?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Continue removing -nb gpu_cpu
[gromacs.git] / src / gromacs / mdlib / nbnxn_kernels / nbnxn_kernel_gpu_ref.cpp
blob309d9a33e421e947944cf81a9e6697bc31d26853
1 /*
2 * This file is part of the GROMACS molecular simulation package.
3 *
4 * Copyright (c) 2012,2013,2014,2015,2016,2017, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
8 *
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
13 *
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
23 *
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
31 *
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
34 */
35 #include "gmxpre.h"
36
37 #include "nbnxn_kernel_gpu_ref.h"
38
39 #include "config.h"
40
41 #include <cmath>
42
43 #include <algorithm>
44
45 #include "gromacs/math/functions.h"
46 #include "gromacs/math/utilities.h"
47 #include "gromacs/math/vec.h"
48 #include "gromacs/mdlib/force.h"
49 #include "gromacs/mdlib/nb_verlet.h"
50 #include "gromacs/mdlib/nbnxn_consts.h"
51 #include "gromacs/mdtypes/md_enums.h"
52 #include "gromacs/pbcutil/ishift.h"
53 #include "gromacs/utility/fatalerror.h"
54
55 #include "nbnxn_kernel_common.h"
56
57 static const int c_numClPerSupercl = c_nbnxnGpuNumClusterPerSupercluster;
58 static const int c_clSize = c_nbnxnGpuClusterSize;
59
60 void
61 nbnxn_kernel_gpu_ref(const nbnxn_pairlist_t *nbl,
62 const nbnxn_atomdata_t *nbat,
63 const interaction_const_t *iconst,
64 rvec *shift_vec,
65 int force_flags,
66 int clearF,
67 real * f,
68 real * fshift,
69 real * Vc,
70 real * Vvdw)
71 {
72 const nbnxn_sci_t *nbln;
73 const real *x;
74 gmx_bool bEner;
75 gmx_bool bEwald;
76 const real *Ftab = nullptr;
77 real rcut2, rvdw2, rlist2;
78 int ntype;
79 real facel;
80 int n;
81 int ish3;
82 int sci;
83 int cj4_ind0, cj4_ind1, cj4_ind;
84 int ci, cj;
85 int ic, jc, ia, ja, is, ifs, js, jfs, im, jm;
86 int n0;
87 int ggid;
88 real shX, shY, shZ;
89 real fscal, tx, ty, tz;
90 real rinvsq;
91 real iq;
92 real qq, vcoul = 0, krsq, vctot;
93 int nti;
94 int tj;
95 real rt, r, eps;
96 real rinvsix;
97 real Vvdwtot;
98 real Vvdw_rep, Vvdw_disp;
99 real ix, iy, iz, fix, fiy, fiz;
100 real jx, jy, jz;
101 real dx, dy, dz, rsq, rinv;
102 int int_bit;
103 real fexcl;
104 real c6, c12;
105 const real * shiftvec;
106 real * vdwparam;
107 int * type;
108 const nbnxn_excl_t *excl[2];
109
110 int npair_tot, npair;
111 int nhwu, nhwu_pruned;
112
113 if (nbl->na_ci != c_clSize)
114 {
115 gmx_fatal(FARGS, "The neighborlist cluster size in the GPU reference kernel is %d, expected it to be %d", nbl->na_ci, c_clSize);
116 }
117
118 if (clearF == enbvClearFYes)
119 {
120 clear_f(nbat, 0, f);
121 }
122
123 bEner = (force_flags & GMX_FORCE_ENERGY);
124
125 bEwald = EEL_FULL(iconst->eeltype);
126 if (bEwald)
127 {
128 Ftab = iconst->tabq_coul_F;
129 }
130
131 rcut2 = iconst->rcoulomb*iconst->rcoulomb;
132 rvdw2 = iconst->rvdw*iconst->rvdw;
133
134 rlist2 = nbl->rlist*nbl->rlist;
135
136 type = nbat->type;
137 facel = iconst->epsfac;
138 shiftvec = shift_vec[0];
139 vdwparam = nbat->nbfp;
140 ntype = nbat->ntype;
141
142 x = nbat->x;
143
144 npair_tot = 0;
145 nhwu = 0;
146 nhwu_pruned = 0;
147
148 for (n = 0; n < nbl->nsci; n++)
149 {
150 nbln = &nbl->sci[n];
151
152 ish3 = 3*nbln->shift;
153 shX = shiftvec[ish3];
154 shY = shiftvec[ish3+1];
155 shZ = shiftvec[ish3+2];
156 cj4_ind0 = nbln->cj4_ind_start;
157 cj4_ind1 = nbln->cj4_ind_end;
158 sci = nbln->sci;
159 vctot = 0;
160 Vvdwtot = 0;
161
162 if (nbln->shift == CENTRAL &&
163 nbl->cj4[cj4_ind0].cj[0] == sci*c_numClPerSupercl)
164 {
165 /* we have the diagonal:
166 * add the charge self interaction energy term
167 */
168 for (im = 0; im < c_numClPerSupercl; im++)
169 {
170 ci = sci*c_numClPerSupercl + im;
171 for (ic = 0; ic < c_clSize; ic++)
172 {
173 ia = ci*c_clSize + ic;
174 iq = x[ia*nbat->xstride+3];
175 vctot += iq*iq;
176 }
177 }
178 if (!bEwald)
179 {
180 vctot *= -facel*0.5*iconst->c_rf;
181 }
182 else
183 {
184 /* last factor 1/sqrt(pi) */
185 vctot *= -facel*iconst->ewaldcoeff_q*M_1_SQRTPI;
186 }
187 }
188
189 for (cj4_ind = cj4_ind0; (cj4_ind < cj4_ind1); cj4_ind++)
190 {
191 excl[0] = &nbl->excl[nbl->cj4[cj4_ind].imei[0].excl_ind];
192 excl[1] = &nbl->excl[nbl->cj4[cj4_ind].imei[1].excl_ind];
193
194 for (jm = 0; jm < c_nbnxnGpuJgroupSize; jm++)
195 {
196 cj = nbl->cj4[cj4_ind].cj[jm];
197
198 for (im = 0; im < c_numClPerSupercl; im++)
199 {
200 /* We're only using the first imask,
201 * but here imei[1].imask is identical.
202 */
203 if ((nbl->cj4[cj4_ind].imei[0].imask >> (jm*c_numClPerSupercl + im)) & 1)
204 {
205 gmx_bool within_rlist;
206
207 ci = sci*c_numClPerSupercl + im;
208
209 within_rlist = FALSE;
210 npair = 0;
211 for (ic = 0; ic < c_clSize; ic++)
212 {
213 ia = ci*c_clSize + ic;
214
215 is = ia*nbat->xstride;
216 ifs = ia*nbat->fstride;
217 ix = shX + x[is+0];
218 iy = shY + x[is+1];
219 iz = shZ + x[is+2];
220 iq = facel*x[is+3];
221 nti = ntype*2*type[ia];
222
223 fix = 0;
224 fiy = 0;
225 fiz = 0;
226
227 for (jc = 0; jc < c_clSize; jc++)
228 {
229 ja = cj*c_clSize + jc;
230
231 if (nbln->shift == CENTRAL &&
232 ci == cj && ja <= ia)
233 {
234 continue;
235 }
236
237 int_bit = ((excl[jc >> 2]->pair[(jc & 3)*c_clSize + ic] >> (jm*c_numClPerSupercl + im)) & 1);
238
239 js = ja*nbat->xstride;
240 jfs = ja*nbat->fstride;
241 jx = x[js+0];
242 jy = x[js+1];
243 jz = x[js+2];
244 dx = ix - jx;
245 dy = iy - jy;
246 dz = iz - jz;
247 rsq = dx*dx + dy*dy + dz*dz;
248 if (rsq < rlist2)
249 {
250 within_rlist = TRUE;
251 }
252 if (rsq >= rcut2)
253 {
254 continue;
255 }
256
257 if (type[ia] != ntype-1 && type[ja] != ntype-1)
258 {
259 npair++;
260 }
261
262 // Ensure distance do not become so small that r^-12 overflows
263 rsq = std::max(rsq, NBNXN_MIN_RSQ);
264
265 rinv = gmx::invsqrt(rsq);
266 rinvsq = rinv*rinv;
267
268 qq = iq*x[js+3];
269 if (!bEwald)
270 {
271 /* Reaction-field */
272 krsq = iconst->k_rf*rsq;
273 fscal = qq*(int_bit*rinv - 2*krsq)*rinvsq;
274 if (bEner)
275 {
276 vcoul = qq*(int_bit*rinv + krsq - iconst->c_rf);
277 }
278 }
279 else
280 {
281 r = rsq*rinv;
282 rt = r*iconst->tabq_scale;
283 n0 = rt;
284 eps = rt - n0;
285
286 fexcl = (1 - eps)*Ftab[n0] + eps*Ftab[n0+1];
287
288 fscal = qq*(int_bit*rinvsq - fexcl)*rinv;
289
290 if (bEner)
291 {
292 vcoul = qq*((int_bit - std::erf(iconst->ewaldcoeff_q*r))*rinv - int_bit*iconst->sh_ewald);
293 }
294 }
295
296 if (rsq < rvdw2)
297 {
298 tj = nti + 2*type[ja];
299
300 /* Vanilla Lennard-Jones cutoff */
301 c6 = vdwparam[tj];
302 c12 = vdwparam[tj+1];
303
304 rinvsix = int_bit*rinvsq*rinvsq*rinvsq;
305 Vvdw_disp = c6*rinvsix;
306 Vvdw_rep = c12*rinvsix*rinvsix;
307 fscal += (Vvdw_rep - Vvdw_disp)*rinvsq;
308
309 if (bEner)
310 {
311 vctot += vcoul;
312
313 Vvdwtot +=
314 (Vvdw_rep - int_bit*c12*iconst->sh_invrc6*iconst->sh_invrc6)/12 -
315 (Vvdw_disp - int_bit*c6*iconst->sh_invrc6)/6;
316 }
317 }
318
319 tx = fscal*dx;
320 ty = fscal*dy;
321 tz = fscal*dz;
322 fix = fix + tx;
323 fiy = fiy + ty;
324 fiz = fiz + tz;
325 f[jfs+0] -= tx;
326 f[jfs+1] -= ty;
327 f[jfs+2] -= tz;
328 }
329
330 f[ifs+0] += fix;
331 f[ifs+1] += fiy;
332 f[ifs+2] += fiz;
333 fshift[ish3] = fshift[ish3] + fix;
334 fshift[ish3+1] = fshift[ish3+1] + fiy;
335 fshift[ish3+2] = fshift[ish3+2] + fiz;
336
337 /* Count in half work-units.
338 * In CUDA one work-unit is 2 warps.
339 */
340 if ((ic+1) % (c_clSize/c_nbnxnGpuClusterpairSplit) == 0)
341 {
342 npair_tot += npair;
343
344 nhwu++;
345 if (within_rlist)
346 {
347 nhwu_pruned++;
348 }
349
350 within_rlist = FALSE;
351 npair = 0;
352 }
353 }
354 }
355 }
356 }
357 }
358
359 if (bEner)
360 {
361 ggid = 0;
362 Vc[ggid] = Vc[ggid] + vctot;
363 Vvdw[ggid] = Vvdw[ggid] + Vvdwtot;
364 }
365 }
366
367 if (debug)
368 {
369 fprintf(debug, "number of half %dx%d atom pairs: %d after pruning: %d fraction %4.2f\n",
370 nbl->na_ci, nbl->na_ci,
371 nhwu, nhwu_pruned, nhwu_pruned/(double)nhwu);
372 fprintf(debug, "generic kernel pair interactions: %d\n",
373 nhwu*nbl->na_ci/2*nbl->na_ci);
374 fprintf(debug, "generic kernel post-prune pair interactions: %d\n",
375 nhwu_pruned*nbl->na_ci/2*nbl->na_ci);
376 fprintf(debug, "generic kernel non-zero pair interactions: %d\n",
377 npair_tot);
378 fprintf(debug, "ratio non-zero/post-prune pair interactions: %4.2f\n",
379 npair_tot/(double)(nhwu_pruned*nbl->na_ci/2*nbl->na_ci));
380 }
381 }
The ultimate molecular dynamics simulation package