src/gromacs/gmxlib/nonbonded/nb_generic.cpp

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
   5  * Copyright (c) 2001-2004, The GROMACS development team.
   6  * Copyright (c) 2012,2014,2015,2017,2018, by the GROMACS development team, led by
   7  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   8  * and including many others, as listed in the AUTHORS file in the
   9  * top-level source directory and at http://www.gromacs.org.
  10  *
  11  * GROMACS is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public License
  13  * as published by the Free Software Foundation; either version 2.1
  14  * of the License, or (at your option) any later version.
  15  *
  16  * GROMACS is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with GROMACS; if not, see
  23  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  24  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  25  *
  26  * If you want to redistribute modifications to GROMACS, please
  27  * consider that scientific software is very special. Version
  28  * control is crucial - bugs must be traceable. We will be happy to
  29  * consider code for inclusion in the official distribution, but
  30  * derived work must not be called official GROMACS. Details are found
  31  * in the README & COPYING files - if they are missing, get the
  32  * official version at http://www.gromacs.org.
  33  *
  34  * To help us fund GROMACS development, we humbly ask that you cite
  35  * the research papers on the package. Check out http://www.gromacs.org.
  36  */
  37 #include "gmxpre.h"
  38
  39 #include "nb_generic.h"
  40
  41 #include <cmath>
  42
  43 #include "gromacs/gmxlib/nrnb.h"
  44 #include "gromacs/gmxlib/nonbonded/nb_kernel.h"
  45 #include "gromacs/gmxlib/nonbonded/nonbonded.h"
  46 #include "gromacs/math/functions.h"
  47 #include "gromacs/math/vec.h"
  48 #include "gromacs/mdtypes/md_enums.h"
  49 #include "gromacs/utility/fatalerror.h"
  50
  51 void
  52 gmx_nb_generic_kernel(t_nblist *                nlist,
  53                       rvec *                    xx,
  54                       rvec *                    ff,
  55                       t_forcerec *              fr,
  56                       t_mdatoms *               mdatoms,
  57                       nb_kernel_data_t *        kernel_data,
  58                       t_nrnb *                  nrnb)
  59 {
  60     int           ntype, table_nelements, ielec, ivdw;
  61     real          facel;
  62     int           n, ii, is3, ii3, k, nj0, nj1, jnr, j3, ggid, nnn, n0;
  63     real          shX, shY, shZ;
  64     real          fscal, felec, fvdw, velec, vvdw, tx, ty, tz;
  65     real          rinvsq;
  66     real          iq;
  67     real          qq, vctot;
  68     int           nti, nvdwparam;
  69     int           tj;
  70     real          rt, r, eps, eps2, Y, F, Geps, Heps2, VV, FF, Fp, fijD, fijR;
  71     real          rinvsix;
  72     real          vvdwtot;
  73     real          vvdw_rep, vvdw_disp;
  74     real          ix, iy, iz, fix, fiy, fiz;
  75     real          jx, jy, jz;
  76     real          dx, dy, dz, rsq, rinv;
  77     real          c6, c12, c6grid, cexp1, cexp2, br;
  78     real *        charge;
  79     real *        shiftvec;
  80     real *        vdwparam, *vdwgridparam;
  81     int *         type;
  82     real *        fshift;
  83     real *        velecgrp;
  84     real *        vvdwgrp;
  85     real          tabscale;
  86     real *        VFtab;
  87     real *        x;
  88     real *        f;
  89     int           ewitab;
  90     real          ewtabscale, eweps, ewrt, ewtabhalfspace;
  91     real *        ewtab;
  92     real          rcoulomb2, rvdw, rvdw2, sh_dispersion, sh_repulsion;
  93     real          rcutoff, rcutoff2;
  94     real          d, d2, sw, dsw, rinvcorr;
  95     real          elec_swV3, elec_swV4, elec_swV5, elec_swF2, elec_swF3, elec_swF4;
  96     real          vdw_swV3, vdw_swV4, vdw_swV5, vdw_swF2, vdw_swF3, vdw_swF4;
  97     real          ewclj, ewclj2, ewclj6, ewcljrsq, poly, exponent, sh_lj_ewald;
  98     gmx_bool      bExactElecCutoff, bExactVdwCutoff, bExactCutoff;
  99     gmx_bool      do_tab;
 100
 101     x                   = xx[0];
 102     f                   = ff[0];
 103     ielec               = nlist->ielec;
 104     ivdw                = nlist->ivdw;
 105
 106     fshift              = fr->fshift[0];
 107     velecgrp            = kernel_data->energygrp_elec;
 108     vvdwgrp             = kernel_data->energygrp_vdw;
 109
 110     do_tab = (ielec == GMX_NBKERNEL_ELEC_CUBICSPLINETABLE ||
 111               ivdw == GMX_NBKERNEL_VDW_CUBICSPLINETABLE);
 112     if (do_tab)
 113     {
 114         tabscale         = kernel_data->table_elec_vdw->scale;
 115         VFtab            = kernel_data->table_elec_vdw->data;
 116     }
 117     else
 118     {
 119         tabscale        = 0;
 120         VFtab           = nullptr;
 121     }
 122
 123     const interaction_const_t *ic = fr->ic;
 124
 125     ewtab               = ic->tabq_coul_FDV0;
 126     ewtabscale          = ic->tabq_scale;
 127     ewtabhalfspace      = 0.5/ewtabscale;
 128
 129     rcoulomb2           = ic->rcoulomb*ic->rcoulomb;
 130     rvdw                = ic->rvdw;
 131     rvdw2               = rvdw*rvdw;
 132     sh_dispersion       = ic->dispersion_shift.cpot;
 133     sh_repulsion        = ic->repulsion_shift.cpot;
 134     sh_lj_ewald         = ic->sh_lj_ewald;
 135
 136     ewclj               = ic->ewaldcoeff_lj;
 137     ewclj2              = ewclj*ewclj;
 138     ewclj6              = ewclj2*ewclj2*ewclj2;
 139
 140     if (ic->coulomb_modifier == eintmodPOTSWITCH)
 141     {
 142         d               = ic->rcoulomb - ic->rcoulomb_switch;
 143         elec_swV3       = -10.0/(d*d*d);
 144         elec_swV4       =  15.0/(d*d*d*d);
 145         elec_swV5       =  -6.0/(d*d*d*d*d);
 146         elec_swF2       = -30.0/(d*d*d);
 147         elec_swF3       =  60.0/(d*d*d*d);
 148         elec_swF4       = -30.0/(d*d*d*d*d);
 149     }
 150     else
 151     {
 152         /* Avoid warnings from stupid compilers (looking at you, Clang!) */
 153         elec_swV3 = elec_swV4 = elec_swV5 = elec_swF2 = elec_swF3 = elec_swF4 = 0.0;
 154     }
 155     if (ic->vdw_modifier == eintmodPOTSWITCH)
 156     {
 157         d               = ic->rvdw - ic->rvdw_switch;
 158         vdw_swV3        = -10.0/(d*d*d);
 159         vdw_swV4        =  15.0/(d*d*d*d);
 160         vdw_swV5        =  -6.0/(d*d*d*d*d);
 161         vdw_swF2        = -30.0/(d*d*d);
 162         vdw_swF3        =  60.0/(d*d*d*d);
 163         vdw_swF4        = -30.0/(d*d*d*d*d);
 164     }
 165     else
 166     {
 167         /* Avoid warnings from stupid compilers (looking at you, Clang!) */
 168         vdw_swV3 = vdw_swV4 = vdw_swV5 = vdw_swF2 = vdw_swF3 = vdw_swF4 = 0.0;
 169     }
 170
 171     bExactElecCutoff    = (ic->coulomb_modifier != eintmodNONE) || ic->eeltype == eelRF_ZERO;
 172     bExactVdwCutoff     = (ic->vdw_modifier != eintmodNONE);
 173     bExactCutoff        = bExactElecCutoff && bExactVdwCutoff;
 174
 175     if (bExactCutoff)
 176     {
 177         rcutoff  = ( ic->rcoulomb > ic->rvdw ) ? ic->rcoulomb : ic->rvdw;
 178         rcutoff2 = rcutoff*rcutoff;
 179     }
 180     else
 181     {
 182         /* Fix warnings for stupid compilers */
 183         rcutoff2 = 1e30;
 184     }
 185
 186     /* avoid compiler warnings for cases that cannot happen */
 187     nnn                 = 0;
 188     eps                 = 0.0;
 189     eps2                = 0.0;
 190
 191     /* 3 VdW parameters for Buckingham, otherwise 2 */
 192     nvdwparam           = (ivdw == GMX_NBKERNEL_VDW_BUCKINGHAM) ? 3 : 2;
 193     table_nelements     = 12;
 194
 195     charge              = mdatoms->chargeA;
 196     type                = mdatoms->typeA;
 197     facel               = fr->ic->epsfac;
 198     shiftvec            = fr->shift_vec[0];
 199     vdwparam            = fr->nbfp;
 200     ntype               = fr->ntype;
 201     vdwgridparam        = fr->ljpme_c6grid;
 202
 203     for (n = 0; (n < nlist->nri); n++)
 204     {
 205         is3              = 3*nlist->shift[n];
 206         shX              = shiftvec[is3];
 207         shY              = shiftvec[is3+1];
 208         shZ              = shiftvec[is3+2];
 209         nj0              = nlist->jindex[n];
 210         nj1              = nlist->jindex[n+1];
 211         ii               = nlist->iinr[n];
 212         ii3              = 3*ii;
 213         ix               = shX + x[ii3+0];
 214         iy               = shY + x[ii3+1];
 215         iz               = shZ + x[ii3+2];
 216         iq               = facel*charge[ii];
 217         nti              = nvdwparam*ntype*type[ii];
 218         vctot            = 0;
 219         vvdwtot          = 0;
 220         fix              = 0;
 221         fiy              = 0;
 222         fiz              = 0;
 223
 224         for (k = nj0; (k < nj1); k++)
 225         {
 226             jnr              = nlist->jjnr[k];
 227             j3               = 3*jnr;
 228             jx               = x[j3+0];
 229             jy               = x[j3+1];
 230             jz               = x[j3+2];
 231             dx               = ix - jx;
 232             dy               = iy - jy;
 233             dz               = iz - jz;
 234             rsq              = dx*dx+dy*dy+dz*dz;
 235             rinv             = gmx::invsqrt(rsq);
 236             rinvsq           = rinv*rinv;
 237             felec            = 0;
 238             fvdw             = 0;
 239             velec            = 0;
 240             vvdw             = 0;
 241
 242             if (bExactCutoff && rsq >= rcutoff2)
 243             {
 244                 continue;
 245             }
 246
 247             if (ielec == GMX_NBKERNEL_ELEC_CUBICSPLINETABLE || ivdw == GMX_NBKERNEL_VDW_CUBICSPLINETABLE)
 248             {
 249                 r                = rsq*rinv;
 250                 rt               = r*tabscale;
 251                 n0               = rt;
 252                 eps              = rt-n0;
 253                 eps2             = eps*eps;
 254                 nnn              = table_nelements*n0;
 255             }
 256
 257             /* Coulomb interaction. ielec==0 means no interaction */
 258             if (ielec != GMX_NBKERNEL_ELEC_NONE)
 259             {
 260                 qq               = iq*charge[jnr];
 261
 262                 switch (ielec)
 263                 {
 264                     case GMX_NBKERNEL_ELEC_NONE:
 265                         break;
 266
 267                     case GMX_NBKERNEL_ELEC_COULOMB:
 268                         /* Vanilla cutoff coulomb */
 269                         velec            = qq*rinv;
 270                         felec            = velec*rinvsq;
 271                         /* The shift for the Coulomb potential is stored in
 272                          * the RF parameter c_rf, which is 0 without shift
 273                          */
 274                         velec           -= qq*ic->c_rf;
 275                         break;
 276
 277                     case GMX_NBKERNEL_ELEC_REACTIONFIELD:
 278                         /* Reaction-field */
 279                         velec            = qq*(rinv + ic->k_rf*rsq-ic->c_rf);
 280                         felec            = qq*(rinv*rinvsq - 2.0*ic->k_rf);
 281                         break;
 282
 283                     case GMX_NBKERNEL_ELEC_CUBICSPLINETABLE:
 284                         /* Tabulated coulomb */
 285                         Y                = VFtab[nnn];
 286                         F                = VFtab[nnn+1];
 287                         Geps             = eps*VFtab[nnn+2];
 288                         Heps2            = eps2*VFtab[nnn+3];
 289                         Fp               = F+Geps+Heps2;
 290                         VV               = Y+eps*Fp;
 291                         FF               = Fp+Geps+2.0*Heps2;
 292                         velec            = qq*VV;
 293                         felec            = -qq*FF*tabscale*rinv;
 294                         break;
 295
 296                     case GMX_NBKERNEL_ELEC_EWALD:
 297                         ewrt             = rsq*rinv*ewtabscale;
 298                         ewitab           = ewrt;
 299                         eweps            = ewrt-ewitab;
 300                         ewitab           = 4*ewitab;
 301                         felec            = ewtab[ewitab]+eweps*ewtab[ewitab+1];
 302                         rinvcorr         = (ic->coulomb_modifier == eintmodPOTSHIFT) ? rinv - ic->sh_ewald : rinv;
 303                         velec            = qq*(rinvcorr-(ewtab[ewitab+2]-ewtabhalfspace*eweps*(ewtab[ewitab]+felec)));
 304                         felec            = qq*rinv*(rinvsq-felec);
 305                         break;
 306
 307                     default:
 308                         gmx_fatal(FARGS, "Death & horror! No generic coulomb interaction for ielec=%d.\n", ielec);
 309                         break;
 310                 }
 311                 if (ic->coulomb_modifier == eintmodPOTSWITCH)
 312                 {
 313                     d                = rsq*rinv - ic->rcoulomb_switch;
 314                     d                = (d > 0.0) ? d : 0.0;
 315                     d2               = d*d;
 316                     sw               = 1.0+d2*d*(elec_swV3+d*(elec_swV4+d*elec_swV5));
 317                     dsw              = d2*(elec_swF2+d*(elec_swF3+d*elec_swF4));
 318                     /* Apply switch function. Note that felec=f/r since it will be multiplied
 319                      * by the i-j displacement vector. This means felec'=f'/r=-(v*sw)'/r=
 320                      * -(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=felec*sw-v*dsw/r
 321                      */
 322                     felec            = felec*sw - rinv*velec*dsw;
 323                     /* Once we have used velec to update felec we can modify velec too */
 324                     velec           *= sw;
 325                 }
 326                 if (bExactElecCutoff)
 327                 {
 328                     felec            = (rsq < rcoulomb2) ? felec : 0.0;
 329                     velec            = (rsq < rcoulomb2) ? velec : 0.0;
 330                 }
 331                 vctot           += velec;
 332             } /* End of coulomb interactions */
 333
 334
 335             /* VdW interaction. ivdw==0 means no interaction */
 336             if (ivdw != GMX_NBKERNEL_VDW_NONE)
 337             {
 338                 tj               = nti+nvdwparam*type[jnr];
 339
 340                 switch (ivdw)
 341                 {
 342                     case GMX_NBKERNEL_VDW_NONE:
 343                         break;
 344
 345                     case GMX_NBKERNEL_VDW_LENNARDJONES:
 346                         /* Vanilla Lennard-Jones cutoff */
 347                         c6               = vdwparam[tj];
 348                         c12              = vdwparam[tj+1];
 349                         rinvsix          = rinvsq*rinvsq*rinvsq;
 350                         vvdw_disp        = c6*rinvsix;
 351                         vvdw_rep         = c12*rinvsix*rinvsix;
 352                         fvdw             = (vvdw_rep-vvdw_disp)*rinvsq;
 353                         if (ic->vdw_modifier == eintmodPOTSHIFT)
 354                         {
 355                             vvdw             = (vvdw_rep + c12*sh_repulsion)/12.0 - (vvdw_disp + c6*sh_dispersion)/6.0;
 356                         }
 357                         else
 358                         {
 359                             vvdw             = vvdw_rep/12.0-vvdw_disp/6.0;
 360                         }
 361                         break;
 362
 363                     case GMX_NBKERNEL_VDW_BUCKINGHAM:
 364                         /* Buckingham */
 365                         c6               = vdwparam[tj];
 366                         cexp1            = vdwparam[tj+1];
 367                         cexp2            = vdwparam[tj+2];
 368
 369                         rinvsix          = rinvsq*rinvsq*rinvsq;
 370                         vvdw_disp        = c6*rinvsix;
 371                         br               = cexp2*rsq*rinv;
 372                         vvdw_rep         = cexp1*std::exp(-br);
 373                         fvdw             = (br*vvdw_rep-vvdw_disp)*rinvsq;
 374                         if (ic->vdw_modifier == eintmodPOTSHIFT)
 375                         {
 376                             vvdw             = (vvdw_rep-cexp1*std::exp(-cexp2*rvdw))-(vvdw_disp + c6*sh_dispersion)/6.0;
 377                         }
 378                         else
 379                         {
 380                             vvdw             = vvdw_rep-vvdw_disp/6.0;
 381                         }
 382                         break;
 383
 384                     case GMX_NBKERNEL_VDW_CUBICSPLINETABLE:
 385                         /* Tabulated VdW */
 386                         c6               = vdwparam[tj];
 387                         c12              = vdwparam[tj+1];
 388                         Y                = VFtab[nnn+4];
 389                         F                = VFtab[nnn+5];
 390                         Geps             = eps*VFtab[nnn+6];
 391                         Heps2            = eps2*VFtab[nnn+7];
 392                         Fp               = F+Geps+Heps2;
 393                         VV               = Y+eps*Fp;
 394                         FF               = Fp+Geps+2.0*Heps2;
 395                         vvdw_disp        = c6*VV;
 396                         fijD             = c6*FF;
 397                         Y                = VFtab[nnn+8];
 398                         F                = VFtab[nnn+9];
 399                         Geps             = eps*VFtab[nnn+10];
 400                         Heps2            = eps2*VFtab[nnn+11];
 401                         Fp               = F+Geps+Heps2;
 402                         VV               = Y+eps*Fp;
 403                         FF               = Fp+Geps+2.0*Heps2;
 404                         vvdw_rep         = c12*VV;
 405                         fijR             = c12*FF;
 406                         fvdw             = -(fijD+fijR)*tabscale*rinv;
 407                         vvdw             = vvdw_disp + vvdw_rep;
 408                         break;
 409
 410
 411                     case GMX_NBKERNEL_VDW_LJEWALD:
 412                         /* LJ-PME */
 413                         rinvsix          = rinvsq*rinvsq*rinvsq;
 414                         ewcljrsq         = ewclj2*rsq;
 415                         exponent         = std::exp(-ewcljrsq);
 416                         poly             = exponent*(1.0 + ewcljrsq + ewcljrsq*ewcljrsq*0.5);
 417                         c6               = vdwparam[tj];
 418                         c12              = vdwparam[tj+1];
 419                         c6grid           = vdwgridparam[tj];
 420                         vvdw_disp        = (c6-c6grid*(1.0-poly))*rinvsix;
 421                         vvdw_rep         = c12*rinvsix*rinvsix;
 422                         fvdw             = (vvdw_rep - vvdw_disp - c6grid*(1.0/6.0)*exponent*ewclj6)*rinvsq;
 423                         if (ic->vdw_modifier == eintmodPOTSHIFT)
 424                         {
 425                             vvdw             = (vvdw_rep + c12*sh_repulsion)/12.0 - (vvdw_disp + c6*sh_dispersion - c6grid*sh_lj_ewald)/6.0;
 426                         }
 427                         else
 428                         {
 429                             vvdw             = vvdw_rep/12.0-vvdw_disp/6.0;
 430                         }
 431                         break;
 432
 433                     default:
 434                         gmx_fatal(FARGS, "Death & horror! No generic VdW interaction for ivdw=%d.\n", ivdw);
 435                         break;
 436                 }
 437                 if (ic->vdw_modifier == eintmodPOTSWITCH)
 438                 {
 439                     d                = rsq*rinv - ic->rvdw_switch;
 440                     d                = (d > 0.0) ? d : 0.0;
 441                     d2               = d*d;
 442                     sw               = 1.0+d2*d*(vdw_swV3+d*(vdw_swV4+d*vdw_swV5));
 443                     dsw              = d2*(vdw_swF2+d*(vdw_swF3+d*vdw_swF4));
 444                     /* See coulomb interaction for the force-switch formula */
 445                     fvdw             = fvdw*sw - rinv*vvdw*dsw;
 446                     vvdw            *= sw;
 447                 }
 448                 if (bExactVdwCutoff)
 449                 {
 450                     fvdw             = (rsq < rvdw2) ? fvdw : 0.0;
 451                     vvdw             = (rsq < rvdw2) ? vvdw : 0.0;
 452                 }
 453                 vvdwtot         += vvdw;
 454             } /* end VdW interactions */
 455
 456             fscal            = felec+fvdw;
 457
 458             tx               = fscal*dx;
 459             ty               = fscal*dy;
 460             tz               = fscal*dz;
 461             fix              = fix + tx;
 462             fiy              = fiy + ty;
 463             fiz              = fiz + tz;
 464             f[j3+0]          = f[j3+0] - tx;
 465             f[j3+1]          = f[j3+1] - ty;
 466             f[j3+2]          = f[j3+2] - tz;
 467         }
 468
 469         f[ii3+0]         = f[ii3+0] + fix;
 470         f[ii3+1]         = f[ii3+1] + fiy;
 471         f[ii3+2]         = f[ii3+2] + fiz;
 472         fshift[is3]      = fshift[is3]+fix;
 473         fshift[is3+1]    = fshift[is3+1]+fiy;
 474         fshift[is3+2]    = fshift[is3+2]+fiz;
 475         ggid             = nlist->gid[n];
 476         velecgrp[ggid]  += vctot;
 477         vvdwgrp[ggid]   += vvdwtot;
 478     }
 479     /* Estimate flops, average for generic kernel:
 480      * 12 flops per outer iteration
 481      * 50 flops per inner iteration
 482      */
 483     inc_nrnb(nrnb, eNR_NBKERNEL_GENERIC, nlist->nri*12 + nlist->jindex[n]*50);
 484 }