2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2010, The GROMACS development team,
6 * check out http://www.gromacs.org for more information.
7 * Copyright (c) 2012, by the GROMACS development team, led by
8 * David van der Spoel, Berk Hess, Erik Lindahl, and including many
9 * others, as listed in the AUTHORS file in the top-level source
10 * directory and at http://www.gromacs.org.
12 * GROMACS is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public License
14 * as published by the Free Software Foundation; either version 2.1
15 * of the License, or (at your option) any later version.
17 * GROMACS is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with GROMACS; if not, see
24 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
25 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 * If you want to redistribute modifications to GROMACS, please
28 * consider that scientific software is very special. Version
29 * control is crucial - bugs must be traceable. We will be happy to
30 * consider code for inclusion in the official distribution, but
31 * derived work must not be called official GROMACS. Details are found
32 * in the README & COPYING files - if they are missing, get the
33 * official version at http://www.gromacs.org.
35 * To help us fund GROMACS development, we humbly ask that you cite
36 * the research papers on the package. Check out http://www.gromacs.org.
48 #include "gmx_fatal.h"
54 #include "gmx_omp_nthreads.h"
55 #include "md_logging.h"
57 /*! Structure with the number of threads for each OpenMP multi-threaded
58 * algorithmic module in mdrun. */
61 int gnth; /*! Global num. of threads per PP or PP+PME process/tMPI thread. */
62 int gnth_pme; /*! Global num. of threads per PME only process/tMPI thread. */
64 int nth[emntNR]; /*! Number of threads for each module, indexed with module_nth_t */
65 gmx_bool initialized; /*! TRUE if the module as been initialized. */
66 } omp_module_nthreads_t;
68 /*! Names of environment variables to set the per module number of threads.
70 * Indexed with the values of module_nth_t.
72 static const char *modth_env_var[emntNR] =
74 "GMX_DEFAULT_NUM_THREADS should never be set",
75 "GMX_DOMDEC_NUM_THREADS", "GMX_PAIRSEARCH_NUM_THREADS",
76 "GMX_NONBONDED_NUM_THREADS", "GMX_BONDED_NUM_THREADS",
77 "GMX_PME_NUM_THREADS", "GMX_UPDATE_NUM_THREADS",
78 "GMX_VSITE_NUM_THREADS",
79 "GMX_LINCS_NUM_THREADS", "GMX_SETTLE_NUM_THREADS"
82 /*! Names of the modules. */
83 static const char *mod_name[emntNR] =
85 "default", "domain decomposition", "pair search", "non-bonded",
86 "bonded", "PME", "update", "LINCS", "SETTLE"
89 /*! Number of threads for each algorithmic module.
91 * File-scope global variable that gets set once in \init_module_nthreads
92 * and queried via gmx_omp_nthreads_get.
94 * All fields are initialized to 0 which should result in errors if
95 * the init call is omitted.
97 static omp_module_nthreads_t modth = { 0, 0, {0, 0, 0, 0, 0, 0, 0, 0, 0}, FALSE};
100 /*! Determine the number of threads for module \mod.
102 * \m takes values form the module_nth_t enum and maps these to the
103 * corresponding value in modth_env_var.
105 * Each number of threads per module takes the default value unless
106 * GMX_*_NUM_THERADS env var is set, case in which its value overrides
109 * The "group" scheme supports OpenMP only in PME and in thise case all but
110 * the PME nthread values default to 1.
112 static int pick_module_nthreads(FILE *fplog, int m,
114 gmx_bool bFullOmpSupport,
126 #endif /* GMX_OPENMP */
128 /* The default should never be set through a GMX_*_NUM_THREADS env var
129 * as it's always equal with gnth. */
130 if (m == emntDefault)
132 return modth.nth[emntDefault];
135 /* check the environment variable */
136 if ((env = getenv(modth_env_var[m])) != NULL)
138 sscanf(env, "%d", &nth);
142 gmx_warning("%s=%d is set, but %s is compiled without OpenMP!",
143 modth_env_var[m], nth, ShortProgram());
146 /* with the verlet codepath, when any GMX_*_NUM_THREADS env var is set,
147 * OMP_NUM_THREADS also has to be set */
148 if (bFullOmpSupport && getenv("OMP_NUM_THREADS") == NULL)
150 gmx_fatal(FARGS, "%s=%d is set, the default number of threads also "
151 "needs to be set with OMP_NUM_THREADS!",
152 modth_env_var[m], nth);
155 /* with the group scheme warn if any env var except PME is set */
156 if (!bFullOmpSupport)
160 gmx_warning("%s=%d is set, but OpenMP multithreading is not "
162 modth_env_var[m], nth, mod_name[m]);
167 /* only babble if we are really overriding with a different value */
168 if ((bSepPME && m == emntPME && nth != modth.gnth_pme) || (nth != modth.gnth))
170 sprintf(sbuf, "%s=%d set, overriding the default number of %s threads",
171 modth_env_var[m], nth, mod_name[m]);
174 fprintf(stderr, "\n%s\n", sbuf);
178 fprintf(fplog, "%s\n", sbuf);
184 /* pick the global PME node nthreads if we are setting the number
185 * of threads in separate PME nodes */
186 nth = (bSepPME && m == emntPME) ? modth.gnth_pme : modth.gnth;
189 return modth.nth[m] = nth;
192 void gmx_omp_nthreads_read_env(int *nthreads_omp)
196 assert(nthreads_omp);
198 if ((env = getenv("OMP_NUM_THREADS")) != NULL)
202 sscanf(env,"%d",&nt_omp);
205 gmx_fatal(FARGS,"OMP_NUM_THREADS is invalid: '%s'",env);
208 if (*nthreads_omp > 0 && nt_omp != *nthreads_omp)
210 gmx_fatal(FARGS,"OMP_NUM_THREADS (%d) and the number of threads requested on the command line (%d) have different values",nt_omp,*nthreads_omp);
213 /* Setting the number of OpenMP threads.
214 * NOTE: with tMPI this function is only called on the master node,
215 * but with MPI on all nodes which means lots of messages on stderr.
217 fprintf(stderr,"Getting the number of OpenMP threads from OMP_NUM_THREADS: %d\n",nt_omp);
218 *nthreads_omp = nt_omp;
222 void gmx_omp_nthreads_init(FILE *fplog, t_commrec *cr,
223 int nthreads_hw_avail,
224 int omp_nthreads_req,
225 int omp_nthreads_pme_req,
226 gmx_bool bThisNodePMEOnly,
227 gmx_bool bFullOmpSupport)
229 int nth, nth_pmeonly, gmx_maxth, nppn;
231 gmx_bool bSepPME, bOMP;
237 #endif /* GMX_OPENMP */
239 /* number of MPI processes/threads per physical node */
240 nppn = cr->nrank_intranode;
242 bSepPME = ( (cr->duty & DUTY_PP) && !(cr->duty & DUTY_PME)) ||
243 (!(cr->duty & DUTY_PP) && (cr->duty & DUTY_PME));
245 #ifdef GMX_THREAD_MPI
246 /* modth is shared among tMPI threads, so for thread safety do the
247 * detection is done on the master only. It is not thread-safe with
248 * multiple simulations, but that's anyway not supported by tMPI. */
252 /* just return if the initialization has already been done */
253 if (modth.initialized)
258 /* With full OpenMP support (verlet scheme) set the number of threads
259 * per process / default:
260 * - 1 if not compiled with OpenMP or
261 * - OMP_NUM_THREADS if the env. var is set, or
262 * - omp_nthreads_req = #of threads requested by the user on the mdrun
263 * command line, otherwise
264 * - take the max number of available threads and distribute them
265 * on the processes/tMPI threads.
266 * ~ The GMX_*_NUM_THREADS env var overrides the number of threads of
267 * the respective module and it has to be used in conjunction with
270 * With the group scheme OpenMP multithreading is only supported in PME,
271 * for all other modules nthreads is set to 1.
272 * The number of PME threads is equal to:
273 * - 1 if not compiled with OpenMP or
274 * - GMX_PME_NUM_THREADS if defined, otherwise
275 * - OMP_NUM_THREADS if defined, otherwise
279 if ((env = getenv("OMP_NUM_THREADS")) != NULL)
281 if (!bOMP && (strncmp(env, "1", 1) != 0))
283 gmx_warning("OMP_NUM_THREADS is set, but %s was compiled without OpenMP support!",
288 nth = gmx_omp_get_max_threads();
291 else if (omp_nthreads_req > 0)
293 nth = omp_nthreads_req;
295 else if (bFullOmpSupport && bOMP)
297 /* max available threads per node */
298 nth = nthreads_hw_avail;
300 /* divide the threads among the MPI processes/tMPI threads */
311 /* now we have the global values, set them:
312 * - 1 if not compiled with OpenMP and for the group scheme
313 * - nth for the verlet scheme when compiled with OpenMP
315 if (bFullOmpSupport && bOMP)
326 if (omp_nthreads_pme_req > 0)
328 modth.gnth_pme = omp_nthreads_pme_req;
332 modth.gnth_pme = nth;
340 /* now set the per-module values */
341 modth.nth[emntDefault] = modth.gnth;
342 pick_module_nthreads(fplog, emntDomdec, SIMMASTER(cr), bFullOmpSupport, bSepPME);
343 pick_module_nthreads(fplog, emntPairsearch, SIMMASTER(cr), bFullOmpSupport, bSepPME);
344 pick_module_nthreads(fplog, emntNonbonded, SIMMASTER(cr), bFullOmpSupport, bSepPME);
345 pick_module_nthreads(fplog, emntBonded, SIMMASTER(cr), bFullOmpSupport, bSepPME);
346 pick_module_nthreads(fplog, emntPME, SIMMASTER(cr), bFullOmpSupport, bSepPME);
347 pick_module_nthreads(fplog, emntUpdate, SIMMASTER(cr), bFullOmpSupport, bSepPME);
348 pick_module_nthreads(fplog, emntVSITE, SIMMASTER(cr), bFullOmpSupport, bSepPME);
349 pick_module_nthreads(fplog, emntLINCS, SIMMASTER(cr), bFullOmpSupport, bSepPME);
350 pick_module_nthreads(fplog, emntSETTLE, SIMMASTER(cr), bFullOmpSupport, bSepPME);
352 /* set the number of threads globally */
355 #ifndef GMX_THREAD_MPI
356 if (bThisNodePMEOnly)
358 gmx_omp_set_num_threads(modth.gnth_pme);
361 #endif /* GMX_THREAD_MPI */
365 gmx_omp_set_num_threads(nth);
369 gmx_omp_set_num_threads(1);
374 modth.initialized = TRUE;
376 #ifdef GMX_THREAD_MPI
377 /* Non-master threads have to wait for the detection to be done. */
380 MPI_Barrier(cr->mpi_comm_mysim);
384 /* inform the user about the settings */
387 #ifdef GMX_THREAD_MPI
388 const char *mpi_str="per tMPI thread";
390 const char *mpi_str="per MPI process";
393 /* for group scheme we print PME threads info only */
396 md_print_info(cr, fplog, "Using %d OpenMP thread%s %s\n",
397 modth.gnth,modth.gnth > 1 ? "s" : "",
398 cr->nnodes > 1 ? mpi_str : "");
400 if (bSepPME && modth.gnth_pme != modth.gnth)
402 md_print_info(cr, fplog, "Using %d OpenMP thread%s %s for PME\n",
403 modth.gnth_pme,modth.gnth_pme > 1 ? "s" : "",
404 cr->nnodes > 1 ? mpi_str : "");
408 /* detect and warn about oversubscription
409 * TODO: enable this for separate PME nodes as well! */
410 if (!bSepPME && cr->rank_pp_intranode == 0)
412 char sbuf[STRLEN], sbuf1[STRLEN], sbuf2[STRLEN];
414 if (modth.gnth*nppn > nthreads_hw_avail)
416 sprintf(sbuf, "threads");
422 #ifdef GMX_THREAD_MPI
423 sprintf(sbuf, "thread-MPI threads");
425 sprintf(sbuf, "MPI processes");
426 sprintf(sbuf1, " per node");
427 sprintf(sbuf2, "On node %d: o", cr->sim_nodeid);
431 md_print_warn(cr, fplog,
432 "WARNING: %sversubscribing the available %d logical CPU cores%s with %d %s.\n"
433 " This will cause considerable performance loss!",
434 sbuf2, nthreads_hw_avail, sbuf1, nppn*modth.gnth, sbuf);
439 int gmx_omp_nthreads_get(int mod)
441 if (mod < 0 || mod >= emntNR)
443 /* invalid module queried */
448 return modth.nth[mod];