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"
63 /* Enum for situations that can occur during log file parsing, the
64 * corresponding string entries can be found in do_the_tests() in
65 * const char* ParseLog[] */
71 eParselogResetProblem,
82 int nPMEnodes; /* number of PME-only nodes used in this test */
83 int nx, ny, nz; /* DD grid */
84 int guessPME; /* if nPMEnodes == -1, this is the guessed number of PME nodes */
85 double *Gcycles; /* This can contain more than one value if doing multiple tests */
89 float *PME_f_load; /* PME mesh/force load average*/
90 float PME_f_load_Av; /* Average average ;) ... */
91 char *mdrun_cmd_line; /* Mdrun command line used for this test */
97 int nr_inputfiles; /* The number of tpr and mdp input files */
98 gmx_large_int_t orig_sim_steps; /* Number of steps to be done in the real simulation */
99 real *rcoulomb; /* The coulomb radii [0...nr_inputfiles] */
100 real *rvdw; /* The vdW radii */
101 real *rlist; /* Neighbourlist cutoff radius */
103 int *nkx, *nky, *nkz;
104 real *fsx, *fsy, *fsz; /* Fourierspacing in x,y,z dimension */
108 static void sep_line(FILE *fp)
110 fprintf(fp, "\n------------------------------------------------------------\n");
114 /* Wrapper for system calls */
115 static int gmx_system_call(char *command)
118 gmx_fatal(FARGS,"No calls to system(3) supported on this platform. Attempted to call:\n'%s'\n",command);
120 return ( system(command) );
125 /* Check if string starts with substring */
126 static gmx_bool str_starts(const char *string, const char *substring)
128 return ( strncmp(string, substring, strlen(substring)) == 0);
132 static void cleandata(t_perf *perfdata, int test_nr)
134 perfdata->Gcycles[test_nr] = 0.0;
135 perfdata->ns_per_day[test_nr] = 0.0;
136 perfdata->PME_f_load[test_nr] = 0.0;
142 static gmx_bool is_equal(real a, real b)
144 real diff, eps=1.0e-7;
149 if (diff < 0.0) diff = -diff;
158 static void finalize(const char *fn_out)
164 fp = fopen(fn_out,"r");
165 fprintf(stdout,"\n\n");
167 while( fgets(buf,STRLEN-1,fp) != NULL )
169 fprintf(stdout,"%s",buf);
172 fprintf(stdout,"\n\n");
176 enum {eFoundNothing, eFoundDDStr, eFoundAccountingStr, eFoundCycleStr};
178 static int parse_logfile(const char *logfile, const char *errfile,
179 t_perf *perfdata, int test_nr, int presteps, gmx_large_int_t cpt_steps,
183 char line[STRLEN], dumstring[STRLEN], dumstring2[STRLEN];
184 const char matchstrdd[]="Domain decomposition grid";
185 const char matchstrcr[]="resetting all time and cycle counters";
186 const char matchstrbal[]="Average PME mesh/force load:";
187 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";
188 const char errSIG[]="signal, stopping at the next";
191 float dum1,dum2,dum3,dum4;
194 gmx_large_int_t resetsteps=-1;
195 gmx_bool bFoundResetStr = FALSE;
196 gmx_bool bResetChecked = FALSE;
199 if (!gmx_fexist(logfile))
201 fprintf(stderr, "WARNING: Could not find logfile %s.\n", logfile);
202 cleandata(perfdata, test_nr);
203 return eParselogNotFound;
206 fp = fopen(logfile, "r");
207 perfdata->PME_f_load[test_nr] = -1.0;
208 perfdata->guessPME = -1;
210 iFound = eFoundNothing;
212 iFound = eFoundDDStr; /* Skip some case statements */
214 while (fgets(line, STRLEN, fp) != NULL)
216 /* Remove leading spaces */
219 /* Check for TERM and INT signals from user: */
220 if ( strstr(line, errSIG) != NULL )
223 cleandata(perfdata, test_nr);
224 return eParselogTerm;
227 /* Check whether cycle resetting worked */
228 if (presteps > 0 && !bFoundResetStr)
230 if (strstr(line, matchstrcr) != NULL)
232 sprintf(dumstring, "step %s", gmx_large_int_pfmt);
233 sscanf(line, dumstring, &resetsteps);
234 bFoundResetStr = TRUE;
235 if (resetsteps == presteps+cpt_steps)
237 bResetChecked = TRUE;
241 sprintf(dumstring , gmx_large_int_pfmt, resetsteps);
242 sprintf(dumstring2, gmx_large_int_pfmt, presteps+cpt_steps);
243 fprintf(stderr, "WARNING: Time step counters were reset at step %s,\n"
244 " though they were supposed to be reset at step %s!\n",
245 dumstring, dumstring2);
250 /* Look for strings that appear in a certain order in the log file: */
254 /* Look for domain decomp grid and separate PME nodes: */
255 if (str_starts(line, matchstrdd))
257 sscanf(line, "Domain decomposition grid %d x %d x %d, separate PME nodes %d",
258 &(perfdata->nx), &(perfdata->ny), &(perfdata->nz), &npme);
259 if (perfdata->nPMEnodes == -1)
260 perfdata->guessPME = npme;
261 else if (perfdata->nPMEnodes != npme)
262 gmx_fatal(FARGS, "PME nodes from command line and output file are not identical");
263 iFound = eFoundDDStr;
265 /* Catch a few errors that might have occured: */
266 else if (str_starts(line, "There is no domain decomposition for"))
269 return eParselogNoDDGrid;
271 else if (str_starts(line, "reading tpx file"))
274 return eParselogTPXVersion;
276 else if (str_starts(line, "The -dd or -npme option request a parallel simulation"))
279 return eParselogNotParallel;
283 /* Look for PME mesh/force balance (not necessarily present, though) */
284 if (str_starts(line, matchstrbal))
285 sscanf(&line[strlen(matchstrbal)], "%f", &(perfdata->PME_f_load[test_nr]));
286 /* Look for matchstring */
287 if (str_starts(line, matchstring))
288 iFound = eFoundAccountingStr;
290 case eFoundAccountingStr:
291 /* Already found matchstring - look for cycle data */
292 if (str_starts(line, "Total "))
294 sscanf(line,"Total %d %lf",&procs,&(perfdata->Gcycles[test_nr]));
295 iFound = eFoundCycleStr;
299 /* Already found cycle data - look for remaining performance info and return */
300 if (str_starts(line, "Performance:"))
302 ndum = sscanf(line,"%s %f %f %f %f", dumstring, &dum1, &dum2, &dum3, &dum4);
303 /* (ns/day) is the second last entry, depending on whether GMX_DETAILED_PERF_STATS was set in print_perf(), nrnb.c */
304 perfdata->ns_per_day[test_nr] = (ndum==5)? dum3 : dum1;
306 if (bResetChecked || presteps == 0)
309 return eParselogResetProblem;
315 /* Close the log file */
318 /* Check why there is no performance data in the log file.
319 * Did a fatal errors occur? */
320 if (gmx_fexist(errfile))
322 fp = fopen(errfile, "r");
323 while (fgets(line, STRLEN, fp) != NULL)
325 if ( str_starts(line, "Fatal error:") )
327 if (fgets(line, STRLEN, fp) != NULL)
328 fprintf(stderr, "\nWARNING: An error occured during this benchmark:\n"
331 cleandata(perfdata, test_nr);
332 return eParselogFatal;
339 fprintf(stderr, "WARNING: Could not find stderr file %s.\n", errfile);
342 /* Giving up ... we could not find out why there is no performance data in
344 fprintf(stdout, "No performance data in log file.\n");
345 cleandata(perfdata, test_nr);
347 return eParselogNoPerfData;
351 static gmx_bool analyze_data(
360 int *index_tpr, /* OUT: Nr of mdp file with best settings */
361 int *npme_optimal) /* OUT: Optimal number of PME nodes */
364 int line=0, line_win=-1;
365 int k_win=-1, i_win=-1, winPME;
366 double s=0.0; /* standard deviation */
369 char str_PME_f_load[13];
370 gmx_bool bCanUseOrigTPR;
371 gmx_bool bRefinedCoul, bRefinedVdW, bRefinedGrid;
377 fprintf(fp, "Summary of successful runs:\n");
378 fprintf(fp, "Line tpr PME nodes Gcycles Av. Std.dev. ns/day PME/f");
380 fprintf(fp, " DD grid");
385 for (k=0; k<ntprs; k++)
387 for (i=0; i<ntests; i++)
389 /* Select the right dataset: */
390 pd = &(perfdata[k][i]);
392 pd->Gcycles_Av = 0.0;
393 pd->PME_f_load_Av = 0.0;
394 pd->ns_per_day_Av = 0.0;
396 if (pd->nPMEnodes == -1)
397 sprintf(strbuf, "(%3d)", pd->guessPME);
399 sprintf(strbuf, " ");
401 /* Get the average run time of a setting */
402 for (j=0; j<nrepeats; j++)
404 pd->Gcycles_Av += pd->Gcycles[j];
405 pd->PME_f_load_Av += pd->PME_f_load[j];
407 pd->Gcycles_Av /= nrepeats;
408 pd->PME_f_load_Av /= nrepeats;
410 for (j=0; j<nrepeats; j++)
412 if (pd->ns_per_day[j] > 0.0)
413 pd->ns_per_day_Av += pd->ns_per_day[j];
416 /* Somehow the performance number was not aquired for this run,
417 * therefor set the average to some negative value: */
418 pd->ns_per_day_Av = -1.0f*nrepeats;
422 pd->ns_per_day_Av /= nrepeats;
425 if (pd->PME_f_load_Av > 0.0)
426 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load_Av);
428 sprintf(str_PME_f_load, "%s", " - ");
431 /* We assume we had a successful run if both averages are positive */
432 if (pd->Gcycles_Av > 0.0 && pd->ns_per_day_Av > 0.0)
434 /* Output statistics if repeats were done */
437 /* Calculate the standard deviation */
439 for (j=0; j<nrepeats; j++)
440 s += pow( pd->Gcycles[j] - pd->Gcycles_Av, 2 );
444 fprintf(fp, "%4d %3d %4d%s %12.3f %12.3f %12.3f %s",
445 line, k, pd->nPMEnodes, strbuf, pd->Gcycles_Av, s,
446 pd->ns_per_day_Av, str_PME_f_load);
448 fprintf(fp, " %3d %3d %3d", pd->nx, pd->ny, pd->nz);
451 /* Store the index of the best run found so far in 'winner': */
452 if ( (k_win == -1) || (pd->Gcycles_Av < perfdata[k_win][i_win].Gcycles_Av) )
464 gmx_fatal(FARGS, "None of the runs was successful! Check %s for problems.", fn);
468 winPME = perfdata[k_win][i_win].nPMEnodes;
472 /* We stuck to a fixed number of PME-only nodes */
473 sprintf(strbuf, "settings No. %d", k_win);
477 /* We have optimized the number of PME-only nodes */
479 sprintf(strbuf, "%s", "the automatic number of PME nodes");
481 sprintf(strbuf, "%d PME nodes", winPME);
483 fprintf(fp, "Best performance was achieved with %s", strbuf);
484 if ((nrepeats > 1) && (ntests > 1))
485 fprintf(fp, " (see line %d)", line_win);
488 /* Only mention settings if they were modified: */
489 bRefinedCoul = !is_equal(info->rcoulomb[k_win], info->rcoulomb[0]);
490 bRefinedVdW = !is_equal(info->rvdw[k_win] , info->rvdw[0] );
491 bRefinedGrid = !(info->nkx[k_win] == info->nkx[0] &&
492 info->nky[k_win] == info->nky[0] &&
493 info->nkz[k_win] == info->nkz[0]);
495 if (bRefinedCoul || bRefinedVdW || bRefinedGrid)
497 fprintf(fp, "Optimized PME settings:\n");
498 bCanUseOrigTPR = FALSE;
502 bCanUseOrigTPR = TRUE;
506 fprintf(fp, " New Coulomb radius: %f nm (was %f nm)\n", info->rcoulomb[k_win], info->rcoulomb[0]);
509 fprintf(fp, " New Van der Waals radius: %f nm (was %f nm)\n", info->rvdw[k_win], info->rvdw[0]);
512 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],
513 info->nkx[0], info->nky[0], info->nkz[0]);
515 if (bCanUseOrigTPR && ntprs > 1)
516 fprintf(fp, "and original PME settings.\n");
520 /* Return the index of the mdp file that showed the highest performance
521 * and the optimal number of PME nodes */
523 *npme_optimal = winPME;
525 return bCanUseOrigTPR;
529 /* Get the commands we need to set up the runs from environment variables */
530 static void get_program_paths(gmx_bool bThreads, char *cmd_mpirun[], char cmd_np[],
531 char *cmd_mdrun[], int repeats)
538 const char def_mpirun[] = "mpirun";
539 const char def_mdrun[] = "mdrun";
540 const char filename[] = "benchtest.log";
541 const char match_mpi[] = "NNODES=";
542 const char match_mdrun[]= "Program: ";
543 const char empty_mpirun[] = "";
544 gmx_bool bMdrun = FALSE;
545 gmx_bool bMPI = FALSE;
548 /* Get the commands we need to set up the runs from environment variables */
551 if ( (cp = getenv("MPIRUN")) != NULL)
552 *cmd_mpirun = strdup(cp);
554 *cmd_mpirun = strdup(def_mpirun);
558 *cmd_mpirun = strdup(empty_mpirun);
561 if ( (cp = getenv("MDRUN" )) != NULL )
562 *cmd_mdrun = strdup(cp);
564 *cmd_mdrun = strdup(def_mdrun);
567 /* If no simulations have to be performed, we are done here */
571 /* Run a small test to see whether mpirun + mdrun work */
572 fprintf(stdout, "Making sure that mdrun can be executed. ");
575 snew(command, strlen(*cmd_mdrun) + strlen(cmd_np) + strlen(filename) + 50);
576 sprintf(command, "%s%s-version -maxh 0.001 1> %s 2>&1", *cmd_mdrun, cmd_np, filename);
580 snew(command, strlen(*cmd_mpirun) + strlen(cmd_np) + strlen(*cmd_mdrun) + strlen(filename) + 50);
581 sprintf(command, "%s%s%s -version -maxh 0.001 1> %s 2>&1", *cmd_mpirun, cmd_np, *cmd_mdrun, filename);
583 fprintf(stdout, "Trying '%s' ... ", command);
584 make_backup(filename);
585 gmx_system_call(command);
587 /* Check if we find the characteristic string in the output: */
588 if (!gmx_fexist(filename))
589 gmx_fatal(FARGS, "Output from test run could not be found.");
591 fp = fopen(filename, "r");
592 /* We need to scan the whole output file, since sometimes the queuing system
593 * also writes stuff to stdout/err */
596 cp2=fgets(line, STRLEN, fp);
599 if ( str_starts(line, match_mdrun) )
601 if ( str_starts(line, match_mpi) )
611 gmx_fatal(FARGS, "Need a threaded version of mdrun. This one\n"
613 "seems to have been compiled with MPI instead.",
621 gmx_fatal(FARGS, "Need an MPI-enabled version of mdrun. This one\n"
623 "seems to have been compiled without MPI support.",
630 gmx_fatal(FARGS, "Cannot execute mdrun. Please check %s for problems!",
634 fprintf(stdout, "passed.\n");
642 static void launch_simulation(
643 gmx_bool bLaunch, /* Should the simulation be launched? */
644 FILE *fp, /* General log file */
645 gmx_bool bThreads, /* whether to use threads */
646 char *cmd_mpirun, /* Command for mpirun */
647 char *cmd_np, /* Switch for -np or -nt or empty */
648 char *cmd_mdrun, /* Command for mdrun */
649 char *args_for_mdrun, /* Arguments for mdrun */
650 const char *simulation_tpr, /* This tpr will be simulated */
651 int nnodes, /* Number of nodes to run on */
652 int nPMEnodes) /* Number of PME nodes to use */
657 /* Make enough space for the system call command,
658 * (100 extra chars for -npme ... etc. options should suffice): */
659 snew(command, strlen(cmd_mpirun)+strlen(cmd_mdrun)+strlen(cmd_np)+strlen(args_for_mdrun)+strlen(simulation_tpr)+100);
661 /* Note that the -passall options requires args_for_mdrun to be at the end
662 * of the command line string */
665 sprintf(command, "%s%s-npme %d -s %s %s",
666 cmd_mdrun, cmd_np, nPMEnodes, simulation_tpr, args_for_mdrun);
670 sprintf(command, "%s%s%s -npme %d -s %s %s",
671 cmd_mpirun, cmd_np, cmd_mdrun, nPMEnodes, simulation_tpr, args_for_mdrun);
674 fprintf(fp, "%s this command line to launch the simulation:\n\n%s", bLaunch? "Using":"Please use", command);
678 /* Now the real thing! */
681 fprintf(stdout, "\nLaunching simulation with best parameters now.\nExecuting '%s'", command);
684 gmx_system_call(command);
689 static void modify_PMEsettings(
690 gmx_large_int_t simsteps, /* Set this value as number of time steps */
691 const char *fn_best_tpr, /* tpr file with the best performance */
692 const char *fn_sim_tpr) /* name of tpr file to be launched */
700 read_tpx_state(fn_best_tpr,ir,&state,NULL,&mtop);
702 /* Set nsteps to the right value */
703 ir->nsteps = simsteps;
705 /* Write the tpr file which will be launched */
706 sprintf(buf, "Writing optimized simulation file %s with nsteps=%s.\n", fn_sim_tpr, gmx_large_int_pfmt);
707 fprintf(stdout,buf,ir->nsteps);
709 write_tpx_state(fn_sim_tpr,ir,&state,&mtop);
715 #define EPME_SWITCHED(e) ((e) == eelPMESWITCH || (e) == eelPMEUSERSWITCH)
717 /* Make additional TPR files with more computational load for the
718 * direct space processors: */
719 static void make_benchmark_tprs(
720 const char *fn_sim_tpr, /* READ : User-provided tpr file */
721 char *fn_bench_tprs[], /* WRITE: Names of benchmark tpr files */
722 gmx_large_int_t benchsteps, /* Number of time steps for benchmark runs */
723 gmx_large_int_t statesteps, /* Step counter in checkpoint file */
724 real rmin, /* Minimal Coulomb radius */
725 real rmax, /* Maximal Coulomb radius */
726 real bScaleRvdw, /* Scale rvdw along with rcoulomb */
727 int *ntprs, /* No. of TPRs to write, each with a different
728 rcoulomb and fourierspacing */
729 t_inputinfo *info, /* Contains information about mdp file options */
730 FILE *fp) /* Write the output here */
736 real nlist_buffer; /* Thickness of the buffer regions for PME-switch potentials */
739 gmx_bool bNote = FALSE;
740 real add; /* Add this to rcoul for the next test */
741 real fac = 1.0; /* Scaling factor for Coulomb radius */
742 real fourierspacing; /* Basic fourierspacing from tpr */
745 sprintf(buf, "Making benchmark tpr file%s with %s time step%s",
746 *ntprs > 1? "s":"", gmx_large_int_pfmt, benchsteps>1?"s":"");
747 fprintf(stdout, buf, benchsteps);
750 sprintf(buf, " (adding %s steps from checkpoint file)", gmx_large_int_pfmt);
751 fprintf(stdout, buf, statesteps);
752 benchsteps += statesteps;
754 fprintf(stdout, ".\n");
758 read_tpx_state(fn_sim_tpr,ir,&state,NULL,&mtop);
760 /* Check if some kind of PME was chosen */
761 if (EEL_PME(ir->coulombtype) == FALSE)
762 gmx_fatal(FARGS, "Can only do optimizations for simulations with %s electrostatics.",
765 /* Check if rcoulomb == rlist, which is necessary for plain PME. */
766 if ( (ir->cutoff_scheme != ecutsVERLET) &&
767 (eelPME == ir->coulombtype) && !(ir->rcoulomb == ir->rlist))
769 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to rlist (%f).",
770 EELTYPE(eelPME), ir->rcoulomb, ir->rlist);
772 /* For other PME types, rcoulomb is allowed to be smaller than rlist */
773 else if (ir->rcoulomb > ir->rlist)
775 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to or smaller than rlist (%f)",
776 EELTYPE(ir->coulombtype), ir->rcoulomb, ir->rlist);
779 if (bScaleRvdw && ir->rvdw != ir->rcoulomb)
781 fprintf(stdout,"NOTE: input rvdw != rcoulomb, will not scale rvdw\n");
785 /* Reduce the number of steps for the benchmarks */
786 info->orig_sim_steps = ir->nsteps;
787 ir->nsteps = benchsteps;
789 /* For PME-switch potentials, keep the radial distance of the buffer region */
790 nlist_buffer = ir->rlist - ir->rcoulomb;
792 /* Determine length of triclinic box vectors */
797 box_size[d] += state.box[d][i]*state.box[d][i];
798 box_size[d] = sqrt(box_size[d]);
801 if (ir->fourier_spacing > 0)
803 info->fsx[0] = ir->fourier_spacing;
804 info->fsy[0] = ir->fourier_spacing;
805 info->fsz[0] = ir->fourier_spacing;
809 /* Reconstruct fourierspacing per dimension from the number of grid points and box size */
810 info->fsx[0] = box_size[XX]/ir->nkx;
811 info->fsy[0] = box_size[YY]/ir->nky;
812 info->fsz[0] = box_size[ZZ]/ir->nkz;
815 /* If no value for the fourierspacing was provided on the command line, we
816 * use the reconstruction from the tpr file */
817 if (ir->fourier_spacing > 0)
819 /* Use the spacing from the tpr */
820 fourierspacing = ir->fourier_spacing;
824 /* Use the maximum observed spacing */
825 fourierspacing = max(max(info->fsx[0],info->fsy[0]),info->fsz[0]);
828 fprintf(stdout, "Calculating PME grid points on the basis of a fourierspacing of %f nm\n", fourierspacing);
830 /* For performance comparisons the number of particles is useful to have */
831 fprintf(fp, " Number of particles : %d\n", mtop.natoms);
833 /* Print information about settings of which some are potentially modified: */
834 fprintf(fp, " Coulomb type : %s\n", EELTYPE(ir->coulombtype));
835 fprintf(fp, " Grid spacing x y z : %f %f %f\n",
836 box_size[XX]/ir->nkx, box_size[YY]/ir->nky, box_size[ZZ]/ir->nkz);
837 fprintf(fp, " Van der Waals type : %s\n", EVDWTYPE(ir->vdwtype));
838 if (EVDW_SWITCHED(ir->vdwtype))
839 fprintf(fp, " rvdw_switch : %f nm\n", ir->rvdw_switch);
840 if (EPME_SWITCHED(ir->coulombtype))
841 fprintf(fp, " rlist : %f nm\n", ir->rlist);
842 if (ir->rlistlong != max_cutoff(ir->rvdw,ir->rcoulomb))
843 fprintf(fp, " rlistlong : %f nm\n", ir->rlistlong);
845 /* Print a descriptive line about the tpr settings tested */
846 fprintf(fp, "\nWill try these real/reciprocal workload settings:\n");
847 fprintf(fp, " No. scaling rcoulomb");
848 fprintf(fp, " nkx nky nkz");
849 fprintf(fp, " spacing");
850 if (evdwCUT == ir->vdwtype)
851 fprintf(fp, " rvdw");
852 if (EPME_SWITCHED(ir->coulombtype))
853 fprintf(fp, " rlist");
854 if ( ir->rlistlong != max_cutoff(ir->rlist,max_cutoff(ir->rvdw,ir->rcoulomb)) )
855 fprintf(fp, " rlistlong");
856 fprintf(fp, " tpr file\n");
858 /* Loop to create the requested number of tpr input files */
859 for (j = 0; j < *ntprs; j++)
861 /* The first .tpr is the provided one, just need to modify nsteps,
862 * so skip the following block */
865 /* Determine which Coulomb radii rc to use in the benchmarks */
866 add = (rmax-rmin)/(*ntprs-1);
867 if (is_equal(rmin,info->rcoulomb[0]))
869 ir->rcoulomb = rmin + j*add;
871 else if (is_equal(rmax,info->rcoulomb[0]))
873 ir->rcoulomb = rmin + (j-1)*add;
877 /* rmin != rcoul != rmax, ergo test between rmin and rmax */
878 add = (rmax-rmin)/(*ntprs-2);
879 ir->rcoulomb = rmin + (j-1)*add;
882 /* Determine the scaling factor fac */
883 fac = ir->rcoulomb/info->rcoulomb[0];
885 /* Scale the Fourier grid spacing */
886 ir->nkx = ir->nky = ir->nkz = 0;
887 calc_grid(NULL,state.box,fourierspacing*fac,&ir->nkx,&ir->nky,&ir->nkz);
889 /* Adjust other radii since various conditions neet to be fulfilled */
890 if (eelPME == ir->coulombtype)
892 /* plain PME, rcoulomb must be equal to rlist */
893 ir->rlist = ir->rcoulomb;
897 /* rlist must be >= rcoulomb, we keep the size of the buffer region */
898 ir->rlist = ir->rcoulomb + nlist_buffer;
901 if (bScaleRvdw && evdwCUT == ir->vdwtype)
903 /* For vdw cutoff, rvdw >= rlist */
904 ir->rvdw = max(info->rvdw[0], ir->rlist);
907 ir->rlistlong = max_cutoff(ir->rlist,max_cutoff(ir->rvdw,ir->rcoulomb));
909 } /* end of "if (j != 0)" */
911 /* for j==0: Save the original settings
912 * for j >0: Save modified radii and Fourier grids */
913 info->rcoulomb[j] = ir->rcoulomb;
914 info->rvdw[j] = ir->rvdw;
915 info->nkx[j] = ir->nkx;
916 info->nky[j] = ir->nky;
917 info->nkz[j] = ir->nkz;
918 info->rlist[j] = ir->rlist;
919 info->rlistlong[j] = ir->rlistlong;
920 info->fsx[j] = fac*fourierspacing;
921 info->fsy[j] = fac*fourierspacing;
922 info->fsz[j] = fac*fourierspacing;
924 /* Write the benchmark tpr file */
925 strncpy(fn_bench_tprs[j],fn_sim_tpr,strlen(fn_sim_tpr)-strlen(".tpr"));
926 sprintf(buf, "_bench%.2d.tpr", j);
927 strcat(fn_bench_tprs[j], buf);
928 fprintf(stdout,"Writing benchmark tpr %s with nsteps=", fn_bench_tprs[j]);
929 fprintf(stdout, gmx_large_int_pfmt, ir->nsteps);
931 fprintf(stdout,", scaling factor %f\n", fac);
933 fprintf(stdout,", unmodified settings\n");
935 write_tpx_state(fn_bench_tprs[j],ir,&state,&mtop);
937 /* Write information about modified tpr settings to log file */
938 fprintf(fp, "%4d%10f%10f", j, fac, ir->rcoulomb);
939 fprintf(fp, "%5d%5d%5d", ir->nkx, ir->nky, ir->nkz);
940 fprintf(fp, " %9f ", info->fsx[j]);
941 if (evdwCUT == ir->vdwtype)
942 fprintf(fp, "%10f", ir->rvdw);
943 if (EPME_SWITCHED(ir->coulombtype))
944 fprintf(fp, "%10f", ir->rlist);
945 if ( info->rlistlong[0] != max_cutoff(info->rlist[0],max_cutoff(info->rvdw[0],info->rcoulomb[0])) )
946 fprintf(fp, "%10f", ir->rlistlong);
947 fprintf(fp, " %-14s\n",fn_bench_tprs[j]);
949 /* Make it clear to the user that some additional settings were modified */
950 if ( !is_equal(ir->rvdw , info->rvdw[0])
951 || !is_equal(ir->rlistlong, info->rlistlong[0]) )
957 fprintf(fp, "\nNote that in addition to the Coulomb radius and the Fourier grid\n"
958 "other input settings were also changed (see table above).\n"
959 "Please check if the modified settings are appropriate.\n");
966 /* Rename the files we want to keep to some meaningful filename and
968 static void cleanup(const t_filenm *fnm, int nfile, int k, int nnodes,
969 int nPMEnodes, int nr, gmx_bool bKeepStderr)
971 char numstring[STRLEN];
972 char newfilename[STRLEN];
978 fprintf(stdout, "Cleaning up, deleting benchmark temp files ...\n");
980 for (i=0; i<nfile; i++)
982 opt = (char *)fnm[i].opt;
983 if ( strcmp(opt, "-p") == 0 )
985 /* do nothing; keep this file */
988 else if (strcmp(opt, "-bg") == 0)
990 /* Give the log file a nice name so one can later see which parameters were used */
993 sprintf(numstring, "_%d", nr);
994 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", opt2fn("-bg",nfile,fnm), k, nnodes, nPMEnodes, numstring);
995 if (gmx_fexist(opt2fn("-bg",nfile,fnm)))
997 fprintf(stdout, "renaming log file to %s\n", newfilename);
998 make_backup(newfilename);
999 rename(opt2fn("-bg",nfile,fnm), newfilename);
1002 else if (strcmp(opt, "-err") == 0)
1004 /* This file contains the output of stderr. We want to keep it in
1005 * cases where there have been problems. */
1006 fn = opt2fn(opt, nfile, fnm);
1007 numstring[0] = '\0';
1009 sprintf(numstring, "_%d", nr);
1010 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", fn, k, nnodes, nPMEnodes, numstring);
1015 fprintf(stdout, "Saving stderr output in %s\n", newfilename);
1016 make_backup(newfilename);
1017 rename(fn, newfilename);
1021 fprintf(stdout, "Deleting %s\n", fn);
1026 /* Delete the files which are created for each benchmark run: (options -b*) */
1027 else if ( (0 == strncmp(opt, "-b", 2)) && (opt2bSet(opt,nfile,fnm) || !is_optional(&fnm[i])) )
1029 fn = opt2fn(opt, nfile, fnm);
1032 fprintf(stdout, "Deleting %s\n", fn);
1040 /* Returns the largest common factor of n1 and n2 */
1041 static int largest_common_factor(int n1, int n2)
1046 for (factor=nmax; factor > 0; factor--)
1048 if ( 0==(n1 % factor) && 0==(n2 % factor) )
1053 return 0; /* one for the compiler */
1056 enum {eNpmeAuto, eNpmeAll, eNpmeReduced, eNpmeSubset, eNpmeNr};
1058 /* Create a list of numbers of PME nodes to test */
1059 static void make_npme_list(
1060 const char *npmevalues_opt, /* Make a complete list with all
1061 * possibilities or a short list that keeps only
1062 * reasonable numbers of PME nodes */
1063 int *nentries, /* Number of entries we put in the nPMEnodes list */
1064 int *nPMEnodes[], /* Each entry contains the value for -npme */
1065 int nnodes, /* Total number of nodes to do the tests on */
1066 int minPMEnodes, /* Minimum number of PME nodes */
1067 int maxPMEnodes) /* Maximum number of PME nodes */
1070 int min_factor=1; /* We request that npp and npme have this minimal
1071 * largest common factor (depends on npp) */
1072 int nlistmax; /* Max. list size */
1073 int nlist; /* Actual number of entries in list */
1077 /* Do we need to check all possible values for -npme or is a reduced list enough? */
1078 if ( 0 == strcmp(npmevalues_opt, "all") )
1082 else if ( 0 == strcmp(npmevalues_opt, "subset") )
1084 eNPME = eNpmeSubset;
1086 else /* "auto" or "range" */
1090 else if (nnodes < 128)
1091 eNPME = eNpmeReduced;
1093 eNPME = eNpmeSubset;
1096 /* Calculate how many entries we could possibly have (in case of -npme all) */
1099 nlistmax = maxPMEnodes - minPMEnodes + 3;
1100 if (0 == minPMEnodes)
1106 /* Now make the actual list which is at most of size nlist */
1107 snew(*nPMEnodes, nlistmax);
1108 nlist = 0; /* start counting again, now the real entries in the list */
1109 for (i = 0; i < nlistmax - 2; i++)
1111 npme = maxPMEnodes - i;
1122 /* For 2d PME we want a common largest factor of at least the cube
1123 * root of the number of PP nodes */
1124 min_factor = (int) pow(npp, 1.0/3.0);
1127 gmx_fatal(FARGS, "Unknown option for eNPME in make_npme_list");
1130 if (largest_common_factor(npp, npme) >= min_factor)
1132 (*nPMEnodes)[nlist] = npme;
1136 /* We always test 0 PME nodes and the automatic number */
1137 *nentries = nlist + 2;
1138 (*nPMEnodes)[nlist ] = 0;
1139 (*nPMEnodes)[nlist+1] = -1;
1141 fprintf(stderr, "Will try the following %d different values for -npme:\n", *nentries);
1142 for (i=0; i<*nentries-1; i++)
1143 fprintf(stderr, "%d, ", (*nPMEnodes)[i]);
1144 fprintf(stderr, "and %d (auto).\n", (*nPMEnodes)[*nentries-1]);
1148 /* Allocate memory to store the performance data */
1149 static void init_perfdata(t_perf *perfdata[], int ntprs, int datasets, int repeats)
1154 for (k=0; k<ntprs; k++)
1156 snew(perfdata[k], datasets);
1157 for (i=0; i<datasets; i++)
1159 for (j=0; j<repeats; j++)
1161 snew(perfdata[k][i].Gcycles , repeats);
1162 snew(perfdata[k][i].ns_per_day, repeats);
1163 snew(perfdata[k][i].PME_f_load, repeats);
1170 /* Check for errors on mdrun -h */
1171 static void make_sure_it_runs(char *mdrun_cmd_line, int length, FILE *fp)
1173 char *command, *msg;
1177 snew(command, length + 15);
1178 snew(msg , length + 500);
1180 fprintf(stdout, "Making shure the benchmarks can be executed ...\n");
1181 sprintf(command, "%s-h -quiet", mdrun_cmd_line);
1182 ret = gmx_system_call(command);
1186 /* To prevent confusion, do not again issue a gmx_fatal here since we already
1187 * get the error message from mdrun itself */
1188 sprintf(msg, "Cannot run the benchmark simulations! Please check the error message of\n"
1189 "mdrun for the source of the problem. Did you provide a command line\n"
1190 "argument that neither g_tune_pme nor mdrun understands? Offending command:\n"
1191 "\n%s\n\n", command);
1193 fprintf(stderr, "%s", msg);
1195 fprintf(fp , "%s", msg);
1205 static void do_the_tests(
1206 FILE *fp, /* General g_tune_pme output file */
1207 char **tpr_names, /* Filenames of the input files to test */
1208 int maxPMEnodes, /* Max fraction of nodes to use for PME */
1209 int minPMEnodes, /* Min fraction of nodes to use for PME */
1210 int npme_fixed, /* If >= -1, test fixed number of PME
1212 const char *npmevalues_opt, /* Which -npme values should be tested */
1213 t_perf **perfdata, /* Here the performace data is stored */
1214 int *pmeentries, /* Entries in the nPMEnodes list */
1215 int repeats, /* Repeat each test this often */
1216 int nnodes, /* Total number of nodes = nPP + nPME */
1217 int nr_tprs, /* Total number of tpr files to test */
1218 gmx_bool bThreads, /* Threads or MPI? */
1219 char *cmd_mpirun, /* mpirun command string */
1220 char *cmd_np, /* "-np", "-n", whatever mpirun needs */
1221 char *cmd_mdrun, /* mdrun command string */
1222 char *cmd_args_bench, /* arguments for mdrun in a string */
1223 const t_filenm *fnm, /* List of filenames from command line */
1224 int nfile, /* Number of files specified on the cmdl. */
1225 int sim_part, /* For checkpointing */
1226 int presteps, /* DLB equilibration steps, is checked */
1227 gmx_large_int_t cpt_steps) /* Time step counter in the checkpoint */
1229 int i,nr,k,ret,count=0,totaltests;
1230 int *nPMEnodes=NULL;
1233 char *command, *cmd_stub;
1235 gmx_bool bResetProblem=FALSE;
1236 gmx_bool bFirst=TRUE;
1239 /* This string array corresponds to the eParselog enum type at the start
1241 const char* ParseLog[] = {"OK.",
1242 "Logfile not found!",
1243 "No timings, logfile truncated?",
1244 "Run was terminated.",
1245 "Counters were not reset properly.",
1246 "No DD grid found for these settings.",
1247 "TPX version conflict!",
1248 "mdrun was not started in parallel!",
1249 "An error occured." };
1250 char str_PME_f_load[13];
1253 /* Allocate space for the mdrun command line. 100 extra characters should
1254 be more than enough for the -npme etcetera arguments */
1255 cmdline_length = strlen(cmd_mpirun)
1258 + strlen(cmd_args_bench)
1259 + strlen(tpr_names[0]) + 100;
1260 snew(command , cmdline_length);
1261 snew(cmd_stub, cmdline_length);
1263 /* Construct the part of the command line that stays the same for all tests: */
1266 sprintf(cmd_stub, "%s%s", cmd_mdrun, cmd_np);
1270 sprintf(cmd_stub, "%s%s%s ", cmd_mpirun, cmd_np, cmd_mdrun);
1273 /* Create a list of numbers of PME nodes to test */
1274 if (npme_fixed < -1)
1276 make_npme_list(npmevalues_opt, pmeentries, &nPMEnodes,
1277 nnodes, minPMEnodes, maxPMEnodes);
1283 nPMEnodes[0] = npme_fixed;
1284 fprintf(stderr, "Will use a fixed number of %d PME-only nodes.\n", nPMEnodes[0]);
1289 fprintf(fp, "\nNo benchmarks done since number of repeats (-r) is 0.\n");
1291 finalize(opt2fn("-p", nfile, fnm));
1295 /* Allocate one dataset for each tpr input file: */
1296 init_perfdata(perfdata, nr_tprs, *pmeentries, repeats);
1298 /*****************************************/
1299 /* Main loop over all tpr files to test: */
1300 /*****************************************/
1301 totaltests = nr_tprs*(*pmeentries)*repeats;
1302 for (k=0; k<nr_tprs;k++)
1304 fprintf(fp, "\nIndividual timings for input file %d (%s):\n", k, tpr_names[k]);
1305 fprintf(fp, "PME nodes Gcycles ns/day PME/f Remark\n");
1306 /* Loop over various numbers of PME nodes: */
1307 for (i = 0; i < *pmeentries; i++)
1309 pd = &perfdata[k][i];
1311 /* Loop over the repeats for each scenario: */
1312 for (nr = 0; nr < repeats; nr++)
1314 pd->nPMEnodes = nPMEnodes[i];
1316 /* Add -npme and -s to the command line and save it. Note that
1317 * the -passall (if set) options requires cmd_args_bench to be
1318 * at the end of the command line string */
1319 snew(pd->mdrun_cmd_line, cmdline_length);
1320 sprintf(pd->mdrun_cmd_line, "%s-npme %d -s %s %s",
1321 cmd_stub, pd->nPMEnodes, tpr_names[k], cmd_args_bench);
1323 /* To prevent that all benchmarks fail due to a show-stopper argument
1324 * on the mdrun command line, we make a quick check with mdrun -h first */
1326 make_sure_it_runs(pd->mdrun_cmd_line, cmdline_length, fp);
1329 /* Do a benchmark simulation: */
1331 sprintf(buf, ", pass %d/%d", nr+1, repeats);
1334 fprintf(stdout, "\n=== Progress %2.0f%%, tpr %d/%d, run %d/%d%s:\n",
1335 (100.0*count)/totaltests,
1336 k+1, nr_tprs, i+1, *pmeentries, buf);
1337 make_backup(opt2fn("-err",nfile,fnm));
1338 sprintf(command, "%s 1> /dev/null 2>%s", pd->mdrun_cmd_line, opt2fn("-err",nfile,fnm));
1339 fprintf(stdout, "%s\n", pd->mdrun_cmd_line);
1340 gmx_system_call(command);
1342 /* Collect the performance data from the log file; also check stderr
1343 * for fatal errors */
1344 ret = parse_logfile(opt2fn("-bg",nfile,fnm), opt2fn("-err",nfile,fnm),
1345 pd, nr, presteps, cpt_steps, nnodes);
1346 if ((presteps > 0) && (ret == eParselogResetProblem))
1347 bResetProblem = TRUE;
1349 if (-1 == pd->nPMEnodes)
1350 sprintf(buf, "(%3d)", pd->guessPME);
1355 if (pd->PME_f_load[nr] > 0.0)
1356 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load[nr]);
1358 sprintf(str_PME_f_load, "%s", " - ");
1360 /* Write the data we got to disk */
1361 fprintf(fp, "%4d%s %12.3f %12.3f %s %s", pd->nPMEnodes,
1362 buf, pd->Gcycles[nr], pd->ns_per_day[nr], str_PME_f_load, ParseLog[ret]);
1363 if (! (ret==eParselogOK || ret==eParselogNoDDGrid || ret==eParselogNotFound) )
1364 fprintf(fp, " Check %s file for problems.", ret==eParselogFatal? "err":"log");
1369 /* Do some cleaning up and delete the files we do not need any more */
1370 cleanup(fnm, nfile, k, nnodes, pd->nPMEnodes, nr, ret==eParselogFatal);
1372 /* If the first run with this number of processors already failed, do not try again: */
1373 if (pd->Gcycles[0] <= 0.0 && repeats > 1)
1375 fprintf(stdout, "Skipping remaining passes of unsuccessful setting, see log file for details.\n");
1376 count += repeats-(nr+1);
1379 } /* end of repeats loop */
1380 } /* end of -npme loop */
1381 } /* end of tpr file loop */
1386 fprintf(fp, "WARNING: The cycle and time step counters could not be reset\n"
1387 "properly. The reason could be that mpirun did not manage to\n"
1388 "export the environment variable GMX_RESET_COUNTER. You might\n"
1389 "have to give a special switch to mpirun for that.\n"
1390 "Alternatively, you can manually set GMX_RESET_COUNTER to the\n"
1391 "value normally provided by -presteps.");
1399 static void check_input(
1406 real maxPMEfraction,
1407 real minPMEfraction,
1409 gmx_large_int_t bench_nsteps,
1410 const t_filenm *fnm,
1420 /* Make sure the input file exists */
1421 if (!gmx_fexist(opt2fn("-s",nfile,fnm)))
1422 gmx_fatal(FARGS, "File %s not found.", opt2fn("-s",nfile,fnm));
1424 /* Make sure that the checkpoint file is not overwritten during benchmarking */
1425 if ( (0 == strcmp(opt2fn("-cpi",nfile,fnm), opt2fn("-bcpo",nfile,fnm)) ) && (sim_part > 1) )
1426 gmx_fatal(FARGS, "Checkpoint input (-cpi) and benchmark checkpoint output (-bcpo) files must not be identical.\n"
1427 "The checkpoint input file must not be overwritten during the benchmarks.\n");
1429 /* Make sure that repeats is >= 0 (if == 0, only write tpr files) */
1431 gmx_fatal(FARGS, "Number of repeats < 0!");
1433 /* Check number of nodes */
1435 gmx_fatal(FARGS, "Number of nodes/threads must be a positive integer.");
1437 /* Automatically choose -ntpr if not set */
1445 /* Set a reasonable scaling factor for rcoulomb */
1447 *rmax = rcoulomb * 1.2;
1449 fprintf(stderr, "Will test %d tpr file%s.\n", *ntprs, *ntprs==1?"":"s");
1454 fprintf(stderr, "Note: Choose ntpr>1 to shift PME load between real and reciprocal space.\n");
1457 /* Make shure that rmin <= rcoulomb <= rmax */
1458 if (*rmin <= 0) *rmin = rcoulomb;
1459 if (*rmax <= 0) *rmax = rcoulomb;
1460 if ( !(*rmin <= *rmax) )
1462 gmx_fatal(FARGS, "Please choose the Coulomb radii such that rmin <= rmax.\n"
1463 "rmin = %g, rmax = %g, actual rcoul from .tpr file = %g\n", *rmin, *rmax, rcoulomb);
1465 /* Add test scenarios if rmin or rmax were set */
1468 if ( !is_equal(*rmin, rcoulomb) && (*ntprs == 1) )
1471 fprintf(stderr, "NOTE: Setting -rmin to %g changed -ntpr to %d\n",
1474 if (!is_equal(*rmax, rcoulomb) && (*ntprs == 1) )
1477 fprintf(stderr, "NOTE: Setting -rmax to %g changed -ntpr to %d\n",
1482 /* If one of rmin, rmax is set, we need 2 tpr files at minimum */
1483 if ( !is_equal(*rmax, rcoulomb) || !is_equal(*rmin, rcoulomb) )
1484 *ntprs = max(*ntprs, 2);
1486 /* If both rmin, rmax are set, we need 3 tpr files at minimum */
1487 if ( !is_equal(*rmax, rcoulomb) && !is_equal(*rmin, rcoulomb) )
1488 *ntprs = max(*ntprs, 3);
1492 fprintf(stderr, "NOTE: Your rmin, rmax setting changed -ntpr to %d\n", *ntprs);
1497 if (is_equal(*rmin,rcoulomb) && is_equal(rcoulomb,*rmax)) /* We have just a single rc */
1499 fprintf(stderr, "WARNING: Resetting -ntpr to 1 since no Coulomb radius scaling is requested.\n"
1500 "Please set rmin < rmax to test Coulomb radii in the [rmin, rmax] interval\n"
1501 "with correspondingly adjusted PME grid settings\n");
1506 /* Check whether max and min fraction are within required values */
1507 if (maxPMEfraction > 0.5 || maxPMEfraction < 0)
1508 gmx_fatal(FARGS, "-max must be between 0 and 0.5");
1509 if (minPMEfraction > 0.5 || minPMEfraction < 0)
1510 gmx_fatal(FARGS, "-min must be between 0 and 0.5");
1511 if (maxPMEfraction < minPMEfraction)
1512 gmx_fatal(FARGS, "-max must be larger or equal to -min");
1514 /* Check whether the number of steps - if it was set - has a reasonable value */
1515 if (bench_nsteps < 0)
1516 gmx_fatal(FARGS, "Number of steps must be positive.");
1518 if (bench_nsteps > 10000 || bench_nsteps < 100)
1520 fprintf(stderr, "WARNING: steps=");
1521 fprintf(stderr, gmx_large_int_pfmt, bench_nsteps);
1522 fprintf(stderr, ". Are you sure you want to perform so %s steps for each benchmark?\n", (bench_nsteps < 100)? "few" : "many");
1527 gmx_fatal(FARGS, "Cannot have a negative number of presteps.\n");
1530 /* Check for rcoulomb scaling if more than one .tpr file is tested */
1533 if (*rmin/rcoulomb < 0.75 || *rmax/rcoulomb > 1.25)
1534 fprintf(stderr, "WARNING: Applying extreme scaling factor. I hope you know what you are doing.\n");
1537 /* If a fixed number of PME nodes is set we do rcoulomb and PME gird tuning
1538 * only. We need to check whether the requested number of PME-only nodes
1540 if (npme_fixed > -1)
1542 /* No more than 50% of all nodes can be assigned as PME-only nodes. */
1543 if (2*npme_fixed > nnodes)
1545 gmx_fatal(FARGS, "Cannot have more than %d PME-only nodes for a total of %d nodes (you chose %d).\n",
1546 nnodes/2, nnodes, npme_fixed);
1548 if ((npme_fixed > 0) && (5*npme_fixed < nnodes))
1550 fprintf(stderr, "WARNING: Only %g percent of the nodes are assigned as PME-only nodes.\n",
1551 100.0*((real)npme_fixed / (real)nnodes));
1553 if (opt2parg_bSet("-min",npargs,pa) || opt2parg_bSet("-max",npargs,pa))
1555 fprintf(stderr, "NOTE: The -min, -max, and -npme options have no effect when a\n"
1556 " fixed number of PME-only nodes is requested with -fix.\n");
1562 /* Returns TRUE when "opt" is needed at launch time */
1563 static gmx_bool is_launch_file(char *opt, gmx_bool bSet)
1565 /* Apart from the input .tpr we need all options that were set
1566 * on the command line and that do not start with -b */
1567 if (0 == strncmp(opt,"-b", 2) || 0 == strncmp(opt,"-s", 2))
1577 /* Returns TRUE when "opt" defines a file which is needed for the benchmarks runs */
1578 static gmx_bool is_bench_file(char *opt, gmx_bool bSet, gmx_bool bOptional, gmx_bool bIsOutput)
1580 /* Apart from the input .tpr, all files starting with "-b" are for
1581 * _b_enchmark files exclusively */
1582 if (0 == strncmp(opt,"-s", 2)) return FALSE;
1583 if (0 == strncmp(opt,"-b", 2) || 0 == strncmp(opt,"-s", 2))
1585 if (!bOptional || bSet)
1595 if (bSet) /* These are additional input files like -cpi -ei */
1603 /* Adds 'buf' to 'str' */
1604 static void add_to_string(char **str, char *buf)
1609 len = strlen(*str) + strlen(buf) + 1;
1615 /* Create the command line for the benchmark as well as for the real run */
1616 static void create_command_line_snippets(
1618 gmx_bool bAppendFiles,
1619 gmx_bool bKeepAndNumCPT,
1620 gmx_bool bResetHWay,
1626 const char *procstring, /* How to pass the number of processors to $MPIRUN */
1627 char *cmd_np[], /* Actual command line snippet, e.g. '-np <N>' */
1628 char *cmd_args_bench[], /* command line arguments for benchmark runs */
1629 char *cmd_args_launch[], /* command line arguments for simulation run */
1630 char extra_args[]) /* Add this to the end of the command line */
1635 char strbuf[STRLEN];
1638 /* strlen needs at least '\0' as a string: */
1639 snew(*cmd_args_bench ,1);
1640 snew(*cmd_args_launch,1);
1641 *cmd_args_launch[0]='\0';
1642 *cmd_args_bench[0] ='\0';
1645 /*******************************************/
1646 /* 1. Process other command line arguments */
1647 /*******************************************/
1650 /* Add equilibration steps to benchmark options */
1651 sprintf(strbuf, "-resetstep %d ", presteps);
1652 add_to_string(cmd_args_bench, strbuf);
1654 /* These switches take effect only at launch time */
1655 if (FALSE == bAppendFiles)
1657 add_to_string(cmd_args_launch, "-noappend ");
1661 add_to_string(cmd_args_launch, "-cpnum ");
1665 add_to_string(cmd_args_launch, "-resethway ");
1668 /********************/
1669 /* 2. Process files */
1670 /********************/
1671 for (i=0; i<nfile; i++)
1673 opt = (char *)fnm[i].opt;
1674 name = opt2fn(opt,nfile,fnm);
1676 /* Strbuf contains the options, now let's sort out where we need that */
1677 sprintf(strbuf, "%s %s ", opt, name);
1679 if ( is_bench_file(opt, opt2bSet(opt,nfile,fnm), is_optional(&fnm[i]), is_output(&fnm[i])) )
1681 /* All options starting with -b* need the 'b' removed,
1682 * therefore overwrite strbuf */
1683 if (0 == strncmp(opt, "-b", 2))
1684 sprintf(strbuf, "-%s %s ", &opt[2], name);
1686 add_to_string(cmd_args_bench, strbuf);
1689 if ( is_launch_file(opt,opt2bSet(opt,nfile,fnm)) )
1690 add_to_string(cmd_args_launch, strbuf);
1693 add_to_string(cmd_args_bench , extra_args);
1694 add_to_string(cmd_args_launch, extra_args);
1698 /* Set option opt */
1699 static void setopt(const char *opt,int nfile,t_filenm fnm[])
1703 for(i=0; (i<nfile); i++)
1704 if (strcmp(opt,fnm[i].opt)==0)
1705 fnm[i].flag |= ffSET;
1709 /* This routine inspects the tpr file and ...
1710 * 1. checks for output files that get triggered by a tpr option. These output
1711 * files are marked as 'set' to allow for a proper cleanup after each
1713 * 2. returns the PME:PP load ratio
1714 * 3. returns rcoulomb from the tpr */
1715 static float inspect_tpr(int nfile, t_filenm fnm[], real *rcoulomb)
1717 gmx_bool bPull; /* Is pulling requested in .tpr file? */
1718 gmx_bool bTpi; /* Is test particle insertion requested? */
1719 gmx_bool bFree; /* Is a free energy simulation requested? */
1720 gmx_bool bNM; /* Is a normal mode analysis requested? */
1726 /* Check tpr file for options that trigger extra output files */
1727 read_tpx_state(opt2fn("-s",nfile,fnm),&ir,&state,NULL,&mtop);
1728 bPull = (epullNO != ir.ePull);
1729 bFree = (efepNO != ir.efep );
1730 bNM = (eiNM == ir.eI );
1731 bTpi = EI_TPI(ir.eI);
1733 /* Set these output files on the tuning command-line */
1736 setopt("-pf" , nfile, fnm);
1737 setopt("-px" , nfile, fnm);
1741 setopt("-dhdl", nfile, fnm);
1745 setopt("-tpi" , nfile, fnm);
1746 setopt("-tpid", nfile, fnm);
1750 setopt("-mtx" , nfile, fnm);
1753 *rcoulomb = ir.rcoulomb;
1755 /* Return the estimate for the number of PME nodes */
1756 return pme_load_estimate(&mtop,&ir,state.box);
1760 static void couple_files_options(int nfile, t_filenm fnm[])
1763 gmx_bool bSet,bBench;
1768 for (i=0; i<nfile; i++)
1770 opt = (char *)fnm[i].opt;
1771 bSet = ((fnm[i].flag & ffSET) != 0);
1772 bBench = (0 == strncmp(opt,"-b", 2));
1774 /* Check optional files */
1775 /* If e.g. -eo is set, then -beo also needs to be set */
1776 if (is_optional(&fnm[i]) && bSet && !bBench)
1778 sprintf(buf, "-b%s", &opt[1]);
1779 setopt(buf,nfile,fnm);
1781 /* If -beo is set, then -eo also needs to be! */
1782 if (is_optional(&fnm[i]) && bSet && bBench)
1784 sprintf(buf, "-%s", &opt[2]);
1785 setopt(buf,nfile,fnm);
1791 static double gettime()
1793 #ifdef HAVE_GETTIMEOFDAY
1797 gettimeofday(&t,NULL);
1799 seconds = (double) t.tv_sec + 1e-6*(double)t.tv_usec;
1805 seconds = time(NULL);
1812 #define BENCHSTEPS (1000)
1814 int gmx_tune_pme(int argc,char *argv[])
1816 const char *desc[] = {
1817 "For a given number [TT]-np[tt] or [TT]-nt[tt] of processors/threads, this program systematically",
1818 "times [TT]mdrun[tt] with various numbers of PME-only nodes and determines",
1819 "which setting is fastest. It will also test whether performance can",
1820 "be enhanced by shifting load from the reciprocal to the real space",
1821 "part of the Ewald sum. ",
1822 "Simply pass your [TT].tpr[tt] file to [TT]g_tune_pme[tt] together with other options",
1823 "for [TT]mdrun[tt] as needed.[PAR]",
1824 "Which executables are used can be set in the environment variables",
1825 "MPIRUN and MDRUN. If these are not present, 'mpirun' and 'mdrun'",
1826 "will be used as defaults. Note that for certain MPI frameworks you",
1827 "need to provide a machine- or hostfile. This can also be passed",
1828 "via the MPIRUN variable, e.g.[PAR]",
1829 "[TT]export MPIRUN=\"/usr/local/mpirun -machinefile hosts\"[tt][PAR]",
1830 "Please call [TT]g_tune_pme[tt] with the normal options you would pass to",
1831 "[TT]mdrun[tt] and add [TT]-np[tt] for the number of processors to perform the",
1832 "tests on, or [TT]-nt[tt] for the number of threads. You can also add [TT]-r[tt]",
1833 "to repeat each test several times to get better statistics. [PAR]",
1834 "[TT]g_tune_pme[tt] can test various real space / reciprocal space workloads",
1835 "for you. With [TT]-ntpr[tt] you control how many extra [TT].tpr[tt] files will be",
1836 "written with enlarged cutoffs and smaller Fourier grids respectively.",
1837 "Typically, the first test (number 0) will be with the settings from the input",
1838 "[TT].tpr[tt] file; the last test (number [TT]ntpr[tt]) will have the Coulomb cutoff",
1839 "specified by [TT]-rmax[tt] with a somwhat smaller PME grid at the same time. ",
1840 "In this last test, the Fourier spacing is multiplied with [TT]rmax[tt]/rcoulomb. ",
1841 "The remaining [TT].tpr[tt] files will have equally-spaced Coulomb radii (and Fourier "
1842 "spacings) between these extremes. [BB]Note[bb] that you can set [TT]-ntpr[tt] to 1",
1843 "if you just seek the optimal number of PME-only nodes; in that case",
1844 "your input [TT].tpr[tt] file will remain unchanged.[PAR]",
1845 "For the benchmark runs, the default of 1000 time steps should suffice for most",
1846 "MD systems. The dynamic load balancing needs about 100 time steps",
1847 "to adapt to local load imbalances, therefore the time step counters",
1848 "are by default reset after 100 steps. For large systems (>1M atoms), as well as ",
1849 "for a higher accuarcy of the measurements, you should set [TT]-resetstep[tt] to a higher value.",
1850 "From the 'DD' load imbalance entries in the md.log output file you",
1851 "can tell after how many steps the load is sufficiently balanced. Example call:[PAR]"
1852 "[TT]g_tune_pme -np 64 -s protein.tpr -launch[tt][PAR]",
1853 "After calling [TT]mdrun[tt] several times, detailed performance information",
1854 "is available in the output file [TT]perf.out.[tt] ",
1855 "[BB]Note[bb] that during the benchmarks, a couple of temporary files are written",
1856 "(options [TT]-b[tt]*), these will be automatically deleted after each test.[PAR]",
1857 "If you want the simulation to be started automatically with the",
1858 "optimized parameters, use the command line option [TT]-launch[tt].[PAR]",
1863 int pmeentries=0; /* How many values for -npme do we actually test for each tpr file */
1864 real maxPMEfraction=0.50;
1865 real minPMEfraction=0.25;
1866 int maxPMEnodes, minPMEnodes;
1867 float guessPMEratio; /* guessed PME:PP ratio based on the tpr file */
1868 float guessPMEnodes;
1869 int npme_fixed=-2; /* If >= -1, use only this number
1870 * of PME-only nodes */
1872 real rmin=0.0,rmax=0.0; /* min and max value for rcoulomb if scaling is requested */
1873 real rcoulomb=-1.0; /* Coulomb radius as set in .tpr file */
1874 gmx_bool bScaleRvdw=TRUE;
1875 gmx_large_int_t bench_nsteps=BENCHSTEPS;
1876 gmx_large_int_t new_sim_nsteps=-1; /* -1 indicates: not set by the user */
1877 gmx_large_int_t cpt_steps=0; /* Step counter in .cpt input file */
1878 int presteps=100; /* Do a full cycle reset after presteps steps */
1879 gmx_bool bOverwrite=FALSE, bKeepTPR;
1880 gmx_bool bLaunch=FALSE;
1881 char *ExtraArgs=NULL;
1882 char **tpr_names=NULL;
1883 const char *simulation_tpr=NULL;
1884 int best_npme, best_tpr;
1885 int sim_part = 1; /* For benchmarks with checkpoint files */
1888 /* Default program names if nothing else is found */
1889 char *cmd_mpirun=NULL, *cmd_mdrun=NULL;
1890 char *cmd_args_bench, *cmd_args_launch;
1893 t_perf **perfdata=NULL;
1899 /* Print out how long the tuning took */
1902 static t_filenm fnm[] = {
1904 { efOUT, "-p", "perf", ffWRITE },
1905 { efLOG, "-err", "bencherr", ffWRITE },
1906 { efTPX, "-so", "tuned", ffWRITE },
1908 { efTPX, NULL, NULL, ffREAD },
1909 { efTRN, "-o", NULL, ffWRITE },
1910 { efXTC, "-x", NULL, ffOPTWR },
1911 { efCPT, "-cpi", NULL, ffOPTRD },
1912 { efCPT, "-cpo", NULL, ffOPTWR },
1913 { efSTO, "-c", "confout", ffWRITE },
1914 { efEDR, "-e", "ener", ffWRITE },
1915 { efLOG, "-g", "md", ffWRITE },
1916 { efXVG, "-dhdl", "dhdl", ffOPTWR },
1917 { efXVG, "-field", "field", ffOPTWR },
1918 { efXVG, "-table", "table", ffOPTRD },
1919 { efXVG, "-tabletf", "tabletf", ffOPTRD },
1920 { efXVG, "-tablep", "tablep", ffOPTRD },
1921 { efXVG, "-tableb", "table", ffOPTRD },
1922 { efTRX, "-rerun", "rerun", ffOPTRD },
1923 { efXVG, "-tpi", "tpi", ffOPTWR },
1924 { efXVG, "-tpid", "tpidist", ffOPTWR },
1925 { efEDI, "-ei", "sam", ffOPTRD },
1926 { efEDO, "-eo", "sam", ffOPTWR },
1927 { efGCT, "-j", "wham", ffOPTRD },
1928 { efGCT, "-jo", "bam", ffOPTWR },
1929 { efXVG, "-ffout", "gct", ffOPTWR },
1930 { efXVG, "-devout", "deviatie", ffOPTWR },
1931 { efXVG, "-runav", "runaver", ffOPTWR },
1932 { efXVG, "-px", "pullx", ffOPTWR },
1933 { efXVG, "-pf", "pullf", ffOPTWR },
1934 { efXVG, "-ro", "rotation", ffOPTWR },
1935 { efLOG, "-ra", "rotangles",ffOPTWR },
1936 { efLOG, "-rs", "rotslabs", ffOPTWR },
1937 { efLOG, "-rt", "rottorque",ffOPTWR },
1938 { efMTX, "-mtx", "nm", ffOPTWR },
1939 { efNDX, "-dn", "dipole", ffOPTWR },
1940 /* Output files that are deleted after each benchmark run */
1941 { efTRN, "-bo", "bench", ffWRITE },
1942 { efXTC, "-bx", "bench", ffWRITE },
1943 { efCPT, "-bcpo", "bench", ffWRITE },
1944 { efSTO, "-bc", "bench", ffWRITE },
1945 { efEDR, "-be", "bench", ffWRITE },
1946 { efLOG, "-bg", "bench", ffWRITE },
1947 { efEDO, "-beo", "bench", ffOPTWR },
1948 { efXVG, "-bdhdl", "benchdhdl",ffOPTWR },
1949 { efXVG, "-bfield", "benchfld" ,ffOPTWR },
1950 { efXVG, "-btpi", "benchtpi", ffOPTWR },
1951 { efXVG, "-btpid", "benchtpid",ffOPTWR },
1952 { efGCT, "-bjo", "bench", ffOPTWR },
1953 { efXVG, "-bffout", "benchgct", ffOPTWR },
1954 { efXVG, "-bdevout","benchdev", ffOPTWR },
1955 { efXVG, "-brunav", "benchrnav",ffOPTWR },
1956 { efXVG, "-bpx", "benchpx", ffOPTWR },
1957 { efXVG, "-bpf", "benchpf", ffOPTWR },
1958 { efXVG, "-bro", "benchrot", ffOPTWR },
1959 { efLOG, "-bra", "benchrota",ffOPTWR },
1960 { efLOG, "-brs", "benchrots",ffOPTWR },
1961 { efLOG, "-brt", "benchrott",ffOPTWR },
1962 { efMTX, "-bmtx", "benchn", ffOPTWR },
1963 { efNDX, "-bdn", "bench", ffOPTWR }
1966 gmx_bool bThreads = FALSE;
1970 const char *procstring[] =
1971 { NULL, "-np", "-n", "none", NULL };
1972 const char *npmevalues_opt[] =
1973 { NULL, "auto", "all", "subset", NULL };
1975 gmx_bool bAppendFiles=TRUE;
1976 gmx_bool bKeepAndNumCPT=FALSE;
1977 gmx_bool bResetCountersHalfWay=FALSE;
1978 gmx_bool bBenchmark=TRUE;
1980 output_env_t oenv=NULL;
1983 /***********************/
1984 /* g_tune_pme options: */
1985 /***********************/
1986 { "-np", FALSE, etINT, {&nnodes},
1987 "Number of nodes to run the tests on (must be > 2 for separate PME nodes)" },
1988 { "-npstring", FALSE, etENUM, {procstring},
1989 "Specify the number of processors to [TT]$MPIRUN[tt] using this string"},
1990 { "-nt", FALSE, etINT, {&nthreads},
1991 "Number of threads to run the tests on (turns MPI & mpirun off)"},
1992 { "-r", FALSE, etINT, {&repeats},
1993 "Repeat each test this often" },
1994 { "-max", FALSE, etREAL, {&maxPMEfraction},
1995 "Max fraction of PME nodes to test with" },
1996 { "-min", FALSE, etREAL, {&minPMEfraction},
1997 "Min fraction of PME nodes to test with" },
1998 { "-npme", FALSE, etENUM, {npmevalues_opt},
1999 "Within -min and -max, benchmark all possible values for [TT]-npme[tt], or just a reasonable subset. "
2000 "Auto neglects -min and -max and chooses reasonable values around a guess for npme derived from the .tpr"},
2001 { "-fix", FALSE, etINT, {&npme_fixed},
2002 "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."},
2003 { "-rmax", FALSE, etREAL, {&rmax},
2004 "If >0, maximal rcoulomb for -ntpr>1 (rcoulomb upscaling results in fourier grid downscaling)" },
2005 { "-rmin", FALSE, etREAL, {&rmin},
2006 "If >0, minimal rcoulomb for -ntpr>1" },
2007 { "-scalevdw", FALSE, etBOOL, {&bScaleRvdw},
2008 "Scale rvdw along with rcoulomb"},
2009 { "-ntpr", FALSE, etINT, {&ntprs},
2010 "Number of [TT].tpr[tt] files to benchmark. Create this many files with different rcoulomb scaling factors depending on -rmin and -rmax. "
2011 "If < 1, automatically choose the number of [TT].tpr[tt] files to test" },
2012 { "-steps", FALSE, etGMX_LARGE_INT, {&bench_nsteps},
2013 "Take timings for this many steps in the benchmark runs" },
2014 { "-resetstep",FALSE, etINT, {&presteps},
2015 "Let dlb equilibrate this many steps before timings are taken (reset cycle counters after this many steps)" },
2016 { "-simsteps", FALSE, etGMX_LARGE_INT, {&new_sim_nsteps},
2017 "If non-negative, perform this many steps in the real run (overwrites nsteps from [TT].tpr[tt], add [TT].cpt[tt] steps)" },
2018 { "-launch", FALSE, etBOOL, {&bLaunch},
2019 "Launch the real simulation after optimization" },
2020 { "-bench", FALSE, etBOOL, {&bBenchmark},
2021 "Run the benchmarks or just create the input [TT].tpr[tt] files?" },
2022 /******************/
2023 /* mdrun options: */
2024 /******************/
2025 /* We let g_tune_pme parse and understand these options, because we need to
2026 * prevent that they appear on the mdrun command line for the benchmarks */
2027 { "-append", FALSE, etBOOL, {&bAppendFiles},
2028 "Append to previous output files when continuing from checkpoint instead of adding the simulation part number to all file names (for launch only)" },
2029 { "-cpnum", FALSE, etBOOL, {&bKeepAndNumCPT},
2030 "Keep and number checkpoint files (launch only)" },
2031 { "-resethway", FALSE, etBOOL, {&bResetCountersHalfWay},
2032 "HIDDENReset the cycle counters after half the number of steps or halfway [TT]-maxh[tt] (launch only)" }
2036 #define NFILE asize(fnm)
2038 CopyRight(stderr,argv[0]);
2040 seconds = gettime();
2042 parse_common_args(&argc,argv,PCA_NOEXIT_ON_ARGS,
2043 NFILE,fnm,asize(pa),pa,asize(desc),desc,
2046 /* Store the remaining unparsed command line entries in a string which
2047 * is then attached to the mdrun command line */
2049 ExtraArgs[0] = '\0';
2050 for (i=1; i<argc; i++) /* argc will now be 1 if everything was understood */
2052 add_to_string(&ExtraArgs, argv[i]);
2053 add_to_string(&ExtraArgs, " ");
2056 if (opt2parg_bSet("-nt",asize(pa),pa))
2059 if (opt2parg_bSet("-npstring",asize(pa),pa))
2060 fprintf(stderr, "WARNING: -npstring has no effect when using threads.\n");
2063 gmx_fatal(FARGS, "Can't run multi-threaded MPI simulation yet!");
2064 /* and now we just set this; a bit of an ugly hack*/
2067 /* Check for PME:PP ratio and whether tpr triggers additional output files */
2068 guessPMEratio = inspect_tpr(NFILE,fnm,&rcoulomb);
2070 /* Automatically set -beo options if -eo is set etc. */
2071 couple_files_options(NFILE,fnm);
2073 /* Construct the command line arguments for benchmark runs
2074 * as well as for the simulation run */
2076 sprintf(bbuf," -nt %d ", nthreads);
2078 sprintf(bbuf," -np %d ", nnodes);
2082 create_command_line_snippets(bThreads,bAppendFiles,bKeepAndNumCPT,bResetCountersHalfWay,presteps,
2083 NFILE,fnm,asize(pa),pa,procstring[0],
2084 &cmd_np, &cmd_args_bench, &cmd_args_launch,
2087 /* Read in checkpoint file if requested */
2089 if(opt2bSet("-cpi",NFILE,fnm))
2092 cr->duty=DUTY_PP; /* makes the following routine happy */
2093 read_checkpoint_simulation_part(opt2fn("-cpi",NFILE,fnm),
2094 &sim_part,&cpt_steps,cr,
2095 FALSE,NFILE,fnm,NULL,NULL);
2098 /* sim_part will now be 1 if no checkpoint file was found */
2100 gmx_fatal(FARGS, "Checkpoint file %s not found!", opt2fn("-cpi",NFILE,fnm));
2103 /* Open performance output file and write header info */
2104 fp = ffopen(opt2fn("-p",NFILE,fnm),"w");
2106 /* Make a quick consistency check of command line parameters */
2107 check_input(nnodes, repeats, &ntprs, &rmin, rcoulomb, &rmax,
2108 maxPMEfraction, minPMEfraction, npme_fixed,
2109 bench_nsteps, fnm, NFILE, sim_part, presteps,
2112 /* Determine the maximum and minimum number of PME nodes to test,
2113 * the actual list of settings is build in do_the_tests(). */
2114 if ((nnodes > 2) && (npme_fixed < -1))
2116 if (0 == strcmp(npmevalues_opt[0], "auto"))
2118 /* Determine the npme range automatically based on the PME:PP load guess */
2119 if (guessPMEratio > 1.0)
2121 /* More PME than PP work, probably we do not need separate PME nodes at all! */
2122 maxPMEnodes=nnodes/2;
2123 minPMEnodes=nnodes/2;
2127 /* PME : PP load is in the range 0..1, let's test around the guess */
2128 guessPMEnodes = nnodes/(1.0 + 1.0/guessPMEratio);
2129 minPMEnodes = floor(0.7*guessPMEnodes);
2130 maxPMEnodes = ceil(1.6*guessPMEnodes);
2131 maxPMEnodes = min(maxPMEnodes, nnodes/2);
2136 /* Determine the npme range based on user input */
2137 maxPMEnodes = floor(maxPMEfraction*nnodes);
2138 minPMEnodes = max(floor(minPMEfraction*nnodes), 0);
2139 fprintf(stdout, "Will try runs with %d ", minPMEnodes);
2140 if (maxPMEnodes != minPMEnodes)
2141 fprintf(stdout, "- %d ", maxPMEnodes);
2142 fprintf(stdout, "PME-only nodes.\n Note that the automatic number of PME-only nodes and no separate PME nodes are always tested.\n");
2151 /* Get the commands we need to set up the runs from environment variables */
2152 get_program_paths(bThreads, &cmd_mpirun, cmd_np, &cmd_mdrun, repeats);
2154 /* Print some header info to file */
2156 fprintf(fp, "\n P E R F O R M A N C E R E S U L T S\n");
2158 fprintf(fp, "%s for Gromacs %s\n", ShortProgram(),GromacsVersion());
2161 fprintf(fp, "Number of nodes : %d\n", nnodes);
2162 fprintf(fp, "The mpirun command is : %s\n", cmd_mpirun);
2163 if ( strcmp(procstring[0], "none") != 0)
2164 fprintf(fp, "Passing # of nodes via : %s\n", procstring[0]);
2166 fprintf(fp, "Not setting number of nodes in system call\n");
2169 fprintf(fp, "Number of threads : %d\n", nnodes);
2171 fprintf(fp, "The mdrun command is : %s\n", cmd_mdrun);
2172 fprintf(fp, "mdrun args benchmarks : %s\n", cmd_args_bench);
2173 fprintf(fp, "Benchmark steps : ");
2174 fprintf(fp, gmx_large_int_pfmt, bench_nsteps);
2176 fprintf(fp, "dlb equilibration steps : %d\n", presteps);
2179 fprintf(fp, "Checkpoint time step : ");
2180 fprintf(fp, gmx_large_int_pfmt, cpt_steps);
2183 fprintf(fp, "mdrun args at launchtime: %s\n", cmd_args_launch);
2185 if (new_sim_nsteps >= 0)
2188 fprintf(stderr, "Note: Simulation input file %s will have ", opt2fn("-so",NFILE,fnm));
2189 fprintf(stderr, gmx_large_int_pfmt, new_sim_nsteps+cpt_steps);
2190 fprintf(stderr, " steps.\n");
2191 fprintf(fp, "Simulation steps : ");
2192 fprintf(fp, gmx_large_int_pfmt, new_sim_nsteps);
2196 fprintf(fp, "Repeats for each test : %d\n", repeats);
2198 if (npme_fixed >= -1)
2200 fprintf(fp, "Fixing -npme at : %d\n", npme_fixed);
2203 fprintf(fp, "Input file : %s\n", opt2fn("-s",NFILE,fnm));
2204 fprintf(fp, " PME/PP load estimate : %g\n", guessPMEratio);
2206 /* Allocate memory for the inputinfo struct: */
2208 info->nr_inputfiles = ntprs;
2209 for (i=0; i<ntprs; i++)
2211 snew(info->rcoulomb , ntprs);
2212 snew(info->rvdw , ntprs);
2213 snew(info->rlist , ntprs);
2214 snew(info->rlistlong, ntprs);
2215 snew(info->nkx , ntprs);
2216 snew(info->nky , ntprs);
2217 snew(info->nkz , ntprs);
2218 snew(info->fsx , ntprs);
2219 snew(info->fsy , ntprs);
2220 snew(info->fsz , ntprs);
2222 /* Make alternative tpr files to test: */
2223 snew(tpr_names, ntprs);
2224 for (i=0; i<ntprs; i++)
2225 snew(tpr_names[i], STRLEN);
2227 /* It can be that ntprs is reduced by make_benchmark_tprs if not enough
2228 * different grids could be found. */
2229 make_benchmark_tprs(opt2fn("-s",NFILE,fnm), tpr_names, bench_nsteps+presteps,
2230 cpt_steps, rmin, rmax, bScaleRvdw, &ntprs, info, fp);
2232 /********************************************************************************/
2233 /* Main loop over all scenarios we need to test: tpr files, PME nodes, repeats */
2234 /********************************************************************************/
2235 snew(perfdata, ntprs);
2238 do_the_tests(fp, tpr_names, maxPMEnodes, minPMEnodes, npme_fixed, npmevalues_opt[0], perfdata, &pmeentries,
2239 repeats, nnodes, ntprs, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2240 cmd_args_bench, fnm, NFILE, sim_part, presteps, cpt_steps);
2242 fprintf(fp, "\nTuning took%8.1f minutes.\n", (gettime()-seconds)/60.0);
2244 /* Analyse the results and give a suggestion for optimal settings: */
2245 bKeepTPR = analyze_data(fp, opt2fn("-p", NFILE, fnm), perfdata, nnodes, ntprs, pmeentries,
2246 repeats, info, &best_tpr, &best_npme);
2248 /* Take the best-performing tpr file and enlarge nsteps to original value */
2249 if ( bKeepTPR && !bOverwrite )
2251 simulation_tpr = opt2fn("-s",NFILE,fnm);
2255 simulation_tpr = opt2fn("-so",NFILE,fnm);
2256 modify_PMEsettings(bOverwrite? (new_sim_nsteps+cpt_steps) :
2257 info->orig_sim_steps, tpr_names[best_tpr],
2261 /* Let's get rid of the temporary benchmark input files */
2262 for (i=0; i<ntprs; i++)
2264 fprintf(stdout, "Deleting temporary benchmark input file %s\n", tpr_names[i]);
2265 remove(tpr_names[i]);
2268 /* Now start the real simulation if the user requested it ... */
2269 launch_simulation(bLaunch, fp, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2270 cmd_args_launch, simulation_tpr, nnodes, best_npme);
2274 /* ... or simply print the performance results to screen: */
2276 finalize(opt2fn("-p", NFILE, fnm));