LATTE is a semi-empirical tight-binding quantum code, developed
primarily at Los Alamos National Labs.
See these links:
https://www.osti.gov/biblio/1526907-los-alamos-transferable-tight-binding-energetics-latte-version
https://github.com/lanl/LATTE
LAMMPS has 2 ways of working with LATTE:
(1) Via its LATTE package and the fix latte command
must run LAMMPS on a single processor, it calls LATTE as a library
(2) Via its MDI package and the code-coupling MDI library
(a) can run LAMMPS and LATTE as stand-alone codes
LAMMPS can be run on any number of procs
LATTE must run on a single proc, but can use OpenMP
(b) can run LAMMPS with LATTE as a plug-in library
must run LAMMPS on a single processor
Examples for use case (1) are in the examples/latte dir. Use case (2)
is illustrated in this dir.
NOTE: If you compare MDI runs in this dir to similar fix latte runs in
examples/latte, the answers for energy and virial will be differnt.
This is b/c the version of LATTE used by the fix latte command within
the LATTE package is older than the version of LATTE used here.
------------------
Building 3 codes needed to run these examples
(1) Download and build MDI
% git clone git@github.com:MolSSI-MDI/MDI_Library.git mdi
% cd mdi
% mkdir build; cd build
% cmake .. # includes support for all langauges (incl Fortran, Python)
% make
(2) Download and build LATTE with MDI support
% git clone git@github.com:lanl/LATTE.git latte
% cd latte
% git checkout skimLATTE-progress # goto branch with MDI support
% cp makefiles/makefile.CHOICES.mdi makefile.CHOICES # so can now edit
% edit makefile.CHOICES settings to have these settings:
MAKELIB = OFF, SHARED = ON, MDI = ON
MDI_PATH must point to CMake build of MDI in (1),
e.g. /home/sjplimp/mdi/build/MDI_Library
comment out 2 LIB lines with CUDA-CUDART_LIBRARY
% make clean
% make # creates liblatte.so and LATTE_DOUBLE with support for MDI
(3) Build LAMMPS with its MDI package
also with the MOLECULE package for these example scripts
Build with traditional make
% cd lammps/lib/mdi
% python Install.py -m mpi # downloads and builds MDI
% cd ../../src
% make yes-mdi yes-molecule
$ make mpi # creates lmp_mpi
% mkdir build; cd build
% cmake -D PKG_MDI=yes -D PKG_MOLECULE=yes ../cmake
% make # creates lmp
Build with CMake
% cd lammps
% mkdir build; cd build
% cmake -D PKG_MDI=yes -D PKG_MOLECULE=yes ../cmake
% make # creates lmp
(4) Copy LAMMPS and LATTE executables into this dir
Copy the LAMMPS executable (lmp_mpi or lmp) into this dir as lmp_mpi.
Copy the LATTE executable LATTE_DOUBLE into this dir.
The run commands below assume you have done this.
(5) Insure LD_LIBRARY_PATH includes the dir where MDI was built in (1)
with its libmdi.so file, e.g. mdi/build/MDI_Library. This is needed
so when LATTE_DOUBLE runs as an executable it will able to find
libmdi.so.
------------------
Notes on LATTE usage
You must run this version of LATTE on a single MPI processor.
However, you can use OpenMP with LATTE. To do this you need to build
LATTE with OpenMP support by editing the makefile.CHOICES file to
include -fopenmp with FFLAGS and LINKFLAGS. Also -lapack and -lblas
need to be added to LIB, and those libraries must be available on your
system. For best performance you should also build LATTE with its
PROGRESS and BML libraries. Building those libs is more complex,
see details here:
https://github.com/lanl/LATTE_SUPER/tree/allMachines/Laptop
At run time, you need to also first set an environment variable for
the number of OpenMP threads to use, e.g.
% export OMP_NUM_THREADS=12
By default LATTE reads the latte.in file for its parameters. That
file specifies other files LATTE will read. With MDI, the driver code
(e.g. LAMMPS) can use the >FNAME command to specify an alternate
filename to use instead of latte.in.
By default LATTE writes out a log.latte file with info about its
calculations. An "OUTFILE= logfile" setting in latte.in can rename
this file.
---------
Run example #1: AIMD
* Run with MPI: 1 proc each
mpirun -np 1 lmp_mpi -mdi "-name LMP -role DRIVER -method MPI" \
-in in.aimd -log log.aimd.lammps.mpi : \
-np 1 LATTE_DOUBLE -mdi "-name LATTE -role ENGINE -method MPI"
* Run with MPI: 2 procs for LAMMPS, 1 for LATTE
mpirun -np 2 lmp_mpi -mdi "-name LMP -role DRIVER -method MPI" \
-in in.aimd -log log.aimd.lammps.mpi : \
-np 1 LATTE_DOUBLE -mdi "-name LATTE -role ENGINE -method MPI"
* Run in plugin mode: 1 proc
lmp_mpi -mdi \
"-name LMP -role DRIVER -method LINK -plugin_path /home/sjplimp/latte/git" \
-in in.aimd.plugin -log log.aimd.lammps.plugin
NOTE: The -plugin_path needs to point to where LATTE was built in step
(2).
---------
Run example #2: sequence of configurations
* Run with MPI: 1 proc each
mpirun -np 1 lmp_mpi -mdi "-name LMP -role DRIVER -method MPI" \
-in in.series -log log.series.lammps.mpi : \
-np 1 LATTE_DOUBLE -mdi "-name LATTE -role ENGINE -method MPI"
* Run with MPI: 2 procs for LAMMPS, 1 for LATTE
mpirun -np 2 lmp_mpi -mdi "-name LMP -role DRIVER -method MPI" \
-in in.series -log log.series.lammps.mpi : \
-np 1 LATTE_DOUBLE -mdi "-name LATTE -role ENGINE -method MPI"
* Run in plugin mode: 1 proc
lmp_mpi -mdi \
"-name LMP -role DRIVER -method LINK -plugin_path /home/sjplimp/latte/git" \
-in in.series.plugin -log log.series.lammps.plugin
NOTE: The -plugin_path needs to point to where LATTE was built in step
(2).
