diff --git a/src/GPU/Install.sh b/src/GPU/Install.sh
index 32ee689be..4f1659989 100644
--- a/src/GPU/Install.sh
+++ b/src/GPU/Install.sh
@@ -1,148 +1,155 @@
 # Install/unInstall package files in LAMMPS
 # edit 2 Makefile.package files to include/exclude GPU info
 # do not install child files if parent does not exist
 
 if (test $1 = 1) then
 
   if (test -e ../Makefile.package) then
     sed -i -e 's/[^ \t]*gpu[^ \t]* //' ../Makefile.package
     sed -i -e 's|^PKG_PATH =[ \t]*|&-L../../lib/gpu |' ../Makefile.package
     sed -i -e 's|^PKG_LIB =[ \t]*|&-lgpu |' ../Makefile.package
     sed -i -e 's|^PKG_SYSINC =[ \t]*|&$(gpu_SYSINC) |' ../Makefile.package
     sed -i -e 's|^PKG_SYSLIB =[ \t]*|&$(gpu_SYSLIB) |' ../Makefile.package
     sed -i -e 's|^PKG_SYSPATH =[ \t]*|&$(gpu_SYSPATH) |' ../Makefile.package
   fi
 
   if (test -e ../Makefile.package.settings) then
     sed -i -e '/^include.*gpu.*$/d' ../Makefile.package.settings
     sed -i '4 i include ..\/..\/lib\/gpu\/Makefile.lammps' ../Makefile.package.settings
   fi
   
   if (test -e ../pair_gayberne.cpp) then
     cp pair_gayberne_gpu.cpp ..
     cp pair_gayberne_gpu.h ..
     cp pair_resquared_gpu.cpp ..
     cp pair_resquared_gpu.h ..
   fi
   
   if (test -e ../pair_lj_cut_coul_long.cpp) then
     cp pair_lj_cut_coul_long_gpu.cpp ..
     cp pair_lj_cut_coul_long_gpu.h ..
   fi
 
   if (test -e ../pair_lj_class2.cpp) then
     cp pair_lj_class2_gpu.cpp ..
     cp pair_lj_class2_gpu.h ..
   fi
 
   if (test -e ../pair_lj_class2_coul_long.cpp) then
     cp pair_lj_class2_coul_long_gpu.cpp ..
     cp pair_lj_class2_coul_long_gpu.h ..
   fi
 
   if (test -e ../pair_lj_charmm_coul_long.cpp) then
     cp pair_lj_charmm_coul_long_gpu.cpp ..
     cp pair_lj_charmm_coul_long_gpu.h ..
   fi
 
+  if (test -e ../pair_coul_long.cpp) then
+    cp pair_coul_long_gpu.cpp ..
+    cp pair_coul_long_gpu.h ..
+  fi
+
   if (test -e ../pair_cg_cmm.cpp) then
     cp pair_cg_cmm_gpu.cpp ..
     cp pair_cg_cmm_gpu.h ..
   fi
 
   if (test -e ../pair_cg_cmm_coul_long.cpp) then
     cp pair_cg_cmm_coul_long_gpu.cpp ..
     cp pair_cg_cmm_coul_long_gpu.h ..
     cp pair_cg_cmm_coul_msm.cpp ..
     cp pair_cg_cmm_coul_msm.h ..
     cp pair_cg_cmm_coul_msm_gpu.cpp ..
     cp pair_cg_cmm_coul_msm_gpu.h ..
   fi
 
   if (test -e ../pppm.cpp) then
     cp pppm_gpu.cpp ..
     cp pppm_gpu.h ..
   fi
 
   cp pair_lj_cut_gpu.cpp ..
   cp pair_morse_gpu.cpp ..
   cp pair_lj96_cut_gpu.cpp ..
   cp pair_lj_expand_gpu.cpp ..
   cp pair_lj_cut_coul_cut_gpu.cpp ..
   cp pair_lj_cut_tgpu.cpp ..
 
   cp fix_gpu.cpp ..
 
   cp pair_lj_cut_gpu.h ..
   cp pair_morse_gpu.h ..
   cp pair_lj96_cut_gpu.h ..
   cp pair_lj_expand_gpu.h ..
   cp pair_lj_cut_coul_cut_gpu.h ..
   cp pair_lj_cut_tgpu.h ..
   
   cp fix_gpu.h ..
   cp gpu_extra.h ..
 
   cp pair_omp_gpu.cpp ..
   cp pair_lj_cut_tgpu.cpp ..
 
   cp pair_omp_gpu.h ..
   cp pair_lj_cut_tgpu.h ..
   
 elif (test $1 = 0) then
 
   if (test -e ../Makefile.package) then
     sed -i -e 's/[^ \t]*gpu[^ \t]* //' ../Makefile.package
   fi
   
   if (test -e ../Makefile.package.settings) then
     sed -i -e '/^include.*gpu.*$/d' ../Makefile.package.settings
   fi
 
   rm -f ../pppm_gpu.cpp
   rm -f ../pair_gayberne_gpu.cpp
   rm -f ../pair_resquared_gpu.cpp
   rm -f ../pair_lj_cut_gpu.cpp
   rm -f ../pair_morse_gpu.cpp
   rm -f ../pair_lj96_cut_gpu.cpp
   rm -f ../pair_lj_expand_gpu.cpp
   rm -f ../pair_lj_cut_coul_cut_gpu.cpp
   rm -f ../pair_lj_cut_coul_long_gpu.cpp
   rm -f ../pair_lj_class2_gpu.cpp
   rm -f ../pair_lj_class2_coul_long_gpu.cpp
   rm -f ../pair_lj_charmm_coul_long_gpu.cpp
   rm -f ../pair_lj_cut_tgpu.cpp
+  rm -f ../pair_coul_long_gpu.cpp
   rm -f ../pair_cg_cmm_gpu.cpp
   rm -f ../pair_cg_cmm_coul_long_gpu.cpp
   rm -f ../pair_cg_cmm_coul_msm.cpp
   rm -f ../pair_cg_cmm_coul_msm_gpu.cpp
 
   rm -f ../fix_gpu.cpp
   rm -f ../pair_omp_gpu.cpp
   rm -f ../pair_lj_cut_tgpu.cpp
 
   rm -f ../pppm_gpu.h
   rm -f ../pair_gayberne_gpu.h
   rm -f ../pair_resquared_gpu.h
   rm -f ../pair_lj_cut_gpu.h
   rm -f ../pair_morse_gpu.h
   rm -f ../pair_lj96_cut_gpu.h
   rm -f ../pair_lj_expand_gpu.h
   rm -f ../pair_lj_cut_coul_cut_gpu.h
   rm -f ../pair_lj_cut_coul_long_gpu.h
   rm -f ../pair_lj_class2_gpu.h
   rm -f ../pair_lj_class2_coul_long_gpu.h
   rm -f ../pair_lj_charmm_coul_long_gpu.h
   rm -f ../pair_lj_cut_tgpu.h
+  rm -f ../pair_coul_long_gpu.h
   rm -f ../pair_cg_cmm_gpu.h
   rm -f ../pair_cg_cmm_coul_long_gpu.h
   rm -f ../pair_cg_cmm_coul_msm.h
   rm -f ../pair_cg_cmm_coul_msm_gpu.h
 
   rm -f ../fix_gpu.h
   rm -f ../gpu_extra.h
   rm -f ../pair_omp_gpu.h
   rm -f ../pair_lj_cut_tgpu.h
   
 fi
 
diff --git a/src/GPU/Package.sh b/src/GPU/Package.sh
index 60aa6820c..2eb20755e 100644
--- a/src/GPU/Package.sh
+++ b/src/GPU/Package.sh
@@ -1,50 +1,56 @@
 # Update package files in LAMMPS
 # cp package file to src if doesn't exist or is different
 # do not copy gayberne files if non-GPU version does not exist
 
 for file in *.cpp *.h; do
   if (test $file = pair_gayberne_gpu.cpp -a ! -e ../pair_gayberne.cpp) then
     continue
   fi
   if (test $file = pair_gayberne_gpu.h -a ! -e ../pair_gayberne.cpp) then
     continue
   fi
   if (test $file = pair_lj_cut_coul_long_gpu.cpp -a ! -e ../pair_lj_cut_coul_long.cpp) then
     continue
   fi
   if (test $file = pair_lj_cut_coul_long_gpu.h -a ! -e ../pair_lj_cut_coul_long.cpp) then
     continue
   fi
+  if (test $file = pair_coul_long_gpu.cpp -a ! -e ../pair_coul_long.cpp) then
+    continue
+  fi
+  if (test $file = pair_coul_long_gpu.h -a ! -e ../pair_coul_long.cpp) then
+    continue
+  fi
   if (test $file = pair_cg_cmm_gpu.cpp -a ! -e ../pair_cg_cmm.cpp) then
     continue
   fi
   if (test $file = pair_cg_cmm_gpu.h -a ! -e ../pair_cg_cmm.cpp) then
     continue
   fi
   if (test $file = pair_cg_cmm_coul_long_gpu.cpp -a ! -e ../pair_cg_cmm_coul_long.cpp) then
     continue
   fi
   if (test $file = pair_cg_cmm_coul_long_gpu.h -a ! -e ../pair_cg_cmm_coul_long.cpp) then
     continue
   fi
   if (test $file = pair_cg_cmm_coul_msm.cpp -a ! -e ../pair_cg_cmm.cpp) then
     continue
   fi
   if (test $file = pair_cg_cmm_coul_msm.h -a ! -e ../pair_cg_cmm.cpp) then
     continue
   fi
   if (test $file = pair_cg_cmm_coul_msm_gpu.cpp -a ! -e ../pair_cg_cmm.cpp) then
     continue
   fi
   if (test $file = pair_cg_cmm_coul_msm_gpu.h -a ! -e ../pair_cg_cmm.cpp) then
     continue
   fi
 
   if (test ! -e ../$file) then
     echo "  creating src/$file"
     cp $file ..
   elif (test "`diff --brief $file ../$file`" != "") then
     echo "  updating src/$file"
     cp $file ..
   fi
 done
diff --git a/src/GPU/pair_cg_cmm_coul_long_gpu.cpp b/src/GPU/pair_cg_cmm_coul_long_gpu.cpp
index 153cb98a9..d19b2d651 100644
--- a/src/GPU/pair_cg_cmm_coul_long_gpu.cpp
+++ b/src/GPU/pair_cg_cmm_coul_long_gpu.cpp
@@ -1,340 +1,340 @@
 /* ----------------------------------------------------------------------
    LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
    http://lammps.sandia.gov, Sandia National Laboratories
    Steve Plimpton, sjplimp@sandia.gov
    
    Copyright (2003) Sandia Corporation.  Under the terms of Contract
    DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
    certain rights in this software.  This software is distributed under 
    the GNU General Public License.
    
    See the README file in the top-level LAMMPS directory.
 ------------------------------------------------------------------------- */
 
 /* ----------------------------------------------------------------------
    Contributing author: Mike Brown (SNL)
 ------------------------------------------------------------------------- */
 
 #include "lmptype.h"
 #include "math.h"
 #include "stdio.h"
 #include "stdlib.h"
 #include "pair_cg_cmm_coul_long_gpu.h"
 #include "atom.h"
 #include "atom_vec.h"
 #include "comm.h"
 #include "force.h"
 #include "neighbor.h"
 #include "neigh_list.h"
 #include "integrate.h"
 #include "memory.h"
 #include "error.h"
 #include "neigh_request.h"
 #include "universe.h"
 #include "update.h"
 #include "domain.h"
 #include "string.h"
 #include "kspace.h"
 #include "gpu_extra.h"
 
 #define MIN(a,b) ((a) < (b) ? (a) : (b))
 #define MAX(a,b) ((a) > (b) ? (a) : (b))
 
 #define EWALD_F   1.12837917
 #define EWALD_P   0.3275911
 #define A1        0.254829592
 #define A2       -0.284496736
 #define A3        1.421413741
 #define A4       -1.453152027
 #define A5        1.061405429
 
 // External functions from cuda library for atom decomposition
 
 int cmml_gpu_init(const int ntypes, double **cutsq, int **cg_type,
 		  double **host_lj1, double **host_lj2, double **host_lj3,
 		  double **host_lj4, double **offset, double *special_lj,
 		  const int nlocal, const int nall, const int max_nbors,
 		  const int maxspecial, const double cell_size, int &gpu_mode,
 		  FILE *screen, double **host_cut_ljsq, double host_cut_coulsq,
 		  double *host_special_coul, const double qqrd2e,
 		  const double g_ewald);
 void cmml_gpu_clear();
 int ** cmml_gpu_compute_n(const int ago, const int inum, const int nall,
 			  double **host_x, int *host_type, double *sublo,
 			  double *subhi, int *tag, int **nspecial,
 			  int **special, const bool eflag, const bool vflag,
 			  const bool eatom, const bool vatom, int &host_start,
 			  int **ilist, int **jnum, const double cpu_time,
 			  bool &success, double *host_q, double *boxlo,
 			  double *prd);
 void cmml_gpu_compute(const int ago, const int inum, const int nall,
 		      double **host_x, int *host_type, int *ilist, int *numj,
 		      int **firstneigh, const bool eflag, const bool vflag,
 		      const bool eatom, const bool vatom, int &host_start,
 		      const double cpu_time, bool &success, double *host_q,
 		      const int nlocal, double *boxlo, double *prd);
 double cmml_gpu_bytes();
 
 using namespace LAMMPS_NS;
 
 /* ---------------------------------------------------------------------- */
 
 PairCGCMMCoulLongGPU::PairCGCMMCoulLongGPU(LAMMPS *lmp) : PairCGCMMCoulLong(lmp), gpu_mode(GPU_PAIR)
 {
   respa_enable = 0;
   cpu_time = 0.0;
 }
 
 /* ----------------------------------------------------------------------
    free all arrays
 ------------------------------------------------------------------------- */
 
 PairCGCMMCoulLongGPU::~PairCGCMMCoulLongGPU()
 {
   cmml_gpu_clear();
 }
 
 /* ---------------------------------------------------------------------- */
 
 void PairCGCMMCoulLongGPU::compute(int eflag, int vflag)
 {
   if (eflag || vflag) ev_setup(eflag,vflag);
   else evflag = vflag_fdotr = 0;
   
   int nall = atom->nlocal + atom->nghost;
   int inum, host_start;
   
   bool success = true;
   int *ilist, *numneigh, **firstneigh;    
   if (gpu_mode == GPU_NEIGH) {
     inum = atom->nlocal;
     firstneigh = cmml_gpu_compute_n(neighbor->ago, inum, nall, atom->x,
 				    atom->type, domain->sublo, domain->subhi,
 				    atom->tag, atom->nspecial, atom->special,
 				    eflag, vflag, eflag_atom, vflag_atom,
 				    host_start, &ilist, &numneigh, cpu_time,
 				    success, atom->q, domain->boxlo,
 				    domain->prd);
   } else {
     inum = list->inum;
     ilist = list->ilist;
     numneigh = list->numneigh;
     firstneigh = list->firstneigh;
     cmml_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
 		     ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
 		     vflag_atom, host_start, cpu_time, success, atom->q,
 		     atom->nlocal, domain->boxlo, domain->prd);
   }
   if (!success)
     error->one("Out of memory on GPGPU");
 
   if (host_start<inum) {
     cpu_time = MPI_Wtime();
     cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
     cpu_time = MPI_Wtime() - cpu_time;
   }
 }
 
 /* ----------------------------------------------------------------------
    init specific to this pair style
 ------------------------------------------------------------------------- */
 
 void PairCGCMMCoulLongGPU::init_style()
 {
   cut_respa = NULL;
 
   if (!atom->q_flag)
-    error->all("Pair style cg/cmm/coul/long requires atom attribute q");
+    error->all("Pair style cg/cmm/coul/long/gpu requires atom attribute q");
   if (force->newton_pair) 
-    error->all("Cannot use newton pair with GPU cg/cmm pair style");
+    error->all("Cannot use newton pair with cg/cmm/coul/long/gpu pair style");
 
   // Repeat cutsq calculation because done after call to init_style
   double maxcut = -1.0;
   double cut;
   for (int i = 1; i <= atom->ntypes; i++) {
     for (int j = i; j <= atom->ntypes; j++) {
       if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
         cut = init_one(i,j);
         cut *= cut;
         if (cut > maxcut)
           maxcut = cut;
         cutsq[i][j] = cutsq[j][i] = cut;
       } else
         cutsq[i][j] = cutsq[j][i] = 0.0;
     }
   }
   double cell_size = sqrt(maxcut) + neighbor->skin;
 
   // insure use of KSpace long-range solver, set g_ewald
 
   if (force->kspace == NULL)
     error->all("Pair style is incompatible with KSpace style");
   g_ewald = force->kspace->g_ewald;
 
   // setup force tables
 
   if (ncoultablebits) init_tables();
 
   int maxspecial=0;
   if (atom->molecular)
     maxspecial=atom->maxspecial;
   int success = cmml_gpu_init(atom->ntypes+1, cutsq, cg_type, lj1, lj2, lj3,
 			      lj4, offset, force->special_lj, atom->nlocal,
 			      atom->nlocal+atom->nghost, 300, maxspecial,
 			      cell_size, gpu_mode, screen, cut_ljsq,
 			      cut_coulsq_global, force->special_coul,
 			      force->qqrd2e, g_ewald);
   GPU_EXTRA::check_flag(success,error,world);
 
   if (gpu_mode != GPU_NEIGH) {
     int irequest = neighbor->request(this);
     neighbor->requests[irequest]->half = 0;
     neighbor->requests[irequest]->full = 1;
   }
 }
 
 /* ---------------------------------------------------------------------- */
 
 double PairCGCMMCoulLongGPU::memory_usage()
 {
   double bytes = Pair::memory_usage();
   return bytes + cmml_gpu_bytes();
 }
 
 /* ---------------------------------------------------------------------- */
 
 void PairCGCMMCoulLongGPU::cpu_compute(int start, int inum, int eflag,
 				       int vflag, int *ilist, int *numneigh,
 				       int **firstneigh)
 {
   int i,j,ii,jj,jnum,itype,jtype,itable;
   double qtmp,xtmp,ytmp,ztmp,delx,dely,delz;
   double fraction,table;
   double r,r2inv,r6inv,forcecoul,forcelj,factor_coul,factor_lj;
   double grij,expm2,prefactor,t,erfc;
   int *jlist;
   double rsq;
 
   double **x = atom->x;
   double **f = atom->f;
   double *q = atom->q;
   int *type = atom->type;
   int nlocal = atom->nlocal;
   int nall = nlocal + atom->nghost;
   double *special_coul = force->special_coul;
   double *special_lj = force->special_lj;
   double qqrd2e = force->qqrd2e;
 
   // loop over neighbors of my atoms
 
   for (ii = start; ii < inum; ii++) {
     i = ilist[ii];
     qtmp = q[i];
     xtmp = x[i][0];
     ytmp = x[i][1];
     ztmp = x[i][2];
     itype = type[i];
     jlist = firstneigh[i];
     jnum = numneigh[i];
 
     for (jj = 0; jj < jnum; jj++) {
       j = jlist[jj];
       factor_lj = special_lj[sbmask(j)];
       factor_coul = special_coul[sbmask(j)];
       j &= NEIGHMASK;
 
       const double delx = xtmp - x[j][0];
       const double dely = ytmp - x[j][1];
       const double delz = ztmp - x[j][2];
       const double rsq = delx*delx + dely*dely + delz*delz;
       const int jtype = type[j];
 
       double evdwl = 0.0;
       double ecoul = 0.0;
       double fpair = 0.0;
 
       if (rsq < cutsq[itype][jtype]) {
         const double r2inv = 1.0/rsq;
         const int cgt=cg_type[itype][jtype];
 
         double forcelj  = 0.0;
         double forcecoul = 0.0;
 
         if (rsq < cut_ljsq[itype][jtype]) {
           forcelj=factor_lj;
           if (eflag) evdwl=factor_lj;
 
           if (cgt == CG_LJ12_4) {
             const double r4inv=r2inv*r2inv;
             forcelj *= r4inv*(lj1[itype][jtype]*r4inv*r4inv
                        - lj2[itype][jtype]);
             if (eflag) {
               evdwl *= r4inv*(lj3[itype][jtype]*r4inv*r4inv
                        - lj4[itype][jtype]) - offset[itype][jtype];
             }
           } else if (cgt == CG_LJ9_6) {
             const double r3inv = r2inv*sqrt(r2inv);
             const double r6inv = r3inv*r3inv;
             forcelj *= r6inv*(lj1[itype][jtype]*r3inv
                        - lj2[itype][jtype]);
             if (eflag) {
               evdwl *= r6inv*(lj3[itype][jtype]*r3inv
                         - lj4[itype][jtype]) - offset[itype][jtype];
             }
           } else {
             const double r6inv = r2inv*r2inv*r2inv;
             forcelj *= r6inv*(lj1[itype][jtype]*r6inv
                        - lj2[itype][jtype]);
             if (eflag) {
               evdwl *= r6inv*(lj3[itype][jtype]*r6inv
                        - lj4[itype][jtype]) - offset[itype][jtype];
             }
           }
         }
 
         if (rsq < cut_coulsq_global) {
           if (!ncoultablebits || rsq <= tabinnersq) {
             const double r = sqrt(rsq);
             const double grij = g_ewald * r;
             const double expm2 = exp(-grij*grij);
             const double t = 1.0 / (1.0 + EWALD_P*grij);
             const double erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;
             const double prefactor = qqrd2e * qtmp*q[j]/r;
             forcecoul = prefactor * (erfc + EWALD_F*grij*expm2);
             if (eflag) ecoul = prefactor*erfc;
             if (factor_coul < 1.0) {
               forcecoul -= (1.0-factor_coul)*prefactor;
               if (eflag) ecoul -= (1.0-factor_coul)*prefactor;
             }
           } else {
             union_int_float_t rsq_lookup;
             rsq_lookup.f = rsq;
             int itable = rsq_lookup.i & ncoulmask;
             itable >>= ncoulshiftbits;
             const double fraction = (rsq_lookup.f - rtable[itable]) *
                                      drtable[itable];
             const double table = ftable[itable] + fraction*dftable[itable];
             forcecoul = qtmp*q[j] * table;
             if (eflag) {
               const double table2 = etable[itable] + fraction*detable[itable];
               ecoul = qtmp*q[j] * table2;
             }
             if (factor_coul < 1.0) {
               const double table2 = ctable[itable] + fraction*dctable[itable];
               const double prefactor = qtmp*q[j] * table2;
               forcecoul -= (1.0-factor_coul)*prefactor;
               if (eflag) ecoul -= (1.0-factor_coul)*prefactor;
             }
           }
         }
         fpair = (forcecoul + forcelj) * r2inv;
 
 	f[i][0] += delx*fpair;
 	f[i][1] += dely*fpair;
 	f[i][2] += delz*fpair;
 
 	if (evflag) ev_tally_full(i,evdwl,ecoul,fpair,delx,dely,delz);
       }
     }
   }
 }
diff --git a/src/GPU/pair_cg_cmm_coul_msm_gpu.cpp b/src/GPU/pair_cg_cmm_coul_msm_gpu.cpp
index e88144bea..d946b739d 100644
--- a/src/GPU/pair_cg_cmm_coul_msm_gpu.cpp
+++ b/src/GPU/pair_cg_cmm_coul_msm_gpu.cpp
@@ -1,309 +1,309 @@
 /* ----------------------------------------------------------------------
    LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
    http://lammps.sandia.gov, Sandia National Laboratories
    Steve Plimpton, sjplimp@sandia.gov
    
    Copyright (2003) Sandia Corporation.  Under the terms of Contract
    DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
    certain rights in this software.  This software is distributed under 
    the GNU General Public License.
    
    See the README file in the top-level LAMMPS directory.
 ------------------------------------------------------------------------- */
 
 /* ----------------------------------------------------------------------
    Contributing author: Mike Brown (SNL)
 ------------------------------------------------------------------------- */
 
 #include "math.h"
 #include "stdio.h"
 #include "stdlib.h"
 #include "pair_cg_cmm_coul_msm_gpu.h"
 #include "atom.h"
 #include "atom_vec.h"
 #include "comm.h"
 #include "force.h"
 #include "neighbor.h"
 #include "neigh_list.h"
 #include "integrate.h"
 #include "memory.h"
 #include "error.h"
 #include "neigh_request.h"
 #include "universe.h"
 #include "update.h"
 #include "domain.h"
 #include "string.h"
 #include "kspace.h"
 #include "gpu_extra.h"
 
 #define MIN(a,b) ((a) < (b) ? (a) : (b))
 #define MAX(a,b) ((a) > (b) ? (a) : (b))
 enum {C3=0,C4=1};
 
 // External functions from cuda library for atom decomposition
 
 int cmmm_gpu_init(const int ntypes, double **cutsq, int **cg_type,
 		  double **host_lj1, double **host_lj2, double **host_lj3,
 		  double **host_lj4, double **offset, double *special_lj,
 		  const int nlocal, const int nall, const int max_nbors,
 		  const int maxspecial, const double cell_size, int &gpu_mode,
 		  FILE *screen, double **host_cut_ljsq, double host_cut_coulsq,
 		  double *host_special_coul, const double qqrd2e,
 		  const int smooth);
 void cmmm_gpu_clear();
 int ** cmmm_gpu_compute_n(const int ago, const int inum, const int nall,
 			  double **host_x, int *host_type, double *sublo,
 			  double *subhi, int *tag, int **nspecial,
 			  int **special, const bool eflag, const bool vflag,
 			  const bool eatom, const bool vatom, int &host_start,
 			  int **ilist, int **jnum, const double cpu_time,
 			  bool &success, double *host_q, double *boxlo,
 			  double *prd);
 void cmmm_gpu_compute(const int ago, const int inum, const int nall,
 		      double **host_x, int *host_type, int *ilist, int *numj,
 		      int **firstneigh, const bool eflag, const bool vflag,
 		      const bool eatom, const bool vatom, int &host_start,
 		      const double cpu_time, bool &success, double *host_q,
 		      const int nlocal, double *boxlo, double *prd);
 double cmmm_gpu_bytes();
 
 using namespace LAMMPS_NS;
 
 /* ---------------------------------------------------------------------- */
 
 PairCGCMMCoulMSMGPU::PairCGCMMCoulMSMGPU(LAMMPS *lmp) : PairCGCMMCoulMSM(lmp),
                                                         gpu_mode(GPU_PAIR)
 {
   respa_enable = 0;
   cpu_time = 0.0;
 }
 
 /* ----------------------------------------------------------------------
    free all arrays
 ------------------------------------------------------------------------- */
 
 PairCGCMMCoulMSMGPU::~PairCGCMMCoulMSMGPU()
 {
   cmmm_gpu_clear();
 }
 
 /* ---------------------------------------------------------------------- */
 
 void PairCGCMMCoulMSMGPU::compute(int eflag, int vflag)
 {
   if (eflag || vflag) ev_setup(eflag,vflag);
   else evflag = vflag_fdotr = 0;
   
   int nall = atom->nlocal + atom->nghost;
   int inum, host_start;
   
   bool success = true;
   int *ilist, *numneigh, **firstneigh;
   if (gpu_mode == GPU_NEIGH) {
     inum = atom->nlocal;
     firstneigh = cmmm_gpu_compute_n(neighbor->ago, inum, nall, atom->x,
 				    atom->type, domain->sublo, domain->subhi,
 				    atom->tag, atom->nspecial, atom->special,
 				    eflag, vflag, eflag_atom, vflag_atom,
 				    host_start, &ilist, &numneigh, cpu_time,
 				    success, atom->q, domain->boxlo,
 				    domain->prd);
   } else {
     inum = list->inum;
     ilist = list->ilist;
     numneigh = list->numneigh;
     firstneigh = list->firstneigh;
     cmmm_gpu_compute(neighbor->ago, inum, nall, atom->x, atom->type,
 		     ilist, numneigh, firstneigh, eflag, vflag, eflag_atom,
 		     vflag_atom, host_start, cpu_time, success, atom->q,
 		     atom->nlocal, domain->boxlo, domain->prd);
   }
   if (!success)
     error->one("Out of memory on GPGPU");
 
   if (host_start<inum) {
     cpu_time = MPI_Wtime();
     cpu_compute(host_start, inum, eflag, vflag, ilist, numneigh, firstneigh);
     cpu_time = MPI_Wtime() - cpu_time;
   }
 }
 
 /* ----------------------------------------------------------------------
    init specific to this pair style
 ------------------------------------------------------------------------- */
 
 void PairCGCMMCoulMSMGPU::init_style()
 {
   PairCGCMMCoulMSM::init_style();
   if (force->newton_pair) 
-    error->all("Cannot use newton pair with GPU cg/cmm pair style");
+    error->all("Cannot use newton pair with cg/cmm/coul/msm/gpu pair style");
 
   // Repeat cutsq calculation because done after call to init_style
   double maxcut = -1.0;
   double cut;
   for (int i = 1; i <= atom->ntypes; i++) {
     for (int j = i; j <= atom->ntypes; j++) {
       if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
         cut = init_one(i,j);
         cut *= cut;
         if (cut > maxcut)
           maxcut = cut;
         cutsq[i][j] = cutsq[j][i] = cut;
       } else
         cutsq[i][j] = cutsq[j][i] = 0.0;
     }
   }
   double cell_size = sqrt(maxcut) + neighbor->skin;
 
   int maxspecial=0;
   if (atom->molecular)
     maxspecial=atom->maxspecial;
   int success = cmmm_gpu_init(atom->ntypes+1, cutsq, cg_type, lj1, lj2, lj3,
 			      lj4, offset, force->special_lj, atom->nlocal,
 			      atom->nlocal+atom->nghost, 300, maxspecial,
 			      cell_size, gpu_mode, screen, cut_ljsq,
 			      cut_coulsq_global, force->special_coul,
 			      force->qqrd2e,_smooth);
   GPU_EXTRA::check_flag(success,error,world);
 
   if (gpu_mode != GPU_NEIGH) {
     int irequest = neighbor->request(this);
     neighbor->requests[irequest]->half = 0;
     neighbor->requests[irequest]->full = 1;
   }
 }
 
 /* ---------------------------------------------------------------------- */
 
 double PairCGCMMCoulMSMGPU::memory_usage()
 {
   double bytes = Pair::memory_usage();
   return bytes + cmmm_gpu_bytes();
 }
 
 /* ---------------------------------------------------------------------- */
 
 void PairCGCMMCoulMSMGPU::cpu_compute(int start, int inum, int eflag,
 				      int vflag, int *ilist, int *numneigh,
 				      int **firstneigh)
 {
   int i,j,ii,jj,jnum,itype,jtype,itable;
   double qtmp,xtmp,ytmp,ztmp,delx,dely,delz;
   double fraction,table;
   double r,r2inv,r6inv,forcecoul,forcelj,factor_coul,factor_lj;
   double grij,expm2,prefactor,t,erfc;
   int *jlist;
   double rsq;
 
   double **x = atom->x;
   double **f = atom->f;
   double *q = atom->q;
   int *type = atom->type;
   int nlocal = atom->nlocal;
   int nall = nlocal + atom->nghost;
   double *special_coul = force->special_coul;
   double *special_lj = force->special_lj;
   double qqrd2e = force->qqrd2e;
 
   // loop over neighbors of my atoms
 
   for (ii = start; ii < inum; ii++) {
     i = ilist[ii];
     qtmp = q[i];
     xtmp = x[i][0];
     ytmp = x[i][1];
     ztmp = x[i][2];
     itype = type[i];
     jlist = firstneigh[i];
     jnum = numneigh[i];
 
     for (jj = 0; jj < jnum; jj++) {
       j = jlist[jj];
       factor_lj = special_lj[sbmask(j)];
       factor_coul = special_coul[sbmask(j)];
       j &= NEIGHMASK;
 
       const double delx = xtmp - x[j][0];
       const double dely = ytmp - x[j][1];
       const double delz = ztmp - x[j][2];
       const double rsq = delx*delx + dely*dely + delz*delz;
       const int jtype = type[j];
 
       double evdwl = 0.0;
       double ecoul = 0.0;
       double fpair = 0.0;
 
       if (rsq < cutsq[itype][jtype]) {
         const double r2inv = 1.0/rsq;
         const int cgt=cg_type[itype][jtype];
 
         double forcelj  = 0.0;
         double forcecoul = 0.0;
 
         if (rsq < cut_ljsq[itype][jtype]) {
           forcelj=factor_lj;
           if (eflag) evdwl=factor_lj;
 
           if (cgt == CG_LJ12_4) {
             const double r4inv=r2inv*r2inv;
             forcelj *= r4inv*(lj1[itype][jtype]*r4inv*r4inv
                        - lj2[itype][jtype]);
             if (eflag) {
               evdwl *= r4inv*(lj3[itype][jtype]*r4inv*r4inv
                        - lj4[itype][jtype]) - offset[itype][jtype];
             }
           } else if (cgt == CG_LJ9_6) {
             const double r3inv = r2inv*sqrt(r2inv);
             const double r6inv = r3inv*r3inv;
             forcelj *= r6inv*(lj1[itype][jtype]*r3inv
                        - lj2[itype][jtype]);
             if (eflag) {
               evdwl *= r6inv*(lj3[itype][jtype]*r3inv
                         - lj4[itype][jtype]) - offset[itype][jtype];
             }
           } else {
             const double r6inv = r2inv*r2inv*r2inv;
             forcelj *= r6inv*(lj1[itype][jtype]*r6inv
                        - lj2[itype][jtype]);
             if (eflag) {
               evdwl *= r6inv*(lj3[itype][jtype]*r6inv
                        - lj4[itype][jtype]) - offset[itype][jtype];
             }
           }
         }
 
         if (rsq < cut_coulsq_global) {
           const double ir = 1.0/sqrt(rsq);
           const double prefactor = qqrd2e * qtmp*q[j];
           const double r2_ia2 = rsq*_ia2;
           const double r4_ia4 = r2_ia2*r2_ia2;
           if (_smooth==C3) {
             forcecoul = prefactor*(_ia3*(-4.375+5.25*r2_ia2-1.875*r4_ia4)-
                                    ir/rsq);
             if (eflag)
               ecoul = prefactor*(ir+_ia*(2.1875-2.1875*r2_ia2+
                                          1.3125*r4_ia4-0.3125*r2_ia2*r4_ia4));
           } else {
             const double r6_ia6 = r2_ia2*r4_ia4;
             forcecoul = prefactor*(_ia3*(-6.5625+11.8125*r2_ia2-8.4375*r4_ia4+
                                          2.1875*r6_ia6)-ir/rsq);
             if (eflag)
               ecoul = prefactor*(ir+_ia*(2.4609375-3.28125*r2_ia2+
                                          2.953125*r4_ia4-1.40625*r6_ia6+
                                          0.2734375*r4_ia4*r4_ia4));
           }
           if (factor_coul < 1.0) {
             forcecoul -= (1.0-factor_coul)*prefactor*ir;
             if (eflag) ecoul -= (1.0-factor_coul)*prefactor*ir;
           }
         }
         fpair = forcecoul + forcelj * r2inv;
 
 	f[i][0] += delx*fpair;
 	f[i][1] += dely*fpair;
 	f[i][2] += delz*fpair;
 	if (evflag) ev_tally_full(i,evdwl,ecoul,fpair,delx,dely,delz);
       }
     }
   }
 }