2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
5 * Copyright (c) 2001-2008, The GROMACS development team,
6 * check out http://www.gromacs.org for more information.
7 * Copyright (c) 2012, by the GROMACS development team, led by
8 * David van der Spoel, Berk Hess, Erik Lindahl, and including many
9 * others, as listed in the AUTHORS file in the top-level source
10 * directory and at http://www.gromacs.org.
12 * GROMACS is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU Lesser General Public License
14 * as published by the Free Software Foundation; either version 2.1
15 * of the License, or (at your option) any later version.
17 * GROMACS is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
20 * Lesser General Public License for more details.
22 * You should have received a copy of the GNU Lesser General Public
23 * License along with GROMACS; if not, see
24 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
25 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
27 * If you want to redistribute modifications to GROMACS, please
28 * consider that scientific software is very special. Version
29 * control is crucial - bugs must be traceable. We will be happy to
30 * consider code for inclusion in the official distribution, but
31 * derived work must not be called official GROMACS. Details are found
32 * in the README & COPYING files - if they are missing, get the
33 * official version at http://www.gromacs.org.
35 * To help us fund GROMACS development, we humbly ask that you cite
36 * the research papers on the package. Check out http://www.gromacs.org.
44 #ifdef HAVE_SYS_TIME_H
60 #include "checkpoint.h"
67 /* Enum for situations that can occur during log file parsing, the
68 * corresponding string entries can be found in do_the_tests() in
69 * const char* ParseLog[] */
75 eParselogResetProblem,
86 int nPMEnodes; /* number of PME-only nodes used in this test */
87 int nx, ny, nz; /* DD grid */
88 int guessPME; /* if nPMEnodes == -1, this is the guessed number of PME nodes */
89 double *Gcycles; /* This can contain more than one value if doing multiple tests */
93 float *PME_f_load; /* PME mesh/force load average*/
94 float PME_f_load_Av; /* Average average ;) ... */
95 char *mdrun_cmd_line; /* Mdrun command line used for this test */
101 int nr_inputfiles; /* The number of tpr and mdp input files */
102 gmx_large_int_t orig_sim_steps; /* Number of steps to be done in the real simulation */
103 gmx_large_int_t orig_init_step; /* Init step for the real simulation */
104 real *rcoulomb; /* The coulomb radii [0...nr_inputfiles] */
105 real *rvdw; /* The vdW radii */
106 real *rlist; /* Neighbourlist cutoff radius */
108 int *nkx, *nky, *nkz;
109 real *fsx, *fsy, *fsz; /* Fourierspacing in x,y,z dimension */
113 static void sep_line(FILE *fp)
115 fprintf(fp, "\n------------------------------------------------------------\n");
119 /* Wrapper for system calls */
120 static int gmx_system_call(char *command)
123 gmx_fatal(FARGS,"No calls to system(3) supported on this platform. Attempted to call:\n'%s'\n",command);
125 return ( system(command) );
130 /* Check if string starts with substring */
131 static gmx_bool str_starts(const char *string, const char *substring)
133 return ( strncmp(string, substring, strlen(substring)) == 0);
137 static void cleandata(t_perf *perfdata, int test_nr)
139 perfdata->Gcycles[test_nr] = 0.0;
140 perfdata->ns_per_day[test_nr] = 0.0;
141 perfdata->PME_f_load[test_nr] = 0.0;
147 static gmx_bool is_equal(real a, real b)
149 real diff, eps=1.0e-7;
154 if (diff < 0.0) diff = -diff;
163 static void finalize(const char *fn_out)
169 fp = fopen(fn_out,"r");
170 fprintf(stdout,"\n\n");
172 while( fgets(buf,STRLEN-1,fp) != NULL )
174 fprintf(stdout,"%s",buf);
177 fprintf(stdout,"\n\n");
181 enum {eFoundNothing, eFoundDDStr, eFoundAccountingStr, eFoundCycleStr};
183 static int parse_logfile(const char *logfile, const char *errfile,
184 t_perf *perfdata, int test_nr, int presteps, gmx_large_int_t cpt_steps,
188 char line[STRLEN], dumstring[STRLEN], dumstring2[STRLEN];
189 const char matchstrdd[]="Domain decomposition grid";
190 const char matchstrcr[]="resetting all time and cycle counters";
191 const char matchstrbal[]="Average PME mesh/force load:";
192 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";
193 const char errSIG[]="signal, stopping at the next";
196 float dum1,dum2,dum3,dum4;
199 gmx_large_int_t resetsteps=-1;
200 gmx_bool bFoundResetStr = FALSE;
201 gmx_bool bResetChecked = FALSE;
204 if (!gmx_fexist(logfile))
206 fprintf(stderr, "WARNING: Could not find logfile %s.\n", logfile);
207 cleandata(perfdata, test_nr);
208 return eParselogNotFound;
211 fp = fopen(logfile, "r");
212 perfdata->PME_f_load[test_nr] = -1.0;
213 perfdata->guessPME = -1;
215 iFound = eFoundNothing;
217 iFound = eFoundDDStr; /* Skip some case statements */
219 while (fgets(line, STRLEN, fp) != NULL)
221 /* Remove leading spaces */
224 /* Check for TERM and INT signals from user: */
225 if ( strstr(line, errSIG) != NULL )
228 cleandata(perfdata, test_nr);
229 return eParselogTerm;
232 /* Check whether cycle resetting worked */
233 if (presteps > 0 && !bFoundResetStr)
235 if (strstr(line, matchstrcr) != NULL)
237 sprintf(dumstring, "step %s", gmx_large_int_pfmt);
238 sscanf(line, dumstring, &resetsteps);
239 bFoundResetStr = TRUE;
240 if (resetsteps == presteps+cpt_steps)
242 bResetChecked = TRUE;
246 sprintf(dumstring , gmx_large_int_pfmt, resetsteps);
247 sprintf(dumstring2, gmx_large_int_pfmt, presteps+cpt_steps);
248 fprintf(stderr, "WARNING: Time step counters were reset at step %s,\n"
249 " though they were supposed to be reset at step %s!\n",
250 dumstring, dumstring2);
255 /* Look for strings that appear in a certain order in the log file: */
259 /* Look for domain decomp grid and separate PME nodes: */
260 if (str_starts(line, matchstrdd))
262 sscanf(line, "Domain decomposition grid %d x %d x %d, separate PME nodes %d",
263 &(perfdata->nx), &(perfdata->ny), &(perfdata->nz), &npme);
264 if (perfdata->nPMEnodes == -1)
265 perfdata->guessPME = npme;
266 else if (perfdata->nPMEnodes != npme)
267 gmx_fatal(FARGS, "PME nodes from command line and output file are not identical");
268 iFound = eFoundDDStr;
270 /* Catch a few errors that might have occured: */
271 else if (str_starts(line, "There is no domain decomposition for"))
274 return eParselogNoDDGrid;
276 else if (str_starts(line, "reading tpx file"))
279 return eParselogTPXVersion;
281 else if (str_starts(line, "The -dd or -npme option request a parallel simulation"))
284 return eParselogNotParallel;
288 /* Look for PME mesh/force balance (not necessarily present, though) */
289 if (str_starts(line, matchstrbal))
290 sscanf(&line[strlen(matchstrbal)], "%f", &(perfdata->PME_f_load[test_nr]));
291 /* Look for matchstring */
292 if (str_starts(line, matchstring))
293 iFound = eFoundAccountingStr;
295 case eFoundAccountingStr:
296 /* Already found matchstring - look for cycle data */
297 if (str_starts(line, "Total "))
299 sscanf(line,"Total %d %lf",&procs,&(perfdata->Gcycles[test_nr]));
300 iFound = eFoundCycleStr;
304 /* Already found cycle data - look for remaining performance info and return */
305 if (str_starts(line, "Performance:"))
307 ndum = sscanf(line,"%s %f %f %f %f", dumstring, &dum1, &dum2, &dum3, &dum4);
308 /* (ns/day) is the second last entry, depending on whether GMX_DETAILED_PERF_STATS was set in print_perf(), nrnb.c */
309 perfdata->ns_per_day[test_nr] = (ndum==5)? dum3 : dum1;
311 if (bResetChecked || presteps == 0)
314 return eParselogResetProblem;
320 /* Close the log file */
323 /* Check why there is no performance data in the log file.
324 * Did a fatal errors occur? */
325 if (gmx_fexist(errfile))
327 fp = fopen(errfile, "r");
328 while (fgets(line, STRLEN, fp) != NULL)
330 if ( str_starts(line, "Fatal error:") )
332 if (fgets(line, STRLEN, fp) != NULL)
333 fprintf(stderr, "\nWARNING: An error occured during this benchmark:\n"
336 cleandata(perfdata, test_nr);
337 return eParselogFatal;
344 fprintf(stderr, "WARNING: Could not find stderr file %s.\n", errfile);
347 /* Giving up ... we could not find out why there is no performance data in
349 fprintf(stdout, "No performance data in log file.\n");
350 cleandata(perfdata, test_nr);
352 return eParselogNoPerfData;
356 static gmx_bool analyze_data(
365 int *index_tpr, /* OUT: Nr of mdp file with best settings */
366 int *npme_optimal) /* OUT: Optimal number of PME nodes */
369 int line=0, line_win=-1;
370 int k_win=-1, i_win=-1, winPME;
371 double s=0.0; /* standard deviation */
374 char str_PME_f_load[13];
375 gmx_bool bCanUseOrigTPR;
376 gmx_bool bRefinedCoul, bRefinedVdW, bRefinedGrid;
382 fprintf(fp, "Summary of successful runs:\n");
383 fprintf(fp, "Line tpr PME nodes Gcycles Av. Std.dev. ns/day PME/f");
385 fprintf(fp, " DD grid");
390 for (k=0; k<ntprs; k++)
392 for (i=0; i<ntests; i++)
394 /* Select the right dataset: */
395 pd = &(perfdata[k][i]);
397 pd->Gcycles_Av = 0.0;
398 pd->PME_f_load_Av = 0.0;
399 pd->ns_per_day_Av = 0.0;
401 if (pd->nPMEnodes == -1)
402 sprintf(strbuf, "(%3d)", pd->guessPME);
404 sprintf(strbuf, " ");
406 /* Get the average run time of a setting */
407 for (j=0; j<nrepeats; j++)
409 pd->Gcycles_Av += pd->Gcycles[j];
410 pd->PME_f_load_Av += pd->PME_f_load[j];
412 pd->Gcycles_Av /= nrepeats;
413 pd->PME_f_load_Av /= nrepeats;
415 for (j=0; j<nrepeats; j++)
417 if (pd->ns_per_day[j] > 0.0)
418 pd->ns_per_day_Av += pd->ns_per_day[j];
421 /* Somehow the performance number was not aquired for this run,
422 * therefor set the average to some negative value: */
423 pd->ns_per_day_Av = -1.0f*nrepeats;
427 pd->ns_per_day_Av /= nrepeats;
430 if (pd->PME_f_load_Av > 0.0)
431 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load_Av);
433 sprintf(str_PME_f_load, "%s", " - ");
436 /* We assume we had a successful run if both averages are positive */
437 if (pd->Gcycles_Av > 0.0 && pd->ns_per_day_Av > 0.0)
439 /* Output statistics if repeats were done */
442 /* Calculate the standard deviation */
444 for (j=0; j<nrepeats; j++)
445 s += pow( pd->Gcycles[j] - pd->Gcycles_Av, 2 );
449 fprintf(fp, "%4d %3d %4d%s %12.3f %12.3f %12.3f %s",
450 line, k, pd->nPMEnodes, strbuf, pd->Gcycles_Av, s,
451 pd->ns_per_day_Av, str_PME_f_load);
453 fprintf(fp, " %3d %3d %3d", pd->nx, pd->ny, pd->nz);
456 /* Store the index of the best run found so far in 'winner': */
457 if ( (k_win == -1) || (pd->Gcycles_Av < perfdata[k_win][i_win].Gcycles_Av) )
469 gmx_fatal(FARGS, "None of the runs was successful! Check %s for problems.", fn);
473 winPME = perfdata[k_win][i_win].nPMEnodes;
477 /* We stuck to a fixed number of PME-only nodes */
478 sprintf(strbuf, "settings No. %d", k_win);
482 /* We have optimized the number of PME-only nodes */
484 sprintf(strbuf, "%s", "the automatic number of PME nodes");
486 sprintf(strbuf, "%d PME nodes", winPME);
488 fprintf(fp, "Best performance was achieved with %s", strbuf);
489 if ((nrepeats > 1) && (ntests > 1))
490 fprintf(fp, " (see line %d)", line_win);
493 /* Only mention settings if they were modified: */
494 bRefinedCoul = !is_equal(info->rcoulomb[k_win], info->rcoulomb[0]);
495 bRefinedVdW = !is_equal(info->rvdw[k_win] , info->rvdw[0] );
496 bRefinedGrid = !(info->nkx[k_win] == info->nkx[0] &&
497 info->nky[k_win] == info->nky[0] &&
498 info->nkz[k_win] == info->nkz[0]);
500 if (bRefinedCoul || bRefinedVdW || bRefinedGrid)
502 fprintf(fp, "Optimized PME settings:\n");
503 bCanUseOrigTPR = FALSE;
507 bCanUseOrigTPR = TRUE;
511 fprintf(fp, " New Coulomb radius: %f nm (was %f nm)\n", info->rcoulomb[k_win], info->rcoulomb[0]);
514 fprintf(fp, " New Van der Waals radius: %f nm (was %f nm)\n", info->rvdw[k_win], info->rvdw[0]);
517 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],
518 info->nkx[0], info->nky[0], info->nkz[0]);
520 if (bCanUseOrigTPR && ntprs > 1)
521 fprintf(fp, "and original PME settings.\n");
525 /* Return the index of the mdp file that showed the highest performance
526 * and the optimal number of PME nodes */
528 *npme_optimal = winPME;
530 return bCanUseOrigTPR;
534 /* Get the commands we need to set up the runs from environment variables */
535 static void get_program_paths(gmx_bool bThreads, char *cmd_mpirun[], char cmd_np[],
536 char *cmd_mdrun[], int repeats)
543 const char def_mpirun[] = "mpirun";
544 const char def_mdrun[] = "mdrun";
545 const char filename[] = "benchtest.log";
546 const char match_mpi[] = "NNODES=";
547 const char match_mdrun[]= "Program: ";
548 const char empty_mpirun[] = "";
549 gmx_bool bMdrun = FALSE;
550 gmx_bool bMPI = FALSE;
553 /* Get the commands we need to set up the runs from environment variables */
556 if ( (cp = getenv("MPIRUN")) != NULL)
557 *cmd_mpirun = strdup(cp);
559 *cmd_mpirun = strdup(def_mpirun);
563 *cmd_mpirun = strdup(empty_mpirun);
566 if ( (cp = getenv("MDRUN" )) != NULL )
567 *cmd_mdrun = strdup(cp);
569 *cmd_mdrun = strdup(def_mdrun);
572 /* If no simulations have to be performed, we are done here */
576 /* Run a small test to see whether mpirun + mdrun work */
577 fprintf(stdout, "Making sure that mdrun can be executed. ");
580 snew(command, strlen(*cmd_mdrun) + strlen(cmd_np) + strlen(filename) + 50);
581 sprintf(command, "%s%s-version -maxh 0.001 1> %s 2>&1", *cmd_mdrun, cmd_np, filename);
585 snew(command, strlen(*cmd_mpirun) + strlen(cmd_np) + strlen(*cmd_mdrun) + strlen(filename) + 50);
586 sprintf(command, "%s%s%s -version -maxh 0.001 1> %s 2>&1", *cmd_mpirun, cmd_np, *cmd_mdrun, filename);
588 fprintf(stdout, "Trying '%s' ... ", command);
589 make_backup(filename);
590 gmx_system_call(command);
592 /* Check if we find the characteristic string in the output: */
593 if (!gmx_fexist(filename))
594 gmx_fatal(FARGS, "Output from test run could not be found.");
596 fp = fopen(filename, "r");
597 /* We need to scan the whole output file, since sometimes the queuing system
598 * also writes stuff to stdout/err */
601 cp2=fgets(line, STRLEN, fp);
604 if ( str_starts(line, match_mdrun) )
606 if ( str_starts(line, match_mpi) )
616 gmx_fatal(FARGS, "Need a threaded version of mdrun. This one\n"
618 "seems to have been compiled with MPI instead.",
626 gmx_fatal(FARGS, "Need an MPI-enabled version of mdrun. This one\n"
628 "seems to have been compiled without MPI support.",
635 gmx_fatal(FARGS, "Cannot execute mdrun. Please check %s for problems!",
639 fprintf(stdout, "passed.\n");
647 static void launch_simulation(
648 gmx_bool bLaunch, /* Should the simulation be launched? */
649 FILE *fp, /* General log file */
650 gmx_bool bThreads, /* whether to use threads */
651 char *cmd_mpirun, /* Command for mpirun */
652 char *cmd_np, /* Switch for -np or -nt or empty */
653 char *cmd_mdrun, /* Command for mdrun */
654 char *args_for_mdrun, /* Arguments for mdrun */
655 const char *simulation_tpr, /* This tpr will be simulated */
656 int nnodes, /* Number of nodes to run on */
657 int nPMEnodes) /* Number of PME nodes to use */
662 /* Make enough space for the system call command,
663 * (100 extra chars for -npme ... etc. options should suffice): */
664 snew(command, strlen(cmd_mpirun)+strlen(cmd_mdrun)+strlen(cmd_np)+strlen(args_for_mdrun)+strlen(simulation_tpr)+100);
666 /* Note that the -passall options requires args_for_mdrun to be at the end
667 * of the command line string */
670 sprintf(command, "%s%s-npme %d -s %s %s",
671 cmd_mdrun, cmd_np, nPMEnodes, simulation_tpr, args_for_mdrun);
675 sprintf(command, "%s%s%s -npme %d -s %s %s",
676 cmd_mpirun, cmd_np, cmd_mdrun, nPMEnodes, simulation_tpr, args_for_mdrun);
679 fprintf(fp, "%s this command line to launch the simulation:\n\n%s", bLaunch? "Using":"Please use", command);
683 /* Now the real thing! */
686 fprintf(stdout, "\nLaunching simulation with best parameters now.\nExecuting '%s'", command);
689 gmx_system_call(command);
694 static void modify_PMEsettings(
695 gmx_large_int_t simsteps, /* Set this value as number of time steps */
696 gmx_large_int_t init_step, /* Set this value as init_step */
697 const char *fn_best_tpr, /* tpr file with the best performance */
698 const char *fn_sim_tpr) /* name of tpr file to be launched */
706 read_tpx_state(fn_best_tpr,ir,&state,NULL,&mtop);
708 /* Reset nsteps and init_step to the value of the input .tpr file */
709 ir->nsteps = simsteps;
710 ir->init_step = init_step;
712 /* Write the tpr file which will be launched */
713 sprintf(buf, "Writing optimized simulation file %s with nsteps=%s.\n", fn_sim_tpr, gmx_large_int_pfmt);
714 fprintf(stdout,buf,ir->nsteps);
716 write_tpx_state(fn_sim_tpr,ir,&state,&mtop);
722 #define EPME_SWITCHED(e) ((e) == eelPMESWITCH || (e) == eelPMEUSERSWITCH)
724 /* Make additional TPR files with more computational load for the
725 * direct space processors: */
726 static void make_benchmark_tprs(
727 const char *fn_sim_tpr, /* READ : User-provided tpr file */
728 char *fn_bench_tprs[], /* WRITE: Names of benchmark tpr files */
729 gmx_large_int_t benchsteps, /* Number of time steps for benchmark runs */
730 gmx_large_int_t statesteps, /* Step counter in checkpoint file */
731 real rmin, /* Minimal Coulomb radius */
732 real rmax, /* Maximal Coulomb radius */
733 real bScaleRvdw, /* Scale rvdw along with rcoulomb */
734 int *ntprs, /* No. of TPRs to write, each with a different
735 rcoulomb and fourierspacing */
736 t_inputinfo *info, /* Contains information about mdp file options */
737 FILE *fp) /* Write the output here */
743 real nlist_buffer; /* Thickness of the buffer regions for PME-switch potentials */
746 gmx_bool bNote = FALSE;
747 real add; /* Add this to rcoul for the next test */
748 real fac = 1.0; /* Scaling factor for Coulomb radius */
749 real fourierspacing; /* Basic fourierspacing from tpr */
752 sprintf(buf, "Making benchmark tpr file%s with %s time step%s",
753 *ntprs > 1? "s":"", gmx_large_int_pfmt, benchsteps>1?"s":"");
754 fprintf(stdout, buf, benchsteps);
757 sprintf(buf, " (adding %s steps from checkpoint file)", gmx_large_int_pfmt);
758 fprintf(stdout, buf, statesteps);
759 benchsteps += statesteps;
761 fprintf(stdout, ".\n");
765 read_tpx_state(fn_sim_tpr,ir,&state,NULL,&mtop);
767 /* Check if some kind of PME was chosen */
768 if (EEL_PME(ir->coulombtype) == FALSE)
769 gmx_fatal(FARGS, "Can only do optimizations for simulations with %s electrostatics.",
772 /* Check if rcoulomb == rlist, which is necessary for plain PME. */
773 if ( (ir->cutoff_scheme != ecutsVERLET) &&
774 (eelPME == ir->coulombtype) && !(ir->rcoulomb == ir->rlist))
776 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to rlist (%f).",
777 EELTYPE(eelPME), ir->rcoulomb, ir->rlist);
779 /* For other PME types, rcoulomb is allowed to be smaller than rlist */
780 else if (ir->rcoulomb > ir->rlist)
782 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to or smaller than rlist (%f)",
783 EELTYPE(ir->coulombtype), ir->rcoulomb, ir->rlist);
786 if (bScaleRvdw && ir->rvdw != ir->rcoulomb)
788 fprintf(stdout,"NOTE: input rvdw != rcoulomb, will not scale rvdw\n");
792 /* Reduce the number of steps for the benchmarks */
793 info->orig_sim_steps = ir->nsteps;
794 ir->nsteps = benchsteps;
795 /* We must not use init_step from the input tpr file for the benchmarks */
796 info->orig_init_step = ir->init_step;
799 /* For PME-switch potentials, keep the radial distance of the buffer region */
800 nlist_buffer = ir->rlist - ir->rcoulomb;
802 /* Determine length of triclinic box vectors */
807 box_size[d] += state.box[d][i]*state.box[d][i];
808 box_size[d] = sqrt(box_size[d]);
811 if (ir->fourier_spacing > 0)
813 info->fsx[0] = ir->fourier_spacing;
814 info->fsy[0] = ir->fourier_spacing;
815 info->fsz[0] = ir->fourier_spacing;
819 /* Reconstruct fourierspacing per dimension from the number of grid points and box size */
820 info->fsx[0] = box_size[XX]/ir->nkx;
821 info->fsy[0] = box_size[YY]/ir->nky;
822 info->fsz[0] = box_size[ZZ]/ir->nkz;
825 /* If no value for the fourierspacing was provided on the command line, we
826 * use the reconstruction from the tpr file */
827 if (ir->fourier_spacing > 0)
829 /* Use the spacing from the tpr */
830 fourierspacing = ir->fourier_spacing;
834 /* Use the maximum observed spacing */
835 fourierspacing = max(max(info->fsx[0],info->fsy[0]),info->fsz[0]);
838 fprintf(stdout, "Calculating PME grid points on the basis of a fourierspacing of %f nm\n", fourierspacing);
840 /* For performance comparisons the number of particles is useful to have */
841 fprintf(fp, " Number of particles : %d\n", mtop.natoms);
843 /* Print information about settings of which some are potentially modified: */
844 fprintf(fp, " Coulomb type : %s\n", EELTYPE(ir->coulombtype));
845 fprintf(fp, " Grid spacing x y z : %f %f %f\n",
846 box_size[XX]/ir->nkx, box_size[YY]/ir->nky, box_size[ZZ]/ir->nkz);
847 fprintf(fp, " Van der Waals type : %s\n", EVDWTYPE(ir->vdwtype));
848 if (EVDW_SWITCHED(ir->vdwtype))
849 fprintf(fp, " rvdw_switch : %f nm\n", ir->rvdw_switch);
850 if (EPME_SWITCHED(ir->coulombtype))
851 fprintf(fp, " rlist : %f nm\n", ir->rlist);
852 if (ir->rlistlong != max_cutoff(ir->rvdw,ir->rcoulomb))
853 fprintf(fp, " rlistlong : %f nm\n", ir->rlistlong);
855 /* Print a descriptive line about the tpr settings tested */
856 fprintf(fp, "\nWill try these real/reciprocal workload settings:\n");
857 fprintf(fp, " No. scaling rcoulomb");
858 fprintf(fp, " nkx nky nkz");
859 fprintf(fp, " spacing");
860 if (evdwCUT == ir->vdwtype)
861 fprintf(fp, " rvdw");
862 if (EPME_SWITCHED(ir->coulombtype))
863 fprintf(fp, " rlist");
864 if ( ir->rlistlong != max_cutoff(ir->rlist,max_cutoff(ir->rvdw,ir->rcoulomb)) )
865 fprintf(fp, " rlistlong");
866 fprintf(fp, " tpr file\n");
868 /* Loop to create the requested number of tpr input files */
869 for (j = 0; j < *ntprs; j++)
871 /* The first .tpr is the provided one, just need to modify nsteps,
872 * so skip the following block */
875 /* Determine which Coulomb radii rc to use in the benchmarks */
876 add = (rmax-rmin)/(*ntprs-1);
877 if (is_equal(rmin,info->rcoulomb[0]))
879 ir->rcoulomb = rmin + j*add;
881 else if (is_equal(rmax,info->rcoulomb[0]))
883 ir->rcoulomb = rmin + (j-1)*add;
887 /* rmin != rcoul != rmax, ergo test between rmin and rmax */
888 add = (rmax-rmin)/(*ntprs-2);
889 ir->rcoulomb = rmin + (j-1)*add;
892 /* Determine the scaling factor fac */
893 fac = ir->rcoulomb/info->rcoulomb[0];
895 /* Scale the Fourier grid spacing */
896 ir->nkx = ir->nky = ir->nkz = 0;
897 calc_grid(NULL,state.box,fourierspacing*fac,&ir->nkx,&ir->nky,&ir->nkz);
899 /* Adjust other radii since various conditions neet to be fulfilled */
900 if (eelPME == ir->coulombtype)
902 /* plain PME, rcoulomb must be equal to rlist */
903 ir->rlist = ir->rcoulomb;
907 /* rlist must be >= rcoulomb, we keep the size of the buffer region */
908 ir->rlist = ir->rcoulomb + nlist_buffer;
911 if (bScaleRvdw && evdwCUT == ir->vdwtype)
913 /* For vdw cutoff, rvdw >= rlist */
914 ir->rvdw = max(info->rvdw[0], ir->rlist);
917 ir->rlistlong = max_cutoff(ir->rlist,max_cutoff(ir->rvdw,ir->rcoulomb));
919 } /* end of "if (j != 0)" */
921 /* for j==0: Save the original settings
922 * for j >0: Save modified radii and Fourier grids */
923 info->rcoulomb[j] = ir->rcoulomb;
924 info->rvdw[j] = ir->rvdw;
925 info->nkx[j] = ir->nkx;
926 info->nky[j] = ir->nky;
927 info->nkz[j] = ir->nkz;
928 info->rlist[j] = ir->rlist;
929 info->rlistlong[j] = ir->rlistlong;
930 info->fsx[j] = fac*fourierspacing;
931 info->fsy[j] = fac*fourierspacing;
932 info->fsz[j] = fac*fourierspacing;
934 /* Write the benchmark tpr file */
935 strncpy(fn_bench_tprs[j],fn_sim_tpr,strlen(fn_sim_tpr)-strlen(".tpr"));
936 sprintf(buf, "_bench%.2d.tpr", j);
937 strcat(fn_bench_tprs[j], buf);
938 fprintf(stdout,"Writing benchmark tpr %s with nsteps=", fn_bench_tprs[j]);
939 fprintf(stdout, gmx_large_int_pfmt, ir->nsteps);
941 fprintf(stdout,", scaling factor %f\n", fac);
943 fprintf(stdout,", unmodified settings\n");
945 write_tpx_state(fn_bench_tprs[j],ir,&state,&mtop);
947 /* Write information about modified tpr settings to log file */
948 fprintf(fp, "%4d%10f%10f", j, fac, ir->rcoulomb);
949 fprintf(fp, "%5d%5d%5d", ir->nkx, ir->nky, ir->nkz);
950 fprintf(fp, " %9f ", info->fsx[j]);
951 if (evdwCUT == ir->vdwtype)
952 fprintf(fp, "%10f", ir->rvdw);
953 if (EPME_SWITCHED(ir->coulombtype))
954 fprintf(fp, "%10f", ir->rlist);
955 if ( info->rlistlong[0] != max_cutoff(info->rlist[0],max_cutoff(info->rvdw[0],info->rcoulomb[0])) )
956 fprintf(fp, "%10f", ir->rlistlong);
957 fprintf(fp, " %-14s\n",fn_bench_tprs[j]);
959 /* Make it clear to the user that some additional settings were modified */
960 if ( !is_equal(ir->rvdw , info->rvdw[0])
961 || !is_equal(ir->rlistlong, info->rlistlong[0]) )
967 fprintf(fp, "\nNote that in addition to the Coulomb radius and the Fourier grid\n"
968 "other input settings were also changed (see table above).\n"
969 "Please check if the modified settings are appropriate.\n");
976 /* Rename the files we want to keep to some meaningful filename and
978 static void cleanup(const t_filenm *fnm, int nfile, int k, int nnodes,
979 int nPMEnodes, int nr, gmx_bool bKeepStderr)
981 char numstring[STRLEN];
982 char newfilename[STRLEN];
988 fprintf(stdout, "Cleaning up, deleting benchmark temp files ...\n");
990 for (i=0; i<nfile; i++)
992 opt = (char *)fnm[i].opt;
993 if ( strcmp(opt, "-p") == 0 )
995 /* do nothing; keep this file */
998 else if (strcmp(opt, "-bg") == 0)
1000 /* Give the log file a nice name so one can later see which parameters were used */
1001 numstring[0] = '\0';
1003 sprintf(numstring, "_%d", nr);
1004 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", opt2fn("-bg",nfile,fnm), k, nnodes, nPMEnodes, numstring);
1005 if (gmx_fexist(opt2fn("-bg",nfile,fnm)))
1007 fprintf(stdout, "renaming log file to %s\n", newfilename);
1008 make_backup(newfilename);
1009 rename(opt2fn("-bg",nfile,fnm), newfilename);
1012 else if (strcmp(opt, "-err") == 0)
1014 /* This file contains the output of stderr. We want to keep it in
1015 * cases where there have been problems. */
1016 fn = opt2fn(opt, nfile, fnm);
1017 numstring[0] = '\0';
1019 sprintf(numstring, "_%d", nr);
1020 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", fn, k, nnodes, nPMEnodes, numstring);
1025 fprintf(stdout, "Saving stderr output in %s\n", newfilename);
1026 make_backup(newfilename);
1027 rename(fn, newfilename);
1031 fprintf(stdout, "Deleting %s\n", fn);
1036 /* Delete the files which are created for each benchmark run: (options -b*) */
1037 else if ( (0 == strncmp(opt, "-b", 2)) && (opt2bSet(opt,nfile,fnm) || !is_optional(&fnm[i])) )
1039 fn = opt2fn(opt, nfile, fnm);
1042 fprintf(stdout, "Deleting %s\n", fn);
1050 /* Returns the largest common factor of n1 and n2 */
1051 static int largest_common_factor(int n1, int n2)
1056 for (factor=nmax; factor > 0; factor--)
1058 if ( 0==(n1 % factor) && 0==(n2 % factor) )
1063 return 0; /* one for the compiler */
1066 enum {eNpmeAuto, eNpmeAll, eNpmeReduced, eNpmeSubset, eNpmeNr};
1068 /* Create a list of numbers of PME nodes to test */
1069 static void make_npme_list(
1070 const char *npmevalues_opt, /* Make a complete list with all
1071 * possibilities or a short list that keeps only
1072 * reasonable numbers of PME nodes */
1073 int *nentries, /* Number of entries we put in the nPMEnodes list */
1074 int *nPMEnodes[], /* Each entry contains the value for -npme */
1075 int nnodes, /* Total number of nodes to do the tests on */
1076 int minPMEnodes, /* Minimum number of PME nodes */
1077 int maxPMEnodes) /* Maximum number of PME nodes */
1080 int min_factor=1; /* We request that npp and npme have this minimal
1081 * largest common factor (depends on npp) */
1082 int nlistmax; /* Max. list size */
1083 int nlist; /* Actual number of entries in list */
1087 /* Do we need to check all possible values for -npme or is a reduced list enough? */
1088 if ( 0 == strcmp(npmevalues_opt, "all") )
1092 else if ( 0 == strcmp(npmevalues_opt, "subset") )
1094 eNPME = eNpmeSubset;
1096 else /* "auto" or "range" */
1100 else if (nnodes < 128)
1101 eNPME = eNpmeReduced;
1103 eNPME = eNpmeSubset;
1106 /* Calculate how many entries we could possibly have (in case of -npme all) */
1109 nlistmax = maxPMEnodes - minPMEnodes + 3;
1110 if (0 == minPMEnodes)
1116 /* Now make the actual list which is at most of size nlist */
1117 snew(*nPMEnodes, nlistmax);
1118 nlist = 0; /* start counting again, now the real entries in the list */
1119 for (i = 0; i < nlistmax - 2; i++)
1121 npme = maxPMEnodes - i;
1132 /* For 2d PME we want a common largest factor of at least the cube
1133 * root of the number of PP nodes */
1134 min_factor = (int) pow(npp, 1.0/3.0);
1137 gmx_fatal(FARGS, "Unknown option for eNPME in make_npme_list");
1140 if (largest_common_factor(npp, npme) >= min_factor)
1142 (*nPMEnodes)[nlist] = npme;
1146 /* We always test 0 PME nodes and the automatic number */
1147 *nentries = nlist + 2;
1148 (*nPMEnodes)[nlist ] = 0;
1149 (*nPMEnodes)[nlist+1] = -1;
1151 fprintf(stderr, "Will try the following %d different values for -npme:\n", *nentries);
1152 for (i=0; i<*nentries-1; i++)
1153 fprintf(stderr, "%d, ", (*nPMEnodes)[i]);
1154 fprintf(stderr, "and %d (auto).\n", (*nPMEnodes)[*nentries-1]);
1158 /* Allocate memory to store the performance data */
1159 static void init_perfdata(t_perf *perfdata[], int ntprs, int datasets, int repeats)
1164 for (k=0; k<ntprs; k++)
1166 snew(perfdata[k], datasets);
1167 for (i=0; i<datasets; i++)
1169 for (j=0; j<repeats; j++)
1171 snew(perfdata[k][i].Gcycles , repeats);
1172 snew(perfdata[k][i].ns_per_day, repeats);
1173 snew(perfdata[k][i].PME_f_load, repeats);
1180 /* Check for errors on mdrun -h */
1181 static void make_sure_it_runs(char *mdrun_cmd_line, int length, FILE *fp)
1183 char *command, *msg;
1187 snew(command, length + 15);
1188 snew(msg , length + 500);
1190 fprintf(stdout, "Making shure the benchmarks can be executed ...\n");
1191 sprintf(command, "%s-h -quiet", mdrun_cmd_line);
1192 ret = gmx_system_call(command);
1196 /* To prevent confusion, do not again issue a gmx_fatal here since we already
1197 * get the error message from mdrun itself */
1198 sprintf(msg, "Cannot run the benchmark simulations! Please check the error message of\n"
1199 "mdrun for the source of the problem. Did you provide a command line\n"
1200 "argument that neither g_tune_pme nor mdrun understands? Offending command:\n"
1201 "\n%s\n\n", command);
1203 fprintf(stderr, "%s", msg);
1205 fprintf(fp , "%s", msg);
1215 static void do_the_tests(
1216 FILE *fp, /* General g_tune_pme output file */
1217 char **tpr_names, /* Filenames of the input files to test */
1218 int maxPMEnodes, /* Max fraction of nodes to use for PME */
1219 int minPMEnodes, /* Min fraction of nodes to use for PME */
1220 int npme_fixed, /* If >= -1, test fixed number of PME
1222 const char *npmevalues_opt, /* Which -npme values should be tested */
1223 t_perf **perfdata, /* Here the performace data is stored */
1224 int *pmeentries, /* Entries in the nPMEnodes list */
1225 int repeats, /* Repeat each test this often */
1226 int nnodes, /* Total number of nodes = nPP + nPME */
1227 int nr_tprs, /* Total number of tpr files to test */
1228 gmx_bool bThreads, /* Threads or MPI? */
1229 char *cmd_mpirun, /* mpirun command string */
1230 char *cmd_np, /* "-np", "-n", whatever mpirun needs */
1231 char *cmd_mdrun, /* mdrun command string */
1232 char *cmd_args_bench, /* arguments for mdrun in a string */
1233 const t_filenm *fnm, /* List of filenames from command line */
1234 int nfile, /* Number of files specified on the cmdl. */
1235 int sim_part, /* For checkpointing */
1236 int presteps, /* DLB equilibration steps, is checked */
1237 gmx_large_int_t cpt_steps) /* Time step counter in the checkpoint */
1239 int i,nr,k,ret,count=0,totaltests;
1240 int *nPMEnodes=NULL;
1243 char *command, *cmd_stub;
1245 gmx_bool bResetProblem=FALSE;
1246 gmx_bool bFirst=TRUE;
1249 /* This string array corresponds to the eParselog enum type at the start
1251 const char* ParseLog[] = {"OK.",
1252 "Logfile not found!",
1253 "No timings, logfile truncated?",
1254 "Run was terminated.",
1255 "Counters were not reset properly.",
1256 "No DD grid found for these settings.",
1257 "TPX version conflict!",
1258 "mdrun was not started in parallel!",
1259 "An error occured." };
1260 char str_PME_f_load[13];
1263 /* Allocate space for the mdrun command line. 100 extra characters should
1264 be more than enough for the -npme etcetera arguments */
1265 cmdline_length = strlen(cmd_mpirun)
1268 + strlen(cmd_args_bench)
1269 + strlen(tpr_names[0]) + 100;
1270 snew(command , cmdline_length);
1271 snew(cmd_stub, cmdline_length);
1273 /* Construct the part of the command line that stays the same for all tests: */
1276 sprintf(cmd_stub, "%s%s", cmd_mdrun, cmd_np);
1280 sprintf(cmd_stub, "%s%s%s ", cmd_mpirun, cmd_np, cmd_mdrun);
1283 /* Create a list of numbers of PME nodes to test */
1284 if (npme_fixed < -1)
1286 make_npme_list(npmevalues_opt, pmeentries, &nPMEnodes,
1287 nnodes, minPMEnodes, maxPMEnodes);
1293 nPMEnodes[0] = npme_fixed;
1294 fprintf(stderr, "Will use a fixed number of %d PME-only nodes.\n", nPMEnodes[0]);
1299 fprintf(fp, "\nNo benchmarks done since number of repeats (-r) is 0.\n");
1301 finalize(opt2fn("-p", nfile, fnm));
1305 /* Allocate one dataset for each tpr input file: */
1306 init_perfdata(perfdata, nr_tprs, *pmeentries, repeats);
1308 /*****************************************/
1309 /* Main loop over all tpr files to test: */
1310 /*****************************************/
1311 totaltests = nr_tprs*(*pmeentries)*repeats;
1312 for (k=0; k<nr_tprs;k++)
1314 fprintf(fp, "\nIndividual timings for input file %d (%s):\n", k, tpr_names[k]);
1315 fprintf(fp, "PME nodes Gcycles ns/day PME/f Remark\n");
1316 /* Loop over various numbers of PME nodes: */
1317 for (i = 0; i < *pmeentries; i++)
1319 pd = &perfdata[k][i];
1321 /* Loop over the repeats for each scenario: */
1322 for (nr = 0; nr < repeats; nr++)
1324 pd->nPMEnodes = nPMEnodes[i];
1326 /* Add -npme and -s to the command line and save it. Note that
1327 * the -passall (if set) options requires cmd_args_bench to be
1328 * at the end of the command line string */
1329 snew(pd->mdrun_cmd_line, cmdline_length);
1330 sprintf(pd->mdrun_cmd_line, "%s-npme %d -s %s %s",
1331 cmd_stub, pd->nPMEnodes, tpr_names[k], cmd_args_bench);
1333 /* To prevent that all benchmarks fail due to a show-stopper argument
1334 * on the mdrun command line, we make a quick check with mdrun -h first */
1336 make_sure_it_runs(pd->mdrun_cmd_line, cmdline_length, fp);
1339 /* Do a benchmark simulation: */
1341 sprintf(buf, ", pass %d/%d", nr+1, repeats);
1344 fprintf(stdout, "\n=== Progress %2.0f%%, tpr %d/%d, run %d/%d%s:\n",
1345 (100.0*count)/totaltests,
1346 k+1, nr_tprs, i+1, *pmeentries, buf);
1347 make_backup(opt2fn("-err",nfile,fnm));
1348 sprintf(command, "%s 1> /dev/null 2>%s", pd->mdrun_cmd_line, opt2fn("-err",nfile,fnm));
1349 fprintf(stdout, "%s\n", pd->mdrun_cmd_line);
1350 gmx_system_call(command);
1352 /* Collect the performance data from the log file; also check stderr
1353 * for fatal errors */
1354 ret = parse_logfile(opt2fn("-bg",nfile,fnm), opt2fn("-err",nfile,fnm),
1355 pd, nr, presteps, cpt_steps, nnodes);
1356 if ((presteps > 0) && (ret == eParselogResetProblem))
1357 bResetProblem = TRUE;
1359 if (-1 == pd->nPMEnodes)
1360 sprintf(buf, "(%3d)", pd->guessPME);
1365 if (pd->PME_f_load[nr] > 0.0)
1366 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load[nr]);
1368 sprintf(str_PME_f_load, "%s", " - ");
1370 /* Write the data we got to disk */
1371 fprintf(fp, "%4d%s %12.3f %12.3f %s %s", pd->nPMEnodes,
1372 buf, pd->Gcycles[nr], pd->ns_per_day[nr], str_PME_f_load, ParseLog[ret]);
1373 if (! (ret==eParselogOK || ret==eParselogNoDDGrid || ret==eParselogNotFound) )
1374 fprintf(fp, " Check %s file for problems.", ret==eParselogFatal? "err":"log");
1379 /* Do some cleaning up and delete the files we do not need any more */
1380 cleanup(fnm, nfile, k, nnodes, pd->nPMEnodes, nr, ret==eParselogFatal);
1382 /* If the first run with this number of processors already failed, do not try again: */
1383 if (pd->Gcycles[0] <= 0.0 && repeats > 1)
1385 fprintf(stdout, "Skipping remaining passes of unsuccessful setting, see log file for details.\n");
1386 count += repeats-(nr+1);
1389 } /* end of repeats loop */
1390 } /* end of -npme loop */
1391 } /* end of tpr file loop */
1396 fprintf(fp, "WARNING: The cycle and time step counters could not be reset properly. ");
1404 static void check_input(
1411 real maxPMEfraction,
1412 real minPMEfraction,
1414 gmx_large_int_t bench_nsteps,
1415 const t_filenm *fnm,
1425 /* Make sure the input file exists */
1426 if (!gmx_fexist(opt2fn("-s",nfile,fnm)))
1427 gmx_fatal(FARGS, "File %s not found.", opt2fn("-s",nfile,fnm));
1429 /* Make sure that the checkpoint file is not overwritten during benchmarking */
1430 if ( (0 == strcmp(opt2fn("-cpi",nfile,fnm), opt2fn("-bcpo",nfile,fnm)) ) && (sim_part > 1) )
1431 gmx_fatal(FARGS, "Checkpoint input (-cpi) and benchmark checkpoint output (-bcpo) files must not be identical.\n"
1432 "The checkpoint input file must not be overwritten during the benchmarks.\n");
1434 /* Make sure that repeats is >= 0 (if == 0, only write tpr files) */
1436 gmx_fatal(FARGS, "Number of repeats < 0!");
1438 /* Check number of nodes */
1440 gmx_fatal(FARGS, "Number of nodes/threads must be a positive integer.");
1442 /* Automatically choose -ntpr if not set */
1450 /* Set a reasonable scaling factor for rcoulomb */
1452 *rmax = rcoulomb * 1.2;
1454 fprintf(stderr, "Will test %d tpr file%s.\n", *ntprs, *ntprs==1?"":"s");
1459 fprintf(stderr, "Note: Choose ntpr>1 to shift PME load between real and reciprocal space.\n");
1462 /* Make shure that rmin <= rcoulomb <= rmax */
1463 if (*rmin <= 0) *rmin = rcoulomb;
1464 if (*rmax <= 0) *rmax = rcoulomb;
1465 if ( !(*rmin <= *rmax) )
1467 gmx_fatal(FARGS, "Please choose the Coulomb radii such that rmin <= rmax.\n"
1468 "rmin = %g, rmax = %g, actual rcoul from .tpr file = %g\n", *rmin, *rmax, rcoulomb);
1470 /* Add test scenarios if rmin or rmax were set */
1473 if ( !is_equal(*rmin, rcoulomb) && (*ntprs == 1) )
1476 fprintf(stderr, "NOTE: Setting -rmin to %g changed -ntpr to %d\n",
1479 if (!is_equal(*rmax, rcoulomb) && (*ntprs == 1) )
1482 fprintf(stderr, "NOTE: Setting -rmax to %g changed -ntpr to %d\n",
1487 /* If one of rmin, rmax is set, we need 2 tpr files at minimum */
1488 if ( !is_equal(*rmax, rcoulomb) || !is_equal(*rmin, rcoulomb) )
1489 *ntprs = max(*ntprs, 2);
1491 /* If both rmin, rmax are set, we need 3 tpr files at minimum */
1492 if ( !is_equal(*rmax, rcoulomb) && !is_equal(*rmin, rcoulomb) )
1493 *ntprs = max(*ntprs, 3);
1497 fprintf(stderr, "NOTE: Your rmin, rmax setting changed -ntpr to %d\n", *ntprs);
1502 if (is_equal(*rmin,rcoulomb) && is_equal(rcoulomb,*rmax)) /* We have just a single rc */
1504 fprintf(stderr, "WARNING: Resetting -ntpr to 1 since no Coulomb radius scaling is requested.\n"
1505 "Please set rmin < rmax to test Coulomb radii in the [rmin, rmax] interval\n"
1506 "with correspondingly adjusted PME grid settings\n");
1511 /* Check whether max and min fraction are within required values */
1512 if (maxPMEfraction > 0.5 || maxPMEfraction < 0)
1513 gmx_fatal(FARGS, "-max must be between 0 and 0.5");
1514 if (minPMEfraction > 0.5 || minPMEfraction < 0)
1515 gmx_fatal(FARGS, "-min must be between 0 and 0.5");
1516 if (maxPMEfraction < minPMEfraction)
1517 gmx_fatal(FARGS, "-max must be larger or equal to -min");
1519 /* Check whether the number of steps - if it was set - has a reasonable value */
1520 if (bench_nsteps < 0)
1521 gmx_fatal(FARGS, "Number of steps must be positive.");
1523 if (bench_nsteps > 10000 || bench_nsteps < 100)
1525 fprintf(stderr, "WARNING: steps=");
1526 fprintf(stderr, gmx_large_int_pfmt, bench_nsteps);
1527 fprintf(stderr, ". Are you sure you want to perform so %s steps for each benchmark?\n", (bench_nsteps < 100)? "few" : "many");
1532 gmx_fatal(FARGS, "Cannot have a negative number of presteps.\n");
1535 /* Check for rcoulomb scaling if more than one .tpr file is tested */
1538 if (*rmin/rcoulomb < 0.75 || *rmax/rcoulomb > 1.25)
1539 fprintf(stderr, "WARNING: Applying extreme scaling factor. I hope you know what you are doing.\n");
1542 /* If a fixed number of PME nodes is set we do rcoulomb and PME gird tuning
1543 * only. We need to check whether the requested number of PME-only nodes
1545 if (npme_fixed > -1)
1547 /* No more than 50% of all nodes can be assigned as PME-only nodes. */
1548 if (2*npme_fixed > nnodes)
1550 gmx_fatal(FARGS, "Cannot have more than %d PME-only nodes for a total of %d nodes (you chose %d).\n",
1551 nnodes/2, nnodes, npme_fixed);
1553 if ((npme_fixed > 0) && (5*npme_fixed < nnodes))
1555 fprintf(stderr, "WARNING: Only %g percent of the nodes are assigned as PME-only nodes.\n",
1556 100.0*((real)npme_fixed / (real)nnodes));
1558 if (opt2parg_bSet("-min",npargs,pa) || opt2parg_bSet("-max",npargs,pa))
1560 fprintf(stderr, "NOTE: The -min, -max, and -npme options have no effect when a\n"
1561 " fixed number of PME-only nodes is requested with -fix.\n");
1567 /* Returns TRUE when "opt" is needed at launch time */
1568 static gmx_bool is_launch_file(char *opt, gmx_bool bSet)
1570 /* Apart from the input .tpr we need all options that were set
1571 * on the command line and that do not start with -b */
1572 if (0 == strncmp(opt,"-b", 2) || 0 == strncmp(opt,"-s", 2))
1582 /* Returns TRUE when "opt" defines a file which is needed for the benchmarks runs */
1583 static gmx_bool is_bench_file(char *opt, gmx_bool bSet, gmx_bool bOptional, gmx_bool bIsOutput)
1585 /* Apart from the input .tpr, all files starting with "-b" are for
1586 * _b_enchmark files exclusively */
1587 if (0 == strncmp(opt,"-s", 2)) return FALSE;
1588 if (0 == strncmp(opt,"-b", 2) || 0 == strncmp(opt,"-s", 2))
1590 if (!bOptional || bSet)
1600 if (bSet) /* These are additional input files like -cpi -ei */
1608 /* Adds 'buf' to 'str' */
1609 static void add_to_string(char **str, char *buf)
1614 len = strlen(*str) + strlen(buf) + 1;
1620 /* Create the command line for the benchmark as well as for the real run */
1621 static void create_command_line_snippets(
1623 gmx_bool bAppendFiles,
1624 gmx_bool bKeepAndNumCPT,
1625 gmx_bool bResetHWay,
1631 const char *procstring, /* How to pass the number of processors to $MPIRUN */
1632 char *cmd_np[], /* Actual command line snippet, e.g. '-np <N>' */
1633 char *cmd_args_bench[], /* command line arguments for benchmark runs */
1634 char *cmd_args_launch[], /* command line arguments for simulation run */
1635 char extra_args[]) /* Add this to the end of the command line */
1640 char strbuf[STRLEN];
1643 /* strlen needs at least '\0' as a string: */
1644 snew(*cmd_args_bench ,1);
1645 snew(*cmd_args_launch,1);
1646 *cmd_args_launch[0]='\0';
1647 *cmd_args_bench[0] ='\0';
1650 /*******************************************/
1651 /* 1. Process other command line arguments */
1652 /*******************************************/
1655 /* Add equilibration steps to benchmark options */
1656 sprintf(strbuf, "-resetstep %d ", presteps);
1657 add_to_string(cmd_args_bench, strbuf);
1659 /* These switches take effect only at launch time */
1660 if (FALSE == bAppendFiles)
1662 add_to_string(cmd_args_launch, "-noappend ");
1666 add_to_string(cmd_args_launch, "-cpnum ");
1670 add_to_string(cmd_args_launch, "-resethway ");
1673 /********************/
1674 /* 2. Process files */
1675 /********************/
1676 for (i=0; i<nfile; i++)
1678 opt = (char *)fnm[i].opt;
1679 name = opt2fn(opt,nfile,fnm);
1681 /* Strbuf contains the options, now let's sort out where we need that */
1682 sprintf(strbuf, "%s %s ", opt, name);
1684 if ( is_bench_file(opt, opt2bSet(opt,nfile,fnm), is_optional(&fnm[i]), is_output(&fnm[i])) )
1686 /* All options starting with -b* need the 'b' removed,
1687 * therefore overwrite strbuf */
1688 if (0 == strncmp(opt, "-b", 2))
1689 sprintf(strbuf, "-%s %s ", &opt[2], name);
1691 add_to_string(cmd_args_bench, strbuf);
1694 if ( is_launch_file(opt,opt2bSet(opt,nfile,fnm)) )
1695 add_to_string(cmd_args_launch, strbuf);
1698 add_to_string(cmd_args_bench , extra_args);
1699 add_to_string(cmd_args_launch, extra_args);
1703 /* Set option opt */
1704 static void setopt(const char *opt,int nfile,t_filenm fnm[])
1708 for(i=0; (i<nfile); i++)
1709 if (strcmp(opt,fnm[i].opt)==0)
1710 fnm[i].flag |= ffSET;
1714 /* This routine inspects the tpr file and ...
1715 * 1. checks for output files that get triggered by a tpr option. These output
1716 * files are marked as 'set' to allow for a proper cleanup after each
1718 * 2. returns the PME:PP load ratio
1719 * 3. returns rcoulomb from the tpr */
1720 static float inspect_tpr(int nfile, t_filenm fnm[], real *rcoulomb)
1722 gmx_bool bPull; /* Is pulling requested in .tpr file? */
1723 gmx_bool bTpi; /* Is test particle insertion requested? */
1724 gmx_bool bFree; /* Is a free energy simulation requested? */
1725 gmx_bool bNM; /* Is a normal mode analysis requested? */
1731 /* Check tpr file for options that trigger extra output files */
1732 read_tpx_state(opt2fn("-s",nfile,fnm),&ir,&state,NULL,&mtop);
1733 bPull = (epullNO != ir.ePull);
1734 bFree = (efepNO != ir.efep );
1735 bNM = (eiNM == ir.eI );
1736 bTpi = EI_TPI(ir.eI);
1738 /* Set these output files on the tuning command-line */
1741 setopt("-pf" , nfile, fnm);
1742 setopt("-px" , nfile, fnm);
1746 setopt("-dhdl", nfile, fnm);
1750 setopt("-tpi" , nfile, fnm);
1751 setopt("-tpid", nfile, fnm);
1755 setopt("-mtx" , nfile, fnm);
1758 *rcoulomb = ir.rcoulomb;
1760 /* Return the estimate for the number of PME nodes */
1761 return pme_load_estimate(&mtop,&ir,state.box);
1765 static void couple_files_options(int nfile, t_filenm fnm[])
1768 gmx_bool bSet,bBench;
1773 for (i=0; i<nfile; i++)
1775 opt = (char *)fnm[i].opt;
1776 bSet = ((fnm[i].flag & ffSET) != 0);
1777 bBench = (0 == strncmp(opt,"-b", 2));
1779 /* Check optional files */
1780 /* If e.g. -eo is set, then -beo also needs to be set */
1781 if (is_optional(&fnm[i]) && bSet && !bBench)
1783 sprintf(buf, "-b%s", &opt[1]);
1784 setopt(buf,nfile,fnm);
1786 /* If -beo is set, then -eo also needs to be! */
1787 if (is_optional(&fnm[i]) && bSet && bBench)
1789 sprintf(buf, "-%s", &opt[2]);
1790 setopt(buf,nfile,fnm);
1796 static double gettime()
1798 #ifdef HAVE_GETTIMEOFDAY
1802 gettimeofday(&t,NULL);
1804 seconds = (double) t.tv_sec + 1e-6*(double)t.tv_usec;
1810 seconds = time(NULL);
1817 #define BENCHSTEPS (1000)
1819 int gmx_tune_pme(int argc,char *argv[])
1821 const char *desc[] = {
1822 "For a given number [TT]-np[tt] or [TT]-nt[tt] of processors/threads, this program systematically",
1823 "times [TT]mdrun[tt] with various numbers of PME-only nodes and determines",
1824 "which setting is fastest. It will also test whether performance can",
1825 "be enhanced by shifting load from the reciprocal to the real space",
1826 "part of the Ewald sum. ",
1827 "Simply pass your [TT].tpr[tt] file to [TT]g_tune_pme[tt] together with other options",
1828 "for [TT]mdrun[tt] as needed.[PAR]",
1829 "Which executables are used can be set in the environment variables",
1830 "MPIRUN and MDRUN. If these are not present, 'mpirun' and 'mdrun'",
1831 "will be used as defaults. Note that for certain MPI frameworks you",
1832 "need to provide a machine- or hostfile. This can also be passed",
1833 "via the MPIRUN variable, e.g.[PAR]",
1834 "[TT]export MPIRUN=\"/usr/local/mpirun -machinefile hosts\"[tt][PAR]",
1835 "Please call [TT]g_tune_pme[tt] with the normal options you would pass to",
1836 "[TT]mdrun[tt] and add [TT]-np[tt] for the number of processors to perform the",
1837 "tests on, or [TT]-nt[tt] for the number of threads. You can also add [TT]-r[tt]",
1838 "to repeat each test several times to get better statistics. [PAR]",
1839 "[TT]g_tune_pme[tt] can test various real space / reciprocal space workloads",
1840 "for you. With [TT]-ntpr[tt] you control how many extra [TT].tpr[tt] files will be",
1841 "written with enlarged cutoffs and smaller Fourier grids respectively.",
1842 "Typically, the first test (number 0) will be with the settings from the input",
1843 "[TT].tpr[tt] file; the last test (number [TT]ntpr[tt]) will have the Coulomb cutoff",
1844 "specified by [TT]-rmax[tt] with a somwhat smaller PME grid at the same time. ",
1845 "In this last test, the Fourier spacing is multiplied with [TT]rmax[tt]/rcoulomb. ",
1846 "The remaining [TT].tpr[tt] files will have equally-spaced Coulomb radii (and Fourier "
1847 "spacings) between these extremes. [BB]Note[bb] that you can set [TT]-ntpr[tt] to 1",
1848 "if you just seek the optimal number of PME-only nodes; in that case",
1849 "your input [TT].tpr[tt] file will remain unchanged.[PAR]",
1850 "For the benchmark runs, the default of 1000 time steps should suffice for most",
1851 "MD systems. The dynamic load balancing needs about 100 time steps",
1852 "to adapt to local load imbalances, therefore the time step counters",
1853 "are by default reset after 100 steps. For large systems (>1M atoms), as well as ",
1854 "for a higher accuarcy of the measurements, you should set [TT]-resetstep[tt] to a higher value.",
1855 "From the 'DD' load imbalance entries in the md.log output file you",
1856 "can tell after how many steps the load is sufficiently balanced. Example call:[PAR]"
1857 "[TT]g_tune_pme -np 64 -s protein.tpr -launch[tt][PAR]",
1858 "After calling [TT]mdrun[tt] several times, detailed performance information",
1859 "is available in the output file [TT]perf.out.[tt] ",
1860 "[BB]Note[bb] that during the benchmarks, a couple of temporary files are written",
1861 "(options [TT]-b[tt]*), these will be automatically deleted after each test.[PAR]",
1862 "If you want the simulation to be started automatically with the",
1863 "optimized parameters, use the command line option [TT]-launch[tt].[PAR]",
1868 int pmeentries=0; /* How many values for -npme do we actually test for each tpr file */
1869 real maxPMEfraction=0.50;
1870 real minPMEfraction=0.25;
1871 int maxPMEnodes, minPMEnodes;
1872 float guessPMEratio; /* guessed PME:PP ratio based on the tpr file */
1873 float guessPMEnodes;
1874 int npme_fixed=-2; /* If >= -1, use only this number
1875 * of PME-only nodes */
1877 real rmin=0.0,rmax=0.0; /* min and max value for rcoulomb if scaling is requested */
1878 real rcoulomb=-1.0; /* Coulomb radius as set in .tpr file */
1879 gmx_bool bScaleRvdw=TRUE;
1880 gmx_large_int_t bench_nsteps=BENCHSTEPS;
1881 gmx_large_int_t new_sim_nsteps=-1; /* -1 indicates: not set by the user */
1882 gmx_large_int_t cpt_steps=0; /* Step counter in .cpt input file */
1883 int presteps=100; /* Do a full cycle reset after presteps steps */
1884 gmx_bool bOverwrite=FALSE, bKeepTPR;
1885 gmx_bool bLaunch=FALSE;
1886 char *ExtraArgs=NULL;
1887 char **tpr_names=NULL;
1888 const char *simulation_tpr=NULL;
1889 int best_npme, best_tpr;
1890 int sim_part = 1; /* For benchmarks with checkpoint files */
1893 /* Default program names if nothing else is found */
1894 char *cmd_mpirun=NULL, *cmd_mdrun=NULL;
1895 char *cmd_args_bench, *cmd_args_launch;
1898 t_perf **perfdata=NULL;
1904 /* Print out how long the tuning took */
1907 static t_filenm fnm[] = {
1909 { efOUT, "-p", "perf", ffWRITE },
1910 { efLOG, "-err", "bencherr", ffWRITE },
1911 { efTPX, "-so", "tuned", ffWRITE },
1913 { efTPX, NULL, NULL, ffREAD },
1914 { efTRN, "-o", NULL, ffWRITE },
1915 { efXTC, "-x", NULL, ffOPTWR },
1916 { efCPT, "-cpi", NULL, ffOPTRD },
1917 { efCPT, "-cpo", NULL, ffOPTWR },
1918 { efSTO, "-c", "confout", ffWRITE },
1919 { efEDR, "-e", "ener", ffWRITE },
1920 { efLOG, "-g", "md", ffWRITE },
1921 { efXVG, "-dhdl", "dhdl", ffOPTWR },
1922 { efXVG, "-field", "field", ffOPTWR },
1923 { efXVG, "-table", "table", ffOPTRD },
1924 { efXVG, "-tabletf", "tabletf", ffOPTRD },
1925 { efXVG, "-tablep", "tablep", ffOPTRD },
1926 { efXVG, "-tableb", "table", ffOPTRD },
1927 { efTRX, "-rerun", "rerun", ffOPTRD },
1928 { efXVG, "-tpi", "tpi", ffOPTWR },
1929 { efXVG, "-tpid", "tpidist", ffOPTWR },
1930 { efEDI, "-ei", "sam", ffOPTRD },
1931 { efEDO, "-eo", "sam", ffOPTWR },
1932 { efGCT, "-j", "wham", ffOPTRD },
1933 { efGCT, "-jo", "bam", ffOPTWR },
1934 { efXVG, "-ffout", "gct", ffOPTWR },
1935 { efXVG, "-devout", "deviatie", ffOPTWR },
1936 { efXVG, "-runav", "runaver", ffOPTWR },
1937 { efXVG, "-px", "pullx", ffOPTWR },
1938 { efXVG, "-pf", "pullf", ffOPTWR },
1939 { efXVG, "-ro", "rotation", ffOPTWR },
1940 { efLOG, "-ra", "rotangles",ffOPTWR },
1941 { efLOG, "-rs", "rotslabs", ffOPTWR },
1942 { efLOG, "-rt", "rottorque",ffOPTWR },
1943 { efMTX, "-mtx", "nm", ffOPTWR },
1944 { efNDX, "-dn", "dipole", ffOPTWR },
1945 /* Output files that are deleted after each benchmark run */
1946 { efTRN, "-bo", "bench", ffWRITE },
1947 { efXTC, "-bx", "bench", ffWRITE },
1948 { efCPT, "-bcpo", "bench", ffWRITE },
1949 { efSTO, "-bc", "bench", ffWRITE },
1950 { efEDR, "-be", "bench", ffWRITE },
1951 { efLOG, "-bg", "bench", ffWRITE },
1952 { efEDO, "-beo", "bench", ffOPTWR },
1953 { efXVG, "-bdhdl", "benchdhdl",ffOPTWR },
1954 { efXVG, "-bfield", "benchfld" ,ffOPTWR },
1955 { efXVG, "-btpi", "benchtpi", ffOPTWR },
1956 { efXVG, "-btpid", "benchtpid",ffOPTWR },
1957 { efGCT, "-bjo", "bench", ffOPTWR },
1958 { efXVG, "-bffout", "benchgct", ffOPTWR },
1959 { efXVG, "-bdevout","benchdev", ffOPTWR },
1960 { efXVG, "-brunav", "benchrnav",ffOPTWR },
1961 { efXVG, "-bpx", "benchpx", ffOPTWR },
1962 { efXVG, "-bpf", "benchpf", ffOPTWR },
1963 { efXVG, "-bro", "benchrot", ffOPTWR },
1964 { efLOG, "-bra", "benchrota",ffOPTWR },
1965 { efLOG, "-brs", "benchrots",ffOPTWR },
1966 { efLOG, "-brt", "benchrott",ffOPTWR },
1967 { efMTX, "-bmtx", "benchn", ffOPTWR },
1968 { efNDX, "-bdn", "bench", ffOPTWR }
1971 gmx_bool bThreads = FALSE;
1975 const char *procstring[] =
1976 { NULL, "-np", "-n", "none", NULL };
1977 const char *npmevalues_opt[] =
1978 { NULL, "auto", "all", "subset", NULL };
1980 gmx_bool bAppendFiles=TRUE;
1981 gmx_bool bKeepAndNumCPT=FALSE;
1982 gmx_bool bResetCountersHalfWay=FALSE;
1983 gmx_bool bBenchmark=TRUE;
1985 output_env_t oenv=NULL;
1988 /***********************/
1989 /* g_tune_pme options: */
1990 /***********************/
1991 { "-np", FALSE, etINT, {&nnodes},
1992 "Number of nodes to run the tests on (must be > 2 for separate PME nodes)" },
1993 { "-npstring", FALSE, etENUM, {procstring},
1994 "Specify the number of processors to [TT]$MPIRUN[tt] using this string"},
1995 { "-nt", FALSE, etINT, {&nthreads},
1996 "Number of threads to run the tests on (turns MPI & mpirun off)"},
1997 { "-r", FALSE, etINT, {&repeats},
1998 "Repeat each test this often" },
1999 { "-max", FALSE, etREAL, {&maxPMEfraction},
2000 "Max fraction of PME nodes to test with" },
2001 { "-min", FALSE, etREAL, {&minPMEfraction},
2002 "Min fraction of PME nodes to test with" },
2003 { "-npme", FALSE, etENUM, {npmevalues_opt},
2004 "Within -min and -max, benchmark all possible values for [TT]-npme[tt], or just a reasonable subset. "
2005 "Auto neglects -min and -max and chooses reasonable values around a guess for npme derived from the .tpr"},
2006 { "-fix", FALSE, etINT, {&npme_fixed},
2007 "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."},
2008 { "-rmax", FALSE, etREAL, {&rmax},
2009 "If >0, maximal rcoulomb for -ntpr>1 (rcoulomb upscaling results in fourier grid downscaling)" },
2010 { "-rmin", FALSE, etREAL, {&rmin},
2011 "If >0, minimal rcoulomb for -ntpr>1" },
2012 { "-scalevdw", FALSE, etBOOL, {&bScaleRvdw},
2013 "Scale rvdw along with rcoulomb"},
2014 { "-ntpr", FALSE, etINT, {&ntprs},
2015 "Number of [TT].tpr[tt] files to benchmark. Create this many files with different rcoulomb scaling factors depending on -rmin and -rmax. "
2016 "If < 1, automatically choose the number of [TT].tpr[tt] files to test" },
2017 { "-steps", FALSE, etGMX_LARGE_INT, {&bench_nsteps},
2018 "Take timings for this many steps in the benchmark runs" },
2019 { "-resetstep",FALSE, etINT, {&presteps},
2020 "Let dlb equilibrate this many steps before timings are taken (reset cycle counters after this many steps)" },
2021 { "-simsteps", FALSE, etGMX_LARGE_INT, {&new_sim_nsteps},
2022 "If non-negative, perform this many steps in the real run (overwrites nsteps from [TT].tpr[tt], add [TT].cpt[tt] steps)" },
2023 { "-launch", FALSE, etBOOL, {&bLaunch},
2024 "Launch the real simulation after optimization" },
2025 { "-bench", FALSE, etBOOL, {&bBenchmark},
2026 "Run the benchmarks or just create the input [TT].tpr[tt] files?" },
2027 /******************/
2028 /* mdrun options: */
2029 /******************/
2030 /* We let g_tune_pme parse and understand these options, because we need to
2031 * prevent that they appear on the mdrun command line for the benchmarks */
2032 { "-append", FALSE, etBOOL, {&bAppendFiles},
2033 "Append to previous output files when continuing from checkpoint instead of adding the simulation part number to all file names (for launch only)" },
2034 { "-cpnum", FALSE, etBOOL, {&bKeepAndNumCPT},
2035 "Keep and number checkpoint files (launch only)" },
2036 { "-resethway", FALSE, etBOOL, {&bResetCountersHalfWay},
2037 "HIDDENReset the cycle counters after half the number of steps or halfway [TT]-maxh[tt] (launch only)" }
2041 #define NFILE asize(fnm)
2043 CopyRight(stderr,argv[0]);
2045 seconds = gettime();
2047 parse_common_args(&argc,argv,PCA_NOEXIT_ON_ARGS,
2048 NFILE,fnm,asize(pa),pa,asize(desc),desc,
2051 /* Store the remaining unparsed command line entries in a string which
2052 * is then attached to the mdrun command line */
2054 ExtraArgs[0] = '\0';
2055 for (i=1; i<argc; i++) /* argc will now be 1 if everything was understood */
2057 add_to_string(&ExtraArgs, argv[i]);
2058 add_to_string(&ExtraArgs, " ");
2061 if (opt2parg_bSet("-nt",asize(pa),pa))
2064 if (opt2parg_bSet("-npstring",asize(pa),pa))
2065 fprintf(stderr, "WARNING: -npstring has no effect when using threads.\n");
2068 gmx_fatal(FARGS, "Can't run multi-threaded MPI simulation yet!");
2069 /* and now we just set this; a bit of an ugly hack*/
2072 /* Check for PME:PP ratio and whether tpr triggers additional output files */
2073 guessPMEratio = inspect_tpr(NFILE,fnm,&rcoulomb);
2075 /* Automatically set -beo options if -eo is set etc. */
2076 couple_files_options(NFILE,fnm);
2078 /* Construct the command line arguments for benchmark runs
2079 * as well as for the simulation run */
2081 sprintf(bbuf," -nt %d ", nthreads);
2083 sprintf(bbuf," -np %d ", nnodes);
2087 create_command_line_snippets(bThreads,bAppendFiles,bKeepAndNumCPT,bResetCountersHalfWay,presteps,
2088 NFILE,fnm,asize(pa),pa,procstring[0],
2089 &cmd_np, &cmd_args_bench, &cmd_args_launch,
2092 /* Read in checkpoint file if requested */
2094 if(opt2bSet("-cpi",NFILE,fnm))
2097 cr->duty=DUTY_PP; /* makes the following routine happy */
2098 read_checkpoint_simulation_part(opt2fn("-cpi",NFILE,fnm),
2099 &sim_part,&cpt_steps,cr,
2100 FALSE,NFILE,fnm,NULL,NULL);
2103 /* sim_part will now be 1 if no checkpoint file was found */
2105 gmx_fatal(FARGS, "Checkpoint file %s not found!", opt2fn("-cpi",NFILE,fnm));
2108 /* Open performance output file and write header info */
2109 fp = ffopen(opt2fn("-p",NFILE,fnm),"w");
2111 /* Make a quick consistency check of command line parameters */
2112 check_input(nnodes, repeats, &ntprs, &rmin, rcoulomb, &rmax,
2113 maxPMEfraction, minPMEfraction, npme_fixed,
2114 bench_nsteps, fnm, NFILE, sim_part, presteps,
2117 /* Determine the maximum and minimum number of PME nodes to test,
2118 * the actual list of settings is build in do_the_tests(). */
2119 if ((nnodes > 2) && (npme_fixed < -1))
2121 if (0 == strcmp(npmevalues_opt[0], "auto"))
2123 /* Determine the npme range automatically based on the PME:PP load guess */
2124 if (guessPMEratio > 1.0)
2126 /* More PME than PP work, probably we do not need separate PME nodes at all! */
2127 maxPMEnodes=nnodes/2;
2128 minPMEnodes=nnodes/2;
2132 /* PME : PP load is in the range 0..1, let's test around the guess */
2133 guessPMEnodes = nnodes/(1.0 + 1.0/guessPMEratio);
2134 minPMEnodes = floor(0.7*guessPMEnodes);
2135 maxPMEnodes = ceil(1.6*guessPMEnodes);
2136 maxPMEnodes = min(maxPMEnodes, nnodes/2);
2141 /* Determine the npme range based on user input */
2142 maxPMEnodes = floor(maxPMEfraction*nnodes);
2143 minPMEnodes = max(floor(minPMEfraction*nnodes), 0);
2144 fprintf(stdout, "Will try runs with %d ", minPMEnodes);
2145 if (maxPMEnodes != minPMEnodes)
2146 fprintf(stdout, "- %d ", maxPMEnodes);
2147 fprintf(stdout, "PME-only nodes.\n Note that the automatic number of PME-only nodes and no separate PME nodes are always tested.\n");
2156 /* Get the commands we need to set up the runs from environment variables */
2157 get_program_paths(bThreads, &cmd_mpirun, cmd_np, &cmd_mdrun, repeats);
2159 /* Print some header info to file */
2161 fprintf(fp, "\n P E R F O R M A N C E R E S U L T S\n");
2163 fprintf(fp, "%s for Gromacs %s\n", ShortProgram(),GromacsVersion());
2166 fprintf(fp, "Number of nodes : %d\n", nnodes);
2167 fprintf(fp, "The mpirun command is : %s\n", cmd_mpirun);
2168 if ( strcmp(procstring[0], "none") != 0)
2169 fprintf(fp, "Passing # of nodes via : %s\n", procstring[0]);
2171 fprintf(fp, "Not setting number of nodes in system call\n");
2174 fprintf(fp, "Number of threads : %d\n", nnodes);
2176 fprintf(fp, "The mdrun command is : %s\n", cmd_mdrun);
2177 fprintf(fp, "mdrun args benchmarks : %s\n", cmd_args_bench);
2178 fprintf(fp, "Benchmark steps : ");
2179 fprintf(fp, gmx_large_int_pfmt, bench_nsteps);
2181 fprintf(fp, "dlb equilibration steps : %d\n", presteps);
2184 fprintf(fp, "Checkpoint time step : ");
2185 fprintf(fp, gmx_large_int_pfmt, cpt_steps);
2188 fprintf(fp, "mdrun args at launchtime: %s\n", cmd_args_launch);
2190 if (new_sim_nsteps >= 0)
2193 fprintf(stderr, "Note: Simulation input file %s will have ", opt2fn("-so",NFILE,fnm));
2194 fprintf(stderr, gmx_large_int_pfmt, new_sim_nsteps+cpt_steps);
2195 fprintf(stderr, " steps.\n");
2196 fprintf(fp, "Simulation steps : ");
2197 fprintf(fp, gmx_large_int_pfmt, new_sim_nsteps);
2201 fprintf(fp, "Repeats for each test : %d\n", repeats);
2203 if (npme_fixed >= -1)
2205 fprintf(fp, "Fixing -npme at : %d\n", npme_fixed);
2208 fprintf(fp, "Input file : %s\n", opt2fn("-s",NFILE,fnm));
2209 fprintf(fp, " PME/PP load estimate : %g\n", guessPMEratio);
2211 /* Allocate memory for the inputinfo struct: */
2213 info->nr_inputfiles = ntprs;
2214 for (i=0; i<ntprs; i++)
2216 snew(info->rcoulomb , ntprs);
2217 snew(info->rvdw , ntprs);
2218 snew(info->rlist , ntprs);
2219 snew(info->rlistlong, ntprs);
2220 snew(info->nkx , ntprs);
2221 snew(info->nky , ntprs);
2222 snew(info->nkz , ntprs);
2223 snew(info->fsx , ntprs);
2224 snew(info->fsy , ntprs);
2225 snew(info->fsz , ntprs);
2227 /* Make alternative tpr files to test: */
2228 snew(tpr_names, ntprs);
2229 for (i=0; i<ntprs; i++)
2230 snew(tpr_names[i], STRLEN);
2232 /* It can be that ntprs is reduced by make_benchmark_tprs if not enough
2233 * different grids could be found. */
2234 make_benchmark_tprs(opt2fn("-s",NFILE,fnm), tpr_names, bench_nsteps+presteps,
2235 cpt_steps, rmin, rmax, bScaleRvdw, &ntprs, info, fp);
2237 /********************************************************************************/
2238 /* Main loop over all scenarios we need to test: tpr files, PME nodes, repeats */
2239 /********************************************************************************/
2240 snew(perfdata, ntprs);
2243 do_the_tests(fp, tpr_names, maxPMEnodes, minPMEnodes, npme_fixed, npmevalues_opt[0], perfdata, &pmeentries,
2244 repeats, nnodes, ntprs, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2245 cmd_args_bench, fnm, NFILE, sim_part, presteps, cpt_steps);
2247 fprintf(fp, "\nTuning took%8.1f minutes.\n", (gettime()-seconds)/60.0);
2249 /* Analyse the results and give a suggestion for optimal settings: */
2250 bKeepTPR = analyze_data(fp, opt2fn("-p", NFILE, fnm), perfdata, nnodes, ntprs, pmeentries,
2251 repeats, info, &best_tpr, &best_npme);
2253 /* Take the best-performing tpr file and enlarge nsteps to original value */
2254 if ( bKeepTPR && !bOverwrite )
2256 simulation_tpr = opt2fn("-s",NFILE,fnm);
2260 simulation_tpr = opt2fn("-so",NFILE,fnm);
2261 modify_PMEsettings(bOverwrite? (new_sim_nsteps+cpt_steps) : info->orig_sim_steps,
2262 info->orig_init_step, tpr_names[best_tpr], simulation_tpr);
2265 /* Let's get rid of the temporary benchmark input files */
2266 for (i=0; i<ntprs; i++)
2268 fprintf(stdout, "Deleting temporary benchmark input file %s\n", tpr_names[i]);
2269 remove(tpr_names[i]);
2272 /* Now start the real simulation if the user requested it ... */
2273 launch_simulation(bLaunch, fp, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2274 cmd_args_launch, simulation_tpr, nnodes, best_npme);
2278 /* ... or simply print the performance results to screen: */
2280 finalize(opt2fn("-p", NFILE, fnm));