2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2004, The GROMACS development team.
6 * Copyright (c) 2013,2014, 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.
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.
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.
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.
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.
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.
37 /* This file is completely threadsafe - keep it that way! */
43 #include "gromacs/domdec/domdec.h"
44 #include "gromacs/fileio/gmxfio.h"
45 #include "gromacs/legacyheaders/constr.h"
46 #include "gromacs/legacyheaders/copyrite.h"
47 #include "gromacs/legacyheaders/gmx_omp_nthreads.h"
48 #include "gromacs/legacyheaders/mdrun.h"
49 #include "gromacs/legacyheaders/nrnb.h"
50 #include "gromacs/legacyheaders/types/commrec.h"
51 #include "gromacs/math/units.h"
52 #include "gromacs/math/vec.h"
53 #include "gromacs/mdlib/bitmask.h"
54 #include "gromacs/pbcutil/pbc.h"
55 #include "gromacs/topology/block.h"
56 #include "gromacs/topology/mtop_util.h"
57 #include "gromacs/utility/fatalerror.h"
58 #include "gromacs/utility/gmxomp.h"
59 #include "gromacs/utility/smalloc.h"
62 int b0; /* first constraint for this thread */
63 int b1; /* b1-1 is the last constraint for this thread */
64 int nind; /* number of indices */
65 int *ind; /* constraint index for updating atom data */
66 int nind_r; /* number of indices */
67 int *ind_r; /* constraint index for updating atom data */
68 int ind_nalloc; /* allocation size of ind and ind_r */
69 tensor vir_r_m_dr; /* temporary variable for virial calculation */
70 real dhdlambda; /* temporary variable for lambda derivative */
73 typedef struct gmx_lincsdata {
74 int ncg; /* the global number of constraints */
75 int ncg_flex; /* the global number of flexible constraints */
76 int ncg_triangle; /* the global number of constraints in triangles */
77 int nIter; /* the number of iterations */
78 int nOrder; /* the order of the matrix expansion */
79 int nc; /* the number of constraints */
80 int nc_alloc; /* the number we allocated memory for */
81 int ncc; /* the number of constraint connections */
82 int ncc_alloc; /* the number we allocated memory for */
83 real matlam; /* the FE lambda value used for filling blc and blmf */
84 real *bllen0; /* the reference distance in topology A */
85 real *ddist; /* the reference distance in top B - the r.d. in top A */
86 int *bla; /* the atom pairs involved in the constraints */
87 real *blc; /* 1/sqrt(invmass1 + invmass2) */
88 real *blc1; /* as blc, but with all masses 1 */
89 int *blnr; /* index into blbnb and blmf */
90 int *blbnb; /* list of constraint connections */
91 int ntriangle; /* the local number of constraints in triangles */
92 int *triangle; /* the list of triangle constraints */
93 int *tri_bits; /* the bits tell if the matrix element should be used */
94 int ncc_triangle; /* the number of constraint connections in triangles */
95 gmx_bool bCommIter; /* communicate before each LINCS interation */
96 real *blmf; /* matrix of mass factors for constraint connections */
97 real *blmf1; /* as blmf, but with all masses 1 */
98 real *bllen; /* the reference bond length */
99 int nth; /* The number of threads doing LINCS */
100 lincs_thread_t *th; /* LINCS thread division */
101 gmx_bitmask_t *atf; /* atom flags for thread parallelization */
102 int atf_nalloc; /* allocation size of atf */
103 /* arrays for temporary storage in the LINCS algorithm */
110 real *mlambda; /* the Lagrange multipliers * -1 */
111 /* storage for the constraint RMS relative deviation output */
115 real *lincs_rmsd_data(struct gmx_lincsdata *lincsd)
117 return lincsd->rmsd_data;
120 real lincs_rmsd(struct gmx_lincsdata *lincsd, gmx_bool bSD2)
122 if (lincsd->rmsd_data[0] > 0)
124 return sqrt(lincsd->rmsd_data[bSD2 ? 2 : 1]/lincsd->rmsd_data[0]);
132 /* Do a set of nrec LINCS matrix multiplications.
133 * This function will return with up to date thread-local
134 * constraint data, without an OpenMP barrier.
136 static void lincs_matrix_expand(const struct gmx_lincsdata *lincsd,
139 real *rhs1, real *rhs2, real *sol)
141 int nrec, rec, b, j, n, nr0, nr1;
143 int ntriangle, tb, bits;
144 const int *blnr = lincsd->blnr, *blbnb = lincsd->blbnb;
145 const int *triangle = lincsd->triangle, *tri_bits = lincsd->tri_bits;
147 ntriangle = lincsd->ntriangle;
148 nrec = lincsd->nOrder;
150 for (rec = 0; rec < nrec; rec++)
153 for (b = b0; b < b1; b++)
156 for (n = blnr[b]; n < blnr[b+1]; n++)
159 mvb = mvb + blcc[n]*rhs1[j];
162 sol[b] = sol[b] + mvb;
167 } /* nrec*(ncons+2*nrtot) flops */
171 /* Perform an extra nrec recursions for only the constraints
172 * involved in rigid triangles.
173 * In this way their accuracy should come close to those of the other
174 * constraints, since traingles of constraints can produce eigenvalues
175 * around 0.7, while the effective eigenvalue for bond constraints
176 * is around 0.4 (and 0.7*0.7=0.5).
178 /* We need to copy the temporary array, since only the elements
179 * for constraints involved in triangles are updated and then
180 * the pointers are swapped. This saving copying the whole arrary.
181 * We need barrier as other threads might still be reading from rhs2.
184 for (b = b0; b < b1; b++)
191 for (rec = 0; rec < nrec; rec++)
193 for (tb = 0; tb < ntriangle; tb++)
200 for (n = nr0; n < nr1; n++)
202 if (bits & (1<<(n-nr0)))
205 mvb = mvb + blcc[n]*rhs1[j];
209 sol[b] = sol[b] + mvb;
215 } /* flops count is missing here */
217 /* We need a barrier here as the calling routine will continue
218 * to operate on the thread-local constraints without barrier.
224 static void lincs_update_atoms_noind(int ncons, const int *bla,
226 const real *fac, rvec *r,
231 real mvb, im1, im2, tmp0, tmp1, tmp2;
235 for (b = 0; b < ncons; b++)
251 } /* 16 ncons flops */
255 for (b = 0; b < ncons; b++)
273 static void lincs_update_atoms_ind(int ncons, const int *ind, const int *bla,
275 const real *fac, rvec *r,
280 real mvb, im1, im2, tmp0, tmp1, tmp2;
284 for (bi = 0; bi < ncons; bi++)
301 } /* 16 ncons flops */
305 for (bi = 0; bi < ncons; bi++)
320 } /* 16 ncons flops */
324 static void lincs_update_atoms(struct gmx_lincsdata *li, int th,
326 const real *fac, rvec *r,
332 /* Single thread, we simply update for all constraints */
333 lincs_update_atoms_noind(li->nc, li->bla, prefac, fac, r, invmass, x);
337 /* Update the atom vector components for our thread local
338 * constraints that only access our local atom range.
339 * This can be done without a barrier.
341 lincs_update_atoms_ind(li->th[th].nind, li->th[th].ind,
342 li->bla, prefac, fac, r, invmass, x);
344 if (li->th[li->nth].nind > 0)
346 /* Update the constraints that operate on atoms
347 * in multiple thread atom blocks on the master thread.
352 lincs_update_atoms_ind(li->th[li->nth].nind,
354 li->bla, prefac, fac, r, invmass, x);
360 /* LINCS projection, works on derivatives of the coordinates */
361 static void do_lincsp(rvec *x, rvec *f, rvec *fp, t_pbc *pbc,
362 struct gmx_lincsdata *lincsd, int th,
364 int econq, gmx_bool bCalcDHDL,
365 gmx_bool bCalcVir, tensor rmdf)
367 int b0, b1, b, i, j, k, n;
368 real tmp0, tmp1, tmp2, mvb;
370 int *bla, *blnr, *blbnb;
372 real *blc, *blmf, *blcc, *rhs1, *rhs2, *sol;
374 b0 = lincsd->th[th].b0;
375 b1 = lincsd->th[th].b1;
380 blbnb = lincsd->blbnb;
381 if (econq != econqForce)
383 /* Use mass-weighted parameters */
389 /* Use non mass-weighted parameters */
391 blmf = lincsd->blmf1;
393 blcc = lincsd->tmpncc;
398 /* Compute normalized i-j vectors */
401 for (b = b0; b < b1; b++)
403 pbc_dx_aiuc(pbc, x[bla[2*b]], x[bla[2*b+1]], dx);
409 for (b = b0; b < b1; b++)
411 rvec_sub(x[bla[2*b]], x[bla[2*b+1]], dx);
413 } /* 16 ncons flops */
417 for (b = b0; b < b1; b++)
424 for (n = blnr[b]; n < blnr[b+1]; n++)
427 blcc[n] = blmf[n]*(tmp0*r[k][0] + tmp1*r[k][1] + tmp2*r[k][2]);
429 mvb = blc[b]*(tmp0*(f[i][0] - f[j][0]) +
430 tmp1*(f[i][1] - f[j][1]) +
431 tmp2*(f[i][2] - f[j][2]));
436 /* Together: 23*ncons + 6*nrtot flops */
438 lincs_matrix_expand(lincsd, b0, b1, blcc, rhs1, rhs2, sol);
439 /* nrec*(ncons+2*nrtot) flops */
441 if (econq == econqDeriv_FlexCon)
443 /* We only want to constraint the flexible constraints,
444 * so we mask out the normal ones by setting sol to 0.
446 for (b = b0; b < b1; b++)
448 if (!(lincsd->bllen0[b] == 0 && lincsd->ddist[b] == 0))
455 /* We multiply sol by blc, so we can use lincs_update_atoms for OpenMP */
456 for (b = b0; b < b1; b++)
461 /* When constraining forces, we should not use mass weighting,
462 * so we pass invmass=NULL, which results in the use of 1 for all atoms.
464 lincs_update_atoms(lincsd, th, 1.0, sol, r,
465 (econq != econqForce) ? invmass : NULL, fp);
472 for (b = b0; b < b1; b++)
474 dhdlambda -= sol[b]*lincsd->ddist[b];
477 lincsd->th[th].dhdlambda = dhdlambda;
482 /* Constraint virial,
483 * determines sum r_bond x delta f,
484 * where delta f is the constraint correction
485 * of the quantity that is being constrained.
487 for (b = b0; b < b1; b++)
489 mvb = lincsd->bllen[b]*sol[b];
490 for (i = 0; i < DIM; i++)
493 for (j = 0; j < DIM; j++)
495 rmdf[i][j] += tmp1*r[b][j];
498 } /* 23 ncons flops */
502 static void do_lincs(rvec *x, rvec *xp, matrix box, t_pbc *pbc,
503 struct gmx_lincsdata *lincsd, int th,
507 real wangle, int *warn,
509 gmx_bool bCalcVir, tensor vir_r_m_dr)
511 int b0, b1, b, i, j, k, n, iter;
512 real tmp0, tmp1, tmp2, mvb, rlen, len, len2, dlen2, wfac;
514 int *bla, *blnr, *blbnb;
516 real *blc, *blmf, *bllen, *blcc, *rhs1, *rhs2, *sol, *blc_sol, *mlambda;
519 b0 = lincsd->th[th].b0;
520 b1 = lincsd->th[th].b1;
525 blbnb = lincsd->blbnb;
528 bllen = lincsd->bllen;
529 blcc = lincsd->tmpncc;
533 blc_sol = lincsd->tmp4;
534 mlambda = lincsd->mlambda;
536 if (DOMAINDECOMP(cr) && cr->dd->constraints)
538 nlocat = dd_constraints_nlocalatoms(cr->dd);
547 /* Compute normalized i-j vectors */
548 for (b = b0; b < b1; b++)
550 pbc_dx_aiuc(pbc, x[bla[2*b]], x[bla[2*b+1]], dx);
554 for (b = b0; b < b1; b++)
556 for (n = blnr[b]; n < blnr[b+1]; n++)
558 blcc[n] = blmf[n]*iprod(r[b], r[blbnb[n]]);
560 pbc_dx_aiuc(pbc, xp[bla[2*b]], xp[bla[2*b+1]], dx);
561 mvb = blc[b]*(iprod(r[b], dx) - bllen[b]);
568 /* Compute normalized i-j vectors */
569 for (b = b0; b < b1; b++)
573 tmp0 = x[i][0] - x[j][0];
574 tmp1 = x[i][1] - x[j][1];
575 tmp2 = x[i][2] - x[j][2];
576 rlen = gmx_invsqrt(tmp0*tmp0+tmp1*tmp1+tmp2*tmp2);
580 } /* 16 ncons flops */
583 for (b = b0; b < b1; b++)
591 for (n = blnr[b]; n < blnr[b+1]; n++)
594 blcc[n] = blmf[n]*(tmp0*r[k][0] + tmp1*r[k][1] + tmp2*r[k][2]);
596 mvb = blc[b]*(tmp0*(xp[i][0] - xp[j][0]) +
597 tmp1*(xp[i][1] - xp[j][1]) +
598 tmp2*(xp[i][2] - xp[j][2]) - len);
603 /* Together: 26*ncons + 6*nrtot flops */
606 lincs_matrix_expand(lincsd, b0, b1, blcc, rhs1, rhs2, sol);
607 /* nrec*(ncons+2*nrtot) flops */
609 for (b = b0; b < b1; b++)
611 mlambda[b] = blc[b]*sol[b];
614 /* Update the coordinates */
615 lincs_update_atoms(lincsd, th, 1.0, mlambda, r, invmass, xp);
618 ******** Correction for centripetal effects ********
621 wfac = cos(DEG2RAD*wangle);
624 for (iter = 0; iter < lincsd->nIter; iter++)
626 if ((lincsd->bCommIter && DOMAINDECOMP(cr) && cr->dd->constraints))
631 /* Communicate the corrected non-local coordinates */
632 if (DOMAINDECOMP(cr))
634 dd_move_x_constraints(cr->dd, box, xp, NULL, FALSE);
640 for (b = b0; b < b1; b++)
645 pbc_dx_aiuc(pbc, xp[bla[2*b]], xp[bla[2*b+1]], dx);
649 rvec_sub(xp[bla[2*b]], xp[bla[2*b+1]], dx);
652 dlen2 = 2*len2 - norm2(dx);
653 if (dlen2 < wfac*len2 && (nlocat == NULL || nlocat[b]))
655 /* not race free - see detailed comment in caller */
660 mvb = blc[b]*(len - dlen2*gmx_invsqrt(dlen2));
668 } /* 20*ncons flops */
670 lincs_matrix_expand(lincsd, b0, b1, blcc, rhs1, rhs2, sol);
671 /* nrec*(ncons+2*nrtot) flops */
673 for (b = b0; b < b1; b++)
680 /* Update the coordinates */
681 lincs_update_atoms(lincsd, th, 1.0, blc_sol, r, invmass, xp);
683 /* nit*ncons*(37+9*nrec) flops */
687 /* Update the velocities */
688 lincs_update_atoms(lincsd, th, invdt, mlambda, r, invmass, v);
692 if (nlocat != NULL && (bCalcDHDL || bCalcVir))
694 /* In lincs_update_atoms threads might cross-read mlambda */
697 /* Only account for local atoms */
698 for (b = b0; b < b1; b++)
700 mlambda[b] *= 0.5*nlocat[b];
709 for (b = b0; b < b1; b++)
711 /* Note that this this is dhdl*dt^2, the dt^2 factor is corrected
712 * later after the contributions are reduced over the threads.
714 dhdl -= lincsd->mlambda[b]*lincsd->ddist[b];
716 lincsd->th[th].dhdlambda = dhdl;
721 /* Constraint virial */
722 for (b = b0; b < b1; b++)
724 tmp0 = -bllen[b]*mlambda[b];
725 for (i = 0; i < DIM; i++)
728 for (j = 0; j < DIM; j++)
730 vir_r_m_dr[i][j] -= tmp1*r[b][j];
733 } /* 22 ncons flops */
737 * 26*ncons + 6*nrtot + nrec*(ncons+2*nrtot)
738 * + nit * (20*ncons + nrec*(ncons+2*nrtot) + 17 ncons)
740 * (26+nrec)*ncons + (6+2*nrec)*nrtot
741 * + nit * ((37+nrec)*ncons + 2*nrec*nrtot)
743 * (63+nrec)*ncons + (6+4*nrec)*nrtot
747 void set_lincs_matrix(struct gmx_lincsdata *li, real *invmass, real lambda)
749 int i, a1, a2, n, k, sign, center;
751 const real invsqrt2 = 0.7071067811865475244;
753 for (i = 0; (i < li->nc); i++)
757 li->blc[i] = gmx_invsqrt(invmass[a1] + invmass[a2]);
758 li->blc1[i] = invsqrt2;
761 /* Construct the coupling coefficient matrix blmf */
763 li->ncc_triangle = 0;
764 for (i = 0; (i < li->nc); i++)
768 for (n = li->blnr[i]; (n < li->blnr[i+1]); n++)
771 if (a1 == li->bla[2*k] || a2 == li->bla[2*k+1])
779 if (a1 == li->bla[2*k] || a1 == li->bla[2*k+1])
789 li->blmf[n] = sign*invmass[center]*li->blc[i]*li->blc[k];
790 li->blmf1[n] = sign*0.5;
791 if (li->ncg_triangle > 0)
793 /* Look for constraint triangles */
794 for (nk = li->blnr[k]; (nk < li->blnr[k+1]); nk++)
797 if (kk != i && kk != k &&
798 (li->bla[2*kk] == end || li->bla[2*kk+1] == end))
800 if (li->ntriangle == 0 ||
801 li->triangle[li->ntriangle-1] < i)
803 /* Add this constraint to the triangle list */
804 li->triangle[li->ntriangle] = i;
805 li->tri_bits[li->ntriangle] = 0;
807 if (li->blnr[i+1] - li->blnr[i] > static_cast<int>(sizeof(li->tri_bits[0])*8 - 1))
809 gmx_fatal(FARGS, "A constraint is connected to %d constraints, this is more than the %d allowed for constraints participating in triangles",
810 li->blnr[i+1] - li->blnr[i],
811 sizeof(li->tri_bits[0])*8-1);
814 li->tri_bits[li->ntriangle-1] |= (1<<(n-li->blnr[i]));
824 fprintf(debug, "Of the %d constraints %d participate in triangles\n",
825 li->nc, li->ntriangle);
826 fprintf(debug, "There are %d couplings of which %d in triangles\n",
827 li->ncc, li->ncc_triangle);
831 * so we know with which lambda value the masses have been set.
836 static int count_triangle_constraints(t_ilist *ilist, t_blocka *at2con)
839 int c0, a00, a01, n1, c1, a10, a11, ac1, n2, c2, a20, a21;
842 t_iatom *ia1, *ia2, *iap;
844 ncon1 = ilist[F_CONSTR].nr/3;
845 ncon_tot = ncon1 + ilist[F_CONSTRNC].nr/3;
847 ia1 = ilist[F_CONSTR].iatoms;
848 ia2 = ilist[F_CONSTRNC].iatoms;
851 for (c0 = 0; c0 < ncon_tot; c0++)
854 iap = constr_iatomptr(ncon1, ia1, ia2, c0);
857 for (n1 = at2con->index[a01]; n1 < at2con->index[a01+1]; n1++)
862 iap = constr_iatomptr(ncon1, ia1, ia2, c1);
873 for (n2 = at2con->index[ac1]; n2 < at2con->index[ac1+1]; n2++)
876 if (c2 != c0 && c2 != c1)
878 iap = constr_iatomptr(ncon1, ia1, ia2, c2);
881 if (a20 == a00 || a21 == a00)
895 return ncon_triangle;
898 static gmx_bool more_than_two_sequential_constraints(const t_ilist *ilist,
899 const t_blocka *at2con)
901 t_iatom *ia1, *ia2, *iap;
902 int ncon1, ncon_tot, c;
904 gmx_bool bMoreThanTwoSequentialConstraints;
906 ncon1 = ilist[F_CONSTR].nr/3;
907 ncon_tot = ncon1 + ilist[F_CONSTRNC].nr/3;
909 ia1 = ilist[F_CONSTR].iatoms;
910 ia2 = ilist[F_CONSTRNC].iatoms;
912 bMoreThanTwoSequentialConstraints = FALSE;
913 for (c = 0; c < ncon_tot && !bMoreThanTwoSequentialConstraints; c++)
915 iap = constr_iatomptr(ncon1, ia1, ia2, c);
918 /* Check if this constraint has constraints connected at both atoms */
919 if (at2con->index[a1+1] - at2con->index[a1] > 1 &&
920 at2con->index[a2+1] - at2con->index[a2] > 1)
922 bMoreThanTwoSequentialConstraints = TRUE;
926 return bMoreThanTwoSequentialConstraints;
929 static int int_comp(const void *a, const void *b)
931 return (*(int *)a) - (*(int *)b);
934 gmx_lincsdata_t init_lincs(FILE *fplog, gmx_mtop_t *mtop,
935 int nflexcon_global, t_blocka *at2con,
936 gmx_bool bPLINCS, int nIter, int nProjOrder)
938 struct gmx_lincsdata *li;
944 fprintf(fplog, "\nInitializing%s LINear Constraint Solver\n",
945 bPLINCS ? " Parallel" : "");
951 gmx_mtop_ftype_count(mtop, F_CONSTR) +
952 gmx_mtop_ftype_count(mtop, F_CONSTRNC);
953 li->ncg_flex = nflexcon_global;
956 li->nOrder = nProjOrder;
958 li->ncg_triangle = 0;
959 li->bCommIter = FALSE;
960 for (mb = 0; mb < mtop->nmolblock; mb++)
962 molt = &mtop->moltype[mtop->molblock[mb].type];
964 mtop->molblock[mb].nmol*
965 count_triangle_constraints(molt->ilist,
966 &at2con[mtop->molblock[mb].type]);
967 if (bPLINCS && li->bCommIter == FALSE)
969 /* Check if we need to communicate not only before LINCS,
970 * but also before each iteration.
971 * The check for only two sequential constraints is only
972 * useful for the common case of H-bond only constraints.
973 * With more effort we could also make it useful for small
974 * molecules with nr. sequential constraints <= nOrder-1.
976 li->bCommIter = (li->nOrder < 1 || more_than_two_sequential_constraints(molt->ilist, &at2con[mtop->molblock[mb].type]));
979 if (debug && bPLINCS)
981 fprintf(debug, "PLINCS communication before each iteration: %d\n",
985 /* LINCS can run on any number of threads.
986 * Currently the number is fixed for the whole simulation,
987 * but it could be set in set_lincs().
989 li->nth = gmx_omp_nthreads_get(emntLINCS);
996 /* Allocate an extra elements for "thread-overlap" constraints */
997 snew(li->th, li->nth+1);
1001 fprintf(debug, "LINCS: using %d threads\n", li->nth);
1004 if (bPLINCS || li->ncg_triangle > 0)
1006 please_cite(fplog, "Hess2008a");
1010 please_cite(fplog, "Hess97a");
1015 fprintf(fplog, "The number of constraints is %d\n", li->ncg);
1018 fprintf(fplog, "There are inter charge-group constraints,\n"
1019 "will communicate selected coordinates each lincs iteration\n");
1021 if (li->ncg_triangle > 0)
1024 "%d constraints are involved in constraint triangles,\n"
1025 "will apply an additional matrix expansion of order %d for couplings\n"
1026 "between constraints inside triangles\n",
1027 li->ncg_triangle, li->nOrder);
1034 /* Sets up the work division over the threads */
1035 static void lincs_thread_setup(struct gmx_lincsdata *li, int natoms)
1037 lincs_thread_t *li_m;
1042 if (natoms > li->atf_nalloc)
1044 li->atf_nalloc = over_alloc_large(natoms);
1045 srenew(li->atf, li->atf_nalloc);
1049 /* Clear the atom flags */
1050 for (a = 0; a < natoms; a++)
1052 bitmask_clear(&atf[a]);
1055 if (li->nth > BITMASK_SIZE)
1057 gmx_fatal(FARGS, "More than %d threads is not supported for LINCS.", BITMASK_SIZE);
1060 for (th = 0; th < li->nth; th++)
1062 lincs_thread_t *li_th;
1065 li_th = &li->th[th];
1067 /* The constraints are divided equally over the threads */
1068 li_th->b0 = (li->nc* th )/li->nth;
1069 li_th->b1 = (li->nc*(th+1))/li->nth;
1071 /* For each atom set a flag for constraints from each */
1072 for (b = li_th->b0; b < li_th->b1; b++)
1074 bitmask_set_bit(&atf[li->bla[b*2]], th);
1075 bitmask_set_bit(&atf[li->bla[b*2+1]], th);
1079 #pragma omp parallel for num_threads(li->nth) schedule(static)
1080 for (th = 0; th < li->nth; th++)
1082 lincs_thread_t *li_th;
1086 li_th = &li->th[th];
1088 if (li_th->b1 - li_th->b0 > li_th->ind_nalloc)
1090 li_th->ind_nalloc = over_alloc_large(li_th->b1-li_th->b0);
1091 srenew(li_th->ind, li_th->ind_nalloc);
1092 srenew(li_th->ind_r, li_th->ind_nalloc);
1095 bitmask_init_low_bits(&mask, th);
1099 for (b = li_th->b0; b < li_th->b1; b++)
1101 /* We let the constraint with the lowest thread index
1102 * operate on atoms with constraints from multiple threads.
1104 if (bitmask_is_disjoint(atf[li->bla[b*2]], mask) &&
1105 bitmask_is_disjoint(atf[li->bla[b*2+1]], mask))
1107 /* Add the constraint to the local atom update index */
1108 li_th->ind[li_th->nind++] = b;
1112 /* Add the constraint to the rest block */
1113 li_th->ind_r[li_th->nind_r++] = b;
1118 /* We need to copy all constraints which have not be assigned
1119 * to a thread to a separate list which will be handled by one thread.
1121 li_m = &li->th[li->nth];
1124 for (th = 0; th < li->nth; th++)
1126 lincs_thread_t *li_th;
1129 li_th = &li->th[th];
1131 if (li_m->nind + li_th->nind_r > li_m->ind_nalloc)
1133 li_m->ind_nalloc = over_alloc_large(li_m->nind+li_th->nind_r);
1134 srenew(li_m->ind, li_m->ind_nalloc);
1137 for (b = 0; b < li_th->nind_r; b++)
1139 li_m->ind[li_m->nind++] = li_th->ind_r[b];
1144 fprintf(debug, "LINCS thread %d: %d constraints\n",
1151 fprintf(debug, "LINCS thread r: %d constraints\n",
1157 void set_lincs(t_idef *idef, t_mdatoms *md,
1158 gmx_bool bDynamics, t_commrec *cr,
1159 struct gmx_lincsdata *li)
1161 int start, natoms, nflexcon;
1164 int i, k, ncc_alloc, ni, con, nconnect, concon;
1166 real lenA = 0, lenB;
1170 /* Zero the thread index ranges.
1171 * Otherwise without local constraints we could return with old ranges.
1173 for (i = 0; i < li->nth; i++)
1181 li->th[li->nth].nind = 0;
1184 /* This is the local topology, so there are only F_CONSTR constraints */
1185 if (idef->il[F_CONSTR].nr == 0)
1187 /* There are no constraints,
1188 * we do not need to fill any data structures.
1195 fprintf(debug, "Building the LINCS connectivity\n");
1198 if (DOMAINDECOMP(cr))
1200 if (cr->dd->constraints)
1202 dd_get_constraint_range(cr->dd, &start, &natoms);
1206 natoms = cr->dd->nat_home;
1213 natoms = md->homenr;
1215 at2con = make_at2con(start, natoms, idef->il, idef->iparams, bDynamics,
1219 if (idef->il[F_CONSTR].nr/3 > li->nc_alloc || li->nc_alloc == 0)
1221 li->nc_alloc = over_alloc_dd(idef->il[F_CONSTR].nr/3);
1222 srenew(li->bllen0, li->nc_alloc);
1223 srenew(li->ddist, li->nc_alloc);
1224 srenew(li->bla, 2*li->nc_alloc);
1225 srenew(li->blc, li->nc_alloc);
1226 srenew(li->blc1, li->nc_alloc);
1227 srenew(li->blnr, li->nc_alloc+1);
1228 srenew(li->bllen, li->nc_alloc);
1229 srenew(li->tmpv, li->nc_alloc);
1230 srenew(li->tmp1, li->nc_alloc);
1231 srenew(li->tmp2, li->nc_alloc);
1232 srenew(li->tmp3, li->nc_alloc);
1233 srenew(li->tmp4, li->nc_alloc);
1234 srenew(li->mlambda, li->nc_alloc);
1235 if (li->ncg_triangle > 0)
1237 /* This is allocating too much, but it is difficult to improve */
1238 srenew(li->triangle, li->nc_alloc);
1239 srenew(li->tri_bits, li->nc_alloc);
1243 iatom = idef->il[F_CONSTR].iatoms;
1245 ncc_alloc = li->ncc_alloc;
1248 ni = idef->il[F_CONSTR].nr/3;
1252 li->blnr[con] = nconnect;
1253 for (i = 0; i < ni; i++)
1258 lenA = idef->iparams[type].constr.dA;
1259 lenB = idef->iparams[type].constr.dB;
1260 /* Skip the flexible constraints when not doing dynamics */
1261 if (bDynamics || lenA != 0 || lenB != 0)
1263 li->bllen0[con] = lenA;
1264 li->ddist[con] = lenB - lenA;
1265 /* Set the length to the topology A length */
1266 li->bllen[con] = li->bllen0[con];
1267 li->bla[2*con] = a1;
1268 li->bla[2*con+1] = a2;
1269 /* Construct the constraint connection matrix blbnb */
1270 for (k = at2con.index[a1-start]; k < at2con.index[a1-start+1]; k++)
1272 concon = at2con.a[k];
1275 if (nconnect >= ncc_alloc)
1277 ncc_alloc = over_alloc_small(nconnect+1);
1278 srenew(li->blbnb, ncc_alloc);
1280 li->blbnb[nconnect++] = concon;
1283 for (k = at2con.index[a2-start]; k < at2con.index[a2-start+1]; k++)
1285 concon = at2con.a[k];
1288 if (nconnect+1 > ncc_alloc)
1290 ncc_alloc = over_alloc_small(nconnect+1);
1291 srenew(li->blbnb, ncc_alloc);
1293 li->blbnb[nconnect++] = concon;
1296 li->blnr[con+1] = nconnect;
1300 /* Order the blbnb matrix to optimize memory access */
1301 qsort(&(li->blbnb[li->blnr[con]]), li->blnr[con+1]-li->blnr[con],
1302 sizeof(li->blbnb[0]), int_comp);
1304 /* Increase the constraint count */
1309 done_blocka(&at2con);
1311 /* This is the real number of constraints,
1312 * without dynamics the flexible constraints are not present.
1316 li->ncc = li->blnr[con];
1319 /* Since the matrix is static, we can free some memory */
1320 ncc_alloc = li->ncc;
1321 srenew(li->blbnb, ncc_alloc);
1324 if (ncc_alloc > li->ncc_alloc)
1326 li->ncc_alloc = ncc_alloc;
1327 srenew(li->blmf, li->ncc_alloc);
1328 srenew(li->blmf1, li->ncc_alloc);
1329 srenew(li->tmpncc, li->ncc_alloc);
1334 fprintf(debug, "Number of constraints is %d, couplings %d\n",
1341 li->th[0].b1 = li->nc;
1345 lincs_thread_setup(li, md->nr);
1348 set_lincs_matrix(li, md->invmass, md->lambda);
1351 static void lincs_warning(FILE *fplog,
1352 gmx_domdec_t *dd, rvec *x, rvec *xprime, t_pbc *pbc,
1353 int ncons, int *bla, real *bllen, real wangle,
1354 int maxwarn, int *warncount)
1358 real wfac, d0, d1, cosine;
1361 wfac = cos(DEG2RAD*wangle);
1363 sprintf(buf, "bonds that rotated more than %g degrees:\n"
1364 " atom 1 atom 2 angle previous, current, constraint length\n",
1366 fprintf(stderr, "%s", buf);
1369 fprintf(fplog, "%s", buf);
1372 for (b = 0; b < ncons; b++)
1378 pbc_dx_aiuc(pbc, x[i], x[j], v0);
1379 pbc_dx_aiuc(pbc, xprime[i], xprime[j], v1);
1383 rvec_sub(x[i], x[j], v0);
1384 rvec_sub(xprime[i], xprime[j], v1);
1388 cosine = iprod(v0, v1)/(d0*d1);
1391 sprintf(buf, " %6d %6d %5.1f %8.4f %8.4f %8.4f\n",
1392 ddglatnr(dd, i), ddglatnr(dd, j),
1393 RAD2DEG*acos(cosine), d0, d1, bllen[b]);
1394 fprintf(stderr, "%s", buf);
1397 fprintf(fplog, "%s", buf);
1399 if (!gmx_isfinite(d1))
1401 gmx_fatal(FARGS, "Bond length not finite.");
1407 if (*warncount > maxwarn)
1409 too_many_constraint_warnings(econtLINCS, *warncount);
1413 static void cconerr(gmx_domdec_t *dd,
1414 int ncons, int *bla, real *bllen, rvec *x, t_pbc *pbc,
1415 real *ncons_loc, real *ssd, real *max, int *imax)
1417 real len, d, ma, ssd2, r2;
1418 int *nlocat, count, b, im;
1421 if (dd && dd->constraints)
1423 nlocat = dd_constraints_nlocalatoms(dd);
1434 for (b = 0; b < ncons; b++)
1438 pbc_dx_aiuc(pbc, x[bla[2*b]], x[bla[2*b+1]], dx);
1442 rvec_sub(x[bla[2*b]], x[bla[2*b+1]], dx);
1445 len = r2*gmx_invsqrt(r2);
1446 d = fabs(len/bllen[b]-1);
1447 if (d > ma && (nlocat == NULL || nlocat[b]))
1459 ssd2 += nlocat[b]*d*d;
1464 *ncons_loc = (nlocat ? 0.5 : 1)*count;
1465 *ssd = (nlocat ? 0.5 : 1)*ssd2;
1470 gmx_bool constrain_lincs(FILE *fplog, gmx_bool bLog, gmx_bool bEner,
1473 struct gmx_lincsdata *lincsd, t_mdatoms *md,
1475 rvec *x, rvec *xprime, rvec *min_proj,
1476 matrix box, t_pbc *pbc,
1477 real lambda, real *dvdlambda,
1478 real invdt, rvec *v,
1479 gmx_bool bCalcVir, tensor vir_r_m_dr,
1482 int maxwarn, int *warncount)
1485 char buf[STRLEN], buf2[22], buf3[STRLEN];
1486 int i, warn, p_imax;
1487 real ncons_loc, p_ssd, p_max = 0;
1493 /* This boolean should be set by a flag passed to this routine.
1494 * We can also easily check if any constraint length is changed,
1495 * if not dH/dlambda=0 and we can also set the boolean to FALSE.
1497 bCalcDHDL = (ir->efep != efepNO && dvdlambda != NULL);
1499 if (lincsd->nc == 0 && cr->dd == NULL)
1503 lincsd->rmsd_data[0] = 0;
1504 if (ir->eI == eiSD2 && v == NULL)
1512 lincsd->rmsd_data[i] = 0;
1518 if (econq == econqCoord)
1520 /* We can't use bCalcDHDL here, since NULL can be passed for dvdlambda
1521 * also with efep!=fepNO.
1523 if (ir->efep != efepNO)
1525 if (md->nMassPerturbed && lincsd->matlam != md->lambda)
1527 set_lincs_matrix(lincsd, md->invmass, md->lambda);
1530 for (i = 0; i < lincsd->nc; i++)
1532 lincsd->bllen[i] = lincsd->bllen0[i] + lambda*lincsd->ddist[i];
1536 if (lincsd->ncg_flex)
1538 /* Set the flexible constraint lengths to the old lengths */
1541 for (i = 0; i < lincsd->nc; i++)
1543 if (lincsd->bllen[i] == 0)
1545 pbc_dx_aiuc(pbc, x[lincsd->bla[2*i]], x[lincsd->bla[2*i+1]], dx);
1546 lincsd->bllen[i] = norm(dx);
1552 for (i = 0; i < lincsd->nc; i++)
1554 if (lincsd->bllen[i] == 0)
1557 sqrt(distance2(x[lincsd->bla[2*i]],
1558 x[lincsd->bla[2*i+1]]));
1566 cconerr(cr->dd, lincsd->nc, lincsd->bla, lincsd->bllen, xprime, pbc,
1567 &ncons_loc, &p_ssd, &p_max, &p_imax);
1570 /* This warn var can be updated by multiple threads
1571 * at the same time. But as we only need to detect
1572 * if a warning occured or not, this is not an issue.
1576 /* The OpenMP parallel region of constrain_lincs for coords */
1577 #pragma omp parallel num_threads(lincsd->nth)
1579 int th = gmx_omp_get_thread_num();
1581 clear_mat(lincsd->th[th].vir_r_m_dr);
1583 do_lincs(x, xprime, box, pbc, lincsd, th,
1586 ir->LincsWarnAngle, &warn,
1588 th == 0 ? vir_r_m_dr : lincsd->th[th].vir_r_m_dr);
1591 if (bLog && fplog && lincsd->nc > 0)
1593 fprintf(fplog, " Rel. Constraint Deviation: RMS MAX between atoms\n");
1594 fprintf(fplog, " Before LINCS %.6f %.6f %6d %6d\n",
1595 sqrt(p_ssd/ncons_loc), p_max,
1596 ddglatnr(cr->dd, lincsd->bla[2*p_imax]),
1597 ddglatnr(cr->dd, lincsd->bla[2*p_imax+1]));
1601 cconerr(cr->dd, lincsd->nc, lincsd->bla, lincsd->bllen, xprime, pbc,
1602 &ncons_loc, &p_ssd, &p_max, &p_imax);
1603 /* Check if we are doing the second part of SD */
1604 if (ir->eI == eiSD2 && v == NULL)
1612 lincsd->rmsd_data[0] = ncons_loc;
1613 lincsd->rmsd_data[i] = p_ssd;
1617 lincsd->rmsd_data[0] = 0;
1618 lincsd->rmsd_data[1] = 0;
1619 lincsd->rmsd_data[2] = 0;
1621 if (bLog && fplog && lincsd->nc > 0)
1624 " After LINCS %.6f %.6f %6d %6d\n\n",
1625 sqrt(p_ssd/ncons_loc), p_max,
1626 ddglatnr(cr->dd, lincsd->bla[2*p_imax]),
1627 ddglatnr(cr->dd, lincsd->bla[2*p_imax+1]));
1634 cconerr(cr->dd, lincsd->nc, lincsd->bla, lincsd->bllen, xprime, pbc,
1635 &ncons_loc, &p_ssd, &p_max, &p_imax);
1638 sprintf(buf3, " in simulation %d", cr->ms->sim);
1644 sprintf(buf, "\nStep %s, time %g (ps) LINCS WARNING%s\n"
1645 "relative constraint deviation after LINCS:\n"
1646 "rms %.6f, max %.6f (between atoms %d and %d)\n",
1647 gmx_step_str(step, buf2), ir->init_t+step*ir->delta_t,
1649 sqrt(p_ssd/ncons_loc), p_max,
1650 ddglatnr(cr->dd, lincsd->bla[2*p_imax]),
1651 ddglatnr(cr->dd, lincsd->bla[2*p_imax+1]));
1654 fprintf(fplog, "%s", buf);
1656 fprintf(stderr, "%s", buf);
1657 lincs_warning(fplog, cr->dd, x, xprime, pbc,
1658 lincsd->nc, lincsd->bla, lincsd->bllen,
1659 ir->LincsWarnAngle, maxwarn, warncount);
1661 bOK = (p_max < 0.5);
1664 if (lincsd->ncg_flex)
1666 for (i = 0; (i < lincsd->nc); i++)
1668 if (lincsd->bllen0[i] == 0 && lincsd->ddist[i] == 0)
1670 lincsd->bllen[i] = 0;
1677 /* The OpenMP parallel region of constrain_lincs for derivatives */
1678 #pragma omp parallel num_threads(lincsd->nth)
1680 int th = gmx_omp_get_thread_num();
1682 do_lincsp(x, xprime, min_proj, pbc, lincsd, th,
1683 md->invmass, econq, bCalcDHDL,
1684 bCalcVir, th == 0 ? vir_r_m_dr : lincsd->th[th].vir_r_m_dr);
1690 /* Reduce the dH/dlambda contributions over the threads */
1695 for (th = 0; th < lincsd->nth; th++)
1697 dhdlambda += lincsd->th[th].dhdlambda;
1701 /* dhdlambda contains dH/dlambda*dt^2, correct for this */
1702 /* TODO This should probably use invdt, so that sd integrator scaling works properly */
1703 dhdlambda /= ir->delta_t*ir->delta_t;
1705 *dvdlambda += dhdlambda;
1708 if (bCalcVir && lincsd->nth > 1)
1710 for (i = 1; i < lincsd->nth; i++)
1712 m_add(vir_r_m_dr, lincsd->th[i].vir_r_m_dr, vir_r_m_dr);
1716 /* count assuming nit=1 */
1717 inc_nrnb(nrnb, eNR_LINCS, lincsd->nc);
1718 inc_nrnb(nrnb, eNR_LINCSMAT, (2+lincsd->nOrder)*lincsd->ncc);
1719 if (lincsd->ntriangle > 0)
1721 inc_nrnb(nrnb, eNR_LINCSMAT, lincsd->nOrder*lincsd->ncc_triangle);
1725 inc_nrnb(nrnb, eNR_CONSTR_V, lincsd->nc*2);
1729 inc_nrnb(nrnb, eNR_CONSTR_VIR, lincsd->nc);