2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2008
5 * Copyright (c) 2012,2013, by the GROMACS development team, led by
6 * David van der Spoel, Berk Hess, Erik Lindahl, and including many
7 * others, as listed in the AUTHORS file in the top-level source
8 * directory and at http://www.gromacs.org.
10 * GROMACS is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public License
12 * as published by the Free Software Foundation; either version 2.1
13 * of the License, or (at your option) any later version.
15 * GROMACS is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with GROMACS; if not, see
22 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
23 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
25 * If you want to redistribute modifications to GROMACS, please
26 * consider that scientific software is very special. Version
27 * control is crucial - bugs must be traceable. We will be happy to
28 * consider code for inclusion in the official distribution, but
29 * derived work must not be called official GROMACS. Details are found
30 * in the README & COPYING files - if they are missing, get the
31 * official version at http://www.gromacs.org.
33 * To help us fund GROMACS development, we humbly ask that you cite
34 * the research papers on the package. Check out http://www.gromacs.org.
46 #include "domdec_network.h"
47 #include "mtop_util.h"
48 #include "gmx_ga2la.h"
50 #include "gmx_omp_nthreads.h"
65 typedef struct gmx_domdec_specat_comm {
66 /* The number of indices to receive during the setup */
68 /* The atoms to send */
69 gmx_specatsend_t spas[DIM][2];
79 /* The range in the local buffer(s) for received atoms */
83 /* The atom indices we need from the surrounding cells.
84 * We can gather the indices over nthread threads.
88 } gmx_domdec_specat_comm_t;
90 typedef struct gmx_domdec_constraints {
93 /* The fully local and connected constraints */
95 /* The global constraint number, only required for clearing gc_req */
99 /* Boolean that tells if a global constraint index has been requested */
101 /* Global to local communicated constraint atom only index */
104 /* Multi-threading stuff */
107 } gmx_domdec_constraints_t;
110 static void dd_move_f_specat(gmx_domdec_t *dd, gmx_domdec_specat_comm_t *spac,
111 rvec *f, rvec *fshift)
113 gmx_specatsend_t *spas;
115 int n, n0, n1, d, dim, dir, i;
118 gmx_bool bPBC, bScrew;
121 for (d = dd->ndim-1; d >= 0; d--)
126 /* Pulse the grid forward and backward */
127 spas = spac->spas[d];
132 /* Send and receive the coordinates */
133 dd_sendrecv2_rvec(dd, d,
134 f+n+n1, n0, vbuf, spas[0].nsend,
135 f+n, n1, vbuf+spas[0].nsend, spas[1].nsend);
136 for (dir = 0; dir < 2; dir++)
138 bPBC = ((dir == 0 && dd->ci[dim] == 0) ||
139 (dir == 1 && dd->ci[dim] == dd->nc[dim]-1));
140 bScrew = (bPBC && dd->bScrewPBC && dim == XX);
142 spas = &spac->spas[d][dir];
143 /* Sum the buffer into the required forces */
144 if (!bPBC || (!bScrew && fshift == NULL))
146 for (i = 0; i < spas->nsend; i++)
148 rvec_inc(f[spas->a[i]], *vbuf);
155 vis[dim] = (dir == 0 ? 1 : -1);
159 /* Sum and add to shift forces */
160 for (i = 0; i < spas->nsend; i++)
162 rvec_inc(f[spas->a[i]], *vbuf);
163 rvec_inc(fshift[is], *vbuf);
169 /* Rotate the forces */
170 for (i = 0; i < spas->nsend; i++)
172 f[spas->a[i]][XX] += (*vbuf)[XX];
173 f[spas->a[i]][YY] -= (*vbuf)[YY];
174 f[spas->a[i]][ZZ] -= (*vbuf)[ZZ];
177 rvec_inc(fshift[is], *vbuf);
187 /* Two cells, so we only need to communicate one way */
188 spas = &spac->spas[d][0];
190 /* Send and receive the coordinates */
191 dd_sendrecv_rvec(dd, d, dddirForward,
192 f+n, spas->nrecv, spac->vbuf, spas->nsend);
193 /* Sum the buffer into the required forces */
194 if (dd->bScrewPBC && dim == XX &&
196 dd->ci[dim] == dd->nc[dim]-1))
198 for (i = 0; i < spas->nsend; i++)
200 /* Rotate the force */
201 f[spas->a[i]][XX] += spac->vbuf[i][XX];
202 f[spas->a[i]][YY] -= spac->vbuf[i][YY];
203 f[spas->a[i]][ZZ] -= spac->vbuf[i][ZZ];
208 for (i = 0; i < spas->nsend; i++)
210 rvec_inc(f[spas->a[i]], spac->vbuf[i]);
217 void dd_move_f_vsites(gmx_domdec_t *dd, rvec *f, rvec *fshift)
221 dd_move_f_specat(dd, dd->vsite_comm, f, fshift);
225 void dd_clear_f_vsites(gmx_domdec_t *dd, rvec *f)
231 for (i = dd->vsite_comm->at_start; i < dd->vsite_comm->at_end; i++)
238 static void dd_move_x_specat(gmx_domdec_t *dd, gmx_domdec_specat_comm_t *spac,
241 rvec *x1, gmx_bool bX1IsCoord)
243 gmx_specatsend_t *spas;
244 rvec *x, *vbuf, *rbuf;
245 int nvec, v, n, nn, ns0, ns1, nr0, nr1, nr, d, dim, dir, i;
246 gmx_bool bPBC, bScrew = FALSE;
247 rvec shift = {0, 0, 0};
256 for (d = 0; d < dd->ndim; d++)
261 /* Pulse the grid forward and backward */
263 for (dir = 0; dir < 2; dir++)
265 if (dir == 0 && dd->ci[dim] == 0)
268 bScrew = (dd->bScrewPBC && dim == XX);
269 copy_rvec(box[dim], shift);
271 else if (dir == 1 && dd->ci[dim] == dd->nc[dim]-1)
274 bScrew = (dd->bScrewPBC && dim == XX);
275 for (i = 0; i < DIM; i++)
277 shift[i] = -box[dim][i];
285 spas = &spac->spas[d][dir];
286 for (v = 0; v < nvec; v++)
288 x = (v == 0 ? x0 : x1);
289 /* Copy the required coordinates to the send buffer */
290 if (!bPBC || (v == 1 && !bX1IsCoord))
293 for (i = 0; i < spas->nsend; i++)
295 copy_rvec(x[spas->a[i]], *vbuf);
301 /* Shift coordinates */
302 for (i = 0; i < spas->nsend; i++)
304 rvec_add(x[spas->a[i]], shift, *vbuf);
310 /* Shift and rotate coordinates */
311 for (i = 0; i < spas->nsend; i++)
313 (*vbuf)[XX] = x[spas->a[i]][XX] + shift[XX];
314 (*vbuf)[YY] = box[YY][YY] - x[spas->a[i]][YY] + shift[YY];
315 (*vbuf)[ZZ] = box[ZZ][ZZ] - x[spas->a[i]][ZZ] + shift[ZZ];
321 /* Send and receive the coordinates */
322 spas = spac->spas[d];
329 dd_sendrecv2_rvec(dd, d,
330 spac->vbuf+ns0, ns1, x0+n, nr1,
331 spac->vbuf, ns0, x0+n+nr1, nr0);
335 /* Communicate both vectors in one buffer */
337 dd_sendrecv2_rvec(dd, d,
338 spac->vbuf+2*ns0, 2*ns1, rbuf, 2*nr1,
339 spac->vbuf, 2*ns0, rbuf+2*nr1, 2*nr0);
340 /* Split the buffer into the two vectors */
342 for (dir = 1; dir >= 0; dir--)
344 nr = spas[dir].nrecv;
345 for (v = 0; v < 2; v++)
347 x = (v == 0 ? x0 : x1);
348 for (i = 0; i < nr; i++)
350 copy_rvec(*rbuf, x[nn+i]);
361 spas = &spac->spas[d][0];
362 /* Copy the required coordinates to the send buffer */
364 for (v = 0; v < nvec; v++)
366 x = (v == 0 ? x0 : x1);
367 if (dd->bScrewPBC && dim == XX &&
368 (dd->ci[XX] == 0 || dd->ci[XX] == dd->nc[XX]-1))
370 /* Here we only perform the rotation, the rest of the pbc
371 * is handled in the constraint or viste routines.
373 for (i = 0; i < spas->nsend; i++)
375 (*vbuf)[XX] = x[spas->a[i]][XX];
376 (*vbuf)[YY] = box[YY][YY] - x[spas->a[i]][YY];
377 (*vbuf)[ZZ] = box[ZZ][ZZ] - x[spas->a[i]][ZZ];
383 for (i = 0; i < spas->nsend; i++)
385 copy_rvec(x[spas->a[i]], *vbuf);
390 /* Send and receive the coordinates */
393 dd_sendrecv_rvec(dd, d, dddirBackward,
394 spac->vbuf, spas->nsend, x0+n, spas->nrecv);
398 /* Communicate both vectors in one buffer */
400 dd_sendrecv_rvec(dd, d, dddirBackward,
401 spac->vbuf, 2*spas->nsend, rbuf, 2*spas->nrecv);
402 /* Split the buffer into the two vectors */
404 for (v = 0; v < 2; v++)
406 x = (v == 0 ? x0 : x1);
407 for (i = 0; i < nr; i++)
409 copy_rvec(*rbuf, x[n+i]);
419 void dd_move_x_constraints(gmx_domdec_t *dd, matrix box,
420 rvec *x0, rvec *x1, gmx_bool bX1IsCoord)
422 if (dd->constraint_comm)
424 dd_move_x_specat(dd, dd->constraint_comm, box, x0, x1, bX1IsCoord);
428 void dd_move_x_vsites(gmx_domdec_t *dd, matrix box, rvec *x)
432 dd_move_x_specat(dd, dd->vsite_comm, box, x, NULL, FALSE);
436 int *dd_constraints_nlocalatoms(gmx_domdec_t *dd)
440 return dd->constraints->con_nlocat;
448 void dd_clear_local_constraint_indices(gmx_domdec_t *dd)
450 gmx_domdec_constraints_t *dc;
453 dc = dd->constraints;
455 for (i = 0; i < dc->ncon; i++)
457 dc->gc_req[dc->con_gl[i]] = 0;
460 if (dd->constraint_comm)
462 gmx_hash_clear_and_optimize(dc->ga2la);
466 void dd_clear_local_vsite_indices(gmx_domdec_t *dd)
472 gmx_hash_clear_and_optimize(dd->ga2la_vsite);
476 static int setup_specat_communication(gmx_domdec_t *dd,
478 gmx_domdec_specat_comm_t *spac,
479 gmx_hash_t ga2la_specat,
482 const char *specat_type,
485 int nsend[2], nlast, nsend_zero[2] = {0, 0}, *nsend_ptr;
486 int d, dim, ndir, dir, nr, ns, i, nrecv_local, n0, start, indr, ind, buf[2];
487 int nat_tot_specat, nat_tot_prev, nalloc_old;
488 gmx_bool bPBC, bFirst;
489 gmx_specatsend_t *spas;
493 fprintf(debug, "Begin setup_specat_communication for %s\n", specat_type);
496 /* nsend[0]: the number of atoms requested by this node only,
497 * we communicate this for more efficients checks
498 * nsend[1]: the total number of requested atoms
503 for (d = dd->ndim-1; d >= 0; d--)
505 /* Pulse the grid forward and backward */
507 bPBC = (dim < dd->npbcdim);
508 if (dd->nc[dim] == 2)
510 /* Only 2 cells, so we only need to communicate once */
517 for (dir = 0; dir < ndir; dir++)
521 ((dir == 0 && dd->ci[dim] == dd->nc[dim] - 1) ||
522 (dir == 1 && dd->ci[dim] == 0)))
524 /* No pbc: the fist/last cell should not request atoms */
525 nsend_ptr = nsend_zero;
531 /* Communicate the number of indices */
532 dd_sendrecv_int(dd, d, dir == 0 ? dddirForward : dddirBackward,
533 nsend_ptr, 2, spac->nreq[d][dir], 2);
534 nr = spac->nreq[d][dir][1];
535 if (nlast+nr > ireq->nalloc)
537 ireq->nalloc = over_alloc_dd(nlast+nr);
538 srenew(ireq->ind, ireq->nalloc);
540 /* Communicate the indices */
541 dd_sendrecv_int(dd, d, dir == 0 ? dddirForward : dddirBackward,
542 ireq->ind, nsend_ptr[1], ireq->ind+nlast, nr);
549 fprintf(debug, "Communicated the counts\n");
552 /* Search for the requested atoms and communicate the indices we have */
553 nat_tot_specat = at_start;
555 for (d = 0; d < dd->ndim; d++)
558 /* Pulse the grid forward and backward */
559 if (dd->dim[d] >= dd->npbcdim || dd->nc[dd->dim[d]] > 2)
567 nat_tot_prev = nat_tot_specat;
568 for (dir = ndir-1; dir >= 0; dir--)
570 if (nat_tot_specat > spac->bSendAtom_nalloc)
572 nalloc_old = spac->bSendAtom_nalloc;
573 spac->bSendAtom_nalloc = over_alloc_dd(nat_tot_specat);
574 srenew(spac->bSendAtom, spac->bSendAtom_nalloc);
575 for (i = nalloc_old; i < spac->bSendAtom_nalloc; i++)
577 spac->bSendAtom[i] = FALSE;
580 spas = &spac->spas[d][dir];
581 n0 = spac->nreq[d][dir][0];
582 nr = spac->nreq[d][dir][1];
585 fprintf(debug, "dim=%d, dir=%d, searching for %d atoms\n",
591 for (i = 0; i < nr; i++)
593 indr = ireq->ind[start+i];
595 /* Check if this is a home atom and if so ind will be set */
596 if (!ga2la_get_home(dd->ga2la, indr, &ind))
598 /* Search in the communicated atoms */
599 ind = gmx_hash_get_minone(ga2la_specat, indr);
603 if (i < n0 || !spac->bSendAtom[ind])
605 if (spas->nsend+1 > spas->a_nalloc)
607 spas->a_nalloc = over_alloc_large(spas->nsend+1);
608 srenew(spas->a, spas->a_nalloc);
610 /* Store the local index so we know which coordinates
613 spas->a[spas->nsend] = ind;
614 spac->bSendAtom[ind] = TRUE;
615 if (spas->nsend+1 > spac->ibuf_nalloc)
617 spac->ibuf_nalloc = over_alloc_large(spas->nsend+1);
618 srenew(spac->ibuf, spac->ibuf_nalloc);
620 /* Store the global index so we can send it now */
621 spac->ibuf[spas->nsend] = indr;
631 /* Clear the local flags */
632 for (i = 0; i < spas->nsend; i++)
634 spac->bSendAtom[spas->a[i]] = FALSE;
636 /* Send and receive the number of indices to communicate */
637 nsend[1] = spas->nsend;
638 dd_sendrecv_int(dd, d, dir == 0 ? dddirBackward : dddirForward,
642 fprintf(debug, "Send to node %d, %d (%d) indices, "
643 "receive from node %d, %d (%d) indices\n",
644 dd->neighbor[d][1-dir], nsend[1], nsend[0],
645 dd->neighbor[d][dir], buf[1], buf[0]);
648 for (i = 0; i < spas->nsend; i++)
650 fprintf(debug, " %d", spac->ibuf[i]+1);
652 fprintf(debug, "\n");
655 nrecv_local += buf[0];
656 spas->nrecv = buf[1];
657 if (nat_tot_specat + spas->nrecv > dd->gatindex_nalloc)
659 dd->gatindex_nalloc =
660 over_alloc_dd(nat_tot_specat + spas->nrecv);
661 srenew(dd->gatindex, dd->gatindex_nalloc);
663 /* Send and receive the indices */
664 dd_sendrecv_int(dd, d, dir == 0 ? dddirBackward : dddirForward,
665 spac->ibuf, spas->nsend,
666 dd->gatindex+nat_tot_specat, spas->nrecv);
667 nat_tot_specat += spas->nrecv;
670 /* Allocate the x/f communication buffers */
671 ns = spac->spas[d][0].nsend;
672 nr = spac->spas[d][0].nrecv;
675 ns += spac->spas[d][1].nsend;
676 nr += spac->spas[d][1].nrecv;
678 if (vbuf_fac*ns > spac->vbuf_nalloc)
680 spac->vbuf_nalloc = over_alloc_dd(vbuf_fac*ns);
681 srenew(spac->vbuf, spac->vbuf_nalloc);
683 if (vbuf_fac == 2 && vbuf_fac*nr > spac->vbuf2_nalloc)
685 spac->vbuf2_nalloc = over_alloc_dd(vbuf_fac*nr);
686 srenew(spac->vbuf2, spac->vbuf2_nalloc);
689 /* Make a global to local index for the communication atoms */
690 for (i = nat_tot_prev; i < nat_tot_specat; i++)
692 gmx_hash_change_or_set(ga2la_specat, dd->gatindex[i], i);
696 /* Check that in the end we got the number of atoms we asked for */
697 if (nrecv_local != ireq->n)
701 fprintf(debug, "Requested %d, received %d (tot recv %d)\n",
702 ireq->n, nrecv_local, nat_tot_specat-at_start);
705 for (i = 0; i < ireq->n; i++)
707 ind = gmx_hash_get_minone(ga2la_specat, ireq->ind[i]);
708 fprintf(debug, " %s%d",
709 (ind >= 0) ? "" : "!",
712 fprintf(debug, "\n");
715 fprintf(stderr, "\nDD cell %d %d %d: Neighboring cells do not have atoms:",
716 dd->ci[XX], dd->ci[YY], dd->ci[ZZ]);
717 for (i = 0; i < ireq->n; i++)
719 if (gmx_hash_get_minone(ga2la_specat, ireq->ind[i]) < 0)
721 fprintf(stderr, " %d", ireq->ind[i]+1);
724 fprintf(stderr, "\n");
725 gmx_fatal(FARGS, "DD cell %d %d %d could only obtain %d of the %d atoms that are connected via %ss from the neighboring cells. This probably means your %s lengths are too long compared to the domain decomposition cell size. Decrease the number of domain decomposition grid cells%s%s.",
726 dd->ci[XX], dd->ci[YY], dd->ci[ZZ],
727 nrecv_local, ireq->n, specat_type,
728 specat_type, add_err,
729 dd->bGridJump ? " or use the -rcon option of mdrun" : "");
732 spac->at_start = at_start;
733 spac->at_end = nat_tot_specat;
737 fprintf(debug, "Done setup_specat_communication\n");
740 return nat_tot_specat;
743 static void walk_out(int con, int con_offset, int a, int offset, int nrec,
744 int ncon1, const t_iatom *ia1, const t_iatom *ia2,
745 const t_blocka *at2con,
746 const gmx_ga2la_t ga2la, gmx_bool bHomeConnect,
747 gmx_domdec_constraints_t *dc,
748 gmx_domdec_specat_comm_t *dcc,
752 int a1_gl, a2_gl, a_loc, i, coni, b;
755 if (dc->gc_req[con_offset+con] == 0)
757 /* Add this non-home constraint to the list */
758 if (dc->ncon+1 > dc->con_nalloc)
760 dc->con_nalloc = over_alloc_large(dc->ncon+1);
761 srenew(dc->con_gl, dc->con_nalloc);
762 srenew(dc->con_nlocat, dc->con_nalloc);
764 dc->con_gl[dc->ncon] = con_offset + con;
765 dc->con_nlocat[dc->ncon] = (bHomeConnect ? 1 : 0);
766 dc->gc_req[con_offset+con] = 1;
767 if (il_local->nr + 3 > il_local->nalloc)
769 il_local->nalloc = over_alloc_dd(il_local->nr+3);
770 srenew(il_local->iatoms, il_local->nalloc);
772 iap = constr_iatomptr(ncon1, ia1, ia2, con);
773 il_local->iatoms[il_local->nr++] = iap[0];
774 a1_gl = offset + iap[1];
775 a2_gl = offset + iap[2];
776 /* The following indexing code can probably be optizimed */
777 if (ga2la_get_home(ga2la, a1_gl, &a_loc))
779 il_local->iatoms[il_local->nr++] = a_loc;
783 /* We set this index later */
784 il_local->iatoms[il_local->nr++] = -a1_gl - 1;
786 if (ga2la_get_home(ga2la, a2_gl, &a_loc))
788 il_local->iatoms[il_local->nr++] = a_loc;
792 /* We set this index later */
793 il_local->iatoms[il_local->nr++] = -a2_gl - 1;
797 /* Check to not ask for the same atom more than once */
798 if (gmx_hash_get_minone(dc->ga2la, offset+a) == -1)
801 /* Add this non-home atom to the list */
802 if (ireq->n+1 > ireq->nalloc)
804 ireq->nalloc = over_alloc_large(ireq->n+1);
805 srenew(ireq->ind, ireq->nalloc);
807 ireq->ind[ireq->n++] = offset + a;
808 /* Temporarily mark with -2, we get the index later */
809 gmx_hash_set(dc->ga2la, offset+a, -2);
814 for (i = at2con->index[a]; i < at2con->index[a+1]; i++)
820 iap = constr_iatomptr(ncon1, ia1, ia2, coni);
829 if (!ga2la_get_home(ga2la, offset+b, &a_loc))
831 walk_out(coni, con_offset, b, offset, nrec-1,
832 ncon1, ia1, ia2, at2con,
833 ga2la, FALSE, dc, dcc, il_local, ireq);
840 static void atoms_to_settles(gmx_domdec_t *dd,
841 const gmx_mtop_t *mtop,
843 const int **at2settle_mt,
844 int cg_start, int cg_end,
849 gmx_mtop_atomlookup_t alook;
852 int cg, a, a_gl, a_glsa, a_gls[3], a_locs[3];
853 int mb, molnr, a_mol, offset;
854 const gmx_molblock_t *molb;
862 alook = gmx_mtop_atomlookup_settle_init(mtop);
864 nral = NRAL(F_SETTLE);
866 for (cg = cg_start; cg < cg_end; cg++)
868 if (GET_CGINFO_SETTLE(cginfo[cg]))
870 for (a = dd->cgindex[cg]; a < dd->cgindex[cg+1]; a++)
872 a_gl = dd->gatindex[a];
874 gmx_mtop_atomnr_to_molblock_ind(alook, a_gl, &mb, &molnr, &a_mol);
875 molb = &mtop->molblock[mb];
877 settle = at2settle_mt[molb->type][a_mol];
881 offset = a_gl - a_mol;
883 ia1 = mtop->moltype[molb->type].ilist[F_SETTLE].iatoms;
887 for (sa = 0; sa < nral; sa++)
889 a_glsa = offset + ia1[settle*(1+nral)+1+sa];
891 a_home[sa] = ga2la_get_home(ga2la, a_glsa, &a_locs[sa]);
894 if (nlocal == 0 && a_gl == a_glsa)
904 if (ils_local->nr+1+nral > ils_local->nalloc)
906 ils_local->nalloc = over_alloc_dd(ils_local->nr+1+nral);
907 srenew(ils_local->iatoms, ils_local->nalloc);
910 ils_local->iatoms[ils_local->nr++] = ia1[settle*4];
912 for (sa = 0; sa < nral; sa++)
914 if (ga2la_get_home(ga2la, a_gls[sa], &a_locs[sa]))
916 ils_local->iatoms[ils_local->nr++] = a_locs[sa];
920 ils_local->iatoms[ils_local->nr++] = -a_gls[sa] - 1;
921 /* Add this non-home atom to the list */
922 if (ireq->n+1 > ireq->nalloc)
924 ireq->nalloc = over_alloc_large(ireq->n+1);
925 srenew(ireq->ind, ireq->nalloc);
927 ireq->ind[ireq->n++] = a_gls[sa];
928 /* A check on double atom requests is
929 * not required for settle.
939 gmx_mtop_atomlookup_destroy(alook);
942 static void atoms_to_constraints(gmx_domdec_t *dd,
943 const gmx_mtop_t *mtop,
945 const t_blocka *at2con_mt, int nrec,
949 const t_blocka *at2con;
951 gmx_mtop_atomlookup_t alook;
953 gmx_molblock_t *molb;
954 t_iatom *ia1, *ia2, *iap;
955 int nhome, cg, a, a_gl, a_mol, a_loc, b_lo, offset, mb, molnr, b_mol, i, con, con_offset;
956 gmx_domdec_constraints_t *dc;
957 gmx_domdec_specat_comm_t *dcc;
959 dc = dd->constraints;
960 dcc = dd->constraint_comm;
964 alook = gmx_mtop_atomlookup_init(mtop);
967 for (cg = 0; cg < dd->ncg_home; cg++)
969 if (GET_CGINFO_CONSTR(cginfo[cg]))
971 for (a = dd->cgindex[cg]; a < dd->cgindex[cg+1]; a++)
973 a_gl = dd->gatindex[a];
975 gmx_mtop_atomnr_to_molblock_ind(alook, a_gl, &mb, &molnr, &a_mol);
976 molb = &mtop->molblock[mb];
978 ncon1 = mtop->moltype[molb->type].ilist[F_CONSTR].nr/NRAL(F_SETTLE);
980 ia1 = mtop->moltype[molb->type].ilist[F_CONSTR].iatoms;
981 ia2 = mtop->moltype[molb->type].ilist[F_CONSTRNC].iatoms;
983 /* Calculate the global constraint number offset for the molecule.
984 * This is only required for the global index to make sure
985 * that we use each constraint only once.
988 dc->molb_con_offset[mb] + molnr*dc->molb_ncon_mol[mb];
990 /* The global atom number offset for this molecule */
991 offset = a_gl - a_mol;
992 at2con = &at2con_mt[molb->type];
993 for (i = at2con->index[a_mol]; i < at2con->index[a_mol+1]; i++)
996 iap = constr_iatomptr(ncon1, ia1, ia2, con);
1005 if (ga2la_get_home(ga2la, offset+b_mol, &a_loc))
1007 /* Add this fully home constraint at the first atom */
1010 if (dc->ncon+1 > dc->con_nalloc)
1012 dc->con_nalloc = over_alloc_large(dc->ncon+1);
1013 srenew(dc->con_gl, dc->con_nalloc);
1014 srenew(dc->con_nlocat, dc->con_nalloc);
1016 dc->con_gl[dc->ncon] = con_offset + con;
1017 dc->con_nlocat[dc->ncon] = 2;
1018 if (ilc_local->nr + 3 > ilc_local->nalloc)
1020 ilc_local->nalloc = over_alloc_dd(ilc_local->nr + 3);
1021 srenew(ilc_local->iatoms, ilc_local->nalloc);
1024 ilc_local->iatoms[ilc_local->nr++] = iap[0];
1025 ilc_local->iatoms[ilc_local->nr++] = (a_gl == iap[1] ? a : b_lo);
1026 ilc_local->iatoms[ilc_local->nr++] = (a_gl == iap[1] ? b_lo : a );
1033 /* We need the nrec constraints coupled to this constraint,
1034 * so we need to walk out of the home cell by nrec+1 atoms,
1035 * since already atom bg is not locally present.
1036 * Therefore we call walk_out with nrec recursions to go
1037 * after this first call.
1039 walk_out(con, con_offset, b_mol, offset, nrec,
1040 ncon1, ia1, ia2, at2con,
1041 dd->ga2la, TRUE, dc, dcc, ilc_local, ireq);
1048 gmx_mtop_atomlookup_destroy(alook);
1053 "Constraints: home %3d border %3d atoms: %3d\n",
1054 nhome, dc->ncon-nhome,
1055 dd->constraint_comm ? ireq->n : 0);
1059 int dd_make_local_constraints(gmx_domdec_t *dd, int at_start,
1060 const gmx_mtop_t *mtop,
1062 gmx_constr_t constr, int nrec,
1065 gmx_domdec_constraints_t *dc;
1066 t_ilist *ilc_local, *ils_local;
1068 const t_blocka *at2con_mt;
1069 const int **at2settle_mt;
1070 gmx_hash_t ga2la_specat;
1074 dc = dd->constraints;
1076 ilc_local = &il_local[F_CONSTR];
1077 ils_local = &il_local[F_SETTLE];
1081 if (dd->constraint_comm)
1083 at2con_mt = atom2constraints_moltype(constr);
1084 ireq = &dd->constraint_comm->ireq[0];
1093 if (dd->bInterCGsettles)
1095 at2settle_mt = atom2settle_moltype(constr);
1100 /* Settle works inside charge groups, we assigned them already */
1101 at2settle_mt = NULL;
1104 if (at2settle_mt == NULL)
1106 atoms_to_constraints(dd, mtop, cginfo, at2con_mt, nrec,
1114 /* Do the constraints, if present, on the first thread.
1115 * Do the settles on all other threads.
1117 t0_set = ((at2con_mt != NULL && dc->nthread > 1) ? 1 : 0);
1119 #pragma omp parallel for num_threads(dc->nthread) schedule(static)
1120 for (thread = 0; thread < dc->nthread; thread++)
1122 if (at2con_mt && thread == 0)
1124 atoms_to_constraints(dd, mtop, cginfo, at2con_mt, nrec,
1128 if (thread >= t0_set)
1134 /* Distribute the settle check+assignments over
1135 * dc->nthread or dc->nthread-1 threads.
1137 cg0 = (dd->ncg_home*(thread-t0_set ))/(dc->nthread-t0_set);
1138 cg1 = (dd->ncg_home*(thread-t0_set+1))/(dc->nthread-t0_set);
1140 if (thread == t0_set)
1146 ilst = &dc->ils[thread];
1150 ireqt = &dd->constraint_comm->ireq[thread];
1156 atoms_to_settles(dd, mtop, cginfo, at2settle_mt,
1162 /* Combine the generate settles and requested indices */
1163 for (thread = 1; thread < dc->nthread; thread++)
1169 if (thread > t0_set)
1171 ilst = &dc->ils[thread];
1172 if (ils_local->nr + ilst->nr > ils_local->nalloc)
1174 ils_local->nalloc = over_alloc_large(ils_local->nr + ilst->nr);
1175 srenew(ils_local->iatoms, ils_local->nalloc);
1177 for (ia = 0; ia < ilst->nr; ia++)
1179 ils_local->iatoms[ils_local->nr+ia] = ilst->iatoms[ia];
1181 ils_local->nr += ilst->nr;
1184 ireqt = &dd->constraint_comm->ireq[thread];
1185 if (ireq->n+ireqt->n > ireq->nalloc)
1187 ireq->nalloc = over_alloc_large(ireq->n+ireqt->n);
1188 srenew(ireq->ind, ireq->nalloc);
1190 for (ia = 0; ia < ireqt->n; ia++)
1192 ireq->ind[ireq->n+ia] = ireqt->ind[ia];
1194 ireq->n += ireqt->n;
1199 fprintf(debug, "Settles: total %3d\n", ils_local->nr/4);
1203 if (dd->constraint_comm)
1208 setup_specat_communication(dd, ireq, dd->constraint_comm,
1209 dd->constraints->ga2la,
1211 "constraint", " or lincs-order");
1213 /* Fill in the missing indices */
1214 ga2la_specat = dd->constraints->ga2la;
1216 nral1 = 1 + NRAL(F_CONSTR);
1217 for (i = 0; i < ilc_local->nr; i += nral1)
1219 iap = ilc_local->iatoms + i;
1220 for (j = 1; j < nral1; j++)
1224 iap[j] = gmx_hash_get_minone(ga2la_specat, -iap[j]-1);
1229 nral1 = 1 + NRAL(F_SETTLE);
1230 for (i = 0; i < ils_local->nr; i += nral1)
1232 iap = ils_local->iatoms + i;
1233 for (j = 1; j < nral1; j++)
1237 iap[j] = gmx_hash_get_minone(ga2la_specat, -iap[j]-1);
1250 int dd_make_local_vsites(gmx_domdec_t *dd, int at_start, t_ilist *lil)
1252 gmx_domdec_specat_comm_t *spac;
1254 gmx_hash_t ga2la_specat;
1255 int ftype, nral, i, j, gat, a;
1260 spac = dd->vsite_comm;
1261 ireq = &spac->ireq[0];
1262 ga2la_specat = dd->ga2la_vsite;
1265 /* Loop over all the home vsites */
1266 for (ftype = 0; ftype < F_NRE; ftype++)
1268 if (interaction_function[ftype].flags & IF_VSITE)
1272 for (i = 0; i < lilf->nr; i += 1+nral)
1274 iatoms = lilf->iatoms + i;
1275 /* Check if we have the other atoms */
1276 for (j = 1; j < 1+nral; j++)
1280 /* This is not a home atom,
1281 * we need to ask our neighbors.
1284 /* Check to not ask for the same atom more than once */
1285 if (gmx_hash_get_minone(dd->ga2la_vsite, a) == -1)
1287 /* Add this non-home atom to the list */
1288 if (ireq->n+1 > ireq->nalloc)
1290 ireq->nalloc = over_alloc_large(ireq->n+1);
1291 srenew(ireq->ind, ireq->nalloc);
1293 ireq->ind[ireq->n++] = a;
1294 /* Temporarily mark with -2,
1295 * we get the index later.
1297 gmx_hash_set(ga2la_specat, a, -2);
1305 at_end = setup_specat_communication(dd, ireq, dd->vsite_comm, ga2la_specat,
1306 at_start, 1, "vsite", "");
1308 /* Fill in the missing indices */
1309 for (ftype = 0; ftype < F_NRE; ftype++)
1311 if (interaction_function[ftype].flags & IF_VSITE)
1315 for (i = 0; i < lilf->nr; i += 1+nral)
1317 iatoms = lilf->iatoms + i;
1318 for (j = 1; j < 1+nral; j++)
1322 iatoms[j] = gmx_hash_get_minone(ga2la_specat, -iatoms[j]-1);
1332 static gmx_domdec_specat_comm_t *specat_comm_init(int nthread)
1334 gmx_domdec_specat_comm_t *spac;
1337 spac->nthread = nthread;
1338 snew(spac->ireq, spac->nthread);
1343 void init_domdec_constraints(gmx_domdec_t *dd,
1345 gmx_constr_t constr)
1347 gmx_domdec_constraints_t *dc;
1348 gmx_molblock_t *molb;
1353 fprintf(debug, "Begin init_domdec_constraints\n");
1356 snew(dd->constraints, 1);
1357 dc = dd->constraints;
1359 snew(dc->molb_con_offset, mtop->nmolblock);
1360 snew(dc->molb_ncon_mol, mtop->nmolblock);
1363 for (mb = 0; mb < mtop->nmolblock; mb++)
1365 molb = &mtop->molblock[mb];
1366 dc->molb_con_offset[mb] = ncon;
1367 dc->molb_ncon_mol[mb] =
1368 mtop->moltype[molb->type].ilist[F_CONSTR].nr/3 +
1369 mtop->moltype[molb->type].ilist[F_CONSTRNC].nr/3;
1370 ncon += molb->nmol*dc->molb_ncon_mol[mb];
1375 snew(dc->gc_req, ncon);
1376 for (c = 0; c < ncon; c++)
1382 /* Use a hash table for the global to local index.
1383 * The number of keys is a rough estimate, it will be optimized later.
1385 dc->ga2la = gmx_hash_init(min(mtop->natoms/20,
1386 mtop->natoms/(2*dd->nnodes)));
1388 dc->nthread = gmx_omp_nthreads_get(emntDomdec);
1389 snew(dc->ils, dc->nthread);
1391 dd->constraint_comm = specat_comm_init(dc->nthread);
1394 void init_domdec_vsites(gmx_domdec_t *dd, int n_intercg_vsite)
1397 gmx_domdec_constraints_t *dc;
1401 fprintf(debug, "Begin init_domdec_vsites\n");
1404 /* Use a hash table for the global to local index.
1405 * The number of keys is a rough estimate, it will be optimized later.
1407 dd->ga2la_vsite = gmx_hash_init(min(n_intercg_vsite/20,
1408 n_intercg_vsite/(2*dd->nnodes)));
1410 dd->vsite_comm = specat_comm_init(1);