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;
151 if (diff < 0.0) diff = -diff;
160 static void finalize(const char *fn_out)
166 fp = fopen(fn_out,"r");
167 fprintf(stdout,"\n\n");
169 while( fgets(buf,STRLEN-1,fp) != NULL )
171 fprintf(stdout,"%s",buf);
174 fprintf(stdout,"\n\n");
178 enum {eFoundNothing, eFoundDDStr, eFoundAccountingStr, eFoundCycleStr};
180 static int parse_logfile(const char *logfile, const char *errfile,
181 t_perf *perfdata, int test_nr, int presteps, gmx_large_int_t cpt_steps,
185 char line[STRLEN], dumstring[STRLEN], dumstring2[STRLEN];
186 const char matchstrdd[]="Domain decomposition grid";
187 const char matchstrcr[]="resetting all time and cycle counters";
188 const char matchstrbal[]="Average PME mesh/force load:";
189 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";
190 const char errSIG[]="signal, stopping at the next";
193 float dum1,dum2,dum3,dum4;
196 gmx_large_int_t resetsteps=-1;
197 gmx_bool bFoundResetStr = FALSE;
198 gmx_bool bResetChecked = FALSE;
201 if (!gmx_fexist(logfile))
203 fprintf(stderr, "WARNING: Could not find logfile %s.\n", logfile);
204 cleandata(perfdata, test_nr);
205 return eParselogNotFound;
208 fp = fopen(logfile, "r");
209 perfdata->PME_f_load[test_nr] = -1.0;
210 perfdata->guessPME = -1;
212 iFound = eFoundNothing;
214 iFound = eFoundDDStr; /* Skip some case statements */
216 while (fgets(line, STRLEN, fp) != NULL)
218 /* Remove leading spaces */
221 /* Check for TERM and INT signals from user: */
222 if ( strstr(line, errSIG) != NULL )
225 cleandata(perfdata, test_nr);
226 return eParselogTerm;
229 /* Check whether cycle resetting worked */
230 if (presteps > 0 && !bFoundResetStr)
232 if (strstr(line, matchstrcr) != NULL)
234 sprintf(dumstring, "step %s", gmx_large_int_pfmt);
235 sscanf(line, dumstring, &resetsteps);
236 bFoundResetStr = TRUE;
237 if (resetsteps == presteps+cpt_steps)
239 bResetChecked = TRUE;
243 sprintf(dumstring , gmx_large_int_pfmt, resetsteps);
244 sprintf(dumstring2, gmx_large_int_pfmt, presteps+cpt_steps);
245 fprintf(stderr, "WARNING: Time step counters were reset at step %s,\n"
246 " though they were supposed to be reset at step %s!\n",
247 dumstring, dumstring2);
252 /* Look for strings that appear in a certain order in the log file: */
256 /* Look for domain decomp grid and separate PME nodes: */
257 if (str_starts(line, matchstrdd))
259 sscanf(line, "Domain decomposition grid %d x %d x %d, separate PME nodes %d",
260 &(perfdata->nx), &(perfdata->ny), &(perfdata->nz), &npme);
261 if (perfdata->nPMEnodes == -1)
262 perfdata->guessPME = npme;
263 else if (perfdata->nPMEnodes != npme)
264 gmx_fatal(FARGS, "PME nodes from command line and output file are not identical");
265 iFound = eFoundDDStr;
267 /* Catch a few errors that might have occured: */
268 else if (str_starts(line, "There is no domain decomposition for"))
271 return eParselogNoDDGrid;
273 else if (str_starts(line, "reading tpx file"))
276 return eParselogTPXVersion;
278 else if (str_starts(line, "The -dd or -npme option request a parallel simulation"))
281 return eParselogNotParallel;
285 /* Look for PME mesh/force balance (not necessarily present, though) */
286 if (str_starts(line, matchstrbal))
287 sscanf(&line[strlen(matchstrbal)], "%f", &(perfdata->PME_f_load[test_nr]));
288 /* Look for matchstring */
289 if (str_starts(line, matchstring))
290 iFound = eFoundAccountingStr;
292 case eFoundAccountingStr:
293 /* Already found matchstring - look for cycle data */
294 if (str_starts(line, "Total "))
296 sscanf(line,"Total %d %lf",&procs,&(perfdata->Gcycles[test_nr]));
297 iFound = eFoundCycleStr;
301 /* Already found cycle data - look for remaining performance info and return */
302 if (str_starts(line, "Performance:"))
304 ndum = sscanf(line,"%s %f %f %f %f", dumstring, &dum1, &dum2, &dum3, &dum4);
305 /* (ns/day) is the second last entry, depending on whether GMX_DETAILED_PERF_STATS was set in print_perf(), nrnb.c */
306 perfdata->ns_per_day[test_nr] = (ndum==5)? dum3 : dum1;
308 if (bResetChecked || presteps == 0)
311 return eParselogResetProblem;
317 /* Close the log file */
320 /* Check why there is no performance data in the log file.
321 * Did a fatal errors occur? */
322 if (gmx_fexist(errfile))
324 fp = fopen(errfile, "r");
325 while (fgets(line, STRLEN, fp) != NULL)
327 if ( str_starts(line, "Fatal error:") )
329 if (fgets(line, STRLEN, fp) != NULL)
330 fprintf(stderr, "\nWARNING: An error occured during this benchmark:\n"
333 cleandata(perfdata, test_nr);
334 return eParselogFatal;
341 fprintf(stderr, "WARNING: Could not find stderr file %s.\n", errfile);
344 /* Giving up ... we could not find out why there is no performance data in
346 fprintf(stdout, "No performance data in log file.\n");
347 cleandata(perfdata, test_nr);
349 return eParselogNoPerfData;
353 static gmx_bool analyze_data(
362 int *index_tpr, /* OUT: Nr of mdp file with best settings */
363 int *npme_optimal) /* OUT: Optimal number of PME nodes */
366 int line=0, line_win=-1;
367 int k_win=-1, i_win=-1, winPME;
368 double s=0.0; /* standard deviation */
371 char str_PME_f_load[13];
372 gmx_bool bCanUseOrigTPR;
373 gmx_bool bRefinedCoul, bRefinedVdW, bRefinedGrid;
379 fprintf(fp, "Summary of successful runs:\n");
380 fprintf(fp, "Line tpr PME nodes Gcycles Av. Std.dev. ns/day PME/f");
382 fprintf(fp, " DD grid");
387 for (k=0; k<ntprs; k++)
389 for (i=0; i<ntests; i++)
391 /* Select the right dataset: */
392 pd = &(perfdata[k][i]);
394 pd->Gcycles_Av = 0.0;
395 pd->PME_f_load_Av = 0.0;
396 pd->ns_per_day_Av = 0.0;
398 if (pd->nPMEnodes == -1)
399 sprintf(strbuf, "(%3d)", pd->guessPME);
401 sprintf(strbuf, " ");
403 /* Get the average run time of a setting */
404 for (j=0; j<nrepeats; j++)
406 pd->Gcycles_Av += pd->Gcycles[j];
407 pd->PME_f_load_Av += pd->PME_f_load[j];
409 pd->Gcycles_Av /= nrepeats;
410 pd->PME_f_load_Av /= nrepeats;
412 for (j=0; j<nrepeats; j++)
414 if (pd->ns_per_day[j] > 0.0)
415 pd->ns_per_day_Av += pd->ns_per_day[j];
418 /* Somehow the performance number was not aquired for this run,
419 * therefor set the average to some negative value: */
420 pd->ns_per_day_Av = -1.0f*nrepeats;
424 pd->ns_per_day_Av /= nrepeats;
427 if (pd->PME_f_load_Av > 0.0)
428 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load_Av);
430 sprintf(str_PME_f_load, "%s", " - ");
433 /* We assume we had a successful run if both averages are positive */
434 if (pd->Gcycles_Av > 0.0 && pd->ns_per_day_Av > 0.0)
436 /* Output statistics if repeats were done */
439 /* Calculate the standard deviation */
441 for (j=0; j<nrepeats; j++)
442 s += pow( pd->Gcycles[j] - pd->Gcycles_Av, 2 );
446 fprintf(fp, "%4d %3d %4d%s %12.3f %12.3f %12.3f %s",
447 line, k, pd->nPMEnodes, strbuf, pd->Gcycles_Av, s,
448 pd->ns_per_day_Av, str_PME_f_load);
450 fprintf(fp, " %3d %3d %3d", pd->nx, pd->ny, pd->nz);
453 /* Store the index of the best run found so far in 'winner': */
454 if ( (k_win == -1) || (pd->Gcycles_Av < perfdata[k_win][i_win].Gcycles_Av) )
466 gmx_fatal(FARGS, "None of the runs was successful! Check %s for problems.", fn);
470 winPME = perfdata[k_win][i_win].nPMEnodes;
474 /* We stuck to a fixed number of PME-only nodes */
475 sprintf(strbuf, "settings No. %d", k_win);
479 /* We have optimized the number of PME-only nodes */
481 sprintf(strbuf, "%s", "the automatic number of PME nodes");
483 sprintf(strbuf, "%d PME nodes", winPME);
485 fprintf(fp, "Best performance was achieved with %s", strbuf);
486 if ((nrepeats > 1) && (ntests > 1))
487 fprintf(fp, " (see line %d)", line_win);
490 /* Only mention settings if they were modified: */
491 bRefinedCoul = !is_equal(info->rcoulomb[k_win], info->rcoulomb[0]);
492 bRefinedVdW = !is_equal(info->rvdw[k_win] , info->rvdw[0] );
493 bRefinedGrid = !(info->nkx[k_win] == info->nkx[0] &&
494 info->nky[k_win] == info->nky[0] &&
495 info->nkz[k_win] == info->nkz[0]);
497 if (bRefinedCoul || bRefinedVdW || bRefinedGrid)
499 fprintf(fp, "Optimized PME settings:\n");
500 bCanUseOrigTPR = FALSE;
504 bCanUseOrigTPR = TRUE;
508 fprintf(fp, " New Coulomb radius: %f nm (was %f nm)\n", info->rcoulomb[k_win], info->rcoulomb[0]);
511 fprintf(fp, " New Van der Waals radius: %f nm (was %f nm)\n", info->rvdw[k_win], info->rvdw[0]);
514 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],
515 info->nkx[0], info->nky[0], info->nkz[0]);
517 if (bCanUseOrigTPR && ntprs > 1)
518 fprintf(fp, "and original PME settings.\n");
522 /* Return the index of the mdp file that showed the highest performance
523 * and the optimal number of PME nodes */
525 *npme_optimal = winPME;
527 return bCanUseOrigTPR;
531 /* Get the commands we need to set up the runs from environment variables */
532 static void get_program_paths(gmx_bool bThreads, char *cmd_mpirun[], char cmd_np[],
533 char *cmd_mdrun[], int repeats)
540 const char def_mpirun[] = "mpirun";
541 const char def_mdrun[] = "mdrun";
542 const char filename[] = "benchtest.log";
543 const char match_mpi[] = "NNODES=";
544 const char match_mdrun[]= "Program: ";
545 const char empty_mpirun[] = "";
546 gmx_bool bMdrun = FALSE;
547 gmx_bool bMPI = FALSE;
550 /* Get the commands we need to set up the runs from environment variables */
553 if ( (cp = getenv("MPIRUN")) != NULL)
554 *cmd_mpirun = strdup(cp);
556 *cmd_mpirun = strdup(def_mpirun);
560 *cmd_mpirun = strdup(empty_mpirun);
563 if ( (cp = getenv("MDRUN" )) != NULL )
564 *cmd_mdrun = strdup(cp);
566 *cmd_mdrun = strdup(def_mdrun);
569 /* If no simulations have to be performed, we are done here */
573 /* Run a small test to see whether mpirun + mdrun work */
574 fprintf(stdout, "Making sure that mdrun can be executed. ");
577 snew(command, strlen(*cmd_mdrun) + strlen(cmd_np) + strlen(filename) + 50);
578 sprintf(command, "%s%s-version -maxh 0.001 1> %s 2>&1", *cmd_mdrun, cmd_np, filename);
582 snew(command, strlen(*cmd_mpirun) + strlen(cmd_np) + strlen(*cmd_mdrun) + strlen(filename) + 50);
583 sprintf(command, "%s%s%s -version -maxh 0.001 1> %s 2>&1", *cmd_mpirun, cmd_np, *cmd_mdrun, filename);
585 fprintf(stdout, "Trying '%s' ... ", command);
586 make_backup(filename);
587 gmx_system_call(command);
589 /* Check if we find the characteristic string in the output: */
590 if (!gmx_fexist(filename))
591 gmx_fatal(FARGS, "Output from test run could not be found.");
593 fp = fopen(filename, "r");
594 /* We need to scan the whole output file, since sometimes the queuing system
595 * also writes stuff to stdout/err */
598 cp2=fgets(line, STRLEN, fp);
601 if ( str_starts(line, match_mdrun) )
603 if ( str_starts(line, match_mpi) )
613 gmx_fatal(FARGS, "Need a threaded version of mdrun. This one\n"
615 "seems to have been compiled with MPI instead.",
623 gmx_fatal(FARGS, "Need an MPI-enabled version of mdrun. This one\n"
625 "seems to have been compiled without MPI support.",
632 gmx_fatal(FARGS, "Cannot execute mdrun. Please check %s for problems!",
636 fprintf(stdout, "passed.\n");
644 static void launch_simulation(
645 gmx_bool bLaunch, /* Should the simulation be launched? */
646 FILE *fp, /* General log file */
647 gmx_bool bThreads, /* whether to use threads */
648 char *cmd_mpirun, /* Command for mpirun */
649 char *cmd_np, /* Switch for -np or -nt or empty */
650 char *cmd_mdrun, /* Command for mdrun */
651 char *args_for_mdrun, /* Arguments for mdrun */
652 const char *simulation_tpr, /* This tpr will be simulated */
653 int nnodes, /* Number of nodes to run on */
654 int nPMEnodes) /* Number of PME nodes to use */
659 /* Make enough space for the system call command,
660 * (100 extra chars for -npme ... etc. options should suffice): */
661 snew(command, strlen(cmd_mpirun)+strlen(cmd_mdrun)+strlen(cmd_np)+strlen(args_for_mdrun)+strlen(simulation_tpr)+100);
663 /* Note that the -passall options requires args_for_mdrun to be at the end
664 * of the command line string */
667 sprintf(command, "%s%s-npme %d -s %s %s",
668 cmd_mdrun, cmd_np, nPMEnodes, simulation_tpr, args_for_mdrun);
672 sprintf(command, "%s%s%s -npme %d -s %s %s",
673 cmd_mpirun, cmd_np, cmd_mdrun, nPMEnodes, simulation_tpr, args_for_mdrun);
676 fprintf(fp, "%s this command line to launch the simulation:\n\n%s", bLaunch? "Using":"Please use", command);
680 /* Now the real thing! */
683 fprintf(stdout, "\nLaunching simulation with best parameters now.\nExecuting '%s'", command);
686 gmx_system_call(command);
691 static void modify_PMEsettings(
692 gmx_large_int_t simsteps, /* Set this value as number of time steps */
693 gmx_large_int_t init_step, /* Set this value as init_step */
694 const char *fn_best_tpr, /* tpr file with the best performance */
695 const char *fn_sim_tpr) /* name of tpr file to be launched */
703 read_tpx_state(fn_best_tpr,ir,&state,NULL,&mtop);
705 /* Reset nsteps and init_step to the value of the input .tpr file */
706 ir->nsteps = simsteps;
707 ir->init_step = init_step;
709 /* Write the tpr file which will be launched */
710 sprintf(buf, "Writing optimized simulation file %s with nsteps=%s.\n", fn_sim_tpr, gmx_large_int_pfmt);
711 fprintf(stdout,buf,ir->nsteps);
713 write_tpx_state(fn_sim_tpr,ir,&state,&mtop);
719 #define EPME_SWITCHED(e) ((e) == eelPMESWITCH || (e) == eelPMEUSERSWITCH)
721 /* Make additional TPR files with more computational load for the
722 * direct space processors: */
723 static void make_benchmark_tprs(
724 const char *fn_sim_tpr, /* READ : User-provided tpr file */
725 char *fn_bench_tprs[], /* WRITE: Names of benchmark tpr files */
726 gmx_large_int_t benchsteps, /* Number of time steps for benchmark runs */
727 gmx_large_int_t statesteps, /* Step counter in checkpoint file */
728 real rmin, /* Minimal Coulomb radius */
729 real rmax, /* Maximal Coulomb radius */
730 real bScaleRvdw, /* Scale rvdw along with rcoulomb */
731 int *ntprs, /* No. of TPRs to write, each with a different
732 rcoulomb and fourierspacing */
733 t_inputinfo *info, /* Contains information about mdp file options */
734 FILE *fp) /* Write the output here */
740 real nlist_buffer; /* Thickness of the buffer regions for PME-switch potentials */
743 gmx_bool bNote = FALSE;
744 real add; /* Add this to rcoul for the next test */
745 real fac = 1.0; /* Scaling factor for Coulomb radius */
746 real fourierspacing; /* Basic fourierspacing from tpr */
749 sprintf(buf, "Making benchmark tpr file%s with %s time step%s",
750 *ntprs > 1? "s":"", gmx_large_int_pfmt, benchsteps>1?"s":"");
751 fprintf(stdout, buf, benchsteps);
754 sprintf(buf, " (adding %s steps from checkpoint file)", gmx_large_int_pfmt);
755 fprintf(stdout, buf, statesteps);
756 benchsteps += statesteps;
758 fprintf(stdout, ".\n");
762 read_tpx_state(fn_sim_tpr,ir,&state,NULL,&mtop);
764 /* Check if some kind of PME was chosen */
765 if (EEL_PME(ir->coulombtype) == FALSE)
766 gmx_fatal(FARGS, "Can only do optimizations for simulations with %s electrostatics.",
769 /* Check if rcoulomb == rlist, which is necessary for plain PME. */
770 if ( (ir->cutoff_scheme != ecutsVERLET) &&
771 (eelPME == ir->coulombtype) && !(ir->rcoulomb == ir->rlist))
773 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to rlist (%f).",
774 EELTYPE(eelPME), ir->rcoulomb, ir->rlist);
776 /* For other PME types, rcoulomb is allowed to be smaller than rlist */
777 else if (ir->rcoulomb > ir->rlist)
779 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to or smaller than rlist (%f)",
780 EELTYPE(ir->coulombtype), ir->rcoulomb, ir->rlist);
783 if (bScaleRvdw && ir->rvdw != ir->rcoulomb)
785 fprintf(stdout,"NOTE: input rvdw != rcoulomb, will not scale rvdw\n");
789 /* Reduce the number of steps for the benchmarks */
790 info->orig_sim_steps = ir->nsteps;
791 ir->nsteps = benchsteps;
792 /* We must not use init_step from the input tpr file for the benchmarks */
793 info->orig_init_step = ir->init_step;
796 /* For PME-switch potentials, keep the radial distance of the buffer region */
797 nlist_buffer = ir->rlist - ir->rcoulomb;
799 /* Determine length of triclinic box vectors */
804 box_size[d] += state.box[d][i]*state.box[d][i];
805 box_size[d] = sqrt(box_size[d]);
808 if (ir->fourier_spacing > 0)
810 info->fsx[0] = ir->fourier_spacing;
811 info->fsy[0] = ir->fourier_spacing;
812 info->fsz[0] = ir->fourier_spacing;
816 /* Reconstruct fourierspacing per dimension from the number of grid points and box size */
817 info->fsx[0] = box_size[XX]/ir->nkx;
818 info->fsy[0] = box_size[YY]/ir->nky;
819 info->fsz[0] = box_size[ZZ]/ir->nkz;
822 /* If no value for the fourierspacing was provided on the command line, we
823 * use the reconstruction from the tpr file */
824 if (ir->fourier_spacing > 0)
826 /* Use the spacing from the tpr */
827 fourierspacing = ir->fourier_spacing;
831 /* Use the maximum observed spacing */
832 fourierspacing = max(max(info->fsx[0],info->fsy[0]),info->fsz[0]);
835 fprintf(stdout, "Calculating PME grid points on the basis of a fourierspacing of %f nm\n", fourierspacing);
837 /* For performance comparisons the number of particles is useful to have */
838 fprintf(fp, " Number of particles : %d\n", mtop.natoms);
840 /* Print information about settings of which some are potentially modified: */
841 fprintf(fp, " Coulomb type : %s\n", EELTYPE(ir->coulombtype));
842 fprintf(fp, " Grid spacing x y z : %f %f %f\n",
843 box_size[XX]/ir->nkx, box_size[YY]/ir->nky, box_size[ZZ]/ir->nkz);
844 fprintf(fp, " Van der Waals type : %s\n", EVDWTYPE(ir->vdwtype));
845 if (EVDW_SWITCHED(ir->vdwtype))
846 fprintf(fp, " rvdw_switch : %f nm\n", ir->rvdw_switch);
847 if (EPME_SWITCHED(ir->coulombtype))
848 fprintf(fp, " rlist : %f nm\n", ir->rlist);
849 if (ir->rlistlong != max_cutoff(ir->rvdw,ir->rcoulomb))
850 fprintf(fp, " rlistlong : %f nm\n", ir->rlistlong);
852 /* Print a descriptive line about the tpr settings tested */
853 fprintf(fp, "\nWill try these real/reciprocal workload settings:\n");
854 fprintf(fp, " No. scaling rcoulomb");
855 fprintf(fp, " nkx nky nkz");
856 fprintf(fp, " spacing");
857 if (evdwCUT == ir->vdwtype)
858 fprintf(fp, " rvdw");
859 if (EPME_SWITCHED(ir->coulombtype))
860 fprintf(fp, " rlist");
861 if ( ir->rlistlong != max_cutoff(ir->rlist,max_cutoff(ir->rvdw,ir->rcoulomb)) )
862 fprintf(fp, " rlistlong");
863 fprintf(fp, " tpr file\n");
865 /* Loop to create the requested number of tpr input files */
866 for (j = 0; j < *ntprs; j++)
868 /* The first .tpr is the provided one, just need to modify nsteps,
869 * so skip the following block */
872 /* Determine which Coulomb radii rc to use in the benchmarks */
873 add = (rmax-rmin)/(*ntprs-1);
874 if (is_equal(rmin,info->rcoulomb[0]))
876 ir->rcoulomb = rmin + j*add;
878 else if (is_equal(rmax,info->rcoulomb[0]))
880 ir->rcoulomb = rmin + (j-1)*add;
884 /* rmin != rcoul != rmax, ergo test between rmin and rmax */
885 add = (rmax-rmin)/(*ntprs-2);
886 ir->rcoulomb = rmin + (j-1)*add;
889 /* Determine the scaling factor fac */
890 fac = ir->rcoulomb/info->rcoulomb[0];
892 /* Scale the Fourier grid spacing */
893 ir->nkx = ir->nky = ir->nkz = 0;
894 calc_grid(NULL,state.box,fourierspacing*fac,&ir->nkx,&ir->nky,&ir->nkz);
896 /* Adjust other radii since various conditions neet to be fulfilled */
897 if (eelPME == ir->coulombtype)
899 /* plain PME, rcoulomb must be equal to rlist */
900 ir->rlist = ir->rcoulomb;
904 /* rlist must be >= rcoulomb, we keep the size of the buffer region */
905 ir->rlist = ir->rcoulomb + nlist_buffer;
908 if (bScaleRvdw && evdwCUT == ir->vdwtype)
910 /* For vdw cutoff, rvdw >= rlist */
911 ir->rvdw = max(info->rvdw[0], ir->rlist);
914 ir->rlistlong = max_cutoff(ir->rlist,max_cutoff(ir->rvdw,ir->rcoulomb));
916 } /* end of "if (j != 0)" */
918 /* for j==0: Save the original settings
919 * for j >0: Save modified radii and Fourier grids */
920 info->rcoulomb[j] = ir->rcoulomb;
921 info->rvdw[j] = ir->rvdw;
922 info->nkx[j] = ir->nkx;
923 info->nky[j] = ir->nky;
924 info->nkz[j] = ir->nkz;
925 info->rlist[j] = ir->rlist;
926 info->rlistlong[j] = ir->rlistlong;
927 info->fsx[j] = fac*fourierspacing;
928 info->fsy[j] = fac*fourierspacing;
929 info->fsz[j] = fac*fourierspacing;
931 /* Write the benchmark tpr file */
932 strncpy(fn_bench_tprs[j],fn_sim_tpr,strlen(fn_sim_tpr)-strlen(".tpr"));
933 sprintf(buf, "_bench%.2d.tpr", j);
934 strcat(fn_bench_tprs[j], buf);
935 fprintf(stdout,"Writing benchmark tpr %s with nsteps=", fn_bench_tprs[j]);
936 fprintf(stdout, gmx_large_int_pfmt, ir->nsteps);
938 fprintf(stdout,", scaling factor %f\n", fac);
940 fprintf(stdout,", unmodified settings\n");
942 write_tpx_state(fn_bench_tprs[j],ir,&state,&mtop);
944 /* Write information about modified tpr settings to log file */
945 fprintf(fp, "%4d%10f%10f", j, fac, ir->rcoulomb);
946 fprintf(fp, "%5d%5d%5d", ir->nkx, ir->nky, ir->nkz);
947 fprintf(fp, " %9f ", info->fsx[j]);
948 if (evdwCUT == ir->vdwtype)
949 fprintf(fp, "%10f", ir->rvdw);
950 if (EPME_SWITCHED(ir->coulombtype))
951 fprintf(fp, "%10f", ir->rlist);
952 if ( info->rlistlong[0] != max_cutoff(info->rlist[0],max_cutoff(info->rvdw[0],info->rcoulomb[0])) )
953 fprintf(fp, "%10f", ir->rlistlong);
954 fprintf(fp, " %-14s\n",fn_bench_tprs[j]);
956 /* Make it clear to the user that some additional settings were modified */
957 if ( !is_equal(ir->rvdw , info->rvdw[0])
958 || !is_equal(ir->rlistlong, info->rlistlong[0]) )
964 fprintf(fp, "\nNote that in addition to the Coulomb radius and the Fourier grid\n"
965 "other input settings were also changed (see table above).\n"
966 "Please check if the modified settings are appropriate.\n");
973 /* Rename the files we want to keep to some meaningful filename and
975 static void cleanup(const t_filenm *fnm, int nfile, int k, int nnodes,
976 int nPMEnodes, int nr, gmx_bool bKeepStderr)
978 char numstring[STRLEN];
979 char newfilename[STRLEN];
985 fprintf(stdout, "Cleaning up, deleting benchmark temp files ...\n");
987 for (i=0; i<nfile; i++)
989 opt = (char *)fnm[i].opt;
990 if ( strcmp(opt, "-p") == 0 )
992 /* do nothing; keep this file */
995 else if (strcmp(opt, "-bg") == 0)
997 /* Give the log file a nice name so one can later see which parameters were used */
1000 sprintf(numstring, "_%d", nr);
1001 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", opt2fn("-bg",nfile,fnm), k, nnodes, nPMEnodes, numstring);
1002 if (gmx_fexist(opt2fn("-bg",nfile,fnm)))
1004 fprintf(stdout, "renaming log file to %s\n", newfilename);
1005 make_backup(newfilename);
1006 rename(opt2fn("-bg",nfile,fnm), newfilename);
1009 else if (strcmp(opt, "-err") == 0)
1011 /* This file contains the output of stderr. We want to keep it in
1012 * cases where there have been problems. */
1013 fn = opt2fn(opt, nfile, fnm);
1014 numstring[0] = '\0';
1016 sprintf(numstring, "_%d", nr);
1017 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", fn, k, nnodes, nPMEnodes, numstring);
1022 fprintf(stdout, "Saving stderr output in %s\n", newfilename);
1023 make_backup(newfilename);
1024 rename(fn, newfilename);
1028 fprintf(stdout, "Deleting %s\n", fn);
1033 /* Delete the files which are created for each benchmark run: (options -b*) */
1034 else if ( (0 == strncmp(opt, "-b", 2)) && (opt2bSet(opt,nfile,fnm) || !is_optional(&fnm[i])) )
1036 fn = opt2fn(opt, nfile, fnm);
1039 fprintf(stdout, "Deleting %s\n", fn);
1047 /* Returns the largest common factor of n1 and n2 */
1048 static int largest_common_factor(int n1, int n2)
1053 for (factor=nmax; factor > 0; factor--)
1055 if ( 0==(n1 % factor) && 0==(n2 % factor) )
1060 return 0; /* one for the compiler */
1063 enum {eNpmeAuto, eNpmeAll, eNpmeReduced, eNpmeSubset, eNpmeNr};
1065 /* Create a list of numbers of PME nodes to test */
1066 static void make_npme_list(
1067 const char *npmevalues_opt, /* Make a complete list with all
1068 * possibilities or a short list that keeps only
1069 * reasonable numbers of PME nodes */
1070 int *nentries, /* Number of entries we put in the nPMEnodes list */
1071 int *nPMEnodes[], /* Each entry contains the value for -npme */
1072 int nnodes, /* Total number of nodes to do the tests on */
1073 int minPMEnodes, /* Minimum number of PME nodes */
1074 int maxPMEnodes) /* Maximum number of PME nodes */
1077 int min_factor=1; /* We request that npp and npme have this minimal
1078 * largest common factor (depends on npp) */
1079 int nlistmax; /* Max. list size */
1080 int nlist; /* Actual number of entries in list */
1084 /* Do we need to check all possible values for -npme or is a reduced list enough? */
1085 if ( 0 == strcmp(npmevalues_opt, "all") )
1089 else if ( 0 == strcmp(npmevalues_opt, "subset") )
1091 eNPME = eNpmeSubset;
1093 else /* "auto" or "range" */
1097 else if (nnodes < 128)
1098 eNPME = eNpmeReduced;
1100 eNPME = eNpmeSubset;
1103 /* Calculate how many entries we could possibly have (in case of -npme all) */
1106 nlistmax = maxPMEnodes - minPMEnodes + 3;
1107 if (0 == minPMEnodes)
1113 /* Now make the actual list which is at most of size nlist */
1114 snew(*nPMEnodes, nlistmax);
1115 nlist = 0; /* start counting again, now the real entries in the list */
1116 for (i = 0; i < nlistmax - 2; i++)
1118 npme = maxPMEnodes - i;
1129 /* For 2d PME we want a common largest factor of at least the cube
1130 * root of the number of PP nodes */
1131 min_factor = (int) pow(npp, 1.0/3.0);
1134 gmx_fatal(FARGS, "Unknown option for eNPME in make_npme_list");
1137 if (largest_common_factor(npp, npme) >= min_factor)
1139 (*nPMEnodes)[nlist] = npme;
1143 /* We always test 0 PME nodes and the automatic number */
1144 *nentries = nlist + 2;
1145 (*nPMEnodes)[nlist ] = 0;
1146 (*nPMEnodes)[nlist+1] = -1;
1148 fprintf(stderr, "Will try the following %d different values for -npme:\n", *nentries);
1149 for (i=0; i<*nentries-1; i++)
1150 fprintf(stderr, "%d, ", (*nPMEnodes)[i]);
1151 fprintf(stderr, "and %d (auto).\n", (*nPMEnodes)[*nentries-1]);
1155 /* Allocate memory to store the performance data */
1156 static void init_perfdata(t_perf *perfdata[], int ntprs, int datasets, int repeats)
1161 for (k=0; k<ntprs; k++)
1163 snew(perfdata[k], datasets);
1164 for (i=0; i<datasets; i++)
1166 for (j=0; j<repeats; j++)
1168 snew(perfdata[k][i].Gcycles , repeats);
1169 snew(perfdata[k][i].ns_per_day, repeats);
1170 snew(perfdata[k][i].PME_f_load, repeats);
1177 /* Check for errors on mdrun -h */
1178 static void make_sure_it_runs(char *mdrun_cmd_line, int length, FILE *fp)
1180 char *command, *msg;
1184 snew(command, length + 15);
1185 snew(msg , length + 500);
1187 fprintf(stdout, "Making shure the benchmarks can be executed ...\n");
1188 sprintf(command, "%s-h -quiet", mdrun_cmd_line);
1189 ret = gmx_system_call(command);
1193 /* To prevent confusion, do not again issue a gmx_fatal here since we already
1194 * get the error message from mdrun itself */
1195 sprintf(msg, "Cannot run the benchmark simulations! Please check the error message of\n"
1196 "mdrun for the source of the problem. Did you provide a command line\n"
1197 "argument that neither g_tune_pme nor mdrun understands? Offending command:\n"
1198 "\n%s\n\n", command);
1200 fprintf(stderr, "%s", msg);
1202 fprintf(fp , "%s", msg);
1212 static void do_the_tests(
1213 FILE *fp, /* General g_tune_pme output file */
1214 char **tpr_names, /* Filenames of the input files to test */
1215 int maxPMEnodes, /* Max fraction of nodes to use for PME */
1216 int minPMEnodes, /* Min fraction of nodes to use for PME */
1217 int npme_fixed, /* If >= -1, test fixed number of PME
1219 const char *npmevalues_opt, /* Which -npme values should be tested */
1220 t_perf **perfdata, /* Here the performace data is stored */
1221 int *pmeentries, /* Entries in the nPMEnodes list */
1222 int repeats, /* Repeat each test this often */
1223 int nnodes, /* Total number of nodes = nPP + nPME */
1224 int nr_tprs, /* Total number of tpr files to test */
1225 gmx_bool bThreads, /* Threads or MPI? */
1226 char *cmd_mpirun, /* mpirun command string */
1227 char *cmd_np, /* "-np", "-n", whatever mpirun needs */
1228 char *cmd_mdrun, /* mdrun command string */
1229 char *cmd_args_bench, /* arguments for mdrun in a string */
1230 const t_filenm *fnm, /* List of filenames from command line */
1231 int nfile, /* Number of files specified on the cmdl. */
1232 int sim_part, /* For checkpointing */
1233 int presteps, /* DLB equilibration steps, is checked */
1234 gmx_large_int_t cpt_steps) /* Time step counter in the checkpoint */
1236 int i,nr,k,ret,count=0,totaltests;
1237 int *nPMEnodes=NULL;
1240 char *command, *cmd_stub;
1242 gmx_bool bResetProblem=FALSE;
1243 gmx_bool bFirst=TRUE;
1246 /* This string array corresponds to the eParselog enum type at the start
1248 const char* ParseLog[] = {"OK.",
1249 "Logfile not found!",
1250 "No timings, logfile truncated?",
1251 "Run was terminated.",
1252 "Counters were not reset properly.",
1253 "No DD grid found for these settings.",
1254 "TPX version conflict!",
1255 "mdrun was not started in parallel!",
1256 "An error occured." };
1257 char str_PME_f_load[13];
1260 /* Allocate space for the mdrun command line. 100 extra characters should
1261 be more than enough for the -npme etcetera arguments */
1262 cmdline_length = strlen(cmd_mpirun)
1265 + strlen(cmd_args_bench)
1266 + strlen(tpr_names[0]) + 100;
1267 snew(command , cmdline_length);
1268 snew(cmd_stub, cmdline_length);
1270 /* Construct the part of the command line that stays the same for all tests: */
1273 sprintf(cmd_stub, "%s%s", cmd_mdrun, cmd_np);
1277 sprintf(cmd_stub, "%s%s%s ", cmd_mpirun, cmd_np, cmd_mdrun);
1280 /* Create a list of numbers of PME nodes to test */
1281 if (npme_fixed < -1)
1283 make_npme_list(npmevalues_opt, pmeentries, &nPMEnodes,
1284 nnodes, minPMEnodes, maxPMEnodes);
1290 nPMEnodes[0] = npme_fixed;
1291 fprintf(stderr, "Will use a fixed number of %d PME-only nodes.\n", nPMEnodes[0]);
1296 fprintf(fp, "\nNo benchmarks done since number of repeats (-r) is 0.\n");
1298 finalize(opt2fn("-p", nfile, fnm));
1302 /* Allocate one dataset for each tpr input file: */
1303 init_perfdata(perfdata, nr_tprs, *pmeentries, repeats);
1305 /*****************************************/
1306 /* Main loop over all tpr files to test: */
1307 /*****************************************/
1308 totaltests = nr_tprs*(*pmeentries)*repeats;
1309 for (k=0; k<nr_tprs;k++)
1311 fprintf(fp, "\nIndividual timings for input file %d (%s):\n", k, tpr_names[k]);
1312 fprintf(fp, "PME nodes Gcycles ns/day PME/f Remark\n");
1313 /* Loop over various numbers of PME nodes: */
1314 for (i = 0; i < *pmeentries; i++)
1316 pd = &perfdata[k][i];
1318 /* Loop over the repeats for each scenario: */
1319 for (nr = 0; nr < repeats; nr++)
1321 pd->nPMEnodes = nPMEnodes[i];
1323 /* Add -npme and -s to the command line and save it. Note that
1324 * the -passall (if set) options requires cmd_args_bench to be
1325 * at the end of the command line string */
1326 snew(pd->mdrun_cmd_line, cmdline_length);
1327 sprintf(pd->mdrun_cmd_line, "%s-npme %d -s %s %s",
1328 cmd_stub, pd->nPMEnodes, tpr_names[k], cmd_args_bench);
1330 /* To prevent that all benchmarks fail due to a show-stopper argument
1331 * on the mdrun command line, we make a quick check with mdrun -h first */
1333 make_sure_it_runs(pd->mdrun_cmd_line, cmdline_length, fp);
1336 /* Do a benchmark simulation: */
1338 sprintf(buf, ", pass %d/%d", nr+1, repeats);
1341 fprintf(stdout, "\n=== Progress %2.0f%%, tpr %d/%d, run %d/%d%s:\n",
1342 (100.0*count)/totaltests,
1343 k+1, nr_tprs, i+1, *pmeentries, buf);
1344 make_backup(opt2fn("-err",nfile,fnm));
1345 sprintf(command, "%s 1> /dev/null 2>%s", pd->mdrun_cmd_line, opt2fn("-err",nfile,fnm));
1346 fprintf(stdout, "%s\n", pd->mdrun_cmd_line);
1347 gmx_system_call(command);
1349 /* Collect the performance data from the log file; also check stderr
1350 * for fatal errors */
1351 ret = parse_logfile(opt2fn("-bg",nfile,fnm), opt2fn("-err",nfile,fnm),
1352 pd, nr, presteps, cpt_steps, nnodes);
1353 if ((presteps > 0) && (ret == eParselogResetProblem))
1354 bResetProblem = TRUE;
1356 if (-1 == pd->nPMEnodes)
1357 sprintf(buf, "(%3d)", pd->guessPME);
1362 if (pd->PME_f_load[nr] > 0.0)
1363 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load[nr]);
1365 sprintf(str_PME_f_load, "%s", " - ");
1367 /* Write the data we got to disk */
1368 fprintf(fp, "%4d%s %12.3f %12.3f %s %s", pd->nPMEnodes,
1369 buf, pd->Gcycles[nr], pd->ns_per_day[nr], str_PME_f_load, ParseLog[ret]);
1370 if (! (ret==eParselogOK || ret==eParselogNoDDGrid || ret==eParselogNotFound) )
1371 fprintf(fp, " Check %s file for problems.", ret==eParselogFatal? "err":"log");
1376 /* Do some cleaning up and delete the files we do not need any more */
1377 cleanup(fnm, nfile, k, nnodes, pd->nPMEnodes, nr, ret==eParselogFatal);
1379 /* If the first run with this number of processors already failed, do not try again: */
1380 if (pd->Gcycles[0] <= 0.0 && repeats > 1)
1382 fprintf(stdout, "Skipping remaining passes of unsuccessful setting, see log file for details.\n");
1383 count += repeats-(nr+1);
1386 } /* end of repeats loop */
1387 } /* end of -npme loop */
1388 } /* end of tpr file loop */
1393 fprintf(fp, "WARNING: The cycle and time step counters could not be reset properly. ");
1401 static void check_input(
1408 real maxPMEfraction,
1409 real minPMEfraction,
1411 gmx_large_int_t bench_nsteps,
1412 const t_filenm *fnm,
1422 /* Make sure the input file exists */
1423 if (!gmx_fexist(opt2fn("-s",nfile,fnm)))
1424 gmx_fatal(FARGS, "File %s not found.", opt2fn("-s",nfile,fnm));
1426 /* Make sure that the checkpoint file is not overwritten during benchmarking */
1427 if ( (0 == strcmp(opt2fn("-cpi",nfile,fnm), opt2fn("-bcpo",nfile,fnm)) ) && (sim_part > 1) )
1428 gmx_fatal(FARGS, "Checkpoint input (-cpi) and benchmark checkpoint output (-bcpo) files must not be identical.\n"
1429 "The checkpoint input file must not be overwritten during the benchmarks.\n");
1431 /* Make sure that repeats is >= 0 (if == 0, only write tpr files) */
1433 gmx_fatal(FARGS, "Number of repeats < 0!");
1435 /* Check number of nodes */
1437 gmx_fatal(FARGS, "Number of nodes/threads must be a positive integer.");
1439 /* Automatically choose -ntpr if not set */
1447 /* Set a reasonable scaling factor for rcoulomb */
1449 *rmax = rcoulomb * 1.2;
1451 fprintf(stderr, "Will test %d tpr file%s.\n", *ntprs, *ntprs==1?"":"s");
1456 fprintf(stderr, "Note: Choose ntpr>1 to shift PME load between real and reciprocal space.\n");
1459 /* Make shure that rmin <= rcoulomb <= rmax */
1460 if (*rmin <= 0) *rmin = rcoulomb;
1461 if (*rmax <= 0) *rmax = rcoulomb;
1462 if ( !(*rmin <= *rmax) )
1464 gmx_fatal(FARGS, "Please choose the Coulomb radii such that rmin <= rmax.\n"
1465 "rmin = %g, rmax = %g, actual rcoul from .tpr file = %g\n", *rmin, *rmax, rcoulomb);
1467 /* Add test scenarios if rmin or rmax were set */
1470 if ( !is_equal(*rmin, rcoulomb) && (*ntprs == 1) )
1473 fprintf(stderr, "NOTE: Setting -rmin to %g changed -ntpr to %d\n",
1476 if (!is_equal(*rmax, rcoulomb) && (*ntprs == 1) )
1479 fprintf(stderr, "NOTE: Setting -rmax to %g changed -ntpr to %d\n",
1484 /* If one of rmin, rmax is set, we need 2 tpr files at minimum */
1485 if ( !is_equal(*rmax, rcoulomb) || !is_equal(*rmin, rcoulomb) )
1486 *ntprs = max(*ntprs, 2);
1488 /* If both rmin, rmax are set, we need 3 tpr files at minimum */
1489 if ( !is_equal(*rmax, rcoulomb) && !is_equal(*rmin, rcoulomb) )
1490 *ntprs = max(*ntprs, 3);
1494 fprintf(stderr, "NOTE: Your rmin, rmax setting changed -ntpr to %d\n", *ntprs);
1499 if (is_equal(*rmin,rcoulomb) && is_equal(rcoulomb,*rmax)) /* We have just a single rc */
1501 fprintf(stderr, "WARNING: Resetting -ntpr to 1 since no Coulomb radius scaling is requested.\n"
1502 "Please set rmin < rmax to test Coulomb radii in the [rmin, rmax] interval\n"
1503 "with correspondingly adjusted PME grid settings\n");
1508 /* Check whether max and min fraction are within required values */
1509 if (maxPMEfraction > 0.5 || maxPMEfraction < 0)
1510 gmx_fatal(FARGS, "-max must be between 0 and 0.5");
1511 if (minPMEfraction > 0.5 || minPMEfraction < 0)
1512 gmx_fatal(FARGS, "-min must be between 0 and 0.5");
1513 if (maxPMEfraction < minPMEfraction)
1514 gmx_fatal(FARGS, "-max must be larger or equal to -min");
1516 /* Check whether the number of steps - if it was set - has a reasonable value */
1517 if (bench_nsteps < 0)
1518 gmx_fatal(FARGS, "Number of steps must be positive.");
1520 if (bench_nsteps > 10000 || bench_nsteps < 100)
1522 fprintf(stderr, "WARNING: steps=");
1523 fprintf(stderr, gmx_large_int_pfmt, bench_nsteps);
1524 fprintf(stderr, ". Are you sure you want to perform so %s steps for each benchmark?\n", (bench_nsteps < 100)? "few" : "many");
1529 gmx_fatal(FARGS, "Cannot have a negative number of presteps.\n");
1532 /* Check for rcoulomb scaling if more than one .tpr file is tested */
1535 if (*rmin/rcoulomb < 0.75 || *rmax/rcoulomb > 1.25)
1536 fprintf(stderr, "WARNING: Applying extreme scaling factor. I hope you know what you are doing.\n");
1539 /* If a fixed number of PME nodes is set we do rcoulomb and PME gird tuning
1540 * only. We need to check whether the requested number of PME-only nodes
1542 if (npme_fixed > -1)
1544 /* No more than 50% of all nodes can be assigned as PME-only nodes. */
1545 if (2*npme_fixed > nnodes)
1547 gmx_fatal(FARGS, "Cannot have more than %d PME-only nodes for a total of %d nodes (you chose %d).\n",
1548 nnodes/2, nnodes, npme_fixed);
1550 if ((npme_fixed > 0) && (5*npme_fixed < nnodes))
1552 fprintf(stderr, "WARNING: Only %g percent of the nodes are assigned as PME-only nodes.\n",
1553 100.0*((real)npme_fixed / (real)nnodes));
1555 if (opt2parg_bSet("-min",npargs,pa) || opt2parg_bSet("-max",npargs,pa))
1557 fprintf(stderr, "NOTE: The -min, -max, and -npme options have no effect when a\n"
1558 " fixed number of PME-only nodes is requested with -fix.\n");
1564 /* Returns TRUE when "opt" is needed at launch time */
1565 static gmx_bool is_launch_file(char *opt, gmx_bool bSet)
1567 /* Apart from the input .tpr we need all options that were set
1568 * on the command line and that do not start with -b */
1569 if (0 == strncmp(opt,"-b", 2) || 0 == strncmp(opt,"-s", 2))
1579 /* Returns TRUE when "opt" defines a file which is needed for the benchmarks runs */
1580 static gmx_bool is_bench_file(char *opt, gmx_bool bSet, gmx_bool bOptional, gmx_bool bIsOutput)
1582 /* Apart from the input .tpr, all files starting with "-b" are for
1583 * _b_enchmark files exclusively */
1584 if (0 == strncmp(opt,"-s", 2)) return FALSE;
1585 if (0 == strncmp(opt,"-b", 2) || 0 == strncmp(opt,"-s", 2))
1587 if (!bOptional || bSet)
1597 if (bSet) /* These are additional input files like -cpi -ei */
1605 /* Adds 'buf' to 'str' */
1606 static void add_to_string(char **str, char *buf)
1611 len = strlen(*str) + strlen(buf) + 1;
1617 /* Create the command line for the benchmark as well as for the real run */
1618 static void create_command_line_snippets(
1620 gmx_bool bAppendFiles,
1621 gmx_bool bKeepAndNumCPT,
1622 gmx_bool bResetHWay,
1628 const char *procstring, /* How to pass the number of processors to $MPIRUN */
1629 char *cmd_np[], /* Actual command line snippet, e.g. '-np <N>' */
1630 char *cmd_args_bench[], /* command line arguments for benchmark runs */
1631 char *cmd_args_launch[], /* command line arguments for simulation run */
1632 char extra_args[]) /* Add this to the end of the command line */
1637 char strbuf[STRLEN];
1640 /* strlen needs at least '\0' as a string: */
1641 snew(*cmd_args_bench ,1);
1642 snew(*cmd_args_launch,1);
1643 *cmd_args_launch[0]='\0';
1644 *cmd_args_bench[0] ='\0';
1647 /*******************************************/
1648 /* 1. Process other command line arguments */
1649 /*******************************************/
1652 /* Add equilibration steps to benchmark options */
1653 sprintf(strbuf, "-resetstep %d ", presteps);
1654 add_to_string(cmd_args_bench, strbuf);
1656 /* These switches take effect only at launch time */
1657 if (FALSE == bAppendFiles)
1659 add_to_string(cmd_args_launch, "-noappend ");
1663 add_to_string(cmd_args_launch, "-cpnum ");
1667 add_to_string(cmd_args_launch, "-resethway ");
1670 /********************/
1671 /* 2. Process files */
1672 /********************/
1673 for (i=0; i<nfile; i++)
1675 opt = (char *)fnm[i].opt;
1676 name = opt2fn(opt,nfile,fnm);
1678 /* Strbuf contains the options, now let's sort out where we need that */
1679 sprintf(strbuf, "%s %s ", opt, name);
1681 if ( is_bench_file(opt, opt2bSet(opt,nfile,fnm), is_optional(&fnm[i]), is_output(&fnm[i])) )
1683 /* All options starting with -b* need the 'b' removed,
1684 * therefore overwrite strbuf */
1685 if (0 == strncmp(opt, "-b", 2))
1686 sprintf(strbuf, "-%s %s ", &opt[2], name);
1688 add_to_string(cmd_args_bench, strbuf);
1691 if ( is_launch_file(opt,opt2bSet(opt,nfile,fnm)) )
1692 add_to_string(cmd_args_launch, strbuf);
1695 add_to_string(cmd_args_bench , extra_args);
1696 add_to_string(cmd_args_launch, extra_args);
1700 /* Set option opt */
1701 static void setopt(const char *opt,int nfile,t_filenm fnm[])
1705 for(i=0; (i<nfile); i++)
1706 if (strcmp(opt,fnm[i].opt)==0)
1707 fnm[i].flag |= ffSET;
1711 /* This routine inspects the tpr file and ...
1712 * 1. checks for output files that get triggered by a tpr option. These output
1713 * files are marked as 'set' to allow for a proper cleanup after each
1715 * 2. returns the PME:PP load ratio
1716 * 3. returns rcoulomb from the tpr */
1717 static float inspect_tpr(int nfile, t_filenm fnm[], real *rcoulomb)
1719 gmx_bool bPull; /* Is pulling requested in .tpr file? */
1720 gmx_bool bTpi; /* Is test particle insertion requested? */
1721 gmx_bool bFree; /* Is a free energy simulation requested? */
1722 gmx_bool bNM; /* Is a normal mode analysis requested? */
1728 /* Check tpr file for options that trigger extra output files */
1729 read_tpx_state(opt2fn("-s",nfile,fnm),&ir,&state,NULL,&mtop);
1730 bPull = (epullNO != ir.ePull);
1731 bFree = (efepNO != ir.efep );
1732 bNM = (eiNM == ir.eI );
1733 bTpi = EI_TPI(ir.eI);
1735 /* Set these output files on the tuning command-line */
1738 setopt("-pf" , nfile, fnm);
1739 setopt("-px" , nfile, fnm);
1743 setopt("-dhdl", nfile, fnm);
1747 setopt("-tpi" , nfile, fnm);
1748 setopt("-tpid", nfile, fnm);
1752 setopt("-mtx" , nfile, fnm);
1755 *rcoulomb = ir.rcoulomb;
1757 /* Return the estimate for the number of PME nodes */
1758 return pme_load_estimate(&mtop,&ir,state.box);
1762 static void couple_files_options(int nfile, t_filenm fnm[])
1765 gmx_bool bSet,bBench;
1770 for (i=0; i<nfile; i++)
1772 opt = (char *)fnm[i].opt;
1773 bSet = ((fnm[i].flag & ffSET) != 0);
1774 bBench = (0 == strncmp(opt,"-b", 2));
1776 /* Check optional files */
1777 /* If e.g. -eo is set, then -beo also needs to be set */
1778 if (is_optional(&fnm[i]) && bSet && !bBench)
1780 sprintf(buf, "-b%s", &opt[1]);
1781 setopt(buf,nfile,fnm);
1783 /* If -beo is set, then -eo also needs to be! */
1784 if (is_optional(&fnm[i]) && bSet && bBench)
1786 sprintf(buf, "-%s", &opt[2]);
1787 setopt(buf,nfile,fnm);
1793 static double gettime()
1795 #ifdef HAVE_GETTIMEOFDAY
1799 gettimeofday(&t,NULL);
1801 seconds = (double) t.tv_sec + 1e-6*(double)t.tv_usec;
1807 seconds = time(NULL);
1814 #define BENCHSTEPS (1000)
1816 int gmx_tune_pme(int argc,char *argv[])
1818 const char *desc[] = {
1819 "For a given number [TT]-np[tt] or [TT]-nt[tt] of processors/threads, this program systematically",
1820 "times [TT]mdrun[tt] with various numbers of PME-only nodes and determines",
1821 "which setting is fastest. It will also test whether performance can",
1822 "be enhanced by shifting load from the reciprocal to the real space",
1823 "part of the Ewald sum. ",
1824 "Simply pass your [TT].tpr[tt] file to [TT]g_tune_pme[tt] together with other options",
1825 "for [TT]mdrun[tt] as needed.[PAR]",
1826 "Which executables are used can be set in the environment variables",
1827 "MPIRUN and MDRUN. If these are not present, 'mpirun' and 'mdrun'",
1828 "will be used as defaults. Note that for certain MPI frameworks you",
1829 "need to provide a machine- or hostfile. This can also be passed",
1830 "via the MPIRUN variable, e.g.[PAR]",
1831 "[TT]export MPIRUN=\"/usr/local/mpirun -machinefile hosts\"[tt][PAR]",
1832 "Please call [TT]g_tune_pme[tt] with the normal options you would pass to",
1833 "[TT]mdrun[tt] and add [TT]-np[tt] for the number of processors to perform the",
1834 "tests on, or [TT]-nt[tt] for the number of threads. You can also add [TT]-r[tt]",
1835 "to repeat each test several times to get better statistics. [PAR]",
1836 "[TT]g_tune_pme[tt] can test various real space / reciprocal space workloads",
1837 "for you. With [TT]-ntpr[tt] you control how many extra [TT].tpr[tt] files will be",
1838 "written with enlarged cutoffs and smaller Fourier grids respectively.",
1839 "Typically, the first test (number 0) will be with the settings from the input",
1840 "[TT].tpr[tt] file; the last test (number [TT]ntpr[tt]) will have the Coulomb cutoff",
1841 "specified by [TT]-rmax[tt] with a somwhat smaller PME grid at the same time. ",
1842 "In this last test, the Fourier spacing is multiplied with [TT]rmax[tt]/rcoulomb. ",
1843 "The remaining [TT].tpr[tt] files will have equally-spaced Coulomb radii (and Fourier "
1844 "spacings) between these extremes. [BB]Note[bb] that you can set [TT]-ntpr[tt] to 1",
1845 "if you just seek the optimal number of PME-only nodes; in that case",
1846 "your input [TT].tpr[tt] file will remain unchanged.[PAR]",
1847 "For the benchmark runs, the default of 1000 time steps should suffice for most",
1848 "MD systems. The dynamic load balancing needs about 100 time steps",
1849 "to adapt to local load imbalances, therefore the time step counters",
1850 "are by default reset after 100 steps. For large systems (>1M atoms), as well as ",
1851 "for a higher accuarcy of the measurements, you should set [TT]-resetstep[tt] to a higher value.",
1852 "From the 'DD' load imbalance entries in the md.log output file you",
1853 "can tell after how many steps the load is sufficiently balanced. Example call:[PAR]"
1854 "[TT]g_tune_pme -np 64 -s protein.tpr -launch[tt][PAR]",
1855 "After calling [TT]mdrun[tt] several times, detailed performance information",
1856 "is available in the output file [TT]perf.out.[tt] ",
1857 "[BB]Note[bb] that during the benchmarks, a couple of temporary files are written",
1858 "(options [TT]-b[tt]*), these will be automatically deleted after each test.[PAR]",
1859 "If you want the simulation to be started automatically with the",
1860 "optimized parameters, use the command line option [TT]-launch[tt].[PAR]",
1865 int pmeentries=0; /* How many values for -npme do we actually test for each tpr file */
1866 real maxPMEfraction=0.50;
1867 real minPMEfraction=0.25;
1868 int maxPMEnodes, minPMEnodes;
1869 float guessPMEratio; /* guessed PME:PP ratio based on the tpr file */
1870 float guessPMEnodes;
1871 int npme_fixed=-2; /* If >= -1, use only this number
1872 * of PME-only nodes */
1874 real rmin=0.0,rmax=0.0; /* min and max value for rcoulomb if scaling is requested */
1875 real rcoulomb=-1.0; /* Coulomb radius as set in .tpr file */
1876 gmx_bool bScaleRvdw=TRUE;
1877 gmx_large_int_t bench_nsteps=BENCHSTEPS;
1878 gmx_large_int_t new_sim_nsteps=-1; /* -1 indicates: not set by the user */
1879 gmx_large_int_t cpt_steps=0; /* Step counter in .cpt input file */
1880 int presteps=100; /* Do a full cycle reset after presteps steps */
1881 gmx_bool bOverwrite=FALSE, bKeepTPR;
1882 gmx_bool bLaunch=FALSE;
1883 char *ExtraArgs=NULL;
1884 char **tpr_names=NULL;
1885 const char *simulation_tpr=NULL;
1886 int best_npme, best_tpr;
1887 int sim_part = 1; /* For benchmarks with checkpoint files */
1890 /* Default program names if nothing else is found */
1891 char *cmd_mpirun=NULL, *cmd_mdrun=NULL;
1892 char *cmd_args_bench, *cmd_args_launch;
1895 t_perf **perfdata=NULL;
1901 /* Print out how long the tuning took */
1904 static t_filenm fnm[] = {
1906 { efOUT, "-p", "perf", ffWRITE },
1907 { efLOG, "-err", "bencherr", ffWRITE },
1908 { efTPX, "-so", "tuned", ffWRITE },
1910 { efTPX, NULL, NULL, ffREAD },
1911 { efTRN, "-o", NULL, ffWRITE },
1912 { efXTC, "-x", NULL, ffOPTWR },
1913 { efCPT, "-cpi", NULL, ffOPTRD },
1914 { efCPT, "-cpo", NULL, ffOPTWR },
1915 { efSTO, "-c", "confout", ffWRITE },
1916 { efEDR, "-e", "ener", ffWRITE },
1917 { efLOG, "-g", "md", ffWRITE },
1918 { efXVG, "-dhdl", "dhdl", ffOPTWR },
1919 { efXVG, "-field", "field", ffOPTWR },
1920 { efXVG, "-table", "table", ffOPTRD },
1921 { efXVG, "-tabletf", "tabletf", ffOPTRD },
1922 { efXVG, "-tablep", "tablep", ffOPTRD },
1923 { efXVG, "-tableb", "table", ffOPTRD },
1924 { efTRX, "-rerun", "rerun", ffOPTRD },
1925 { efXVG, "-tpi", "tpi", ffOPTWR },
1926 { efXVG, "-tpid", "tpidist", ffOPTWR },
1927 { efEDI, "-ei", "sam", ffOPTRD },
1928 { efEDO, "-eo", "sam", ffOPTWR },
1929 { efGCT, "-j", "wham", ffOPTRD },
1930 { efGCT, "-jo", "bam", ffOPTWR },
1931 { efXVG, "-ffout", "gct", ffOPTWR },
1932 { efXVG, "-devout", "deviatie", ffOPTWR },
1933 { efXVG, "-runav", "runaver", ffOPTWR },
1934 { efXVG, "-px", "pullx", ffOPTWR },
1935 { efXVG, "-pf", "pullf", ffOPTWR },
1936 { efXVG, "-ro", "rotation", ffOPTWR },
1937 { efLOG, "-ra", "rotangles",ffOPTWR },
1938 { efLOG, "-rs", "rotslabs", ffOPTWR },
1939 { efLOG, "-rt", "rottorque",ffOPTWR },
1940 { efMTX, "-mtx", "nm", ffOPTWR },
1941 { efNDX, "-dn", "dipole", ffOPTWR },
1942 /* Output files that are deleted after each benchmark run */
1943 { efTRN, "-bo", "bench", ffWRITE },
1944 { efXTC, "-bx", "bench", ffWRITE },
1945 { efCPT, "-bcpo", "bench", ffWRITE },
1946 { efSTO, "-bc", "bench", ffWRITE },
1947 { efEDR, "-be", "bench", ffWRITE },
1948 { efLOG, "-bg", "bench", ffWRITE },
1949 { efEDO, "-beo", "bench", ffOPTWR },
1950 { efXVG, "-bdhdl", "benchdhdl",ffOPTWR },
1951 { efXVG, "-bfield", "benchfld" ,ffOPTWR },
1952 { efXVG, "-btpi", "benchtpi", ffOPTWR },
1953 { efXVG, "-btpid", "benchtpid",ffOPTWR },
1954 { efGCT, "-bjo", "bench", ffOPTWR },
1955 { efXVG, "-bffout", "benchgct", ffOPTWR },
1956 { efXVG, "-bdevout","benchdev", ffOPTWR },
1957 { efXVG, "-brunav", "benchrnav",ffOPTWR },
1958 { efXVG, "-bpx", "benchpx", ffOPTWR },
1959 { efXVG, "-bpf", "benchpf", ffOPTWR },
1960 { efXVG, "-bro", "benchrot", ffOPTWR },
1961 { efLOG, "-bra", "benchrota",ffOPTWR },
1962 { efLOG, "-brs", "benchrots",ffOPTWR },
1963 { efLOG, "-brt", "benchrott",ffOPTWR },
1964 { efMTX, "-bmtx", "benchn", ffOPTWR },
1965 { efNDX, "-bdn", "bench", ffOPTWR }
1968 gmx_bool bThreads = FALSE;
1972 const char *procstring[] =
1973 { NULL, "-np", "-n", "none", NULL };
1974 const char *npmevalues_opt[] =
1975 { NULL, "auto", "all", "subset", NULL };
1977 gmx_bool bAppendFiles=TRUE;
1978 gmx_bool bKeepAndNumCPT=FALSE;
1979 gmx_bool bResetCountersHalfWay=FALSE;
1980 gmx_bool bBenchmark=TRUE;
1982 output_env_t oenv=NULL;
1985 /***********************/
1986 /* g_tune_pme options: */
1987 /***********************/
1988 { "-np", FALSE, etINT, {&nnodes},
1989 "Number of nodes to run the tests on (must be > 2 for separate PME nodes)" },
1990 { "-npstring", FALSE, etENUM, {procstring},
1991 "Specify the number of processors to [TT]$MPIRUN[tt] using this string"},
1992 { "-nt", FALSE, etINT, {&nthreads},
1993 "Number of threads to run the tests on (turns MPI & mpirun off)"},
1994 { "-r", FALSE, etINT, {&repeats},
1995 "Repeat each test this often" },
1996 { "-max", FALSE, etREAL, {&maxPMEfraction},
1997 "Max fraction of PME nodes to test with" },
1998 { "-min", FALSE, etREAL, {&minPMEfraction},
1999 "Min fraction of PME nodes to test with" },
2000 { "-npme", FALSE, etENUM, {npmevalues_opt},
2001 "Within -min and -max, benchmark all possible values for [TT]-npme[tt], or just a reasonable subset. "
2002 "Auto neglects -min and -max and chooses reasonable values around a guess for npme derived from the .tpr"},
2003 { "-fix", FALSE, etINT, {&npme_fixed},
2004 "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."},
2005 { "-rmax", FALSE, etREAL, {&rmax},
2006 "If >0, maximal rcoulomb for -ntpr>1 (rcoulomb upscaling results in fourier grid downscaling)" },
2007 { "-rmin", FALSE, etREAL, {&rmin},
2008 "If >0, minimal rcoulomb for -ntpr>1" },
2009 { "-scalevdw", FALSE, etBOOL, {&bScaleRvdw},
2010 "Scale rvdw along with rcoulomb"},
2011 { "-ntpr", FALSE, etINT, {&ntprs},
2012 "Number of [TT].tpr[tt] files to benchmark. Create this many files with different rcoulomb scaling factors depending on -rmin and -rmax. "
2013 "If < 1, automatically choose the number of [TT].tpr[tt] files to test" },
2014 { "-steps", FALSE, etGMX_LARGE_INT, {&bench_nsteps},
2015 "Take timings for this many steps in the benchmark runs" },
2016 { "-resetstep",FALSE, etINT, {&presteps},
2017 "Let dlb equilibrate this many steps before timings are taken (reset cycle counters after this many steps)" },
2018 { "-simsteps", FALSE, etGMX_LARGE_INT, {&new_sim_nsteps},
2019 "If non-negative, perform this many steps in the real run (overwrites nsteps from [TT].tpr[tt], add [TT].cpt[tt] steps)" },
2020 { "-launch", FALSE, etBOOL, {&bLaunch},
2021 "Launch the real simulation after optimization" },
2022 { "-bench", FALSE, etBOOL, {&bBenchmark},
2023 "Run the benchmarks or just create the input [TT].tpr[tt] files?" },
2024 /******************/
2025 /* mdrun options: */
2026 /******************/
2027 /* We let g_tune_pme parse and understand these options, because we need to
2028 * prevent that they appear on the mdrun command line for the benchmarks */
2029 { "-append", FALSE, etBOOL, {&bAppendFiles},
2030 "Append to previous output files when continuing from checkpoint instead of adding the simulation part number to all file names (for launch only)" },
2031 { "-cpnum", FALSE, etBOOL, {&bKeepAndNumCPT},
2032 "Keep and number checkpoint files (launch only)" },
2033 { "-resethway", FALSE, etBOOL, {&bResetCountersHalfWay},
2034 "HIDDENReset the cycle counters after half the number of steps or halfway [TT]-maxh[tt] (launch only)" }
2038 #define NFILE asize(fnm)
2040 CopyRight(stderr,argv[0]);
2042 seconds = gettime();
2044 parse_common_args(&argc,argv,PCA_NOEXIT_ON_ARGS,
2045 NFILE,fnm,asize(pa),pa,asize(desc),desc,
2048 /* Store the remaining unparsed command line entries in a string which
2049 * is then attached to the mdrun command line */
2051 ExtraArgs[0] = '\0';
2052 for (i=1; i<argc; i++) /* argc will now be 1 if everything was understood */
2054 add_to_string(&ExtraArgs, argv[i]);
2055 add_to_string(&ExtraArgs, " ");
2058 if (opt2parg_bSet("-nt",asize(pa),pa))
2061 if (opt2parg_bSet("-npstring",asize(pa),pa))
2062 fprintf(stderr, "WARNING: -npstring has no effect when using threads.\n");
2065 gmx_fatal(FARGS, "Can't run multi-threaded MPI simulation yet!");
2066 /* and now we just set this; a bit of an ugly hack*/
2069 /* Check for PME:PP ratio and whether tpr triggers additional output files */
2070 guessPMEratio = inspect_tpr(NFILE,fnm,&rcoulomb);
2072 /* Automatically set -beo options if -eo is set etc. */
2073 couple_files_options(NFILE,fnm);
2075 /* Construct the command line arguments for benchmark runs
2076 * as well as for the simulation run */
2078 sprintf(bbuf," -nt %d ", nthreads);
2080 sprintf(bbuf," -np %d ", nnodes);
2084 create_command_line_snippets(bThreads,bAppendFiles,bKeepAndNumCPT,bResetCountersHalfWay,presteps,
2085 NFILE,fnm,asize(pa),pa,procstring[0],
2086 &cmd_np, &cmd_args_bench, &cmd_args_launch,
2089 /* Read in checkpoint file if requested */
2091 if(opt2bSet("-cpi",NFILE,fnm))
2094 cr->duty=DUTY_PP; /* makes the following routine happy */
2095 read_checkpoint_simulation_part(opt2fn("-cpi",NFILE,fnm),
2096 &sim_part,&cpt_steps,cr,
2097 FALSE,NFILE,fnm,NULL,NULL);
2100 /* sim_part will now be 1 if no checkpoint file was found */
2102 gmx_fatal(FARGS, "Checkpoint file %s not found!", opt2fn("-cpi",NFILE,fnm));
2105 /* Open performance output file and write header info */
2106 fp = ffopen(opt2fn("-p",NFILE,fnm),"w");
2108 /* Make a quick consistency check of command line parameters */
2109 check_input(nnodes, repeats, &ntprs, &rmin, rcoulomb, &rmax,
2110 maxPMEfraction, minPMEfraction, npme_fixed,
2111 bench_nsteps, fnm, NFILE, sim_part, presteps,
2114 /* Determine the maximum and minimum number of PME nodes to test,
2115 * the actual list of settings is build in do_the_tests(). */
2116 if ((nnodes > 2) && (npme_fixed < -1))
2118 if (0 == strcmp(npmevalues_opt[0], "auto"))
2120 /* Determine the npme range automatically based on the PME:PP load guess */
2121 if (guessPMEratio > 1.0)
2123 /* More PME than PP work, probably we do not need separate PME nodes at all! */
2124 maxPMEnodes=nnodes/2;
2125 minPMEnodes=nnodes/2;
2129 /* PME : PP load is in the range 0..1, let's test around the guess */
2130 guessPMEnodes = nnodes/(1.0 + 1.0/guessPMEratio);
2131 minPMEnodes = floor(0.7*guessPMEnodes);
2132 maxPMEnodes = ceil(1.6*guessPMEnodes);
2133 maxPMEnodes = min(maxPMEnodes, nnodes/2);
2138 /* Determine the npme range based on user input */
2139 maxPMEnodes = floor(maxPMEfraction*nnodes);
2140 minPMEnodes = max(floor(minPMEfraction*nnodes), 0);
2141 fprintf(stdout, "Will try runs with %d ", minPMEnodes);
2142 if (maxPMEnodes != minPMEnodes)
2143 fprintf(stdout, "- %d ", maxPMEnodes);
2144 fprintf(stdout, "PME-only nodes.\n Note that the automatic number of PME-only nodes and no separate PME nodes are always tested.\n");
2153 /* Get the commands we need to set up the runs from environment variables */
2154 get_program_paths(bThreads, &cmd_mpirun, cmd_np, &cmd_mdrun, repeats);
2156 /* Print some header info to file */
2158 fprintf(fp, "\n P E R F O R M A N C E R E S U L T S\n");
2160 fprintf(fp, "%s for Gromacs %s\n", ShortProgram(),GromacsVersion());
2163 fprintf(fp, "Number of nodes : %d\n", nnodes);
2164 fprintf(fp, "The mpirun command is : %s\n", cmd_mpirun);
2165 if ( strcmp(procstring[0], "none") != 0)
2166 fprintf(fp, "Passing # of nodes via : %s\n", procstring[0]);
2168 fprintf(fp, "Not setting number of nodes in system call\n");
2171 fprintf(fp, "Number of threads : %d\n", nnodes);
2173 fprintf(fp, "The mdrun command is : %s\n", cmd_mdrun);
2174 fprintf(fp, "mdrun args benchmarks : %s\n", cmd_args_bench);
2175 fprintf(fp, "Benchmark steps : ");
2176 fprintf(fp, gmx_large_int_pfmt, bench_nsteps);
2178 fprintf(fp, "dlb equilibration steps : %d\n", presteps);
2181 fprintf(fp, "Checkpoint time step : ");
2182 fprintf(fp, gmx_large_int_pfmt, cpt_steps);
2185 fprintf(fp, "mdrun args at launchtime: %s\n", cmd_args_launch);
2187 if (new_sim_nsteps >= 0)
2190 fprintf(stderr, "Note: Simulation input file %s will have ", opt2fn("-so",NFILE,fnm));
2191 fprintf(stderr, gmx_large_int_pfmt, new_sim_nsteps+cpt_steps);
2192 fprintf(stderr, " steps.\n");
2193 fprintf(fp, "Simulation steps : ");
2194 fprintf(fp, gmx_large_int_pfmt, new_sim_nsteps);
2198 fprintf(fp, "Repeats for each test : %d\n", repeats);
2200 if (npme_fixed >= -1)
2202 fprintf(fp, "Fixing -npme at : %d\n", npme_fixed);
2205 fprintf(fp, "Input file : %s\n", opt2fn("-s",NFILE,fnm));
2206 fprintf(fp, " PME/PP load estimate : %g\n", guessPMEratio);
2208 /* Allocate memory for the inputinfo struct: */
2210 info->nr_inputfiles = ntprs;
2211 for (i=0; i<ntprs; i++)
2213 snew(info->rcoulomb , ntprs);
2214 snew(info->rvdw , ntprs);
2215 snew(info->rlist , ntprs);
2216 snew(info->rlistlong, ntprs);
2217 snew(info->nkx , ntprs);
2218 snew(info->nky , ntprs);
2219 snew(info->nkz , ntprs);
2220 snew(info->fsx , ntprs);
2221 snew(info->fsy , ntprs);
2222 snew(info->fsz , ntprs);
2224 /* Make alternative tpr files to test: */
2225 snew(tpr_names, ntprs);
2226 for (i=0; i<ntprs; i++)
2227 snew(tpr_names[i], STRLEN);
2229 /* It can be that ntprs is reduced by make_benchmark_tprs if not enough
2230 * different grids could be found. */
2231 make_benchmark_tprs(opt2fn("-s",NFILE,fnm), tpr_names, bench_nsteps+presteps,
2232 cpt_steps, rmin, rmax, bScaleRvdw, &ntprs, info, fp);
2234 /********************************************************************************/
2235 /* Main loop over all scenarios we need to test: tpr files, PME nodes, repeats */
2236 /********************************************************************************/
2237 snew(perfdata, ntprs);
2240 do_the_tests(fp, tpr_names, maxPMEnodes, minPMEnodes, npme_fixed, npmevalues_opt[0], perfdata, &pmeentries,
2241 repeats, nnodes, ntprs, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2242 cmd_args_bench, fnm, NFILE, sim_part, presteps, cpt_steps);
2244 fprintf(fp, "\nTuning took%8.1f minutes.\n", (gettime()-seconds)/60.0);
2246 /* Analyse the results and give a suggestion for optimal settings: */
2247 bKeepTPR = analyze_data(fp, opt2fn("-p", NFILE, fnm), perfdata, nnodes, ntprs, pmeentries,
2248 repeats, info, &best_tpr, &best_npme);
2250 /* Take the best-performing tpr file and enlarge nsteps to original value */
2251 if ( bKeepTPR && !bOverwrite )
2253 simulation_tpr = opt2fn("-s",NFILE,fnm);
2257 simulation_tpr = opt2fn("-so",NFILE,fnm);
2258 modify_PMEsettings(bOverwrite? (new_sim_nsteps+cpt_steps) : info->orig_sim_steps,
2259 info->orig_init_step, tpr_names[best_tpr], simulation_tpr);
2262 /* Let's get rid of the temporary benchmark input files */
2263 for (i=0; i<ntprs; i++)
2265 fprintf(stdout, "Deleting temporary benchmark input file %s\n", tpr_names[i]);
2266 remove(tpr_names[i]);
2269 /* Now start the real simulation if the user requested it ... */
2270 launch_simulation(bLaunch, fp, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2271 cmd_args_launch, simulation_tpr, nnodes, best_npme);
2275 /* ... or simply print the performance results to screen: */
2277 finalize(opt2fn("-p", NFILE, fnm));