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