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_consts.h"
80 #include "gromacs/mdlib/nbnxn_search.h"
81 #include "gromacs/pbcutil/pbc.h"
82 #include "gromacs/pulling/pull.h"
83 #include "gromacs/pulling/pull_rotation.h"
84 #include "gromacs/swap/swapcoords.h"
85 #include "gromacs/timing/wallcycle.h"
86 #include "gromacs/topology/mtop_util.h"
87 #include "gromacs/utility/gmxassert.h"
88 #include "gromacs/utility/gmxmpi.h"
89 #include "gromacs/utility/smalloc.h"
100 gmx_integrator_t *func;
103 /* The array should match the eI array in include/types/enums.h */
104 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}};
106 gmx_int64_t deform_init_init_step_tpx;
107 matrix deform_init_box_tpx;
108 tMPI_Thread_mutex_t deform_init_box_mutex = TMPI_THREAD_MUTEX_INITIALIZER;
111 #ifdef GMX_THREAD_MPI
112 /* The minimum number of atoms per tMPI thread. With fewer atoms than this,
113 * the number of threads will get lowered.
115 #define MIN_ATOMS_PER_MPI_THREAD 90
116 #define MIN_ATOMS_PER_GPU 900
118 struct mdrunner_arglist
133 const char *dddlb_opt;
138 const char *nbpu_opt;
140 gmx_int64_t nsteps_cmdline;
155 /* The function used for spawning threads. Extracts the mdrunner()
156 arguments from its one argument and calls mdrunner(), after making
158 static void mdrunner_start_fn(void *arg)
160 struct mdrunner_arglist *mda = (struct mdrunner_arglist*)arg;
161 struct mdrunner_arglist mc = *mda; /* copy the arg list to make sure
162 that it's thread-local. This doesn't
163 copy pointed-to items, of course,
164 but those are all const. */
165 t_commrec *cr; /* we need a local version of this */
169 fnm = dup_tfn(mc.nfile, mc.fnm);
171 cr = reinitialize_commrec_for_this_thread(mc.cr);
178 mdrunner(&mc.hw_opt, fplog, cr, mc.nfile, fnm, mc.oenv,
179 mc.bVerbose, mc.bCompact, mc.nstglobalcomm,
180 mc.ddxyz, mc.dd_node_order, mc.rdd,
181 mc.rconstr, mc.dddlb_opt, mc.dlb_scale,
182 mc.ddcsx, mc.ddcsy, mc.ddcsz,
183 mc.nbpu_opt, mc.nstlist_cmdline,
184 mc.nsteps_cmdline, mc.nstepout, mc.resetstep,
185 mc.nmultisim, mc.repl_ex_nst, mc.repl_ex_nex, mc.repl_ex_seed, mc.pforce,
186 mc.cpt_period, mc.max_hours, mc.imdport, mc.Flags);
189 /* called by mdrunner() to start a specific number of threads (including
190 the main thread) for thread-parallel runs. This in turn calls mdrunner()
192 All options besides nthreads are the same as for mdrunner(). */
193 static t_commrec *mdrunner_start_threads(gmx_hw_opt_t *hw_opt,
194 FILE *fplog, t_commrec *cr, int nfile,
195 const t_filenm fnm[], const output_env_t oenv, gmx_bool bVerbose,
196 gmx_bool bCompact, int nstglobalcomm,
197 ivec ddxyz, int dd_node_order, real rdd, real rconstr,
198 const char *dddlb_opt, real dlb_scale,
199 const char *ddcsx, const char *ddcsy, const char *ddcsz,
200 const char *nbpu_opt, int nstlist_cmdline,
201 gmx_int64_t nsteps_cmdline,
202 int nstepout, int resetstep,
203 int nmultisim, int repl_ex_nst, int repl_ex_nex, int repl_ex_seed,
204 real pforce, real cpt_period, real max_hours,
208 struct mdrunner_arglist *mda;
209 t_commrec *crn; /* the new commrec */
212 /* first check whether we even need to start tMPI */
213 if (hw_opt->nthreads_tmpi < 2)
218 /* a few small, one-time, almost unavoidable memory leaks: */
220 fnmn = dup_tfn(nfile, fnm);
222 /* fill the data structure to pass as void pointer to thread start fn */
223 /* hw_opt contains pointers, which should all be NULL at this stage */
224 mda->hw_opt = *hw_opt;
230 mda->bVerbose = bVerbose;
231 mda->bCompact = bCompact;
232 mda->nstglobalcomm = nstglobalcomm;
233 mda->ddxyz[XX] = ddxyz[XX];
234 mda->ddxyz[YY] = ddxyz[YY];
235 mda->ddxyz[ZZ] = ddxyz[ZZ];
236 mda->dd_node_order = dd_node_order;
238 mda->rconstr = rconstr;
239 mda->dddlb_opt = dddlb_opt;
240 mda->dlb_scale = dlb_scale;
244 mda->nbpu_opt = nbpu_opt;
245 mda->nstlist_cmdline = nstlist_cmdline;
246 mda->nsteps_cmdline = nsteps_cmdline;
247 mda->nstepout = nstepout;
248 mda->resetstep = resetstep;
249 mda->nmultisim = nmultisim;
250 mda->repl_ex_nst = repl_ex_nst;
251 mda->repl_ex_nex = repl_ex_nex;
252 mda->repl_ex_seed = repl_ex_seed;
253 mda->pforce = pforce;
254 mda->cpt_period = cpt_period;
255 mda->max_hours = max_hours;
258 /* now spawn new threads that start mdrunner_start_fn(), while
259 the main thread returns, we set thread affinity later */
260 ret = tMPI_Init_fn(TRUE, hw_opt->nthreads_tmpi, TMPI_AFFINITY_NONE,
261 mdrunner_start_fn, (void*)(mda) );
262 if (ret != TMPI_SUCCESS)
267 crn = reinitialize_commrec_for_this_thread(cr);
272 static int get_tmpi_omp_thread_division(const gmx_hw_info_t *hwinfo,
273 const gmx_hw_opt_t *hw_opt,
279 /* There are no separate PME nodes here, as we ensured in
280 * check_and_update_hw_opt that nthreads_tmpi>0 with PME nodes
281 * and a conditional ensures we would not have ended up here.
282 * Note that separate PME nodes might be switched on later.
286 nthreads_tmpi = ngpu;
287 if (nthreads_tot > 0 && nthreads_tot < nthreads_tmpi)
289 nthreads_tmpi = nthreads_tot;
292 else if (hw_opt->nthreads_omp > 0)
294 /* Here we could oversubscribe, when we do, we issue a warning later */
295 nthreads_tmpi = std::max(1, nthreads_tot/hw_opt->nthreads_omp);
299 /* TODO choose nthreads_omp based on hardware topology
300 when we have a hardware topology detection library */
301 /* In general, when running up to 4 threads, OpenMP should be faster.
302 * Note: on AMD Bulldozer we should avoid running OpenMP over two dies.
303 * On Intel>=Nehalem running OpenMP on a single CPU is always faster,
304 * even on two CPUs it's usually faster (but with many OpenMP threads
305 * it could be faster not to use HT, currently we always use HT).
306 * On Nehalem/Westmere we want to avoid running 16 threads over
307 * two CPUs with HT, so we need a limit<16; thus we use 12.
308 * A reasonable limit for Intel Sandy and Ivy bridge,
309 * not knowing the topology, is 16 threads.
310 * Below we check for Intel and AVX, which for now includes
311 * Sandy/Ivy Bridge, Has/Broadwell. By checking for AVX instead of
312 * model numbers we ensure also future Intel CPUs are covered.
314 const int nthreads_omp_always_faster = 4;
315 const int nthreads_omp_always_faster_Nehalem = 12;
316 const int nthreads_omp_always_faster_Intel_AVX = 16;
320 (gmx_cpuid_vendor(hwinfo->cpuid_info) == GMX_CPUID_VENDOR_INTEL &&
321 gmx_cpuid_feature(hwinfo->cpuid_info, GMX_CPUID_FEATURE_X86_AVX));
323 if (nthreads_tot <= nthreads_omp_always_faster ||
324 ((gmx_cpuid_is_intel_nehalem(hwinfo->cpuid_info) && nthreads_tot <= nthreads_omp_always_faster_Nehalem) ||
325 (bIntelAVX && nthreads_tot <= nthreads_omp_always_faster_Intel_AVX)))
327 /* Use pure OpenMP parallelization */
332 /* Don't use OpenMP parallelization */
333 nthreads_tmpi = nthreads_tot;
337 return nthreads_tmpi;
341 /* Get the number of threads to use for thread-MPI based on how many
342 * were requested, which algorithms we're using,
343 * and how many particles there are.
344 * At the point we have already called check_and_update_hw_opt.
345 * Thus all options should be internally consistent and consistent
346 * with the hardware, except that ntmpi could be larger than #GPU.
348 static int get_nthreads_mpi(const gmx_hw_info_t *hwinfo,
349 gmx_hw_opt_t *hw_opt,
350 t_inputrec *inputrec, gmx_mtop_t *mtop,
354 int nthreads_hw, nthreads_tot_max, nthreads_tmpi, nthreads_new, ngpu;
355 int min_atoms_per_mpi_thread;
358 if (hw_opt->nthreads_tmpi > 0)
360 /* Trivial, return right away */
361 return hw_opt->nthreads_tmpi;
364 nthreads_hw = hwinfo->nthreads_hw_avail;
366 /* How many total (#tMPI*#OpenMP) threads can we start? */
367 if (hw_opt->nthreads_tot > 0)
369 nthreads_tot_max = hw_opt->nthreads_tot;
373 nthreads_tot_max = nthreads_hw;
376 bCanUseGPU = (inputrec->cutoff_scheme == ecutsVERLET &&
377 hwinfo->gpu_info.n_dev_compatible > 0);
381 ngpu = hwinfo->gpu_info.n_dev_compatible;
388 if (inputrec->cutoff_scheme == ecutsGROUP)
390 /* We checked this before, but it doesn't hurt to do it once more */
391 assert(hw_opt->nthreads_omp == 1);
395 get_tmpi_omp_thread_division(hwinfo, hw_opt, nthreads_tot_max, ngpu);
397 if (inputrec->eI == eiNM || EI_TPI(inputrec->eI))
399 /* Dims/steps are divided over the nodes iso splitting the atoms */
400 min_atoms_per_mpi_thread = 0;
406 min_atoms_per_mpi_thread = MIN_ATOMS_PER_GPU;
410 min_atoms_per_mpi_thread = MIN_ATOMS_PER_MPI_THREAD;
414 /* Check if an algorithm does not support parallel simulation. */
415 if (nthreads_tmpi != 1 &&
416 ( inputrec->eI == eiLBFGS ||
417 inputrec->coulombtype == eelEWALD ) )
421 md_print_warn(cr, fplog, "The integration or electrostatics algorithm doesn't support parallel runs. Using a single thread-MPI thread.\n");
422 if (hw_opt->nthreads_tmpi > nthreads_tmpi)
424 gmx_fatal(FARGS, "You asked for more than 1 thread-MPI thread, but an algorithm doesn't support that");
427 else if (mtop->natoms/nthreads_tmpi < min_atoms_per_mpi_thread)
429 /* the thread number was chosen automatically, but there are too many
430 threads (too few atoms per thread) */
431 nthreads_new = std::max(1, mtop->natoms/min_atoms_per_mpi_thread);
433 /* Avoid partial use of Hyper-Threading */
434 if (gmx_cpuid_x86_smt(hwinfo->cpuid_info) == GMX_CPUID_X86_SMT_ENABLED &&
435 nthreads_new > nthreads_hw/2 && nthreads_new < nthreads_hw)
437 nthreads_new = nthreads_hw/2;
440 /* Avoid large prime numbers in the thread count */
441 if (nthreads_new >= 6)
443 /* Use only 6,8,10 with additional factors of 2 */
447 while (3*fac*2 <= nthreads_new)
452 nthreads_new = (nthreads_new/fac)*fac;
457 if (nthreads_new == 5)
463 nthreads_tmpi = nthreads_new;
465 fprintf(stderr, "\n");
466 fprintf(stderr, "NOTE: Parallelization is limited by the small number of atoms,\n");
467 fprintf(stderr, " only starting %d thread-MPI threads.\n", nthreads_tmpi);
468 fprintf(stderr, " You can use the -nt and/or -ntmpi option to optimize the number of threads.\n\n");
471 return nthreads_tmpi;
473 #endif /* GMX_THREAD_MPI */
476 /* We determine the extra cost of the non-bonded kernels compared to
477 * a reference nstlist value of 10 (which is the default in grompp).
479 static const int nbnxnReferenceNstlist = 10;
480 /* The values to try when switching */
481 const int nstlist_try[] = { 20, 25, 40 };
482 #define NNSTL sizeof(nstlist_try)/sizeof(nstlist_try[0])
483 /* Increase nstlist until the non-bonded cost increases more than listfac_ok,
484 * but never more than listfac_max.
485 * A standard (protein+)water system at 300K with PME ewald_rtol=1e-5
486 * needs 1.28 at rcoulomb=0.9 and 1.24 at rcoulomb=1.0 to get to nstlist=40.
487 * Note that both CPU and GPU factors are conservative. Performance should
488 * not go down due to this tuning, except with a relatively slow GPU.
489 * On the other hand, at medium/high parallelization or with fast GPUs
490 * nstlist will not be increased enough to reach optimal performance.
492 /* CPU: pair-search is about a factor 1.5 slower than the non-bonded kernel */
493 static const float nbnxn_cpu_listfac_ok = 1.05;
494 static const float nbnxn_cpu_listfac_max = 1.09;
495 /* GPU: pair-search is a factor 1.5-3 slower than the non-bonded kernel */
496 static const float nbnxn_gpu_listfac_ok = 1.20;
497 static const float nbnxn_gpu_listfac_max = 1.30;
499 /* Try to increase nstlist when using the Verlet cut-off scheme */
500 static void increase_nstlist(FILE *fp, t_commrec *cr,
501 t_inputrec *ir, int nstlist_cmdline,
502 const gmx_mtop_t *mtop, matrix box,
505 float listfac_ok, listfac_max;
506 int nstlist_orig, nstlist_prev;
507 verletbuf_list_setup_t ls;
508 real rlistWithReferenceNstlist, rlist_inc, rlist_ok, rlist_max;
509 real rlist_new, rlist_prev;
510 size_t nstlist_ind = 0;
512 gmx_bool bBox, bDD, bCont;
513 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";
514 const char *nve_err = "Can not increase nstlist because an NVE ensemble is used";
515 const char *vbd_err = "Can not increase nstlist because verlet-buffer-tolerance is not set or used";
516 const char *box_err = "Can not increase nstlist because the box is too small";
517 const char *dd_err = "Can not increase nstlist because of domain decomposition limitations";
519 const float oneThird = 1.0f / 3.0f;
521 if (nstlist_cmdline <= 0)
523 if (ir->nstlist == 1)
525 /* The user probably set nstlist=1 for a reason,
526 * don't mess with the settings.
531 if (fp != NULL && bGPU && ir->nstlist < nstlist_try[0])
533 fprintf(fp, nstl_gpu, ir->nstlist);
536 while (nstlist_ind < NNSTL && ir->nstlist >= nstlist_try[nstlist_ind])
540 if (nstlist_ind == NNSTL)
542 /* There are no larger nstlist value to try */
547 if (EI_MD(ir->eI) && ir->etc == etcNO)
551 fprintf(stderr, "%s\n", nve_err);
555 fprintf(fp, "%s\n", nve_err);
561 if (ir->verletbuf_tol == 0 && bGPU)
563 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");
566 if (ir->verletbuf_tol < 0)
570 fprintf(stderr, "%s\n", vbd_err);
574 fprintf(fp, "%s\n", vbd_err);
582 listfac_ok = nbnxn_gpu_listfac_ok;
583 listfac_max = nbnxn_gpu_listfac_max;
587 listfac_ok = nbnxn_cpu_listfac_ok;
588 listfac_max = nbnxn_cpu_listfac_max;
591 nstlist_orig = ir->nstlist;
592 if (nstlist_cmdline > 0)
596 sprintf(buf, "Getting nstlist=%d from command line option",
599 ir->nstlist = nstlist_cmdline;
602 verletbuf_get_list_setup(bGPU, &ls);
604 /* Allow rlist to make the list a given factor larger than the list
605 * would be with the reference value for nstlist (10).
607 nstlist_prev = ir->nstlist;
608 ir->nstlist = nbnxnReferenceNstlist;
609 calc_verlet_buffer_size(mtop, det(box), ir, -1, &ls, NULL,
610 &rlistWithReferenceNstlist);
611 ir->nstlist = nstlist_prev;
613 /* Determine the pair list size increase due to zero interactions */
614 rlist_inc = nbnxn_get_rlist_effective_inc(ls.cluster_size_j,
615 mtop->natoms/det(box));
616 rlist_ok = (rlistWithReferenceNstlist + rlist_inc)*pow(listfac_ok, oneThird) - rlist_inc;
617 rlist_max = (rlistWithReferenceNstlist + rlist_inc)*pow(listfac_max, oneThird) - rlist_inc;
620 fprintf(debug, "nstlist tuning: rlist_inc %.3f rlist_ok %.3f rlist_max %.3f\n",
621 rlist_inc, rlist_ok, rlist_max);
624 nstlist_prev = nstlist_orig;
625 rlist_prev = ir->rlist;
628 if (nstlist_cmdline <= 0)
630 ir->nstlist = nstlist_try[nstlist_ind];
633 /* Set the pair-list buffer size in ir */
634 calc_verlet_buffer_size(mtop, det(box), ir, -1, &ls, NULL, &rlist_new);
636 /* Does rlist fit in the box? */
637 bBox = (sqr(rlist_new) < max_cutoff2(ir->ePBC, box));
639 if (bBox && DOMAINDECOMP(cr))
641 /* Check if rlist fits in the domain decomposition */
642 if (inputrec2nboundeddim(ir) < DIM)
644 gmx_incons("Changing nstlist with domain decomposition and unbounded dimensions is not implemented yet");
646 copy_mat(box, state_tmp.box);
647 bDD = change_dd_cutoff(cr, &state_tmp, ir, rlist_new);
652 fprintf(debug, "nstlist %d rlist %.3f bBox %d bDD %d\n",
653 ir->nstlist, rlist_new, bBox, bDD);
658 if (nstlist_cmdline <= 0)
660 if (bBox && bDD && rlist_new <= rlist_max)
662 /* Increase nstlist */
663 nstlist_prev = ir->nstlist;
664 rlist_prev = rlist_new;
665 bCont = (nstlist_ind+1 < NNSTL && rlist_new < rlist_ok);
669 /* Stick with the previous nstlist */
670 ir->nstlist = nstlist_prev;
671 rlist_new = rlist_prev;
683 gmx_warning(!bBox ? box_err : dd_err);
686 fprintf(fp, "\n%s\n", bBox ? box_err : dd_err);
688 ir->nstlist = nstlist_orig;
690 else if (ir->nstlist != nstlist_orig || rlist_new != ir->rlist)
692 sprintf(buf, "Changing nstlist from %d to %d, rlist from %g to %g",
693 nstlist_orig, ir->nstlist,
694 ir->rlist, rlist_new);
697 fprintf(stderr, "%s\n\n", buf);
701 fprintf(fp, "%s\n\n", buf);
703 ir->rlist = rlist_new;
704 ir->rlistlong = rlist_new;
708 static void prepare_verlet_scheme(FILE *fplog,
712 const gmx_mtop_t *mtop,
716 /* For NVE simulations, we will retain the initial list buffer */
717 if (ir->verletbuf_tol > 0 && !(EI_MD(ir->eI) && ir->etc == etcNO))
719 /* Update the Verlet buffer size for the current run setup */
720 verletbuf_list_setup_t ls;
723 /* Here we assume SIMD-enabled kernels are being used. But as currently
724 * calc_verlet_buffer_size gives the same results for 4x8 and 4x4
725 * and 4x2 gives a larger buffer than 4x4, this is ok.
727 verletbuf_get_list_setup(bUseGPU, &ls);
729 calc_verlet_buffer_size(mtop, det(box), ir, -1, &ls, NULL, &rlist_new);
731 if (rlist_new != ir->rlist)
735 fprintf(fplog, "\nChanging rlist from %g to %g for non-bonded %dx%d atom kernels\n\n",
736 ir->rlist, rlist_new,
737 ls.cluster_size_i, ls.cluster_size_j);
739 ir->rlist = rlist_new;
740 ir->rlistlong = rlist_new;
744 if (nstlist_cmdline > 0 && (!EI_DYNAMICS(ir->eI) || ir->verletbuf_tol <= 0))
746 gmx_fatal(FARGS, "Can not set nstlist without %s",
747 !EI_DYNAMICS(ir->eI) ? "dynamics" : "verlet-buffer-tolerance");
750 if (EI_DYNAMICS(ir->eI))
752 /* Set or try nstlist values */
753 increase_nstlist(fplog, cr, ir, nstlist_cmdline, mtop, box, bUseGPU);
757 static void print_hw_opt(FILE *fp, const gmx_hw_opt_t *hw_opt)
759 fprintf(fp, "hw_opt: nt %d ntmpi %d ntomp %d ntomp_pme %d gpu_id '%s'\n",
760 hw_opt->nthreads_tot,
761 hw_opt->nthreads_tmpi,
762 hw_opt->nthreads_omp,
763 hw_opt->nthreads_omp_pme,
764 hw_opt->gpu_opt.gpu_id != NULL ? hw_opt->gpu_opt.gpu_id : "");
767 /* Checks we can do when we don't (yet) know the cut-off scheme */
768 static void check_and_update_hw_opt_1(gmx_hw_opt_t *hw_opt,
769 gmx_bool bIsSimMaster)
771 gmx_omp_nthreads_read_env(&hw_opt->nthreads_omp, bIsSimMaster);
773 #ifndef GMX_THREAD_MPI
774 if (hw_opt->nthreads_tot > 0)
776 gmx_fatal(FARGS, "Setting the total number of threads is only supported with thread-MPI and GROMACS was compiled without thread-MPI");
778 if (hw_opt->nthreads_tmpi > 0)
780 gmx_fatal(FARGS, "Setting the number of thread-MPI threads is only supported with thread-MPI and GROMACS was compiled without thread-MPI");
785 if (hw_opt->nthreads_omp > 1)
787 gmx_fatal(FARGS, "More than 1 OpenMP thread requested, but GROMACS was compiled without OpenMP support");
789 hw_opt->nthreads_omp = 1;
792 if (hw_opt->nthreads_tot > 0 && hw_opt->nthreads_omp_pme <= 0)
794 /* We have the same number of OpenMP threads for PP and PME processes,
795 * thus we can perform several consistency checks.
797 if (hw_opt->nthreads_tmpi > 0 &&
798 hw_opt->nthreads_omp > 0 &&
799 hw_opt->nthreads_tot != hw_opt->nthreads_tmpi*hw_opt->nthreads_omp)
801 gmx_fatal(FARGS, "The total number of threads requested (%d) does not match the thread-MPI threads (%d) times the OpenMP threads (%d) requested",
802 hw_opt->nthreads_tot, hw_opt->nthreads_tmpi, hw_opt->nthreads_omp);
805 if (hw_opt->nthreads_tmpi > 0 &&
806 hw_opt->nthreads_tot % hw_opt->nthreads_tmpi != 0)
808 gmx_fatal(FARGS, "The total number of threads requested (%d) is not divisible by the number of thread-MPI threads requested (%d)",
809 hw_opt->nthreads_tot, hw_opt->nthreads_tmpi);
812 if (hw_opt->nthreads_omp > 0 &&
813 hw_opt->nthreads_tot % hw_opt->nthreads_omp != 0)
815 gmx_fatal(FARGS, "The total number of threads requested (%d) is not divisible by the number of OpenMP threads requested (%d)",
816 hw_opt->nthreads_tot, hw_opt->nthreads_omp);
819 if (hw_opt->nthreads_tmpi > 0 &&
820 hw_opt->nthreads_omp <= 0)
822 hw_opt->nthreads_omp = hw_opt->nthreads_tot/hw_opt->nthreads_tmpi;
827 if (hw_opt->nthreads_omp > 1)
829 gmx_fatal(FARGS, "OpenMP threads are requested, but GROMACS was compiled without OpenMP support");
833 if (hw_opt->nthreads_omp_pme > 0 && hw_opt->nthreads_omp <= 0)
835 gmx_fatal(FARGS, "You need to specify -ntomp in addition to -ntomp_pme");
838 if (hw_opt->nthreads_tot == 1)
840 hw_opt->nthreads_tmpi = 1;
842 if (hw_opt->nthreads_omp > 1)
844 gmx_fatal(FARGS, "You requested %d OpenMP threads with %d total threads",
845 hw_opt->nthreads_tmpi, hw_opt->nthreads_tot);
847 hw_opt->nthreads_omp = 1;
850 if (hw_opt->nthreads_omp_pme <= 0 && hw_opt->nthreads_omp > 0)
852 hw_opt->nthreads_omp_pme = hw_opt->nthreads_omp;
855 /* Parse GPU IDs, if provided.
856 * We check consistency with the tMPI thread count later.
858 gmx_parse_gpu_ids(&hw_opt->gpu_opt);
860 #ifdef GMX_THREAD_MPI
861 if (hw_opt->gpu_opt.n_dev_use > 0
863 hw_opt->nthreads_tmpi == 0)
865 /* Set the number of MPI threads equal to the number of GPUs */
866 hw_opt->nthreads_tmpi = hw_opt->gpu_opt.n_dev_use;
868 if (hw_opt->nthreads_tot > 0 &&
869 hw_opt->nthreads_tmpi > hw_opt->nthreads_tot)
871 /* We have more GPUs than total threads requested.
872 * We choose to (later) generate a mismatch error,
873 * instead of launching more threads than requested.
875 hw_opt->nthreads_tmpi = hw_opt->nthreads_tot;
882 print_hw_opt(debug, hw_opt);
886 /* Checks we can do when we know the cut-off scheme */
887 static void check_and_update_hw_opt_2(gmx_hw_opt_t *hw_opt,
890 if (cutoff_scheme == ecutsGROUP)
892 /* We only have OpenMP support for PME only nodes */
893 if (hw_opt->nthreads_omp > 1)
895 gmx_fatal(FARGS, "OpenMP threads have been requested with cut-off scheme %s, but these are only supported with cut-off scheme %s",
896 ecutscheme_names[cutoff_scheme],
897 ecutscheme_names[ecutsVERLET]);
899 hw_opt->nthreads_omp = 1;
902 if (hw_opt->nthreads_omp_pme <= 0 && hw_opt->nthreads_omp > 0)
904 hw_opt->nthreads_omp_pme = hw_opt->nthreads_omp;
909 print_hw_opt(debug, hw_opt);
914 /* Override the value in inputrec with value passed on the command line (if any) */
915 static void override_nsteps_cmdline(FILE *fplog,
916 gmx_int64_t nsteps_cmdline,
923 /* override with anything else than the default -2 */
924 if (nsteps_cmdline > -2)
926 char sbuf_steps[STEPSTRSIZE];
927 char sbuf_msg[STRLEN];
929 ir->nsteps = nsteps_cmdline;
930 if (EI_DYNAMICS(ir->eI) && nsteps_cmdline != -1)
932 sprintf(sbuf_msg, "Overriding nsteps with value passed on the command line: %s steps, %.3g ps",
933 gmx_step_str(nsteps_cmdline, sbuf_steps),
934 fabs(nsteps_cmdline*ir->delta_t));
938 sprintf(sbuf_msg, "Overriding nsteps with value passed on the command line: %s steps",
939 gmx_step_str(nsteps_cmdline, sbuf_steps));
942 md_print_warn(cr, fplog, "%s\n", sbuf_msg);
944 else if (nsteps_cmdline < -2)
946 gmx_fatal(FARGS, "Invalid nsteps value passed on the command line: %d",
949 /* Do nothing if nsteps_cmdline == -2 */
952 int mdrunner(gmx_hw_opt_t *hw_opt,
953 FILE *fplog, t_commrec *cr, int nfile,
954 const t_filenm fnm[], const output_env_t oenv, gmx_bool bVerbose,
955 gmx_bool bCompact, int nstglobalcomm,
956 ivec ddxyz, int dd_node_order, real rdd, real rconstr,
957 const char *dddlb_opt, real dlb_scale,
958 const char *ddcsx, const char *ddcsy, const char *ddcsz,
959 const char *nbpu_opt, int nstlist_cmdline,
960 gmx_int64_t nsteps_cmdline, int nstepout, int resetstep,
961 int gmx_unused nmultisim, int repl_ex_nst, int repl_ex_nex,
962 int repl_ex_seed, real pforce, real cpt_period, real max_hours,
963 int imdport, unsigned long Flags)
965 gmx_bool bForceUseGPU, bTryUseGPU, bRerunMD, bCantUseGPU;
966 t_inputrec *inputrec;
967 t_state *state = NULL;
969 gmx_ddbox_t ddbox = {0};
970 int npme_major, npme_minor;
972 gmx_mtop_t *mtop = NULL;
973 t_mdatoms *mdatoms = NULL;
974 t_forcerec *fr = NULL;
975 t_fcdata *fcd = NULL;
976 real ewaldcoeff_q = 0;
977 real ewaldcoeff_lj = 0;
978 struct gmx_pme_t **pmedata = NULL;
979 gmx_vsite_t *vsite = NULL;
981 int nChargePerturbed = -1, nTypePerturbed = 0, status;
982 gmx_wallcycle_t wcycle;
984 gmx_walltime_accounting_t walltime_accounting = NULL;
986 gmx_int64_t reset_counters;
987 gmx_edsam_t ed = NULL;
988 int nthreads_pme = 1;
990 gmx_membed_t membed = NULL;
991 gmx_hw_info_t *hwinfo = NULL;
992 /* The master rank decides early on bUseGPU and broadcasts this later */
993 gmx_bool bUseGPU = FALSE;
995 /* CAUTION: threads may be started later on in this function, so
996 cr doesn't reflect the final parallel state right now */
1000 if (Flags & MD_APPENDFILES)
1005 bRerunMD = (Flags & MD_RERUN);
1006 bForceUseGPU = (strncmp(nbpu_opt, "gpu", 3) == 0);
1007 bTryUseGPU = (strncmp(nbpu_opt, "auto", 4) == 0) || bForceUseGPU;
1008 /* Rerun execution time is dominated by I/O and pair search, so
1009 * GPUs are not very useful, plus they do not support more than
1010 * one energy group. Don't select them when they can't be used,
1011 * unless the user requested it, then fatal_error is called later.
1013 * TODO it would be nice to notify the user that if this check
1014 * causes GPUs not to be used that this is what is happening, and
1015 * why, but that will be easier to do after some future
1017 bCantUseGPU = bRerunMD && (inputrec->opts.ngener > 1);
1018 bTryUseGPU = bTryUseGPU && !(bCantUseGPU && !bForceUseGPU);
1020 /* Detect hardware, gather information. This is an operation that is
1021 * global for this process (MPI rank). */
1022 hwinfo = gmx_detect_hardware(fplog, cr, bTryUseGPU);
1024 gmx_print_detected_hardware(fplog, cr, hwinfo);
1028 /* Print references after all software/hardware printing */
1029 please_cite(fplog, "Hess2008b");
1030 please_cite(fplog, "Spoel2005a");
1031 please_cite(fplog, "Lindahl2001a");
1032 please_cite(fplog, "Berendsen95a");
1038 /* Read (nearly) all data required for the simulation */
1039 read_tpx_state(ftp2fn(efTPR, nfile, fnm), inputrec, state, NULL, mtop);
1041 if (inputrec->cutoff_scheme == ecutsVERLET)
1043 /* Here the master rank decides if all ranks will use GPUs */
1044 bUseGPU = (hwinfo->gpu_info.n_dev_compatible > 0 ||
1045 getenv("GMX_EMULATE_GPU") != NULL);
1047 /* TODO add GPU kernels for this and replace this check by:
1048 * (bUseGPU && (ir->vdwtype == evdwPME &&
1049 * ir->ljpme_combination_rule == eljpmeLB))
1050 * update the message text and the content of nbnxn_acceleration_supported.
1053 !nbnxn_acceleration_supported(fplog, cr, inputrec, bUseGPU))
1055 /* Fallback message printed by nbnxn_acceleration_supported */
1058 gmx_fatal(FARGS, "GPU acceleration requested, but not supported with the given input settings");
1063 prepare_verlet_scheme(fplog, cr,
1064 inputrec, nstlist_cmdline, mtop, state->box,
1069 if (nstlist_cmdline > 0)
1071 gmx_fatal(FARGS, "Can not set nstlist with the group cut-off scheme");
1074 if (hwinfo->gpu_info.n_dev_compatible > 0)
1076 md_print_warn(cr, fplog,
1077 "NOTE: GPU(s) found, but the current simulation can not use GPUs\n"
1078 " To use a GPU, set the mdp option: cutoff-scheme = Verlet\n");
1083 gmx_fatal(FARGS, "GPU requested, but can't be used without cutoff-scheme=Verlet");
1086 #ifdef GMX_TARGET_BGQ
1087 md_print_warn(cr, fplog,
1088 "NOTE: There is no SIMD implementation of the group scheme kernels on\n"
1089 " BlueGene/Q. You will observe better performance from using the\n"
1090 " Verlet cut-off scheme.\n");
1094 if (inputrec->eI == eiSD2)
1096 md_print_warn(cr, fplog, "The stochastic dynamics integrator %s is deprecated, since\n"
1097 "it is slower than integrator %s and is slightly less accurate\n"
1098 "with constraints. Use the %s integrator.",
1099 ei_names[inputrec->eI], ei_names[eiSD1], ei_names[eiSD1]);
1103 /* Check and update the hardware options for internal consistency */
1104 check_and_update_hw_opt_1(hw_opt, SIMMASTER(cr));
1106 /* Early check for externally set process affinity. */
1107 gmx_check_thread_affinity_set(fplog, cr,
1108 hw_opt, hwinfo->nthreads_hw_avail, FALSE);
1112 #ifdef GMX_THREAD_MPI
1113 if (cr->npmenodes > 0 && hw_opt->nthreads_tmpi <= 0)
1115 gmx_fatal(FARGS, "You need to explicitly specify the number of MPI threads (-ntmpi) when using separate PME ranks");
1119 if (hw_opt->nthreads_omp_pme != hw_opt->nthreads_omp &&
1122 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");
1126 #ifdef GMX_THREAD_MPI
1129 /* Since the master knows the cut-off scheme, update hw_opt for this.
1130 * This is done later for normal MPI and also once more with tMPI
1131 * for all tMPI ranks.
1133 check_and_update_hw_opt_2(hw_opt, inputrec->cutoff_scheme);
1135 /* NOW the threads will be started: */
1136 hw_opt->nthreads_tmpi = get_nthreads_mpi(hwinfo,
1140 if (hw_opt->nthreads_tot > 0 && hw_opt->nthreads_omp <= 0)
1142 hw_opt->nthreads_omp = hw_opt->nthreads_tot/hw_opt->nthreads_tmpi;
1145 if (hw_opt->nthreads_tmpi > 1)
1147 t_commrec *cr_old = cr;
1148 /* now start the threads. */
1149 cr = mdrunner_start_threads(hw_opt, fplog, cr_old, nfile, fnm,
1150 oenv, bVerbose, bCompact, nstglobalcomm,
1151 ddxyz, dd_node_order, rdd, rconstr,
1152 dddlb_opt, dlb_scale, ddcsx, ddcsy, ddcsz,
1153 nbpu_opt, nstlist_cmdline,
1154 nsteps_cmdline, nstepout, resetstep, nmultisim,
1155 repl_ex_nst, repl_ex_nex, repl_ex_seed, pforce,
1156 cpt_period, max_hours,
1158 /* the main thread continues here with a new cr. We don't deallocate
1159 the old cr because other threads may still be reading it. */
1162 gmx_comm("Failed to spawn threads");
1167 /* END OF CAUTION: cr is now reliable */
1169 /* g_membed initialisation *
1170 * Because we change the mtop, init_membed is called before the init_parallel *
1171 * (in case we ever want to make it run in parallel) */
1172 if (opt2bSet("-membed", nfile, fnm))
1176 fprintf(stderr, "Initializing membed");
1178 membed = init_membed(fplog, nfile, fnm, mtop, inputrec, state, cr, &cpt_period);
1183 /* now broadcast everything to the non-master nodes/threads: */
1184 init_parallel(cr, inputrec, mtop);
1186 /* The master rank decided on the use of GPUs,
1187 * broadcast this information to all ranks.
1189 gmx_bcast_sim(sizeof(bUseGPU), &bUseGPU, cr);
1194 pr_inputrec(fplog, 0, "Input Parameters", inputrec, FALSE);
1195 fprintf(fplog, "\n");
1198 /* now make sure the state is initialized and propagated */
1199 set_state_entries(state, inputrec);
1201 /* A parallel command line option consistency check that we can
1202 only do after any threads have started. */
1204 (ddxyz[XX] > 1 || ddxyz[YY] > 1 || ddxyz[ZZ] > 1 || cr->npmenodes > 0))
1207 "The -dd or -npme option request a parallel simulation, "
1209 "but %s was compiled without threads or MPI enabled"
1211 #ifdef GMX_THREAD_MPI
1212 "but the number of threads (option -nt) is 1"
1214 "but %s was not started through mpirun/mpiexec or only one rank was requested through mpirun/mpiexec"
1217 , output_env_get_program_display_name(oenv)
1222 (EI_ENERGY_MINIMIZATION(inputrec->eI) || eiNM == inputrec->eI))
1224 gmx_fatal(FARGS, "The .mdp file specified an energy mininization or normal mode algorithm, and these are not compatible with mdrun -rerun");
1227 if (can_use_allvsall(inputrec, TRUE, cr, fplog) && DOMAINDECOMP(cr))
1229 gmx_fatal(FARGS, "All-vs-all loops do not work with domain decomposition, use a single MPI rank");
1232 if (!(EEL_PME(inputrec->coulombtype) || EVDW_PME(inputrec->vdwtype)))
1234 if (cr->npmenodes > 0)
1236 gmx_fatal_collective(FARGS, cr, NULL,
1237 "PME-only ranks are requested, but the system does not use PME for electrostatics or LJ");
1243 if (bUseGPU && cr->npmenodes < 0)
1245 /* With GPUs we don't automatically use PME-only ranks. PME ranks can
1246 * improve performance with many threads per GPU, since our OpenMP
1247 * scaling is bad, but it's difficult to automate the setup.
1255 fcRegisterSteps(inputrec->nsteps, inputrec->init_step);
1259 /* NMR restraints must be initialized before load_checkpoint,
1260 * since with time averaging the history is added to t_state.
1261 * For proper consistency check we therefore need to extend
1263 * So the PME-only nodes (if present) will also initialize
1264 * the distance restraints.
1268 /* This needs to be called before read_checkpoint to extend the state */
1269 init_disres(fplog, mtop, inputrec, cr, fcd, state, repl_ex_nst > 0);
1271 init_orires(fplog, mtop, state->x, inputrec, cr, &(fcd->orires),
1274 if (DEFORM(*inputrec))
1276 /* Store the deform reference box before reading the checkpoint */
1279 copy_mat(state->box, box);
1283 gmx_bcast(sizeof(box), box, cr);
1285 /* Because we do not have the update struct available yet
1286 * in which the reference values should be stored,
1287 * we store them temporarily in static variables.
1288 * This should be thread safe, since they are only written once
1289 * and with identical values.
1291 tMPI_Thread_mutex_lock(&deform_init_box_mutex);
1292 deform_init_init_step_tpx = inputrec->init_step;
1293 copy_mat(box, deform_init_box_tpx);
1294 tMPI_Thread_mutex_unlock(&deform_init_box_mutex);
1297 if (opt2bSet("-cpi", nfile, fnm))
1299 /* Check if checkpoint file exists before doing continuation.
1300 * This way we can use identical input options for the first and subsequent runs...
1302 if (gmx_fexist_master(opt2fn_master("-cpi", nfile, fnm, cr), cr) )
1304 load_checkpoint(opt2fn_master("-cpi", nfile, fnm, cr), &fplog,
1306 inputrec, state, &bReadEkin,
1307 (Flags & MD_APPENDFILES),
1308 (Flags & MD_APPENDFILESSET));
1312 Flags |= MD_READ_EKIN;
1317 if (MASTER(cr) && (Flags & MD_APPENDFILES))
1319 gmx_log_open(ftp2fn(efLOG, nfile, fnm), cr,
1323 /* override nsteps with value from cmdline */
1324 override_nsteps_cmdline(fplog, nsteps_cmdline, inputrec, cr);
1328 copy_mat(state->box, box);
1333 gmx_bcast(sizeof(box), box, cr);
1336 /* Essential dynamics */
1337 if (opt2bSet("-ei", nfile, fnm))
1339 /* Open input and output files, allocate space for ED data structure */
1340 ed = ed_open(mtop->natoms, &state->edsamstate, nfile, fnm, Flags, oenv, cr);
1343 if (PAR(cr) && !(EI_TPI(inputrec->eI) ||
1344 inputrec->eI == eiNM))
1346 cr->dd = init_domain_decomposition(fplog, cr, Flags, ddxyz, rdd, rconstr,
1347 dddlb_opt, dlb_scale,
1348 ddcsx, ddcsy, ddcsz,
1351 &ddbox, &npme_major, &npme_minor);
1353 make_dd_communicators(fplog, cr, dd_node_order);
1355 /* Set overallocation to avoid frequent reallocation of arrays */
1356 set_over_alloc_dd(TRUE);
1360 /* PME, if used, is done on all nodes with 1D decomposition */
1362 cr->duty = (DUTY_PP | DUTY_PME);
1366 if (inputrec->ePBC == epbcSCREW)
1369 "pbc=%s is only implemented with domain decomposition",
1370 epbc_names[inputrec->ePBC]);
1376 /* After possible communicator splitting in make_dd_communicators.
1377 * we can set up the intra/inter node communication.
1379 gmx_setup_nodecomm(fplog, cr);
1382 /* Initialize per-physical-node MPI process/thread ID and counters. */
1383 gmx_init_intranode_counters(cr);
1387 md_print_info(cr, fplog,
1388 "This is simulation %d out of %d running as a composite GROMACS\n"
1389 "multi-simulation job. Setup for this simulation:\n\n",
1390 cr->ms->sim, cr->ms->nsim);
1392 md_print_info(cr, fplog, "Using %d MPI %s\n",
1394 #ifdef GMX_THREAD_MPI
1395 cr->nnodes == 1 ? "thread" : "threads"
1397 cr->nnodes == 1 ? "process" : "processes"
1403 /* Check and update hw_opt for the cut-off scheme */
1404 check_and_update_hw_opt_2(hw_opt, inputrec->cutoff_scheme);
1406 gmx_omp_nthreads_init(fplog, cr,
1407 hwinfo->nthreads_hw_avail,
1408 hw_opt->nthreads_omp,
1409 hw_opt->nthreads_omp_pme,
1410 (cr->duty & DUTY_PP) == 0,
1411 inputrec->cutoff_scheme == ecutsVERLET);
1414 if (integrator[inputrec->eI].func != do_tpi &&
1415 inputrec->cutoff_scheme == ecutsVERLET)
1417 gmx_feenableexcept();
1423 /* Select GPU id's to use */
1424 gmx_select_gpu_ids(fplog, cr, &hwinfo->gpu_info, bForceUseGPU,
1429 /* Ignore (potentially) manually selected GPUs */
1430 hw_opt->gpu_opt.n_dev_use = 0;
1433 /* check consistency across ranks of things like SIMD
1434 * support and number of GPUs selected */
1435 gmx_check_hw_runconf_consistency(fplog, hwinfo, cr, hw_opt, bUseGPU);
1437 if (DOMAINDECOMP(cr))
1439 /* When we share GPUs over ranks, we need to know this for the DLB */
1440 dd_setup_dlb_resource_sharing(cr, hwinfo, hw_opt);
1443 /* getting number of PP/PME threads
1444 PME: env variable should be read only on one node to make sure it is
1445 identical everywhere;
1447 /* TODO nthreads_pp is only used for pinning threads.
1448 * This is a temporary solution until we have a hw topology library.
1450 nthreads_pp = gmx_omp_nthreads_get(emntNonbonded);
1451 nthreads_pme = gmx_omp_nthreads_get(emntPME);
1453 wcycle = wallcycle_init(fplog, resetstep, cr, nthreads_pp, nthreads_pme);
1457 /* Master synchronizes its value of reset_counters with all nodes
1458 * including PME only nodes */
1459 reset_counters = wcycle_get_reset_counters(wcycle);
1460 gmx_bcast_sim(sizeof(reset_counters), &reset_counters, cr);
1461 wcycle_set_reset_counters(wcycle, reset_counters);
1465 if (cr->duty & DUTY_PP)
1467 bcast_state(cr, state);
1469 /* Initiate forcerecord */
1471 fr->hwinfo = hwinfo;
1472 fr->gpu_opt = &hw_opt->gpu_opt;
1473 init_forcerec(fplog, oenv, fr, fcd, inputrec, mtop, cr, box,
1474 opt2fn("-table", nfile, fnm),
1475 opt2fn("-tabletf", nfile, fnm),
1476 opt2fn("-tablep", nfile, fnm),
1477 opt2fn("-tableb", nfile, fnm),
1482 /* version for PCA_NOT_READ_NODE (see md.c) */
1483 /*init_forcerec(fplog,fr,fcd,inputrec,mtop,cr,box,FALSE,
1484 "nofile","nofile","nofile","nofile",FALSE,pforce);
1487 /* Initialize QM-MM */
1490 init_QMMMrec(cr, mtop, inputrec, fr);
1493 /* Initialize the mdatoms structure.
1494 * mdatoms is not filled with atom data,
1495 * as this can not be done now with domain decomposition.
1497 mdatoms = init_mdatoms(fplog, mtop, inputrec->efep != efepNO);
1499 /* Initialize the virtual site communication */
1500 vsite = init_vsite(mtop, cr, FALSE);
1502 calc_shifts(box, fr->shift_vec);
1504 /* With periodic molecules the charge groups should be whole at start up
1505 * and the virtual sites should not be far from their proper positions.
1507 if (!inputrec->bContinuation && MASTER(cr) &&
1508 !(inputrec->ePBC != epbcNONE && inputrec->bPeriodicMols))
1510 /* Make molecules whole at start of run */
1511 if (fr->ePBC != epbcNONE)
1513 do_pbc_first_mtop(fplog, inputrec->ePBC, box, mtop, state->x);
1517 /* Correct initial vsite positions are required
1518 * for the initial distribution in the domain decomposition
1519 * and for the initial shell prediction.
1521 construct_vsites_mtop(vsite, mtop, state->x);
1525 if (EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype))
1527 ewaldcoeff_q = fr->ewaldcoeff_q;
1528 ewaldcoeff_lj = fr->ewaldcoeff_lj;
1529 pmedata = &fr->pmedata;
1538 /* This is a PME only node */
1540 /* We don't need the state */
1543 ewaldcoeff_q = calc_ewaldcoeff_q(inputrec->rcoulomb, inputrec->ewald_rtol);
1544 ewaldcoeff_lj = calc_ewaldcoeff_lj(inputrec->rvdw, inputrec->ewald_rtol_lj);
1548 if (hw_opt->thread_affinity != threadaffOFF)
1550 /* Before setting affinity, check whether the affinity has changed
1551 * - which indicates that probably the OpenMP library has changed it
1552 * since we first checked).
1554 gmx_check_thread_affinity_set(fplog, cr,
1555 hw_opt, hwinfo->nthreads_hw_avail, TRUE);
1557 /* Set the CPU affinity */
1558 gmx_set_thread_affinity(fplog, cr, hw_opt, hwinfo);
1561 /* Initiate PME if necessary,
1562 * either on all nodes or on dedicated PME nodes only. */
1563 if (EEL_PME(inputrec->coulombtype) || EVDW_PME(inputrec->vdwtype))
1567 nChargePerturbed = mdatoms->nChargePerturbed;
1568 if (EVDW_PME(inputrec->vdwtype))
1570 nTypePerturbed = mdatoms->nTypePerturbed;
1573 if (cr->npmenodes > 0)
1575 /* The PME only nodes need to know nChargePerturbed(FEP on Q) and nTypePerturbed(FEP on LJ)*/
1576 gmx_bcast_sim(sizeof(nChargePerturbed), &nChargePerturbed, cr);
1577 gmx_bcast_sim(sizeof(nTypePerturbed), &nTypePerturbed, cr);
1580 if (cr->duty & DUTY_PME)
1582 status = gmx_pme_init(pmedata, cr, npme_major, npme_minor, inputrec,
1583 mtop ? mtop->natoms : 0, nChargePerturbed, nTypePerturbed,
1584 (Flags & MD_REPRODUCIBLE), nthreads_pme);
1587 gmx_fatal(FARGS, "Error %d initializing PME", status);
1593 if (integrator[inputrec->eI].func == do_md)
1595 /* Turn on signal handling on all nodes */
1597 * (A user signal from the PME nodes (if any)
1598 * is communicated to the PP nodes.
1600 signal_handler_install();
1603 if (cr->duty & DUTY_PP)
1605 /* Assumes uniform use of the number of OpenMP threads */
1606 walltime_accounting = walltime_accounting_init(gmx_omp_nthreads_get(emntDefault));
1608 if (inputrec->bPull)
1610 /* Initialize pull code */
1611 init_pull(fplog, inputrec, nfile, fnm, mtop, cr, oenv, inputrec->fepvals->init_lambda,
1612 EI_DYNAMICS(inputrec->eI) && MASTER(cr), Flags);
1617 /* Initialize enforced rotation code */
1618 init_rot(fplog, inputrec, nfile, fnm, cr, state->x, box, mtop, oenv,
1622 if (inputrec->eSwapCoords != eswapNO)
1624 /* Initialize ion swapping code */
1625 init_swapcoords(fplog, bVerbose, inputrec, opt2fn_master("-swap", nfile, fnm, cr),
1626 mtop, state->x, state->box, &state->swapstate, cr, oenv, Flags);
1629 constr = init_constraints(fplog, mtop, inputrec, ed, state, cr);
1631 if (DOMAINDECOMP(cr))
1633 GMX_RELEASE_ASSERT(fr, "fr was NULL while cr->duty was DUTY_PP");
1634 dd_init_bondeds(fplog, cr->dd, mtop, vsite, inputrec,
1635 Flags & MD_DDBONDCHECK, fr->cginfo_mb);
1637 set_dd_parameters(fplog, cr->dd, dlb_scale, inputrec, &ddbox);
1639 setup_dd_grid(fplog, cr->dd);
1642 /* Now do whatever the user wants us to do (how flexible...) */
1643 integrator[inputrec->eI].func(fplog, cr, nfile, fnm,
1644 oenv, bVerbose, bCompact,
1647 nstepout, inputrec, mtop,
1649 mdatoms, nrnb, wcycle, ed, fr,
1650 repl_ex_nst, repl_ex_nex, repl_ex_seed,
1652 cpt_period, max_hours,
1655 walltime_accounting);
1657 if (inputrec->bPull)
1659 finish_pull(inputrec->pull);
1664 finish_rot(inputrec->rot);
1670 GMX_RELEASE_ASSERT(pmedata, "pmedata was NULL while cr->duty was not DUTY_PP");
1672 walltime_accounting = walltime_accounting_init(gmx_omp_nthreads_get(emntPME));
1673 gmx_pmeonly(*pmedata, cr, nrnb, wcycle, walltime_accounting, ewaldcoeff_q, ewaldcoeff_lj, inputrec);
1676 wallcycle_stop(wcycle, ewcRUN);
1678 /* Finish up, write some stuff
1679 * if rerunMD, don't write last frame again
1681 finish_run(fplog, cr,
1682 inputrec, nrnb, wcycle, walltime_accounting,
1683 fr ? fr->nbv : NULL,
1684 EI_DYNAMICS(inputrec->eI) && !MULTISIM(cr));
1687 /* Free GPU memory and context */
1688 free_gpu_resources(fr, cr, &hwinfo->gpu_info, fr ? fr->gpu_opt : NULL);
1690 if (opt2bSet("-membed", nfile, fnm))
1695 gmx_hardware_info_free(hwinfo);
1697 /* Does what it says */
1698 print_date_and_time(fplog, cr->nodeid, "Finished mdrun", gmx_gettime());
1699 walltime_accounting_destroy(walltime_accounting);
1701 /* Close logfile already here if we were appending to it */
1702 if (MASTER(cr) && (Flags & MD_APPENDFILES))
1704 gmx_log_close(fplog);
1707 rc = (int)gmx_get_stop_condition();
1711 #ifdef GMX_THREAD_MPI
1712 /* we need to join all threads. The sub-threads join when they
1713 exit this function, but the master thread needs to be told to
1715 if (PAR(cr) && MASTER(cr))