2 * This file is part of the GROMACS molecular simulation package.
4 * Copyright (c) 2009,2010,2011,2012,2013,2014, by the GROMACS development team, led by
5 * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
6 * and including many others, as listed in the AUTHORS file in the
7 * top-level source directory and at http://www.gromacs.org.
9 * GROMACS is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public License
11 * as published by the Free Software Foundation; either version 2.1
12 * of the License, or (at your option) any later version.
14 * GROMACS is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with GROMACS; if not, see
21 * http://www.gnu.org/licenses, or write to the Free Software Foundation,
22 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
24 * If you want to redistribute modifications to GROMACS, please
25 * consider that scientific software is very special. Version
26 * control is crucial - bugs must be traceable. We will be happy to
27 * consider code for inclusion in the official distribution, but
28 * derived work must not be called official GROMACS. Details are found
29 * in the README & COPYING files - if they are missing, get the
30 * official version at http://www.gromacs.org.
32 * To help us fund GROMACS development, we humbly ask that you cite
33 * the research papers on the package. Check out http://www.gromacs.org.
42 #ifdef HAVE_SYS_TIME_H
46 #include "gromacs/commandline/pargs.h"
47 #include "gromacs/fileio/tpxio.h"
48 #include "gromacs/gmxana/gmx_ana.h"
49 #include "gromacs/legacyheaders/calcgrid.h"
50 #include "gromacs/legacyheaders/checkpoint.h"
51 #include "gromacs/legacyheaders/inputrec.h"
52 #include "gromacs/legacyheaders/macros.h"
53 #include "gromacs/legacyheaders/names.h"
54 #include "gromacs/legacyheaders/perf_est.h"
55 #include "gromacs/legacyheaders/readinp.h"
56 #include "gromacs/legacyheaders/typedefs.h"
57 #include "gromacs/legacyheaders/types/commrec.h"
58 #include "gromacs/math/utilities.h"
59 #include "gromacs/math/vec.h"
60 #include "gromacs/timing/walltime_accounting.h"
61 #include "gromacs/utility/baseversion.h"
62 #include "gromacs/utility/cstringutil.h"
63 #include "gromacs/utility/fatalerror.h"
64 #include "gromacs/utility/smalloc.h"
66 /* Enum for situations that can occur during log file parsing, the
67 * corresponding string entries can be found in do_the_tests() in
68 * const char* ParseLog[] */
74 eParselogResetProblem,
78 eParselogLargePrimeFactor,
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_int64_t orig_sim_steps; /* Number of steps to be done in the real simulation */
103 gmx_int64_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 void remove_if_exists(const char *fn)
151 fprintf(stdout, "Deleting %s\n", fn);
157 static void finalize(const char *fn_out)
163 fp = fopen(fn_out, "r");
164 fprintf(stdout, "\n\n");
166 while (fgets(buf, STRLEN-1, fp) != NULL)
168 fprintf(stdout, "%s", buf);
171 fprintf(stdout, "\n\n");
176 eFoundNothing, eFoundDDStr, eFoundAccountingStr, eFoundCycleStr
179 static int parse_logfile(const char *logfile, const char *errfile,
180 t_perf *perfdata, int test_nr, int presteps, gmx_int64_t cpt_steps,
184 char line[STRLEN], dumstring[STRLEN], dumstring2[STRLEN];
185 const char matchstrdd[] = "Domain decomposition grid";
186 const char matchstrcr[] = "resetting all time and cycle counters";
187 const char matchstrbal[] = "Average PME mesh/force load:";
188 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";
189 const char errSIG[] = "signal, stopping at the next";
191 float dum1, dum2, dum3, dum4;
194 gmx_int64_t resetsteps = -1;
195 gmx_bool bFoundResetStr = FALSE;
196 gmx_bool bResetChecked = FALSE;
199 if (!gmx_fexist(logfile))
201 fprintf(stderr, "WARNING: Could not find logfile %s.\n", logfile);
202 cleandata(perfdata, test_nr);
203 return eParselogNotFound;
206 fp = fopen(logfile, "r");
207 perfdata->PME_f_load[test_nr] = -1.0;
208 perfdata->guessPME = -1;
210 iFound = eFoundNothing;
213 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_SCNd64);
235 sscanf(line, dumstring, &resetsteps);
236 bFoundResetStr = TRUE;
237 if (resetsteps == presteps+cpt_steps)
239 bResetChecked = TRUE;
243 sprintf(dumstring, "%"GMX_PRId64, resetsteps);
244 sprintf(dumstring2, "%"GMX_PRId64, 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 ranks %d",
260 &(perfdata->nx), &(perfdata->ny), &(perfdata->nz), &npme);
261 if (perfdata->nPMEnodes == -1)
263 perfdata->guessPME = npme;
265 else if (perfdata->nPMEnodes != npme)
267 gmx_fatal(FARGS, "PME ranks from command line and output file are not identical");
269 iFound = eFoundDDStr;
271 /* Catch a few errors that might have occured: */
272 else if (str_starts(line, "There is no domain decomposition for"))
275 return eParselogNoDDGrid;
277 else if (str_starts(line, "The number of ranks you selected"))
280 return eParselogLargePrimeFactor;
282 else if (str_starts(line, "reading tpx file"))
285 return eParselogTPXVersion;
287 else if (str_starts(line, "The -dd or -npme option request a parallel simulation"))
290 return eParselogNotParallel;
294 /* Look for PME mesh/force balance (not necessarily present, though) */
295 if (str_starts(line, matchstrbal))
297 sscanf(&line[strlen(matchstrbal)], "%f", &(perfdata->PME_f_load[test_nr]));
299 /* Look for matchstring */
300 if (str_starts(line, matchstring))
302 iFound = eFoundAccountingStr;
305 case eFoundAccountingStr:
306 /* Already found matchstring - look for cycle data */
307 if (str_starts(line, "Total "))
309 sscanf(line, "Total %*f %lf", &(perfdata->Gcycles[test_nr]));
310 iFound = eFoundCycleStr;
314 /* Already found cycle data - look for remaining performance info and return */
315 if (str_starts(line, "Performance:"))
317 ndum = sscanf(line, "%s %f %f %f %f", dumstring, &dum1, &dum2, &dum3, &dum4);
318 /* (ns/day) is the second last entry, depending on whether GMX_DETAILED_PERF_STATS was set in print_perf(), nrnb.c */
319 perfdata->ns_per_day[test_nr] = (ndum == 5) ? dum3 : dum1;
321 if (bResetChecked || presteps == 0)
327 return eParselogResetProblem;
334 /* Close the log file */
337 /* Check why there is no performance data in the log file.
338 * Did a fatal errors occur? */
339 if (gmx_fexist(errfile))
341 fp = fopen(errfile, "r");
342 while (fgets(line, STRLEN, fp) != NULL)
344 if (str_starts(line, "Fatal error:") )
346 if (fgets(line, STRLEN, fp) != NULL)
348 fprintf(stderr, "\nWARNING: An error occured during this benchmark:\n"
352 cleandata(perfdata, test_nr);
353 return eParselogFatal;
360 fprintf(stderr, "WARNING: Could not find stderr file %s.\n", errfile);
363 /* Giving up ... we could not find out why there is no performance data in
365 fprintf(stdout, "No performance data in log file.\n");
366 cleandata(perfdata, test_nr);
368 return eParselogNoPerfData;
372 static gmx_bool analyze_data(
381 int *index_tpr, /* OUT: Nr of mdp file with best settings */
382 int *npme_optimal) /* OUT: Optimal number of PME nodes */
385 int line = 0, line_win = -1;
386 int k_win = -1, i_win = -1, winPME;
387 double s = 0.0; /* standard deviation */
390 char str_PME_f_load[13];
391 gmx_bool bCanUseOrigTPR;
392 gmx_bool bRefinedCoul, bRefinedVdW, bRefinedGrid;
398 fprintf(fp, "Summary of successful runs:\n");
399 fprintf(fp, "Line tpr PME ranks Gcycles Av. Std.dev. ns/day PME/f");
402 fprintf(fp, " DD grid");
408 for (k = 0; k < ntprs; k++)
410 for (i = 0; i < ntests; i++)
412 /* Select the right dataset: */
413 pd = &(perfdata[k][i]);
415 pd->Gcycles_Av = 0.0;
416 pd->PME_f_load_Av = 0.0;
417 pd->ns_per_day_Av = 0.0;
419 if (pd->nPMEnodes == -1)
421 sprintf(strbuf, "(%3d)", pd->guessPME);
425 sprintf(strbuf, " ");
428 /* Get the average run time of a setting */
429 for (j = 0; j < nrepeats; j++)
431 pd->Gcycles_Av += pd->Gcycles[j];
432 pd->PME_f_load_Av += pd->PME_f_load[j];
434 pd->Gcycles_Av /= nrepeats;
435 pd->PME_f_load_Av /= nrepeats;
437 for (j = 0; j < nrepeats; j++)
439 if (pd->ns_per_day[j] > 0.0)
441 pd->ns_per_day_Av += pd->ns_per_day[j];
445 /* Somehow the performance number was not aquired for this run,
446 * therefor set the average to some negative value: */
447 pd->ns_per_day_Av = -1.0f*nrepeats;
451 pd->ns_per_day_Av /= nrepeats;
454 if (pd->PME_f_load_Av > 0.0)
456 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load_Av);
460 sprintf(str_PME_f_load, "%s", " - ");
464 /* We assume we had a successful run if both averages are positive */
465 if (pd->Gcycles_Av > 0.0 && pd->ns_per_day_Av > 0.0)
467 /* Output statistics if repeats were done */
470 /* Calculate the standard deviation */
472 for (j = 0; j < nrepeats; j++)
474 s += pow( pd->Gcycles[j] - pd->Gcycles_Av, 2 );
479 fprintf(fp, "%4d %3d %4d%s %12.3f %12.3f %12.3f %s",
480 line, k, pd->nPMEnodes, strbuf, pd->Gcycles_Av, s,
481 pd->ns_per_day_Av, str_PME_f_load);
484 fprintf(fp, " %3d %3d %3d", pd->nx, pd->ny, pd->nz);
488 /* Store the index of the best run found so far in 'winner': */
489 if ( (k_win == -1) || (pd->Gcycles_Av < perfdata[k_win][i_win].Gcycles_Av) )
502 gmx_fatal(FARGS, "None of the runs was successful! Check %s for problems.", fn);
507 winPME = perfdata[k_win][i_win].nPMEnodes;
511 /* We stuck to a fixed number of PME-only nodes */
512 sprintf(strbuf, "settings No. %d", k_win);
516 /* We have optimized the number of PME-only nodes */
519 sprintf(strbuf, "%s", "the automatic number of PME ranks");
523 sprintf(strbuf, "%d PME ranks", winPME);
526 fprintf(fp, "Best performance was achieved with %s", strbuf);
527 if ((nrepeats > 1) && (ntests > 1))
529 fprintf(fp, " (see line %d)", line_win);
533 /* Only mention settings if they were modified: */
534 bRefinedCoul = !gmx_within_tol(info->rcoulomb[k_win], info->rcoulomb[0], GMX_REAL_EPS);
535 bRefinedVdW = !gmx_within_tol(info->rvdw[k_win], info->rvdw[0], GMX_REAL_EPS);
536 bRefinedGrid = !(info->nkx[k_win] == info->nkx[0] &&
537 info->nky[k_win] == info->nky[0] &&
538 info->nkz[k_win] == info->nkz[0]);
540 if (bRefinedCoul || bRefinedVdW || bRefinedGrid)
542 fprintf(fp, "Optimized PME settings:\n");
543 bCanUseOrigTPR = FALSE;
547 bCanUseOrigTPR = TRUE;
552 fprintf(fp, " New Coulomb radius: %f nm (was %f nm)\n", info->rcoulomb[k_win], info->rcoulomb[0]);
557 fprintf(fp, " New Van der Waals radius: %f nm (was %f nm)\n", info->rvdw[k_win], info->rvdw[0]);
562 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],
563 info->nkx[0], info->nky[0], info->nkz[0]);
566 if (bCanUseOrigTPR && ntprs > 1)
568 fprintf(fp, "and original PME settings.\n");
573 /* Return the index of the mdp file that showed the highest performance
574 * and the optimal number of PME nodes */
576 *npme_optimal = winPME;
578 return bCanUseOrigTPR;
582 /* Get the commands we need to set up the runs from environment variables */
583 static void get_program_paths(gmx_bool bThreads, char *cmd_mpirun[], char *cmd_mdrun[])
587 const char def_mpirun[] = "mpirun";
588 const char def_mdrun[] = "mdrun";
590 const char empty_mpirun[] = "";
592 /* Get the commands we need to set up the runs from environment variables */
595 if ( (cp = getenv("MPIRUN")) != NULL)
597 *cmd_mpirun = gmx_strdup(cp);
601 *cmd_mpirun = gmx_strdup(def_mpirun);
606 *cmd_mpirun = gmx_strdup(empty_mpirun);
609 if ( (cp = getenv("MDRUN" )) != NULL)
611 *cmd_mdrun = gmx_strdup(cp);
615 *cmd_mdrun = gmx_strdup(def_mdrun);
619 /* Check that the commands will run mdrun (perhaps via mpirun) by
620 * running a very quick test simulation. Requires MPI environment to
621 * be available if applicable. */
622 static void check_mdrun_works(gmx_bool bThreads,
623 const char *cmd_mpirun,
625 const char *cmd_mdrun)
627 char *command = NULL;
631 const char filename[] = "benchtest.log";
633 /* This string should always be identical to the one in copyrite.c,
634 * gmx_print_version_info() in the defined(GMX_MPI) section */
635 const char match_mpi[] = "MPI library: MPI";
636 const char match_mdrun[] = "Executable: ";
637 gmx_bool bMdrun = FALSE;
638 gmx_bool bMPI = FALSE;
640 /* Run a small test to see whether mpirun + mdrun work */
641 fprintf(stdout, "Making sure that mdrun can be executed. ");
644 snew(command, strlen(cmd_mdrun) + strlen(cmd_np) + strlen(filename) + 50);
645 sprintf(command, "%s%s-version -maxh 0.001 1> %s 2>&1", cmd_mdrun, cmd_np, filename);
649 snew(command, strlen(cmd_mpirun) + strlen(cmd_np) + strlen(cmd_mdrun) + strlen(filename) + 50);
650 sprintf(command, "%s%s%s -version -maxh 0.001 1> %s 2>&1", cmd_mpirun, cmd_np, cmd_mdrun, filename);
652 fprintf(stdout, "Trying '%s' ... ", command);
653 make_backup(filename);
654 gmx_system_call(command);
656 /* Check if we find the characteristic string in the output: */
657 if (!gmx_fexist(filename))
659 gmx_fatal(FARGS, "Output from test run could not be found.");
662 fp = fopen(filename, "r");
663 /* We need to scan the whole output file, since sometimes the queuing system
664 * also writes stuff to stdout/err */
667 cp = fgets(line, STRLEN, fp);
670 if (str_starts(line, match_mdrun) )
674 if (str_starts(line, match_mpi) )
686 gmx_fatal(FARGS, "Need a threaded version of mdrun. This one\n"
688 "seems to have been compiled with MPI instead.",
696 gmx_fatal(FARGS, "Need an MPI-enabled version of mdrun. This one\n"
698 "seems to have been compiled without MPI support.",
705 gmx_fatal(FARGS, "Cannot execute mdrun. Please check %s for problems!",
709 fprintf(stdout, "passed.\n");
717 static void launch_simulation(
718 gmx_bool bLaunch, /* Should the simulation be launched? */
719 FILE *fp, /* General log file */
720 gmx_bool bThreads, /* whether to use threads */
721 char *cmd_mpirun, /* Command for mpirun */
722 char *cmd_np, /* Switch for -np or -ntmpi or empty */
723 char *cmd_mdrun, /* Command for mdrun */
724 char *args_for_mdrun, /* Arguments for mdrun */
725 const char *simulation_tpr, /* This tpr will be simulated */
726 int nPMEnodes) /* Number of PME nodes to use */
731 /* Make enough space for the system call command,
732 * (100 extra chars for -npme ... etc. options should suffice): */
733 snew(command, strlen(cmd_mpirun)+strlen(cmd_mdrun)+strlen(cmd_np)+strlen(args_for_mdrun)+strlen(simulation_tpr)+100);
735 /* Note that the -passall options requires args_for_mdrun to be at the end
736 * of the command line string */
739 sprintf(command, "%s%s-npme %d -s %s %s",
740 cmd_mdrun, cmd_np, nPMEnodes, simulation_tpr, args_for_mdrun);
744 sprintf(command, "%s%s%s -npme %d -s %s %s",
745 cmd_mpirun, cmd_np, cmd_mdrun, nPMEnodes, simulation_tpr, args_for_mdrun);
748 fprintf(fp, "%s this command line to launch the simulation:\n\n%s", bLaunch ? "Using" : "Please use", command);
752 /* Now the real thing! */
755 fprintf(stdout, "\nLaunching simulation with best parameters now.\nExecuting '%s'", command);
758 gmx_system_call(command);
763 static void modify_PMEsettings(
764 gmx_int64_t simsteps, /* Set this value as number of time steps */
765 gmx_int64_t init_step, /* Set this value as init_step */
766 const char *fn_best_tpr, /* tpr file with the best performance */
767 const char *fn_sim_tpr) /* name of tpr file to be launched */
775 read_tpx_state(fn_best_tpr, ir, &state, NULL, &mtop);
777 /* Reset nsteps and init_step to the value of the input .tpr file */
778 ir->nsteps = simsteps;
779 ir->init_step = init_step;
781 /* Write the tpr file which will be launched */
782 sprintf(buf, "Writing optimized simulation file %s with nsteps=%s.\n", fn_sim_tpr, "%"GMX_PRId64);
783 fprintf(stdout, buf, ir->nsteps);
785 write_tpx_state(fn_sim_tpr, ir, &state, &mtop);
790 static gmx_bool can_scale_rvdw(int vdwtype)
792 return (evdwCUT == vdwtype ||
796 #define EPME_SWITCHED(e) ((e) == eelPMESWITCH || (e) == eelPMEUSERSWITCH)
798 /* Make additional TPR files with more computational load for the
799 * direct space processors: */
800 static void make_benchmark_tprs(
801 const char *fn_sim_tpr, /* READ : User-provided tpr file */
802 char *fn_bench_tprs[], /* WRITE: Names of benchmark tpr files */
803 gmx_int64_t benchsteps, /* Number of time steps for benchmark runs */
804 gmx_int64_t statesteps, /* Step counter in checkpoint file */
805 real rmin, /* Minimal Coulomb radius */
806 real rmax, /* Maximal Coulomb radius */
807 real bScaleRvdw, /* Scale rvdw along with rcoulomb */
808 int *ntprs, /* No. of TPRs to write, each with a different
809 rcoulomb and fourierspacing */
810 t_inputinfo *info, /* Contains information about mdp file options */
811 FILE *fp) /* Write the output here */
817 real nlist_buffer; /* Thickness of the buffer regions for PME-switch potentials */
820 gmx_bool bNote = FALSE;
821 real add; /* Add this to rcoul for the next test */
822 real fac = 1.0; /* Scaling factor for Coulomb radius */
823 real fourierspacing; /* Basic fourierspacing from tpr */
826 sprintf(buf, "Making benchmark tpr file%s with %s time step%s",
827 *ntprs > 1 ? "s" : "", "%"GMX_PRId64, benchsteps > 1 ? "s" : "");
828 fprintf(stdout, buf, benchsteps);
831 sprintf(buf, " (adding %s steps from checkpoint file)", "%"GMX_PRId64);
832 fprintf(stdout, buf, statesteps);
833 benchsteps += statesteps;
835 fprintf(stdout, ".\n");
839 read_tpx_state(fn_sim_tpr, ir, &state, NULL, &mtop);
841 /* Check if some kind of PME was chosen */
842 if (EEL_PME(ir->coulombtype) == FALSE)
844 gmx_fatal(FARGS, "Can only do optimizations for simulations with %s electrostatics.",
848 /* Check if rcoulomb == rlist, which is necessary for plain PME. */
849 if ( (ir->cutoff_scheme != ecutsVERLET) &&
850 (eelPME == ir->coulombtype) && !(ir->rcoulomb == ir->rlist))
852 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to rlist (%f).",
853 EELTYPE(eelPME), ir->rcoulomb, ir->rlist);
855 /* For other PME types, rcoulomb is allowed to be smaller than rlist */
856 else if (ir->rcoulomb > ir->rlist)
858 gmx_fatal(FARGS, "%s requires rcoulomb (%f) to be equal to or smaller than rlist (%f)",
859 EELTYPE(ir->coulombtype), ir->rcoulomb, ir->rlist);
862 if (bScaleRvdw && ir->rvdw != ir->rcoulomb)
864 fprintf(stdout, "NOTE: input rvdw != rcoulomb, will not scale rvdw\n");
868 /* Reduce the number of steps for the benchmarks */
869 info->orig_sim_steps = ir->nsteps;
870 ir->nsteps = benchsteps;
871 /* We must not use init_step from the input tpr file for the benchmarks */
872 info->orig_init_step = ir->init_step;
875 /* For PME-switch potentials, keep the radial distance of the buffer region */
876 nlist_buffer = ir->rlist - ir->rcoulomb;
878 /* Determine length of triclinic box vectors */
879 for (d = 0; d < DIM; d++)
882 for (i = 0; i < DIM; i++)
884 box_size[d] += state.box[d][i]*state.box[d][i];
886 box_size[d] = sqrt(box_size[d]);
889 if (ir->fourier_spacing > 0)
891 info->fsx[0] = ir->fourier_spacing;
892 info->fsy[0] = ir->fourier_spacing;
893 info->fsz[0] = ir->fourier_spacing;
897 /* Reconstruct fourierspacing per dimension from the number of grid points and box size */
898 info->fsx[0] = box_size[XX]/ir->nkx;
899 info->fsy[0] = box_size[YY]/ir->nky;
900 info->fsz[0] = box_size[ZZ]/ir->nkz;
903 /* If no value for the fourierspacing was provided on the command line, we
904 * use the reconstruction from the tpr file */
905 if (ir->fourier_spacing > 0)
907 /* Use the spacing from the tpr */
908 fourierspacing = ir->fourier_spacing;
912 /* Use the maximum observed spacing */
913 fourierspacing = max(max(info->fsx[0], info->fsy[0]), info->fsz[0]);
916 fprintf(stdout, "Calculating PME grid points on the basis of a fourierspacing of %f nm\n", fourierspacing);
918 /* For performance comparisons the number of particles is useful to have */
919 fprintf(fp, " Number of particles : %d\n", mtop.natoms);
921 /* Print information about settings of which some are potentially modified: */
922 fprintf(fp, " Coulomb type : %s\n", EELTYPE(ir->coulombtype));
923 fprintf(fp, " Grid spacing x y z : %f %f %f\n",
924 box_size[XX]/ir->nkx, box_size[YY]/ir->nky, box_size[ZZ]/ir->nkz);
925 fprintf(fp, " Van der Waals type : %s\n", EVDWTYPE(ir->vdwtype));
926 if (ir_vdw_switched(ir))
928 fprintf(fp, " rvdw_switch : %f nm\n", ir->rvdw_switch);
930 if (EPME_SWITCHED(ir->coulombtype))
932 fprintf(fp, " rlist : %f nm\n", ir->rlist);
934 if (ir->rlistlong != max_cutoff(ir->rvdw, ir->rcoulomb))
936 fprintf(fp, " rlistlong : %f nm\n", ir->rlistlong);
939 /* Print a descriptive line about the tpr settings tested */
940 fprintf(fp, "\nWill try these real/reciprocal workload settings:\n");
941 fprintf(fp, " No. scaling rcoulomb");
942 fprintf(fp, " nkx nky nkz");
943 fprintf(fp, " spacing");
944 if (can_scale_rvdw(ir->vdwtype))
946 fprintf(fp, " rvdw");
948 if (EPME_SWITCHED(ir->coulombtype))
950 fprintf(fp, " rlist");
952 if (ir->rlistlong != max_cutoff(ir->rlist, max_cutoff(ir->rvdw, ir->rcoulomb)) )
954 fprintf(fp, " rlistlong");
956 fprintf(fp, " tpr file\n");
958 /* Loop to create the requested number of tpr input files */
959 for (j = 0; j < *ntprs; j++)
961 /* The first .tpr is the provided one, just need to modify nsteps,
962 * so skip the following block */
965 /* Determine which Coulomb radii rc to use in the benchmarks */
966 add = (rmax-rmin)/(*ntprs-1);
967 if (gmx_within_tol(rmin, info->rcoulomb[0], GMX_REAL_EPS))
969 ir->rcoulomb = rmin + j*add;
971 else if (gmx_within_tol(rmax, info->rcoulomb[0], GMX_REAL_EPS))
973 ir->rcoulomb = rmin + (j-1)*add;
977 /* rmin != rcoul != rmax, ergo test between rmin and rmax */
978 add = (rmax-rmin)/(*ntprs-2);
979 ir->rcoulomb = rmin + (j-1)*add;
982 /* Determine the scaling factor fac */
983 fac = ir->rcoulomb/info->rcoulomb[0];
985 /* Scale the Fourier grid spacing */
986 ir->nkx = ir->nky = ir->nkz = 0;
987 calc_grid(NULL, state.box, fourierspacing*fac, &ir->nkx, &ir->nky, &ir->nkz);
989 /* Adjust other radii since various conditions need to be fulfilled */
990 if (eelPME == ir->coulombtype)
992 /* plain PME, rcoulomb must be equal to rlist */
993 ir->rlist = ir->rcoulomb;
997 /* rlist must be >= rcoulomb, we keep the size of the buffer region */
998 ir->rlist = ir->rcoulomb + nlist_buffer;
1001 if (bScaleRvdw && can_scale_rvdw(ir->vdwtype))
1003 if (ecutsVERLET == ir->cutoff_scheme ||
1004 evdwPME == ir->vdwtype)
1006 /* With either the Verlet cutoff-scheme or LJ-PME,
1007 the van der Waals radius must always equal the
1009 ir->rvdw = ir->rcoulomb;
1013 /* For vdw cutoff, rvdw >= rlist */
1014 ir->rvdw = max(info->rvdw[0], ir->rlist);
1018 ir->rlistlong = max_cutoff(ir->rlist, max_cutoff(ir->rvdw, ir->rcoulomb));
1020 } /* end of "if (j != 0)" */
1022 /* for j==0: Save the original settings
1023 * for j >0: Save modified radii and Fourier grids */
1024 info->rcoulomb[j] = ir->rcoulomb;
1025 info->rvdw[j] = ir->rvdw;
1026 info->nkx[j] = ir->nkx;
1027 info->nky[j] = ir->nky;
1028 info->nkz[j] = ir->nkz;
1029 info->rlist[j] = ir->rlist;
1030 info->rlistlong[j] = ir->rlistlong;
1031 info->fsx[j] = fac*fourierspacing;
1032 info->fsy[j] = fac*fourierspacing;
1033 info->fsz[j] = fac*fourierspacing;
1035 /* Write the benchmark tpr file */
1036 strncpy(fn_bench_tprs[j], fn_sim_tpr, strlen(fn_sim_tpr)-strlen(".tpr"));
1037 sprintf(buf, "_bench%.2d.tpr", j);
1038 strcat(fn_bench_tprs[j], buf);
1039 fprintf(stdout, "Writing benchmark tpr %s with nsteps=", fn_bench_tprs[j]);
1040 fprintf(stdout, "%"GMX_PRId64, ir->nsteps);
1043 fprintf(stdout, ", scaling factor %f\n", fac);
1047 fprintf(stdout, ", unmodified settings\n");
1050 write_tpx_state(fn_bench_tprs[j], ir, &state, &mtop);
1052 /* Write information about modified tpr settings to log file */
1053 fprintf(fp, "%4d%10f%10f", j, fac, ir->rcoulomb);
1054 fprintf(fp, "%5d%5d%5d", ir->nkx, ir->nky, ir->nkz);
1055 fprintf(fp, " %9f ", info->fsx[j]);
1056 if (can_scale_rvdw(ir->vdwtype))
1058 fprintf(fp, "%10f", ir->rvdw);
1060 if (EPME_SWITCHED(ir->coulombtype))
1062 fprintf(fp, "%10f", ir->rlist);
1064 if (info->rlistlong[0] != max_cutoff(info->rlist[0], max_cutoff(info->rvdw[0], info->rcoulomb[0])) )
1066 fprintf(fp, "%10f", ir->rlistlong);
1068 fprintf(fp, " %-14s\n", fn_bench_tprs[j]);
1070 /* Make it clear to the user that some additional settings were modified */
1071 if (!gmx_within_tol(ir->rvdw, info->rvdw[0], GMX_REAL_EPS)
1072 || !gmx_within_tol(ir->rlistlong, info->rlistlong[0], GMX_REAL_EPS) )
1079 fprintf(fp, "\nNote that in addition to the Coulomb radius and the Fourier grid\n"
1080 "other input settings were also changed (see table above).\n"
1081 "Please check if the modified settings are appropriate.\n");
1089 /* Rename the files we want to keep to some meaningful filename and
1090 * delete the rest */
1091 static void cleanup(const t_filenm *fnm, int nfile, int k, int nnodes,
1092 int nPMEnodes, int nr, gmx_bool bKeepStderr)
1094 char numstring[STRLEN];
1095 char newfilename[STRLEN];
1096 const char *fn = NULL;
1101 fprintf(stdout, "Cleaning up, deleting benchmark temp files ...\n");
1103 for (i = 0; i < nfile; i++)
1105 opt = (char *)fnm[i].opt;
1106 if (strcmp(opt, "-p") == 0)
1108 /* do nothing; keep this file */
1111 else if (strcmp(opt, "-bg") == 0)
1113 /* Give the log file a nice name so one can later see which parameters were used */
1114 numstring[0] = '\0';
1117 sprintf(numstring, "_%d", nr);
1119 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", opt2fn("-bg", nfile, fnm), k, nnodes, nPMEnodes, numstring);
1120 if (gmx_fexist(opt2fn("-bg", nfile, fnm)))
1122 fprintf(stdout, "renaming log file to %s\n", newfilename);
1123 make_backup(newfilename);
1124 rename(opt2fn("-bg", nfile, fnm), newfilename);
1127 else if (strcmp(opt, "-err") == 0)
1129 /* This file contains the output of stderr. We want to keep it in
1130 * cases where there have been problems. */
1131 fn = opt2fn(opt, nfile, fnm);
1132 numstring[0] = '\0';
1135 sprintf(numstring, "_%d", nr);
1137 sprintf(newfilename, "%s_no%d_np%d_npme%d%s", fn, k, nnodes, nPMEnodes, numstring);
1142 fprintf(stdout, "Saving stderr output in %s\n", newfilename);
1143 make_backup(newfilename);
1144 rename(fn, newfilename);
1148 fprintf(stdout, "Deleting %s\n", fn);
1153 /* Delete the files which are created for each benchmark run: (options -b*) */
1154 else if ( (0 == strncmp(opt, "-b", 2)) && (opt2bSet(opt, nfile, fnm) || !is_optional(&fnm[i])) )
1156 remove_if_exists(opt2fn(opt, nfile, fnm));
1163 eNpmeAuto, eNpmeAll, eNpmeReduced, eNpmeSubset, eNpmeNr
1166 /* Create a list of numbers of PME nodes to test */
1167 static void make_npme_list(
1168 const char *npmevalues_opt, /* Make a complete list with all
1169 * possibilities or a short list that keeps only
1170 * reasonable numbers of PME nodes */
1171 int *nentries, /* Number of entries we put in the nPMEnodes list */
1172 int *nPMEnodes[], /* Each entry contains the value for -npme */
1173 int nnodes, /* Total number of nodes to do the tests on */
1174 int minPMEnodes, /* Minimum number of PME nodes */
1175 int maxPMEnodes) /* Maximum number of PME nodes */
1178 int min_factor = 1; /* We request that npp and npme have this minimal
1179 * largest common factor (depends on npp) */
1180 int nlistmax; /* Max. list size */
1181 int nlist; /* Actual number of entries in list */
1185 /* Do we need to check all possible values for -npme or is a reduced list enough? */
1186 if (0 == strcmp(npmevalues_opt, "all") )
1190 else if (0 == strcmp(npmevalues_opt, "subset") )
1192 eNPME = eNpmeSubset;
1194 else /* "auto" or "range" */
1200 else if (nnodes < 128)
1202 eNPME = eNpmeReduced;
1206 eNPME = eNpmeSubset;
1210 /* Calculate how many entries we could possibly have (in case of -npme all) */
1213 nlistmax = maxPMEnodes - minPMEnodes + 3;
1214 if (0 == minPMEnodes)
1224 /* Now make the actual list which is at most of size nlist */
1225 snew(*nPMEnodes, nlistmax);
1226 nlist = 0; /* start counting again, now the real entries in the list */
1227 for (i = 0; i < nlistmax - 2; i++)
1229 npme = maxPMEnodes - i;
1240 /* For 2d PME we want a common largest factor of at least the cube
1241 * root of the number of PP nodes */
1242 min_factor = (int) pow(npp, 1.0/3.0);
1245 gmx_fatal(FARGS, "Unknown option for eNPME in make_npme_list");
1248 if (gmx_greatest_common_divisor(npp, npme) >= min_factor)
1250 (*nPMEnodes)[nlist] = npme;
1254 /* We always test 0 PME nodes and the automatic number */
1255 *nentries = nlist + 2;
1256 (*nPMEnodes)[nlist ] = 0;
1257 (*nPMEnodes)[nlist+1] = -1;
1259 fprintf(stderr, "Will try the following %d different values for -npme:\n", *nentries);
1260 for (i = 0; i < *nentries-1; i++)
1262 fprintf(stderr, "%d, ", (*nPMEnodes)[i]);
1264 fprintf(stderr, "and %d (auto).\n", (*nPMEnodes)[*nentries-1]);
1268 /* Allocate memory to store the performance data */
1269 static void init_perfdata(t_perf *perfdata[], int ntprs, int datasets, int repeats)
1274 for (k = 0; k < ntprs; k++)
1276 snew(perfdata[k], datasets);
1277 for (i = 0; i < datasets; i++)
1279 for (j = 0; j < repeats; j++)
1281 snew(perfdata[k][i].Gcycles, repeats);
1282 snew(perfdata[k][i].ns_per_day, repeats);
1283 snew(perfdata[k][i].PME_f_load, repeats);
1290 /* Check for errors on mdrun -h */
1291 static void make_sure_it_runs(char *mdrun_cmd_line, int length, FILE *fp,
1292 const t_filenm *fnm, int nfile)
1294 const char *fn = NULL;
1295 char *command, *msg;
1299 snew(command, length + 15);
1300 snew(msg, length + 500);
1302 fprintf(stdout, "Making sure the benchmarks can be executed by running just 1 step...\n");
1303 sprintf(command, "%s -nsteps 1 -quiet", mdrun_cmd_line);
1304 fprintf(stdout, "Executing '%s' ...\n", command);
1305 ret = gmx_system_call(command);
1309 /* To prevent confusion, do not again issue a gmx_fatal here since we already
1310 * get the error message from mdrun itself */
1312 "Cannot run the first benchmark simulation! Please check the error message of\n"
1313 "mdrun for the source of the problem. Did you provide a command line\n"
1314 "argument that neither gmx tune_pme nor mdrun understands? If you're\n"
1315 "sure your command line should work, you can bypass this check with \n"
1316 "gmx tune_pme -nocheck. The failing command was:\n"
1317 "\n%s\n\n", command);
1319 fprintf(stderr, "%s", msg);
1321 fprintf(fp, "%s", msg);
1325 fprintf(stdout, "Benchmarks can be executed!\n");
1327 /* Clean up the benchmark output files we just created */
1328 fprintf(stdout, "Cleaning up ...\n");
1329 remove_if_exists(opt2fn("-bc", nfile, fnm));
1330 remove_if_exists(opt2fn("-be", nfile, fnm));
1331 remove_if_exists(opt2fn("-bcpo", nfile, fnm));
1332 remove_if_exists(opt2fn("-bg", nfile, fnm));
1339 static void do_the_tests(
1340 FILE *fp, /* General g_tune_pme output file */
1341 char **tpr_names, /* Filenames of the input files to test */
1342 int maxPMEnodes, /* Max fraction of nodes to use for PME */
1343 int minPMEnodes, /* Min fraction of nodes to use for PME */
1344 int npme_fixed, /* If >= -1, test fixed number of PME
1346 const char *npmevalues_opt, /* Which -npme values should be tested */
1347 t_perf **perfdata, /* Here the performance data is stored */
1348 int *pmeentries, /* Entries in the nPMEnodes list */
1349 int repeats, /* Repeat each test this often */
1350 int nnodes, /* Total number of nodes = nPP + nPME */
1351 int nr_tprs, /* Total number of tpr files to test */
1352 gmx_bool bThreads, /* Threads or MPI? */
1353 char *cmd_mpirun, /* mpirun command string */
1354 char *cmd_np, /* "-np", "-n", whatever mpirun needs */
1355 char *cmd_mdrun, /* mdrun command string */
1356 char *cmd_args_bench, /* arguments for mdrun in a string */
1357 const t_filenm *fnm, /* List of filenames from command line */
1358 int nfile, /* Number of files specified on the cmdl. */
1359 int presteps, /* DLB equilibration steps, is checked */
1360 gmx_int64_t cpt_steps, /* Time step counter in the checkpoint */
1361 gmx_bool bCheck) /* Check whether benchmark mdrun works */
1363 int i, nr, k, ret, count = 0, totaltests;
1364 int *nPMEnodes = NULL;
1367 char *command, *cmd_stub;
1369 gmx_bool bResetProblem = FALSE;
1370 gmx_bool bFirst = TRUE;
1373 /* This string array corresponds to the eParselog enum type at the start
1375 const char* ParseLog[] = {
1377 "Logfile not found!",
1378 "No timings, logfile truncated?",
1379 "Run was terminated.",
1380 "Counters were not reset properly.",
1381 "No DD grid found for these settings.",
1382 "TPX version conflict!",
1383 "mdrun was not started in parallel!",
1384 "Number of PP ranks has a prime factor that is too large.",
1387 char str_PME_f_load[13];
1390 /* Allocate space for the mdrun command line. 100 extra characters should
1391 be more than enough for the -npme etcetera arguments */
1392 cmdline_length = strlen(cmd_mpirun)
1395 + strlen(cmd_args_bench)
1396 + strlen(tpr_names[0]) + 100;
1397 snew(command, cmdline_length);
1398 snew(cmd_stub, cmdline_length);
1400 /* Construct the part of the command line that stays the same for all tests: */
1403 sprintf(cmd_stub, "%s%s", cmd_mdrun, cmd_np);
1407 sprintf(cmd_stub, "%s%s%s ", cmd_mpirun, cmd_np, cmd_mdrun);
1410 /* Create a list of numbers of PME nodes to test */
1411 if (npme_fixed < -1)
1413 make_npme_list(npmevalues_opt, pmeentries, &nPMEnodes,
1414 nnodes, minPMEnodes, maxPMEnodes);
1420 nPMEnodes[0] = npme_fixed;
1421 fprintf(stderr, "Will use a fixed number of %d PME-only ranks.\n", nPMEnodes[0]);
1426 fprintf(fp, "\nNo benchmarks done since number of repeats (-r) is 0.\n");
1428 finalize(opt2fn("-p", nfile, fnm));
1432 /* Allocate one dataset for each tpr input file: */
1433 init_perfdata(perfdata, nr_tprs, *pmeentries, repeats);
1435 /*****************************************/
1436 /* Main loop over all tpr files to test: */
1437 /*****************************************/
1438 totaltests = nr_tprs*(*pmeentries)*repeats;
1439 for (k = 0; k < nr_tprs; k++)
1441 fprintf(fp, "\nIndividual timings for input file %d (%s):\n", k, tpr_names[k]);
1442 fprintf(fp, "PME ranks Gcycles ns/day PME/f Remark\n");
1443 /* Loop over various numbers of PME nodes: */
1444 for (i = 0; i < *pmeentries; i++)
1446 pd = &perfdata[k][i];
1448 /* Loop over the repeats for each scenario: */
1449 for (nr = 0; nr < repeats; nr++)
1451 pd->nPMEnodes = nPMEnodes[i];
1453 /* Add -npme and -s to the command line and save it. Note that
1454 * the -passall (if set) options requires cmd_args_bench to be
1455 * at the end of the command line string */
1456 snew(pd->mdrun_cmd_line, cmdline_length);
1457 sprintf(pd->mdrun_cmd_line, "%s-npme %d -s %s %s",
1458 cmd_stub, pd->nPMEnodes, tpr_names[k], cmd_args_bench);
1460 /* To prevent that all benchmarks fail due to a show-stopper argument
1461 * on the mdrun command line, we make a quick check first.
1462 * This check can be turned off in cases where the automatically chosen
1463 * number of PME-only ranks leads to a number of PP ranks for which no
1464 * decomposition can be found (e.g. for large prime numbers) */
1465 if (bFirst && bCheck)
1467 make_sure_it_runs(pd->mdrun_cmd_line, cmdline_length, fp, fnm, nfile);
1471 /* Do a benchmark simulation: */
1474 sprintf(buf, ", pass %d/%d", nr+1, repeats);
1480 fprintf(stdout, "\n=== Progress %2.0f%%, tpr %d/%d, run %d/%d%s:\n",
1481 (100.0*count)/totaltests,
1482 k+1, nr_tprs, i+1, *pmeentries, buf);
1483 make_backup(opt2fn("-err", nfile, fnm));
1484 sprintf(command, "%s 1> /dev/null 2>%s", pd->mdrun_cmd_line, opt2fn("-err", nfile, fnm));
1485 fprintf(stdout, "%s\n", pd->mdrun_cmd_line);
1486 gmx_system_call(command);
1488 /* Collect the performance data from the log file; also check stderr
1489 * for fatal errors */
1490 ret = parse_logfile(opt2fn("-bg", nfile, fnm), opt2fn("-err", nfile, fnm),
1491 pd, nr, presteps, cpt_steps, nnodes);
1492 if ((presteps > 0) && (ret == eParselogResetProblem))
1494 bResetProblem = TRUE;
1497 if (-1 == pd->nPMEnodes)
1499 sprintf(buf, "(%3d)", pd->guessPME);
1507 if (pd->PME_f_load[nr] > 0.0)
1509 sprintf(str_PME_f_load, "%12.3f", pd->PME_f_load[nr]);
1513 sprintf(str_PME_f_load, "%s", " - ");
1516 /* Write the data we got to disk */
1517 fprintf(fp, "%4d%s %12.3f %12.3f %s %s", pd->nPMEnodes,
1518 buf, pd->Gcycles[nr], pd->ns_per_day[nr], str_PME_f_load, ParseLog[ret]);
1519 if (!(ret == eParselogOK || ret == eParselogNoDDGrid || ret == eParselogNotFound) )
1521 fprintf(fp, " Check %s file for problems.", ret == eParselogFatal ? "err" : "log");
1527 /* Do some cleaning up and delete the files we do not need any more */
1528 cleanup(fnm, nfile, k, nnodes, pd->nPMEnodes, nr, ret == eParselogFatal);
1530 /* If the first run with this number of processors already failed, do not try again: */
1531 if (pd->Gcycles[0] <= 0.0 && repeats > 1)
1533 fprintf(stdout, "Skipping remaining passes of unsuccessful setting, see log file for details.\n");
1534 count += repeats-(nr+1);
1537 } /* end of repeats loop */
1538 } /* end of -npme loop */
1539 } /* end of tpr file loop */
1544 fprintf(fp, "WARNING: The cycle and time step counters could not be reset properly. ");
1552 static void check_input(
1559 real maxPMEfraction,
1560 real minPMEfraction,
1562 gmx_int64_t bench_nsteps,
1563 const t_filenm *fnm,
1573 /* Make sure the input file exists */
1574 if (!gmx_fexist(opt2fn("-s", nfile, fnm)))
1576 gmx_fatal(FARGS, "File %s not found.", opt2fn("-s", nfile, fnm));
1579 /* Make sure that the checkpoint file is not overwritten during benchmarking */
1580 if ( (0 == strcmp(opt2fn("-cpi", nfile, fnm), opt2fn("-bcpo", nfile, fnm)) ) && (sim_part > 1) )
1582 gmx_fatal(FARGS, "Checkpoint input (-cpi) and benchmark checkpoint output (-bcpo) files must not be identical.\n"
1583 "The checkpoint input file must not be overwritten during the benchmarks.\n");
1586 /* Make sure that repeats is >= 0 (if == 0, only write tpr files) */
1589 gmx_fatal(FARGS, "Number of repeats < 0!");
1592 /* Check number of nodes */
1595 gmx_fatal(FARGS, "Number of ranks/threads must be a positive integer.");
1598 /* Automatically choose -ntpr if not set */
1608 /* Set a reasonable scaling factor for rcoulomb */
1611 *rmax = rcoulomb * 1.2;
1614 fprintf(stderr, "Will test %d tpr file%s.\n", *ntprs, *ntprs == 1 ? "" : "s");
1620 fprintf(stderr, "Note: Choose ntpr>1 to shift PME load between real and reciprocal space.\n");
1624 /* Make shure that rmin <= rcoulomb <= rmax */
1633 if (!(*rmin <= *rmax) )
1635 gmx_fatal(FARGS, "Please choose the Coulomb radii such that rmin <= rmax.\n"
1636 "rmin = %g, rmax = %g, actual rcoul from .tpr file = %g\n", *rmin, *rmax, rcoulomb);
1638 /* Add test scenarios if rmin or rmax were set */
1641 if (!gmx_within_tol(*rmin, rcoulomb, GMX_REAL_EPS) && (*ntprs == 1) )
1644 fprintf(stderr, "NOTE: Setting -rmin to %g changed -ntpr to %d\n",
1647 if (!gmx_within_tol(*rmax, rcoulomb, GMX_REAL_EPS) && (*ntprs == 1) )
1650 fprintf(stderr, "NOTE: Setting -rmax to %g changed -ntpr to %d\n",
1655 /* If one of rmin, rmax is set, we need 2 tpr files at minimum */
1656 if (!gmx_within_tol(*rmax, rcoulomb, GMX_REAL_EPS) || !gmx_within_tol(*rmin, rcoulomb, GMX_REAL_EPS) )
1658 *ntprs = max(*ntprs, 2);
1661 /* If both rmin, rmax are set, we need 3 tpr files at minimum */
1662 if (!gmx_within_tol(*rmax, rcoulomb, GMX_REAL_EPS) && !gmx_within_tol(*rmin, rcoulomb, GMX_REAL_EPS) )
1664 *ntprs = max(*ntprs, 3);
1669 fprintf(stderr, "NOTE: Your rmin, rmax setting changed -ntpr to %d\n", *ntprs);
1674 if (gmx_within_tol(*rmin, rcoulomb, GMX_REAL_EPS) && gmx_within_tol(rcoulomb, *rmax, GMX_REAL_EPS)) /* We have just a single rc */
1676 fprintf(stderr, "WARNING: Resetting -ntpr to 1 since no Coulomb radius scaling is requested.\n"
1677 "Please set rmin < rmax to test Coulomb radii in the [rmin, rmax] interval\n"
1678 "with correspondingly adjusted PME grid settings\n");
1683 /* Check whether max and min fraction are within required values */
1684 if (maxPMEfraction > 0.5 || maxPMEfraction < 0)
1686 gmx_fatal(FARGS, "-max must be between 0 and 0.5");
1688 if (minPMEfraction > 0.5 || minPMEfraction < 0)
1690 gmx_fatal(FARGS, "-min must be between 0 and 0.5");
1692 if (maxPMEfraction < minPMEfraction)
1694 gmx_fatal(FARGS, "-max must be larger or equal to -min");
1697 /* Check whether the number of steps - if it was set - has a reasonable value */
1698 if (bench_nsteps < 0)
1700 gmx_fatal(FARGS, "Number of steps must be positive.");
1703 if (bench_nsteps > 10000 || bench_nsteps < 100)
1705 fprintf(stderr, "WARNING: steps=");
1706 fprintf(stderr, "%"GMX_PRId64, bench_nsteps);
1707 fprintf(stderr, ". Are you sure you want to perform so %s steps for each benchmark?\n", (bench_nsteps < 100) ? "few" : "many");
1712 gmx_fatal(FARGS, "Cannot have a negative number of presteps.\n");
1715 /* Check for rcoulomb scaling if more than one .tpr file is tested */
1718 if (*rmin/rcoulomb < 0.75 || *rmax/rcoulomb > 1.25)
1720 fprintf(stderr, "WARNING: Applying extreme scaling factor. I hope you know what you are doing.\n");
1724 /* If a fixed number of PME nodes is set we do rcoulomb and PME gird tuning
1725 * only. We need to check whether the requested number of PME-only nodes
1727 if (npme_fixed > -1)
1729 /* No more than 50% of all nodes can be assigned as PME-only nodes. */
1730 if (2*npme_fixed > nnodes)
1732 gmx_fatal(FARGS, "Cannot have more than %d PME-only ranks for a total of %d ranks (you chose %d).\n",
1733 nnodes/2, nnodes, npme_fixed);
1735 if ((npme_fixed > 0) && (5*npme_fixed < nnodes))
1737 fprintf(stderr, "WARNING: Only %g percent of the ranks are assigned as PME-only ranks.\n",
1738 100.0*((real)npme_fixed / (real)nnodes));
1740 if (opt2parg_bSet("-min", npargs, pa) || opt2parg_bSet("-max", npargs, pa))
1742 fprintf(stderr, "NOTE: The -min, -max, and -npme options have no effect when a\n"
1743 " fixed number of PME-only ranks is requested with -fix.\n");
1749 /* Returns TRUE when "opt" is needed at launch time */
1750 static gmx_bool is_launch_file(char *opt, gmx_bool bSet)
1752 if (0 == strncmp(opt, "-swap", 5))
1757 /* Apart from the input .tpr and the output log files we need all options that
1758 * were set on the command line and that do not start with -b */
1759 if (0 == strncmp(opt, "-b", 2) || 0 == strncmp(opt, "-s", 2)
1760 || 0 == strncmp(opt, "-err", 4) || 0 == strncmp(opt, "-p", 2) )
1769 /* Returns TRUE when "opt" defines a file which is needed for the benchmarks runs */
1770 static gmx_bool is_bench_file(char *opt, gmx_bool bSet, gmx_bool bOptional, gmx_bool bIsOutput)
1772 /* Apart from the input .tpr, all files starting with "-b" are for
1773 * _b_enchmark files exclusively */
1774 if (0 == strncmp(opt, "-s", 2))
1779 if (0 == strncmp(opt, "-b", 2) || 0 == strncmp(opt, "-s", 2))
1781 if (!bOptional || bSet)
1798 if (bSet) /* These are additional input files like -cpi -ei */
1811 /* Adds 'buf' to 'str' */
1812 static void add_to_string(char **str, char *buf)
1817 len = strlen(*str) + strlen(buf) + 1;
1823 /* Create the command line for the benchmark as well as for the real run */
1824 static void create_command_line_snippets(
1825 gmx_bool bAppendFiles,
1826 gmx_bool bKeepAndNumCPT,
1827 gmx_bool bResetHWay,
1831 char *cmd_args_bench[], /* command line arguments for benchmark runs */
1832 char *cmd_args_launch[], /* command line arguments for simulation run */
1833 char extra_args[]) /* Add this to the end of the command line */
1838 char strbuf[STRLEN];
1841 /* strlen needs at least '\0' as a string: */
1842 snew(*cmd_args_bench, 1);
1843 snew(*cmd_args_launch, 1);
1844 *cmd_args_launch[0] = '\0';
1845 *cmd_args_bench[0] = '\0';
1848 /*******************************************/
1849 /* 1. Process other command line arguments */
1850 /*******************************************/
1853 /* Add equilibration steps to benchmark options */
1854 sprintf(strbuf, "-resetstep %d ", presteps);
1855 add_to_string(cmd_args_bench, strbuf);
1857 /* These switches take effect only at launch time */
1858 if (FALSE == bAppendFiles)
1860 add_to_string(cmd_args_launch, "-noappend ");
1864 add_to_string(cmd_args_launch, "-cpnum ");
1868 add_to_string(cmd_args_launch, "-resethway ");
1871 /********************/
1872 /* 2. Process files */
1873 /********************/
1874 for (i = 0; i < nfile; i++)
1876 opt = (char *)fnm[i].opt;
1877 name = opt2fn(opt, nfile, fnm);
1879 /* Strbuf contains the options, now let's sort out where we need that */
1880 sprintf(strbuf, "%s %s ", opt, name);
1882 if (is_bench_file(opt, opt2bSet(opt, nfile, fnm), is_optional(&fnm[i]), is_output(&fnm[i])) )
1884 /* All options starting with -b* need the 'b' removed,
1885 * therefore overwrite strbuf */
1886 if (0 == strncmp(opt, "-b", 2))
1888 sprintf(strbuf, "-%s %s ", &opt[2], name);
1891 add_to_string(cmd_args_bench, strbuf);
1894 if (is_launch_file(opt, opt2bSet(opt, nfile, fnm)) )
1896 add_to_string(cmd_args_launch, strbuf);
1900 add_to_string(cmd_args_bench, extra_args);
1901 add_to_string(cmd_args_launch, extra_args);
1905 /* Set option opt */
1906 static void setopt(const char *opt, int nfile, t_filenm fnm[])
1910 for (i = 0; (i < nfile); i++)
1912 if (strcmp(opt, fnm[i].opt) == 0)
1914 fnm[i].flag |= ffSET;
1920 /* This routine inspects the tpr file and ...
1921 * 1. checks for output files that get triggered by a tpr option. These output
1922 * files are marked as 'set' to allow for a proper cleanup after each
1924 * 2. returns the PME:PP load ratio
1925 * 3. returns rcoulomb from the tpr */
1926 static float inspect_tpr(int nfile, t_filenm fnm[], real *rcoulomb)
1928 gmx_bool bPull; /* Is pulling requested in .tpr file? */
1929 gmx_bool bTpi; /* Is test particle insertion requested? */
1930 gmx_bool bFree; /* Is a free energy simulation requested? */
1931 gmx_bool bNM; /* Is a normal mode analysis requested? */
1932 gmx_bool bSwap; /* Is water/ion position swapping requested? */
1938 /* Check tpr file for options that trigger extra output files */
1939 read_tpx_state(opt2fn("-s", nfile, fnm), &ir, &state, NULL, &mtop);
1940 bPull = (epullNO != ir.ePull);
1941 bFree = (efepNO != ir.efep );
1942 bNM = (eiNM == ir.eI );
1943 bSwap = (eswapNO != ir.eSwapCoords);
1944 bTpi = EI_TPI(ir.eI);
1946 /* Set these output files on the tuning command-line */
1949 setopt("-pf", nfile, fnm);
1950 setopt("-px", nfile, fnm);
1954 setopt("-dhdl", nfile, fnm);
1958 setopt("-tpi", nfile, fnm);
1959 setopt("-tpid", nfile, fnm);
1963 setopt("-mtx", nfile, fnm);
1967 setopt("-swap", nfile, fnm);
1970 *rcoulomb = ir.rcoulomb;
1972 /* Return the estimate for the number of PME nodes */
1973 return pme_load_estimate(&mtop, &ir, state.box);
1977 static void couple_files_options(int nfile, t_filenm fnm[])
1980 gmx_bool bSet, bBench;
1985 for (i = 0; i < nfile; i++)
1987 opt = (char *)fnm[i].opt;
1988 bSet = ((fnm[i].flag & ffSET) != 0);
1989 bBench = (0 == strncmp(opt, "-b", 2));
1991 /* Check optional files */
1992 /* If e.g. -eo is set, then -beo also needs to be set */
1993 if (is_optional(&fnm[i]) && bSet && !bBench)
1995 sprintf(buf, "-b%s", &opt[1]);
1996 setopt(buf, nfile, fnm);
1998 /* If -beo is set, then -eo also needs to be! */
1999 if (is_optional(&fnm[i]) && bSet && bBench)
2001 sprintf(buf, "-%s", &opt[2]);
2002 setopt(buf, nfile, fnm);
2008 #define BENCHSTEPS (1000)
2010 int gmx_tune_pme(int argc, char *argv[])
2012 const char *desc[] = {
2013 "For a given number [TT]-np[tt] or [TT]-ntmpi[tt] of ranks, [THISMODULE] systematically",
2014 "times [gmx-mdrun] with various numbers of PME-only ranks and determines",
2015 "which setting is fastest. It will also test whether performance can",
2016 "be enhanced by shifting load from the reciprocal to the real space",
2017 "part of the Ewald sum. ",
2018 "Simply pass your [TT].tpr[tt] file to [THISMODULE] together with other options",
2019 "for [gmx-mdrun] as needed.[PAR]",
2020 "Which executables are used can be set in the environment variables",
2021 "MPIRUN and MDRUN. If these are not present, 'mpirun' and 'mdrun'",
2022 "will be used as defaults. Note that for certain MPI frameworks you",
2023 "need to provide a machine- or hostfile. This can also be passed",
2024 "via the MPIRUN variable, e.g.[PAR]",
2025 "[TT]export MPIRUN=\"/usr/local/mpirun -machinefile hosts\"[tt][PAR]",
2026 "Before doing the actual benchmark runs, [THISMODULE] will do a quick",
2027 "check whether mdrun works as expected with the provided parallel settings",
2028 "if the [TT]-check[tt] option is activated (the default).",
2029 "Please call [THISMODULE] with the normal options you would pass to",
2030 "[gmx-mdrun] and add [TT]-np[tt] for the number of ranks to perform the",
2031 "tests on, or [TT]-ntmpi[tt] for the number of threads. You can also add [TT]-r[tt]",
2032 "to repeat each test several times to get better statistics. [PAR]",
2033 "[THISMODULE] can test various real space / reciprocal space workloads",
2034 "for you. With [TT]-ntpr[tt] you control how many extra [TT].tpr[tt] files will be",
2035 "written with enlarged cutoffs and smaller Fourier grids respectively.",
2036 "Typically, the first test (number 0) will be with the settings from the input",
2037 "[TT].tpr[tt] file; the last test (number [TT]ntpr[tt]) will have the Coulomb cutoff",
2038 "specified by [TT]-rmax[tt] with a somewhat smaller PME grid at the same time. ",
2039 "In this last test, the Fourier spacing is multiplied with [TT]rmax[tt]/rcoulomb. ",
2040 "The remaining [TT].tpr[tt] files will have equally-spaced Coulomb radii (and Fourier "
2041 "spacings) between these extremes. [BB]Note[bb] that you can set [TT]-ntpr[tt] to 1",
2042 "if you just seek the optimal number of PME-only ranks; in that case",
2043 "your input [TT].tpr[tt] file will remain unchanged.[PAR]",
2044 "For the benchmark runs, the default of 1000 time steps should suffice for most",
2045 "MD systems. The dynamic load balancing needs about 100 time steps",
2046 "to adapt to local load imbalances, therefore the time step counters",
2047 "are by default reset after 100 steps. For large systems (>1M atoms), as well as ",
2048 "for a higher accuracy of the measurements, you should set [TT]-resetstep[tt] to a higher value.",
2049 "From the 'DD' load imbalance entries in the md.log output file you",
2050 "can tell after how many steps the load is sufficiently balanced. Example call:[PAR]"
2051 "[TT]gmx tune_pme -np 64 -s protein.tpr -launch[tt][PAR]",
2052 "After calling [gmx-mdrun] several times, detailed performance information",
2053 "is available in the output file [TT]perf.out[tt].",
2054 "[BB]Note[bb] that during the benchmarks, a couple of temporary files are written",
2055 "(options [TT]-b[tt]*), these will be automatically deleted after each test.[PAR]",
2056 "If you want the simulation to be started automatically with the",
2057 "optimized parameters, use the command line option [TT]-launch[tt].[PAR]",
2062 int pmeentries = 0; /* How many values for -npme do we actually test for each tpr file */
2063 real maxPMEfraction = 0.50;
2064 real minPMEfraction = 0.25;
2065 int maxPMEnodes, minPMEnodes;
2066 float guessPMEratio; /* guessed PME:PP ratio based on the tpr file */
2067 float guessPMEnodes;
2068 int npme_fixed = -2; /* If >= -1, use only this number
2069 * of PME-only nodes */
2071 real rmin = 0.0, rmax = 0.0; /* min and max value for rcoulomb if scaling is requested */
2072 real rcoulomb = -1.0; /* Coulomb radius as set in .tpr file */
2073 gmx_bool bScaleRvdw = TRUE;
2074 gmx_int64_t bench_nsteps = BENCHSTEPS;
2075 gmx_int64_t new_sim_nsteps = -1; /* -1 indicates: not set by the user */
2076 gmx_int64_t cpt_steps = 0; /* Step counter in .cpt input file */
2077 int presteps = 100; /* Do a full cycle reset after presteps steps */
2078 gmx_bool bOverwrite = FALSE, bKeepTPR;
2079 gmx_bool bLaunch = FALSE;
2080 char *ExtraArgs = NULL;
2081 char **tpr_names = NULL;
2082 const char *simulation_tpr = NULL;
2083 int best_npme, best_tpr;
2084 int sim_part = 1; /* For benchmarks with checkpoint files */
2087 /* Default program names if nothing else is found */
2088 char *cmd_mpirun = NULL, *cmd_mdrun = NULL;
2089 char *cmd_args_bench, *cmd_args_launch;
2090 char *cmd_np = NULL;
2092 t_perf **perfdata = NULL;
2098 /* Print out how long the tuning took */
2101 static t_filenm fnm[] = {
2103 { efOUT, "-p", "perf", ffWRITE },
2104 { efLOG, "-err", "bencherr", ffWRITE },
2105 { efTPR, "-so", "tuned", ffWRITE },
2107 { efTPR, NULL, NULL, ffREAD },
2108 { efTRN, "-o", NULL, ffWRITE },
2109 { efCOMPRESSED, "-x", NULL, ffOPTWR },
2110 { efCPT, "-cpi", NULL, ffOPTRD },
2111 { efCPT, "-cpo", NULL, ffOPTWR },
2112 { efSTO, "-c", "confout", ffWRITE },
2113 { efEDR, "-e", "ener", ffWRITE },
2114 { efLOG, "-g", "md", ffWRITE },
2115 { efXVG, "-dhdl", "dhdl", ffOPTWR },
2116 { efXVG, "-field", "field", ffOPTWR },
2117 { efXVG, "-table", "table", ffOPTRD },
2118 { efXVG, "-tabletf", "tabletf", ffOPTRD },
2119 { efXVG, "-tablep", "tablep", ffOPTRD },
2120 { efXVG, "-tableb", "table", ffOPTRD },
2121 { efTRX, "-rerun", "rerun", ffOPTRD },
2122 { efXVG, "-tpi", "tpi", ffOPTWR },
2123 { efXVG, "-tpid", "tpidist", ffOPTWR },
2124 { efEDI, "-ei", "sam", ffOPTRD },
2125 { efXVG, "-eo", "edsam", ffOPTWR },
2126 { efXVG, "-devout", "deviatie", ffOPTWR },
2127 { efXVG, "-runav", "runaver", ffOPTWR },
2128 { efXVG, "-px", "pullx", ffOPTWR },
2129 { efXVG, "-pf", "pullf", ffOPTWR },
2130 { efXVG, "-ro", "rotation", ffOPTWR },
2131 { efLOG, "-ra", "rotangles", ffOPTWR },
2132 { efLOG, "-rs", "rotslabs", ffOPTWR },
2133 { efLOG, "-rt", "rottorque", ffOPTWR },
2134 { efMTX, "-mtx", "nm", ffOPTWR },
2135 { efNDX, "-dn", "dipole", ffOPTWR },
2136 { efXVG, "-swap", "swapions", ffOPTWR },
2137 /* Output files that are deleted after each benchmark run */
2138 { efTRN, "-bo", "bench", ffWRITE },
2139 { efXTC, "-bx", "bench", ffWRITE },
2140 { efCPT, "-bcpo", "bench", ffWRITE },
2141 { efSTO, "-bc", "bench", ffWRITE },
2142 { efEDR, "-be", "bench", ffWRITE },
2143 { efLOG, "-bg", "bench", ffWRITE },
2144 { efXVG, "-beo", "benchedo", ffOPTWR },
2145 { efXVG, "-bdhdl", "benchdhdl", ffOPTWR },
2146 { efXVG, "-bfield", "benchfld", ffOPTWR },
2147 { efXVG, "-btpi", "benchtpi", ffOPTWR },
2148 { efXVG, "-btpid", "benchtpid", ffOPTWR },
2149 { efXVG, "-bdevout", "benchdev", ffOPTWR },
2150 { efXVG, "-brunav", "benchrnav", ffOPTWR },
2151 { efXVG, "-bpx", "benchpx", ffOPTWR },
2152 { efXVG, "-bpf", "benchpf", ffOPTWR },
2153 { efXVG, "-bro", "benchrot", ffOPTWR },
2154 { efLOG, "-bra", "benchrota", ffOPTWR },
2155 { efLOG, "-brs", "benchrots", ffOPTWR },
2156 { efLOG, "-brt", "benchrott", ffOPTWR },
2157 { efMTX, "-bmtx", "benchn", ffOPTWR },
2158 { efNDX, "-bdn", "bench", ffOPTWR },
2159 { efXVG, "-bswap", "benchswp", ffOPTWR }
2162 gmx_bool bThreads = FALSE;
2166 const char *procstring[] =
2167 { NULL, "-np", "-n", "none", NULL };
2168 const char *npmevalues_opt[] =
2169 { NULL, "auto", "all", "subset", NULL };
2171 gmx_bool bAppendFiles = TRUE;
2172 gmx_bool bKeepAndNumCPT = FALSE;
2173 gmx_bool bResetCountersHalfWay = FALSE;
2174 gmx_bool bBenchmark = TRUE;
2175 gmx_bool bCheck = TRUE;
2177 output_env_t oenv = NULL;
2180 /***********************/
2181 /* g_tune_pme options: */
2182 /***********************/
2183 { "-np", FALSE, etINT, {&nnodes},
2184 "Number of ranks to run the tests on (must be > 2 for separate PME ranks)" },
2185 { "-npstring", FALSE, etENUM, {procstring},
2186 "Specify the number of ranks to [TT]$MPIRUN[tt] using this string"},
2187 { "-ntmpi", FALSE, etINT, {&nthreads},
2188 "Number of MPI-threads to run the tests on (turns MPI & mpirun off)"},
2189 { "-r", FALSE, etINT, {&repeats},
2190 "Repeat each test this often" },
2191 { "-max", FALSE, etREAL, {&maxPMEfraction},
2192 "Max fraction of PME ranks to test with" },
2193 { "-min", FALSE, etREAL, {&minPMEfraction},
2194 "Min fraction of PME ranks to test with" },
2195 { "-npme", FALSE, etENUM, {npmevalues_opt},
2196 "Within -min and -max, benchmark all possible values for [TT]-npme[tt], or just a reasonable subset. "
2197 "Auto neglects -min and -max and chooses reasonable values around a guess for npme derived from the .tpr"},
2198 { "-fix", FALSE, etINT, {&npme_fixed},
2199 "If >= -1, do not vary the number of PME-only ranks, instead use this fixed value and only vary rcoulomb and the PME grid spacing."},
2200 { "-rmax", FALSE, etREAL, {&rmax},
2201 "If >0, maximal rcoulomb for -ntpr>1 (rcoulomb upscaling results in fourier grid downscaling)" },
2202 { "-rmin", FALSE, etREAL, {&rmin},
2203 "If >0, minimal rcoulomb for -ntpr>1" },
2204 { "-scalevdw", FALSE, etBOOL, {&bScaleRvdw},
2205 "Scale rvdw along with rcoulomb"},
2206 { "-ntpr", FALSE, etINT, {&ntprs},
2207 "Number of [TT].tpr[tt] files to benchmark. Create this many files with different rcoulomb scaling factors depending on -rmin and -rmax. "
2208 "If < 1, automatically choose the number of [TT].tpr[tt] files to test" },
2209 { "-steps", FALSE, etINT64, {&bench_nsteps},
2210 "Take timings for this many steps in the benchmark runs" },
2211 { "-resetstep", FALSE, etINT, {&presteps},
2212 "Let dlb equilibrate this many steps before timings are taken (reset cycle counters after this many steps)" },
2213 { "-simsteps", FALSE, etINT64, {&new_sim_nsteps},
2214 "If non-negative, perform this many steps in the real run (overwrites nsteps from [TT].tpr[tt], add [TT].cpt[tt] steps)" },
2215 { "-launch", FALSE, etBOOL, {&bLaunch},
2216 "Launch the real simulation after optimization" },
2217 { "-bench", FALSE, etBOOL, {&bBenchmark},
2218 "Run the benchmarks or just create the input [TT].tpr[tt] files?" },
2219 { "-check", FALSE, etBOOL, {&bCheck},
2220 "Before the benchmark runs, check whether mdrun works in parallel" },
2221 /******************/
2222 /* mdrun options: */
2223 /******************/
2224 /* We let g_tune_pme parse and understand these options, because we need to
2225 * prevent that they appear on the mdrun command line for the benchmarks */
2226 { "-append", FALSE, etBOOL, {&bAppendFiles},
2227 "Append to previous output files when continuing from checkpoint instead of adding the simulation part number to all file names (for launch only)" },
2228 { "-cpnum", FALSE, etBOOL, {&bKeepAndNumCPT},
2229 "Keep and number checkpoint files (launch only)" },
2230 { "-resethway", FALSE, etBOOL, {&bResetCountersHalfWay},
2231 "HIDDENReset the cycle counters after half the number of steps or halfway [TT]-maxh[tt] (launch only)" }
2234 #define NFILE asize(fnm)
2236 seconds = gmx_gettime();
2238 if (!parse_common_args(&argc, argv, PCA_NOEXIT_ON_ARGS,
2239 NFILE, fnm, asize(pa), pa, asize(desc), desc,
2245 /* Store the remaining unparsed command line entries in a string which
2246 * is then attached to the mdrun command line */
2248 ExtraArgs[0] = '\0';
2249 for (i = 1; i < argc; i++) /* argc will now be 1 if everything was understood */
2251 add_to_string(&ExtraArgs, argv[i]);
2252 add_to_string(&ExtraArgs, " ");
2255 if (opt2parg_bSet("-ntmpi", asize(pa), pa))
2258 if (opt2parg_bSet("-npstring", asize(pa), pa))
2260 fprintf(stderr, "WARNING: -npstring has no effect when using threads.\n");
2265 gmx_fatal(FARGS, "Can't run multi-threaded MPI simulation yet!");
2267 /* and now we just set this; a bit of an ugly hack*/
2270 /* Check for PME:PP ratio and whether tpr triggers additional output files */
2271 guessPMEratio = inspect_tpr(NFILE, fnm, &rcoulomb);
2273 /* Automatically set -beo options if -eo is set etc. */
2274 couple_files_options(NFILE, fnm);
2276 /* Construct the command line arguments for benchmark runs
2277 * as well as for the simulation run */
2280 sprintf(bbuf, " -ntmpi %d ", nthreads);
2284 /* This string will be used for MPI runs and will appear after the
2285 * mpirun command. */
2286 if (strcmp(procstring[0], "none") != 0)
2288 sprintf(bbuf, " %s %d ", procstring[0], nnodes);
2298 create_command_line_snippets(bAppendFiles, bKeepAndNumCPT, bResetCountersHalfWay, presteps,
2299 NFILE, fnm, &cmd_args_bench, &cmd_args_launch, ExtraArgs);
2301 /* Read in checkpoint file if requested */
2303 if (opt2bSet("-cpi", NFILE, fnm))
2306 cr->duty = DUTY_PP; /* makes the following routine happy */
2307 read_checkpoint_simulation_part(opt2fn("-cpi", NFILE, fnm),
2308 &sim_part, &cpt_steps, cr,
2309 FALSE, NFILE, fnm, NULL, NULL);
2312 /* sim_part will now be 1 if no checkpoint file was found */
2315 gmx_fatal(FARGS, "Checkpoint file %s not found!", opt2fn("-cpi", NFILE, fnm));
2319 /* Open performance output file and write header info */
2320 fp = gmx_ffopen(opt2fn("-p", NFILE, fnm), "w");
2322 /* Make a quick consistency check of command line parameters */
2323 check_input(nnodes, repeats, &ntprs, &rmin, rcoulomb, &rmax,
2324 maxPMEfraction, minPMEfraction, npme_fixed,
2325 bench_nsteps, fnm, NFILE, sim_part, presteps,
2328 /* Determine the maximum and minimum number of PME nodes to test,
2329 * the actual list of settings is build in do_the_tests(). */
2330 if ((nnodes > 2) && (npme_fixed < -1))
2332 if (0 == strcmp(npmevalues_opt[0], "auto"))
2334 /* Determine the npme range automatically based on the PME:PP load guess */
2335 if (guessPMEratio > 1.0)
2337 /* More PME than PP work, probably we do not need separate PME nodes at all! */
2338 maxPMEnodes = nnodes/2;
2339 minPMEnodes = nnodes/2;
2343 /* PME : PP load is in the range 0..1, let's test around the guess */
2344 guessPMEnodes = nnodes/(1.0 + 1.0/guessPMEratio);
2345 minPMEnodes = floor(0.7*guessPMEnodes);
2346 maxPMEnodes = ceil(1.6*guessPMEnodes);
2347 maxPMEnodes = min(maxPMEnodes, nnodes/2);
2352 /* Determine the npme range based on user input */
2353 maxPMEnodes = floor(maxPMEfraction*nnodes);
2354 minPMEnodes = max(floor(minPMEfraction*nnodes), 0);
2355 fprintf(stdout, "Will try runs with %d ", minPMEnodes);
2356 if (maxPMEnodes != minPMEnodes)
2358 fprintf(stdout, "- %d ", maxPMEnodes);
2360 fprintf(stdout, "PME-only ranks.\n Note that the automatic number of PME-only ranks and no separate PME ranks are always tested.\n");
2369 /* Get the commands we need to set up the runs from environment variables */
2370 get_program_paths(bThreads, &cmd_mpirun, &cmd_mdrun);
2371 if (bBenchmark && repeats > 0)
2373 check_mdrun_works(bThreads, cmd_mpirun, cmd_np, cmd_mdrun);
2376 /* Print some header info to file */
2378 fprintf(fp, "\n P E R F O R M A N C E R E S U L T S\n");
2380 fprintf(fp, "%s for Gromacs %s\n", output_env_get_program_display_name(oenv),
2384 fprintf(fp, "Number of ranks : %d\n", nnodes);
2385 fprintf(fp, "The mpirun command is : %s\n", cmd_mpirun);
2386 if (strcmp(procstring[0], "none") != 0)
2388 fprintf(fp, "Passing # of ranks via : %s\n", procstring[0]);
2392 fprintf(fp, "Not setting number of ranks in system call\n");
2397 fprintf(fp, "Number of threads : %d\n", nnodes);
2400 fprintf(fp, "The mdrun command is : %s\n", cmd_mdrun);
2401 fprintf(fp, "mdrun args benchmarks : %s\n", cmd_args_bench);
2402 fprintf(fp, "Benchmark steps : ");
2403 fprintf(fp, "%"GMX_PRId64, bench_nsteps);
2405 fprintf(fp, "dlb equilibration steps : %d\n", presteps);
2408 fprintf(fp, "Checkpoint time step : ");
2409 fprintf(fp, "%"GMX_PRId64, cpt_steps);
2412 fprintf(fp, "mdrun args at launchtime: %s\n", cmd_args_launch);
2414 if (new_sim_nsteps >= 0)
2417 fprintf(stderr, "Note: Simulation input file %s will have ", opt2fn("-so", NFILE, fnm));
2418 fprintf(stderr, "%"GMX_PRId64, new_sim_nsteps+cpt_steps);
2419 fprintf(stderr, " steps.\n");
2420 fprintf(fp, "Simulation steps : ");
2421 fprintf(fp, "%"GMX_PRId64, new_sim_nsteps);
2426 fprintf(fp, "Repeats for each test : %d\n", repeats);
2429 if (npme_fixed >= -1)
2431 fprintf(fp, "Fixing -npme at : %d\n", npme_fixed);
2434 fprintf(fp, "Input file : %s\n", opt2fn("-s", NFILE, fnm));
2435 fprintf(fp, " PME/PP load estimate : %g\n", guessPMEratio);
2437 /* Allocate memory for the inputinfo struct: */
2439 info->nr_inputfiles = ntprs;
2440 for (i = 0; i < ntprs; i++)
2442 snew(info->rcoulomb, ntprs);
2443 snew(info->rvdw, ntprs);
2444 snew(info->rlist, ntprs);
2445 snew(info->rlistlong, ntprs);
2446 snew(info->nkx, ntprs);
2447 snew(info->nky, ntprs);
2448 snew(info->nkz, ntprs);
2449 snew(info->fsx, ntprs);
2450 snew(info->fsy, ntprs);
2451 snew(info->fsz, ntprs);
2453 /* Make alternative tpr files to test: */
2454 snew(tpr_names, ntprs);
2455 for (i = 0; i < ntprs; i++)
2457 snew(tpr_names[i], STRLEN);
2460 /* It can be that ntprs is reduced by make_benchmark_tprs if not enough
2461 * different grids could be found. */
2462 make_benchmark_tprs(opt2fn("-s", NFILE, fnm), tpr_names, bench_nsteps+presteps,
2463 cpt_steps, rmin, rmax, bScaleRvdw, &ntprs, info, fp);
2465 /********************************************************************************/
2466 /* Main loop over all scenarios we need to test: tpr files, PME nodes, repeats */
2467 /********************************************************************************/
2468 snew(perfdata, ntprs);
2471 do_the_tests(fp, tpr_names, maxPMEnodes, minPMEnodes, npme_fixed, npmevalues_opt[0], perfdata, &pmeentries,
2472 repeats, nnodes, ntprs, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2473 cmd_args_bench, fnm, NFILE, presteps, cpt_steps, bCheck);
2475 fprintf(fp, "\nTuning took%8.1f minutes.\n", (gmx_gettime()-seconds)/60.0);
2477 /* Analyse the results and give a suggestion for optimal settings: */
2478 bKeepTPR = analyze_data(fp, opt2fn("-p", NFILE, fnm), perfdata, nnodes, ntprs, pmeentries,
2479 repeats, info, &best_tpr, &best_npme);
2481 /* Take the best-performing tpr file and enlarge nsteps to original value */
2482 if (bKeepTPR && !bOverwrite)
2484 simulation_tpr = opt2fn("-s", NFILE, fnm);
2488 simulation_tpr = opt2fn("-so", NFILE, fnm);
2489 modify_PMEsettings(bOverwrite ? (new_sim_nsteps+cpt_steps) : info->orig_sim_steps,
2490 info->orig_init_step, tpr_names[best_tpr], simulation_tpr);
2493 /* Let's get rid of the temporary benchmark input files */
2494 for (i = 0; i < ntprs; i++)
2496 fprintf(stdout, "Deleting temporary benchmark input file %s\n", tpr_names[i]);
2497 remove(tpr_names[i]);
2500 /* Now start the real simulation if the user requested it ... */
2501 launch_simulation(bLaunch, fp, bThreads, cmd_mpirun, cmd_np, cmd_mdrun,
2502 cmd_args_launch, simulation_tpr, best_npme);
2506 /* ... or simply print the performance results to screen: */
2509 finalize(opt2fn("-p", NFILE, fnm));