git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@12147 f3b2605a-c512-4ea7-a41b-209d697bcdaa

src/GPU/Install.sh

@@ -43,6 +43,10 @@ action pair_buck_gpu.cpp pair_buck.cpp
 action pair_buck_gpu.h pair_buck.cpp
 action pair_colloid_gpu.cpp pair_colloid.cpp
 action pair_colloid_gpu.h pair_colloid.cpp
+action pair_coul_cut_gpu.cpp
+action pair_coul_cut_gpu.h
+action pair_coul_debye_gpu.cpp
+action pair_coul_debye_gpu.h
 action pair_coul_dsf_gpu.cpp
 action pair_coul_dsf_gpu.h
 action pair_coul_long_gpu.cpp pair_coul_long.cpp

src/GPU/pair_born_gpu.cpp

@@ -47,6 +47,10 @@ int born_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                   const int inum, const int nall, const int max_nbors,
                   const int maxspecial, const double cell_size, 
                   int &gpu_mode, FILE *screen);
+void born_gpu_reinit(const int ntypes, double **host_rhoinv,
+                     double **host_born1, double **host_born2, double **host_born3,
+                     double **host_a, double **host_c, double **host_d,
+                     double **offset);
 void born_gpu_clear();
 int ** born_gpu_compute_n(const int ago, const int inum_full, 
                           const int nall, double **host_x, int *host_type, 
@@ -67,7 +71,6 @@ double born_gpu_bytes();
 PairBornGPU::PairBornGPU(LAMMPS *lmp) : PairBorn(lmp), gpu_mode(GPU_FORCE)
 {
   respa_enable = 0;
-  reinitflag = 0;
   cpu_time = 0.0;
   GPU_EXTRA::gpu_ready(lmp->modify, lmp->error); 
 }
@@ -165,6 +168,16 @@ void PairBornGPU::init_style()
 
 /* ---------------------------------------------------------------------- */
 
+void PairBornGPU::reinit()
+{
+  Pair::reinit();
+  
+  born_gpu_reinit(atom->ntypes+1, rhoinv, born1, born2, born3,
+                  a, c, d, offset);
+}
+
+/* ---------------------------------------------------------------------- */
+
 double PairBornGPU::memory_usage()
 {
   double bytes = Pair::memory_usage();

src/GPU/pair_born_gpu.h

@@ -31,6 +31,7 @@ class PairBornGPU : public PairBorn {
   void cpu_compute(int, int, int, int, int *, int *, int **);
   void compute(int, int);
   void init_style();
+  void reinit();
   double memory_usage();
 
  enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };

src/GPU/pair_buck_gpu.cpp

@@ -45,6 +45,9 @@ int buck_gpu_init(const int ntypes, double **cutsq, double **host_rhoinv,
                   double **offset, double *special_lj, const int inum,
                   const int nall, const int max_nbors,  const int maxspecial,
                   const double cell_size, int &gpu_mode, FILE *screen);
+void buck_gpu_reinit(const int ntypes, double **cutsq, double **host_rhoinv,
+                  double **host_buck1, double **host_buck2,
+                  double **host_a, double **host_c, double **offset);
 void buck_gpu_clear();
 int ** buck_gpu_compute_n(const int ago, const int inum_full, const int nall,
                           double **host_x, int *host_type, double *sublo,
@@ -65,7 +68,6 @@ double buck_gpu_bytes();
 PairBuckGPU::PairBuckGPU(LAMMPS *lmp) : PairBuck(lmp), gpu_mode(GPU_FORCE)
 {
   respa_enable = 0;
-  reinitflag = 0;
   cpu_time = 0.0;
   GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
 }
@@ -162,6 +164,16 @@ void PairBuckGPU::init_style()
 
 /* ---------------------------------------------------------------------- */
 
+void PairBuckGPU::reinit()
+{
+  Pair::reinit();
+  
+  buck_gpu_reinit(atom->ntypes+1, cutsq, rhoinv, buck1, buck2,
+                  a, c, offset);
+}
+
+/* ---------------------------------------------------------------------- */
+
 double PairBuckGPU::memory_usage()
 {
   double bytes = Pair::memory_usage();

src/GPU/pair_buck_gpu.h

@@ -31,6 +31,7 @@ class PairBuckGPU : public PairBuck {
   void cpu_compute(int, int, int, int, int *, int *, int **);
   void compute(int, int);
   void init_style();
+  void reinit();
   double memory_usage();
 
  enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };

src/GPU/pair_coul_long_gpu.cpp

@@ -48,10 +48,12 @@ using namespace LAMMPS_NS;
 
 // External functions from cuda library for atom decomposition
 
-int cl_gpu_init(const int nlocal, const int nall, const int max_nbors,
+int cl_gpu_init(const int ntypes, double **scale,
+                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_coulsq, double *host_special_coul,
                 const double qqrd2e, const double g_ewald);
+void cl_gpu_reinit(const int ntypes, double **scale);
 void cl_gpu_clear();
 int ** cl_gpu_compute_n(const int ago, const int inum,
                         const int nall, double **host_x, int *host_type,
@@ -75,7 +77,6 @@ PairCoulLongGPU::PairCoulLongGPU(LAMMPS *lmp) :
   PairCoulLong(lmp), gpu_mode(GPU_FORCE)
 {
   respa_enable = 0;
-  reinitflag = 0;
   cpu_time = 0.0;
   GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
 }
@@ -144,7 +145,7 @@ void PairCoulLongGPU::init_style()
     error->all(FLERR,"Cannot use newton pair with coul/long/gpu pair style");
 
   // Repeat cutsq calculation because done after call to init_style
-  double cell_size = sqrt(cut_coul) + neighbor->skin;
+  double cell_size = cut_coul + neighbor->skin;
 
   cut_coulsq = cut_coul * cut_coul;
 
@@ -161,7 +162,8 @@ void PairCoulLongGPU::init_style()
   int maxspecial=0;
   if (atom->molecular)
     maxspecial=atom->maxspecial;
-  int success = cl_gpu_init(atom->nlocal, atom->nlocal+atom->nghost, 300,
+  int success = cl_gpu_init(atom->ntypes+1, scale,
+                            atom->nlocal, atom->nlocal+atom->nghost, 300,
                             maxspecial, cell_size, gpu_mode, screen, cut_coulsq,
                             force->special_coul, force->qqrd2e, g_ewald);
 
@@ -176,6 +178,15 @@ void PairCoulLongGPU::init_style()
 
 /* ---------------------------------------------------------------------- */
 
+void PairCoulLongGPU::reinit()
+{
+  Pair::reinit();
+  
+  cl_gpu_reinit(atom->ntypes+1, scale);
+}
+
+/* ---------------------------------------------------------------------- */
+
 double PairCoulLongGPU::memory_usage()
 {
   double bytes = Pair::memory_usage();

src/GPU/pair_coul_long_gpu.h

@@ -31,6 +31,7 @@ class PairCoulLongGPU : public PairCoulLong {
   void cpu_compute(int, int, int, int, int *, int *, int **);
   void compute(int, int);
   void init_style();
+  void reinit();
   double memory_usage();
 
  enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };

src/GPU/pair_gauss_gpu.cpp

@@ -44,6 +44,8 @@ int gauss_gpu_init(const int ntypes, double **cutsq, double **host_a,
                    double **b, 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);
+int gauss_gpu_reinit(const int ntypes, double **cutsq, double **host_a,
+                   double **b, double **offset);
 void gauss_gpu_clear();
 int ** gauss_gpu_compute_n(const int ago, const int inum,
                            const int nall, double **host_x, int *host_type, 
@@ -64,7 +66,6 @@ double gauss_gpu_bytes();
 PairGaussGPU::PairGaussGPU(LAMMPS *lmp) : PairGauss(lmp), gpu_mode(GPU_FORCE)
 {
   respa_enable = 0;
-  reinitflag = 0;
   cpu_time = 0.0;
   GPU_EXTRA::gpu_ready(lmp->modify, lmp->error); 
 }
@@ -161,6 +162,15 @@ void PairGaussGPU::init_style()
 
 /* ---------------------------------------------------------------------- */
 
+void PairGaussGPU::reinit()
+{
+  Pair::reinit();
+  
+  gauss_gpu_reinit(atom->ntypes+1, cutsq, a, b, offset);
+}
+
+/* ---------------------------------------------------------------------- */
+
 double PairGaussGPU::memory_usage()
 {
   double bytes = Pair::memory_usage();

src/GPU/pair_gauss_gpu.h

@@ -31,6 +31,7 @@ class PairGaussGPU : public PairGauss {
   void cpu_compute(int, int, int, int, int *, int *, int **);
   void compute(int, int);
   void init_style();
+  void reinit();
   double memory_usage();
 
  enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };

src/GPU/pair_lj_cut_gpu.cpp

@@ -44,6 +44,11 @@ int ljl_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                  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);
+
+int ljl_gpu_reinit(const int ntypes, double **cutsq, double **host_lj1,
+                   double **host_lj2, double **host_lj3, double **host_lj4,
+                   double **offset);
+
 void ljl_gpu_clear();
 int ** ljl_gpu_compute_n(const int ago, const int inum,
                          const int nall, double **host_x, int *host_type,
@@ -64,7 +69,6 @@ double ljl_gpu_bytes();
 PairLJCutGPU::PairLJCutGPU(LAMMPS *lmp) : PairLJCut(lmp), gpu_mode(GPU_FORCE)
 {
   respa_enable = 0;
-  reinitflag = 0;
   cpu_time = 0.0;
   GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
 }
@@ -87,7 +91,7 @@ void PairLJCutGPU::compute(int eflag, int vflag)
 
   int nall = atom->nlocal + atom->nghost;
   int inum, host_start;
-
+  
   bool success = true;
   int *ilist, *numneigh, **firstneigh;
   if (gpu_mode != GPU_FORCE) {
@@ -163,6 +167,15 @@ void PairLJCutGPU::init_style()
 
 /* ---------------------------------------------------------------------- */
 
+void PairLJCutGPU::reinit()
+{
+  Pair::reinit();
+  
+  ljl_gpu_reinit(atom->ntypes+1, cutsq, lj1, lj2, lj3, lj4, offset);
+}
+
+/* ---------------------------------------------------------------------- */
+
 double PairLJCutGPU::memory_usage()
 {
   double bytes = Pair::memory_usage();

src/GPU/pair_lj_cut_gpu.h

@@ -31,6 +31,7 @@ class PairLJCutGPU : public PairLJCut {
   void cpu_compute(int, int, int, int, int *, int *, int **);
   void compute(int, int);
   void init_style();
+  void reinit();
   double memory_usage();
 
  enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };

src/GPU/pair_lj_expand_gpu.cpp

@@ -45,6 +45,9 @@ int lje_gpu_init(const int ntypes, double **cutsq, double **host_lj1,
                  const int nlocal, const int nall, const int max_nbors,
                  const int maxspecial, const double cell_size, int &gpu_mode,
                  FILE *screen);
+int lje_gpu_reinit(const int ntypes, double **cutsq, double **host_lj1,
+                   double **host_lj2, double **host_lj3, double **host_lj4,
+                   double **offset, double **shift);
 void lje_gpu_clear();
 int ** lje_gpu_compute_n(const int ago, const int inum, const int nall,
                          double **host_x, int *host_type, double *sublo,
@@ -65,7 +68,6 @@ double lje_gpu_bytes();
 PairLJExpandGPU::PairLJExpandGPU(LAMMPS *lmp) : PairLJExpand(lmp), gpu_mode(GPU_FORCE)
 {
   respa_enable = 0;
-  reinitflag = 0;
   cpu_time = 0.0;
   GPU_EXTRA::gpu_ready(lmp->modify, lmp->error);
 }
@@ -162,6 +164,15 @@ void PairLJExpandGPU::init_style()
 
 /* ---------------------------------------------------------------------- */
 
+void PairLJExpandGPU::reinit()
+{
+  Pair::reinit();
+  
+  lje_gpu_reinit(atom->ntypes+1, cutsq, lj1, lj2, lj3, lj4, offset, shift);
+}
+
+/* ---------------------------------------------------------------------- */
+
 double PairLJExpandGPU::memory_usage()
 {
   double bytes = Pair::memory_usage();

src/GPU/pair_lj_expand_gpu.h

@@ -31,6 +31,7 @@ class PairLJExpandGPU : public PairLJExpand {
   void cpu_compute(int, int, int, int, int *, int *, int **);
   void compute(int, int);
   void init_style();
+  void reinit();
   double memory_usage();
 
  enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };

src/GPU/pair_soft_gpu.cpp

@@ -45,6 +45,8 @@ int soft_gpu_init(const int ntypes, double **cutsq, double **prefactor,
                    double **cut, 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);
+void soft_gpu_reinit(const int ntypes, double **cutsq, double **host_prefactor,
+                     double **host_cut);
 void soft_gpu_clear();
 int ** soft_gpu_compute_n(const int ago, const int inum,
                            const int nall, double **host_x, int *host_type, 
@@ -68,7 +70,6 @@ using namespace MathConst;
 PairSoftGPU::PairSoftGPU(LAMMPS *lmp) : PairSoft(lmp), gpu_mode(GPU_FORCE)
 {
   respa_enable = 0;
-  reinitflag = 0;
   cpu_time = 0.0;
   GPU_EXTRA::gpu_ready(lmp->modify, lmp->error); 
 }
@@ -165,6 +166,15 @@ void PairSoftGPU::init_style()
 
 /* ---------------------------------------------------------------------- */
 
+void PairSoftGPU::reinit()
+{
+  Pair::reinit();
+  
+  soft_gpu_reinit(atom->ntypes+1, cutsq, prefactor, cut);
+}
+
+/* ---------------------------------------------------------------------- */
+
 double PairSoftGPU::memory_usage()
 {
   double bytes = Pair::memory_usage();

src/GPU/pair_soft_gpu.h

@@ -31,6 +31,7 @@ class PairSoftGPU : public PairSoft {
   void cpu_compute(int, int, int, int, int *, int *, int **);
   void compute(int, int);
   void init_style();
+  void reinit();
   double memory_usage();
 
  enum { GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH };

src/comm_tiled.cpp

@@ -84,12 +84,39 @@ void CommTiled::init()
 void CommTiled::setup()
 {
   // error on triclinic or multi?
-  // set nswap
+  // set nswap = 2*dim
   // setup neighbor proc info for exchange()
   // setup nsendproc and nrecvproc bounts
   // setup sendproc and recvproc lists
-  // setup sendbox
+  // setup sendboxes
   // reallocate requests and statuses
+
+  // check that cutoff is <= 1/2 of periodic box len?
+
+  // loop over dims
+  //  left:
+  //    construct ghost boxes
+  //      differnet in x,y,z
+  //      account for ghost borders in y,z
+  //      account for PBC by shifting
+  //      split into multiple boxes if straddles PBC
+  //    drop boxes down RCB tree
+  //      count unique procs they cover
+  //      what about self if crosses PBC
+  //      for each proc they cover:
+  //        compute box I send it to left
+  //        is a message I will recv from right (don't care about box)
+  //    for ghost-extended boxes
+  //      do not count procs that do not overlap my owned box at all
+  //      only touching edge of my owned box does not count
+  //      in this case list I send to and recv from may be different?
+  //  same thing to right
+
+  // what need from decomp (RCB):
+  // dropbox: return list of procs with overlap and overlapping boxes
+  //   return n, proclist, boxlist
+  // otherbox: bbox of another proc
+  // dropatom: return what proc owns the atom coord
 }
 
 /* ----------------------------------------------------------------------
@@ -249,6 +276,14 @@ void CommTiled::reverse_comm()
 
 void CommTiled::exchange()
 {
+  // loop over atoms
+  //   if not outside my box, continue
+  //   find which proc it is in
+  //   find which one of my touching procs it is, else lost
+  //   make sure all atoms are "lost" that should be (e.g. outside non-PBC)
+  //   add to list to send to that proc
+  // loop over touching procs
+  //   send buffer to them
 }
 
 /* ----------------------------------------------------------------------

src/pair_coul_debye.cpp

@@ -85,7 +85,8 @@ void PairCoulDebye::compute(int eflag, int vflag)
         r = sqrt(rsq);
         rinv = 1.0/r;
         screening = exp(-kappa*r);
-        forcecoul = qqrd2e * qtmp*q[j] * screening * (kappa + rinv);
+        forcecoul = qqrd2e * scale[itype][jtype] *
+          qtmp*q[j] * screening * (kappa + rinv);
         fpair = factor_coul*forcecoul * r2inv;
 
         f[i][0] += delx*fpair;
@@ -97,7 +98,8 @@ void PairCoulDebye::compute(int eflag, int vflag)
           f[j][2] -= delz*fpair;
         }
 
-        if (eflag) ecoul = factor_coul * qqrd2e * qtmp*q[j] * rinv * screening;
+        if (eflag) ecoul = factor_coul * qqrd2e *
+            scale[itype][jtype] * qtmp*q[j] * rinv * screening;
 
         if (evflag) ev_tally(i,j,nlocal,newton_pair,
                              0.0,ecoul,fpair,delx,dely,delz);