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     if (ielec == GMX_NBKERNEL_ELEC_EWALD)
 126     {
 127         ewtab               = ic->tabq_coul_FDV0;
 128         ewtabscale          = ic->tabq_scale;
 129         ewtabhalfspace      = 0.5/ewtabscale;
 130     }
 131     else
 132     {
 133         ewtab          = nullptr;
 134         ewtabhalfspace = ewtabscale = 0;
 135     }
 136
 137     rcoulomb2           = ic->rcoulomb*ic->rcoulomb;
 138     rvdw                = ic->rvdw;
 139     rvdw2               = rvdw*rvdw;
 140     sh_dispersion       = ic->dispersion_shift.cpot;
 141     sh_repulsion        = ic->repulsion_shift.cpot;
 142     sh_lj_ewald         = ic->sh_lj_ewald;
 143
 144     ewclj               = ic->ewaldcoeff_lj;
 145     ewclj2              = ewclj*ewclj;
 146     ewclj6              = ewclj2*ewclj2*ewclj2;
 147
 148     if (ic->coulomb_modifier == eintmodPOTSWITCH)
 149     {
 150         d               = ic->rcoulomb - ic->rcoulomb_switch;
 151         elec_swV3       = -10.0/(d*d*d);
 152         elec_swV4       =  15.0/(d*d*d*d);
 153         elec_swV5       =  -6.0/(d*d*d*d*d);
 154         elec_swF2       = -30.0/(d*d*d);
 155         elec_swF3       =  60.0/(d*d*d*d);
 156         elec_swF4       = -30.0/(d*d*d*d*d);
 157     }
 158     else
 159     {
 160         /* Avoid warnings from stupid compilers (looking at you, Clang!) */
 161         elec_swV3 = elec_swV4 = elec_swV5 = elec_swF2 = elec_swF3 = elec_swF4 = 0.0;
 162     }
 163     if (ic->vdw_modifier == eintmodPOTSWITCH)
 164     {
 165         d               = ic->rvdw - ic->rvdw_switch;
 166         vdw_swV3        = -10.0/(d*d*d);
 167         vdw_swV4        =  15.0/(d*d*d*d);
 168         vdw_swV5        =  -6.0/(d*d*d*d*d);
 169         vdw_swF2        = -30.0/(d*d*d);
 170         vdw_swF3        =  60.0/(d*d*d*d);
 171         vdw_swF4        = -30.0/(d*d*d*d*d);
 172     }
 173     else
 174     {
 175         /* Avoid warnings from stupid compilers (looking at you, Clang!) */
 176         vdw_swV3 = vdw_swV4 = vdw_swV5 = vdw_swF2 = vdw_swF3 = vdw_swF4 = 0.0;
 177     }
 178
 179     bExactElecCutoff    = (ic->coulomb_modifier != eintmodNONE) || ic->eeltype == eelRF_ZERO;
 180     bExactVdwCutoff     = (ic->vdw_modifier != eintmodNONE);
 181     bExactCutoff        = bExactElecCutoff && bExactVdwCutoff;
 182
 183     if (bExactCutoff)
 184     {
 185         rcutoff  = ( ic->rcoulomb > ic->rvdw ) ? ic->rcoulomb : ic->rvdw;
 186         rcutoff2 = rcutoff*rcutoff;
 187     }
 188     else
 189     {
 190         /* Fix warnings for stupid compilers */
 191         rcutoff2 = 1e30;
 192     }
 193
 194     /* avoid compiler warnings for cases that cannot happen */
 195     nnn                 = 0;
 196     eps                 = 0.0;
 197     eps2                = 0.0;
 198
 199     /* 3 VdW parameters for Buckingham, otherwise 2 */
 200     nvdwparam           = (ivdw == GMX_NBKERNEL_VDW_BUCKINGHAM) ? 3 : 2;
 201     table_nelements     = 12;
 202
 203     charge              = mdatoms->chargeA;
 204     type                = mdatoms->typeA;
 205     facel               = fr->ic->epsfac;
 206     shiftvec            = fr->shift_vec[0];
 207     vdwparam            = fr->nbfp;
 208     ntype               = fr->ntype;
 209     vdwgridparam        = fr->ljpme_c6grid;
 210
 211     for (n = 0; (n < nlist->nri); n++)
 212     {
 213         is3              = 3*nlist->shift[n];
 214         shX              = shiftvec[is3];
 215         shY              = shiftvec[is3+1];
 216         shZ              = shiftvec[is3+2];
 217         nj0              = nlist->jindex[n];
 218         nj1              = nlist->jindex[n+1];
 219         ii               = nlist->iinr[n];
 220         ii3              = 3*ii;
 221         ix               = shX + x[ii3+0];
 222         iy               = shY + x[ii3+1];
 223         iz               = shZ + x[ii3+2];
 224         iq               = facel*charge[ii];
 225         nti              = nvdwparam*ntype*type[ii];
 226         vctot            = 0;
 227         vvdwtot          = 0;
 228         fix              = 0;
 229         fiy              = 0;
 230         fiz              = 0;
 231
 232         for (k = nj0; (k < nj1); k++)
 233         {
 234             jnr              = nlist->jjnr[k];
 235             j3               = 3*jnr;
 236             jx               = x[j3+0];
 237             jy               = x[j3+1];
 238             jz               = x[j3+2];
 239             dx               = ix - jx;
 240             dy               = iy - jy;
 241             dz               = iz - jz;
 242             rsq              = dx*dx+dy*dy+dz*dz;
 243             rinv             = gmx::invsqrt(rsq);
 244             rinvsq           = rinv*rinv;
 245             felec            = 0;
 246             fvdw             = 0;
 247             velec            = 0;
 248             vvdw             = 0;
 249
 250             if (bExactCutoff && rsq >= rcutoff2)
 251             {
 252                 continue;
 253             }
 254
 255             if (ielec == GMX_NBKERNEL_ELEC_CUBICSPLINETABLE || ivdw == GMX_NBKERNEL_VDW_CUBICSPLINETABLE)
 256             {
 257                 r                = rsq*rinv;
 258                 rt               = r*tabscale;
 259                 n0               = rt;
 260                 eps              = rt-n0;
 261                 eps2             = eps*eps;
 262                 nnn              = table_nelements*n0;
 263             }
 264
 265             /* Coulomb interaction. ielec==0 means no interaction */
 266             if (ielec != GMX_NBKERNEL_ELEC_NONE)
 267             {
 268                 qq               = iq*charge[jnr];
 269
 270                 switch (ielec)
 271                 {
 272                     case GMX_NBKERNEL_ELEC_NONE:
 273                         break;
 274
 275                     case GMX_NBKERNEL_ELEC_COULOMB:
 276                         /* Vanilla cutoff coulomb */
 277                         velec            = qq*rinv;
 278                         felec            = velec*rinvsq;
 279                         /* The shift for the Coulomb potential is stored in
 280                          * the RF parameter c_rf, which is 0 without shift
 281                          */
 282                         velec           -= qq*ic->c_rf;
 283                         break;
 284
 285                     case GMX_NBKERNEL_ELEC_REACTIONFIELD:
 286                         /* Reaction-field */
 287                         velec            = qq*(rinv + ic->k_rf*rsq-ic->c_rf);
 288                         felec            = qq*(rinv*rinvsq - 2.0*ic->k_rf);
 289                         break;
 290
 291                     case GMX_NBKERNEL_ELEC_CUBICSPLINETABLE:
 292                         /* Tabulated coulomb */
 293                         Y                = VFtab[nnn];
 294                         F                = VFtab[nnn+1];
 295                         Geps             = eps*VFtab[nnn+2];
 296                         Heps2            = eps2*VFtab[nnn+3];
 297                         Fp               = F+Geps+Heps2;
 298                         VV               = Y+eps*Fp;
 299                         FF               = Fp+Geps+2.0*Heps2;
 300                         velec            = qq*VV;
 301                         felec            = -qq*FF*tabscale*rinv;
 302                         break;
 303
 304                     case GMX_NBKERNEL_ELEC_EWALD:
 305                         ewrt             = rsq*rinv*ewtabscale;
 306                         ewitab           = ewrt;
 307                         eweps            = ewrt-ewitab;
 308                         ewitab           = 4*ewitab;
 309                         felec            = ewtab[ewitab]+eweps*ewtab[ewitab+1];
 310                         rinvcorr         = (ic->coulomb_modifier == eintmodPOTSHIFT) ? rinv - ic->sh_ewald : rinv;
 311                         velec            = qq*(rinvcorr-(ewtab[ewitab+2]-ewtabhalfspace*eweps*(ewtab[ewitab]+felec)));
 312                         felec            = qq*rinv*(rinvsq-felec);
 313                         break;
 314
 315                     default:
 316                         gmx_fatal(FARGS, "Death & horror! No generic coulomb interaction for ielec=%d.\n", ielec);
 317                 }
 318                 if (ic->coulomb_modifier == eintmodPOTSWITCH)
 319                 {
 320                     d                = rsq*rinv - ic->rcoulomb_switch;
 321                     d                = (d > 0.0) ? d : 0.0;
 322                     d2               = d*d;
 323                     sw               = 1.0+d2*d*(elec_swV3+d*(elec_swV4+d*elec_swV5));
 324                     dsw              = d2*(elec_swF2+d*(elec_swF3+d*elec_swF4));
 325                     /* Apply switch function. Note that felec=f/r since it will be multiplied
 326                      * by the i-j displacement vector. This means felec'=f'/r=-(v*sw)'/r=
 327                      * -(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=felec*sw-v*dsw/r
 328                      */
 329                     felec            = felec*sw - rinv*velec*dsw;
 330                     /* Once we have used velec to update felec we can modify velec too */
 331                     velec           *= sw;
 332                 }
 333                 if (bExactElecCutoff)
 334                 {
 335                     felec            = (rsq < rcoulomb2) ? felec : 0.0;
 336                     velec            = (rsq < rcoulomb2) ? velec : 0.0;
 337                 }
 338                 vctot           += velec;
 339             } /* End of coulomb interactions */
 340
 341
 342             /* VdW interaction. ivdw==0 means no interaction */
 343             if (ivdw != GMX_NBKERNEL_VDW_NONE)
 344             {
 345                 tj               = nti+nvdwparam*type[jnr];
 346
 347                 switch (ivdw)
 348                 {
 349                     case GMX_NBKERNEL_VDW_NONE:
 350                         break;
 351
 352                     case GMX_NBKERNEL_VDW_LENNARDJONES:
 353                         /* Vanilla Lennard-Jones cutoff */
 354                         c6               = vdwparam[tj];
 355                         c12              = vdwparam[tj+1];
 356                         rinvsix          = rinvsq*rinvsq*rinvsq;
 357                         vvdw_disp        = c6*rinvsix;
 358                         vvdw_rep         = c12*rinvsix*rinvsix;
 359                         fvdw             = (vvdw_rep-vvdw_disp)*rinvsq;
 360                         if (ic->vdw_modifier == eintmodPOTSHIFT)
 361                         {
 362                             vvdw             = (vvdw_rep + c12*sh_repulsion)/12.0 - (vvdw_disp + c6*sh_dispersion)/6.0;
 363                         }
 364                         else
 365                         {
 366                             vvdw             = vvdw_rep/12.0-vvdw_disp/6.0;
 367                         }
 368                         break;
 369
 370                     case GMX_NBKERNEL_VDW_BUCKINGHAM:
 371                         /* Buckingham */
 372                         c6               = vdwparam[tj];
 373                         cexp1            = vdwparam[tj+1];
 374                         cexp2            = vdwparam[tj+2];
 375
 376                         rinvsix          = rinvsq*rinvsq*rinvsq;
 377                         vvdw_disp        = c6*rinvsix;
 378                         br               = cexp2*rsq*rinv;
 379                         vvdw_rep         = cexp1*std::exp(-br);
 380                         fvdw             = (br*vvdw_rep-vvdw_disp)*rinvsq;
 381                         if (ic->vdw_modifier == eintmodPOTSHIFT)
 382                         {
 383                             vvdw             = (vvdw_rep-cexp1*std::exp(-cexp2*rvdw))-(vvdw_disp + c6*sh_dispersion)/6.0;
 384                         }
 385                         else
 386                         {
 387                             vvdw             = vvdw_rep-vvdw_disp/6.0;
 388                         }
 389                         break;
 390
 391                     case GMX_NBKERNEL_VDW_CUBICSPLINETABLE:
 392                         /* Tabulated VdW */
 393                         c6               = vdwparam[tj];
 394                         c12              = vdwparam[tj+1];
 395                         Y                = VFtab[nnn+4];
 396                         F                = VFtab[nnn+5];
 397                         Geps             = eps*VFtab[nnn+6];
 398                         Heps2            = eps2*VFtab[nnn+7];
 399                         Fp               = F+Geps+Heps2;
 400                         VV               = Y+eps*Fp;
 401                         FF               = Fp+Geps+2.0*Heps2;
 402                         vvdw_disp        = c6*VV;
 403                         fijD             = c6*FF;
 404                         Y                = VFtab[nnn+8];
 405                         F                = VFtab[nnn+9];
 406                         Geps             = eps*VFtab[nnn+10];
 407                         Heps2            = eps2*VFtab[nnn+11];
 408                         Fp               = F+Geps+Heps2;
 409                         VV               = Y+eps*Fp;
 410                         FF               = Fp+Geps+2.0*Heps2;
 411                         vvdw_rep         = c12*VV;
 412                         fijR             = c12*FF;
 413                         fvdw             = -(fijD+fijR)*tabscale*rinv;
 414                         vvdw             = vvdw_disp + vvdw_rep;
 415                         break;
 416
 417
 418                     case GMX_NBKERNEL_VDW_LJEWALD:
 419                         /* LJ-PME */
 420                         rinvsix          = rinvsq*rinvsq*rinvsq;
 421                         ewcljrsq         = ewclj2*rsq;
 422                         exponent         = std::exp(-ewcljrsq);
 423                         poly             = exponent*(1.0 + ewcljrsq + ewcljrsq*ewcljrsq*0.5);
 424                         c6               = vdwparam[tj];
 425                         c12              = vdwparam[tj+1];
 426                         c6grid           = vdwgridparam[tj];
 427                         vvdw_disp        = (c6-c6grid*(1.0-poly))*rinvsix;
 428                         vvdw_rep         = c12*rinvsix*rinvsix;
 429                         fvdw             = (vvdw_rep - vvdw_disp - c6grid*(1.0/6.0)*exponent*ewclj6)*rinvsq;
 430                         if (ic->vdw_modifier == eintmodPOTSHIFT)
 431                         {
 432                             vvdw             = (vvdw_rep + c12*sh_repulsion)/12.0 - (vvdw_disp + c6*sh_dispersion - c6grid*sh_lj_ewald)/6.0;
 433                         }
 434                         else
 435                         {
 436                             vvdw             = vvdw_rep/12.0-vvdw_disp/6.0;
 437                         }
 438                         break;
 439
 440                     default:
 441                         gmx_fatal(FARGS, "Death & horror! No generic VdW interaction for ivdw=%d.\n", ivdw);
 442                 }
 443                 if (ic->vdw_modifier == eintmodPOTSWITCH)
 444                 {
 445                     d                = rsq*rinv - ic->rvdw_switch;
 446                     d                = (d > 0.0) ? d : 0.0;
 447                     d2               = d*d;
 448                     sw               = 1.0+d2*d*(vdw_swV3+d*(vdw_swV4+d*vdw_swV5));
 449                     dsw              = d2*(vdw_swF2+d*(vdw_swF3+d*vdw_swF4));
 450                     /* See coulomb interaction for the force-switch formula */
 451                     fvdw             = fvdw*sw - rinv*vvdw*dsw;
 452                     vvdw            *= sw;
 453                 }
 454                 if (bExactVdwCutoff)
 455                 {
 456                     fvdw             = (rsq < rvdw2) ? fvdw : 0.0;
 457                     vvdw             = (rsq < rvdw2) ? vvdw : 0.0;
 458                 }
 459                 vvdwtot         += vvdw;
 460             } /* end VdW interactions */
 461
 462             fscal            = felec+fvdw;
 463
 464             tx               = fscal*dx;
 465             ty               = fscal*dy;
 466             tz               = fscal*dz;
 467             fix              = fix + tx;
 468             fiy              = fiy + ty;
 469             fiz              = fiz + tz;
 470             f[j3+0]          = f[j3+0] - tx;
 471             f[j3+1]          = f[j3+1] - ty;
 472             f[j3+2]          = f[j3+2] - tz;
 473         }
 474
 475         f[ii3+0]         = f[ii3+0] + fix;
 476         f[ii3+1]         = f[ii3+1] + fiy;
 477         f[ii3+2]         = f[ii3+2] + fiz;
 478         fshift[is3]      = fshift[is3]+fix;
 479         fshift[is3+1]    = fshift[is3+1]+fiy;
 480         fshift[is3+2]    = fshift[is3+2]+fiz;
 481         ggid             = nlist->gid[n];
 482         velecgrp[ggid]  += vctot;
 483         vvdwgrp[ggid]   += vvdwtot;
 484     }
 485     /* Estimate flops, average for generic kernel:
 486      * 12 flops per outer iteration
 487      * 50 flops per inner iteration
 488      */
 489     inc_nrnb(nrnb, eNR_NBKERNEL_GENERIC, nlist->nri*12 + nlist->jindex[n]*50);
 490 }