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: W. Michael Brown (Intel)
17 ------------------------------------------------------------------------- */
18
19 #include "bond_harmonic_intel.h"
20
21 #include "atom.h"
22 #include "comm.h"
23 #include "error.h"
24 #include "force.h"
25 #include "memory.h"
26 #include "modify.h"
27 #include "neighbor.h"
28 #include "suffix.h"
29
30 #include <cmath>
31 #include <cstring>
32
33 #include "omp_compat.h"
34
35 using namespace LAMMPS_NS;
36
37 typedef struct { int a,b,t; } int3_t;
38
39 /* ---------------------------------------------------------------------- */
40
BondHarmonicIntel(LAMMPS * lmp)41 BondHarmonicIntel::BondHarmonicIntel(LAMMPS *lmp) : BondHarmonic(lmp)
42 {
43 suffix_flag |= Suffix::INTEL;
44 }
45
46 /* ---------------------------------------------------------------------- */
47
~BondHarmonicIntel()48 BondHarmonicIntel::~BondHarmonicIntel()
49 {
50 }
51
52 /* ---------------------------------------------------------------------- */
53
compute(int eflag,int vflag)54 void BondHarmonicIntel::compute(int eflag, int vflag)
55 {
56 #ifdef _LMP_INTEL_OFFLOAD
57 if (_use_base) {
58 BondHarmonic::compute(eflag, vflag);
59 return;
60 }
61 #endif
62
63 if (fix->precision() == FixIntel::PREC_MODE_MIXED)
64 compute<float,double>(eflag, vflag, fix->get_mixed_buffers(),
65 force_const_single);
66 else if (fix->precision() == FixIntel::PREC_MODE_DOUBLE)
67 compute<double,double>(eflag, vflag, fix->get_double_buffers(),
68 force_const_double);
69 else
70 compute<float,float>(eflag, vflag, fix->get_single_buffers(),
71 force_const_single);
72 }
73
74 /* ---------------------------------------------------------------------- */
75
76 template <class flt_t, class acc_t>
compute(int eflag,int vflag,IntelBuffers<flt_t,acc_t> * buffers,const ForceConst<flt_t> & fc)77 void BondHarmonicIntel::compute(int eflag, int vflag,
78 IntelBuffers<flt_t,acc_t> *buffers,
79 const ForceConst<flt_t> &fc)
80 {
81 ev_init(eflag,vflag);
82 if (vflag_atom)
83 error->all(FLERR,"INTEL package does not support per-atom stress");
84
85 if (evflag) {
86 if (vflag && !eflag) {
87 if (force->newton_bond)
88 eval<0,1,1>(vflag, buffers, fc);
89 else
90 eval<0,1,0>(vflag, buffers, fc);
91 } else {
92 if (force->newton_bond)
93 eval<1,1,1>(vflag, buffers, fc);
94 else
95 eval<1,1,0>(vflag, buffers, fc);
96 }
97 } else {
98 if (force->newton_bond)
99 eval<0,0,1>(vflag, buffers, fc);
100 else
101 eval<0,0,0>(vflag, buffers, fc);
102 }
103 }
104
105 template <int EFLAG, int VFLAG, int NEWTON_BOND, class flt_t, class acc_t>
eval(const int vflag,IntelBuffers<flt_t,acc_t> * buffers,const ForceConst<flt_t> & fc)106 void BondHarmonicIntel::eval(const int vflag,
107 IntelBuffers<flt_t,acc_t> *buffers,
108 const ForceConst<flt_t> &fc)
109 {
110 const int inum = neighbor->nbondlist;
111 if (inum == 0) return;
112
113 ATOM_T * _noalias const x = buffers->get_x(0);
114 const int nlocal = atom->nlocal;
115 const int nall = nlocal + atom->nghost;
116
117 int f_stride;
118 if (NEWTON_BOND) f_stride = buffers->get_stride(nall);
119 else f_stride = buffers->get_stride(nlocal);
120
121 int tc;
122 FORCE_T * _noalias f_start;
123 acc_t * _noalias ev_global;
124 IP_PRE_get_buffers(0, buffers, fix, tc, f_start, ev_global);
125 const int nthreads = tc;
126
127 acc_t oebond, ov0, ov1, ov2, ov3, ov4, ov5;
128 if (EFLAG) oebond = (acc_t)0.0;
129 if (VFLAG && vflag) {
130 ov0 = ov1 = ov2 = ov3 = ov4 = ov5 = (acc_t)0.0;
131 }
132
133 #if defined(_OPENMP)
134 #pragma omp parallel LMP_DEFAULT_NONE \
135 shared(f_start,f_stride,fc) \
136 reduction(+:oebond,ov0,ov1,ov2,ov3,ov4,ov5)
137 #endif
138 {
139 int nfrom, npl, nto, tid;
140 #ifdef LMP_INTEL_USE_SIMDOFF
141 IP_PRE_omp_range_id(nfrom, nto, tid, inum, nthreads);
142 #else
143 IP_PRE_omp_stride_id(nfrom, npl, nto, tid, inum, nthreads);
144 #endif
145
146 FORCE_T * _noalias const f = f_start + (tid * f_stride);
147 if (fix->need_zero(tid))
148 memset(f, 0, f_stride * sizeof(FORCE_T));
149
150 const int3_t * _noalias const bondlist =
151 (int3_t *) neighbor->bondlist[0];
152
153 #ifdef LMP_INTEL_USE_SIMDOFF
154 acc_t sebond, sv0, sv1, sv2, sv3, sv4, sv5;
155 if (EFLAG) sebond = (acc_t)0.0;
156 if (VFLAG && vflag) {
157 sv0 = sv1 = sv2 = sv3 = sv4 = sv5 = (acc_t)0.0;
158 }
159 #if defined(USE_OMP_SIMD)
160 #pragma omp simd reduction(+:sebond, sv0, sv1, sv2, sv3, sv4, sv5)
161 #else
162 #pragma simd reduction(+:sebond, sv0, sv1, sv2, sv3, sv4, sv5)
163 #endif
164 for (int n = nfrom; n < nto; n ++) {
165 #else
166 for (int n = nfrom; n < nto; n += npl) {
167 #endif
168 const int i1 = bondlist[n].a;
169 const int i2 = bondlist[n].b;
170 const int type = bondlist[n].t;
171
172 const flt_t delx = x[i1].x - x[i2].x;
173 const flt_t dely = x[i1].y - x[i2].y;
174 const flt_t delz = x[i1].z - x[i2].z;
175
176 const flt_t rsq = delx*delx + dely*dely + delz*delz;
177 const flt_t r = sqrt(rsq);
178 const flt_t dr = r - fc.fc[type].r0;
179 const flt_t rk = fc.fc[type].k * dr;
180
181 // force & energy
182
183 flt_t fbond;
184 if (r > (flt_t)0.0) fbond = (flt_t)-2.0*rk/r;
185 else fbond = (flt_t)0.0;
186
187 flt_t ebond;
188 if (EFLAG) ebond = rk*dr;
189
190 // apply force to each of 2 atoms
191 #ifdef LMP_INTEL_USE_SIMDOFF
192 #if defined(USE_OMP_SIMD)
193 #pragma omp ordered simd
194 #else
195 #pragma simdoff
196 #endif
197 #endif
198 {
199 if (NEWTON_BOND || i1 < nlocal) {
200 f[i1].x += delx*fbond;
201 f[i1].y += dely*fbond;
202 f[i1].z += delz*fbond;
203 }
204
205 if (NEWTON_BOND || i2 < nlocal) {
206 f[i2].x -= delx*fbond;
207 f[i2].y -= dely*fbond;
208 f[i2].z -= delz*fbond;
209 }
210 }
211
212 if (EFLAG || VFLAG) {
213 #ifdef LMP_INTEL_USE_SIMDOFF
214 IP_PRE_ev_tally_bond(EFLAG, VFLAG, eatom, vflag, ebond, i1, i2,
215 fbond, delx, dely, delz, sebond, f,
216 NEWTON_BOND, nlocal, sv0, sv1, sv2, sv3,
217 sv4, sv5);
218 #else
219 IP_PRE_ev_tally_bond(EFLAG, VFLAG, eatom, vflag, ebond, i1, i2,
220 fbond, delx, dely, delz, oebond, f,
221 NEWTON_BOND, nlocal, ov0, ov1, ov2, ov3,
222 ov4, ov5);
223 #endif
224 }
225 } // for n
226 #ifdef LMP_INTEL_USE_SIMDOFF
227 if (EFLAG) oebond += sebond;
228 if (VFLAG && vflag) {
229 ov0 += sv0; ov1 += sv1; ov2 += sv2;
230 ov3 += sv3; ov4 += sv4; ov5 += sv5;
231 }
232 #endif
233 } // omp parallel
234
235 if (EFLAG) energy += oebond;
236 if (VFLAG && vflag) {
237 virial[0] += ov0; virial[1] += ov1; virial[2] += ov2;
238 virial[3] += ov3; virial[4] += ov4; virial[5] += ov5;
239 }
240
241 fix->set_reduce_flag();
242 }
243
244 /* ---------------------------------------------------------------------- */
245
246 void BondHarmonicIntel::init_style()
247 {
248 BondHarmonic::init_style();
249
250 int ifix = modify->find_fix("package_intel");
251 if (ifix < 0)
252 error->all(FLERR,
253 "The 'package intel' command is required for /intel styles");
254 fix = static_cast<FixIntel *>(modify->fix[ifix]);
255
256 #ifdef _LMP_INTEL_OFFLOAD
257 _use_base = 0;
258 if (fix->offload_balance() != 0.0) {
259 _use_base = 1;
260 return;
261 }
262 #endif
263
264 fix->bond_init_check();
265
266 if (fix->precision() == FixIntel::PREC_MODE_MIXED)
267 pack_force_const(force_const_single, fix->get_mixed_buffers());
268 else if (fix->precision() == FixIntel::PREC_MODE_DOUBLE)
269 pack_force_const(force_const_double, fix->get_double_buffers());
270 else
271 pack_force_const(force_const_single, fix->get_single_buffers());
272 }
273
274 /* ---------------------------------------------------------------------- */
275
276 template <class flt_t, class acc_t>
277 void BondHarmonicIntel::pack_force_const(ForceConst<flt_t> &fc,
278 IntelBuffers<flt_t,acc_t> * /*buffers*/)
279 {
280 const int bp1 = atom->nbondtypes + 1;
281 fc.set_ntypes(bp1,memory);
282
283 for (int i = 1; i < bp1; i++) {
284 fc.fc[i].k = k[i];
285 fc.fc[i].r0 = r0[i];
286 }
287 }
288
289 /* ---------------------------------------------------------------------- */
290
291 template <class flt_t>
292 void BondHarmonicIntel::ForceConst<flt_t>::set_ntypes(const int nbondtypes,
293 Memory *memory) {
294 if (nbondtypes != _nbondtypes) {
295 if (_nbondtypes > 0)
296 _memory->destroy(fc);
297
298 if (nbondtypes > 0)
299 _memory->create(fc,nbondtypes,"bondharmonicintel.fc");
300 }
301 _nbondtypes = nbondtypes;
302 _memory = memory;
303 }
304