17 +432,19 @@ The ``test_pair_style`` tester is used with 4 categories of test inputs:
pair style is defined, but the computation of the pair style contributions
is disabled.
The ``test_bond_style`` and ``test_angle_style`` are set up in a similar
fashion and share support functions with the pair style tester. The final
group of tests in this section is for fix styles that add/manipulate forces
and velocities, e.g. for time integration, thermostats and more.
The ``test_bond_style``, ``test_angle_style``, ``test_dihedral_style``, and
``test_improper_style`` tester programs are set up in a similar fashion and
share support functions with the pair style tester. The final group of
tests in this section is for fix styles that add/manipulate forces and
velocities, e.g. for time integration, thermostats and more.
Adding a new test is easiest done by copying and modifying an existing test
for a style that is similar to one to be tested. The file name should follow
the naming conventions described above and after copying the file, the first
step is to replace the style names where needed. The coefficient values
do not have to be meaningful, just in a reasonable range for the given system.
It does not matter if some forces are large, for as long as they do not diverge.
Adding a new test is easiest done by copying and modifying an existing YAML
file for a style that is similar to one to be tested. The file name should
follow the naming conventions described above and after copying the file,
the first step is to replace the style names where needed. The coefficient
values do not have to be meaningful, just in a reasonable range for the
given system. It does not matter if some forces are large, for as long as
they do not diverge.
The template input files define a large number of index variables at the top
that can be modified inside the YAML file to control the behavior. For example,
@ -472,7 +472,7 @@ Note that this disables computing the kspace contribution, but still will run
the setup. The "gewald" parameter should be set explicitly to speed up the run.
For styles with long-range electrostatics, typically two tests are added one using
the (slower) analytic approximation of the erfc() function and the other using
the tabulated coulomb, to test both code paths. The reference results in the YAML
the tabulated coulomb, to test both code paths. The reference results in the YAML
files then should be compared manually, if they agree well enough within the limits
of those two approximations.