2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2005,2006,2007,2008,2009,2010,2011,2012,2013,2014,2015,2016,2017,2018,2019, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
53 #include "gromacs/domdec/collect.h"
54 #include "gromacs/domdec/dlb.h"
55 #include "gromacs/domdec/dlbtiming.h"
56 #include "gromacs/domdec/domdec_network.h"
57 #include "gromacs/domdec/ga2la.h"
58 #include "gromacs/domdec/gpuhaloexchange.h"
59 #include "gromacs/domdec/options.h"
60 #include "gromacs/domdec/partition.h"
61 #include "gromacs/gmxlib/network.h"
62 #include "gromacs/gmxlib/nrnb.h"
63 #include "gromacs/gpu_utils/gpu_utils.h"
64 #include "gromacs/hardware/hw_info.h"
65 #include "gromacs/listed_forces/manage_threading.h"
66 #include "gromacs/math/vec.h"
67 #include "gromacs/math/vectypes.h"
68 #include "gromacs/mdlib/calc_verletbuf.h"
69 #include "gromacs/mdlib/constr.h"
70 #include "gromacs/mdlib/constraintrange.h"
71 #include "gromacs/mdlib/updategroups.h"
72 #include "gromacs/mdlib/vsite.h"
73 #include "gromacs/mdtypes/commrec.h"
74 #include "gromacs/mdtypes/forceoutput.h"
75 #include "gromacs/mdtypes/inputrec.h"
76 #include "gromacs/mdtypes/mdrunoptions.h"
77 #include "gromacs/mdtypes/state.h"
78 #include "gromacs/pbcutil/ishift.h"
79 #include "gromacs/pbcutil/pbc.h"
80 #include "gromacs/pulling/pull.h"
81 #include "gromacs/timing/wallcycle.h"
82 #include "gromacs/topology/block.h"
83 #include "gromacs/topology/idef.h"
84 #include "gromacs/topology/ifunc.h"
85 #include "gromacs/topology/mtop_lookup.h"
86 #include "gromacs/topology/mtop_util.h"
87 #include "gromacs/topology/topology.h"
88 #include "gromacs/utility/basedefinitions.h"
89 #include "gromacs/utility/basenetwork.h"
90 #include "gromacs/utility/cstringutil.h"
91 #include "gromacs/utility/exceptions.h"
92 #include "gromacs/utility/fatalerror.h"
93 #include "gromacs/utility/gmxmpi.h"
94 #include "gromacs/utility/logger.h"
95 #include "gromacs/utility/real.h"
96 #include "gromacs/utility/smalloc.h"
97 #include "gromacs/utility/strconvert.h"
98 #include "gromacs/utility/stringstream.h"
99 #include "gromacs/utility/stringutil.h"
100 #include "gromacs/utility/textwriter.h"
102 #include "atomdistribution.h"
104 #include "cellsizes.h"
105 #include "distribute.h"
106 #include "domdec_constraints.h"
107 #include "domdec_internal.h"
108 #include "domdec_setup.h"
109 #include "domdec_vsite.h"
110 #include "redistribute.h"
113 // TODO remove this when moving domdec into gmx namespace
114 using gmx::DdRankOrder;
115 using gmx::DlbOption;
116 using gmx::DomdecOptions;
118 static const char *edlbs_names[int(DlbState::nr)] = { "off", "auto", "locked", "on", "on" };
120 /* The size per atom group of the cggl_flag buffer in gmx_domdec_comm_t */
123 /* The flags for the cggl_flag buffer in gmx_domdec_comm_t */
124 #define DD_FLAG_NRCG 65535
125 #define DD_FLAG_FW(d) (1<<(16+(d)*2))
126 #define DD_FLAG_BW(d) (1<<(16+(d)*2+1))
128 /* The DD zone order */
129 static const ivec dd_zo[DD_MAXZONE] =
130 {{0, 0, 0}, {1, 0, 0}, {1, 1, 0}, {0, 1, 0}, {0, 1, 1}, {0, 0, 1}, {1, 0, 1}, {1, 1, 1}};
132 /* The non-bonded zone-pair setup for domain decomposition
133 * The first number is the i-zone, the second number the first j-zone seen by
134 * this i-zone, the third number the last+1 j-zone seen by this i-zone.
135 * As is, this is for 3D decomposition, where there are 4 i-zones.
136 * With 2D decomposition use only the first 2 i-zones and a last+1 j-zone of 4.
137 * With 1D decomposition use only the first i-zone and a last+1 j-zone of 2.
140 ddNonbondedZonePairRanges[DD_MAXIZONE][3] = {{0, 0, 8},
147 #define dd_index(n,i) ((((i)[ZZ]*(n)[YY] + (i)[YY])*(n)[XX]) + (i)[XX])
149 static void index2xyz(ivec nc,int ind,ivec xyz)
151 xyz[XX] = ind % nc[XX];
152 xyz[YY] = (ind / nc[XX]) % nc[YY];
153 xyz[ZZ] = ind / (nc[YY]*nc[XX]);
157 static void ddindex2xyz(const ivec nc, int ind, ivec xyz)
159 xyz[XX] = ind / (nc[YY]*nc[ZZ]);
160 xyz[YY] = (ind / nc[ZZ]) % nc[YY];
161 xyz[ZZ] = ind % nc[ZZ];
164 static int ddcoord2ddnodeid(gmx_domdec_t *dd, ivec c)
168 const CartesianRankSetup &cartSetup = dd->comm->cartesianRankSetup;
169 const int ddindex = dd_index(dd->nc, c);
170 if (cartSetup.bCartesianPP_PME)
172 ddnodeid = cartSetup.ddindex2ddnodeid[ddindex];
174 else if (cartSetup.bCartesianPP)
177 MPI_Cart_rank(dd->mpi_comm_all, c, &ddnodeid);
188 int ddglatnr(const gmx_domdec_t *dd, int i)
198 if (i >= dd->comm->atomRanges.numAtomsTotal())
200 gmx_fatal(FARGS, "glatnr called with %d, which is larger than the local number of atoms (%d)", i, dd->comm->atomRanges.numAtomsTotal());
202 atnr = dd->globalAtomIndices[i] + 1;
208 t_block *dd_charge_groups_global(gmx_domdec_t *dd)
210 return &dd->comm->cgs_gl;
213 gmx::ArrayRef<const gmx::RangePartitioning> getUpdateGroupingPerMoleculetype(const gmx_domdec_t &dd)
215 GMX_RELEASE_ASSERT(dd.comm, "Need a valid dd.comm");
216 return dd.comm->systemInfo.updateGroupingPerMoleculetype;
219 void dd_store_state(gmx_domdec_t *dd, t_state *state)
223 if (state->ddp_count != dd->ddp_count)
225 gmx_incons("The MD state does not match the domain decomposition state");
228 state->cg_gl.resize(dd->ncg_home);
229 for (i = 0; i < dd->ncg_home; i++)
231 state->cg_gl[i] = dd->globalAtomGroupIndices[i];
234 state->ddp_count_cg_gl = dd->ddp_count;
237 gmx_domdec_zones_t *domdec_zones(gmx_domdec_t *dd)
239 return &dd->comm->zones;
242 void dd_get_ns_ranges(const gmx_domdec_t *dd, int icg,
243 int *jcg0, int *jcg1, ivec shift0, ivec shift1)
245 gmx_domdec_zones_t *zones;
248 zones = &dd->comm->zones;
251 while (icg >= zones->izone[izone].cg1)
260 else if (izone < zones->nizone)
262 *jcg0 = zones->izone[izone].jcg0;
266 gmx_fatal(FARGS, "DD icg %d out of range: izone (%d) >= nizone (%d)",
267 icg, izone, zones->nizone);
270 *jcg1 = zones->izone[izone].jcg1;
272 for (d = 0; d < dd->ndim; d++)
275 shift0[dim] = zones->izone[izone].shift0[dim];
276 shift1[dim] = zones->izone[izone].shift1[dim];
277 if (dd->comm->tric_dir[dim] || (isDlbOn(dd->comm) && d > 0))
279 /* A conservative approach, this can be optimized */
286 int dd_numHomeAtoms(const gmx_domdec_t &dd)
288 return dd.comm->atomRanges.numHomeAtoms();
291 int dd_natoms_mdatoms(const gmx_domdec_t *dd)
293 /* We currently set mdatoms entries for all atoms:
294 * local + non-local + communicated for vsite + constraints
297 return dd->comm->atomRanges.numAtomsTotal();
300 int dd_natoms_vsite(const gmx_domdec_t *dd)
302 return dd->comm->atomRanges.end(DDAtomRanges::Type::Vsites);
305 void dd_get_constraint_range(const gmx_domdec_t *dd, int *at_start, int *at_end)
307 *at_start = dd->comm->atomRanges.start(DDAtomRanges::Type::Constraints);
308 *at_end = dd->comm->atomRanges.end(DDAtomRanges::Type::Constraints);
311 void dd_move_x(gmx_domdec_t *dd,
313 gmx::ArrayRef<gmx::RVec> x,
314 gmx_wallcycle *wcycle)
316 wallcycle_start(wcycle, ewcMOVEX);
319 gmx_domdec_comm_t *comm;
320 gmx_domdec_comm_dim_t *cd;
321 rvec shift = {0, 0, 0};
322 gmx_bool bPBC, bScrew;
327 nat_tot = comm->atomRanges.numHomeAtoms();
328 for (int d = 0; d < dd->ndim; d++)
330 bPBC = (dd->ci[dd->dim[d]] == 0);
331 bScrew = (bPBC && dd->unitCellInfo.haveScrewPBC && dd->dim[d] == XX);
334 copy_rvec(box[dd->dim[d]], shift);
337 for (const gmx_domdec_ind_t &ind : cd->ind)
339 DDBufferAccess<gmx::RVec> sendBufferAccess(comm->rvecBuffer, ind.nsend[nzone + 1]);
340 gmx::ArrayRef<gmx::RVec> &sendBuffer = sendBufferAccess.buffer;
344 for (int j : ind.index)
346 sendBuffer[n] = x[j];
352 for (int j : ind.index)
354 /* We need to shift the coordinates */
355 for (int d = 0; d < DIM; d++)
357 sendBuffer[n][d] = x[j][d] + shift[d];
364 for (int j : ind.index)
367 sendBuffer[n][XX] = x[j][XX] + shift[XX];
369 * This operation requires a special shift force
370 * treatment, which is performed in calc_vir.
372 sendBuffer[n][YY] = box[YY][YY] - x[j][YY];
373 sendBuffer[n][ZZ] = box[ZZ][ZZ] - x[j][ZZ];
378 DDBufferAccess<gmx::RVec> receiveBufferAccess(comm->rvecBuffer2, cd->receiveInPlace ? 0 : ind.nrecv[nzone + 1]);
380 gmx::ArrayRef<gmx::RVec> receiveBuffer;
381 if (cd->receiveInPlace)
383 receiveBuffer = gmx::arrayRefFromArray(x.data() + nat_tot, ind.nrecv[nzone + 1]);
387 receiveBuffer = receiveBufferAccess.buffer;
389 /* Send and receive the coordinates */
390 ddSendrecv(dd, d, dddirBackward,
391 sendBuffer, receiveBuffer);
393 if (!cd->receiveInPlace)
396 for (int zone = 0; zone < nzone; zone++)
398 for (int i = ind.cell2at0[zone]; i < ind.cell2at1[zone]; i++)
400 x[i] = receiveBuffer[j++];
404 nat_tot += ind.nrecv[nzone+1];
409 wallcycle_stop(wcycle, ewcMOVEX);
412 void dd_move_f(gmx_domdec_t *dd,
413 gmx::ForceWithShiftForces *forceWithShiftForces,
414 gmx_wallcycle *wcycle)
416 wallcycle_start(wcycle, ewcMOVEF);
418 gmx::ArrayRef<gmx::RVec> f = forceWithShiftForces->force();
419 gmx::ArrayRef<gmx::RVec> fshift = forceWithShiftForces->shiftForces();
421 gmx_domdec_comm_t &comm = *dd->comm;
422 int nzone = comm.zones.n/2;
423 int nat_tot = comm.atomRanges.end(DDAtomRanges::Type::Zones);
424 for (int d = dd->ndim-1; d >= 0; d--)
426 /* Only forces in domains near the PBC boundaries need to
427 consider PBC in the treatment of fshift */
428 const bool shiftForcesNeedPbc = (forceWithShiftForces->computeVirial() && dd->ci[dd->dim[d]] == 0);
429 const bool applyScrewPbc = (shiftForcesNeedPbc && dd->unitCellInfo.haveScrewPBC && dd->dim[d] == XX);
430 /* Determine which shift vector we need */
431 ivec vis = { 0, 0, 0 };
433 const int is = IVEC2IS(vis);
435 /* Loop over the pulses */
436 const gmx_domdec_comm_dim_t &cd = comm.cd[d];
437 for (int p = cd.numPulses() - 1; p >= 0; p--)
439 const gmx_domdec_ind_t &ind = cd.ind[p];
440 DDBufferAccess<gmx::RVec> receiveBufferAccess(comm.rvecBuffer, ind.nsend[nzone + 1]);
441 gmx::ArrayRef<gmx::RVec> &receiveBuffer = receiveBufferAccess.buffer;
443 nat_tot -= ind.nrecv[nzone+1];
445 DDBufferAccess<gmx::RVec> sendBufferAccess(comm.rvecBuffer2, cd.receiveInPlace ? 0 : ind.nrecv[nzone + 1]);
447 gmx::ArrayRef<gmx::RVec> sendBuffer;
448 if (cd.receiveInPlace)
450 sendBuffer = gmx::arrayRefFromArray(f.data() + nat_tot, ind.nrecv[nzone + 1]);
454 sendBuffer = sendBufferAccess.buffer;
456 for (int zone = 0; zone < nzone; zone++)
458 for (int i = ind.cell2at0[zone]; i < ind.cell2at1[zone]; i++)
460 sendBuffer[j++] = f[i];
464 /* Communicate the forces */
465 ddSendrecv(dd, d, dddirForward,
466 sendBuffer, receiveBuffer);
467 /* Add the received forces */
469 if (!shiftForcesNeedPbc)
471 for (int j : ind.index)
473 for (int d = 0; d < DIM; d++)
475 f[j][d] += receiveBuffer[n][d];
480 else if (!applyScrewPbc)
482 for (int j : ind.index)
484 for (int d = 0; d < DIM; d++)
486 f[j][d] += receiveBuffer[n][d];
488 /* Add this force to the shift force */
489 for (int d = 0; d < DIM; d++)
491 fshift[is][d] += receiveBuffer[n][d];
498 for (int j : ind.index)
500 /* Rotate the force */
501 f[j][XX] += receiveBuffer[n][XX];
502 f[j][YY] -= receiveBuffer[n][YY];
503 f[j][ZZ] -= receiveBuffer[n][ZZ];
504 if (shiftForcesNeedPbc)
506 /* Add this force to the shift force */
507 for (int d = 0; d < DIM; d++)
509 fshift[is][d] += receiveBuffer[n][d];
518 wallcycle_stop(wcycle, ewcMOVEF);
521 /* Convenience function for extracting a real buffer from an rvec buffer
523 * To reduce the number of temporary communication buffers and avoid
524 * cache polution, we reuse gmx::RVec buffers for storing reals.
525 * This functions return a real buffer reference with the same number
526 * of elements as the gmx::RVec buffer (so 1/3 of the size in bytes).
528 static gmx::ArrayRef<real>
529 realArrayRefFromRvecArrayRef(gmx::ArrayRef<gmx::RVec> arrayRef)
531 return gmx::arrayRefFromArray(as_rvec_array(arrayRef.data())[0],
535 void dd_atom_spread_real(gmx_domdec_t *dd, real v[])
538 gmx_domdec_comm_t *comm;
539 gmx_domdec_comm_dim_t *cd;
544 nat_tot = comm->atomRanges.numHomeAtoms();
545 for (int d = 0; d < dd->ndim; d++)
548 for (const gmx_domdec_ind_t &ind : cd->ind)
550 /* Note: We provision for RVec instead of real, so a factor of 3
551 * more than needed. The buffer actually already has this size
552 * and we pass a plain pointer below, so this does not matter.
554 DDBufferAccess<gmx::RVec> sendBufferAccess(comm->rvecBuffer, ind.nsend[nzone + 1]);
555 gmx::ArrayRef<real> sendBuffer = realArrayRefFromRvecArrayRef(sendBufferAccess.buffer);
557 for (int j : ind.index)
559 sendBuffer[n++] = v[j];
562 DDBufferAccess<gmx::RVec> receiveBufferAccess(comm->rvecBuffer2, cd->receiveInPlace ? 0 : ind.nrecv[nzone + 1]);
564 gmx::ArrayRef<real> receiveBuffer;
565 if (cd->receiveInPlace)
567 receiveBuffer = gmx::arrayRefFromArray(v + nat_tot, ind.nrecv[nzone + 1]);
571 receiveBuffer = realArrayRefFromRvecArrayRef(receiveBufferAccess.buffer);
573 /* Send and receive the data */
574 ddSendrecv(dd, d, dddirBackward,
575 sendBuffer, receiveBuffer);
576 if (!cd->receiveInPlace)
579 for (int zone = 0; zone < nzone; zone++)
581 for (int i = ind.cell2at0[zone]; i < ind.cell2at1[zone]; i++)
583 v[i] = receiveBuffer[j++];
587 nat_tot += ind.nrecv[nzone+1];
593 void dd_atom_sum_real(gmx_domdec_t *dd, real v[])
596 gmx_domdec_comm_t *comm;
597 gmx_domdec_comm_dim_t *cd;
601 nzone = comm->zones.n/2;
602 nat_tot = comm->atomRanges.end(DDAtomRanges::Type::Zones);
603 for (int d = dd->ndim-1; d >= 0; d--)
606 for (int p = cd->numPulses() - 1; p >= 0; p--)
608 const gmx_domdec_ind_t &ind = cd->ind[p];
610 /* Note: We provision for RVec instead of real, so a factor of 3
611 * more than needed. The buffer actually already has this size
612 * and we typecast, so this works as intended.
614 DDBufferAccess<gmx::RVec> receiveBufferAccess(comm->rvecBuffer, ind.nsend[nzone + 1]);
615 gmx::ArrayRef<real> receiveBuffer = realArrayRefFromRvecArrayRef(receiveBufferAccess.buffer);
616 nat_tot -= ind.nrecv[nzone + 1];
618 DDBufferAccess<gmx::RVec> sendBufferAccess(comm->rvecBuffer2, cd->receiveInPlace ? 0 : ind.nrecv[nzone + 1]);
620 gmx::ArrayRef<real> sendBuffer;
621 if (cd->receiveInPlace)
623 sendBuffer = gmx::arrayRefFromArray(v + nat_tot, ind.nrecv[nzone + 1]);
627 sendBuffer = realArrayRefFromRvecArrayRef(sendBufferAccess.buffer);
629 for (int zone = 0; zone < nzone; zone++)
631 for (int i = ind.cell2at0[zone]; i < ind.cell2at1[zone]; i++)
633 sendBuffer[j++] = v[i];
637 /* Communicate the forces */
638 ddSendrecv(dd, d, dddirForward,
639 sendBuffer, receiveBuffer);
640 /* Add the received forces */
642 for (int j : ind.index)
644 v[j] += receiveBuffer[n];
652 real dd_cutoff_multibody(const gmx_domdec_t *dd)
654 const gmx_domdec_comm_t &comm = *dd->comm;
655 const DDSystemInfo &systemInfo = comm.systemInfo;
658 if (systemInfo.haveInterDomainMultiBodyBondeds)
660 if (comm.cutoff_mbody > 0)
662 r = comm.cutoff_mbody;
666 /* cutoff_mbody=0 means we do not have DLB */
667 r = comm.cellsize_min[dd->dim[0]];
668 for (int di = 1; di < dd->ndim; di++)
670 r = std::min(r, comm.cellsize_min[dd->dim[di]]);
672 if (comm.systemInfo.filterBondedCommunication)
674 r = std::max(r, comm.cutoff_mbody);
678 r = std::min(r, systemInfo.cutoff);
686 real dd_cutoff_twobody(const gmx_domdec_t *dd)
690 r_mb = dd_cutoff_multibody(dd);
692 return std::max(dd->comm->systemInfo.cutoff, r_mb);
696 static void dd_cart_coord2pmecoord(const DDRankSetup &ddRankSetup,
697 const CartesianRankSetup &cartSetup,
701 const int nc = ddRankSetup.numPPCells[cartSetup.cartpmedim];
702 const int ntot = cartSetup.ntot[cartSetup.cartpmedim];
703 copy_ivec(coord, coord_pme);
704 coord_pme[cartSetup.cartpmedim] =
705 nc + (coord[cartSetup.cartpmedim]*(ntot - nc) + (ntot - nc)/2)/nc;
708 /* Returns the PME rank index in 0...npmenodes-1 for the PP cell with index ddCellIndex */
709 static int ddindex2pmeindex(const DDRankSetup &ddRankSetup,
710 const int ddCellIndex)
712 const int npp = ddRankSetup.numPPRanks;
713 const int npme = ddRankSetup.numRanksDoingPme;
715 /* Here we assign a PME node to communicate with this DD node
716 * by assuming that the major index of both is x.
717 * We add npme/2 to obtain an even distribution.
719 return (ddCellIndex*npme + npme/2)/npp;
722 static std::vector<int> dd_interleaved_pme_ranks(const DDRankSetup &ddRankSetup)
724 std::vector<int> pmeRanks(ddRankSetup.numRanksDoingPme);
727 for (int i = 0; i < ddRankSetup.numPPRanks; i++)
729 const int p0 = ddindex2pmeindex(ddRankSetup, i);
730 const int p1 = ddindex2pmeindex(ddRankSetup, i + 1);
731 if (i + 1 == ddRankSetup.numPPRanks || p1 > p0)
735 fprintf(debug, "pme_rank[%d] = %d\n", n, i+1+n);
737 pmeRanks[n] = i + 1 + n;
745 static int gmx_ddcoord2pmeindex(const t_commrec *cr, int x, int y, int z)
753 if (dd->comm->bCartesian) {
754 gmx_ddindex2xyz(dd->nc,ddindex,coords);
755 dd_coords2pmecoords(dd,coords,coords_pme);
756 copy_ivec(dd->ntot,nc);
757 nc[dd->cartpmedim] -= dd->nc[dd->cartpmedim];
758 coords_pme[dd->cartpmedim] -= dd->nc[dd->cartpmedim];
760 slab = (coords_pme[XX]*nc[YY] + coords_pme[YY])*nc[ZZ] + coords_pme[ZZ];
762 slab = (ddindex*cr->npmenodes + cr->npmenodes/2)/dd->nnodes;
768 slab = ddindex2pmeindex(dd->comm->ddRankSetup, dd_index(dd->nc, coords));
773 static int ddcoord2simnodeid(const t_commrec *cr, int x, int y, int z)
775 const CartesianRankSetup &cartSetup = cr->dd->comm->cartesianRankSetup;
776 ivec coords = { x, y, z };
779 if (cartSetup.bCartesianPP_PME)
782 MPI_Cart_rank(cr->mpi_comm_mysim, coords, &nodeid);
787 int ddindex = dd_index(cr->dd->nc, coords);
788 if (cartSetup.bCartesianPP)
790 nodeid = cartSetup.ddindex2simnodeid[ddindex];
794 if (cr->dd->comm->ddRankSetup.usePmeOnlyRanks)
796 nodeid = ddindex + gmx_ddcoord2pmeindex(cr, x, y, z);
808 static int dd_simnode2pmenode(const DDRankSetup &ddRankSetup,
809 const CartesianRankSetup &cartSetup,
810 gmx::ArrayRef<const int> pmeRanks,
811 const t_commrec gmx_unused *cr,
812 const int sim_nodeid)
816 /* This assumes a uniform x domain decomposition grid cell size */
817 if (cartSetup.bCartesianPP_PME)
820 ivec coord, coord_pme;
821 MPI_Cart_coords(cr->mpi_comm_mysim, sim_nodeid, DIM, coord);
822 if (coord[cartSetup.cartpmedim] < ddRankSetup.numPPCells[cartSetup.cartpmedim])
824 /* This is a PP rank */
825 dd_cart_coord2pmecoord(ddRankSetup, cartSetup, coord, coord_pme);
826 MPI_Cart_rank(cr->mpi_comm_mysim, coord_pme, &pmenode);
830 else if (cartSetup.bCartesianPP)
832 if (sim_nodeid < ddRankSetup.numPPRanks)
834 pmenode = ddRankSetup.numPPRanks + ddindex2pmeindex(ddRankSetup, sim_nodeid);
839 /* This assumes DD cells with identical x coordinates
840 * are numbered sequentially.
842 if (pmeRanks.empty())
844 if (sim_nodeid < ddRankSetup.numPPRanks)
846 /* The DD index equals the nodeid */
847 pmenode = ddRankSetup.numPPRanks + ddindex2pmeindex(ddRankSetup, sim_nodeid);
853 while (sim_nodeid > pmeRanks[i])
857 if (sim_nodeid < pmeRanks[i])
859 pmenode = pmeRanks[i];
867 NumPmeDomains getNumPmeDomains(const gmx_domdec_t *dd)
872 dd->comm->ddRankSetup.npmenodes_x,
873 dd->comm->ddRankSetup.npmenodes_y
884 std::vector<int> get_pme_ddranks(const t_commrec *cr,
887 const DDRankSetup &ddRankSetup = cr->dd->comm->ddRankSetup;
888 const CartesianRankSetup &cartSetup = cr->dd->comm->cartesianRankSetup;
889 GMX_RELEASE_ASSERT(ddRankSetup.usePmeOnlyRanks, "This function should only be called when PME-only ranks are in use");
890 std::vector<int> ddranks;
891 ddranks.reserve((ddRankSetup.numPPRanks + ddRankSetup.numRanksDoingPme - 1)/ddRankSetup.numRanksDoingPme);
893 for (int x = 0; x < ddRankSetup.numPPCells[XX]; x++)
895 for (int y = 0; y < ddRankSetup.numPPCells[YY]; y++)
897 for (int z = 0; z < ddRankSetup.numPPCells[ZZ]; z++)
899 if (cartSetup.bCartesianPP_PME)
901 ivec coord = { x, y, z };
903 dd_cart_coord2pmecoord(ddRankSetup, cartSetup, coord, coord_pme);
904 if (cr->dd->ci[XX] == coord_pme[XX] &&
905 cr->dd->ci[YY] == coord_pme[YY] &&
906 cr->dd->ci[ZZ] == coord_pme[ZZ])
908 ddranks.push_back(ddcoord2simnodeid(cr, x, y, z));
913 /* The slab corresponds to the nodeid in the PME group */
914 if (gmx_ddcoord2pmeindex(cr, x, y, z) == pmenodeid)
916 ddranks.push_back(ddcoord2simnodeid(cr, x, y, z));
925 static gmx_bool receive_vir_ener(const gmx_domdec_t *dd,
926 gmx::ArrayRef<const int> pmeRanks,
929 gmx_bool bReceive = TRUE;
931 const DDRankSetup &ddRankSetup = dd->comm->ddRankSetup;
932 if (ddRankSetup.usePmeOnlyRanks)
934 const CartesianRankSetup &cartSetup = dd->comm->cartesianRankSetup;
935 if (cartSetup.bCartesianPP_PME)
938 int pmenode = dd_simnode2pmenode(ddRankSetup, cartSetup, pmeRanks, cr, cr->sim_nodeid);
940 MPI_Cart_coords(cr->mpi_comm_mysim, cr->sim_nodeid, DIM, coords);
941 coords[cartSetup.cartpmedim]++;
942 if (coords[cartSetup.cartpmedim] < dd->nc[cartSetup.cartpmedim])
945 MPI_Cart_rank(cr->mpi_comm_mysim, coords, &rank);
946 if (dd_simnode2pmenode(ddRankSetup, cartSetup, pmeRanks, cr, rank) == pmenode)
948 /* This is not the last PP node for pmenode */
953 GMX_RELEASE_ASSERT(false, "Without MPI we should not have Cartesian PP-PME with #PMEnodes < #DDnodes");
958 int pmenode = dd_simnode2pmenode(ddRankSetup, cartSetup, pmeRanks, cr, cr->sim_nodeid);
959 if (cr->sim_nodeid+1 < cr->nnodes &&
960 dd_simnode2pmenode(ddRankSetup, cartSetup, pmeRanks, cr, cr->sim_nodeid + 1) == pmenode)
962 /* This is not the last PP node for pmenode */
971 static void set_slb_pme_dim_f(gmx_domdec_t *dd, int dim, real **dim_f)
973 gmx_domdec_comm_t *comm;
978 snew(*dim_f, dd->nc[dim]+1);
980 for (i = 1; i < dd->nc[dim]; i++)
982 if (comm->slb_frac[dim])
984 (*dim_f)[i] = (*dim_f)[i-1] + comm->slb_frac[dim][i-1];
988 (*dim_f)[i] = static_cast<real>(i)/static_cast<real>(dd->nc[dim]);
991 (*dim_f)[dd->nc[dim]] = 1;
994 static void init_ddpme(gmx_domdec_t *dd, gmx_ddpme_t *ddpme, int dimind)
996 const DDRankSetup &ddRankSetup = dd->comm->ddRankSetup;
998 if (dimind == 0 && dd->dim[0] == YY && ddRankSetup.npmenodes_x == 1)
1004 ddpme->dim = dimind;
1006 ddpme->dim_match = (ddpme->dim == dd->dim[dimind]);
1008 ddpme->nslab = (ddpme->dim == 0 ?
1009 ddRankSetup.npmenodes_x :
1010 ddRankSetup.npmenodes_y);
1012 if (ddpme->nslab <= 1)
1017 const int nso = ddRankSetup.numRanksDoingPme/ddpme->nslab;
1018 /* Determine for each PME slab the PP location range for dimension dim */
1019 snew(ddpme->pp_min, ddpme->nslab);
1020 snew(ddpme->pp_max, ddpme->nslab);
1021 for (int slab = 0; slab < ddpme->nslab; slab++)
1023 ddpme->pp_min[slab] = dd->nc[dd->dim[dimind]] - 1;
1024 ddpme->pp_max[slab] = 0;
1026 for (int i = 0; i < dd->nnodes; i++)
1029 ddindex2xyz(dd->nc, i, xyz);
1030 /* For y only use our y/z slab.
1031 * This assumes that the PME x grid size matches the DD grid size.
1033 if (dimind == 0 || xyz[XX] == dd->ci[XX])
1035 const int pmeindex = ddindex2pmeindex(ddRankSetup, i);
1039 slab = pmeindex/nso;
1043 slab = pmeindex % ddpme->nslab;
1045 ddpme->pp_min[slab] = std::min(ddpme->pp_min[slab], xyz[dimind]);
1046 ddpme->pp_max[slab] = std::max(ddpme->pp_max[slab], xyz[dimind]);
1050 set_slb_pme_dim_f(dd, ddpme->dim, &ddpme->slb_dim_f);
1053 int dd_pme_maxshift_x(const gmx_domdec_t *dd)
1055 const DDRankSetup &ddRankSetup = dd->comm->ddRankSetup;
1057 if (ddRankSetup.ddpme[0].dim == XX)
1059 return ddRankSetup.ddpme[0].maxshift;
1067 int dd_pme_maxshift_y(const gmx_domdec_t *dd)
1069 const DDRankSetup &ddRankSetup = dd->comm->ddRankSetup;
1071 if (ddRankSetup.ddpme[0].dim == YY)
1073 return ddRankSetup.ddpme[0].maxshift;
1075 else if (ddRankSetup.npmedecompdim >= 2 && ddRankSetup.ddpme[1].dim == YY)
1077 return ddRankSetup.ddpme[1].maxshift;
1085 bool ddHaveSplitConstraints(const gmx_domdec_t &dd)
1087 return dd.comm->systemInfo.haveSplitConstraints;
1090 void dd_cycles_add(const gmx_domdec_t *dd, float cycles, int ddCycl)
1092 /* Note that the cycles value can be incorrect, either 0 or some
1093 * extremely large value, when our thread migrated to another core
1094 * with an unsynchronized cycle counter. If this happens less often
1095 * that once per nstlist steps, this will not cause issues, since
1096 * we later subtract the maximum value from the sum over nstlist steps.
1097 * A zero count will slightly lower the total, but that's a small effect.
1098 * Note that the main purpose of the subtraction of the maximum value
1099 * is to avoid throwing off the load balancing when stalls occur due
1100 * e.g. system activity or network congestion.
1102 dd->comm->cycl[ddCycl] += cycles;
1103 dd->comm->cycl_n[ddCycl]++;
1104 if (cycles > dd->comm->cycl_max[ddCycl])
1106 dd->comm->cycl_max[ddCycl] = cycles;
1111 static void make_load_communicator(gmx_domdec_t *dd, int dim_ind, ivec loc)
1116 gmx_bool bPartOfGroup = FALSE;
1118 dim = dd->dim[dim_ind];
1119 copy_ivec(loc, loc_c);
1120 for (i = 0; i < dd->nc[dim]; i++)
1123 rank = dd_index(dd->nc, loc_c);
1124 if (rank == dd->rank)
1126 /* This process is part of the group */
1127 bPartOfGroup = TRUE;
1130 MPI_Comm_split(dd->mpi_comm_all, bPartOfGroup ? 0 : MPI_UNDEFINED, dd->rank,
1134 dd->comm->mpi_comm_load[dim_ind] = c_row;
1135 if (!isDlbDisabled(dd->comm))
1137 DDCellsizesWithDlb &cellsizes = dd->comm->cellsizesWithDlb[dim_ind];
1139 if (dd->ci[dim] == dd->master_ci[dim])
1141 /* This is the root process of this row */
1142 cellsizes.rowMaster = std::make_unique<RowMaster>();
1144 RowMaster &rowMaster = *cellsizes.rowMaster;
1145 rowMaster.cellFrac.resize(ddCellFractionBufferSize(dd, dim_ind));
1146 rowMaster.oldCellFrac.resize(dd->nc[dim] + 1);
1147 rowMaster.isCellMin.resize(dd->nc[dim]);
1150 rowMaster.bounds.resize(dd->nc[dim]);
1152 rowMaster.buf_ncd.resize(dd->nc[dim]);
1156 /* This is not a root process, we only need to receive cell_f */
1157 cellsizes.fracRow.resize(ddCellFractionBufferSize(dd, dim_ind));
1160 if (dd->ci[dim] == dd->master_ci[dim])
1162 snew(dd->comm->load[dim_ind].load, dd->nc[dim]*DD_NLOAD_MAX);
1168 void dd_setup_dlb_resource_sharing(t_commrec *cr,
1172 int physicalnode_id_hash;
1174 MPI_Comm mpi_comm_pp_physicalnode;
1176 if (!thisRankHasDuty(cr, DUTY_PP) || gpu_id < 0)
1178 /* Only ranks with short-ranged tasks (currently) use GPUs.
1179 * If we don't have GPUs assigned, there are no resources to share.
1184 physicalnode_id_hash = gmx_physicalnode_id_hash();
1190 fprintf(debug, "dd_setup_dd_dlb_gpu_sharing:\n");
1191 fprintf(debug, "DD PP rank %d physical node hash %d gpu_id %d\n",
1192 dd->rank, physicalnode_id_hash, gpu_id);
1194 /* Split the PP communicator over the physical nodes */
1195 /* TODO: See if we should store this (before), as it's also used for
1196 * for the nodecomm summation.
1198 // TODO PhysicalNodeCommunicator could be extended/used to handle
1199 // the need for per-node per-group communicators.
1200 MPI_Comm_split(dd->mpi_comm_all, physicalnode_id_hash, dd->rank,
1201 &mpi_comm_pp_physicalnode);
1202 MPI_Comm_split(mpi_comm_pp_physicalnode, gpu_id, dd->rank,
1203 &dd->comm->mpi_comm_gpu_shared);
1204 MPI_Comm_free(&mpi_comm_pp_physicalnode);
1205 MPI_Comm_size(dd->comm->mpi_comm_gpu_shared, &dd->comm->nrank_gpu_shared);
1209 fprintf(debug, "nrank_gpu_shared %d\n", dd->comm->nrank_gpu_shared);
1212 /* Note that some ranks could share a GPU, while others don't */
1214 if (dd->comm->nrank_gpu_shared == 1)
1216 MPI_Comm_free(&dd->comm->mpi_comm_gpu_shared);
1219 GMX_UNUSED_VALUE(cr);
1220 GMX_UNUSED_VALUE(gpu_id);
1224 static void make_load_communicators(gmx_domdec_t gmx_unused *dd)
1227 int dim0, dim1, i, j;
1232 fprintf(debug, "Making load communicators\n");
1235 dd->comm->load = new domdec_load_t[std::max(dd->ndim, 1)];
1236 snew(dd->comm->mpi_comm_load, std::max(dd->ndim, 1));
1244 make_load_communicator(dd, 0, loc);
1248 for (i = 0; i < dd->nc[dim0]; i++)
1251 make_load_communicator(dd, 1, loc);
1257 for (i = 0; i < dd->nc[dim0]; i++)
1261 for (j = 0; j < dd->nc[dim1]; j++)
1264 make_load_communicator(dd, 2, loc);
1271 fprintf(debug, "Finished making load communicators\n");
1276 /*! \brief Sets up the relation between neighboring domains and zones */
1277 static void setup_neighbor_relations(gmx_domdec_t *dd)
1279 int d, dim, i, j, m;
1281 gmx_domdec_zones_t *zones;
1282 gmx_domdec_ns_ranges_t *izone;
1283 GMX_ASSERT(dd->ndim >= 0, "Must have non-negative number of dimensions for DD");
1285 for (d = 0; d < dd->ndim; d++)
1288 copy_ivec(dd->ci, tmp);
1289 tmp[dim] = (tmp[dim] + 1) % dd->nc[dim];
1290 dd->neighbor[d][0] = ddcoord2ddnodeid(dd, tmp);
1291 copy_ivec(dd->ci, tmp);
1292 tmp[dim] = (tmp[dim] - 1 + dd->nc[dim]) % dd->nc[dim];
1293 dd->neighbor[d][1] = ddcoord2ddnodeid(dd, tmp);
1296 fprintf(debug, "DD rank %d neighbor ranks in dir %d are + %d - %d\n",
1299 dd->neighbor[d][1]);
1303 int nzone = (1 << dd->ndim);
1304 GMX_ASSERT(dd->ndim < DIM, "Invalid number of dimensions");
1305 int nizone = (1 << std::max(dd->ndim - 1, 0));
1306 assert(nizone >= 1 && nizone <= DD_MAXIZONE);
1308 zones = &dd->comm->zones;
1310 for (i = 0; i < nzone; i++)
1313 clear_ivec(zones->shift[i]);
1314 for (d = 0; d < dd->ndim; d++)
1316 zones->shift[i][dd->dim[d]] = dd_zo[i][m++];
1321 for (i = 0; i < nzone; i++)
1323 for (d = 0; d < DIM; d++)
1325 s[d] = dd->ci[d] - zones->shift[i][d];
1330 else if (s[d] >= dd->nc[d])
1336 zones->nizone = nizone;
1337 for (i = 0; i < zones->nizone; i++)
1339 assert(ddNonbondedZonePairRanges[i][0] == i);
1341 izone = &zones->izone[i];
1342 /* dd_zp3 is for 3D decomposition, for fewer dimensions use only
1343 * j-zones up to nzone.
1345 izone->j0 = std::min(ddNonbondedZonePairRanges[i][1], nzone);
1346 izone->j1 = std::min(ddNonbondedZonePairRanges[i][2], nzone);
1347 for (dim = 0; dim < DIM; dim++)
1349 if (dd->nc[dim] == 1)
1351 /* All shifts should be allowed */
1352 izone->shift0[dim] = -1;
1353 izone->shift1[dim] = 1;
1357 /* Determine the min/max j-zone shift wrt the i-zone */
1358 izone->shift0[dim] = 1;
1359 izone->shift1[dim] = -1;
1360 for (j = izone->j0; j < izone->j1; j++)
1362 int shift_diff = zones->shift[j][dim] - zones->shift[i][dim];
1363 if (shift_diff < izone->shift0[dim])
1365 izone->shift0[dim] = shift_diff;
1367 if (shift_diff > izone->shift1[dim])
1369 izone->shift1[dim] = shift_diff;
1376 if (!isDlbDisabled(dd->comm))
1378 dd->comm->cellsizesWithDlb.resize(dd->ndim);
1381 if (dd->comm->ddSettings.recordLoad)
1383 make_load_communicators(dd);
1387 static void make_pp_communicator(const gmx::MDLogger &mdlog,
1389 t_commrec gmx_unused *cr,
1390 bool gmx_unused reorder)
1393 gmx_domdec_comm_t *comm = dd->comm;
1394 CartesianRankSetup &cartSetup = comm->cartesianRankSetup;
1396 if (cartSetup.bCartesianPP)
1398 /* Set up cartesian communication for the particle-particle part */
1399 GMX_LOG(mdlog.info).appendTextFormatted(
1400 "Will use a Cartesian communicator: %d x %d x %d",
1401 dd->nc[XX], dd->nc[YY], dd->nc[ZZ]);
1404 for (int i = 0; i < DIM; i++)
1409 MPI_Cart_create(cr->mpi_comm_mygroup, DIM, dd->nc, periods, static_cast<int>(reorder),
1411 /* We overwrite the old communicator with the new cartesian one */
1412 cr->mpi_comm_mygroup = comm_cart;
1415 dd->mpi_comm_all = cr->mpi_comm_mygroup;
1416 MPI_Comm_rank(dd->mpi_comm_all, &dd->rank);
1418 if (cartSetup.bCartesianPP_PME)
1420 /* Since we want to use the original cartesian setup for sim,
1421 * and not the one after split, we need to make an index.
1423 cartSetup.ddindex2ddnodeid.resize(dd->nnodes);
1424 cartSetup.ddindex2ddnodeid[dd_index(dd->nc, dd->ci)] = dd->rank;
1425 gmx_sumi(dd->nnodes, cartSetup.ddindex2ddnodeid.data(), cr);
1426 /* Get the rank of the DD master,
1427 * above we made sure that the master node is a PP node.
1438 MPI_Allreduce(&rank, &dd->masterrank, 1, MPI_INT, MPI_SUM, dd->mpi_comm_all);
1440 else if (cartSetup.bCartesianPP)
1442 if (!comm->ddRankSetup.usePmeOnlyRanks)
1444 /* The PP communicator is also
1445 * the communicator for this simulation
1447 cr->mpi_comm_mysim = cr->mpi_comm_mygroup;
1449 cr->nodeid = dd->rank;
1451 MPI_Cart_coords(dd->mpi_comm_all, dd->rank, DIM, dd->ci);
1453 /* We need to make an index to go from the coordinates
1454 * to the nodeid of this simulation.
1456 cartSetup.ddindex2simnodeid.resize(dd->nnodes);
1457 std::vector<int> buf(dd->nnodes);
1458 if (thisRankHasDuty(cr, DUTY_PP))
1460 buf[dd_index(dd->nc, dd->ci)] = cr->sim_nodeid;
1462 /* Communicate the ddindex to simulation nodeid index */
1463 MPI_Allreduce(buf.data(), cartSetup.ddindex2simnodeid.data(), dd->nnodes, MPI_INT, MPI_SUM,
1464 cr->mpi_comm_mysim);
1466 /* Determine the master coordinates and rank.
1467 * The DD master should be the same node as the master of this sim.
1469 for (int i = 0; i < dd->nnodes; i++)
1471 if (cartSetup.ddindex2simnodeid[i] == 0)
1473 ddindex2xyz(dd->nc, i, dd->master_ci);
1474 MPI_Cart_rank(dd->mpi_comm_all, dd->master_ci, &dd->masterrank);
1479 fprintf(debug, "The master rank is %d\n", dd->masterrank);
1484 /* No Cartesian communicators */
1485 /* We use the rank in dd->comm->all as DD index */
1486 ddindex2xyz(dd->nc, dd->rank, dd->ci);
1487 /* The simulation master nodeid is 0, so the DD master rank is also 0 */
1489 clear_ivec(dd->master_ci);
1493 GMX_LOG(mdlog.info).appendTextFormatted(
1494 "Domain decomposition rank %d, coordinates %d %d %d\n",
1495 dd->rank, dd->ci[XX], dd->ci[YY], dd->ci[ZZ]);
1499 "Domain decomposition rank %d, coordinates %d %d %d\n\n",
1500 dd->rank, dd->ci[XX], dd->ci[YY], dd->ci[ZZ]);
1504 static void receive_ddindex2simnodeid(gmx_domdec_t *dd,
1508 CartesianRankSetup &cartSetup = dd->comm->cartesianRankSetup;
1510 if (!cartSetup.bCartesianPP_PME && cartSetup.bCartesianPP)
1512 cartSetup.ddindex2simnodeid.resize(dd->nnodes);
1513 std::vector<int> buf(dd->nnodes);
1514 if (thisRankHasDuty(cr, DUTY_PP))
1516 buf[dd_index(dd->nc, dd->ci)] = cr->sim_nodeid;
1518 /* Communicate the ddindex to simulation nodeid index */
1519 MPI_Allreduce(buf.data(), cartSetup.ddindex2simnodeid.data(), dd->nnodes, MPI_INT, MPI_SUM,
1520 cr->mpi_comm_mysim);
1523 GMX_UNUSED_VALUE(dd);
1524 GMX_UNUSED_VALUE(cr);
1528 static CartesianRankSetup
1529 split_communicator(const gmx::MDLogger &mdlog,
1531 const DdRankOrder ddRankOrder,
1532 bool gmx_unused reorder,
1533 const DDRankSetup &ddRankSetup,
1535 std::vector<int> *pmeRanks)
1537 CartesianRankSetup cartSetup;
1539 cartSetup.bCartesianPP = (ddRankOrder == DdRankOrder::cartesian);
1540 cartSetup.bCartesianPP_PME = false;
1542 const ivec &numDDCells = ddRankSetup.numPPCells;
1543 /* Initially we set ntot to the number of PP cells */
1544 copy_ivec(numDDCells, cartSetup.ntot);
1546 if (cartSetup.bCartesianPP)
1548 const int numDDCellsTot = ddRankSetup.numPPRanks;
1550 for (int i = 1; i < DIM; i++)
1552 bDiv[i] = ((ddRankSetup.numRanksDoingPme*numDDCells[i]) % numDDCellsTot == 0);
1554 if (bDiv[YY] || bDiv[ZZ])
1556 cartSetup.bCartesianPP_PME = TRUE;
1557 /* If we have 2D PME decomposition, which is always in x+y,
1558 * we stack the PME only nodes in z.
1559 * Otherwise we choose the direction that provides the thinnest slab
1560 * of PME only nodes as this will have the least effect
1561 * on the PP communication.
1562 * But for the PME communication the opposite might be better.
1564 if (bDiv[ZZ] && (ddRankSetup.npmenodes_y > 1 ||
1566 numDDCells[YY] > numDDCells[ZZ]))
1568 cartSetup.cartpmedim = ZZ;
1572 cartSetup.cartpmedim = YY;
1574 cartSetup.ntot[cartSetup.cartpmedim]
1575 += (ddRankSetup.numRanksDoingPme*numDDCells[cartSetup.cartpmedim])/numDDCellsTot;
1579 GMX_LOG(mdlog.info).appendTextFormatted(
1580 "Number of PME-only ranks (%d) is not a multiple of nx*ny (%d*%d) or nx*nz (%d*%d)",
1581 ddRankSetup.numRanksDoingPme,
1582 numDDCells[XX], numDDCells[YY],
1583 numDDCells[XX], numDDCells[ZZ]);
1584 GMX_LOG(mdlog.info).appendText("Will not use a Cartesian communicator for PP <-> PME\n");
1588 if (cartSetup.bCartesianPP_PME)
1594 GMX_LOG(mdlog.info).appendTextFormatted(
1595 "Will use a Cartesian communicator for PP <-> PME: %d x %d x %d",
1596 cartSetup.ntot[XX], cartSetup.ntot[YY], cartSetup.ntot[ZZ]);
1598 for (int i = 0; i < DIM; i++)
1603 MPI_Cart_create(cr->mpi_comm_mysim, DIM, cartSetup.ntot, periods, static_cast<int>(reorder),
1605 MPI_Comm_rank(comm_cart, &rank);
1606 if (MASTER(cr) && rank != 0)
1608 gmx_fatal(FARGS, "MPI rank 0 was renumbered by MPI_Cart_create, we do not allow this");
1611 /* With this assigment we loose the link to the original communicator
1612 * which will usually be MPI_COMM_WORLD, unless have multisim.
1614 cr->mpi_comm_mysim = comm_cart;
1615 cr->sim_nodeid = rank;
1617 MPI_Cart_coords(cr->mpi_comm_mysim, cr->sim_nodeid, DIM, ddCellIndex);
1619 GMX_LOG(mdlog.info).appendTextFormatted(
1620 "Cartesian rank %d, coordinates %d %d %d\n",
1621 cr->sim_nodeid, ddCellIndex[XX], ddCellIndex[YY], ddCellIndex[ZZ]);
1623 if (ddCellIndex[cartSetup.cartpmedim] < numDDCells[cartSetup.cartpmedim])
1627 if (!ddRankSetup.usePmeOnlyRanks ||
1628 ddCellIndex[cartSetup.cartpmedim] >= numDDCells[cartSetup.cartpmedim])
1630 cr->duty = DUTY_PME;
1633 /* Split the sim communicator into PP and PME only nodes */
1634 MPI_Comm_split(cr->mpi_comm_mysim,
1635 getThisRankDuties(cr),
1636 dd_index(cartSetup.ntot, ddCellIndex),
1637 &cr->mpi_comm_mygroup);
1639 GMX_UNUSED_VALUE(ddCellIndex);
1644 switch (ddRankOrder)
1646 case DdRankOrder::pp_pme:
1647 GMX_LOG(mdlog.info).appendText("Order of the ranks: PP first, PME last");
1649 case DdRankOrder::interleave:
1650 /* Interleave the PP-only and PME-only ranks */
1651 GMX_LOG(mdlog.info).appendText("Interleaving PP and PME ranks");
1652 *pmeRanks = dd_interleaved_pme_ranks(ddRankSetup);
1654 case DdRankOrder::cartesian:
1657 gmx_fatal(FARGS, "Invalid ddRankOrder=%d", static_cast<int>(ddRankOrder));
1660 if (dd_simnode2pmenode(ddRankSetup, cartSetup, *pmeRanks, cr, cr->sim_nodeid) == -1)
1662 cr->duty = DUTY_PME;
1669 /* Split the sim communicator into PP and PME only nodes */
1670 MPI_Comm_split(cr->mpi_comm_mysim,
1671 getThisRankDuties(cr),
1673 &cr->mpi_comm_mygroup);
1674 MPI_Comm_rank(cr->mpi_comm_mygroup, &cr->nodeid);
1678 GMX_LOG(mdlog.info).appendTextFormatted(
1679 "This rank does only %s work.\n",
1680 thisRankHasDuty(cr, DUTY_PP) ? "particle-particle" : "PME-mesh");
1685 /*! \brief Makes the PP communicator and the PME communicator, when needed
1687 * Returns the Cartesian rank setup.
1688 * Sets \p cr->mpi_comm_mygroup
1689 * For PP ranks, sets the DD PP cell index in \p ddCellIndex.
1690 * With separate PME ranks in interleaved order, set the PME ranks in \p pmeRanks.
1692 static CartesianRankSetup
1693 makeGroupCommunicators(const gmx::MDLogger &mdlog,
1694 const DDSettings &ddSettings,
1695 const DdRankOrder ddRankOrder,
1696 const DDRankSetup &ddRankSetup,
1699 std::vector<int> *pmeRanks)
1701 CartesianRankSetup cartSetup;
1703 if (ddRankSetup.usePmeOnlyRanks)
1705 /* Split the communicator into a PP and PME part */
1707 split_communicator(mdlog, cr, ddRankOrder, ddSettings.useCartesianReorder,
1708 ddRankSetup, ddCellIndex, pmeRanks);
1712 /* All nodes do PP and PME */
1713 /* We do not require separate communicators */
1714 cr->mpi_comm_mygroup = cr->mpi_comm_mysim;
1716 cartSetup.bCartesianPP = false;
1717 cartSetup.bCartesianPP_PME = false;
1723 /*! \brief For PP ranks, sets or makes the communicator
1725 * For PME ranks get the rank id.
1726 * For PP only ranks, sets the PME-only rank.
1728 static void setupGroupCommunication(const gmx::MDLogger &mdlog,
1729 const DDSettings &ddSettings,
1730 gmx::ArrayRef<const int> pmeRanks,
1734 const DDRankSetup &ddRankSetup = dd->comm->ddRankSetup;
1735 const CartesianRankSetup &cartSetup = dd->comm->cartesianRankSetup;
1737 if (thisRankHasDuty(cr, DUTY_PP))
1739 /* Copy or make a new PP communicator */
1741 /* We (possibly) reordered the nodes in split_communicator,
1742 * so it is no longer required in make_pp_communicator.
1744 const bool useCartesianReorder =
1745 (ddSettings.useCartesianReorder &&
1746 !cartSetup.bCartesianPP_PME);
1748 make_pp_communicator(mdlog, dd, cr, useCartesianReorder);
1752 receive_ddindex2simnodeid(dd, cr);
1755 if (!thisRankHasDuty(cr, DUTY_PME))
1757 /* Set up the commnuication to our PME node */
1758 dd->pme_nodeid = dd_simnode2pmenode(ddRankSetup, cartSetup, pmeRanks, cr, cr->sim_nodeid);
1759 dd->pme_receive_vir_ener = receive_vir_ener(dd, pmeRanks, cr);
1762 fprintf(debug, "My pme_nodeid %d receive ener %s\n",
1763 dd->pme_nodeid, gmx::boolToString(dd->pme_receive_vir_ener));
1768 dd->pme_nodeid = -1;
1771 /* We can not use DDMASTER(dd), because dd->masterrank is set later */
1774 dd->ma = std::make_unique<AtomDistribution>(dd->nc,
1775 dd->comm->cgs_gl.nr,
1776 dd->comm->cgs_gl.index[dd->comm->cgs_gl.nr]);
1780 static real *get_slb_frac(const gmx::MDLogger &mdlog,
1781 const char *dir, int nc, const char *size_string)
1783 real *slb_frac, tot;
1788 if (nc > 1 && size_string != nullptr)
1790 GMX_LOG(mdlog.info).appendTextFormatted(
1791 "Using static load balancing for the %s direction", dir);
1794 for (i = 0; i < nc; i++)
1797 sscanf(size_string, "%20lf%n", &dbl, &n);
1800 gmx_fatal(FARGS, "Incorrect or not enough DD cell size entries for direction %s: '%s'", dir, size_string);
1807 std::string relativeCellSizes = "Relative cell sizes:";
1808 for (i = 0; i < nc; i++)
1811 relativeCellSizes += gmx::formatString(" %5.3f", slb_frac[i]);
1813 GMX_LOG(mdlog.info).appendText(relativeCellSizes);
1819 static int multi_body_bondeds_count(const gmx_mtop_t *mtop)
1822 gmx_mtop_ilistloop_t iloop = gmx_mtop_ilistloop_init(mtop);
1824 while (const InteractionLists *ilists = gmx_mtop_ilistloop_next(iloop, &nmol))
1826 for (auto &ilist : extractILists(*ilists, IF_BOND))
1828 if (NRAL(ilist.functionType) > 2)
1830 n += nmol*(ilist.iatoms.size()/ilistStride(ilist));
1838 static int dd_getenv(const gmx::MDLogger &mdlog,
1839 const char *env_var, int def)
1845 val = getenv(env_var);
1848 if (sscanf(val, "%20d", &nst) <= 0)
1852 GMX_LOG(mdlog.info).appendTextFormatted(
1853 "Found env.var. %s = %s, using value %d",
1860 static void check_dd_restrictions(const gmx_domdec_t *dd,
1861 const t_inputrec *ir,
1862 const gmx::MDLogger &mdlog)
1864 if (ir->ePBC == epbcSCREW &&
1865 (dd->nc[XX] == 1 || dd->nc[YY] > 1 || dd->nc[ZZ] > 1))
1867 gmx_fatal(FARGS, "With pbc=%s can only do domain decomposition in the x-direction", epbc_names[ir->ePBC]);
1870 if (ir->nstlist == 0)
1872 gmx_fatal(FARGS, "Domain decomposition does not work with nstlist=0");
1875 if (ir->comm_mode == ecmANGULAR && ir->ePBC != epbcNONE)
1877 GMX_LOG(mdlog.warning).appendText("comm-mode angular will give incorrect results when the comm group partially crosses a periodic boundary");
1881 static real average_cellsize_min(const gmx_ddbox_t &ddbox,
1882 const ivec numDomains)
1884 real r = ddbox.box_size[XX];
1885 for (int d = 0; d < DIM; d++)
1887 if (numDomains[d] > 1)
1889 /* Check using the initial average cell size */
1890 r = std::min(r, ddbox.box_size[d]*ddbox.skew_fac[d]/numDomains[d]);
1897 /*! \brief Depending on the DLB initial value return the DLB switched off state or issue an error.
1899 static DlbState forceDlbOffOrBail(DlbState cmdlineDlbState,
1900 const std::string &reasonStr,
1901 const gmx::MDLogger &mdlog)
1903 std::string dlbNotSupportedErr = "Dynamic load balancing requested, but ";
1904 std::string dlbDisableNote = "NOTE: disabling dynamic load balancing as ";
1906 if (cmdlineDlbState == DlbState::onUser)
1908 gmx_fatal(FARGS, "%s", (dlbNotSupportedErr + reasonStr).c_str());
1910 else if (cmdlineDlbState == DlbState::offCanTurnOn)
1912 GMX_LOG(mdlog.info).appendText(dlbDisableNote + reasonStr);
1914 return DlbState::offForever;
1917 /*! \brief Return the dynamic load balancer's initial state based on initial conditions and user inputs.
1919 * This function parses the parameters of "-dlb" command line option setting
1920 * corresponding state values. Then it checks the consistency of the determined
1921 * state with other run parameters and settings. As a result, the initial state
1922 * may be altered or an error may be thrown if incompatibility of options is detected.
1924 * \param [in] mdlog Logger.
1925 * \param [in] dlbOption Enum value for the DLB option.
1926 * \param [in] bRecordLoad True if the load balancer is recording load information.
1927 * \param [in] mdrunOptions Options for mdrun.
1928 * \param [in] ir Pointer mdrun to input parameters.
1929 * \returns DLB initial/startup state.
1931 static DlbState determineInitialDlbState(const gmx::MDLogger &mdlog,
1932 DlbOption dlbOption, gmx_bool bRecordLoad,
1933 const gmx::MdrunOptions &mdrunOptions,
1934 const t_inputrec *ir)
1936 DlbState dlbState = DlbState::offCanTurnOn;
1940 case DlbOption::turnOnWhenUseful: dlbState = DlbState::offCanTurnOn; break;
1941 case DlbOption::no: dlbState = DlbState::offUser; break;
1942 case DlbOption::yes: dlbState = DlbState::onUser; break;
1943 default: gmx_incons("Invalid dlbOption enum value");
1946 /* Reruns don't support DLB: bail or override auto mode */
1947 if (mdrunOptions.rerun)
1949 std::string reasonStr = "it is not supported in reruns.";
1950 return forceDlbOffOrBail(dlbState, reasonStr, mdlog);
1953 /* Unsupported integrators */
1954 if (!EI_DYNAMICS(ir->eI))
1956 auto reasonStr = gmx::formatString("it is only supported with dynamics, not with integrator '%s'.", EI(ir->eI));
1957 return forceDlbOffOrBail(dlbState, reasonStr, mdlog);
1960 /* Without cycle counters we can't time work to balance on */
1963 std::string reasonStr = "cycle counters unsupported or not enabled in the operating system kernel.";
1964 return forceDlbOffOrBail(dlbState, reasonStr, mdlog);
1967 if (mdrunOptions.reproducible)
1969 std::string reasonStr = "you started a reproducible run.";
1972 case DlbState::offUser:
1974 case DlbState::offForever:
1975 GMX_RELEASE_ASSERT(false, "DlbState::offForever is not a valid initial state");
1977 case DlbState::offCanTurnOn:
1978 return forceDlbOffOrBail(dlbState, reasonStr, mdlog);
1979 case DlbState::onCanTurnOff:
1980 GMX_RELEASE_ASSERT(false, "DlbState::offCanTurnOff is not a valid initial state");
1982 case DlbState::onUser:
1983 return forceDlbOffOrBail(dlbState, reasonStr + " In load balanced runs binary reproducibility cannot be ensured.", mdlog);
1985 gmx_fatal(FARGS, "Death horror: undefined case (%d) for load balancing choice", static_cast<int>(dlbState));
1992 /* Sets the order of the DD dimensions, returns the number of DD dimensions */
1993 static int set_dd_dim(const ivec numDDCells,
1994 const DDSettings &ddSettings,
1998 if (ddSettings.useDDOrderZYX)
2000 /* Decomposition order z,y,x */
2001 for (int dim = DIM - 1; dim >= 0; dim--)
2003 if (numDDCells[dim] > 1)
2011 /* Decomposition order x,y,z */
2012 for (int dim = 0; dim < DIM; dim++)
2014 if (numDDCells[dim] > 1)
2023 /* Set dim[0] to avoid extra checks on ndim in several places */
2030 static gmx_domdec_comm_t *init_dd_comm()
2032 gmx_domdec_comm_t *comm = new gmx_domdec_comm_t;
2034 comm->n_load_have = 0;
2035 comm->n_load_collect = 0;
2037 comm->haveTurnedOffDlb = false;
2039 for (int i = 0; i < static_cast<int>(DDAtomRanges::Type::Number); i++)
2041 comm->sum_nat[i] = 0;
2045 comm->load_step = 0;
2048 clear_ivec(comm->load_lim);
2052 /* This should be replaced by a unique pointer */
2053 comm->balanceRegion = ddBalanceRegionAllocate();
2058 /* Returns whether mtop contains constraints and/or vsites */
2059 static bool systemHasConstraintsOrVsites(const gmx_mtop_t &mtop)
2061 auto ilistLoop = gmx_mtop_ilistloop_init(mtop);
2063 while (const InteractionLists *ilists = gmx_mtop_ilistloop_next(ilistLoop, &nmol))
2065 if (!extractILists(*ilists, IF_CONSTRAINT | IF_VSITE).empty())
2074 static void setupUpdateGroups(const gmx::MDLogger &mdlog,
2075 const gmx_mtop_t &mtop,
2076 const t_inputrec &inputrec,
2077 const real cutoffMargin,
2078 DDSystemInfo *systemInfo)
2080 /* When we have constraints and/or vsites, it is beneficial to use
2081 * update groups (when possible) to allow independent update of groups.
2083 if (!systemHasConstraintsOrVsites(mtop))
2085 /* No constraints or vsites, atoms can be updated independently */
2089 systemInfo->updateGroupingPerMoleculetype = gmx::makeUpdateGroups(mtop);
2090 systemInfo->useUpdateGroups =
2091 (!systemInfo->updateGroupingPerMoleculetype.empty() &&
2092 getenv("GMX_NO_UPDATEGROUPS") == nullptr);
2094 if (systemInfo->useUpdateGroups)
2096 int numUpdateGroups = 0;
2097 for (const auto &molblock : mtop.molblock)
2099 numUpdateGroups += molblock.nmol*systemInfo->updateGroupingPerMoleculetype[molblock.type].numBlocks();
2102 systemInfo->maxUpdateGroupRadius =
2103 computeMaxUpdateGroupRadius(mtop,
2104 systemInfo->updateGroupingPerMoleculetype,
2105 maxReferenceTemperature(inputrec));
2107 /* To use update groups, the large domain-to-domain cutoff distance
2108 * should be compatible with the box size.
2110 systemInfo->useUpdateGroups = (atomToAtomIntoDomainToDomainCutoff(*systemInfo, 0) < cutoffMargin);
2112 if (systemInfo->useUpdateGroups)
2114 GMX_LOG(mdlog.info).appendTextFormatted(
2115 "Using update groups, nr %d, average size %.1f atoms, max. radius %.3f nm\n",
2117 mtop.natoms/static_cast<double>(numUpdateGroups),
2118 systemInfo->maxUpdateGroupRadius);
2122 GMX_LOG(mdlog.info).appendTextFormatted("The combination of rlist and box size prohibits the use of update groups\n");
2123 systemInfo->updateGroupingPerMoleculetype.clear();
2128 UnitCellInfo::UnitCellInfo(const t_inputrec &ir) :
2129 npbcdim(ePBC2npbcdim(ir.ePBC)),
2130 numBoundedDimensions(inputrec2nboundeddim(&ir)),
2131 ddBoxIsDynamic(numBoundedDimensions < DIM || inputrecDynamicBox(&ir)),
2132 haveScrewPBC(ir.ePBC == epbcSCREW)
2136 /*! \brief Generate the simulation system information */
2138 getSystemInfo(const gmx::MDLogger &mdlog,
2140 const DomdecOptions &options,
2141 const gmx_mtop_t *mtop,
2142 const t_inputrec *ir,
2144 gmx::ArrayRef<const gmx::RVec> xGlobal)
2146 const real tenPercentMargin = 1.1;
2148 DDSystemInfo systemInfo;
2150 /* We need to decide on update groups early, as this affects communication distances */
2151 systemInfo.useUpdateGroups = false;
2152 if (ir->cutoff_scheme == ecutsVERLET)
2154 real cutoffMargin = std::sqrt(max_cutoff2(ir->ePBC, box)) - ir->rlist;
2155 setupUpdateGroups(mdlog, *mtop, *ir, cutoffMargin, &systemInfo);
2158 // TODO: Check whether all bondeds are within update groups
2159 systemInfo.haveInterDomainBondeds = (mtop->natoms > gmx_mtop_num_molecules(*mtop) ||
2160 mtop->bIntermolecularInteractions);
2161 systemInfo.haveInterDomainMultiBodyBondeds = (multi_body_bondeds_count(mtop) > 0);
2163 if (systemInfo.useUpdateGroups)
2165 systemInfo.haveSplitConstraints = false;
2166 systemInfo.haveSplitSettles = false;
2170 systemInfo.haveSplitConstraints = gmx::inter_charge_group_constraints(*mtop);
2171 systemInfo.haveSplitSettles = gmx::inter_charge_group_settles(*mtop);
2176 /* Set the cut-off to some very large value,
2177 * so we don't need if statements everywhere in the code.
2178 * We use sqrt, since the cut-off is squared in some places.
2180 systemInfo.cutoff = GMX_CUTOFF_INF;
2184 systemInfo.cutoff = atomToAtomIntoDomainToDomainCutoff(systemInfo, ir->rlist);
2186 systemInfo.minCutoffForMultiBody = 0;
2188 /* Determine the minimum cell size limit, affected by many factors */
2189 systemInfo.cellsizeLimit = 0;
2190 systemInfo.filterBondedCommunication = false;
2192 /* We do not allow home atoms to move beyond the neighboring domain
2193 * between domain decomposition steps, which limits the cell size.
2194 * Get an estimate of cell size limit due to atom displacement.
2195 * In most cases this is a large overestimate, because it assumes
2196 * non-interaction atoms.
2197 * We set the chance to 1 in a trillion steps.
2199 constexpr real c_chanceThatAtomMovesBeyondDomain = 1e-12;
2200 const real limitForAtomDisplacement =
2201 minCellSizeForAtomDisplacement(*mtop, *ir,
2202 systemInfo.updateGroupingPerMoleculetype,
2203 c_chanceThatAtomMovesBeyondDomain);
2204 GMX_LOG(mdlog.info).appendTextFormatted(
2205 "Minimum cell size due to atom displacement: %.3f nm",
2206 limitForAtomDisplacement);
2208 systemInfo.cellsizeLimit = std::max(systemInfo.cellsizeLimit,
2209 limitForAtomDisplacement);
2211 /* TODO: PME decomposition currently requires atoms not to be more than
2212 * 2/3 of comm->cutoff, which is >=rlist, outside of their domain.
2213 * In nearly all cases, limitForAtomDisplacement will be smaller
2214 * than 2/3*rlist, so the PME requirement is satisfied.
2215 * But it would be better for both correctness and performance
2216 * to use limitForAtomDisplacement instead of 2/3*comm->cutoff.
2217 * Note that we would need to improve the pairlist buffer case.
2220 if (systemInfo.haveInterDomainBondeds)
2222 if (options.minimumCommunicationRange > 0)
2224 systemInfo.minCutoffForMultiBody =
2225 atomToAtomIntoDomainToDomainCutoff(systemInfo, options.minimumCommunicationRange);
2226 if (options.useBondedCommunication)
2228 systemInfo.filterBondedCommunication = (systemInfo.minCutoffForMultiBody > systemInfo.cutoff);
2232 systemInfo.cutoff = std::max(systemInfo.cutoff,
2233 systemInfo.minCutoffForMultiBody);
2236 else if (ir->bPeriodicMols)
2238 /* Can not easily determine the required cut-off */
2239 GMX_LOG(mdlog.warning).appendText("NOTE: Periodic molecules are present in this system. Because of this, the domain decomposition algorithm cannot easily determine the minimum cell size that it requires for treating bonded interactions. Instead, domain decomposition will assume that half the non-bonded cut-off will be a suitable lower bound.");
2240 systemInfo.minCutoffForMultiBody = systemInfo.cutoff/2;
2248 dd_bonded_cg_distance(mdlog, mtop, ir, as_rvec_array(xGlobal.data()), box,
2249 options.checkBondedInteractions,
2252 gmx_bcast(sizeof(r_2b), &r_2b, cr);
2253 gmx_bcast(sizeof(r_mb), &r_mb, cr);
2255 /* We use an initial margin of 10% for the minimum cell size,
2256 * except when we are just below the non-bonded cut-off.
2258 if (options.useBondedCommunication)
2260 if (std::max(r_2b, r_mb) > systemInfo.cutoff)
2262 const real r_bonded = std::max(r_2b, r_mb);
2263 systemInfo.minCutoffForMultiBody = tenPercentMargin*r_bonded;
2264 /* This is the (only) place where we turn on the filtering */
2265 systemInfo.filterBondedCommunication = true;
2269 const real r_bonded = r_mb;
2270 systemInfo.minCutoffForMultiBody = std::min(tenPercentMargin*r_bonded,
2273 /* We determine cutoff_mbody later */
2274 systemInfo.increaseMultiBodyCutoff = true;
2278 /* No special bonded communication,
2279 * simply increase the DD cut-off.
2281 systemInfo.minCutoffForMultiBody = tenPercentMargin*std::max(r_2b, r_mb);
2282 systemInfo.cutoff = std::max(systemInfo.cutoff,
2283 systemInfo.minCutoffForMultiBody);
2286 GMX_LOG(mdlog.info).appendTextFormatted(
2287 "Minimum cell size due to bonded interactions: %.3f nm",
2288 systemInfo.minCutoffForMultiBody);
2290 systemInfo.cellsizeLimit = std::max(systemInfo.cellsizeLimit,
2291 systemInfo.minCutoffForMultiBody);
2294 systemInfo.constraintCommunicationRange = 0;
2295 if (systemInfo.haveSplitConstraints && options.constraintCommunicationRange <= 0)
2297 /* There is a cell size limit due to the constraints (P-LINCS) */
2298 systemInfo.constraintCommunicationRange = gmx::constr_r_max(mdlog, mtop, ir);
2299 GMX_LOG(mdlog.info).appendTextFormatted(
2300 "Estimated maximum distance required for P-LINCS: %.3f nm",
2301 systemInfo.constraintCommunicationRange);
2302 if (systemInfo.constraintCommunicationRange > systemInfo.cellsizeLimit)
2304 GMX_LOG(mdlog.info).appendText("This distance will limit the DD cell size, you can override this with -rcon");
2307 else if (options.constraintCommunicationRange > 0)
2309 /* Here we do not check for dd->splitConstraints.
2310 * because one can also set a cell size limit for virtual sites only
2311 * and at this point we don't know yet if there are intercg v-sites.
2313 GMX_LOG(mdlog.info).appendTextFormatted(
2314 "User supplied maximum distance required for P-LINCS: %.3f nm",
2315 options.constraintCommunicationRange);
2316 systemInfo.constraintCommunicationRange = options.constraintCommunicationRange;
2318 systemInfo.cellsizeLimit = std::max(systemInfo.cellsizeLimit,
2319 systemInfo.constraintCommunicationRange);
2324 /*! \brief Set the cell size and interaction limits, as well as the DD grid
2326 * Also computes the initial ddbox.
2329 getDDGridSetup(const gmx::MDLogger &mdlog,
2331 const DomdecOptions &options,
2332 const DDSettings &ddSettings,
2333 const DDSystemInfo &systemInfo,
2334 const gmx_mtop_t *mtop,
2335 const t_inputrec *ir,
2337 gmx::ArrayRef<const gmx::RVec> xGlobal,
2340 DDGridSetup ddGridSetup;
2342 if (options.numCells[XX] > 0)
2344 copy_ivec(options.numCells, ddGridSetup.numDomains);
2345 set_ddbox_cr(*cr, &ddGridSetup.numDomains, *ir, box, xGlobal, ddbox);
2347 if (options.numPmeRanks >= 0)
2349 ddGridSetup.numPmeOnlyRanks = options.numPmeRanks;
2353 /* When the DD grid is set explicitly and -npme is set to auto,
2354 * don't use PME ranks. We check later if the DD grid is
2355 * compatible with the total number of ranks.
2357 ddGridSetup.numPmeOnlyRanks = 0;
2362 set_ddbox_cr(*cr, nullptr, *ir, box, xGlobal, ddbox);
2364 /* We need to choose the optimal DD grid and possibly PME nodes */
2366 dd_choose_grid(mdlog, cr, ir, mtop, box, ddbox,
2367 options.numPmeRanks,
2368 ddSettings.request1DAnd1Pulse,
2369 !isDlbDisabled(ddSettings.initialDlbState),
2373 if (ddGridSetup.numDomains[XX] == 0)
2376 gmx_bool bC = (systemInfo.haveSplitConstraints &&
2377 systemInfo.constraintCommunicationRange > systemInfo.minCutoffForMultiBody);
2378 sprintf(buf, "Change the number of ranks or mdrun option %s%s%s",
2379 !bC ? "-rdd" : "-rcon",
2380 ddSettings.initialDlbState != DlbState::offUser ? " or -dds" : "",
2381 bC ? " or your LINCS settings" : "");
2383 gmx_fatal_collective(FARGS, cr->mpi_comm_mysim, MASTER(cr),
2384 "There is no domain decomposition for %d ranks that is compatible with the given box and a minimum cell size of %g nm\n"
2386 "Look in the log file for details on the domain decomposition",
2387 cr->nnodes - ddGridSetup.numPmeOnlyRanks,
2388 ddGridSetup.cellsizeLimit,
2393 const real acs = average_cellsize_min(*ddbox, ddGridSetup.numDomains);
2394 if (acs < systemInfo.cellsizeLimit)
2396 if (options.numCells[XX] <= 0)
2398 GMX_RELEASE_ASSERT(false, "dd_choose_grid() should return a grid that satisfies the cell size limits");
2402 gmx_fatal_collective(FARGS, cr->mpi_comm_mysim, MASTER(cr),
2403 "The initial cell size (%f) is smaller than the cell size limit (%f), change options -dd, -rdd or -rcon, see the log file for details",
2404 acs, systemInfo.cellsizeLimit);
2408 const int numPPRanks = ddGridSetup.numDomains[XX]*ddGridSetup.numDomains[YY]*ddGridSetup.numDomains[ZZ];
2409 if (cr->nnodes - numPPRanks != ddGridSetup.numPmeOnlyRanks)
2411 gmx_fatal_collective(FARGS, cr->mpi_comm_mysim, MASTER(cr),
2412 "The size of the domain decomposition grid (%d) does not match the number of ranks (%d). The total number of ranks is %d",
2413 numPPRanks, cr->nnodes - ddGridSetup.numPmeOnlyRanks, cr->nnodes);
2415 if (ddGridSetup.numPmeOnlyRanks > numPPRanks)
2417 gmx_fatal_collective(FARGS, cr->mpi_comm_mysim, MASTER(cr),
2418 "The number of separate PME ranks (%d) is larger than the number of PP ranks (%d), this is not supported.", ddGridSetup.numPmeOnlyRanks, numPPRanks);
2421 ddGridSetup.numDDDimensions = set_dd_dim(ddGridSetup.numDomains, ddSettings,
2422 ddGridSetup.ddDimensions);
2427 /*! \brief Set the cell size and interaction limits, as well as the DD grid */
2429 getDDRankSetup(const gmx::MDLogger &mdlog,
2431 const DDGridSetup &ddGridSetup,
2432 const t_inputrec &ir)
2434 GMX_LOG(mdlog.info).appendTextFormatted(
2435 "Domain decomposition grid %d x %d x %d, separate PME ranks %d",
2436 ddGridSetup.numDomains[XX], ddGridSetup.numDomains[YY], ddGridSetup.numDomains[ZZ],
2437 ddGridSetup.numPmeOnlyRanks);
2439 DDRankSetup ddRankSetup;
2441 ddRankSetup.numPPRanks = cr->nnodes - ddGridSetup.numPmeOnlyRanks;
2442 copy_ivec(ddGridSetup.numDomains, ddRankSetup.numPPCells);
2444 ddRankSetup.usePmeOnlyRanks = (ddGridSetup.numPmeOnlyRanks > 0);
2445 if (ddRankSetup.usePmeOnlyRanks)
2447 ddRankSetup.numRanksDoingPme = ddGridSetup.numPmeOnlyRanks;
2451 ddRankSetup.numRanksDoingPme = ddGridSetup.numDomains[XX]*ddGridSetup.numDomains[YY]*ddGridSetup.numDomains[ZZ];
2454 if (EEL_PME(ir.coulombtype) || EVDW_PME(ir.vdwtype))
2456 /* The following choices should match those
2457 * in comm_cost_est in domdec_setup.c.
2458 * Note that here the checks have to take into account
2459 * that the decomposition might occur in a different order than xyz
2460 * (for instance through the env.var. GMX_DD_ORDER_ZYX),
2461 * in which case they will not match those in comm_cost_est,
2462 * but since that is mainly for testing purposes that's fine.
2464 if (ddGridSetup.numDDDimensions >= 2 &&
2465 ddGridSetup.ddDimensions[0] == XX &&
2466 ddGridSetup.ddDimensions[1] == YY &&
2467 ddRankSetup.numRanksDoingPme > ddGridSetup.numDomains[XX] &&
2468 ddRankSetup.numRanksDoingPme % ddGridSetup.numDomains[XX] == 0 &&
2469 getenv("GMX_PMEONEDD") == nullptr)
2471 ddRankSetup.npmedecompdim = 2;
2472 ddRankSetup.npmenodes_x = ddGridSetup.numDomains[XX];
2473 ddRankSetup.npmenodes_y = ddRankSetup.numRanksDoingPme/ddRankSetup.npmenodes_x;
2477 /* In case nc is 1 in both x and y we could still choose to
2478 * decompose pme in y instead of x, but we use x for simplicity.
2480 ddRankSetup.npmedecompdim = 1;
2481 if (ddGridSetup.ddDimensions[0] == YY)
2483 ddRankSetup.npmenodes_x = 1;
2484 ddRankSetup.npmenodes_y = ddRankSetup.numRanksDoingPme;
2488 ddRankSetup.npmenodes_x = ddRankSetup.numRanksDoingPme;
2489 ddRankSetup.npmenodes_y = 1;
2492 GMX_LOG(mdlog.info).appendTextFormatted(
2493 "PME domain decomposition: %d x %d x %d",
2494 ddRankSetup.npmenodes_x, ddRankSetup.npmenodes_y, 1);
2498 ddRankSetup.npmedecompdim = 0;
2499 ddRankSetup.npmenodes_x = 0;
2500 ddRankSetup.npmenodes_y = 0;
2506 /*! \brief Set the cell size and interaction limits */
2507 static void set_dd_limits(const gmx::MDLogger &mdlog,
2508 t_commrec *cr, gmx_domdec_t *dd,
2509 const DomdecOptions &options,
2510 const DDSettings &ddSettings,
2511 const DDSystemInfo &systemInfo,
2512 const DDGridSetup &ddGridSetup,
2513 const gmx_mtop_t *mtop,
2514 const t_inputrec *ir,
2515 const gmx_ddbox_t &ddbox)
2517 gmx_domdec_comm_t *comm = dd->comm;
2518 comm->ddSettings = ddSettings;
2520 /* Initialize to GPU share count to 0, might change later */
2521 comm->nrank_gpu_shared = 0;
2523 comm->dlbState = comm->ddSettings.initialDlbState;
2524 dd_dlb_set_should_check_whether_to_turn_dlb_on(dd, TRUE);
2525 /* To consider turning DLB on after 2*nstlist steps we need to check
2526 * at partitioning count 3. Thus we need to increase the first count by 2.
2528 comm->ddPartioningCountFirstDlbOff += 2;
2530 comm->bPMELoadBalDLBLimits = FALSE;
2532 /* Allocate the charge group/atom sorting struct */
2533 comm->sort = std::make_unique<gmx_domdec_sort_t>();
2535 comm->systemInfo = systemInfo;
2537 if (systemInfo.useUpdateGroups)
2539 /* Note: We would like to use dd->nnodes for the atom count estimate,
2540 * but that is not yet available here. But this anyhow only
2541 * affect performance up to the second dd_partition_system call.
2543 const int homeAtomCountEstimate = mtop->natoms/cr->nnodes;
2544 comm->updateGroupsCog =
2545 std::make_unique<gmx::UpdateGroupsCog>(*mtop,
2546 systemInfo.updateGroupingPerMoleculetype,
2547 maxReferenceTemperature(*ir),
2548 homeAtomCountEstimate);
2551 comm->cgs_gl = gmx_mtop_global_cgs(mtop);
2553 /* Set the DD setup given by ddGridSetup */
2554 copy_ivec(ddGridSetup.numDomains, dd->nc);
2555 dd->ndim = ddGridSetup.numDDDimensions;
2556 copy_ivec(ddGridSetup.ddDimensions, dd->dim);
2558 dd->nnodes = dd->nc[XX]*dd->nc[YY]*dd->nc[ZZ];
2560 snew(comm->slb_frac, DIM);
2561 if (isDlbDisabled(comm))
2563 comm->slb_frac[XX] = get_slb_frac(mdlog, "x", dd->nc[XX], options.cellSizeX);
2564 comm->slb_frac[YY] = get_slb_frac(mdlog, "y", dd->nc[YY], options.cellSizeY);
2565 comm->slb_frac[ZZ] = get_slb_frac(mdlog, "z", dd->nc[ZZ], options.cellSizeZ);
2568 /* Set the multi-body cut-off and cellsize limit for DLB */
2569 comm->cutoff_mbody = systemInfo.minCutoffForMultiBody;
2570 comm->cellsize_limit = systemInfo.cellsizeLimit;
2571 if (systemInfo.haveInterDomainBondeds && systemInfo.increaseMultiBodyCutoff)
2573 if (systemInfo.filterBondedCommunication || !isDlbDisabled(comm))
2575 /* Set the bonded communication distance to halfway
2576 * the minimum and the maximum,
2577 * since the extra communication cost is nearly zero.
2579 real acs = average_cellsize_min(ddbox, dd->nc);
2580 comm->cutoff_mbody = 0.5*(systemInfo.minCutoffForMultiBody + acs);
2581 if (!isDlbDisabled(comm))
2583 /* Check if this does not limit the scaling */
2584 comm->cutoff_mbody = std::min(comm->cutoff_mbody,
2585 options.dlbScaling*acs);
2587 if (!systemInfo.filterBondedCommunication)
2589 /* Without bBondComm do not go beyond the n.b. cut-off */
2590 comm->cutoff_mbody = std::min(comm->cutoff_mbody, systemInfo.cutoff);
2591 if (comm->cellsize_limit >= systemInfo.cutoff)
2593 /* We don't loose a lot of efficieny
2594 * when increasing it to the n.b. cut-off.
2595 * It can even be slightly faster, because we need
2596 * less checks for the communication setup.
2598 comm->cutoff_mbody = systemInfo.cutoff;
2601 /* Check if we did not end up below our original limit */
2602 comm->cutoff_mbody = std::max(comm->cutoff_mbody,
2603 systemInfo.minCutoffForMultiBody);
2605 if (comm->cutoff_mbody > comm->cellsize_limit)
2607 comm->cellsize_limit = comm->cutoff_mbody;
2610 /* Without DLB and cutoff_mbody<cutoff, cutoff_mbody is dynamic */
2615 fprintf(debug, "Bonded atom communication beyond the cut-off: %s\n"
2616 "cellsize limit %f\n",
2617 gmx::boolToString(systemInfo.filterBondedCommunication),
2618 comm->cellsize_limit);
2623 check_dd_restrictions(dd, ir, mdlog);
2627 static char *init_bLocalCG(const gmx_mtop_t *mtop)
2632 ncg = ncg_mtop(mtop);
2633 snew(bLocalCG, ncg);
2634 for (cg = 0; cg < ncg; cg++)
2636 bLocalCG[cg] = FALSE;
2642 void dd_init_bondeds(FILE *fplog,
2644 const gmx_mtop_t *mtop,
2645 const gmx_vsite_t *vsite,
2646 const t_inputrec *ir,
2647 gmx_bool bBCheck, cginfo_mb_t *cginfo_mb)
2649 gmx_domdec_comm_t *comm;
2651 dd_make_reverse_top(fplog, dd, mtop, vsite, ir, bBCheck);
2655 if (comm->systemInfo.filterBondedCommunication)
2657 /* Communicate atoms beyond the cut-off for bonded interactions */
2660 comm->cglink = make_charge_group_links(mtop, dd, cginfo_mb);
2662 comm->bLocalCG = init_bLocalCG(mtop);
2666 /* Only communicate atoms based on cut-off */
2667 comm->cglink = nullptr;
2668 comm->bLocalCG = nullptr;
2672 static void writeSettings(gmx::TextWriter *log,
2674 const gmx_mtop_t *mtop,
2675 const t_inputrec *ir,
2676 gmx_bool bDynLoadBal,
2678 const gmx_ddbox_t *ddbox)
2680 gmx_domdec_comm_t *comm;
2689 log->writeString("The maximum number of communication pulses is:");
2690 for (d = 0; d < dd->ndim; d++)
2692 log->writeStringFormatted(" %c %d", dim2char(dd->dim[d]), comm->cd[d].np_dlb);
2694 log->ensureLineBreak();
2695 log->writeLineFormatted("The minimum size for domain decomposition cells is %.3f nm", comm->cellsize_limit);
2696 log->writeLineFormatted("The requested allowed shrink of DD cells (option -dds) is: %.2f", dlb_scale);
2697 log->writeString("The allowed shrink of domain decomposition cells is:");
2698 for (d = 0; d < DIM; d++)
2702 if (d >= ddbox->npbcdim && dd->nc[d] == 2)
2709 comm->cellsize_min_dlb[d]/
2710 (ddbox->box_size[d]*ddbox->skew_fac[d]/dd->nc[d]);
2712 log->writeStringFormatted(" %c %.2f", dim2char(d), shrink);
2715 log->ensureLineBreak();
2719 set_dd_cell_sizes_slb(dd, ddbox, setcellsizeslbPULSE_ONLY, np);
2720 log->writeString("The initial number of communication pulses is:");
2721 for (d = 0; d < dd->ndim; d++)
2723 log->writeStringFormatted(" %c %d", dim2char(dd->dim[d]), np[dd->dim[d]]);
2725 log->ensureLineBreak();
2726 log->writeString("The initial domain decomposition cell size is:");
2727 for (d = 0; d < DIM; d++)
2731 log->writeStringFormatted(" %c %.2f nm",
2732 dim2char(d), dd->comm->cellsize_min[d]);
2735 log->ensureLineBreak();
2739 const bool haveInterDomainVsites =
2740 (countInterUpdategroupVsites(*mtop, comm->systemInfo.updateGroupingPerMoleculetype) != 0);
2742 if (comm->systemInfo.haveInterDomainBondeds ||
2743 haveInterDomainVsites ||
2744 comm->systemInfo.haveSplitConstraints ||
2745 comm->systemInfo.haveSplitSettles)
2747 std::string decompUnits;
2748 if (comm->systemInfo.useUpdateGroups)
2750 decompUnits = "atom groups";
2754 decompUnits = "atoms";
2757 log->writeLineFormatted("The maximum allowed distance for %s involved in interactions is:", decompUnits.c_str());
2758 log->writeLineFormatted("%40s %-7s %6.3f nm", "non-bonded interactions", "", comm->systemInfo.cutoff);
2762 limit = dd->comm->cellsize_limit;
2766 if (dd->unitCellInfo.ddBoxIsDynamic)
2768 log->writeLine("(the following are initial values, they could change due to box deformation)");
2770 limit = dd->comm->cellsize_min[XX];
2771 for (d = 1; d < DIM; d++)
2773 limit = std::min(limit, dd->comm->cellsize_min[d]);
2777 if (comm->systemInfo.haveInterDomainBondeds)
2779 log->writeLineFormatted("%40s %-7s %6.3f nm",
2780 "two-body bonded interactions", "(-rdd)",
2781 std::max(comm->systemInfo.cutoff, comm->cutoff_mbody));
2782 log->writeLineFormatted("%40s %-7s %6.3f nm",
2783 "multi-body bonded interactions", "(-rdd)",
2784 (comm->systemInfo.filterBondedCommunication || isDlbOn(dd->comm)) ? comm->cutoff_mbody : std::min(comm->systemInfo.cutoff, limit));
2786 if (haveInterDomainVsites)
2788 log->writeLineFormatted("%40s %-7s %6.3f nm",
2789 "virtual site constructions", "(-rcon)", limit);
2791 if (comm->systemInfo.haveSplitConstraints || comm->systemInfo.haveSplitSettles)
2793 std::string separation = gmx::formatString("atoms separated by up to %d constraints",
2795 log->writeLineFormatted("%40s %-7s %6.3f nm\n",
2796 separation.c_str(), "(-rcon)", limit);
2798 log->ensureLineBreak();
2802 static void logSettings(const gmx::MDLogger &mdlog,
2804 const gmx_mtop_t *mtop,
2805 const t_inputrec *ir,
2807 const gmx_ddbox_t *ddbox)
2809 gmx::StringOutputStream stream;
2810 gmx::TextWriter log(&stream);
2811 writeSettings(&log, dd, mtop, ir, isDlbOn(dd->comm), dlb_scale, ddbox);
2812 if (dd->comm->dlbState == DlbState::offCanTurnOn)
2815 log.ensureEmptyLine();
2816 log.writeLine("When dynamic load balancing gets turned on, these settings will change to:");
2818 writeSettings(&log, dd, mtop, ir, true, dlb_scale, ddbox);
2820 GMX_LOG(mdlog.info).asParagraph().appendText(stream.toString());
2823 static void set_cell_limits_dlb(const gmx::MDLogger &mdlog,
2826 const t_inputrec *ir,
2827 const gmx_ddbox_t *ddbox)
2829 gmx_domdec_comm_t *comm;
2830 int d, dim, npulse, npulse_d_max, npulse_d;
2835 bNoCutOff = (ir->rvdw == 0 || ir->rcoulomb == 0);
2837 /* Determine the maximum number of comm. pulses in one dimension */
2839 comm->cellsize_limit = std::max(comm->cellsize_limit, comm->cutoff_mbody);
2841 /* Determine the maximum required number of grid pulses */
2842 if (comm->cellsize_limit >= comm->systemInfo.cutoff)
2844 /* Only a single pulse is required */
2847 else if (!bNoCutOff && comm->cellsize_limit > 0)
2849 /* We round down slightly here to avoid overhead due to the latency
2850 * of extra communication calls when the cut-off
2851 * would be only slightly longer than the cell size.
2852 * Later cellsize_limit is redetermined,
2853 * so we can not miss interactions due to this rounding.
2855 npulse = static_cast<int>(0.96 + comm->systemInfo.cutoff/comm->cellsize_limit);
2859 /* There is no cell size limit */
2860 npulse = std::max(dd->nc[XX]-1, std::max(dd->nc[YY]-1, dd->nc[ZZ]-1));
2863 if (!bNoCutOff && npulse > 1)
2865 /* See if we can do with less pulses, based on dlb_scale */
2867 for (d = 0; d < dd->ndim; d++)
2870 npulse_d = static_cast<int>(1 + dd->nc[dim]*comm->systemInfo.cutoff
2871 /(ddbox->box_size[dim]*ddbox->skew_fac[dim]*dlb_scale));
2872 npulse_d_max = std::max(npulse_d_max, npulse_d);
2874 npulse = std::min(npulse, npulse_d_max);
2877 /* This env var can override npulse */
2878 d = dd_getenv(mdlog, "GMX_DD_NPULSE", 0);
2885 comm->bVacDLBNoLimit = (ir->ePBC == epbcNONE);
2886 for (d = 0; d < dd->ndim; d++)
2888 if (comm->ddSettings.request1DAnd1Pulse)
2890 comm->cd[d].np_dlb = 1;
2894 comm->cd[d].np_dlb = std::min(npulse, dd->nc[dd->dim[d]]-1);
2895 comm->maxpulse = std::max(comm->maxpulse, comm->cd[d].np_dlb);
2897 if (comm->cd[d].np_dlb < dd->nc[dd->dim[d]]-1)
2899 comm->bVacDLBNoLimit = FALSE;
2903 /* cellsize_limit is set for LINCS in init_domain_decomposition */
2904 if (!comm->bVacDLBNoLimit)
2906 comm->cellsize_limit = std::max(comm->cellsize_limit,
2907 comm->systemInfo.cutoff/comm->maxpulse);
2909 comm->cellsize_limit = std::max(comm->cellsize_limit, comm->cutoff_mbody);
2910 /* Set the minimum cell size for each DD dimension */
2911 for (d = 0; d < dd->ndim; d++)
2913 if (comm->bVacDLBNoLimit ||
2914 comm->cd[d].np_dlb*comm->cellsize_limit >= comm->systemInfo.cutoff)
2916 comm->cellsize_min_dlb[dd->dim[d]] = comm->cellsize_limit;
2920 comm->cellsize_min_dlb[dd->dim[d]] =
2921 comm->systemInfo.cutoff/comm->cd[d].np_dlb;
2924 if (comm->cutoff_mbody <= 0)
2926 comm->cutoff_mbody = std::min(comm->systemInfo.cutoff, comm->cellsize_limit);
2934 gmx_bool dd_bonded_molpbc(const gmx_domdec_t *dd, int ePBC)
2936 /* If each molecule is a single charge group
2937 * or we use domain decomposition for each periodic dimension,
2938 * we do not need to take pbc into account for the bonded interactions.
2940 return (ePBC != epbcNONE && dd->comm->systemInfo.haveInterDomainBondeds &&
2943 (dd->nc[ZZ] > 1 || ePBC == epbcXY)));
2946 /*! \brief Sets grid size limits and PP-PME setup, prints settings to log */
2947 static void set_ddgrid_parameters(const gmx::MDLogger &mdlog,
2948 gmx_domdec_t *dd, real dlb_scale,
2949 const gmx_mtop_t *mtop, const t_inputrec *ir,
2950 const gmx_ddbox_t *ddbox)
2952 gmx_domdec_comm_t *comm = dd->comm;
2953 DDRankSetup &ddRankSetup = comm->ddRankSetup;
2955 if (EEL_PME(ir->coulombtype) || EVDW_PME(ir->vdwtype))
2957 init_ddpme(dd, &ddRankSetup.ddpme[0], 0);
2958 if (ddRankSetup.npmedecompdim >= 2)
2960 init_ddpme(dd, &ddRankSetup.ddpme[1], 1);
2965 ddRankSetup.numRanksDoingPme = 0;
2966 if (dd->pme_nodeid >= 0)
2968 gmx_fatal_collective(FARGS, dd->mpi_comm_all, DDMASTER(dd),
2969 "Can not have separate PME ranks without PME electrostatics");
2975 fprintf(debug, "The DD cut-off is %f\n", comm->systemInfo.cutoff);
2977 if (!isDlbDisabled(comm))
2979 set_cell_limits_dlb(mdlog, dd, dlb_scale, ir, ddbox);
2982 logSettings(mdlog, dd, mtop, ir, dlb_scale, ddbox);
2985 if (ir->ePBC == epbcNONE)
2987 vol_frac = 1 - 1/static_cast<double>(dd->nnodes);
2992 (1 + comm_box_frac(dd->nc, comm->systemInfo.cutoff, ddbox))/static_cast<double>(dd->nnodes);
2996 fprintf(debug, "Volume fraction for all DD zones: %f\n", vol_frac);
2998 int natoms_tot = comm->cgs_gl.index[comm->cgs_gl.nr];
3000 dd->ga2la = new gmx_ga2la_t(natoms_tot,
3001 static_cast<int>(vol_frac*natoms_tot));
3004 /*! \brief Get some important DD parameters which can be modified by env.vars */
3006 getDDSettings(const gmx::MDLogger &mdlog,
3007 const DomdecOptions &options,
3008 const gmx::MdrunOptions &mdrunOptions,
3009 const t_inputrec &ir)
3011 DDSettings ddSettings;
3013 ddSettings.useSendRecv2 = (dd_getenv(mdlog, "GMX_DD_USE_SENDRECV2", 0) != 0);
3014 ddSettings.dlb_scale_lim = dd_getenv(mdlog, "GMX_DLB_MAX_BOX_SCALING", 10);
3015 // TODO GPU halo exchange requires a 1D single-pulse DD, and when
3016 // it is properly integrated the hack with GMX_GPU_DD_COMMS should
3018 ddSettings.request1DAnd1Pulse = (bool(dd_getenv(mdlog, "GMX_DD_1D_1PULSE", 0)) ||
3019 (bool(getenv("GMX_GPU_DD_COMMS") != nullptr &&
3021 (GMX_GPU == GMX_GPU_CUDA))));
3022 ddSettings.useDDOrderZYX = bool(dd_getenv(mdlog, "GMX_DD_ORDER_ZYX", 0));
3023 ddSettings.useCartesianReorder = bool(dd_getenv(mdlog, "GMX_NO_CART_REORDER", 1));
3024 ddSettings.eFlop = dd_getenv(mdlog, "GMX_DLB_BASED_ON_FLOPS", 0);
3025 const int recload = dd_getenv(mdlog, "GMX_DD_RECORD_LOAD", 1);
3026 ddSettings.nstDDDump = dd_getenv(mdlog, "GMX_DD_NST_DUMP", 0);
3027 ddSettings.nstDDDumpGrid = dd_getenv(mdlog, "GMX_DD_NST_DUMP_GRID", 0);
3028 ddSettings.DD_debug = dd_getenv(mdlog, "GMX_DD_DEBUG", 0);
3030 if (ddSettings.useSendRecv2)
3032 GMX_LOG(mdlog.info).appendText("Will use two sequential MPI_Sendrecv calls instead of two simultaneous non-blocking MPI_Irecv and MPI_Isend pairs for constraint and vsite communication");
3035 if (ddSettings.eFlop)
3037 GMX_LOG(mdlog.info).appendText("Will load balance based on FLOP count");
3038 ddSettings.recordLoad = true;
3042 ddSettings.recordLoad = (wallcycle_have_counter() && recload > 0);
3045 ddSettings.initialDlbState =
3046 determineInitialDlbState(mdlog, options.dlbOption, ddSettings.recordLoad, mdrunOptions, &ir);
3047 GMX_LOG(mdlog.info).appendTextFormatted("Dynamic load balancing: %s",
3048 edlbs_names[static_cast<int>(ddSettings.initialDlbState)]);
3053 gmx_domdec_t::gmx_domdec_t(const t_inputrec &ir) :
3058 gmx_domdec_t *init_domain_decomposition(const gmx::MDLogger &mdlog,
3060 const DomdecOptions &options,
3061 const gmx::MdrunOptions &mdrunOptions,
3062 const gmx_mtop_t *mtop,
3063 const t_inputrec *ir,
3065 gmx::ArrayRef<const gmx::RVec> xGlobal,
3066 gmx::LocalAtomSetManager *atomSets)
3068 GMX_LOG(mdlog.info).appendTextFormatted(
3069 "\nInitializing Domain Decomposition on %d ranks", cr->nnodes);
3071 DDSettings ddSettings = getDDSettings(mdlog, options, mdrunOptions, *ir);
3072 if (ddSettings.eFlop > 1)
3074 /* Ensure that we have different random flop counts on different ranks */
3075 srand(1 + cr->nodeid);
3078 DDSystemInfo systemInfo = getSystemInfo(mdlog, cr, options, mtop, ir, box, xGlobal);
3080 gmx_ddbox_t ddbox = {0};
3081 DDGridSetup ddGridSetup = getDDGridSetup(mdlog, cr, options, ddSettings, systemInfo,
3082 mtop, ir, box, xGlobal, &ddbox);
3084 cr->npmenodes = ddGridSetup.numPmeOnlyRanks;
3086 DDRankSetup ddRankSetup = getDDRankSetup(mdlog, cr, ddGridSetup, *ir);
3088 /* Generate the group communicator, also decides the duty of each rank */
3089 ivec ddCellIndex = { 0, 0, 0 };
3090 std::vector<int> pmeRanks;
3091 CartesianRankSetup cartSetup =
3092 makeGroupCommunicators(mdlog, ddSettings, options.rankOrder,
3094 ddCellIndex, &pmeRanks);
3096 gmx_domdec_t *dd = new gmx_domdec_t(*ir);
3098 copy_ivec(ddCellIndex, dd->ci);
3100 dd->comm = init_dd_comm();
3102 dd->comm->ddRankSetup = ddRankSetup;
3103 dd->comm->cartesianRankSetup = cartSetup;
3105 set_dd_limits(mdlog, cr, dd, options,
3106 ddSettings, systemInfo, ddGridSetup,
3110 setupGroupCommunication(mdlog, ddSettings, pmeRanks, cr, dd);
3112 if (thisRankHasDuty(cr, DUTY_PP))
3114 set_ddgrid_parameters(mdlog, dd, options.dlbScaling, mtop, ir, &ddbox);
3116 setup_neighbor_relations(dd);
3119 /* Set overallocation to avoid frequent reallocation of arrays */
3120 set_over_alloc_dd(TRUE);
3122 dd->atomSets = atomSets;
3127 static gmx_bool test_dd_cutoff(t_commrec *cr,
3129 gmx::ArrayRef<const gmx::RVec> x,
3130 real cutoffRequested)
3140 set_ddbox(*dd, false, box, true, x, &ddbox);
3144 for (d = 0; d < dd->ndim; d++)
3148 inv_cell_size = DD_CELL_MARGIN*dd->nc[dim]/ddbox.box_size[dim];
3149 if (dd->unitCellInfo.ddBoxIsDynamic)
3151 inv_cell_size *= DD_PRES_SCALE_MARGIN;
3154 np = 1 + static_cast<int>(cutoffRequested*inv_cell_size*ddbox.skew_fac[dim]);
3156 if (!isDlbDisabled(dd->comm) && (dim < ddbox.npbcdim) && (dd->comm->cd[d].np_dlb > 0))
3158 if (np > dd->comm->cd[d].np_dlb)
3163 /* If a current local cell size is smaller than the requested
3164 * cut-off, we could still fix it, but this gets very complicated.
3165 * Without fixing here, we might actually need more checks.
3167 real cellSizeAlongDim = (dd->comm->cell_x1[dim] - dd->comm->cell_x0[dim])*ddbox.skew_fac[dim];
3168 if (cellSizeAlongDim*dd->comm->cd[d].np_dlb < cutoffRequested)
3175 if (!isDlbDisabled(dd->comm))
3177 /* If DLB is not active yet, we don't need to check the grid jumps.
3178 * Actually we shouldn't, because then the grid jump data is not set.
3180 if (isDlbOn(dd->comm) &&
3181 check_grid_jump(0, dd, cutoffRequested, &ddbox, FALSE))
3186 gmx_sumi(1, &LocallyLimited, cr);
3188 if (LocallyLimited > 0)
3197 gmx_bool change_dd_cutoff(t_commrec *cr,
3199 gmx::ArrayRef<const gmx::RVec> x,
3200 real cutoffRequested)
3202 gmx_bool bCutoffAllowed;
3204 bCutoffAllowed = test_dd_cutoff(cr, box, x, cutoffRequested);
3208 cr->dd->comm->systemInfo.cutoff = cutoffRequested;
3211 return bCutoffAllowed;