3 * This source code is part of
7 * GROningen MAchine for Chemical Simulations
10 * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
11 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
12 * Copyright (c) 2001-2004, The GROMACS development team,
13 * check out http://www.gromacs.org for more information.
15 * This program is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public License
17 * as published by the Free Software Foundation; either version 2
18 * of the License, or (at your option) any later version.
20 * If you want to redistribute modifications, please consider that
21 * scientific software is very special. Version control is crucial -
22 * bugs must be traceable. We will be happy to consider code for
23 * inclusion in the official distribution, but derived work must not
24 * be called official GROMACS. Details are found in the README & COPYING
25 * files - if they are missing, get the official version at www.gromacs.org.
27 * To help us fund GROMACS development, we humbly ask that you cite
28 * the papers on the package - you can find them in the top README file.
30 * For more info, check our website at http://www.gromacs.org
33 * GROningen Mixture of Alchemy and Childrens' Stories
40 #include "gmx_fatal.h"
57 #include "mpelogging.h"
59 /* The source code in this file should be thread-safe.
60 Please keep it that way. */
62 gmx_bool gmx_mpi_initialized(void)
74 int gmx_setup(int *argc,char **argv,int *nnodes)
77 gmx_call("gmx_setup");
81 int resultlen; /* actual length of node name */
85 char mpi_hostname[MPI_MAX_PROCESSOR_NAME];
87 /* Call the MPI routines */
90 (void) fah_MPI_Init(argc,&argv);
92 (void) MPI_Init(argc,&argv);
95 (void) MPI_Comm_size( MPI_COMM_WORLD, &mpi_num_nodes );
96 (void) MPI_Comm_rank( MPI_COMM_WORLD, &mpi_my_rank );
97 (void) MPI_Get_processor_name( mpi_hostname, &resultlen );
101 /* MPE logging routines. Get event IDs from MPE: */
103 ev_timestep1 = MPE_Log_get_event_number( );
104 ev_timestep2 = MPE_Log_get_event_number( );
105 ev_force_start = MPE_Log_get_event_number( );
106 ev_force_finish = MPE_Log_get_event_number( );
107 ev_do_fnbf_start = MPE_Log_get_event_number( );
108 ev_do_fnbf_finish = MPE_Log_get_event_number( );
109 ev_ns_start = MPE_Log_get_event_number( );
110 ev_ns_finish = MPE_Log_get_event_number( );
111 ev_calc_bonds_start = MPE_Log_get_event_number( );
112 ev_calc_bonds_finish = MPE_Log_get_event_number( );
113 ev_global_stat_start = MPE_Log_get_event_number( );
114 ev_global_stat_finish = MPE_Log_get_event_number( );
115 ev_virial_start = MPE_Log_get_event_number( );
116 ev_virial_finish = MPE_Log_get_event_number( );
118 /* Enforced rotation */
119 ev_flexll_start = MPE_Log_get_event_number( );
120 ev_flexll_finish = MPE_Log_get_event_number( );
121 ev_add_rot_forces_start = MPE_Log_get_event_number( );
122 ev_add_rot_forces_finish = MPE_Log_get_event_number( );
123 ev_rotcycles_start = MPE_Log_get_event_number( );
124 ev_rotcycles_finish = MPE_Log_get_event_number( );
125 ev_forcecycles_start = MPE_Log_get_event_number( );
126 ev_forcecycles_finish = MPE_Log_get_event_number( );
128 /* Shift related events */
129 ev_shift_start = MPE_Log_get_event_number( );
130 ev_shift_finish = MPE_Log_get_event_number( );
131 ev_unshift_start = MPE_Log_get_event_number( );
132 ev_unshift_finish = MPE_Log_get_event_number( );
133 ev_mk_mshift_start = MPE_Log_get_event_number( );
134 ev_mk_mshift_finish = MPE_Log_get_event_number( );
136 /* PME related events */
137 ev_pme_start = MPE_Log_get_event_number( );
138 ev_pme_finish = MPE_Log_get_event_number( );
139 ev_spread_on_grid_start = MPE_Log_get_event_number( );
140 ev_spread_on_grid_finish = MPE_Log_get_event_number( );
141 ev_sum_qgrid_start = MPE_Log_get_event_number( );
142 ev_sum_qgrid_finish = MPE_Log_get_event_number( );
143 ev_gmxfft3d_start = MPE_Log_get_event_number( );
144 ev_gmxfft3d_finish = MPE_Log_get_event_number( );
145 ev_solve_pme_start = MPE_Log_get_event_number( );
146 ev_solve_pme_finish = MPE_Log_get_event_number( );
147 ev_gather_f_bsplines_start = MPE_Log_get_event_number( );
148 ev_gather_f_bsplines_finish= MPE_Log_get_event_number( );
149 ev_reduce_start = MPE_Log_get_event_number( );
150 ev_reduce_finish = MPE_Log_get_event_number( );
151 ev_rscatter_start = MPE_Log_get_event_number( );
152 ev_rscatter_finish = MPE_Log_get_event_number( );
153 ev_alltoall_start = MPE_Log_get_event_number( );
154 ev_alltoall_finish = MPE_Log_get_event_number( );
155 ev_pmeredist_start = MPE_Log_get_event_number( );
156 ev_pmeredist_finish = MPE_Log_get_event_number( );
157 ev_init_pme_start = MPE_Log_get_event_number( );
158 ev_init_pme_finish = MPE_Log_get_event_number( );
159 ev_send_coordinates_start = MPE_Log_get_event_number( );
160 ev_send_coordinates_finish = MPE_Log_get_event_number( );
161 ev_update_fr_start = MPE_Log_get_event_number( );
162 ev_update_fr_finish = MPE_Log_get_event_number( );
163 ev_clear_rvecs_start = MPE_Log_get_event_number( );
164 ev_clear_rvecs_finish = MPE_Log_get_event_number( );
165 ev_update_start = MPE_Log_get_event_number( );
166 ev_update_finish = MPE_Log_get_event_number( );
167 ev_output_start = MPE_Log_get_event_number( );
168 ev_output_finish = MPE_Log_get_event_number( );
169 ev_sum_lrforces_start = MPE_Log_get_event_number( );
170 ev_sum_lrforces_finish = MPE_Log_get_event_number( );
171 ev_sort_start = MPE_Log_get_event_number( );
172 ev_sort_finish = MPE_Log_get_event_number( );
173 ev_sum_qgrid_start = MPE_Log_get_event_number( );
174 ev_sum_qgrid_finish = MPE_Log_get_event_number( );
176 /* Essential dynamics related events */
177 ev_edsam_start = MPE_Log_get_event_number( );
178 ev_edsam_finish = MPE_Log_get_event_number( );
179 ev_get_coords_start = MPE_Log_get_event_number( );
180 ev_get_coords_finish = MPE_Log_get_event_number( );
181 ev_ed_apply_cons_start = MPE_Log_get_event_number( );
182 ev_ed_apply_cons_finish = MPE_Log_get_event_number( );
183 ev_fit_to_reference_start = MPE_Log_get_event_number( );
184 ev_fit_to_reference_finish = MPE_Log_get_event_number( );
186 /* describe events: */
187 if ( mpi_my_rank == 0 )
190 MPE_Describe_state(ev_timestep1, ev_timestep2, "timestep START", "magenta" );
191 MPE_Describe_state(ev_force_start, ev_force_finish, "force", "cornflower blue" );
192 MPE_Describe_state(ev_do_fnbf_start, ev_do_fnbf_finish, "do_fnbf", "navy" );
193 MPE_Describe_state(ev_ns_start, ev_ns_finish, "neighbor search", "tomato" );
194 MPE_Describe_state(ev_calc_bonds_start, ev_calc_bonds_finish, "bonded forces", "slate blue" );
195 MPE_Describe_state(ev_global_stat_start, ev_global_stat_finish, "global stat", "firebrick3");
196 MPE_Describe_state(ev_update_fr_start, ev_update_fr_finish, "update forcerec", "goldenrod");
197 MPE_Describe_state(ev_clear_rvecs_start, ev_clear_rvecs_finish, "clear rvecs", "bisque");
198 MPE_Describe_state(ev_update_start, ev_update_finish, "update", "cornsilk");
199 MPE_Describe_state(ev_output_start, ev_output_finish, "output", "black");
200 MPE_Describe_state(ev_virial_start, ev_virial_finish, "calc_virial", "thistle4");
202 /* Enforced rotation */
203 MPE_Describe_state(ev_flexll_start, ev_flexll_finish, "flex lowlevel", "navajo white");
204 MPE_Describe_state(ev_add_rot_forces_start, ev_add_rot_forces_finish, "add rot forces", "green");
205 MPE_Describe_state(ev_rotcycles_start, ev_rotcycles_finish, "count rot cyc", "moccasin");
206 MPE_Describe_state(ev_forcecycles_start, ev_forcecycles_finish, "count force cyc", "powder blue");
208 /* PME related events */
209 MPE_Describe_state(ev_pme_start, ev_pme_finish, "doing PME", "grey" );
210 MPE_Describe_state(ev_spread_on_grid_start, ev_spread_on_grid_finish, "spread", "dark orange" );
211 MPE_Describe_state(ev_sum_qgrid_start, ev_sum_qgrid_finish, "sum qgrid", "slate blue");
212 MPE_Describe_state(ev_gmxfft3d_start, ev_gmxfft3d_finish, "fft3d", "snow2" );
213 MPE_Describe_state(ev_solve_pme_start, ev_solve_pme_finish, "solve PME", "indian red" );
214 MPE_Describe_state(ev_gather_f_bsplines_start, ev_gather_f_bsplines_finish, "bsplines", "light sea green" );
215 MPE_Describe_state(ev_reduce_start, ev_reduce_finish, "reduce", "cyan1" );
216 MPE_Describe_state(ev_rscatter_start, ev_rscatter_finish, "rscatter", "cyan3" );
217 MPE_Describe_state(ev_alltoall_start, ev_alltoall_finish, "alltoall", "LightCyan4" );
218 MPE_Describe_state(ev_pmeredist_start, ev_pmeredist_finish, "pmeredist", "thistle" );
219 MPE_Describe_state(ev_init_pme_start, ev_init_pme_finish, "init PME", "snow4");
220 MPE_Describe_state(ev_send_coordinates_start, ev_send_coordinates_finish, "send_coordinates","blue");
221 MPE_Describe_state(ev_sum_lrforces_start, ev_sum_lrforces_finish, "sum_LRforces", "lime green");
222 MPE_Describe_state(ev_sort_start, ev_sort_finish, "sort pme atoms", "brown");
223 MPE_Describe_state(ev_sum_qgrid_start, ev_sum_qgrid_finish, "sum charge grid", "medium orchid");
225 /* Shift related events */
226 MPE_Describe_state(ev_shift_start, ev_shift_finish, "shift", "orange");
227 MPE_Describe_state(ev_unshift_start, ev_unshift_finish, "unshift", "dark orange");
228 MPE_Describe_state(ev_mk_mshift_start, ev_mk_mshift_finish, "mk_mshift", "maroon");
230 /* Essential dynamics related events */
231 MPE_Describe_state(ev_edsam_start, ev_edsam_finish, "EDSAM", "deep sky blue");
232 MPE_Describe_state(ev_get_coords_start, ev_get_coords_finish, "ED get coords", "steel blue");
233 MPE_Describe_state(ev_ed_apply_cons_start, ev_ed_apply_cons_finish, "ED apply constr", "forest green");
234 MPE_Describe_state(ev_fit_to_reference_start, ev_fit_to_reference_finish, "ED fit to ref", "lavender");
241 fprintf(stderr,"NNODES=%d, MYRANK=%d, HOSTNAME=%s\n",
242 mpi_num_nodes,mpi_my_rank,mpi_hostname);
245 *nnodes=mpi_num_nodes;
251 int gmx_node_num(void)
257 (void) MPI_Comm_size(MPI_COMM_WORLD, &i);
262 int gmx_node_rank(void)
268 (void) MPI_Comm_rank(MPI_COMM_WORLD, &i);
273 void gmx_setup_nodecomm(FILE *fplog,t_commrec *cr)
276 int n,rank,resultlen,hostnum,i,j,ng,ni;
278 char mpi_hostname[MPI_MAX_PROCESSOR_NAME],num[MPI_MAX_PROCESSOR_NAME];
281 /* Many MPI implementations do not optimize MPI_Allreduce
282 * (and probably also other global communication calls)
283 * for multi-core nodes connected by a network.
284 * We can optimize such communication by using one MPI call
285 * within each node and one between the nodes.
286 * For MVAPICH2 and Intel MPI this reduces the time for
287 * the global_stat communication by 25%
288 * for 2x2-core 3 GHz Woodcrest connected by mixed DDR/SDR Infiniband.
289 * B. Hess, November 2007
296 if (getenv("GMX_NO_NODECOMM") == NULL) {
298 MPI_Comm_size(cr->mpi_comm_mygroup,&n);
299 MPI_Comm_rank(cr->mpi_comm_mygroup,&rank);
300 MPI_Get_processor_name(mpi_hostname,&resultlen);
301 /* This procedure can only differentiate nodes with host names
302 * that end on unique numbers.
306 /* Only parse the host name up to the first dot */
307 while(i < resultlen && mpi_hostname[i] != '.') {
308 if (isdigit(mpi_hostname[i])) {
309 num[j++] = mpi_hostname[i];
317 /* Use only the last 9 decimals, so we don't overflow an int */
318 hostnum = strtol(num + max(0,j-9), NULL, 10);
323 "In gmx_setup_nodecomm: splitting communicator of size %d\n",
325 fprintf(debug,"In gmx_setup_nodecomm: hostname '%s', hostnum %d\n",
326 mpi_hostname,hostnum);
329 /* The intra-node communicator, split on node number */
330 MPI_Comm_split(cr->mpi_comm_mygroup,hostnum,rank,&nc->comm_intra);
331 MPI_Comm_rank(nc->comm_intra,&nc->rank_intra);
333 fprintf(debug,"In gmx_setup_nodecomm: node rank %d rank_intra %d\n",
334 rank,nc->rank_intra);
336 /* The inter-node communicator, split on rank_intra.
337 * We actually only need the one for rank=0,
338 * but it is easier to create them all.
340 MPI_Comm_split(cr->mpi_comm_mygroup,nc->rank_intra,rank,&nc->comm_inter);
341 /* Check if this really created two step communication */
342 MPI_Comm_size(nc->comm_inter,&ng);
343 MPI_Comm_size(nc->comm_intra,&ni);
345 fprintf(debug,"In gmx_setup_nodecomm: groups %d, my group size %d\n",
348 if ((ng > 1 && ng < n) || (ni > 1 && ni < n)) {
351 fprintf(fplog,"Using two step summing over %d groups of on average %.1f processes\n\n",ng,(real)n/(real)ng);
352 if (nc->rank_intra > 0)
353 MPI_Comm_free(&nc->comm_inter);
355 /* One group or all processes in a separate group, use normal summing */
356 MPI_Comm_free(&nc->comm_inter);
357 MPI_Comm_free(&nc->comm_intra);
364 void gmx_barrier(const t_commrec *cr)
367 gmx_call("gmx_barrier");
369 MPI_Barrier(cr->mpi_comm_mygroup);
373 void gmx_abort(int noderank,int nnodes,int errorno)
376 gmx_call("gmx_abort");
379 fprintf(stderr,"Halting program %s\n",ShortProgram());
385 fprintf(stderr,"Halting parallel program %s on CPU %d out of %d\n",
386 ShortProgram(),noderank,nnodes);
390 fprintf(stderr,"Halting program %s\n",ShortProgram());
394 MPI_Abort(MPI_COMM_WORLD,errorno);
400 void gmx_bcast(int nbytes,void *b,const t_commrec *cr)
403 gmx_call("gmx_bast");
405 MPI_Bcast(b,nbytes,MPI_BYTE,MASTERRANK(cr),cr->mpi_comm_mygroup);
409 void gmx_bcast_sim(int nbytes,void *b,const t_commrec *cr)
412 gmx_call("gmx_bast");
414 MPI_Bcast(b,nbytes,MPI_BYTE,MASTERRANK(cr),cr->mpi_comm_mysim);
418 void gmx_sumd(int nr,double r[],const t_commrec *cr)
421 gmx_call("gmx_sumd");
423 #if defined(MPI_IN_PLACE_EXISTS) || defined(GMX_THREADS)
425 if (cr->nc.rank_intra == 0)
427 /* Use two step summing. */
428 MPI_Reduce(MPI_IN_PLACE,r,nr,MPI_DOUBLE,MPI_SUM,0,
430 /* Sum the roots of the internal (intra) buffers. */
431 MPI_Allreduce(MPI_IN_PLACE,r,nr,MPI_DOUBLE,MPI_SUM,
436 /* This is here because of the silly MPI specification
437 that MPI_IN_PLACE should be put in sendbuf instead of recvbuf */
438 MPI_Reduce(r,NULL,nr,MPI_DOUBLE,MPI_SUM,0,cr->nc.comm_intra);
440 MPI_Bcast(r,nr,MPI_DOUBLE,0,cr->nc.comm_intra);
444 MPI_Allreduce(MPI_IN_PLACE,r,nr,MPI_DOUBLE,MPI_SUM,
445 cr->mpi_comm_mygroup);
450 if (nr > cr->mpb->dbuf_alloc) {
451 cr->mpb->dbuf_alloc = nr;
452 srenew(cr->mpb->dbuf,cr->mpb->dbuf_alloc);
455 /* Use two step summing */
456 MPI_Allreduce(r,cr->mpb->dbuf,nr,MPI_DOUBLE,MPI_SUM,cr->nc.comm_intra);
457 if (cr->nc.rank_intra == 0) {
458 /* Sum with the buffers reversed */
459 MPI_Allreduce(cr->mpb->dbuf,r,nr,MPI_DOUBLE,MPI_SUM,
462 MPI_Bcast(r,nr,MPI_DOUBLE,0,cr->nc.comm_intra);
464 MPI_Allreduce(r,cr->mpb->dbuf,nr,MPI_DOUBLE,MPI_SUM,
465 cr->mpi_comm_mygroup);
467 r[i] = cr->mpb->dbuf[i];
473 void gmx_sumf(int nr,float r[],const t_commrec *cr)
476 gmx_call("gmx_sumf");
478 #if defined(MPI_IN_PLACE_EXISTS) || defined(GMX_THREADS)
480 /* Use two step summing. */
481 if (cr->nc.rank_intra == 0)
483 MPI_Reduce(MPI_IN_PLACE,r,nr,MPI_FLOAT,MPI_SUM,0,
485 /* Sum the roots of the internal (intra) buffers */
486 MPI_Allreduce(MPI_IN_PLACE,r,nr,MPI_FLOAT,MPI_SUM,
491 /* This is here because of the silly MPI specification
492 that MPI_IN_PLACE should be put in sendbuf instead of recvbuf */
493 MPI_Reduce(r,NULL,nr,MPI_FLOAT,MPI_SUM,0,cr->nc.comm_intra);
495 MPI_Bcast(r,nr,MPI_FLOAT,0,cr->nc.comm_intra);
499 MPI_Allreduce(MPI_IN_PLACE,r,nr,MPI_FLOAT,MPI_SUM,cr->mpi_comm_mygroup);
504 if (nr > cr->mpb->fbuf_alloc) {
505 cr->mpb->fbuf_alloc = nr;
506 srenew(cr->mpb->fbuf,cr->mpb->fbuf_alloc);
509 /* Use two step summing */
510 MPI_Allreduce(r,cr->mpb->fbuf,nr,MPI_FLOAT,MPI_SUM,cr->nc.comm_intra);
511 if (cr->nc.rank_intra == 0) {
512 /* Sum with the buffers reversed */
513 MPI_Allreduce(cr->mpb->fbuf,r,nr,MPI_FLOAT,MPI_SUM,
516 MPI_Bcast(r,nr,MPI_FLOAT,0,cr->nc.comm_intra);
518 MPI_Allreduce(r,cr->mpb->fbuf,nr,MPI_FLOAT,MPI_SUM,
519 cr->mpi_comm_mygroup);
521 r[i] = cr->mpb->fbuf[i];
527 void gmx_sumi(int nr,int r[],const t_commrec *cr)
530 gmx_call("gmx_sumi");
532 #if defined(MPI_IN_PLACE_EXISTS) || defined(GMX_THREADS)
534 /* Use two step summing */
535 if (cr->nc.rank_intra == 0)
537 MPI_Reduce(MPI_IN_PLACE,r,nr,MPI_INT,MPI_SUM,0,cr->nc.comm_intra);
538 /* Sum with the buffers reversed */
539 MPI_Allreduce(MPI_IN_PLACE,r,nr,MPI_INT,MPI_SUM,cr->nc.comm_inter);
543 /* This is here because of the silly MPI specification
544 that MPI_IN_PLACE should be put in sendbuf instead of recvbuf */
545 MPI_Reduce(r,NULL,nr,MPI_INT,MPI_SUM,0,cr->nc.comm_intra);
547 MPI_Bcast(r,nr,MPI_INT,0,cr->nc.comm_intra);
551 MPI_Allreduce(MPI_IN_PLACE,r,nr,MPI_INT,MPI_SUM,cr->mpi_comm_mygroup);
556 if (nr > cr->mpb->ibuf_alloc) {
557 cr->mpb->ibuf_alloc = nr;
558 srenew(cr->mpb->ibuf,cr->mpb->ibuf_alloc);
561 /* Use two step summing */
562 MPI_Allreduce(r,cr->mpb->ibuf,nr,MPI_INT,MPI_SUM,cr->nc.comm_intra);
563 if (cr->nc.rank_intra == 0) {
564 /* Sum with the buffers reversed */
565 MPI_Allreduce(cr->mpb->ibuf,r,nr,MPI_INT,MPI_SUM,cr->nc.comm_inter);
567 MPI_Bcast(r,nr,MPI_INT,0,cr->nc.comm_intra);
569 MPI_Allreduce(r,cr->mpb->ibuf,nr,MPI_INT,MPI_SUM,cr->mpi_comm_mygroup);
571 r[i] = cr->mpb->ibuf[i];
577 void gmx_sumli(int nr,gmx_large_int_t r[],const t_commrec *cr)
580 gmx_call("gmx_sumli");
582 #if defined(MPI_IN_PLACE_EXISTS) || defined(GMX_THREADS)
584 /* Use two step summing */
585 if (cr->nc.rank_intra == 0)
587 MPI_Reduce(MPI_IN_PLACE,r,nr,GMX_MPI_LARGE_INT,MPI_SUM,0,
589 /* Sum with the buffers reversed */
590 MPI_Allreduce(MPI_IN_PLACE,r,nr,GMX_MPI_LARGE_INT,MPI_SUM,
595 /* This is here because of the silly MPI specification
596 that MPI_IN_PLACE should be put in sendbuf instead of recvbuf */
597 MPI_Reduce(r,NULL,nr,GMX_MPI_LARGE_INT,MPI_SUM,0,cr->nc.comm_intra);
599 MPI_Bcast(r,nr,GMX_MPI_LARGE_INT,0,cr->nc.comm_intra);
603 MPI_Allreduce(MPI_IN_PLACE,r,nr,GMX_MPI_LARGE_INT,MPI_SUM,cr->mpi_comm_mygroup);
608 if (nr > cr->mpb->ibuf_alloc) {
609 cr->mpb->ibuf_alloc = nr;
610 srenew(cr->mpb->ibuf,cr->mpb->ibuf_alloc);
613 /* Use two step summing */
614 MPI_Allreduce(r,cr->mpb->ibuf,nr,GMX_MPI_LARGE_INT,MPI_SUM,
616 if (cr->nc.rank_intra == 0) {
617 /* Sum with the buffers reversed */
618 MPI_Allreduce(cr->mpb->ibuf,r,nr,GMX_MPI_LARGE_INT,MPI_SUM,
621 MPI_Bcast(r,nr,GMX_MPI_LARGE_INT,0,cr->nc.comm_intra);
623 MPI_Allreduce(r,cr->mpb->ibuf,nr,GMX_MPI_LARGE_INT,MPI_SUM,
624 cr->mpi_comm_mygroup);
626 r[i] = cr->mpb->ibuf[i];
635 void gmx_sumd_comm(int nr,double r[],MPI_Comm mpi_comm)
637 #if defined(MPI_IN_PLACE_EXISTS) || defined(GMX_THREADS)
638 MPI_Allreduce(MPI_IN_PLACE,r,nr,MPI_DOUBLE,MPI_SUM,mpi_comm);
640 /* this function is only used in code that is not performance critical,
641 (during setup, when comm_rec is not the appropriate communication
642 structure), so this isn't as bad as it looks. */
647 MPI_Allreduce(r,buf,nr,MPI_DOUBLE,MPI_SUM,mpi_comm);
656 void gmx_sumf_comm(int nr,float r[],MPI_Comm mpi_comm)
658 #if defined(MPI_IN_PLACE_EXISTS) || defined(GMX_THREADS)
659 MPI_Allreduce(MPI_IN_PLACE,r,nr,MPI_FLOAT,MPI_SUM,mpi_comm);
661 /* this function is only used in code that is not performance critical,
662 (during setup, when comm_rec is not the appropriate communication
663 structure), so this isn't as bad as it looks. */
668 MPI_Allreduce(r,buf,nr,MPI_FLOAT,MPI_SUM,mpi_comm);
676 void gmx_sumd_sim(int nr,double r[],const gmx_multisim_t *ms)
679 gmx_call("gmx_sumd_sim");
681 gmx_sumd_comm(nr,r,ms->mpi_comm_masters);
685 void gmx_sumf_sim(int nr,float r[],const gmx_multisim_t *ms)
688 gmx_call("gmx_sumf_sim");
690 gmx_sumf_comm(nr,r,ms->mpi_comm_masters);
694 void gmx_sumi_sim(int nr,int r[], const gmx_multisim_t *ms)
697 gmx_call("gmx_sumi_sim");
699 #if defined(MPI_IN_PLACE_EXISTS) || defined(GMX_THREADS)
700 MPI_Allreduce(MPI_IN_PLACE,r,nr,MPI_INT,MPI_SUM,ms->mpi_comm_masters);
702 /* this is thread-unsafe, but it will do for now: */
705 if (nr > ms->mpb->ibuf_alloc) {
706 ms->mpb->ibuf_alloc = nr;
707 srenew(ms->mpb->ibuf,ms->mpb->ibuf_alloc);
709 MPI_Allreduce(r,ms->mpb->ibuf,nr,MPI_INT,MPI_SUM,ms->mpi_comm_masters);
711 r[i] = ms->mpb->ibuf[i];
716 void gmx_sumli_sim(int nr,gmx_large_int_t r[], const gmx_multisim_t *ms)
719 gmx_call("gmx_sumli_sim");
721 #if defined(MPI_IN_PLACE_EXISTS) || defined(GMX_THREADS)
722 MPI_Allreduce(MPI_IN_PLACE,r,nr,GMX_MPI_LARGE_INT,MPI_SUM,
723 ms->mpi_comm_masters);
725 /* this is thread-unsafe, but it will do for now: */
728 if (nr > ms->mpb->ibuf_alloc) {
729 ms->mpb->ibuf_alloc = nr;
730 srenew(ms->mpb->ibuf,ms->mpb->ibuf_alloc);
732 MPI_Allreduce(r,ms->mpb->ibuf,nr,GMX_MPI_LARGE_INT,MPI_SUM,
733 ms->mpi_comm_masters);
735 r[i] = ms->mpb->ibuf[i];
741 void gmx_finalize(void)
744 gmx_call("gmx_finalize");
748 /* just as a check; we don't want to finalize twice */
750 MPI_Finalized(&finalized);
754 /* We sync the processes here to try to avoid problems
755 * with buggy MPI implementations that could cause
756 * unfinished processes to terminate.
758 MPI_Barrier(MPI_COMM_WORLD);
761 if (DOMAINDECOMP(cr)) {
762 if (cr->npmenodes > 0 || cr->dd->bCartesian)
763 MPI_Comm_free(&cr->mpi_comm_mygroup);
764 if (cr->dd->bCartesian)
765 MPI_Comm_free(&cr->mpi_comm_mysim);
769 /* Apparently certain mpich implementations cause problems
770 * with MPI_Finalize. In that case comment out MPI_Finalize.
773 fprintf(debug,"Will call MPI_Finalize now\n");
775 ret = MPI_Finalize();
777 fprintf(debug,"Return code from MPI_Finalize = %d\n",ret);