LAMMPS

Overview

LAMMPS is a general-purpose molecular dynamics software package for massively parallel computers. It is written in an exceptionally clean style that makes it one of the most popular codes for users to extend and it currently has dozens of user-developed extensions.

LAMMPS was trivially ported to Blue Gene/Q shortly after the first installations came online and it has run on the entire Sequoia system using millions of MPI processes.

For details about the code and its usage, see the LAMMPS Home Page. This page is dedicated to information pertaining to Blue Gene/Q at the ALCF.

Using LAMMPS at ALCF

ALCF does not officially support LAMMPS software but provides a site installation user convenience.

How to Obtain the Code

LAMMPS is an open-source code which can be download at http://lammps.sandia.gov/download.html

Site Installation

On Mira, the Blue Gene/Q test system, the 17Nov2015 version of LAMMPS has been built with the XL compiler tool chain and is installed in /soft/applications/lammps/17Dec15/. Some older versions are also available and the Makefiles provided below can be used for local installations of LAMMPS with different user packages and/or with local modifications (with appropriate edits, as necessary).

Building on Blue Gene/Q

After  [LAMMPS] has been downloaded and g/unzipped, you should see a directory whose name is of the form lammps-<version>. Go to the directory lammps-<version>/src/MAKE/MACHINES and two files, as described below, will be found. The first is called Makefile.bgq.

# bgq = IBM Blue Gene/Q, multiple compiler options, native MPI, ALCF FFTW3

SHELL = /bin/bash
.SUFFIXES: .cpp .u

# ---------------------------------------------------------------------
# build rules and dependencies
# do not edit this section
# select which compiler by editing Makefile.bgq.details

include ../MAKE/MACHINES/Makefile.bgq.details

include Makefile.package.settings
include Makefile.package

EXTRA_INC = $(LMP_INC) $(PKG_INC) $(MPI_INC) $(FFT_INC) $(JPG_INC) $(PKG_SYSINC)
EXTRA_PATH = $(PKG_PATH) $(MPI_PATH) $(FFT_PATH) $(JPG_PATH) $(PKG_SYSPATH)
EXTRA_LIB = $(PKG_LIB) $(MPI_LIB) $(FFT_LIB) $(JPG_LIB) $(PKG_SYSLIB)
EXTRA_CPP_DEPENDS = $(PKG_CPP_DEPENDS)
EXTRA_LINK_DEPENDS = $(PKG_LINK_DEPENDS)

# Path to src files

vpath %.cpp ..
vpath %.h ..

# Link target

$(EXE): $(OBJ) $(EXTRA_LINK_DEPENDS)
        $(LINK) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(EXTRA_LIB) $(LIB) -o $(EXE)
        $(SIZE) $(EXE)

# Library targets

lib:    $(OBJ) $(EXTRA_LINK_DEPENDS)
        $(ARCHIVE) $(ARFLAGS) $(EXE) $(OBJ)

shlib:  $(OBJ) $(EXTRA_LINK_DEPENDS)
        $(CC) $(CCFLAGS) $(SHFLAGS) $(SHLIBFLAGS) $(EXTRA_PATH) -o $(EXE) \
        $(OBJ) $(EXTRA_LIB) $(LIB)

# Compilation rules

%.o:%.cpp $(EXTRA_CPP_DEPENDS)
        $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<

%.d:%.cpp $(EXTRA_CPP_DEPENDS)
        $(CC) $(CCFLAGS) $(EXTRA_INC) $(DEPFLAGS) $< > $@

%.o:%.cu $(EXTRA_CPP_DEPENDS)
        $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<

# Individual dependencies

depend : fastdep.exe $(SRC)
        @./fastdep.exe $(EXTRA_INC) -- $^ > .depend || exit 1

fastdep.exe: ../DEPEND/fastdep.c
        cc -0 -o $@ $<

sinclude .depend

The second file is called Makefile.bgq.details and contains specifics for each compiler (xl, gcc, and llvm).

# multiple compiler options for BGQ

# ---------------------------------------------------------------------
# compiler/linker settings
# specify flags and libraries needed for your compiler

# uncomment one and only one of the following three lines
# to choose a compiler toolchain

#COMPILER = GCC
#COMPILER = LLVM
COMPILER = XLC

ifeq ($(COMPILER),XLC)
CC       = mpixlcxx_r
CCFLAGS  = -O3 -qarch=qp -qtune=qp -qsmp=omp -qsimd=auto -qhot=level=2 -qprefetch -qunroll=yes
FC       = mpixlf90_r
FFLAGS   = -O3 -qarch=qp -qtune=qp -qsimd=auto -qhot=level=2 -qprefetch -qunroll=yes -qsmp=omp -qextname -qnosave
DEPFLAGS = -M -qmakedep=gcc
FFTW     = /soft/libraries/alcf/current/xl/FFTW3
endif

ifeq ($(COMPILER),GCC)
CC       = mpicxx
CCFLAGS  = -O3 -fopenmp
FC       = mpif90
FFLAGS   = -O3 -fopenmp
DEPFLAGS = -M
FFTW     = /soft/libraries/alcf/current/gcc/FFTW3
endif

ifeq ($(COMPILER),LLVM)
CC       = mpiclang++
CCFLAGS  = -O3 -fopenmp
DEPFLAGS = -M
FC       = /bin/false
FFLAGS   = LLVM does not have a Fortran front-end!
FFTW     = /soft/libraries/alcf/current/xl/FFTW3
endif

LINK      = $(CC)
LINKFLAGS = $(CCFLAGS)

ifeq ($(COMPILER),XLC)
  MASS_LIB    = ${IBM_MAIN_DIR}/xlmass/bg/7.3/bglib64
  XLF_LIB     = ${IBM_MAIN_DIR}/xlf/bg/14.1/bglib64
  XLSMP_LIB   = ${IBM_MAIN_DIR}/xlsmp/bg/3.1/bglib64
  LIB        += -L${MASS_LIB} -L${XLF_LIB} -L${XLSMP_LIB}
  LIB        += -lmassv -lmass
  LIB        += -lxlf90_r -lxlsmp -lxlopt -lxlfmath -lxl
endif

ifeq ($(COMPILER),GCC)
# libm is definitely slower than libmass...
  LIB += -lm -lgfortran
endif

ifeq ($(COMPILER),LLVM)
    SLEEF_DIR = /home/projects/llvm/sleef
    LIB += -L${SLEEF_DIR}/lib -lsleef
endif

SIZE       = size

ARCHIVE    = ar
ARFLAGS    = -rc

# BGQ should not use shared libraries

SHFLAGS    =
SHLIBFLAGS =

# ---------------------------------------------------------------------
# LAMMPS-specific settings, all OPTIONAL
# specify settings for LAMMPS features you will use
# if you change any -D setting, do full re-compile after "make clean"

# LAMMPS ifdef settings
# see possible settings in Section 2.2 (step 4) of manual

LMP_INC = -DLAMMPS_GZIP

# MPI library
# see discussion in Section 2.2 (step 5) of manual
# MPI wrapper compiler/linker can provide this info
# can point to dummy MPI library in src/STUBS as in Makefile.serial
# use -D MPICH and OMPI settings in INC to avoid C++ lib conflicts
# INC = path for mpi.h, MPI compiler settings
# PATH = path for MPI library
# LIB = name of MPI library

MPI_INC    =
MPI_PATH   =
MPI_LIB    =

MPI_INC += -DMPICH_SKIP_MPICXX
MPI_LIB +=

# FFT library
# see discussion in Section 2.2 (step 6) of manaul
# can be left blank to use provided KISS FFT library
# INC = -DFFT setting, e.g. -DFFT_FFTW, FFT compiler settings
# PATH = path for FFT library
# LIB = name of FFT library

FFT_INC  = -I$(FFTW)/include -DFFT_FFTW3 -DFFT_SINGLE
FFT_PATH = -L$(FFTW)/lib
FFT_LIB  = -lfftw3f

# JPEG and/or PNG library
# see discussion in Section 2.2 (step 7) of manual
# only needed if -DLAMMPS_JPEG or -DLAMMPS_PNG listed with LMP_INC
# INC = path(s) for jpeglib.h and/or png.h
# PATH = path(s) for JPEG library and/or PNG library
# LIB = name(s) of JPEG library and/or PNG library

JPG_INC  =       
JPG_PATH =
JPG_LIB  =

depend : fastdep.exe $(SRC)
        @./fastdep.exe $(EXTRA_INC) -- $^ > .depend || exit 1

fastdep.exe: ../DEPEND/fastdep.c
        cc -0 -o $@ $<

sinclude .depend

The compiler of choice can be selected by uncommenting the corresponding line at the top of Makefile.bgq.details and updating your .soft file to include the appropriate mpiwrapper (e.g. +mpiwrapper-xl.legacy.ndebug). LAMMPS can then be compiled from the lammps-<version>/src directory using "make bgq -j 8".

Running Jobs on Blue Gene/Q

The following is a valid invocation of LAMMPS on the ALCF Blue Gene/Q systems, provided that this is the input file.

   qsub -t 1:00:00 -n 1024 -A project_name --mode c32 ./lmp_bgq -in in_274.MS -log log.lammps

This is a valid invocation of a LAMMPS build with user-omp included.

   qsub -t 1:00:00 -n 1024 -A project_name --mode c16 --env OMP_NUM_THREADS=4 ./lmp_bgq -sf omp -in in_274.MS -log log.lammps

Performance on Blue Gene/Q

All timings here are for the following compilation of the 6Dec2012 version of the LAMMPS source code without modifications.

For comparative performance for other systems, see http://lammps.sandia.gov/bench.html#billion.

   make yes-user-omp no-gpu bgq -j16

A Makefile equivalent to the one above for the XLC compiler was used for these runs.

Lennard-Jones Melt

Input File

This is a modified version of the LAMMPS benchmark in.lj that has 32 million atoms instead of 32 thousand.

   # 3d Lennard-Jones melt

   variable	x index 1
   variable	y index 1
   variable	z index 1

   variable	xx equal 200*$x
   variable	yy equal 200*$y
   variable	zz equal 200*$z

   units		lj
   atom_style	atomic

   lattice		fcc 0.8442
   region		box block 0 ${xx} 0 ${yy} 0 ${zz}
   create_box	1 box
   create_atoms	1 box
   mass		1 1.0

   velocity	all create 1.44 87287 loop geom

   pair_style	lj/cut 2.5
   pair_coeff	1 1 1.0 1.0 2.5

   neighbor	0.3 bin
   neigh_modify	delay 0 every 20 check no

   fix		1 all nve

   run		10000

Job Submission

Jobs were submitted as follows:

   qsub -t 30 -n ${N} --mode c${C} --env OMP_NUM_THREADS=${T} \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.lj2 \
   -log lj2.xl-omp.n${N}.c${C}.omp${T}.log.scaling 

Timing Data

Output timings are negligible (less than 0.001% of the total) and are not reported in this table.

 (Nodes,Cores,Threads)

Pair

Neigh

Comm

Other

(128,16,1)

903.985

136.865

58.5272

37.3184

(128,16,2)

591.522

88.5962

53.0196

38.0991

(128,16,4)

406.937

55.3114

47.0328

46.4155

(128,32,1)

586.601

88.6527

38.3137

20.3007

(128,32,2)

383.127

53.891

38.7898

22.2251

(128,64,1)

383.198

55.4978

38.4279

13.3232

(256,16,1)

456.906

69.0528

33.8805

18.8491

(256,16,2)

291.551

43.9156

32.8679

19.5998

(256,16,4)

197.754

29.1659

32.0378

21.8867

(256,32,1)

294.855

44.1746

30.8948

10.0103

(256,32,2)

187.76

27.0357

29.0489

10.863

(256,64,1)

183.973

26.9979

27.8938

6.43218

(512,16,1)

232.018

34.7407

22.766

9.28841

(512,16,2)

146.897

22.2967

19.1644

9.57644

(512,16,4)

96.4826

13.4836

16.7823

10.1924

(512,32,1)

145.837

22.2908

19.5539

5.09412

(512,32,2)

92.8522

13.4444

18.5853

5.5271

(512,64,1)

92.2212

13.3299

24.1934

3.27035

 

Lennard-Jones Melt (bigger)

One can compare these results to the data here: http://lammps.sandia.gov/bench.html#billion.

Input File

This is a modified version of the included LAMMPS benchmark in.lj that has many more atoms. In the input file below, N=512, 1024, and 2048.

# 3d Lennard-Jones melt

variable	x index 1
variable	y index 1
variable	z index 1

variable	xx equal N*$x
variable	yy equal N*$y
variable	zz equal N*$z

units		lj
atom_style	atomic

lattice		fcc 0.8442
region		box block 0 ${xx} 0 ${yy} 0 ${zz}
create_box	1 box
create_atoms	1 box
mass		1 1.0

velocity	all create 1.44 87287 loop geom

pair_style	lj/cut 2.5
pair_coeff	1 1 1.0 1.0 2.5

neighbor	0.3 bin
neigh_modify	delay 0 every 20 check no

fix		1 all nve

run		1000

Job Submission

Jobs were submitted as follows:

   #!/bin/bash

   INPUT=lj512

   CORES=16
   THREADS=4
   for n in 1024 2048 4096 ; do 
    qsub -t 60 -n $n --mode c${CORES} --env OMP_NUM_THREADS=${THREADS} \
    /home/jhammond/LAMMPS/lammps-6Dec12/src/lmp_bgq \
    -sf omp -in in.${INPUT} -log ${INPUT}.xl-omp.n$n.c${CORES}.omp${THREADS}.log.scaling
   done

   CORES=32
   THREADS=2
   for n in 1024 2048 4096 ; do 
    qsub -t 60 -n $n --mode c${CORES} --env OMP_NUM_THREADS=${THREADS} \
    /home/jhammond/LAMMPS/lammps-6Dec12/src/lmp_bgq \
    -sf omp -in in.${INPUT} -log ${INPUT}.xl-omp.n$n.c${CORES}.omp${THREADS}.log.scaling
   done

   CORES=64
   THREADS=1
   for n in 1024 2048 4096 ; do 
    qsub -t 60 -n $n --mode c${CORES} --env OMP_NUM_THREADS=${THREADS} \
    /home/jhammond/LAMMPS/lammps-6Dec12/src/lmp_bgq \
    -sf omp -in in.${INPUT} -log ${INPUT}.xl-omp.n$n.c${CORES}.omp${THREADS}.log.scaling
   done

Timing Data

Output timings are negligible (0.01% or less) and are not reported in this table.

Below is data for half a billion atoms:

Atoms

(Nodes,Cores,Threads)

Pair

Neigh

Comm

Other

536,870,912

(1024,16,4)

83.5958

11.2556

10.7224

9.34939

536,870,912

(1024,32,2)

81.0333

10.9222

9.82558

4.74534

536,870,912

(1024,64,1)

82.1304

11.1569

8.84917

3.29235

536,870,912

(2048,16,4)

40.7869

5.55647

6.7153

4.73016

536,870,912

(2048,32,2)

39.836

5.44656

4.94552

2.37361

536,870,912

(2048,64,1)

40.5356

5.61214

5.78408

1.60949

536,870,912

(4096,16,4)

20.848

2.83956

4.10531

2.16598

536,870,912

(4096,32,2)

19.5241

2.7665

3.80719

1.12133

536,870,912

(4096,64,1)

19.6054

3.00689

3.7096

0.67681

 

Below is data for four billion atoms:

Atoms

(Nodes,Cores,Threads)

Pair

Neigh

Comm

Other

4,294,967,296

(1024,16,4)

707.573

92.2235

62.9546

83.083

4,294,967,296

(1024,32,2)

671.345

90.9449

58.0772

41.7037

4,294,967,296

(1024,64,1)

703.872

94.9056

72.8057

22.2946

4,294,967,296

(2048,16,4)

337.606

45.5278

38.1299

36.5494

4,294,967,296

(2048,32,2)

337.402

45.0738

36.9501

18.2013

4,294,967,296

(2048,64,1)

347.444

48.9484

37.7461

11.4642

4,294,967,296

(4096,16,4)

169.747

22.7482

25.6915

17.1804

4,294,967,296

(4096,32,2)

165.149

22.0062

20.3155

8.97527

4,294,967,296

(4096,64,1)

167.666

24.7743

19.3463

5.89172

 

Below is data for 34 billion atoms:

Atoms

(Nodes,Cores,Threads)

Total

Pair

Neigh

Comm

Other

34,359,738,368

(2048,32,2)

3402.54

2691.24

354.89

204.005

152.142

34,359,738,368

(4096,16,4)

1863.18

1392.4

183.232

149.22

138.112

34,359,738,368

(4096,32,2)

1751.78

1372.02

179.872

125.818

73.9034

34,359,738,368

(4096,64,1)

1747.83

1366.11

191.007

140.985

49.3224

 

Copper Lattice

Input File

This is a modified version of the included LAMMPS benchmark in.eam that has 32 million atoms instead of 32 thousand.

   # bulk Cu lattice

   variable	x index 1
   variable	y index 1
   variable	z index 1

   variable	xx equal 200*$x
   variable	yy equal 200*$y
   variable	zz equal 200*$z

   units		metal
   atom_style	atomic

   lattice		fcc 3.615
   region		box block 0 ${xx} 0 ${yy} 0 ${zz}
   create_box	1 box
   create_atoms	1 box

   pair_style	eam
   pair_coeff	1 1 Cu_u3.eam

   velocity	all create 1600.0 376847 loop geom

   neighbor	1.0 bin
   neigh_modify    every 1 delay 5 check yes

   fix		1 all nve

   timestep	0.005
   thermo		50

   run		100

Job Submission

Jobs were submitted as follows:

   #!/bin/bash
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c16 --env OMP_NUM_THREADS=4 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.eam2 -log eam2.xl-omp.n$n.c16.omp4.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c32 --env OMP_NUM_THREADS=2 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.eam2 -log eam2.xl-omp.n$n.c32.omp2.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c64 --env OMP_NUM_THREADS=1 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.eam2 -log eam2.xl-omp.n$n.c64.omp1.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c32 --env OMP_NUM_THREADS=1 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.eam2 -log eam2.xl-omp.n$n.c32.omp1.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c16 --env OMP_NUM_THREADS=1 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.eam2 -log eam2.xl-omp.n$n.c16.omp1.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c16 --env OMP_NUM_THREADS=2 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.eam2 -log eam2.xl-omp.n$n.c16.omp2.log.scaling ; done

Timing Data

Output timings are negligible (less than 0.1% of the total) and are not reported in this table.

(Nodes,Cores,Threads)

Pair

Neigh

Comm

Other

(128,16,1)

33.6279

3.88927

0.87631

0.450641

(128,16,2)

19.7062

2.51144

0.717301

0.470908

(128,16,4)

11.9051

1.5783

0.714246

0.534397

(128,32,1)

20.8353

2.49203

0.663389

0.275988

(128,32,2)

12.2091

1.51635

0.624944

0.303134

(128,64,1)

13.4671

1.58481

0.648977

0.222134

(256,16,1)

16.9983

1.99589

0.560821

0.241915

(256,16,2)

9.97638

1.26302

0.458407

0.243051

(256,16,4)

6.02135

0.79092

0.439217

0.2724

(256,32,1)

10.3428

1.23439

0.440537

0.176839

(256,32,2)

6.12204

0.764731

0.386978

0.149776

(256,64,1)

6.39477

0.746718

0.415273

0.120801

(512,16,1)

8.48893

0.974289

0.361599

0.156456

(512,16,2)

5.04149

0.631331

0.288922

0.1471

(512,16,4)

3.09269

0.391045

0.258176

0.13799

(512,32,1)

5.15135

0.61045

0.275696

0.0958401

(512,32,2)

3.06

0.379021

0.241103

0.0908437

(512,64,1)

3.28059

0.363596

0.30915

0.104848

 

Rhodopsin Model

Input File

The input file was not modified in.rhodo for these tests.

Job Submission

   for n in 1 8 16 32 64 ; do qsub -t 60 -n $n --mode c64 --env OMP_NUM_THREADS=1 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.rhodo -log rhodo.xl-omp.n$n.c64.omp1.log.scaling ; done

Timing Data

Output timings are negligible (less than 0.5% of the total) and are not reported in this table.

(Nodes,Cores,Threads)

Pair

Bond

Kspce

Neigh

Comm

Other

(1,64,1)

6.64919

0.219783

1.87901

1.20142

0.217869

0.390536

(8,64,1)

0.91955

0.0325569

0.885943

0.164351

0.155724

0.132105

(16,64,1)

0.4807

0.0190747

0.792541

0.0819856

0.143043

0.114512

(32,64,1)

0.252456

0.0116682

0.686376

0.0427477

0.133941

0.102508

(64,64,1)

0.139422

0.00776608

0.644641

0.0235872

0.136908

0.103329

 

Tersoff Potential for Silicon

This benchmark data was provided by John Russell.

Input File

   # bulk Si via Tersoff

   units	metal
   atom_style	atomic

   lattice	diamond 5.431
   region	box block 0 200 0 200 0 200
   create_box	1 box
   create_atoms	1 box

   pair_style	tersoff
   pair_coeff	* * Si.tersoff Si
   mass         1 28.06

   velocity	all create 1000.0 376847 loop geom

   neighbor	1.0 bin
   neigh_modify delay 5 every 1

   fix		1 all nve

   timestep	0.001

   run		100

Job Submission

   #!/bin/bash
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c16 --env OMP_NUM_THREADS=4 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.tersoff -log tersoff.xl-omp.n$n.c16.omp4.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c32 --env OMP_NUM_THREADS=2 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.tersoff -log tersoff.xl-omp.n$n.c32.omp2.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c64 --env OMP_NUM_THREADS=1 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.tersoff -log tersoff.xl-omp.n$n.c64.omp1.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c32 --env OMP_NUM_THREADS=1 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.tersoff -log tersoff.xl-omp.n$n.c32.omp1.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c16 --env OMP_NUM_THREADS=1 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.tersoff -log tersoff.xl-omp.n$n.c16.omp1.log.scaling ; done
   for n in 128 256 512 ; do qsub -t 30 -n $n --mode c16 --env OMP_NUM_THREADS=2 \
   /soft/applications/lammps/6Dec12/xl-omp/lmp_bgq -sf omp -in in.tersoff -log tersoff.xl-omp.n$n.c16.omp2.log.scaling ; done

Timing Data

Output timings are negligible (less than 0.5% of the total) and are not reported in this table.

(Nodes,Cores,Threads)

Total

Pair

Neigh

Comm

Other

(128,16,1)

120.395

115.822

0.955202

2.62538

0.98920

(128,16,2)

69.0233

65.1012

0.680769

2.07471

1.16119

(128,16,4)

48.8062

46.0290

0.423408

1.24533

1.10657

(128,32,1)

70.7485

67.1883

0.655726

2.19065

0.71006

(128,32,2)

50.6281

47.7650

0.403046

1.60761

0.84739

(128,64,1)

56.5516

53.3761

0.413241

2.12397

0.63501

(256,16,1)

60.3178

57.7389

0.481374

1.59623

0.49934

(256,16,2)

34.4745

32.4862

0.321297

1.08914

0.57580

(256,16,4)

24.6075

23.0436

0.205511

0.79283

0.56431

(256,32,1)

36.4726

34.3706

0.327798

1.36564

0.40637

(256,32,2)

25.6327

23.8550

0.200893

1.02836

0.54710

(256,64,1)

29.5581

27.6274

0.205788

1.37371

0.34871

(512,16,1)

30.4686

28.9168

0.240024

1.01865

0.29159

(512,16,2)

17.3410

16.2272

0.161762

0.65465

0.29597

(512,16,4)

12.4033

11.5303

0.102741

0.48994

0.27932

(512,32,1)

18.4566

17.1675

0.165885

0.87474

0.24626

(512,32,2)

13.0919

12.0992

0.101187

0.63272

0.25748

(512,64,1)

15.3649

13.9857

0.099590

0.93050

0.34516

 

Building on Blue Gene/P

See the Blue Gene/Q section and modify the paths as appropriate.

Ensembles

An efficient implementation of the ensemble jobs using the LAMMPS library interface, including support for parallel PLUMED jobs, was developed by Luke Westby, Mladen Rasic, Adrian Lange, and Jeff Hammond in the summer of 2012.

The use of this code will be documented here as it requires minimal patching of the LAMMPS source. The non-public Subversion repository is located at https://repo.anl-external.org/repos/lammps-ensembles/. For access to the code, contact Jeff Hammond.