use local variables for more efficient force accumulation

src/MANYBODY/pair_sw.cpp

@@ -106,6 +106,8 @@ void PairSW::compute(int eflag, int vflag)
   numneigh = list->numneigh;
   firstneigh = list->firstneigh;
 
+  double fxtmp,fytmp,fztmp;
+
   // loop over full neighbor list of my atoms
 
   for (ii = 0; ii < inum; ii++) {
@@ -115,6 +117,7 @@ void PairSW::compute(int eflag, int vflag)
     xtmp = x[i][0];
     ytmp = x[i][1];
     ztmp = x[i][2];
+    fxtmp = fytmp = fztmp = 0.0;
 
     // two-body interactions, skip half of them
 
@@ -154,12 +157,11 @@ void PairSW::compute(int eflag, int vflag)
         if (x[j][2] == ztmp && x[j][1] == ytmp && x[j][0] < xtmp) continue;
       }
 
-
       twobody(&params[ijparam],rsq,fpair,eflag,evdwl);
 
-      f[i][0] += delx*fpair;
-      f[i][1] += dely*fpair;
-      f[i][2] += delz*fpair;
+      fxtmp += delx*fpair;
+      fytmp += dely*fpair;
+      fztmp += delz*fpair;
       f[j][0] -= delx*fpair;
       f[j][1] -= dely*fpair;
       f[j][2] -= delz*fpair;
@@ -179,6 +181,9 @@ void PairSW::compute(int eflag, int vflag)
       delr1[2] = x[j][2] - ztmp;
       rsq1 = delr1[0]*delr1[0] + delr1[1]*delr1[1] + delr1[2]*delr1[2];
 
+      double fjxtmp,fjytmp,fjztmp;
+      fjxtmp = fjytmp = fjztmp = 0.0;
+
       for (kk = jj+1; kk < numshort; kk++) {
         k = neighshort[kk];
         ktype = map[type[k]];
@@ -193,19 +198,25 @@ void PairSW::compute(int eflag, int vflag)
         threebody(&params[ijparam],&params[ikparam],&params[ijkparam],
                   rsq1,rsq2,delr1,delr2,fj,fk,eflag,evdwl);
 
-        f[i][0] -= fj[0] + fk[0];
-        f[i][1] -= fj[1] + fk[1];
-        f[i][2] -= fj[2] + fk[2];
-        f[j][0] += fj[0];
-        f[j][1] += fj[1];
-        f[j][2] += fj[2];
+        fxtmp -= fj[0] + fk[0];
+        fytmp -= fj[1] + fk[1];
+        fztmp -= fj[2] + fk[2];
+        fjxtmp += fj[0];
+        fjytmp += fj[1];
+        fjztmp += fj[2];
         f[k][0] += fk[0];
         f[k][1] += fk[1];
         f[k][2] += fk[2];
 
         if (evflag) ev_tally3(i,j,k,evdwl,0.0,fj,fk,delr1,delr2);
       }
+      f[j][0] += fjxtmp;
+      f[j][1] += fjytmp;
+      f[j][2] += fjztmp;
     }
+    f[i][0] += fxtmp;
+    f[i][1] += fytmp;
+    f[i][2] += fztmp;
   }
 
   if (vflag_fdotr) virial_fdotr_compute();