1 // clang-format off
2 /* ----------------------------------------------------------------------
3    LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
4    https://www.lammps.org/, Sandia National Laboratories
5    Steve Plimpton, sjplimp@sandia.gov
6 
7    Copyright (2003) Sandia Corporation.  Under the terms of Contract
8    DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
9    certain rights in this software.  This software is distributed under
10    the GNU General Public License.
11 
12    See the README file in the top-level LAMMPS directory.
13 ------------------------------------------------------------------------- */
14 
15 /* ----------------------------------------------------------------------
16    Contributing author: Axel Kohlmeyer (Temple U)
17 ------------------------------------------------------------------------- */
18 
19 #include "dihedral_cosine_shift_exp_omp.h"
20 
21 #include "atom.h"
22 #include "comm.h"
23 #include "force.h"
24 #include "neighbor.h"
25 
26 #include <cmath>
27 
28 #include "omp_compat.h"
29 #include "suffix.h"
30 using namespace LAMMPS_NS;
31 
32 #define TOLERANCE 0.05
33 #define SMALL     0.001
34 
35 /* ---------------------------------------------------------------------- */
36 
DihedralCosineShiftExpOMP(class LAMMPS * lmp)37 DihedralCosineShiftExpOMP::DihedralCosineShiftExpOMP(class LAMMPS *lmp)
38   : DihedralCosineShiftExp(lmp), ThrOMP(lmp,THR_DIHEDRAL)
39 {
40   suffix_flag |= Suffix::OMP;
41 }
42 
43 /* ---------------------------------------------------------------------- */
44 
compute(int eflag,int vflag)45 void DihedralCosineShiftExpOMP::compute(int eflag, int vflag)
46 {
47   ev_init(eflag,vflag);
48 
49   const int nall = atom->nlocal + atom->nghost;
50   const int nthreads = comm->nthreads;
51   const int inum = neighbor->ndihedrallist;
52 
53 #if defined(_OPENMP)
54 #pragma omp parallel LMP_DEFAULT_NONE LMP_SHARED(eflag,vflag)
55 #endif
56   {
57     int ifrom, ito, tid;
58 
59     loop_setup_thr(ifrom, ito, tid, inum, nthreads);
60     ThrData *thr = fix->get_thr(tid);
61     thr->timer(Timer::START);
62     ev_setup_thr(eflag, vflag, nall, eatom, vatom, cvatom, thr);
63 
64     if (inum > 0) {
65       if (evflag) {
66         if (eflag) {
67           if (force->newton_bond) eval<1,1,1>(ifrom, ito, thr);
68           else eval<1,1,0>(ifrom, ito, thr);
69         } else {
70           if (force->newton_bond) eval<1,0,1>(ifrom, ito, thr);
71           else eval<1,0,0>(ifrom, ito, thr);
72         }
73       } else {
74         if (force->newton_bond) eval<0,0,1>(ifrom, ito, thr);
75         else eval<0,0,0>(ifrom, ito, thr);
76       }
77     }
78     thr->timer(Timer::BOND);
79     reduce_thr(this, eflag, vflag, thr);
80   } // end of omp parallel region
81 }
82 
83 template <int EVFLAG, int EFLAG, int NEWTON_BOND>
eval(int nfrom,int nto,ThrData * const thr)84 void DihedralCosineShiftExpOMP::eval(int nfrom, int nto, ThrData * const thr)
85 {
86 
87   int i1,i2,i3,i4,n,type;
88   double vb1x,vb1y,vb1z,vb2x,vb2y,vb2z,vb3x,vb3y,vb3z,vb2xm,vb2ym,vb2zm;
89   double edihedral,f1[3],f2[3],f3[3],f4[3];
90   double ax,ay,az,bx,by,bz,rasq,rbsq,rgsq,rg,rginv,ra2inv,rb2inv,rabinv;
91   double df,fg,hg,fga,hgb,gaa,gbb;
92   double dtfx,dtfy,dtfz,dtgx,dtgy,dtgz,dthx,dthy,dthz;
93   double c,s,sx2,sy2,sz2;
94   double cccpsss,cssmscc,exp2;
95 
96   edihedral = 0.0;
97 
98   const dbl3_t * _noalias const x = (dbl3_t *) atom->x[0];
99   dbl3_t * _noalias const f = (dbl3_t *) thr->get_f()[0];
100   const int5_t * _noalias const dihedrallist = (int5_t *) neighbor->dihedrallist[0];
101   const int nlocal = atom->nlocal;
102 
103   for (n = nfrom; n < nto; n++) {
104     i1 = dihedrallist[n].a;
105     i2 = dihedrallist[n].b;
106     i3 = dihedrallist[n].c;
107     i4 = dihedrallist[n].d;
108     type = dihedrallist[n].t;
109 
110     // 1st bond
111 
112     vb1x = x[i1].x - x[i2].x;
113     vb1y = x[i1].y - x[i2].y;
114     vb1z = x[i1].z - x[i2].z;
115 
116     // 2nd bond
117 
118     vb2x = x[i3].x - x[i2].x;
119     vb2y = x[i3].y - x[i2].y;
120     vb2z = x[i3].z - x[i2].z;
121 
122     vb2xm = -vb2x;
123     vb2ym = -vb2y;
124     vb2zm = -vb2z;
125 
126     // 3rd bond
127 
128     vb3x = x[i4].x - x[i3].x;
129     vb3y = x[i4].y - x[i3].y;
130     vb3z = x[i4].z - x[i3].z;
131 
132     // c,s calculation
133 
134     ax = vb1y*vb2zm - vb1z*vb2ym;
135     ay = vb1z*vb2xm - vb1x*vb2zm;
136     az = vb1x*vb2ym - vb1y*vb2xm;
137     bx = vb3y*vb2zm - vb3z*vb2ym;
138     by = vb3z*vb2xm - vb3x*vb2zm;
139     bz = vb3x*vb2ym - vb3y*vb2xm;
140 
141     rasq = ax*ax + ay*ay + az*az;
142     rbsq = bx*bx + by*by + bz*bz;
143     rgsq = vb2xm*vb2xm + vb2ym*vb2ym + vb2zm*vb2zm;
144     rg = sqrt(rgsq);
145 
146     rginv = ra2inv = rb2inv = 0.0;
147     if (rg > 0) rginv = 1.0/rg;
148     if (rasq > 0) ra2inv = 1.0/rasq;
149     if (rbsq > 0) rb2inv = 1.0/rbsq;
150     rabinv = sqrt(ra2inv*rb2inv);
151 
152     c = (ax*bx + ay*by + az*bz)*rabinv;
153     s = rg*rabinv*(ax*vb3x + ay*vb3y + az*vb3z);
154 
155     // error check
156 
157     if (c > 1.0 + TOLERANCE || c < (-1.0 - TOLERANCE))
158       problem(FLERR, i1, i2, i3, i4);
159 
160     if (c > 1.0) c = 1.0;
161     if (c < -1.0) c = -1.0;
162 
163     double aa=a[type];
164     double uumin=umin[type];
165 
166     cccpsss = c*cost[type]+s*sint[type];
167     cssmscc = c*sint[type]-s*cost[type];
168 
169     if (doExpansion[type]) {
170       //  |a|<0.001 so use expansions relative precision <1e-5
171       if (EFLAG) edihedral = -0.125*(1+cccpsss)*(4+aa*(cccpsss-1))*uumin;
172       df=0.5*uumin*( cssmscc + 0.5*aa*cccpsss);
173     } else {
174       exp2=exp(0.5*aa*(1+cccpsss));
175       if (EFLAG) edihedral = opt1[type]*(1-exp2);
176       df= 0.5*opt1[type]*aa* ( exp2*cssmscc );
177     }
178 
179     fg = vb1x*vb2xm + vb1y*vb2ym + vb1z*vb2zm;
180     hg = vb3x*vb2xm + vb3y*vb2ym + vb3z*vb2zm;
181     fga = fg*ra2inv*rginv;
182     hgb = hg*rb2inv*rginv;
183     gaa = -ra2inv*rg;
184     gbb = rb2inv*rg;
185 
186     dtfx = gaa*ax;
187     dtfy = gaa*ay;
188     dtfz = gaa*az;
189     dtgx = fga*ax - hgb*bx;
190     dtgy = fga*ay - hgb*by;
191     dtgz = fga*az - hgb*bz;
192     dthx = gbb*bx;
193     dthy = gbb*by;
194     dthz = gbb*bz;
195 
196     sx2 = df*dtgx;
197     sy2 = df*dtgy;
198     sz2 = df*dtgz;
199 
200     f1[0] = df*dtfx;
201     f1[1] = df*dtfy;
202     f1[2] = df*dtfz;
203 
204     f2[0] = sx2 - f1[0];
205     f2[1] = sy2 - f1[1];
206     f2[2] = sz2 - f1[2];
207 
208     f4[0] = df*dthx;
209     f4[1] = df*dthy;
210     f4[2] = df*dthz;
211 
212     f3[0] = -sx2 - f4[0];
213     f3[1] = -sy2 - f4[1];
214     f3[2] = -sz2 - f4[2];
215 
216     // apply force to each of 4 atoms
217 
218     if (NEWTON_BOND || i1 < nlocal) {
219       f[i1].x += f1[0];
220       f[i1].y += f1[1];
221       f[i1].z += f1[2];
222     }
223 
224     if (NEWTON_BOND || i2 < nlocal) {
225       f[i2].x += f2[0];
226       f[i2].y += f2[1];
227       f[i2].z += f2[2];
228     }
229 
230     if (NEWTON_BOND || i3 < nlocal) {
231       f[i3].x += f3[0];
232       f[i3].y += f3[1];
233       f[i3].z += f3[2];
234     }
235 
236     if (NEWTON_BOND || i4 < nlocal) {
237       f[i4].x += f4[0];
238       f[i4].y += f4[1];
239       f[i4].z += f4[2];
240     }
241 
242     if (EVFLAG)
243       ev_tally_thr(this,i1,i2,i3,i4,nlocal,NEWTON_BOND,edihedral,f1,f3,f4,
244                    vb1x,vb1y,vb1z,vb2x,vb2y,vb2z,vb3x,vb3y,vb3z,thr);
245   }
246 }
247