2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2012,2013,2014,2015,2016,2017,2018,2019, 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.
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.
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.
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.
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.
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.
37 #include "nbnxn_kernel_gpu_ref.h"
45 #include "gromacs/math/functions.h"
46 #include "gromacs/math/utilities.h"
47 #include "gromacs/math/vec.h"
48 #include "gromacs/mdlib/force_flags.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"
55 #include "nbnxn_kernel_common.h"
57 static const int c_numClPerSupercl = c_nbnxnGpuNumClusterPerSupercluster;
58 static const int c_clSize = c_nbnxnGpuClusterSize;
61 nbnxn_kernel_gpu_ref(const NbnxnPairlistGpu *nbl,
62 const nbnxn_atomdata_t *nbat,
63 const interaction_const_t *iconst,
72 const nbnxn_sci_t *nbln;
76 const real *Ftab = nullptr;
77 real rcut2, rvdw2, rlist2;
83 int cj4_ind0, cj4_ind1, cj4_ind;
85 int ic, jc, ia, ja, is, ifs, js, jfs, im, jm;
89 real fscal, tx, ty, tz;
92 real qq, vcoul = 0, krsq, vctot;
98 real Vvdw_rep, Vvdw_disp;
99 real ix, iy, iz, fix, fiy, fiz;
101 real dx, dy, dz, rsq, rinv;
105 const real * shiftvec;
108 const nbnxn_excl_t *excl[2];
110 int npair_tot, npair;
111 int nhwu, nhwu_pruned;
113 if (nbl->na_ci != c_clSize)
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);
118 if (clearF == enbvClearFYes)
123 bEner = ((force_flags & GMX_FORCE_ENERGY) != 0);
125 bEwald = EEL_FULL(iconst->eeltype);
128 Ftab = iconst->tabq_coul_F;
131 rcut2 = iconst->rcoulomb*iconst->rcoulomb;
132 rvdw2 = iconst->rvdw*iconst->rvdw;
134 rlist2 = nbl->rlist*nbl->rlist;
137 facel = iconst->epsfac;
138 shiftvec = shift_vec[0];
139 vdwparam = nbat->nbfp;
148 for (n = 0; n < nbl->nsci; n++)
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;
162 if (nbln->shift == CENTRAL &&
163 nbl->cj4[cj4_ind0].cj[0] == sci*c_numClPerSupercl)
165 /* we have the diagonal:
166 * add the charge self interaction energy term
168 for (im = 0; im < c_numClPerSupercl; im++)
170 ci = sci*c_numClPerSupercl + im;
171 for (ic = 0; ic < c_clSize; ic++)
173 ia = ci*c_clSize + ic;
174 iq = x[ia*nbat->xstride+3];
180 vctot *= -facel*0.5*iconst->c_rf;
184 /* last factor 1/sqrt(pi) */
185 vctot *= -facel*iconst->ewaldcoeff_q*M_1_SQRTPI;
189 for (cj4_ind = cj4_ind0; (cj4_ind < cj4_ind1); cj4_ind++)
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];
194 for (jm = 0; jm < c_nbnxnGpuJgroupSize; jm++)
196 cj = nbl->cj4[cj4_ind].cj[jm];
198 for (im = 0; im < c_numClPerSupercl; im++)
200 /* We're only using the first imask,
201 * but here imei[1].imask is identical.
203 if ((nbl->cj4[cj4_ind].imei[0].imask >> (jm*c_numClPerSupercl + im)) & 1)
205 gmx_bool within_rlist;
207 ci = sci*c_numClPerSupercl + im;
209 within_rlist = FALSE;
211 for (ic = 0; ic < c_clSize; ic++)
213 ia = ci*c_clSize + ic;
215 is = ia*nbat->xstride;
216 ifs = ia*nbat->fstride;
221 nti = ntype*2*type[ia];
227 for (jc = 0; jc < c_clSize; jc++)
229 ja = cj*c_clSize + jc;
231 if (nbln->shift == CENTRAL &&
232 ci == cj && ja <= ia)
237 constexpr int clusterPerSplit = c_nbnxnGpuClusterSize/c_nbnxnGpuClusterpairSplit;
238 int_bit = ((excl[jc/clusterPerSplit]->pair[(jc & (clusterPerSplit - 1))*c_clSize + ic]
239 >> (jm*c_numClPerSupercl + im)) & 1);
241 js = ja*nbat->xstride;
242 jfs = ja*nbat->fstride;
249 rsq = dx*dx + dy*dy + dz*dz;
259 if (type[ia] != ntype-1 && type[ja] != ntype-1)
264 // Ensure distance do not become so small that r^-12 overflows
265 rsq = std::max(rsq, NBNXN_MIN_RSQ);
267 rinv = gmx::invsqrt(rsq);
274 krsq = iconst->k_rf*rsq;
275 fscal = qq*(int_bit*rinv - 2*krsq)*rinvsq;
278 vcoul = qq*(int_bit*rinv + krsq - iconst->c_rf);
284 rt = r*iconst->tabq_scale;
285 n0 = static_cast<int>(rt);
288 fexcl = (1 - eps)*Ftab[n0] + eps*Ftab[n0+1];
290 fscal = qq*(int_bit*rinvsq - fexcl)*rinv;
294 vcoul = qq*((int_bit - std::erf(iconst->ewaldcoeff_q*r))*rinv - int_bit*iconst->sh_ewald);
300 tj = nti + 2*type[ja];
302 /* Vanilla Lennard-Jones cutoff */
304 c12 = vdwparam[tj+1];
306 rinvsix = int_bit*rinvsq*rinvsq*rinvsq;
307 Vvdw_disp = c6*rinvsix;
308 Vvdw_rep = c12*rinvsix*rinvsix;
309 fscal += (Vvdw_rep - Vvdw_disp)*rinvsq;
316 (Vvdw_rep - int_bit*c12*iconst->sh_invrc6*iconst->sh_invrc6)/12 -
317 (Vvdw_disp - int_bit*c6*iconst->sh_invrc6)/6;
335 fshift[ish3] = fshift[ish3] + fix;
336 fshift[ish3+1] = fshift[ish3+1] + fiy;
337 fshift[ish3+2] = fshift[ish3+2] + fiz;
339 /* Count in half work-units.
340 * In CUDA one work-unit is 2 warps.
342 if ((ic+1) % (c_clSize/c_nbnxnGpuClusterpairSplit) == 0)
352 within_rlist = FALSE;
364 Vc[ggid] = Vc[ggid] + vctot;
365 Vvdw[ggid] = Vvdw[ggid] + Vvdwtot;
371 fprintf(debug, "number of half %dx%d atom pairs: %d after pruning: %d fraction %4.2f\n",
372 nbl->na_ci, nbl->na_ci,
373 nhwu, nhwu_pruned, nhwu_pruned/static_cast<double>(nhwu));
374 fprintf(debug, "generic kernel pair interactions: %d\n",
375 nhwu*nbl->na_ci/2*nbl->na_ci);
376 fprintf(debug, "generic kernel post-prune pair interactions: %d\n",
377 nhwu_pruned*nbl->na_ci/2*nbl->na_ci);
378 fprintf(debug, "generic kernel non-zero pair interactions: %d\n",
380 fprintf(debug, "ratio non-zero/post-prune pair interactions: %4.2f\n",
381 npair_tot/static_cast<double>(nhwu_pruned*gmx::exactDiv(nbl->na_ci, 2)*nbl->na_ci));