diff --git a/bench/GPU/README b/bench/FERMI/README
similarity index 81%
rename from bench/GPU/README
rename to bench/FERMI/README
index 03aeaf8ef9..5244aa8916 100644
--- a/bench/GPU/README
+++ b/bench/FERMI/README
@@ -1,7 +1,9 @@
-These are input scripts used to run GPU versions of several of the
-benchmarks in the top-level bench directory. The results of running
-these scripts on different machines are shown on the GPU section of
-the Benchmark page of the LAMMPS WWW site (lammps.sandia.gov/bench).
+These are input scripts used to run versions of several of the
+benchmarks in the top-level bench directory using the GPU and
+USER-CUDA accelerator packages. The results of running these scripts
+on two different machines (a desktop with 2 Tesla GPUs and the ORNL
+Titan supercomputer) are shown on the "GPU (Fermi)" section of the
+Benchmark page of the LAMMPS WWW site: lammps.sandia.gov/bench.
Examples are shown below of how to run these scripts. This assumes
you have built 3 executables with both the GPU and USER-CUDA packages
@@ -59,3 +61,8 @@ many MPI tasks per compute node the problem will run on, and the "g"
setting determines how many GPUs per compute node the problem will run
on, i.e. 1 or 2 in this case. For the USER-CUDA package, the number
of MPI tasks and GPUs (both per compute node) must be equal.
+
+------------------------------------------------------------------------
+
+If the script has "titan" in its name, it was run on the Titan supercomputer
+at ORNL.
diff --git a/bench/GPU/in.eam.cpu b/bench/FERMI/in.eam.cpu
similarity index 100%
rename from bench/GPU/in.eam.cpu
rename to bench/FERMI/in.eam.cpu
diff --git a/bench/GPU/in.eam.cuda b/bench/FERMI/in.eam.cuda
similarity index 100%
rename from bench/GPU/in.eam.cuda
rename to bench/FERMI/in.eam.cuda
diff --git a/bench/GPU/in.eam.gpu b/bench/FERMI/in.eam.gpu
similarity index 100%
rename from bench/GPU/in.eam.gpu
rename to bench/FERMI/in.eam.gpu
diff --git a/bench/FERMI/in.eam.titan b/bench/FERMI/in.eam.titan
new file mode 100644
index 0000000000..871e5df4d0
--- /dev/null
+++ b/bench/FERMI/in.eam.titan
@@ -0,0 +1,37 @@
+# bulk Cu lattice
+
+newton off
+package gpu force/neigh 0 0 1
+processors * * * grid numa
+
+variable x index 1
+variable y index 1
+variable z index 1
+
+variable xx equal 20*$x
+variable yy equal 20*$y
+variable zz equal 20*$z
+
+units metal
+atom_style atomic
+
+lattice fcc 3.615
+region box block 0 ${xx} 0 ${yy} 0 ${zz}
+create_box 1 box
+create_atoms 1 box
+
+pair_style eam/gpu
+pair_coeff 1 1 Cu_u3.eam
+
+velocity all create 1600.0 376847 loop geom
+
+neighbor 1.0 bin
+neigh_modify every 1 delay 5 check yes
+
+fix 1 all nve
+
+timestep 0.005
+thermo 50
+
+run 15
+run 100
diff --git a/bench/GPU/in.lj.cpu b/bench/FERMI/in.lj.cpu
similarity index 100%
rename from bench/GPU/in.lj.cpu
rename to bench/FERMI/in.lj.cpu
diff --git a/bench/GPU/in.lj.cuda b/bench/FERMI/in.lj.cuda
similarity index 100%
rename from bench/GPU/in.lj.cuda
rename to bench/FERMI/in.lj.cuda
diff --git a/bench/GPU/in.lj.gpu b/bench/FERMI/in.lj.gpu
similarity index 100%
rename from bench/GPU/in.lj.gpu
rename to bench/FERMI/in.lj.gpu
diff --git a/bench/FERMI/in.lj.titan b/bench/FERMI/in.lj.titan
new file mode 100644
index 0000000000..2b2e8b11e5
--- /dev/null
+++ b/bench/FERMI/in.lj.titan
@@ -0,0 +1,35 @@
+# 3d Lennard-Jones melt
+
+newton off
+package gpu force/neigh 0 0 1
+processors * * * grid numa
+
+variable x index 1
+variable y index 1
+variable z index 1
+
+variable xx equal 20*$x
+variable yy equal 20*$y
+variable zz equal 20*$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/gpu 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 15
+run 100
diff --git a/bench/GPU/in.rhodo.cpu b/bench/FERMI/in.rhodo.cpu
similarity index 100%
rename from bench/GPU/in.rhodo.cpu
rename to bench/FERMI/in.rhodo.cpu
diff --git a/bench/GPU/in.rhodo.cuda b/bench/FERMI/in.rhodo.cuda
similarity index 100%
rename from bench/GPU/in.rhodo.cuda
rename to bench/FERMI/in.rhodo.cuda
diff --git a/bench/GPU/in.rhodo.gpu b/bench/FERMI/in.rhodo.gpu
similarity index 100%
rename from bench/GPU/in.rhodo.gpu
rename to bench/FERMI/in.rhodo.gpu
diff --git a/bench/FERMI/in.rhodo.scaled.titan b/bench/FERMI/in.rhodo.scaled.titan
new file mode 100644
index 0000000000..00ee7eaefb
--- /dev/null
+++ b/bench/FERMI/in.rhodo.scaled.titan
@@ -0,0 +1,39 @@
+# Rhodopsin model
+
+newton off
+package gpu force/neigh 0 0 1
+processors * * * grid numa
+
+variable x index 1
+variable y index 1
+variable z index 1
+
+units real
+neigh_modify delay 5 every 1
+
+atom_style full
+atom_modify map hash
+bond_style harmonic
+angle_style charmm
+dihedral_style charmm
+improper_style harmonic
+pair_style lj/charmm/coul/long/gpu 8.0 ${cutoff}
+pair_modify mix arithmetic
+kspace_style pppm/gpu 1e-4
+
+read_data data.rhodo
+
+replicate $x $y $z
+
+fix 1 all shake 0.0001 5 0 m 1.0 a 232
+fix 2 all npt temp 300.0 300.0 100.0 &
+ z 0.0 0.0 1000.0 mtk no pchain 0 tchain 1
+
+special_bonds charmm
+
+thermo 50
+# thermo_style multi
+timestep 2.0
+
+run 15
+run 100
diff --git a/bench/FERMI/in.rhodo.split.titan b/bench/FERMI/in.rhodo.split.titan
new file mode 100644
index 0000000000..83c6ef4eaa
--- /dev/null
+++ b/bench/FERMI/in.rhodo.split.titan
@@ -0,0 +1,42 @@
+# Rhodopsin model
+
+newton off
+package gpu force/neigh 0 0 1
+partition yes 1 processors * * * grid twolevel ${grid} * * * &
+ part 1 2 multiple
+partition yes 2 processors * * * part 1 2 multiple
+
+variable x index 1
+variable y index 1
+variable z index 1
+
+units real
+neigh_modify delay 5 every 1
+
+atom_style full
+atom_modify map hash
+bond_style harmonic
+angle_style charmm
+dihedral_style charmm
+improper_style harmonic
+pair_style lj/charmm/coul/long/gpu 8.0 ${cutoff}
+pair_modify mix arithmetic
+kspace_style pppm/gpu 1e-4
+
+read_data data.rhodo
+
+replicate $x $y $z
+
+fix 1 all shake 0.0001 5 0 m 1.0 a 232
+fix 2 all npt temp 300.0 300.0 100.0 &
+ z 0.0 0.0 1000.0 mtk no pchain 0 tchain 1
+
+special_bonds charmm
+
+thermo 50
+# thermo_style multi
+timestep 2.0
+
+run_style verlet/split
+run 15
+run 100
diff --git a/bench/KEPLER/Makefile.cpu b/bench/KEPLER/Makefile.cpu
new file mode 100755
index 0000000000..f27d7e9301
--- /dev/null
+++ b/bench/KEPLER/Makefile.cpu
@@ -0,0 +1,108 @@
+# linux = Shannon Linux box, Intel icc, OpenMPI, KISS FFTW
+
+SHELL = /bin/sh
+
+# ---------------------------------------------------------------------
+# compiler/linker settings
+# specify flags and libraries needed for your compiler
+
+CC = icc
+CCFLAGS = -O
+SHFLAGS = -fPIC
+DEPFLAGS = -M
+
+LINK = icc
+LINKFLAGS = -O
+LIB = -lstdc++
+SIZE = size
+
+ARCHIVE = ar
+ARFLAGS = -rc
+SHLIBFLAGS = -shared
+
+# ---------------------------------------------------------------------
+# LAMMPS-specific settings
+# specify settings for LAMMPS features you will use
+# if you change any -D setting, do full re-compile after "make clean"
+
+# LAMMPS ifdef settings, OPTIONAL
+# see possible settings in doc/Section_start.html#2_2 (step 4)
+
+LMP_INC =
+
+# MPI library, REQUIRED
+# see discussion in doc/Section_start.html#2_2 (step 5)
+# can point to dummy MPI library in src/STUBS as in Makefile.serial
+# INC = path for mpi.h, MPI compiler settings
+# PATH = path for MPI library
+# LIB = name of MPI library
+
+MPI_INC = -I/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/include/
+MPI_PATH = -L/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/lib
+MPI_LIB = -lmpi
+
+# FFT library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 6)
+# 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 =
+FFT_PATH =
+FFT_LIB =
+
+# JPEG and/or PNG library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 7)
+# 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 = -ljpeg
+
+# ---------------------------------------------------------------------
+# build rules and dependencies
+# no need to edit this section
+
+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) $(PKG_SYSLIB)
+
+# Path to src files
+
+vpath %.cpp ..
+vpath %.h ..
+
+# Link target
+
+$(EXE): $(OBJ)
+ $(LINK) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(EXTRA_LIB) $(LIB) -o $(EXE)
+ $(SIZE) $(EXE)
+
+# Library targets
+
+lib: $(OBJ)
+ $(ARCHIVE) $(ARFLAGS) $(EXE) $(OBJ)
+
+shlib: $(OBJ)
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(SHLIBFLAGS) $(EXTRA_PATH) -o $(EXE) \
+ $(OBJ) $(EXTRA_LIB) $(LIB)
+
+# Compilation rules
+
+%.o:%.cpp
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<
+
+%.d:%.cpp
+ $(CC) $(CCFLAGS) $(EXTRA_INC) $(DEPFLAGS) $< > $@
+
+# Individual dependencies
+
+DEPENDS = $(OBJ:.o=.d)
+sinclude $(DEPENDS)
diff --git a/bench/KEPLER/Makefile.cuda b/bench/KEPLER/Makefile.cuda
new file mode 100755
index 0000000000..f27d7e9301
--- /dev/null
+++ b/bench/KEPLER/Makefile.cuda
@@ -0,0 +1,108 @@
+# linux = Shannon Linux box, Intel icc, OpenMPI, KISS FFTW
+
+SHELL = /bin/sh
+
+# ---------------------------------------------------------------------
+# compiler/linker settings
+# specify flags and libraries needed for your compiler
+
+CC = icc
+CCFLAGS = -O
+SHFLAGS = -fPIC
+DEPFLAGS = -M
+
+LINK = icc
+LINKFLAGS = -O
+LIB = -lstdc++
+SIZE = size
+
+ARCHIVE = ar
+ARFLAGS = -rc
+SHLIBFLAGS = -shared
+
+# ---------------------------------------------------------------------
+# LAMMPS-specific settings
+# specify settings for LAMMPS features you will use
+# if you change any -D setting, do full re-compile after "make clean"
+
+# LAMMPS ifdef settings, OPTIONAL
+# see possible settings in doc/Section_start.html#2_2 (step 4)
+
+LMP_INC =
+
+# MPI library, REQUIRED
+# see discussion in doc/Section_start.html#2_2 (step 5)
+# can point to dummy MPI library in src/STUBS as in Makefile.serial
+# INC = path for mpi.h, MPI compiler settings
+# PATH = path for MPI library
+# LIB = name of MPI library
+
+MPI_INC = -I/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/include/
+MPI_PATH = -L/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/lib
+MPI_LIB = -lmpi
+
+# FFT library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 6)
+# 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 =
+FFT_PATH =
+FFT_LIB =
+
+# JPEG and/or PNG library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 7)
+# 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 = -ljpeg
+
+# ---------------------------------------------------------------------
+# build rules and dependencies
+# no need to edit this section
+
+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) $(PKG_SYSLIB)
+
+# Path to src files
+
+vpath %.cpp ..
+vpath %.h ..
+
+# Link target
+
+$(EXE): $(OBJ)
+ $(LINK) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(EXTRA_LIB) $(LIB) -o $(EXE)
+ $(SIZE) $(EXE)
+
+# Library targets
+
+lib: $(OBJ)
+ $(ARCHIVE) $(ARFLAGS) $(EXE) $(OBJ)
+
+shlib: $(OBJ)
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(SHLIBFLAGS) $(EXTRA_PATH) -o $(EXE) \
+ $(OBJ) $(EXTRA_LIB) $(LIB)
+
+# Compilation rules
+
+%.o:%.cpp
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<
+
+%.d:%.cpp
+ $(CC) $(CCFLAGS) $(EXTRA_INC) $(DEPFLAGS) $< > $@
+
+# Individual dependencies
+
+DEPENDS = $(OBJ:.o=.d)
+sinclude $(DEPENDS)
diff --git a/bench/KEPLER/Makefile.gpu b/bench/KEPLER/Makefile.gpu
new file mode 100755
index 0000000000..f27d7e9301
--- /dev/null
+++ b/bench/KEPLER/Makefile.gpu
@@ -0,0 +1,108 @@
+# linux = Shannon Linux box, Intel icc, OpenMPI, KISS FFTW
+
+SHELL = /bin/sh
+
+# ---------------------------------------------------------------------
+# compiler/linker settings
+# specify flags and libraries needed for your compiler
+
+CC = icc
+CCFLAGS = -O
+SHFLAGS = -fPIC
+DEPFLAGS = -M
+
+LINK = icc
+LINKFLAGS = -O
+LIB = -lstdc++
+SIZE = size
+
+ARCHIVE = ar
+ARFLAGS = -rc
+SHLIBFLAGS = -shared
+
+# ---------------------------------------------------------------------
+# LAMMPS-specific settings
+# specify settings for LAMMPS features you will use
+# if you change any -D setting, do full re-compile after "make clean"
+
+# LAMMPS ifdef settings, OPTIONAL
+# see possible settings in doc/Section_start.html#2_2 (step 4)
+
+LMP_INC =
+
+# MPI library, REQUIRED
+# see discussion in doc/Section_start.html#2_2 (step 5)
+# can point to dummy MPI library in src/STUBS as in Makefile.serial
+# INC = path for mpi.h, MPI compiler settings
+# PATH = path for MPI library
+# LIB = name of MPI library
+
+MPI_INC = -I/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/include/
+MPI_PATH = -L/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/lib
+MPI_LIB = -lmpi
+
+# FFT library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 6)
+# 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 =
+FFT_PATH =
+FFT_LIB =
+
+# JPEG and/or PNG library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 7)
+# 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 = -ljpeg
+
+# ---------------------------------------------------------------------
+# build rules and dependencies
+# no need to edit this section
+
+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) $(PKG_SYSLIB)
+
+# Path to src files
+
+vpath %.cpp ..
+vpath %.h ..
+
+# Link target
+
+$(EXE): $(OBJ)
+ $(LINK) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(EXTRA_LIB) $(LIB) -o $(EXE)
+ $(SIZE) $(EXE)
+
+# Library targets
+
+lib: $(OBJ)
+ $(ARCHIVE) $(ARFLAGS) $(EXE) $(OBJ)
+
+shlib: $(OBJ)
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(SHLIBFLAGS) $(EXTRA_PATH) -o $(EXE) \
+ $(OBJ) $(EXTRA_LIB) $(LIB)
+
+# Compilation rules
+
+%.o:%.cpp
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<
+
+%.d:%.cpp
+ $(CC) $(CCFLAGS) $(EXTRA_INC) $(DEPFLAGS) $< > $@
+
+# Individual dependencies
+
+DEPENDS = $(OBJ:.o=.d)
+sinclude $(DEPENDS)
diff --git a/bench/KEPLER/Makefile.gpu.double b/bench/KEPLER/Makefile.gpu.double
new file mode 100644
index 0000000000..19dae5544d
--- /dev/null
+++ b/bench/KEPLER/Makefile.gpu.double
@@ -0,0 +1,50 @@
+# /* ----------------------------------------------------------------------
+# Generic Linux Makefile for CUDA
+# - Change CUDA_ARCH for your GPU
+# ------------------------------------------------------------------------- */
+
+# which file will be copied to Makefile.lammps
+
+EXTRAMAKE = Makefile.lammps.standard
+
+CUDA_HOME = /home/projects/cuda/6.0.37
+NVCC = nvcc
+
+# Kepler CUDA
+CUDA_ARCH = -arch=sm_35
+# Tesla CUDA
+#CUDA_ARCH = -arch=sm_21
+# newer CUDA
+#CUDA_ARCH = -arch=sm_13
+# older CUDA
+#CUDA_ARCH = -arch=sm_10 -DCUDA_PRE_THREE
+
+# this setting should match LAMMPS Makefile
+# one of LAMMPS_SMALLBIG (default), LAMMPS_BIGBIG and LAMMPS_SMALLSMALL
+
+LMP_INC = -DLAMMPS_SMALLBIG
+
+# precision for GPU calculations
+# -D_SINGLE_SINGLE # Single precision for all calculations
+# -D_DOUBLE_DOUBLE # Double precision for all calculations
+# -D_SINGLE_DOUBLE # Accumulation of forces, etc. in double
+
+CUDA_PRECISION = -D_DOUBLE_DOUBLE
+
+CUDA_INCLUDE = -I$(CUDA_HOME)/include
+CUDA_LIB = -L$(CUDA_HOME)/lib64
+CUDA_OPTS = -DUNIX -O3 -Xptxas -v --use_fast_math
+
+CUDR_CPP = mpic++ -DMPI_GERYON -DUCL_NO_EXIT -DMPICH_IGNORE_CXX_SEEK
+CUDR_OPTS = -O2 # -xHost -no-prec-div -ansi-alias
+
+BIN_DIR = ./
+OBJ_DIR = ./
+LIB_DIR = ./
+AR = ar
+BSH = /bin/sh
+
+CUDPP_OPT = -DUSE_CUDPP -Icudpp_mini
+
+include Nvidia.makefile
+
diff --git a/bench/KEPLER/Makefile.gpu.mixed b/bench/KEPLER/Makefile.gpu.mixed
new file mode 100644
index 0000000000..97ca9201c7
--- /dev/null
+++ b/bench/KEPLER/Makefile.gpu.mixed
@@ -0,0 +1,50 @@
+# /* ----------------------------------------------------------------------
+# Generic Linux Makefile for CUDA
+# - Change CUDA_ARCH for your GPU
+# ------------------------------------------------------------------------- */
+
+# which file will be copied to Makefile.lammps
+
+EXTRAMAKE = Makefile.lammps.standard
+
+CUDA_HOME = /home/projects/cuda/6.0.37
+NVCC = nvcc
+
+# Kepler CUDA
+CUDA_ARCH = -arch=sm_35
+# Tesla CUDA
+#CUDA_ARCH = -arch=sm_21
+# newer CUDA
+#CUDA_ARCH = -arch=sm_13
+# older CUDA
+#CUDA_ARCH = -arch=sm_10 -DCUDA_PRE_THREE
+
+# this setting should match LAMMPS Makefile
+# one of LAMMPS_SMALLBIG (default), LAMMPS_BIGBIG and LAMMPS_SMALLSMALL
+
+LMP_INC = -DLAMMPS_SMALLBIG
+
+# precision for GPU calculations
+# -D_SINGLE_SINGLE # Single precision for all calculations
+# -D_DOUBLE_DOUBLE # Double precision for all calculations
+# -D_SINGLE_DOUBLE # Accumulation of forces, etc. in double
+
+CUDA_PRECISION = -D_SINGLE_DOUBLE
+
+CUDA_INCLUDE = -I$(CUDA_HOME)/include
+CUDA_LIB = -L$(CUDA_HOME)/lib64
+CUDA_OPTS = -DUNIX -O3 -Xptxas -v --use_fast_math
+
+CUDR_CPP = mpic++ -DMPI_GERYON -DUCL_NO_EXIT -DMPICH_IGNORE_CXX_SEEK
+CUDR_OPTS = -O2 # -xHost -no-prec-div -ansi-alias
+
+BIN_DIR = ./
+OBJ_DIR = ./
+LIB_DIR = ./
+AR = ar
+BSH = /bin/sh
+
+CUDPP_OPT = -DUSE_CUDPP -Icudpp_mini
+
+include Nvidia.makefile
+
diff --git a/bench/KEPLER/Makefile.gpu.single b/bench/KEPLER/Makefile.gpu.single
new file mode 100644
index 0000000000..f8005aaac7
--- /dev/null
+++ b/bench/KEPLER/Makefile.gpu.single
@@ -0,0 +1,50 @@
+# /* ----------------------------------------------------------------------
+# Generic Linux Makefile for CUDA
+# - Change CUDA_ARCH for your GPU
+# ------------------------------------------------------------------------- */
+
+# which file will be copied to Makefile.lammps
+
+EXTRAMAKE = Makefile.lammps.standard
+
+CUDA_HOME = /home/projects/cuda/6.0.37
+NVCC = nvcc
+
+# Kepler CUDA
+CUDA_ARCH = -arch=sm_35
+# Tesla CUDA
+#CUDA_ARCH = -arch=sm_21
+# newer CUDA
+#CUDA_ARCH = -arch=sm_13
+# older CUDA
+#CUDA_ARCH = -arch=sm_10 -DCUDA_PRE_THREE
+
+# this setting should match LAMMPS Makefile
+# one of LAMMPS_SMALLBIG (default), LAMMPS_BIGBIG and LAMMPS_SMALLSMALL
+
+LMP_INC = -DLAMMPS_SMALLBIG
+
+# precision for GPU calculations
+# -D_SINGLE_SINGLE # Single precision for all calculations
+# -D_DOUBLE_DOUBLE # Double precision for all calculations
+# -D_SINGLE_DOUBLE # Accumulation of forces, etc. in double
+
+CUDA_PRECISION = -D_SINGLE_SINGLE
+
+CUDA_INCLUDE = -I$(CUDA_HOME)/include
+CUDA_LIB = -L$(CUDA_HOME)/lib64
+CUDA_OPTS = -DUNIX -O3 -Xptxas -v --use_fast_math
+
+CUDR_CPP = mpic++ -DMPI_GERYON -DUCL_NO_EXIT -DMPICH_IGNORE_CXX_SEEK
+CUDR_OPTS = -O2 # -xHost -no-prec-div -ansi-alias
+
+BIN_DIR = ./
+OBJ_DIR = ./
+LIB_DIR = ./
+AR = ar
+BSH = /bin/sh
+
+CUDPP_OPT = -DUSE_CUDPP -Icudpp_mini
+
+include Nvidia.makefile
+
diff --git a/bench/KEPLER/Makefile.intel.cpu b/bench/KEPLER/Makefile.intel.cpu
new file mode 100755
index 0000000000..49a5f12b8f
--- /dev/null
+++ b/bench/KEPLER/Makefile.intel.cpu
@@ -0,0 +1,109 @@
+# linux = Shannon Linux box, Intel icc, OpenMPI, KISS FFTW
+
+SHELL = /bin/sh
+
+# ---------------------------------------------------------------------
+# compiler/linker settings
+# specify flags and libraries needed for your compiler
+
+CC = icc
+CCFLAGS = -O3 -openmp -DLAMMPS_MEMALIGN=64 -no-offload \
+ -xHost -fno-alias -ansi-alias -restrict -override-limits
+SHFLAGS = -fPIC
+DEPFLAGS = -M
+
+LINK = icc
+LINKFLAGS = -O -openmp
+LIB = -lstdc++
+SIZE = size
+
+ARCHIVE = ar
+ARFLAGS = -rc
+SHLIBFLAGS = -shared
+
+# ---------------------------------------------------------------------
+# LAMMPS-specific settings
+# specify settings for LAMMPS features you will use
+# if you change any -D setting, do full re-compile after "make clean"
+
+# LAMMPS ifdef settings, OPTIONAL
+# see possible settings in doc/Section_start.html#2_2 (step 4)
+
+LMP_INC = -DLAMMPS_GZIP -DLAMMPS_JPEG
+
+# MPI library, REQUIRED
+# see discussion in doc/Section_start.html#2_2 (step 5)
+# can point to dummy MPI library in src/STUBS as in Makefile.serial
+# INC = path for mpi.h, MPI compiler settings
+# PATH = path for MPI library
+# LIB = name of MPI library
+
+MPI_INC = -I/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/include/
+MPI_PATH = -L/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/lib
+MPI_LIB = -lmpi
+
+# FFT library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 6)
+# 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 =
+FFT_PATH =
+FFT_LIB =
+
+# JPEG and/or PNG library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 7)
+# 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 = -ljpeg
+
+# ---------------------------------------------------------------------
+# build rules and dependencies
+# no need to edit this section
+
+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) $(PKG_SYSLIB)
+
+# Path to src files
+
+vpath %.cpp ..
+vpath %.h ..
+
+# Link target
+
+$(EXE): $(OBJ)
+ $(LINK) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(EXTRA_LIB) $(LIB) -o $(EXE)
+ $(SIZE) $(EXE)
+
+# Library targets
+
+lib: $(OBJ)
+ $(ARCHIVE) $(ARFLAGS) $(EXE) $(OBJ)
+
+shlib: $(OBJ)
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(SHLIBFLAGS) $(EXTRA_PATH) -o $(EXE) \
+ $(OBJ) $(EXTRA_LIB) $(LIB)
+
+# Compilation rules
+
+%.o:%.cpp
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<
+
+%.d:%.cpp
+ $(CC) $(CCFLAGS) $(EXTRA_INC) $(DEPFLAGS) $< > $@
+
+# Individual dependencies
+
+DEPENDS = $(OBJ:.o=.d)
+sinclude $(DEPENDS)
diff --git a/bench/KEPLER/Makefile.kokkos.cuda b/bench/KEPLER/Makefile.kokkos.cuda
new file mode 100755
index 0000000000..3f75694885
--- /dev/null
+++ b/bench/KEPLER/Makefile.kokkos.cuda
@@ -0,0 +1,113 @@
+# linux = Shannon Linux box, Intel icc, OpenMPI, KISS FFTW
+
+SHELL = /bin/sh
+
+# ---------------------------------------------------------------------
+# compiler/linker settings
+# specify flags and libraries needed for your compiler
+
+CC = nvcc
+CCFLAGS = -O3 -arch=sm_35
+SHFLAGS = -fPIC
+DEPFLAGS = -M
+
+LINK = mpicxx
+LINKFLAGS = -O
+LIB = -lstdc++
+SIZE = size
+
+ARCHIVE = ar
+ARFLAGS = -rc
+SHLIBFLAGS = -shared
+
+OMP = yes
+CUDA = yes
+
+# ---------------------------------------------------------------------
+# LAMMPS-specific settings
+# specify settings for LAMMPS features you will use
+# if you change any -D setting, do full re-compile after "make clean"
+
+# LAMMPS ifdef settings, OPTIONAL
+# see possible settings in doc/Section_start.html#2_2 (step 4)
+
+LMP_INC =
+
+# MPI library, REQUIRED
+# see discussion in doc/Section_start.html#2_2 (step 5)
+# can point to dummy MPI library in src/STUBS as in Makefile.serial
+# INC = path for mpi.h, MPI compiler settings
+# PATH = path for MPI library
+# LIB = name of MPI library
+
+MPI_INC = -I/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/include/
+MPI_PATH = -L/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/lib
+MPI_LIB = -lmpi
+
+# FFT library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 6)
+# 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 =
+FFT_PATH =
+FFT_LIB =
+
+# JPEG and/or PNG library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 7)
+# 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 = -ljpeg
+
+# ---------------------------------------------------------------------
+# build rules and dependencies
+# no need to edit this section
+
+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) $(PKG_SYSLIB)
+
+# Path to src files
+
+vpath %.cpp ..
+vpath %.h ..
+
+# Link target
+
+$(EXE): $(OBJ)
+ $(LINK) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(EXTRA_LIB) $(LIB) -o $(EXE)
+ $(SIZE) $(EXE)
+
+# Library targets
+
+lib: $(OBJ)
+ $(ARCHIVE) $(ARFLAGS) $(EXE) $(OBJ)
+
+shlib: $(OBJ)
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(SHLIBFLAGS) $(EXTRA_PATH) -o $(EXE) \
+ $(OBJ) $(EXTRA_LIB) $(LIB)
+
+# Compilation rules
+%.o:%.cu
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<
+
+%.o:%.cpp
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<
+
+%.d:%.cpp
+ $(CC) $(CCFLAGS) $(EXTRA_INC) $(DEPFLAGS) $< > $@
+
+# Individual dependencies
+
+DEPENDS = $(OBJ:.o=.d)
+sinclude $(DEPENDS)
diff --git a/bench/KEPLER/Makefile.kokkos.omp b/bench/KEPLER/Makefile.kokkos.omp
new file mode 100755
index 0000000000..61efa1ff69
--- /dev/null
+++ b/bench/KEPLER/Makefile.kokkos.omp
@@ -0,0 +1,110 @@
+# linux = Shannon Linux box, Intel icc, OpenMPI, KISS FFTW
+
+SHELL = /bin/sh
+
+# ---------------------------------------------------------------------
+# compiler/linker settings
+# specify flags and libraries needed for your compiler
+
+CC = icc
+CCFLAGS = -O
+SHFLAGS = -fPIC
+DEPFLAGS = -M
+
+LINK = icc
+LINKFLAGS = -O
+LIB = -lstdc++
+SIZE = size
+
+ARCHIVE = ar
+ARFLAGS = -rc
+SHLIBFLAGS = -shared
+
+OMP = yes
+
+# ---------------------------------------------------------------------
+# LAMMPS-specific settings
+# specify settings for LAMMPS features you will use
+# if you change any -D setting, do full re-compile after "make clean"
+
+# LAMMPS ifdef settings, OPTIONAL
+# see possible settings in doc/Section_start.html#2_2 (step 4)
+
+LMP_INC =
+
+# MPI library, REQUIRED
+# see discussion in doc/Section_start.html#2_2 (step 5)
+# can point to dummy MPI library in src/STUBS as in Makefile.serial
+# INC = path for mpi.h, MPI compiler settings
+# PATH = path for MPI library
+# LIB = name of MPI library
+
+MPI_INC = -I/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/include/
+MPI_PATH = -L/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/lib
+MPI_LIB = -lmpi
+
+# FFT library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 6)
+# 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 =
+FFT_PATH =
+FFT_LIB =
+
+# JPEG and/or PNG library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 7)
+# 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 = -ljpeg
+
+# ---------------------------------------------------------------------
+# build rules and dependencies
+# no need to edit this section
+
+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) $(PKG_SYSLIB)
+
+# Path to src files
+
+vpath %.cpp ..
+vpath %.h ..
+
+# Link target
+
+$(EXE): $(OBJ)
+ $(LINK) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(EXTRA_LIB) $(LIB) -o $(EXE)
+ $(SIZE) $(EXE)
+
+# Library targets
+
+lib: $(OBJ)
+ $(ARCHIVE) $(ARFLAGS) $(EXE) $(OBJ)
+
+shlib: $(OBJ)
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(SHLIBFLAGS) $(EXTRA_PATH) -o $(EXE) \
+ $(OBJ) $(EXTRA_LIB) $(LIB)
+
+# Compilation rules
+
+%.o:%.cpp
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<
+
+%.d:%.cpp
+ $(CC) $(CCFLAGS) $(EXTRA_INC) $(DEPFLAGS) $< > $@
+
+# Individual dependencies
+
+DEPENDS = $(OBJ:.o=.d)
+sinclude $(DEPENDS)
diff --git a/bench/KEPLER/Makefile.omp b/bench/KEPLER/Makefile.omp
new file mode 100755
index 0000000000..41c430f0fa
--- /dev/null
+++ b/bench/KEPLER/Makefile.omp
@@ -0,0 +1,108 @@
+# linux = Shannon Linux box, Intel icc, OpenMPI, KISS FFTW
+
+SHELL = /bin/sh
+
+# ---------------------------------------------------------------------
+# compiler/linker settings
+# specify flags and libraries needed for your compiler
+
+CC = icc
+CCFLAGS = -O3 -openmp -ansi-alias
+SHFLAGS = -fPIC
+DEPFLAGS = -M
+
+LINK = icc
+LINKFLAGS = -O -openmp
+LIB = -lstdc++
+SIZE = size
+
+ARCHIVE = ar
+ARFLAGS = -rc
+SHLIBFLAGS = -shared
+
+# ---------------------------------------------------------------------
+# LAMMPS-specific settings
+# specify settings for LAMMPS features you will use
+# if you change any -D setting, do full re-compile after "make clean"
+
+# LAMMPS ifdef settings, OPTIONAL
+# see possible settings in doc/Section_start.html#2_2 (step 4)
+
+LMP_INC =
+
+# MPI library, REQUIRED
+# see discussion in doc/Section_start.html#2_2 (step 5)
+# can point to dummy MPI library in src/STUBS as in Makefile.serial
+# INC = path for mpi.h, MPI compiler settings
+# PATH = path for MPI library
+# LIB = name of MPI library
+
+MPI_INC = -I/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/include/
+MPI_PATH = -L/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/lib
+MPI_LIB = -lmpi
+
+# FFT library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 6)
+# 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 =
+FFT_PATH =
+FFT_LIB =
+
+# JPEG and/or PNG library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 7)
+# 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 = -ljpeg
+
+# ---------------------------------------------------------------------
+# build rules and dependencies
+# no need to edit this section
+
+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) $(PKG_SYSLIB)
+
+# Path to src files
+
+vpath %.cpp ..
+vpath %.h ..
+
+# Link target
+
+$(EXE): $(OBJ)
+ $(LINK) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(EXTRA_LIB) $(LIB) -o $(EXE)
+ $(SIZE) $(EXE)
+
+# Library targets
+
+lib: $(OBJ)
+ $(ARCHIVE) $(ARFLAGS) $(EXE) $(OBJ)
+
+shlib: $(OBJ)
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(SHLIBFLAGS) $(EXTRA_PATH) -o $(EXE) \
+ $(OBJ) $(EXTRA_LIB) $(LIB)
+
+# Compilation rules
+
+%.o:%.cpp
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<
+
+%.d:%.cpp
+ $(CC) $(CCFLAGS) $(EXTRA_INC) $(DEPFLAGS) $< > $@
+
+# Individual dependencies
+
+DEPENDS = $(OBJ:.o=.d)
+sinclude $(DEPENDS)
diff --git a/bench/KEPLER/Makefile.opt b/bench/KEPLER/Makefile.opt
new file mode 100755
index 0000000000..f27d7e9301
--- /dev/null
+++ b/bench/KEPLER/Makefile.opt
@@ -0,0 +1,108 @@
+# linux = Shannon Linux box, Intel icc, OpenMPI, KISS FFTW
+
+SHELL = /bin/sh
+
+# ---------------------------------------------------------------------
+# compiler/linker settings
+# specify flags and libraries needed for your compiler
+
+CC = icc
+CCFLAGS = -O
+SHFLAGS = -fPIC
+DEPFLAGS = -M
+
+LINK = icc
+LINKFLAGS = -O
+LIB = -lstdc++
+SIZE = size
+
+ARCHIVE = ar
+ARFLAGS = -rc
+SHLIBFLAGS = -shared
+
+# ---------------------------------------------------------------------
+# LAMMPS-specific settings
+# specify settings for LAMMPS features you will use
+# if you change any -D setting, do full re-compile after "make clean"
+
+# LAMMPS ifdef settings, OPTIONAL
+# see possible settings in doc/Section_start.html#2_2 (step 4)
+
+LMP_INC =
+
+# MPI library, REQUIRED
+# see discussion in doc/Section_start.html#2_2 (step 5)
+# can point to dummy MPI library in src/STUBS as in Makefile.serial
+# INC = path for mpi.h, MPI compiler settings
+# PATH = path for MPI library
+# LIB = name of MPI library
+
+MPI_INC = -I/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/include/
+MPI_PATH = -L/home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37/lib
+MPI_LIB = -lmpi
+
+# FFT library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 6)
+# 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 =
+FFT_PATH =
+FFT_LIB =
+
+# JPEG and/or PNG library, OPTIONAL
+# see discussion in doc/Section_start.html#2_2 (step 7)
+# 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 = -ljpeg
+
+# ---------------------------------------------------------------------
+# build rules and dependencies
+# no need to edit this section
+
+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) $(PKG_SYSLIB)
+
+# Path to src files
+
+vpath %.cpp ..
+vpath %.h ..
+
+# Link target
+
+$(EXE): $(OBJ)
+ $(LINK) $(LINKFLAGS) $(EXTRA_PATH) $(OBJ) $(EXTRA_LIB) $(LIB) -o $(EXE)
+ $(SIZE) $(EXE)
+
+# Library targets
+
+lib: $(OBJ)
+ $(ARCHIVE) $(ARFLAGS) $(EXE) $(OBJ)
+
+shlib: $(OBJ)
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(SHLIBFLAGS) $(EXTRA_PATH) -o $(EXE) \
+ $(OBJ) $(EXTRA_LIB) $(LIB)
+
+# Compilation rules
+
+%.o:%.cpp
+ $(CC) $(CCFLAGS) $(SHFLAGS) $(EXTRA_INC) -c $<
+
+%.d:%.cpp
+ $(CC) $(CCFLAGS) $(EXTRA_INC) $(DEPFLAGS) $< > $@
+
+# Individual dependencies
+
+DEPENDS = $(OBJ:.o=.d)
+sinclude $(DEPENDS)
diff --git a/bench/KEPLER/README b/bench/KEPLER/README
new file mode 100644
index 0000000000..feee3e0b41
--- /dev/null
+++ b/bench/KEPLER/README
@@ -0,0 +1,71 @@
+These are build and input and run scripts used to run the LJ benchmark
+in the top-level bench directory using all the various accelerator
+packages currently available in LAMMPS. The results of running these
+benchmarks on a GPU cluster with Kepler GPUs are shown on the "GPU
+(Kepler)" section of the Benchmark page of the LAMMPS WWW site:
+lammps.sandia.gov/bench.
+
+The specifics of the benchmark machine are as follows:
+
+It is a small GPU cluster at Sandia National Labs called "shannon". It
+has 32 nodes, each with two 8-core Sandy Bridge Xeon CPUs (E5-2670,
+2.6GHz, HT deactivated), for a total of 512 cores. Twenty-four of the
+nodes have two NVIDIA Kepler GPUs (K20x, 2688 732 MHz cores). LAMMPS
+was compiled with the Intel icc compiler, using module
+openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37.
+
+------------------------------------------------------------------------
+
+You can, of course, build LAMMPS yourself with any of the accelerator
+packages for your platform.
+
+The build.py script will build LAMMPS for the various accelerlator
+packages using the Makefile.* files in this dir, which you can edit if
+necessary for your platform. You must set the "lmpdir" variable at
+the top of build.py to the home directory of LAMMPS as installed on
+your system. Note that the build.py script hardcodes the arch setting
+for the USER-CUDA package, which should be matched to the GPUs on your
+system, e.g. sm_35 for Kepler GPUs. For the GPU package, this setting
+is in the Makefile.gpu.* files, as is the CUDA_HOME variable which
+should point to where NVIDIA Cuda software is installed on your
+system.
+
+Once the Makefiles are in place, then typing, for example,
+
+python build.py cpu gpu
+
+will build executables for the CPU (no accelerators), and 3 variants
+(double, mixed, single precision) of the GPU package. See the list of
+possible targets at the top of the build.py script.
+
+Note that the build.py script will un-install all packages in your
+LAMMPS directory, then only install the ones needed for the benchmark.
+The Makefile.* files in this dir are copied into lammps/src/MAKE, as a
+dummy Makefile.foo, so they will not conflict with makefiles that may
+already be there. The build.py script also builds the auxiliary GPU
+and USER-CUDA library as needed.
+
+LAMMPS executables that are generated by build.py are copied into this
+directory when the script finishes each build.
+
+------------------------------------------------------------------------
+
+The in.* files have settings for the benchmark appopriate to each
+accelerator package. Many of them, including the problem size,
+and number of timsteps, must be set as command-line arguments
+when the input script is run.
+
+------------------------------------------------------------------------
+
+The run*.sh scripts have sample mpirun commands for running the input
+scripts on a single node or on multiple nodes for the strong and weak
+scaling results shown on the benchmark web page. These scripts are
+provided for illustration purposes, to show what command-line
+arguments are used with each accelerator package, in combination with
+settings in the input scripts themselves.
+
+Note that we generate these run scripts, either for interactive or
+batch submission, via Python scripts which produce a long list of runs
+to exercise a combination of options. To perform a quick benchmark
+calculation on your platform, you will typically only want to run a
+few commands out of any of the run*.sh scripts.
diff --git a/bench/KEPLER/build.py b/bench/KEPLER/build.py
new file mode 100644
index 0000000000..39b7617a0f
--- /dev/null
+++ b/bench/KEPLER/build.py
@@ -0,0 +1,181 @@
+#!/usr/local/bin/python
+
+# Syntax: build.py target1 target2 ...
+# targets:
+# cpu, opt, omp,
+# gpu/double, gpu/mixed, gpu/single,
+# cuda/double, cuda/mixed, cuda/single,
+# intel/cpu, intel/phi,
+# kokkos/omp, kokkos/phi, kokkos/cuda
+# gpu = gpu/double + gpu/mixed + gpu/single
+# cuda = cuda/double + cuda/mixed + cuda/single
+# intel = intel/cpu + intel/phi
+# kokkos = kokkos/omp + kokkos/phi + kokkos/cuda
+# all = cpu + opt + omp + gpu + cuda + intel + kokkos
+
+# create exectuables for different packages
+# MUST set lmpdir to path of LAMMPS home directory
+
+import sys,commands,os
+
+lmpdir = "~/lammps"
+
+# build LAMMPS
+# copy makefile into src/MAKE as Makefile.foo, then remove it
+
+def build_lammps(makefile,pkg):
+ print "Building LAMMPS with %s and %s packages ..." % (makefile,pkg)
+ commands.getoutput("cp %s %s/src/MAKE/Makefile.foo" % (makefile,lmpdir))
+ cwd = os.getcwd()
+ os.chdir(os.path.expanduser(lmpdir + "/src"))
+ str = "make clean-foo"
+ txt = commands.getoutput(str)
+ str = "make no-all"
+ txt = commands.getoutput(str)
+ for package in pkg:
+ str = "make yes-%s" % package
+ txt = commands.getoutput(str)
+ print txt
+ str = "make -j 16 foo"
+ txt = commands.getoutput(str)
+ os.remove("MAKE/Makefile.foo")
+ os.chdir(cwd)
+
+# build GPU library in LAMMPS
+# copy makefile into lib/gpu as Makefile.foo, then remove it
+
+def build_gpu(makefile):
+ print "Building GPU lib with %s ..." % makefile
+ commands.getoutput("cp %s %s/lib/gpu/Makefile.foo" % (makefile,lmpdir))
+ cwd = os.getcwd()
+ os.chdir(os.path.expanduser(lmpdir + "/lib/gpu"))
+ str = "make -f Makefile.foo clean"
+ txt = commands.getoutput(str)
+ str = "make -j 16 -f Makefile.foo"
+ txt = commands.getoutput(str)
+ os.remove("Makefile.foo")
+ os.chdir(cwd)
+
+# build CUDA library in LAMMPS
+# set precision and arch explicitly as options to make in lib/cuda
+
+def build_cuda(precision,arch):
+ print "Building USER-CUDA lib with %s and arch sm_%d ..." % (precision,arch)
+ cwd = os.getcwd()
+ os.chdir(os.path.expanduser(lmpdir + "/lib/cuda"))
+ str = "make clean"
+ txt = commands.getoutput(str)
+ if precision == "double": pflag = 2
+ elif precision == "mixed": pflag = 4
+ elif precision == "single": pflag = 1
+ str = "make -j 16 precision=%d arch=%s" % (pflag,arch)
+ txt = commands.getoutput(str)
+
+ os.chdir(cwd)
+
+# main program
+# convert target keywords into target flags
+
+cpu = opt = omp = 0
+gpu_double = gpu_mixed = gpu_single = 0
+cuda_double = cuda_mixed = cuda_single = 0
+intel_cpu = intel_phi = 0
+kokkos_omp = kokkos_phi = kokkos_cuda = 0
+
+targets = sys.argv[1:]
+for target in targets:
+ if target == "cpu": cpu = 1
+ elif target == "opt": opt = 1
+ elif target == "omp": omp = 1
+ elif target == "gpu/double": gpu_double = 1
+ elif target == "gpu/mixed": gpu_mixed = 1
+ elif target == "gpu/single": gpu_single = 1
+ elif target == "gpu": gpu_double = gpu_mixed = gpu_single = 1
+ elif target == "cuda/double": cuda_double = 1
+ elif target == "cuda/mixed": cuda_mixed = 1
+ elif target == "cuda/single": cuda_single = 1
+ elif target == "cuda": cuda_double = cuda_mixed = cuda_single = 1
+ elif target == "intel/cpu": intel_cpu = 1
+ elif target == "intel/phi": intel_phi = 1
+ elif target == "intel": intel_cpu = intel_phi = 1
+ elif target == "kokkos/omp": kokkos_omp = 1
+ elif target == "kokkos/phi": kokkos_phi = 1
+ elif target == "kokkos/cuda": kokkos_cuda = 1
+ elif target == "kokkos": kokkos_omp = kokkos_phi = kokkos_cuda = 1
+ else: print "Target",target,"is unknown"
+
+# CPU
+
+if cpu:
+ build_lammps(makefile = "Makefile.cpu", pkg = [])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_cpu" % lmpdir)
+
+# OPT
+
+if opt:
+ build_lammps(makefile = "Makefile.opt", pkg = ["opt"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_opt" % lmpdir)
+
+# OMP
+
+if omp:
+ build_lammps(makefile = "Makefile.omp", pkg = ["user-omp"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_omp" % lmpdir)
+
+# GPU, 3 precisions
+
+if gpu_double:
+ build_gpu(makefile = "Makefile.gpu.double")
+ build_lammps(makefile = "Makefile.gpu", pkg = ["gpu"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_gpu_double" % lmpdir)
+
+if gpu_mixed:
+ build_gpu(makefile = "Makefile.gpu.mixed")
+ build_lammps(makefile = "Makefile.gpu", pkg = ["gpu"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_gpu_mixed" % lmpdir)
+
+if gpu_single:
+ build_gpu(makefile = "Makefile.gpu.single")
+ build_lammps(makefile = "Makefile.gpu", pkg = ["gpu"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_gpu_single" % lmpdir)
+
+# CUDA, 3 precisions
+
+if cuda_double:
+ build_cuda(precision = "double", arch = 35)
+ build_lammps(makefile = "Makefile.cuda", pkg = ["kspace","user-cuda"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_cuda_double" % lmpdir)
+
+if cuda_mixed:
+ build_cuda(precision = "mixed", arch = 35)
+ build_lammps(makefile = "Makefile.cuda", pkg = ["kspace","user-cuda"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_cuda_mixed" % lmpdir)
+
+if cuda_single:
+ build_cuda(precision = "single", arch = 35)
+ build_lammps(makefile = "Makefile.cuda", pkg = ["kspace","user-cuda"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_cuda_single" % lmpdir)
+
+# INTEL, CPU and Phi
+
+if intel_cpu:
+ build_lammps(makefile = "Makefile.intel.cpu", pkg = ["user-intel"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_intel_cpu" % lmpdir)
+
+if intel_phi:
+ build_lammps(makefile = "Makefile.intel.phi", pkg = ["user-intel","user-omp"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_intel_phi" % lmpdir)
+
+# KOKKOS, all variants
+
+if kokkos_omp:
+ build_lammps(makefile = "Makefile.kokkos.omp", pkg = ["kokkos"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_kokkos_omp" % lmpdir)
+
+if kokkos_phi:
+ build_lammps(makefile = "Makefile.kokkos.phi", pkg = ["kokkos"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_kokkos_phi" % lmpdir)
+
+if kokkos_cuda:
+ build_lammps(makefile = "Makefile.kokkos.cuda", pkg = ["kokkos"])
+ print commands.getoutput("mv %s/src/lmp_foo ./lmp_kokkos_cuda" % lmpdir)
diff --git a/bench/KEPLER/in.lj.cpu b/bench/KEPLER/in.lj.cpu
new file mode 100644
index 0000000000..ab6988e286
--- /dev/null
+++ b/bench/KEPLER/in.lj.cpu
@@ -0,0 +1,22 @@
+# 3d Lennard-Jones melt
+
+units lj
+atom_style atomic
+
+lattice fcc 0.8442
+region box block 0 $x 0 $y 0 $z
+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 $t
diff --git a/bench/KEPLER/in.lj.cuda b/bench/KEPLER/in.lj.cuda
new file mode 100644
index 0000000000..fcf6e37bc7
--- /dev/null
+++ b/bench/KEPLER/in.lj.cuda
@@ -0,0 +1,26 @@
+# 3d Lennard-Jones melt
+
+# set number of GPUs
+
+package cuda gpu/node $g
+
+units lj
+atom_style atomic
+
+lattice fcc 0.8442
+region box block 0 $x 0 $y 0 $z
+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 $t
diff --git a/bench/KEPLER/in.lj.gpu b/bench/KEPLER/in.lj.gpu
new file mode 100644
index 0000000000..6cc2831365
--- /dev/null
+++ b/bench/KEPLER/in.lj.gpu
@@ -0,0 +1,29 @@
+# 3d Lennard-Jones melt
+
+# newton off is required for GPU package
+# set number of GPUs
+
+newton off
+if "$g == 1" then "package gpu force/neigh 0 0 1"
+if "$g == 2" then "package gpu force/neigh 0 1 1"
+
+units lj
+atom_style atomic
+
+lattice fcc 0.8442
+region box block 0 $x 0 $y 0 $z
+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 $t
diff --git a/bench/KEPLER/in.lj.intel.cpu b/bench/KEPLER/in.lj.intel.cpu
new file mode 100644
index 0000000000..f8634ead8e
--- /dev/null
+++ b/bench/KEPLER/in.lj.intel.cpu
@@ -0,0 +1,26 @@
+# 3d Lennard-Jones melt
+
+# set precision for USER-INTEL package
+
+package intel 1 $a balance 0.0
+
+units lj
+atom_style atomic
+
+lattice fcc 0.8442
+region box block 0 $x 0 $y 0 $z
+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 $t
diff --git a/bench/KEPLER/in.lj.kokkos.cuda b/bench/KEPLER/in.lj.kokkos.cuda
new file mode 100644
index 0000000000..ae403e70c0
--- /dev/null
+++ b/bench/KEPLER/in.lj.kokkos.cuda
@@ -0,0 +1,29 @@
+# 3d Lennard-Jones melt
+
+# set neighlist style and comm mode
+
+package kokkos neigh full comm/exchange $c comm/forward $c
+newton off
+
+units lj
+atom_style atomic
+
+lattice fcc 0.8442
+region box block 0 $x 0 $y 0 $z
+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
+
+# set neighbor bins to cutoff + skin
+
+neighbor 0.3 bin
+neigh_modify delay 0 every 20 check no binsize 2.8
+
+fix 1 all nve
+
+run $t
diff --git a/bench/KEPLER/in.lj.kokkos.omp b/bench/KEPLER/in.lj.kokkos.omp
new file mode 100644
index 0000000000..34bfddcbdd
--- /dev/null
+++ b/bench/KEPLER/in.lj.kokkos.omp
@@ -0,0 +1,26 @@
+# 3d Lennard-Jones melt
+
+# set neighlist style and comm mode
+
+package kokkos neigh $n comm/exchange $c comm/forward $c
+
+units lj
+atom_style atomic
+
+lattice fcc 0.8442
+region box block 0 $x 0 $y 0 $z
+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 $t
diff --git a/bench/KEPLER/in.lj.omp b/bench/KEPLER/in.lj.omp
new file mode 100644
index 0000000000..3c80af3e39
--- /dev/null
+++ b/bench/KEPLER/in.lj.omp
@@ -0,0 +1,26 @@
+# 3d Lennard-Jones melt
+
+# set number of threads per MPI task
+
+package omp $h
+
+units lj
+atom_style atomic
+
+lattice fcc 0.8442
+region box block 0 $x 0 $y 0 $z
+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 $t
diff --git a/bench/KEPLER/in.lj.opt b/bench/KEPLER/in.lj.opt
new file mode 100644
index 0000000000..ab6988e286
--- /dev/null
+++ b/bench/KEPLER/in.lj.opt
@@ -0,0 +1,22 @@
+# 3d Lennard-Jones melt
+
+units lj
+atom_style atomic
+
+lattice fcc 0.8442
+region box block 0 $x 0 $y 0 $z
+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 $t
diff --git a/bench/KEPLER/run_cpu.sh b/bench/KEPLER/run_cpu.sh
new file mode 100644
index 0000000000..b08d606c92
--- /dev/null
+++ b/bench/KEPLER/run_cpu.sh
@@ -0,0 +1,434 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -N 1 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.1
+mpirun -N 1 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.1
+mpirun -N 1 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.1
+mpirun -N 1 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.1
+mpirun -N 1 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.1
+mpirun -N 1 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.1
+mpirun -N 1 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.1
+mpirun -N 1 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.1
+mpirun -N 1 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.1
+mpirun -N 1 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.1
+mpirun -N 1 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.1
+mpirun -N 1 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.1
+mpirun -N 1 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.1
+
+mpirun -N 2 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.2
+mpirun -N 2 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.2
+mpirun -N 2 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.2
+mpirun -N 2 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.2
+mpirun -N 2 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.2
+mpirun -N 2 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.2
+mpirun -N 2 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.2
+mpirun -N 2 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.2
+mpirun -N 2 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.2
+mpirun -N 2 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.2
+mpirun -N 2 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.2
+mpirun -N 2 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.2
+mpirun -N 2 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.2
+
+mpirun -N 3 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.3
+mpirun -N 3 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.3
+mpirun -N 3 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.3
+mpirun -N 3 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.3
+mpirun -N 3 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.3
+mpirun -N 3 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.3
+mpirun -N 3 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.3
+mpirun -N 3 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.3
+mpirun -N 3 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.3
+mpirun -N 3 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.3
+mpirun -N 3 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.3
+mpirun -N 3 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.3
+mpirun -N 3 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.3
+
+mpirun -N 4 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.4
+mpirun -N 4 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.4
+mpirun -N 4 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.4
+mpirun -N 4 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.4
+mpirun -N 4 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.4
+mpirun -N 4 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.4
+mpirun -N 4 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.4
+mpirun -N 4 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.4
+mpirun -N 4 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.4
+mpirun -N 4 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.4
+mpirun -N 4 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.4
+mpirun -N 4 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.4
+mpirun -N 4 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.4
+
+mpirun -N 5 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.5
+mpirun -N 5 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.5
+mpirun -N 5 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.5
+mpirun -N 5 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.5
+mpirun -N 5 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.5
+mpirun -N 5 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.5
+mpirun -N 5 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.5
+mpirun -N 5 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.5
+mpirun -N 5 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.5
+mpirun -N 5 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.5
+mpirun -N 5 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.5
+mpirun -N 5 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.5
+mpirun -N 5 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.5
+
+mpirun -N 6 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.6
+mpirun -N 6 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.6
+mpirun -N 6 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.6
+mpirun -N 6 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.6
+mpirun -N 6 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.6
+mpirun -N 6 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.6
+mpirun -N 6 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.6
+mpirun -N 6 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.6
+mpirun -N 6 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.6
+mpirun -N 6 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.6
+mpirun -N 6 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.6
+mpirun -N 6 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.6
+mpirun -N 6 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.6
+
+mpirun -N 7 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.7
+mpirun -N 7 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.7
+mpirun -N 7 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.7
+mpirun -N 7 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.7
+mpirun -N 7 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.7
+mpirun -N 7 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.7
+mpirun -N 7 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.7
+mpirun -N 7 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.7
+mpirun -N 7 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.7
+mpirun -N 7 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.7
+mpirun -N 7 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.7
+mpirun -N 7 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.7
+mpirun -N 7 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.7
+
+mpirun -N 8 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.8
+mpirun -N 8 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.8
+mpirun -N 8 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.8
+mpirun -N 8 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.8
+mpirun -N 8 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.8
+mpirun -N 8 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.8
+mpirun -N 8 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.8
+mpirun -N 8 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.8
+mpirun -N 8 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.8
+mpirun -N 8 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.8
+mpirun -N 8 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.8
+mpirun -N 8 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.8
+mpirun -N 8 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.8
+
+mpirun -N 9 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.9
+mpirun -N 9 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.9
+mpirun -N 9 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.9
+mpirun -N 9 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.9
+mpirun -N 9 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.9
+mpirun -N 9 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.9
+mpirun -N 9 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.9
+mpirun -N 9 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.9
+mpirun -N 9 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.9
+mpirun -N 9 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.9
+mpirun -N 9 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.9
+mpirun -N 9 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.9
+mpirun -N 9 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.9
+
+mpirun -N 10 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.10
+mpirun -N 10 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.10
+mpirun -N 10 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.10
+mpirun -N 10 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.10
+mpirun -N 10 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.10
+mpirun -N 10 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.10
+mpirun -N 10 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.10
+mpirun -N 10 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.10
+mpirun -N 10 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.10
+mpirun -N 10 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.10
+mpirun -N 10 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.10
+mpirun -N 10 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.10
+mpirun -N 10 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.10
+
+mpirun -N 11 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.11
+mpirun -N 11 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.11
+mpirun -N 11 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.11
+mpirun -N 11 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.11
+mpirun -N 11 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.11
+mpirun -N 11 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.11
+mpirun -N 11 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.11
+mpirun -N 11 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.11
+mpirun -N 11 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.11
+mpirun -N 11 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.11
+mpirun -N 11 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.11
+mpirun -N 11 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.11
+mpirun -N 11 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.11
+
+mpirun -N 12 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.12
+mpirun -N 12 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.12
+mpirun -N 12 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.12
+mpirun -N 12 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.12
+mpirun -N 12 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.12
+mpirun -N 12 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.12
+mpirun -N 12 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.12
+mpirun -N 12 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.12
+mpirun -N 12 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.12
+mpirun -N 12 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.12
+mpirun -N 12 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.12
+mpirun -N 12 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.12
+mpirun -N 12 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.12
+
+mpirun -N 13 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.13
+mpirun -N 13 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.13
+mpirun -N 13 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.13
+mpirun -N 13 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.13
+mpirun -N 13 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.13
+mpirun -N 13 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.13
+mpirun -N 13 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.13
+mpirun -N 13 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.13
+mpirun -N 13 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.13
+mpirun -N 13 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.13
+mpirun -N 13 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.13
+mpirun -N 13 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.13
+mpirun -N 13 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.13
+
+mpirun -N 14 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.14
+mpirun -N 14 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.14
+mpirun -N 14 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.14
+mpirun -N 14 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.14
+mpirun -N 14 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.14
+mpirun -N 14 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.14
+mpirun -N 14 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.14
+mpirun -N 14 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.14
+mpirun -N 14 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.14
+mpirun -N 14 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.14
+mpirun -N 14 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.14
+mpirun -N 14 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.14
+mpirun -N 14 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.14
+
+mpirun -N 15 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.15
+mpirun -N 15 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.15
+mpirun -N 15 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.15
+mpirun -N 15 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.15
+mpirun -N 15 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.15
+mpirun -N 15 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.15
+mpirun -N 15 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.15
+mpirun -N 15 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.15
+mpirun -N 15 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.15
+mpirun -N 15 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.15
+mpirun -N 15 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.15
+mpirun -N 15 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.15
+mpirun -N 15 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.15
+
+mpirun -N 16 ./lmp_cpu -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2K.16
+mpirun -N 16 ./lmp_cpu -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4K.16
+mpirun -N 16 ./lmp_cpu -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8K.16
+mpirun -N 16 ./lmp_cpu -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.16K.16
+mpirun -N 16 ./lmp_cpu -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.32K.16
+mpirun -N 16 ./lmp_cpu -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.64K.16
+mpirun -N 16 ./lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.128K.16
+mpirun -N 16 ./lmp_cpu -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.256K.16
+mpirun -N 16 ./lmp_cpu -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.512K.16
+mpirun -N 16 ./lmp_cpu -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.1024K.16
+mpirun -N 16 ./lmp_cpu -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.2048K.16
+mpirun -N 16 ./lmp_cpu -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.4096K.16
+mpirun -N 16 ./lmp_cpu -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cpu
+mv log.lammps log.28Jun14.lj.cpu.8192K.16
diff --git a/bench/KEPLER/run_gpu.sh b/bench/KEPLER/run_gpu.sh
new file mode 100644
index 0000000000..33faeccc69
--- /dev/null
+++ b/bench/KEPLER/run_gpu.sh
@@ -0,0 +1,1658 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.1
+mpirun -N 1 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.1
+
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.2
+
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.3
+mpirun -N 3 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.3
+
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.4
+
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.5
+mpirun -N 5 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.5
+
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.6
+
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.7
+mpirun -N 7 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.7
+
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.8
+
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.9
+mpirun -N 9 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.9
+
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.10
+
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.11
+mpirun -N 11 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.11
+
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.12
+
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.13
+mpirun -N 13 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.13
+
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.14
+
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.15
+mpirun -N 15 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.15
+
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.1.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.1.16
+
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.2.2
+mpirun -N 2 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.2.2
+
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.2.4
+mpirun -N 4 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.2.4
+
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.2.6
+mpirun -N 6 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.2.6
+
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.2.8
+mpirun -N 8 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.2.8
+
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.2.10
+mpirun -N 10 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.2.10
+
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.2.12
+mpirun -N 12 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.2.12
+
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.2.14
+mpirun -N 14 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.2.14
+
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.4K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.8K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.16K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.32K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.64K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.128K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.256K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.512K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.1024K.2.16
+mpirun -N 16 ./lmp_gpu_double -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.double.2048K.2.16
+
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.1
+mpirun -N 1 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.1
+
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.2
+
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.3
+mpirun -N 3 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.3
+
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.4
+
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.5
+mpirun -N 5 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.5
+
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.6
+
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.7
+mpirun -N 7 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.7
+
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.8
+
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.9
+mpirun -N 9 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.9
+
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.10
+
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.11
+mpirun -N 11 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.11
+
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.12
+
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.13
+mpirun -N 13 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.13
+
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.14
+
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.15
+mpirun -N 15 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.15
+
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.1.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.1.16
+
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.2.2
+mpirun -N 2 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.2.2
+
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.2.4
+mpirun -N 4 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.2.4
+
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.2.6
+mpirun -N 6 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.2.6
+
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.2.8
+mpirun -N 8 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.2.8
+
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.2.10
+mpirun -N 10 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.2.10
+
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.2.12
+mpirun -N 12 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.2.12
+
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.2.14
+mpirun -N 14 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.2.14
+
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.4K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.8K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.16K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.32K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.64K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.128K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.256K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.512K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.1024K.2.16
+mpirun -N 16 ./lmp_gpu_mixed -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.mixed.2048K.2.16
+
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.1
+mpirun -N 1 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.1
+
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.2
+
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.3
+mpirun -N 3 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.3
+
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.4
+
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.5
+mpirun -N 5 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.5
+
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.6
+
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.7
+mpirun -N 7 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.7
+
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.8
+
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.9
+mpirun -N 9 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.9
+
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.10
+
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.11
+mpirun -N 11 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.11
+
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.12
+
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.13
+mpirun -N 13 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.13
+
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.14
+
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.15
+mpirun -N 15 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.15
+
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.1.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.1.16
+
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.2.2
+mpirun -N 2 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.2.2
+
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.2.4
+mpirun -N 4 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.2.4
+
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.2.6
+mpirun -N 6 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.2.6
+
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.2.8
+mpirun -N 8 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.2.8
+
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.2.10
+mpirun -N 10 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.2.10
+
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.2.12
+mpirun -N 12 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.2.12
+
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.2.14
+mpirun -N 14 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.2.14
+
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.4K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.8K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.16K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.32K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.64K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.128K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.256K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.512K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.1024K.2.16
+mpirun -N 16 ./lmp_gpu_single -sf gpu -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.gpu
+mv log.lammps log.28Jun14.lj.gpu.single.2048K.2.16
diff --git a/bench/KEPLER/run_intel_cpu.sh b/bench/KEPLER/run_intel_cpu.sh
new file mode 100644
index 0000000000..0cc44a6205
--- /dev/null
+++ b/bench/KEPLER/run_intel_cpu.sh
@@ -0,0 +1,83 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.2K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.4K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.8K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.16K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.32K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.64K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.128K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.256K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.512K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.1024K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.2048K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.4096K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 1 -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.single.8192K.16
+
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.2K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.4K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.8K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.16K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.32K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.64K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.128K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.256K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.512K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.1024K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.2048K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.4096K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 3 -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.mixed.8192K.16
+
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.2K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.4K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.8K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.16K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.32K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.64K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.128K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.256K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.512K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.1024K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.2048K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.4096K.16
+mpirun -np 16 lmp_intel_cpu -sf intel -v a 2 -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.intel.cpu
+mv log.lammps log.28Jun14.lj.intel.cpu.double.8192K.16
diff --git a/bench/KEPLER/run_kokkos_cuda.sh b/bench/KEPLER/run_kokkos_cuda.sh
new file mode 100644
index 0000000000..778cd9acfb
--- /dev/null
+++ b/bench/KEPLER/run_kokkos_cuda.sh
@@ -0,0 +1,650 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.1
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 1 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.1
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.2
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 2 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.2
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.3
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 3 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.3
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.4
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 4 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.4
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.5
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 5 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.5
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.6
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 6 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.6
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.7
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 7 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.7
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.8
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 8 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.8
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.9
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 9 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.9
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.10
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 10 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.10
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.11
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 11 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.11
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.12
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 12 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.12
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.13
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 13 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.13
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.14
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 14 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.14
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.15
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 15 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.15
+
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.1.16
+mpirun -np 1 ./lmp_kokkos_cuda -k on g 1 t 16 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.1.16
+
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.2.1
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 1 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.2.1
+
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.2.2
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 2 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.2.2
+
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.2.3
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 3 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.2.3
+
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.2.4
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 4 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.2.4
+
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.2.5
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 5 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.2.5
+
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.2.6
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 6 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.2.6
+
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.2.7
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 7 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.2.7
+
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.16K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.32K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.64K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.128K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.256K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.512K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.1024K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.2048K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.4096K.2.8
+mpirun -np 2 ./lmp_kokkos_cuda -k on g 2 t 8 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.cuda
+mv log.lammps log.28Jun14.lj.kokkos.cuda.8192K.2.8
diff --git a/bench/KEPLER/run_kokkos_omp.sh b/bench/KEPLER/run_kokkos_omp.sh
new file mode 100644
index 0000000000..b1fa78c79d
--- /dev/null
+++ b/bench/KEPLER/run_kokkos_omp.sh
@@ -0,0 +1,137 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.2K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.4K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.8K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.16K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.32K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.64K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.128K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.256K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.512K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.1024K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.2048K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.4096K.1.8
+mpirun -np 1 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.8192K.1.8
+
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.2K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.4K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.8K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.16K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.32K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.64K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.128K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.256K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.512K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.1024K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.2048K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.4096K.2.8
+mpirun -np 2 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 8 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.8192K.2.8
+
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.2K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.4K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.8K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.16K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.32K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.64K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.128K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.256K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.512K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.1024K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.2048K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.4096K.4.4
+mpirun -np 4 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 4 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.8192K.4.4
+
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.2K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.4K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.8K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.16K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.32K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.64K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.128K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.256K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.512K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.1024K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.2048K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.4096K.8.2
+mpirun -np 8 -bind-to socket -map-by socket -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 2 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omp
+mv log.lammps log.28Jun14.lj.kokkos.omp.8192K.8.2
+
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.2K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.4K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.8K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.16K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.32K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.64K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.128K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.256K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.512K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.1024K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.2048K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.4096K.16.1
+mpirun -np 16 -bind-to core -map-by core -x KMP_AFFINITY=scatter ./lmp_kokkos_omp -k on t 1 -sf kk -v c device -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.kokkos.omphalf
+mv log.lammps log.28Jun14.lj.kokkos.omp.8192K.16.1
diff --git a/bench/KEPLER/run_opt.sh b/bench/KEPLER/run_opt.sh
new file mode 100644
index 0000000000..30d80a999f
--- /dev/null
+++ b/bench/KEPLER/run_opt.sh
@@ -0,0 +1,29 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -np 16 lmp_opt -sf opt -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.2K.16
+mpirun -np 16 lmp_opt -sf opt -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.4K.16
+mpirun -np 16 lmp_opt -sf opt -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.8K.16
+mpirun -np 16 lmp_opt -sf opt -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.16K.16
+mpirun -np 16 lmp_opt -sf opt -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.32K.16
+mpirun -np 16 lmp_opt -sf opt -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.64K.16
+mpirun -np 16 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.128K.16
+mpirun -np 16 lmp_opt -sf opt -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.256K.16
+mpirun -np 16 lmp_opt -sf opt -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.512K.16
+mpirun -np 16 lmp_opt -sf opt -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.1024K.16
+mpirun -np 16 lmp_opt -sf opt -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.2048K.16
+mpirun -np 16 lmp_opt -sf opt -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.4096K.16
+mpirun -np 16 lmp_opt -sf opt -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.opt
+mv log.lammps log.28Jun14.lj.opt.8192K.16
diff --git a/bench/KEPLER/run_user_cuda.sh b/bench/KEPLER/run_user_cuda.sh
new file mode 100644
index 0000000000..7e76177bbb
--- /dev/null
+++ b/bench/KEPLER/run_user_cuda.sh
@@ -0,0 +1,164 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.2K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.4K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.8K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.16K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.32K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.64K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.128K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.256K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.512K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.1024K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.2048K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.4096K.1
+mpirun -N 1 ./lmp_cuda_double -c on -sf cuda -v g 1 -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.8192K.1
+
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.2K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.4K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.8K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.16K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.32K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.64K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.128K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.256K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.512K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.1024K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.2048K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.4096K.2
+mpirun -N 2 ./lmp_cuda_double -c on -sf cuda -v g 2 -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.double.8192K.2
+
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.2K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.4K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.8K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.16K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.32K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.64K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.128K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.256K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.512K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.1024K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.2048K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.4096K.1
+mpirun -N 1 ./lmp_cuda_mixed -c on -sf cuda -v g 1 -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.8192K.1
+
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.2K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.4K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.8K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.16K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.32K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.64K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.128K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.256K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.512K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.1024K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.2048K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.4096K.2
+mpirun -N 2 ./lmp_cuda_mixed -c on -sf cuda -v g 2 -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.mixed.8192K.2
+
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.2K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.4K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.8K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.16K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.32K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.64K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.128K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.256K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.512K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.1024K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.2048K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.4096K.1
+mpirun -N 1 ./lmp_cuda_single -c on -sf cuda -v g 1 -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.8192K.1
+
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 8 -v y 8 -v z 8 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.2K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 8 -v y 8 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.4K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 8 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.8K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 16 -v y 16 -v z 16 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.16K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 16 -v y 16 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.32K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 16 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.64K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.128K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 32 -v y 32 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.256K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 32 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.512K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 64 -v y 64 -v z 64 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.1024K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 64 -v y 64 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.2048K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 64 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.4096K.2
+mpirun -N 2 ./lmp_cuda_single -c on -sf cuda -v g 2 -v x 128 -v y 128 -v z 128 -v t 100 < in.lj.cuda
+mv log.lammps log.28Jun14.lj.cuda.single.8192K.2
diff --git a/bench/KEPLER/run_user_omp.sh b/bench/KEPLER/run_user_omp.sh
new file mode 100644
index 0000000000..0b71f15b55
--- /dev/null
+++ b/bench/KEPLER/run_user_omp.sh
@@ -0,0 +1,137 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -bind-to none -N 1 ./lmp_omp -v x 8 -v y 8 -v z 8 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 8 -v y 8 -v z 16 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 8 -v y 16 -v z 16 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 16 -v y 16 -v z 16 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.16K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 16 -v y 16 -v z 32 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.32K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 16 -v y 32 -v z 32 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.64K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 32 -v y 32 -v z 32 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.128K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 32 -v y 32 -v z 64 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.256K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 32 -v y 64 -v z 64 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.512K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 64 -v y 64 -v z 64 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.1024K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 64 -v y 64 -v z 128 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2048K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 64 -v y 128 -v z 128 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4096K.1.16
+mpirun -bind-to none -N 1 ./lmp_omp -v x 128 -v y 128 -v z 128 -v t 100 -v h 16 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8192K.1.16
+
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 8 -v y 8 -v z 8 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 8 -v y 8 -v z 16 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 8 -v y 16 -v z 16 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 16 -v y 16 -v z 16 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.16K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 16 -v y 16 -v z 32 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.32K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 16 -v y 32 -v z 32 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.64K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 32 -v y 32 -v z 32 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.128K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 32 -v y 32 -v z 64 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.256K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 32 -v y 64 -v z 64 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.512K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 64 -v y 64 -v z 64 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.1024K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 64 -v y 64 -v z 128 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2048K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 64 -v y 128 -v z 128 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4096K.2.8
+mpirun -bind-to socket -N 2 ./lmp_omp -v x 128 -v y 128 -v z 128 -v t 100 -v h 8 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8192K.2.8
+
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 8 -v y 8 -v z 8 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 8 -v y 8 -v z 16 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 8 -v y 16 -v z 16 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 16 -v y 16 -v z 16 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.16K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 16 -v y 16 -v z 32 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.32K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 16 -v y 32 -v z 32 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.64K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 32 -v y 32 -v z 32 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.128K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 32 -v y 32 -v z 64 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.256K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 32 -v y 64 -v z 64 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.512K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 64 -v y 64 -v z 64 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.1024K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 64 -v y 64 -v z 128 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2048K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 64 -v y 128 -v z 128 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4096K.4.4
+mpirun -bind-to socket -N 4 ./lmp_omp -v x 128 -v y 128 -v z 128 -v t 100 -v h 4 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8192K.4.4
+
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 8 -v y 8 -v z 8 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 8 -v y 8 -v z 16 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 8 -v y 16 -v z 16 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 16 -v y 16 -v z 16 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.16K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 16 -v y 16 -v z 32 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.32K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 16 -v y 32 -v z 32 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.64K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 32 -v y 32 -v z 32 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.128K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 32 -v y 32 -v z 64 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.256K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 32 -v y 64 -v z 64 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.512K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 64 -v y 64 -v z 64 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.1024K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 64 -v y 64 -v z 128 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2048K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 64 -v y 128 -v z 128 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4096K.8.2
+mpirun -bind-to socket -N 8 ./lmp_omp -v x 128 -v y 128 -v z 128 -v t 100 -v h 2 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8192K.8.2
+
+mpirun -bind-to core -N 16 ./lmp_omp -v x 8 -v y 8 -v z 8 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 8 -v y 8 -v z 16 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 8 -v y 16 -v z 16 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 16 -v y 16 -v z 16 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.16K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 16 -v y 16 -v z 32 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.32K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 16 -v y 32 -v z 32 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.64K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 32 -v y 32 -v z 32 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.128K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 32 -v y 32 -v z 64 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.256K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 32 -v y 64 -v z 64 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.512K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 64 -v y 64 -v z 64 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.1024K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 64 -v y 64 -v z 128 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.2048K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 64 -v y 128 -v z 128 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.4096K.16.1
+mpirun -bind-to core -N 16 ./lmp_omp -v x 128 -v y 128 -v z 128 -v t 100 -v h 1 -sf omp < in.lj.omp
+mv log.lammps log.28Jun14.lj.omp.8192K.16.1
diff --git a/bench/README b/bench/README
index 4bd97de124..3c02a6b176 100644
--- a/bench/README
+++ b/bench/README
@@ -4,19 +4,19 @@ This directory contains 5 benchmark problems which are discussed in
the Benchmark section of the LAMMPS documentation, and on the
Benchmark page of the LAMMPS WWW site (lammps.sandia.gov/bench).
-This directory also has 2 sub-directories:
+This directory also has several sub-directories:
-GPU GPU versions of 3 of these benchmarks
-POTENTIALS benchmarks scripts for various potentials in LAMMPS
+FERMI benchmark scripts for desktop machine with Fermi GPUs (Tesla)
+KEPLER benchmark scripts for GPU cluster with Kepler GPUs
+POTENTIALS benchmarks scripts for various potentials in LAMMPS
-The scripts and results in both of these directories are discussed on
-the Benchmark page of the LAMMPS WWW site (lammps.sandia.gov/bench) as
-well. The two directories have their own README files which you
-should refer to before running the scripts.
+The results for all of these benchmarks are displayed and discussed on
+the Benchmark page of the LAMMPS WWW site: lammps.sandia.gov/bench.
-The remainer of this file refers to the 5 problems in the top-level
-of this directory and how to run them on CPUs, either in serial
-or parallel.
+The remainder of this file refers to the 5 problems in the top-level
+of this directory and how to run them on CPUs, either in serial or
+parallel. The directories have their own README files which you
+should refer to before running those scripts.
----------------------------------------------------------------------
diff --git a/doc/Section_accelerate.html b/doc/Section_accelerate.html
index 3e9658f75e..224ead93fa 100644
--- a/doc/Section_accelerate.html
+++ b/doc/Section_accelerate.html
@@ -520,10 +520,16 @@ LAMMPS.
(a) Build the GPU library
The GPU library is in lammps/lib/gpu. Select a Makefile.machine (in
-lib/gpu) appropriate for your system.
+lib/gpu) appropriate for your system. You should pay special
+attention to 3 settings in this makefile.
-Before building the library, you can set its precision by editing the
-CUDA_PREC setting in Makefile.machine, as follows:
+
- CUDA_HOME = needs to be where NVIDIA Cuda software is installed on your system
+
- CUDA_ARCH = needs to be appropriate to your GPUs
+
- CUDA_PREC = precision (double, mixed, single) you desire
+
+See lib/gpu/Makefile.linux.double for examples of the ARCH settings
+for different GPU choices, e.g. Fermi vs Kepler. It also lists the
+possible precision settings:
CUDA_PREC = -D_SINGLE_SINGLE # Single precision for all calculations
CUDA_PREC = -D_DOUBLE_DOUBLE # Double precision for all calculations
@@ -745,6 +751,7 @@ the library is built.
N = 3 for positions in double precision
N = 4 for positions and velocities in double precision
arch=M to set GPU compute capability
+ M = 35 for Kepler GPUs
M = 20 for CC2.0 (GF100/110, e.g. C2050,GTX580,GTX470) (default)
M = 21 for CC2.1 (GF104/114, e.g. GTX560, GTX460, GTX450)
M = 13 for CC1.3 (GF200, e.g. C1060, GTX285)
diff --git a/doc/Section_accelerate.txt b/doc/Section_accelerate.txt
index 649a448f13..de7c9473bc 100644
--- a/doc/Section_accelerate.txt
+++ b/doc/Section_accelerate.txt
@@ -513,10 +513,16 @@ LAMMPS.
(a) Build the GPU library
The GPU library is in lammps/lib/gpu. Select a Makefile.machine (in
-lib/gpu) appropriate for your system.
+lib/gpu) appropriate for your system. You should pay special
+attention to 3 settings in this makefile.
-Before building the library, you can set its precision by editing the
-CUDA_PREC setting in Makefile.machine, as follows:
+CUDA_HOME = needs to be where NVIDIA Cuda software is installed on your system
+CUDA_ARCH = needs to be appropriate to your GPUs
+CUDA_PREC = precision (double, mixed, single) you desire :ul
+
+See lib/gpu/Makefile.linux.double for examples of the ARCH settings
+for different GPU choices, e.g. Fermi vs Kepler. It also lists the
+possible precision settings:
CUDA_PREC = -D_SINGLE_SINGLE # Single precision for all calculations
CUDA_PREC = -D_DOUBLE_DOUBLE # Double precision for all calculations
@@ -738,6 +744,7 @@ the library is built.
N = 3 for positions in double precision
N = 4 for positions and velocities in double precision
{arch=M} to set GPU compute capability
+ M = 35 for Kepler GPUs
M = 20 for CC2.0 (GF100/110, e.g. C2050,GTX580,GTX470) (default)
M = 21 for CC2.1 (GF104/114, e.g. GTX560, GTX460, GTX450)
M = 13 for CC1.3 (GF200, e.g. C1060, GTX285)
diff --git a/doc/Section_start.html b/doc/Section_start.html
index 6448a72a76..b75d7cc223 100644
--- a/doc/Section_start.html
+++ b/doc/Section_start.html
@@ -1222,9 +1222,9 @@ supercomputer there may be dozens or 1000s of physical nodes.
Either the full word or an abbreviation can be used for the keywords.
Note that the keywords do not use a leading minus sign. I.e. the
keyword is "t", not "-t". Also note that each of the keywords has a
-default setting. More explanation as to when to use these options and
-what settings to use on different platforms is given in this
-section.
+default setting. Example of when to use these options and what
+settings to use on different platforms is given in Section
+5.8.
- d or device
- g or gpus
diff --git a/doc/Section_start.txt b/doc/Section_start.txt
index 89270d4846..70afcad0ae 100644
--- a/doc/Section_start.txt
+++ b/doc/Section_start.txt
@@ -1216,9 +1216,9 @@ supercomputer there may be dozens or 1000s of physical nodes.
Either the full word or an abbreviation can be used for the keywords.
Note that the keywords do not use a leading minus sign. I.e. the
keyword is "t", not "-t". Also note that each of the keywords has a
-default setting. More explanation as to when to use these options and
-what settings to use on different platforms is given in "this
-section"_Section_accerlerate.html#acc_8.
+default setting. Example of when to use these options and what
+settings to use on different platforms is given in "Section
+5.8"_Section_accelerate.html#acc_8.
d or device
g or gpus
diff --git a/lib/cuda/Makefile.defaults b/lib/cuda/Makefile.defaults
index 896983a7a4..e0f496a855 100644
--- a/lib/cuda/Makefile.defaults
+++ b/lib/cuda/Makefile.defaults
@@ -5,7 +5,7 @@ precision ?= 4
#verbose setting: 0 no, 1 yes
verbose ?= 1
-#GPU architecture (compute capability): 13, 20, 21
+#GPU architecture (compute capability): 13, 20, 21, 30, 35
arch ?= 21
#Using cufft (should not be changed)
diff --git a/lib/cuda/Makefile.lammps b/lib/cuda/Makefile.lammps
index 17efca469a..a9ce5fdf53 100644
--- a/lib/cuda/Makefile.lammps
+++ b/lib/cuda/Makefile.lammps
@@ -5,4 +5,3 @@ CUDA_USRLIB_CONDITIONAL := -L$(CUDA_HOME)/lib -L$(CUDA_HOME)/lib64 -lcufft
user-cuda_SYSINC = ${CUDA_FLAGS}
user-cuda_SYSLIB = -lcuda -lcudart
user-cuda_SYSPATH = $(CUDA_USRLIB_CONDITIONAL)
-
diff --git a/lib/gpu/Makefile.linux.double b/lib/gpu/Makefile.linux.double
index 1574cd9568..defbe2bc9c 100644
--- a/lib/gpu/Makefile.linux.double
+++ b/lib/gpu/Makefile.linux.double
@@ -10,6 +10,8 @@ EXTRAMAKE = Makefile.lammps.standard
CUDA_HOME = /usr/local/cuda
NVCC = nvcc
+# Kepler CUDA
+#CUDA_ARCH = -arch=sm_35
# Tesla CUDA
CUDA_ARCH = -arch=sm_21
# newer CUDA
diff --git a/lib/gpu/Makefile.linux.mixed b/lib/gpu/Makefile.linux.mixed
index fc01eea46f..42fc8e9740 100644
--- a/lib/gpu/Makefile.linux.mixed
+++ b/lib/gpu/Makefile.linux.mixed
@@ -8,8 +8,11 @@
EXTRAMAKE = Makefile.lammps.standard
CUDA_HOME = /usr/local/cuda
+CUDA_HOME = /home/projects/openmpi/1.8.1/intel/13.1.SP1.106/cuda/6.0.37
NVCC = nvcc
+# Kepler CUDA
+#CUDA_ARCH = -arch=sm_35
# Tesla CUDA
CUDA_ARCH = -arch=sm_21
# newer CUDA
diff --git a/lib/gpu/Makefile.linux.single b/lib/gpu/Makefile.linux.single
index d0034b0b1f..4048bc8788 100644
--- a/lib/gpu/Makefile.linux.single
+++ b/lib/gpu/Makefile.linux.single
@@ -10,6 +10,8 @@ EXTRAMAKE = Makefile.lammps.standard
CUDA_HOME = /usr/local/cuda
NVCC = nvcc
+# Kepler CUDA
+#CUDA_ARCH = -arch=sm_35
# Tesla CUDA
CUDA_ARCH = -arch=sm_21
# newer CUDA