git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@8689 f3b2605a-c512-4ea7-a41b-209d697bcdaa

src/GPU/Install.sh

@@ -20,7 +20,27 @@ if (test $1 = 1) then
 include ..\/..\/lib\/gpu\/Makefile.lammps
 ' ../Makefile.package.settings
   fi
+  
+  if (test -e ../pair_dipole_sf.cpp) then
+    cp pair_dipole_sf_gpu.cpp ..
+    cp pair_dipole_sf_gpu.h ..
+  fi
 
+  if (test -e ../pair_dipole_cut.cpp) then
+    cp pair_dipole_cut_gpu.cpp ..
+    cp pair_dipole_cut_gpu.h ..
+  fi
+  
+  if (test -e ../pair_yukawa_colloid.cpp) then
+    cp pair_yukawa_colloid_gpu.cpp ..
+    cp pair_yukawa_colloid_gpu.h ..
+  fi
+  
+  if (test -e ../pair_colloid.cpp) then
+    cp pair_colloid_gpu.cpp ..
+    cp pair_colloid_gpu.h ..
+  fi
+  
   if (test -e ../pair_yukawa.cpp) then
     cp pair_yukawa_gpu.cpp ..
     cp pair_yukawa_gpu.h ..
@@ -54,7 +74,7 @@ include ..\/..\/lib\/gpu\/Makefile.lammps
     cp pair_eam_fs_gpu.cpp ..
     cp pair_eam_fs_gpu.h ..
   fi
-  
+
   if (test -e ../pair_gayberne.cpp) then
     cp pair_gayberne_gpu.cpp ..
     cp pair_gayberne_gpu.h ..
@@ -107,7 +127,16 @@ include ..\/..\/lib\/gpu\/Makefile.lammps
   cp pair_lj96_cut_gpu.cpp ..
   cp pair_lj_expand_gpu.cpp ..
   cp pair_lj_cut_coul_cut_gpu.cpp ..
-
+#  cp pair_lj_cut_coul_dsf.cpp ..
+#  cp pair_lj_cut_coul_dsf_gpu.cpp ..
+#  cp pair_coul_dsf.cpp ..
+#  cp pair_coul_dsf_gpu.cpp ..
+  cp pair_lj_cut_coul_debye_gpu.cpp ..
+  cp pair_born_gpu.cpp ..
+  cp pair_born_coul_wolf_gpu.cpp ..
+  cp pair_born_coul_long_gpu.cpp ..
+  cp pair_gauss_gpu.cpp ..
+  
   cp fix_gpu.cpp ..
 
   cp pair_lj_cut_gpu.h ..
@@ -115,6 +144,15 @@ include ..\/..\/lib\/gpu\/Makefile.lammps
   cp pair_lj96_cut_gpu.h ..
   cp pair_lj_expand_gpu.h ..
   cp pair_lj_cut_coul_cut_gpu.h ..
+#  cp pair_lj_cut_coul_dsf.h ..
+#  cp pair_lj_cut_coul_dsf_gpu.h ..
+#  cp pair_coul_dsf.h ..
+#  cp pair_coul_dsf_gpu.h ..
+  cp pair_lj_cut_coul_debye_gpu.h ..
+  cp pair_born_gpu.h ..
+  cp pair_born_coul_wolf_gpu.h ..
+  cp pair_born_coul_long_gpu.h ..
+  cp pair_gauss_gpu.h ..
   
   cp fix_gpu.h ..
   cp gpu_extra.h ..
@@ -151,6 +189,19 @@ elif (test $1 = 0) then
   rm -f ../pair_resquared_gpu.cpp
   rm -f ../pair_table_gpu.cpp
   rm -f ../pair_yukawa_gpu.cpp
+  rm -f ../pair_born_gpu.cpp
+  rm -f ../pair_born_coul_wolf_gpu.cpp
+  rm -f ../pair_born_coul_long_gpu.cpp
+  rm -f ../pair_gauss_gpu.cpp
+  rm -f ../pair_colloid_gpu.cpp
+  rm -f ../pair_yukawa_colloid_gpu.cpp
+  rm -f ../pair_dipole_cut_gpu.cpp
+  rm -f ../pair_dipole_sf_gpu.cpp
+  rm -f ../pair_lj_cut_coul_dsf.cpp
+  rm -f ../pair_lj_cut_coul_dsf_gpu.cpp
+  rm -f ../pair_coul_dsf.cpp
+  rm -f ../pair_coul_dsf_gpu.cpp
+  rm -f ../pair_lj_cut_coul_debye_gpu.cpp
   rm -f ../pppm_gpu.cpp
 
   rm -f ../fix_gpu.cpp
@@ -177,6 +228,19 @@ elif (test $1 = 0) then
   rm -f ../pair_resquared_gpu.h
   rm -f ../pair_table_gpu.h
   rm -f ../pair_yukawa_gpu.h
+  rm -f ../pair_born_gpu.h
+  rm -f ../pair_born_coul_wolf_gpu.h
+  rm -f ../pair_born_coul_long_gpu.h
+  rm -f ../pair_gauss_gpu.h
+  rm -f ../pair_colloid_gpu.h
+  rm -f ../pair_yukawa_colloid_gpu.h
+  rm -f ../pair_dipole_cut_gpu.h
+  rm -f ../pair_dipole_sf_gpu.h
+  rm -f ../pair_lj_cut_coul_dsf.h
+  rm -f ../pair_lj_cut_coul_dsf_gpu.h
+  rm -f ../pair_coul_dsf.h
+  rm -f ../pair_coul_dsf_gpu.h
+  rm -f ../pair_lj_cut_coul_debye_gpu.h
   rm -f ../pppm_gpu.h
 
   rm -f ../fix_gpu.h

src/GPU/Package.sh

@@ -9,6 +9,12 @@ for file in *.cpp *.h; do
   if (test $file = pair_gayberne_gpu.h -a ! -e ../pair_gayberne.cpp) then
     continue
   fi
+  if (test $file = pair_resquared_gpu.cpp -a ! -e ../pair_resquared.cpp) then
+    continue
+  fi
+  if (test $file = pair_resquared_gpu.h -a ! -e ../pair_resquared.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
@@ -33,7 +39,88 @@ for file in *.cpp *.h; do
   if (test $file = pair_lj_sdk_coul_long_gpu.h -a ! -e ../pair_lj_sdk_coul_long.cpp) then
     continue
   fi
-
+  if (test $file = pair_dipole_sf_gpu.cpp -a ! -e ../pair_dipole_sf.cpp) then
+    continue
+  fi
+  if (test $file = pair_dipole_sf_gpu.h -a ! -e ../pair_dipole_sf.cpp) then
+    continue
+  fi
+  if (test $file = pair_dipole_cut_gpu.cpp -a ! -e ../pair_dipole_cut.cpp) then
+    continue
+  fi
+  if (test $file = pair_dipole_cut_gpu.h -a ! -e ../pair_dipole_cut.cpp) then
+    continue
+  fi
+  if (test $file = pair_yukawa_colloid_gpu.cpp -a ! -e ../pair_yukawa_colloid.cpp) then
+    continue
+  fi
+  if (test $file = pair_yukawa_colloid_gpu.h -a ! -e ../pair_yukawa_colloid.cpp) then
+    continue
+  fi
+  if (test $file = pair_colloid_gpu.cpp -a ! -e ../pair_colloid.cpp) then
+    continue
+  fi
+  if (test $file = pair_colloid_gpu.h -a ! -e ../pair_colloid.cpp) then
+    continue
+  fi
+  if (test $file = pair_buck_coul_long_gpu.cpp -a ! -e ../pair_buck_coul_long.cpp) then
+    continue
+  fi
+  if (test $file = pair_buck_coul_long_gpu.h -a ! -e ../pair_buck_coul_long.cpp) then
+    continue
+  fi
+  if (test $file = pair_born_coul_long_gpu.cpp -a ! -e ../pair_born_coul_long.cpp) then
+    continue
+  fi
+  if (test $file = pair_born_coul_long_gpu.h -a ! -e ../pair_born_coul_long.cpp) then
+    continue
+  fi
+  if (test $file = pair_eam_gpu.cpp -a ! -e ../pair_eam.cpp) then
+    continue
+  fi
+  if (test $file = pair_eam_gpu.h -a ! -e ../pair_eam.cpp) then
+    continue
+  fi
+  if (test $file = pair_eam_alloy_gpu.cpp -a ! -e ../pair_eam_alloy.cpp) then
+    continue
+  fi
+  if (test $file = pair_eam_alloy_gpu.h -a ! -e ../pair_eam_alloy.cpp) then
+    continue
+  fi
+  if (test $file = pair_eam_fs_gpu.cpp -a ! -e ../pair_eam_fs.cpp) then
+    continue
+  fi
+  if (test $file = pair_eam_fs_gpu.h -a ! -e ../pair_eam_fs.cpp) then
+    continue
+  fi
+  if (test $file = pair_lj_class2_gpu.cpp -a ! -e ../pair_lj_class2.cpp) then
+    continue
+  fi
+  if (test $file = pair_lj_class2_coul_long_gpu.cpp -a ! -e ../pair_lj_class2_coul_long.cpp) then
+    continue
+  fi
+  if (test $file = pair_lj_class2_coul_long_gpu.h -a ! -e ../pair_lj_class2_coul_long.cpp) then
+    continue
+  fi
+  if (test $file = pair_lj_charmm_coul_long_gpu.cpp -a ! -e ../pair_lj_charmm_coul_long.cpp) then
+    continue
+  fi
+  if (test $file = pair_lj_charmm_coul_long_gpu.h -a ! -e ../pair_lj_charmm_coul_long.cpp) then
+    continue
+  fi
+  if (test $file = pair_lj_cut_coul_dsf_gpu.cpp -a ! -e ../pair_lj_cut_coul_dsf.cpp) then
+    continue
+  fi
+  if (test $file = pair_lj_cut_coul_dsf_gpu.h -a ! -e ../pair_lj_cut_coul_dsf.cpp) then
+    continue
+  fi
+  if (test $file = pppm_gpu.cpp -a ! -e ../pppm.cpp) then
+    continue
+  fi
+  if (test $file = pppm_gpu.h -a ! -e ../pppm.cpp) then
+    continue
+  fi
+  
   if (test ! -e ../$file) then
     echo "  creating src/$file"
     cp $file ..

src/GPU/fix_gpu.cpp

@@ -35,7 +35,8 @@ enum{GPU_FORCE, GPU_NEIGH, GPU_HYB_NEIGH};
 extern int lmp_init_device(MPI_Comm world, MPI_Comm replica,
                            const int first_gpu, const int last_gpu,
                            const int gpu_mode, const double particle_split,
-                           const int nthreads, const int t_per_atom);
+                           const int nthreads, const int t_per_atom,
+                           const double cell_size);
 extern void lmp_clear_device();
 extern double lmp_gpu_forces(double **f, double **tor, double *eatom,
                              double **vatom, double *virial, double &ecoul);
@@ -76,15 +77,23 @@ FixGPU::FixGPU(LAMMPS *lmp, int narg, char **arg) :
 
   int nthreads = 1;
   int threads_per_atom = -1;
-  if (narg == 9) {
+  double cell_size = -1;
-    if (strcmp(arg[7],"threads_per_atom") == 0)
+
-      threads_per_atom = atoi(arg[8]);
+  int iarg = 7;
-    else if (strcmp(arg[7],"nthreads") == 0)
+  while (iarg < narg) {
-      nthreads = atoi(arg[8]);
+    if (iarg+2 > narg) error->all(FLERR,"Illegal fix GPU command");
+
+    if (strcmp(arg[iarg],"threads_per_atom") == 0)
+      threads_per_atom = atoi(arg[iarg+1]);
+    else if (strcmp(arg[iarg],"nthreads") == 0)
+      nthreads = atoi(arg[iarg+1]);
+    else if (strcmp(arg[iarg],"cellsize") == 0)
+      cell_size = atof(arg[iarg+1]);
     else
       error->all(FLERR,"Illegal fix GPU command");
-  } else if (narg != 7)
+
-    error->all(FLERR,"Illegal fix GPU command");
+    iarg += 2;
+  }
 
   if (nthreads < 1)
     error->all(FLERR,"Illegal fix GPU command");
@@ -96,7 +105,7 @@ FixGPU::FixGPU(LAMMPS *lmp, int narg, char **arg) :
 
   int gpu_flag = lmp_init_device(universe->uworld, world, first_gpu, last_gpu,
                                  _gpu_mode, _particle_split, nthreads,
-                                 threads_per_atom);
+                                 threads_per_atom, cell_size);
   GPU_EXTRA::check_flag(gpu_flag,error,world);
 }
 

src/GPU/pppm_gpu.cpp

@@ -85,7 +85,7 @@ void PPPM_GPU_API(forces)(double **f);
 
 /* ---------------------------------------------------------------------- */
 
-PPPMGPU::PPPMGPU(LAMMPS *lmp, int narg, char **arg) : PPPMOld(lmp, narg, arg)
+PPPMGPU::PPPMGPU(LAMMPS *lmp, int narg, char **arg) : PPPM(lmp, narg, arg)
 {
   if (narg != 1) error->all(FLERR,"Illegal kspace_style pppm/gpu command");
 
@@ -111,10 +111,15 @@ PPPMGPU::~PPPMGPU()
 
 void PPPMGPU::init()
 {
-  PPPMOld::init();
+  PPPM::init();
 
-  if (strcmp(update->integrate_style,"verlet/split") == 0)
+  if (differentiation_flag == 1)
+    error->all(FLERR,"Cannot (yet) do analytic differentiation with pppm/gpu.");
+
+  if (strcmp(update->integrate_style,"verlet/split") == 0) {
     kspace_split=true;
+    old_nlocal = 0;
+  }
 
   if (kspace_split && universe->iworld == 0) {
     im_real_space = true;
@@ -153,21 +158,31 @@ void PPPMGPU::init()
 
 void PPPMGPU::compute(int eflag, int vflag)
 {
-  if (im_real_space) return;
+  int nago;
+  if (kspace_split) {
+    if (im_real_space) return;
+    if (atom->nlocal > old_nlocal) {
+      nago=0;
+      old_nlocal = atom->nlocal;
+    } else
+      nago=1;
+  } else
+    nago=neighbor->ago;
 
   // set energy/virial flags
   // invoke allocate_peratom() if needed for first time
 
   if (eflag || vflag) ev_setup(eflag,vflag);
-  else evflag = eflag_global = vflag_global = eflag_atom = vflag_atom = 0;
+  else evflag = evflag_atom = eflag_global = vflag_global = 
+        eflag_atom = vflag_atom = 0;
 
-  if (!peratom_allocate_flag && (eflag_atom || vflag_atom)) {
+  if (evflag_atom && !peratom_allocate_flag) {
     allocate_peratom();
     peratom_allocate_flag = 1;
   }
 
   bool success = true;
-  int flag=PPPM_GPU_API(spread)(neighbor->ago, atom->nlocal, atom->nlocal +
+  int flag=PPPM_GPU_API(spread)(nago, atom->nlocal, atom->nlocal +
                              atom->nghost, atom->x, atom->type, success,
                              atom->q, domain->boxlo, delxinv, delyinv,
                              delzinv);
@@ -241,7 +256,7 @@ void PPPMGPU::compute(int eflag, int vflag)
 
     if (vflag_atom) {
       for (i = 0; i < nlocal; i++)
-        for (j = 0; j < 6; j++) vatom[i][j] *= 0.5*q[i]*qscale;
+        for (j = 0; j < 6; j++) vatom[i][j] *= 0.5*qscale;
     }
   }
 
@@ -300,7 +315,10 @@ void PPPMGPU::allocate()
 
   memory->create(gf_b,order,"pppm:gf_b");
   memory->create2d_offset(rho1d,3,-order/2,order/2,"pppm:rho1d");
+  memory->create2d_offset(drho1d,3,-order/2,order/2,"pppm:drho1d");
   memory->create2d_offset(rho_coeff,order,(1-order)/2,order/2,"pppm:rho_coeff");
+  memory->create2d_offset(drho_coeff,order,(1-order)/2,order/2,
+                          "pppm:drho_coeff");
 
   // create 2 FFTs and a Remap
   // 1st FFT keeps data in FFT decompostion
@@ -349,7 +367,9 @@ void PPPMGPU::deallocate()
 
   memory->destroy(gf_b);
   memory->destroy2d_offset(rho1d,-order/2);
+  memory->destroy2d_offset(drho1d,-order/2);
   memory->destroy2d_offset(rho_coeff,(1-order)/2);
+  memory->destroy2d_offset(drho_coeff,(1-order)/2);
 
   delete fft1;
   delete fft2;
@@ -527,10 +547,20 @@ void PPPMGPU::brick2fft()
 }
 
 /* ----------------------------------------------------------------------
-   ghost-swap to fill ghost cells of my brick with field values
+   Same as base class - needed to call GPU version of fillbrick_.
 ------------------------------------------------------------------------- */
 
 void PPPMGPU::fillbrick()
+{
+  if (differentiation_flag == 1) fillbrick_ad();
+  else fillbrick_ik();
+}
+
+/* ----------------------------------------------------------------------
+   ghost-swap to fill ghost cells of my brick with field values
+------------------------------------------------------------------------- */
+
+void PPPMGPU::fillbrick_ik()
 {
   int i,n,ix,iy,iz;
   MPI_Request request;
@@ -727,10 +757,20 @@ void PPPMGPU::fillbrick()
 }
 
 /* ----------------------------------------------------------------------
-   FFT-based Poisson solver
+   Same code as base class - necessary to call GPU version of poisson_ik
 ------------------------------------------------------------------------- */
 
 void PPPMGPU::poisson()
+{
+  if (differentiation_flag == 1) poisson_ad();
+  else poisson_ik();
+}
+
+/* ----------------------------------------------------------------------
+   FFT-based Poisson solver
+------------------------------------------------------------------------- */
+
+void PPPMGPU::poisson_ik()
 {
   int i,j,k,n;
   double eng;
@@ -925,11 +965,11 @@ double PPPMGPU::memory_usage()
 
 int PPPMGPU::timing(int n, double &time3d, double &time1d) {
   if (im_real_space) {
-    time3d = 0.0;
+    time3d = 1.0;
-    time1d = 0.0;
+    time1d = 1.0;
     return 4;
   }
-  PPPMOld::timing(n,time3d,time1d);
+  PPPM::timing(n,time3d,time1d);
   return 4;
 }
 
@@ -940,5 +980,5 @@ int PPPMGPU::timing(int n, double &time3d, double &time1d) {
 void PPPMGPU::setup()
 {
   if (im_real_space) return;
-  PPPMOld::setup();
+  PPPM::setup();
 }

src/GPU/pppm_gpu.h

@@ -20,31 +20,34 @@ KSpaceStyle(pppm/gpu,PPPMGPU)
 #ifndef LMP_PPPM_GPU_H
 #define LMP_PPPM_GPU_H
 
-#include "pppm_old.h"
+#include "pppm.h"
 
 namespace LAMMPS_NS {
 
-class PPPMGPU : public PPPMOld {
+class PPPMGPU : public PPPM {
  public:
   PPPMGPU(class LAMMPS *, int, char **);
   virtual ~PPPMGPU();
-  virtual void init();
+  void init();
-  virtual void setup();
+  void setup();
-  virtual void compute(int, int);
+  void compute(int, int);
-  virtual int timing(int, double &, double &);
+  int timing(int, double &, double &);
-  virtual double memory_usage();
+  double memory_usage();
 
  protected:
 
   FFT_SCALAR ***density_brick_gpu, ***vd_brick;
   bool kspace_split, im_real_space;
 
-  virtual void allocate();
+  void allocate();
-  virtual void deallocate();
+  void deallocate();
-  virtual void brick2fft();
+  void brick2fft();
-  virtual void fillbrick();
+  void fillbrick();
-  virtual void poisson();
+  void fillbrick_ik();
+  void poisson();
+  void poisson_ik();
 
+  int old_nlocal;
   double poisson_time;
 
   FFT_SCALAR ***create_3d_offset(int, int, int, int, int, int, const char *,
@@ -65,6 +68,10 @@ Self-explanatory.  Check the input script syntax and compare to the
 documentation for the command.  You can use -echo screen as a
 command-line option when running LAMMPS to see the offending line.
 
+E: Cannot (yet) do analytic differentiation with pppm/gpu.
+
+Self-explanatory.
+
 E: Cannot use order greater than 8 with pppm/gpu.
 
 Self-explanatory.