Building LAMMPS with FFTW

Sachin Shanbhag
488 DSL, sshanbhag@fsu.edu

sandia.gov
 When should you use this tutorial?
● You use LAMMPS as you MD program
● You have to consider long range Coulombic interaction
● You don't know where your fft files are OR
● You don't have permissions to access them OR
● You simply want to create a fftw library in your local
account
 Download LAMMPS
● Get the program from
http://lammps.sandia.gov/download.html
● Select C++, and hit download now
● This gets you lammps.tar.gz
● Untar-zip it using the command in an appropriate
directory
tar -zxvf lammps.tar.gz
● mv lammps-xxx to lammps, for simplicity
 Download FFTW
● Download the 2.1.x version of fftw from fftw.org
● Untar/zip the downloaded file (I downloaded v2.1.5)
tar -zxvf fftw-2.1.5.tar.gz
● make a temporary directory called localfft
● cd fftw-2.1.5/
● In fftw-2.1.5/ type
./configure --prefix=[ABSPATH]/localfft
 Build FFTW
● Here [ABSPATH]is the absolute path to this directory.
– This really depends upon where you want to place the library in
your local account
● In my case it is, for now (I'll move it later),
/panfs/panasas1/users/sachins/Desktop/
● Type the following commands to build the fftw library
make; make install
● This will make three subdirectories in the localfft
directory
● I moved the localfft directory into my
lammps/src/ directory
 Building the LAMMPS Executable
● Detailed instructions available at:
http://lammps.sandia.gov/doc/Section_sta
rt.html#2_2
● Right now, lets do a simple compilation:
– single processor, no MPI
– with freely available FFTW library
● cd lammps/src/STUBS/
– make (builds libmpi.a)
● cd lammps/src/MAKE/
– comment out/delete LINKFORT, FORTLIB, and SYSLIB in
Makefile.g++
 Building the Executable
● The only part I changed in Makefile.g++ was
CC = g++
CCFLAGS = -g -DFFT_FFTW -I../STUBS \
-I../localfft/include $(PKGINC)
DEPFLAGS = -M
LINK = g++
LINKFLAGS = -O -L../STUBS -L../localfft/lib $(PKGPATH)
USRLIB = -lfftw -lmpi $(PKGLIB)
ARCHIVE = ar
ARFLAGS = -rc
SIZE = size

● cd lammps/src/
– make yes-kspace
– make g++
● You should have an executable called lmp_g++
 Testing if it works
● I used the peptide example in the examples
directory
– It needs fftw to work
● cd lammps/examples/peptide
 Testing if it works
sachins% cat in.peptide
# Solvated 5-mer peptide

units real
atom_style full

pair_style lj/charmm/coul/long 8.0

10.0 10.0
bond_style harmonic
angle_style charmm
dihedral_style charmm
improper_style harmonic
kspace_style pppm 0.0001

...
 Run the program
sachins% lmp_g++ < in.peptide
LAMMPS (7 Jul 2009)
Scanning data file ...
3 = max bonds/atom
6 = max angles/atom
14 = max dihedrals/atom
1 = max impropers/atom

<blah> <blah>

PPPM initialization …
G vector = 0.268721
grid = 15 15 15
stencil order = 5
RMS precision = 4.86052e-05
brick FFT buffer size/proc = 10648 3375 5808
Setting up run …

<blah> <blah>