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

fix spelling error and reformat paragraph

Browse Source
This commit is contained in:
Axel Kohlmeyer
2021-08-29 22:08:49 -04:00
parent 5d03028028
commit 284ed98fb8
1 changed files with 13 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
doc/src/pair_snap.rst
View File

@ -191,22 +191,19 @@ pair_coeff command, to avoid ambiguity in the number of coefficients.
The keywords *chunksize* and *parallelthresh* are only applicable when
using the pair style *snap* with the KOKKOS package on GPUs and are
ignored otherwise.
The *chunksize* keyword controls
the number of atoms in each pass used to compute the bispectrum
components and is used to avoid running out of memory. For example
if there are 8192 atoms in the simulation and the *chunksize*
is set to 4096, the bispectrum calculation will be broken up
into two passes (running on a single GPU).
The *parallelthresh* keyword controls
a crossover threshold for performing extra parallelism. For
small systems, exposing additional parallism can be beneficial when
there is not enough work to fully saturate the GPU threads otherwise.
However, the extra parallelism also leads to more divergence
and can hurt performance when the system is already large enough to
saturate the GPU threads. Extra parallelism will be performed if the
*chunksize* (or total number of atoms per GPU) is smaller than
*parallelthresh*.
ignored otherwise. The *chunksize* keyword controls the number of atoms
in each pass used to compute the bispectrum components and is used to
avoid running out of memory. For example if there are 8192 atoms in the
simulation and the *chunksize* is set to 4096, the bispectrum
calculation will be broken up into two passes (running on a single GPU).
The *parallelthresh* keyword controls a crossover threshold for
performing extra parallelism. For small systems, exposing additional
parallelism can be beneficial when there is not enough work to fully
saturate the GPU threads otherwise. However, the extra parallelism also
leads to more divergence and can hurt performance when the system is
already large enough to saturate the GPU threads. Extra parallelism
will be performed if the *chunksize* (or total number of atoms per GPU)
is smaller than *parallelthresh*.
Detailed definitions for all the other keywords
are given on the :doc:`compute sna/atom <compute_sna_atom>` doc page.