src/gromacs/gmxlib/nonbonded/nb_generic.c

   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, 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 #ifdef HAVE_CONFIG_H
  38 #include <config.h>
  39 #endif
  40
  41 #include <math.h>
  42
  43 #include "types/simple.h"
  44 #include "vec.h"
  45 #include "typedefs.h"
  46 #include "nb_generic.h"
  47 #include "nrnb.h"
  48
  49 #include "nonbonded.h"
  50 #include "nb_kernel.h"
  51
  52
  53 void
  54 gmx_nb_generic_kernel(t_nblist *                nlist,
  55                       rvec *                    xx,
  56                       rvec *                    ff,
  57                       t_forcerec *              fr,
  58                       t_mdatoms *               mdatoms,
  59                       nb_kernel_data_t *        kernel_data,
  60                       t_nrnb *                  nrnb)
  61 {
  62     int           nri, ntype, table_nelements, ielec, ivdw;
  63     real          facel, gbtabscale;
  64     int           n, ii, is3, ii3, k, nj0, nj1, jnr, j3, ggid, nnn, n0;
  65     real          shX, shY, shZ;
  66     real          fscal, felec, fvdw, velec, vvdw, tx, ty, tz;
  67     real          rinvsq;
  68     real          iq;
  69     real          qq, vctot;
  70     int           nti, nvdwparam;
  71     int           tj;
  72     real          rt, r, eps, eps2, Y, F, Geps, Heps2, VV, FF, Fp, fijD, fijR;
  73     real          rinvsix;
  74     real          vvdwtot;
  75     real          vvdw_rep, vvdw_disp;
  76     real          ix, iy, iz, fix, fiy, fiz;
  77     real          jx, jy, jz;
  78     real          dx, dy, dz, rsq, rinv;
  79     real          c6, c12, c6grid, cexp1, cexp2, br;
  80     real *        charge;
  81     real *        shiftvec;
  82     real *        vdwparam, *vdwgridparam;
  83     int *         shift;
  84     int *         type;
  85     real *        fshift;
  86     real *        velecgrp;
  87     real *        vvdwgrp;
  88     real          tabscale;
  89     real *        VFtab;
  90     real *        x;
  91     real *        f;
  92     int           ewitab;
  93     real          ewtabscale, eweps, sh_ewald, ewrt, ewtabhalfspace;
  94     real *        ewtab;
  95     real          rcoulomb2, rvdw, rvdw2, sh_dispersion, sh_repulsion;
  96     real          rcutoff, rcutoff2;
  97     real          rswitch_elec, rswitch_vdw, d, d2, sw, dsw, rinvcorr;
  98     real          elec_swV3, elec_swV4, elec_swV5, elec_swF2, elec_swF3, elec_swF4;
  99     real          vdw_swV3, vdw_swV4, vdw_swV5, vdw_swF2, vdw_swF3, vdw_swF4;
 100     real          ewclj, ewclj2, ewclj6, ewcljrsq, poly, exponent, sh_lj_ewald;
 101     gmx_bool      bExactElecCutoff, bExactVdwCutoff, bExactCutoff;
 102
 103     x                   = xx[0];
 104     f                   = ff[0];
 105     ielec               = nlist->ielec;
 106     ivdw                = nlist->ivdw;
 107
 108     fshift              = fr->fshift[0];
 109     velecgrp            = kernel_data->energygrp_elec;
 110     vvdwgrp             = kernel_data->energygrp_vdw;
 111     tabscale            = kernel_data->table_elec_vdw->scale;
 112     VFtab               = kernel_data->table_elec_vdw->data;
 113
 114     sh_ewald            = fr->ic->sh_ewald;
 115     ewtab               = fr->ic->tabq_coul_FDV0;
 116     ewtabscale          = fr->ic->tabq_scale;
 117     ewtabhalfspace      = 0.5/ewtabscale;
 118
 119     rcoulomb2           = fr->rcoulomb*fr->rcoulomb;
 120     rvdw                = fr->rvdw;
 121     rvdw2               = rvdw*rvdw;
 122     sh_dispersion       = fr->ic->dispersion_shift.cpot;
 123     sh_repulsion        = fr->ic->repulsion_shift.cpot;
 124     sh_lj_ewald         = fr->ic->sh_lj_ewald;
 125
 126     ewclj               = fr->ewaldcoeff_lj;
 127     ewclj2              = ewclj*ewclj;
 128     ewclj6              = ewclj2*ewclj2*ewclj2;
 129
 130     if (fr->coulomb_modifier == eintmodPOTSWITCH)
 131     {
 132         d               = fr->rcoulomb-fr->rcoulomb_switch;
 133         elec_swV3       = -10.0/(d*d*d);
 134         elec_swV4       =  15.0/(d*d*d*d);
 135         elec_swV5       =  -6.0/(d*d*d*d*d);
 136         elec_swF2       = -30.0/(d*d*d);
 137         elec_swF3       =  60.0/(d*d*d*d);
 138         elec_swF4       = -30.0/(d*d*d*d*d);
 139     }
 140     else
 141     {
 142         /* Avoid warnings from stupid compilers (looking at you, Clang!) */
 143         elec_swV3 = elec_swV4 = elec_swV5 = elec_swF2 = elec_swF3 = elec_swF4 = 0.0;
 144     }
 145     if (fr->vdw_modifier == eintmodPOTSWITCH)
 146     {
 147         d               = fr->rvdw-fr->rvdw_switch;
 148         vdw_swV3        = -10.0/(d*d*d);
 149         vdw_swV4        =  15.0/(d*d*d*d);
 150         vdw_swV5        =  -6.0/(d*d*d*d*d);
 151         vdw_swF2        = -30.0/(d*d*d);
 152         vdw_swF3        =  60.0/(d*d*d*d);
 153         vdw_swF4        = -30.0/(d*d*d*d*d);
 154     }
 155     else
 156     {
 157         /* Avoid warnings from stupid compilers (looking at you, Clang!) */
 158         vdw_swV3 = vdw_swV4 = vdw_swV5 = vdw_swF2 = vdw_swF3 = vdw_swF4 = 0.0;
 159     }
 160
 161     bExactElecCutoff    = (fr->coulomb_modifier != eintmodNONE) || fr->eeltype == eelRF_ZERO;
 162     bExactVdwCutoff     = (fr->vdw_modifier != eintmodNONE);
 163     bExactCutoff        = bExactElecCutoff && bExactVdwCutoff;
 164
 165     if (bExactCutoff)
 166     {
 167         rcutoff  = ( fr->rcoulomb > fr->rvdw ) ? fr->rcoulomb : fr->rvdw;
 168         rcutoff2 = rcutoff*rcutoff;
 169     }
 170     else
 171     {
 172         /* Fix warnings for stupid compilers */
 173         rcutoff = rcutoff2 = 1e30;
 174     }
 175
 176     /* avoid compiler warnings for cases that cannot happen */
 177     nnn                 = 0;
 178     eps                 = 0.0;
 179     eps2                = 0.0;
 180
 181     /* 3 VdW parameters for Buckingham, otherwise 2 */
 182     nvdwparam           = (ivdw == GMX_NBKERNEL_VDW_BUCKINGHAM) ? 3 : 2;
 183     table_nelements     = 12;
 184
 185     charge              = mdatoms->chargeA;
 186     type                = mdatoms->typeA;
 187     facel               = fr->epsfac;
 188     shiftvec            = fr->shift_vec[0];
 189     vdwparam            = fr->nbfp;
 190     ntype               = fr->ntype;
 191     vdwgridparam        = fr->ljpme_c6grid;
 192
 193     for (n = 0; (n < nlist->nri); n++)
 194     {
 195         is3              = 3*nlist->shift[n];
 196         shX              = shiftvec[is3];
 197         shY              = shiftvec[is3+1];
 198         shZ              = shiftvec[is3+2];
 199         nj0              = nlist->jindex[n];
 200         nj1              = nlist->jindex[n+1];
 201         ii               = nlist->iinr[n];
 202         ii3              = 3*ii;
 203         ix               = shX + x[ii3+0];
 204         iy               = shY + x[ii3+1];
 205         iz               = shZ + x[ii3+2];
 206         iq               = facel*charge[ii];
 207         nti              = nvdwparam*ntype*type[ii];
 208         vctot            = 0;
 209         vvdwtot          = 0;
 210         fix              = 0;
 211         fiy              = 0;
 212         fiz              = 0;
 213
 214         for (k = nj0; (k < nj1); k++)
 215         {
 216             jnr              = nlist->jjnr[k];
 217             j3               = 3*jnr;
 218             jx               = x[j3+0];
 219             jy               = x[j3+1];
 220             jz               = x[j3+2];
 221             dx               = ix - jx;
 222             dy               = iy - jy;
 223             dz               = iz - jz;
 224             rsq              = dx*dx+dy*dy+dz*dz;
 225             rinv             = gmx_invsqrt(rsq);
 226             rinvsq           = rinv*rinv;
 227             felec            = 0;
 228             fvdw             = 0;
 229             velec            = 0;
 230             vvdw             = 0;
 231
 232             if (bExactCutoff && rsq >= rcutoff2)
 233             {
 234                 continue;
 235             }
 236
 237             if (ielec == GMX_NBKERNEL_ELEC_CUBICSPLINETABLE || ivdw == GMX_NBKERNEL_VDW_CUBICSPLINETABLE)
 238             {
 239                 r                = rsq*rinv;
 240                 rt               = r*tabscale;
 241                 n0               = rt;
 242                 eps              = rt-n0;
 243                 eps2             = eps*eps;
 244                 nnn              = table_nelements*n0;
 245             }
 246
 247             /* Coulomb interaction. ielec==0 means no interaction */
 248             if (ielec != GMX_NBKERNEL_ELEC_NONE)
 249             {
 250                 qq               = iq*charge[jnr];
 251
 252                 switch (ielec)
 253                 {
 254                     case GMX_NBKERNEL_ELEC_NONE:
 255                         break;
 256
 257                     case GMX_NBKERNEL_ELEC_COULOMB:
 258                         /* Vanilla cutoff coulomb */
 259                         velec            = qq*rinv;
 260                         felec            = velec*rinvsq;
 261                         /* The shift for the Coulomb potential is stored in
 262                          * the RF parameter c_rf, which is 0 without shift
 263                          */
 264                         velec           -= qq*fr->ic->c_rf;
 265                         break;
 266
 267                     case GMX_NBKERNEL_ELEC_REACTIONFIELD:
 268                         /* Reaction-field */
 269                         velec            = qq*(rinv+fr->k_rf*rsq-fr->c_rf);
 270                         felec            = qq*(rinv*rinvsq-2.0*fr->k_rf);
 271                         break;
 272
 273                     case GMX_NBKERNEL_ELEC_CUBICSPLINETABLE:
 274                         /* Tabulated coulomb */
 275                         Y                = VFtab[nnn];
 276                         F                = VFtab[nnn+1];
 277                         Geps             = eps*VFtab[nnn+2];
 278                         Heps2            = eps2*VFtab[nnn+3];
 279                         Fp               = F+Geps+Heps2;
 280                         VV               = Y+eps*Fp;
 281                         FF               = Fp+Geps+2.0*Heps2;
 282                         velec            = qq*VV;
 283                         felec            = -qq*FF*tabscale*rinv;
 284                         break;
 285
 286                     case GMX_NBKERNEL_ELEC_GENERALIZEDBORN:
 287                         /* GB */
 288                         gmx_fatal(FARGS, "Death & horror! GB generic interaction not implemented.\n");
 289                         break;
 290
 291                     case GMX_NBKERNEL_ELEC_EWALD:
 292                         ewrt             = rsq*rinv*ewtabscale;
 293                         ewitab           = ewrt;
 294                         eweps            = ewrt-ewitab;
 295                         ewitab           = 4*ewitab;
 296                         felec            = ewtab[ewitab]+eweps*ewtab[ewitab+1];
 297                         rinvcorr         = (fr->coulomb_modifier == eintmodPOTSHIFT) ? rinv-fr->ic->sh_ewald : rinv;
 298                         velec            = qq*(rinvcorr-(ewtab[ewitab+2]-ewtabhalfspace*eweps*(ewtab[ewitab]+felec)));
 299                         felec            = qq*rinv*(rinvsq-felec);
 300                         break;
 301
 302                     default:
 303                         gmx_fatal(FARGS, "Death & horror! No generic coulomb interaction for ielec=%d.\n", ielec);
 304                         break;
 305                 }
 306                 if (fr->coulomb_modifier == eintmodPOTSWITCH)
 307                 {
 308                     d                = rsq*rinv-fr->rcoulomb_switch;
 309                     d                = (d > 0.0) ? d : 0.0;
 310                     d2               = d*d;
 311                     sw               = 1.0+d2*d*(elec_swV3+d*(elec_swV4+d*elec_swV5));
 312                     dsw              = d2*(elec_swF2+d*(elec_swF3+d*elec_swF4));
 313                     /* Apply switch function. Note that felec=f/r since it will be multiplied
 314                      * by the i-j displacement vector. This means felec'=f'/r=-(v*sw)'/r=
 315                      * -(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=felec*sw-v*dsw/r
 316                      */
 317                     felec            = felec*sw - rinv*velec*dsw;
 318                     /* Once we have used velec to update felec we can modify velec too */
 319                     velec           *= sw;
 320                 }
 321                 if (bExactElecCutoff)
 322                 {
 323                     felec            = (rsq < rcoulomb2) ? felec : 0.0;
 324                     velec            = (rsq < rcoulomb2) ? velec : 0.0;
 325                 }
 326                 vctot           += velec;
 327             } /* End of coulomb interactions */
 328
 329
 330             /* VdW interaction. ivdw==0 means no interaction */
 331             if (ivdw != GMX_NBKERNEL_VDW_NONE)
 332             {
 333                 tj               = nti+nvdwparam*type[jnr];
 334
 335                 switch (ivdw)
 336                 {
 337                     case GMX_NBKERNEL_VDW_NONE:
 338                         break;
 339
 340                     case GMX_NBKERNEL_VDW_LENNARDJONES:
 341                         /* Vanilla Lennard-Jones cutoff */
 342                         c6               = vdwparam[tj];
 343                         c12              = vdwparam[tj+1];
 344                         rinvsix          = rinvsq*rinvsq*rinvsq;
 345                         vvdw_disp        = c6*rinvsix;
 346                         vvdw_rep         = c12*rinvsix*rinvsix;
 347                         fvdw             = (vvdw_rep-vvdw_disp)*rinvsq;
 348                         if (fr->vdw_modifier == eintmodPOTSHIFT)
 349                         {
 350                             vvdw             = (vvdw_rep + c12*sh_repulsion)/12.0 - (vvdw_disp + c6*sh_dispersion)/6.0;
 351                         }
 352                         else
 353                         {
 354                             vvdw             = vvdw_rep/12.0-vvdw_disp/6.0;
 355                         }
 356                         break;
 357
 358                     case GMX_NBKERNEL_VDW_BUCKINGHAM:
 359                         /* Buckingham */
 360                         c6               = vdwparam[tj];
 361                         cexp1            = vdwparam[tj+1];
 362                         cexp2            = vdwparam[tj+2];
 363
 364                         rinvsix          = rinvsq*rinvsq*rinvsq;
 365                         vvdw_disp        = c6*rinvsix;
 366                         br               = cexp2*rsq*rinv;
 367                         vvdw_rep         = cexp1*exp(-br);
 368                         fvdw             = (br*vvdw_rep-vvdw_disp)*rinvsq;
 369                         if (fr->vdw_modifier == eintmodPOTSHIFT)
 370                         {
 371                             vvdw             = (vvdw_rep-cexp1*exp(-cexp2*rvdw))-(vvdw_disp + c6*sh_dispersion)/6.0;
 372                         }
 373                         else
 374                         {
 375                             vvdw             = vvdw_rep-vvdw_disp/6.0;
 376                         }
 377                         break;
 378
 379                     case GMX_NBKERNEL_VDW_CUBICSPLINETABLE:
 380                         /* Tabulated VdW */
 381                         c6               = vdwparam[tj];
 382                         c12              = vdwparam[tj+1];
 383                         Y                = VFtab[nnn+4];
 384                         F                = VFtab[nnn+5];
 385                         Geps             = eps*VFtab[nnn+6];
 386                         Heps2            = eps2*VFtab[nnn+7];
 387                         Fp               = F+Geps+Heps2;
 388                         VV               = Y+eps*Fp;
 389                         FF               = Fp+Geps+2.0*Heps2;
 390                         vvdw_disp        = c6*VV;
 391                         fijD             = c6*FF;
 392                         Y                = VFtab[nnn+8];
 393                         F                = VFtab[nnn+9];
 394                         Geps             = eps*VFtab[nnn+10];
 395                         Heps2            = eps2*VFtab[nnn+11];
 396                         Fp               = F+Geps+Heps2;
 397                         VV               = Y+eps*Fp;
 398                         FF               = Fp+Geps+2.0*Heps2;
 399                         vvdw_rep         = c12*VV;
 400                         fijR             = c12*FF;
 401                         fvdw             = -(fijD+fijR)*tabscale*rinv;
 402                         vvdw             = vvdw_disp + vvdw_rep;
 403                         break;
 404
 405
 406                     case GMX_NBKERNEL_VDW_LJEWALD:
 407                         /* LJ-PME */
 408                         rinvsix          = rinvsq*rinvsq*rinvsq;
 409                         ewcljrsq         = ewclj2*rsq;
 410                         exponent         = exp(-ewcljrsq);
 411                         poly             = exponent*(1.0 + ewcljrsq + ewcljrsq*ewcljrsq*0.5);
 412                         c6               = vdwparam[tj];
 413                         c12              = vdwparam[tj+1];
 414                         c6grid           = vdwgridparam[tj];
 415                         vvdw_disp        = (c6-c6grid*(1.0-poly))*rinvsix;
 416                         vvdw_rep         = c12*rinvsix*rinvsix;
 417                         fvdw             = (vvdw_rep - vvdw_disp - c6grid*(1.0/6.0)*exponent*ewclj6)*rinvsq;
 418                         if (fr->vdw_modifier == eintmodPOTSHIFT)
 419                         {
 420                             vvdw             = (vvdw_rep + c12*sh_repulsion)/12.0 - (vvdw_disp + c6*sh_dispersion + c6grid*sh_lj_ewald)/6.0;
 421                         }
 422                         else
 423                         {
 424                             vvdw             = vvdw_rep/12.0-vvdw_disp/6.0;
 425                         }
 426                         break;
 427
 428                     default:
 429                         gmx_fatal(FARGS, "Death & horror! No generic VdW interaction for ivdw=%d.\n", ivdw);
 430                         break;
 431                 }
 432                 if (fr->vdw_modifier == eintmodPOTSWITCH)
 433                 {
 434                     d                = rsq*rinv-fr->rvdw_switch;
 435                     d                = (d > 0.0) ? d : 0.0;
 436                     d2               = d*d;
 437                     sw               = 1.0+d2*d*(vdw_swV3+d*(vdw_swV4+d*vdw_swV5));
 438                     dsw              = d2*(vdw_swF2+d*(vdw_swF3+d*vdw_swF4));
 439                     /* See coulomb interaction for the force-switch formula */
 440                     fvdw             = fvdw*sw - rinv*vvdw*dsw;
 441                     vvdw            *= sw;
 442                 }
 443                 if (bExactVdwCutoff)
 444                 {
 445                     fvdw             = (rsq < rvdw2) ? fvdw : 0.0;
 446                     vvdw             = (rsq < rvdw2) ? vvdw : 0.0;
 447                 }
 448                 vvdwtot         += vvdw;
 449             } /* end VdW interactions */
 450
 451             fscal            = felec+fvdw;
 452
 453             tx               = fscal*dx;
 454             ty               = fscal*dy;
 455             tz               = fscal*dz;
 456             fix              = fix + tx;
 457             fiy              = fiy + ty;
 458             fiz              = fiz + tz;
 459             f[j3+0]          = f[j3+0] - tx;
 460             f[j3+1]          = f[j3+1] - ty;
 461             f[j3+2]          = f[j3+2] - tz;
 462         }
 463
 464         f[ii3+0]         = f[ii3+0] + fix;
 465         f[ii3+1]         = f[ii3+1] + fiy;
 466         f[ii3+2]         = f[ii3+2] + fiz;
 467         fshift[is3]      = fshift[is3]+fix;
 468         fshift[is3+1]    = fshift[is3+1]+fiy;
 469         fshift[is3+2]    = fshift[is3+2]+fiz;
 470         ggid             = nlist->gid[n];
 471         velecgrp[ggid]  += vctot;
 472         vvdwgrp[ggid]   += vvdwtot;
 473     }
 474     /* Estimate flops, average for generic kernel:
 475      * 12 flops per outer iteration
 476      * 50 flops per inner iteration
 477      */
 478     inc_nrnb(nrnb, eNR_NBKERNEL_GENERIC, nlist->nri*12 + nlist->jindex[n]*50);
 479 }