3 * This source code is part of
7 * GROningen MAchine for Chemical Simulations
9 * Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
10 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
11 * Copyright (c) 2001-2008, The GROMACS development team,
12 * check out http://www.gromacs.org for more information.
14 * This program is free software; you can redistribute it and/or
15 * modify it under the terms of the GNU General Public License
16 * as published by the Free Software Foundation; either version 2
17 * of the License, or (at your option) any later version.
19 * If you want to redistribute modifications, please consider that
20 * scientific software is very special. Version control is crucial -
21 * bugs must be traceable. We will be happy to consider code for
22 * inclusion in the official distribution, but derived work must not
23 * be called official GROMACS. Details are found in the README & COPYING
24 * files - if they are missing, get the official version at www.gromacs.org.
26 * To help us fund GROMACS development, we humbly ask that you cite
27 * the papers on the package - you can find them in the top README file.
29 * For more info, check our website at http://www.gromacs.org
32 * Gallium Rubidium Oxygen Manganese Argon Carbon Silicon
40 #ifdef HAVE_SYS_TIME_H
56 #include "checkpoint.h"
64 /* Enum for situations that can occur during log file parsing, the
65 * corresponding string entries can be found in do_the_tests() in
66 * const char* ParseLog[] */
72 eParselogResetProblem,
83 int nPMEnodes; /* number of PME-only nodes used in this test */
84 int nx, ny, nz; /* DD grid */
85 int guessPME; /* if nPMEnodes == -1, this is the guessed number of PME nodes */
86 double *Gcycles; /* This can contain more than one value if doing multiple tests */
90 float *PME_f_load; /* PME mesh/force load average*/
91 float PME_f_load_Av; /* Average average ;) ... */
92 char *mdrun_cmd_line; /* Mdrun command line used for this test */
98 int nr_inputfiles; /* The number of tpr and mdp input files */
99 gmx_large_int_t orig_sim_steps; /* Number of steps to be done in the real simulation */
100 gmx_large_int_t orig_init_step; /* Init step for the real simulation */
101 real *rcoulomb; /* The coulomb radii [0...nr_inputfiles] */
102 real *rvdw; /* The vdW radii */
103 real *rlist; /* Neighbourlist cutoff radius */
105 int *nkx, *nky, *nkz;
106 real *fsx, *fsy, *fsz; /* Fourierspacing in x,y,z dimension */
110 static void sep_line(FILE *fp)
112 fprintf(fp, "\n------------------------------------------------------------\n");
116 /* Wrapper for system calls */
117 static int gmx_system_call(char *command)
120 gmx_fatal(FARGS, "No calls to system(3) supported on this platform. Attempted to call:\n'%s'\n", command);
122 return ( system(command) );
127 /* Check if string starts with substring */
128 static gmx_bool str_starts(const char *string, const char *substring)
130 return ( strncmp(string, substring, strlen(substring)) == 0);
134 static void cleandata(t_perf *perfdata, int test_nr)
136 perfdata->Gcycles[test_nr] = 0.0;
137 perfdata->ns_per_day[test_nr] = 0.0;
138 perfdata->PME_f_load[test_nr] = 0.0;
144 static gmx_bool is_equal(real a, real b)
146 real diff, eps = 1.0e-7;
167 static void finalize(const char *fn_out)
173 fp = fopen(fn_out, "r");
174 fprintf(stdout, "\n\n");
176 while (fgets(buf, STRLEN-1, fp) != NULL)
178 fprintf(stdout, "%s", buf);
181 fprintf(stdout, "\n\n");
186 eFoundNothing, eFoundDDStr, eFoundAccountingStr, eFoundCycleStr
189 static int parse_logfile(const char *logfile, const char *errfile,
190 t_perf *perfdata, int test_nr, int presteps, gmx_large_int_t cpt_steps,
194 char line[STRLEN], dumstring[STRLEN], dumstring2[STRLEN];
195 const char matchstrdd[] = "Domain decomposition grid";
196 const char matchstrcr[] = "resetting all time and cycle counters";
197 const char matchstrbal[] = "Average PME mesh/force load:";
198 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";
199 const char errSIG[] = "signal, stopping at the next";
202 float dum1, dum2, dum3, dum4;
205 gmx_large_int_t resetsteps = -1;
206 gmx_bool bFoundResetStr = FALSE;
207 gmx_bool bResetChecked = FALSE;
210 if (!gmx_fexist(logfile))
212 fprintf(stderr, "WARNING: Could not find logfile %s.\n", logfile);
213 cleandata(perfdata, test_nr);
214 return eParselogNotFound;
217 fp = fopen(logfile, "r");
218 perfdata->PME_f_load[test_nr] = -1.0;
219 perfdata->guessPME = -1;
221 iFound = eFoundNothing;
224 iFound = eFoundDDStr; /* Skip some case statements */
227 while (fgets(line, STRLEN, fp) != NULL)
229 /* Remove leading spaces */
232 /* Check for TERM and INT signals from user: */
233 if (strstr(line, errSIG) != NULL)
236 cleandata(perfdata, test_nr);
237 return eParselogTerm;
240 /* Check whether cycle resetting worked */
241 if (presteps > 0 && !bFoundResetStr)
243 if (strstr(line, matchstrcr) != NULL)
245 sprintf(dumstring, "step %s", gmx_large_int_pfmt);
246 sscanf(line, dumstring, &resetsteps);
247 bFoundResetStr = TRUE;
248 if (resetsteps == presteps+cpt_steps)
250 bResetChecked = TRUE;
254 sprintf(dumstring, gmx_large_int_pfmt, resetsteps);
255 sprintf(dumstring2, gmx_large_int_pfmt, presteps+cpt_steps);
256 fprintf(stderr, "WARNING: Time step counters were reset at step %s,\n"
257 " though they were supposed to be reset at step %s!\n",
258 dumstring, dumstring2);
263 /* Look for strings that appear in a certain order in the log file: */
267 /* Look for domain decomp grid and separate PME nodes: */
268 if (str_starts(line, matchstrdd))
270 sscanf(line, "Domain decomposition grid %d x %d x %d, separate PME nodes %d",
271 &(perfdata->nx), &(perfdata->ny), &(perfdata->nz), &npme);
272 if (perfdata->nPMEnodes == -1)
274 perfdata->guessPME = npme;
276 else if (perfdata->nPMEnodes != npme)
278 gmx_fatal(FARGS, "PME nodes from command line and output file are not identical");
280 iFound = eFoundDDStr;
282 /* Catch a few errors that might have occured: */
283 else if (str_starts(line, "There is no domain decomposition for"))
286 return eParselogNoDDGrid;
288 else if (str_starts(line, "reading tpx file"))
291 return eParselogTPXVersion;
293 else if (str_starts(line, "The -dd or -npme option request a parallel simulation"))
296 return eParselogNotParallel;
300 /* Look for PME mesh/force balance (not necessarily present, though) */
301 if (str_starts(line, matchstrbal))
303 sscanf(&line[strlen(matchstrbal)], "%f", &(perfdata->PME_f_load[test_nr]));
305 /* Look for matchstring */
306 if (str_starts(line, matchstring))
308 iFound = eFoundAccountingStr;
311 case eFoundAccountingStr:
312 /* Already found matchstring - look for cycle data */
313 if (str_starts(line, "Total "))
315 sscanf(line, "Total %d %lf", &procs, &(perfdata->Gcycles[test_nr]));
316 iFound = eFoundCycleStr;
320 /* Already found cycle data - look for remaining performance info and return */
321 if (str_starts(line, "Performance:"))
323 ndum = sscanf(line, "%s %f %f %f %f", dumstring, &dum1, &dum2, &dum3, &dum4);
324 /* (ns/day) is the second last entry, depending on whether GMX_DETAILED_PERF_STATS was set in print_perf(), nrnb.c */
325 perfdata->ns_per_day[test_nr] = (ndum == 5) ? dum3 : dum1;
327 if (bResetChecked || presteps == 0)
333 return eParselogResetProblem;
340 /* Close the log file */
343 /* Check why there is no performance data in the log file.
344 * Did a fatal errors occur? */
345 if (gmx_fexist(errfile))
347 fp = fopen(errfile, "r");
348 while (fgets(line, STRLEN, fp) != NULL)
350 if (str_starts(line, "Fatal error:") )
352 if (fgets(line, STRLEN, fp) != NULL)
354 fprintf(stderr, "\nWARNING: An error occured during this benchmark:\n"
358 cleandata(perfdata, test_nr);
359 return eParselogFatal;
366 fprintf(stderr, "WARNING: Could not find stderr file %s.\n", errfile);
369 /* Giving up ... we could not find out why there is no performance data in
371 fprintf(stdout, "No performance data in log file.\n");
372 cleandata(perfdata, test_nr);
374 return eParselogNoPerfData;
378 static gmx_bool analyze_data(
387 int *index_tpr, /* OUT: Nr of mdp file with best settings */
388 int *npme_optimal) /* OUT: Optimal number of PME nodes */
391 int line = 0, line_win = -1;
392 int k_win = -1, i_win = -1, winPME;
393 double s = 0.0; /* standard deviation */
396 char str_PME_f_load[13];
397 gmx_bool bCanUseOrigTPR;
398 gmx_bool bRefinedCoul, bRefinedVdW, bRefinedGrid;
404 fprintf(fp, "Summary of successful runs:\n");
405 fprintf(fp, "Line tpr PME nodes Gcycles Av. Std.dev. ns/day PME/f");
408 fprintf(fp, " DD grid");
414 for (k = 0; k < ntprs; k++)
416 for (i = 0; i < ntests; i++)
418 /* Select the right dataset: */
419 pd = &(perfdata[k][i]);
421 pd->Gcycles_Av = 0.0;
422 pd->PME_f_load_Av = 0.0;
423 pd->ns_per_day_Av = 0.0;
425 if (pd->nPMEnodes == -1)
427 sprintf(strbuf, "(%3d)", pd->guessPME);
431 sprintf(strbuf, " ");
434 /* Get the average run time of a setting */
435 for (j = 0; j < nrepeats; j++)
437 pd->Gcycles_Av += pd->Gcycles[j];
438 pd->PME_f_load_Av += pd->PME_f_load[j];
440 pd->Gcycles_Av /= nrepeats;
441 pd->PME_f_load_Av /= nrepeats;
443 for (j = 0; j < nrepeats; j++)
445 if (pd->ns_per_day[j] > 0.0)
447 pd->ns_per_day_Av += pd->ns_per_day[j];
451 /* Somehow the performance number was not aquired for this run,
452 * therefor set the average to some negative value: */
453 pd->ns_per_day_Av = -1.0f*nrepeats;
457 pd->ns_per_day_Av /= nrepeats;
460 if (pd->PME_f_load_Av > 0.0)
462 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load_Av);
466 sprintf(str_PME_f_load, "%s", " - ");
470 /* We assume we had a successful run if both averages are positive */
471 if (pd->Gcycles_Av > 0.0 && pd->ns_per_day_Av > 0.0)
473 /* Output statistics if repeats were done */
476 /* Calculate the standard deviation */
478 for (j = 0; j < nrepeats; j++)
480 s += pow( pd->Gcycles[j] - pd->Gcycles_Av, 2 );
485 fprintf(fp, "%4d %3d %4d%s %12.3f %12.3f %12.3f %s",
486 line, k, pd->nPMEnodes, strbuf, pd->Gcycles_Av, s,
487 pd->ns_per_day_Av, str_PME_f_load);
490 fprintf(fp, " %3d %3d %3d", pd->nx, pd->ny, pd->nz);
494 /* Store the index of the best run found so far in 'winner': */
495 if ( (k_win == -1) || (pd->Gcycles_Av < perfdata[k_win][i_win].Gcycles_Av) )
508 gmx_fatal(FARGS, "None of the runs was successful! Check %s for problems.", fn);
513 winPME = perfdata[k_win][i_win].nPMEnodes;
517 /* We stuck to a fixed number of PME-only nodes */
518 sprintf(strbuf, "settings No. %d", k_win);
522 /* We have optimized the number of PME-only nodes */
525 sprintf(strbuf, "%s", "the automatic number of PME nodes");
529 sprintf(strbuf, "%d PME nodes", winPME);
532 fprintf(fp, "Best performance was achieved with %s", strbuf);
533 if ((nrepeats > 1) && (ntests > 1))
535 fprintf(fp, " (see line %d)", line_win);
539 /* Only mention settings if they were modified: */
540 bRefinedCoul = !is_equal(info->rcoulomb[k_win], info->rcoulomb[0]);
541 bRefinedVdW = !is_equal(info->rvdw[k_win], info->rvdw[0] );
542 bRefinedGrid = !(info->nkx[k_win] == info->nkx[0] &&
543 info->nky[k_win] == info->nky[0] &&
544 info->nkz[k_win] == info->nkz[0]);
546 if (bRefinedCoul || bRefinedVdW || bRefinedGrid)
548 fprintf(fp, "Optimized PME settings:\n");
549 bCanUseOrigTPR = FALSE;
553 bCanUseOrigTPR = TRUE;
558 fprintf(fp, " New Coulomb radius: %f nm (was %f nm)\n", info->rcoulomb[k_win], info->rcoulomb[0]);
563 fprintf(fp, " New Van der Waals radius: %f nm (was %f nm)\n", info->rvdw[k_win], info->rvdw[0]);
568 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],
569 info->nkx[0], info->nky[0], info->nkz[0]);
572 if (bCanUseOrigTPR && ntprs > 1)
574 fprintf(fp, "and original PME settings.\n");
579 /* Return the index of the mdp file that showed the highest performance
580 * and the optimal number of PME nodes */
582 *npme_optimal = winPME;
584 return bCanUseOrigTPR;
588 /* Get the commands we need to set up the runs from environment variables */
589 static void get_program_paths(gmx_bool bThreads, char *cmd_mpirun[], char cmd_np[],
590 char *cmd_mdrun[], int repeats)
594 const char def_mpirun[] = "mpirun";
595 const char def_mdrun[] = "mdrun";
597 const char empty_mpirun[] = "";
599 /* Get the commands we need to set up the runs from environment variables */
602 if ( (cp = getenv("MPIRUN")) != NULL)
604 *cmd_mpirun = strdup(cp);
608 *cmd_mpirun = strdup(def_mpirun);
613 *cmd_mpirun = strdup(empty_mpirun);
616 if ( (cp = getenv("MDRUN" )) != NULL)
618 *cmd_mdrun = strdup(cp);
622 *cmd_mdrun = strdup(def_mdrun);
626 /* Check that the commands will run mdrun (perhaps via mpirun) by
627 * running a very quick test simulation. Requires MPI environment to
628 * be available if applicable. */
629 static void check_mdrun_works(gmx_bool bThreads,
630 const char *cmd_mpirun,
632 const char *cmd_mdrun)
634 char *command = NULL;
638 const char filename[] = "benchtest.log";
640 /* This string should always be identical to the one in copyrite.c,
641 * gmx_print_version_info() in the defined(GMX_MPI) section */
642 const char match_mpi[] = "MPI library: MPI";
643 const char match_mdrun[] = "Program: ";
644 gmx_bool bMdrun = FALSE;
645 gmx_bool bMPI = FALSE;
647 /* Run a small test to see whether mpirun + mdrun work */
648 fprintf(stdout, "Making sure that mdrun can be executed. ");
651 snew(command, strlen(cmd_mdrun) + strlen(cmd_np) + strlen(filename) + 50);
652 sprintf(command, "%s%s-version -maxh 0.001 1> %s 2>&1", cmd_mdrun, cmd_np, filename);
656 snew(command, strlen(cmd_mpirun) + strlen(cmd_np) + strlen(cmd_mdrun) + strlen(filename) + 50);
657 sprintf(command, "%s%s%s -version -maxh 0.001 1> %s 2>&1", cmd_mpirun, cmd_np, cmd_mdrun, filename);
659 fprintf(stdout, "Trying '%s' ... ", command);
660 make_backup(filename);
661 gmx_system_call(command);
663 /* Check if we find the characteristic string in the output: */
664 if (!gmx_fexist(filename))
666 gmx_fatal(FARGS, "Output from test run could not be found.");
669 fp = fopen(filename, "r");
670 /* We need to scan the whole output file, since sometimes the queuing system
671 * also writes stuff to stdout/err */
674 cp = fgets(line, STRLEN, fp);
677 if (str_starts(line, match_mdrun) )
681 if (str_starts(line, match_mpi) )
693 gmx_fatal(FARGS, "Need a threaded version of mdrun. This one\n"
695 "seems to have been compiled with MPI instead.",
703 gmx_fatal(FARGS, "Need an MPI-enabled version of mdrun. This one\n"
705 "seems to have been compiled without MPI support.",
712 gmx_fatal(FARGS, "Cannot execute mdrun. Please check %s for problems!",
716 fprintf(stdout, "passed.\n");
724 static void launch_simulation(
725 gmx_bool bLaunch, /* Should the simulation be launched? */
726 FILE *fp, /* General log file */
727 gmx_bool bThreads, /* whether to use threads */
728 char *cmd_mpirun, /* Command for mpirun */
729 char *cmd_np, /* Switch for -np or -ntmpi or empty */
730 char *cmd_mdrun, /* Command for mdrun */
731 char *args_for_mdrun, /* Arguments for mdrun */
732 const char *simulation_tpr, /* This tpr will be simulated */
733 int nPMEnodes) /* Number of PME nodes to use */
738 /* Make enough space for the system call command,
739 * (100 extra chars for -npme ... etc. options should suffice): */
740 snew(command, strlen(cmd_mpirun)+strlen(cmd_mdrun)+strlen(cmd_np)+strlen(args_for_mdrun)+strlen(simulation_tpr)+100);
742 /* Note that the -passall options requires args_for_mdrun to be at the end
743 * of the command line string */
746 sprintf(command, "%s%s-npme %d -s %s %s",
747 cmd_mdrun, cmd_np, nPMEnodes, simulation_tpr, args_for_mdrun);
751 sprintf(command, "%s%s%s -npme %d -s %s %s",
752 cmd_mpirun, cmd_np, cmd_mdrun, nPMEnodes, simulation_tpr, args_for_mdrun);
755 fprintf(fp, "%s this command line to launch the simulation:\n\n%s", bLaunch ? "Using" : "Please use", command);
759 /* Now the real thing! */
762 fprintf(stdout, "\nLaunching simulation with best parameters now.\nExecuting '%s'", command);
765 gmx_system_call(command);
770 static void modify_PMEsettings(
771 gmx_large_int_t simsteps, /* Set this value as number of time steps */
772 gmx_large_int_t init_step, /* Set this value as init_step */
773 const char *fn_best_tpr, /* tpr file with the best performance */
774 const char *fn_sim_tpr) /* name of tpr file to be launched */
782 read_tpx_state(fn_best_tpr, ir, &state, NULL, &mtop);
784 /* Reset nsteps and init_step to the value of the input .tpr file */
785 ir->nsteps = simsteps;
786 ir->init_step = init_step;
788 /* Write the tpr file which will be launched */
789 sprintf(buf, "Writing optimized simulation file %s with nsteps=%s.\n", fn_sim_tpr, gmx_large_int_pfmt);
790 fprintf(stdout, buf, ir->nsteps);
792 write_tpx_state(fn_sim_tpr, ir, &state, &mtop);
798 #define EPME_SWITCHED(e) ((e) == eelPMESWITCH || (e) == eelPMEUSERSWITCH)
800 /* Make additional TPR files with more computational load for the
801 * direct space processors: */
802 static void make_benchmark_tprs(
803 const char *fn_sim_tpr, /* READ : User-provided tpr file */
804 char *fn_bench_tprs[], /* WRITE: Names of benchmark tpr files */
805 gmx_large_int_t benchsteps, /* Number of time steps for benchmark runs */
806 gmx_large_int_t statesteps, /* Step counter in checkpoint file */
807 real rmin, /* Minimal Coulomb radius */
808 real rmax, /* Maximal Coulomb radius */
809 real bScaleRvdw, /* Scale rvdw along with rcoulomb */
810 int *ntprs, /* No. of TPRs to write, each with a different
811 rcoulomb and fourierspacing */
812 t_inputinfo *info, /* Contains information about mdp file options */
813 FILE *fp) /* Write the output here */
819 real nlist_buffer; /* Thickness of the buffer regions for PME-switch potentials */
822 gmx_bool bNote = FALSE;
823 real add; /* Add this to rcoul for the next test */
824 real fac = 1.0; /* Scaling factor for Coulomb radius */
825 real fourierspacing; /* Basic fourierspacing from tpr */
828 sprintf(buf, "Making benchmark tpr file%s with %s time step%s",
829 *ntprs > 1 ? "s" : "", gmx_large_int_pfmt, benchsteps > 1 ? "s" : "");
830 fprintf(stdout, buf, benchsteps);
833 sprintf(buf, " (adding %s steps from checkpoint file)", gmx_large_int_pfmt);
834 fprintf(stdout, buf, statesteps);
835 benchsteps += statesteps;
837 fprintf(stdout, ".\n");
841 read_tpx_state(fn_sim_tpr, ir, &state, NULL, &mtop);
843 /* Check if some kind of PME was chosen */
844 if (EEL_PME(ir->coulombtype) == FALSE)
846 gmx_fatal(FARGS, "Can only do optimizations for simulations with %s electrostatics.",
850 /* Check if rcoulomb == rlist, which is necessary for plain PME. */
851 if ( (ir->cutoff_scheme != ecutsVERLET) &&
852 (eelPME == ir->coulombtype) && !(ir->rcoulomb == ir->rlist))
854 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to rlist (%f).",
855 EELTYPE(eelPME), ir->rcoulomb, ir->rlist);
857 /* For other PME types, rcoulomb is allowed to be smaller than rlist */
858 else if (ir->rcoulomb > ir->rlist)
860 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to or smaller than rlist (%f)",
861 EELTYPE(ir->coulombtype), ir->rcoulomb, ir->rlist);
864 if (bScaleRvdw && ir->rvdw != ir->rcoulomb)
866 fprintf(stdout, "NOTE: input rvdw != rcoulomb, will not scale rvdw\n");
870 /* Reduce the number of steps for the benchmarks */
871 info->orig_sim_steps = ir->nsteps;
872 ir->nsteps = benchsteps;
873 /* We must not use init_step from the input tpr file for the benchmarks */
874 info->orig_init_step = ir->init_step;
877 /* For PME-switch potentials, keep the radial distance of the buffer region */
878 nlist_buffer = ir->rlist - ir->rcoulomb;
880 /* Determine length of triclinic box vectors */
881 for (d = 0; d < DIM; d++)
884 for (i = 0; i < DIM; i++)
886 box_size[d] += state.box[d][i]*state.box[d][i];
888 box_size[d] = sqrt(box_size[d]);
891 if (ir->fourier_spacing > 0)
893 info->fsx[0] = ir->fourier_spacing;
894 info->fsy[0] = ir->fourier_spacing;
895 info->fsz[0] = ir->fourier_spacing;
899 /* Reconstruct fourierspacing per dimension from the number of grid points and box size */
900 info->fsx[0] = box_size[XX]/ir->nkx;
901 info->fsy[0] = box_size[YY]/ir->nky;
902 info->fsz[0] = box_size[ZZ]/ir->nkz;
905 /* If no value for the fourierspacing was provided on the command line, we
906 * use the reconstruction from the tpr file */
907 if (ir->fourier_spacing > 0)
909 /* Use the spacing from the tpr */
910 fourierspacing = ir->fourier_spacing;
914 /* Use the maximum observed spacing */
915 fourierspacing = max(max(info->fsx[0], info->fsy[0]), info->fsz[0]);
918 fprintf(stdout, "Calculating PME grid points on the basis of a fourierspacing of %f nm\n", fourierspacing);
920 /* For performance comparisons the number of particles is useful to have */
921 fprintf(fp, " Number of particles : %d\n", mtop.natoms);
923 /* Print information about settings of which some are potentially modified: */
924 fprintf(fp, " Coulomb type : %s\n", EELTYPE(ir->coulombtype));
925 fprintf(fp, " Grid spacing x y z : %f %f %f\n",
926 box_size[XX]/ir->nkx, box_size[YY]/ir->nky, box_size[ZZ]/ir->nkz);
927 fprintf(fp, " Van der Waals type : %s\n", EVDWTYPE(ir->vdwtype));
928 if (EVDW_SWITCHED(ir->vdwtype))
930 fprintf(fp, " rvdw_switch : %f nm\n", ir->rvdw_switch);
932 if (EPME_SWITCHED(ir->coulombtype))
934 fprintf(fp, " rlist : %f nm\n", ir->rlist);
936 if (ir->rlistlong != max_cutoff(ir->rvdw, ir->rcoulomb))
938 fprintf(fp, " rlistlong : %f nm\n", ir->rlistlong);
941 /* Print a descriptive line about the tpr settings tested */
942 fprintf(fp, "\nWill try these real/reciprocal workload settings:\n");
943 fprintf(fp, " No. scaling rcoulomb");
944 fprintf(fp, " nkx nky nkz");
945 fprintf(fp, " spacing");
946 if (evdwCUT == ir->vdwtype)
948 fprintf(fp, " rvdw");
950 if (EPME_SWITCHED(ir->coulombtype))
952 fprintf(fp, " rlist");
954 if (ir->rlistlong != max_cutoff(ir->rlist, max_cutoff(ir->rvdw, ir->rcoulomb)) )
956 fprintf(fp, " rlistlong");
958 fprintf(fp, " tpr file\n");
960 /* Loop to create the requested number of tpr input files */
961 for (j = 0; j < *ntprs; j++)
963 /* The first .tpr is the provided one, just need to modify nsteps,
964 * so skip the following block */
967 /* Determine which Coulomb radii rc to use in the benchmarks */
968 add = (rmax-rmin)/(*ntprs-1);
969 if (is_equal(rmin, info->rcoulomb[0]))
971 ir->rcoulomb = rmin + j*add;
973 else if (is_equal(rmax, info->rcoulomb[0]))
975 ir->rcoulomb = rmin + (j-1)*add;
979 /* rmin != rcoul != rmax, ergo test between rmin and rmax */
980 add = (rmax-rmin)/(*ntprs-2);
981 ir->rcoulomb = rmin + (j-1)*add;
984 /* Determine the scaling factor fac */
985 fac = ir->rcoulomb/info->rcoulomb[0];
987 /* Scale the Fourier grid spacing */
988 ir->nkx = ir->nky = ir->nkz = 0;
989 calc_grid(NULL, state.box, fourierspacing*fac, &ir->nkx, &ir->nky, &ir->nkz);
991 /* Adjust other radii since various conditions neet to be fulfilled */
992 if (eelPME == ir->coulombtype)
994 /* plain PME, rcoulomb must be equal to rlist */
995 ir->rlist = ir->rcoulomb;
999 /* rlist must be >= rcoulomb, we keep the size of the buffer region */
1000 ir->rlist = ir->rcoulomb + nlist_buffer;
1003 if (bScaleRvdw && evdwCUT == ir->vdwtype)
1005 /* For vdw cutoff, rvdw >= rlist */
1006 ir->rvdw = max(info->rvdw[0], ir->rlist);
1009 ir->rlistlong = max_cutoff(ir->rlist, max_cutoff(ir->rvdw, ir->rcoulomb));
1011 } /* end of "if (j != 0)" */
1013 /* for j==0: Save the original settings
1014 * for j >0: Save modified radii and Fourier grids */
1015 info->rcoulomb[j] = ir->rcoulomb;
1016 info->rvdw[j] = ir->rvdw;
1017 info->nkx[j] = ir->nkx;
1018 info->nky[j] = ir->nky;
1019 info->nkz[j] = ir->nkz;
1020 info->rlist[j] = ir->rlist;
1021 info->rlistlong[j] = ir->rlistlong;
1022 info->fsx[j] = fac*fourierspacing;
1023 info->fsy[j] = fac*fourierspacing;
1024 info->fsz[j] = fac*fourierspacing;
1026 /* Write the benchmark tpr file */
1027 strncpy(fn_bench_tprs[j], fn_sim_tpr, strlen(fn_sim_tpr)-strlen(".tpr"));
1028 sprintf(buf, "_bench%.2d.tpr", j);
1029 strcat(fn_bench_tprs[j], buf);
1030 fprintf(stdout, "Writing benchmark tpr %s with nsteps=", fn_bench_tprs[j]);
1031 fprintf(stdout, gmx_large_int_pfmt, ir->nsteps);
1034 fprintf(stdout, ", scaling factor %f\n", fac);
1038 fprintf(stdout, ", unmodified settings\n");
1041 write_tpx_state(fn_bench_tprs[j], ir, &state, &mtop);
1043 /* Write information about modified tpr settings to log file */
1044 fprintf(fp, "%4d%10f%10f", j, fac, ir->rcoulomb);
1045 fprintf(fp, "%5d%5d%5d", ir->nkx, ir->nky, ir->nkz);
1046 fprintf(fp, " %9f ", info->fsx[j]);
1047 if (evdwCUT == ir->vdwtype)
1049 fprintf(fp, "%10f", ir->rvdw);
1051 if (EPME_SWITCHED(ir->coulombtype))
1053 fprintf(fp, "%10f", ir->rlist);
1055 if (info->rlistlong[0] != max_cutoff(info->rlist[0], max_cutoff(info->rvdw[0], info->rcoulomb[0])) )
1057 fprintf(fp, "%10f", ir->rlistlong);
1059 fprintf(fp, " %-14s\n", fn_bench_tprs[j]);
1061 /* Make it clear to the user that some additional settings were modified */
1062 if (!is_equal(ir->rvdw, info->rvdw[0])
1063 || !is_equal(ir->rlistlong, info->rlistlong[0]) )
1070 fprintf(fp, "\nNote that in addition to the Coulomb radius and the Fourier grid\n"
1071 "other input settings were also changed (see table above).\n"
1072 "Please check if the modified settings are appropriate.\n");
1080 /* Rename the files we want to keep to some meaningful filename and
1081 * delete the rest */
1082 static void cleanup(const t_filenm *fnm, int nfile, int k, int nnodes,
1083 int nPMEnodes, int nr, gmx_bool bKeepStderr)
1085 char numstring[STRLEN];
1086 char newfilename[STRLEN];
1087 const char *fn = NULL;
1092 fprintf(stdout, "Cleaning up, deleting benchmark temp files ...\n");
1094 for (i = 0; i < nfile; i++)
1096 opt = (char *)fnm[i].opt;
1097 if (strcmp(opt, "-p") == 0)
1099 /* do nothing; keep this file */
1102 else if (strcmp(opt, "-bg") == 0)
1104 /* Give the log file a nice name so one can later see which parameters were used */
1105 numstring[0] = '\0';
1108 sprintf(numstring, "_%d", nr);
1110 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", opt2fn("-bg", nfile, fnm), k, nnodes, nPMEnodes, numstring);
1111 if (gmx_fexist(opt2fn("-bg", nfile, fnm)))
1113 fprintf(stdout, "renaming log file to %s\n", newfilename);
1114 make_backup(newfilename);
1115 rename(opt2fn("-bg", nfile, fnm), newfilename);
1118 else if (strcmp(opt, "-err") == 0)
1120 /* This file contains the output of stderr. We want to keep it in
1121 * cases where there have been problems. */
1122 fn = opt2fn(opt, nfile, fnm);
1123 numstring[0] = '\0';
1126 sprintf(numstring, "_%d", nr);
1128 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", fn, k, nnodes, nPMEnodes, numstring);
1133 fprintf(stdout, "Saving stderr output in %s\n", newfilename);
1134 make_backup(newfilename);
1135 rename(fn, newfilename);
1139 fprintf(stdout, "Deleting %s\n", fn);
1144 /* Delete the files which are created for each benchmark run: (options -b*) */
1145 else if ( (0 == strncmp(opt, "-b", 2)) && (opt2bSet(opt, nfile, fnm) || !is_optional(&fnm[i])) )
1147 fn = opt2fn(opt, nfile, fnm);
1150 fprintf(stdout, "Deleting %s\n", fn);
1158 /* Returns the largest common factor of n1 and n2 */
1159 static int largest_common_factor(int n1, int n2)
1164 for (factor = nmax; factor > 0; factor--)
1166 if (0 == (n1 % factor) && 0 == (n2 % factor) )
1171 return 0; /* one for the compiler */
1175 eNpmeAuto, eNpmeAll, eNpmeReduced, eNpmeSubset, eNpmeNr
1178 /* Create a list of numbers of PME nodes to test */
1179 static void make_npme_list(
1180 const char *npmevalues_opt, /* Make a complete list with all
1181 * possibilities or a short list that keeps only
1182 * reasonable numbers of PME nodes */
1183 int *nentries, /* Number of entries we put in the nPMEnodes list */
1184 int *nPMEnodes[], /* Each entry contains the value for -npme */
1185 int nnodes, /* Total number of nodes to do the tests on */
1186 int minPMEnodes, /* Minimum number of PME nodes */
1187 int maxPMEnodes) /* Maximum number of PME nodes */
1190 int min_factor = 1; /* We request that npp and npme have this minimal
1191 * largest common factor (depends on npp) */
1192 int nlistmax; /* Max. list size */
1193 int nlist; /* Actual number of entries in list */
1197 /* Do we need to check all possible values for -npme or is a reduced list enough? */
1198 if (0 == strcmp(npmevalues_opt, "all") )
1202 else if (0 == strcmp(npmevalues_opt, "subset") )
1204 eNPME = eNpmeSubset;
1206 else /* "auto" or "range" */
1212 else if (nnodes < 128)
1214 eNPME = eNpmeReduced;
1218 eNPME = eNpmeSubset;
1222 /* Calculate how many entries we could possibly have (in case of -npme all) */
1225 nlistmax = maxPMEnodes - minPMEnodes + 3;
1226 if (0 == minPMEnodes)
1236 /* Now make the actual list which is at most of size nlist */
1237 snew(*nPMEnodes, nlistmax);
1238 nlist = 0; /* start counting again, now the real entries in the list */
1239 for (i = 0; i < nlistmax - 2; i++)
1241 npme = maxPMEnodes - i;
1252 /* For 2d PME we want a common largest factor of at least the cube
1253 * root of the number of PP nodes */
1254 min_factor = (int) pow(npp, 1.0/3.0);
1257 gmx_fatal(FARGS, "Unknown option for eNPME in make_npme_list");
1260 if (largest_common_factor(npp, npme) >= min_factor)
1262 (*nPMEnodes)[nlist] = npme;
1266 /* We always test 0 PME nodes and the automatic number */
1267 *nentries = nlist + 2;
1268 (*nPMEnodes)[nlist ] = 0;
1269 (*nPMEnodes)[nlist+1] = -1;
1271 fprintf(stderr, "Will try the following %d different values for -npme:\n", *nentries);
1272 for (i = 0; i < *nentries-1; i++)
1274 fprintf(stderr, "%d, ", (*nPMEnodes)[i]);
1276 fprintf(stderr, "and %d (auto).\n", (*nPMEnodes)[*nentries-1]);
1280 /* Allocate memory to store the performance data */
1281 static void init_perfdata(t_perf *perfdata[], int ntprs, int datasets, int repeats)
1286 for (k = 0; k < ntprs; k++)
1288 snew(perfdata[k], datasets);
1289 for (i = 0; i < datasets; i++)
1291 for (j = 0; j < repeats; j++)
1293 snew(perfdata[k][i].Gcycles, repeats);
1294 snew(perfdata[k][i].ns_per_day, repeats);
1295 snew(perfdata[k][i].PME_f_load, repeats);
1302 /* Check for errors on mdrun -h */
1303 static void make_sure_it_runs(char *mdrun_cmd_line, int length, FILE *fp)
1305 char *command, *msg;
1309 snew(command, length + 15);
1310 snew(msg, length + 500);
1312 fprintf(stdout, "Making sure the benchmarks can be executed ...\n");
1313 /* FIXME: mdrun -h no longer actually does anything useful.
1314 * It unconditionally prints the help, ignoring all other options. */
1315 sprintf(command, "%s-h -quiet", mdrun_cmd_line);
1316 ret = gmx_system_call(command);
1320 /* To prevent confusion, do not again issue a gmx_fatal here since we already
1321 * get the error message from mdrun itself */
1322 sprintf(msg, "Cannot run the benchmark simulations! Please check the error message of\n"
1323 "mdrun for the source of the problem. Did you provide a command line\n"
1324 "argument that neither g_tune_pme nor mdrun understands? Offending command:\n"
1325 "\n%s\n\n", command);
1327 fprintf(stderr, "%s", msg);
1329 fprintf(fp, "%s", msg);
1339 static void do_the_tests(
1340 FILE *fp, /* General g_tune_pme output file */
1341 char **tpr_names, /* Filenames of the input files to test */
1342 int maxPMEnodes, /* Max fraction of nodes to use for PME */
1343 int minPMEnodes, /* Min fraction of nodes to use for PME */
1344 int npme_fixed, /* If >= -1, test fixed number of PME
1346 const char *npmevalues_opt, /* Which -npme values should be tested */
1347 t_perf **perfdata, /* Here the performace data is stored */
1348 int *pmeentries, /* Entries in the nPMEnodes list */
1349 int repeats, /* Repeat each test this often */
1350 int nnodes, /* Total number of nodes = nPP + nPME */
1351 int nr_tprs, /* Total number of tpr files to test */
1352 gmx_bool bThreads, /* Threads or MPI? */
1353 char *cmd_mpirun, /* mpirun command string */
1354 char *cmd_np, /* "-np", "-n", whatever mpirun needs */
1355 char *cmd_mdrun, /* mdrun command string */
1356 char *cmd_args_bench, /* arguments for mdrun in a string */
1357 const t_filenm *fnm, /* List of filenames from command line */
1358 int nfile, /* Number of files specified on the cmdl. */
1359 int presteps, /* DLB equilibration steps, is checked */
1360 gmx_large_int_t cpt_steps) /* Time step counter in the checkpoint */
1362 int i, nr, k, ret, count = 0, totaltests;
1363 int *nPMEnodes = NULL;
1366 char *command, *cmd_stub;
1368 gmx_bool bResetProblem = FALSE;
1369 gmx_bool bFirst = TRUE;
1372 /* This string array corresponds to the eParselog enum type at the start
1374 const char* ParseLog[] = {
1376 "Logfile not found!",
1377 "No timings, logfile truncated?",
1378 "Run was terminated.",
1379 "Counters were not reset properly.",
1380 "No DD grid found for these settings.",
1381 "TPX version conflict!",
1382 "mdrun was not started in parallel!",
1385 char str_PME_f_load[13];
1388 /* Allocate space for the mdrun command line. 100 extra characters should
1389 be more than enough for the -npme etcetera arguments */
1390 cmdline_length = strlen(cmd_mpirun)
1393 + strlen(cmd_args_bench)
1394 + strlen(tpr_names[0]) + 100;
1395 snew(command, cmdline_length);
1396 snew(cmd_stub, cmdline_length);
1398 /* Construct the part of the command line that stays the same for all tests: */
1401 sprintf(cmd_stub, "%s%s", cmd_mdrun, cmd_np);
1405 sprintf(cmd_stub, "%s%s%s ", cmd_mpirun, cmd_np, cmd_mdrun);
1408 /* Create a list of numbers of PME nodes to test */
1409 if (npme_fixed < -1)
1411 make_npme_list(npmevalues_opt, pmeentries, &nPMEnodes,
1412 nnodes, minPMEnodes, maxPMEnodes);
1418 nPMEnodes[0] = npme_fixed;
1419 fprintf(stderr, "Will use a fixed number of %d PME-only nodes.\n", nPMEnodes[0]);
1424 fprintf(fp, "\nNo benchmarks done since number of repeats (-r) is 0.\n");
1426 finalize(opt2fn("-p", nfile, fnm));
1430 /* Allocate one dataset for each tpr input file: */
1431 init_perfdata(perfdata, nr_tprs, *pmeentries, repeats);
1433 /*****************************************/
1434 /* Main loop over all tpr files to test: */
1435 /*****************************************/
1436 totaltests = nr_tprs*(*pmeentries)*repeats;
1437 for (k = 0; k < nr_tprs; k++)
1439 fprintf(fp, "\nIndividual timings for input file %d (%s):\n", k, tpr_names[k]);
1440 fprintf(fp, "PME nodes Gcycles ns/day PME/f Remark\n");
1441 /* Loop over various numbers of PME nodes: */
1442 for (i = 0; i < *pmeentries; i++)
1444 pd = &perfdata[k][i];
1446 /* Loop over the repeats for each scenario: */
1447 for (nr = 0; nr < repeats; nr++)
1449 pd->nPMEnodes = nPMEnodes[i];
1451 /* Add -npme and -s to the command line and save it. Note that
1452 * the -passall (if set) options requires cmd_args_bench to be
1453 * at the end of the command line string */
1454 snew(pd->mdrun_cmd_line, cmdline_length);
1455 sprintf(pd->mdrun_cmd_line, "%s-npme %d -s %s %s",
1456 cmd_stub, pd->nPMEnodes, tpr_names[k], cmd_args_bench);
1458 /* To prevent that all benchmarks fail due to a show-stopper argument
1459 * on the mdrun command line, we make a quick check with mdrun -h first */
1462 make_sure_it_runs(pd->mdrun_cmd_line, cmdline_length, fp);
1466 /* Do a benchmark simulation: */
1469 sprintf(buf, ", pass %d/%d", nr+1, repeats);
1475 fprintf(stdout, "\n=== Progress %2.0f%%, tpr %d/%d, run %d/%d%s:\n",
1476 (100.0*count)/totaltests,
1477 k+1, nr_tprs, i+1, *pmeentries, buf);
1478 make_backup(opt2fn("-err", nfile, fnm));
1479 sprintf(command, "%s 1> /dev/null 2>%s", pd->mdrun_cmd_line, opt2fn("-err", nfile, fnm));
1480 fprintf(stdout, "%s\n", pd->mdrun_cmd_line);
1481 gmx_system_call(command);
1483 /* Collect the performance data from the log file; also check stderr
1484 * for fatal errors */
1485 ret = parse_logfile(opt2fn("-bg", nfile, fnm), opt2fn("-err", nfile, fnm),
1486 pd, nr, presteps, cpt_steps, nnodes);
1487 if ((presteps > 0) && (ret == eParselogResetProblem))
1489 bResetProblem = TRUE;
1492 if (-1 == pd->nPMEnodes)
1494 sprintf(buf, "(%3d)", pd->guessPME);
1502 if (pd->PME_f_load[nr] > 0.0)
1504 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load[nr]);
1508 sprintf(str_PME_f_load, "%s", " - ");
1511 /* Write the data we got to disk */
1512 fprintf(fp, "%4d%s %12.3f %12.3f %s %s", pd->nPMEnodes,
1513 buf, pd->Gcycles[nr], pd->ns_per_day[nr], str_PME_f_load, ParseLog[ret]);
1514 if (!(ret == eParselogOK || ret == eParselogNoDDGrid || ret == eParselogNotFound) )
1516 fprintf(fp, " Check %s file for problems.", ret == eParselogFatal ? "err" : "log");
1522 /* Do some cleaning up and delete the files we do not need any more */
1523 cleanup(fnm, nfile, k, nnodes, pd->nPMEnodes, nr, ret == eParselogFatal);
1525 /* If the first run with this number of processors already failed, do not try again: */
1526 if (pd->Gcycles[0] <= 0.0 && repeats > 1)
1528 fprintf(stdout, "Skipping remaining passes of unsuccessful setting, see log file for details.\n");
1529 count += repeats-(nr+1);
1532 } /* end of repeats loop */
1533 } /* end of -npme loop */
1534 } /* end of tpr file loop */
1539 fprintf(fp, "WARNING: The cycle and time step counters could not be reset properly. ");
1547 static void check_input(
1554 real maxPMEfraction,
1555 real minPMEfraction,
1557 gmx_large_int_t bench_nsteps,
1558 const t_filenm *fnm,
1568 /* Make sure the input file exists */
1569 if (!gmx_fexist(opt2fn("-s", nfile, fnm)))
1571 gmx_fatal(FARGS, "File %s not found.", opt2fn("-s", nfile, fnm));
1574 /* Make sure that the checkpoint file is not overwritten during benchmarking */
1575 if ( (0 == strcmp(opt2fn("-cpi", nfile, fnm), opt2fn("-bcpo", nfile, fnm)) ) && (sim_part > 1) )
1577 gmx_fatal(FARGS, "Checkpoint input (-cpi) and benchmark checkpoint output (-bcpo) files must not be identical.\n"
1578 "The checkpoint input file must not be overwritten during the benchmarks.\n");
1581 /* Make sure that repeats is >= 0 (if == 0, only write tpr files) */
1584 gmx_fatal(FARGS, "Number of repeats < 0!");
1587 /* Check number of nodes */
1590 gmx_fatal(FARGS, "Number of nodes/threads must be a positive integer.");
1593 /* Automatically choose -ntpr if not set */
1603 /* Set a reasonable scaling factor for rcoulomb */
1606 *rmax = rcoulomb * 1.2;
1609 fprintf(stderr, "Will test %d tpr file%s.\n", *ntprs, *ntprs == 1 ? "" : "s");
1615 fprintf(stderr, "Note: Choose ntpr>1 to shift PME load between real and reciprocal space.\n");
1619 /* Make shure that rmin <= rcoulomb <= rmax */
1628 if (!(*rmin <= *rmax) )
1630 gmx_fatal(FARGS, "Please choose the Coulomb radii such that rmin <= rmax.\n"
1631 "rmin = %g, rmax = %g, actual rcoul from .tpr file = %g\n", *rmin, *rmax, rcoulomb);
1633 /* Add test scenarios if rmin or rmax were set */
1636 if (!is_equal(*rmin, rcoulomb) && (*ntprs == 1) )
1639 fprintf(stderr, "NOTE: Setting -rmin to %g changed -ntpr to %d\n",
1642 if (!is_equal(*rmax, rcoulomb) && (*ntprs == 1) )
1645 fprintf(stderr, "NOTE: Setting -rmax to %g changed -ntpr to %d\n",
1650 /* If one of rmin, rmax is set, we need 2 tpr files at minimum */
1651 if (!is_equal(*rmax, rcoulomb) || !is_equal(*rmin, rcoulomb) )
1653 *ntprs = max(*ntprs, 2);
1656 /* If both rmin, rmax are set, we need 3 tpr files at minimum */
1657 if (!is_equal(*rmax, rcoulomb) && !is_equal(*rmin, rcoulomb) )
1659 *ntprs = max(*ntprs, 3);
1664 fprintf(stderr, "NOTE: Your rmin, rmax setting changed -ntpr to %d\n", *ntprs);
1669 if (is_equal(*rmin, rcoulomb) && is_equal(rcoulomb, *rmax)) /* We have just a single rc */
1671 fprintf(stderr, "WARNING: Resetting -ntpr to 1 since no Coulomb radius scaling is requested.\n"
1672 "Please set rmin < rmax to test Coulomb radii in the [rmin, rmax] interval\n"
1673 "with correspondingly adjusted PME grid settings\n");
1678 /* Check whether max and min fraction are within required values */
1679 if (maxPMEfraction > 0.5 || maxPMEfraction < 0)
1681 gmx_fatal(FARGS, "-max must be between 0 and 0.5");
1683 if (minPMEfraction > 0.5 || minPMEfraction < 0)
1685 gmx_fatal(FARGS, "-min must be between 0 and 0.5");
1687 if (maxPMEfraction < minPMEfraction)
1689 gmx_fatal(FARGS, "-max must be larger or equal to -min");
1692 /* Check whether the number of steps - if it was set - has a reasonable value */
1693 if (bench_nsteps < 0)
1695 gmx_fatal(FARGS, "Number of steps must be positive.");
1698 if (bench_nsteps > 10000 || bench_nsteps < 100)
1700 fprintf(stderr, "WARNING: steps=");
1701 fprintf(stderr, gmx_large_int_pfmt, bench_nsteps);
1702 fprintf(stderr, ". Are you sure you want to perform so %s steps for each benchmark?\n", (bench_nsteps < 100) ? "few" : "many");
1707 gmx_fatal(FARGS, "Cannot have a negative number of presteps.\n");
1710 /* Check for rcoulomb scaling if more than one .tpr file is tested */
1713 if (*rmin/rcoulomb < 0.75 || *rmax/rcoulomb > 1.25)
1715 fprintf(stderr, "WARNING: Applying extreme scaling factor. I hope you know what you are doing.\n");
1719 /* If a fixed number of PME nodes is set we do rcoulomb and PME gird tuning
1720 * only. We need to check whether the requested number of PME-only nodes
1722 if (npme_fixed > -1)
1724 /* No more than 50% of all nodes can be assigned as PME-only nodes. */
1725 if (2*npme_fixed > nnodes)
1727 gmx_fatal(FARGS, "Cannot have more than %d PME-only nodes for a total of %d nodes (you chose %d).\n",
1728 nnodes/2, nnodes, npme_fixed);
1730 if ((npme_fixed > 0) && (5*npme_fixed < nnodes))
1732 fprintf(stderr, "WARNING: Only %g percent of the nodes are assigned as PME-only nodes.\n",
1733 100.0*((real)npme_fixed / (real)nnodes));
1735 if (opt2parg_bSet("-min", npargs, pa) || opt2parg_bSet("-max", npargs, pa))
1737 fprintf(stderr, "NOTE: The -min, -max, and -npme options have no effect when a\n"
1738 " fixed number of PME-only nodes is requested with -fix.\n");
1744 /* Returns TRUE when "opt" is needed at launch time */
1745 static gmx_bool is_launch_file(char *opt, gmx_bool bSet)
1747 /* Apart from the input .tpr and the output log files we need all options that
1748 * were set on the command line and that do not start with -b */
1749 if (0 == strncmp(opt, "-b", 2) || 0 == strncmp(opt, "-s", 2)
1750 || 0 == strncmp(opt, "-err", 4) || 0 == strncmp(opt, "-p", 2) )
1759 /* Returns TRUE when "opt" defines a file which is needed for the benchmarks runs */
1760 static gmx_bool is_bench_file(char *opt, gmx_bool bSet, gmx_bool bOptional, gmx_bool bIsOutput)
1762 /* Apart from the input .tpr, all files starting with "-b" are for
1763 * _b_enchmark files exclusively */
1764 if (0 == strncmp(opt, "-s", 2))
1769 if (0 == strncmp(opt, "-b", 2) || 0 == strncmp(opt, "-s", 2))
1771 if (!bOptional || bSet)
1788 if (bSet) /* These are additional input files like -cpi -ei */
1801 /* Adds 'buf' to 'str' */
1802 static void add_to_string(char **str, char *buf)
1807 len = strlen(*str) + strlen(buf) + 1;
1813 /* Create the command line for the benchmark as well as for the real run */
1814 static void create_command_line_snippets(
1815 gmx_bool bAppendFiles,
1816 gmx_bool bKeepAndNumCPT,
1817 gmx_bool bResetHWay,
1821 char *cmd_args_bench[], /* command line arguments for benchmark runs */
1822 char *cmd_args_launch[], /* command line arguments for simulation run */
1823 char extra_args[]) /* Add this to the end of the command line */
1828 char strbuf[STRLEN];
1831 /* strlen needs at least '\0' as a string: */
1832 snew(*cmd_args_bench, 1);
1833 snew(*cmd_args_launch, 1);
1834 *cmd_args_launch[0] = '\0';
1835 *cmd_args_bench[0] = '\0';
1838 /*******************************************/
1839 /* 1. Process other command line arguments */
1840 /*******************************************/
1843 /* Add equilibration steps to benchmark options */
1844 sprintf(strbuf, "-resetstep %d ", presteps);
1845 add_to_string(cmd_args_bench, strbuf);
1847 /* These switches take effect only at launch time */
1848 if (FALSE == bAppendFiles)
1850 add_to_string(cmd_args_launch, "-noappend ");
1854 add_to_string(cmd_args_launch, "-cpnum ");
1858 add_to_string(cmd_args_launch, "-resethway ");
1861 /********************/
1862 /* 2. Process files */
1863 /********************/
1864 for (i = 0; i < nfile; i++)
1866 opt = (char *)fnm[i].opt;
1867 name = opt2fn(opt, nfile, fnm);
1869 /* Strbuf contains the options, now let's sort out where we need that */
1870 sprintf(strbuf, "%s %s ", opt, name);
1872 if (is_bench_file(opt, opt2bSet(opt, nfile, fnm), is_optional(&fnm[i]), is_output(&fnm[i])) )
1874 /* All options starting with -b* need the 'b' removed,
1875 * therefore overwrite strbuf */
1876 if (0 == strncmp(opt, "-b", 2))
1878 sprintf(strbuf, "-%s %s ", &opt[2], name);
1881 add_to_string(cmd_args_bench, strbuf);
1884 if (is_launch_file(opt, opt2bSet(opt, nfile, fnm)) )
1886 add_to_string(cmd_args_launch, strbuf);
1890 add_to_string(cmd_args_bench, extra_args);
1891 add_to_string(cmd_args_launch, extra_args);
1895 /* Set option opt */
1896 static void setopt(const char *opt, int nfile, t_filenm fnm[])
1900 for (i = 0; (i < nfile); i++)
1902 if (strcmp(opt, fnm[i].opt) == 0)
1904 fnm[i].flag |= ffSET;
1910 /* This routine inspects the tpr file and ...
1911 * 1. checks for output files that get triggered by a tpr option. These output
1912 * files are marked as 'set' to allow for a proper cleanup after each
1914 * 2. returns the PME:PP load ratio
1915 * 3. returns rcoulomb from the tpr */
1916 static float inspect_tpr(int nfile, t_filenm fnm[], real *rcoulomb)
1918 gmx_bool bPull; /* Is pulling requested in .tpr file? */
1919 gmx_bool bTpi; /* Is test particle insertion requested? */
1920 gmx_bool bFree; /* Is a free energy simulation requested? */
1921 gmx_bool bNM; /* Is a normal mode analysis requested? */
1927 /* Check tpr file for options that trigger extra output files */
1928 read_tpx_state(opt2fn("-s", nfile, fnm), &ir, &state, NULL, &mtop);
1929 bPull = (epullNO != ir.ePull);
1930 bFree = (efepNO != ir.efep );
1931 bNM = (eiNM == ir.eI );
1932 bTpi = EI_TPI(ir.eI);
1934 /* Set these output files on the tuning command-line */
1937 setopt("-pf", nfile, fnm);
1938 setopt("-px", nfile, fnm);
1942 setopt("-dhdl", nfile, fnm);
1946 setopt("-tpi", nfile, fnm);
1947 setopt("-tpid", nfile, fnm);
1951 setopt("-mtx", nfile, fnm);
1954 *rcoulomb = ir.rcoulomb;
1956 /* Return the estimate for the number of PME nodes */
1957 return pme_load_estimate(&mtop, &ir, state.box);
1961 static void couple_files_options(int nfile, t_filenm fnm[])
1964 gmx_bool bSet, bBench;
1969 for (i = 0; i < nfile; i++)
1971 opt = (char *)fnm[i].opt;
1972 bSet = ((fnm[i].flag & ffSET) != 0);
1973 bBench = (0 == strncmp(opt, "-b", 2));
1975 /* Check optional files */
1976 /* If e.g. -eo is set, then -beo also needs to be set */
1977 if (is_optional(&fnm[i]) && bSet && !bBench)
1979 sprintf(buf, "-b%s", &opt[1]);
1980 setopt(buf, nfile, fnm);
1982 /* If -beo is set, then -eo also needs to be! */
1983 if (is_optional(&fnm[i]) && bSet && bBench)
1985 sprintf(buf, "-%s", &opt[2]);
1986 setopt(buf, nfile, fnm);
1992 static double gettime()
1994 #ifdef HAVE_GETTIMEOFDAY
1998 gettimeofday(&t, NULL);
2000 seconds = (double) t.tv_sec + 1e-6*(double)t.tv_usec;
2006 seconds = time(NULL);
2013 #define BENCHSTEPS (1000)
2015 int gmx_tune_pme(int argc, char *argv[])
2017 const char *desc[] = {
2018 "For a given number [TT]-np[tt] or [TT]-ntmpi[tt] of processors/threads, this program systematically",
2019 "times [TT]mdrun[tt] with various numbers of PME-only nodes and determines",
2020 "which setting is fastest. It will also test whether performance can",
2021 "be enhanced by shifting load from the reciprocal to the real space",
2022 "part of the Ewald sum. ",
2023 "Simply pass your [TT].tpr[tt] file to [TT]g_tune_pme[tt] together with other options",
2024 "for [TT]mdrun[tt] as needed.[PAR]",
2025 "Which executables are used can be set in the environment variables",
2026 "MPIRUN and MDRUN. If these are not present, 'mpirun' and 'mdrun'",
2027 "will be used as defaults. Note that for certain MPI frameworks you",
2028 "need to provide a machine- or hostfile. This can also be passed",
2029 "via the MPIRUN variable, e.g.[PAR]",
2030 "[TT]export MPIRUN=\"/usr/local/mpirun -machinefile hosts\"[tt][PAR]",
2031 "Please call [TT]g_tune_pme[tt] with the normal options you would pass to",
2032 "[TT]mdrun[tt] and add [TT]-np[tt] for the number of processors to perform the",
2033 "tests on, or [TT]-ntmpi[tt] for the number of threads. You can also add [TT]-r[tt]",
2034 "to repeat each test several times to get better statistics. [PAR]",
2035 "[TT]g_tune_pme[tt] can test various real space / reciprocal space workloads",
2036 "for you. With [TT]-ntpr[tt] you control how many extra [TT].tpr[tt] files will be",
2037 "written with enlarged cutoffs and smaller Fourier grids respectively.",
2038 "Typically, the first test (number 0) will be with the settings from the input",
2039 "[TT].tpr[tt] file; the last test (number [TT]ntpr[tt]) will have the Coulomb cutoff",
2040 "specified by [TT]-rmax[tt] with a somwhat smaller PME grid at the same time. ",
2041 "In this last test, the Fourier spacing is multiplied with [TT]rmax[tt]/rcoulomb. ",
2042 "The remaining [TT].tpr[tt] files will have equally-spaced Coulomb radii (and Fourier "
2043 "spacings) between these extremes. [BB]Note[bb] that you can set [TT]-ntpr[tt] to 1",
2044 "if you just seek the optimal number of PME-only nodes; in that case",
2045 "your input [TT].tpr[tt] file will remain unchanged.[PAR]",
2046 "For the benchmark runs, the default of 1000 time steps should suffice for most",
2047 "MD systems. The dynamic load balancing needs about 100 time steps",
2048 "to adapt to local load imbalances, therefore the time step counters",
2049 "are by default reset after 100 steps. For large systems (>1M atoms), as well as ",
2050 "for a higher accuarcy of the measurements, you should set [TT]-resetstep[tt] to a higher value.",
2051 "From the 'DD' load imbalance entries in the md.log output file you",
2052 "can tell after how many steps the load is sufficiently balanced. Example call:[PAR]"
2053 "[TT]g_tune_pme -np 64 -s protein.tpr -launch[tt][PAR]",
2054 "After calling [TT]mdrun[tt] several times, detailed performance information",
2055 "is available in the output file [TT]perf.out.[tt] ",
2056 "[BB]Note[bb] that during the benchmarks, a couple of temporary files are written",
2057 "(options [TT]-b[tt]*), these will be automatically deleted after each test.[PAR]",
2058 "If you want the simulation to be started automatically with the",
2059 "optimized parameters, use the command line option [TT]-launch[tt].[PAR]",
2064 int pmeentries = 0; /* How many values for -npme do we actually test for each tpr file */
2065 real maxPMEfraction = 0.50;
2066 real minPMEfraction = 0.25;
2067 int maxPMEnodes, minPMEnodes;
2068 float guessPMEratio; /* guessed PME:PP ratio based on the tpr file */
2069 float guessPMEnodes;
2070 int npme_fixed = -2; /* If >= -1, use only this number
2071 * of PME-only nodes */
2073 real rmin = 0.0, rmax = 0.0; /* min and max value for rcoulomb if scaling is requested */
2074 real rcoulomb = -1.0; /* Coulomb radius as set in .tpr file */
2075 gmx_bool bScaleRvdw = TRUE;
2076 gmx_large_int_t bench_nsteps = BENCHSTEPS;
2077 gmx_large_int_t new_sim_nsteps = -1; /* -1 indicates: not set by the user */
2078 gmx_large_int_t cpt_steps = 0; /* Step counter in .cpt input file */
2079 int presteps = 100; /* Do a full cycle reset after presteps steps */
2080 gmx_bool bOverwrite = FALSE, bKeepTPR;
2081 gmx_bool bLaunch = FALSE;
2082 char *ExtraArgs = NULL;
2083 char **tpr_names = NULL;
2084 const char *simulation_tpr = NULL;
2085 int best_npme, best_tpr;
2086 int sim_part = 1; /* For benchmarks with checkpoint files */
2089 /* Default program names if nothing else is found */
2090 char *cmd_mpirun = NULL, *cmd_mdrun = NULL;
2091 char *cmd_args_bench, *cmd_args_launch;
2092 char *cmd_np = NULL;
2094 t_perf **perfdata = NULL;
2100 /* Print out how long the tuning took */
2103 static t_filenm fnm[] = {
2105 { efOUT, "-p", "perf", ffWRITE },
2106 { efLOG, "-err", "bencherr", ffWRITE },
2107 { efTPX, "-so", "tuned", ffWRITE },
2109 { efTPX, NULL, NULL, ffREAD },
2110 { efTRN, "-o", NULL, ffWRITE },
2111 { efXTC, "-x", NULL, ffOPTWR },
2112 { efCPT, "-cpi", NULL, ffOPTRD },
2113 { efCPT, "-cpo", NULL, ffOPTWR },
2114 { efSTO, "-c", "confout", ffWRITE },
2115 { efEDR, "-e", "ener", ffWRITE },
2116 { efLOG, "-g", "md", ffWRITE },
2117 { efXVG, "-dhdl", "dhdl", ffOPTWR },
2118 { efXVG, "-field", "field", ffOPTWR },
2119 { efXVG, "-table", "table", ffOPTRD },
2120 { efXVG, "-tabletf", "tabletf", ffOPTRD },
2121 { efXVG, "-tablep", "tablep", ffOPTRD },
2122 { efXVG, "-tableb", "table", ffOPTRD },
2123 { efTRX, "-rerun", "rerun", ffOPTRD },
2124 { efXVG, "-tpi", "tpi", ffOPTWR },
2125 { efXVG, "-tpid", "tpidist", ffOPTWR },
2126 { efEDI, "-ei", "sam", ffOPTRD },
2127 { efXVG, "-eo", "edsam", ffOPTWR },
2128 { efGCT, "-j", "wham", ffOPTRD },
2129 { efGCT, "-jo", "bam", ffOPTWR },
2130 { efXVG, "-ffout", "gct", ffOPTWR },
2131 { efXVG, "-devout", "deviatie", ffOPTWR },
2132 { efXVG, "-runav", "runaver", ffOPTWR },
2133 { efXVG, "-px", "pullx", ffOPTWR },
2134 { efXVG, "-pf", "pullf", ffOPTWR },
2135 { efXVG, "-ro", "rotation", ffOPTWR },
2136 { efLOG, "-ra", "rotangles", ffOPTWR },
2137 { efLOG, "-rs", "rotslabs", ffOPTWR },
2138 { efLOG, "-rt", "rottorque", ffOPTWR },
2139 { efMTX, "-mtx", "nm", ffOPTWR },
2140 { efNDX, "-dn", "dipole", ffOPTWR },
2141 /* Output files that are deleted after each benchmark run */
2142 { efTRN, "-bo", "bench", ffWRITE },
2143 { efXTC, "-bx", "bench", ffWRITE },
2144 { efCPT, "-bcpo", "bench", ffWRITE },
2145 { efSTO, "-bc", "bench", ffWRITE },
2146 { efEDR, "-be", "bench", ffWRITE },
2147 { efLOG, "-bg", "bench", ffWRITE },
2148 { efXVG, "-beo", "benchedo", ffOPTWR },
2149 { efXVG, "-bdhdl", "benchdhdl", ffOPTWR },
2150 { efXVG, "-bfield", "benchfld", ffOPTWR },
2151 { efXVG, "-btpi", "benchtpi", ffOPTWR },
2152 { efXVG, "-btpid", "benchtpid", ffOPTWR },
2153 { efGCT, "-bjo", "bench", ffOPTWR },
2154 { efXVG, "-bffout", "benchgct", ffOPTWR },
2155 { efXVG, "-bdevout", "benchdev", ffOPTWR },
2156 { efXVG, "-brunav", "benchrnav", ffOPTWR },
2157 { efXVG, "-bpx", "benchpx", ffOPTWR },
2158 { efXVG, "-bpf", "benchpf", ffOPTWR },
2159 { efXVG, "-bro", "benchrot", ffOPTWR },
2160 { efLOG, "-bra", "benchrota", ffOPTWR },
2161 { efLOG, "-brs", "benchrots", ffOPTWR },
2162 { efLOG, "-brt", "benchrott", ffOPTWR },
2163 { efMTX, "-bmtx", "benchn", ffOPTWR },
2164 { efNDX, "-bdn", "bench", ffOPTWR }
2167 gmx_bool bThreads = FALSE;
2171 const char *procstring[] =
2172 { NULL, "-np", "-n", "none", NULL };
2173 const char *npmevalues_opt[] =
2174 { NULL, "auto", "all", "subset", NULL };
2176 gmx_bool bAppendFiles = TRUE;
2177 gmx_bool bKeepAndNumCPT = FALSE;
2178 gmx_bool bResetCountersHalfWay = FALSE;
2179 gmx_bool bBenchmark = TRUE;
2181 output_env_t oenv = NULL;
2184 /***********************/
2185 /* g_tune_pme options: */
2186 /***********************/
2187 { "-np", FALSE, etINT, {&nnodes},
2188 "Number of nodes to run the tests on (must be > 2 for separate PME nodes)" },
2189 { "-npstring", FALSE, etENUM, {procstring},
2190 "Specify the number of processors to [TT]$MPIRUN[tt] using this string"},
2191 { "-ntmpi", FALSE, etINT, {&nthreads},
2192 "Number of MPI-threads to run the tests on (turns MPI & mpirun off)"},
2193 { "-r", FALSE, etINT, {&repeats},
2194 "Repeat each test this often" },
2195 { "-max", FALSE, etREAL, {&maxPMEfraction},
2196 "Max fraction of PME nodes to test with" },
2197 { "-min", FALSE, etREAL, {&minPMEfraction},
2198 "Min fraction of PME nodes to test with" },
2199 { "-npme", FALSE, etENUM, {npmevalues_opt},
2200 "Within -min and -max, benchmark all possible values for [TT]-npme[tt], or just a reasonable subset. "
2201 "Auto neglects -min and -max and chooses reasonable values around a guess for npme derived from the .tpr"},
2202 { "-fix", FALSE, etINT, {&npme_fixed},
2203 "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."},
2204 { "-rmax", FALSE, etREAL, {&rmax},
2205 "If >0, maximal rcoulomb for -ntpr>1 (rcoulomb upscaling results in fourier grid downscaling)" },
2206 { "-rmin", FALSE, etREAL, {&rmin},
2207 "If >0, minimal rcoulomb for -ntpr>1" },
2208 { "-scalevdw", FALSE, etBOOL, {&bScaleRvdw},
2209 "Scale rvdw along with rcoulomb"},
2210 { "-ntpr", FALSE, etINT, {&ntprs},
2211 "Number of [TT].tpr[tt] files to benchmark. Create this many files with different rcoulomb scaling factors depending on -rmin and -rmax. "
2212 "If < 1, automatically choose the number of [TT].tpr[tt] files to test" },
2213 { "-steps", FALSE, etGMX_LARGE_INT, {&bench_nsteps},
2214 "Take timings for this many steps in the benchmark runs" },
2215 { "-resetstep", FALSE, etINT, {&presteps},
2216 "Let dlb equilibrate this many steps before timings are taken (reset cycle counters after this many steps)" },
2217 { "-simsteps", FALSE, etGMX_LARGE_INT, {&new_sim_nsteps},
2218 "If non-negative, perform this many steps in the real run (overwrites nsteps from [TT].tpr[tt], add [TT].cpt[tt] steps)" },
2219 { "-launch", FALSE, etBOOL, {&bLaunch},
2220 "Launch the real simulation after optimization" },
2221 { "-bench", FALSE, etBOOL, {&bBenchmark},
2222 "Run the benchmarks or just create the input [TT].tpr[tt] files?" },
2223 /******************/
2224 /* mdrun options: */
2225 /******************/
2226 /* We let g_tune_pme parse and understand these options, because we need to
2227 * prevent that they appear on the mdrun command line for the benchmarks */
2228 { "-append", FALSE, etBOOL, {&bAppendFiles},
2229 "Append to previous output files when continuing from checkpoint instead of adding the simulation part number to all file names (for launch only)" },
2230 { "-cpnum", FALSE, etBOOL, {&bKeepAndNumCPT},
2231 "Keep and number checkpoint files (launch only)" },
2232 { "-resethway", FALSE, etBOOL, {&bResetCountersHalfWay},
2233 "HIDDENReset the cycle counters after half the number of steps or halfway [TT]-maxh[tt] (launch only)" }
2236 #define NFILE asize(fnm)
2238 seconds = gettime();
2240 if (!parse_common_args(&argc, argv, PCA_NOEXIT_ON_ARGS,
2241 NFILE, fnm, asize(pa), pa, asize(desc), desc,
2247 /* Store the remaining unparsed command line entries in a string which
2248 * is then attached to the mdrun command line */
2250 ExtraArgs[0] = '\0';
2251 for (i = 1; i < argc; i++) /* argc will now be 1 if everything was understood */
2253 add_to_string(&ExtraArgs, argv[i]);
2254 add_to_string(&ExtraArgs, " ");
2257 if (opt2parg_bSet("-ntmpi", asize(pa), pa))
2260 if (opt2parg_bSet("-npstring", asize(pa), pa))
2262 fprintf(stderr, "WARNING: -npstring has no effect when using threads.\n");
2267 gmx_fatal(FARGS, "Can't run multi-threaded MPI simulation yet!");
2269 /* and now we just set this; a bit of an ugly hack*/
2272 /* Check for PME:PP ratio and whether tpr triggers additional output files */
2273 guessPMEratio = inspect_tpr(NFILE, fnm, &rcoulomb);
2275 /* Automatically set -beo options if -eo is set etc. */
2276 couple_files_options(NFILE, fnm);
2278 /* Construct the command line arguments for benchmark runs
2279 * as well as for the simulation run */
2282 sprintf(bbuf, " -ntmpi %d ", nthreads);
2286 /* This string will be used for MPI runs and will appear after the
2287 * mpirun command. */
2288 if (strcmp(procstring[0], "none") != 0)
2290 sprintf(bbuf, " %s %d ", procstring[0], nnodes);
2300 create_command_line_snippets(bAppendFiles, bKeepAndNumCPT, bResetCountersHalfWay, presteps,
2301 NFILE, fnm, &cmd_args_bench, &cmd_args_launch, ExtraArgs);
2303 /* Read in checkpoint file if requested */
2305 if (opt2bSet("-cpi", NFILE, fnm))
2308 cr->duty = DUTY_PP; /* makes the following routine happy */
2309 read_checkpoint_simulation_part(opt2fn("-cpi", NFILE, fnm),
2310 &sim_part, &cpt_steps, cr,
2311 FALSE, NFILE, fnm, NULL, NULL);
2314 /* sim_part will now be 1 if no checkpoint file was found */
2317 gmx_fatal(FARGS, "Checkpoint file %s not found!", opt2fn("-cpi", NFILE, fnm));
2321 /* Open performance output file and write header info */
2322 fp = ffopen(opt2fn("-p", NFILE, fnm), "w");
2324 /* Make a quick consistency check of command line parameters */
2325 check_input(nnodes, repeats, &ntprs, &rmin, rcoulomb, &rmax,
2326 maxPMEfraction, minPMEfraction, npme_fixed,
2327 bench_nsteps, fnm, NFILE, sim_part, presteps,
2330 /* Determine the maximum and minimum number of PME nodes to test,
2331 * the actual list of settings is build in do_the_tests(). */
2332 if ((nnodes > 2) && (npme_fixed < -1))
2334 if (0 == strcmp(npmevalues_opt[0], "auto"))
2336 /* Determine the npme range automatically based on the PME:PP load guess */
2337 if (guessPMEratio > 1.0)
2339 /* More PME than PP work, probably we do not need separate PME nodes at all! */
2340 maxPMEnodes = nnodes/2;
2341 minPMEnodes = nnodes/2;
2345 /* PME : PP load is in the range 0..1, let's test around the guess */
2346 guessPMEnodes = nnodes/(1.0 + 1.0/guessPMEratio);
2347 minPMEnodes = floor(0.7*guessPMEnodes);
2348 maxPMEnodes = ceil(1.6*guessPMEnodes);
2349 maxPMEnodes = min(maxPMEnodes, nnodes/2);
2354 /* Determine the npme range based on user input */
2355 maxPMEnodes = floor(maxPMEfraction*nnodes);
2356 minPMEnodes = max(floor(minPMEfraction*nnodes), 0);
2357 fprintf(stdout, "Will try runs with %d ", minPMEnodes);
2358 if (maxPMEnodes != minPMEnodes)
2360 fprintf(stdout, "- %d ", maxPMEnodes);
2362 fprintf(stdout, "PME-only nodes.\n Note that the automatic number of PME-only nodes and no separate PME nodes are always tested.\n");
2371 /* Get the commands we need to set up the runs from environment variables */
2372 get_program_paths(bThreads, &cmd_mpirun, cmd_np, &cmd_mdrun, repeats);
2373 if (bBenchmark && repeats > 0)
2375 check_mdrun_works(bThreads, cmd_mpirun, cmd_np, cmd_mdrun);
2378 /* Print some header info to file */
2380 fprintf(fp, "\n P E R F O R M A N C E R E S U L T S\n");
2382 fprintf(fp, "%s for Gromacs %s\n", ShortProgram(), GromacsVersion());
2385 fprintf(fp, "Number of nodes : %d\n", nnodes);
2386 fprintf(fp, "The mpirun command is : %s\n", cmd_mpirun);
2387 if (strcmp(procstring[0], "none") != 0)
2389 fprintf(fp, "Passing # of nodes via : %s\n", procstring[0]);
2393 fprintf(fp, "Not setting number of nodes in system call\n");
2398 fprintf(fp, "Number of threads : %d\n", nnodes);
2401 fprintf(fp, "The mdrun command is : %s\n", cmd_mdrun);
2402 fprintf(fp, "mdrun args benchmarks : %s\n", cmd_args_bench);
2403 fprintf(fp, "Benchmark steps : ");
2404 fprintf(fp, gmx_large_int_pfmt, bench_nsteps);
2406 fprintf(fp, "dlb equilibration steps : %d\n", presteps);
2409 fprintf(fp, "Checkpoint time step : ");
2410 fprintf(fp, gmx_large_int_pfmt, cpt_steps);
2413 fprintf(fp, "mdrun args at launchtime: %s\n", cmd_args_launch);
2415 if (new_sim_nsteps >= 0)
2418 fprintf(stderr, "Note: Simulation input file %s will have ", opt2fn("-so", NFILE, fnm));
2419 fprintf(stderr, gmx_large_int_pfmt, new_sim_nsteps+cpt_steps);
2420 fprintf(stderr, " steps.\n");
2421 fprintf(fp, "Simulation steps : ");
2422 fprintf(fp, gmx_large_int_pfmt, new_sim_nsteps);
2427 fprintf(fp, "Repeats for each test : %d\n", repeats);
2430 if (npme_fixed >= -1)
2432 fprintf(fp, "Fixing -npme at : %d\n", npme_fixed);
2435 fprintf(fp, "Input file : %s\n", opt2fn("-s", NFILE, fnm));
2436 fprintf(fp, " PME/PP load estimate : %g\n", guessPMEratio);
2438 /* Allocate memory for the inputinfo struct: */
2440 info->nr_inputfiles = ntprs;
2441 for (i = 0; i < ntprs; i++)
2443 snew(info->rcoulomb, ntprs);
2444 snew(info->rvdw, ntprs);
2445 snew(info->rlist, ntprs);
2446 snew(info->rlistlong, ntprs);
2447 snew(info->nkx, ntprs);
2448 snew(info->nky, ntprs);
2449 snew(info->nkz, ntprs);
2450 snew(info->fsx, ntprs);
2451 snew(info->fsy, ntprs);
2452 snew(info->fsz, ntprs);
2454 /* Make alternative tpr files to test: */
2455 snew(tpr_names, ntprs);
2456 for (i = 0; i < ntprs; i++)
2458 snew(tpr_names[i], STRLEN);
2461 /* It can be that ntprs is reduced by make_benchmark_tprs if not enough
2462 * different grids could be found. */
2463 make_benchmark_tprs(opt2fn("-s", NFILE, fnm), tpr_names, bench_nsteps+presteps,
2464 cpt_steps, rmin, rmax, bScaleRvdw, &ntprs, info, fp);
2466 /********************************************************************************/
2467 /* Main loop over all scenarios we need to test: tpr files, PME nodes, repeats */
2468 /********************************************************************************/
2469 snew(perfdata, ntprs);
2472 do_the_tests(fp, tpr_names, maxPMEnodes, minPMEnodes, npme_fixed, npmevalues_opt[0], perfdata, &pmeentries,
2473 repeats, nnodes, ntprs, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2474 cmd_args_bench, fnm, NFILE, presteps, cpt_steps);
2476 fprintf(fp, "\nTuning took%8.1f minutes.\n", (gettime()-seconds)/60.0);
2478 /* Analyse the results and give a suggestion for optimal settings: */
2479 bKeepTPR = analyze_data(fp, opt2fn("-p", NFILE, fnm), perfdata, nnodes, ntprs, pmeentries,
2480 repeats, info, &best_tpr, &best_npme);
2482 /* Take the best-performing tpr file and enlarge nsteps to original value */
2483 if (bKeepTPR && !bOverwrite)
2485 simulation_tpr = opt2fn("-s", NFILE, fnm);
2489 simulation_tpr = opt2fn("-so", NFILE, fnm);
2490 modify_PMEsettings(bOverwrite ? (new_sim_nsteps+cpt_steps) : info->orig_sim_steps,
2491 info->orig_init_step, tpr_names[best_tpr], simulation_tpr);
2494 /* Let's get rid of the temporary benchmark input files */
2495 for (i = 0; i < ntprs; i++)
2497 fprintf(stdout, "Deleting temporary benchmark input file %s\n", tpr_names[i]);
2498 remove(tpr_names[i]);
2501 /* Now start the real simulation if the user requested it ... */
2502 launch_simulation(bLaunch, fp, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2503 cmd_args_launch, simulation_tpr, best_npme);
2507 /* ... or simply print the performance results to screen: */
2510 finalize(opt2fn("-p", NFILE, fnm));