1 /* -*- mode: c; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4; c-file-style: "stroustrup"; -*-
4 * This source code is part of
8 * GROningen MAchine for Chemical Simulations
11 * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
12 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
13 * Copyright (c) 2001-2004, The GROMACS development team,
14 * check out http://www.gromacs.org for more information.
16 * This program is free software; you can redistribute it and/or
17 * modify it under the terms of the GNU General Public License
18 * as published by the Free Software Foundation; either version 2
19 * of the License, or (at your option) any later version.
21 * If you want to redistribute modifications, please consider that
22 * scientific software is very special. Version control is crucial -
23 * bugs must be traceable. We will be happy to consider code for
24 * inclusion in the official distribution, but derived work must not
25 * be called official GROMACS. Details are found in the README & COPYING
26 * files - if they are missing, get the official version at www.gromacs.org.
28 * To help us fund GROMACS development, we humbly ask that you cite
29 * the papers on the package - you can find them in the top README file.
31 * For more info, check our website at http://www.gromacs.org
34 * GROwing Monsters And Cloning Shrimps
47 #include "gmx_fatal.h"
52 #include "sighandler.h"
61 #define PP_PME_CHARGE (1<<0)
62 #define PP_PME_CHARGEB (1<<1)
63 #define PP_PME_COORD (1<<2)
64 #define PP_PME_FEP (1<<3)
65 #define PP_PME_ENER_VIR (1<<4)
66 #define PP_PME_FINISH (1<<5)
67 #define PP_PME_SWITCH (1<<6)
69 #define PME_PP_SIGSTOP (1<<0)
70 #define PME_PP_SIGSTOPNSS (1<<1)
72 typedef struct gmx_pme_pp {
74 MPI_Comm mpi_comm_mysim;
76 int nnode; /* The number of PP node to communicate with */
77 int *node; /* The PP node ranks */
78 int node_peer; /* The peer PP node rank */
79 int *nat; /* The number of atom for each PP node */
80 int flags_charge; /* The flags sent along with the last charges */
92 typedef struct gmx_pme_comm_n_box {
100 ivec grid_size; /* For PME grid tuning */
101 real ewaldcoeff; /* For PME grid tuning */
102 } gmx_pme_comm_n_box_t;
109 gmx_stop_cond_t stop_cond;
110 } gmx_pme_comm_vir_ene_t;
115 gmx_pme_pp_t gmx_pme_pp_init(t_commrec *cr)
117 struct gmx_pme_pp *pme_pp;
123 pme_pp->mpi_comm_mysim = cr->mpi_comm_mysim;
124 MPI_Comm_rank(cr->mpi_comm_mygroup, &rank);
125 get_pme_ddnodes(cr, rank, &pme_pp->nnode, &pme_pp->node, &pme_pp->node_peer);
126 snew(pme_pp->nat, pme_pp->nnode);
127 snew(pme_pp->req, 2*pme_pp->nnode);
128 snew(pme_pp->stat, 2*pme_pp->nnode);
130 pme_pp->flags_charge = 0;
136 /* This should be faster with a real non-blocking MPI implementation */
137 /* #define GMX_PME_DELAYED_WAIT */
139 static void gmx_pme_send_q_x_wait(gmx_domdec_t *dd)
144 MPI_Waitall(dd->nreq_pme, dd->req_pme, MPI_STATUSES_IGNORE);
150 static void gmx_pme_send_q_x(t_commrec *cr, int flags,
151 real *chargeA, real *chargeB,
154 int maxshift_x, int maxshift_y,
155 gmx_large_int_t step)
158 gmx_pme_comm_n_box_t *cnb;
166 fprintf(debug, "PP node %d sending to PME node %d: %d%s%s\n",
167 cr->sim_nodeid, dd->pme_nodeid, n,
168 flags & PP_PME_CHARGE ? " charges" : "",
169 flags & PP_PME_COORD ? " coordinates" : "");
172 #ifdef GMX_PME_DELAYED_WAIT
173 /* When can not use cnb until pending communication has finished */
174 gmx_pme_send_x_q_wait(dd);
177 if (dd->pme_receive_vir_ener)
179 /* Peer PP node: communicate all data */
188 cnb->maxshift_x = maxshift_x;
189 cnb->maxshift_y = maxshift_y;
190 cnb->lambda = lambda;
192 if (flags & PP_PME_COORD)
194 copy_mat(box, cnb->box);
197 MPI_Isend(cnb, sizeof(*cnb), MPI_BYTE,
198 dd->pme_nodeid, 0, cr->mpi_comm_mysim,
199 &dd->req_pme[dd->nreq_pme++]);
202 else if (flags & PP_PME_CHARGE)
205 /* Communicate only the number of atoms */
206 MPI_Isend(&n, sizeof(n), MPI_BYTE,
207 dd->pme_nodeid, 0, cr->mpi_comm_mysim,
208 &dd->req_pme[dd->nreq_pme++]);
215 if (flags & PP_PME_CHARGE)
217 MPI_Isend(chargeA, n*sizeof(real), MPI_BYTE,
218 dd->pme_nodeid, 1, cr->mpi_comm_mysim,
219 &dd->req_pme[dd->nreq_pme++]);
221 if (flags & PP_PME_CHARGEB)
223 MPI_Isend(chargeB, n*sizeof(real), MPI_BYTE,
224 dd->pme_nodeid, 2, cr->mpi_comm_mysim,
225 &dd->req_pme[dd->nreq_pme++]);
227 if (flags & PP_PME_COORD)
229 MPI_Isend(x[0], n*sizeof(rvec), MPI_BYTE,
230 dd->pme_nodeid, 3, cr->mpi_comm_mysim,
231 &dd->req_pme[dd->nreq_pme++]);
235 #ifndef GMX_PME_DELAYED_WAIT
236 /* Wait for the data to arrive */
237 /* We can skip this wait as we are sure x and q will not be modified
238 * before the next call to gmx_pme_send_x_q or gmx_pme_receive_f.
240 gmx_pme_send_q_x_wait(dd);
245 void gmx_pme_send_q(t_commrec *cr,
246 gmx_bool bFreeEnergy, real *chargeA, real *chargeB,
247 int maxshift_x, int maxshift_y)
251 flags = PP_PME_CHARGE;
254 flags |= PP_PME_CHARGEB;
257 gmx_pme_send_q_x(cr, flags,
258 chargeA, chargeB, NULL, NULL, 0, maxshift_x, maxshift_y, -1);
261 void gmx_pme_send_x(t_commrec *cr, matrix box, rvec *x,
262 gmx_bool bFreeEnergy, real lambda,
264 gmx_large_int_t step)
268 flags = PP_PME_COORD;
275 flags |= PP_PME_ENER_VIR;
278 gmx_pme_send_q_x(cr, flags, NULL, NULL, box, x, lambda, 0, 0, step);
281 void gmx_pme_send_finish(t_commrec *cr)
285 flags = PP_PME_FINISH;
287 gmx_pme_send_q_x(cr, flags, NULL, NULL, NULL, NULL, 0, 0, 0, -1);
290 void gmx_pme_send_switch(t_commrec *cr, ivec grid_size, real ewaldcoeff)
293 gmx_pme_comm_n_box_t cnb;
295 if (cr->dd->pme_receive_vir_ener)
297 cnb.flags = PP_PME_SWITCH;
298 copy_ivec(grid_size, cnb.grid_size);
299 cnb.ewaldcoeff = ewaldcoeff;
301 /* We send this, uncommon, message blocking to simplify the code */
302 MPI_Send(&cnb, sizeof(cnb), MPI_BYTE,
303 cr->dd->pme_nodeid, 0, cr->mpi_comm_mysim);
308 int gmx_pme_recv_q_x(struct gmx_pme_pp *pme_pp,
309 real **chargeA, real **chargeB,
310 matrix box, rvec **x, rvec **f,
311 int *maxshift_x, int *maxshift_y,
312 gmx_bool *bFreeEnergy, real *lambda,
314 gmx_large_int_t *step,
315 ivec grid_size, real *ewaldcoeff)
317 gmx_pme_comm_n_box_t cnb;
318 int nat = 0, q, messages, sender;
323 /* avoid compiler warning about unused variable without MPI support */
328 /* Receive the send count, box and time step from the peer PP node */
329 MPI_Recv(&cnb, sizeof(cnb), MPI_BYTE,
330 pme_pp->node_peer, 0,
331 pme_pp->mpi_comm_mysim, MPI_STATUS_IGNORE);
335 fprintf(debug, "PME only node receiving:%s%s%s%s\n",
336 (cnb.flags & PP_PME_CHARGE) ? " charges" : "",
337 (cnb.flags & PP_PME_COORD ) ? " coordinates" : "",
338 (cnb.flags & PP_PME_FINISH) ? " finish" : "",
339 (cnb.flags & PP_PME_SWITCH) ? " switch" : "");
342 if (cnb.flags & PP_PME_SWITCH)
344 /* Special case, receive the new parameters and return */
345 copy_ivec(cnb.grid_size, grid_size);
346 *ewaldcoeff = cnb.ewaldcoeff;
351 if (cnb.flags & PP_PME_CHARGE)
353 /* Receive the send counts from the other PP nodes */
354 for (sender = 0; sender < pme_pp->nnode; sender++)
356 if (pme_pp->node[sender] == pme_pp->node_peer)
358 pme_pp->nat[sender] = cnb.natoms;
362 MPI_Irecv(&(pme_pp->nat[sender]), sizeof(pme_pp->nat[0]),
364 pme_pp->node[sender], 0,
365 pme_pp->mpi_comm_mysim, &pme_pp->req[messages++]);
368 MPI_Waitall(messages, pme_pp->req, pme_pp->stat);
372 for (sender = 0; sender < pme_pp->nnode; sender++)
374 nat += pme_pp->nat[sender];
377 if (nat > pme_pp->nalloc)
379 pme_pp->nalloc = over_alloc_dd(nat);
380 srenew(pme_pp->chargeA, pme_pp->nalloc);
381 if (cnb.flags & PP_PME_CHARGEB)
383 srenew(pme_pp->chargeB, pme_pp->nalloc);
385 srenew(pme_pp->x, pme_pp->nalloc);
386 srenew(pme_pp->f, pme_pp->nalloc);
389 /* maxshift is sent when the charges are sent */
390 *maxshift_x = cnb.maxshift_x;
391 *maxshift_y = cnb.maxshift_y;
393 /* Receive the charges in place */
394 for (q = 0; q < ((cnb.flags & PP_PME_CHARGEB) ? 2 : 1); q++)
398 charge_pp = pme_pp->chargeA;
402 charge_pp = pme_pp->chargeB;
405 for (sender = 0; sender < pme_pp->nnode; sender++)
407 if (pme_pp->nat[sender] > 0)
409 MPI_Irecv(charge_pp+nat,
410 pme_pp->nat[sender]*sizeof(real),
412 pme_pp->node[sender], 1+q,
413 pme_pp->mpi_comm_mysim,
414 &pme_pp->req[messages++]);
415 nat += pme_pp->nat[sender];
418 fprintf(debug, "Received from PP node %d: %d "
420 pme_pp->node[sender], pme_pp->nat[sender]);
426 pme_pp->flags_charge = cnb.flags;
429 if (cnb.flags & PP_PME_COORD)
431 if (!(pme_pp->flags_charge & PP_PME_CHARGE))
433 gmx_incons("PME-only node received coordinates before charges"
437 /* The box, FE flag and lambda are sent along with the coordinates
439 copy_mat(cnb.box, box);
440 *bFreeEnergy = (cnb.flags & PP_PME_FEP);
441 *lambda = cnb.lambda;
442 *bEnerVir = (cnb.flags & PP_PME_ENER_VIR);
444 if (*bFreeEnergy && !(pme_pp->flags_charge & PP_PME_CHARGEB))
446 gmx_incons("PME-only node received free energy request, but "
447 "did not receive B-state charges");
450 /* Receive the coordinates in place */
452 for (sender = 0; sender < pme_pp->nnode; sender++)
454 if (pme_pp->nat[sender] > 0)
456 MPI_Irecv(pme_pp->x[nat], pme_pp->nat[sender]*sizeof(rvec),
458 pme_pp->node[sender], 3,
459 pme_pp->mpi_comm_mysim, &pme_pp->req[messages++]);
460 nat += pme_pp->nat[sender];
463 fprintf(debug, "Received from PP node %d: %d "
465 pme_pp->node[sender], pme_pp->nat[sender]);
471 /* Wait for the coordinates and/or charges to arrive */
472 MPI_Waitall(messages, pme_pp->req, pme_pp->stat);
475 while (!(cnb.flags & (PP_PME_COORD | PP_PME_FINISH)));
480 *chargeA = pme_pp->chargeA;
481 *chargeB = pme_pp->chargeB;
486 return ((cnb.flags & PP_PME_FINISH) ? -1 : nat);
489 static void receive_virial_energy(t_commrec *cr,
490 matrix vir, real *energy, real *dvdlambda,
493 gmx_pme_comm_vir_ene_t cve;
495 if (cr->dd->pme_receive_vir_ener)
500 "PP node %d receiving from PME node %d: virial and energy\n",
501 cr->sim_nodeid, cr->dd->pme_nodeid);
504 MPI_Recv(&cve, sizeof(cve), MPI_BYTE, cr->dd->pme_nodeid, 1, cr->mpi_comm_mysim,
507 memset(&cve, 0, sizeof(cve));
510 m_add(vir, cve.vir, vir);
511 *energy = cve.energy;
512 *dvdlambda += cve.dvdlambda;
513 *pme_cycles = cve.cycles;
515 if (cve.stop_cond != gmx_stop_cond_none)
517 gmx_set_stop_condition(cve.stop_cond);
527 void gmx_pme_receive_f(t_commrec *cr,
528 rvec f[], matrix vir,
529 real *energy, real *dvdlambda,
534 #ifdef GMX_PME_DELAYED_WAIT
535 /* Wait for the x request to finish */
536 gmx_pme_send_q_x_wait(cr->dd);
539 natoms = cr->dd->nat_home;
541 if (natoms > cr->dd->pme_recv_f_alloc)
543 cr->dd->pme_recv_f_alloc = over_alloc_dd(natoms);
544 srenew(cr->dd->pme_recv_f_buf, cr->dd->pme_recv_f_alloc);
548 MPI_Recv(cr->dd->pme_recv_f_buf[0],
549 natoms*sizeof(rvec), MPI_BYTE,
550 cr->dd->pme_nodeid, 0, cr->mpi_comm_mysim,
554 for (i = 0; i < natoms; i++)
556 rvec_inc(f[i], cr->dd->pme_recv_f_buf[i]);
560 receive_virial_energy(cr, vir, energy, dvdlambda, pme_cycles);
563 void gmx_pme_send_force_vir_ener(struct gmx_pme_pp *pme_pp,
565 real energy, real dvdlambda,
568 gmx_pme_comm_vir_ene_t cve;
569 int messages, ind_start, ind_end, receiver;
573 /* Now the evaluated forces have to be transferred to the PP nodes */
576 for (receiver = 0; receiver < pme_pp->nnode; receiver++)
579 ind_end = ind_start + pme_pp->nat[receiver];
581 if (MPI_Isend(f[ind_start], (ind_end-ind_start)*sizeof(rvec), MPI_BYTE,
582 pme_pp->node[receiver], 0,
583 pme_pp->mpi_comm_mysim, &pme_pp->req[messages++]) != 0)
585 gmx_comm("MPI_Isend failed in do_pmeonly");
590 /* send virial and energy to our last PP node */
591 copy_mat(vir, cve.vir);
593 cve.dvdlambda = dvdlambda;
594 /* check for the signals to send back to a PP node */
595 cve.stop_cond = gmx_get_stop_condition();
601 fprintf(debug, "PME node sending to PP node %d: virial and energy\n",
605 MPI_Isend(&cve, sizeof(cve), MPI_BYTE,
606 pme_pp->node_peer, 1,
607 pme_pp->mpi_comm_mysim, &pme_pp->req[messages++]);
609 /* Wait for the forces to arrive */
610 MPI_Waitall(messages, pme_pp->req, pme_pp->stat);