1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
4 * This source code is part of
8 * GROningen MAchine for Chemical Simulations
10 * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
11 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
12 * Copyright (c) 2001-2008, The GROMACS development team,
13 * check out http://www.gromacs.org for more information.
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version 2
18 * of the License, or (at your option) any later version.
20 * If you want to redistribute modifications, please consider that
21 * scientific software is very special. Version control is crucial -
22 * bugs must be traceable. We will be happy to consider code for
23 * inclusion in the official distribution, but derived work must not
24 * be called official GROMACS. Details are found in the README & COPYING
25 * files - if they are missing, get the official version at www.gromacs.org.
27 * To help us fund GROMACS development, we humbly ask that you cite
28 * the papers on the package - you can find them in the top README file.
30 * For more info, check our website at http://www.gromacs.org
33 * Gallium Rubidium Oxygen Manganese Argon Carbon Silicon
48 #include "gromacs/fileio/pdbio.h"
54 #include "gmx_fatal.h"
55 #include "mtop_util.h"
60 #include "gromacs/utility/gmxmpi.h"
64 # include "genborn_sse2_double.h"
65 # include "genborn_allvsall_sse2_double.h"
67 # include "genborn_sse2_single.h"
68 # include "genborn_allvsall_sse2_single.h"
69 # endif /* GMX_DOUBLE */
70 #endif /* SSE or AVX present */
72 #include "genborn_allvsall.h"
74 /*#define DISABLE_SSE*/
83 typedef struct gbtmpnbls {
89 /* This function is exactly the same as the one in bondfree.c. The reason
90 * it is copied here is that the bonded gb-interactions are evaluated
91 * not in calc_bonds, but rather in calc_gb_forces
93 static int pbc_rvec_sub(const t_pbc *pbc, const rvec xi, const rvec xj, rvec dx)
97 return pbc_dx_aiuc(pbc, xi, xj, dx);
101 rvec_sub(xi, xj, dx);
106 int init_gb_nblist(int natoms, t_nblist *nl)
108 nl->maxnri = natoms*4;
118 /*nl->nltype = nltype;*/
120 srenew(nl->iinr, nl->maxnri);
121 srenew(nl->gid, nl->maxnri);
122 srenew(nl->shift, nl->maxnri);
123 srenew(nl->jindex, nl->maxnri+1);
130 void gb_pd_send(t_commrec *cr, real *send_data, int nr)
134 int *index, *sendc, *disp;
136 snew(sendc, cr->nnodes);
137 snew(disp, cr->nnodes);
139 index = pd_index(cr);
142 /* Setup count/index arrays */
143 for (i = 0; i < cr->nnodes; i++)
145 sendc[i] = index[i+1]-index[i];
149 /* Do communication */
150 MPI_Gatherv(send_data+index[cur], sendc[cur], GMX_MPI_REAL, send_data, sendc,
151 disp, GMX_MPI_REAL, 0, cr->mpi_comm_mygroup);
152 MPI_Bcast(send_data, nr, GMX_MPI_REAL, 0, cr->mpi_comm_mygroup);
158 int init_gb_plist(t_params *p_list)
161 p_list->param = NULL;
168 int init_gb_still(const t_commrec *cr,
169 const t_atomtypes *atype, t_idef *idef, t_atoms *atoms,
170 gmx_genborn_t *born, int natoms)
173 int i, j, i1, i2, k, m, nbond, nang, ia, ib, ic, id, nb, idx, idx2, at;
177 real r, ri, rj, ri2, ri3, rj2, r2, r3, r4, rk, ratio, term, h, doffset;
178 real p1, p2, p3, factor, cosine, rab, rbc;
185 snew(born->gpol_still_work, natoms+3);
191 pd_at_range(cr, &at0, &at1);
193 for (i = 0; i < natoms; i++)
210 doffset = born->gb_doffset;
212 for (i = 0; i < natoms; i++)
214 born->gpol_globalindex[i] = born->vsolv_globalindex[i] =
215 born->gb_radius_globalindex[i] = 0;
218 /* Compute atomic solvation volumes for Still method */
219 for (i = 0; i < natoms; i++)
221 ri = atype->gb_radius[atoms->atom[i].type];
222 born->gb_radius_globalindex[i] = ri;
224 born->vsolv_globalindex[i] = (4*M_PI/3)*r3;
227 for (j = 0; j < idef->il[F_GB12].nr; j += 3)
229 m = idef->il[F_GB12].iatoms[j];
230 ia = idef->il[F_GB12].iatoms[j+1];
231 ib = idef->il[F_GB12].iatoms[j+2];
233 r = 1.01*idef->iparams[m].gb.st;
235 ri = atype->gb_radius[atoms->atom[ia].type];
236 rj = atype->gb_radius[atoms->atom[ib].type];
242 ratio = (rj2-ri2-r*r)/(2*ri*r);
244 term = (M_PI/3.0)*h*h*(3.0*ri-h);
252 born->vsolv_globalindex[ia] -= term;
255 ratio = (ri2-rj2-r*r)/(2*rj*r);
257 term = (M_PI/3.0)*h*h*(3.0*rj-h);
265 born->vsolv_globalindex[ib] -= term;
271 gmx_sum(natoms, vsol, cr);
273 for (i = 0; i < natoms; i++)
275 born->vsolv_globalindex[i] = born->vsolv_globalindex[i]-vsol[i];
279 /* Get the self-, 1-2 and 1-3 polarization energies for analytical Still
282 for (j = 0; j < natoms; j++)
284 if (born->use_globalindex[j] == 1)
286 born->gpol_globalindex[j] = -0.5*ONE_4PI_EPS0/
287 (atype->gb_radius[atoms->atom[j].type]-doffset+STILL_P1);
292 for (j = 0; j < idef->il[F_GB12].nr; j += 3)
294 m = idef->il[F_GB12].iatoms[j];
295 ia = idef->il[F_GB12].iatoms[j+1];
296 ib = idef->il[F_GB12].iatoms[j+2];
298 r = idef->iparams[m].gb.st;
304 gp[ia] += STILL_P2*born->vsolv_globalindex[ib]/r4;
305 gp[ib] += STILL_P2*born->vsolv_globalindex[ia]/r4;
309 born->gpol_globalindex[ia] = born->gpol_globalindex[ia]+
310 STILL_P2*born->vsolv_globalindex[ib]/r4;
311 born->gpol_globalindex[ib] = born->gpol_globalindex[ib]+
312 STILL_P2*born->vsolv_globalindex[ia]/r4;
317 for (j = 0; j < idef->il[F_GB13].nr; j += 3)
319 m = idef->il[F_GB13].iatoms[j];
320 ia = idef->il[F_GB13].iatoms[j+1];
321 ib = idef->il[F_GB13].iatoms[j+2];
323 r = idef->iparams[m].gb.st;
328 gp[ia] += STILL_P3*born->vsolv[ib]/r4;
329 gp[ib] += STILL_P3*born->vsolv[ia]/r4;
333 born->gpol_globalindex[ia] = born->gpol_globalindex[ia]+
334 STILL_P3*born->vsolv_globalindex[ib]/r4;
335 born->gpol_globalindex[ib] = born->gpol_globalindex[ib]+
336 STILL_P3*born->vsolv_globalindex[ia]/r4;
342 gmx_sum(natoms, gp, cr);
344 for (i = 0; i < natoms; i++)
346 born->gpol_globalindex[i] = born->gpol_globalindex[i]+gp[i];
356 /* Initialize all GB datastructs and compute polarization energies */
357 int init_gb(gmx_genborn_t **p_born,
358 const t_commrec *cr, t_forcerec *fr, const t_inputrec *ir,
359 const gmx_mtop_t *mtop, int gb_algorithm)
361 int i, j, m, ai, aj, jj, natoms, nalloc;
362 real rai, sk, p, doffset;
366 gmx_localtop_t *localtop;
368 natoms = mtop->natoms;
370 atoms = gmx_mtop_global_atoms(mtop);
371 localtop = gmx_mtop_generate_local_top(mtop, ir);
378 snew(born->drobc, natoms);
379 snew(born->bRad, natoms);
381 /* Allocate memory for the global data arrays */
382 snew(born->param_globalindex, natoms+3);
383 snew(born->gpol_globalindex, natoms+3);
384 snew(born->vsolv_globalindex, natoms+3);
385 snew(born->gb_radius_globalindex, natoms+3);
386 snew(born->use_globalindex, natoms+3);
388 snew(fr->invsqrta, natoms);
389 snew(fr->dvda, natoms);
392 fr->dadx_rawptr = NULL;
394 born->gpol_still_work = NULL;
395 born->gpol_hct_work = NULL;
397 /* snew(born->asurf,natoms); */
398 /* snew(born->dasurf,natoms); */
400 /* Initialize the gb neighbourlist */
401 init_gb_nblist(natoms, &(fr->gblist));
403 /* Do the Vsites exclusions (if any) */
404 for (i = 0; i < natoms; i++)
406 jj = atoms.atom[i].type;
407 if (mtop->atomtypes.gb_radius[atoms.atom[i].type] > 0)
409 born->use_globalindex[i] = 1;
413 born->use_globalindex[i] = 0;
416 /* If we have a Vsite, put vs_globalindex[i]=0 */
417 if (C6 (fr->nbfp, fr->ntype, jj, jj) == 0 &&
418 C12(fr->nbfp, fr->ntype, jj, jj) == 0 &&
419 atoms.atom[i].q == 0)
421 born->use_globalindex[i] = 0;
425 /* Copy algorithm parameters from inputrecord to local structure */
426 born->obc_alpha = ir->gb_obc_alpha;
427 born->obc_beta = ir->gb_obc_beta;
428 born->obc_gamma = ir->gb_obc_gamma;
429 born->gb_doffset = ir->gb_dielectric_offset;
430 born->gb_epsilon_solvent = ir->gb_epsilon_solvent;
431 born->epsilon_r = ir->epsilon_r;
433 doffset = born->gb_doffset;
435 /* Set the surface tension */
436 born->sa_surface_tension = ir->sa_surface_tension;
438 /* If Still model, initialise the polarisation energies */
439 if (gb_algorithm == egbSTILL)
441 init_gb_still(cr, &(mtop->atomtypes), &(localtop->idef), &atoms,
446 /* If HCT/OBC, precalculate the sk*atype->S_hct factors */
447 else if (gb_algorithm == egbHCT || gb_algorithm == egbOBC)
450 snew(born->gpol_hct_work, natoms+3);
452 for (i = 0; i < natoms; i++)
454 if (born->use_globalindex[i] == 1)
456 rai = mtop->atomtypes.gb_radius[atoms.atom[i].type]-doffset;
457 sk = rai * mtop->atomtypes.S_hct[atoms.atom[i].type];
458 born->param_globalindex[i] = sk;
459 born->gb_radius_globalindex[i] = rai;
463 born->param_globalindex[i] = 0;
464 born->gb_radius_globalindex[i] = 0;
469 /* Allocate memory for work arrays for temporary use */
470 snew(born->work, natoms+4);
471 snew(born->count, natoms);
472 snew(born->nblist_work, natoms);
474 /* Domain decomposition specific stuff */
483 calc_gb_rad_still(t_commrec *cr, t_forcerec *fr, int natoms, gmx_localtop_t *top,
484 rvec x[], t_nblist *nl,
485 gmx_genborn_t *born, t_mdatoms *md)
487 int i, k, n, nj0, nj1, ai, aj, type;
490 real gpi, dr, dr2, dr4, idr4, rvdw, ratio, ccf, theta, term, rai, raj;
491 real ix1, iy1, iz1, jx1, jy1, jz1, dx11, dy11, dz11;
492 real rinv, idr2, idr6, vaj, dccf, cosq, sinq, prod, gpi2;
494 real vai, prod_ai, icf4, icf6;
496 factor = 0.5*ONE_4PI_EPS0;
499 for (i = 0; i < born->nr; i++)
501 born->gpol_still_work[i] = 0;
504 for (i = 0; i < nl->nri; i++)
509 nj1 = nl->jindex[i+1];
511 /* Load shifts for this list */
512 shift = nl->shift[i];
513 shX = fr->shift_vec[shift][0];
514 shY = fr->shift_vec[shift][1];
515 shZ = fr->shift_vec[shift][2];
519 rai = top->atomtypes.gb_radius[md->typeA[ai]];
520 vai = born->vsolv[ai];
521 prod_ai = STILL_P4*vai;
523 /* Load atom i coordinates, add shift vectors */
524 ix1 = shX + x[ai][0];
525 iy1 = shY + x[ai][1];
526 iz1 = shZ + x[ai][2];
528 for (k = nj0; k < nj1 && nl->jjnr[k] >= 0; k++)
539 dr2 = dx11*dx11+dy11*dy11+dz11*dz11;
540 rinv = gmx_invsqrt(dr2);
545 raj = top->atomtypes.gb_radius[md->typeA[aj]];
549 ratio = dr2 / (rvdw * rvdw);
550 vaj = born->vsolv[aj];
552 if (ratio > STILL_P5INV)
559 theta = ratio*STILL_PIP5;
561 term = 0.5*(1.0-cosq);
563 sinq = 1.0 - cosq*cosq;
564 dccf = 2.0*term*sinq*gmx_invsqrt(sinq)*theta;
569 icf6 = (4*ccf-dccf)*idr6;
570 born->gpol_still_work[aj] += prod_ai*icf4;
573 /* Save ai->aj and aj->ai chain rule terms */
574 fr->dadx[n++] = prod*icf6;
575 fr->dadx[n++] = prod_ai*icf6;
577 born->gpol_still_work[ai] += gpi;
580 /* Parallel summations */
583 gmx_sum(natoms, born->gpol_still_work, cr);
585 else if (DOMAINDECOMP(cr))
587 dd_atom_sum_real(cr->dd, born->gpol_still_work);
590 /* Calculate the radii */
591 for (i = 0; i < fr->natoms_force; i++) /* PELA born->nr */
593 if (born->use[i] != 0)
595 gpi = born->gpol[i]+born->gpol_still_work[i];
597 born->bRad[i] = factor*gmx_invsqrt(gpi2);
598 fr->invsqrta[i] = gmx_invsqrt(born->bRad[i]);
602 /* Extra communication required for DD */
603 if (DOMAINDECOMP(cr))
605 dd_atom_spread_real(cr->dd, born->bRad);
606 dd_atom_spread_real(cr->dd, fr->invsqrta);
615 calc_gb_rad_hct(t_commrec *cr, t_forcerec *fr, int natoms, gmx_localtop_t *top,
616 rvec x[], t_nblist *nl,
617 gmx_genborn_t *born, t_mdatoms *md)
619 int i, k, n, ai, aj, nj0, nj1, at0, at1;
622 real rai, raj, gpi, dr2, dr, sk, sk_ai, sk2, sk2_ai, lij, uij, diff2, tmp, sum_ai;
623 real rad, min_rad, rinv, rai_inv;
624 real ix1, iy1, iz1, jx1, jy1, jz1, dx11, dy11, dz11;
625 real lij2, uij2, lij3, uij3, t1, t2, t3;
626 real lij_inv, dlij, duij, sk2_rinv, prod, log_term;
627 real doffset, raj_inv, dadx_val;
630 doffset = born->gb_doffset;
631 gb_radius = born->gb_radius;
633 for (i = 0; i < born->nr; i++)
635 born->gpol_hct_work[i] = 0;
638 /* Keep the compiler happy */
642 for (i = 0; i < nl->nri; i++)
647 nj1 = nl->jindex[i+1];
649 /* Load shifts for this list */
650 shift = nl->shift[i];
651 shX = fr->shift_vec[shift][0];
652 shY = fr->shift_vec[shift][1];
653 shZ = fr->shift_vec[shift][2];
658 sk_ai = born->param[ai];
659 sk2_ai = sk_ai*sk_ai;
661 /* Load atom i coordinates, add shift vectors */
662 ix1 = shX + x[ai][0];
663 iy1 = shY + x[ai][1];
664 iz1 = shZ + x[ai][2];
668 for (k = nj0; k < nj1 && nl->jjnr[k] >= 0; k++)
680 dr2 = dx11*dx11+dy11*dy11+dz11*dz11;
681 rinv = gmx_invsqrt(dr2);
684 sk = born->param[aj];
687 /* aj -> ai interaction */
708 lij_inv = gmx_invsqrt(lij2);
711 prod = 0.25*sk2_rinv;
713 log_term = log(uij*lij_inv);
715 tmp = lij-uij + 0.25*dr*diff2 + (0.5*rinv)*log_term +
720 tmp = tmp + 2.0 * (rai_inv-lij);
723 t1 = 0.5*lij2 + prod*lij3 - 0.25*(lij*rinv+lij3*dr);
724 t2 = -0.5*uij2 - 0.25*sk2_rinv*uij3 + 0.25*(uij*rinv+uij3*dr);
725 t3 = 0.125*(1.0+sk2_rinv*rinv)*(-diff2)+0.25*log_term*rinv*rinv;
727 dadx_val = (dlij*t1+t2+t3)*rinv; /* rb2 is moved to chainrule */
728 /* fr->dadx[n++] = (dlij*t1+duij*t2+t3)*rinv; */
729 /* rb2 is moved to chainrule */
737 fr->dadx[n++] = dadx_val;
740 /* ai -> aj interaction */
741 if (raj < dr + sk_ai)
743 lij = 1.0/(dr-sk_ai);
756 uij = 1.0/(dr+sk_ai);
762 lij_inv = gmx_invsqrt(lij2);
763 sk2 = sk2_ai; /* sk2_ai = sk_ai * sk_ai in i loop above */
765 prod = 0.25 * sk2_rinv;
767 /* log_term = table_log(uij*lij_inv,born->log_table,
768 LOG_TABLE_ACCURACY); */
769 log_term = log(uij*lij_inv);
771 tmp = lij-uij + 0.25*dr*diff2 + (0.5*rinv)*log_term +
776 tmp = tmp + 2.0 * (raj_inv-lij);
780 t1 = 0.5*lij2 + prod*lij3 - 0.25*(lij*rinv+lij3*dr);
781 t2 = -0.5*uij2 - 0.25*sk2_rinv*uij3 + 0.25*(uij*rinv+uij3*dr);
782 t3 = 0.125*(1.0+sk2_rinv*rinv)*(-diff2)+0.25*log_term*rinv*rinv;
784 dadx_val = (dlij*t1+t2+t3)*rinv; /* rb2 is moved to chainrule */
785 /* fr->dadx[n++] = (dlij*t1+duij*t2+t3)*rinv; */ /* rb2 is moved to chainrule */
787 born->gpol_hct_work[aj] += 0.5*tmp;
793 fr->dadx[n++] = dadx_val;
796 born->gpol_hct_work[ai] += sum_ai;
799 /* Parallel summations */
802 gmx_sum(natoms, born->gpol_hct_work, cr);
804 else if (DOMAINDECOMP(cr))
806 dd_atom_sum_real(cr->dd, born->gpol_hct_work);
809 for (i = 0; i < fr->natoms_force; i++) /* PELA born->nr */
811 if (born->use[i] != 0)
813 rai = top->atomtypes.gb_radius[md->typeA[i]]-doffset;
814 sum_ai = 1.0/rai - born->gpol_hct_work[i];
815 min_rad = rai + doffset;
818 born->bRad[i] = rad > min_rad ? rad : min_rad;
819 fr->invsqrta[i] = gmx_invsqrt(born->bRad[i]);
823 /* Extra communication required for DD */
824 if (DOMAINDECOMP(cr))
826 dd_atom_spread_real(cr->dd, born->bRad);
827 dd_atom_spread_real(cr->dd, fr->invsqrta);
835 calc_gb_rad_obc(t_commrec *cr, t_forcerec *fr, int natoms, gmx_localtop_t *top,
836 rvec x[], t_nblist *nl, gmx_genborn_t *born, t_mdatoms *md)
838 int i, k, ai, aj, nj0, nj1, n, at0, at1;
841 real rai, raj, gpi, dr2, dr, sk, sk2, lij, uij, diff2, tmp, sum_ai;
842 real rad, min_rad, sum_ai2, sum_ai3, tsum, tchain, rinv, rai_inv, lij_inv, rai_inv2;
843 real log_term, prod, sk2_rinv, sk_ai, sk2_ai;
844 real ix1, iy1, iz1, jx1, jy1, jz1, dx11, dy11, dz11;
845 real lij2, uij2, lij3, uij3, dlij, duij, t1, t2, t3;
846 real doffset, raj_inv, dadx_val;
849 /* Keep the compiler happy */
854 doffset = born->gb_doffset;
855 gb_radius = born->gb_radius;
857 for (i = 0; i < born->nr; i++)
859 born->gpol_hct_work[i] = 0;
862 for (i = 0; i < nl->nri; i++)
867 nj1 = nl->jindex[i+1];
869 /* Load shifts for this list */
870 shift = nl->shift[i];
871 shX = fr->shift_vec[shift][0];
872 shY = fr->shift_vec[shift][1];
873 shZ = fr->shift_vec[shift][2];
878 sk_ai = born->param[ai];
879 sk2_ai = sk_ai*sk_ai;
881 /* Load atom i coordinates, add shift vectors */
882 ix1 = shX + x[ai][0];
883 iy1 = shY + x[ai][1];
884 iz1 = shZ + x[ai][2];
888 for (k = nj0; k < nj1 && nl->jjnr[k] >= 0; k++)
900 dr2 = dx11*dx11+dy11*dy11+dz11*dz11;
901 rinv = gmx_invsqrt(dr2);
904 /* sk is precalculated in init_gb() */
905 sk = born->param[aj];
908 /* aj -> ai interaction */
928 lij_inv = gmx_invsqrt(lij2);
931 prod = 0.25*sk2_rinv;
933 log_term = log(uij*lij_inv);
935 tmp = lij-uij + 0.25*dr*diff2 + (0.5*rinv)*log_term + prod*(-diff2);
939 tmp = tmp + 2.0 * (rai_inv-lij);
943 t1 = 0.5*lij2 + prod*lij3 - 0.25*(lij*rinv+lij3*dr);
944 t2 = -0.5*uij2 - 0.25*sk2_rinv*uij3 + 0.25*(uij*rinv+uij3*dr);
945 t3 = 0.125*(1.0+sk2_rinv*rinv)*(-diff2)+0.25*log_term*rinv*rinv;
947 dadx_val = (dlij*t1+t2+t3)*rinv; /* rb2 is moved to chainrule */
955 fr->dadx[n++] = dadx_val;
957 /* ai -> aj interaction */
958 if (raj < dr + sk_ai)
960 lij = 1.0/(dr-sk_ai);
973 uij = 1.0/(dr+sk_ai);
979 lij_inv = gmx_invsqrt(lij2);
980 sk2 = sk2_ai; /* sk2_ai = sk_ai * sk_ai in i loop above */
982 prod = 0.25 * sk2_rinv;
984 /* log_term = table_log(uij*lij_inv,born->log_table,LOG_TABLE_ACCURACY); */
985 log_term = log(uij*lij_inv);
987 tmp = lij-uij + 0.25*dr*diff2 + (0.5*rinv)*log_term + prod*(-diff2);
991 tmp = tmp + 2.0 * (raj_inv-lij);
994 t1 = 0.5*lij2 + prod*lij3 - 0.25*(lij*rinv+lij3*dr);
995 t2 = -0.5*uij2 - 0.25*sk2_rinv*uij3 + 0.25*(uij*rinv+uij3*dr);
996 t3 = 0.125*(1.0+sk2_rinv*rinv)*(-diff2)+0.25*log_term*rinv*rinv;
998 dadx_val = (dlij*t1+t2+t3)*rinv; /* rb2 is moved to chainrule */
1000 born->gpol_hct_work[aj] += 0.5*tmp;
1007 fr->dadx[n++] = dadx_val;
1010 born->gpol_hct_work[ai] += sum_ai;
1014 /* Parallel summations */
1017 gmx_sum(natoms, born->gpol_hct_work, cr);
1019 else if (DOMAINDECOMP(cr))
1021 dd_atom_sum_real(cr->dd, born->gpol_hct_work);
1024 for (i = 0; i < fr->natoms_force; i++) /* PELA born->nr */
1026 if (born->use[i] != 0)
1028 rai = top->atomtypes.gb_radius[md->typeA[i]];
1032 sum_ai = rai * born->gpol_hct_work[i];
1033 sum_ai2 = sum_ai * sum_ai;
1034 sum_ai3 = sum_ai2 * sum_ai;
1036 tsum = tanh(born->obc_alpha*sum_ai-born->obc_beta*sum_ai2+born->obc_gamma*sum_ai3);
1037 born->bRad[i] = rai_inv - tsum*rai_inv2;
1038 born->bRad[i] = 1.0 / born->bRad[i];
1040 fr->invsqrta[i] = gmx_invsqrt(born->bRad[i]);
1042 tchain = rai * (born->obc_alpha-2*born->obc_beta*sum_ai+3*born->obc_gamma*sum_ai2);
1043 born->drobc[i] = (1.0-tsum*tsum)*tchain*rai_inv2;
1047 /* Extra (local) communication required for DD */
1048 if (DOMAINDECOMP(cr))
1050 dd_atom_spread_real(cr->dd, born->bRad);
1051 dd_atom_spread_real(cr->dd, fr->invsqrta);
1052 dd_atom_spread_real(cr->dd, born->drobc);
1061 int calc_gb_rad(t_commrec *cr, t_forcerec *fr, t_inputrec *ir, gmx_localtop_t *top,
1062 rvec x[], t_nblist *nl, gmx_genborn_t *born, t_mdatoms *md, t_nrnb *nrnb)
1068 if (fr->bAllvsAll && fr->dadx == NULL)
1070 /* We might need up to 8 atoms of padding before and after,
1071 * and another 4 units to guarantee SSE alignment.
1073 fr->nalloc_dadx = 2*(md->homenr+12)*(md->nr/2+1+12);
1074 snew(fr->dadx_rawptr, fr->nalloc_dadx);
1075 fr->dadx = (real *) (((size_t) fr->dadx_rawptr + 16) & (~((size_t) 15)));
1079 /* In the SSE-enabled gb-loops, when writing to dadx, we
1080 * always write 2*4 elements at a time, even in the case with only
1081 * 1-3 j particles, where we only really need to write 2*(1-3)
1082 * elements. This is because we want dadx to be aligned to a 16-
1083 * byte boundary, and being able to use _mm_store/load_ps
1085 ndadx = 2 * (nl->nrj + 3*nl->nri);
1087 /* First, reallocate the dadx array, we need 3 extra for SSE */
1088 if (ndadx + 3 > fr->nalloc_dadx)
1090 fr->nalloc_dadx = over_alloc_large(ndadx) + 3;
1091 srenew(fr->dadx_rawptr, fr->nalloc_dadx);
1092 fr->dadx = (real *) (((size_t) fr->dadx_rawptr + 16) & (~((size_t) 15)));
1098 cnt = md->homenr*(md->nr/2+1);
1100 if (ir->gb_algorithm == egbSTILL)
1102 #if 0 && defined (GMX_X86_SSE2)
1103 if (fr->use_acceleration)
1106 genborn_allvsall_calc_still_radii_sse2_double(fr, md, born, top, x[0], cr, &fr->AllvsAll_workgb);
1108 genborn_allvsall_calc_still_radii_sse2_single(fr, md, born, top, x[0], cr, &fr->AllvsAll_workgb);
1113 genborn_allvsall_calc_still_radii(fr, md, born, top, x[0], cr, &fr->AllvsAll_workgb);
1116 genborn_allvsall_calc_still_radii(fr, md, born, top, x[0], cr, &fr->AllvsAll_workgb);
1118 /* 13 flops in outer loop, 47 flops in inner loop */
1119 inc_nrnb(nrnb, eNR_BORN_AVA_RADII_STILL, md->homenr*13+cnt*47);
1121 else if (ir->gb_algorithm == egbHCT || ir->gb_algorithm == egbOBC)
1123 #if 0 && defined (GMX_X86_SSE2)
1124 if (fr->use_acceleration)
1127 genborn_allvsall_calc_hct_obc_radii_sse2_double(fr, md, born, ir->gb_algorithm, top, x[0], cr, &fr->AllvsAll_workgb);
1129 genborn_allvsall_calc_hct_obc_radii_sse2_single(fr, md, born, ir->gb_algorithm, top, x[0], cr, &fr->AllvsAll_workgb);
1134 genborn_allvsall_calc_hct_obc_radii(fr, md, born, ir->gb_algorithm, top, x[0], cr, &fr->AllvsAll_workgb);
1137 genborn_allvsall_calc_hct_obc_radii(fr, md, born, ir->gb_algorithm, top, x[0], cr, &fr->AllvsAll_workgb);
1139 /* 24 flops in outer loop, 183 in inner */
1140 inc_nrnb(nrnb, eNR_BORN_AVA_RADII_HCT_OBC, md->homenr*24+cnt*183);
1144 gmx_fatal(FARGS, "Bad gb algorithm for all-vs-all interactions");
1149 /* Switch for determining which algorithm to use for Born radii calculation */
1152 #if 0 && defined (GMX_X86_SSE2)
1153 /* x86 or x86-64 with GCC inline assembly and/or SSE intrinsics */
1154 switch (ir->gb_algorithm)
1157 if (fr->use_acceleration)
1159 calc_gb_rad_still_sse2_double(cr, fr, born->nr, top, atype, x[0], nl, born);
1163 calc_gb_rad_still(cr, fr, born->nr, top, atype, x, nl, born, md);
1167 if (fr->use_acceleration)
1169 calc_gb_rad_hct_obc_sse2_double(cr, fr, born->nr, top, atype, x[0], nl, born, md, ir->gb_algorithm);
1173 calc_gb_rad_hct(cr, fr, born->nr, top, atype, x, nl, born, md);
1177 if (fr->use_acceleration)
1179 calc_gb_rad_hct_obc_sse2_double(cr, fr, born->nr, top, atype, x[0], nl, born, md, ir->gb_algorithm);
1183 calc_gb_rad_obc(cr, fr, born->nr, top, atype, x, nl, born, md);
1188 gmx_fatal(FARGS, "Unknown double precision sse-enabled algorithm for Born radii calculation: %d", ir->gb_algorithm);
1191 switch (ir->gb_algorithm)
1194 calc_gb_rad_still(cr, fr, born->nr, top, x, nl, born, md);
1197 calc_gb_rad_hct(cr, fr, born->nr, top, x, nl, born, md);
1200 calc_gb_rad_obc(cr, fr, born->nr, top, x, nl, born, md);
1204 gmx_fatal(FARGS, "Unknown double precision algorithm for Born radii calculation: %d", ir->gb_algorithm);
1211 #if 0 && defined (GMX_X86_SSE2)
1212 /* x86 or x86-64 with GCC inline assembly and/or SSE intrinsics */
1213 switch (ir->gb_algorithm)
1216 if (fr->use_acceleration)
1218 calc_gb_rad_still_sse2_single(cr, fr, born->nr, top, x[0], nl, born);
1222 calc_gb_rad_still(cr, fr, born->nr, top, x, nl, born, md);
1226 if (fr->use_acceleration)
1228 calc_gb_rad_hct_obc_sse2_single(cr, fr, born->nr, top, x[0], nl, born, md, ir->gb_algorithm);
1232 calc_gb_rad_hct(cr, fr, born->nr, top, x, nl, born, md);
1237 if (fr->use_acceleration)
1239 calc_gb_rad_hct_obc_sse2_single(cr, fr, born->nr, top, x[0], nl, born, md, ir->gb_algorithm);
1243 calc_gb_rad_obc(cr, fr, born->nr, top, x, nl, born, md);
1248 gmx_fatal(FARGS, "Unknown sse-enabled algorithm for Born radii calculation: %d", ir->gb_algorithm);
1252 switch (ir->gb_algorithm)
1255 calc_gb_rad_still(cr, fr, born->nr, top, x, nl, born, md);
1258 calc_gb_rad_hct(cr, fr, born->nr, top, x, nl, born, md);
1261 calc_gb_rad_obc(cr, fr, born->nr, top, x, nl, born, md);
1265 gmx_fatal(FARGS, "Unknown algorithm for Born radii calculation: %d", ir->gb_algorithm);
1268 #endif /* Single precision sse */
1270 #endif /* Double or single precision */
1272 if (fr->bAllvsAll == FALSE)
1274 switch (ir->gb_algorithm)
1277 /* 17 flops per outer loop iteration, 47 flops per inner loop */
1278 inc_nrnb(nrnb, eNR_BORN_RADII_STILL, nl->nri*17+nl->nrj*47);
1282 /* 61 (assuming 10 for tanh) flops for outer loop iteration, 183 flops per inner loop */
1283 inc_nrnb(nrnb, eNR_BORN_RADII_HCT_OBC, nl->nri*61+nl->nrj*183);
1296 real gb_bonds_tab(rvec x[], rvec f[], rvec fshift[], real *charge, real *p_gbtabscale,
1297 real *invsqrta, real *dvda, real *GBtab, t_idef *idef, real epsilon_r,
1298 real gb_epsilon_solvent, real facel, const t_pbc *pbc, const t_graph *graph)
1300 int i, j, n0, m, nnn, type, ai, aj;
1306 real isaprod, qq, gbscale, gbtabscale, Y, F, Geps, Heps2, Fp, VV, FF, rt, eps, eps2;
1307 real vgb, fgb, vcoul, fijC, dvdatmp, fscal, dvdaj;
1313 t_iatom *forceatoms;
1315 /* Scale the electrostatics by gb_epsilon_solvent */
1316 facel = facel * ((1.0/epsilon_r) - 1.0/gb_epsilon_solvent);
1318 gbtabscale = *p_gbtabscale;
1321 for (j = F_GB12; j <= F_GB14; j++)
1323 forceatoms = idef->il[j].iatoms;
1325 for (i = 0; i < idef->il[j].nr; )
1327 /* To avoid reading in the interaction type, we just increment i to pass over
1328 * the types in the forceatoms array, this saves some memory accesses
1331 ai = forceatoms[i++];
1332 aj = forceatoms[i++];
1334 ki = pbc_rvec_sub(pbc, x[ai], x[aj], dx);
1335 rsq11 = iprod(dx, dx);
1337 isai = invsqrta[ai];
1338 iq = (-1)*facel*charge[ai];
1340 rinv11 = gmx_invsqrt(rsq11);
1341 isaj = invsqrta[aj];
1342 isaprod = isai*isaj;
1343 qq = isaprod*iq*charge[aj];
1344 gbscale = isaprod*gbtabscale;
1353 Geps = eps*GBtab[nnn+2];
1354 Heps2 = eps2*GBtab[nnn+3];
1357 FF = Fp+Geps+2.0*Heps2;
1359 fijC = qq*FF*gbscale;
1360 dvdatmp = -(vgb+fijC*r)*0.5;
1361 dvda[aj] = dvda[aj] + dvdatmp*isaj*isaj;
1362 dvda[ai] = dvda[ai] + dvdatmp*isai*isai;
1363 vctot = vctot + vgb;
1364 fgb = -(fijC)*rinv11;
1368 ivec_sub(SHIFT_IVEC(graph, ai), SHIFT_IVEC(graph, aj), dt);
1372 for (m = 0; (m < DIM); m++) /* 15 */
1377 fshift[ki][m] += fscal;
1378 fshift[CENTRAL][m] -= fscal;
1386 real calc_gb_selfcorrections(t_commrec *cr, int natoms,
1387 real *charge, gmx_genborn_t *born, real *dvda, double facel)
1389 int i, ai, at0, at1;
1390 real rai, e, derb, q, q2, fi, rai_inv, vtot;
1394 pd_at_range(cr, &at0, &at1);
1396 else if (DOMAINDECOMP(cr))
1399 at1 = cr->dd->nat_home;
1408 /* Scale the electrostatics by gb_epsilon_solvent */
1409 facel = facel * ((1.0/born->epsilon_r) - 1.0/born->gb_epsilon_solvent);
1413 /* Apply self corrections */
1414 for (i = at0; i < at1; i++)
1418 if (born->use[ai] == 1)
1420 rai = born->bRad[ai];
1426 derb = 0.5*e*rai_inv*rai_inv;
1427 dvda[ai] += derb*rai;
1436 real calc_gb_nonpolar(t_commrec *cr, t_forcerec *fr, int natoms, gmx_genborn_t *born, gmx_localtop_t *top,
1437 real *dvda, t_mdatoms *md)
1439 int ai, i, at0, at1;
1440 real e, es, rai, rbi, term, probe, tmp, factor;
1441 real rbi_inv, rbi_inv2;
1443 /* To keep the compiler happy */
1448 pd_at_range(cr, &at0, &at1);
1450 else if (DOMAINDECOMP(cr))
1453 at1 = cr->dd->nat_home;
1461 /* factor is the surface tension */
1462 factor = born->sa_surface_tension;
1465 // The surface tension factor is 0.0049 for Still model, 0.0054 for HCT/OBC
1466 if(gb_algorithm==egbSTILL)
1468 factor=0.0049*100*CAL2JOULE;
1472 factor=0.0054*100*CAL2JOULE;
1475 /* if(gb_algorithm==egbHCT || gb_algorithm==egbOBC) */
1481 for (i = at0; i < at1; i++)
1485 if (born->use[ai] == 1)
1487 rai = top->atomtypes.gb_radius[md->typeA[ai]];
1488 rbi_inv = fr->invsqrta[ai];
1489 rbi_inv2 = rbi_inv * rbi_inv;
1490 tmp = (rai*rbi_inv2)*(rai*rbi_inv2);
1492 e = factor*term*(rai+probe)*(rai+probe)*tmp;
1493 dvda[ai] = dvda[ai] - 6*e*rbi_inv2;
1503 real calc_gb_chainrule(int natoms, t_nblist *nl, real *dadx, real *dvda, rvec x[], rvec t[], rvec fshift[],
1504 rvec shift_vec[], int gb_algorithm, gmx_genborn_t *born)
1506 int i, k, n, ai, aj, nj0, nj1, n0, n1;
1509 real fgb, fij, rb2, rbi, fix1, fiy1, fiz1;
1510 real ix1, iy1, iz1, jx1, jy1, jz1, dx11, dy11, dz11, rsq11;
1511 real rinv11, tx, ty, tz, rbai, rbaj, fgb_ai;
1521 if (gb_algorithm == egbSTILL)
1523 for (i = n0; i < n1; i++)
1525 rbi = born->bRad[i];
1526 rb[i] = (2 * rbi * rbi * dvda[i])/ONE_4PI_EPS0;
1529 else if (gb_algorithm == egbHCT)
1531 for (i = n0; i < n1; i++)
1533 rbi = born->bRad[i];
1534 rb[i] = rbi * rbi * dvda[i];
1537 else if (gb_algorithm == egbOBC)
1539 for (i = n0; i < n1; i++)
1541 rbi = born->bRad[i];
1542 rb[i] = rbi * rbi * born->drobc[i] * dvda[i];
1546 for (i = 0; i < nl->nri; i++)
1550 nj0 = nl->jindex[i];
1551 nj1 = nl->jindex[i+1];
1553 /* Load shifts for this list */
1554 shift = nl->shift[i];
1555 shX = shift_vec[shift][0];
1556 shY = shift_vec[shift][1];
1557 shZ = shift_vec[shift][2];
1559 /* Load atom i coordinates, add shift vectors */
1560 ix1 = shX + x[ai][0];
1561 iy1 = shY + x[ai][1];
1562 iz1 = shZ + x[ai][2];
1570 for (k = nj0; k < nj1 && nl->jjnr[k] >= 0; k++)
1584 fgb = rbai*dadx[n++];
1585 fgb_ai = rbaj*dadx[n++];
1587 /* Total force between ai and aj is the sum of ai->aj and aj->ai */
1598 /* Update force on atom aj */
1599 t[aj][0] = t[aj][0] - tx;
1600 t[aj][1] = t[aj][1] - ty;
1601 t[aj][2] = t[aj][2] - tz;
1604 /* Update force and shift forces on atom ai */
1605 t[ai][0] = t[ai][0] + fix1;
1606 t[ai][1] = t[ai][1] + fiy1;
1607 t[ai][2] = t[ai][2] + fiz1;
1609 fshift[shift][0] = fshift[shift][0] + fix1;
1610 fshift[shift][1] = fshift[shift][1] + fiy1;
1611 fshift[shift][2] = fshift[shift][2] + fiz1;
1620 calc_gb_forces(t_commrec *cr, t_mdatoms *md, gmx_genborn_t *born, gmx_localtop_t *top,
1621 rvec x[], rvec f[], t_forcerec *fr, t_idef *idef, int gb_algorithm, int sa_algorithm, t_nrnb *nrnb,
1622 const t_pbc *pbc, const t_graph *graph, gmx_enerdata_t *enerd)
1629 const t_pbc *pbc_null;
1640 if (sa_algorithm == esaAPPROX)
1642 /* Do a simple ACE type approximation for the non-polar solvation */
1643 enerd->term[F_NPSOLVATION] += calc_gb_nonpolar(cr, fr, born->nr, born, top, fr->dvda, md);
1646 /* Calculate the bonded GB-interactions using either table or analytical formula */
1647 enerd->term[F_GBPOL] += gb_bonds_tab(x, f, fr->fshift, md->chargeA, &(fr->gbtabscale),
1648 fr->invsqrta, fr->dvda, fr->gbtab.data, idef, born->epsilon_r, born->gb_epsilon_solvent, fr->epsfac, pbc_null, graph);
1650 /* Calculate self corrections to the GB energies - currently only A state used! (FIXME) */
1651 enerd->term[F_GBPOL] += calc_gb_selfcorrections(cr, born->nr, md->chargeA, born, fr->dvda,fr->epsfac);
1653 /* If parallel, sum the derivative of the potential w.r.t the born radii */
1656 gmx_sum(md->nr, fr->dvda, cr);
1658 else if (DOMAINDECOMP(cr))
1660 dd_atom_sum_real(cr->dd, fr->dvda);
1661 dd_atom_spread_real(cr->dd, fr->dvda);
1666 #if 0 && defined (GMX_X86_SSE2)
1667 if (fr->use_acceleration)
1670 genborn_allvsall_calc_chainrule_sse2_double(fr, md, born, x[0], f[0], gb_algorithm, fr->AllvsAll_workgb);
1672 genborn_allvsall_calc_chainrule_sse2_single(fr, md, born, x[0], f[0], gb_algorithm, fr->AllvsAll_workgb);
1677 genborn_allvsall_calc_chainrule(fr, md, born, x[0], f[0], gb_algorithm, fr->AllvsAll_workgb);
1680 genborn_allvsall_calc_chainrule(fr, md, born, x[0], f[0], gb_algorithm, fr->AllvsAll_workgb);
1682 cnt = md->homenr*(md->nr/2+1);
1683 /* 9 flops for outer loop, 15 for inner */
1684 inc_nrnb(nrnb, eNR_BORN_AVA_CHAINRULE, md->homenr*9+cnt*15);
1688 #if 0 && defined (GMX_X86_SSE2)
1689 if (fr->use_acceleration)
1692 calc_gb_chainrule_sse2_double(fr->natoms_force, &(fr->gblist), fr->dadx, fr->dvda, x[0],
1693 f[0], fr->fshift[0], fr->shift_vec[0], gb_algorithm, born, md);
1695 calc_gb_chainrule_sse2_single(fr->natoms_force, &(fr->gblist), fr->dadx, fr->dvda, x[0],
1696 f[0], fr->fshift[0], fr->shift_vec[0], gb_algorithm, born, md);
1701 calc_gb_chainrule(fr->natoms_force, &(fr->gblist), fr->dadx, fr->dvda,
1702 x, f, fr->fshift, fr->shift_vec, gb_algorithm, born, md);
1705 calc_gb_chainrule(fr->natoms_force, &(fr->gblist), fr->dadx, fr->dvda,
1706 x, f, fr->fshift, fr->shift_vec, gb_algorithm, born);
1711 /* 9 flops for outer loop, 15 for inner */
1712 inc_nrnb(nrnb, eNR_BORN_CHAINRULE, fr->gblist.nri*9+fr->gblist.nrj*15);
1716 static void add_j_to_gblist(gbtmpnbl_t *list, int aj)
1718 if (list->naj >= list->aj_nalloc)
1720 list->aj_nalloc = over_alloc_large(list->naj+1);
1721 srenew(list->aj, list->aj_nalloc);
1724 list->aj[list->naj++] = aj;
1727 static gbtmpnbl_t *find_gbtmplist(struct gbtmpnbls *lists, int shift)
1731 /* Search the list with the same shift, if there is one */
1733 while (ind < lists->nlist && shift != lists->list[ind].shift)
1737 if (ind == lists->nlist)
1739 if (lists->nlist == lists->list_nalloc)
1741 lists->list_nalloc++;
1742 srenew(lists->list, lists->list_nalloc);
1743 for (i = lists->nlist; i < lists->list_nalloc; i++)
1745 lists->list[i].aj = NULL;
1746 lists->list[i].aj_nalloc = 0;
1751 lists->list[lists->nlist].shift = shift;
1752 lists->list[lists->nlist].naj = 0;
1756 return &lists->list[ind];
1759 static void add_bondeds_to_gblist(t_ilist *il,
1760 gmx_bool bMolPBC, t_pbc *pbc, t_graph *g, rvec *x,
1761 struct gbtmpnbls *nls)
1763 int ind, j, ai, aj, shift, found;
1769 for (ind = 0; ind < il->nr; ind += 3)
1771 ai = il->iatoms[ind+1];
1772 aj = il->iatoms[ind+2];
1777 rvec_sub(x[ai], x[aj], dx);
1778 ivec_sub(SHIFT_IVEC(g, ai), SHIFT_IVEC(g, aj), dt);
1779 shift = IVEC2IS(dt);
1783 shift = pbc_dx_aiuc(pbc, x[ai], x[aj], dx);
1786 /* Find the list for this shift or create one */
1787 list = find_gbtmplist(&nls[ai], shift);
1791 /* So that we do not add the same bond twice.
1792 * This happens with some constraints between 1-3 atoms
1793 * that are in the bond-list but should not be in the GB nb-list */
1794 for (j = 0; j < list->naj; j++)
1796 if (list->aj[j] == aj)
1806 gmx_incons("ai == aj");
1809 add_j_to_gblist(list, aj);
1815 compare_int (const void * a, const void * b)
1817 return ( *(int*)a - *(int*)b );
1822 int make_gb_nblist(t_commrec *cr, int gb_algorithm,
1823 rvec x[], matrix box,
1824 t_forcerec *fr, t_idef *idef, t_graph *graph, gmx_genborn_t *born)
1826 int i, l, ii, j, k, n, nj0, nj1, ai, aj, at0, at1, found, shift, s;
1831 struct gbtmpnbls *nls;
1832 gbtmpnbl_t *list = NULL;
1834 set_pbc(&pbc, fr->ePBC, box);
1835 nls = born->nblist_work;
1837 for (i = 0; i < born->nr; i++)
1844 set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
1847 switch (gb_algorithm)
1851 /* Loop over 1-2, 1-3 and 1-4 interactions */
1852 for (j = F_GB12; j <= F_GB14; j++)
1854 add_bondeds_to_gblist(&idef->il[j], fr->bMolPBC, &pbc, graph, x, nls);
1858 /* Loop over 1-4 interactions */
1859 add_bondeds_to_gblist(&idef->il[F_GB14], fr->bMolPBC, &pbc, graph, x, nls);
1862 gmx_incons("Unknown GB algorithm");
1865 /* Loop over the VDWQQ and VDW nblists to set up the nonbonded part of the GB list */
1866 for (n = 0; (n < fr->nnblists); n++)
1868 for (i = 0; (i < eNL_NR); i++)
1870 nblist = &(fr->nblists[n].nlist_sr[i]);
1872 if (nblist->nri > 0 && (i == eNL_VDWQQ || i == eNL_QQ))
1874 for (j = 0; j < nblist->nri; j++)
1876 ai = nblist->iinr[j];
1877 shift = nblist->shift[j];
1879 /* Find the list for this shift or create one */
1880 list = find_gbtmplist(&nls[ai], shift);
1882 nj0 = nblist->jindex[j];
1883 nj1 = nblist->jindex[j+1];
1885 /* Add all the j-atoms in the non-bonded list to the GB list */
1886 for (k = nj0; k < nj1; k++)
1888 add_j_to_gblist(list, nblist->jjnr[k]);
1895 /* Zero out some counters */
1899 fr->gblist.jindex[0] = fr->gblist.nri;
1901 for (i = 0; i < fr->natoms_force; i++)
1903 for (s = 0; s < nls[i].nlist; s++)
1905 list = &nls[i].list[s];
1907 /* Only add those atoms that actually have neighbours */
1908 if (born->use[i] != 0)
1910 fr->gblist.iinr[fr->gblist.nri] = i;
1911 fr->gblist.shift[fr->gblist.nri] = list->shift;
1914 for (k = 0; k < list->naj; k++)
1916 /* Memory allocation for jjnr */
1917 if (fr->gblist.nrj >= fr->gblist.maxnrj)
1919 fr->gblist.maxnrj += over_alloc_large(fr->gblist.maxnrj);
1923 fprintf(debug, "Increasing GB neighbourlist j size to %d\n", fr->gblist.maxnrj);
1926 srenew(fr->gblist.jjnr, fr->gblist.maxnrj);
1930 if (i == list->aj[k])
1932 gmx_incons("i == list->aj[k]");
1934 fr->gblist.jjnr[fr->gblist.nrj++] = list->aj[k];
1937 fr->gblist.jindex[fr->gblist.nri] = fr->gblist.nrj;
1944 for (i = 0; i < fr->gblist.nri; i++)
1946 nj0 = fr->gblist.jindex[i];
1947 nj1 = fr->gblist.jindex[i+1];
1948 ai = fr->gblist.iinr[i];
1951 for (j = nj0; j < nj1; j++)
1953 if (fr->gblist.jjnr[j] < ai)
1955 fr->gblist.jjnr[j] += fr->natoms_force;
1958 qsort(fr->gblist.jjnr+nj0, nj1-nj0, sizeof(int), compare_int);
1960 for (j = nj0; j < nj1; j++)
1962 if (fr->gblist.jjnr[j] >= fr->natoms_force)
1964 fr->gblist.jjnr[j] -= fr->natoms_force;
1974 void make_local_gb(const t_commrec *cr, gmx_genborn_t *born, int gb_algorithm)
1977 gmx_domdec_t *dd = NULL;
1979 if (DOMAINDECOMP(cr))
1987 /* Single node or particle decomp (global==local), just copy pointers and return */
1988 if (gb_algorithm == egbSTILL)
1990 born->gpol = born->gpol_globalindex;
1991 born->vsolv = born->vsolv_globalindex;
1992 born->gb_radius = born->gb_radius_globalindex;
1996 born->param = born->param_globalindex;
1997 born->gb_radius = born->gb_radius_globalindex;
2000 born->use = born->use_globalindex;
2005 /* Reallocation of local arrays if necessary */
2006 /* fr->natoms_force is equal to dd->nat_tot */
2007 if (DOMAINDECOMP(cr) && dd->nat_tot > born->nalloc)
2011 nalloc = dd->nat_tot;
2013 /* Arrays specific to different gb algorithms */
2014 if (gb_algorithm == egbSTILL)
2016 srenew(born->gpol, nalloc+3);
2017 srenew(born->vsolv, nalloc+3);
2018 srenew(born->gb_radius, nalloc+3);
2019 for (i = born->nalloc; (i < nalloc+3); i++)
2023 born->gb_radius[i] = 0;
2028 srenew(born->param, nalloc+3);
2029 srenew(born->gb_radius, nalloc+3);
2030 for (i = born->nalloc; (i < nalloc+3); i++)
2033 born->gb_radius[i] = 0;
2037 /* All gb-algorithms use the array for vsites exclusions */
2038 srenew(born->use, nalloc+3);
2039 for (i = born->nalloc; (i < nalloc+3); i++)
2044 born->nalloc = nalloc;
2047 /* With dd, copy algorithm specific arrays */
2048 if (gb_algorithm == egbSTILL)
2050 for (i = at0; i < at1; i++)
2052 born->gpol[i] = born->gpol_globalindex[dd->gatindex[i]];
2053 born->vsolv[i] = born->vsolv_globalindex[dd->gatindex[i]];
2054 born->gb_radius[i] = born->gb_radius_globalindex[dd->gatindex[i]];
2055 born->use[i] = born->use_globalindex[dd->gatindex[i]];
2060 for (i = at0; i < at1; i++)
2062 born->param[i] = born->param_globalindex[dd->gatindex[i]];
2063 born->gb_radius[i] = born->gb_radius_globalindex[dd->gatindex[i]];
2064 born->use[i] = born->use_globalindex[dd->gatindex[i]];