diff --git a/bench/KEPLER/Makefile.omp b/bench/KEPLER/Makefile.omp
index 41c430f0f..d8ac8e1ae 100755
--- a/bench/KEPLER/Makefile.omp
+++ b/bench/KEPLER/Makefile.omp
@@ -1,108 +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
+CCFLAGS =	-O3 -openmp -restrict -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
index f27d7e930..a1855156a 100755
--- a/bench/KEPLER/Makefile.opt
+++ b/bench/KEPLER/Makefile.opt
@@ -1,108 +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
+CCFLAGS =	-O -restrict
 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
index feee3e0b4..0bf19945f 100644
--- a/bench/KEPLER/README
+++ b/bench/KEPLER/README
@@ -1,71 +1,68 @@
 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.
+packages installed 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 in.* files can be run with any of the accelerator packages,
+if you specify the appropriate command-line switches.  These
+include switches to set the problem size and number of timesteps
+to 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.
+arguments are used with each accelerator package.
 
 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.
+batch submission, via Python scripts which often produces 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/run_cpu.sh b/bench/KEPLER/run_cpu.sh
index b08d606c9..78fe313a2 100644
--- a/bench/KEPLER/run_cpu.sh
+++ b/bench/KEPLER/run_cpu.sh
@@ -1,434 +1,29 @@
 #!/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 -np 1 lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cpu.128K.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 -np 2 lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cpu.128K.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 -np 4 lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cpu.128K.4
 
-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 -np 6 lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cpu.128K.6
 
-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 -np 8 lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cpu.128K.8
 
-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 -np 10 lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cpu.128K.10
 
-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 -np 12 lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cpu.128K.12
 
-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 -np 14 lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cpu.128K.14
 
-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
+mpirun -np 16 lmp_cpu -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cpu.128K.16
diff --git a/bench/KEPLER/run_cuda.sh b/bench/KEPLER/run_cuda.sh
new file mode 100644
index 000000000..6805a0892
--- /dev/null
+++ b/bench/KEPLER/run_cuda.sh
@@ -0,0 +1,20 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -N 1 lmp_cuda_double -c on -sf cuda -pk cuda 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cuda.double.128K.1
+
+mpirun -N 2 lmp_cuda_double -c on -sf cuda -pk cuda 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cuda.double.128K.2
+
+mpirun -N 1 lmp_cuda_mixed -c on -sf cuda -pk cuda 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cuda.mixed.128K.1
+
+mpirun -N 2 lmp_cuda_mixed -c on -sf cuda -pk cuda 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cuda.mixed.128K.2
+
+mpirun -N 1 lmp_cuda_single -c on -sf cuda -pk cuda 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cuda.single.128K.1
+
+mpirun -N 2 lmp_cuda_single -c on -sf cuda -pk cuda 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.cuda.single.128K.2
diff --git a/bench/KEPLER/run_gpu.sh b/bench/KEPLER/run_gpu.sh
index 33faeccc6..80210f988 100644
--- a/bench/KEPLER/run_gpu.sh
+++ b/bench/KEPLER/run_gpu.sh
@@ -1,1658 +1,155 @@
 #!/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 -np 1 lmp_gpu_single -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.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 -np 2 lmp_gpu_single -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.2.1
 
-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 -np 2 lmp_gpu_single -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.2.2
 
-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 -np 4 lmp_gpu_single -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.4.1
 
-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 -np 4 lmp_gpu_single -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.4.2
 
-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 -np 6 lmp_gpu_single -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.6.1
 
-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 -np 6 lmp_gpu_single -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.6.2
 
-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 -np 8 lmp_gpu_single -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.8.1
 
-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 -np 8 lmp_gpu_single -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.8.2
 
-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 -np 10 lmp_gpu_single -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.10.1
 
-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 -np 10 lmp_gpu_single -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.10.2
 
-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 -np 12 lmp_gpu_single -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.12.1
 
-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 -np 12 lmp_gpu_single -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.12.2
 
-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 -np 14 lmp_gpu_single -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.14.1
 
-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 -np 14 lmp_gpu_single -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.14.2
 
-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 -np 16 lmp_gpu_single -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.16.1
 
-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 -np 16 lmp_gpu_single -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.single.128K.16.2
 
-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 -np 1 lmp_gpu_mixed -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.1.1
 
-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 -np 2 lmp_gpu_mixed -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.2.1
 
-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 -np 2 lmp_gpu_mixed -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.2.2
 
-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 -np 4 lmp_gpu_mixed -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.4.1
 
-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 -np 4 lmp_gpu_mixed -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.4.2
 
-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 -np 6 lmp_gpu_mixed -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.6.1
 
-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 -np 6 lmp_gpu_mixed -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.6.2
 
-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 -np 8 lmp_gpu_mixed -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.8.1
 
-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 -np 8 lmp_gpu_mixed -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.8.2
 
-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 -np 10 lmp_gpu_mixed -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.10.1
 
-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 -np 10 lmp_gpu_mixed -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.10.2
 
-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 -np 12 lmp_gpu_mixed -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.12.1
 
-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 -np 12 lmp_gpu_mixed -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.12.2
 
-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 -np 14 lmp_gpu_mixed -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.14.1
 
-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 -np 14 lmp_gpu_mixed -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.14.2
 
-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 -np 16 lmp_gpu_mixed -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.16.1
 
-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 -np 16 lmp_gpu_mixed -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.mixed.128K.16.2
 
-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 -np 1 lmp_gpu_double -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.1.1
 
-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 -np 2 lmp_gpu_double -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.2.1
 
-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 -np 2 lmp_gpu_double -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.2.2
 
-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 -np 4 lmp_gpu_double -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.4.1
 
-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 -np 4 lmp_gpu_double -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.4.2
 
-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 -np 6 lmp_gpu_double -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.6.1
 
-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 -np 6 lmp_gpu_double -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.6.2
 
-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 -np 8 lmp_gpu_double -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.8.1
 
-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 -np 8 lmp_gpu_double -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.8.2
 
-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 -np 10 lmp_gpu_double -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.10.1
 
-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 -np 10 lmp_gpu_double -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.10.2
 
-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 -np 12 lmp_gpu_double -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.12.1
 
-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 -np 12 lmp_gpu_double -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.12.2
 
-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 -np 14 lmp_gpu_double -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.14.1
 
-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 -np 14 lmp_gpu_double -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.14.2
 
-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 -np 16 lmp_gpu_double -sf gpu -pk gpu 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.16.1
 
-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
+mpirun -np 16 lmp_gpu_double -sf gpu -pk gpu 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.gpu.double.128K.16.2
diff --git a/bench/KEPLER/run_intel_cpu.sh b/bench/KEPLER/run_intel_cpu.sh
index 0cc44a620..f185dfe05 100644
--- a/bench/KEPLER/run_intel_cpu.sh
+++ b/bench/KEPLER/run_intel_cpu.sh
@@ -1,83 +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
+mpirun -np 1 lmp_intel_cpu -sf intel -pk intel 1 prec single -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.single.128K.1
+
+mpirun -np 2 lmp_intel_cpu -sf intel -pk intel 1 prec single -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.single.128K.2
+
+mpirun -np 4 lmp_intel_cpu -sf intel -pk intel 1 prec single -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.single.128K.4
+
+mpirun -np 6 lmp_intel_cpu -sf intel -pk intel 1 prec single -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.single.128K.6
+
+mpirun -np 8 lmp_intel_cpu -sf intel -pk intel 1 prec single -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.single.128K.8
+
+mpirun -np 10 lmp_intel_cpu -sf intel -pk intel 1 prec single -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.single.128K.10
+
+mpirun -np 12 lmp_intel_cpu -sf intel -pk intel 1 prec single -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.single.128K.12
+
+mpirun -np 14 lmp_intel_cpu -sf intel -pk intel 1 prec single -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.single.128K.14
+
+mpirun -np 16 lmp_intel_cpu -sf intel -pk intel 1 prec single -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.single.128K.16
+
+mpirun -np 1 lmp_intel_cpu -sf intel -pk intel 1 prec mixed -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.mixed.128K.1
+
+mpirun -np 2 lmp_intel_cpu -sf intel -pk intel 1 prec mixed -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.mixed.128K.2
+
+mpirun -np 4 lmp_intel_cpu -sf intel -pk intel 1 prec mixed -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.mixed.128K.4
+
+mpirun -np 6 lmp_intel_cpu -sf intel -pk intel 1 prec mixed -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.mixed.128K.6
+
+mpirun -np 8 lmp_intel_cpu -sf intel -pk intel 1 prec mixed -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.mixed.128K.8
+
+mpirun -np 10 lmp_intel_cpu -sf intel -pk intel 1 prec mixed -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.mixed.128K.10
+
+mpirun -np 12 lmp_intel_cpu -sf intel -pk intel 1 prec mixed -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.mixed.128K.12
+
+mpirun -np 14 lmp_intel_cpu -sf intel -pk intel 1 prec mixed -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.mixed.128K.14
+
+mpirun -np 16 lmp_intel_cpu -sf intel -pk intel 1 prec mixed -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.mixed.128K.16
+
+mpirun -np 1 lmp_intel_cpu -sf intel -pk intel 1 prec double -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.double.128K.1
+
+mpirun -np 2 lmp_intel_cpu -sf intel -pk intel 1 prec double -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.double.128K.2
+
+mpirun -np 4 lmp_intel_cpu -sf intel -pk intel 1 prec double -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.double.128K.4
+
+mpirun -np 6 lmp_intel_cpu -sf intel -pk intel 1 prec double -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.double.128K.6
+
+mpirun -np 8 lmp_intel_cpu -sf intel -pk intel 1 prec double -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.double.128K.8
+
+mpirun -np 10 lmp_intel_cpu -sf intel -pk intel 1 prec double -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.double.128K.10
+
+mpirun -np 12 lmp_intel_cpu -sf intel -pk intel 1 prec double -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.double.128K.12
+
+mpirun -np 14 lmp_intel_cpu -sf intel -pk intel 1 prec double -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.double.128K.14
+
+mpirun -np 16 lmp_intel_cpu -sf intel -pk intel 1 prec double -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.intel.cpu.double.128K.16
diff --git a/bench/KEPLER/run_kokkos_cuda.sh b/bench/KEPLER/run_kokkos_cuda.sh
index 778cd9acf..c0fb80bd0 100644
--- a/bench/KEPLER/run_kokkos_cuda.sh
+++ b/bench/KEPLER/run_kokkos_cuda.sh
@@ -1,650 +1,74 @@
 #!/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 1 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 2 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 3 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 4 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 5 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 6 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 7 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 8 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 9 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 10 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 11 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 12 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 13 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 14 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 15 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 1 lmp_kokkos_cuda -k on g 1 t 16 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 1 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 2 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 3 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 4 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 5 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 6 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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 7 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.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
+mpirun -np 2 lmp_kokkos_cuda -k on g 2 t 8 -sf kk -pk kokkos binsize 2.8 comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.cuda.128K.2.8
diff --git a/bench/KEPLER/run_kokkos_omp.sh b/bench/KEPLER/run_kokkos_omp.sh
index b1fa78c79..29af431c9 100644
--- a/bench/KEPLER/run_kokkos_omp.sh
+++ b/bench/KEPLER/run_kokkos_omp.sh
@@ -1,137 +1,17 @@
 #!/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 full -bind-to socket -map-by socket -x KMP_AFFINITY=scatter lmp_kokkos_omp -k on t 16 -sf kk -pk kokkos neigh full newton off comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.omp.128K.1.16
 
-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 full -bind-to socket -map-by socket -x KMP_AFFINITY=scatter lmp_kokkos_omp -k on t 8 -sf kk -pk kokkos neigh full newton off comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.omp.128K.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 full -bind-to socket -map-by socket -x KMP_AFFINITY=scatter lmp_kokkos_omp -k on t 4 -sf kk -pk kokkos neigh full newton off comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.omp.128K.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 full -bind-to socket -map-by socket -x KMP_AFFINITY=scatter lmp_kokkos_omp -k on t 2 -sf kk -pk kokkos neigh full newton off comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.omp.128K.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
+mpirun -np half -bind-to socket -map-by socket -x KMP_AFFINITY=scatter lmp_kokkos_omp -k on t 1 -sf kk -pk kokkos neigh half newton on comm device -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.kokkos.omp.128K.16.1
diff --git a/bench/KEPLER/run_omp.sh b/bench/KEPLER/run_omp.sh
new file mode 100644
index 000000000..8e91ad894
--- /dev/null
+++ b/bench/KEPLER/run_omp.sh
@@ -0,0 +1,17 @@
+#!/bin/bash
+#SBATCH -N 1 --time=12:00:00
+
+mpirun -np 1 lmp_omp -sf omp -pk omp 16 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.omp.128K.1.16
+
+mpirun -np 2 lmp_omp -sf omp -pk omp 8 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.omp.128K.2.8
+
+mpirun -np 4 lmp_omp -sf omp -pk omp 4 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.omp.128K.4.4
+
+mpirun -np 8 lmp_omp -sf omp -pk omp 2 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.omp.128K.8.2
+
+mpirun -np 16 lmp_omp -sf omp -pk omp 1 -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.omp.128K.16.1
diff --git a/bench/KEPLER/run_opt.sh b/bench/KEPLER/run_opt.sh
index 30d80a999..c2772aabc 100644
--- a/bench/KEPLER/run_opt.sh
+++ b/bench/KEPLER/run_opt.sh
@@ -1,29 +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
+mpirun -np 1 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.opt.128K.1
+
+mpirun -np 2 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.opt.128K.2
+
+mpirun -np 4 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.opt.128K.4
+
+mpirun -np 6 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.opt.128K.6
+
+mpirun -np 8 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.opt.128K.8
+
+mpirun -np 10 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.opt.128K.10
+
+mpirun -np 12 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.opt.128K.12
+
+mpirun -np 14 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.opt.128K.14
+
+mpirun -np 16 lmp_opt -sf opt -v x 32 -v y 32 -v z 32 -v t 100 < in.lj
+mv log.lammps log.10Sep14.lj.opt.128K.16
diff --git a/bench/KEPLER/run_user_cuda.sh b/bench/KEPLER/run_user_cuda.sh
deleted file mode 100644
index 7e76177bb..000000000
--- a/bench/KEPLER/run_user_cuda.sh
+++ /dev/null
@@ -1,164 +0,0 @@
-#!/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
deleted file mode 100644
index 0b71f15b5..000000000
--- a/bench/KEPLER/run_user_omp.sh
+++ /dev/null
@@ -1,137 +0,0 @@
-#!/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