tx258/lammps-gran-kokkos
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

updated the doc page for the new parameters

Browse Source
This commit is contained in:
Trung Nguyen
2025-06-15 15:20:56 -05:00
parent b3423e9b7a
commit b5809aff67
1 changed files with 21 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
doc/src/package.rst
View File

@ -123,6 +123,10 @@ Syntax
Nteamsize = # of threads per block used for the pair compute kernel
*nbin/atoms/per/bin = Natomsperbin
Natomsperbin = # of atoms per bin used for neighbor list builds
*nbor/block/size = blocksize
blocksize = # of GPU threads per block for the flat neighbor build method
*bond/block/size = blocksize
blocksize = # of GPU threads per block for the bond force computation
*omp* args = Nthreads keyword value ...
Nthreads = # of OpenMP threads to associate with each MPI process
zero or more keyword/value pairs may be appended
@ -592,14 +596,14 @@ keyword above.
The *gpu/aware* keyword chooses whether GPU-aware MPI will be used. When
this keyword is set to *on*, buffers in GPU memory are passed directly
through MPI send/receive calls. This reduces overhead of first copying
the data to the host CPU. However GPU-aware MPI is not supported on all
the data to the host CPU. However GPU-aware MPI is not supported on all
systems, which can lead to segmentation faults and would require using a
value of *off*\ . If LAMMPS can safely detect that GPU-aware MPI is not
available (currently only possible with OpenMPI v2.0.0 or later), then
the *gpu/aware* keyword is automatically set to *off* by default. When
the *gpu/aware* keyword is set to *off* while any of the *comm*
keywords are set to *device*, the value for these *comm* keywords will
be automatically changed to *no*\ . This setting has no effect if not
be automatically changed to *no*\ . This setting has no effect if not
running on GPUs or if using only one MPI rank. GPU-aware MPI is available
for OpenMPI 1.8 (or later versions), Mvapich2 1.9 (or later) when the
"MV2_USE_CUDA" environment variable is set to "1", CrayMPI, and IBM
@ -614,6 +618,10 @@ other force computations on the host CPU. The *comm* flags, along with the
This can result in better performance for certain configurations and
system sizes.
The following parameters allow users to tune the overall performance
depending on the simulated systems. If not explicitly specified,
their values will be set internally by the KOKKOS package.
The *threads/per/atom* keyword sets the number of GPU vector lanes per atom
used to perform force calculations. This keyword is only applicable
when *neigh/thread* is set to *on*. For large cutoffs or with a small number
@ -629,7 +637,17 @@ when *neigh/thread* is set to *on*. The default value of this parameter
is determined based on the GPU architecture at runtime.
The *nbin/atoms/per/bin* keyword sets the number of atoms per bin
used for the neighbor list builds on the GPU. The default value is 16.
used for the neighbor list builds on the GPU, which then determines
the number of GPU threads per bin. The default value of this parameter is 16.
The *nbor/block/size* keyword sets the number of GPU threads per block
used for the neighbor list builds on the GPU using the flat method (i.e.,
each thread finds the neighbor list of an atom). If not specified, then
the GPU threads are assigned to the bins.
The *bond/block/size* keyword sets the number of GPU threads per block
used for launching the bond force kernel on the GPU. The default value
of this parameter is determined based on the GPU architecture at runtime.
----------