2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2009,2010,2011,2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
41 #ifdef HAVE_SYS_TIME_H
47 #include "gromacs/commandline/pargs.h"
52 #include "gromacs/fileio/tpxio.h"
56 #include "checkpoint.h"
62 #include "gromacs/timing/walltime_accounting.h"
65 /* Enum for situations that can occur during log file parsing, the
66 * corresponding string entries can be found in do_the_tests() in
67 * const char* ParseLog[] */
73 eParselogResetProblem,
77 eParselogLargePrimeFactor,
85 int nPMEnodes; /* number of PME-only nodes used in this test */
86 int nx, ny, nz; /* DD grid */
87 int guessPME; /* if nPMEnodes == -1, this is the guessed number of PME nodes */
88 double *Gcycles; /* This can contain more than one value if doing multiple tests */
92 float *PME_f_load; /* PME mesh/force load average*/
93 float PME_f_load_Av; /* Average average ;) ... */
94 char *mdrun_cmd_line; /* Mdrun command line used for this test */
100 int nr_inputfiles; /* The number of tpr and mdp input files */
101 gmx_int64_t orig_sim_steps; /* Number of steps to be done in the real simulation */
102 gmx_int64_t orig_init_step; /* Init step for the real simulation */
103 real *rcoulomb; /* The coulomb radii [0...nr_inputfiles] */
104 real *rvdw; /* The vdW radii */
105 real *rlist; /* Neighbourlist cutoff radius */
107 int *nkx, *nky, *nkz;
108 real *fsx, *fsy, *fsz; /* Fourierspacing in x,y,z dimension */
112 static void sep_line(FILE *fp)
114 fprintf(fp, "\n------------------------------------------------------------\n");
118 /* Wrapper for system calls */
119 static int gmx_system_call(char *command)
122 gmx_fatal(FARGS, "No calls to system(3) supported on this platform. Attempted to call:\n'%s'\n", command);
124 return ( system(command) );
129 /* Check if string starts with substring */
130 static gmx_bool str_starts(const char *string, const char *substring)
132 return ( strncmp(string, substring, strlen(substring)) == 0);
136 static void cleandata(t_perf *perfdata, int test_nr)
138 perfdata->Gcycles[test_nr] = 0.0;
139 perfdata->ns_per_day[test_nr] = 0.0;
140 perfdata->PME_f_load[test_nr] = 0.0;
146 static gmx_bool is_equal(real a, real b)
148 real diff, eps = 1.0e-7;
169 static void finalize(const char *fn_out)
175 fp = fopen(fn_out, "r");
176 fprintf(stdout, "\n\n");
178 while (fgets(buf, STRLEN-1, fp) != NULL)
180 fprintf(stdout, "%s", buf);
183 fprintf(stdout, "\n\n");
188 eFoundNothing, eFoundDDStr, eFoundAccountingStr, eFoundCycleStr
191 static int parse_logfile(const char *logfile, const char *errfile,
192 t_perf *perfdata, int test_nr, int presteps, gmx_int64_t cpt_steps,
196 char line[STRLEN], dumstring[STRLEN], dumstring2[STRLEN];
197 const char matchstrdd[] = "Domain decomposition grid";
198 const char matchstrcr[] = "resetting all time and cycle counters";
199 const char matchstrbal[] = "Average PME mesh/force load:";
200 const char matchstring[] = "R E A L C Y C L E A N D T I M E A C C O U N T I N G";
201 const char errSIG[] = "signal, stopping at the next";
204 float dum1, dum2, dum3, dum4;
207 gmx_int64_t resetsteps = -1;
208 gmx_bool bFoundResetStr = FALSE;
209 gmx_bool bResetChecked = FALSE;
212 if (!gmx_fexist(logfile))
214 fprintf(stderr, "WARNING: Could not find logfile %s.\n", logfile);
215 cleandata(perfdata, test_nr);
216 return eParselogNotFound;
219 fp = fopen(logfile, "r");
220 perfdata->PME_f_load[test_nr] = -1.0;
221 perfdata->guessPME = -1;
223 iFound = eFoundNothing;
226 iFound = eFoundDDStr; /* Skip some case statements */
229 while (fgets(line, STRLEN, fp) != NULL)
231 /* Remove leading spaces */
234 /* Check for TERM and INT signals from user: */
235 if (strstr(line, errSIG) != NULL)
238 cleandata(perfdata, test_nr);
239 return eParselogTerm;
242 /* Check whether cycle resetting worked */
243 if (presteps > 0 && !bFoundResetStr)
245 if (strstr(line, matchstrcr) != NULL)
247 sprintf(dumstring, "step %s", "%"GMX_SCNd64);
248 sscanf(line, dumstring, &resetsteps);
249 bFoundResetStr = TRUE;
250 if (resetsteps == presteps+cpt_steps)
252 bResetChecked = TRUE;
256 sprintf(dumstring, "%"GMX_PRId64, resetsteps);
257 sprintf(dumstring2, "%"GMX_PRId64, presteps+cpt_steps);
258 fprintf(stderr, "WARNING: Time step counters were reset at step %s,\n"
259 " though they were supposed to be reset at step %s!\n",
260 dumstring, dumstring2);
265 /* Look for strings that appear in a certain order in the log file: */
269 /* Look for domain decomp grid and separate PME nodes: */
270 if (str_starts(line, matchstrdd))
272 sscanf(line, "Domain decomposition grid %d x %d x %d, separate PME nodes %d",
273 &(perfdata->nx), &(perfdata->ny), &(perfdata->nz), &npme);
274 if (perfdata->nPMEnodes == -1)
276 perfdata->guessPME = npme;
278 else if (perfdata->nPMEnodes != npme)
280 gmx_fatal(FARGS, "PME nodes from command line and output file are not identical");
282 iFound = eFoundDDStr;
284 /* Catch a few errors that might have occured: */
285 else if (str_starts(line, "There is no domain decomposition for"))
288 return eParselogNoDDGrid;
290 else if (str_starts(line, "The number of nodes you selected"))
293 return eParselogLargePrimeFactor;
295 else if (str_starts(line, "reading tpx file"))
298 return eParselogTPXVersion;
300 else if (str_starts(line, "The -dd or -npme option request a parallel simulation"))
303 return eParselogNotParallel;
307 /* Look for PME mesh/force balance (not necessarily present, though) */
308 if (str_starts(line, matchstrbal))
310 sscanf(&line[strlen(matchstrbal)], "%f", &(perfdata->PME_f_load[test_nr]));
312 /* Look for matchstring */
313 if (str_starts(line, matchstring))
315 iFound = eFoundAccountingStr;
318 case eFoundAccountingStr:
319 /* Already found matchstring - look for cycle data */
320 if (str_starts(line, "Total "))
322 sscanf(line, "Total %d %lf", &procs, &(perfdata->Gcycles[test_nr]));
323 iFound = eFoundCycleStr;
327 /* Already found cycle data - look for remaining performance info and return */
328 if (str_starts(line, "Performance:"))
330 ndum = sscanf(line, "%s %f %f %f %f", dumstring, &dum1, &dum2, &dum3, &dum4);
331 /* (ns/day) is the second last entry, depending on whether GMX_DETAILED_PERF_STATS was set in print_perf(), nrnb.c */
332 perfdata->ns_per_day[test_nr] = (ndum == 5) ? dum3 : dum1;
334 if (bResetChecked || presteps == 0)
340 return eParselogResetProblem;
347 /* Close the log file */
350 /* Check why there is no performance data in the log file.
351 * Did a fatal errors occur? */
352 if (gmx_fexist(errfile))
354 fp = fopen(errfile, "r");
355 while (fgets(line, STRLEN, fp) != NULL)
357 if (str_starts(line, "Fatal error:") )
359 if (fgets(line, STRLEN, fp) != NULL)
361 fprintf(stderr, "\nWARNING: An error occured during this benchmark:\n"
365 cleandata(perfdata, test_nr);
366 return eParselogFatal;
373 fprintf(stderr, "WARNING: Could not find stderr file %s.\n", errfile);
376 /* Giving up ... we could not find out why there is no performance data in
378 fprintf(stdout, "No performance data in log file.\n");
379 cleandata(perfdata, test_nr);
381 return eParselogNoPerfData;
385 static gmx_bool analyze_data(
394 int *index_tpr, /* OUT: Nr of mdp file with best settings */
395 int *npme_optimal) /* OUT: Optimal number of PME nodes */
398 int line = 0, line_win = -1;
399 int k_win = -1, i_win = -1, winPME;
400 double s = 0.0; /* standard deviation */
403 char str_PME_f_load[13];
404 gmx_bool bCanUseOrigTPR;
405 gmx_bool bRefinedCoul, bRefinedVdW, bRefinedGrid;
411 fprintf(fp, "Summary of successful runs:\n");
412 fprintf(fp, "Line tpr PME nodes Gcycles Av. Std.dev. ns/day PME/f");
415 fprintf(fp, " DD grid");
421 for (k = 0; k < ntprs; k++)
423 for (i = 0; i < ntests; i++)
425 /* Select the right dataset: */
426 pd = &(perfdata[k][i]);
428 pd->Gcycles_Av = 0.0;
429 pd->PME_f_load_Av = 0.0;
430 pd->ns_per_day_Av = 0.0;
432 if (pd->nPMEnodes == -1)
434 sprintf(strbuf, "(%3d)", pd->guessPME);
438 sprintf(strbuf, " ");
441 /* Get the average run time of a setting */
442 for (j = 0; j < nrepeats; j++)
444 pd->Gcycles_Av += pd->Gcycles[j];
445 pd->PME_f_load_Av += pd->PME_f_load[j];
447 pd->Gcycles_Av /= nrepeats;
448 pd->PME_f_load_Av /= nrepeats;
450 for (j = 0; j < nrepeats; j++)
452 if (pd->ns_per_day[j] > 0.0)
454 pd->ns_per_day_Av += pd->ns_per_day[j];
458 /* Somehow the performance number was not aquired for this run,
459 * therefor set the average to some negative value: */
460 pd->ns_per_day_Av = -1.0f*nrepeats;
464 pd->ns_per_day_Av /= nrepeats;
467 if (pd->PME_f_load_Av > 0.0)
469 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load_Av);
473 sprintf(str_PME_f_load, "%s", " - ");
477 /* We assume we had a successful run if both averages are positive */
478 if (pd->Gcycles_Av > 0.0 && pd->ns_per_day_Av > 0.0)
480 /* Output statistics if repeats were done */
483 /* Calculate the standard deviation */
485 for (j = 0; j < nrepeats; j++)
487 s += pow( pd->Gcycles[j] - pd->Gcycles_Av, 2 );
492 fprintf(fp, "%4d %3d %4d%s %12.3f %12.3f %12.3f %s",
493 line, k, pd->nPMEnodes, strbuf, pd->Gcycles_Av, s,
494 pd->ns_per_day_Av, str_PME_f_load);
497 fprintf(fp, " %3d %3d %3d", pd->nx, pd->ny, pd->nz);
501 /* Store the index of the best run found so far in 'winner': */
502 if ( (k_win == -1) || (pd->Gcycles_Av < perfdata[k_win][i_win].Gcycles_Av) )
515 gmx_fatal(FARGS, "None of the runs was successful! Check %s for problems.", fn);
520 winPME = perfdata[k_win][i_win].nPMEnodes;
524 /* We stuck to a fixed number of PME-only nodes */
525 sprintf(strbuf, "settings No. %d", k_win);
529 /* We have optimized the number of PME-only nodes */
532 sprintf(strbuf, "%s", "the automatic number of PME nodes");
536 sprintf(strbuf, "%d PME nodes", winPME);
539 fprintf(fp, "Best performance was achieved with %s", strbuf);
540 if ((nrepeats > 1) && (ntests > 1))
542 fprintf(fp, " (see line %d)", line_win);
546 /* Only mention settings if they were modified: */
547 bRefinedCoul = !is_equal(info->rcoulomb[k_win], info->rcoulomb[0]);
548 bRefinedVdW = !is_equal(info->rvdw[k_win], info->rvdw[0] );
549 bRefinedGrid = !(info->nkx[k_win] == info->nkx[0] &&
550 info->nky[k_win] == info->nky[0] &&
551 info->nkz[k_win] == info->nkz[0]);
553 if (bRefinedCoul || bRefinedVdW || bRefinedGrid)
555 fprintf(fp, "Optimized PME settings:\n");
556 bCanUseOrigTPR = FALSE;
560 bCanUseOrigTPR = TRUE;
565 fprintf(fp, " New Coulomb radius: %f nm (was %f nm)\n", info->rcoulomb[k_win], info->rcoulomb[0]);
570 fprintf(fp, " New Van der Waals radius: %f nm (was %f nm)\n", info->rvdw[k_win], info->rvdw[0]);
575 fprintf(fp, " New Fourier grid xyz: %d %d %d (was %d %d %d)\n", info->nkx[k_win], info->nky[k_win], info->nkz[k_win],
576 info->nkx[0], info->nky[0], info->nkz[0]);
579 if (bCanUseOrigTPR && ntprs > 1)
581 fprintf(fp, "and original PME settings.\n");
586 /* Return the index of the mdp file that showed the highest performance
587 * and the optimal number of PME nodes */
589 *npme_optimal = winPME;
591 return bCanUseOrigTPR;
595 /* Get the commands we need to set up the runs from environment variables */
596 static void get_program_paths(gmx_bool bThreads, char *cmd_mpirun[], char *cmd_mdrun[])
600 const char def_mpirun[] = "mpirun";
601 const char def_mdrun[] = "mdrun";
603 const char empty_mpirun[] = "";
605 /* Get the commands we need to set up the runs from environment variables */
608 if ( (cp = getenv("MPIRUN")) != NULL)
610 *cmd_mpirun = strdup(cp);
614 *cmd_mpirun = strdup(def_mpirun);
619 *cmd_mpirun = strdup(empty_mpirun);
622 if ( (cp = getenv("MDRUN" )) != NULL)
624 *cmd_mdrun = strdup(cp);
628 *cmd_mdrun = strdup(def_mdrun);
632 /* Check that the commands will run mdrun (perhaps via mpirun) by
633 * running a very quick test simulation. Requires MPI environment to
634 * be available if applicable. */
635 static void check_mdrun_works(gmx_bool bThreads,
636 const char *cmd_mpirun,
638 const char *cmd_mdrun)
640 char *command = NULL;
644 const char filename[] = "benchtest.log";
646 /* This string should always be identical to the one in copyrite.c,
647 * gmx_print_version_info() in the defined(GMX_MPI) section */
648 const char match_mpi[] = "MPI library: MPI";
649 const char match_mdrun[] = "Program: ";
650 gmx_bool bMdrun = FALSE;
651 gmx_bool bMPI = FALSE;
653 /* Run a small test to see whether mpirun + mdrun work */
654 fprintf(stdout, "Making sure that mdrun can be executed. ");
657 snew(command, strlen(cmd_mdrun) + strlen(cmd_np) + strlen(filename) + 50);
658 sprintf(command, "%s%s-version -maxh 0.001 1> %s 2>&1", cmd_mdrun, cmd_np, filename);
662 snew(command, strlen(cmd_mpirun) + strlen(cmd_np) + strlen(cmd_mdrun) + strlen(filename) + 50);
663 sprintf(command, "%s%s%s -version -maxh 0.001 1> %s 2>&1", cmd_mpirun, cmd_np, cmd_mdrun, filename);
665 fprintf(stdout, "Trying '%s' ... ", command);
666 make_backup(filename);
667 gmx_system_call(command);
669 /* Check if we find the characteristic string in the output: */
670 if (!gmx_fexist(filename))
672 gmx_fatal(FARGS, "Output from test run could not be found.");
675 fp = fopen(filename, "r");
676 /* We need to scan the whole output file, since sometimes the queuing system
677 * also writes stuff to stdout/err */
680 cp = fgets(line, STRLEN, fp);
683 if (str_starts(line, match_mdrun) )
687 if (str_starts(line, match_mpi) )
699 gmx_fatal(FARGS, "Need a threaded version of mdrun. This one\n"
701 "seems to have been compiled with MPI instead.",
709 gmx_fatal(FARGS, "Need an MPI-enabled version of mdrun. This one\n"
711 "seems to have been compiled without MPI support.",
718 gmx_fatal(FARGS, "Cannot execute mdrun. Please check %s for problems!",
722 fprintf(stdout, "passed.\n");
730 static void launch_simulation(
731 gmx_bool bLaunch, /* Should the simulation be launched? */
732 FILE *fp, /* General log file */
733 gmx_bool bThreads, /* whether to use threads */
734 char *cmd_mpirun, /* Command for mpirun */
735 char *cmd_np, /* Switch for -np or -ntmpi or empty */
736 char *cmd_mdrun, /* Command for mdrun */
737 char *args_for_mdrun, /* Arguments for mdrun */
738 const char *simulation_tpr, /* This tpr will be simulated */
739 int nPMEnodes) /* Number of PME nodes to use */
744 /* Make enough space for the system call command,
745 * (100 extra chars for -npme ... etc. options should suffice): */
746 snew(command, strlen(cmd_mpirun)+strlen(cmd_mdrun)+strlen(cmd_np)+strlen(args_for_mdrun)+strlen(simulation_tpr)+100);
748 /* Note that the -passall options requires args_for_mdrun to be at the end
749 * of the command line string */
752 sprintf(command, "%s%s-npme %d -s %s %s",
753 cmd_mdrun, cmd_np, nPMEnodes, simulation_tpr, args_for_mdrun);
757 sprintf(command, "%s%s%s -npme %d -s %s %s",
758 cmd_mpirun, cmd_np, cmd_mdrun, nPMEnodes, simulation_tpr, args_for_mdrun);
761 fprintf(fp, "%s this command line to launch the simulation:\n\n%s", bLaunch ? "Using" : "Please use", command);
765 /* Now the real thing! */
768 fprintf(stdout, "\nLaunching simulation with best parameters now.\nExecuting '%s'", command);
771 gmx_system_call(command);
776 static void modify_PMEsettings(
777 gmx_int64_t simsteps, /* Set this value as number of time steps */
778 gmx_int64_t init_step, /* Set this value as init_step */
779 const char *fn_best_tpr, /* tpr file with the best performance */
780 const char *fn_sim_tpr) /* name of tpr file to be launched */
788 read_tpx_state(fn_best_tpr, ir, &state, NULL, &mtop);
790 /* Reset nsteps and init_step to the value of the input .tpr file */
791 ir->nsteps = simsteps;
792 ir->init_step = init_step;
794 /* Write the tpr file which will be launched */
795 sprintf(buf, "Writing optimized simulation file %s with nsteps=%s.\n", fn_sim_tpr, "%"GMX_PRId64);
796 fprintf(stdout, buf, ir->nsteps);
798 write_tpx_state(fn_sim_tpr, ir, &state, &mtop);
804 #define EPME_SWITCHED(e) ((e) == eelPMESWITCH || (e) == eelPMEUSERSWITCH)
806 /* Make additional TPR files with more computational load for the
807 * direct space processors: */
808 static void make_benchmark_tprs(
809 const char *fn_sim_tpr, /* READ : User-provided tpr file */
810 char *fn_bench_tprs[], /* WRITE: Names of benchmark tpr files */
811 gmx_int64_t benchsteps, /* Number of time steps for benchmark runs */
812 gmx_int64_t statesteps, /* Step counter in checkpoint file */
813 real rmin, /* Minimal Coulomb radius */
814 real rmax, /* Maximal Coulomb radius */
815 real bScaleRvdw, /* Scale rvdw along with rcoulomb */
816 int *ntprs, /* No. of TPRs to write, each with a different
817 rcoulomb and fourierspacing */
818 t_inputinfo *info, /* Contains information about mdp file options */
819 FILE *fp) /* Write the output here */
825 real nlist_buffer; /* Thickness of the buffer regions for PME-switch potentials */
828 gmx_bool bNote = FALSE;
829 real add; /* Add this to rcoul for the next test */
830 real fac = 1.0; /* Scaling factor for Coulomb radius */
831 real fourierspacing; /* Basic fourierspacing from tpr */
834 sprintf(buf, "Making benchmark tpr file%s with %s time step%s",
835 *ntprs > 1 ? "s" : "", "%"GMX_PRId64, benchsteps > 1 ? "s" : "");
836 fprintf(stdout, buf, benchsteps);
839 sprintf(buf, " (adding %s steps from checkpoint file)", "%"GMX_PRId64);
840 fprintf(stdout, buf, statesteps);
841 benchsteps += statesteps;
843 fprintf(stdout, ".\n");
847 read_tpx_state(fn_sim_tpr, ir, &state, NULL, &mtop);
849 /* Check if some kind of PME was chosen */
850 if (EEL_PME(ir->coulombtype) == FALSE)
852 gmx_fatal(FARGS, "Can only do optimizations for simulations with %s electrostatics.",
856 /* Check if rcoulomb == rlist, which is necessary for plain PME. */
857 if ( (ir->cutoff_scheme != ecutsVERLET) &&
858 (eelPME == ir->coulombtype) && !(ir->rcoulomb == ir->rlist))
860 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to rlist (%f).",
861 EELTYPE(eelPME), ir->rcoulomb, ir->rlist);
863 /* For other PME types, rcoulomb is allowed to be smaller than rlist */
864 else if (ir->rcoulomb > ir->rlist)
866 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to or smaller than rlist (%f)",
867 EELTYPE(ir->coulombtype), ir->rcoulomb, ir->rlist);
870 if (bScaleRvdw && ir->rvdw != ir->rcoulomb)
872 fprintf(stdout, "NOTE: input rvdw != rcoulomb, will not scale rvdw\n");
876 /* Reduce the number of steps for the benchmarks */
877 info->orig_sim_steps = ir->nsteps;
878 ir->nsteps = benchsteps;
879 /* We must not use init_step from the input tpr file for the benchmarks */
880 info->orig_init_step = ir->init_step;
883 /* For PME-switch potentials, keep the radial distance of the buffer region */
884 nlist_buffer = ir->rlist - ir->rcoulomb;
886 /* Determine length of triclinic box vectors */
887 for (d = 0; d < DIM; d++)
890 for (i = 0; i < DIM; i++)
892 box_size[d] += state.box[d][i]*state.box[d][i];
894 box_size[d] = sqrt(box_size[d]);
897 if (ir->fourier_spacing > 0)
899 info->fsx[0] = ir->fourier_spacing;
900 info->fsy[0] = ir->fourier_spacing;
901 info->fsz[0] = ir->fourier_spacing;
905 /* Reconstruct fourierspacing per dimension from the number of grid points and box size */
906 info->fsx[0] = box_size[XX]/ir->nkx;
907 info->fsy[0] = box_size[YY]/ir->nky;
908 info->fsz[0] = box_size[ZZ]/ir->nkz;
911 /* If no value for the fourierspacing was provided on the command line, we
912 * use the reconstruction from the tpr file */
913 if (ir->fourier_spacing > 0)
915 /* Use the spacing from the tpr */
916 fourierspacing = ir->fourier_spacing;
920 /* Use the maximum observed spacing */
921 fourierspacing = max(max(info->fsx[0], info->fsy[0]), info->fsz[0]);
924 fprintf(stdout, "Calculating PME grid points on the basis of a fourierspacing of %f nm\n", fourierspacing);
926 /* For performance comparisons the number of particles is useful to have */
927 fprintf(fp, " Number of particles : %d\n", mtop.natoms);
929 /* Print information about settings of which some are potentially modified: */
930 fprintf(fp, " Coulomb type : %s\n", EELTYPE(ir->coulombtype));
931 fprintf(fp, " Grid spacing x y z : %f %f %f\n",
932 box_size[XX]/ir->nkx, box_size[YY]/ir->nky, box_size[ZZ]/ir->nkz);
933 fprintf(fp, " Van der Waals type : %s\n", EVDWTYPE(ir->vdwtype));
934 if (ir_vdw_switched(ir))
936 fprintf(fp, " rvdw_switch : %f nm\n", ir->rvdw_switch);
938 if (EPME_SWITCHED(ir->coulombtype))
940 fprintf(fp, " rlist : %f nm\n", ir->rlist);
942 if (ir->rlistlong != max_cutoff(ir->rvdw, ir->rcoulomb))
944 fprintf(fp, " rlistlong : %f nm\n", ir->rlistlong);
947 /* Print a descriptive line about the tpr settings tested */
948 fprintf(fp, "\nWill try these real/reciprocal workload settings:\n");
949 fprintf(fp, " No. scaling rcoulomb");
950 fprintf(fp, " nkx nky nkz");
951 fprintf(fp, " spacing");
952 if (evdwCUT == ir->vdwtype)
954 fprintf(fp, " rvdw");
956 if (EPME_SWITCHED(ir->coulombtype))
958 fprintf(fp, " rlist");
960 if (ir->rlistlong != max_cutoff(ir->rlist, max_cutoff(ir->rvdw, ir->rcoulomb)) )
962 fprintf(fp, " rlistlong");
964 fprintf(fp, " tpr file\n");
966 /* Loop to create the requested number of tpr input files */
967 for (j = 0; j < *ntprs; j++)
969 /* The first .tpr is the provided one, just need to modify nsteps,
970 * so skip the following block */
973 /* Determine which Coulomb radii rc to use in the benchmarks */
974 add = (rmax-rmin)/(*ntprs-1);
975 if (is_equal(rmin, info->rcoulomb[0]))
977 ir->rcoulomb = rmin + j*add;
979 else if (is_equal(rmax, info->rcoulomb[0]))
981 ir->rcoulomb = rmin + (j-1)*add;
985 /* rmin != rcoul != rmax, ergo test between rmin and rmax */
986 add = (rmax-rmin)/(*ntprs-2);
987 ir->rcoulomb = rmin + (j-1)*add;
990 /* Determine the scaling factor fac */
991 fac = ir->rcoulomb/info->rcoulomb[0];
993 /* Scale the Fourier grid spacing */
994 ir->nkx = ir->nky = ir->nkz = 0;
995 calc_grid(NULL, state.box, fourierspacing*fac, &ir->nkx, &ir->nky, &ir->nkz);
997 /* Adjust other radii since various conditions need to be fulfilled */
998 if (eelPME == ir->coulombtype)
1000 /* plain PME, rcoulomb must be equal to rlist */
1001 ir->rlist = ir->rcoulomb;
1005 /* rlist must be >= rcoulomb, we keep the size of the buffer region */
1006 ir->rlist = ir->rcoulomb + nlist_buffer;
1009 if (bScaleRvdw && evdwCUT == ir->vdwtype)
1011 if (ecutsVERLET == ir->cutoff_scheme)
1013 /* With Verlet, the van der Waals radius must always equal the Coulomb radius */
1014 ir->rvdw = ir->rcoulomb;
1018 /* For vdw cutoff, rvdw >= rlist */
1019 ir->rvdw = max(info->rvdw[0], ir->rlist);
1023 ir->rlistlong = max_cutoff(ir->rlist, max_cutoff(ir->rvdw, ir->rcoulomb));
1025 } /* end of "if (j != 0)" */
1027 /* for j==0: Save the original settings
1028 * for j >0: Save modified radii and Fourier grids */
1029 info->rcoulomb[j] = ir->rcoulomb;
1030 info->rvdw[j] = ir->rvdw;
1031 info->nkx[j] = ir->nkx;
1032 info->nky[j] = ir->nky;
1033 info->nkz[j] = ir->nkz;
1034 info->rlist[j] = ir->rlist;
1035 info->rlistlong[j] = ir->rlistlong;
1036 info->fsx[j] = fac*fourierspacing;
1037 info->fsy[j] = fac*fourierspacing;
1038 info->fsz[j] = fac*fourierspacing;
1040 /* Write the benchmark tpr file */
1041 strncpy(fn_bench_tprs[j], fn_sim_tpr, strlen(fn_sim_tpr)-strlen(".tpr"));
1042 sprintf(buf, "_bench%.2d.tpr", j);
1043 strcat(fn_bench_tprs[j], buf);
1044 fprintf(stdout, "Writing benchmark tpr %s with nsteps=", fn_bench_tprs[j]);
1045 fprintf(stdout, "%"GMX_PRId64, ir->nsteps);
1048 fprintf(stdout, ", scaling factor %f\n", fac);
1052 fprintf(stdout, ", unmodified settings\n");
1055 write_tpx_state(fn_bench_tprs[j], ir, &state, &mtop);
1057 /* Write information about modified tpr settings to log file */
1058 fprintf(fp, "%4d%10f%10f", j, fac, ir->rcoulomb);
1059 fprintf(fp, "%5d%5d%5d", ir->nkx, ir->nky, ir->nkz);
1060 fprintf(fp, " %9f ", info->fsx[j]);
1061 if (evdwCUT == ir->vdwtype)
1063 fprintf(fp, "%10f", ir->rvdw);
1065 if (EPME_SWITCHED(ir->coulombtype))
1067 fprintf(fp, "%10f", ir->rlist);
1069 if (info->rlistlong[0] != max_cutoff(info->rlist[0], max_cutoff(info->rvdw[0], info->rcoulomb[0])) )
1071 fprintf(fp, "%10f", ir->rlistlong);
1073 fprintf(fp, " %-14s\n", fn_bench_tprs[j]);
1075 /* Make it clear to the user that some additional settings were modified */
1076 if (!is_equal(ir->rvdw, info->rvdw[0])
1077 || !is_equal(ir->rlistlong, info->rlistlong[0]) )
1084 fprintf(fp, "\nNote that in addition to the Coulomb radius and the Fourier grid\n"
1085 "other input settings were also changed (see table above).\n"
1086 "Please check if the modified settings are appropriate.\n");
1094 /* Rename the files we want to keep to some meaningful filename and
1095 * delete the rest */
1096 static void cleanup(const t_filenm *fnm, int nfile, int k, int nnodes,
1097 int nPMEnodes, int nr, gmx_bool bKeepStderr)
1099 char numstring[STRLEN];
1100 char newfilename[STRLEN];
1101 const char *fn = NULL;
1106 fprintf(stdout, "Cleaning up, deleting benchmark temp files ...\n");
1108 for (i = 0; i < nfile; i++)
1110 opt = (char *)fnm[i].opt;
1111 if (strcmp(opt, "-p") == 0)
1113 /* do nothing; keep this file */
1116 else if (strcmp(opt, "-bg") == 0)
1118 /* Give the log file a nice name so one can later see which parameters were used */
1119 numstring[0] = '\0';
1122 sprintf(numstring, "_%d", nr);
1124 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", opt2fn("-bg", nfile, fnm), k, nnodes, nPMEnodes, numstring);
1125 if (gmx_fexist(opt2fn("-bg", nfile, fnm)))
1127 fprintf(stdout, "renaming log file to %s\n", newfilename);
1128 make_backup(newfilename);
1129 rename(opt2fn("-bg", nfile, fnm), newfilename);
1132 else if (strcmp(opt, "-err") == 0)
1134 /* This file contains the output of stderr. We want to keep it in
1135 * cases where there have been problems. */
1136 fn = opt2fn(opt, nfile, fnm);
1137 numstring[0] = '\0';
1140 sprintf(numstring, "_%d", nr);
1142 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", fn, k, nnodes, nPMEnodes, numstring);
1147 fprintf(stdout, "Saving stderr output in %s\n", newfilename);
1148 make_backup(newfilename);
1149 rename(fn, newfilename);
1153 fprintf(stdout, "Deleting %s\n", fn);
1158 /* Delete the files which are created for each benchmark run: (options -b*) */
1159 else if ( (0 == strncmp(opt, "-b", 2)) && (opt2bSet(opt, nfile, fnm) || !is_optional(&fnm[i])) )
1161 fn = opt2fn(opt, nfile, fnm);
1164 fprintf(stdout, "Deleting %s\n", fn);
1172 /* Returns the largest common factor of n1 and n2 */
1173 static int largest_common_factor(int n1, int n2)
1178 for (factor = nmax; factor > 0; factor--)
1180 if (0 == (n1 % factor) && 0 == (n2 % factor) )
1185 return 0; /* one for the compiler */
1189 eNpmeAuto, eNpmeAll, eNpmeReduced, eNpmeSubset, eNpmeNr
1192 /* Create a list of numbers of PME nodes to test */
1193 static void make_npme_list(
1194 const char *npmevalues_opt, /* Make a complete list with all
1195 * possibilities or a short list that keeps only
1196 * reasonable numbers of PME nodes */
1197 int *nentries, /* Number of entries we put in the nPMEnodes list */
1198 int *nPMEnodes[], /* Each entry contains the value for -npme */
1199 int nnodes, /* Total number of nodes to do the tests on */
1200 int minPMEnodes, /* Minimum number of PME nodes */
1201 int maxPMEnodes) /* Maximum number of PME nodes */
1204 int min_factor = 1; /* We request that npp and npme have this minimal
1205 * largest common factor (depends on npp) */
1206 int nlistmax; /* Max. list size */
1207 int nlist; /* Actual number of entries in list */
1211 /* Do we need to check all possible values for -npme or is a reduced list enough? */
1212 if (0 == strcmp(npmevalues_opt, "all") )
1216 else if (0 == strcmp(npmevalues_opt, "subset") )
1218 eNPME = eNpmeSubset;
1220 else /* "auto" or "range" */
1226 else if (nnodes < 128)
1228 eNPME = eNpmeReduced;
1232 eNPME = eNpmeSubset;
1236 /* Calculate how many entries we could possibly have (in case of -npme all) */
1239 nlistmax = maxPMEnodes - minPMEnodes + 3;
1240 if (0 == minPMEnodes)
1250 /* Now make the actual list which is at most of size nlist */
1251 snew(*nPMEnodes, nlistmax);
1252 nlist = 0; /* start counting again, now the real entries in the list */
1253 for (i = 0; i < nlistmax - 2; i++)
1255 npme = maxPMEnodes - i;
1266 /* For 2d PME we want a common largest factor of at least the cube
1267 * root of the number of PP nodes */
1268 min_factor = (int) pow(npp, 1.0/3.0);
1271 gmx_fatal(FARGS, "Unknown option for eNPME in make_npme_list");
1274 if (largest_common_factor(npp, npme) >= min_factor)
1276 (*nPMEnodes)[nlist] = npme;
1280 /* We always test 0 PME nodes and the automatic number */
1281 *nentries = nlist + 2;
1282 (*nPMEnodes)[nlist ] = 0;
1283 (*nPMEnodes)[nlist+1] = -1;
1285 fprintf(stderr, "Will try the following %d different values for -npme:\n", *nentries);
1286 for (i = 0; i < *nentries-1; i++)
1288 fprintf(stderr, "%d, ", (*nPMEnodes)[i]);
1290 fprintf(stderr, "and %d (auto).\n", (*nPMEnodes)[*nentries-1]);
1294 /* Allocate memory to store the performance data */
1295 static void init_perfdata(t_perf *perfdata[], int ntprs, int datasets, int repeats)
1300 for (k = 0; k < ntprs; k++)
1302 snew(perfdata[k], datasets);
1303 for (i = 0; i < datasets; i++)
1305 for (j = 0; j < repeats; j++)
1307 snew(perfdata[k][i].Gcycles, repeats);
1308 snew(perfdata[k][i].ns_per_day, repeats);
1309 snew(perfdata[k][i].PME_f_load, repeats);
1316 /* Check for errors on mdrun -h */
1317 static void make_sure_it_runs(char *mdrun_cmd_line, int length, FILE *fp)
1319 char *command, *msg;
1323 snew(command, length + 15);
1324 snew(msg, length + 500);
1326 fprintf(stdout, "Making sure the benchmarks can be executed ...\n");
1327 /* FIXME: mdrun -h no longer actually does anything useful.
1328 * It unconditionally prints the help, ignoring all other options. */
1329 sprintf(command, "%s-h -quiet", mdrun_cmd_line);
1330 ret = gmx_system_call(command);
1334 /* To prevent confusion, do not again issue a gmx_fatal here since we already
1335 * get the error message from mdrun itself */
1336 sprintf(msg, "Cannot run the benchmark simulations! Please check the error message of\n"
1337 "mdrun for the source of the problem. Did you provide a command line\n"
1338 "argument that neither g_tune_pme nor mdrun understands? Offending command:\n"
1339 "\n%s\n\n", command);
1341 fprintf(stderr, "%s", msg);
1343 fprintf(fp, "%s", msg);
1353 static void do_the_tests(
1354 FILE *fp, /* General g_tune_pme output file */
1355 char **tpr_names, /* Filenames of the input files to test */
1356 int maxPMEnodes, /* Max fraction of nodes to use for PME */
1357 int minPMEnodes, /* Min fraction of nodes to use for PME */
1358 int npme_fixed, /* If >= -1, test fixed number of PME
1360 const char *npmevalues_opt, /* Which -npme values should be tested */
1361 t_perf **perfdata, /* Here the performace data is stored */
1362 int *pmeentries, /* Entries in the nPMEnodes list */
1363 int repeats, /* Repeat each test this often */
1364 int nnodes, /* Total number of nodes = nPP + nPME */
1365 int nr_tprs, /* Total number of tpr files to test */
1366 gmx_bool bThreads, /* Threads or MPI? */
1367 char *cmd_mpirun, /* mpirun command string */
1368 char *cmd_np, /* "-np", "-n", whatever mpirun needs */
1369 char *cmd_mdrun, /* mdrun command string */
1370 char *cmd_args_bench, /* arguments for mdrun in a string */
1371 const t_filenm *fnm, /* List of filenames from command line */
1372 int nfile, /* Number of files specified on the cmdl. */
1373 int presteps, /* DLB equilibration steps, is checked */
1374 gmx_int64_t cpt_steps) /* Time step counter in the checkpoint */
1376 int i, nr, k, ret, count = 0, totaltests;
1377 int *nPMEnodes = NULL;
1380 char *command, *cmd_stub;
1382 gmx_bool bResetProblem = FALSE;
1383 gmx_bool bFirst = TRUE;
1386 /* This string array corresponds to the eParselog enum type at the start
1388 const char* ParseLog[] = {
1390 "Logfile not found!",
1391 "No timings, logfile truncated?",
1392 "Run was terminated.",
1393 "Counters were not reset properly.",
1394 "No DD grid found for these settings.",
1395 "TPX version conflict!",
1396 "mdrun was not started in parallel!",
1397 "Number of PP nodes has a prime factor that is too large.",
1400 char str_PME_f_load[13];
1403 /* Allocate space for the mdrun command line. 100 extra characters should
1404 be more than enough for the -npme etcetera arguments */
1405 cmdline_length = strlen(cmd_mpirun)
1408 + strlen(cmd_args_bench)
1409 + strlen(tpr_names[0]) + 100;
1410 snew(command, cmdline_length);
1411 snew(cmd_stub, cmdline_length);
1413 /* Construct the part of the command line that stays the same for all tests: */
1416 sprintf(cmd_stub, "%s%s", cmd_mdrun, cmd_np);
1420 sprintf(cmd_stub, "%s%s%s ", cmd_mpirun, cmd_np, cmd_mdrun);
1423 /* Create a list of numbers of PME nodes to test */
1424 if (npme_fixed < -1)
1426 make_npme_list(npmevalues_opt, pmeentries, &nPMEnodes,
1427 nnodes, minPMEnodes, maxPMEnodes);
1433 nPMEnodes[0] = npme_fixed;
1434 fprintf(stderr, "Will use a fixed number of %d PME-only nodes.\n", nPMEnodes[0]);
1439 fprintf(fp, "\nNo benchmarks done since number of repeats (-r) is 0.\n");
1441 finalize(opt2fn("-p", nfile, fnm));
1445 /* Allocate one dataset for each tpr input file: */
1446 init_perfdata(perfdata, nr_tprs, *pmeentries, repeats);
1448 /*****************************************/
1449 /* Main loop over all tpr files to test: */
1450 /*****************************************/
1451 totaltests = nr_tprs*(*pmeentries)*repeats;
1452 for (k = 0; k < nr_tprs; k++)
1454 fprintf(fp, "\nIndividual timings for input file %d (%s):\n", k, tpr_names[k]);
1455 fprintf(fp, "PME nodes Gcycles ns/day PME/f Remark\n");
1456 /* Loop over various numbers of PME nodes: */
1457 for (i = 0; i < *pmeentries; i++)
1459 pd = &perfdata[k][i];
1461 /* Loop over the repeats for each scenario: */
1462 for (nr = 0; nr < repeats; nr++)
1464 pd->nPMEnodes = nPMEnodes[i];
1466 /* Add -npme and -s to the command line and save it. Note that
1467 * the -passall (if set) options requires cmd_args_bench to be
1468 * at the end of the command line string */
1469 snew(pd->mdrun_cmd_line, cmdline_length);
1470 sprintf(pd->mdrun_cmd_line, "%s-npme %d -s %s %s",
1471 cmd_stub, pd->nPMEnodes, tpr_names[k], cmd_args_bench);
1473 /* To prevent that all benchmarks fail due to a show-stopper argument
1474 * on the mdrun command line, we make a quick check with mdrun -h first */
1477 make_sure_it_runs(pd->mdrun_cmd_line, cmdline_length, fp);
1481 /* Do a benchmark simulation: */
1484 sprintf(buf, ", pass %d/%d", nr+1, repeats);
1490 fprintf(stdout, "\n=== Progress %2.0f%%, tpr %d/%d, run %d/%d%s:\n",
1491 (100.0*count)/totaltests,
1492 k+1, nr_tprs, i+1, *pmeentries, buf);
1493 make_backup(opt2fn("-err", nfile, fnm));
1494 sprintf(command, "%s 1> /dev/null 2>%s", pd->mdrun_cmd_line, opt2fn("-err", nfile, fnm));
1495 fprintf(stdout, "%s\n", pd->mdrun_cmd_line);
1496 gmx_system_call(command);
1498 /* Collect the performance data from the log file; also check stderr
1499 * for fatal errors */
1500 ret = parse_logfile(opt2fn("-bg", nfile, fnm), opt2fn("-err", nfile, fnm),
1501 pd, nr, presteps, cpt_steps, nnodes);
1502 if ((presteps > 0) && (ret == eParselogResetProblem))
1504 bResetProblem = TRUE;
1507 if (-1 == pd->nPMEnodes)
1509 sprintf(buf, "(%3d)", pd->guessPME);
1517 if (pd->PME_f_load[nr] > 0.0)
1519 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load[nr]);
1523 sprintf(str_PME_f_load, "%s", " - ");
1526 /* Write the data we got to disk */
1527 fprintf(fp, "%4d%s %12.3f %12.3f %s %s", pd->nPMEnodes,
1528 buf, pd->Gcycles[nr], pd->ns_per_day[nr], str_PME_f_load, ParseLog[ret]);
1529 if (!(ret == eParselogOK || ret == eParselogNoDDGrid || ret == eParselogNotFound) )
1531 fprintf(fp, " Check %s file for problems.", ret == eParselogFatal ? "err" : "log");
1537 /* Do some cleaning up and delete the files we do not need any more */
1538 cleanup(fnm, nfile, k, nnodes, pd->nPMEnodes, nr, ret == eParselogFatal);
1540 /* If the first run with this number of processors already failed, do not try again: */
1541 if (pd->Gcycles[0] <= 0.0 && repeats > 1)
1543 fprintf(stdout, "Skipping remaining passes of unsuccessful setting, see log file for details.\n");
1544 count += repeats-(nr+1);
1547 } /* end of repeats loop */
1548 } /* end of -npme loop */
1549 } /* end of tpr file loop */
1554 fprintf(fp, "WARNING: The cycle and time step counters could not be reset properly. ");
1562 static void check_input(
1569 real maxPMEfraction,
1570 real minPMEfraction,
1572 gmx_int64_t bench_nsteps,
1573 const t_filenm *fnm,
1583 /* Make sure the input file exists */
1584 if (!gmx_fexist(opt2fn("-s", nfile, fnm)))
1586 gmx_fatal(FARGS, "File %s not found.", opt2fn("-s", nfile, fnm));
1589 /* Make sure that the checkpoint file is not overwritten during benchmarking */
1590 if ( (0 == strcmp(opt2fn("-cpi", nfile, fnm), opt2fn("-bcpo", nfile, fnm)) ) && (sim_part > 1) )
1592 gmx_fatal(FARGS, "Checkpoint input (-cpi) and benchmark checkpoint output (-bcpo) files must not be identical.\n"
1593 "The checkpoint input file must not be overwritten during the benchmarks.\n");
1596 /* Make sure that repeats is >= 0 (if == 0, only write tpr files) */
1599 gmx_fatal(FARGS, "Number of repeats < 0!");
1602 /* Check number of nodes */
1605 gmx_fatal(FARGS, "Number of nodes/threads must be a positive integer.");
1608 /* Automatically choose -ntpr if not set */
1618 /* Set a reasonable scaling factor for rcoulomb */
1621 *rmax = rcoulomb * 1.2;
1624 fprintf(stderr, "Will test %d tpr file%s.\n", *ntprs, *ntprs == 1 ? "" : "s");
1630 fprintf(stderr, "Note: Choose ntpr>1 to shift PME load between real and reciprocal space.\n");
1634 /* Make shure that rmin <= rcoulomb <= rmax */
1643 if (!(*rmin <= *rmax) )
1645 gmx_fatal(FARGS, "Please choose the Coulomb radii such that rmin <= rmax.\n"
1646 "rmin = %g, rmax = %g, actual rcoul from .tpr file = %g\n", *rmin, *rmax, rcoulomb);
1648 /* Add test scenarios if rmin or rmax were set */
1651 if (!is_equal(*rmin, rcoulomb) && (*ntprs == 1) )
1654 fprintf(stderr, "NOTE: Setting -rmin to %g changed -ntpr to %d\n",
1657 if (!is_equal(*rmax, rcoulomb) && (*ntprs == 1) )
1660 fprintf(stderr, "NOTE: Setting -rmax to %g changed -ntpr to %d\n",
1665 /* If one of rmin, rmax is set, we need 2 tpr files at minimum */
1666 if (!is_equal(*rmax, rcoulomb) || !is_equal(*rmin, rcoulomb) )
1668 *ntprs = max(*ntprs, 2);
1671 /* If both rmin, rmax are set, we need 3 tpr files at minimum */
1672 if (!is_equal(*rmax, rcoulomb) && !is_equal(*rmin, rcoulomb) )
1674 *ntprs = max(*ntprs, 3);
1679 fprintf(stderr, "NOTE: Your rmin, rmax setting changed -ntpr to %d\n", *ntprs);
1684 if (is_equal(*rmin, rcoulomb) && is_equal(rcoulomb, *rmax)) /* We have just a single rc */
1686 fprintf(stderr, "WARNING: Resetting -ntpr to 1 since no Coulomb radius scaling is requested.\n"
1687 "Please set rmin < rmax to test Coulomb radii in the [rmin, rmax] interval\n"
1688 "with correspondingly adjusted PME grid settings\n");
1693 /* Check whether max and min fraction are within required values */
1694 if (maxPMEfraction > 0.5 || maxPMEfraction < 0)
1696 gmx_fatal(FARGS, "-max must be between 0 and 0.5");
1698 if (minPMEfraction > 0.5 || minPMEfraction < 0)
1700 gmx_fatal(FARGS, "-min must be between 0 and 0.5");
1702 if (maxPMEfraction < minPMEfraction)
1704 gmx_fatal(FARGS, "-max must be larger or equal to -min");
1707 /* Check whether the number of steps - if it was set - has a reasonable value */
1708 if (bench_nsteps < 0)
1710 gmx_fatal(FARGS, "Number of steps must be positive.");
1713 if (bench_nsteps > 10000 || bench_nsteps < 100)
1715 fprintf(stderr, "WARNING: steps=");
1716 fprintf(stderr, "%"GMX_PRId64, bench_nsteps);
1717 fprintf(stderr, ". Are you sure you want to perform so %s steps for each benchmark?\n", (bench_nsteps < 100) ? "few" : "many");
1722 gmx_fatal(FARGS, "Cannot have a negative number of presteps.\n");
1725 /* Check for rcoulomb scaling if more than one .tpr file is tested */
1728 if (*rmin/rcoulomb < 0.75 || *rmax/rcoulomb > 1.25)
1730 fprintf(stderr, "WARNING: Applying extreme scaling factor. I hope you know what you are doing.\n");
1734 /* If a fixed number of PME nodes is set we do rcoulomb and PME gird tuning
1735 * only. We need to check whether the requested number of PME-only nodes
1737 if (npme_fixed > -1)
1739 /* No more than 50% of all nodes can be assigned as PME-only nodes. */
1740 if (2*npme_fixed > nnodes)
1742 gmx_fatal(FARGS, "Cannot have more than %d PME-only nodes for a total of %d nodes (you chose %d).\n",
1743 nnodes/2, nnodes, npme_fixed);
1745 if ((npme_fixed > 0) && (5*npme_fixed < nnodes))
1747 fprintf(stderr, "WARNING: Only %g percent of the nodes are assigned as PME-only nodes.\n",
1748 100.0*((real)npme_fixed / (real)nnodes));
1750 if (opt2parg_bSet("-min", npargs, pa) || opt2parg_bSet("-max", npargs, pa))
1752 fprintf(stderr, "NOTE: The -min, -max, and -npme options have no effect when a\n"
1753 " fixed number of PME-only nodes is requested with -fix.\n");
1759 /* Returns TRUE when "opt" is needed at launch time */
1760 static gmx_bool is_launch_file(char *opt, gmx_bool bSet)
1762 if (0 == strncmp(opt, "-swap", 5))
1767 /* Apart from the input .tpr and the output log files we need all options that
1768 * were set on the command line and that do not start with -b */
1769 if (0 == strncmp(opt, "-b", 2) || 0 == strncmp(opt, "-s", 2)
1770 || 0 == strncmp(opt, "-err", 4) || 0 == strncmp(opt, "-p", 2) )
1779 /* Returns TRUE when "opt" defines a file which is needed for the benchmarks runs */
1780 static gmx_bool is_bench_file(char *opt, gmx_bool bSet, gmx_bool bOptional, gmx_bool bIsOutput)
1782 /* Apart from the input .tpr, all files starting with "-b" are for
1783 * _b_enchmark files exclusively */
1784 if (0 == strncmp(opt, "-s", 2))
1789 if (0 == strncmp(opt, "-b", 2) || 0 == strncmp(opt, "-s", 2))
1791 if (!bOptional || bSet)
1808 if (bSet) /* These are additional input files like -cpi -ei */
1821 /* Adds 'buf' to 'str' */
1822 static void add_to_string(char **str, char *buf)
1827 len = strlen(*str) + strlen(buf) + 1;
1833 /* Create the command line for the benchmark as well as for the real run */
1834 static void create_command_line_snippets(
1835 gmx_bool bAppendFiles,
1836 gmx_bool bKeepAndNumCPT,
1837 gmx_bool bResetHWay,
1841 char *cmd_args_bench[], /* command line arguments for benchmark runs */
1842 char *cmd_args_launch[], /* command line arguments for simulation run */
1843 char extra_args[]) /* Add this to the end of the command line */
1848 char strbuf[STRLEN];
1851 /* strlen needs at least '\0' as a string: */
1852 snew(*cmd_args_bench, 1);
1853 snew(*cmd_args_launch, 1);
1854 *cmd_args_launch[0] = '\0';
1855 *cmd_args_bench[0] = '\0';
1858 /*******************************************/
1859 /* 1. Process other command line arguments */
1860 /*******************************************/
1863 /* Add equilibration steps to benchmark options */
1864 sprintf(strbuf, "-resetstep %d ", presteps);
1865 add_to_string(cmd_args_bench, strbuf);
1867 /* These switches take effect only at launch time */
1868 if (FALSE == bAppendFiles)
1870 add_to_string(cmd_args_launch, "-noappend ");
1874 add_to_string(cmd_args_launch, "-cpnum ");
1878 add_to_string(cmd_args_launch, "-resethway ");
1881 /********************/
1882 /* 2. Process files */
1883 /********************/
1884 for (i = 0; i < nfile; i++)
1886 opt = (char *)fnm[i].opt;
1887 name = opt2fn(opt, nfile, fnm);
1889 /* Strbuf contains the options, now let's sort out where we need that */
1890 sprintf(strbuf, "%s %s ", opt, name);
1892 if (is_bench_file(opt, opt2bSet(opt, nfile, fnm), is_optional(&fnm[i]), is_output(&fnm[i])) )
1894 /* All options starting with -b* need the 'b' removed,
1895 * therefore overwrite strbuf */
1896 if (0 == strncmp(opt, "-b", 2))
1898 sprintf(strbuf, "-%s %s ", &opt[2], name);
1901 add_to_string(cmd_args_bench, strbuf);
1904 if (is_launch_file(opt, opt2bSet(opt, nfile, fnm)) )
1906 add_to_string(cmd_args_launch, strbuf);
1910 add_to_string(cmd_args_bench, extra_args);
1911 add_to_string(cmd_args_launch, extra_args);
1915 /* Set option opt */
1916 static void setopt(const char *opt, int nfile, t_filenm fnm[])
1920 for (i = 0; (i < nfile); i++)
1922 if (strcmp(opt, fnm[i].opt) == 0)
1924 fnm[i].flag |= ffSET;
1930 /* This routine inspects the tpr file and ...
1931 * 1. checks for output files that get triggered by a tpr option. These output
1932 * files are marked as 'set' to allow for a proper cleanup after each
1934 * 2. returns the PME:PP load ratio
1935 * 3. returns rcoulomb from the tpr */
1936 static float inspect_tpr(int nfile, t_filenm fnm[], real *rcoulomb)
1938 gmx_bool bPull; /* Is pulling requested in .tpr file? */
1939 gmx_bool bTpi; /* Is test particle insertion requested? */
1940 gmx_bool bFree; /* Is a free energy simulation requested? */
1941 gmx_bool bNM; /* Is a normal mode analysis requested? */
1942 gmx_bool bSwap; /* Is water/ion position swapping requested? */
1948 /* Check tpr file for options that trigger extra output files */
1949 read_tpx_state(opt2fn("-s", nfile, fnm), &ir, &state, NULL, &mtop);
1950 bPull = (epullNO != ir.ePull);
1951 bFree = (efepNO != ir.efep );
1952 bNM = (eiNM == ir.eI );
1953 bSwap = (eswapNO != ir.eSwapCoords);
1954 bTpi = EI_TPI(ir.eI);
1956 /* Set these output files on the tuning command-line */
1959 setopt("-pf", nfile, fnm);
1960 setopt("-px", nfile, fnm);
1964 setopt("-dhdl", nfile, fnm);
1968 setopt("-tpi", nfile, fnm);
1969 setopt("-tpid", nfile, fnm);
1973 setopt("-mtx", nfile, fnm);
1977 setopt("-swap", nfile, fnm);
1980 *rcoulomb = ir.rcoulomb;
1982 /* Return the estimate for the number of PME nodes */
1983 return pme_load_estimate(&mtop, &ir, state.box);
1987 static void couple_files_options(int nfile, t_filenm fnm[])
1990 gmx_bool bSet, bBench;
1995 for (i = 0; i < nfile; i++)
1997 opt = (char *)fnm[i].opt;
1998 bSet = ((fnm[i].flag & ffSET) != 0);
1999 bBench = (0 == strncmp(opt, "-b", 2));
2001 /* Check optional files */
2002 /* If e.g. -eo is set, then -beo also needs to be set */
2003 if (is_optional(&fnm[i]) && bSet && !bBench)
2005 sprintf(buf, "-b%s", &opt[1]);
2006 setopt(buf, nfile, fnm);
2008 /* If -beo is set, then -eo also needs to be! */
2009 if (is_optional(&fnm[i]) && bSet && bBench)
2011 sprintf(buf, "-%s", &opt[2]);
2012 setopt(buf, nfile, fnm);
2018 #define BENCHSTEPS (1000)
2020 int gmx_tune_pme(int argc, char *argv[])
2022 const char *desc[] = {
2023 "For a given number [TT]-np[tt] or [TT]-ntmpi[tt] of processors/threads, [THISMODULE] systematically",
2024 "times [gmx-mdrun] with various numbers of PME-only nodes and determines",
2025 "which setting is fastest. It will also test whether performance can",
2026 "be enhanced by shifting load from the reciprocal to the real space",
2027 "part of the Ewald sum. ",
2028 "Simply pass your [TT].tpr[tt] file to [THISMODULE] together with other options",
2029 "for [gmx-mdrun] as needed.[PAR]",
2030 "Which executables are used can be set in the environment variables",
2031 "MPIRUN and MDRUN. If these are not present, 'mpirun' and 'mdrun'",
2032 "will be used as defaults. Note that for certain MPI frameworks you",
2033 "need to provide a machine- or hostfile. This can also be passed",
2034 "via the MPIRUN variable, e.g.[PAR]",
2035 "[TT]export MPIRUN=\"/usr/local/mpirun -machinefile hosts\"[tt][PAR]",
2036 "Please call [THISMODULE] with the normal options you would pass to",
2037 "[gmx-mdrun] and add [TT]-np[tt] for the number of processors to perform the",
2038 "tests on, or [TT]-ntmpi[tt] for the number of threads. You can also add [TT]-r[tt]",
2039 "to repeat each test several times to get better statistics. [PAR]",
2040 "[THISMODULE] can test various real space / reciprocal space workloads",
2041 "for you. With [TT]-ntpr[tt] you control how many extra [TT].tpr[tt] files will be",
2042 "written with enlarged cutoffs and smaller Fourier grids respectively.",
2043 "Typically, the first test (number 0) will be with the settings from the input",
2044 "[TT].tpr[tt] file; the last test (number [TT]ntpr[tt]) will have the Coulomb cutoff",
2045 "specified by [TT]-rmax[tt] with a somwhat smaller PME grid at the same time. ",
2046 "In this last test, the Fourier spacing is multiplied with [TT]rmax[tt]/rcoulomb. ",
2047 "The remaining [TT].tpr[tt] files will have equally-spaced Coulomb radii (and Fourier "
2048 "spacings) between these extremes. [BB]Note[bb] that you can set [TT]-ntpr[tt] to 1",
2049 "if you just seek the optimal number of PME-only nodes; in that case",
2050 "your input [TT].tpr[tt] file will remain unchanged.[PAR]",
2051 "For the benchmark runs, the default of 1000 time steps should suffice for most",
2052 "MD systems. The dynamic load balancing needs about 100 time steps",
2053 "to adapt to local load imbalances, therefore the time step counters",
2054 "are by default reset after 100 steps. For large systems (>1M atoms), as well as ",
2055 "for a higher accuarcy of the measurements, you should set [TT]-resetstep[tt] to a higher value.",
2056 "From the 'DD' load imbalance entries in the md.log output file you",
2057 "can tell after how many steps the load is sufficiently balanced. Example call:[PAR]"
2058 "[TT]gmx tune_pme -np 64 -s protein.tpr -launch[tt][PAR]",
2059 "After calling [gmx-mdrun] several times, detailed performance information",
2060 "is available in the output file [TT]perf.out[tt].",
2061 "[BB]Note[bb] that during the benchmarks, a couple of temporary files are written",
2062 "(options [TT]-b[tt]*), these will be automatically deleted after each test.[PAR]",
2063 "If you want the simulation to be started automatically with the",
2064 "optimized parameters, use the command line option [TT]-launch[tt].[PAR]",
2069 int pmeentries = 0; /* How many values for -npme do we actually test for each tpr file */
2070 real maxPMEfraction = 0.50;
2071 real minPMEfraction = 0.25;
2072 int maxPMEnodes, minPMEnodes;
2073 float guessPMEratio; /* guessed PME:PP ratio based on the tpr file */
2074 float guessPMEnodes;
2075 int npme_fixed = -2; /* If >= -1, use only this number
2076 * of PME-only nodes */
2078 real rmin = 0.0, rmax = 0.0; /* min and max value for rcoulomb if scaling is requested */
2079 real rcoulomb = -1.0; /* Coulomb radius as set in .tpr file */
2080 gmx_bool bScaleRvdw = TRUE;
2081 gmx_int64_t bench_nsteps = BENCHSTEPS;
2082 gmx_int64_t new_sim_nsteps = -1; /* -1 indicates: not set by the user */
2083 gmx_int64_t cpt_steps = 0; /* Step counter in .cpt input file */
2084 int presteps = 100; /* Do a full cycle reset after presteps steps */
2085 gmx_bool bOverwrite = FALSE, bKeepTPR;
2086 gmx_bool bLaunch = FALSE;
2087 char *ExtraArgs = NULL;
2088 char **tpr_names = NULL;
2089 const char *simulation_tpr = NULL;
2090 int best_npme, best_tpr;
2091 int sim_part = 1; /* For benchmarks with checkpoint files */
2094 /* Default program names if nothing else is found */
2095 char *cmd_mpirun = NULL, *cmd_mdrun = NULL;
2096 char *cmd_args_bench, *cmd_args_launch;
2097 char *cmd_np = NULL;
2099 t_perf **perfdata = NULL;
2105 /* Print out how long the tuning took */
2108 static t_filenm fnm[] = {
2110 { efOUT, "-p", "perf", ffWRITE },
2111 { efLOG, "-err", "bencherr", ffWRITE },
2112 { efTPX, "-so", "tuned", ffWRITE },
2114 { efTPX, NULL, NULL, ffREAD },
2115 { efTRN, "-o", NULL, ffWRITE },
2116 { efCOMPRESSED, "-x", NULL, ffOPTWR },
2117 { efCPT, "-cpi", NULL, ffOPTRD },
2118 { efCPT, "-cpo", NULL, ffOPTWR },
2119 { efSTO, "-c", "confout", ffWRITE },
2120 { efEDR, "-e", "ener", ffWRITE },
2121 { efLOG, "-g", "md", ffWRITE },
2122 { efXVG, "-dhdl", "dhdl", ffOPTWR },
2123 { efXVG, "-field", "field", ffOPTWR },
2124 { efXVG, "-table", "table", ffOPTRD },
2125 { efXVG, "-tabletf", "tabletf", ffOPTRD },
2126 { efXVG, "-tablep", "tablep", ffOPTRD },
2127 { efXVG, "-tableb", "table", ffOPTRD },
2128 { efTRX, "-rerun", "rerun", ffOPTRD },
2129 { efXVG, "-tpi", "tpi", ffOPTWR },
2130 { efXVG, "-tpid", "tpidist", ffOPTWR },
2131 { efEDI, "-ei", "sam", ffOPTRD },
2132 { efXVG, "-eo", "edsam", ffOPTWR },
2133 { efXVG, "-devout", "deviatie", ffOPTWR },
2134 { efXVG, "-runav", "runaver", ffOPTWR },
2135 { efXVG, "-px", "pullx", ffOPTWR },
2136 { efXVG, "-pf", "pullf", ffOPTWR },
2137 { efXVG, "-ro", "rotation", ffOPTWR },
2138 { efLOG, "-ra", "rotangles", ffOPTWR },
2139 { efLOG, "-rs", "rotslabs", ffOPTWR },
2140 { efLOG, "-rt", "rottorque", ffOPTWR },
2141 { efMTX, "-mtx", "nm", ffOPTWR },
2142 { efNDX, "-dn", "dipole", ffOPTWR },
2143 { efXVG, "-swap", "swapions", ffOPTWR },
2144 /* Output files that are deleted after each benchmark run */
2145 { efTRN, "-bo", "bench", ffWRITE },
2146 { efXTC, "-bx", "bench", ffWRITE },
2147 { efCPT, "-bcpo", "bench", ffWRITE },
2148 { efSTO, "-bc", "bench", ffWRITE },
2149 { efEDR, "-be", "bench", ffWRITE },
2150 { efLOG, "-bg", "bench", ffWRITE },
2151 { efXVG, "-beo", "benchedo", ffOPTWR },
2152 { efXVG, "-bdhdl", "benchdhdl", ffOPTWR },
2153 { efXVG, "-bfield", "benchfld", ffOPTWR },
2154 { efXVG, "-btpi", "benchtpi", ffOPTWR },
2155 { efXVG, "-btpid", "benchtpid", ffOPTWR },
2156 { efXVG, "-bdevout", "benchdev", ffOPTWR },
2157 { efXVG, "-brunav", "benchrnav", ffOPTWR },
2158 { efXVG, "-bpx", "benchpx", ffOPTWR },
2159 { efXVG, "-bpf", "benchpf", ffOPTWR },
2160 { efXVG, "-bro", "benchrot", ffOPTWR },
2161 { efLOG, "-bra", "benchrota", ffOPTWR },
2162 { efLOG, "-brs", "benchrots", ffOPTWR },
2163 { efLOG, "-brt", "benchrott", ffOPTWR },
2164 { efMTX, "-bmtx", "benchn", ffOPTWR },
2165 { efNDX, "-bdn", "bench", ffOPTWR },
2166 { efXVG, "-bswap", "benchswp", ffOPTWR }
2169 gmx_bool bThreads = FALSE;
2173 const char *procstring[] =
2174 { NULL, "-np", "-n", "none", NULL };
2175 const char *npmevalues_opt[] =
2176 { NULL, "auto", "all", "subset", NULL };
2178 gmx_bool bAppendFiles = TRUE;
2179 gmx_bool bKeepAndNumCPT = FALSE;
2180 gmx_bool bResetCountersHalfWay = FALSE;
2181 gmx_bool bBenchmark = TRUE;
2183 output_env_t oenv = NULL;
2186 /***********************/
2187 /* g_tune_pme options: */
2188 /***********************/
2189 { "-np", FALSE, etINT, {&nnodes},
2190 "Number of nodes to run the tests on (must be > 2 for separate PME nodes)" },
2191 { "-npstring", FALSE, etENUM, {procstring},
2192 "Specify the number of processors to [TT]$MPIRUN[tt] using this string"},
2193 { "-ntmpi", FALSE, etINT, {&nthreads},
2194 "Number of MPI-threads to run the tests on (turns MPI & mpirun off)"},
2195 { "-r", FALSE, etINT, {&repeats},
2196 "Repeat each test this often" },
2197 { "-max", FALSE, etREAL, {&maxPMEfraction},
2198 "Max fraction of PME nodes to test with" },
2199 { "-min", FALSE, etREAL, {&minPMEfraction},
2200 "Min fraction of PME nodes to test with" },
2201 { "-npme", FALSE, etENUM, {npmevalues_opt},
2202 "Within -min and -max, benchmark all possible values for [TT]-npme[tt], or just a reasonable subset. "
2203 "Auto neglects -min and -max and chooses reasonable values around a guess for npme derived from the .tpr"},
2204 { "-fix", FALSE, etINT, {&npme_fixed},
2205 "If >= -1, do not vary the number of PME-only nodes, instead use this fixed value and only vary rcoulomb and the PME grid spacing."},
2206 { "-rmax", FALSE, etREAL, {&rmax},
2207 "If >0, maximal rcoulomb for -ntpr>1 (rcoulomb upscaling results in fourier grid downscaling)" },
2208 { "-rmin", FALSE, etREAL, {&rmin},
2209 "If >0, minimal rcoulomb for -ntpr>1" },
2210 { "-scalevdw", FALSE, etBOOL, {&bScaleRvdw},
2211 "Scale rvdw along with rcoulomb"},
2212 { "-ntpr", FALSE, etINT, {&ntprs},
2213 "Number of [TT].tpr[tt] files to benchmark. Create this many files with different rcoulomb scaling factors depending on -rmin and -rmax. "
2214 "If < 1, automatically choose the number of [TT].tpr[tt] files to test" },
2215 { "-steps", FALSE, etINT64, {&bench_nsteps},
2216 "Take timings for this many steps in the benchmark runs" },
2217 { "-resetstep", FALSE, etINT, {&presteps},
2218 "Let dlb equilibrate this many steps before timings are taken (reset cycle counters after this many steps)" },
2219 { "-simsteps", FALSE, etINT64, {&new_sim_nsteps},
2220 "If non-negative, perform this many steps in the real run (overwrites nsteps from [TT].tpr[tt], add [TT].cpt[tt] steps)" },
2221 { "-launch", FALSE, etBOOL, {&bLaunch},
2222 "Launch the real simulation after optimization" },
2223 { "-bench", FALSE, etBOOL, {&bBenchmark},
2224 "Run the benchmarks or just create the input [TT].tpr[tt] files?" },
2225 /******************/
2226 /* mdrun options: */
2227 /******************/
2228 /* We let g_tune_pme parse and understand these options, because we need to
2229 * prevent that they appear on the mdrun command line for the benchmarks */
2230 { "-append", FALSE, etBOOL, {&bAppendFiles},
2231 "Append to previous output files when continuing from checkpoint instead of adding the simulation part number to all file names (for launch only)" },
2232 { "-cpnum", FALSE, etBOOL, {&bKeepAndNumCPT},
2233 "Keep and number checkpoint files (launch only)" },
2234 { "-resethway", FALSE, etBOOL, {&bResetCountersHalfWay},
2235 "HIDDENReset the cycle counters after half the number of steps or halfway [TT]-maxh[tt] (launch only)" }
2238 #define NFILE asize(fnm)
2240 seconds = gmx_gettime();
2242 if (!parse_common_args(&argc, argv, PCA_NOEXIT_ON_ARGS,
2243 NFILE, fnm, asize(pa), pa, asize(desc), desc,
2249 /* Store the remaining unparsed command line entries in a string which
2250 * is then attached to the mdrun command line */
2252 ExtraArgs[0] = '\0';
2253 for (i = 1; i < argc; i++) /* argc will now be 1 if everything was understood */
2255 add_to_string(&ExtraArgs, argv[i]);
2256 add_to_string(&ExtraArgs, " ");
2259 if (opt2parg_bSet("-ntmpi", asize(pa), pa))
2262 if (opt2parg_bSet("-npstring", asize(pa), pa))
2264 fprintf(stderr, "WARNING: -npstring has no effect when using threads.\n");
2269 gmx_fatal(FARGS, "Can't run multi-threaded MPI simulation yet!");
2271 /* and now we just set this; a bit of an ugly hack*/
2274 /* Check for PME:PP ratio and whether tpr triggers additional output files */
2275 guessPMEratio = inspect_tpr(NFILE, fnm, &rcoulomb);
2277 /* Automatically set -beo options if -eo is set etc. */
2278 couple_files_options(NFILE, fnm);
2280 /* Construct the command line arguments for benchmark runs
2281 * as well as for the simulation run */
2284 sprintf(bbuf, " -ntmpi %d ", nthreads);
2288 /* This string will be used for MPI runs and will appear after the
2289 * mpirun command. */
2290 if (strcmp(procstring[0], "none") != 0)
2292 sprintf(bbuf, " %s %d ", procstring[0], nnodes);
2302 create_command_line_snippets(bAppendFiles, bKeepAndNumCPT, bResetCountersHalfWay, presteps,
2303 NFILE, fnm, &cmd_args_bench, &cmd_args_launch, ExtraArgs);
2305 /* Read in checkpoint file if requested */
2307 if (opt2bSet("-cpi", NFILE, fnm))
2310 cr->duty = DUTY_PP; /* makes the following routine happy */
2311 read_checkpoint_simulation_part(opt2fn("-cpi", NFILE, fnm),
2312 &sim_part, &cpt_steps, cr,
2313 FALSE, NFILE, fnm, NULL, NULL);
2316 /* sim_part will now be 1 if no checkpoint file was found */
2319 gmx_fatal(FARGS, "Checkpoint file %s not found!", opt2fn("-cpi", NFILE, fnm));
2323 /* Open performance output file and write header info */
2324 fp = gmx_ffopen(opt2fn("-p", NFILE, fnm), "w");
2326 /* Make a quick consistency check of command line parameters */
2327 check_input(nnodes, repeats, &ntprs, &rmin, rcoulomb, &rmax,
2328 maxPMEfraction, minPMEfraction, npme_fixed,
2329 bench_nsteps, fnm, NFILE, sim_part, presteps,
2332 /* Determine the maximum and minimum number of PME nodes to test,
2333 * the actual list of settings is build in do_the_tests(). */
2334 if ((nnodes > 2) && (npme_fixed < -1))
2336 if (0 == strcmp(npmevalues_opt[0], "auto"))
2338 /* Determine the npme range automatically based on the PME:PP load guess */
2339 if (guessPMEratio > 1.0)
2341 /* More PME than PP work, probably we do not need separate PME nodes at all! */
2342 maxPMEnodes = nnodes/2;
2343 minPMEnodes = nnodes/2;
2347 /* PME : PP load is in the range 0..1, let's test around the guess */
2348 guessPMEnodes = nnodes/(1.0 + 1.0/guessPMEratio);
2349 minPMEnodes = floor(0.7*guessPMEnodes);
2350 maxPMEnodes = ceil(1.6*guessPMEnodes);
2351 maxPMEnodes = min(maxPMEnodes, nnodes/2);
2356 /* Determine the npme range based on user input */
2357 maxPMEnodes = floor(maxPMEfraction*nnodes);
2358 minPMEnodes = max(floor(minPMEfraction*nnodes), 0);
2359 fprintf(stdout, "Will try runs with %d ", minPMEnodes);
2360 if (maxPMEnodes != minPMEnodes)
2362 fprintf(stdout, "- %d ", maxPMEnodes);
2364 fprintf(stdout, "PME-only nodes.\n Note that the automatic number of PME-only nodes and no separate PME nodes are always tested.\n");
2373 /* Get the commands we need to set up the runs from environment variables */
2374 get_program_paths(bThreads, &cmd_mpirun, &cmd_mdrun);
2375 if (bBenchmark && repeats > 0)
2377 check_mdrun_works(bThreads, cmd_mpirun, cmd_np, cmd_mdrun);
2380 /* Print some header info to file */
2382 fprintf(fp, "\n P E R F O R M A N C E R E S U L T S\n");
2384 fprintf(fp, "%s for Gromacs %s\n", ShortProgram(), GromacsVersion());
2387 fprintf(fp, "Number of nodes : %d\n", nnodes);
2388 fprintf(fp, "The mpirun command is : %s\n", cmd_mpirun);
2389 if (strcmp(procstring[0], "none") != 0)
2391 fprintf(fp, "Passing # of nodes via : %s\n", procstring[0]);
2395 fprintf(fp, "Not setting number of nodes in system call\n");
2400 fprintf(fp, "Number of threads : %d\n", nnodes);
2403 fprintf(fp, "The mdrun command is : %s\n", cmd_mdrun);
2404 fprintf(fp, "mdrun args benchmarks : %s\n", cmd_args_bench);
2405 fprintf(fp, "Benchmark steps : ");
2406 fprintf(fp, "%"GMX_PRId64, bench_nsteps);
2408 fprintf(fp, "dlb equilibration steps : %d\n", presteps);
2411 fprintf(fp, "Checkpoint time step : ");
2412 fprintf(fp, "%"GMX_PRId64, cpt_steps);
2415 fprintf(fp, "mdrun args at launchtime: %s\n", cmd_args_launch);
2417 if (new_sim_nsteps >= 0)
2420 fprintf(stderr, "Note: Simulation input file %s will have ", opt2fn("-so", NFILE, fnm));
2421 fprintf(stderr, "%"GMX_PRId64, new_sim_nsteps+cpt_steps);
2422 fprintf(stderr, " steps.\n");
2423 fprintf(fp, "Simulation steps : ");
2424 fprintf(fp, "%"GMX_PRId64, new_sim_nsteps);
2429 fprintf(fp, "Repeats for each test : %d\n", repeats);
2432 if (npme_fixed >= -1)
2434 fprintf(fp, "Fixing -npme at : %d\n", npme_fixed);
2437 fprintf(fp, "Input file : %s\n", opt2fn("-s", NFILE, fnm));
2438 fprintf(fp, " PME/PP load estimate : %g\n", guessPMEratio);
2440 /* Allocate memory for the inputinfo struct: */
2442 info->nr_inputfiles = ntprs;
2443 for (i = 0; i < ntprs; i++)
2445 snew(info->rcoulomb, ntprs);
2446 snew(info->rvdw, ntprs);
2447 snew(info->rlist, ntprs);
2448 snew(info->rlistlong, ntprs);
2449 snew(info->nkx, ntprs);
2450 snew(info->nky, ntprs);
2451 snew(info->nkz, ntprs);
2452 snew(info->fsx, ntprs);
2453 snew(info->fsy, ntprs);
2454 snew(info->fsz, ntprs);
2456 /* Make alternative tpr files to test: */
2457 snew(tpr_names, ntprs);
2458 for (i = 0; i < ntprs; i++)
2460 snew(tpr_names[i], STRLEN);
2463 /* It can be that ntprs is reduced by make_benchmark_tprs if not enough
2464 * different grids could be found. */
2465 make_benchmark_tprs(opt2fn("-s", NFILE, fnm), tpr_names, bench_nsteps+presteps,
2466 cpt_steps, rmin, rmax, bScaleRvdw, &ntprs, info, fp);
2468 /********************************************************************************/
2469 /* Main loop over all scenarios we need to test: tpr files, PME nodes, repeats */
2470 /********************************************************************************/
2471 snew(perfdata, ntprs);
2474 do_the_tests(fp, tpr_names, maxPMEnodes, minPMEnodes, npme_fixed, npmevalues_opt[0], perfdata, &pmeentries,
2475 repeats, nnodes, ntprs, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2476 cmd_args_bench, fnm, NFILE, presteps, cpt_steps);
2478 fprintf(fp, "\nTuning took%8.1f minutes.\n", (gmx_gettime()-seconds)/60.0);
2480 /* Analyse the results and give a suggestion for optimal settings: */
2481 bKeepTPR = analyze_data(fp, opt2fn("-p", NFILE, fnm), perfdata, nnodes, ntprs, pmeentries,
2482 repeats, info, &best_tpr, &best_npme);
2484 /* Take the best-performing tpr file and enlarge nsteps to original value */
2485 if (bKeepTPR && !bOverwrite)
2487 simulation_tpr = opt2fn("-s", NFILE, fnm);
2491 simulation_tpr = opt2fn("-so", NFILE, fnm);
2492 modify_PMEsettings(bOverwrite ? (new_sim_nsteps+cpt_steps) : info->orig_sim_steps,
2493 info->orig_init_step, tpr_names[best_tpr], simulation_tpr);
2496 /* Let's get rid of the temporary benchmark input files */
2497 for (i = 0; i < ntprs; i++)
2499 fprintf(stdout, "Deleting temporary benchmark input file %s\n", tpr_names[i]);
2500 remove(tpr_names[i]);
2503 /* Now start the real simulation if the user requested it ... */
2504 launch_simulation(bLaunch, fp, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2505 cmd_args_launch, simulation_tpr, best_npme);
2509 /* ... or simply print the performance results to screen: */
2512 finalize(opt2fn("-p", NFILE, fnm));