Bug Summary

File:programs/mdrun/runner.c
Location:line 1729, column 53
Description:Access to field 'cginfo_mb' results in a dereference of a null pointer (loaded from variable 'fr')

Annotated Source Code

1/*
2 * This file is part of the GROMACS molecular simulation package.
3 *
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, 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.
10 *
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.
15 *
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.
20 *
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.
25 *
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.
33 *
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.
36 */
37#ifdef HAVE_CONFIG_H1
38#include <config.h>
39#endif
40
41#include <assert.h>
42#include <signal.h>
43#include <stdlib.h>
44#include <string.h>
45#ifdef HAVE_UNISTD_H
46#include <unistd.h>
47#endif
48
49#include "typedefs.h"
50#include "copyrite.h"
51#include "force.h"
52#include "mdrun.h"
53#include "md_logging.h"
54#include "md_support.h"
55#include "network.h"
56#include "names.h"
57#include "disre.h"
58#include "orires.h"
59#include "pme.h"
60#include "mdatoms.h"
61#include "repl_ex.h"
62#include "deform.h"
63#include "qmmm.h"
64#include "domdec.h"
65#include "coulomb.h"
66#include "constr.h"
67#include "mvdata.h"
68#include "checkpoint.h"
69#include "mtop_util.h"
70#include "sighandler.h"
71#include "txtdump.h"
72#include "gmx_detect_hardware.h"
73#include "gmx_omp_nthreads.h"
74#include "gromacs/gmxpreprocess/calc_verletbuf.h"
75#include "membed.h"
76#include "macros.h"
77#include "gmx_thread_affinity.h"
78#include "inputrec.h"
79#include "main.h"
80
81#include "gromacs/essentialdynamics/edsam.h"
82#include "gromacs/fileio/tpxio.h"
83#include "gromacs/math/vec.h"
84#include "gromacs/mdlib/nbnxn_search.h"
85#include "gromacs/mdlib/nbnxn_consts.h"
86#include "gromacs/pulling/pull.h"
87#include "gromacs/pulling/pull_rotation.h"
88#include "gromacs/swap/swapcoords.h"
89#include "gromacs/timing/wallcycle.h"
90#include "gromacs/utility/gmxmpi.h"
91#include "gromacs/utility/smalloc.h"
92
93#ifdef GMX_FAHCORE
94#include "corewrap.h"
95#endif
96
97#include "gpu_utils.h"
98#include "nbnxn_cuda_data_mgmt.h"
99
100typedef struct {
101 gmx_integrator_t *func;
102} gmx_intp_t;
103
104/* The array should match the eI array in include/types/enums.h */
105const 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
107gmx_int64_t deform_init_init_step_tpx;
108matrix deform_init_box_tpx;
109tMPI_Thread_mutex_t deform_init_box_mutex = TMPI_THREAD_MUTEX_INITIALIZER{ {0}, ((void*)0) };
110
111
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.
115 */
116#define MIN_ATOMS_PER_MPI_THREAD90 90
117#define MIN_ATOMS_PER_GPU900 900
118
119struct mdrunner_arglist
120{
121 gmx_hw_opt_t hw_opt;
122 FILE *fplog;
123 t_commrec *cr;
124 int nfile;
125 const t_filenm *fnm;
126 output_env_t oenv;
127 gmx_bool bVerbose;
128 gmx_bool bCompact;
129 int nstglobalcomm;
130 ivec ddxyz;
131 int dd_node_order;
132 real rdd;
133 real rconstr;
134 const char *dddlb_opt;
135 real dlb_scale;
136 const char *ddcsx;
137 const char *ddcsy;
138 const char *ddcsz;
139 const char *nbpu_opt;
140 int nstlist_cmdline;
141 gmx_int64_t nsteps_cmdline;
142 int nstepout;
143 int resetstep;
144 int nmultisim;
145 int repl_ex_nst;
146 int repl_ex_nex;
147 int repl_ex_seed;
148 real pforce;
149 real cpt_period;
150 real max_hours;
151 const char *deviceOptions;
152 int imdport;
153 unsigned long Flags;
154};
155
156
157/* The function used for spawning threads. Extracts the mdrunner()
158 arguments from its one argument and calls mdrunner(), after making
159 a commrec. */
160static void mdrunner_start_fn(void *arg)
161{
162 struct mdrunner_arglist *mda = (struct mdrunner_arglist*)arg;
163 struct mdrunner_arglist mc = *mda; /* copy the arg list to make sure
164 that it's thread-local. This doesn't
165 copy pointed-to items, of course,
166 but those are all const. */
167 t_commrec *cr; /* we need a local version of this */
168 FILE *fplog = NULL((void*)0);
169 t_filenm *fnm;
170
171 fnm = dup_tfn(mc.nfile, mc.fnm);
172
173 cr = reinitialize_commrec_for_this_thread(mc.cr);
174
175 if (MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)))
176 {
177 fplog = mc.fplog;
178 }
179
180 mdrunner(&mc.hw_opt, fplog, cr, mc.nfile, fnm, mc.oenv,
181 mc.bVerbose, mc.bCompact, mc.nstglobalcomm,
182 mc.ddxyz, mc.dd_node_order, mc.rdd,
183 mc.rconstr, mc.dddlb_opt, mc.dlb_scale,
184 mc.ddcsx, mc.ddcsy, mc.ddcsz,
185 mc.nbpu_opt, mc.nstlist_cmdline,
186 mc.nsteps_cmdline, mc.nstepout, mc.resetstep,
187 mc.nmultisim, mc.repl_ex_nst, mc.repl_ex_nex, mc.repl_ex_seed, mc.pforce,
188 mc.cpt_period, mc.max_hours, mc.deviceOptions, mc.imdport, mc.Flags);
189}
190
191/* called by mdrunner() to start a specific number of threads (including
192 the main thread) for thread-parallel runs. This in turn calls mdrunner()
193 for each thread.
194 All options besides nthreads are the same as for mdrunner(). */
195static t_commrec *mdrunner_start_threads(gmx_hw_opt_t *hw_opt,
196 FILE *fplog, t_commrec *cr, int nfile,
197 const t_filenm fnm[], const output_env_t oenv, gmx_bool bVerbose,
198 gmx_bool bCompact, int nstglobalcomm,
199 ivec ddxyz, int dd_node_order, real rdd, real rconstr,
200 const char *dddlb_opt, real dlb_scale,
201 const char *ddcsx, const char *ddcsy, const char *ddcsz,
202 const char *nbpu_opt, int nstlist_cmdline,
203 gmx_int64_t nsteps_cmdline,
204 int nstepout, int resetstep,
205 int nmultisim, int repl_ex_nst, int repl_ex_nex, int repl_ex_seed,
206 real pforce, real cpt_period, real max_hours,
207 const char *deviceOptions, unsigned long Flags)
208{
209 int ret;
210 struct mdrunner_arglist *mda;
211 t_commrec *crn; /* the new commrec */
212 t_filenm *fnmn;
213
214 /* first check whether we even need to start tMPI */
215 if (hw_opt->nthreads_tmpi < 2)
216 {
217 return cr;
218 }
219
220 /* a few small, one-time, almost unavoidable memory leaks: */
221 snew(mda, 1)(mda) = save_calloc("mda", "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 221, (1), sizeof(*(mda)));
222 fnmn = dup_tfn(nfile, fnm);
223
224 /* fill the data structure to pass as void pointer to thread start fn */
225 /* hw_opt contains pointers, which should all be NULL at this stage */
226 mda->hw_opt = *hw_opt;
227 mda->fplog = fplog;
228 mda->cr = cr;
229 mda->nfile = nfile;
230 mda->fnm = fnmn;
231 mda->oenv = oenv;
232 mda->bVerbose = bVerbose;
233 mda->bCompact = bCompact;
234 mda->nstglobalcomm = nstglobalcomm;
235 mda->ddxyz[XX0] = ddxyz[XX0];
236 mda->ddxyz[YY1] = ddxyz[YY1];
237 mda->ddxyz[ZZ2] = ddxyz[ZZ2];
238 mda->dd_node_order = dd_node_order;
239 mda->rdd = rdd;
240 mda->rconstr = rconstr;
241 mda->dddlb_opt = dddlb_opt;
242 mda->dlb_scale = dlb_scale;
243 mda->ddcsx = ddcsx;
244 mda->ddcsy = ddcsy;
245 mda->ddcsz = ddcsz;
246 mda->nbpu_opt = nbpu_opt;
247 mda->nstlist_cmdline = nstlist_cmdline;
248 mda->nsteps_cmdline = nsteps_cmdline;
249 mda->nstepout = nstepout;
250 mda->resetstep = resetstep;
251 mda->nmultisim = nmultisim;
252 mda->repl_ex_nst = repl_ex_nst;
253 mda->repl_ex_nex = repl_ex_nex;
254 mda->repl_ex_seed = repl_ex_seed;
255 mda->pforce = pforce;
256 mda->cpt_period = cpt_period;
257 mda->max_hours = max_hours;
258 mda->deviceOptions = deviceOptions;
259 mda->Flags = Flags;
260
261 /* now spawn new threads that start mdrunner_start_fn(), while
262 the main thread returns, we set thread affinity later */
263 ret = tMPI_Init_fn(TRUE1, hw_opt->nthreads_tmpi, TMPI_AFFINITY_NONE,
264 mdrunner_start_fn, (void*)(mda) );
265 if (ret != TMPI_SUCCESS)
266 {
267 return NULL((void*)0);
268 }
269
270 crn = reinitialize_commrec_for_this_thread(cr);
271 return crn;
272}
273
274
275static int get_tmpi_omp_thread_division(const gmx_hw_info_t *hwinfo,
276 const gmx_hw_opt_t *hw_opt,
277 int nthreads_tot,
278 int ngpu)
279{
280 int nthreads_tmpi;
281
282 /* There are no separate PME nodes here, as we ensured in
283 * check_and_update_hw_opt that nthreads_tmpi>0 with PME nodes
284 * and a conditional ensures we would not have ended up here.
285 * Note that separate PME nodes might be switched on later.
286 */
287 if (ngpu > 0)
288 {
289 nthreads_tmpi = ngpu;
290 if (nthreads_tot > 0 && nthreads_tot < nthreads_tmpi)
291 {
292 nthreads_tmpi = nthreads_tot;
293 }
294 }
295 else if (hw_opt->nthreads_omp > 0)
296 {
297 /* Here we could oversubscribe, when we do, we issue a warning later */
298 nthreads_tmpi = max(1, nthreads_tot/hw_opt->nthreads_omp)(((1) > (nthreads_tot/hw_opt->nthreads_omp)) ? (1) : (nthreads_tot
/hw_opt->nthreads_omp) );
299 }
300 else
301 {
302 /* TODO choose nthreads_omp based on hardware topology
303 when we have a hardware topology detection library */
304 /* In general, when running up to 4 threads, OpenMP should be faster.
305 * Note: on AMD Bulldozer we should avoid running OpenMP over two dies.
306 * On Intel>=Nehalem running OpenMP on a single CPU is always faster,
307 * even on two CPUs it's usually faster (but with many OpenMP threads
308 * it could be faster not to use HT, currently we always use HT).
309 * On Nehalem/Westmere we want to avoid running 16 threads over
310 * two CPUs with HT, so we need a limit<16; thus we use 12.
311 * A reasonable limit for Intel Sandy and Ivy bridge,
312 * not knowing the topology, is 16 threads.
313 */
314 const int nthreads_omp_always_faster = 4;
315 const int nthreads_omp_always_faster_Nehalem = 12;
316 const int nthreads_omp_always_faster_SandyBridge = 16;
317 const int first_model_Nehalem = 0x1A;
318 const int first_model_SandyBridge = 0x2A;
319 gmx_bool bIntel_Family6;
320
321 bIntel_Family6 =
322 (gmx_cpuid_vendor(hwinfo->cpuid_info) == GMX_CPUID_VENDOR_INTEL &&
323 gmx_cpuid_family(hwinfo->cpuid_info) == 6);
324
325 if (nthreads_tot <= nthreads_omp_always_faster ||
326 (bIntel_Family6 &&
327 ((gmx_cpuid_model(hwinfo->cpuid_info) >= nthreads_omp_always_faster_Nehalem && nthreads_tot <= nthreads_omp_always_faster_Nehalem) ||
328 (gmx_cpuid_model(hwinfo->cpuid_info) >= nthreads_omp_always_faster_SandyBridge && nthreads_tot <= nthreads_omp_always_faster_SandyBridge))))
329 {
330 /* Use pure OpenMP parallelization */
331 nthreads_tmpi = 1;
332 }
333 else
334 {
335 /* Don't use OpenMP parallelization */
336 nthreads_tmpi = nthreads_tot;
337 }
338 }
339
340 return nthreads_tmpi;
341}
342
343
344/* Get the number of threads to use for thread-MPI based on how many
345 * were requested, which algorithms we're using,
346 * and how many particles there are.
347 * At the point we have already called check_and_update_hw_opt.
348 * Thus all options should be internally consistent and consistent
349 * with the hardware, except that ntmpi could be larger than #GPU.
350 */
351static int get_nthreads_mpi(const gmx_hw_info_t *hwinfo,
352 gmx_hw_opt_t *hw_opt,
353 t_inputrec *inputrec, gmx_mtop_t *mtop,
354 const t_commrec *cr,
355 FILE *fplog)
356{
357 int nthreads_hw, nthreads_tot_max, nthreads_tmpi, nthreads_new, ngpu;
358 int min_atoms_per_mpi_thread;
359 char *env;
360 char sbuf[STRLEN4096];
361 gmx_bool bCanUseGPU;
362
363 if (hw_opt->nthreads_tmpi > 0)
364 {
365 /* Trivial, return right away */
366 return hw_opt->nthreads_tmpi;
367 }
368
369 nthreads_hw = hwinfo->nthreads_hw_avail;
370
371 /* How many total (#tMPI*#OpenMP) threads can we start? */
372 if (hw_opt->nthreads_tot > 0)
373 {
374 nthreads_tot_max = hw_opt->nthreads_tot;
375 }
376 else
377 {
378 nthreads_tot_max = nthreads_hw;
379 }
380
381 bCanUseGPU = (inputrec->cutoff_scheme == ecutsVERLET &&
382 hwinfo->gpu_info.ncuda_dev_compatible > 0);
383 if (bCanUseGPU)
384 {
385 ngpu = hwinfo->gpu_info.ncuda_dev_compatible;
386 }
387 else
388 {
389 ngpu = 0;
390 }
391
392 if (inputrec->cutoff_scheme == ecutsGROUP)
393 {
394 /* We checked this before, but it doesn't hurt to do it once more */
395 assert(hw_opt->nthreads_omp == 1)((void) (0));
396 }
397
398 nthreads_tmpi =
399 get_tmpi_omp_thread_division(hwinfo, hw_opt, nthreads_tot_max, ngpu);
400
401 if (inputrec->eI == eiNM || EI_TPI(inputrec->eI)((inputrec->eI) == eiTPI || (inputrec->eI) == eiTPIC))
402 {
403 /* Dims/steps are divided over the nodes iso splitting the atoms */
404 min_atoms_per_mpi_thread = 0;
405 }
406 else
407 {
408 if (bCanUseGPU)
409 {
410 min_atoms_per_mpi_thread = MIN_ATOMS_PER_GPU900;
411 }
412 else
413 {
414 min_atoms_per_mpi_thread = MIN_ATOMS_PER_MPI_THREAD90;
415 }
416 }
417
418 /* Check if an algorithm does not support parallel simulation. */
419 if (nthreads_tmpi != 1 &&
420 ( inputrec->eI == eiLBFGS ||
421 inputrec->coulombtype == eelEWALD ) )
422 {
423 nthreads_tmpi = 1;
424
425 md_print_warn(cr, fplog, "The integration or electrostatics algorithm doesn't support parallel runs. Using a single thread-MPI thread.\n");
426 if (hw_opt->nthreads_tmpi > nthreads_tmpi)
427 {
428 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 428, "You asked for more than 1 thread-MPI thread, but an algorithm doesn't support that");
429 }
430 }
431 else if (mtop->natoms/nthreads_tmpi < min_atoms_per_mpi_thread)
432 {
433 /* the thread number was chosen automatically, but there are too many
434 threads (too few atoms per thread) */
435 nthreads_new = max(1, mtop->natoms/min_atoms_per_mpi_thread)(((1) > (mtop->natoms/min_atoms_per_mpi_thread)) ? (1) :
(mtop->natoms/min_atoms_per_mpi_thread) );
436
437 /* Avoid partial use of Hyper-Threading */
438 if (gmx_cpuid_x86_smt(hwinfo->cpuid_info) == GMX_CPUID_X86_SMT_ENABLED &&
439 nthreads_new > nthreads_hw/2 && nthreads_new < nthreads_hw)
440 {
441 nthreads_new = nthreads_hw/2;
442 }
443
444 /* Avoid large prime numbers in the thread count */
445 if (nthreads_new >= 6)
446 {
447 /* Use only 6,8,10 with additional factors of 2 */
448 int fac;
449
450 fac = 2;
451 while (3*fac*2 <= nthreads_new)
452 {
453 fac *= 2;
454 }
455
456 nthreads_new = (nthreads_new/fac)*fac;
457 }
458 else
459 {
460 /* Avoid 5 */
461 if (nthreads_new == 5)
462 {
463 nthreads_new = 4;
464 }
465 }
466
467 nthreads_tmpi = nthreads_new;
468
469 fprintf(stderrstderr, "\n");
470 fprintf(stderrstderr, "NOTE: Parallelization is limited by the small number of atoms,\n");
471 fprintf(stderrstderr, " only starting %d thread-MPI threads.\n", nthreads_tmpi);
472 fprintf(stderrstderr, " You can use the -nt and/or -ntmpi option to optimize the number of threads.\n\n");
473 }
474
475 return nthreads_tmpi;
476}
477#endif /* GMX_THREAD_MPI */
478
479
480/* We determine the extra cost of the non-bonded kernels compared to
481 * a reference nstlist value of 10 (which is the default in grompp).
482 */
483static const int nbnxn_reference_nstlist = 10;
484/* The values to try when switching */
485const int nstlist_try[] = { 20, 25, 40 };
486#define NNSTLsizeof(nstlist_try)/sizeof(nstlist_try[0]) sizeof(nstlist_try)/sizeof(nstlist_try[0])
487/* Increase nstlist until the non-bonded cost increases more than listfac_ok,
488 * but never more than listfac_max.
489 * A standard (protein+)water system at 300K with PME ewald_rtol=1e-5
490 * needs 1.28 at rcoulomb=0.9 and 1.24 at rcoulomb=1.0 to get to nstlist=40.
491 * Note that both CPU and GPU factors are conservative. Performance should
492 * not go down due to this tuning, except with a relatively slow GPU.
493 * On the other hand, at medium/high parallelization or with fast GPUs
494 * nstlist will not be increased enough to reach optimal performance.
495 */
496/* CPU: pair-search is about a factor 1.5 slower than the non-bonded kernel */
497static const float nbnxn_cpu_listfac_ok = 1.05;
498static const float nbnxn_cpu_listfac_max = 1.09;
499/* GPU: pair-search is a factor 1.5-3 slower than the non-bonded kernel */
500static const float nbnxn_gpu_listfac_ok = 1.20;
501static const float nbnxn_gpu_listfac_max = 1.30;
502
503/* Try to increase nstlist when using the Verlet cut-off scheme */
504static void increase_nstlist(FILE *fp, t_commrec *cr,
505 t_inputrec *ir, int nstlist_cmdline,
506 const gmx_mtop_t *mtop, matrix box,
507 gmx_bool bGPU)
508{
509 float listfac_ok, listfac_max;
510 int nstlist_orig, nstlist_prev;
511 verletbuf_list_setup_t ls;
512 real rlist_nstlist10, rlist_inc, rlist_ok, rlist_max;
513 real rlist_new, rlist_prev;
514 int nstlist_ind = 0;
515 t_state state_tmp;
516 gmx_bool bBox, bDD, bCont;
517 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";
518 const char *nve_err = "Can not increase nstlist because an NVE ensemble is used";
519 const char *vbd_err = "Can not increase nstlist because verlet-buffer-tolerance is not set or used";
520 const char *box_err = "Can not increase nstlist because the box is too small";
521 const char *dd_err = "Can not increase nstlist because of domain decomposition limitations";
522 char buf[STRLEN4096];
523
524 if (nstlist_cmdline <= 0)
525 {
526 if (fp != NULL((void*)0) && bGPU && ir->nstlist < nstlist_try[0])
527 {
528 fprintf(fp, nstl_gpu, ir->nstlist);
529 }
530 nstlist_ind = 0;
531 while (nstlist_ind < NNSTLsizeof(nstlist_try)/sizeof(nstlist_try[0]) && ir->nstlist >= nstlist_try[nstlist_ind])
532 {
533 nstlist_ind++;
534 }
535 if (nstlist_ind == NNSTLsizeof(nstlist_try)/sizeof(nstlist_try[0]))
536 {
537 /* There are no larger nstlist value to try */
538 return;
539 }
540 }
541
542 if (EI_MD(ir->eI)((ir->eI) == eiMD || ((ir->eI) == eiVV || (ir->eI) ==
eiVVAK)) && ir->etc == etcNO)
543 {
544 if (MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)))
545 {
546 fprintf(stderrstderr, "%s\n", nve_err);
547 }
548 if (fp != NULL((void*)0))
549 {
550 fprintf(fp, "%s\n", nve_err);
551 }
552
553 return;
554 }
555
556 if (ir->verletbuf_tol == 0 && bGPU)
557 {
558 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 558, "You are using an old tpr file with a GPU, please generate a new tpr file with an up to date version of grompp");
559 }
560
561 if (ir->verletbuf_tol < 0)
562 {
563 if (MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)))
564 {
565 fprintf(stderrstderr, "%s\n", vbd_err);
566 }
567 if (fp != NULL((void*)0))
568 {
569 fprintf(fp, "%s\n", vbd_err);
570 }
571
572 return;
573 }
574
575 if (bGPU)
576 {
577 listfac_ok = nbnxn_gpu_listfac_ok;
578 listfac_max = nbnxn_gpu_listfac_max;
579 }
580 else
581 {
582 listfac_ok = nbnxn_cpu_listfac_ok;
583 listfac_max = nbnxn_cpu_listfac_max;
584 }
585
586 nstlist_orig = ir->nstlist;
587 if (nstlist_cmdline > 0)
588 {
589 if (fp)
590 {
591 sprintf(buf, "Getting nstlist=%d from command line option",
592 nstlist_cmdline);
593 }
594 ir->nstlist = nstlist_cmdline;
595 }
596
597 verletbuf_get_list_setup(bGPU, &ls);
598
599 /* Allow rlist to make the list a given factor larger than the list
600 * would be with nstlist=10.
601 */
602 nstlist_prev = ir->nstlist;
603 ir->nstlist = 10;
604 calc_verlet_buffer_size(mtop, det(box), ir, -1, &ls, NULL((void*)0),
605 &rlist_nstlist10);
606 ir->nstlist = nstlist_prev;
607
608 /* Determine the pair list size increase due to zero interactions */
609 rlist_inc = nbnxn_get_rlist_effective_inc(ls.cluster_size_j,
610 mtop->natoms/det(box));
611 rlist_ok = (rlist_nstlist10 + rlist_inc)*pow(listfac_ok, 1.0/3.0) - rlist_inc;
612 rlist_max = (rlist_nstlist10 + rlist_inc)*pow(listfac_max, 1.0/3.0) - rlist_inc;
613 if (debug)
614 {
615 fprintf(debug, "nstlist tuning: rlist_inc %.3f rlist_ok %.3f rlist_max %.3f\n",
616 rlist_inc, rlist_ok, rlist_max);
617 }
618
619 nstlist_prev = nstlist_orig;
620 rlist_prev = ir->rlist;
621 do
622 {
623 if (nstlist_cmdline <= 0)
624 {
625 ir->nstlist = nstlist_try[nstlist_ind];
626 }
627
628 /* Set the pair-list buffer size in ir */
629 calc_verlet_buffer_size(mtop, det(box), ir, -1, &ls, NULL((void*)0), &rlist_new);
630
631 /* Does rlist fit in the box? */
632 bBox = (sqr(rlist_new) < max_cutoff2(ir->ePBC, box));
633 bDD = TRUE1;
634 if (bBox && DOMAINDECOMP(cr)(((cr)->dd != ((void*)0)) && ((cr)->nnodes >
1)))
635 {
636 /* Check if rlist fits in the domain decomposition */
637 if (inputrec2nboundeddim(ir) < DIM3)
638 {
639 gmx_incons("Changing nstlist with domain decomposition and unbounded dimensions is not implemented yet")_gmx_error("incons", "Changing nstlist with domain decomposition and unbounded dimensions is not implemented yet"
, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c",
639);
640 }
641 copy_mat(box, state_tmp.box);
642 bDD = change_dd_cutoff(cr, &state_tmp, ir, rlist_new);
643 }
644
645 if (debug)
646 {
647 fprintf(debug, "nstlist %d rlist %.3f bBox %d bDD %d\n",
648 ir->nstlist, rlist_new, bBox, bDD);
649 }
650
651 bCont = FALSE0;
652
653 if (nstlist_cmdline <= 0)
654 {
655 if (bBox && bDD && rlist_new <= rlist_max)
656 {
657 /* Increase nstlist */
658 nstlist_prev = ir->nstlist;
659 rlist_prev = rlist_new;
660 bCont = (nstlist_ind+1 < NNSTLsizeof(nstlist_try)/sizeof(nstlist_try[0]) && rlist_new < rlist_ok);
661 }
662 else
663 {
664 /* Stick with the previous nstlist */
665 ir->nstlist = nstlist_prev;
666 rlist_new = rlist_prev;
667 bBox = TRUE1;
668 bDD = TRUE1;
669 }
670 }
671
672 nstlist_ind++;
673 }
674 while (bCont);
675
676 if (!bBox || !bDD)
677 {
678 gmx_warning(!bBox ? box_err : dd_err);
679 if (fp != NULL((void*)0))
680 {
681 fprintf(fp, "\n%s\n", bBox ? box_err : dd_err);
682 }
683 ir->nstlist = nstlist_orig;
684 }
685 else if (ir->nstlist != nstlist_orig || rlist_new != ir->rlist)
686 {
687 sprintf(buf, "Changing nstlist from %d to %d, rlist from %g to %g",
688 nstlist_orig, ir->nstlist,
689 ir->rlist, rlist_new);
690 if (MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)))
691 {
692 fprintf(stderrstderr, "%s\n\n", buf);
693 }
694 if (fp != NULL((void*)0))
695 {
696 fprintf(fp, "%s\n\n", buf);
697 }
698 ir->rlist = rlist_new;
699 ir->rlistlong = rlist_new;
700 }
701}
702
703static void prepare_verlet_scheme(FILE *fplog,
704 t_commrec *cr,
705 t_inputrec *ir,
706 int nstlist_cmdline,
707 const gmx_mtop_t *mtop,
708 matrix box,
709 gmx_bool bUseGPU)
710{
711 /* For NVE simulations, we will retain the initial list buffer */
712 if (ir->verletbuf_tol > 0 && !(EI_MD(ir->eI)((ir->eI) == eiMD || ((ir->eI) == eiVV || (ir->eI) ==
eiVVAK)) && ir->etc == etcNO))
713 {
714 /* Update the Verlet buffer size for the current run setup */
715 verletbuf_list_setup_t ls;
716 real rlist_new;
717
718 /* Here we assume SIMD-enabled kernels are being used. But as currently
719 * calc_verlet_buffer_size gives the same results for 4x8 and 4x4
720 * and 4x2 gives a larger buffer than 4x4, this is ok.
721 */
722 verletbuf_get_list_setup(bUseGPU, &ls);
723
724 calc_verlet_buffer_size(mtop, det(box), ir, -1, &ls, NULL((void*)0), &rlist_new);
725
726 if (rlist_new != ir->rlist)
727 {
728 if (fplog != NULL((void*)0))
729 {
730 fprintf(fplog, "\nChanging rlist from %g to %g for non-bonded %dx%d atom kernels\n\n",
731 ir->rlist, rlist_new,
732 ls.cluster_size_i, ls.cluster_size_j);
733 }
734 ir->rlist = rlist_new;
735 ir->rlistlong = rlist_new;
736 }
737 }
738
739 if (nstlist_cmdline > 0 && (!EI_DYNAMICS(ir->eI)(((ir->eI) == eiMD || ((ir->eI) == eiVV || (ir->eI) ==
eiVVAK)) || ((ir->eI) == eiSD1 || (ir->eI) == eiSD2) ||
(ir->eI) == eiBD) || ir->verletbuf_tol <= 0))
740 {
741 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 741, "Can not set nstlist without %s",
742 !EI_DYNAMICS(ir->eI)(((ir->eI) == eiMD || ((ir->eI) == eiVV || (ir->eI) ==
eiVVAK)) || ((ir->eI) == eiSD1 || (ir->eI) == eiSD2) ||
(ir->eI) == eiBD) ? "dynamics" : "verlet-buffer-tolerance");
743 }
744
745 if (EI_DYNAMICS(ir->eI)(((ir->eI) == eiMD || ((ir->eI) == eiVV || (ir->eI) ==
eiVVAK)) || ((ir->eI) == eiSD1 || (ir->eI) == eiSD2) ||
(ir->eI) == eiBD))
746 {
747 /* Set or try nstlist values */
748 increase_nstlist(fplog, cr, ir, nstlist_cmdline, mtop, box, bUseGPU);
749 }
750}
751
752static void convert_to_verlet_scheme(FILE *fplog,
753 t_inputrec *ir,
754 gmx_mtop_t *mtop, real box_vol)
755{
756 char *conv_mesg = "Converting input file with group cut-off scheme to the Verlet cut-off scheme";
757
758 md_print_warn(NULL((void*)0), fplog, "%s\n", conv_mesg);
759
760 ir->cutoff_scheme = ecutsVERLET;
761 ir->verletbuf_tol = 0.005;
762
763 if (ir->rcoulomb != ir->rvdw)
764 {
765 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 765, "The VdW and Coulomb cut-offs are different, whereas the Verlet scheme only supports equal cut-offs");
766 }
767
768 if (ir->vdwtype == evdwUSER || EEL_USER(ir->coulombtype)((ir->coulombtype) == eelUSER || (ir->coulombtype) == eelPMEUSER
|| (ir->coulombtype) == (eelPMEUSERSWITCH)))
769 {
770 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 770, "User non-bonded potentials are not (yet) supported with the Verlet scheme");
771 }
772 else if (ir_vdw_switched(ir) || ir_coulomb_switched(ir))
773 {
774 if (ir_vdw_switched(ir) && ir->vdw_modifier == eintmodNONE)
775 {
776 ir->vdwtype = evdwCUT;
777
778 switch (ir->vdwtype)
779 {
780 case evdwSHIFT: ir->vdw_modifier = eintmodFORCESWITCH; break;
781 case evdwSWITCH: ir->vdw_modifier = eintmodPOTSWITCH; break;
782 default: gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 782, "The Verlet scheme does not support Van der Waals interactions of type '%s'", evdw_names[ir->vdwtype]);
783 }
784 }
785 if (ir_coulomb_switched(ir) && ir->coulomb_modifier == eintmodNONE)
786 {
787 if (EEL_FULL(ir->coulombtype)((((ir->coulombtype) == eelPME || (ir->coulombtype) == eelPMESWITCH
|| (ir->coulombtype) == eelPMEUSER || (ir->coulombtype
) == eelPMEUSERSWITCH || (ir->coulombtype) == eelP3M_AD) ||
(ir->coulombtype) == eelEWALD) || (ir->coulombtype) ==
eelPOISSON))
788 {
789 /* With full electrostatic only PME can be switched */
790 ir->coulombtype = eelPME;
791 ir->coulomb_modifier = eintmodPOTSHIFT;
792 }
793 else
794 {
795 md_print_warn(NULL((void*)0), fplog, "NOTE: Replacing %s electrostatics with reaction-field with epsilon-rf=inf\n", eel_names[ir->coulombtype]);
796 ir->coulombtype = eelRF;
797 ir->epsilon_rf = 0.0;
798 ir->coulomb_modifier = eintmodPOTSHIFT;
799 }
800 }
801
802 /* We set the pair energy error tolerance to a small number.
803 * Note that this is only for testing. For production the user
804 * should think about this and set the mdp options.
805 */
806 ir->verletbuf_tol = 1e-4;
807 }
808
809 if (inputrec2nboundeddim(ir) != 3)
810 {
811 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 811, "Can only convert old tpr files to the Verlet cut-off scheme with 3D pbc");
812 }
813
814 if (ir->efep != efepNO || ir->implicit_solvent != eisNO)
815 {
816 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 816, "Will not convert old tpr files to the Verlet cut-off scheme with free-energy calculations or implicit solvent");
817 }
818
819 if (EI_DYNAMICS(ir->eI)(((ir->eI) == eiMD || ((ir->eI) == eiVV || (ir->eI) ==
eiVVAK)) || ((ir->eI) == eiSD1 || (ir->eI) == eiSD2) ||
(ir->eI) == eiBD) && !(EI_MD(ir->eI)((ir->eI) == eiMD || ((ir->eI) == eiVV || (ir->eI) ==
eiVVAK)) && ir->etc == etcNO))
820 {
821 verletbuf_list_setup_t ls;
822
823 verletbuf_get_list_setup(FALSE0, &ls);
824 calc_verlet_buffer_size(mtop, box_vol, ir, -1, &ls, NULL((void*)0), &ir->rlist);
825 }
826 else
827 {
828 real rlist_fac;
829
830 if (EI_MD(ir->eI)((ir->eI) == eiMD || ((ir->eI) == eiVV || (ir->eI) ==
eiVVAK)))
831 {
832 rlist_fac = 1 + verlet_buffer_ratio_NVE_T0;
833 }
834 else
835 {
836 rlist_fac = 1 + verlet_buffer_ratio_nodynamics;
837 }
838 ir->verletbuf_tol = -1;
839 ir->rlist = rlist_fac*max(ir->rvdw, ir->rcoulomb)(((ir->rvdw) > (ir->rcoulomb)) ? (ir->rvdw) : (ir
->rcoulomb) );
840 }
841
842 gmx_mtop_remove_chargegroups(mtop);
843}
844
845static void print_hw_opt(FILE *fp, const gmx_hw_opt_t *hw_opt)
846{
847 fprintf(fp, "hw_opt: nt %d ntmpi %d ntomp %d ntomp_pme %d gpu_id '%s'\n",
848 hw_opt->nthreads_tot,
849 hw_opt->nthreads_tmpi,
850 hw_opt->nthreads_omp,
851 hw_opt->nthreads_omp_pme,
852 hw_opt->gpu_opt.gpu_id != NULL((void*)0) ? hw_opt->gpu_opt.gpu_id : "");
853}
854
855/* Checks we can do when we don't (yet) know the cut-off scheme */
856static void check_and_update_hw_opt_1(gmx_hw_opt_t *hw_opt,
857 gmx_bool bIsSimMaster)
858{
859 gmx_omp_nthreads_read_env(&hw_opt->nthreads_omp, bIsSimMaster);
860
861#ifndef GMX_THREAD_MPI
862 if (hw_opt->nthreads_tot > 0)
863 {
864 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 864, "Setting the total number of threads is only supported with thread-MPI and Gromacs was compiled without thread-MPI");
865 }
866 if (hw_opt->nthreads_tmpi > 0)
867 {
868 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 868, "Setting the number of thread-MPI threads is only supported with thread-MPI and Gromacs was compiled without thread-MPI");
869 }
870#endif
871
872#ifndef GMX_OPENMP
873 if (hw_opt->nthreads_omp > 1)
874 {
875 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 875, "More than 1 OpenMP thread requested, but Gromacs was compiled without OpenMP support");
876 }
877 hw_opt->nthreads_omp = 1;
878#endif
879
880 if (hw_opt->nthreads_tot > 0 && hw_opt->nthreads_omp_pme <= 0)
881 {
882 /* We have the same number of OpenMP threads for PP and PME processes,
883 * thus we can perform several consistency checks.
884 */
885 if (hw_opt->nthreads_tmpi > 0 &&
886 hw_opt->nthreads_omp > 0 &&
887 hw_opt->nthreads_tot != hw_opt->nthreads_tmpi*hw_opt->nthreads_omp)
888 {
889 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 889, "The total number of threads requested (%d) does not match the thread-MPI threads (%d) times the OpenMP threads (%d) requested",
890 hw_opt->nthreads_tot, hw_opt->nthreads_tmpi, hw_opt->nthreads_omp);
891 }
892
893 if (hw_opt->nthreads_tmpi > 0 &&
894 hw_opt->nthreads_tot % hw_opt->nthreads_tmpi != 0)
895 {
896 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 896, "The total number of threads requested (%d) is not divisible by the number of thread-MPI threads requested (%d)",
897 hw_opt->nthreads_tot, hw_opt->nthreads_tmpi);
898 }
899
900 if (hw_opt->nthreads_omp > 0 &&
901 hw_opt->nthreads_tot % hw_opt->nthreads_omp != 0)
902 {
903 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 903, "The total number of threads requested (%d) is not divisible by the number of OpenMP threads requested (%d)",
904 hw_opt->nthreads_tot, hw_opt->nthreads_omp);
905 }
906
907 if (hw_opt->nthreads_tmpi > 0 &&
908 hw_opt->nthreads_omp <= 0)
909 {
910 hw_opt->nthreads_omp = hw_opt->nthreads_tot/hw_opt->nthreads_tmpi;
911 }
912 }
913
914#ifndef GMX_OPENMP
915 if (hw_opt->nthreads_omp > 1)
916 {
917 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 917, "OpenMP threads are requested, but Gromacs was compiled without OpenMP support");
918 }
919#endif
920
921 if (hw_opt->nthreads_omp_pme > 0 && hw_opt->nthreads_omp <= 0)
922 {
923 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 923, "You need to specify -ntomp in addition to -ntomp_pme");
924 }
925
926 if (hw_opt->nthreads_tot == 1)
927 {
928 hw_opt->nthreads_tmpi = 1;
929
930 if (hw_opt->nthreads_omp > 1)
931 {
932 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 932, "You requested %d OpenMP threads with %d total threads",
933 hw_opt->nthreads_tmpi, hw_opt->nthreads_tot);
934 }
935 hw_opt->nthreads_omp = 1;
936 }
937
938 if (hw_opt->nthreads_omp_pme <= 0 && hw_opt->nthreads_omp > 0)
939 {
940 hw_opt->nthreads_omp_pme = hw_opt->nthreads_omp;
941 }
942
943 /* Parse GPU IDs, if provided.
944 * We check consistency with the tMPI thread count later.
945 */
946 gmx_parse_gpu_ids(&hw_opt->gpu_opt);
947
948#ifdef GMX_THREAD_MPI
949 if (hw_opt->gpu_opt.ncuda_dev_use > 0 && hw_opt->nthreads_tmpi == 0)
950 {
951 /* Set the number of MPI threads equal to the number of GPUs */
952 hw_opt->nthreads_tmpi = hw_opt->gpu_opt.ncuda_dev_use;
953
954 if (hw_opt->nthreads_tot > 0 &&
955 hw_opt->nthreads_tmpi > hw_opt->nthreads_tot)
956 {
957 /* We have more GPUs than total threads requested.
958 * We choose to (later) generate a mismatch error,
959 * instead of launching more threads than requested.
960 */
961 hw_opt->nthreads_tmpi = hw_opt->nthreads_tot;
962 }
963 }
964#endif
965
966 if (debug)
967 {
968 print_hw_opt(debug, hw_opt);
969 }
970}
971
972/* Checks we can do when we know the cut-off scheme */
973static void check_and_update_hw_opt_2(gmx_hw_opt_t *hw_opt,
974 int cutoff_scheme)
975{
976 if (cutoff_scheme == ecutsGROUP)
977 {
978 /* We only have OpenMP support for PME only nodes */
979 if (hw_opt->nthreads_omp > 1)
980 {
981 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 981, "OpenMP threads have been requested with cut-off scheme %s, but these are only supported with cut-off scheme %s",
982 ecutscheme_names[cutoff_scheme],
983 ecutscheme_names[ecutsVERLET]);
984 }
985 hw_opt->nthreads_omp = 1;
986 }
987
988 if (hw_opt->nthreads_omp_pme <= 0 && hw_opt->nthreads_omp > 0)
989 {
990 hw_opt->nthreads_omp_pme = hw_opt->nthreads_omp;
991 }
992
993 if (debug)
994 {
995 print_hw_opt(debug, hw_opt);
996 }
997}
998
999
1000/* Override the value in inputrec with value passed on the command line (if any) */
1001static void override_nsteps_cmdline(FILE *fplog,
1002 gmx_int64_t nsteps_cmdline,
1003 t_inputrec *ir,
1004 const t_commrec *cr)
1005{
1006 char sbuf[STEPSTRSIZE22];
1007
1008 assert(ir)((void) (0));
1009 assert(cr)((void) (0));
1010
1011 /* override with anything else than the default -2 */
1012 if (nsteps_cmdline > -2)
1013 {
1014 char stmp[STRLEN4096];
1015
1016 ir->nsteps = nsteps_cmdline;
1017 if (EI_DYNAMICS(ir->eI)(((ir->eI) == eiMD || ((ir->eI) == eiVV || (ir->eI) ==
eiVVAK)) || ((ir->eI) == eiSD1 || (ir->eI) == eiSD2) ||
(ir->eI) == eiBD))
1018 {
1019 sprintf(stmp, "Overriding nsteps with value passed on the command line: %s steps, %.3f ps",
1020 gmx_step_str(nsteps_cmdline, sbuf),
1021 nsteps_cmdline*ir->delta_t);
1022 }
1023 else
1024 {
1025 sprintf(stmp, "Overriding nsteps with value passed on the command line: %s steps",
1026 gmx_step_str(nsteps_cmdline, sbuf));
1027 }
1028
1029 md_print_warn(cr, fplog, "%s\n", stmp);
1030 }
1031}
1032
1033/* Frees GPU memory and destroys the CUDA context.
1034 *
1035 * Note that this function needs to be called even if GPUs are not used
1036 * in this run because the PME ranks have no knowledge of whether GPUs
1037 * are used or not, but all ranks need to enter the barrier below.
1038 */
1039static void free_gpu_resources(const t_forcerec *fr,
1040 const t_commrec *cr)
1041{
1042 gmx_bool bIsPPrankUsingGPU;
1043 char gpu_err_str[STRLEN4096];
1044
1045 bIsPPrankUsingGPU = (cr->duty & DUTY_PP(1<<0)) && fr->nbv != NULL((void*)0) && fr->nbv->bUseGPU;
1046
1047 if (bIsPPrankUsingGPU)
1048 {
1049 /* free nbnxn data in GPU memory */
1050 nbnxn_cuda_free(fr->nbv->cu_nbv);
1051
1052 /* With tMPI we need to wait for all ranks to finish deallocation before
1053 * destroying the context in free_gpu() as some ranks may be sharing
1054 * GPU and context.
1055 * Note: as only PP ranks need to free GPU resources, so it is safe to
1056 * not call the barrier on PME ranks.
1057 */
1058#ifdef GMX_THREAD_MPI
1059 if (PAR(cr)((cr)->nnodes > 1))
1060 {
1061 gmx_barrier(cr);
1062 }
1063#endif /* GMX_THREAD_MPI */
1064
1065 /* uninitialize GPU (by destroying the context) */
1066 if (!free_gpu(gpu_err_str))
1067 {
1068 gmx_warning("On node %d failed to free GPU #%d: %s",
1069 cr->nodeid, get_current_gpu_device_id(), gpu_err_str);
1070 }
1071 }
1072}
1073
1074int mdrunner(gmx_hw_opt_t *hw_opt,
1075 FILE *fplog, t_commrec *cr, int nfile,
1076 const t_filenm fnm[], const output_env_t oenv, gmx_bool bVerbose,
1077 gmx_bool bCompact, int nstglobalcomm,
1078 ivec ddxyz, int dd_node_order, real rdd, real rconstr,
1079 const char *dddlb_opt, real dlb_scale,
1080 const char *ddcsx, const char *ddcsy, const char *ddcsz,
1081 const char *nbpu_opt, int nstlist_cmdline,
1082 gmx_int64_t nsteps_cmdline, int nstepout, int resetstep,
1083 int gmx_unused__attribute__ ((unused)) nmultisim, int repl_ex_nst, int repl_ex_nex,
1084 int repl_ex_seed, real pforce, real cpt_period, real max_hours,
1085 const char *deviceOptions, int imdport, unsigned long Flags)
1086{
1087 gmx_bool bForceUseGPU, bTryUseGPU;
1088 double nodetime = 0, realtime;
1089 t_inputrec *inputrec;
1090 t_state *state = NULL((void*)0);
1091 matrix box;
1092 gmx_ddbox_t ddbox = {0};
1093 int npme_major, npme_minor;
1094 real tmpr1, tmpr2;
1095 t_nrnb *nrnb;
1096 gmx_mtop_t *mtop = NULL((void*)0);
1097 t_mdatoms *mdatoms = NULL((void*)0);
1098 t_forcerec *fr = NULL((void*)0);
1
'fr' initialized to a null pointer value
→
1099 t_fcdata *fcd = NULL((void*)0);
1100 real ewaldcoeff_q = 0;
1101 real ewaldcoeff_lj = 0;
1102 gmx_pme_t *pmedata = NULL((void*)0);
1103 gmx_vsite_t *vsite = NULL((void*)0);
1104 gmx_constr_t constr;
1105 int i, m, nChargePerturbed = -1, nTypePerturbed = 0, status, nalloc;
1106 char *gro;
1107 gmx_wallcycle_t wcycle;
1108 gmx_bool bReadEkin;
1109 int list;
1110 gmx_walltime_accounting_t walltime_accounting = NULL((void*)0);
1111 int rc;
1112 gmx_int64_t reset_counters;
1113 gmx_edsam_t ed = NULL((void*)0);
1114 t_commrec *cr_old = cr;
1115 int nthreads_pme = 1;
1116 int nthreads_pp = 1;
1117 gmx_membed_t membed = NULL((void*)0);
1118 gmx_hw_info_t *hwinfo = NULL((void*)0);
1119 /* The master rank decides early on bUseGPU and broadcasts this later */
1120 gmx_bool bUseGPU = FALSE0;
1121
1122 /* CAUTION: threads may be started later on in this function, so
1123 cr doesn't reflect the final parallel state right now */
1124 snew(inputrec, 1)(inputrec) = save_calloc("inputrec", "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1124, (1), sizeof(*(inputrec)));
1125 snew(mtop, 1)(mtop) = save_calloc("mtop", "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1125, (1), sizeof(*(mtop)));
1126
1127 if (Flags & MD_APPENDFILES(1<<15))
2
←
Taking false branch
→
1128 {
1129 fplog = NULL((void*)0);
1130 }
1131
1132 bForceUseGPU = (strncmp(nbpu_opt, "gpu", 3)(__extension__ (__builtin_constant_p (3) && ((__builtin_constant_p
(nbpu_opt) && strlen (nbpu_opt) < ((size_t) (3)))
|| (__builtin_constant_p ("gpu") && strlen ("gpu") <
((size_t) (3)))) ? __extension__ ({ size_t __s1_len, __s2_len
; (__builtin_constant_p (nbpu_opt) && __builtin_constant_p
("gpu") && (__s1_len = strlen (nbpu_opt), __s2_len =
strlen ("gpu"), (!((size_t)(const void *)((nbpu_opt) + 1) - (
size_t)(const void *)(nbpu_opt) == 1) || __s1_len >= 4) &&
(!((size_t)(const void *)(("gpu") + 1) - (size_t)(const void
*)("gpu") == 1) || __s2_len >= 4)) ? __builtin_strcmp (nbpu_opt
, "gpu") : (__builtin_constant_p (nbpu_opt) && ((size_t
)(const void *)((nbpu_opt) + 1) - (size_t)(const void *)(nbpu_opt
) == 1) && (__s1_len = strlen (nbpu_opt), __s1_len <
4) ? (__builtin_constant_p ("gpu") && ((size_t)(const
void *)(("gpu") + 1) - (size_t)(const void *)("gpu") == 1) ?
__builtin_strcmp (nbpu_opt, "gpu") : (__extension__ ({ const
unsigned char *__s2 = (const unsigned char *) (const char *)
("gpu"); int __result = (((const unsigned char *) (const char
*) (nbpu_opt))[0] - __s2[0]); if (__s1_len > 0 &&
__result == 0) { __result = (((const unsigned char *) (const
char *) (nbpu_opt))[1] - __s2[1]); if (__s1_len > 1 &&
__result == 0) { __result = (((const unsigned char *) (const
char *) (nbpu_opt))[2] - __s2[2]); if (__s1_len > 2 &&
__result == 0) __result = (((const unsigned char *) (const char
*) (nbpu_opt))[3] - __s2[3]); } } __result; }))) : (__builtin_constant_p
("gpu") && ((size_t)(const void *)(("gpu") + 1) - (size_t
)(const void *)("gpu") == 1) && (__s2_len = strlen ("gpu"
), __s2_len < 4) ? (__builtin_constant_p (nbpu_opt) &&
((size_t)(const void *)((nbpu_opt) + 1) - (size_t)(const void
*)(nbpu_opt) == 1) ? __builtin_strcmp (nbpu_opt, "gpu") : (-
(__extension__ ({ const unsigned char *__s2 = (const unsigned
char *) (const char *) (nbpu_opt); int __result = (((const unsigned
char *) (const char *) ("gpu"))[0] - __s2[0]); if (__s2_len >
0 && __result == 0) { __result = (((const unsigned char
*) (const char *) ("gpu"))[1] - __s2[1]); if (__s2_len > 1
&& __result == 0) { __result = (((const unsigned char
*) (const char *) ("gpu"))[2] - __s2[2]); if (__s2_len > 2
&& __result == 0) __result = (((const unsigned char *
) (const char *) ("gpu"))[3] - __s2[3]); } } __result; })))) :
__builtin_strcmp (nbpu_opt, "gpu")))); }) : strncmp (nbpu_opt
, "gpu", 3))) == 0);
1133 bTryUseGPU = (strncmp(nbpu_opt, "auto", 4)(__extension__ (__builtin_constant_p (4) && ((__builtin_constant_p
(nbpu_opt) && strlen (nbpu_opt) < ((size_t) (4)))
|| (__builtin_constant_p ("auto") && strlen ("auto")
< ((size_t) (4)))) ? __extension__ ({ size_t __s1_len, __s2_len
; (__builtin_constant_p (nbpu_opt) && __builtin_constant_p
("auto") && (__s1_len = strlen (nbpu_opt), __s2_len =
strlen ("auto"), (!((size_t)(const void *)((nbpu_opt) + 1) -
(size_t)(const void *)(nbpu_opt) == 1) || __s1_len >= 4) &&
(!((size_t)(const void *)(("auto") + 1) - (size_t)(const void
*)("auto") == 1) || __s2_len >= 4)) ? __builtin_strcmp (nbpu_opt
, "auto") : (__builtin_constant_p (nbpu_opt) && ((size_t
)(const void *)((nbpu_opt) + 1) - (size_t)(const void *)(nbpu_opt
) == 1) && (__s1_len = strlen (nbpu_opt), __s1_len <
4) ? (__builtin_constant_p ("auto") && ((size_t)(const
void *)(("auto") + 1) - (size_t)(const void *)("auto") == 1)
? __builtin_strcmp (nbpu_opt, "auto") : (__extension__ ({ const
unsigned char *__s2 = (const unsigned char *) (const char *)
("auto"); int __result = (((const unsigned char *) (const char
*) (nbpu_opt))[0] - __s2[0]); if (__s1_len > 0 &&
__result == 0) { __result = (((const unsigned char *) (const
char *) (nbpu_opt))[1] - __s2[1]); if (__s1_len > 1 &&
__result == 0) { __result = (((const unsigned char *) (const
char *) (nbpu_opt))[2] - __s2[2]); if (__s1_len > 2 &&
__result == 0) __result = (((const unsigned char *) (const char
*) (nbpu_opt))[3] - __s2[3]); } } __result; }))) : (__builtin_constant_p
("auto") && ((size_t)(const void *)(("auto") + 1) - (
size_t)(const void *)("auto") == 1) && (__s2_len = strlen
("auto"), __s2_len < 4) ? (__builtin_constant_p (nbpu_opt
) && ((size_t)(const void *)((nbpu_opt) + 1) - (size_t
)(const void *)(nbpu_opt) == 1) ? __builtin_strcmp (nbpu_opt,
"auto") : (- (__extension__ ({ const unsigned char *__s2 = (
const unsigned char *) (const char *) (nbpu_opt); int __result
= (((const unsigned char *) (const char *) ("auto"))[0] - __s2
[0]); if (__s2_len > 0 && __result == 0) { __result
= (((const unsigned char *) (const char *) ("auto"))[1] - __s2
[1]); if (__s2_len > 1 && __result == 0) { __result
= (((const unsigned char *) (const char *) ("auto"))[2] - __s2
[2]); if (__s2_len > 2 && __result == 0) __result =
(((const unsigned char *) (const char *) ("auto"))[3] - __s2
[3]); } } __result; })))) : __builtin_strcmp (nbpu_opt, "auto"
)))); }) : strncmp (nbpu_opt, "auto", 4))) == 0) || bForceUseGPU;
1134
1135 /* Detect hardware, gather information. This is an operation that is
1136 * global for this process (MPI rank). */
1137 hwinfo = gmx_detect_hardware(fplog, cr, bTryUseGPU);
1138
1139
1140 snew(state, 1)(state) = save_calloc("state", "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1140, (1), sizeof(*(state)));
1141 if (SIMMASTER(cr)(((((cr)->nodeid == 0) || !((cr)->nnodes > 1)) &&
((cr)->duty & (1<<0))) || !((cr)->nnodes >
1)))
3
←
Taking true branch
→
1142 {
1143 /* Read (nearly) all data required for the simulation */
1144 read_tpx_state(ftp2fn(efTPX, nfile, fnm), inputrec, state, NULL((void*)0), mtop);
1145
1146 if (inputrec->cutoff_scheme != ecutsVERLET &&
1147 ((Flags & MD_TESTVERLET(1<<22)) || getenv("GMX_VERLET_SCHEME") != NULL((void*)0)))
1148 {
1149 convert_to_verlet_scheme(fplog, inputrec, mtop, det(state->box));
1150 }
1151
1152 if (inputrec->cutoff_scheme == ecutsVERLET)
4
←
Taking true branch
→
1153 {
1154 /* Here the master rank decides if all ranks will use GPUs */
1155 bUseGPU = (hwinfo->gpu_info.ncuda_dev_compatible > 0 ||
1156 getenv("GMX_EMULATE_GPU") != NULL((void*)0));
1157
1158 /* TODO add GPU kernels for this and replace this check by:
1159 * (bUseGPU && (ir->vdwtype == evdwPME &&
1160 * ir->ljpme_combination_rule == eljpmeLB))
1161 * update the message text and the content of nbnxn_acceleration_supported.
1162 */
1163 if (bUseGPU &&
1164 !nbnxn_acceleration_supported(fplog, cr, inputrec, bUseGPU))
1165 {
1166 /* Fallback message printed by nbnxn_acceleration_supported */
1167 if (bForceUseGPU)
1168 {
1169 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1169, "GPU acceleration requested, but not supported with the given input settings");
1170 }
1171 bUseGPU = FALSE0;
1172 }
1173
1174 prepare_verlet_scheme(fplog, cr,
1175 inputrec, nstlist_cmdline, mtop, state->box,
1176 bUseGPU);
1177 }
1178 else
1179 {
1180 if (nstlist_cmdline > 0)
1181 {
1182 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1182, "Can not set nstlist with the group cut-off scheme");
1183 }
1184
1185 if (hwinfo->gpu_info.ncuda_dev_compatible > 0)
1186 {
1187 md_print_warn(cr, fplog,
1188 "NOTE: GPU(s) found, but the current simulation can not use GPUs\n"
1189 " To use a GPU, set the mdp option: cutoff-scheme = Verlet\n"
1190 " (for quick performance testing you can use the -testverlet option)\n");
1191 }
1192
1193 if (bForceUseGPU)
1194 {
1195 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1195, "GPU requested, but can't be used without cutoff-scheme=Verlet");
1196 }
1197
1198#ifdef GMX_TARGET_BGQ
1199 md_print_warn(cr, fplog,
1200 "NOTE: There is no SIMD implementation of the group scheme kernels on\n"
1201 " BlueGene/Q. You will observe better performance from using the\n"
1202 " Verlet cut-off scheme.\n");
1203#endif
1204 }
1205 }
1206
1207 /* Check and update the hardware options for internal consistency */
1208 check_and_update_hw_opt_1(hw_opt, SIMMASTER(cr)(((((cr)->nodeid == 0) || !((cr)->nnodes > 1)) &&
((cr)->duty & (1<<0))) || !((cr)->nnodes >
1)));
1209
1210 /* Early check for externally set process affinity. */
1211 gmx_check_thread_affinity_set(fplog, cr,
1212 hw_opt, hwinfo->nthreads_hw_avail, FALSE0);
1213 if (SIMMASTER(cr)(((((cr)->nodeid == 0) || !((cr)->nnodes > 1)) &&
((cr)->duty & (1<<0))) || !((cr)->nnodes >
1)))
5
←
Taking true branch
→
1214 {
1215
1216#ifdef GMX_THREAD_MPI
1217 if (cr->npmenodes > 0 && hw_opt->nthreads_tmpi <= 0)
1218 {
1219 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1219, "You need to explicitly specify the number of MPI threads (-ntmpi) when using separate PME nodes");
1220 }
1221#endif
1222
1223 if (hw_opt->nthreads_omp_pme != hw_opt->nthreads_omp &&
1224 cr->npmenodes <= 0)
1225 {
1226 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1226, "You need to explicitly specify the number of PME nodes (-npme) when using different number of OpenMP threads for PP and PME nodes");
1227 }
1228 }
1229
1230#ifdef GMX_THREAD_MPI
1231 if (SIMMASTER(cr)(((((cr)->nodeid == 0) || !((cr)->nnodes > 1)) &&
((cr)->duty & (1<<0))) || !((cr)->nnodes >
1)))
6
←
Taking true branch
→
1232 {
1233 /* Since the master knows the cut-off scheme, update hw_opt for this.
1234 * This is done later for normal MPI and also once more with tMPI
1235 * for all tMPI ranks.
1236 */
1237 check_and_update_hw_opt_2(hw_opt, inputrec->cutoff_scheme);
1238
1239 /* NOW the threads will be started: */
1240 hw_opt->nthreads_tmpi = get_nthreads_mpi(hwinfo,
1241 hw_opt,
1242 inputrec, mtop,
1243 cr, fplog);
1244 if (hw_opt->nthreads_tot > 0 && hw_opt->nthreads_omp <= 0)
1245 {
1246 hw_opt->nthreads_omp = hw_opt->nthreads_tot/hw_opt->nthreads_tmpi;
1247 }
1248
1249 if (hw_opt->nthreads_tmpi > 1)
7
←
Taking true branch
→
1250 {
1251 /* now start the threads. */
1252 cr = mdrunner_start_threads(hw_opt, fplog, cr_old, nfile, fnm,
1253 oenv, bVerbose, bCompact, nstglobalcomm,
1254 ddxyz, dd_node_order, rdd, rconstr,
1255 dddlb_opt, dlb_scale, ddcsx, ddcsy, ddcsz,
1256 nbpu_opt, nstlist_cmdline,
1257 nsteps_cmdline, nstepout, resetstep, nmultisim,
1258 repl_ex_nst, repl_ex_nex, repl_ex_seed, pforce,
1259 cpt_period, max_hours, deviceOptions,
1260 Flags);
1261 /* the main thread continues here with a new cr. We don't deallocate
1262 the old cr because other threads may still be reading it. */
1263 if (cr == NULL((void*)0))
8
←
Assuming 'cr' is not equal to null
→
9
←
Taking false branch
→
1264 {
1265 gmx_comm("Failed to spawn threads")_gmx_error("comm", "Failed to spawn threads", "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1265);
1266 }
1267 }
1268 }
1269#endif
1270 /* END OF CAUTION: cr is now reliable */
1271
1272 /* g_membed initialisation *
1273 * Because we change the mtop, init_membed is called before the init_parallel *
1274 * (in case we ever want to make it run in parallel) */
1275 if (opt2bSet("-membed", nfile, fnm))
10
←
Taking false branch
→
1276 {
1277 if (MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)))
1278 {
1279 fprintf(stderrstderr, "Initializing membed");
1280 }
1281 membed = init_membed(fplog, nfile, fnm, mtop, inputrec, state, cr, &cpt_period);
1282 }
1283
1284 if (PAR(cr)((cr)->nnodes > 1))
11
←
Taking false branch
→
1285 {
1286 /* now broadcast everything to the non-master nodes/threads: */
1287 init_parallel(cr, inputrec, mtop);
1288 }
1289 if (fplog != NULL((void*)0))
12
←
Assuming 'fplog' is equal to null
→
13
←
Taking false branch
→
1290 {
1291 pr_inputrec(fplog, 0, "Input Parameters", inputrec, FALSE0);
1292 }
1293
1294 /* now make sure the state is initialized and propagated */
1295 set_state_entries(state, inputrec);
1296
1297 /* A parallel command line option consistency check that we can
1298 only do after any threads have started. */
1299 if (!PAR(cr)((cr)->nnodes > 1) &&
14
←
Taking false branch
→
1300 (ddxyz[XX0] > 1 || ddxyz[YY1] > 1 || ddxyz[ZZ2] > 1 || cr->npmenodes > 0))
1301 {
1302 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1302,
1303 "The -dd or -npme option request a parallel simulation, "
1304#ifndef GMX_MPI
1305 "but %s was compiled without threads or MPI enabled"
1306#else
1307#ifdef GMX_THREAD_MPI
1308 "but the number of threads (option -nt) is 1"
1309#else
1310 "but %s was not started through mpirun/mpiexec or only one process was requested through mpirun/mpiexec"
1311#endif
1312#endif
1313 , ShortProgram()
1314 );
1315 }
1316
1317 if ((Flags & MD_RERUN(1<<4)) &&
1318 (EI_ENERGY_MINIMIZATION(inputrec->eI)((inputrec->eI) == eiSteep || (inputrec->eI) == eiCG ||
(inputrec->eI) == eiLBFGS) || eiNM == inputrec->eI))
1319 {
1320 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1320, "The .mdp file specified an energy mininization or normal mode algorithm, and these are not compatible with mdrun -rerun");
1321 }
1322
1323 if (can_use_allvsall(inputrec, TRUE1, cr, fplog) && DOMAINDECOMP(cr)(((cr)->dd != ((void*)0)) && ((cr)->nnodes >
1)))
1324 {
1325 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1325, "All-vs-all loops do not work with domain decomposition, use a single MPI rank");
1326 }
1327
1328 if (!(EEL_PME(inputrec->coulombtype)((inputrec->coulombtype) == eelPME || (inputrec->coulombtype
) == eelPMESWITCH || (inputrec->coulombtype) == eelPMEUSER
|| (inputrec->coulombtype) == eelPMEUSERSWITCH || (inputrec
->coulombtype) == eelP3M_AD) || EVDW_PME(inputrec->vdwtype)((inputrec->vdwtype) == evdwPME)))
15
←
Taking true branch
→
1329 {
1330 if (cr->npmenodes > 0)
16
←
Taking false branch
→
1331 {
1332 gmx_fatal_collective(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1332, cr, NULL((void*)0),
1333 "PME nodes are requested, but the system does not use PME electrostatics or LJ-PME");
1334 }
1335
1336 cr->npmenodes = 0;
1337 }
1338
1339#ifdef GMX_FAHCORE
1340 if (MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)))
1341 {
1342 fcRegisterSteps(inputrec->nsteps, inputrec->init_step);
1343 }
1344#endif
1345
1346 /* NMR restraints must be initialized before load_checkpoint,
1347 * since with time averaging the history is added to t_state.
1348 * For proper consistency check we therefore need to extend
1349 * t_state here.
1350 * So the PME-only nodes (if present) will also initialize
1351 * the distance restraints.
1352 */
1353 snew(fcd, 1)(fcd) = save_calloc("fcd", "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1353, (1), sizeof(*(fcd)));
1354
1355 /* This needs to be called before read_checkpoint to extend the state */
1356 init_disres(fplog, mtop, inputrec, cr, fcd, state, repl_ex_nst > 0);
17
←
Assuming 'repl_ex_nst' is <= 0
→
1357
1358 init_orires(fplog, mtop, state->x, inputrec, cr, &(fcd->orires),
1359 state);
1360
1361 if (DEFORM(*inputrec)((*inputrec).deform[0][0] != 0 || (*inputrec).deform[1][1] !=
0 || (*inputrec).deform[2][2] != 0 || (*inputrec).deform[1][
0] != 0 || (*inputrec).deform[2][0] != 0 || (*inputrec).deform
[2][1] != 0))
18
←
Taking false branch
→
1362 {
1363 /* Store the deform reference box before reading the checkpoint */
1364 if (SIMMASTER(cr)(((((cr)->nodeid == 0) || !((cr)->nnodes > 1)) &&
((cr)->duty & (1<<0))) || !((cr)->nnodes >
1)))
1365 {
1366 copy_mat(state->box, box);
1367 }
1368 if (PAR(cr)((cr)->nnodes > 1))
1369 {
1370 gmx_bcast(sizeof(box), box, cr);
1371 }
1372 /* Because we do not have the update struct available yet
1373 * in which the reference values should be stored,
1374 * we store them temporarily in static variables.
1375 * This should be thread safe, since they are only written once
1376 * and with identical values.
1377 */
1378 tMPI_Thread_mutex_lock(&deform_init_box_mutex);
1379 deform_init_init_step_tpx = inputrec->init_step;
1380 copy_mat(box, deform_init_box_tpx);
1381 tMPI_Thread_mutex_unlock(&deform_init_box_mutex);
1382 }
1383
1384 if (opt2bSet("-cpi", nfile, fnm))
19
←
Taking false branch
→
1385 {
1386 /* Check if checkpoint file exists before doing continuation.
1387 * This way we can use identical input options for the first and subsequent runs...
1388 */
1389 if (gmx_fexist_master(opt2fn_master("-cpi", nfile, fnm, cr), cr) )
1390 {
1391 load_checkpoint(opt2fn_master("-cpi", nfile, fnm, cr), &fplog,
1392 cr, ddxyz,
1393 inputrec, state, &bReadEkin,
1394 (Flags & MD_APPENDFILES(1<<15)),
1395 (Flags & MD_APPENDFILESSET(1<<21)));
1396
1397 if (bReadEkin)
1398 {
1399 Flags |= MD_READ_EKIN(1<<17);
1400 }
1401 }
1402 }
1403
1404 if (((MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)) || (Flags & MD_SEPPOT(1<<7))) && (Flags & MD_APPENDFILES(1<<15)))
1405#ifdef GMX_THREAD_MPI
1406 /* With thread MPI only the master node/thread exists in mdrun.c,
1407 * therefore non-master nodes need to open the "seppot" log file here.
1408 */
1409 || (!MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)) && (Flags & MD_SEPPOT(1<<7)))
1410#endif
1411 )
1412 {
1413 gmx_log_open(ftp2fn(efLOG, nfile, fnm), cr, !(Flags & MD_SEPPOT(1<<7)),
1414 Flags, &fplog);
1415 }
1416
1417 /* override nsteps with value from cmdline */
1418 override_nsteps_cmdline(fplog, nsteps_cmdline, inputrec, cr);
1419
1420 if (SIMMASTER(cr)(((((cr)->nodeid == 0) || !((cr)->nnodes > 1)) &&
((cr)->duty & (1<<0))) || !((cr)->nnodes >
1)))
20
←
Taking true branch
→
1421 {
1422 copy_mat(state->box, box);
1423 }
1424
1425 if (PAR(cr)((cr)->nnodes > 1))
21
←
Taking false branch
→
1426 {
1427 gmx_bcast(sizeof(box), box, cr);
1428 }
1429
1430 /* Essential dynamics */
1431 if (opt2bSet("-ei", nfile, fnm))
22
←
Taking false branch
→
1432 {
1433 /* Open input and output files, allocate space for ED data structure */
1434 ed = ed_open(mtop->natoms, &state->edsamstate, nfile, fnm, Flags, oenv, cr);
1435 }
1436
1437 if (PAR(cr)((cr)->nnodes > 1) && !(EI_TPI(inputrec->eI)((inputrec->eI) == eiTPI || (inputrec->eI) == eiTPIC) ||
1438 inputrec->eI == eiNM))
1439 {
1440 cr->dd = init_domain_decomposition(fplog, cr, Flags, ddxyz, rdd, rconstr,
1441 dddlb_opt, dlb_scale,
1442 ddcsx, ddcsy, ddcsz,
1443 mtop, inputrec,
1444 box, state->x,
1445 &ddbox, &npme_major, &npme_minor);
1446
1447 make_dd_communicators(fplog, cr, dd_node_order);
1448
1449 /* Set overallocation to avoid frequent reallocation of arrays */
1450 set_over_alloc_dd(TRUE1);
1451 }
1452 else
1453 {
1454 /* PME, if used, is done on all nodes with 1D decomposition */
1455 cr->npmenodes = 0;
1456 cr->duty = (DUTY_PP(1<<0) | DUTY_PME(1<<1));
1457 npme_major = 1;
1458 npme_minor = 1;
1459
1460 if (inputrec->ePBC == epbcSCREW)
23
←
Taking false branch
→
1461 {
1462 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1462,
1463 "pbc=%s is only implemented with domain decomposition",
1464 epbc_names[inputrec->ePBC]);
1465 }
1466 }
1467
1468 if (PAR(cr)((cr)->nnodes > 1))
24
←
Taking false branch
→
1469 {
1470 /* After possible communicator splitting in make_dd_communicators.
1471 * we can set up the intra/inter node communication.
1472 */
1473 gmx_setup_nodecomm(fplog, cr);
1474 }
1475
1476 /* Initialize per-physical-node MPI process/thread ID and counters. */
1477 gmx_init_intranode_counters(cr);
1478
1479#ifdef GMX_MPI
1480 md_print_info(cr, fplog, "Using %d MPI %s\n",
1481 cr->nnodes,
1482#ifdef GMX_THREAD_MPI
1483 cr->nnodes == 1 ? "thread" : "threads"
25
←
'?' condition is false
→
1484#else
1485 cr->nnodes == 1 ? "process" : "processes"
1486#endif
1487 );
1488 fflush(stderrstderr);
1489#endif
1490
1491 /* Check and update hw_opt for the cut-off scheme */
1492 check_and_update_hw_opt_2(hw_opt, inputrec->cutoff_scheme);
1493
1494 gmx_omp_nthreads_init(fplog, cr,
1495 hwinfo->nthreads_hw_avail,
1496 hw_opt->nthreads_omp,
1497 hw_opt->nthreads_omp_pme,
1498 (cr->duty & DUTY_PP(1<<0)) == 0,
1499 inputrec->cutoff_scheme == ecutsVERLET);
1500
1501 if (PAR(cr)((cr)->nnodes > 1))
26
←
Taking false branch
→
1502 {
1503 /* The master rank decided on the use of GPUs,
1504 * broadcast this information to all ranks.
1505 */
1506 gmx_bcast_sim(sizeof(bUseGPU), &bUseGPU, cr);
1507 }
1508
1509 if (bUseGPU)
27
←
Taking false branch
→
1510 {
1511 if (cr->npmenodes == -1)
1512 {
1513 /* Don't automatically use PME-only nodes with GPUs */
1514 cr->npmenodes = 0;
1515 }
1516
1517 /* Select GPU id's to use */
1518 gmx_select_gpu_ids(fplog, cr, &hwinfo->gpu_info, bForceUseGPU,
1519 &hw_opt->gpu_opt);
1520 }
1521 else
1522 {
1523 /* Ignore (potentially) manually selected GPUs */
1524 hw_opt->gpu_opt.ncuda_dev_use = 0;
1525 }
1526
1527 /* check consistency across ranks of things like SIMD
1528 * support and number of GPUs selected */
1529 gmx_check_hw_runconf_consistency(fplog, hwinfo, cr, hw_opt, bUseGPU);
1530
1531 if (DOMAINDECOMP(cr)(((cr)->dd != ((void*)0)) && ((cr)->nnodes >
1)))
1532 {
1533 /* When we share GPUs over ranks, we need to know this for the DLB */
1534 dd_setup_dlb_resource_sharing(cr, hwinfo, hw_opt);
1535 }
1536
1537 /* getting number of PP/PME threads
1538 PME: env variable should be read only on one node to make sure it is
1539 identical everywhere;
1540 */
1541 /* TODO nthreads_pp is only used for pinning threads.
1542 * This is a temporary solution until we have a hw topology library.
1543 */
1544 nthreads_pp = gmx_omp_nthreads_get(emntNonbonded);
1545 nthreads_pme = gmx_omp_nthreads_get(emntPME);
1546
1547 wcycle = wallcycle_init(fplog, resetstep, cr, nthreads_pp, nthreads_pme);
1548
1549 if (PAR(cr)((cr)->nnodes > 1))
28
←
Taking false branch
→
1550 {
1551 /* Master synchronizes its value of reset_counters with all nodes
1552 * including PME only nodes */
1553 reset_counters = wcycle_get_reset_counters(wcycle);
1554 gmx_bcast_sim(sizeof(reset_counters), &reset_counters, cr);
1555 wcycle_set_reset_counters(wcycle, reset_counters);
1556 }
1557
1558 snew(nrnb, 1)(nrnb) = save_calloc("nrnb", "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1558, (1), sizeof(*(nrnb)));
1559 if (cr->duty & DUTY_PP(1<<0))
29
←
Taking false branch
→
1560 {
1561 bcast_state(cr, state);
1562
1563 /* Initiate forcerecord */
1564 fr = mk_forcerec();
1565 fr->hwinfo = hwinfo;
1566 fr->gpu_opt = &hw_opt->gpu_opt;
1567 init_forcerec(fplog, oenv, fr, fcd, inputrec, mtop, cr, box,
1568 opt2fn("-table", nfile, fnm),
1569 opt2fn("-tabletf", nfile, fnm),
1570 opt2fn("-tablep", nfile, fnm),
1571 opt2fn("-tableb", nfile, fnm),
1572 nbpu_opt,
1573 FALSE0,
1574 pforce);
1575
1576 /* version for PCA_NOT_READ_NODE (see md.c) */
1577 /*init_forcerec(fplog,fr,fcd,inputrec,mtop,cr,box,FALSE,
1578 "nofile","nofile","nofile","nofile",FALSE,pforce);
1579 */
1580 fr->bSepDVDL = ((Flags & MD_SEPPOT(1<<7)) == MD_SEPPOT(1<<7));
1581
1582 /* Initialize QM-MM */
1583 if (fr->bQMMM)
1584 {
1585 init_QMMMrec(cr, mtop, inputrec, fr);
1586 }
1587
1588 /* Initialize the mdatoms structure.
1589 * mdatoms is not filled with atom data,
1590 * as this can not be done now with domain decomposition.
1591 */
1592 mdatoms = init_mdatoms(fplog, mtop, inputrec->efep != efepNO);
1593
1594 /* Initialize the virtual site communication */
1595 vsite = init_vsite(mtop, cr, FALSE0);
1596
1597 calc_shifts(box, fr->shift_vec);
1598
1599 /* With periodic molecules the charge groups should be whole at start up
1600 * and the virtual sites should not be far from their proper positions.
1601 */
1602 if (!inputrec->bContinuation && MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)) &&
1603 !(inputrec->ePBC != epbcNONE && inputrec->bPeriodicMols))
1604 {
1605 /* Make molecules whole at start of run */
1606 if (fr->ePBC != epbcNONE)
1607 {
1608 do_pbc_first_mtop(fplog, inputrec->ePBC, box, mtop, state->x);
1609 }
1610 if (vsite)
1611 {
1612 /* Correct initial vsite positions are required
1613 * for the initial distribution in the domain decomposition
1614 * and for the initial shell prediction.
1615 */
1616 construct_vsites_mtop(vsite, mtop, state->x);
1617 }
1618 }
1619
1620 if (EEL_PME(fr->eeltype)((fr->eeltype) == eelPME || (fr->eeltype) == eelPMESWITCH
|| (fr->eeltype) == eelPMEUSER || (fr->eeltype) == eelPMEUSERSWITCH
|| (fr->eeltype) == eelP3M_AD) || EVDW_PME(fr->vdwtype)((fr->vdwtype) == evdwPME))
1621 {
1622 ewaldcoeff_q = fr->ewaldcoeff_q;
1623 ewaldcoeff_lj = fr->ewaldcoeff_lj;
1624 pmedata = &fr->pmedata;
1625 }
1626 else
1627 {
1628 pmedata = NULL((void*)0);
1629 }
1630 }
1631 else
1632 {
1633 /* This is a PME only node */
1634
1635 /* We don't need the state */
1636 done_state(state);
1637
1638 ewaldcoeff_q = calc_ewaldcoeff_q(inputrec->rcoulomb, inputrec->ewald_rtol);
1639 ewaldcoeff_lj = calc_ewaldcoeff_lj(inputrec->rvdw, inputrec->ewald_rtol_lj);
1640 snew(pmedata, 1)(pmedata) = save_calloc("pmedata", "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1640, (1), sizeof(*(pmedata)));
1641 }
1642
1643 if (hw_opt->thread_affinity != threadaffOFF)
30
←
Taking false branch
→
1644 {
1645 /* Before setting affinity, check whether the affinity has changed
1646 * - which indicates that probably the OpenMP library has changed it
1647 * since we first checked).
1648 */
1649 gmx_check_thread_affinity_set(fplog, cr,
1650 hw_opt, hwinfo->nthreads_hw_avail, TRUE1);
1651
1652 /* Set the CPU affinity */
1653 gmx_set_thread_affinity(fplog, cr, hw_opt, hwinfo);
1654 }
1655
1656 /* Initiate PME if necessary,
1657 * either on all nodes or on dedicated PME nodes only. */
1658 if (EEL_PME(inputrec->coulombtype)((inputrec->coulombtype) == eelPME || (inputrec->coulombtype
) == eelPMESWITCH || (inputrec->coulombtype) == eelPMEUSER
|| (inputrec->coulombtype) == eelPMEUSERSWITCH || (inputrec
->coulombtype) == eelP3M_AD) || EVDW_PME(inputrec->vdwtype)((inputrec->vdwtype) == evdwPME))
1659 {
1660 if (mdatoms)
31
←
Taking false branch
→
1661 {
1662 nChargePerturbed = mdatoms->nChargePerturbed;
1663 if (EVDW_PME(inputrec->vdwtype)((inputrec->vdwtype) == evdwPME))
1664 {
1665 nTypePerturbed = mdatoms->nTypePerturbed;
1666 }
1667 }
1668 if (cr->npmenodes > 0)
32
←
Taking false branch
→
1669 {
1670 /* The PME only nodes need to know nChargePerturbed(FEP on Q) and nTypePerturbed(FEP on LJ)*/
1671 gmx_bcast_sim(sizeof(nChargePerturbed), &nChargePerturbed, cr);
1672 gmx_bcast_sim(sizeof(nTypePerturbed), &nTypePerturbed, cr);
1673 }
1674
1675 if (cr->duty & DUTY_PME(1<<1))
33
←
Taking true branch
→
1676 {
1677 status = gmx_pme_init(pmedata, cr, npme_major, npme_minor, inputrec,
1678 mtop ? mtop->natoms : 0, nChargePerturbed, nTypePerturbed,
34
←
'?' condition is true
→
1679 (Flags & MD_REPRODUCIBLE(1<<13)), nthreads_pme);
1680 if (status != 0)
35
←
Assuming 'status' is equal to 0
→
36
←
Taking false branch
→
1681 {
1682 gmx_fatal(FARGS0, "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1682, "Error %d initializing PME", status);
1683 }
1684 }
1685 }
1686
1687
1688 if (integrator[inputrec->eI].func == do_md)
37
←
Taking false branch
→
1689 {
1690 /* Turn on signal handling on all nodes */
1691 /*
1692 * (A user signal from the PME nodes (if any)
1693 * is communicated to the PP nodes.
1694 */
1695 signal_handler_install();
1696 }
1697
1698 if (cr->duty & DUTY_PP(1<<0))
38
←
Taking true branch
→
1699 {
1700 /* Assumes uniform use of the number of OpenMP threads */
1701 walltime_accounting = walltime_accounting_init(gmx_omp_nthreads_get(emntDefault));
1702
1703 if (inputrec->ePull != epullNO)
39
←
Taking false branch
→
1704 {
1705 /* Initialize pull code */
1706 init_pull(fplog, inputrec, nfile, fnm, mtop, cr, oenv, inputrec->fepvals->init_lambda,
1707 EI_DYNAMICS(inputrec->eI)(((inputrec->eI) == eiMD || ((inputrec->eI) == eiVV || (
inputrec->eI) == eiVVAK)) || ((inputrec->eI) == eiSD1 ||
(inputrec->eI) == eiSD2) || (inputrec->eI) == eiBD) && MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)), Flags);
1708 }
1709
1710 if (inputrec->bRot)
40
←
Taking false branch
→
1711 {
1712 /* Initialize enforced rotation code */
1713 init_rot(fplog, inputrec, nfile, fnm, cr, state->x, box, mtop, oenv,
1714 bVerbose, Flags);
1715 }
1716
1717 if (inputrec->eSwapCoords != eswapNO)
41
←
Taking false branch
→
1718 {
1719 /* Initialize ion swapping code */
1720 init_swapcoords(fplog, bVerbose, inputrec, opt2fn_master("-swap", nfile, fnm, cr),
1721 mtop, state->x, state->box, &state->swapstate, cr, oenv, Flags);
1722 }
1723
1724 constr = init_constraints(fplog, mtop, inputrec, ed, state, cr);
1725
1726 if (DOMAINDECOMP(cr)(((cr)->dd != ((void*)0)) && ((cr)->nnodes >
1)))
42
←
Taking true branch
→
1727 {
1728 dd_init_bondeds(fplog, cr->dd, mtop, vsite, inputrec,
1729 Flags & MD_DDBONDCHECK(1<<10), fr->cginfo_mb);
43
←
Access to field 'cginfo_mb' results in a dereference of a null pointer (loaded from variable 'fr')
1730
1731 set_dd_parameters(fplog, cr->dd, dlb_scale, inputrec, &ddbox);
1732
1733 setup_dd_grid(fplog, cr->dd);
1734 }
1735
1736 /* Now do whatever the user wants us to do (how flexible...) */
1737 integrator[inputrec->eI].func(fplog, cr, nfile, fnm,
1738 oenv, bVerbose, bCompact,
1739 nstglobalcomm,
1740 vsite, constr,
1741 nstepout, inputrec, mtop,
1742 fcd, state,
1743 mdatoms, nrnb, wcycle, ed, fr,
1744 repl_ex_nst, repl_ex_nex, repl_ex_seed,
1745 membed,
1746 cpt_period, max_hours,
1747 deviceOptions,
1748 imdport,
1749 Flags,
1750 walltime_accounting);
1751
1752 if (inputrec->ePull != epullNO)
1753 {
1754 finish_pull(inputrec->pull);
1755 }
1756
1757 if (inputrec->bRot)
1758 {
1759 finish_rot(inputrec->rot);
1760 }
1761
1762 }
1763 else
1764 {
1765 /* do PME only */
1766 walltime_accounting = walltime_accounting_init(gmx_omp_nthreads_get(emntPME));
1767 gmx_pmeonly(*pmedata, cr, nrnb, wcycle, walltime_accounting, ewaldcoeff_q, ewaldcoeff_lj, inputrec);
1768 }
1769
1770 wallcycle_stop(wcycle, ewcRUN);
1771
1772 /* Finish up, write some stuff
1773 * if rerunMD, don't write last frame again
1774 */
1775 finish_run(fplog, cr,
1776 inputrec, nrnb, wcycle, walltime_accounting,
1777 fr != NULL((void*)0) && fr->nbv != NULL((void*)0) && fr->nbv->bUseGPU ?
1778 nbnxn_cuda_get_timings(fr->nbv->cu_nbv) : NULL((void*)0),
1779 EI_DYNAMICS(inputrec->eI)(((inputrec->eI) == eiMD || ((inputrec->eI) == eiVV || (
inputrec->eI) == eiVVAK)) || ((inputrec->eI) == eiSD1 ||
(inputrec->eI) == eiSD2) || (inputrec->eI) == eiBD) && !MULTISIM(cr)((cr)->ms));
1780
1781
1782 /* Free GPU memory and context */
1783 free_gpu_resources(fr, cr);
1784
1785 if (opt2bSet("-membed", nfile, fnm))
1786 {
1787 sfree(membed)save_free("membed", "/home/alexxy/Develop/gromacs/src/programs/mdrun/runner.c"
, 1787, (membed));
1788 }
1789
1790 gmx_hardware_info_free(hwinfo);
1791
1792 /* Does what it says */
1793 print_date_and_time(fplog, cr->nodeid, "Finished mdrun", gmx_gettime());
1794 walltime_accounting_destroy(walltime_accounting);
1795
1796 /* Close logfile already here if we were appending to it */
1797 if (MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)) && (Flags & MD_APPENDFILES(1<<15)))
1798 {
1799 gmx_log_close(fplog);
1800 }
1801
1802 rc = (int)gmx_get_stop_condition();
1803
1804#ifdef GMX_THREAD_MPI
1805 /* we need to join all threads. The sub-threads join when they
1806 exit this function, but the master thread needs to be told to
1807 wait for that. */
1808 if (PAR(cr)((cr)->nnodes > 1) && MASTER(cr)(((cr)->nodeid == 0) || !((cr)->nnodes > 1)))
1809 {
1810 tMPI_Finalize();
1811 }
1812#endif
1813
1814 return rc;
1815}