2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2004, The GROMACS development team.
6 * Copyright (c) 2011,2012,2013,2014,2015, by the GROMACS development team, led by
7 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
8 * and including many others, as listed in the AUTHORS file in the
9 * top-level source directory and at http://www.gromacs.org.
11 * GROMACS is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU Lesser General Public License
13 * as published by the Free Software Foundation; either version 2.1
14 * of the License, or (at your option) any later version.
16 * GROMACS is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * Lesser General Public License for more details.
21 * You should have received a copy of the GNU Lesser General Public
22 * License along with GROMACS; if not, see
23 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
24 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
26 * If you want to redistribute modifications to GROMACS, please
27 * consider that scientific software is very special. Version
28 * control is crucial - bugs must be traceable. We will be happy to
29 * consider code for inclusion in the official distribution, but
30 * derived work must not be called official GROMACS. Details are found
31 * in the README & COPYING files - if they are missing, get the
32 * official version at http://www.gromacs.org.
34 * To help us fund GROMACS development, we humbly ask that you cite
35 * the research papers on the package. Check out http://www.gromacs.org.
49 #include "gromacs/domdec/domdec.h"
50 #include "gromacs/essentialdynamics/edsam.h"
51 #include "gromacs/ewald/pme.h"
52 #include "gromacs/fileio/tpxio.h"
53 #include "gromacs/gmxlib/gpu_utils/gpu_utils.h"
54 #include "gromacs/legacyheaders/checkpoint.h"
55 #include "gromacs/legacyheaders/constr.h"
56 #include "gromacs/legacyheaders/copyrite.h"
57 #include "gromacs/legacyheaders/disre.h"
58 #include "gromacs/legacyheaders/force.h"
59 #include "gromacs/legacyheaders/gmx_detect_hardware.h"
60 #include "gromacs/legacyheaders/gmx_omp_nthreads.h"
61 #include "gromacs/legacyheaders/gmx_thread_affinity.h"
62 #include "gromacs/legacyheaders/inputrec.h"
63 #include "gromacs/legacyheaders/main.h"
64 #include "gromacs/legacyheaders/md_logging.h"
65 #include "gromacs/legacyheaders/md_support.h"
66 #include "gromacs/legacyheaders/mdatoms.h"
67 #include "gromacs/legacyheaders/mdrun.h"
68 #include "gromacs/legacyheaders/names.h"
69 #include "gromacs/legacyheaders/network.h"
70 #include "gromacs/legacyheaders/oenv.h"
71 #include "gromacs/legacyheaders/orires.h"
72 #include "gromacs/legacyheaders/qmmm.h"
73 #include "gromacs/legacyheaders/sighandler.h"
74 #include "gromacs/legacyheaders/txtdump.h"
75 #include "gromacs/legacyheaders/typedefs.h"
76 #include "gromacs/math/calculate-ewald-splitting-coefficient.h"
77 #include "gromacs/math/vec.h"
78 #include "gromacs/mdlib/calc_verletbuf.h"
79 #include "gromacs/mdlib/nbnxn_search.h"
80 #include "gromacs/pbcutil/pbc.h"
81 #include "gromacs/pulling/pull.h"
82 #include "gromacs/pulling/pull_rotation.h"
83 #include "gromacs/swap/swapcoords.h"
84 #include "gromacs/timing/wallcycle.h"
85 #include "gromacs/topology/mtop_util.h"
86 #include "gromacs/utility/cstringutil.h"
87 #include "gromacs/utility/gmxassert.h"
88 #include "gromacs/utility/gmxmpi.h"
89 #include "gromacs/utility/smalloc.h"
94 #include "resource-division.h"
101 gmx_integrator_t *func;
104 /* The array should match the eI array in include/types/enums.h */
105 const gmx_intp_t integrator[eiNR] = { {do_md}, {do_steep}, {do_cg}, {do_md}, {do_md}, {do_nm}, {do_lbfgs}, {do_tpi}, {do_tpi}, {do_md}, {do_md}, {do_md}};
107 gmx_int64_t deform_init_init_step_tpx;
108 matrix deform_init_box_tpx;
109 tMPI_Thread_mutex_t deform_init_box_mutex = TMPI_THREAD_MUTEX_INITIALIZER;
112 #ifdef GMX_THREAD_MPI
113 /* The minimum number of atoms per tMPI thread. With fewer atoms than this,
114 * the number of threads will get lowered.
116 #define MIN_ATOMS_PER_MPI_THREAD 90
117 #define MIN_ATOMS_PER_GPU 900
119 struct mdrunner_arglist
134 const char *dddlb_opt;
139 const char *nbpu_opt;
141 gmx_int64_t nsteps_cmdline;
156 /* The function used for spawning threads. Extracts the mdrunner()
157 arguments from its one argument and calls mdrunner(), after making
159 static void mdrunner_start_fn(void *arg)
161 struct mdrunner_arglist *mda = (struct mdrunner_arglist*)arg;
162 struct mdrunner_arglist mc = *mda; /* copy the arg list to make sure
163 that it's thread-local. This doesn't
164 copy pointed-to items, of course,
165 but those are all const. */
166 t_commrec *cr; /* we need a local version of this */
170 fnm = dup_tfn(mc.nfile, mc.fnm);
172 cr = reinitialize_commrec_for_this_thread(mc.cr);
179 mdrunner(&mc.hw_opt, fplog, cr, mc.nfile, fnm, mc.oenv,
180 mc.bVerbose, mc.bCompact, mc.nstglobalcomm,
181 mc.ddxyz, mc.dd_node_order, mc.rdd,
182 mc.rconstr, mc.dddlb_opt, mc.dlb_scale,
183 mc.ddcsx, mc.ddcsy, mc.ddcsz,
184 mc.nbpu_opt, mc.nstlist_cmdline,
185 mc.nsteps_cmdline, mc.nstepout, mc.resetstep,
186 mc.nmultisim, mc.repl_ex_nst, mc.repl_ex_nex, mc.repl_ex_seed, mc.pforce,
187 mc.cpt_period, mc.max_hours, mc.imdport, mc.Flags);
190 /* called by mdrunner() to start a specific number of threads (including
191 the main thread) for thread-parallel runs. This in turn calls mdrunner()
193 All options besides nthreads are the same as for mdrunner(). */
194 static t_commrec *mdrunner_start_threads(gmx_hw_opt_t *hw_opt,
195 FILE *fplog, t_commrec *cr, int nfile,
196 const t_filenm fnm[], const output_env_t oenv, gmx_bool bVerbose,
197 gmx_bool bCompact, int nstglobalcomm,
198 ivec ddxyz, int dd_node_order, real rdd, real rconstr,
199 const char *dddlb_opt, real dlb_scale,
200 const char *ddcsx, const char *ddcsy, const char *ddcsz,
201 const char *nbpu_opt, int nstlist_cmdline,
202 gmx_int64_t nsteps_cmdline,
203 int nstepout, int resetstep,
204 int nmultisim, int repl_ex_nst, int repl_ex_nex, int repl_ex_seed,
205 real pforce, real cpt_period, real max_hours,
209 struct mdrunner_arglist *mda;
210 t_commrec *crn; /* the new commrec */
213 /* first check whether we even need to start tMPI */
214 if (hw_opt->nthreads_tmpi < 2)
219 /* a few small, one-time, almost unavoidable memory leaks: */
221 fnmn = dup_tfn(nfile, fnm);
223 /* fill the data structure to pass as void pointer to thread start fn */
224 /* hw_opt contains pointers, which should all be NULL at this stage */
225 mda->hw_opt = *hw_opt;
231 mda->bVerbose = bVerbose;
232 mda->bCompact = bCompact;
233 mda->nstglobalcomm = nstglobalcomm;
234 mda->ddxyz[XX] = ddxyz[XX];
235 mda->ddxyz[YY] = ddxyz[YY];
236 mda->ddxyz[ZZ] = ddxyz[ZZ];
237 mda->dd_node_order = dd_node_order;
239 mda->rconstr = rconstr;
240 mda->dddlb_opt = dddlb_opt;
241 mda->dlb_scale = dlb_scale;
245 mda->nbpu_opt = nbpu_opt;
246 mda->nstlist_cmdline = nstlist_cmdline;
247 mda->nsteps_cmdline = nsteps_cmdline;
248 mda->nstepout = nstepout;
249 mda->resetstep = resetstep;
250 mda->nmultisim = nmultisim;
251 mda->repl_ex_nst = repl_ex_nst;
252 mda->repl_ex_nex = repl_ex_nex;
253 mda->repl_ex_seed = repl_ex_seed;
254 mda->pforce = pforce;
255 mda->cpt_period = cpt_period;
256 mda->max_hours = max_hours;
259 /* now spawn new threads that start mdrunner_start_fn(), while
260 the main thread returns, we set thread affinity later */
261 ret = tMPI_Init_fn(TRUE, hw_opt->nthreads_tmpi, TMPI_AFFINITY_NONE,
262 mdrunner_start_fn, (void*)(mda) );
263 if (ret != TMPI_SUCCESS)
268 crn = reinitialize_commrec_for_this_thread(cr);
272 #endif /* GMX_THREAD_MPI */
275 /* We determine the extra cost of the non-bonded kernels compared to
276 * a reference nstlist value of 10 (which is the default in grompp).
278 static const int nbnxnReferenceNstlist = 10;
279 /* The values to try when switching */
280 const int nstlist_try[] = { 20, 25, 40 };
281 #define NNSTL sizeof(nstlist_try)/sizeof(nstlist_try[0])
282 /* Increase nstlist until the non-bonded cost increases more than listfac_ok,
283 * but never more than listfac_max.
284 * A standard (protein+)water system at 300K with PME ewald_rtol=1e-5
285 * needs 1.28 at rcoulomb=0.9 and 1.24 at rcoulomb=1.0 to get to nstlist=40.
286 * Note that both CPU and GPU factors are conservative. Performance should
287 * not go down due to this tuning, except with a relatively slow GPU.
288 * On the other hand, at medium/high parallelization or with fast GPUs
289 * nstlist will not be increased enough to reach optimal performance.
291 /* CPU: pair-search is about a factor 1.5 slower than the non-bonded kernel */
292 static const float nbnxn_cpu_listfac_ok = 1.05;
293 static const float nbnxn_cpu_listfac_max = 1.09;
294 /* GPU: pair-search is a factor 1.5-3 slower than the non-bonded kernel */
295 static const float nbnxn_gpu_listfac_ok = 1.20;
296 static const float nbnxn_gpu_listfac_max = 1.30;
298 /* Try to increase nstlist when using the Verlet cut-off scheme */
299 static void increase_nstlist(FILE *fp, t_commrec *cr,
300 t_inputrec *ir, int nstlist_cmdline,
301 const gmx_mtop_t *mtop, matrix box,
304 float listfac_ok, listfac_max;
305 int nstlist_orig, nstlist_prev;
306 verletbuf_list_setup_t ls;
307 real rlistWithReferenceNstlist, rlist_inc, rlist_ok, rlist_max;
308 real rlist_new, rlist_prev;
309 size_t nstlist_ind = 0;
311 gmx_bool bBox, bDD, bCont;
312 const char *nstl_gpu = "\nFor optimal performance with a GPU nstlist (now %d) should be larger.\nThe optimum depends on your CPU and GPU resources.\nYou might want to try several nstlist values.\n";
313 const char *nve_err = "Can not increase nstlist because an NVE ensemble is used";
314 const char *vbd_err = "Can not increase nstlist because verlet-buffer-tolerance is not set or used";
315 const char *box_err = "Can not increase nstlist because the box is too small";
316 const char *dd_err = "Can not increase nstlist because of domain decomposition limitations";
318 const float oneThird = 1.0f / 3.0f;
320 if (nstlist_cmdline <= 0)
322 if (ir->nstlist == 1)
324 /* The user probably set nstlist=1 for a reason,
325 * don't mess with the settings.
330 if (fp != NULL && bGPU && ir->nstlist < nstlist_try[0])
332 fprintf(fp, nstl_gpu, ir->nstlist);
335 while (nstlist_ind < NNSTL && ir->nstlist >= nstlist_try[nstlist_ind])
339 if (nstlist_ind == NNSTL)
341 /* There are no larger nstlist value to try */
346 if (EI_MD(ir->eI) && ir->etc == etcNO)
350 fprintf(stderr, "%s\n", nve_err);
354 fprintf(fp, "%s\n", nve_err);
360 if (ir->verletbuf_tol == 0 && bGPU)
362 gmx_fatal(FARGS, "You are using an old tpr file with a GPU, please generate a new tpr file with an up to date version of grompp");
365 if (ir->verletbuf_tol < 0)
369 fprintf(stderr, "%s\n", vbd_err);
373 fprintf(fp, "%s\n", vbd_err);
381 listfac_ok = nbnxn_gpu_listfac_ok;
382 listfac_max = nbnxn_gpu_listfac_max;
386 listfac_ok = nbnxn_cpu_listfac_ok;
387 listfac_max = nbnxn_cpu_listfac_max;
390 nstlist_orig = ir->nstlist;
391 if (nstlist_cmdline > 0)
395 sprintf(buf, "Getting nstlist=%d from command line option",
398 ir->nstlist = nstlist_cmdline;
401 verletbuf_get_list_setup(TRUE, bGPU, &ls);
403 /* Allow rlist to make the list a given factor larger than the list
404 * would be with the reference value for nstlist (10).
406 nstlist_prev = ir->nstlist;
407 ir->nstlist = nbnxnReferenceNstlist;
408 calc_verlet_buffer_size(mtop, det(box), ir, -1, &ls, NULL,
409 &rlistWithReferenceNstlist);
410 ir->nstlist = nstlist_prev;
412 /* Determine the pair list size increase due to zero interactions */
413 rlist_inc = nbnxn_get_rlist_effective_inc(ls.cluster_size_j,
414 mtop->natoms/det(box));
415 rlist_ok = (rlistWithReferenceNstlist + rlist_inc)*pow(listfac_ok, oneThird) - rlist_inc;
416 rlist_max = (rlistWithReferenceNstlist + rlist_inc)*pow(listfac_max, oneThird) - rlist_inc;
419 fprintf(debug, "nstlist tuning: rlist_inc %.3f rlist_ok %.3f rlist_max %.3f\n",
420 rlist_inc, rlist_ok, rlist_max);
423 nstlist_prev = nstlist_orig;
424 rlist_prev = ir->rlist;
427 if (nstlist_cmdline <= 0)
429 ir->nstlist = nstlist_try[nstlist_ind];
432 /* Set the pair-list buffer size in ir */
433 calc_verlet_buffer_size(mtop, det(box), ir, -1, &ls, NULL, &rlist_new);
435 /* Does rlist fit in the box? */
436 bBox = (sqr(rlist_new) < max_cutoff2(ir->ePBC, box));
438 if (bBox && DOMAINDECOMP(cr))
440 /* Check if rlist fits in the domain decomposition */
441 if (inputrec2nboundeddim(ir) < DIM)
443 gmx_incons("Changing nstlist with domain decomposition and unbounded dimensions is not implemented yet");
445 copy_mat(box, state_tmp.box);
446 bDD = change_dd_cutoff(cr, &state_tmp, ir, rlist_new);
451 fprintf(debug, "nstlist %d rlist %.3f bBox %d bDD %d\n",
452 ir->nstlist, rlist_new, bBox, bDD);
457 if (nstlist_cmdline <= 0)
459 if (bBox && bDD && rlist_new <= rlist_max)
461 /* Increase nstlist */
462 nstlist_prev = ir->nstlist;
463 rlist_prev = rlist_new;
464 bCont = (nstlist_ind+1 < NNSTL && rlist_new < rlist_ok);
468 /* Stick with the previous nstlist */
469 ir->nstlist = nstlist_prev;
470 rlist_new = rlist_prev;
482 gmx_warning(!bBox ? box_err : dd_err);
485 fprintf(fp, "\n%s\n", bBox ? box_err : dd_err);
487 ir->nstlist = nstlist_orig;
489 else if (ir->nstlist != nstlist_orig || rlist_new != ir->rlist)
491 sprintf(buf, "Changing nstlist from %d to %d, rlist from %g to %g",
492 nstlist_orig, ir->nstlist,
493 ir->rlist, rlist_new);
496 fprintf(stderr, "%s\n\n", buf);
500 fprintf(fp, "%s\n\n", buf);
502 ir->rlist = rlist_new;
503 ir->rlistlong = rlist_new;
507 static void prepare_verlet_scheme(FILE *fplog,
511 const gmx_mtop_t *mtop,
515 /* For NVE simulations, we will retain the initial list buffer */
516 if (ir->verletbuf_tol > 0 && !(EI_MD(ir->eI) && ir->etc == etcNO))
518 /* Update the Verlet buffer size for the current run setup */
519 verletbuf_list_setup_t ls;
522 /* Here we assume SIMD-enabled kernels are being used. But as currently
523 * calc_verlet_buffer_size gives the same results for 4x8 and 4x4
524 * and 4x2 gives a larger buffer than 4x4, this is ok.
526 verletbuf_get_list_setup(TRUE, bUseGPU, &ls);
528 calc_verlet_buffer_size(mtop, det(box), ir, -1, &ls, NULL, &rlist_new);
530 if (rlist_new != ir->rlist)
534 fprintf(fplog, "\nChanging rlist from %g to %g for non-bonded %dx%d atom kernels\n\n",
535 ir->rlist, rlist_new,
536 ls.cluster_size_i, ls.cluster_size_j);
538 ir->rlist = rlist_new;
539 ir->rlistlong = rlist_new;
543 if (nstlist_cmdline > 0 && (!EI_DYNAMICS(ir->eI) || ir->verletbuf_tol <= 0))
545 gmx_fatal(FARGS, "Can not set nstlist without %s",
546 !EI_DYNAMICS(ir->eI) ? "dynamics" : "verlet-buffer-tolerance");
549 if (EI_DYNAMICS(ir->eI))
551 /* Set or try nstlist values */
552 increase_nstlist(fplog, cr, ir, nstlist_cmdline, mtop, box, bUseGPU);
556 /* Override the value in inputrec with value passed on the command line (if any) */
557 static void override_nsteps_cmdline(FILE *fplog,
558 gmx_int64_t nsteps_cmdline,
565 /* override with anything else than the default -2 */
566 if (nsteps_cmdline > -2)
568 char sbuf_steps[STEPSTRSIZE];
569 char sbuf_msg[STRLEN];
571 ir->nsteps = nsteps_cmdline;
572 if (EI_DYNAMICS(ir->eI) && nsteps_cmdline != -1)
574 sprintf(sbuf_msg, "Overriding nsteps with value passed on the command line: %s steps, %.3g ps",
575 gmx_step_str(nsteps_cmdline, sbuf_steps),
576 fabs(nsteps_cmdline*ir->delta_t));
580 sprintf(sbuf_msg, "Overriding nsteps with value passed on the command line: %s steps",
581 gmx_step_str(nsteps_cmdline, sbuf_steps));
584 md_print_warn(cr, fplog, "%s\n", sbuf_msg);
586 else if (nsteps_cmdline < -2)
588 gmx_fatal(FARGS, "Invalid nsteps value passed on the command line: %d",
591 /* Do nothing if nsteps_cmdline == -2 */
594 int mdrunner(gmx_hw_opt_t *hw_opt,
595 FILE *fplog, t_commrec *cr, int nfile,
596 const t_filenm fnm[], const output_env_t oenv, gmx_bool bVerbose,
597 gmx_bool bCompact, int nstglobalcomm,
598 ivec ddxyz, int dd_node_order, real rdd, real rconstr,
599 const char *dddlb_opt, real dlb_scale,
600 const char *ddcsx, const char *ddcsy, const char *ddcsz,
601 const char *nbpu_opt, int nstlist_cmdline,
602 gmx_int64_t nsteps_cmdline, int nstepout, int resetstep,
603 int gmx_unused nmultisim, int repl_ex_nst, int repl_ex_nex,
604 int repl_ex_seed, real pforce, real cpt_period, real max_hours,
605 int imdport, unsigned long Flags)
607 gmx_bool bForceUseGPU, bTryUseGPU, bRerunMD, bCantUseGPU;
608 t_inputrec *inputrec;
609 t_state *state = NULL;
611 gmx_ddbox_t ddbox = {0};
612 int npme_major, npme_minor;
614 gmx_mtop_t *mtop = NULL;
615 t_mdatoms *mdatoms = NULL;
616 t_forcerec *fr = NULL;
617 t_fcdata *fcd = NULL;
618 real ewaldcoeff_q = 0;
619 real ewaldcoeff_lj = 0;
620 struct gmx_pme_t **pmedata = NULL;
621 gmx_vsite_t *vsite = NULL;
623 int nChargePerturbed = -1, nTypePerturbed = 0, status;
624 gmx_wallcycle_t wcycle;
626 gmx_walltime_accounting_t walltime_accounting = NULL;
628 gmx_int64_t reset_counters;
629 gmx_edsam_t ed = NULL;
630 int nthreads_pme = 1;
632 gmx_membed_t membed = NULL;
633 gmx_hw_info_t *hwinfo = NULL;
634 /* The master rank decides early on bUseGPU and broadcasts this later */
635 gmx_bool bUseGPU = FALSE;
637 /* CAUTION: threads may be started later on in this function, so
638 cr doesn't reflect the final parallel state right now */
642 if (Flags & MD_APPENDFILES)
647 bRerunMD = (Flags & MD_RERUN);
648 bForceUseGPU = (strncmp(nbpu_opt, "gpu", 3) == 0);
649 bTryUseGPU = (strncmp(nbpu_opt, "auto", 4) == 0) || bForceUseGPU;
650 /* Rerun execution time is dominated by I/O and pair search, so
651 * GPUs are not very useful, plus they do not support more than
652 * one energy group. Don't select them when they can't be used,
653 * unless the user requested it, then fatal_error is called later.
655 * TODO it would be nice to notify the user that if this check
656 * causes GPUs not to be used that this is what is happening, and
657 * why, but that will be easier to do after some future
659 bCantUseGPU = bRerunMD && (inputrec->opts.ngener > 1);
660 bTryUseGPU = bTryUseGPU && !(bCantUseGPU && !bForceUseGPU);
662 /* Detect hardware, gather information. This is an operation that is
663 * global for this process (MPI rank). */
664 hwinfo = gmx_detect_hardware(fplog, cr, bTryUseGPU);
666 gmx_print_detected_hardware(fplog, cr, hwinfo);
670 /* Print references after all software/hardware printing */
671 please_cite(fplog, "Pall2015");
672 please_cite(fplog, "Pronk2013");
673 please_cite(fplog, "Hess2008b");
674 please_cite(fplog, "Spoel2005a");
675 please_cite(fplog, "Lindahl2001a");
676 please_cite(fplog, "Berendsen95a");
682 /* Read (nearly) all data required for the simulation */
683 read_tpx_state(ftp2fn(efTPR, nfile, fnm), inputrec, state, NULL, mtop);
685 if (inputrec->cutoff_scheme == ecutsVERLET)
687 /* Here the master rank decides if all ranks will use GPUs */
688 bUseGPU = (hwinfo->gpu_info.n_dev_compatible > 0 ||
689 getenv("GMX_EMULATE_GPU") != NULL);
691 /* TODO add GPU kernels for this and replace this check by:
692 * (bUseGPU && (ir->vdwtype == evdwPME &&
693 * ir->ljpme_combination_rule == eljpmeLB))
694 * update the message text and the content of nbnxn_acceleration_supported.
697 !nbnxn_acceleration_supported(fplog, cr, inputrec, bUseGPU))
699 /* Fallback message printed by nbnxn_acceleration_supported */
702 gmx_fatal(FARGS, "GPU acceleration requested, but not supported with the given input settings");
707 prepare_verlet_scheme(fplog, cr,
708 inputrec, nstlist_cmdline, mtop, state->box,
713 if (nstlist_cmdline > 0)
715 gmx_fatal(FARGS, "Can not set nstlist with the group cut-off scheme");
718 if (hwinfo->gpu_info.n_dev_compatible > 0)
720 md_print_warn(cr, fplog,
721 "NOTE: GPU(s) found, but the current simulation can not use GPUs\n"
722 " To use a GPU, set the mdp option: cutoff-scheme = Verlet\n");
727 gmx_fatal(FARGS, "GPU requested, but can't be used without cutoff-scheme=Verlet");
730 #ifdef GMX_TARGET_BGQ
731 md_print_warn(cr, fplog,
732 "NOTE: There is no SIMD implementation of the group scheme kernels on\n"
733 " BlueGene/Q. You will observe better performance from using the\n"
734 " Verlet cut-off scheme.\n");
738 if (inputrec->eI == eiSD2)
740 md_print_warn(cr, fplog, "The stochastic dynamics integrator %s is deprecated, since\n"
741 "it is slower than integrator %s and is slightly less accurate\n"
742 "with constraints. Use the %s integrator.",
743 ei_names[inputrec->eI], ei_names[eiSD1], ei_names[eiSD1]);
747 /* Check and update the hardware options for internal consistency */
748 check_and_update_hw_opt_1(hw_opt, SIMMASTER(cr));
750 /* Early check for externally set process affinity. */
751 gmx_check_thread_affinity_set(fplog, cr,
752 hw_opt, hwinfo->nthreads_hw_avail, FALSE);
756 #ifdef GMX_THREAD_MPI
757 if (cr->npmenodes > 0 && hw_opt->nthreads_tmpi <= 0)
759 gmx_fatal(FARGS, "You need to explicitly specify the number of MPI threads (-ntmpi) when using separate PME ranks");
763 if (hw_opt->nthreads_omp_pme != hw_opt->nthreads_omp &&
766 gmx_fatal(FARGS, "You need to explicitly specify the number of PME ranks (-npme) when using different number of OpenMP threads for PP and PME ranks");
770 #ifdef GMX_THREAD_MPI
773 /* Since the master knows the cut-off scheme, update hw_opt for this.
774 * This is done later for normal MPI and also once more with tMPI
775 * for all tMPI ranks.
777 check_and_update_hw_opt_2(hw_opt, inputrec->cutoff_scheme);
779 /* NOW the threads will be started: */
780 hw_opt->nthreads_tmpi = get_nthreads_mpi(hwinfo,
784 if (hw_opt->nthreads_tot > 0 && hw_opt->nthreads_omp <= 0)
786 hw_opt->nthreads_omp = hw_opt->nthreads_tot/hw_opt->nthreads_tmpi;
789 if (hw_opt->nthreads_tmpi > 1)
791 t_commrec *cr_old = cr;
792 /* now start the threads. */
793 cr = mdrunner_start_threads(hw_opt, fplog, cr_old, nfile, fnm,
794 oenv, bVerbose, bCompact, nstglobalcomm,
795 ddxyz, dd_node_order, rdd, rconstr,
796 dddlb_opt, dlb_scale, ddcsx, ddcsy, ddcsz,
797 nbpu_opt, nstlist_cmdline,
798 nsteps_cmdline, nstepout, resetstep, nmultisim,
799 repl_ex_nst, repl_ex_nex, repl_ex_seed, pforce,
800 cpt_period, max_hours,
802 /* the main thread continues here with a new cr. We don't deallocate
803 the old cr because other threads may still be reading it. */
806 gmx_comm("Failed to spawn threads");
811 /* END OF CAUTION: cr is now reliable */
813 /* g_membed initialisation *
814 * Because we change the mtop, init_membed is called before the init_parallel *
815 * (in case we ever want to make it run in parallel) */
816 if (opt2bSet("-membed", nfile, fnm))
820 fprintf(stderr, "Initializing membed");
822 membed = init_membed(fplog, nfile, fnm, mtop, inputrec, state, cr, &cpt_period);
827 /* now broadcast everything to the non-master nodes/threads: */
828 init_parallel(cr, inputrec, mtop);
830 /* The master rank decided on the use of GPUs,
831 * broadcast this information to all ranks.
833 gmx_bcast_sim(sizeof(bUseGPU), &bUseGPU, cr);
838 pr_inputrec(fplog, 0, "Input Parameters", inputrec, FALSE);
839 fprintf(fplog, "\n");
842 /* now make sure the state is initialized and propagated */
843 set_state_entries(state, inputrec);
845 /* A parallel command line option consistency check that we can
846 only do after any threads have started. */
848 (ddxyz[XX] > 1 || ddxyz[YY] > 1 || ddxyz[ZZ] > 1 || cr->npmenodes > 0))
851 "The -dd or -npme option request a parallel simulation, "
853 "but %s was compiled without threads or MPI enabled"
855 #ifdef GMX_THREAD_MPI
856 "but the number of threads (option -nt) is 1"
858 "but %s was not started through mpirun/mpiexec or only one rank was requested through mpirun/mpiexec"
861 , output_env_get_program_display_name(oenv)
866 (EI_ENERGY_MINIMIZATION(inputrec->eI) || eiNM == inputrec->eI))
868 gmx_fatal(FARGS, "The .mdp file specified an energy mininization or normal mode algorithm, and these are not compatible with mdrun -rerun");
871 if (can_use_allvsall(inputrec, TRUE, cr, fplog) && DOMAINDECOMP(cr))
873 gmx_fatal(FARGS, "All-vs-all loops do not work with domain decomposition, use a single MPI rank");
876 if (!(EEL_PME(inputrec->coulombtype) || EVDW_PME(inputrec->vdwtype)))
878 if (cr->npmenodes > 0)
880 gmx_fatal_collective(FARGS, cr, NULL,
881 "PME-only ranks are requested, but the system does not use PME for electrostatics or LJ");
887 if (bUseGPU && cr->npmenodes < 0)
889 /* With GPUs we don't automatically use PME-only ranks. PME ranks can
890 * improve performance with many threads per GPU, since our OpenMP
891 * scaling is bad, but it's difficult to automate the setup.
899 fcRegisterSteps(inputrec->nsteps, inputrec->init_step);
903 /* NMR restraints must be initialized before load_checkpoint,
904 * since with time averaging the history is added to t_state.
905 * For proper consistency check we therefore need to extend
907 * So the PME-only nodes (if present) will also initialize
908 * the distance restraints.
912 /* This needs to be called before read_checkpoint to extend the state */
913 init_disres(fplog, mtop, inputrec, cr, fcd, state, repl_ex_nst > 0);
915 init_orires(fplog, mtop, state->x, inputrec, cr, &(fcd->orires),
918 if (DEFORM(*inputrec))
920 /* Store the deform reference box before reading the checkpoint */
923 copy_mat(state->box, box);
927 gmx_bcast(sizeof(box), box, cr);
929 /* Because we do not have the update struct available yet
930 * in which the reference values should be stored,
931 * we store them temporarily in static variables.
932 * This should be thread safe, since they are only written once
933 * and with identical values.
935 tMPI_Thread_mutex_lock(&deform_init_box_mutex);
936 deform_init_init_step_tpx = inputrec->init_step;
937 copy_mat(box, deform_init_box_tpx);
938 tMPI_Thread_mutex_unlock(&deform_init_box_mutex);
941 if (opt2bSet("-cpi", nfile, fnm))
943 /* Check if checkpoint file exists before doing continuation.
944 * This way we can use identical input options for the first and subsequent runs...
946 if (gmx_fexist_master(opt2fn_master("-cpi", nfile, fnm, cr), cr) )
948 load_checkpoint(opt2fn_master("-cpi", nfile, fnm, cr), &fplog,
950 inputrec, state, &bReadEkin,
951 (Flags & MD_APPENDFILES),
952 (Flags & MD_APPENDFILESSET));
956 Flags |= MD_READ_EKIN;
961 if (MASTER(cr) && (Flags & MD_APPENDFILES))
963 gmx_log_open(ftp2fn(efLOG, nfile, fnm), cr,
967 /* override nsteps with value from cmdline */
968 override_nsteps_cmdline(fplog, nsteps_cmdline, inputrec, cr);
972 copy_mat(state->box, box);
977 gmx_bcast(sizeof(box), box, cr);
980 /* Essential dynamics */
981 if (opt2bSet("-ei", nfile, fnm))
983 /* Open input and output files, allocate space for ED data structure */
984 ed = ed_open(mtop->natoms, &state->edsamstate, nfile, fnm, Flags, oenv, cr);
987 if (PAR(cr) && !(EI_TPI(inputrec->eI) ||
988 inputrec->eI == eiNM))
990 cr->dd = init_domain_decomposition(fplog, cr, Flags, ddxyz, rdd, rconstr,
991 dddlb_opt, dlb_scale,
995 &ddbox, &npme_major, &npme_minor);
997 make_dd_communicators(fplog, cr, dd_node_order);
999 /* Set overallocation to avoid frequent reallocation of arrays */
1000 set_over_alloc_dd(TRUE);
1004 /* PME, if used, is done on all nodes with 1D decomposition */
1006 cr->duty = (DUTY_PP | DUTY_PME);
1010 if (inputrec->ePBC == epbcSCREW)
1013 "pbc=%s is only implemented with domain decomposition",
1014 epbc_names[inputrec->ePBC]);
1020 /* After possible communicator splitting in make_dd_communicators.
1021 * we can set up the intra/inter node communication.
1023 gmx_setup_nodecomm(fplog, cr);
1026 /* Initialize per-physical-node MPI process/thread ID and counters. */
1027 gmx_init_intranode_counters(cr);
1031 md_print_info(cr, fplog,
1032 "This is simulation %d out of %d running as a composite GROMACS\n"
1033 "multi-simulation job. Setup for this simulation:\n\n",
1034 cr->ms->sim, cr->ms->nsim);
1036 md_print_info(cr, fplog, "Using %d MPI %s\n",
1038 #ifdef GMX_THREAD_MPI
1039 cr->nnodes == 1 ? "thread" : "threads"
1041 cr->nnodes == 1 ? "process" : "processes"
1047 /* Check and update hw_opt for the cut-off scheme */
1048 check_and_update_hw_opt_2(hw_opt, inputrec->cutoff_scheme);
1050 gmx_omp_nthreads_init(fplog, cr,
1051 hwinfo->nthreads_hw_avail,
1052 hw_opt->nthreads_omp,
1053 hw_opt->nthreads_omp_pme,
1054 (cr->duty & DUTY_PP) == 0,
1055 inputrec->cutoff_scheme == ecutsVERLET);
1058 if (integrator[inputrec->eI].func != do_tpi &&
1059 inputrec->cutoff_scheme == ecutsVERLET)
1061 gmx_feenableexcept();
1067 /* Select GPU id's to use */
1068 gmx_select_gpu_ids(fplog, cr, &hwinfo->gpu_info, bForceUseGPU,
1073 /* Ignore (potentially) manually selected GPUs */
1074 hw_opt->gpu_opt.n_dev_use = 0;
1077 /* check consistency across ranks of things like SIMD
1078 * support and number of GPUs selected */
1079 gmx_check_hw_runconf_consistency(fplog, hwinfo, cr, hw_opt, bUseGPU);
1081 /* Now that we know the setup is consistent, check for efficiency */
1082 check_resource_division_efficiency(hwinfo, hw_opt, Flags & MD_NTOMPSET,
1085 if (DOMAINDECOMP(cr))
1087 /* When we share GPUs over ranks, we need to know this for the DLB */
1088 dd_setup_dlb_resource_sharing(cr, hwinfo, hw_opt);
1091 /* getting number of PP/PME threads
1092 PME: env variable should be read only on one node to make sure it is
1093 identical everywhere;
1095 /* TODO nthreads_pp is only used for pinning threads.
1096 * This is a temporary solution until we have a hw topology library.
1098 nthreads_pp = gmx_omp_nthreads_get(emntNonbonded);
1099 nthreads_pme = gmx_omp_nthreads_get(emntPME);
1101 wcycle = wallcycle_init(fplog, resetstep, cr, nthreads_pp, nthreads_pme);
1105 /* Master synchronizes its value of reset_counters with all nodes
1106 * including PME only nodes */
1107 reset_counters = wcycle_get_reset_counters(wcycle);
1108 gmx_bcast_sim(sizeof(reset_counters), &reset_counters, cr);
1109 wcycle_set_reset_counters(wcycle, reset_counters);
1113 if (cr->duty & DUTY_PP)
1115 bcast_state(cr, state);
1117 /* Initiate forcerecord */
1119 fr->hwinfo = hwinfo;
1120 fr->gpu_opt = &hw_opt->gpu_opt;
1121 init_forcerec(fplog, oenv, fr, fcd, inputrec, mtop, cr, box,
1122 opt2fn("-table", nfile, fnm),
1123 opt2fn("-tabletf", nfile, fnm),
1124 opt2fn("-tablep", nfile, fnm),
1125 opt2fn("-tableb", nfile, fnm),
1130 /* version for PCA_NOT_READ_NODE (see md.c) */
1131 /*init_forcerec(fplog,fr,fcd,inputrec,mtop,cr,box,FALSE,
1132 "nofile","nofile","nofile","nofile",FALSE,pforce);
1135 /* Initialize QM-MM */
1138 init_QMMMrec(cr, mtop, inputrec, fr);
1141 /* Initialize the mdatoms structure.
1142 * mdatoms is not filled with atom data,
1143 * as this can not be done now with domain decomposition.
1145 mdatoms = init_mdatoms(fplog, mtop, inputrec->efep != efepNO);
1147 /* Initialize the virtual site communication */
1148 vsite = init_vsite(mtop, cr, FALSE);
1150 calc_shifts(box, fr->shift_vec);
1152 /* With periodic molecules the charge groups should be whole at start up
1153 * and the virtual sites should not be far from their proper positions.
1155 if (!inputrec->bContinuation && MASTER(cr) &&
1156 !(inputrec->ePBC != epbcNONE && inputrec->bPeriodicMols))
1158 /* Make molecules whole at start of run */
1159 if (fr->ePBC != epbcNONE)
1161 do_pbc_first_mtop(fplog, inputrec->ePBC, box, mtop, state->x);
1165 /* Correct initial vsite positions are required
1166 * for the initial distribution in the domain decomposition
1167 * and for the initial shell prediction.
1169 construct_vsites_mtop(vsite, mtop, state->x);
1173 if (EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype))
1175 ewaldcoeff_q = fr->ewaldcoeff_q;
1176 ewaldcoeff_lj = fr->ewaldcoeff_lj;
1177 pmedata = &fr->pmedata;
1186 /* This is a PME only node */
1188 /* We don't need the state */
1191 ewaldcoeff_q = calc_ewaldcoeff_q(inputrec->rcoulomb, inputrec->ewald_rtol);
1192 ewaldcoeff_lj = calc_ewaldcoeff_lj(inputrec->rvdw, inputrec->ewald_rtol_lj);
1196 if (hw_opt->thread_affinity != threadaffOFF)
1198 /* Before setting affinity, check whether the affinity has changed
1199 * - which indicates that probably the OpenMP library has changed it
1200 * since we first checked).
1202 gmx_check_thread_affinity_set(fplog, cr,
1203 hw_opt, hwinfo->nthreads_hw_avail, TRUE);
1205 /* Set the CPU affinity */
1206 gmx_set_thread_affinity(fplog, cr, hw_opt, hwinfo);
1209 /* Initiate PME if necessary,
1210 * either on all nodes or on dedicated PME nodes only. */
1211 if (EEL_PME(inputrec->coulombtype) || EVDW_PME(inputrec->vdwtype))
1215 nChargePerturbed = mdatoms->nChargePerturbed;
1216 if (EVDW_PME(inputrec->vdwtype))
1218 nTypePerturbed = mdatoms->nTypePerturbed;
1221 if (cr->npmenodes > 0)
1223 /* The PME only nodes need to know nChargePerturbed(FEP on Q) and nTypePerturbed(FEP on LJ)*/
1224 gmx_bcast_sim(sizeof(nChargePerturbed), &nChargePerturbed, cr);
1225 gmx_bcast_sim(sizeof(nTypePerturbed), &nTypePerturbed, cr);
1228 if (cr->duty & DUTY_PME)
1230 status = gmx_pme_init(pmedata, cr, npme_major, npme_minor, inputrec,
1231 mtop ? mtop->natoms : 0, nChargePerturbed, nTypePerturbed,
1232 (Flags & MD_REPRODUCIBLE), nthreads_pme);
1235 gmx_fatal(FARGS, "Error %d initializing PME", status);
1241 if (integrator[inputrec->eI].func == do_md)
1243 /* Turn on signal handling on all nodes */
1245 * (A user signal from the PME nodes (if any)
1246 * is communicated to the PP nodes.
1248 signal_handler_install();
1251 if (cr->duty & DUTY_PP)
1253 /* Assumes uniform use of the number of OpenMP threads */
1254 walltime_accounting = walltime_accounting_init(gmx_omp_nthreads_get(emntDefault));
1256 if (inputrec->bPull)
1258 /* Initialize pull code */
1259 inputrec->pull_work =
1260 init_pull(fplog, inputrec->pull, inputrec, nfile, fnm,
1261 mtop, cr, oenv, inputrec->fepvals->init_lambda,
1262 EI_DYNAMICS(inputrec->eI) && MASTER(cr), Flags);
1267 /* Initialize enforced rotation code */
1268 init_rot(fplog, inputrec, nfile, fnm, cr, state->x, box, mtop, oenv,
1272 if (inputrec->eSwapCoords != eswapNO)
1274 /* Initialize ion swapping code */
1275 init_swapcoords(fplog, bVerbose, inputrec, opt2fn_master("-swap", nfile, fnm, cr),
1276 mtop, state->x, state->box, &state->swapstate, cr, oenv, Flags);
1279 constr = init_constraints(fplog, mtop, inputrec, ed, state, cr);
1281 if (DOMAINDECOMP(cr))
1283 GMX_RELEASE_ASSERT(fr, "fr was NULL while cr->duty was DUTY_PP");
1284 dd_init_bondeds(fplog, cr->dd, mtop, vsite, inputrec,
1285 Flags & MD_DDBONDCHECK, fr->cginfo_mb);
1287 set_dd_parameters(fplog, cr->dd, dlb_scale, inputrec, &ddbox);
1289 setup_dd_grid(fplog, cr->dd);
1292 /* Now do whatever the user wants us to do (how flexible...) */
1293 integrator[inputrec->eI].func(fplog, cr, nfile, fnm,
1294 oenv, bVerbose, bCompact,
1297 nstepout, inputrec, mtop,
1299 mdatoms, nrnb, wcycle, ed, fr,
1300 repl_ex_nst, repl_ex_nex, repl_ex_seed,
1302 cpt_period, max_hours,
1305 walltime_accounting);
1307 if (inputrec->bPull)
1309 finish_pull(inputrec->pull_work);
1314 finish_rot(inputrec->rot);
1320 GMX_RELEASE_ASSERT(pmedata, "pmedata was NULL while cr->duty was not DUTY_PP");
1322 walltime_accounting = walltime_accounting_init(gmx_omp_nthreads_get(emntPME));
1323 gmx_pmeonly(*pmedata, cr, nrnb, wcycle, walltime_accounting, ewaldcoeff_q, ewaldcoeff_lj, inputrec);
1326 wallcycle_stop(wcycle, ewcRUN);
1328 /* Finish up, write some stuff
1329 * if rerunMD, don't write last frame again
1331 finish_run(fplog, cr,
1332 inputrec, nrnb, wcycle, walltime_accounting,
1333 fr ? fr->nbv : NULL,
1334 EI_DYNAMICS(inputrec->eI) && !MULTISIM(cr));
1337 /* Free GPU memory and context */
1338 free_gpu_resources(fr, cr, &hwinfo->gpu_info, fr ? fr->gpu_opt : NULL);
1340 if (opt2bSet("-membed", nfile, fnm))
1345 gmx_hardware_info_free(hwinfo);
1347 /* Does what it says */
1348 print_date_and_time(fplog, cr->nodeid, "Finished mdrun", gmx_gettime());
1349 walltime_accounting_destroy(walltime_accounting);
1351 /* Close logfile already here if we were appending to it */
1352 if (MASTER(cr) && (Flags & MD_APPENDFILES))
1354 gmx_log_close(fplog);
1357 rc = (int)gmx_get_stop_condition();
1361 #ifdef GMX_THREAD_MPI
1362 /* we need to join all threads. The sub-threads join when they
1363 exit this function, but the master thread needs to be told to
1365 if (PAR(cr) && MASTER(cr))