git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@9002 f3b2605a-c512-4ea7-a41b-209d697bcdaa

examples/COUPLE/README

@@ -32,7 +32,7 @@ lammps_spparks	    grain-growth Monte Carlo with strain via MD,
 		    coupling to SPPARKS kinetic MC code
 library		    collection of useful inter-code communication routines
 simple		    simple example of driver code calling LAMMPS as library
-fortran             a wrapper on the LAMMPS library API that
+fortran             a simple wrapper on the LAMMPS library API that
  		      can be called from Fortran
 fortran2            a more sophisticated wrapper on the LAMMPS library API that
  		      can be called from Fortran

src/REPLICA/verlet_split.cpp

@@ -1,4 +1,4 @@
-/* -------------------------------------------------------------------------
+/* ----------------------------------------------------------------------
    LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
    http://lammps.sandia.gov, Sandia National Laboratories
    Steve Plimpton, sjplimp@sandia.gov
@@ -32,6 +32,7 @@
 #include "kspace.h"
 #include "output.h"
 #include "update.h"
+#include "fix.h"
 #include "modify.h"
 #include "timer.h"
 #include "memory.h"
@@ -260,7 +261,8 @@ void VerletSplit::setup_minimal(int flag)
 
 void VerletSplit::run(int n)
 {
-  int nflag,ntimestep,sortflag;
+  bigint ntimestep;
+  int nflag,sortflag;
 
   // sync both partitions before start timer
 
@@ -272,6 +274,13 @@ void VerletSplit::run(int n)
 
   rk_setup();
 
+  // check if OpenMP support fix defined
+
+  Fix *fix_omp;
+  int ifix = modify->find_fix("package_omp");
+  if (ifix < 0) fix_omp = NULL;
+  else fix_omp = modify->fix[ifix];
+
   // flags for timestepping iterations
 
   int n_post_integrate = modify->n_post_integrate;
@@ -360,6 +369,11 @@ void VerletSplit::run(int n)
       }
 
     } else {
+
+      // run FixOMP as sole pre_force fix, if defined
+
+      if (fix_omp) fix_omp->pre_force(vflag);
+
       if (force->kspace) {
         timer->stamp();
         force->kspace->compute(eflag,vflag);

src/USER-OMP/Install.sh

@@ -1,10 +1,9 @@
 # Install/unInstall package files in LAMMPS
 # do not install child files if parent does not exist
 
-for file in *_omp.cpp *_omp.h  pppm*proxy.h pppm*proxy.cpp; do
+for file in *_omp.cpp *_omp.h ; do
     # let us see if the "rain man" can count the toothpicks...
-   ofile=`echo $file | sed  -e s,_pppm_tip4p_omp,_long_tip4p_omp, \
+   ofile=`echo $file | sed  \
-   -e s,pppm.\\*_proxy,pppm_omp, -e s,_pppm_omp,_long_omp, \
    -e s,\\\\\\(.\\*\\\\\\)_omp\\\\.h,\\\\1.h, \
    -e s,\\\\\\(.\\*\\\\\\)_omp\\\\.cpp,\\\\1.cpp,`
   if (test $1 = 1) then

src/USER-OMP/Package.sh

@@ -4,10 +4,9 @@
 # not exist. Do remove OpenMP style files that have no matching
 # non-OpenMP version installed, e.g. after a package has been
 # removed
-for file in *_omp.cpp *_omp.h pppm*proxy.h pppm*proxy.cpp thr_data.h thr_data.cpp; do
+for file in *_omp.cpp *_omp.h thr_data.h thr_data.cpp; do
   # let us see if the "rain man" can count the toothpicks...
-   ofile=`echo $file | sed  -e s,_pppm_tip4p_omp,_long_tip4p_omp, \
+   ofile=`echo $file | sed  \
-   -e s,pppm.\\*_proxy,pppm_omp, -e s,_pppm_omp,_long_omp, \
    -e s,\\\\\\(.\\*\\\\\\)_omp\\\\.\\\\\\(h\\\\\\|cpp\\\\\\),\\\\1.\\\\2,`
   if (test $file = "thr_omp.h") || (test $file = "thr_omp.cpp") \
       || (test $file = "thr_data.h") || (test $file = "thr_data.cpp") then

src/USER-OMP/fix_omp.cpp

@@ -22,6 +22,7 @@
 #include "force.h"
 #include "neighbor.h"
 #include "neigh_request.h"
+#include "universe.h"
 #include "update.h"
 #include "integrate.h"
 #include "min.h"
@@ -194,13 +195,24 @@ void FixOMP::init()
   if (strstr(update->integrate_style,"respa") != NULL)
     error->all(FLERR,"Cannot use r-RESPA with /omp styles");
 
-  int check_hybrid;
+  int check_hybrid, kspace_split;
   last_pair_hybrid = NULL;
   last_omp_style = NULL;
   const char *last_omp_name = NULL;
   const char *last_hybrid_name = NULL;
   const char *last_force_name = NULL;
 
+  // support for verlet/split operation.
+  // kspace_split == 0 : regular processing
+  // kspace_split < 0  : master partition, does not do kspace
+  // kspace_split > 0  : slave partition, only does kspace
+  if (strstr(update->integrate_style,"verlet/split") != NULL) {
+    if (universe->iworld == 0) kspace_split = -1;
+    else kspace_split = 1;
+  } else {
+    kspace_split = 0;
+  }
+
 // determine which is the last force style with OpenMP
 // support as this is the one that has to reduce the forces
 
@@ -229,6 +241,7 @@ void FixOMP::init()
     }							      \
   }
 
+  if (kspace_split <= 0) {
     CheckStyleForOMP(pair);
     CheckHybridForOMP(pair,Pair);
     if (check_hybrid) {
@@ -247,8 +260,11 @@ void FixOMP::init()
 
     CheckStyleForOMP(improper);
     CheckHybridForOMP(improper,Improper);
+  }
 
+  if (kspace_split >= 0) {
     CheckStyleForOMP(kspace);
+  }
 
 #undef CheckStyleForOMP
 #undef CheckHybridForOMP

src/USER-OMP/thr_omp.cpp

@@ -147,7 +147,7 @@ void ThrOMP::ev_setup_thr(int eflag, int vflag, int nall, double *eatom,
    ---------------------------------------------------------------------- */
 
 void ThrOMP::reduce_thr(void *style, const int eflag, const int vflag,
-                        ThrData *const thr, const int nproxy)
+                        ThrData *const thr)
 {
   const int nlocal = lmp->atom->nlocal;
   const int nghost = lmp->atom->nghost;
@@ -172,12 +172,7 @@ void ThrOMP::reduce_thr(void *style, const int eflag, const int vflag,
 
     if (pair->vflag_fdotr) {
 
-      if (style == fix->last_pair_hybrid) {
+      if (fix->last_pair_hybrid) return;
-        // pair_style hybrid will compute fdotr for us
-        // but we first need to reduce the forces
-        data_reduce_thr(&(f[0][0]), nall, nthreads, 3, tid);
-        need_force_reduce = 0;
-      }
 
       // this is a non-hybrid pair style. compute per thread fdotr
       if (fix->last_pair_hybrid == NULL) {
@@ -205,67 +200,7 @@ void ThrOMP::reduce_thr(void *style, const int eflag, const int vflag,
             thr->virial_pair[i] = 0.0;
           }
       }
-      if (eflag & 2) {
-        data_reduce_thr(&(pair->eatom[0]), nall, nthreads, 1, tid);
-      }
-      if (vflag & 4) {
-        data_reduce_thr(&(pair->vatom[0][0]), nall, nthreads, 6, tid);
-      }
-    }
-  }
-    break;
 
-  case THR_PAIR|THR_PROXY: {
-    Pair * const pair = lmp->force->pair;
-
-    if (tid >= nproxy && pair->vflag_fdotr) {
-
-      if (fix->last_pair_hybrid) {
-        if (tid == nproxy)
-          lmp->error->all(FLERR,
-                          "Cannot use hybrid pair style with kspace proxy");
-        else return;
-      }
-
-      // this is a non-hybrid pair style. compute per thread fdotr
-      if (fix->last_pair_hybrid == NULL) {
-        if (lmp->neighbor->includegroup == 0)
-          thr->virial_fdotr_compute(x, nlocal, nghost, -1);
-        else
-          thr->virial_fdotr_compute(x, nlocal, nghost, nfirst);
-      }
-    }
-
-    if (evflag) {
-#if defined(_OPENMP)
-#pragma omp critical
-#endif
-      {
-        if (tid < nproxy) {
-          // nothing to collect for kspace proxy threads
-          // just reset pair accumulators to 0.0.
-          if (eflag & 1) {
-            thr->eng_vdwl = 0.0;
-            thr->eng_coul = 0.0;
-          }
-          if (vflag & 3)
-            for (int i=0; i < 6; ++i) {
-              thr->virial_pair[i] = 0.0;
-            }
-        } else {
-          if (eflag & 1) {
-            pair->eng_vdwl += thr->eng_vdwl;
-            pair->eng_coul += thr->eng_coul;
-            thr->eng_vdwl = 0.0;
-            thr->eng_coul = 0.0;
-          }
-          if (vflag & 3)
-            for (int i=0; i < 6; ++i) {
-              pair->virial[i] += thr->virial_pair[i];
-              thr->virial_pair[i] = 0.0;
-            }
-        }
-      }
       if (eflag & 2) {
         data_reduce_thr(&(pair->eatom[0]), nall, nthreads, 1, tid);
       }
@@ -439,9 +374,8 @@ void ThrOMP::reduce_thr(void *style, const int eflag, const int vflag,
     }
     break;
 
-  case THR_KSPACE|THR_PROXY: // fallthrough
   case THR_KSPACE:
-    // nothing to do
+    // nothing to do. XXX need to add support for per-atom info
     break;
 
   default:

src/USER-OMP/thr_omp.h

@@ -58,7 +58,8 @@ class ThrOMP {
 
   enum {THR_NONE=0,THR_PAIR=1,THR_BOND=1<<1,THR_ANGLE=1<<2,
         THR_DIHEDRAL=1<<3,THR_IMPROPER=1<<4,THR_KSPACE=1<<5,
-        THR_CHARMM=1<<6,THR_PROXY=1<<7,THR_HYBRID=1<<8,THR_FIX=1<<9};
+        THR_CHARMM=1<<6, /*THR_PROXY=1<<7,*/ THR_HYBRID=1<<8,
+        THR_FIX=1<<9};
 
  protected:
   // extra ev_tally setup work for threaded styles
@@ -71,7 +72,7 @@ class ThrOMP {
 
   // reduce per thread data as needed
   void reduce_thr(void * const style, const int eflag, const int vflag,
-                  ThrData * const thr, const int nproxy=0);
+                  ThrData * const thr);
 
   // thread safe variant error abort support.
   // signals an error condition in any thread by making
@@ -167,14 +168,14 @@ class ThrOMP {
 
 // set loop range thread id, and force array offset for threaded runs.
 static inline void loop_setup_thr(int &ifrom, int &ito, int &tid,
-                                  int inum, int nthreads, int nproxy=0)
+                                  int inum, int nthreads)
 {
 #if defined(_OPENMP)
   tid = omp_get_thread_num();
 
   // each thread works on a fixed chunk of atoms.
-  const int idelta = 1 + inum/(nthreads-nproxy);
+  const int idelta = 1 + inum/nthreads;
-  ifrom = (tid-nproxy)*idelta;
+  ifrom = tid*idelta;
   ito   = ((ifrom + idelta) > inum) ? inum : ifrom + idelta;
 #else
   tid = 0;