endif()
# Machinery for running the external project
-set(EXTERNAL_FFTW_VERSION 3.3.2)
+set(EXTERNAL_FFTW_VERSION 3.3.3)
+ set(GMX_BUILD_OWN_FFTW_URL "http://www.fftw.org/fftw-${EXTERNAL_FFTW_VERSION}.tar.gz" CACHE PATH "URL from where to download fftw, (use an absolute path when offline)")
+ mark_as_advanced(GMX_BUILD_OWN_FFTW_URL)
+set(EXTERNAL_FFTW_MD5SUM 0a05ca9c7b3bfddc8278e7c40791a1c2)
+set (EXTERNAL_FFTW_BUILD_TARGET fftwBuild)
include(ExternalProject)
# TODO in master branch - show this warning only on the first run
# by using gmx_check_if_changed_result from I21b791ab8e4f3 when
# TODO if/when CMake fixes http://www.cmake.org/Bug/view.php?id=14330
# (ie. at least version > 2.8.11.2), consider reverting to using an
# md5sum check to avoid needing the above warning
- ExternalProject_add(fftwBuild
+ ExternalProject_add(${EXTERNAL_FFTW_BUILD_TARGET}
- URL "http://www.fftw.org/fftw-${EXTERNAL_FFTW_VERSION}.tar.gz"
+ URL "${GMX_BUILD_OWN_FFTW_URL}"
CONFIGURE_COMMAND <SOURCE_DIR>/configure --prefix=<INSTALL_DIR> --libdir=<INSTALL_DIR>/lib --disable-fortran
${GMX_BUILD_OWN_FFTW_SHARED_FLAG} ${GMX_BUILD_OWN_FFTW_OPTIMIZATION_CONFIGURATION}
${GMX_BUILD_OWN_FFTW_PREC}
--- /dev/null
- * Copyright (c) 2013, by the GROMACS development team, led by
+/*
+ * This file is part of the GROMACS molecular simulation package.
+ *
+ * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
+ * Copyright (c) 2001-2004, The GROMACS development team.
- double vvhh, vv, v, h, hh2, vv2, varv, hh, varh, tt, cv, cp, alpha, kappa, dcp, et, varet;
++ * Copyright (c) 2013,2014, by the GROMACS development team, led by
+ * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
+ * and including many others, as listed in the AUTHORS file in the
+ * top-level source directory and at http://www.gromacs.org.
+ *
+ * GROMACS is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public License
+ * as published by the Free Software Foundation; either version 2.1
+ * of the License, or (at your option) any later version.
+ *
+ * GROMACS is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with GROMACS; if not, see
+ * http://www.gnu.org/licenses, or write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * If you want to redistribute modifications to GROMACS, please
+ * consider that scientific software is very special. Version
+ * control is crucial - bugs must be traceable. We will be happy to
+ * consider code for inclusion in the official distribution, but
+ * derived work must not be called official GROMACS. Details are found
+ * in the README & COPYING files - if they are missing, get the
+ * official version at http://www.gromacs.org.
+ *
+ * To help us fund GROMACS development, we humbly ask that you cite
+ * the research papers on the package. Check out http://www.gromacs.org.
+ */
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <string.h>
+#include <math.h>
+#include <ctype.h>
+
+#include "typedefs.h"
+#include "gmx_fatal.h"
+#include "vec.h"
+#include "string2.h"
+#include "smalloc.h"
+#include "gromacs/fileio/enxio.h"
+#include "gromacs/commandline/pargs.h"
+#include "names.h"
+#include "copyrite.h"
+#include "macros.h"
+#include "xvgr.h"
+#include "gstat.h"
+#include "physics.h"
+#include "gromacs/fileio/tpxio.h"
+#include "gromacs/fileio/trxio.h"
+#include "viewit.h"
+#include "mtop_util.h"
+#include "gmx_ana.h"
+#include "mdebin.h"
+
+static real minthird = -1.0/3.0, minsixth = -1.0/6.0;
+
+typedef struct {
+ real sum;
+ real sum2;
+} exactsum_t;
+
+typedef struct {
+ real *ener;
+ exactsum_t *es;
+ gmx_bool bExactStat;
+ double av;
+ double rmsd;
+ double ee;
+ double slope;
+} enerdat_t;
+
+typedef struct {
+ gmx_int64_t nsteps;
+ gmx_int64_t npoints;
+ int nframes;
+ int *step;
+ int *steps;
+ int *points;
+ enerdat_t *s;
+} enerdata_t;
+
+static double mypow(double x, double y)
+{
+ if (x > 0)
+ {
+ return pow(x, y);
+ }
+ else
+ {
+ return 0.0;
+ }
+}
+
+static int *select_it(int nre, char *nm[], int *nset)
+{
+ gmx_bool *bE;
+ int n, k, j, i;
+ int *set;
+ gmx_bool bVerbose = TRUE;
+
+ if ((getenv("VERBOSE")) != NULL)
+ {
+ bVerbose = FALSE;
+ }
+
+ fprintf(stderr, "Select the terms you want from the following list\n");
+ fprintf(stderr, "End your selection with 0\n");
+
+ if (bVerbose)
+ {
+ for (k = 0; (k < nre); )
+ {
+ for (j = 0; (j < 4) && (k < nre); j++, k++)
+ {
+ fprintf(stderr, " %3d=%14s", k+1, nm[k]);
+ }
+ fprintf(stderr, "\n");
+ }
+ }
+
+ snew(bE, nre);
+ do
+ {
+ if (1 != scanf("%d", &n))
+ {
+ gmx_fatal(FARGS, "Error reading user input");
+ }
+ if ((n > 0) && (n <= nre))
+ {
+ bE[n-1] = TRUE;
+ }
+ }
+ while (n != 0);
+
+ snew(set, nre);
+ for (i = (*nset) = 0; (i < nre); i++)
+ {
+ if (bE[i])
+ {
+ set[(*nset)++] = i;
+ }
+ }
+
+ sfree(bE);
+
+ return set;
+}
+
+static void chomp(char *buf)
+{
+ int len = strlen(buf);
+
+ while ((len > 0) && (buf[len-1] == '\n'))
+ {
+ buf[len-1] = '\0';
+ len--;
+ }
+}
+
+static int *select_by_name(int nre, gmx_enxnm_t *nm, int *nset)
+{
+ gmx_bool *bE;
+ int n, k, kk, j, i, nmatch, nind, nss;
+ int *set;
+ gmx_bool bEOF, bVerbose = TRUE, bLong = FALSE;
+ char *ptr, buf[STRLEN];
+ const char *fm4 = "%3d %-14s";
+ const char *fm2 = "%3d %-34s";
+ char **newnm = NULL;
+
+ if ((getenv("VERBOSE")) != NULL)
+ {
+ bVerbose = FALSE;
+ }
+
+ fprintf(stderr, "\n");
+ fprintf(stderr, "Select the terms you want from the following list by\n");
+ fprintf(stderr, "selecting either (part of) the name or the number or a combination.\n");
+ fprintf(stderr, "End your selection with an empty line or a zero.\n");
+ fprintf(stderr, "-------------------------------------------------------------------\n");
+
+ snew(newnm, nre);
+ j = 0;
+ for (k = 0; k < nre; k++)
+ {
+ newnm[k] = strdup(nm[k].name);
+ /* Insert dashes in all the names */
+ while ((ptr = strchr(newnm[k], ' ')) != NULL)
+ {
+ *ptr = '-';
+ }
+ if (bVerbose)
+ {
+ if (j == 0)
+ {
+ if (k > 0)
+ {
+ fprintf(stderr, "\n");
+ }
+ bLong = FALSE;
+ for (kk = k; kk < k+4; kk++)
+ {
+ if (kk < nre && strlen(nm[kk].name) > 14)
+ {
+ bLong = TRUE;
+ }
+ }
+ }
+ else
+ {
+ fprintf(stderr, " ");
+ }
+ if (!bLong)
+ {
+ fprintf(stderr, fm4, k+1, newnm[k]);
+ j++;
+ if (j == 4)
+ {
+ j = 0;
+ }
+ }
+ else
+ {
+ fprintf(stderr, fm2, k+1, newnm[k]);
+ j++;
+ if (j == 2)
+ {
+ j = 0;
+ }
+ }
+ }
+ }
+ if (bVerbose)
+ {
+ fprintf(stderr, "\n\n");
+ }
+
+ snew(bE, nre);
+
+ bEOF = FALSE;
+ while (!bEOF && (fgets2(buf, STRLEN-1, stdin)))
+ {
+ /* Remove newlines */
+ chomp(buf);
+
+ /* Remove spaces */
+ trim(buf);
+
+ /* Empty line means end of input */
+ bEOF = (strlen(buf) == 0);
+ if (!bEOF)
+ {
+ ptr = buf;
+ do
+ {
+ if (!bEOF)
+ {
+ /* First try to read an integer */
+ nss = sscanf(ptr, "%d", &nind);
+ if (nss == 1)
+ {
+ /* Zero means end of input */
+ if (nind == 0)
+ {
+ bEOF = TRUE;
+ }
+ else if ((1 <= nind) && (nind <= nre))
+ {
+ bE[nind-1] = TRUE;
+ }
+ else
+ {
+ fprintf(stderr, "number %d is out of range\n", nind);
+ }
+ }
+ else
+ {
+ /* Now try to read a string */
+ i = strlen(ptr);
+ nmatch = 0;
+ for (nind = 0; nind < nre; nind++)
+ {
+ if (gmx_strcasecmp(newnm[nind], ptr) == 0)
+ {
+ bE[nind] = TRUE;
+ nmatch++;
+ }
+ }
+ if (nmatch == 0)
+ {
+ i = strlen(ptr);
+ nmatch = 0;
+ for (nind = 0; nind < nre; nind++)
+ {
+ if (gmx_strncasecmp(newnm[nind], ptr, i) == 0)
+ {
+ bE[nind] = TRUE;
+ nmatch++;
+ }
+ }
+ if (nmatch == 0)
+ {
+ fprintf(stderr, "String '%s' does not match anything\n", ptr);
+ }
+ }
+ }
+ }
+ /* Look for the first space, and remove spaces from there */
+ if ((ptr = strchr(ptr, ' ')) != NULL)
+ {
+ trim(ptr);
+ }
+ }
+ while (!bEOF && (ptr && (strlen(ptr) > 0)));
+ }
+ }
+
+ snew(set, nre);
+ for (i = (*nset) = 0; (i < nre); i++)
+ {
+ if (bE[i])
+ {
+ set[(*nset)++] = i;
+ }
+ }
+
+ sfree(bE);
+
+ if (*nset == 0)
+ {
+ gmx_fatal(FARGS, "No energy terms selected");
+ }
+
+ for (i = 0; (i < nre); i++)
+ {
+ sfree(newnm[i]);
+ }
+ sfree(newnm);
+
+ return set;
+}
+
+static void get_dhdl_parms(const char *topnm, t_inputrec *ir)
+{
+ gmx_mtop_t mtop;
+ int natoms;
+ t_iatom *iatom;
+ matrix box;
+
+ /* all we need is the ir to be able to write the label */
+ read_tpx(topnm, ir, box, &natoms, NULL, NULL, NULL, &mtop);
+}
+
+static void get_orires_parms(const char *topnm,
+ int *nor, int *nex, int **label, real **obs)
+{
+ gmx_mtop_t mtop;
+ gmx_localtop_t *top;
+ t_inputrec ir;
+ t_iparams *ip;
+ int natoms, i;
+ t_iatom *iatom;
+ int nb;
+ matrix box;
+
+ read_tpx(topnm, &ir, box, &natoms, NULL, NULL, NULL, &mtop);
+ top = gmx_mtop_generate_local_top(&mtop, &ir);
+
+ ip = top->idef.iparams;
+ iatom = top->idef.il[F_ORIRES].iatoms;
+
+ /* Count how many distance restraint there are... */
+ nb = top->idef.il[F_ORIRES].nr;
+ if (nb == 0)
+ {
+ gmx_fatal(FARGS, "No orientation restraints in topology!\n");
+ }
+
+ *nor = nb/3;
+ *nex = 0;
+ snew(*label, *nor);
+ snew(*obs, *nor);
+ for (i = 0; i < nb; i += 3)
+ {
+ (*label)[i/3] = ip[iatom[i]].orires.label;
+ (*obs)[i/3] = ip[iatom[i]].orires.obs;
+ if (ip[iatom[i]].orires.ex >= *nex)
+ {
+ *nex = ip[iatom[i]].orires.ex+1;
+ }
+ }
+ fprintf(stderr, "Found %d orientation restraints with %d experiments",
+ *nor, *nex);
+}
+
+static int get_bounds(const char *topnm,
+ real **bounds, int **index, int **dr_pair, int *npairs,
+ gmx_mtop_t *mtop, gmx_localtop_t **ltop, t_inputrec *ir)
+{
+ gmx_localtop_t *top;
+ t_functype *functype;
+ t_iparams *ip;
+ int natoms, i, j, k, type, ftype, natom;
+ t_ilist *disres;
+ t_iatom *iatom;
+ real *b;
+ int *ind, *pair;
+ int nb, label1;
+ matrix box;
+
+ read_tpx(topnm, ir, box, &natoms, NULL, NULL, NULL, mtop);
+ snew(*ltop, 1);
+ top = gmx_mtop_generate_local_top(mtop, ir);
+ *ltop = top;
+
+ functype = top->idef.functype;
+ ip = top->idef.iparams;
+
+ /* Count how many distance restraint there are... */
+ nb = top->idef.il[F_DISRES].nr;
+ if (nb == 0)
+ {
+ gmx_fatal(FARGS, "No distance restraints in topology!\n");
+ }
+
+ /* Allocate memory */
+ snew(b, nb);
+ snew(ind, nb);
+ snew(pair, nb+1);
+
+ /* Fill the bound array */
+ nb = 0;
+ for (i = 0; (i < top->idef.ntypes); i++)
+ {
+ ftype = functype[i];
+ if (ftype == F_DISRES)
+ {
+
+ label1 = ip[i].disres.label;
+ b[nb] = ip[i].disres.up1;
+ ind[nb] = label1;
+ nb++;
+ }
+ }
+ *bounds = b;
+
+ /* Fill the index array */
+ label1 = -1;
+ disres = &(top->idef.il[F_DISRES]);
+ iatom = disres->iatoms;
+ for (i = j = k = 0; (i < disres->nr); )
+ {
+ type = iatom[i];
+ ftype = top->idef.functype[type];
+ natom = interaction_function[ftype].nratoms+1;
+ if (label1 != top->idef.iparams[type].disres.label)
+ {
+ pair[j] = k;
+ label1 = top->idef.iparams[type].disres.label;
+ j++;
+ }
+ k++;
+ i += natom;
+ }
+ pair[j] = k;
+ *npairs = k;
+ if (j != nb)
+ {
+ gmx_incons("get_bounds for distance restraints");
+ }
+
+ *index = ind;
+ *dr_pair = pair;
+
+ return nb;
+}
+
+static void calc_violations(real rt[], real rav3[], int nb, int index[],
+ real bounds[], real *viol, double *st, double *sa)
+{
+ const real sixth = 1.0/6.0;
+ int i, j;
+ double rsum, rav, sumaver, sumt;
+
+ sumaver = 0;
+ sumt = 0;
+ for (i = 0; (i < nb); i++)
+ {
+ rsum = 0.0;
+ rav = 0.0;
+ for (j = index[i]; (j < index[i+1]); j++)
+ {
+ if (viol)
+ {
+ viol[j] += mypow(rt[j], -3.0);
+ }
+ rav += sqr(rav3[j]);
+ rsum += mypow(rt[j], -6);
+ }
+ rsum = max(0.0, mypow(rsum, -sixth)-bounds[i]);
+ rav = max(0.0, mypow(rav, -sixth)-bounds[i]);
+
+ sumt += rsum;
+ sumaver += rav;
+ }
+ *st = sumt;
+ *sa = sumaver;
+}
+
+static void analyse_disre(const char *voutfn, int nframes,
+ real violaver[], real bounds[], int index[],
+ int pair[], int nbounds,
+ const output_env_t oenv)
+{
+ FILE *vout;
+ double sum, sumt, sumaver;
+ int i, j;
+
+ /* Subtract bounds from distances, to calculate violations */
+ calc_violations(violaver, violaver,
+ nbounds, pair, bounds, NULL, &sumt, &sumaver);
+
+#ifdef DEBUG
+ fprintf(stdout, "\nSum of violations averaged over simulation: %g nm\n",
+ sumaver);
+ fprintf(stdout, "Largest violation averaged over simulation: %g nm\n\n",
+ sumt);
+#endif
+ vout = xvgropen(voutfn, "r\\S-3\\N average violations", "DR Index", "nm",
+ oenv);
+ sum = 0.0;
+ sumt = 0.0;
+ for (i = 0; (i < nbounds); i++)
+ {
+ /* Do ensemble averaging */
+ sumaver = 0;
+ for (j = pair[i]; (j < pair[i+1]); j++)
+ {
+ sumaver += sqr(violaver[j]/nframes);
+ }
+ sumaver = max(0.0, mypow(sumaver, minsixth)-bounds[i]);
+
+ sumt += sumaver;
+ sum = max(sum, sumaver);
+ fprintf(vout, "%10d %10.5e\n", index[i], sumaver);
+ }
+#ifdef DEBUG
+ for (j = 0; (j < dr.ndr); j++)
+ {
+ fprintf(vout, "%10d %10.5e\n", j, mypow(violaver[j]/nframes, minthird));
+ }
+#endif
+ ffclose(vout);
+
+ fprintf(stdout, "\nSum of violations averaged over simulation: %g nm\n",
+ sumt);
+ fprintf(stdout, "Largest violation averaged over simulation: %g nm\n\n", sum);
+
+ do_view(oenv, voutfn, "-graphtype bar");
+}
+
+static void einstein_visco(const char *fn, const char *fni, int nsets,
+ int nframes, real **sum,
+ real V, real T, int nsteps, double time[],
+ const output_env_t oenv)
+{
+ FILE *fp0, *fp1;
+ double av[4], avold[4];
+ double fac, dt, di;
+ int i, j, m, nf4;
+
+ if (nframes < 1)
+ {
+ return;
+ }
+
+ dt = (time[1]-time[0]);
+ nf4 = nframes/4+1;
+
+ for (i = 0; i <= nsets; i++)
+ {
+ avold[i] = 0;
+ }
+ fp0 = xvgropen(fni, "Shear viscosity integral",
+ "Time (ps)", "(kg m\\S-1\\N s\\S-1\\N ps)", oenv);
+ fp1 = xvgropen(fn, "Shear viscosity using Einstein relation",
+ "Time (ps)", "(kg m\\S-1\\N s\\S-1\\N)", oenv);
+ for (i = 1; i < nf4; i++)
+ {
+ fac = dt*nframes/nsteps;
+ for (m = 0; m <= nsets; m++)
+ {
+ av[m] = 0;
+ }
+ for (j = 0; j < nframes-i; j++)
+ {
+ for (m = 0; m < nsets; m++)
+ {
+ di = sqr(fac*(sum[m][j+i]-sum[m][j]));
+
+ av[m] += di;
+ av[nsets] += di/nsets;
+ }
+ }
+ /* Convert to SI for the viscosity */
+ fac = (V*NANO*NANO*NANO*PICO*1e10)/(2*BOLTZMANN*T)/(nframes-i);
+ fprintf(fp0, "%10g", time[i]-time[0]);
+ for (m = 0; (m <= nsets); m++)
+ {
+ av[m] = fac*av[m];
+ fprintf(fp0, " %10g", av[m]);
+ }
+ fprintf(fp0, "\n");
+ fprintf(fp1, "%10g", 0.5*(time[i]+time[i-1])-time[0]);
+ for (m = 0; (m <= nsets); m++)
+ {
+ fprintf(fp1, " %10g", (av[m]-avold[m])/dt);
+ avold[m] = av[m];
+ }
+ fprintf(fp1, "\n");
+ }
+ ffclose(fp0);
+ ffclose(fp1);
+}
+
+typedef struct {
+ gmx_int64_t np;
+ double sum;
+ double sav;
+ double sav2;
+} ee_sum_t;
+
+typedef struct {
+ int b;
+ ee_sum_t sum;
+ gmx_int64_t nst;
+ gmx_int64_t nst_min;
+} ener_ee_t;
+
+static void clear_ee_sum(ee_sum_t *ees)
+{
+ ees->sav = 0;
+ ees->sav2 = 0;
+ ees->np = 0;
+ ees->sum = 0;
+}
+
+static void add_ee_sum(ee_sum_t *ees, double sum, int np)
+{
+ ees->np += np;
+ ees->sum += sum;
+}
+
+static void add_ee_av(ee_sum_t *ees)
+{
+ double av;
+
+ av = ees->sum/ees->np;
+ ees->sav += av;
+ ees->sav2 += av*av;
+ ees->np = 0;
+ ees->sum = 0;
+}
+
+static double calc_ee2(int nb, ee_sum_t *ees)
+{
+ return (ees->sav2/nb - dsqr(ees->sav/nb))/(nb - 1);
+}
+
+static void set_ee_av(ener_ee_t *eee)
+{
+ if (debug)
+ {
+ char buf[STEPSTRSIZE];
+ fprintf(debug, "Storing average for err.est.: %s steps\n",
+ gmx_step_str(eee->nst, buf));
+ }
+ add_ee_av(&eee->sum);
+ eee->b++;
+ if (eee->b == 1 || eee->nst < eee->nst_min)
+ {
+ eee->nst_min = eee->nst;
+ }
+ eee->nst = 0;
+}
+
+static void calc_averages(int nset, enerdata_t *edat, int nbmin, int nbmax)
+{
+ int nb, i, f, nee;
+ double sum, sum2, sump, see2;
+ gmx_int64_t steps, np, p, bound_nb;
+ enerdat_t *ed;
+ exactsum_t *es;
+ gmx_bool bAllZero;
+ double x, sx, sy, sxx, sxy;
+ ener_ee_t *eee;
+
+ /* Check if we have exact statistics over all points */
+ for (i = 0; i < nset; i++)
+ {
+ ed = &edat->s[i];
+ ed->bExactStat = FALSE;
+ if (edat->npoints > 0)
+ {
+ /* All energy file sum entries 0 signals no exact sums.
+ * But if all energy values are 0, we still have exact sums.
+ */
+ bAllZero = TRUE;
+ for (f = 0; f < edat->nframes && !ed->bExactStat; f++)
+ {
+ if (ed->ener[i] != 0)
+ {
+ bAllZero = FALSE;
+ }
+ ed->bExactStat = (ed->es[f].sum != 0);
+ }
+ if (bAllZero)
+ {
+ ed->bExactStat = TRUE;
+ }
+ }
+ }
+
+ snew(eee, nbmax+1);
+ for (i = 0; i < nset; i++)
+ {
+ ed = &edat->s[i];
+
+ sum = 0;
+ sum2 = 0;
+ np = 0;
+ sx = 0;
+ sy = 0;
+ sxx = 0;
+ sxy = 0;
+ for (nb = nbmin; nb <= nbmax; nb++)
+ {
+ eee[nb].b = 0;
+ clear_ee_sum(&eee[nb].sum);
+ eee[nb].nst = 0;
+ eee[nb].nst_min = 0;
+ }
+ for (f = 0; f < edat->nframes; f++)
+ {
+ es = &ed->es[f];
+
+ if (ed->bExactStat)
+ {
+ /* Add the sum and the sum of variances to the totals. */
+ p = edat->points[f];
+ sump = es->sum;
+ sum2 += es->sum2;
+ if (np > 0)
+ {
+ sum2 += dsqr(sum/np - (sum + es->sum)/(np + p))
+ *np*(np + p)/p;
+ }
+ }
+ else
+ {
+ /* Add a single value to the sum and sum of squares. */
+ p = 1;
+ sump = ed->ener[f];
+ sum2 += dsqr(sump);
+ }
+
+ /* sum has to be increased after sum2 */
+ np += p;
+ sum += sump;
+
+ /* For the linear regression use variance 1/p.
+ * Note that sump is the sum, not the average, so we don't need p*.
+ */
+ x = edat->step[f] - 0.5*(edat->steps[f] - 1);
+ sx += p*x;
+ sy += sump;
+ sxx += p*x*x;
+ sxy += x*sump;
+
+ for (nb = nbmin; nb <= nbmax; nb++)
+ {
+ /* Check if the current end step is closer to the desired
+ * block boundary than the next end step.
+ */
+ bound_nb = (edat->step[0]-1)*nb + edat->nsteps*(eee[nb].b+1);
+ if (eee[nb].nst > 0 &&
+ bound_nb - edat->step[f-1]*nb < edat->step[f]*nb - bound_nb)
+ {
+ set_ee_av(&eee[nb]);
+ }
+ if (f == 0)
+ {
+ eee[nb].nst = 1;
+ }
+ else
+ {
+ eee[nb].nst += edat->step[f] - edat->step[f-1];
+ }
+ if (ed->bExactStat)
+ {
+ add_ee_sum(&eee[nb].sum, es->sum, edat->points[f]);
+ }
+ else
+ {
+ add_ee_sum(&eee[nb].sum, edat->s[i].ener[f], 1);
+ }
+ bound_nb = (edat->step[0]-1)*nb + edat->nsteps*(eee[nb].b+1);
+ if (edat->step[f]*nb >= bound_nb)
+ {
+ set_ee_av(&eee[nb]);
+ }
+ }
+ }
+
+ edat->s[i].av = sum/np;
+ if (ed->bExactStat)
+ {
+ edat->s[i].rmsd = sqrt(sum2/np);
+ }
+ else
+ {
+ edat->s[i].rmsd = sqrt(sum2/np - dsqr(edat->s[i].av));
+ }
+
+ if (edat->nframes > 1)
+ {
+ edat->s[i].slope = (np*sxy - sx*sy)/(np*sxx - sx*sx);
+ }
+ else
+ {
+ edat->s[i].slope = 0;
+ }
+
+ nee = 0;
+ see2 = 0;
+ for (nb = nbmin; nb <= nbmax; nb++)
+ {
+ /* Check if we actually got nb blocks and if the smallest
+ * block is not shorter than 80% of the average.
+ */
+ if (debug)
+ {
+ char buf1[STEPSTRSIZE], buf2[STEPSTRSIZE];
+ fprintf(debug, "Requested %d blocks, we have %d blocks, min %s nsteps %s\n",
+ nb, eee[nb].b,
+ gmx_step_str(eee[nb].nst_min, buf1),
+ gmx_step_str(edat->nsteps, buf2));
+ }
+ if (eee[nb].b == nb && 5*nb*eee[nb].nst_min >= 4*edat->nsteps)
+ {
+ see2 += calc_ee2(nb, &eee[nb].sum);
+ nee++;
+ }
+ }
+ if (nee > 0)
+ {
+ edat->s[i].ee = sqrt(see2/nee);
+ }
+ else
+ {
+ edat->s[i].ee = -1;
+ }
+ }
+ sfree(eee);
+}
+
+static enerdata_t *calc_sum(int nset, enerdata_t *edat, int nbmin, int nbmax)
+{
+ enerdata_t *esum;
+ enerdat_t *s;
+ int f, i;
+ double sum;
+
+ snew(esum, 1);
+ *esum = *edat;
+ snew(esum->s, 1);
+ s = &esum->s[0];
+ snew(s->ener, esum->nframes);
+ snew(s->es, esum->nframes);
+
+ s->bExactStat = TRUE;
+ s->slope = 0;
+ for (i = 0; i < nset; i++)
+ {
+ if (!edat->s[i].bExactStat)
+ {
+ s->bExactStat = FALSE;
+ }
+ s->slope += edat->s[i].slope;
+ }
+
+ for (f = 0; f < edat->nframes; f++)
+ {
+ sum = 0;
+ for (i = 0; i < nset; i++)
+ {
+ sum += edat->s[i].ener[f];
+ }
+ s->ener[f] = sum;
+ sum = 0;
+ for (i = 0; i < nset; i++)
+ {
+ sum += edat->s[i].es[f].sum;
+ }
+ s->es[f].sum = sum;
+ s->es[f].sum2 = 0;
+ }
+
+ calc_averages(1, esum, nbmin, nbmax);
+
+ return esum;
+}
+
+static char *ee_pr(double ee, char *buf)
+{
+ char tmp[100];
+ double rnd;
+
+ if (ee < 0)
+ {
+ sprintf(buf, "%s", "--");
+ }
+ else
+ {
+ /* Round to two decimals by printing. */
+ sprintf(tmp, "%.1e", ee);
+ sscanf(tmp, "%lf", &rnd);
+ sprintf(buf, "%g", rnd);
+ }
+
+ return buf;
+}
+
+static void remove_drift(int nset, int nbmin, int nbmax, real dt, enerdata_t *edat)
+{
+/* Remove the drift by performing a fit to y = ax+b.
+ Uses 5 iterations. */
+ int i, j, k;
+ double delta, d, sd, sd2;
+
+ edat->npoints = edat->nframes;
+ edat->nsteps = edat->nframes;
+
+ for (k = 0; (k < 5); k++)
+ {
+ for (i = 0; (i < nset); i++)
+ {
+ delta = edat->s[i].slope*dt;
+
+ if (NULL != debug)
+ {
+ fprintf(debug, "slope for set %d is %g\n", i, edat->s[i].slope);
+ }
+
+ for (j = 0; (j < edat->nframes); j++)
+ {
+ edat->s[i].ener[j] -= j*delta;
+ edat->s[i].es[j].sum = 0;
+ edat->s[i].es[j].sum2 = 0;
+ }
+ }
+ calc_averages(nset, edat, nbmin, nbmax);
+ }
+}
+
+static void calc_fluctuation_props(FILE *fp,
+ gmx_bool bDriftCorr, real dt,
+ int nset, int nmol,
+ char **leg, enerdata_t *edat,
+ int nbmin, int nbmax)
+{
+ int i, j;
- vvhh = alpha = kappa = cp = dcp = cv = NOTSET;
++ double vv, v, h, varv, hh, varh, tt, cv, cp, alpha, kappa, dcp, et, varet;
+ double NANO3;
+ enum {
+ eVol, eEnth, eTemp, eEtot, eNR
+ };
+ const char *my_ener[] = { "Volume", "Enthalpy", "Temperature", "Total Energy" };
+ int ii[eNR];
+
+ NANO3 = NANO*NANO*NANO;
+ if (!bDriftCorr)
+ {
+ fprintf(fp, "\nYou may want to use the -driftcorr flag in order to correct\n"
+ "for spurious drift in the graphs. Note that this is not\n"
+ "a substitute for proper equilibration and sampling!\n");
+ }
+ else
+ {
+ remove_drift(nset, nbmin, nbmax, dt, edat);
+ }
+ for (i = 0; (i < eNR); i++)
+ {
+ for (ii[i] = 0; (ii[i] < nset &&
+ (gmx_strcasecmp(leg[ii[i]], my_ener[i]) != 0)); ii[i]++)
+ {
+ ;
+ }
+/* if (ii[i] < nset)
+ fprintf(fp,"Found %s data.\n",my_ener[i]);
+ */ }
+ /* Compute it all! */
- vvhh = 0;
++ alpha = kappa = cp = dcp = cv = NOTSET;
+
+ /* Temperature */
+ tt = NOTSET;
+ if (ii[eTemp] < nset)
+ {
+ tt = edat->s[ii[eTemp]].av;
+ }
+ /* Volume */
+ vv = varv = NOTSET;
+ if ((ii[eVol] < nset) && (ii[eTemp] < nset))
+ {
+ vv = edat->s[ii[eVol]].av*NANO3;
+ varv = dsqr(edat->s[ii[eVol]].rmsd*NANO3);
+ kappa = (varv/vv)/(BOLTZMANN*tt);
+ }
+ /* Enthalpy */
+ hh = varh = NOTSET;
+ if ((ii[eEnth] < nset) && (ii[eTemp] < nset))
+ {
+ hh = KILO*edat->s[ii[eEnth]].av/AVOGADRO;
+ varh = dsqr(KILO*edat->s[ii[eEnth]].rmsd/AVOGADRO);
+ cp = AVOGADRO*((varh/nmol)/(BOLTZMANN*tt*tt));
+ }
+ /* Total energy */
+ et = varet = NOTSET;
+ if ((ii[eEtot] < nset) && (hh == NOTSET) && (tt != NOTSET))
+ {
+ /* Only compute cv in constant volume runs, which we can test
+ by checking whether the enthalpy was computed.
+ */
+ et = edat->s[ii[eEtot]].av;
+ varet = sqr(edat->s[ii[eEtot]].rmsd);
+ cv = KILO*((varet/nmol)/(BOLTZ*tt*tt));
+ }
+ /* Alpha, dcp */
+ if ((ii[eVol] < nset) && (ii[eEnth] < nset) && (ii[eTemp] < nset))
+ {
- v = edat->s[ii[eVol]].ener[j]*NANO3;
- h = KILO*edat->s[ii[eEnth]].ener[j]/AVOGADRO;
- vvhh += (v*h);
++ double v_sum, h_sum, vh_sum, v_aver, h_aver, vh_aver;
++ vh_sum = v_sum = h_sum = 0;
+ for (j = 0; (j < edat->nframes); j++)
+ {
- vvhh /= edat->nframes;
- alpha = (vvhh-vv*hh)/(vv*BOLTZMANN*tt*tt);
- dcp = (vv*AVOGADRO/nmol)*tt*sqr(alpha)/(kappa);
++ v = edat->s[ii[eVol]].ener[j]*NANO3;
++ h = KILO*edat->s[ii[eEnth]].ener[j]/AVOGADRO;
++ v_sum += v;
++ h_sum += h;
++ vh_sum += (v*h);
+ }
- if (vvhh != NOTSET)
- {
- fprintf(fp, "vvhh = %10g (m^3 J)\n", vvhh);
- }
++ vh_aver = vh_sum / edat->nframes;
++ v_aver = v_sum / edat->nframes;
++ h_aver = h_sum / edat->nframes;
++ alpha = (vh_aver-v_aver*h_aver)/(v_aver*BOLTZMANN*tt*tt);
++ dcp = (v_aver*AVOGADRO/nmol)*tt*sqr(alpha)/(kappa);
+ }
+
+ if (tt != NOTSET)
+ {
+ if (nmol < 2)
+ {
+ fprintf(fp, "\nWARNING: nmol = %d, this may not be what you want.\n",
+ nmol);
+ }
+ fprintf(fp, "\nTemperature dependent fluctuation properties at T = %g.\n", tt);
+ fprintf(fp, "\nHeat capacities obtained from fluctuations do *not* include\n");
+ fprintf(fp, "quantum corrections. If you want to get a more accurate estimate\n");
+ fprintf(fp, "please use the g_dos program.\n\n");
+ fprintf(fp, "WARNING: Please verify that your simulations are converged and perform\n"
+ "a block-averaging error analysis (not implemented in g_energy yet)\n");
+
+ if (debug != NULL)
+ {
+ if (varv != NOTSET)
+ {
+ fprintf(fp, "varv = %10g (m^6)\n", varv*AVOGADRO/nmol);
+ }
+ }
+ if (vv != NOTSET)
+ {
+ fprintf(fp, "Volume = %10g m^3/mol\n",
+ vv*AVOGADRO/nmol);
+ }
+ if (varh != NOTSET)
+ {
+ fprintf(fp, "Enthalpy = %10g kJ/mol\n",
+ hh*AVOGADRO/(KILO*nmol));
+ }
+ if (alpha != NOTSET)
+ {
+ fprintf(fp, "Coefficient of Thermal Expansion Alpha_P = %10g (1/K)\n",
+ alpha);
+ }
+ if (kappa != NOTSET)
+ {
+ fprintf(fp, "Isothermal Compressibility Kappa = %10g (J/m^3)\n",
+ kappa);
+ fprintf(fp, "Adiabatic bulk modulus = %10g (m^3/J)\n",
+ 1.0/kappa);
+ }
+ if (cp != NOTSET)
+ {
+ fprintf(fp, "Heat capacity at constant pressure Cp = %10g J/mol K\n",
+ cp);
+ }
+ if (cv != NOTSET)
+ {
+ fprintf(fp, "Heat capacity at constant volume Cv = %10g J/mol K\n",
+ cv);
+ }
+ if (dcp != NOTSET)
+ {
+ fprintf(fp, "Cp-Cv = %10g J/mol K\n",
+ dcp);
+ }
+ please_cite(fp, "Allen1987a");
+ }
+ else
+ {
+ fprintf(fp, "You should select the temperature in order to obtain fluctuation properties.\n");
+ }
+}
+
+static void analyse_ener(gmx_bool bCorr, const char *corrfn,
+ gmx_bool bFee, gmx_bool bSum, gmx_bool bFluct,
+ gmx_bool bVisco, const char *visfn, int nmol,
+ gmx_int64_t start_step, double start_t,
+ gmx_int64_t step, double t,
+ double time[], real reftemp,
+ enerdata_t *edat,
+ int nset, int set[], gmx_bool *bIsEner,
+ char **leg, gmx_enxnm_t *enm,
+ real Vaver, real ezero,
+ int nbmin, int nbmax,
+ const output_env_t oenv)
+{
+ FILE *fp;
+ /* Check out the printed manual for equations! */
+ double Dt, aver, stddev, errest, delta_t, totaldrift;
+ enerdata_t *esum = NULL;
+ real xxx, integral, intBulk, Temp = 0, Pres = 0;
+ real sfrac, oldfrac, diffsum, diffav, fstep, pr_aver, pr_stddev, pr_errest;
+ double beta = 0, expE, expEtot, *fee = NULL;
+ gmx_int64_t nsteps;
+ int nexact, nnotexact;
+ double x1m, x1mk;
+ int i, j, k, nout;
+ real chi2;
+ char buf[256], eebuf[100];
+
+ nsteps = step - start_step + 1;
+ if (nsteps < 1)
+ {
+ fprintf(stdout, "Not enough steps (%s) for statistics\n",
+ gmx_step_str(nsteps, buf));
+ }
+ else
+ {
+ /* Calculate the time difference */
+ delta_t = t - start_t;
+
+ fprintf(stdout, "\nStatistics over %s steps [ %.4f through %.4f ps ], %d data sets\n",
+ gmx_step_str(nsteps, buf), start_t, t, nset);
+
+ calc_averages(nset, edat, nbmin, nbmax);
+
+ if (bSum)
+ {
+ esum = calc_sum(nset, edat, nbmin, nbmax);
+ }
+
+ if (edat->npoints == 0)
+ {
+ nexact = 0;
+ nnotexact = nset;
+ }
+ else
+ {
+ nexact = 0;
+ nnotexact = 0;
+ for (i = 0; (i < nset); i++)
+ {
+ if (edat->s[i].bExactStat)
+ {
+ nexact++;
+ }
+ else
+ {
+ nnotexact++;
+ }
+ }
+ }
+
+ if (nnotexact == 0)
+ {
+ fprintf(stdout, "All statistics are over %s points\n",
+ gmx_step_str(edat->npoints, buf));
+ }
+ else if (nexact == 0 || edat->npoints == edat->nframes)
+ {
+ fprintf(stdout, "All statistics are over %d points (frames)\n",
+ edat->nframes);
+ }
+ else
+ {
+ fprintf(stdout, "The term%s", nnotexact == 1 ? "" : "s");
+ for (i = 0; (i < nset); i++)
+ {
+ if (!edat->s[i].bExactStat)
+ {
+ fprintf(stdout, " '%s'", leg[i]);
+ }
+ }
+ fprintf(stdout, " %s has statistics over %d points (frames)\n",
+ nnotexact == 1 ? "is" : "are", edat->nframes);
+ fprintf(stdout, "All other statistics are over %s points\n",
+ gmx_step_str(edat->npoints, buf));
+ }
+ fprintf(stdout, "\n");
+
+ fprintf(stdout, "%-24s %10s %10s %10s %10s",
+ "Energy", "Average", "Err.Est.", "RMSD", "Tot-Drift");
+ if (bFee)
+ {
+ fprintf(stdout, " %10s\n", "-kT ln<e^(E/kT)>");
+ }
+ else
+ {
+ fprintf(stdout, "\n");
+ }
+ fprintf(stdout, "-------------------------------------------------------------------------------\n");
+
+ /* Initiate locals, only used with -sum */
+ expEtot = 0;
+ if (bFee)
+ {
+ beta = 1.0/(BOLTZ*reftemp);
+ snew(fee, nset);
+ }
+ for (i = 0; (i < nset); i++)
+ {
+ aver = edat->s[i].av;
+ stddev = edat->s[i].rmsd;
+ errest = edat->s[i].ee;
+
+ if (bFee)
+ {
+ expE = 0;
+ for (j = 0; (j < edat->nframes); j++)
+ {
+ expE += exp(beta*(edat->s[i].ener[j] - aver)/nmol);
+ }
+ if (bSum)
+ {
+ expEtot += expE/edat->nframes;
+ }
+
+ fee[i] = log(expE/edat->nframes)/beta + aver/nmol;
+ }
+ if (strstr(leg[i], "empera") != NULL)
+ {
+ Temp = aver;
+ }
+ else if (strstr(leg[i], "olum") != NULL)
+ {
+ Vaver = aver;
+ }
+ else if (strstr(leg[i], "essure") != NULL)
+ {
+ Pres = aver;
+ }
+ if (bIsEner[i])
+ {
+ pr_aver = aver/nmol-ezero;
+ pr_stddev = stddev/nmol;
+ pr_errest = errest/nmol;
+ }
+ else
+ {
+ pr_aver = aver;
+ pr_stddev = stddev;
+ pr_errest = errest;
+ }
+
+ /* Multiply the slope in steps with the number of steps taken */
+ totaldrift = (edat->nsteps - 1)*edat->s[i].slope;
+ if (bIsEner[i])
+ {
+ totaldrift /= nmol;
+ }
+
+ fprintf(stdout, "%-24s %10g %10s %10g %10g",
+ leg[i], pr_aver, ee_pr(pr_errest, eebuf), pr_stddev, totaldrift);
+ if (bFee)
+ {
+ fprintf(stdout, " %10g", fee[i]);
+ }
+
+ fprintf(stdout, " (%s)\n", enm[set[i]].unit);
+
+ if (bFluct)
+ {
+ for (j = 0; (j < edat->nframes); j++)
+ {
+ edat->s[i].ener[j] -= aver;
+ }
+ }
+ }
+ if (bSum)
+ {
+ totaldrift = (edat->nsteps - 1)*esum->s[0].slope;
+ fprintf(stdout, "%-24s %10g %10s %10s %10g (%s)",
+ "Total", esum->s[0].av/nmol, ee_pr(esum->s[0].ee/nmol, eebuf),
+ "--", totaldrift/nmol, enm[set[0]].unit);
+ /* pr_aver,pr_stddev,a,totaldrift */
+ if (bFee)
+ {
+ fprintf(stdout, " %10g %10g\n",
+ log(expEtot)/beta + esum->s[0].av/nmol, log(expEtot)/beta);
+ }
+ else
+ {
+ fprintf(stdout, "\n");
+ }
+ }
+
+ /* Do correlation function */
+ if (edat->nframes > 1)
+ {
+ Dt = delta_t/(edat->nframes - 1);
+ }
+ else
+ {
+ Dt = 0;
+ }
+ if (bVisco)
+ {
+ const char* leg[] = { "Shear", "Bulk" };
+ real factor;
+ real **eneset;
+ real **enesum;
+
+ /* Assume pressure tensor is in Pxx Pxy Pxz Pyx Pyy Pyz Pzx Pzy Pzz */
+
+ /* Symmetrise tensor! (and store in first three elements)
+ * And subtract average pressure!
+ */
+ snew(eneset, 12);
+ for (i = 0; i < 12; i++)
+ {
+ snew(eneset[i], edat->nframes);
+ }
+ snew(enesum, 3);
+ for (i = 0; i < 3; i++)
+ {
+ snew(enesum[i], edat->nframes);
+ }
+ for (i = 0; (i < edat->nframes); i++)
+ {
+ eneset[0][i] = 0.5*(edat->s[1].ener[i]+edat->s[3].ener[i]);
+ eneset[1][i] = 0.5*(edat->s[2].ener[i]+edat->s[6].ener[i]);
+ eneset[2][i] = 0.5*(edat->s[5].ener[i]+edat->s[7].ener[i]);
+ for (j = 3; j <= 11; j++)
+ {
+ eneset[j][i] = edat->s[j].ener[i];
+ }
+ eneset[11][i] -= Pres;
+ enesum[0][i] = 0.5*(edat->s[1].es[i].sum+edat->s[3].es[i].sum);
+ enesum[1][i] = 0.5*(edat->s[2].es[i].sum+edat->s[6].es[i].sum);
+ enesum[2][i] = 0.5*(edat->s[5].es[i].sum+edat->s[7].es[i].sum);
+ }
+
+ einstein_visco("evisco.xvg", "eviscoi.xvg",
+ 3, edat->nframes, enesum, Vaver, Temp, nsteps, time, oenv);
+
+ /*do_autocorr(corrfn,buf,nenergy,3,eneset,Dt,eacNormal,TRUE);*/
+ /* Do it for shear viscosity */
+ strcpy(buf, "Shear Viscosity");
+ low_do_autocorr(corrfn, oenv, buf, edat->nframes, 3,
+ (edat->nframes+1)/2, eneset, Dt,
+ eacNormal, 1, TRUE, FALSE, FALSE, 0.0, 0.0, 0);
+
+ /* Now for bulk viscosity */
+ strcpy(buf, "Bulk Viscosity");
+ low_do_autocorr(corrfn, oenv, buf, edat->nframes, 1,
+ (edat->nframes+1)/2, &(eneset[11]), Dt,
+ eacNormal, 1, TRUE, FALSE, FALSE, 0.0, 0.0, 0);
+
+ factor = (Vaver*1e-26/(BOLTZMANN*Temp))*Dt;
+ fp = xvgropen(visfn, buf, "Time (ps)", "\\8h\\4 (cp)", oenv);
+ xvgr_legend(fp, asize(leg), leg, oenv);
+
+ /* Use trapezium rule for integration */
+ integral = 0;
+ intBulk = 0;
+ nout = get_acfnout();
+ if ((nout < 2) || (nout >= edat->nframes/2))
+ {
+ nout = edat->nframes/2;
+ }
+ for (i = 1; (i < nout); i++)
+ {
+ integral += 0.5*(eneset[0][i-1] + eneset[0][i])*factor;
+ intBulk += 0.5*(eneset[11][i-1] + eneset[11][i])*factor;
+ fprintf(fp, "%10g %10g %10g\n", (i*Dt), integral, intBulk);
+ }
+ ffclose(fp);
+ }
+ else if (bCorr)
+ {
+ if (bFluct)
+ {
+ strcpy(buf, "Autocorrelation of Energy Fluctuations");
+ }
+ else
+ {
+ strcpy(buf, "Energy Autocorrelation");
+ }
+#if 0
+ do_autocorr(corrfn, oenv, buf, edat->nframes,
+ bSum ? 1 : nset,
+ bSum ? &edat->s[nset-1].ener : eneset,
+ (delta_t/edat->nframes), eacNormal, FALSE);
+#endif
+ }
+ }
+}
+
+static void print_time(FILE *fp, double t)
+{
+ fprintf(fp, "%12.6f", t);
+}
+
+static void print1(FILE *fp, gmx_bool bDp, real e)
+{
+ if (bDp)
+ {
+ fprintf(fp, " %16.12f", e);
+ }
+ else
+ {
+ fprintf(fp, " %10.6f", e);
+ }
+}
+
+static void fec(const char *ene2fn, const char *runavgfn,
+ real reftemp, int nset, int set[], char *leg[],
+ enerdata_t *edat, double time[],
+ const output_env_t oenv)
+{
+ const char * ravgleg[] = {
+ "\\8D\\4E = E\\sB\\N-E\\sA\\N",
+ "<e\\S-\\8D\\4E/kT\\N>\\s0..t\\N"
+ };
+ FILE *fp;
+ ener_file_t enx;
+ int nre, timecheck, step, nenergy, nenergy2, maxenergy;
+ int i, j;
+ gmx_bool bCont;
+ real aver, beta;
+ real **eneset2;
+ double dE, sum;
+ gmx_enxnm_t *enm = NULL;
+ t_enxframe *fr;
+ char buf[22];
+
+ /* read second energy file */
+ snew(fr, 1);
+ enm = NULL;
+ enx = open_enx(ene2fn, "r");
+ do_enxnms(enx, &(fr->nre), &enm);
+
+ snew(eneset2, nset+1);
+ nenergy2 = 0;
+ maxenergy = 0;
+ timecheck = 0;
+ do
+ {
+ /* This loop searches for the first frame (when -b option is given),
+ * or when this has been found it reads just one energy frame
+ */
+ do
+ {
+ bCont = do_enx(enx, fr);
+
+ if (bCont)
+ {
+ timecheck = check_times(fr->t);
+ }
+
+ }
+ while (bCont && (timecheck < 0));
+
+ /* Store energies for analysis afterwards... */
+ if ((timecheck == 0) && bCont)
+ {
+ if (fr->nre > 0)
+ {
+ if (nenergy2 >= maxenergy)
+ {
+ maxenergy += 1000;
+ for (i = 0; i <= nset; i++)
+ {
+ srenew(eneset2[i], maxenergy);
+ }
+ }
+ if (fr->t != time[nenergy2])
+ {
+ fprintf(stderr, "\nWARNING time mismatch %g!=%g at frame %s\n",
+ fr->t, time[nenergy2], gmx_step_str(fr->step, buf));
+ }
+ for (i = 0; i < nset; i++)
+ {
+ eneset2[i][nenergy2] = fr->ener[set[i]].e;
+ }
+ nenergy2++;
+ }
+ }
+ }
+ while (bCont && (timecheck == 0));
+
+ /* check */
+ if (edat->nframes != nenergy2)
+ {
+ fprintf(stderr, "\nWARNING file length mismatch %d!=%d\n",
+ edat->nframes, nenergy2);
+ }
+ nenergy = min(edat->nframes, nenergy2);
+
+ /* calculate fe difference dF = -kT ln < exp(-(E_B-E_A)/kT) >_A */
+ fp = NULL;
+ if (runavgfn)
+ {
+ fp = xvgropen(runavgfn, "Running average free energy difference",
+ "Time (" unit_time ")", "\\8D\\4E (" unit_energy ")", oenv);
+ xvgr_legend(fp, asize(ravgleg), ravgleg, oenv);
+ }
+ fprintf(stdout, "\n%-24s %10s\n",
+ "Energy", "dF = -kT ln < exp(-(EB-EA)/kT) >A");
+ sum = 0;
+ beta = 1.0/(BOLTZ*reftemp);
+ for (i = 0; i < nset; i++)
+ {
+ if (gmx_strcasecmp(leg[i], enm[set[i]].name) != 0)
+ {
+ fprintf(stderr, "\nWARNING energy set name mismatch %s!=%s\n",
+ leg[i], enm[set[i]].name);
+ }
+ for (j = 0; j < nenergy; j++)
+ {
+ dE = eneset2[i][j] - edat->s[i].ener[j];
+ sum += exp(-dE*beta);
+ if (fp)
+ {
+ fprintf(fp, "%10g %10g %10g\n",
+ time[j], dE, -BOLTZ*reftemp*log(sum/(j+1)) );
+ }
+ }
+ aver = -BOLTZ*reftemp*log(sum/nenergy);
+ fprintf(stdout, "%-24s %10g\n", leg[i], aver);
+ }
+ if (fp)
+ {
+ ffclose(fp);
+ }
+ sfree(fr);
+}
+
+
+static void do_dhdl(t_enxframe *fr, t_inputrec *ir, FILE **fp_dhdl,
+ const char *filename, gmx_bool bDp,
+ int *blocks, int *hists, int *samples, int *nlambdas,
+ const output_env_t oenv)
+{
+ const char *dhdl = "dH/d\\lambda", *deltag = "\\DeltaH", *lambda = "\\lambda";
+ char title[STRLEN], label_x[STRLEN], label_y[STRLEN], legend[STRLEN];
+ char buf[STRLEN];
+ gmx_bool first = FALSE;
+ int nblock_hist = 0, nblock_dh = 0, nblock_dhcoll = 0;
+ int i, j, k;
+ /* coll data */
+ double temp = 0, start_time = 0, delta_time = 0, start_lambda = 0, delta_lambda = 0;
+ static int setnr = 0;
+ double *native_lambda_vec = NULL;
+ const char **lambda_components = NULL;
+ int n_lambda_vec = 0;
+ gmx_bool changing_lambda = FALSE;
+ int lambda_fep_state;
+
+ /* now count the blocks & handle the global dh data */
+ for (i = 0; i < fr->nblock; i++)
+ {
+ if (fr->block[i].id == enxDHHIST)
+ {
+ nblock_hist++;
+ }
+ else if (fr->block[i].id == enxDH)
+ {
+ nblock_dh++;
+ }
+ else if (fr->block[i].id == enxDHCOLL)
+ {
+ nblock_dhcoll++;
+ if ( (fr->block[i].nsub < 1) ||
+ (fr->block[i].sub[0].type != xdr_datatype_double) ||
+ (fr->block[i].sub[0].nr < 5))
+ {
+ gmx_fatal(FARGS, "Unexpected block data");
+ }
+
+ /* read the data from the DHCOLL block */
+ temp = fr->block[i].sub[0].dval[0];
+ start_time = fr->block[i].sub[0].dval[1];
+ delta_time = fr->block[i].sub[0].dval[2];
+ start_lambda = fr->block[i].sub[0].dval[3];
+ delta_lambda = fr->block[i].sub[0].dval[4];
+ changing_lambda = (delta_lambda != 0);
+ if (fr->block[i].nsub > 1)
+ {
+ lambda_fep_state = fr->block[i].sub[1].ival[0];
+ if (n_lambda_vec == 0)
+ {
+ n_lambda_vec = fr->block[i].sub[1].ival[1];
+ }
+ else
+ {
+ if (n_lambda_vec != fr->block[i].sub[1].ival[1])
+ {
+ gmx_fatal(FARGS,
+ "Unexpected change of basis set in lambda");
+ }
+ }
+ if (lambda_components == NULL)
+ {
+ snew(lambda_components, n_lambda_vec);
+ }
+ if (native_lambda_vec == NULL)
+ {
+ snew(native_lambda_vec, n_lambda_vec);
+ }
+ for (j = 0; j < n_lambda_vec; j++)
+ {
+ native_lambda_vec[j] = fr->block[i].sub[0].dval[5+j];
+ lambda_components[j] =
+ efpt_singular_names[fr->block[i].sub[1].ival[2+j]];
+ }
+ }
+ }
+ }
+
+ if (nblock_hist == 0 && nblock_dh == 0)
+ {
+ /* don't do anything */
+ return;
+ }
+ if (nblock_hist > 0 && nblock_dh > 0)
+ {
+ gmx_fatal(FARGS, "This energy file contains both histogram dhdl data and non-histogram dhdl data. Don't know what to do.");
+ }
+ if (!*fp_dhdl)
+ {
+ if (nblock_dh > 0)
+ {
+ /* we have standard, non-histogram data --
+ call open_dhdl to open the file */
+ /* TODO this is an ugly hack that needs to be fixed: this will only
+ work if the order of data is always the same and if we're
+ only using the g_energy compiled with the mdrun that produced
+ the ener.edr. */
+ *fp_dhdl = open_dhdl(filename, ir, oenv);
+ }
+ else
+ {
+ sprintf(title, "N(%s)", deltag);
+ sprintf(label_x, "%s (%s)", deltag, unit_energy);
+ sprintf(label_y, "Samples");
+ *fp_dhdl = xvgropen_type(filename, title, label_x, label_y, exvggtXNY, oenv);
+ sprintf(buf, "T = %g (K), %s = %g", temp, lambda, start_lambda);
+ xvgr_subtitle(*fp_dhdl, buf, oenv);
+ }
+ }
+
+ (*hists) += nblock_hist;
+ (*blocks) += nblock_dh;
+ (*nlambdas) = nblock_hist+nblock_dh;
+
+ /* write the data */
+ if (nblock_hist > 0)
+ {
+ gmx_int64_t sum = 0;
+ /* histograms */
+ for (i = 0; i < fr->nblock; i++)
+ {
+ t_enxblock *blk = &(fr->block[i]);
+ if (blk->id == enxDHHIST)
+ {
+ double foreign_lambda, dx;
+ gmx_int64_t x0;
+ int nhist, derivative;
+
+ /* check the block types etc. */
+ if ( (blk->nsub < 2) ||
+ (blk->sub[0].type != xdr_datatype_double) ||
+ (blk->sub[1].type != xdr_datatype_int64) ||
+ (blk->sub[0].nr < 2) ||
+ (blk->sub[1].nr < 2) )
+ {
+ gmx_fatal(FARGS, "Unexpected block data in file");
+ }
+ foreign_lambda = blk->sub[0].dval[0];
+ dx = blk->sub[0].dval[1];
+ nhist = blk->sub[1].lval[0];
+ derivative = blk->sub[1].lval[1];
+ for (j = 0; j < nhist; j++)
+ {
+ const char *lg[1];
+ x0 = blk->sub[1].lval[2+j];
+
+ if (!derivative)
+ {
+ sprintf(legend, "N(%s(%s=%g) | %s=%g)",
+ deltag, lambda, foreign_lambda,
+ lambda, start_lambda);
+ }
+ else
+ {
+ sprintf(legend, "N(%s | %s=%g)",
+ dhdl, lambda, start_lambda);
+ }
+
+ lg[0] = legend;
+ xvgr_new_dataset(*fp_dhdl, setnr, 1, lg, oenv);
+ setnr++;
+ for (k = 0; k < blk->sub[j+2].nr; k++)
+ {
+ int hist;
+ double xmin, xmax;
+
+ hist = blk->sub[j+2].ival[k];
+ xmin = (x0+k)*dx;
+ xmax = (x0+k+1)*dx;
+ fprintf(*fp_dhdl, "%g %d\n%g %d\n", xmin, hist,
+ xmax, hist);
+ sum += hist;
+ }
+ /* multiple histogram data blocks in one histogram
+ mean that the second one is the reverse of the first one:
+ for dhdl derivatives, it's important to know both the
+ maximum and minimum values */
+ dx = -dx;
+ }
+ }
+ }
+ (*samples) += (int)(sum/nblock_hist);
+ }
+ else
+ {
+ /* raw dh */
+ int len = 0;
+ char **setnames = NULL;
+ int nnames = nblock_dh;
+
+ for (i = 0; i < fr->nblock; i++)
+ {
+ t_enxblock *blk = &(fr->block[i]);
+ if (blk->id == enxDH)
+ {
+ if (len == 0)
+ {
+ len = blk->sub[2].nr;
+ }
+ else
+ {
+ if (len != blk->sub[2].nr)
+ {
+ gmx_fatal(FARGS, "Length inconsistency in dhdl data");
+ }
+ }
+ }
+ }
+ (*samples) += len;
+
+ for (i = 0; i < len; i++)
+ {
+ double time = start_time + delta_time*i;
+
+ fprintf(*fp_dhdl, "%.4f ", time);
+
+ for (j = 0; j < fr->nblock; j++)
+ {
+ t_enxblock *blk = &(fr->block[j]);
+ if (blk->id == enxDH)
+ {
+ double value;
+ if (blk->sub[2].type == xdr_datatype_float)
+ {
+ value = blk->sub[2].fval[i];
+ }
+ else
+ {
+ value = blk->sub[2].dval[i];
+ }
+ /* we need to decide which data type it is based on the count*/
+
+ if (j == 1 && ir->bExpanded)
+ {
+ fprintf(*fp_dhdl, "%4d", (int)value); /* if expanded ensembles and zero, this is a state value, it's an integer. We need a cleaner conditional than if j==1! */
+ }
+ else
+ {
+ if (bDp)
+ {
+ fprintf(*fp_dhdl, " %#.12g", value); /* print normal precision */
+ }
+ else
+ {
+ fprintf(*fp_dhdl, " %#.8g", value); /* print normal precision */
+ }
+ }
+ }
+ }
+ fprintf(*fp_dhdl, "\n");
+ }
+ }
+}
+
+
+int gmx_energy(int argc, char *argv[])
+{
+ const char *desc[] = {
+ "[THISMODULE] extracts energy components or distance restraint",
+ "data from an energy file. The user is prompted to interactively",
+ "select the desired energy terms.[PAR]",
+
+ "Average, RMSD, and drift are calculated with full precision from the",
+ "simulation (see printed manual). Drift is calculated by performing",
+ "a least-squares fit of the data to a straight line. The reported total drift",
+ "is the difference of the fit at the first and last point.",
+ "An error estimate of the average is given based on a block averages",
+ "over 5 blocks using the full-precision averages. The error estimate",
+ "can be performed over multiple block lengths with the options",
+ "[TT]-nbmin[tt] and [TT]-nbmax[tt].",
+ "[BB]Note[bb] that in most cases the energy files contains averages over all",
+ "MD steps, or over many more points than the number of frames in",
+ "energy file. This makes the [THISMODULE] statistics output more accurate",
+ "than the [TT].xvg[tt] output. When exact averages are not present in the energy",
+ "file, the statistics mentioned above are simply over the single, per-frame",
+ "energy values.[PAR]",
+
+ "The term fluctuation gives the RMSD around the least-squares fit.[PAR]",
+
+ "Some fluctuation-dependent properties can be calculated provided",
+ "the correct energy terms are selected, and that the command line option",
+ "[TT]-fluct_props[tt] is given. The following properties",
+ "will be computed:[BR]",
+ "Property Energy terms needed[BR]",
+ "---------------------------------------------------[BR]",
+ "Heat capacity C[SUB]p[sub] (NPT sims): Enthalpy, Temp [BR]",
+ "Heat capacity C[SUB]v[sub] (NVT sims): Etot, Temp [BR]",
+ "Thermal expansion coeff. (NPT): Enthalpy, Vol, Temp[BR]",
+ "Isothermal compressibility: Vol, Temp [BR]",
+ "Adiabatic bulk modulus: Vol, Temp [BR]",
+ "---------------------------------------------------[BR]",
+ "You always need to set the number of molecules [TT]-nmol[tt].",
+ "The C[SUB]p[sub]/C[SUB]v[sub] computations do [BB]not[bb] include any corrections",
+ "for quantum effects. Use the [gmx-dos] program if you need that (and you do).[PAR]"
+ "When the [TT]-viol[tt] option is set, the time averaged",
+ "violations are plotted and the running time-averaged and",
+ "instantaneous sum of violations are recalculated. Additionally",
+ "running time-averaged and instantaneous distances between",
+ "selected pairs can be plotted with the [TT]-pairs[tt] option.[PAR]",
+
+ "Options [TT]-ora[tt], [TT]-ort[tt], [TT]-oda[tt], [TT]-odr[tt] and",
+ "[TT]-odt[tt] are used for analyzing orientation restraint data.",
+ "The first two options plot the orientation, the last three the",
+ "deviations of the orientations from the experimental values.",
+ "The options that end on an 'a' plot the average over time",
+ "as a function of restraint. The options that end on a 't'",
+ "prompt the user for restraint label numbers and plot the data",
+ "as a function of time. Option [TT]-odr[tt] plots the RMS",
+ "deviation as a function of restraint.",
+ "When the run used time or ensemble averaged orientation restraints,",
+ "option [TT]-orinst[tt] can be used to analyse the instantaneous,",
+ "not ensemble-averaged orientations and deviations instead of",
+ "the time and ensemble averages.[PAR]",
+
+ "Option [TT]-oten[tt] plots the eigenvalues of the molecular order",
+ "tensor for each orientation restraint experiment. With option",
+ "[TT]-ovec[tt] also the eigenvectors are plotted.[PAR]",
+
+ "Option [TT]-odh[tt] extracts and plots the free energy data",
+ "(Hamiltoian differences and/or the Hamiltonian derivative dhdl)",
+ "from the [TT]ener.edr[tt] file.[PAR]",
+
+ "With [TT]-fee[tt] an estimate is calculated for the free-energy",
+ "difference with an ideal gas state: [BR]",
+ " [GRK]Delta[grk] A = A(N,V,T) - A[SUB]idealgas[sub](N,V,T) = kT [LN][CHEVRON][EXP]U[SUB]pot[sub]/kT[exp][chevron][ln][BR]",
+ " [GRK]Delta[grk] G = G(N,p,T) - G[SUB]idealgas[sub](N,p,T) = kT [LN][CHEVRON][EXP]U[SUB]pot[sub]/kT[exp][chevron][ln][BR]",
+ "where k is Boltzmann's constant, T is set by [TT]-fetemp[tt] and",
+ "the average is over the ensemble (or time in a trajectory).",
+ "Note that this is in principle",
+ "only correct when averaging over the whole (Boltzmann) ensemble",
+ "and using the potential energy. This also allows for an entropy",
+ "estimate using:[BR]",
+ " [GRK]Delta[grk] S(N,V,T) = S(N,V,T) - S[SUB]idealgas[sub](N,V,T) = ([CHEVRON]U[SUB]pot[sub][chevron] - [GRK]Delta[grk] A)/T[BR]",
+ " [GRK]Delta[grk] S(N,p,T) = S(N,p,T) - S[SUB]idealgas[sub](N,p,T) = ([CHEVRON]U[SUB]pot[sub][chevron] + pV - [GRK]Delta[grk] G)/T",
+ "[PAR]",
+
+ "When a second energy file is specified ([TT]-f2[tt]), a free energy",
+ "difference is calculated [BR] dF = -kT [LN][CHEVRON][EXP]-(E[SUB]B[sub]-E[SUB]A[sub])/kT[exp][chevron][SUB]A[sub][ln] ,",
+ "where E[SUB]A[sub] and E[SUB]B[sub] are the energies from the first and second energy",
+ "files, and the average is over the ensemble A. The running average",
+ "of the free energy difference is printed to a file specified by [TT]-ravg[tt].",
+ "[BB]Note[bb] that the energies must both be calculated from the same trajectory."
+
+ };
+ static gmx_bool bSum = FALSE, bFee = FALSE, bPrAll = FALSE, bFluct = FALSE, bDriftCorr = FALSE;
+ static gmx_bool bDp = FALSE, bMutot = FALSE, bOrinst = FALSE, bOvec = FALSE, bFluctProps = FALSE;
+ static int skip = 0, nmol = 1, nbmin = 5, nbmax = 5;
+ static real reftemp = 300.0, ezero = 0;
+ t_pargs pa[] = {
+ { "-fee", FALSE, etBOOL, {&bFee},
+ "Do a free energy estimate" },
+ { "-fetemp", FALSE, etREAL, {&reftemp},
+ "Reference temperature for free energy calculation" },
+ { "-zero", FALSE, etREAL, {&ezero},
+ "Subtract a zero-point energy" },
+ { "-sum", FALSE, etBOOL, {&bSum},
+ "Sum the energy terms selected rather than display them all" },
+ { "-dp", FALSE, etBOOL, {&bDp},
+ "Print energies in high precision" },
+ { "-nbmin", FALSE, etINT, {&nbmin},
+ "Minimum number of blocks for error estimate" },
+ { "-nbmax", FALSE, etINT, {&nbmax},
+ "Maximum number of blocks for error estimate" },
+ { "-mutot", FALSE, etBOOL, {&bMutot},
+ "Compute the total dipole moment from the components" },
+ { "-skip", FALSE, etINT, {&skip},
+ "Skip number of frames between data points" },
+ { "-aver", FALSE, etBOOL, {&bPrAll},
+ "Also print the exact average and rmsd stored in the energy frames (only when 1 term is requested)" },
+ { "-nmol", FALSE, etINT, {&nmol},
+ "Number of molecules in your sample: the energies are divided by this number" },
+ { "-fluct_props", FALSE, etBOOL, {&bFluctProps},
+ "Compute properties based on energy fluctuations, like heat capacity" },
+ { "-driftcorr", FALSE, etBOOL, {&bDriftCorr},
+ "Useful only for calculations of fluctuation properties. The drift in the observables will be subtracted before computing the fluctuation properties."},
+ { "-fluc", FALSE, etBOOL, {&bFluct},
+ "Calculate autocorrelation of energy fluctuations rather than energy itself" },
+ { "-orinst", FALSE, etBOOL, {&bOrinst},
+ "Analyse instantaneous orientation data" },
+ { "-ovec", FALSE, etBOOL, {&bOvec},
+ "Also plot the eigenvectors with [TT]-oten[tt]" }
+ };
+ const char * drleg[] = {
+ "Running average",
+ "Instantaneous"
+ };
+ static const char *setnm[] = {
+ "Pres-XX", "Pres-XY", "Pres-XZ", "Pres-YX", "Pres-YY",
+ "Pres-YZ", "Pres-ZX", "Pres-ZY", "Pres-ZZ", "Temperature",
+ "Volume", "Pressure"
+ };
+
+ FILE *out = NULL, *fp_pairs = NULL, *fort = NULL, *fodt = NULL, *foten = NULL;
+ FILE *fp_dhdl = NULL;
+ FILE **drout;
+ ener_file_t fp;
+ int timecheck = 0;
+ gmx_mtop_t mtop;
+ gmx_localtop_t *top = NULL;
+ t_inputrec ir;
+ t_energy **ee;
+ enerdata_t edat;
+ gmx_enxnm_t *enm = NULL;
+ t_enxframe *frame, *fr = NULL;
+ int cur = 0;
+#define NEXT (1-cur)
+ int nre, teller, teller_disre, nfr;
+ gmx_int64_t start_step;
+ int nor = 0, nex = 0, norfr = 0, enx_i = 0;
+ real start_t;
+ real *bounds = NULL, *violaver = NULL, *oobs = NULL, *orient = NULL, *odrms = NULL;
+ int *index = NULL, *pair = NULL, norsel = 0, *orsel = NULL, *or_label = NULL;
+ int nbounds = 0, npairs;
+ gmx_bool bDisRe, bDRAll, bORA, bORT, bODA, bODR, bODT, bORIRE, bOTEN, bDHDL;
+ gmx_bool bFoundStart, bCont, bEDR, bVisco;
+ double sum, sumaver, sumt, ener, dbl;
+ double *time = NULL;
+ real Vaver;
+ int *set = NULL, i, j, k, nset, sss;
+ gmx_bool *bIsEner = NULL;
+ char **pairleg, **odtleg, **otenleg;
+ char **leg = NULL;
+ char **nms;
+ char *anm_j, *anm_k, *resnm_j, *resnm_k;
+ int resnr_j, resnr_k;
+ const char *orinst_sub = "@ subtitle \"instantaneous\"\n";
+ char buf[256];
+ output_env_t oenv;
+ t_enxblock *blk = NULL;
+ t_enxblock *blk_disre = NULL;
+ int ndisre = 0;
+ int dh_blocks = 0, dh_hists = 0, dh_samples = 0, dh_lambdas = 0;
+
+ t_filenm fnm[] = {
+ { efEDR, "-f", NULL, ffREAD },
+ { efEDR, "-f2", NULL, ffOPTRD },
+ { efTPX, "-s", NULL, ffOPTRD },
+ { efXVG, "-o", "energy", ffWRITE },
+ { efXVG, "-viol", "violaver", ffOPTWR },
+ { efXVG, "-pairs", "pairs", ffOPTWR },
+ { efXVG, "-ora", "orienta", ffOPTWR },
+ { efXVG, "-ort", "orientt", ffOPTWR },
+ { efXVG, "-oda", "orideva", ffOPTWR },
+ { efXVG, "-odr", "oridevr", ffOPTWR },
+ { efXVG, "-odt", "oridevt", ffOPTWR },
+ { efXVG, "-oten", "oriten", ffOPTWR },
+ { efXVG, "-corr", "enecorr", ffOPTWR },
+ { efXVG, "-vis", "visco", ffOPTWR },
+ { efXVG, "-ravg", "runavgdf", ffOPTWR },
+ { efXVG, "-odh", "dhdl", ffOPTWR }
+ };
+#define NFILE asize(fnm)
+ int npargs;
+ t_pargs *ppa;
+
+ npargs = asize(pa);
+ ppa = add_acf_pargs(&npargs, pa);
+ if (!parse_common_args(&argc, argv,
+ PCA_CAN_VIEW | PCA_CAN_BEGIN | PCA_CAN_END | PCA_BE_NICE,
+ NFILE, fnm, npargs, ppa, asize(desc), desc, 0, NULL, &oenv))
+ {
+ return 0;
+ }
+
+ bDRAll = opt2bSet("-pairs", NFILE, fnm);
+ bDisRe = opt2bSet("-viol", NFILE, fnm) || bDRAll;
+ bORA = opt2bSet("-ora", NFILE, fnm);
+ bORT = opt2bSet("-ort", NFILE, fnm);
+ bODA = opt2bSet("-oda", NFILE, fnm);
+ bODR = opt2bSet("-odr", NFILE, fnm);
+ bODT = opt2bSet("-odt", NFILE, fnm);
+ bORIRE = bORA || bORT || bODA || bODR || bODT;
+ bOTEN = opt2bSet("-oten", NFILE, fnm);
+ bDHDL = opt2bSet("-odh", NFILE, fnm);
+
+ nset = 0;
+
+ snew(frame, 2);
+ fp = open_enx(ftp2fn(efEDR, NFILE, fnm), "r");
+ do_enxnms(fp, &nre, &enm);
+
+ Vaver = -1;
+
+ bVisco = opt2bSet("-vis", NFILE, fnm);
+
+ if ((!bDisRe) && (!bDHDL))
+ {
+ if (bVisco)
+ {
+ nset = asize(setnm);
+ snew(set, nset);
+ /* This is nasty code... To extract Pres tensor, Volume and Temperature */
+ for (j = 0; j < nset; j++)
+ {
+ for (i = 0; i < nre; i++)
+ {
+ if (strstr(enm[i].name, setnm[j]))
+ {
+ set[j] = i;
+ break;
+ }
+ }
+ if (i == nre)
+ {
+ if (gmx_strcasecmp(setnm[j], "Volume") == 0)
+ {
+ printf("Enter the box volume (" unit_volume "): ");
+ if (1 != scanf("%lf", &dbl))
+ {
+ gmx_fatal(FARGS, "Error reading user input");
+ }
+ Vaver = dbl;
+ }
+ else
+ {
+ gmx_fatal(FARGS, "Could not find term %s for viscosity calculation",
+ setnm[j]);
+ }
+ }
+ }
+ }
+ else
+ {
+ set = select_by_name(nre, enm, &nset);
+ }
+ /* Print all the different units once */
+ sprintf(buf, "(%s)", enm[set[0]].unit);
+ for (i = 1; i < nset; i++)
+ {
+ for (j = 0; j < i; j++)
+ {
+ if (strcmp(enm[set[i]].unit, enm[set[j]].unit) == 0)
+ {
+ break;
+ }
+ }
+ if (j == i)
+ {
+ strcat(buf, ", (");
+ strcat(buf, enm[set[i]].unit);
+ strcat(buf, ")");
+ }
+ }
+ out = xvgropen(opt2fn("-o", NFILE, fnm), "Gromacs Energies", "Time (ps)", buf,
+ oenv);
+
+ snew(leg, nset+1);
+ for (i = 0; (i < nset); i++)
+ {
+ leg[i] = enm[set[i]].name;
+ }
+ if (bSum)
+ {
+ leg[nset] = strdup("Sum");
+ xvgr_legend(out, nset+1, (const char**)leg, oenv);
+ }
+ else
+ {
+ xvgr_legend(out, nset, (const char**)leg, oenv);
+ }
+
+ snew(bIsEner, nset);
+ for (i = 0; (i < nset); i++)
+ {
+ bIsEner[i] = FALSE;
+ for (j = 0; (j <= F_ETOT); j++)
+ {
+ bIsEner[i] = bIsEner[i] ||
+ (gmx_strcasecmp(interaction_function[j].longname, leg[i]) == 0);
+ }
+ }
+
+ if (bPrAll && nset > 1)
+ {
+ gmx_fatal(FARGS, "Printing averages can only be done when a single set is selected");
+ }
+
+ time = NULL;
+
+ if (bORIRE || bOTEN)
+ {
+ get_orires_parms(ftp2fn(efTPX, NFILE, fnm), &nor, &nex, &or_label, &oobs);
+ }
+
+ if (bORIRE)
+ {
+ if (bOrinst)
+ {
+ enx_i = enxORI;
+ }
+ else
+ {
+ enx_i = enxOR;
+ }
+
+ if (bORA || bODA)
+ {
+ snew(orient, nor);
+ }
+ if (bODR)
+ {
+ snew(odrms, nor);
+ }
+ if (bORT || bODT)
+ {
+ fprintf(stderr, "Select the orientation restraint labels you want (-1 is all)\n");
+ fprintf(stderr, "End your selection with 0\n");
+ j = -1;
+ orsel = NULL;
+ do
+ {
+ j++;
+ srenew(orsel, j+1);
+ if (1 != scanf("%d", &(orsel[j])))
+ {
+ gmx_fatal(FARGS, "Error reading user input");
+ }
+ }
+ while (orsel[j] > 0);
+ if (orsel[0] == -1)
+ {
+ fprintf(stderr, "Selecting all %d orientation restraints\n", nor);
+ norsel = nor;
+ srenew(orsel, nor);
+ for (i = 0; i < nor; i++)
+ {
+ orsel[i] = i;
+ }
+ }
+ else
+ {
+ /* Build the selection */
+ norsel = 0;
+ for (i = 0; i < j; i++)
+ {
+ for (k = 0; k < nor; k++)
+ {
+ if (or_label[k] == orsel[i])
+ {
+ orsel[norsel] = k;
+ norsel++;
+ break;
+ }
+ }
+ if (k == nor)
+ {
+ fprintf(stderr, "Orientation restraint label %d not found\n",
+ orsel[i]);
+ }
+ }
+ }
+ snew(odtleg, norsel);
+ for (i = 0; i < norsel; i++)
+ {
+ snew(odtleg[i], 256);
+ sprintf(odtleg[i], "%d", or_label[orsel[i]]);
+ }
+ if (bORT)
+ {
+ fort = xvgropen(opt2fn("-ort", NFILE, fnm), "Calculated orientations",
+ "Time (ps)", "", oenv);
+ if (bOrinst)
+ {
+ fprintf(fort, "%s", orinst_sub);
+ }
+ xvgr_legend(fort, norsel, (const char**)odtleg, oenv);
+ }
+ if (bODT)
+ {
+ fodt = xvgropen(opt2fn("-odt", NFILE, fnm),
+ "Orientation restraint deviation",
+ "Time (ps)", "", oenv);
+ if (bOrinst)
+ {
+ fprintf(fodt, "%s", orinst_sub);
+ }
+ xvgr_legend(fodt, norsel, (const char**)odtleg, oenv);
+ }
+ }
+ }
+ if (bOTEN)
+ {
+ foten = xvgropen(opt2fn("-oten", NFILE, fnm),
+ "Order tensor", "Time (ps)", "", oenv);
+ snew(otenleg, bOvec ? nex*12 : nex*3);
+ for (i = 0; i < nex; i++)
+ {
+ for (j = 0; j < 3; j++)
+ {
+ sprintf(buf, "eig%d", j+1);
+ otenleg[(bOvec ? 12 : 3)*i+j] = strdup(buf);
+ }
+ if (bOvec)
+ {
+ for (j = 0; j < 9; j++)
+ {
+ sprintf(buf, "vec%d%s", j/3+1, j%3 == 0 ? "x" : (j%3 == 1 ? "y" : "z"));
+ otenleg[12*i+3+j] = strdup(buf);
+ }
+ }
+ }
+ xvgr_legend(foten, bOvec ? nex*12 : nex*3, (const char**)otenleg, oenv);
+ }
+ }
+ else if (bDisRe)
+ {
+ nbounds = get_bounds(ftp2fn(efTPX, NFILE, fnm), &bounds, &index, &pair, &npairs,
+ &mtop, &top, &ir);
+ snew(violaver, npairs);
+ out = xvgropen(opt2fn("-o", NFILE, fnm), "Sum of Violations",
+ "Time (ps)", "nm", oenv);
+ xvgr_legend(out, 2, drleg, oenv);
+ if (bDRAll)
+ {
+ fp_pairs = xvgropen(opt2fn("-pairs", NFILE, fnm), "Pair Distances",
+ "Time (ps)", "Distance (nm)", oenv);
+ if (output_env_get_print_xvgr_codes(oenv))
+ {
+ fprintf(fp_pairs, "@ subtitle \"averaged (tau=%g) and instantaneous\"\n",
+ ir.dr_tau);
+ }
+ }
+ }
+ else if (bDHDL)
+ {
+ get_dhdl_parms(ftp2fn(efTPX, NFILE, fnm), &ir);
+ }
+
+ /* Initiate energies and set them to zero */
+ edat.nsteps = 0;
+ edat.npoints = 0;
+ edat.nframes = 0;
+ edat.step = NULL;
+ edat.steps = NULL;
+ edat.points = NULL;
+ snew(edat.s, nset);
+
+ /* Initiate counters */
+ teller = 0;
+ teller_disre = 0;
+ bFoundStart = FALSE;
+ start_step = 0;
+ start_t = 0;
+ do
+ {
+ /* This loop searches for the first frame (when -b option is given),
+ * or when this has been found it reads just one energy frame
+ */
+ do
+ {
+ bCont = do_enx(fp, &(frame[NEXT]));
+ if (bCont)
+ {
+ timecheck = check_times(frame[NEXT].t);
+ }
+ }
+ while (bCont && (timecheck < 0));
+
+ if ((timecheck == 0) && bCont)
+ {
+ /* We read a valid frame, so we can use it */
+ fr = &(frame[NEXT]);
+
+ if (fr->nre > 0)
+ {
+ /* The frame contains energies, so update cur */
+ cur = NEXT;
+
+ if (edat.nframes % 1000 == 0)
+ {
+ srenew(edat.step, edat.nframes+1000);
+ memset(&(edat.step[edat.nframes]), 0, 1000*sizeof(edat.step[0]));
+ srenew(edat.steps, edat.nframes+1000);
+ memset(&(edat.steps[edat.nframes]), 0, 1000*sizeof(edat.steps[0]));
+ srenew(edat.points, edat.nframes+1000);
+ memset(&(edat.points[edat.nframes]), 0, 1000*sizeof(edat.points[0]));
+
+ for (i = 0; i < nset; i++)
+ {
+ srenew(edat.s[i].ener, edat.nframes+1000);
+ memset(&(edat.s[i].ener[edat.nframes]), 0,
+ 1000*sizeof(edat.s[i].ener[0]));
+ srenew(edat.s[i].es, edat.nframes+1000);
+ memset(&(edat.s[i].es[edat.nframes]), 0,
+ 1000*sizeof(edat.s[i].es[0]));
+ }
+ }
+
+ nfr = edat.nframes;
+ edat.step[nfr] = fr->step;
+
+ if (!bFoundStart)
+ {
+ bFoundStart = TRUE;
+ /* Initiate the previous step data */
+ start_step = fr->step;
+ start_t = fr->t;
+ /* Initiate the energy sums */
+ edat.steps[nfr] = 1;
+ edat.points[nfr] = 1;
+ for (i = 0; i < nset; i++)
+ {
+ sss = set[i];
+ edat.s[i].es[nfr].sum = fr->ener[sss].e;
+ edat.s[i].es[nfr].sum2 = 0;
+ }
+ edat.nsteps = 1;
+ edat.npoints = 1;
+ }
+ else
+ {
+ edat.steps[nfr] = fr->nsteps;
+ {
+ if (fr->step - start_step + 1 == edat.nsteps + fr->nsteps)
+ {
+ if (fr->nsum <= 1)
+ {
+ edat.points[nfr] = 1;
+ for (i = 0; i < nset; i++)
+ {
+ sss = set[i];
+ edat.s[i].es[nfr].sum = fr->ener[sss].e;
+ edat.s[i].es[nfr].sum2 = 0;
+ }
+ edat.npoints += 1;
+ }
+ else
+ {
+ edat.points[nfr] = fr->nsum;
+ for (i = 0; i < nset; i++)
+ {
+ sss = set[i];
+ edat.s[i].es[nfr].sum = fr->ener[sss].esum;
+ edat.s[i].es[nfr].sum2 = fr->ener[sss].eav;
+ }
+ edat.npoints += fr->nsum;
+ }
+ }
+ else
+ {
+ /* The interval does not match fr->nsteps:
+ * can not do exact averages.
+ */
+ edat.npoints = 0;
+ }
+ edat.nsteps = fr->step - start_step + 1;
+ }
+ }
+ for (i = 0; i < nset; i++)
+ {
+ edat.s[i].ener[nfr] = fr->ener[set[i]].e;
+ }
+ }
+ /*
+ * Define distance restraint legends. Can only be done after
+ * the first frame has been read... (Then we know how many there are)
+ */
+ blk_disre = find_block_id_enxframe(fr, enxDISRE, NULL);
+ if (bDisRe && bDRAll && !leg && blk_disre)
+ {
+ t_iatom *fa;
+ t_iparams *ip;
+
+ fa = top->idef.il[F_DISRES].iatoms;
+ ip = top->idef.iparams;
+ if (blk_disre->nsub != 2 ||
+ (blk_disre->sub[0].nr != blk_disre->sub[1].nr) )
+ {
+ gmx_incons("Number of disre sub-blocks not equal to 2");
+ }
+
+ ndisre = blk_disre->sub[0].nr;
+ if (ndisre != top->idef.il[F_DISRES].nr/3)
+ {
+ gmx_fatal(FARGS, "Number of disre pairs in the energy file (%d) does not match the number in the run input file (%d)\n",
+ ndisre, top->idef.il[F_DISRES].nr/3);
+ }
+ snew(pairleg, ndisre);
+ for (i = 0; i < ndisre; i++)
+ {
+ snew(pairleg[i], 30);
+ j = fa[3*i+1];
+ k = fa[3*i+2];
+ gmx_mtop_atominfo_global(&mtop, j, &anm_j, &resnr_j, &resnm_j);
+ gmx_mtop_atominfo_global(&mtop, k, &anm_k, &resnr_k, &resnm_k);
+ sprintf(pairleg[i], "%d %s %d %s (%d)",
+ resnr_j, anm_j, resnr_k, anm_k,
+ ip[fa[3*i]].disres.label);
+ }
+ set = select_it(ndisre, pairleg, &nset);
+ snew(leg, 2*nset);
+ for (i = 0; (i < nset); i++)
+ {
+ snew(leg[2*i], 32);
+ sprintf(leg[2*i], "a %s", pairleg[set[i]]);
+ snew(leg[2*i+1], 32);
+ sprintf(leg[2*i+1], "i %s", pairleg[set[i]]);
+ }
+ xvgr_legend(fp_pairs, 2*nset, (const char**)leg, oenv);
+ }
+
+ /*
+ * Store energies for analysis afterwards...
+ */
+ if (!bDisRe && !bDHDL && (fr->nre > 0))
+ {
+ if (edat.nframes % 1000 == 0)
+ {
+ srenew(time, edat.nframes+1000);
+ }
+ time[edat.nframes] = fr->t;
+ edat.nframes++;
+ }
+ /*
+ * Printing time, only when we do not want to skip frames
+ */
+ if (!skip || teller % skip == 0)
+ {
+ if (bDisRe)
+ {
+ /*******************************************
+ * D I S T A N C E R E S T R A I N T S
+ *******************************************/
+ if (ndisre > 0)
+ {
+ #ifndef GMX_DOUBLE
+ float *disre_rt = blk_disre->sub[0].fval;
+ float *disre_rm3tav = blk_disre->sub[1].fval;
+ #else
+ double *disre_rt = blk_disre->sub[0].dval;
+ double *disre_rm3tav = blk_disre->sub[1].dval;
+ #endif
+
+ print_time(out, fr->t);
+ if (violaver == NULL)
+ {
+ snew(violaver, ndisre);
+ }
+
+ /* Subtract bounds from distances, to calculate violations */
+ calc_violations(disre_rt, disre_rm3tav,
+ nbounds, pair, bounds, violaver, &sumt, &sumaver);
+
+ fprintf(out, " %8.4f %8.4f\n", sumaver, sumt);
+ if (bDRAll)
+ {
+ print_time(fp_pairs, fr->t);
+ for (i = 0; (i < nset); i++)
+ {
+ sss = set[i];
+ fprintf(fp_pairs, " %8.4f", mypow(disre_rm3tav[sss], minthird));
+ fprintf(fp_pairs, " %8.4f", disre_rt[sss]);
+ }
+ fprintf(fp_pairs, "\n");
+ }
+ teller_disre++;
+ }
+ }
+ else if (bDHDL)
+ {
+ do_dhdl(fr, &ir, &fp_dhdl, opt2fn("-odh", NFILE, fnm), bDp, &dh_blocks, &dh_hists, &dh_samples, &dh_lambdas, oenv);
+ }
+
+ /*******************************************
+ * E N E R G I E S
+ *******************************************/
+ else
+ {
+ if (fr->nre > 0)
+ {
+ if (bPrAll)
+ {
+ /* We skip frames with single points (usually only the first frame),
+ * since they would result in an average plot with outliers.
+ */
+ if (fr->nsum > 1)
+ {
+ print_time(out, fr->t);
+ print1(out, bDp, fr->ener[set[0]].e);
+ print1(out, bDp, fr->ener[set[0]].esum/fr->nsum);
+ print1(out, bDp, sqrt(fr->ener[set[0]].eav/fr->nsum));
+ fprintf(out, "\n");
+ }
+ }
+ else
+ {
+ print_time(out, fr->t);
+ if (bSum)
+ {
+ sum = 0;
+ for (i = 0; i < nset; i++)
+ {
+ sum += fr->ener[set[i]].e;
+ }
+ print1(out, bDp, sum/nmol-ezero);
+ }
+ else
+ {
+ for (i = 0; (i < nset); i++)
+ {
+ if (bIsEner[i])
+ {
+ print1(out, bDp, (fr->ener[set[i]].e)/nmol-ezero);
+ }
+ else
+ {
+ print1(out, bDp, fr->ener[set[i]].e);
+ }
+ }
+ }
+ fprintf(out, "\n");
+ }
+ }
+ blk = find_block_id_enxframe(fr, enx_i, NULL);
+ if (bORIRE && blk)
+ {
+#ifndef GMX_DOUBLE
+ xdr_datatype dt = xdr_datatype_float;
+#else
+ xdr_datatype dt = xdr_datatype_double;
+#endif
+ real *vals;
+
+ if ( (blk->nsub != 1) || (blk->sub[0].type != dt) )
+ {
+ gmx_fatal(FARGS, "Orientational restraints read in incorrectly");
+ }
+#ifndef GMX_DOUBLE
+ vals = blk->sub[0].fval;
+#else
+ vals = blk->sub[0].dval;
+#endif
+
+ if (blk->sub[0].nr != (size_t)nor)
+ {
+ gmx_fatal(FARGS, "Number of orientation restraints in energy file (%d) does not match with the topology (%d)", blk->sub[0].nr);
+ }
+ if (bORA || bODA)
+ {
+ for (i = 0; i < nor; i++)
+ {
+ orient[i] += vals[i];
+ }
+ }
+ if (bODR)
+ {
+ for (i = 0; i < nor; i++)
+ {
+ odrms[i] += sqr(vals[i]-oobs[i]);
+ }
+ }
+ if (bORT)
+ {
+ fprintf(fort, " %10f", fr->t);
+ for (i = 0; i < norsel; i++)
+ {
+ fprintf(fort, " %g", vals[orsel[i]]);
+ }
+ fprintf(fort, "\n");
+ }
+ if (bODT)
+ {
+ fprintf(fodt, " %10f", fr->t);
+ for (i = 0; i < norsel; i++)
+ {
+ fprintf(fodt, " %g", vals[orsel[i]]-oobs[orsel[i]]);
+ }
+ fprintf(fodt, "\n");
+ }
+ norfr++;
+ }
+ blk = find_block_id_enxframe(fr, enxORT, NULL);
+ if (bOTEN && blk)
+ {
+#ifndef GMX_DOUBLE
+ xdr_datatype dt = xdr_datatype_float;
+#else
+ xdr_datatype dt = xdr_datatype_double;
+#endif
+ real *vals;
+
+ if ( (blk->nsub != 1) || (blk->sub[0].type != dt) )
+ {
+ gmx_fatal(FARGS, "Orientational restraints read in incorrectly");
+ }
+#ifndef GMX_DOUBLE
+ vals = blk->sub[0].fval;
+#else
+ vals = blk->sub[0].dval;
+#endif
+
+ if (blk->sub[0].nr != (size_t)(nex*12))
+ {
+ gmx_fatal(FARGS, "Number of orientation experiments in energy file (%g) does not match with the topology (%d)",
+ blk->sub[0].nr/12, nex);
+ }
+ fprintf(foten, " %10f", fr->t);
+ for (i = 0; i < nex; i++)
+ {
+ for (j = 0; j < (bOvec ? 12 : 3); j++)
+ {
+ fprintf(foten, " %g", vals[i*12+j]);
+ }
+ }
+ fprintf(foten, "\n");
+ }
+ }
+ }
+ teller++;
+ }
+ }
+ while (bCont && (timecheck == 0));
+
+ fprintf(stderr, "\n");
+ close_enx(fp);
+ if (out)
+ {
+ ffclose(out);
+ }
+
+ if (bDRAll)
+ {
+ ffclose(fp_pairs);
+ }
+
+ if (bORT)
+ {
+ ffclose(fort);
+ }
+ if (bODT)
+ {
+ ffclose(fodt);
+ }
+ if (bORA)
+ {
+ out = xvgropen(opt2fn("-ora", NFILE, fnm),
+ "Average calculated orientations",
+ "Restraint label", "", oenv);
+ if (bOrinst)
+ {
+ fprintf(out, "%s", orinst_sub);
+ }
+ for (i = 0; i < nor; i++)
+ {
+ fprintf(out, "%5d %g\n", or_label[i], orient[i]/norfr);
+ }
+ ffclose(out);
+ }
+ if (bODA)
+ {
+ out = xvgropen(opt2fn("-oda", NFILE, fnm),
+ "Average restraint deviation",
+ "Restraint label", "", oenv);
+ if (bOrinst)
+ {
+ fprintf(out, "%s", orinst_sub);
+ }
+ for (i = 0; i < nor; i++)
+ {
+ fprintf(out, "%5d %g\n", or_label[i], orient[i]/norfr-oobs[i]);
+ }
+ ffclose(out);
+ }
+ if (bODR)
+ {
+ out = xvgropen(opt2fn("-odr", NFILE, fnm),
+ "RMS orientation restraint deviations",
+ "Restraint label", "", oenv);
+ if (bOrinst)
+ {
+ fprintf(out, "%s", orinst_sub);
+ }
+ for (i = 0; i < nor; i++)
+ {
+ fprintf(out, "%5d %g\n", or_label[i], sqrt(odrms[i]/norfr));
+ }
+ ffclose(out);
+ }
+ if (bOTEN)
+ {
+ ffclose(foten);
+ }
+
+ if (bDisRe)
+ {
+ analyse_disre(opt2fn("-viol", NFILE, fnm),
+ teller_disre, violaver, bounds, index, pair, nbounds, oenv);
+ }
+ else if (bDHDL)
+ {
+ if (fp_dhdl)
+ {
+ ffclose(fp_dhdl);
+ printf("\n\nWrote %d lambda values with %d samples as ",
+ dh_lambdas, dh_samples);
+ if (dh_hists > 0)
+ {
+ printf("%d dH histograms ", dh_hists);
+ }
+ if (dh_blocks > 0)
+ {
+ printf("%d dH data blocks ", dh_blocks);
+ }
+ printf("to %s\n", opt2fn("-odh", NFILE, fnm));
+
+ }
+ else
+ {
+ gmx_fatal(FARGS, "No dH data in %s\n", opt2fn("-f", NFILE, fnm));
+ }
+
+ }
+ else
+ {
+ double dt = (frame[cur].t-start_t)/(edat.nframes-1);
+ analyse_ener(opt2bSet("-corr", NFILE, fnm), opt2fn("-corr", NFILE, fnm),
+ bFee, bSum, opt2parg_bSet("-nmol", npargs, ppa),
+ bVisco, opt2fn("-vis", NFILE, fnm),
+ nmol,
+ start_step, start_t, frame[cur].step, frame[cur].t,
+ time, reftemp, &edat,
+ nset, set, bIsEner, leg, enm, Vaver, ezero, nbmin, nbmax,
+ oenv);
+ if (bFluctProps)
+ {
+ calc_fluctuation_props(stdout, bDriftCorr, dt, nset, nmol, leg, &edat,
+ nbmin, nbmax);
+ }
+ }
+ if (opt2bSet("-f2", NFILE, fnm))
+ {
+ fec(opt2fn("-f2", NFILE, fnm), opt2fn("-ravg", NFILE, fnm),
+ reftemp, nset, set, leg, &edat, time, oenv);
+ }
+
+ {
+ const char *nxy = "-nxy";
+
+ do_view(oenv, opt2fn("-o", NFILE, fnm), nxy);
+ do_view(oenv, opt2fn_null("-ravg", NFILE, fnm), nxy);
+ do_view(oenv, opt2fn_null("-ora", NFILE, fnm), nxy);
+ do_view(oenv, opt2fn_null("-ort", NFILE, fnm), nxy);
+ do_view(oenv, opt2fn_null("-oda", NFILE, fnm), nxy);
+ do_view(oenv, opt2fn_null("-odr", NFILE, fnm), nxy);
+ do_view(oenv, opt2fn_null("-odt", NFILE, fnm), nxy);
+ do_view(oenv, opt2fn_null("-oten", NFILE, fnm), nxy);
+ do_view(oenv, opt2fn_null("-odh", NFILE, fnm), nxy);
+ }
+
+ return 0;
+}
--- /dev/null
- * Copyright (c) 2013, by the GROMACS development team, led by
+/*
+ * This file is part of the GROMACS molecular simulation package.
+ *
+ * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
+ * Copyright (c) 2001-2004, The GROMACS development team.
++ * Copyright (c) 2013,2014, by the GROMACS development team, led by
+ * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
+ * and including many others, as listed in the AUTHORS file in the
+ * top-level source directory and at http://www.gromacs.org.
+ *
+ * GROMACS is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public License
+ * as published by the Free Software Foundation; either version 2.1
+ * of the License, or (at your option) any later version.
+ *
+ * GROMACS is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with GROMACS; if not, see
+ * http://www.gnu.org/licenses, or write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * If you want to redistribute modifications to GROMACS, please
+ * consider that scientific software is very special. Version
+ * control is crucial - bugs must be traceable. We will be happy to
+ * consider code for inclusion in the official distribution, but
+ * derived work must not be called official GROMACS. Details are found
+ * in the README & COPYING files - if they are missing, get the
+ * official version at http://www.gromacs.org.
+ *
+ * To help us fund GROMACS development, we humbly ask that you cite
+ * the research papers on the package. Check out http://www.gromacs.org.
+ */
+
+#ifndef _sim_util_h
+#define _sim_util_h
+
+#include "typedefs.h"
+#include "mdebin.h"
+#include "update.h"
+#include "vcm.h"
+#include "../fileio/enxio.h"
+#include "../fileio/mdoutf.h"
+#include "../timing/walltime_accounting.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef struct gmx_global_stat *gmx_global_stat_t;
+
+void do_pbc_first(FILE *log, matrix box, t_forcerec *fr,
+ t_graph *graph, rvec x[]);
+
+void do_pbc_first_mtop(FILE *fplog, int ePBC, matrix box,
+ gmx_mtop_t *mtop, rvec x[]);
+
+void do_pbc_mtop(FILE *fplog, int ePBC, matrix box,
+ gmx_mtop_t *mtop, rvec x[]);
+
+
+
+/* ROUTINES from stat.c */
+gmx_global_stat_t global_stat_init(t_inputrec *ir);
+
+void global_stat_destroy(gmx_global_stat_t gs);
+
+void global_stat(FILE *log, gmx_global_stat_t gs,
+ t_commrec *cr, gmx_enerdata_t *enerd,
+ tensor fvir, tensor svir, rvec mu_tot,
+ t_inputrec *inputrec,
+ gmx_ekindata_t *ekind,
+ gmx_constr_t constr, t_vcm *vcm,
+ int nsig, real *sig,
+ gmx_mtop_t *top_global, t_state *state_local,
+ gmx_bool bSumEkinhOld, int flags);
+/* Communicate statistics over cr->mpi_comm_mysim */
+
+int do_per_step(gmx_int64_t step, gmx_int64_t nstep);
+/* Return TRUE if io should be done */
+
+/* ROUTINES from sim_util.c */
+
+void print_time(FILE *out, gmx_walltime_accounting_t walltime_accounting,
+ gmx_int64_t step, t_inputrec *ir, t_commrec *cr);
+
+void print_date_and_time(FILE *log, int pid, const char *title,
+ const gmx_walltime_accounting_t walltime_accounting);
+
++void print_start(FILE *fplog, t_commrec *cr,
++ gmx_walltime_accounting_t walltime_accounting,
++ const char *name);
++
+void finish_run(FILE *log, t_commrec *cr,
+ t_inputrec *inputrec,
+ t_nrnb nrnb[], gmx_wallcycle_t wcycle,
+ gmx_walltime_accounting_t walltime_accounting,
+ wallclock_gpu_t *gputimes,
+ gmx_bool bWriteStat);
+
+void calc_enervirdiff(FILE *fplog, int eDispCorr, t_forcerec *fr);
+
+void calc_dispcorr(FILE *fplog, t_inputrec *ir, t_forcerec *fr,
+ gmx_int64_t step, int natoms,
+ matrix box, real lambda, tensor pres, tensor virial,
+ real *prescorr, real *enercorr, real *dvdlcorr);
+
+void initialize_lambdas(FILE *fplog, t_inputrec *ir, int *fep_state, real *lambda, double *lam0);
+
+void do_constrain_first(FILE *log, gmx_constr_t constr,
+ t_inputrec *inputrec, t_mdatoms *md,
+ t_state *state, t_commrec *cr, t_nrnb *nrnb,
+ t_forcerec *fr, gmx_localtop_t *top);
+
+void init_md(FILE *fplog,
+ t_commrec *cr, t_inputrec *ir, const output_env_t oenv,
+ double *t, double *t0,
+ real *lambda, int *fep_state, double *lam0,
+ t_nrnb *nrnb, gmx_mtop_t *mtop,
+ gmx_update_t *upd,
+ int nfile, const t_filenm fnm[],
+ gmx_mdoutf_t *outf, t_mdebin **mdebin,
+ tensor force_vir, tensor shake_vir,
+ rvec mu_tot,
+ gmx_bool *bSimAnn, t_vcm **vcm, unsigned long Flags);
+/* Routine in sim_util.c */
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _sim_util_h */
--- /dev/null
- char buf[STRLEN];
-
+/*
+ * This file is part of the GROMACS molecular simulation package.
+ *
+ * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
+ * Copyright (c) 2001-2004, The GROMACS development team.
+ * Copyright (c) 2013,2014, by the GROMACS development team, led by
+ * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
+ * and including many others, as listed in the AUTHORS file in the
+ * top-level source directory and at http://www.gromacs.org.
+ *
+ * GROMACS is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public License
+ * as published by the Free Software Foundation; either version 2.1
+ * of the License, or (at your option) any later version.
+ *
+ * GROMACS is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with GROMACS; if not, see
+ * http://www.gnu.org/licenses, or write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * If you want to redistribute modifications to GROMACS, please
+ * consider that scientific software is very special. Version
+ * control is crucial - bugs must be traceable. We will be happy to
+ * consider code for inclusion in the official distribution, but
+ * derived work must not be called official GROMACS. Details are found
+ * in the README & COPYING files - if they are missing, get the
+ * official version at http://www.gromacs.org.
+ *
+ * To help us fund GROMACS development, we humbly ask that you cite
+ * the research papers on the package. Check out http://www.gromacs.org.
+ */
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <string.h>
+#include <time.h>
+#include <math.h>
+#include "sysstuff.h"
+#include "string2.h"
+#include "network.h"
+#include "smalloc.h"
+#include "nrnb.h"
+#include "main.h"
+#include "force.h"
+#include "macros.h"
+#include "random.h"
+#include "names.h"
+#include "gmx_fatal.h"
+#include "txtdump.h"
+#include "typedefs.h"
+#include "update.h"
+#include "constr.h"
+#include "vec.h"
+#include "tgroup.h"
+#include "mdebin.h"
+#include "vsite.h"
+#include "force.h"
+#include "mdrun.h"
+#include "md_support.h"
++#include "sim_util.h"
+#include "domdec.h"
+#include "partdec.h"
+#include "mdatoms.h"
+#include "ns.h"
+#include "mtop_util.h"
+#include "pme.h"
+#include "bondf.h"
+#include "gmx_omp_nthreads.h"
+#include "md_logging.h"
+
+#include "gromacs/fileio/confio.h"
+#include "gromacs/fileio/trajectory_writing.h"
+#include "gromacs/linearalgebra/mtxio.h"
+#include "gromacs/linearalgebra/sparsematrix.h"
+#include "gromacs/timing/wallcycle.h"
+#include "gromacs/timing/walltime_accounting.h"
+
+typedef struct {
+ t_state s;
+ rvec *f;
+ real epot;
+ real fnorm;
+ real fmax;
+ int a_fmax;
+} em_state_t;
+
+static em_state_t *init_em_state()
+{
+ em_state_t *ems;
+
+ snew(ems, 1);
+
+ /* does this need to be here? Should the array be declared differently (staticaly)in the state definition? */
+ snew(ems->s.lambda, efptNR);
+
+ return ems;
+}
+
+static void print_em_start(FILE *fplog,
+ t_commrec *cr,
+ gmx_walltime_accounting_t walltime_accounting,
+ gmx_wallcycle_t wcycle,
+ const char *name)
+{
-
- sprintf(buf, "Started %s", name);
- print_date_and_time(fplog, cr->nodeid, buf, NULL);
-
+ walltime_accounting_start(walltime_accounting);
+ wallcycle_start(wcycle, ewcRUN);
++ print_start(fplog, cr, walltime_accounting, name);
+}
+static void em_time_end(gmx_walltime_accounting_t walltime_accounting,
+ gmx_wallcycle_t wcycle)
+{
+ wallcycle_stop(wcycle, ewcRUN);
+
+ walltime_accounting_end(walltime_accounting);
+}
+
+static void sp_header(FILE *out, const char *minimizer, real ftol, int nsteps)
+{
+ fprintf(out, "\n");
+ fprintf(out, "%s:\n", minimizer);
+ fprintf(out, " Tolerance (Fmax) = %12.5e\n", ftol);
+ fprintf(out, " Number of steps = %12d\n", nsteps);
+}
+
+static void warn_step(FILE *fp, real ftol, gmx_bool bLastStep, gmx_bool bConstrain)
+{
+ char buffer[2048];
+ if (bLastStep)
+ {
+ sprintf(buffer,
+ "\nEnergy minimization reached the maximum number "
+ "of steps before the forces reached the requested "
+ "precision Fmax < %g.\n", ftol);
+ }
+ else
+ {
+ sprintf(buffer,
+ "\nEnergy minimization has stopped, but the forces have "
+ "not converged to the requested precision Fmax < %g (which "
+ "may not be possible for your system). It stopped "
+ "because the algorithm tried to make a new step whose size "
+ "was too small, or there was no change in the energy since "
+ "last step. Either way, we regard the minimization as "
+ "converged to within the available machine precision, "
+ "given your starting configuration and EM parameters.\n%s%s",
+ ftol,
+ sizeof(real) < sizeof(double) ?
+ "\nDouble precision normally gives you higher accuracy, but "
+ "this is often not needed for preparing to run molecular "
+ "dynamics.\n" :
+ "",
+ bConstrain ?
+ "You might need to increase your constraint accuracy, or turn\n"
+ "off constraints altogether (set constraints = none in mdp file)\n" :
+ "");
+ }
+ fputs(wrap_lines(buffer, 78, 0, FALSE), fp);
+}
+
+
+
+static void print_converged(FILE *fp, const char *alg, real ftol,
+ gmx_int64_t count, gmx_bool bDone, gmx_int64_t nsteps,
+ real epot, real fmax, int nfmax, real fnorm)
+{
+ char buf[STEPSTRSIZE];
+
+ if (bDone)
+ {
+ fprintf(fp, "\n%s converged to Fmax < %g in %s steps\n",
+ alg, ftol, gmx_step_str(count, buf));
+ }
+ else if (count < nsteps)
+ {
+ fprintf(fp, "\n%s converged to machine precision in %s steps,\n"
+ "but did not reach the requested Fmax < %g.\n",
+ alg, gmx_step_str(count, buf), ftol);
+ }
+ else
+ {
+ fprintf(fp, "\n%s did not converge to Fmax < %g in %s steps.\n",
+ alg, ftol, gmx_step_str(count, buf));
+ }
+
+#ifdef GMX_DOUBLE
+ fprintf(fp, "Potential Energy = %21.14e\n", epot);
+ fprintf(fp, "Maximum force = %21.14e on atom %d\n", fmax, nfmax+1);
+ fprintf(fp, "Norm of force = %21.14e\n", fnorm);
+#else
+ fprintf(fp, "Potential Energy = %14.7e\n", epot);
+ fprintf(fp, "Maximum force = %14.7e on atom %d\n", fmax, nfmax+1);
+ fprintf(fp, "Norm of force = %14.7e\n", fnorm);
+#endif
+}
+
+static void get_f_norm_max(t_commrec *cr,
+ t_grpopts *opts, t_mdatoms *mdatoms, rvec *f,
+ real *fnorm, real *fmax, int *a_fmax)
+{
+ double fnorm2, *sum;
+ real fmax2, fmax2_0, fam;
+ int la_max, a_max, start, end, i, m, gf;
+
+ /* This routine finds the largest force and returns it.
+ * On parallel machines the global max is taken.
+ */
+ fnorm2 = 0;
+ fmax2 = 0;
+ la_max = -1;
+ gf = 0;
+ start = mdatoms->start;
+ end = mdatoms->homenr + start;
+ if (mdatoms->cFREEZE)
+ {
+ for (i = start; i < end; i++)
+ {
+ gf = mdatoms->cFREEZE[i];
+ fam = 0;
+ for (m = 0; m < DIM; m++)
+ {
+ if (!opts->nFreeze[gf][m])
+ {
+ fam += sqr(f[i][m]);
+ }
+ }
+ fnorm2 += fam;
+ if (fam > fmax2)
+ {
+ fmax2 = fam;
+ la_max = i;
+ }
+ }
+ }
+ else
+ {
+ for (i = start; i < end; i++)
+ {
+ fam = norm2(f[i]);
+ fnorm2 += fam;
+ if (fam > fmax2)
+ {
+ fmax2 = fam;
+ la_max = i;
+ }
+ }
+ }
+
+ if (la_max >= 0 && DOMAINDECOMP(cr))
+ {
+ a_max = cr->dd->gatindex[la_max];
+ }
+ else
+ {
+ a_max = la_max;
+ }
+ if (PAR(cr))
+ {
+ snew(sum, 2*cr->nnodes+1);
+ sum[2*cr->nodeid] = fmax2;
+ sum[2*cr->nodeid+1] = a_max;
+ sum[2*cr->nnodes] = fnorm2;
+ gmx_sumd(2*cr->nnodes+1, sum, cr);
+ fnorm2 = sum[2*cr->nnodes];
+ /* Determine the global maximum */
+ for (i = 0; i < cr->nnodes; i++)
+ {
+ if (sum[2*i] > fmax2)
+ {
+ fmax2 = sum[2*i];
+ a_max = (int)(sum[2*i+1] + 0.5);
+ }
+ }
+ sfree(sum);
+ }
+
+ if (fnorm)
+ {
+ *fnorm = sqrt(fnorm2);
+ }
+ if (fmax)
+ {
+ *fmax = sqrt(fmax2);
+ }
+ if (a_fmax)
+ {
+ *a_fmax = a_max;
+ }
+}
+
+static void get_state_f_norm_max(t_commrec *cr,
+ t_grpopts *opts, t_mdatoms *mdatoms,
+ em_state_t *ems)
+{
+ get_f_norm_max(cr, opts, mdatoms, ems->f, &ems->fnorm, &ems->fmax, &ems->a_fmax);
+}
+
+void init_em(FILE *fplog, const char *title,
+ t_commrec *cr, t_inputrec *ir,
+ t_state *state_global, gmx_mtop_t *top_global,
+ em_state_t *ems, gmx_localtop_t **top,
+ rvec **f, rvec **f_global,
+ t_nrnb *nrnb, rvec mu_tot,
+ t_forcerec *fr, gmx_enerdata_t **enerd,
+ t_graph **graph, t_mdatoms *mdatoms, gmx_global_stat_t *gstat,
+ gmx_vsite_t *vsite, gmx_constr_t constr,
+ int nfile, const t_filenm fnm[],
+ gmx_mdoutf_t *outf, t_mdebin **mdebin)
+{
+ int start, homenr, i;
+ real dvdl_constr;
+
+ if (fplog)
+ {
+ fprintf(fplog, "Initiating %s\n", title);
+ }
+
+ state_global->ngtc = 0;
+
+ /* Initialize lambda variables */
+ initialize_lambdas(fplog, ir, &(state_global->fep_state), state_global->lambda, NULL);
+
+ init_nrnb(nrnb);
+
+ if (DOMAINDECOMP(cr))
+ {
+ *top = dd_init_local_top(top_global);
+
+ dd_init_local_state(cr->dd, state_global, &ems->s);
+
+ *f = NULL;
+
+ /* Distribute the charge groups over the nodes from the master node */
+ dd_partition_system(fplog, ir->init_step, cr, TRUE, 1,
+ state_global, top_global, ir,
+ &ems->s, &ems->f, mdatoms, *top,
+ fr, vsite, NULL, constr,
+ nrnb, NULL, FALSE);
+ dd_store_state(cr->dd, &ems->s);
+
+ if (ir->nstfout)
+ {
+ snew(*f_global, top_global->natoms);
+ }
+ else
+ {
+ *f_global = NULL;
+ }
+ *graph = NULL;
+ }
+ else
+ {
+ snew(*f, top_global->natoms);
+
+ /* Just copy the state */
+ ems->s = *state_global;
+ snew(ems->s.x, ems->s.nalloc);
+ snew(ems->f, ems->s.nalloc);
+ for (i = 0; i < state_global->natoms; i++)
+ {
+ copy_rvec(state_global->x[i], ems->s.x[i]);
+ }
+ copy_mat(state_global->box, ems->s.box);
+
+ if (PAR(cr) && ir->eI != eiNM)
+ {
+ /* Initialize the particle decomposition and split the topology */
+ *top = split_system(fplog, top_global, ir, cr);
+
+ pd_cg_range(cr, &fr->cg0, &fr->hcg);
+ }
+ else
+ {
+ *top = gmx_mtop_generate_local_top(top_global, ir);
+ }
+ *f_global = *f;
+
+ forcerec_set_excl_load(fr, *top, cr);
+
+ setup_bonded_threading(fr, &(*top)->idef);
+
+ if (ir->ePBC != epbcNONE && !fr->bMolPBC)
+ {
+ *graph = mk_graph(fplog, &((*top)->idef), 0, top_global->natoms, FALSE, FALSE);
+ }
+ else
+ {
+ *graph = NULL;
+ }
+
+ if (PARTDECOMP(cr))
+ {
+ pd_at_range(cr, &start, &homenr);
+ homenr -= start;
+ }
+ else
+ {
+ start = 0;
+ homenr = top_global->natoms;
+ }
+ atoms2md(top_global, ir, 0, NULL, start, homenr, mdatoms);
+ update_mdatoms(mdatoms, state_global->lambda[efptFEP]);
+
+ if (vsite)
+ {
+ set_vsite_top(vsite, *top, mdatoms, cr);
+ }
+ }
+
+ if (constr)
+ {
+ if (ir->eConstrAlg == econtSHAKE &&
+ gmx_mtop_ftype_count(top_global, F_CONSTR) > 0)
+ {
+ gmx_fatal(FARGS, "Can not do energy minimization with %s, use %s\n",
+ econstr_names[econtSHAKE], econstr_names[econtLINCS]);
+ }
+
+ if (!DOMAINDECOMP(cr))
+ {
+ set_constraints(constr, *top, ir, mdatoms, cr);
+ }
+
+ if (!ir->bContinuation)
+ {
+ /* Constrain the starting coordinates */
+ dvdl_constr = 0;
+ constrain(PAR(cr) ? NULL : fplog, TRUE, TRUE, constr, &(*top)->idef,
+ ir, NULL, cr, -1, 0, mdatoms,
+ ems->s.x, ems->s.x, NULL, fr->bMolPBC, ems->s.box,
+ ems->s.lambda[efptFEP], &dvdl_constr,
+ NULL, NULL, nrnb, econqCoord, FALSE, 0, 0);
+ }
+ }
+
+ if (PAR(cr))
+ {
+ *gstat = global_stat_init(ir);
+ }
+
+ *outf = init_mdoutf(nfile, fnm, 0, cr, ir, top_global, NULL);
+
+ snew(*enerd, 1);
+ init_enerdata(top_global->groups.grps[egcENER].nr, ir->fepvals->n_lambda,
+ *enerd);
+
+ if (mdebin != NULL)
+ {
+ /* Init bin for energy stuff */
+ *mdebin = init_mdebin(mdoutf_get_fp_ene(*outf), top_global, ir, NULL);
+ }
+
+ clear_rvec(mu_tot);
+ calc_shifts(ems->s.box, fr->shift_vec);
+}
+
+static void finish_em(t_commrec *cr, gmx_mdoutf_t outf,
+ gmx_walltime_accounting_t walltime_accounting,
+ gmx_wallcycle_t wcycle)
+{
+ if (!(cr->duty & DUTY_PME))
+ {
+ /* Tell the PME only node to finish */
+ gmx_pme_send_finish(cr);
+ }
+
+ done_mdoutf(outf);
+
+ em_time_end(walltime_accounting, wcycle);
+}
+
+static void swap_em_state(em_state_t *ems1, em_state_t *ems2)
+{
+ em_state_t tmp;
+
+ tmp = *ems1;
+ *ems1 = *ems2;
+ *ems2 = tmp;
+}
+
+static void copy_em_coords(em_state_t *ems, t_state *state)
+{
+ int i;
+
+ for (i = 0; (i < state->natoms); i++)
+ {
+ copy_rvec(ems->s.x[i], state->x[i]);
+ }
+}
+
+static void write_em_traj(FILE *fplog, t_commrec *cr,
+ gmx_mdoutf_t outf,
+ gmx_bool bX, gmx_bool bF, const char *confout,
+ gmx_mtop_t *top_global,
+ t_inputrec *ir, gmx_int64_t step,
+ em_state_t *state,
+ t_state *state_global, rvec *f_global)
+{
+ int mdof_flags;
+
+ if ((bX || bF || confout != NULL) && !DOMAINDECOMP(cr))
+ {
+ copy_em_coords(state, state_global);
+ f_global = state->f;
+ }
+
+ mdof_flags = 0;
+ if (bX)
+ {
+ mdof_flags |= MDOF_X;
+ }
+ if (bF)
+ {
+ mdof_flags |= MDOF_F;
+ }
+ mdoutf_write_to_trajectory_files(fplog, cr, outf, mdof_flags,
+ top_global, step, (double)step,
+ &state->s, state_global, state->f, f_global);
+
+ if (confout != NULL && MASTER(cr))
+ {
+ if (ir->ePBC != epbcNONE && !ir->bPeriodicMols && DOMAINDECOMP(cr))
+ {
+ /* Make molecules whole only for confout writing */
+ do_pbc_mtop(fplog, ir->ePBC, state_global->box, top_global,
+ state_global->x);
+ }
+
+ write_sto_conf_mtop(confout,
+ *top_global->name, top_global,
+ state_global->x, NULL, ir->ePBC, state_global->box);
+ }
+}
+
+static void do_em_step(t_commrec *cr, t_inputrec *ir, t_mdatoms *md,
+ gmx_bool bMolPBC,
+ em_state_t *ems1, real a, rvec *f, em_state_t *ems2,
+ gmx_constr_t constr, gmx_localtop_t *top,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_int64_t count)
+
+{
+ t_state *s1, *s2;
+ int i;
+ int start, end;
+ rvec *x1, *x2;
+ real dvdl_constr;
+
+ s1 = &ems1->s;
+ s2 = &ems2->s;
+
+ if (DOMAINDECOMP(cr) && s1->ddp_count != cr->dd->ddp_count)
+ {
+ gmx_incons("state mismatch in do_em_step");
+ }
+
+ s2->flags = s1->flags;
+
+ if (s2->nalloc != s1->nalloc)
+ {
+ s2->nalloc = s1->nalloc;
+ srenew(s2->x, s1->nalloc);
+ srenew(ems2->f, s1->nalloc);
+ if (s2->flags & (1<<estCGP))
+ {
+ srenew(s2->cg_p, s1->nalloc);
+ }
+ }
+
+ s2->natoms = s1->natoms;
+ copy_mat(s1->box, s2->box);
+ /* Copy free energy state */
+ for (i = 0; i < efptNR; i++)
+ {
+ s2->lambda[i] = s1->lambda[i];
+ }
+ copy_mat(s1->box, s2->box);
+
+ start = md->start;
+ end = md->start + md->homenr;
+
+ x1 = s1->x;
+ x2 = s2->x;
+
+#pragma omp parallel num_threads(gmx_omp_nthreads_get(emntUpdate))
+ {
+ int gf, i, m;
+
+ gf = 0;
+#pragma omp for schedule(static) nowait
+ for (i = start; i < end; i++)
+ {
+ if (md->cFREEZE)
+ {
+ gf = md->cFREEZE[i];
+ }
+ for (m = 0; m < DIM; m++)
+ {
+ if (ir->opts.nFreeze[gf][m])
+ {
+ x2[i][m] = x1[i][m];
+ }
+ else
+ {
+ x2[i][m] = x1[i][m] + a*f[i][m];
+ }
+ }
+ }
+
+ if (s2->flags & (1<<estCGP))
+ {
+ /* Copy the CG p vector */
+ x1 = s1->cg_p;
+ x2 = s2->cg_p;
+#pragma omp for schedule(static) nowait
+ for (i = start; i < end; i++)
+ {
+ copy_rvec(x1[i], x2[i]);
+ }
+ }
+
+ if (DOMAINDECOMP(cr))
+ {
+ s2->ddp_count = s1->ddp_count;
+ if (s2->cg_gl_nalloc < s1->cg_gl_nalloc)
+ {
+#pragma omp barrier
+ s2->cg_gl_nalloc = s1->cg_gl_nalloc;
+ srenew(s2->cg_gl, s2->cg_gl_nalloc);
+#pragma omp barrier
+ }
+ s2->ncg_gl = s1->ncg_gl;
+#pragma omp for schedule(static) nowait
+ for (i = 0; i < s2->ncg_gl; i++)
+ {
+ s2->cg_gl[i] = s1->cg_gl[i];
+ }
+ s2->ddp_count_cg_gl = s1->ddp_count_cg_gl;
+ }
+ }
+
+ if (constr)
+ {
+ wallcycle_start(wcycle, ewcCONSTR);
+ dvdl_constr = 0;
+ constrain(NULL, TRUE, TRUE, constr, &top->idef,
+ ir, NULL, cr, count, 0, md,
+ s1->x, s2->x, NULL, bMolPBC, s2->box,
+ s2->lambda[efptBONDED], &dvdl_constr,
+ NULL, NULL, nrnb, econqCoord, FALSE, 0, 0);
+ wallcycle_stop(wcycle, ewcCONSTR);
+ }
+}
+
+static void em_dd_partition_system(FILE *fplog, int step, t_commrec *cr,
+ gmx_mtop_t *top_global, t_inputrec *ir,
+ em_state_t *ems, gmx_localtop_t *top,
+ t_mdatoms *mdatoms, t_forcerec *fr,
+ gmx_vsite_t *vsite, gmx_constr_t constr,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle)
+{
+ /* Repartition the domain decomposition */
+ wallcycle_start(wcycle, ewcDOMDEC);
+ dd_partition_system(fplog, step, cr, FALSE, 1,
+ NULL, top_global, ir,
+ &ems->s, &ems->f,
+ mdatoms, top, fr, vsite, NULL, constr,
+ nrnb, wcycle, FALSE);
+ dd_store_state(cr->dd, &ems->s);
+ wallcycle_stop(wcycle, ewcDOMDEC);
+}
+
+static void evaluate_energy(FILE *fplog, t_commrec *cr,
+ gmx_mtop_t *top_global,
+ em_state_t *ems, gmx_localtop_t *top,
+ t_inputrec *inputrec,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_global_stat_t gstat,
+ gmx_vsite_t *vsite, gmx_constr_t constr,
+ t_fcdata *fcd,
+ t_graph *graph, t_mdatoms *mdatoms,
+ t_forcerec *fr, rvec mu_tot,
+ gmx_enerdata_t *enerd, tensor vir, tensor pres,
+ gmx_int64_t count, gmx_bool bFirst)
+{
+ real t;
+ gmx_bool bNS;
+ int nabnsb;
+ tensor force_vir, shake_vir, ekin;
+ real dvdl_constr, prescorr, enercorr, dvdlcorr;
+ real terminate = 0;
+
+ /* Set the time to the initial time, the time does not change during EM */
+ t = inputrec->init_t;
+
+ if (bFirst ||
+ (DOMAINDECOMP(cr) && ems->s.ddp_count < cr->dd->ddp_count))
+ {
+ /* This the first state or an old state used before the last ns */
+ bNS = TRUE;
+ }
+ else
+ {
+ bNS = FALSE;
+ if (inputrec->nstlist > 0)
+ {
+ bNS = TRUE;
+ }
+ else if (inputrec->nstlist == -1)
+ {
+ nabnsb = natoms_beyond_ns_buffer(inputrec, fr, &top->cgs, NULL, ems->s.x);
+ if (PAR(cr))
+ {
+ gmx_sumi(1, &nabnsb, cr);
+ }
+ bNS = (nabnsb > 0);
+ }
+ }
+
+ if (vsite)
+ {
+ construct_vsites(vsite, ems->s.x, 1, NULL,
+ top->idef.iparams, top->idef.il,
+ fr->ePBC, fr->bMolPBC, graph, cr, ems->s.box);
+ }
+
+ if (DOMAINDECOMP(cr))
+ {
+ if (bNS)
+ {
+ /* Repartition the domain decomposition */
+ em_dd_partition_system(fplog, count, cr, top_global, inputrec,
+ ems, top, mdatoms, fr, vsite, constr,
+ nrnb, wcycle);
+ }
+ }
+
+ /* Calc force & energy on new trial position */
+ /* do_force always puts the charge groups in the box and shifts again
+ * We do not unshift, so molecules are always whole in congrad.c
+ */
+ do_force(fplog, cr, inputrec,
+ count, nrnb, wcycle, top, &top_global->groups,
+ ems->s.box, ems->s.x, &ems->s.hist,
+ ems->f, force_vir, mdatoms, enerd, fcd,
+ ems->s.lambda, graph, fr, vsite, mu_tot, t, NULL, NULL, TRUE,
+ GMX_FORCE_STATECHANGED | GMX_FORCE_ALLFORCES |
+ GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY |
+ (bNS ? GMX_FORCE_NS | GMX_FORCE_DO_LR : 0));
+
+ /* Clear the unused shake virial and pressure */
+ clear_mat(shake_vir);
+ clear_mat(pres);
+
+ /* Communicate stuff when parallel */
+ if (PAR(cr) && inputrec->eI != eiNM)
+ {
+ wallcycle_start(wcycle, ewcMoveE);
+
+ global_stat(fplog, gstat, cr, enerd, force_vir, shake_vir, mu_tot,
+ inputrec, NULL, NULL, NULL, 1, &terminate,
+ top_global, &ems->s, FALSE,
+ CGLO_ENERGY |
+ CGLO_PRESSURE |
+ CGLO_CONSTRAINT |
+ CGLO_FIRSTITERATE);
+
+ wallcycle_stop(wcycle, ewcMoveE);
+ }
+
+ /* Calculate long range corrections to pressure and energy */
+ calc_dispcorr(fplog, inputrec, fr, count, top_global->natoms, ems->s.box, ems->s.lambda[efptVDW],
+ pres, force_vir, &prescorr, &enercorr, &dvdlcorr);
+ enerd->term[F_DISPCORR] = enercorr;
+ enerd->term[F_EPOT] += enercorr;
+ enerd->term[F_PRES] += prescorr;
+ enerd->term[F_DVDL] += dvdlcorr;
+
+ ems->epot = enerd->term[F_EPOT];
+
+ if (constr)
+ {
+ /* Project out the constraint components of the force */
+ wallcycle_start(wcycle, ewcCONSTR);
+ dvdl_constr = 0;
+ constrain(NULL, FALSE, FALSE, constr, &top->idef,
+ inputrec, NULL, cr, count, 0, mdatoms,
+ ems->s.x, ems->f, ems->f, fr->bMolPBC, ems->s.box,
+ ems->s.lambda[efptBONDED], &dvdl_constr,
+ NULL, &shake_vir, nrnb, econqForceDispl, FALSE, 0, 0);
+ if (fr->bSepDVDL && fplog)
+ {
+ gmx_print_sepdvdl(fplog, "Constraints", t, dvdl_constr);
+ }
+ enerd->term[F_DVDL_CONSTR] += dvdl_constr;
+ m_add(force_vir, shake_vir, vir);
+ wallcycle_stop(wcycle, ewcCONSTR);
+ }
+ else
+ {
+ copy_mat(force_vir, vir);
+ }
+
+ clear_mat(ekin);
+ enerd->term[F_PRES] =
+ calc_pres(fr->ePBC, inputrec->nwall, ems->s.box, ekin, vir, pres);
+
+ sum_dhdl(enerd, ems->s.lambda, inputrec->fepvals);
+
+ if (EI_ENERGY_MINIMIZATION(inputrec->eI))
+ {
+ get_state_f_norm_max(cr, &(inputrec->opts), mdatoms, ems);
+ }
+}
+
+static double reorder_partsum(t_commrec *cr, t_grpopts *opts, t_mdatoms *mdatoms,
+ gmx_mtop_t *mtop,
+ em_state_t *s_min, em_state_t *s_b)
+{
+ rvec *fm, *fb, *fmg;
+ t_block *cgs_gl;
+ int ncg, *cg_gl, *index, c, cg, i, a0, a1, a, gf, m;
+ double partsum;
+ unsigned char *grpnrFREEZE;
+
+ if (debug)
+ {
+ fprintf(debug, "Doing reorder_partsum\n");
+ }
+
+ fm = s_min->f;
+ fb = s_b->f;
+
+ cgs_gl = dd_charge_groups_global(cr->dd);
+ index = cgs_gl->index;
+
+ /* Collect fm in a global vector fmg.
+ * This conflicts with the spirit of domain decomposition,
+ * but to fully optimize this a much more complicated algorithm is required.
+ */
+ snew(fmg, mtop->natoms);
+
+ ncg = s_min->s.ncg_gl;
+ cg_gl = s_min->s.cg_gl;
+ i = 0;
+ for (c = 0; c < ncg; c++)
+ {
+ cg = cg_gl[c];
+ a0 = index[cg];
+ a1 = index[cg+1];
+ for (a = a0; a < a1; a++)
+ {
+ copy_rvec(fm[i], fmg[a]);
+ i++;
+ }
+ }
+ gmx_sum(mtop->natoms*3, fmg[0], cr);
+
+ /* Now we will determine the part of the sum for the cgs in state s_b */
+ ncg = s_b->s.ncg_gl;
+ cg_gl = s_b->s.cg_gl;
+ partsum = 0;
+ i = 0;
+ gf = 0;
+ grpnrFREEZE = mtop->groups.grpnr[egcFREEZE];
+ for (c = 0; c < ncg; c++)
+ {
+ cg = cg_gl[c];
+ a0 = index[cg];
+ a1 = index[cg+1];
+ for (a = a0; a < a1; a++)
+ {
+ if (mdatoms->cFREEZE && grpnrFREEZE)
+ {
+ gf = grpnrFREEZE[i];
+ }
+ for (m = 0; m < DIM; m++)
+ {
+ if (!opts->nFreeze[gf][m])
+ {
+ partsum += (fb[i][m] - fmg[a][m])*fb[i][m];
+ }
+ }
+ i++;
+ }
+ }
+
+ sfree(fmg);
+
+ return partsum;
+}
+
+static real pr_beta(t_commrec *cr, t_grpopts *opts, t_mdatoms *mdatoms,
+ gmx_mtop_t *mtop,
+ em_state_t *s_min, em_state_t *s_b)
+{
+ rvec *fm, *fb;
+ double sum;
+ int gf, i, m;
+
+ /* This is just the classical Polak-Ribiere calculation of beta;
+ * it looks a bit complicated since we take freeze groups into account,
+ * and might have to sum it in parallel runs.
+ */
+
+ if (!DOMAINDECOMP(cr) ||
+ (s_min->s.ddp_count == cr->dd->ddp_count &&
+ s_b->s.ddp_count == cr->dd->ddp_count))
+ {
+ fm = s_min->f;
+ fb = s_b->f;
+ sum = 0;
+ gf = 0;
+ /* This part of code can be incorrect with DD,
+ * since the atom ordering in s_b and s_min might differ.
+ */
+ for (i = mdatoms->start; i < mdatoms->start+mdatoms->homenr; i++)
+ {
+ if (mdatoms->cFREEZE)
+ {
+ gf = mdatoms->cFREEZE[i];
+ }
+ for (m = 0; m < DIM; m++)
+ {
+ if (!opts->nFreeze[gf][m])
+ {
+ sum += (fb[i][m] - fm[i][m])*fb[i][m];
+ }
+ }
+ }
+ }
+ else
+ {
+ /* We need to reorder cgs while summing */
+ sum = reorder_partsum(cr, opts, mdatoms, mtop, s_min, s_b);
+ }
+ if (PAR(cr))
+ {
+ gmx_sumd(1, &sum, cr);
+ }
+
+ return sum/sqr(s_min->fnorm);
+}
+
+double do_cg(FILE *fplog, t_commrec *cr,
+ int nfile, const t_filenm fnm[],
+ const output_env_t gmx_unused oenv, gmx_bool bVerbose, gmx_bool gmx_unused bCompact,
+ int gmx_unused nstglobalcomm,
+ gmx_vsite_t *vsite, gmx_constr_t constr,
+ int gmx_unused stepout,
+ t_inputrec *inputrec,
+ gmx_mtop_t *top_global, t_fcdata *fcd,
+ t_state *state_global,
+ t_mdatoms *mdatoms,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_edsam_t gmx_unused ed,
+ t_forcerec *fr,
+ int gmx_unused repl_ex_nst, int gmx_unused repl_ex_nex, int gmx_unused repl_ex_seed,
+ gmx_membed_t gmx_unused membed,
+ real gmx_unused cpt_period, real gmx_unused max_hours,
+ const char gmx_unused *deviceOptions,
+ unsigned long gmx_unused Flags,
+ gmx_walltime_accounting_t walltime_accounting)
+{
+ const char *CG = "Polak-Ribiere Conjugate Gradients";
+
+ em_state_t *s_min, *s_a, *s_b, *s_c;
+ gmx_localtop_t *top;
+ gmx_enerdata_t *enerd;
+ rvec *f;
+ gmx_global_stat_t gstat;
+ t_graph *graph;
+ rvec *f_global, *p, *sf, *sfm;
+ double gpa, gpb, gpc, tmp, sum[2], minstep;
+ real fnormn;
+ real stepsize;
+ real a, b, c, beta = 0.0;
+ real epot_repl = 0;
+ real pnorm;
+ t_mdebin *mdebin;
+ gmx_bool converged, foundlower;
+ rvec mu_tot;
+ gmx_bool do_log = FALSE, do_ene = FALSE, do_x, do_f;
+ tensor vir, pres;
+ int number_steps, neval = 0, nstcg = inputrec->nstcgsteep;
+ gmx_mdoutf_t outf;
+ int i, m, gf, step, nminstep;
+ real terminate = 0;
+
+ step = 0;
+
+ s_min = init_em_state();
+ s_a = init_em_state();
+ s_b = init_em_state();
+ s_c = init_em_state();
+
+ /* Init em and store the local state in s_min */
+ init_em(fplog, CG, cr, inputrec,
+ state_global, top_global, s_min, &top, &f, &f_global,
+ nrnb, mu_tot, fr, &enerd, &graph, mdatoms, &gstat, vsite, constr,
+ nfile, fnm, &outf, &mdebin);
+
+ /* Print to log file */
+ print_em_start(fplog, cr, walltime_accounting, wcycle, CG);
+
+ /* Max number of steps */
+ number_steps = inputrec->nsteps;
+
+ if (MASTER(cr))
+ {
+ sp_header(stderr, CG, inputrec->em_tol, number_steps);
+ }
+ if (fplog)
+ {
+ sp_header(fplog, CG, inputrec->em_tol, number_steps);
+ }
+
+ /* Call the force routine and some auxiliary (neighboursearching etc.) */
+ /* do_force always puts the charge groups in the box and shifts again
+ * We do not unshift, so molecules are always whole in congrad.c
+ */
+ evaluate_energy(fplog, cr,
+ top_global, s_min, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, -1, TRUE);
+ where();
+
+ if (MASTER(cr))
+ {
+ /* Copy stuff to the energy bin for easy printing etc. */
+ upd_mdebin(mdebin, FALSE, FALSE, (double)step,
+ mdatoms->tmass, enerd, &s_min->s, inputrec->fepvals, inputrec->expandedvals, s_min->s.box,
+ NULL, NULL, vir, pres, NULL, mu_tot, constr);
+
+ print_ebin_header(fplog, step, step, s_min->s.lambda[efptFEP]);
+ print_ebin(mdoutf_get_fp_ene(outf), TRUE, FALSE, FALSE, fplog, step, step, eprNORMAL,
+ TRUE, mdebin, fcd, &(top_global->groups), &(inputrec->opts));
+ }
+ where();
+
+ /* Estimate/guess the initial stepsize */
+ stepsize = inputrec->em_stepsize/s_min->fnorm;
+
+ if (MASTER(cr))
+ {
+ fprintf(stderr, " F-max = %12.5e on atom %d\n",
+ s_min->fmax, s_min->a_fmax+1);
+ fprintf(stderr, " F-Norm = %12.5e\n",
+ s_min->fnorm/sqrt(state_global->natoms));
+ fprintf(stderr, "\n");
+ /* and copy to the log file too... */
+ fprintf(fplog, " F-max = %12.5e on atom %d\n",
+ s_min->fmax, s_min->a_fmax+1);
+ fprintf(fplog, " F-Norm = %12.5e\n",
+ s_min->fnorm/sqrt(state_global->natoms));
+ fprintf(fplog, "\n");
+ }
+ /* Start the loop over CG steps.
+ * Each successful step is counted, and we continue until
+ * we either converge or reach the max number of steps.
+ */
+ converged = FALSE;
+ for (step = 0; (number_steps < 0 || (number_steps >= 0 && step <= number_steps)) && !converged; step++)
+ {
+
+ /* start taking steps in a new direction
+ * First time we enter the routine, beta=0, and the direction is
+ * simply the negative gradient.
+ */
+
+ /* Calculate the new direction in p, and the gradient in this direction, gpa */
+ p = s_min->s.cg_p;
+ sf = s_min->f;
+ gpa = 0;
+ gf = 0;
+ for (i = mdatoms->start; i < mdatoms->start+mdatoms->homenr; i++)
+ {
+ if (mdatoms->cFREEZE)
+ {
+ gf = mdatoms->cFREEZE[i];
+ }
+ for (m = 0; m < DIM; m++)
+ {
+ if (!inputrec->opts.nFreeze[gf][m])
+ {
+ p[i][m] = sf[i][m] + beta*p[i][m];
+ gpa -= p[i][m]*sf[i][m];
+ /* f is negative gradient, thus the sign */
+ }
+ else
+ {
+ p[i][m] = 0;
+ }
+ }
+ }
+
+ /* Sum the gradient along the line across CPUs */
+ if (PAR(cr))
+ {
+ gmx_sumd(1, &gpa, cr);
+ }
+
+ /* Calculate the norm of the search vector */
+ get_f_norm_max(cr, &(inputrec->opts), mdatoms, p, &pnorm, NULL, NULL);
+
+ /* Just in case stepsize reaches zero due to numerical precision... */
+ if (stepsize <= 0)
+ {
+ stepsize = inputrec->em_stepsize/pnorm;
+ }
+
+ /*
+ * Double check the value of the derivative in the search direction.
+ * If it is positive it must be due to the old information in the
+ * CG formula, so just remove that and start over with beta=0.
+ * This corresponds to a steepest descent step.
+ */
+ if (gpa > 0)
+ {
+ beta = 0;
+ step--; /* Don't count this step since we are restarting */
+ continue; /* Go back to the beginning of the big for-loop */
+ }
+
+ /* Calculate minimum allowed stepsize, before the average (norm)
+ * relative change in coordinate is smaller than precision
+ */
+ minstep = 0;
+ for (i = mdatoms->start; i < mdatoms->start+mdatoms->homenr; i++)
+ {
+ for (m = 0; m < DIM; m++)
+ {
+ tmp = fabs(s_min->s.x[i][m]);
+ if (tmp < 1.0)
+ {
+ tmp = 1.0;
+ }
+ tmp = p[i][m]/tmp;
+ minstep += tmp*tmp;
+ }
+ }
+ /* Add up from all CPUs */
+ if (PAR(cr))
+ {
+ gmx_sumd(1, &minstep, cr);
+ }
+
+ minstep = GMX_REAL_EPS/sqrt(minstep/(3*state_global->natoms));
+
+ if (stepsize < minstep)
+ {
+ converged = TRUE;
+ break;
+ }
+
+ /* Write coordinates if necessary */
+ do_x = do_per_step(step, inputrec->nstxout);
+ do_f = do_per_step(step, inputrec->nstfout);
+
+ write_em_traj(fplog, cr, outf, do_x, do_f, NULL,
+ top_global, inputrec, step,
+ s_min, state_global, f_global);
+
+ /* Take a step downhill.
+ * In theory, we should minimize the function along this direction.
+ * That is quite possible, but it turns out to take 5-10 function evaluations
+ * for each line. However, we dont really need to find the exact minimum -
+ * it is much better to start a new CG step in a modified direction as soon
+ * as we are close to it. This will save a lot of energy evaluations.
+ *
+ * In practice, we just try to take a single step.
+ * If it worked (i.e. lowered the energy), we increase the stepsize but
+ * the continue straight to the next CG step without trying to find any minimum.
+ * If it didn't work (higher energy), there must be a minimum somewhere between
+ * the old position and the new one.
+ *
+ * Due to the finite numerical accuracy, it turns out that it is a good idea
+ * to even accept a SMALL increase in energy, if the derivative is still downhill.
+ * This leads to lower final energies in the tests I've done. / Erik
+ */
+ s_a->epot = s_min->epot;
+ a = 0.0;
+ c = a + stepsize; /* reference position along line is zero */
+
+ if (DOMAINDECOMP(cr) && s_min->s.ddp_count < cr->dd->ddp_count)
+ {
+ em_dd_partition_system(fplog, step, cr, top_global, inputrec,
+ s_min, top, mdatoms, fr, vsite, constr,
+ nrnb, wcycle);
+ }
+
+ /* Take a trial step (new coords in s_c) */
+ do_em_step(cr, inputrec, mdatoms, fr->bMolPBC, s_min, c, s_min->s.cg_p, s_c,
+ constr, top, nrnb, wcycle, -1);
+
+ neval++;
+ /* Calculate energy for the trial step */
+ evaluate_energy(fplog, cr,
+ top_global, s_c, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, -1, FALSE);
+
+ /* Calc derivative along line */
+ p = s_c->s.cg_p;
+ sf = s_c->f;
+ gpc = 0;
+ for (i = mdatoms->start; i < mdatoms->start+mdatoms->homenr; i++)
+ {
+ for (m = 0; m < DIM; m++)
+ {
+ gpc -= p[i][m]*sf[i][m]; /* f is negative gradient, thus the sign */
+ }
+ }
+ /* Sum the gradient along the line across CPUs */
+ if (PAR(cr))
+ {
+ gmx_sumd(1, &gpc, cr);
+ }
+
+ /* This is the max amount of increase in energy we tolerate */
+ tmp = sqrt(GMX_REAL_EPS)*fabs(s_a->epot);
+
+ /* Accept the step if the energy is lower, or if it is not significantly higher
+ * and the line derivative is still negative.
+ */
+ if (s_c->epot < s_a->epot || (gpc < 0 && s_c->epot < (s_a->epot + tmp)))
+ {
+ foundlower = TRUE;
+ /* Great, we found a better energy. Increase step for next iteration
+ * if we are still going down, decrease it otherwise
+ */
+ if (gpc < 0)
+ {
+ stepsize *= 1.618034; /* The golden section */
+ }
+ else
+ {
+ stepsize *= 0.618034; /* 1/golden section */
+ }
+ }
+ else
+ {
+ /* New energy is the same or higher. We will have to do some work
+ * to find a smaller value in the interval. Take smaller step next time!
+ */
+ foundlower = FALSE;
+ stepsize *= 0.618034;
+ }
+
+
+
+
+ /* OK, if we didn't find a lower value we will have to locate one now - there must
+ * be one in the interval [a=0,c].
+ * The same thing is valid here, though: Don't spend dozens of iterations to find
+ * the line minimum. We try to interpolate based on the derivative at the endpoints,
+ * and only continue until we find a lower value. In most cases this means 1-2 iterations.
+ *
+ * I also have a safeguard for potentially really patological functions so we never
+ * take more than 20 steps before we give up ...
+ *
+ * If we already found a lower value we just skip this step and continue to the update.
+ */
+ if (!foundlower)
+ {
+ nminstep = 0;
+
+ do
+ {
+ /* Select a new trial point.
+ * If the derivatives at points a & c have different sign we interpolate to zero,
+ * otherwise just do a bisection.
+ */
+ if (gpa < 0 && gpc > 0)
+ {
+ b = a + gpa*(a-c)/(gpc-gpa);
+ }
+ else
+ {
+ b = 0.5*(a+c);
+ }
+
+ /* safeguard if interpolation close to machine accuracy causes errors:
+ * never go outside the interval
+ */
+ if (b <= a || b >= c)
+ {
+ b = 0.5*(a+c);
+ }
+
+ if (DOMAINDECOMP(cr) && s_min->s.ddp_count != cr->dd->ddp_count)
+ {
+ /* Reload the old state */
+ em_dd_partition_system(fplog, -1, cr, top_global, inputrec,
+ s_min, top, mdatoms, fr, vsite, constr,
+ nrnb, wcycle);
+ }
+
+ /* Take a trial step to this new point - new coords in s_b */
+ do_em_step(cr, inputrec, mdatoms, fr->bMolPBC, s_min, b, s_min->s.cg_p, s_b,
+ constr, top, nrnb, wcycle, -1);
+
+ neval++;
+ /* Calculate energy for the trial step */
+ evaluate_energy(fplog, cr,
+ top_global, s_b, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, -1, FALSE);
+
+ /* p does not change within a step, but since the domain decomposition
+ * might change, we have to use cg_p of s_b here.
+ */
+ p = s_b->s.cg_p;
+ sf = s_b->f;
+ gpb = 0;
+ for (i = mdatoms->start; i < mdatoms->start+mdatoms->homenr; i++)
+ {
+ for (m = 0; m < DIM; m++)
+ {
+ gpb -= p[i][m]*sf[i][m]; /* f is negative gradient, thus the sign */
+ }
+ }
+ /* Sum the gradient along the line across CPUs */
+ if (PAR(cr))
+ {
+ gmx_sumd(1, &gpb, cr);
+ }
+
+ if (debug)
+ {
+ fprintf(debug, "CGE: EpotA %f EpotB %f EpotC %f gpb %f\n",
+ s_a->epot, s_b->epot, s_c->epot, gpb);
+ }
+
+ epot_repl = s_b->epot;
+
+ /* Keep one of the intervals based on the value of the derivative at the new point */
+ if (gpb > 0)
+ {
+ /* Replace c endpoint with b */
+ swap_em_state(s_b, s_c);
+ c = b;
+ gpc = gpb;
+ }
+ else
+ {
+ /* Replace a endpoint with b */
+ swap_em_state(s_b, s_a);
+ a = b;
+ gpa = gpb;
+ }
+
+ /*
+ * Stop search as soon as we find a value smaller than the endpoints.
+ * Never run more than 20 steps, no matter what.
+ */
+ nminstep++;
+ }
+ while ((epot_repl > s_a->epot || epot_repl > s_c->epot) &&
+ (nminstep < 20));
+
+ if (fabs(epot_repl - s_min->epot) < fabs(s_min->epot)*GMX_REAL_EPS ||
+ nminstep >= 20)
+ {
+ /* OK. We couldn't find a significantly lower energy.
+ * If beta==0 this was steepest descent, and then we give up.
+ * If not, set beta=0 and restart with steepest descent before quitting.
+ */
+ if (beta == 0.0)
+ {
+ /* Converged */
+ converged = TRUE;
+ break;
+ }
+ else
+ {
+ /* Reset memory before giving up */
+ beta = 0.0;
+ continue;
+ }
+ }
+
+ /* Select min energy state of A & C, put the best in B.
+ */
+ if (s_c->epot < s_a->epot)
+ {
+ if (debug)
+ {
+ fprintf(debug, "CGE: C (%f) is lower than A (%f), moving C to B\n",
+ s_c->epot, s_a->epot);
+ }
+ swap_em_state(s_b, s_c);
+ gpb = gpc;
+ b = c;
+ }
+ else
+ {
+ if (debug)
+ {
+ fprintf(debug, "CGE: A (%f) is lower than C (%f), moving A to B\n",
+ s_a->epot, s_c->epot);
+ }
+ swap_em_state(s_b, s_a);
+ gpb = gpa;
+ b = a;
+ }
+
+ }
+ else
+ {
+ if (debug)
+ {
+ fprintf(debug, "CGE: Found a lower energy %f, moving C to B\n",
+ s_c->epot);
+ }
+ swap_em_state(s_b, s_c);
+ gpb = gpc;
+ b = c;
+ }
+
+ /* new search direction */
+ /* beta = 0 means forget all memory and restart with steepest descents. */
+ if (nstcg && ((step % nstcg) == 0))
+ {
+ beta = 0.0;
+ }
+ else
+ {
+ /* s_min->fnorm cannot be zero, because then we would have converged
+ * and broken out.
+ */
+
+ /* Polak-Ribiere update.
+ * Change to fnorm2/fnorm2_old for Fletcher-Reeves
+ */
+ beta = pr_beta(cr, &inputrec->opts, mdatoms, top_global, s_min, s_b);
+ }
+ /* Limit beta to prevent oscillations */
+ if (fabs(beta) > 5.0)
+ {
+ beta = 0.0;
+ }
+
+
+ /* update positions */
+ swap_em_state(s_min, s_b);
+ gpa = gpb;
+
+ /* Print it if necessary */
+ if (MASTER(cr))
+ {
+ if (bVerbose)
+ {
+ fprintf(stderr, "\rStep %d, Epot=%12.6e, Fnorm=%9.3e, Fmax=%9.3e (atom %d)\n",
+ step, s_min->epot, s_min->fnorm/sqrt(state_global->natoms),
+ s_min->fmax, s_min->a_fmax+1);
+ }
+ /* Store the new (lower) energies */
+ upd_mdebin(mdebin, FALSE, FALSE, (double)step,
+ mdatoms->tmass, enerd, &s_min->s, inputrec->fepvals, inputrec->expandedvals, s_min->s.box,
+ NULL, NULL, vir, pres, NULL, mu_tot, constr);
+
+ do_log = do_per_step(step, inputrec->nstlog);
+ do_ene = do_per_step(step, inputrec->nstenergy);
+ if (do_log)
+ {
+ print_ebin_header(fplog, step, step, s_min->s.lambda[efptFEP]);
+ }
+ print_ebin(mdoutf_get_fp_ene(outf), do_ene, FALSE, FALSE,
+ do_log ? fplog : NULL, step, step, eprNORMAL,
+ TRUE, mdebin, fcd, &(top_global->groups), &(inputrec->opts));
+ }
+
+ /* Stop when the maximum force lies below tolerance.
+ * If we have reached machine precision, converged is already set to true.
+ */
+ converged = converged || (s_min->fmax < inputrec->em_tol);
+
+ } /* End of the loop */
+
+ if (converged)
+ {
+ step--; /* we never took that last step in this case */
+
+ }
+ if (s_min->fmax > inputrec->em_tol)
+ {
+ if (MASTER(cr))
+ {
+ warn_step(stderr, inputrec->em_tol, step-1 == number_steps, FALSE);
+ warn_step(fplog, inputrec->em_tol, step-1 == number_steps, FALSE);
+ }
+ converged = FALSE;
+ }
+
+ if (MASTER(cr))
+ {
+ /* If we printed energy and/or logfile last step (which was the last step)
+ * we don't have to do it again, but otherwise print the final values.
+ */
+ if (!do_log)
+ {
+ /* Write final value to log since we didn't do anything the last step */
+ print_ebin_header(fplog, step, step, s_min->s.lambda[efptFEP]);
+ }
+ if (!do_ene || !do_log)
+ {
+ /* Write final energy file entries */
+ print_ebin(mdoutf_get_fp_ene(outf), !do_ene, FALSE, FALSE,
+ !do_log ? fplog : NULL, step, step, eprNORMAL,
+ TRUE, mdebin, fcd, &(top_global->groups), &(inputrec->opts));
+ }
+ }
+
+ /* Print some stuff... */
+ if (MASTER(cr))
+ {
+ fprintf(stderr, "\nwriting lowest energy coordinates.\n");
+ }
+
+ /* IMPORTANT!
+ * For accurate normal mode calculation it is imperative that we
+ * store the last conformation into the full precision binary trajectory.
+ *
+ * However, we should only do it if we did NOT already write this step
+ * above (which we did if do_x or do_f was true).
+ */
+ do_x = !do_per_step(step, inputrec->nstxout);
+ do_f = (inputrec->nstfout > 0 && !do_per_step(step, inputrec->nstfout));
+
+ write_em_traj(fplog, cr, outf, do_x, do_f, ftp2fn(efSTO, nfile, fnm),
+ top_global, inputrec, step,
+ s_min, state_global, f_global);
+
+ fnormn = s_min->fnorm/sqrt(state_global->natoms);
+
+ if (MASTER(cr))
+ {
+ print_converged(stderr, CG, inputrec->em_tol, step, converged, number_steps,
+ s_min->epot, s_min->fmax, s_min->a_fmax, fnormn);
+ print_converged(fplog, CG, inputrec->em_tol, step, converged, number_steps,
+ s_min->epot, s_min->fmax, s_min->a_fmax, fnormn);
+
+ fprintf(fplog, "\nPerformed %d energy evaluations in total.\n", neval);
+ }
+
+ finish_em(cr, outf, walltime_accounting, wcycle);
+
+ /* To print the actual number of steps we needed somewhere */
+ walltime_accounting_set_nsteps_done(walltime_accounting, step);
+
+ return 0;
+} /* That's all folks */
+
+
+double do_lbfgs(FILE *fplog, t_commrec *cr,
+ int nfile, const t_filenm fnm[],
+ const output_env_t gmx_unused oenv, gmx_bool bVerbose, gmx_bool gmx_unused bCompact,
+ int gmx_unused nstglobalcomm,
+ gmx_vsite_t *vsite, gmx_constr_t constr,
+ int gmx_unused stepout,
+ t_inputrec *inputrec,
+ gmx_mtop_t *top_global, t_fcdata *fcd,
+ t_state *state,
+ t_mdatoms *mdatoms,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_edsam_t gmx_unused ed,
+ t_forcerec *fr,
+ int gmx_unused repl_ex_nst, int gmx_unused repl_ex_nex, int gmx_unused repl_ex_seed,
+ gmx_membed_t gmx_unused membed,
+ real gmx_unused cpt_period, real gmx_unused max_hours,
+ const char gmx_unused *deviceOptions,
+ unsigned long gmx_unused Flags,
+ gmx_walltime_accounting_t walltime_accounting)
+{
+ static const char *LBFGS = "Low-Memory BFGS Minimizer";
+ em_state_t ems;
+ gmx_localtop_t *top;
+ gmx_enerdata_t *enerd;
+ rvec *f;
+ gmx_global_stat_t gstat;
+ t_graph *graph;
+ rvec *f_global;
+ int ncorr, nmaxcorr, point, cp, neval, nminstep;
+ double stepsize, gpa, gpb, gpc, tmp, minstep;
+ real *rho, *alpha, *ff, *xx, *p, *s, *lastx, *lastf, **dx, **dg;
+ real *xa, *xb, *xc, *fa, *fb, *fc, *xtmp, *ftmp;
+ real a, b, c, maxdelta, delta;
+ real diag, Epot0, Epot, EpotA, EpotB, EpotC;
+ real dgdx, dgdg, sq, yr, beta;
+ t_mdebin *mdebin;
+ gmx_bool converged, first;
+ rvec mu_tot;
+ real fnorm, fmax;
+ gmx_bool do_log, do_ene, do_x, do_f, foundlower, *frozen;
+ tensor vir, pres;
+ int start, end, number_steps;
+ gmx_mdoutf_t outf;
+ int i, k, m, n, nfmax, gf, step;
+ int mdof_flags;
+ /* not used */
+ real terminate;
+
+ if (PAR(cr))
+ {
+ gmx_fatal(FARGS, "Cannot do parallel L-BFGS Minimization - yet.\n");
+ }
+
+ if (NULL != constr)
+ {
+ gmx_fatal(FARGS, "The combination of constraints and L-BFGS minimization is not implemented. Either do not use constraints, or use another minimizer (e.g. steepest descent).");
+ }
+
+ n = 3*state->natoms;
+ nmaxcorr = inputrec->nbfgscorr;
+
+ /* Allocate memory */
+ /* Use pointers to real so we dont have to loop over both atoms and
+ * dimensions all the time...
+ * x/f are allocated as rvec *, so make new x0/f0 pointers-to-real
+ * that point to the same memory.
+ */
+ snew(xa, n);
+ snew(xb, n);
+ snew(xc, n);
+ snew(fa, n);
+ snew(fb, n);
+ snew(fc, n);
+ snew(frozen, n);
+
+ snew(p, n);
+ snew(lastx, n);
+ snew(lastf, n);
+ snew(rho, nmaxcorr);
+ snew(alpha, nmaxcorr);
+
+ snew(dx, nmaxcorr);
+ for (i = 0; i < nmaxcorr; i++)
+ {
+ snew(dx[i], n);
+ }
+
+ snew(dg, nmaxcorr);
+ for (i = 0; i < nmaxcorr; i++)
+ {
+ snew(dg[i], n);
+ }
+
+ step = 0;
+ neval = 0;
+
+ /* Init em */
+ init_em(fplog, LBFGS, cr, inputrec,
+ state, top_global, &ems, &top, &f, &f_global,
+ nrnb, mu_tot, fr, &enerd, &graph, mdatoms, &gstat, vsite, constr,
+ nfile, fnm, &outf, &mdebin);
+ /* Do_lbfgs is not completely updated like do_steep and do_cg,
+ * so we free some memory again.
+ */
+ sfree(ems.s.x);
+ sfree(ems.f);
+
+ xx = (real *)state->x;
+ ff = (real *)f;
+
+ start = mdatoms->start;
+ end = mdatoms->homenr + start;
+
+ /* Print to log file */
+ print_em_start(fplog, cr, walltime_accounting, wcycle, LBFGS);
+
+ do_log = do_ene = do_x = do_f = TRUE;
+
+ /* Max number of steps */
+ number_steps = inputrec->nsteps;
+
+ /* Create a 3*natoms index to tell whether each degree of freedom is frozen */
+ gf = 0;
+ for (i = start; i < end; i++)
+ {
+ if (mdatoms->cFREEZE)
+ {
+ gf = mdatoms->cFREEZE[i];
+ }
+ for (m = 0; m < DIM; m++)
+ {
+ frozen[3*i+m] = inputrec->opts.nFreeze[gf][m];
+ }
+ }
+ if (MASTER(cr))
+ {
+ sp_header(stderr, LBFGS, inputrec->em_tol, number_steps);
+ }
+ if (fplog)
+ {
+ sp_header(fplog, LBFGS, inputrec->em_tol, number_steps);
+ }
+
+ if (vsite)
+ {
+ construct_vsites(vsite, state->x, 1, NULL,
+ top->idef.iparams, top->idef.il,
+ fr->ePBC, fr->bMolPBC, graph, cr, state->box);
+ }
+
+ /* Call the force routine and some auxiliary (neighboursearching etc.) */
+ /* do_force always puts the charge groups in the box and shifts again
+ * We do not unshift, so molecules are always whole
+ */
+ neval++;
+ ems.s.x = state->x;
+ ems.f = f;
+ evaluate_energy(fplog, cr,
+ top_global, &ems, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, -1, TRUE);
+ where();
+
+ if (MASTER(cr))
+ {
+ /* Copy stuff to the energy bin for easy printing etc. */
+ upd_mdebin(mdebin, FALSE, FALSE, (double)step,
+ mdatoms->tmass, enerd, state, inputrec->fepvals, inputrec->expandedvals, state->box,
+ NULL, NULL, vir, pres, NULL, mu_tot, constr);
+
+ print_ebin_header(fplog, step, step, state->lambda[efptFEP]);
+ print_ebin(mdoutf_get_fp_ene(outf), TRUE, FALSE, FALSE, fplog, step, step, eprNORMAL,
+ TRUE, mdebin, fcd, &(top_global->groups), &(inputrec->opts));
+ }
+ where();
+
+ /* This is the starting energy */
+ Epot = enerd->term[F_EPOT];
+
+ fnorm = ems.fnorm;
+ fmax = ems.fmax;
+ nfmax = ems.a_fmax;
+
+ /* Set the initial step.
+ * since it will be multiplied by the non-normalized search direction
+ * vector (force vector the first time), we scale it by the
+ * norm of the force.
+ */
+
+ if (MASTER(cr))
+ {
+ fprintf(stderr, "Using %d BFGS correction steps.\n\n", nmaxcorr);
+ fprintf(stderr, " F-max = %12.5e on atom %d\n", fmax, nfmax+1);
+ fprintf(stderr, " F-Norm = %12.5e\n", fnorm/sqrt(state->natoms));
+ fprintf(stderr, "\n");
+ /* and copy to the log file too... */
+ fprintf(fplog, "Using %d BFGS correction steps.\n\n", nmaxcorr);
+ fprintf(fplog, " F-max = %12.5e on atom %d\n", fmax, nfmax+1);
+ fprintf(fplog, " F-Norm = %12.5e\n", fnorm/sqrt(state->natoms));
+ fprintf(fplog, "\n");
+ }
+
+ point = 0;
+ for (i = 0; i < n; i++)
+ {
+ if (!frozen[i])
+ {
+ dx[point][i] = ff[i]; /* Initial search direction */
+ }
+ else
+ {
+ dx[point][i] = 0;
+ }
+ }
+
+ stepsize = 1.0/fnorm;
+ converged = FALSE;
+
+ /* Start the loop over BFGS steps.
+ * Each successful step is counted, and we continue until
+ * we either converge or reach the max number of steps.
+ */
+
+ ncorr = 0;
+
+ /* Set the gradient from the force */
+ converged = FALSE;
+ for (step = 0; (number_steps < 0 || (number_steps >= 0 && step <= number_steps)) && !converged; step++)
+ {
+
+ /* Write coordinates if necessary */
+ do_x = do_per_step(step, inputrec->nstxout);
+ do_f = do_per_step(step, inputrec->nstfout);
+
+ mdof_flags = 0;
+ if (do_x)
+ {
+ mdof_flags |= MDOF_X;
+ }
+
+ if (do_f)
+ {
+ mdof_flags |= MDOF_F;
+ }
+
+ mdoutf_write_to_trajectory_files(fplog, cr, outf, mdof_flags,
+ top_global, step, (real)step, state, state, f, f);
+
+ /* Do the linesearching in the direction dx[point][0..(n-1)] */
+
+ /* pointer to current direction - point=0 first time here */
+ s = dx[point];
+
+ /* calculate line gradient */
+ for (gpa = 0, i = 0; i < n; i++)
+ {
+ gpa -= s[i]*ff[i];
+ }
+
+ /* Calculate minimum allowed stepsize, before the average (norm)
+ * relative change in coordinate is smaller than precision
+ */
+ for (minstep = 0, i = 0; i < n; i++)
+ {
+ tmp = fabs(xx[i]);
+ if (tmp < 1.0)
+ {
+ tmp = 1.0;
+ }
+ tmp = s[i]/tmp;
+ minstep += tmp*tmp;
+ }
+ minstep = GMX_REAL_EPS/sqrt(minstep/n);
+
+ if (stepsize < minstep)
+ {
+ converged = TRUE;
+ break;
+ }
+
+ /* Store old forces and coordinates */
+ for (i = 0; i < n; i++)
+ {
+ lastx[i] = xx[i];
+ lastf[i] = ff[i];
+ }
+ Epot0 = Epot;
+
+ first = TRUE;
+
+ for (i = 0; i < n; i++)
+ {
+ xa[i] = xx[i];
+ }
+
+ /* Take a step downhill.
+ * In theory, we should minimize the function along this direction.
+ * That is quite possible, but it turns out to take 5-10 function evaluations
+ * for each line. However, we dont really need to find the exact minimum -
+ * it is much better to start a new BFGS step in a modified direction as soon
+ * as we are close to it. This will save a lot of energy evaluations.
+ *
+ * In practice, we just try to take a single step.
+ * If it worked (i.e. lowered the energy), we increase the stepsize but
+ * the continue straight to the next BFGS step without trying to find any minimum.
+ * If it didn't work (higher energy), there must be a minimum somewhere between
+ * the old position and the new one.
+ *
+ * Due to the finite numerical accuracy, it turns out that it is a good idea
+ * to even accept a SMALL increase in energy, if the derivative is still downhill.
+ * This leads to lower final energies in the tests I've done. / Erik
+ */
+ foundlower = FALSE;
+ EpotA = Epot0;
+ a = 0.0;
+ c = a + stepsize; /* reference position along line is zero */
+
+ /* Check stepsize first. We do not allow displacements
+ * larger than emstep.
+ */
+ do
+ {
+ c = a + stepsize;
+ maxdelta = 0;
+ for (i = 0; i < n; i++)
+ {
+ delta = c*s[i];
+ if (delta > maxdelta)
+ {
+ maxdelta = delta;
+ }
+ }
+ if (maxdelta > inputrec->em_stepsize)
+ {
+ stepsize *= 0.1;
+ }
+ }
+ while (maxdelta > inputrec->em_stepsize);
+
+ /* Take a trial step */
+ for (i = 0; i < n; i++)
+ {
+ xc[i] = lastx[i] + c*s[i];
+ }
+
+ neval++;
+ /* Calculate energy for the trial step */
+ ems.s.x = (rvec *)xc;
+ ems.f = (rvec *)fc;
+ evaluate_energy(fplog, cr,
+ top_global, &ems, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, step, FALSE);
+ EpotC = ems.epot;
+
+ /* Calc derivative along line */
+ for (gpc = 0, i = 0; i < n; i++)
+ {
+ gpc -= s[i]*fc[i]; /* f is negative gradient, thus the sign */
+ }
+ /* Sum the gradient along the line across CPUs */
+ if (PAR(cr))
+ {
+ gmx_sumd(1, &gpc, cr);
+ }
+
+ /* This is the max amount of increase in energy we tolerate */
+ tmp = sqrt(GMX_REAL_EPS)*fabs(EpotA);
+
+ /* Accept the step if the energy is lower, or if it is not significantly higher
+ * and the line derivative is still negative.
+ */
+ if (EpotC < EpotA || (gpc < 0 && EpotC < (EpotA+tmp)))
+ {
+ foundlower = TRUE;
+ /* Great, we found a better energy. Increase step for next iteration
+ * if we are still going down, decrease it otherwise
+ */
+ if (gpc < 0)
+ {
+ stepsize *= 1.618034; /* The golden section */
+ }
+ else
+ {
+ stepsize *= 0.618034; /* 1/golden section */
+ }
+ }
+ else
+ {
+ /* New energy is the same or higher. We will have to do some work
+ * to find a smaller value in the interval. Take smaller step next time!
+ */
+ foundlower = FALSE;
+ stepsize *= 0.618034;
+ }
+
+ /* OK, if we didn't find a lower value we will have to locate one now - there must
+ * be one in the interval [a=0,c].
+ * The same thing is valid here, though: Don't spend dozens of iterations to find
+ * the line minimum. We try to interpolate based on the derivative at the endpoints,
+ * and only continue until we find a lower value. In most cases this means 1-2 iterations.
+ *
+ * I also have a safeguard for potentially really patological functions so we never
+ * take more than 20 steps before we give up ...
+ *
+ * If we already found a lower value we just skip this step and continue to the update.
+ */
+
+ if (!foundlower)
+ {
+
+ nminstep = 0;
+ do
+ {
+ /* Select a new trial point.
+ * If the derivatives at points a & c have different sign we interpolate to zero,
+ * otherwise just do a bisection.
+ */
+
+ if (gpa < 0 && gpc > 0)
+ {
+ b = a + gpa*(a-c)/(gpc-gpa);
+ }
+ else
+ {
+ b = 0.5*(a+c);
+ }
+
+ /* safeguard if interpolation close to machine accuracy causes errors:
+ * never go outside the interval
+ */
+ if (b <= a || b >= c)
+ {
+ b = 0.5*(a+c);
+ }
+
+ /* Take a trial step */
+ for (i = 0; i < n; i++)
+ {
+ xb[i] = lastx[i] + b*s[i];
+ }
+
+ neval++;
+ /* Calculate energy for the trial step */
+ ems.s.x = (rvec *)xb;
+ ems.f = (rvec *)fb;
+ evaluate_energy(fplog, cr,
+ top_global, &ems, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, step, FALSE);
+ EpotB = ems.epot;
+
+ fnorm = ems.fnorm;
+
+ for (gpb = 0, i = 0; i < n; i++)
+ {
+ gpb -= s[i]*fb[i]; /* f is negative gradient, thus the sign */
+
+ }
+ /* Sum the gradient along the line across CPUs */
+ if (PAR(cr))
+ {
+ gmx_sumd(1, &gpb, cr);
+ }
+
+ /* Keep one of the intervals based on the value of the derivative at the new point */
+ if (gpb > 0)
+ {
+ /* Replace c endpoint with b */
+ EpotC = EpotB;
+ c = b;
+ gpc = gpb;
+ /* swap coord pointers b/c */
+ xtmp = xb;
+ ftmp = fb;
+ xb = xc;
+ fb = fc;
+ xc = xtmp;
+ fc = ftmp;
+ }
+ else
+ {
+ /* Replace a endpoint with b */
+ EpotA = EpotB;
+ a = b;
+ gpa = gpb;
+ /* swap coord pointers a/b */
+ xtmp = xb;
+ ftmp = fb;
+ xb = xa;
+ fb = fa;
+ xa = xtmp;
+ fa = ftmp;
+ }
+
+ /*
+ * Stop search as soon as we find a value smaller than the endpoints,
+ * or if the tolerance is below machine precision.
+ * Never run more than 20 steps, no matter what.
+ */
+ nminstep++;
+ }
+ while ((EpotB > EpotA || EpotB > EpotC) && (nminstep < 20));
+
+ if (fabs(EpotB-Epot0) < GMX_REAL_EPS || nminstep >= 20)
+ {
+ /* OK. We couldn't find a significantly lower energy.
+ * If ncorr==0 this was steepest descent, and then we give up.
+ * If not, reset memory to restart as steepest descent before quitting.
+ */
+ if (ncorr == 0)
+ {
+ /* Converged */
+ converged = TRUE;
+ break;
+ }
+ else
+ {
+ /* Reset memory */
+ ncorr = 0;
+ /* Search in gradient direction */
+ for (i = 0; i < n; i++)
+ {
+ dx[point][i] = ff[i];
+ }
+ /* Reset stepsize */
+ stepsize = 1.0/fnorm;
+ continue;
+ }
+ }
+
+ /* Select min energy state of A & C, put the best in xx/ff/Epot
+ */
+ if (EpotC < EpotA)
+ {
+ Epot = EpotC;
+ /* Use state C */
+ for (i = 0; i < n; i++)
+ {
+ xx[i] = xc[i];
+ ff[i] = fc[i];
+ }
+ stepsize = c;
+ }
+ else
+ {
+ Epot = EpotA;
+ /* Use state A */
+ for (i = 0; i < n; i++)
+ {
+ xx[i] = xa[i];
+ ff[i] = fa[i];
+ }
+ stepsize = a;
+ }
+
+ }
+ else
+ {
+ /* found lower */
+ Epot = EpotC;
+ /* Use state C */
+ for (i = 0; i < n; i++)
+ {
+ xx[i] = xc[i];
+ ff[i] = fc[i];
+ }
+ stepsize = c;
+ }
+
+ /* Update the memory information, and calculate a new
+ * approximation of the inverse hessian
+ */
+
+ /* Have new data in Epot, xx, ff */
+ if (ncorr < nmaxcorr)
+ {
+ ncorr++;
+ }
+
+ for (i = 0; i < n; i++)
+ {
+ dg[point][i] = lastf[i]-ff[i];
+ dx[point][i] *= stepsize;
+ }
+
+ dgdg = 0;
+ dgdx = 0;
+ for (i = 0; i < n; i++)
+ {
+ dgdg += dg[point][i]*dg[point][i];
+ dgdx += dg[point][i]*dx[point][i];
+ }
+
+ diag = dgdx/dgdg;
+
+ rho[point] = 1.0/dgdx;
+ point++;
+
+ if (point >= nmaxcorr)
+ {
+ point = 0;
+ }
+
+ /* Update */
+ for (i = 0; i < n; i++)
+ {
+ p[i] = ff[i];
+ }
+
+ cp = point;
+
+ /* Recursive update. First go back over the memory points */
+ for (k = 0; k < ncorr; k++)
+ {
+ cp--;
+ if (cp < 0)
+ {
+ cp = ncorr-1;
+ }
+
+ sq = 0;
+ for (i = 0; i < n; i++)
+ {
+ sq += dx[cp][i]*p[i];
+ }
+
+ alpha[cp] = rho[cp]*sq;
+
+ for (i = 0; i < n; i++)
+ {
+ p[i] -= alpha[cp]*dg[cp][i];
+ }
+ }
+
+ for (i = 0; i < n; i++)
+ {
+ p[i] *= diag;
+ }
+
+ /* And then go forward again */
+ for (k = 0; k < ncorr; k++)
+ {
+ yr = 0;
+ for (i = 0; i < n; i++)
+ {
+ yr += p[i]*dg[cp][i];
+ }
+
+ beta = rho[cp]*yr;
+ beta = alpha[cp]-beta;
+
+ for (i = 0; i < n; i++)
+ {
+ p[i] += beta*dx[cp][i];
+ }
+
+ cp++;
+ if (cp >= ncorr)
+ {
+ cp = 0;
+ }
+ }
+
+ for (i = 0; i < n; i++)
+ {
+ if (!frozen[i])
+ {
+ dx[point][i] = p[i];
+ }
+ else
+ {
+ dx[point][i] = 0;
+ }
+ }
+
+ stepsize = 1.0;
+
+ /* Test whether the convergence criterion is met */
+ get_f_norm_max(cr, &(inputrec->opts), mdatoms, f, &fnorm, &fmax, &nfmax);
+
+ /* Print it if necessary */
+ if (MASTER(cr))
+ {
+ if (bVerbose)
+ {
+ fprintf(stderr, "\rStep %d, Epot=%12.6e, Fnorm=%9.3e, Fmax=%9.3e (atom %d)\n",
+ step, Epot, fnorm/sqrt(state->natoms), fmax, nfmax+1);
+ }
+ /* Store the new (lower) energies */
+ upd_mdebin(mdebin, FALSE, FALSE, (double)step,
+ mdatoms->tmass, enerd, state, inputrec->fepvals, inputrec->expandedvals, state->box,
+ NULL, NULL, vir, pres, NULL, mu_tot, constr);
+ do_log = do_per_step(step, inputrec->nstlog);
+ do_ene = do_per_step(step, inputrec->nstenergy);
+ if (do_log)
+ {
+ print_ebin_header(fplog, step, step, state->lambda[efptFEP]);
+ }
+ print_ebin(mdoutf_get_fp_ene(outf), do_ene, FALSE, FALSE,
+ do_log ? fplog : NULL, step, step, eprNORMAL,
+ TRUE, mdebin, fcd, &(top_global->groups), &(inputrec->opts));
+ }
+
+ /* Stop when the maximum force lies below tolerance.
+ * If we have reached machine precision, converged is already set to true.
+ */
+
+ converged = converged || (fmax < inputrec->em_tol);
+
+ } /* End of the loop */
+
+ if (converged)
+ {
+ step--; /* we never took that last step in this case */
+
+ }
+ if (fmax > inputrec->em_tol)
+ {
+ if (MASTER(cr))
+ {
+ warn_step(stderr, inputrec->em_tol, step-1 == number_steps, FALSE);
+ warn_step(fplog, inputrec->em_tol, step-1 == number_steps, FALSE);
+ }
+ converged = FALSE;
+ }
+
+ /* If we printed energy and/or logfile last step (which was the last step)
+ * we don't have to do it again, but otherwise print the final values.
+ */
+ if (!do_log) /* Write final value to log since we didn't do anythin last step */
+ {
+ print_ebin_header(fplog, step, step, state->lambda[efptFEP]);
+ }
+ if (!do_ene || !do_log) /* Write final energy file entries */
+ {
+ print_ebin(mdoutf_get_fp_ene(outf), !do_ene, FALSE, FALSE,
+ !do_log ? fplog : NULL, step, step, eprNORMAL,
+ TRUE, mdebin, fcd, &(top_global->groups), &(inputrec->opts));
+ }
+
+ /* Print some stuff... */
+ if (MASTER(cr))
+ {
+ fprintf(stderr, "\nwriting lowest energy coordinates.\n");
+ }
+
+ /* IMPORTANT!
+ * For accurate normal mode calculation it is imperative that we
+ * store the last conformation into the full precision binary trajectory.
+ *
+ * However, we should only do it if we did NOT already write this step
+ * above (which we did if do_x or do_f was true).
+ */
+ do_x = !do_per_step(step, inputrec->nstxout);
+ do_f = !do_per_step(step, inputrec->nstfout);
+ write_em_traj(fplog, cr, outf, do_x, do_f, ftp2fn(efSTO, nfile, fnm),
+ top_global, inputrec, step,
+ &ems, state, f);
+
+ if (MASTER(cr))
+ {
+ print_converged(stderr, LBFGS, inputrec->em_tol, step, converged,
+ number_steps, Epot, fmax, nfmax, fnorm/sqrt(state->natoms));
+ print_converged(fplog, LBFGS, inputrec->em_tol, step, converged,
+ number_steps, Epot, fmax, nfmax, fnorm/sqrt(state->natoms));
+
+ fprintf(fplog, "\nPerformed %d energy evaluations in total.\n", neval);
+ }
+
+ finish_em(cr, outf, walltime_accounting, wcycle);
+
+ /* To print the actual number of steps we needed somewhere */
+ walltime_accounting_set_nsteps_done(walltime_accounting, step);
+
+ return 0;
+} /* That's all folks */
+
+
+double do_steep(FILE *fplog, t_commrec *cr,
+ int nfile, const t_filenm fnm[],
+ const output_env_t gmx_unused oenv, gmx_bool bVerbose, gmx_bool gmx_unused bCompact,
+ int gmx_unused nstglobalcomm,
+ gmx_vsite_t *vsite, gmx_constr_t constr,
+ int gmx_unused stepout,
+ t_inputrec *inputrec,
+ gmx_mtop_t *top_global, t_fcdata *fcd,
+ t_state *state_global,
+ t_mdatoms *mdatoms,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_edsam_t gmx_unused ed,
+ t_forcerec *fr,
+ int gmx_unused repl_ex_nst, int gmx_unused repl_ex_nex, int gmx_unused repl_ex_seed,
+ gmx_membed_t gmx_unused membed,
+ real gmx_unused cpt_period, real gmx_unused max_hours,
+ const char gmx_unused *deviceOptions,
+ unsigned long gmx_unused Flags,
+ gmx_walltime_accounting_t walltime_accounting)
+{
+ const char *SD = "Steepest Descents";
+ em_state_t *s_min, *s_try;
+ rvec *f_global;
+ gmx_localtop_t *top;
+ gmx_enerdata_t *enerd;
+ rvec *f;
+ gmx_global_stat_t gstat;
+ t_graph *graph;
+ real stepsize, constepsize;
+ real ustep, fnormn;
+ gmx_mdoutf_t outf;
+ t_mdebin *mdebin;
+ gmx_bool bDone, bAbort, do_x, do_f;
+ tensor vir, pres;
+ rvec mu_tot;
+ int nsteps;
+ int count = 0;
+ int steps_accepted = 0;
+ /* not used */
+ real terminate = 0;
+
+ s_min = init_em_state();
+ s_try = init_em_state();
+
+ /* Init em and store the local state in s_try */
+ init_em(fplog, SD, cr, inputrec,
+ state_global, top_global, s_try, &top, &f, &f_global,
+ nrnb, mu_tot, fr, &enerd, &graph, mdatoms, &gstat, vsite, constr,
+ nfile, fnm, &outf, &mdebin);
+
+ /* Print to log file */
+ print_em_start(fplog, cr, walltime_accounting, wcycle, SD);
+
+ /* Set variables for stepsize (in nm). This is the largest
+ * step that we are going to make in any direction.
+ */
+ ustep = inputrec->em_stepsize;
+ stepsize = 0;
+
+ /* Max number of steps */
+ nsteps = inputrec->nsteps;
+
+ if (MASTER(cr))
+ {
+ /* Print to the screen */
+ sp_header(stderr, SD, inputrec->em_tol, nsteps);
+ }
+ if (fplog)
+ {
+ sp_header(fplog, SD, inputrec->em_tol, nsteps);
+ }
+
+ /**** HERE STARTS THE LOOP ****
+ * count is the counter for the number of steps
+ * bDone will be TRUE when the minimization has converged
+ * bAbort will be TRUE when nsteps steps have been performed or when
+ * the stepsize becomes smaller than is reasonable for machine precision
+ */
+ count = 0;
+ bDone = FALSE;
+ bAbort = FALSE;
+ while (!bDone && !bAbort)
+ {
+ bAbort = (nsteps >= 0) && (count == nsteps);
+
+ /* set new coordinates, except for first step */
+ if (count > 0)
+ {
+ do_em_step(cr, inputrec, mdatoms, fr->bMolPBC,
+ s_min, stepsize, s_min->f, s_try,
+ constr, top, nrnb, wcycle, count);
+ }
+
+ evaluate_energy(fplog, cr,
+ top_global, s_try, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, count, count == 0);
+
+ if (MASTER(cr))
+ {
+ print_ebin_header(fplog, count, count, s_try->s.lambda[efptFEP]);
+ }
+
+ if (count == 0)
+ {
+ s_min->epot = s_try->epot + 1;
+ }
+
+ /* Print it if necessary */
+ if (MASTER(cr))
+ {
+ if (bVerbose)
+ {
+ fprintf(stderr, "Step=%5d, Dmax= %6.1e nm, Epot= %12.5e Fmax= %11.5e, atom= %d%c",
+ count, ustep, s_try->epot, s_try->fmax, s_try->a_fmax+1,
+ (s_try->epot < s_min->epot) ? '\n' : '\r');
+ }
+
+ if (s_try->epot < s_min->epot)
+ {
+ /* Store the new (lower) energies */
+ upd_mdebin(mdebin, FALSE, FALSE, (double)count,
+ mdatoms->tmass, enerd, &s_try->s, inputrec->fepvals, inputrec->expandedvals,
+ s_try->s.box, NULL, NULL, vir, pres, NULL, mu_tot, constr);
+ print_ebin(mdoutf_get_fp_ene(outf), TRUE,
+ do_per_step(steps_accepted, inputrec->nstdisreout),
+ do_per_step(steps_accepted, inputrec->nstorireout),
+ fplog, count, count, eprNORMAL, TRUE,
+ mdebin, fcd, &(top_global->groups), &(inputrec->opts));
+ fflush(fplog);
+ }
+ }
+
+ /* Now if the new energy is smaller than the previous...
+ * or if this is the first step!
+ * or if we did random steps!
+ */
+
+ if ( (count == 0) || (s_try->epot < s_min->epot) )
+ {
+ steps_accepted++;
+
+ /* Test whether the convergence criterion is met... */
+ bDone = (s_try->fmax < inputrec->em_tol);
+
+ /* Copy the arrays for force, positions and energy */
+ /* The 'Min' array always holds the coords and forces of the minimal
+ sampled energy */
+ swap_em_state(s_min, s_try);
+ if (count > 0)
+ {
+ ustep *= 1.2;
+ }
+
+ /* Write to trn, if necessary */
+ do_x = do_per_step(steps_accepted, inputrec->nstxout);
+ do_f = do_per_step(steps_accepted, inputrec->nstfout);
+ write_em_traj(fplog, cr, outf, do_x, do_f, NULL,
+ top_global, inputrec, count,
+ s_min, state_global, f_global);
+ }
+ else
+ {
+ /* If energy is not smaller make the step smaller... */
+ ustep *= 0.5;
+
+ if (DOMAINDECOMP(cr) && s_min->s.ddp_count != cr->dd->ddp_count)
+ {
+ /* Reload the old state */
+ em_dd_partition_system(fplog, count, cr, top_global, inputrec,
+ s_min, top, mdatoms, fr, vsite, constr,
+ nrnb, wcycle);
+ }
+ }
+
+ /* Determine new step */
+ stepsize = ustep/s_min->fmax;
+
+ /* Check if stepsize is too small, with 1 nm as a characteristic length */
+#ifdef GMX_DOUBLE
+ if (count == nsteps || ustep < 1e-12)
+#else
+ if (count == nsteps || ustep < 1e-6)
+#endif
+ {
+ if (MASTER(cr))
+ {
+ warn_step(stderr, inputrec->em_tol, count == nsteps, constr != NULL);
+ warn_step(fplog, inputrec->em_tol, count == nsteps, constr != NULL);
+ }
+ bAbort = TRUE;
+ }
+
+ count++;
+ } /* End of the loop */
+
+ /* Print some shit... */
+ if (MASTER(cr))
+ {
+ fprintf(stderr, "\nwriting lowest energy coordinates.\n");
+ }
+ write_em_traj(fplog, cr, outf, TRUE, inputrec->nstfout, ftp2fn(efSTO, nfile, fnm),
+ top_global, inputrec, count,
+ s_min, state_global, f_global);
+
+ fnormn = s_min->fnorm/sqrt(state_global->natoms);
+
+ if (MASTER(cr))
+ {
+ print_converged(stderr, SD, inputrec->em_tol, count, bDone, nsteps,
+ s_min->epot, s_min->fmax, s_min->a_fmax, fnormn);
+ print_converged(fplog, SD, inputrec->em_tol, count, bDone, nsteps,
+ s_min->epot, s_min->fmax, s_min->a_fmax, fnormn);
+ }
+
+ finish_em(cr, outf, walltime_accounting, wcycle);
+
+ /* To print the actual number of steps we needed somewhere */
+ inputrec->nsteps = count;
+
+ walltime_accounting_set_nsteps_done(walltime_accounting, count);
+
+ return 0;
+} /* That's all folks */
+
+
+double do_nm(FILE *fplog, t_commrec *cr,
+ int nfile, const t_filenm fnm[],
+ const output_env_t gmx_unused oenv, gmx_bool bVerbose, gmx_bool gmx_unused bCompact,
+ int gmx_unused nstglobalcomm,
+ gmx_vsite_t *vsite, gmx_constr_t constr,
+ int gmx_unused stepout,
+ t_inputrec *inputrec,
+ gmx_mtop_t *top_global, t_fcdata *fcd,
+ t_state *state_global,
+ t_mdatoms *mdatoms,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_edsam_t gmx_unused ed,
+ t_forcerec *fr,
+ int gmx_unused repl_ex_nst, int gmx_unused repl_ex_nex, int gmx_unused repl_ex_seed,
+ gmx_membed_t gmx_unused membed,
+ real gmx_unused cpt_period, real gmx_unused max_hours,
+ const char gmx_unused *deviceOptions,
+ unsigned long gmx_unused Flags,
+ gmx_walltime_accounting_t walltime_accounting)
+{
+ const char *NM = "Normal Mode Analysis";
+ gmx_mdoutf_t outf;
+ int natoms, atom, d;
+ int nnodes, node;
+ rvec *f_global;
+ gmx_localtop_t *top;
+ gmx_enerdata_t *enerd;
+ rvec *f;
+ gmx_global_stat_t gstat;
+ t_graph *graph;
+ real t, t0, lambda, lam0;
+ gmx_bool bNS;
+ tensor vir, pres;
+ rvec mu_tot;
+ rvec *fneg, *dfdx;
+ gmx_bool bSparse; /* use sparse matrix storage format */
+ size_t sz = 0;
+ gmx_sparsematrix_t * sparse_matrix = NULL;
+ real * full_matrix = NULL;
+ em_state_t * state_work;
+
+ /* added with respect to mdrun */
+ int i, j, k, row, col;
+ real der_range = 10.0*sqrt(GMX_REAL_EPS);
+ real x_min;
+ real fnorm, fmax;
+
+ if (constr != NULL)
+ {
+ gmx_fatal(FARGS, "Constraints present with Normal Mode Analysis, this combination is not supported");
+ }
+
+ state_work = init_em_state();
+
+ /* Init em and store the local state in state_minimum */
+ init_em(fplog, NM, cr, inputrec,
+ state_global, top_global, state_work, &top,
+ &f, &f_global,
+ nrnb, mu_tot, fr, &enerd, &graph, mdatoms, &gstat, vsite, constr,
+ nfile, fnm, &outf, NULL);
+
+ natoms = top_global->natoms;
+ snew(fneg, natoms);
+ snew(dfdx, natoms);
+
+#ifndef GMX_DOUBLE
+ if (MASTER(cr))
+ {
+ fprintf(stderr,
+ "NOTE: This version of Gromacs has been compiled in single precision,\n"
+ " which MIGHT not be accurate enough for normal mode analysis.\n"
+ " Gromacs now uses sparse matrix storage, so the memory requirements\n"
+ " are fairly modest even if you recompile in double precision.\n\n");
+ }
+#endif
+
+ /* Check if we can/should use sparse storage format.
+ *
+ * Sparse format is only useful when the Hessian itself is sparse, which it
+ * will be when we use a cutoff.
+ * For small systems (n<1000) it is easier to always use full matrix format, though.
+ */
+ if (EEL_FULL(fr->eeltype) || fr->rlist == 0.0)
+ {
+ md_print_info(cr, fplog, "Non-cutoff electrostatics used, forcing full Hessian format.\n");
+ bSparse = FALSE;
+ }
+ else if (top_global->natoms < 1000)
+ {
+ md_print_info(cr, fplog, "Small system size (N=%d), using full Hessian format.\n", top_global->natoms);
+ bSparse = FALSE;
+ }
+ else
+ {
+ md_print_info(cr, fplog, "Using compressed symmetric sparse Hessian format.\n");
+ bSparse = TRUE;
+ }
+
+ if (MASTER(cr))
+ {
+ sz = DIM*top_global->natoms;
+
+ fprintf(stderr, "Allocating Hessian memory...\n\n");
+
+ if (bSparse)
+ {
+ sparse_matrix = gmx_sparsematrix_init(sz);
+ sparse_matrix->compressed_symmetric = TRUE;
+ }
+ else
+ {
+ snew(full_matrix, sz*sz);
+ }
+ }
+
+ /* Initial values */
+ t0 = inputrec->init_t;
+ lam0 = inputrec->fepvals->init_lambda;
+ t = t0;
+ lambda = lam0;
+
+ init_nrnb(nrnb);
+
+ where();
+
+ /* Write start time and temperature */
+ print_em_start(fplog, cr, walltime_accounting, wcycle, NM);
+
+ /* fudge nr of steps to nr of atoms */
+ inputrec->nsteps = natoms*2;
+
+ if (MASTER(cr))
+ {
+ fprintf(stderr, "starting normal mode calculation '%s'\n%d steps.\n\n",
+ *(top_global->name), (int)inputrec->nsteps);
+ }
+
+ nnodes = cr->nnodes;
+
+ /* Make evaluate_energy do a single node force calculation */
+ cr->nnodes = 1;
+ evaluate_energy(fplog, cr,
+ top_global, state_work, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, -1, TRUE);
+ cr->nnodes = nnodes;
+
+ /* if forces are not small, warn user */
+ get_state_f_norm_max(cr, &(inputrec->opts), mdatoms, state_work);
+
+ md_print_info(cr, fplog, "Maximum force:%12.5e\n", state_work->fmax);
+ if (state_work->fmax > 1.0e-3)
+ {
+ md_print_info(cr, fplog,
+ "The force is probably not small enough to "
+ "ensure that you are at a minimum.\n"
+ "Be aware that negative eigenvalues may occur\n"
+ "when the resulting matrix is diagonalized.\n\n");
+ }
+
+ /***********************************************************
+ *
+ * Loop over all pairs in matrix
+ *
+ * do_force called twice. Once with positive and
+ * once with negative displacement
+ *
+ ************************************************************/
+
+ /* Steps are divided one by one over the nodes */
+ for (atom = cr->nodeid; atom < natoms; atom += nnodes)
+ {
+
+ for (d = 0; d < DIM; d++)
+ {
+ x_min = state_work->s.x[atom][d];
+
+ state_work->s.x[atom][d] = x_min - der_range;
+
+ /* Make evaluate_energy do a single node force calculation */
+ cr->nnodes = 1;
+ evaluate_energy(fplog, cr,
+ top_global, state_work, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, atom*2, FALSE);
+
+ for (i = 0; i < natoms; i++)
+ {
+ copy_rvec(state_work->f[i], fneg[i]);
+ }
+
+ state_work->s.x[atom][d] = x_min + der_range;
+
+ evaluate_energy(fplog, cr,
+ top_global, state_work, top,
+ inputrec, nrnb, wcycle, gstat,
+ vsite, constr, fcd, graph, mdatoms, fr,
+ mu_tot, enerd, vir, pres, atom*2+1, FALSE);
+ cr->nnodes = nnodes;
+
+ /* x is restored to original */
+ state_work->s.x[atom][d] = x_min;
+
+ for (j = 0; j < natoms; j++)
+ {
+ for (k = 0; (k < DIM); k++)
+ {
+ dfdx[j][k] =
+ -(state_work->f[j][k] - fneg[j][k])/(2*der_range);
+ }
+ }
+
+ if (!MASTER(cr))
+ {
+#ifdef GMX_MPI
+#ifdef GMX_DOUBLE
+#define mpi_type MPI_DOUBLE
+#else
+#define mpi_type MPI_FLOAT
+#endif
+ MPI_Send(dfdx[0], natoms*DIM, mpi_type, MASTERNODE(cr), cr->nodeid,
+ cr->mpi_comm_mygroup);
+#endif
+ }
+ else
+ {
+ for (node = 0; (node < nnodes && atom+node < natoms); node++)
+ {
+ if (node > 0)
+ {
+#ifdef GMX_MPI
+ MPI_Status stat;
+ MPI_Recv(dfdx[0], natoms*DIM, mpi_type, node, node,
+ cr->mpi_comm_mygroup, &stat);
+#undef mpi_type
+#endif
+ }
+
+ row = (atom + node)*DIM + d;
+
+ for (j = 0; j < natoms; j++)
+ {
+ for (k = 0; k < DIM; k++)
+ {
+ col = j*DIM + k;
+
+ if (bSparse)
+ {
+ if (col >= row && dfdx[j][k] != 0.0)
+ {
+ gmx_sparsematrix_increment_value(sparse_matrix,
+ row, col, dfdx[j][k]);
+ }
+ }
+ else
+ {
+ full_matrix[row*sz+col] = dfdx[j][k];
+ }
+ }
+ }
+ }
+ }
+
+ if (bVerbose && fplog)
+ {
+ fflush(fplog);
+ }
+ }
+ /* write progress */
+ if (MASTER(cr) && bVerbose)
+ {
+ fprintf(stderr, "\rFinished step %d out of %d",
+ min(atom+nnodes, natoms), natoms);
+ fflush(stderr);
+ }
+ }
+
+ if (MASTER(cr))
+ {
+ fprintf(stderr, "\n\nWriting Hessian...\n");
+ gmx_mtxio_write(ftp2fn(efMTX, nfile, fnm), sz, sz, full_matrix, sparse_matrix);
+ }
+
+ finish_em(cr, outf, walltime_accounting, wcycle);
+
+ walltime_accounting_set_nsteps_done(walltime_accounting, natoms*2);
+
+ return 0;
+}
--- /dev/null
+/*
+ * This file is part of the GROMACS molecular simulation package.
+ *
+ * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
+ * Copyright (c) 2001-2004, The GROMACS development team.
+ * Copyright (c) 2013,2014, by the GROMACS development team, led by
+ * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
+ * and including many others, as listed in the AUTHORS file in the
+ * top-level source directory and at http://www.gromacs.org.
+ *
+ * GROMACS is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public License
+ * as published by the Free Software Foundation; either version 2.1
+ * of the License, or (at your option) any later version.
+ *
+ * GROMACS is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with GROMACS; if not, see
+ * http://www.gnu.org/licenses, or write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * If you want to redistribute modifications to GROMACS, please
+ * consider that scientific software is very special. Version
+ * control is crucial - bugs must be traceable. We will be happy to
+ * consider code for inclusion in the official distribution, but
+ * derived work must not be called official GROMACS. Details are found
+ * in the README & COPYING files - if they are missing, get the
+ * official version at http://www.gromacs.org.
+ *
+ * To help us fund GROMACS development, we humbly ask that you cite
+ * the research papers on the package. Check out http://www.gromacs.org.
+ */
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <stdio.h>
+#ifdef HAVE_SYS_TIME_H
+#include <sys/time.h>
+#endif
+#include <math.h>
+#include "typedefs.h"
+#include "string2.h"
+#include "smalloc.h"
+#include "names.h"
+#include "mvdata.h"
+#include "txtdump.h"
+#include "pbc.h"
+#include "chargegroup.h"
+#include "vec.h"
+#include "nrnb.h"
+#include "mshift.h"
+#include "mdrun.h"
+#include "sim_util.h"
+#include "update.h"
+#include "physics.h"
+#include "main.h"
+#include "mdatoms.h"
+#include "force.h"
+#include "bondf.h"
+#include "pme.h"
+#include "disre.h"
+#include "orires.h"
+#include "network.h"
+#include "calcmu.h"
+#include "constr.h"
+#include "xvgr.h"
+#include "copyrite.h"
+#include "gmx_random.h"
+#include "domdec.h"
+#include "partdec.h"
+#include "genborn.h"
+#include "nbnxn_atomdata.h"
+#include "nbnxn_search.h"
+#include "nbnxn_kernels/nbnxn_kernel_ref.h"
+#include "nbnxn_kernels/simd_4xn/nbnxn_kernel_simd_4xn.h"
+#include "nbnxn_kernels/simd_2xnn/nbnxn_kernel_simd_2xnn.h"
+#include "nbnxn_kernels/nbnxn_kernel_gpu_ref.h"
+
+#include "gromacs/timing/wallcycle.h"
+#include "gromacs/timing/walltime_accounting.h"
+#include "gromacs/utility/gmxmpi.h"
+#include "gromacs/essentialdynamics/edsam.h"
+#include "gromacs/pulling/pull.h"
+#include "gromacs/pulling/pull_rotation.h"
+
+#include "adress.h"
+#include "qmmm.h"
+
+#include "nbnxn_cuda_data_mgmt.h"
+#include "nbnxn_cuda/nbnxn_cuda.h"
+
+void print_time(FILE *out,
+ gmx_walltime_accounting_t walltime_accounting,
+ gmx_int64_t step,
+ t_inputrec *ir,
+ t_commrec gmx_unused *cr)
+{
+ time_t finish;
+ char timebuf[STRLEN];
+ double dt, elapsed_seconds, time_per_step;
+ char buf[48];
+
+#ifndef GMX_THREAD_MPI
+ if (!PAR(cr))
+#endif
+ {
+ fprintf(out, "\r");
+ }
+ fprintf(out, "step %s", gmx_step_str(step, buf));
+ if ((step >= ir->nstlist))
+ {
+ double seconds_since_epoch = gmx_gettime();
+ elapsed_seconds = seconds_since_epoch - walltime_accounting_get_start_time_stamp(walltime_accounting);
+ time_per_step = elapsed_seconds/(step - ir->init_step + 1);
+ dt = (ir->nsteps + ir->init_step - step) * time_per_step;
+
+ if (ir->nsteps >= 0)
+ {
+ if (dt >= 300)
+ {
+ finish = (time_t) (seconds_since_epoch + dt);
+ gmx_ctime_r(&finish, timebuf, STRLEN);
+ sprintf(buf, "%s", timebuf);
+ buf[strlen(buf)-1] = '\0';
+ fprintf(out, ", will finish %s", buf);
+ }
+ else
+ {
+ fprintf(out, ", remaining wall clock time: %5d s ", (int)dt);
+ }
+ }
+ else
+ {
+ fprintf(out, " performance: %.1f ns/day ",
+ ir->delta_t/1000*24*60*60/time_per_step);
+ }
+ }
+#ifndef GMX_THREAD_MPI
+ if (PAR(cr))
+ {
+ fprintf(out, "\n");
+ }
+#endif
+
+ fflush(out);
+}
+
+void print_date_and_time(FILE *fplog, int nodeid, const char *title,
+ const gmx_walltime_accounting_t walltime_accounting)
+{
+ int i;
+ char timebuf[STRLEN];
+ char time_string[STRLEN];
+ time_t tmptime;
+
+ if (fplog)
+ {
+ if (walltime_accounting != NULL)
+ {
+ tmptime = (time_t) walltime_accounting_get_start_time_stamp(walltime_accounting);
+ gmx_ctime_r(&tmptime, timebuf, STRLEN);
+ }
+ else
+ {
+ tmptime = (time_t) gmx_gettime();
+ gmx_ctime_r(&tmptime, timebuf, STRLEN);
+ }
+ for (i = 0; timebuf[i] >= ' '; i++)
+ {
+ time_string[i] = timebuf[i];
+ }
+ time_string[i] = '\0';
+
+ fprintf(fplog, "%s on node %d %s\n", title, nodeid, time_string);
+ }
+}
+
++void print_start(FILE *fplog, t_commrec *cr,
++ gmx_walltime_accounting_t walltime_accounting,
++ const char *name)
++{
++ char buf[STRLEN];
++
++ sprintf(buf, "Started %s", name);
++ print_date_and_time(fplog, cr->nodeid, buf, walltime_accounting);
++}
++
+static void sum_forces(int start, int end, rvec f[], rvec flr[])
+{
+ int i;
+
+ if (gmx_debug_at)
+ {
+ pr_rvecs(debug, 0, "fsr", f+start, end-start);
+ pr_rvecs(debug, 0, "flr", flr+start, end-start);
+ }
+ for (i = start; (i < end); i++)
+ {
+ rvec_inc(f[i], flr[i]);
+ }
+}
+
+/*
+ * calc_f_el calculates forces due to an electric field.
+ *
+ * force is kJ mol^-1 nm^-1 = e * kJ mol^-1 nm^-1 / e
+ *
+ * Et[] contains the parameters for the time dependent
+ * part of the field (not yet used).
+ * Ex[] contains the parameters for
+ * the spatial dependent part of the field. You can have cool periodic
+ * fields in principle, but only a constant field is supported
+ * now.
+ * The function should return the energy due to the electric field
+ * (if any) but for now returns 0.
+ *
+ * WARNING:
+ * There can be problems with the virial.
+ * Since the field is not self-consistent this is unavoidable.
+ * For neutral molecules the virial is correct within this approximation.
+ * For neutral systems with many charged molecules the error is small.
+ * But for systems with a net charge or a few charged molecules
+ * the error can be significant when the field is high.
+ * Solution: implement a self-consitent electric field into PME.
+ */
+static void calc_f_el(FILE *fp, int start, int homenr,
+ real charge[], rvec f[],
+ t_cosines Ex[], t_cosines Et[], double t)
+{
+ rvec Ext;
+ real t0;
+ int i, m;
+
+ for (m = 0; (m < DIM); m++)
+ {
+ if (Et[m].n > 0)
+ {
+ if (Et[m].n == 3)
+ {
+ t0 = Et[m].a[1];
+ Ext[m] = cos(Et[m].a[0]*(t-t0))*exp(-sqr(t-t0)/(2.0*sqr(Et[m].a[2])));
+ }
+ else
+ {
+ Ext[m] = cos(Et[m].a[0]*t);
+ }
+ }
+ else
+ {
+ Ext[m] = 1.0;
+ }
+ if (Ex[m].n > 0)
+ {
+ /* Convert the field strength from V/nm to MD-units */
+ Ext[m] *= Ex[m].a[0]*FIELDFAC;
+ for (i = start; (i < start+homenr); i++)
+ {
+ f[i][m] += charge[i]*Ext[m];
+ }
+ }
+ else
+ {
+ Ext[m] = 0;
+ }
+ }
+ if (fp != NULL)
+ {
+ fprintf(fp, "%10g %10g %10g %10g #FIELD\n", t,
+ Ext[XX]/FIELDFAC, Ext[YY]/FIELDFAC, Ext[ZZ]/FIELDFAC);
+ }
+}
+
+static void calc_virial(int start, int homenr, rvec x[], rvec f[],
+ tensor vir_part, t_graph *graph, matrix box,
+ t_nrnb *nrnb, const t_forcerec *fr, int ePBC)
+{
+ int i, j;
+ tensor virtest;
+
+ /* The short-range virial from surrounding boxes */
+ clear_mat(vir_part);
+ calc_vir(SHIFTS, fr->shift_vec, fr->fshift, vir_part, ePBC == epbcSCREW, box);
+ inc_nrnb(nrnb, eNR_VIRIAL, SHIFTS);
+
+ /* Calculate partial virial, for local atoms only, based on short range.
+ * Total virial is computed in global_stat, called from do_md
+ */
+ f_calc_vir(start, start+homenr, x, f, vir_part, graph, box);
+ inc_nrnb(nrnb, eNR_VIRIAL, homenr);
+
+ /* Add position restraint contribution */
+ for (i = 0; i < DIM; i++)
+ {
+ vir_part[i][i] += fr->vir_diag_posres[i];
+ }
+
+ /* Add wall contribution */
+ for (i = 0; i < DIM; i++)
+ {
+ vir_part[i][ZZ] += fr->vir_wall_z[i];
+ }
+
+ if (debug)
+ {
+ pr_rvecs(debug, 0, "vir_part", vir_part, DIM);
+ }
+}
+
+static void posres_wrapper(FILE *fplog,
+ int flags,
+ gmx_bool bSepDVDL,
+ t_inputrec *ir,
+ t_nrnb *nrnb,
+ gmx_localtop_t *top,
+ matrix box, rvec x[],
+ gmx_enerdata_t *enerd,
+ real *lambda,
+ t_forcerec *fr)
+{
+ t_pbc pbc;
+ real v, dvdl;
+ int i;
+
+ /* Position restraints always require full pbc */
+ set_pbc(&pbc, ir->ePBC, box);
+ dvdl = 0;
+ v = posres(top->idef.il[F_POSRES].nr, top->idef.il[F_POSRES].iatoms,
+ top->idef.iparams_posres,
+ (const rvec*)x, fr->f_novirsum, fr->vir_diag_posres,
+ ir->ePBC == epbcNONE ? NULL : &pbc,
+ lambda[efptRESTRAINT], &dvdl,
+ fr->rc_scaling, fr->ePBC, fr->posres_com, fr->posres_comB);
+ if (bSepDVDL)
+ {
+ gmx_print_sepdvdl(fplog, interaction_function[F_POSRES].longname, v, dvdl);
+ }
+ enerd->term[F_POSRES] += v;
+ /* If just the force constant changes, the FEP term is linear,
+ * but if k changes, it is not.
+ */
+ enerd->dvdl_nonlin[efptRESTRAINT] += dvdl;
+ inc_nrnb(nrnb, eNR_POSRES, top->idef.il[F_POSRES].nr/2);
+
+ if ((ir->fepvals->n_lambda > 0) && (flags & GMX_FORCE_DHDL))
+ {
+ for (i = 0; i < enerd->n_lambda; i++)
+ {
+ real dvdl_dum, lambda_dum;
+
+ lambda_dum = (i == 0 ? lambda[efptRESTRAINT] : ir->fepvals->all_lambda[efptRESTRAINT][i-1]);
+ v = posres(top->idef.il[F_POSRES].nr, top->idef.il[F_POSRES].iatoms,
+ top->idef.iparams_posres,
+ (const rvec*)x, NULL, NULL,
+ ir->ePBC == epbcNONE ? NULL : &pbc, lambda_dum, &dvdl,
+ fr->rc_scaling, fr->ePBC, fr->posres_com, fr->posres_comB);
+ enerd->enerpart_lambda[i] += v;
+ }
+ }
+}
+
+static void fbposres_wrapper(t_inputrec *ir,
+ t_nrnb *nrnb,
+ gmx_localtop_t *top,
+ matrix box, rvec x[],
+ gmx_enerdata_t *enerd,
+ t_forcerec *fr)
+{
+ t_pbc pbc;
+ real v;
+
+ /* Flat-bottomed position restraints always require full pbc */
+ set_pbc(&pbc, ir->ePBC, box);
+ v = fbposres(top->idef.il[F_FBPOSRES].nr, top->idef.il[F_FBPOSRES].iatoms,
+ top->idef.iparams_fbposres,
+ (const rvec*)x, fr->f_novirsum, fr->vir_diag_posres,
+ ir->ePBC == epbcNONE ? NULL : &pbc,
+ fr->rc_scaling, fr->ePBC, fr->posres_com);
+ enerd->term[F_FBPOSRES] += v;
+ inc_nrnb(nrnb, eNR_FBPOSRES, top->idef.il[F_FBPOSRES].nr/2);
+}
+
+static void pull_potential_wrapper(FILE *fplog,
+ gmx_bool bSepDVDL,
+ t_commrec *cr,
+ t_inputrec *ir,
+ matrix box, rvec x[],
+ rvec f[],
+ tensor vir_force,
+ t_mdatoms *mdatoms,
+ gmx_enerdata_t *enerd,
+ real *lambda,
+ double t)
+{
+ t_pbc pbc;
+ real dvdl;
+
+ /* Calculate the center of mass forces, this requires communication,
+ * which is why pull_potential is called close to other communication.
+ * The virial contribution is calculated directly,
+ * which is why we call pull_potential after calc_virial.
+ */
+ set_pbc(&pbc, ir->ePBC, box);
+ dvdl = 0;
+ enerd->term[F_COM_PULL] +=
+ pull_potential(ir->ePull, ir->pull, mdatoms, &pbc,
+ cr, t, lambda[efptRESTRAINT], x, f, vir_force, &dvdl);
+ if (bSepDVDL)
+ {
+ gmx_print_sepdvdl(fplog, "Com pull", enerd->term[F_COM_PULL], dvdl);
+ }
+ enerd->dvdl_lin[efptRESTRAINT] += dvdl;
+}
+
+static void pme_receive_force_ener(FILE *fplog,
+ gmx_bool bSepDVDL,
+ t_commrec *cr,
+ gmx_wallcycle_t wcycle,
+ gmx_enerdata_t *enerd,
+ t_forcerec *fr)
+{
+ real e_q, e_lj, v, dvdl_q, dvdl_lj;
+ float cycles_ppdpme, cycles_seppme;
+
+ cycles_ppdpme = wallcycle_stop(wcycle, ewcPPDURINGPME);
+ dd_cycles_add(cr->dd, cycles_ppdpme, ddCyclPPduringPME);
+
+ /* In case of node-splitting, the PP nodes receive the long-range
+ * forces, virial and energy from the PME nodes here.
+ */
+ wallcycle_start(wcycle, ewcPP_PMEWAITRECVF);
+ dvdl_q = 0;
+ dvdl_lj = 0;
+ gmx_pme_receive_f(cr, fr->f_novirsum, fr->vir_el_recip, &e_q,
+ fr->vir_lj_recip, &e_lj, &dvdl_q, &dvdl_lj,
+ &cycles_seppme);
+ if (bSepDVDL)
+ {
+ gmx_print_sepdvdl(fplog, "Electrostatic PME mesh", e_q, dvdl_q);
+ gmx_print_sepdvdl(fplog, "Lennard-Jones PME mesh", e_lj, dvdl_lj);
+ }
+ enerd->term[F_COUL_RECIP] += e_q;
+ enerd->term[F_LJ_RECIP] += e_lj;
+ enerd->dvdl_lin[efptCOUL] += dvdl_q;
+ enerd->dvdl_lin[efptVDW] += dvdl_lj;
+
+ if (wcycle)
+ {
+ dd_cycles_add(cr->dd, cycles_seppme, ddCyclPME);
+ }
+ wallcycle_stop(wcycle, ewcPP_PMEWAITRECVF);
+}
+
+static void print_large_forces(FILE *fp, t_mdatoms *md, t_commrec *cr,
+ gmx_int64_t step, real pforce, rvec *x, rvec *f)
+{
+ int i;
+ real pf2, fn2;
+ char buf[STEPSTRSIZE];
+
+ pf2 = sqr(pforce);
+ for (i = md->start; i < md->start+md->homenr; i++)
+ {
+ fn2 = norm2(f[i]);
+ /* We also catch NAN, if the compiler does not optimize this away. */
+ if (fn2 >= pf2 || fn2 != fn2)
+ {
+ fprintf(fp, "step %s atom %6d x %8.3f %8.3f %8.3f force %12.5e\n",
+ gmx_step_str(step, buf),
+ ddglatnr(cr->dd, i), x[i][XX], x[i][YY], x[i][ZZ], sqrt(fn2));
+ }
+ }
+}
+
+static void post_process_forces(t_commrec *cr,
+ gmx_int64_t step,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_localtop_t *top,
+ matrix box, rvec x[],
+ rvec f[],
+ tensor vir_force,
+ t_mdatoms *mdatoms,
+ t_graph *graph,
+ t_forcerec *fr, gmx_vsite_t *vsite,
+ int flags)
+{
+ if (fr->bF_NoVirSum)
+ {
+ if (vsite)
+ {
+ /* Spread the mesh force on virtual sites to the other particles...
+ * This is parallellized. MPI communication is performed
+ * if the constructing atoms aren't local.
+ */
+ wallcycle_start(wcycle, ewcVSITESPREAD);
+ spread_vsite_f(vsite, x, fr->f_novirsum, NULL,
+ (flags & GMX_FORCE_VIRIAL), fr->vir_el_recip,
+ nrnb,
+ &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
+ wallcycle_stop(wcycle, ewcVSITESPREAD);
+ }
+ if (flags & GMX_FORCE_VIRIAL)
+ {
+ /* Now add the forces, this is local */
+ if (fr->bDomDec)
+ {
+ sum_forces(0, fr->f_novirsum_n, f, fr->f_novirsum);
+ }
+ else
+ {
+ sum_forces(mdatoms->start, mdatoms->start+mdatoms->homenr,
+ f, fr->f_novirsum);
+ }
+ if (EEL_FULL(fr->eeltype))
+ {
+ /* Add the mesh contribution to the virial */
+ m_add(vir_force, fr->vir_el_recip, vir_force);
+ }
+ if (EVDW_PME(fr->vdwtype))
+ {
+ /* Add the mesh contribution to the virial */
+ m_add(vir_force, fr->vir_lj_recip, vir_force);
+ }
+ if (debug)
+ {
+ pr_rvecs(debug, 0, "vir_force", vir_force, DIM);
+ }
+ }
+ }
+
+ if (fr->print_force >= 0)
+ {
+ print_large_forces(stderr, mdatoms, cr, step, fr->print_force, x, f);
+ }
+}
+
+static void do_nb_verlet(t_forcerec *fr,
+ interaction_const_t *ic,
+ gmx_enerdata_t *enerd,
+ int flags, int ilocality,
+ int clearF,
+ t_nrnb *nrnb,
+ gmx_wallcycle_t wcycle)
+{
+ int nnbl, kernel_type, enr_nbnxn_kernel_ljc, enr_nbnxn_kernel_lj;
+ char *env;
+ nonbonded_verlet_group_t *nbvg;
+ gmx_bool bCUDA;
+
+ if (!(flags & GMX_FORCE_NONBONDED))
+ {
+ /* skip non-bonded calculation */
+ return;
+ }
+
+ nbvg = &fr->nbv->grp[ilocality];
+
+ /* CUDA kernel launch overhead is already timed separately */
+ if (fr->cutoff_scheme != ecutsVERLET)
+ {
+ gmx_incons("Invalid cut-off scheme passed!");
+ }
+
+ bCUDA = (nbvg->kernel_type == nbnxnk8x8x8_CUDA);
+
+ if (!bCUDA)
+ {
+ wallcycle_sub_start(wcycle, ewcsNONBONDED);
+ }
+ switch (nbvg->kernel_type)
+ {
+ case nbnxnk4x4_PlainC:
+ nbnxn_kernel_ref(&nbvg->nbl_lists,
+ nbvg->nbat, ic,
+ fr->shift_vec,
+ flags,
+ clearF,
+ fr->fshift[0],
+ enerd->grpp.ener[egCOULSR],
+ fr->bBHAM ?
+ enerd->grpp.ener[egBHAMSR] :
+ enerd->grpp.ener[egLJSR]);
+ break;
+
+ case nbnxnk4xN_SIMD_4xN:
+ nbnxn_kernel_simd_4xn(&nbvg->nbl_lists,
+ nbvg->nbat, ic,
+ nbvg->ewald_excl,
+ fr->shift_vec,
+ flags,
+ clearF,
+ fr->fshift[0],
+ enerd->grpp.ener[egCOULSR],
+ fr->bBHAM ?
+ enerd->grpp.ener[egBHAMSR] :
+ enerd->grpp.ener[egLJSR]);
+ break;
+ case nbnxnk4xN_SIMD_2xNN:
+ nbnxn_kernel_simd_2xnn(&nbvg->nbl_lists,
+ nbvg->nbat, ic,
+ nbvg->ewald_excl,
+ fr->shift_vec,
+ flags,
+ clearF,
+ fr->fshift[0],
+ enerd->grpp.ener[egCOULSR],
+ fr->bBHAM ?
+ enerd->grpp.ener[egBHAMSR] :
+ enerd->grpp.ener[egLJSR]);
+ break;
+
+ case nbnxnk8x8x8_CUDA:
+ nbnxn_cuda_launch_kernel(fr->nbv->cu_nbv, nbvg->nbat, flags, ilocality);
+ break;
+
+ case nbnxnk8x8x8_PlainC:
+ nbnxn_kernel_gpu_ref(nbvg->nbl_lists.nbl[0],
+ nbvg->nbat, ic,
+ fr->shift_vec,
+ flags,
+ clearF,
+ nbvg->nbat->out[0].f,
+ fr->fshift[0],
+ enerd->grpp.ener[egCOULSR],
+ fr->bBHAM ?
+ enerd->grpp.ener[egBHAMSR] :
+ enerd->grpp.ener[egLJSR]);
+ break;
+
+ default:
+ gmx_incons("Invalid nonbonded kernel type passed!");
+
+ }
+ if (!bCUDA)
+ {
+ wallcycle_sub_stop(wcycle, ewcsNONBONDED);
+ }
+
+ if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
+ {
+ enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_RF;
+ }
+ else if ((!bCUDA && nbvg->ewald_excl == ewaldexclAnalytical) ||
+ (bCUDA && nbnxn_cuda_is_kernel_ewald_analytical(fr->nbv->cu_nbv)))
+ {
+ enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_EWALD;
+ }
+ else
+ {
+ enr_nbnxn_kernel_ljc = eNR_NBNXN_LJ_TAB;
+ }
+ enr_nbnxn_kernel_lj = eNR_NBNXN_LJ;
+ if (flags & GMX_FORCE_ENERGY)
+ {
+ /* In eNR_??? the nbnxn F+E kernels are always the F kernel + 1 */
+ enr_nbnxn_kernel_ljc += 1;
+ enr_nbnxn_kernel_lj += 1;
+ }
+
+ inc_nrnb(nrnb, enr_nbnxn_kernel_ljc,
+ nbvg->nbl_lists.natpair_ljq);
+ inc_nrnb(nrnb, enr_nbnxn_kernel_lj,
+ nbvg->nbl_lists.natpair_lj);
+ inc_nrnb(nrnb, enr_nbnxn_kernel_ljc-eNR_NBNXN_LJ_RF+eNR_NBNXN_RF,
+ nbvg->nbl_lists.natpair_q);
+}
+
+void do_force_cutsVERLET(FILE *fplog, t_commrec *cr,
+ t_inputrec *inputrec,
+ gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_localtop_t *top,
+ gmx_groups_t gmx_unused *groups,
+ matrix box, rvec x[], history_t *hist,
+ rvec f[],
+ tensor vir_force,
+ t_mdatoms *mdatoms,
+ gmx_enerdata_t *enerd, t_fcdata *fcd,
+ real *lambda, t_graph *graph,
+ t_forcerec *fr, interaction_const_t *ic,
+ gmx_vsite_t *vsite, rvec mu_tot,
+ double t, FILE *field, gmx_edsam_t ed,
+ gmx_bool bBornRadii,
+ int flags)
+{
+ int cg0, cg1, i, j;
+ int start, homenr;
+ int nb_kernel_type;
+ double mu[2*DIM];
+ gmx_bool bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
+ gmx_bool bDoLongRange, bDoForces, bSepLRF, bUseGPU, bUseOrEmulGPU;
+ gmx_bool bDiffKernels = FALSE;
+ matrix boxs;
+ rvec vzero, box_diag;
+ real e, v, dvdl;
+ float cycles_pme, cycles_force, cycles_wait_gpu;
+ nonbonded_verlet_t *nbv;
+
+ cycles_force = 0;
+ cycles_wait_gpu = 0;
+ nbv = fr->nbv;
+ nb_kernel_type = fr->nbv->grp[0].kernel_type;
+
+ start = mdatoms->start;
+ homenr = mdatoms->homenr;
+
+ bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
+
+ clear_mat(vir_force);
+
+ cg0 = 0;
+ if (DOMAINDECOMP(cr))
+ {
+ cg1 = cr->dd->ncg_tot;
+ }
+ else
+ {
+ cg1 = top->cgs.nr;
+ }
+ if (fr->n_tpi > 0)
+ {
+ cg1--;
+ }
+
+ bStateChanged = (flags & GMX_FORCE_STATECHANGED);
+ bNS = (flags & GMX_FORCE_NS) && (fr->bAllvsAll == FALSE);
+ bFillGrid = (bNS && bStateChanged);
+ bCalcCGCM = (bFillGrid && !DOMAINDECOMP(cr));
+ bDoLongRange = (fr->bTwinRange && bNS && (flags & GMX_FORCE_DO_LR));
+ bDoForces = (flags & GMX_FORCE_FORCES);
+ bSepLRF = (bDoLongRange && bDoForces && (flags & GMX_FORCE_SEPLRF));
+ bUseGPU = fr->nbv->bUseGPU;
+ bUseOrEmulGPU = bUseGPU || (nbv->grp[0].kernel_type == nbnxnk8x8x8_PlainC);
+
+ if (bStateChanged)
+ {
+ update_forcerec(fr, box);
+
+ if (NEED_MUTOT(*inputrec))
+ {
+ /* Calculate total (local) dipole moment in a temporary common array.
+ * This makes it possible to sum them over nodes faster.
+ */
+ calc_mu(start, homenr,
+ x, mdatoms->chargeA, mdatoms->chargeB, mdatoms->nChargePerturbed,
+ mu, mu+DIM);
+ }
+ }
+
+ if (fr->ePBC != epbcNONE)
+ {
+ /* Compute shift vectors every step,
+ * because of pressure coupling or box deformation!
+ */
+ if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
+ {
+ calc_shifts(box, fr->shift_vec);
+ }
+
+ if (bCalcCGCM)
+ {
+ put_atoms_in_box_omp(fr->ePBC, box, homenr, x);
+ inc_nrnb(nrnb, eNR_SHIFTX, homenr);
+ }
+ else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph)
+ {
+ unshift_self(graph, box, x);
+ }
+ }
+
+ nbnxn_atomdata_copy_shiftvec(flags & GMX_FORCE_DYNAMICBOX,
+ fr->shift_vec, nbv->grp[0].nbat);
+
+#ifdef GMX_MPI
+ if (!(cr->duty & DUTY_PME))
+ {
+ /* Send particle coordinates to the pme nodes.
+ * Since this is only implemented for domain decomposition
+ * and domain decomposition does not use the graph,
+ * we do not need to worry about shifting.
+ */
+
+ int pme_flags = 0;
+
+ wallcycle_start(wcycle, ewcPP_PMESENDX);
+
+ bBS = (inputrec->nwall == 2);
+ if (bBS)
+ {
+ copy_mat(box, boxs);
+ svmul(inputrec->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
+ }
+
+ if (EEL_PME(fr->eeltype))
+ {
+ pme_flags |= GMX_PME_DO_COULOMB;
+ }
+
+ if (EVDW_PME(fr->vdwtype))
+ {
+ pme_flags |= GMX_PME_DO_LJ;
+ if (fr->ljpme_combination_rule == eljpmeLB)
+ {
+ pme_flags |= GMX_PME_LJ_LB;
+ }
+ }
+
+ gmx_pme_send_coordinates(cr, bBS ? boxs : box, x,
+ mdatoms->nChargePerturbed, mdatoms->nTypePerturbed, lambda[efptCOUL], lambda[efptVDW],
+ (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)),
+ pme_flags, step);
+
+ wallcycle_stop(wcycle, ewcPP_PMESENDX);
+ }
+#endif /* GMX_MPI */
+
+ /* do gridding for pair search */
+ if (bNS)
+ {
+ if (graph && bStateChanged)
+ {
+ /* Calculate intramolecular shift vectors to make molecules whole */
+ mk_mshift(fplog, graph, fr->ePBC, box, x);
+ }
+
+ clear_rvec(vzero);
+ box_diag[XX] = box[XX][XX];
+ box_diag[YY] = box[YY][YY];
+ box_diag[ZZ] = box[ZZ][ZZ];
+
+ wallcycle_start(wcycle, ewcNS);
+ if (!fr->bDomDec)
+ {
+ wallcycle_sub_start(wcycle, ewcsNBS_GRID_LOCAL);
+ nbnxn_put_on_grid(nbv->nbs, fr->ePBC, box,
+ 0, vzero, box_diag,
+ 0, mdatoms->homenr, -1, fr->cginfo, x,
+ 0, NULL,
+ nbv->grp[eintLocal].kernel_type,
+ nbv->grp[eintLocal].nbat);
+ wallcycle_sub_stop(wcycle, ewcsNBS_GRID_LOCAL);
+ }
+ else
+ {
+ wallcycle_sub_start(wcycle, ewcsNBS_GRID_NONLOCAL);
+ nbnxn_put_on_grid_nonlocal(nbv->nbs, domdec_zones(cr->dd),
+ fr->cginfo, x,
+ nbv->grp[eintNonlocal].kernel_type,
+ nbv->grp[eintNonlocal].nbat);
+ wallcycle_sub_stop(wcycle, ewcsNBS_GRID_NONLOCAL);
+ }
+
+ if (nbv->ngrp == 1 ||
+ nbv->grp[eintNonlocal].nbat == nbv->grp[eintLocal].nbat)
+ {
+ nbnxn_atomdata_set(nbv->grp[eintLocal].nbat, eatAll,
+ nbv->nbs, mdatoms, fr->cginfo);
+ }
+ else
+ {
+ nbnxn_atomdata_set(nbv->grp[eintLocal].nbat, eatLocal,
+ nbv->nbs, mdatoms, fr->cginfo);
+ nbnxn_atomdata_set(nbv->grp[eintNonlocal].nbat, eatAll,
+ nbv->nbs, mdatoms, fr->cginfo);
+ }
+ wallcycle_stop(wcycle, ewcNS);
+ }
+
+ /* initialize the GPU atom data and copy shift vector */
+ if (bUseGPU)
+ {
+ if (bNS)
+ {
+ wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
+ nbnxn_cuda_init_atomdata(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
+ wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
+ }
+
+ wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
+ nbnxn_cuda_upload_shiftvec(nbv->cu_nbv, nbv->grp[eintLocal].nbat);
+ wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
+ }
+
+ /* do local pair search */
+ if (bNS)
+ {
+ wallcycle_start_nocount(wcycle, ewcNS);
+ wallcycle_sub_start(wcycle, ewcsNBS_SEARCH_LOCAL);
+ nbnxn_make_pairlist(nbv->nbs, nbv->grp[eintLocal].nbat,
+ &top->excls,
+ ic->rlist,
+ nbv->min_ci_balanced,
+ &nbv->grp[eintLocal].nbl_lists,
+ eintLocal,
+ nbv->grp[eintLocal].kernel_type,
+ nrnb);
+ wallcycle_sub_stop(wcycle, ewcsNBS_SEARCH_LOCAL);
+
+ if (bUseGPU)
+ {
+ /* initialize local pair-list on the GPU */
+ nbnxn_cuda_init_pairlist(nbv->cu_nbv,
+ nbv->grp[eintLocal].nbl_lists.nbl[0],
+ eintLocal);
+ }
+ wallcycle_stop(wcycle, ewcNS);
+ }
+ else
+ {
+ wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
+ wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
+ nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatLocal, FALSE, x,
+ nbv->grp[eintLocal].nbat);
+ wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
+ wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
+ }
+
+ if (bUseGPU)
+ {
+ wallcycle_start(wcycle, ewcLAUNCH_GPU_NB);
+ /* launch local nonbonded F on GPU */
+ do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFNo,
+ nrnb, wcycle);
+ wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
+ }
+
+ /* Communicate coordinates and sum dipole if necessary +
+ do non-local pair search */
+ if (DOMAINDECOMP(cr))
+ {
+ bDiffKernels = (nbv->grp[eintNonlocal].kernel_type !=
+ nbv->grp[eintLocal].kernel_type);
+
+ if (bDiffKernels)
+ {
+ /* With GPU+CPU non-bonded calculations we need to copy
+ * the local coordinates to the non-local nbat struct
+ * (in CPU format) as the non-local kernel call also
+ * calculates the local - non-local interactions.
+ */
+ wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
+ wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
+ nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatLocal, TRUE, x,
+ nbv->grp[eintNonlocal].nbat);
+ wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
+ wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
+ }
+
+ if (bNS)
+ {
+ wallcycle_start_nocount(wcycle, ewcNS);
+ wallcycle_sub_start(wcycle, ewcsNBS_SEARCH_NONLOCAL);
+
+ if (bDiffKernels)
+ {
+ nbnxn_grid_add_simple(nbv->nbs, nbv->grp[eintNonlocal].nbat);
+ }
+
+ nbnxn_make_pairlist(nbv->nbs, nbv->grp[eintNonlocal].nbat,
+ &top->excls,
+ ic->rlist,
+ nbv->min_ci_balanced,
+ &nbv->grp[eintNonlocal].nbl_lists,
+ eintNonlocal,
+ nbv->grp[eintNonlocal].kernel_type,
+ nrnb);
+
+ wallcycle_sub_stop(wcycle, ewcsNBS_SEARCH_NONLOCAL);
+
+ if (nbv->grp[eintNonlocal].kernel_type == nbnxnk8x8x8_CUDA)
+ {
+ /* initialize non-local pair-list on the GPU */
+ nbnxn_cuda_init_pairlist(nbv->cu_nbv,
+ nbv->grp[eintNonlocal].nbl_lists.nbl[0],
+ eintNonlocal);
+ }
+ wallcycle_stop(wcycle, ewcNS);
+ }
+ else
+ {
+ wallcycle_start(wcycle, ewcMOVEX);
+ dd_move_x(cr->dd, box, x);
+
+ /* When we don't need the total dipole we sum it in global_stat */
+ if (bStateChanged && NEED_MUTOT(*inputrec))
+ {
+ gmx_sumd(2*DIM, mu, cr);
+ }
+ wallcycle_stop(wcycle, ewcMOVEX);
+
+ wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
+ wallcycle_sub_start(wcycle, ewcsNB_X_BUF_OPS);
+ nbnxn_atomdata_copy_x_to_nbat_x(nbv->nbs, eatNonlocal, FALSE, x,
+ nbv->grp[eintNonlocal].nbat);
+ wallcycle_sub_stop(wcycle, ewcsNB_X_BUF_OPS);
+ cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
+ }
+
+ if (bUseGPU && !bDiffKernels)
+ {
+ wallcycle_start(wcycle, ewcLAUNCH_GPU_NB);
+ /* launch non-local nonbonded F on GPU */
+ do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFNo,
+ nrnb, wcycle);
+ cycles_force += wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
+ }
+ }
+
+ if (bUseGPU)
+ {
+ /* launch D2H copy-back F */
+ wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
+ if (DOMAINDECOMP(cr) && !bDiffKernels)
+ {
+ nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintNonlocal].nbat,
+ flags, eatNonlocal);
+ }
+ nbnxn_cuda_launch_cpyback(nbv->cu_nbv, nbv->grp[eintLocal].nbat,
+ flags, eatLocal);
+ cycles_force += wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
+ }
+
+ if (bStateChanged && NEED_MUTOT(*inputrec))
+ {
+ if (PAR(cr))
+ {
+ gmx_sumd(2*DIM, mu, cr);
+ }
+
+ for (i = 0; i < 2; i++)
+ {
+ for (j = 0; j < DIM; j++)
+ {
+ fr->mu_tot[i][j] = mu[i*DIM + j];
+ }
+ }
+ }
+ if (fr->efep == efepNO)
+ {
+ copy_rvec(fr->mu_tot[0], mu_tot);
+ }
+ else
+ {
+ for (j = 0; j < DIM; j++)
+ {
+ mu_tot[j] =
+ (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] +
+ lambda[efptCOUL]*fr->mu_tot[1][j];
+ }
+ }
+
+ /* Reset energies */
+ reset_enerdata(fr, bNS, enerd, MASTER(cr));
+ clear_rvecs(SHIFTS, fr->fshift);
+
+ if (DOMAINDECOMP(cr))
+ {
+ if (!(cr->duty & DUTY_PME))
+ {
+ wallcycle_start(wcycle, ewcPPDURINGPME);
+ dd_force_flop_start(cr->dd, nrnb);
+ }
+ }
+
+ if (inputrec->bRot)
+ {
+ /* Enforced rotation has its own cycle counter that starts after the collective
+ * coordinates have been communicated. It is added to ddCyclF to allow
+ * for proper load-balancing */
+ wallcycle_start(wcycle, ewcROT);
+ do_rotation(cr, inputrec, box, x, t, step, wcycle, bNS);
+ wallcycle_stop(wcycle, ewcROT);
+ }
+
+ /* Start the force cycle counter.
+ * This counter is stopped in do_forcelow_level.
+ * No parallel communication should occur while this counter is running,
+ * since that will interfere with the dynamic load balancing.
+ */
+ wallcycle_start(wcycle, ewcFORCE);
+ if (bDoForces)
+ {
+ /* Reset forces for which the virial is calculated separately:
+ * PME/Ewald forces if necessary */
+ if (fr->bF_NoVirSum)
+ {
+ if (flags & GMX_FORCE_VIRIAL)
+ {
+ fr->f_novirsum = fr->f_novirsum_alloc;
+ if (fr->bDomDec)
+ {
+ clear_rvecs(fr->f_novirsum_n, fr->f_novirsum);
+ }
+ else
+ {
+ clear_rvecs(homenr, fr->f_novirsum+start);
+ }
+ }
+ else
+ {
+ /* We are not calculating the pressure so we do not need
+ * a separate array for forces that do not contribute
+ * to the pressure.
+ */
+ fr->f_novirsum = f;
+ }
+ }
+
+ /* Clear the short- and long-range forces */
+ clear_rvecs(fr->natoms_force_constr, f);
+ if (bSepLRF && do_per_step(step, inputrec->nstcalclr))
+ {
+ clear_rvecs(fr->natoms_force_constr, fr->f_twin);
+ }
+
+ clear_rvec(fr->vir_diag_posres);
+ }
+
+ if (inputrec->ePull == epullCONSTRAINT)
+ {
+ clear_pull_forces(inputrec->pull);
+ }
+
+ /* We calculate the non-bonded forces, when done on the CPU, here.
+ * We do this before calling do_force_lowlevel, as in there bondeds
+ * forces are calculated before PME, which does communication.
+ * With this order, non-bonded and bonded force calculation imbalance
+ * can be balanced out by the domain decomposition load balancing.
+ */
+
+ if (!bUseOrEmulGPU)
+ {
+ /* Maybe we should move this into do_force_lowlevel */
+ do_nb_verlet(fr, ic, enerd, flags, eintLocal, enbvClearFYes,
+ nrnb, wcycle);
+ }
+
+ if (!bUseOrEmulGPU || bDiffKernels)
+ {
+ int aloc;
+
+ if (DOMAINDECOMP(cr))
+ {
+ do_nb_verlet(fr, ic, enerd, flags, eintNonlocal,
+ bDiffKernels ? enbvClearFYes : enbvClearFNo,
+ nrnb, wcycle);
+ }
+
+ if (!bUseOrEmulGPU)
+ {
+ aloc = eintLocal;
+ }
+ else
+ {
+ aloc = eintNonlocal;
+ }
+
+ /* Add all the non-bonded force to the normal force array.
+ * This can be split into a local a non-local part when overlapping
+ * communication with calculation with domain decomposition.
+ */
+ cycles_force += wallcycle_stop(wcycle, ewcFORCE);
+ wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
+ wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
+ nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatAll, nbv->grp[aloc].nbat, f);
+ wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
+ cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
+ wallcycle_start_nocount(wcycle, ewcFORCE);
+
+ /* if there are multiple fshift output buffers reduce them */
+ if ((flags & GMX_FORCE_VIRIAL) &&
+ nbv->grp[aloc].nbl_lists.nnbl > 1)
+ {
+ nbnxn_atomdata_add_nbat_fshift_to_fshift(nbv->grp[aloc].nbat,
+ fr->fshift);
+ }
+ }
+
+ /* update QMMMrec, if necessary */
+ if (fr->bQMMM)
+ {
+ update_QMMMrec(cr, fr, x, mdatoms, box, top);
+ }
+
+ if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
+ {
+ posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
+ enerd, lambda, fr);
+ }
+
+ if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_FBPOSRES].nr > 0)
+ {
+ fbposres_wrapper(inputrec, nrnb, top, box, x, enerd, fr);
+ }
+
+ /* Compute the bonded and non-bonded energies and optionally forces */
+ do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
+ cr, nrnb, wcycle, mdatoms,
+ x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, top, fr->born,
+ &(top->atomtypes), bBornRadii, box,
+ inputrec->fepvals, lambda, graph, &(top->excls), fr->mu_tot,
+ flags, &cycles_pme);
+
+ if (bSepLRF)
+ {
+ if (do_per_step(step, inputrec->nstcalclr))
+ {
+ /* Add the long range forces to the short range forces */
+ for (i = 0; i < fr->natoms_force_constr; i++)
+ {
+ rvec_add(fr->f_twin[i], f[i], f[i]);
+ }
+ }
+ }
+
+ cycles_force += wallcycle_stop(wcycle, ewcFORCE);
+
+ if (ed)
+ {
+ do_flood(cr, inputrec, x, f, ed, box, step, bNS);
+ }
+
+ if (bUseOrEmulGPU && !bDiffKernels)
+ {
+ /* wait for non-local forces (or calculate in emulation mode) */
+ if (DOMAINDECOMP(cr))
+ {
+ if (bUseGPU)
+ {
+ float cycles_tmp;
+
+ wallcycle_start(wcycle, ewcWAIT_GPU_NB_NL);
+ nbnxn_cuda_wait_gpu(nbv->cu_nbv,
+ nbv->grp[eintNonlocal].nbat,
+ flags, eatNonlocal,
+ enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
+ fr->fshift);
+ cycles_tmp = wallcycle_stop(wcycle, ewcWAIT_GPU_NB_NL);
+ cycles_wait_gpu += cycles_tmp;
+ cycles_force += cycles_tmp;
+ }
+ else
+ {
+ wallcycle_start_nocount(wcycle, ewcFORCE);
+ do_nb_verlet(fr, ic, enerd, flags, eintNonlocal, enbvClearFYes,
+ nrnb, wcycle);
+ cycles_force += wallcycle_stop(wcycle, ewcFORCE);
+ }
+ wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
+ wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
+ /* skip the reduction if there was no non-local work to do */
+ if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
+ {
+ nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatNonlocal,
+ nbv->grp[eintNonlocal].nbat, f);
+ }
+ wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
+ cycles_force += wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
+ }
+ }
+
+ if (bDoForces)
+ {
+ /* Communicate the forces */
+ if (PAR(cr))
+ {
+ wallcycle_start(wcycle, ewcMOVEF);
+ if (DOMAINDECOMP(cr))
+ {
+ dd_move_f(cr->dd, f, fr->fshift);
+ /* Do we need to communicate the separate force array
+ * for terms that do not contribute to the single sum virial?
+ * Position restraints and electric fields do not introduce
+ * inter-cg forces, only full electrostatics methods do.
+ * When we do not calculate the virial, fr->f_novirsum = f,
+ * so we have already communicated these forces.
+ */
+ if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
+ (flags & GMX_FORCE_VIRIAL))
+ {
+ dd_move_f(cr->dd, fr->f_novirsum, NULL);
+ }
+ if (bSepLRF)
+ {
+ /* We should not update the shift forces here,
+ * since f_twin is already included in f.
+ */
+ dd_move_f(cr->dd, fr->f_twin, NULL);
+ }
+ }
+ wallcycle_stop(wcycle, ewcMOVEF);
+ }
+ }
+
+ if (bUseOrEmulGPU)
+ {
+ /* wait for local forces (or calculate in emulation mode) */
+ if (bUseGPU)
+ {
+ wallcycle_start(wcycle, ewcWAIT_GPU_NB_L);
+ nbnxn_cuda_wait_gpu(nbv->cu_nbv,
+ nbv->grp[eintLocal].nbat,
+ flags, eatLocal,
+ enerd->grpp.ener[egLJSR], enerd->grpp.ener[egCOULSR],
+ fr->fshift);
+ cycles_wait_gpu += wallcycle_stop(wcycle, ewcWAIT_GPU_NB_L);
+
+ /* now clear the GPU outputs while we finish the step on the CPU */
+
+ wallcycle_start_nocount(wcycle, ewcLAUNCH_GPU_NB);
+ nbnxn_cuda_clear_outputs(nbv->cu_nbv, flags);
+ wallcycle_stop(wcycle, ewcLAUNCH_GPU_NB);
+ }
+ else
+ {
+ wallcycle_start_nocount(wcycle, ewcFORCE);
+ do_nb_verlet(fr, ic, enerd, flags, eintLocal,
+ DOMAINDECOMP(cr) ? enbvClearFNo : enbvClearFYes,
+ nrnb, wcycle);
+ wallcycle_stop(wcycle, ewcFORCE);
+ }
+ wallcycle_start(wcycle, ewcNB_XF_BUF_OPS);
+ wallcycle_sub_start(wcycle, ewcsNB_F_BUF_OPS);
+ if (nbv->grp[eintLocal].nbl_lists.nbl[0]->nsci > 0)
+ {
+ /* skip the reduction if there was no non-local work to do */
+ nbnxn_atomdata_add_nbat_f_to_f(nbv->nbs, eatLocal,
+ nbv->grp[eintLocal].nbat, f);
+ }
+ wallcycle_sub_stop(wcycle, ewcsNB_F_BUF_OPS);
+ wallcycle_stop(wcycle, ewcNB_XF_BUF_OPS);
+ }
+
+ if (DOMAINDECOMP(cr))
+ {
+ dd_force_flop_stop(cr->dd, nrnb);
+ if (wcycle)
+ {
+ dd_cycles_add(cr->dd, cycles_force-cycles_pme, ddCyclF);
+ if (bUseGPU)
+ {
+ dd_cycles_add(cr->dd, cycles_wait_gpu, ddCyclWaitGPU);
+ }
+ }
+ }
+
+ if (bDoForces)
+ {
+ if (IR_ELEC_FIELD(*inputrec))
+ {
+ /* Compute forces due to electric field */
+ calc_f_el(MASTER(cr) ? field : NULL,
+ start, homenr, mdatoms->chargeA, fr->f_novirsum,
+ inputrec->ex, inputrec->et, t);
+ }
+
+ /* If we have NoVirSum forces, but we do not calculate the virial,
+ * we sum fr->f_novirum=f later.
+ */
+ if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
+ {
+ wallcycle_start(wcycle, ewcVSITESPREAD);
+ spread_vsite_f(vsite, x, f, fr->fshift, FALSE, NULL, nrnb,
+ &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
+ wallcycle_stop(wcycle, ewcVSITESPREAD);
+
+ if (bSepLRF)
+ {
+ wallcycle_start(wcycle, ewcVSITESPREAD);
+ spread_vsite_f(vsite, x, fr->f_twin, NULL, FALSE, NULL,
+ nrnb,
+ &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
+ wallcycle_stop(wcycle, ewcVSITESPREAD);
+ }
+ }
+
+ if (flags & GMX_FORCE_VIRIAL)
+ {
+ /* Calculation of the virial must be done after vsites! */
+ calc_virial(mdatoms->start, mdatoms->homenr, x, f,
+ vir_force, graph, box, nrnb, fr, inputrec->ePBC);
+ }
+ }
+
+ if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
+ {
+ pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
+ f, vir_force, mdatoms, enerd, lambda, t);
+ }
+
+ /* Add the forces from enforced rotation potentials (if any) */
+ if (inputrec->bRot)
+ {
+ wallcycle_start(wcycle, ewcROTadd);
+ enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr, step, t);
+ wallcycle_stop(wcycle, ewcROTadd);
+ }
+
+ if (PAR(cr) && !(cr->duty & DUTY_PME))
+ {
+ /* In case of node-splitting, the PP nodes receive the long-range
+ * forces, virial and energy from the PME nodes here.
+ */
+ pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
+ }
+
+ if (bDoForces)
+ {
+ post_process_forces(cr, step, nrnb, wcycle,
+ top, box, x, f, vir_force, mdatoms, graph, fr, vsite,
+ flags);
+ }
+
+ /* Sum the potential energy terms from group contributions */
+ sum_epot(&(enerd->grpp), enerd->term);
+}
+
+void do_force_cutsGROUP(FILE *fplog, t_commrec *cr,
+ t_inputrec *inputrec,
+ gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_localtop_t *top,
+ gmx_groups_t *groups,
+ matrix box, rvec x[], history_t *hist,
+ rvec f[],
+ tensor vir_force,
+ t_mdatoms *mdatoms,
+ gmx_enerdata_t *enerd, t_fcdata *fcd,
+ real *lambda, t_graph *graph,
+ t_forcerec *fr, gmx_vsite_t *vsite, rvec mu_tot,
+ double t, FILE *field, gmx_edsam_t ed,
+ gmx_bool bBornRadii,
+ int flags)
+{
+ int cg0, cg1, i, j;
+ int start, homenr;
+ double mu[2*DIM];
+ gmx_bool bSepDVDL, bStateChanged, bNS, bFillGrid, bCalcCGCM, bBS;
+ gmx_bool bDoLongRangeNS, bDoForces, bDoPotential, bSepLRF;
+ gmx_bool bDoAdressWF;
+ matrix boxs;
+ rvec vzero, box_diag;
+ real e, v, dvdlambda[efptNR];
+ t_pbc pbc;
+ float cycles_pme, cycles_force;
+
+ start = mdatoms->start;
+ homenr = mdatoms->homenr;
+
+ bSepDVDL = (fr->bSepDVDL && do_per_step(step, inputrec->nstlog));
+
+ clear_mat(vir_force);
+
+ if (PARTDECOMP(cr))
+ {
+ pd_cg_range(cr, &cg0, &cg1);
+ }
+ else
+ {
+ cg0 = 0;
+ if (DOMAINDECOMP(cr))
+ {
+ cg1 = cr->dd->ncg_tot;
+ }
+ else
+ {
+ cg1 = top->cgs.nr;
+ }
+ if (fr->n_tpi > 0)
+ {
+ cg1--;
+ }
+ }
+
+ bStateChanged = (flags & GMX_FORCE_STATECHANGED);
+ bNS = (flags & GMX_FORCE_NS) && (fr->bAllvsAll == FALSE);
+ /* Should we update the long-range neighborlists at this step? */
+ bDoLongRangeNS = fr->bTwinRange && bNS;
+ /* Should we perform the long-range nonbonded evaluation inside the neighborsearching? */
+ bFillGrid = (bNS && bStateChanged);
+ bCalcCGCM = (bFillGrid && !DOMAINDECOMP(cr));
+ bDoForces = (flags & GMX_FORCE_FORCES);
+ bDoPotential = (flags & GMX_FORCE_ENERGY);
+ bSepLRF = ((inputrec->nstcalclr > 1) && bDoForces &&
+ (flags & GMX_FORCE_SEPLRF) && (flags & GMX_FORCE_DO_LR));
+
+ /* should probably move this to the forcerec since it doesn't change */
+ bDoAdressWF = ((fr->adress_type != eAdressOff));
+
+ if (bStateChanged)
+ {
+ update_forcerec(fr, box);
+
+ if (NEED_MUTOT(*inputrec))
+ {
+ /* Calculate total (local) dipole moment in a temporary common array.
+ * This makes it possible to sum them over nodes faster.
+ */
+ calc_mu(start, homenr,
+ x, mdatoms->chargeA, mdatoms->chargeB, mdatoms->nChargePerturbed,
+ mu, mu+DIM);
+ }
+ }
+
+ if (fr->ePBC != epbcNONE)
+ {
+ /* Compute shift vectors every step,
+ * because of pressure coupling or box deformation!
+ */
+ if ((flags & GMX_FORCE_DYNAMICBOX) && bStateChanged)
+ {
+ calc_shifts(box, fr->shift_vec);
+ }
+
+ if (bCalcCGCM)
+ {
+ put_charge_groups_in_box(fplog, cg0, cg1, fr->ePBC, box,
+ &(top->cgs), x, fr->cg_cm);
+ inc_nrnb(nrnb, eNR_CGCM, homenr);
+ inc_nrnb(nrnb, eNR_RESETX, cg1-cg0);
+ }
+ else if (EI_ENERGY_MINIMIZATION(inputrec->eI) && graph)
+ {
+ unshift_self(graph, box, x);
+ }
+ }
+ else if (bCalcCGCM)
+ {
+ calc_cgcm(fplog, cg0, cg1, &(top->cgs), x, fr->cg_cm);
+ inc_nrnb(nrnb, eNR_CGCM, homenr);
+ }
+
+ if (bCalcCGCM)
+ {
+ if (PAR(cr))
+ {
+ move_cgcm(fplog, cr, fr->cg_cm);
+ }
+ if (gmx_debug_at)
+ {
+ pr_rvecs(debug, 0, "cgcm", fr->cg_cm, top->cgs.nr);
+ }
+ }
+
+#ifdef GMX_MPI
+ if (!(cr->duty & DUTY_PME))
+ {
+ /* Send particle coordinates to the pme nodes.
+ * Since this is only implemented for domain decomposition
+ * and domain decomposition does not use the graph,
+ * we do not need to worry about shifting.
+ */
+
+ int pme_flags = 0;
+
+ wallcycle_start(wcycle, ewcPP_PMESENDX);
+
+ bBS = (inputrec->nwall == 2);
+ if (bBS)
+ {
+ copy_mat(box, boxs);
+ svmul(inputrec->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
+ }
+
+ if (EEL_PME(fr->eeltype))
+ {
+ pme_flags |= GMX_PME_DO_COULOMB;
+ }
+
+ if (EVDW_PME(fr->vdwtype))
+ {
+ pme_flags |= GMX_PME_DO_LJ;
+ if (fr->ljpme_combination_rule == eljpmeLB)
+ {
+ pme_flags |= GMX_PME_LJ_LB;
+ }
+ }
+
+ gmx_pme_send_coordinates(cr, bBS ? boxs : box, x,
+ mdatoms->nChargePerturbed, mdatoms->nTypePerturbed, lambda[efptCOUL], lambda[efptVDW],
+ (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY)),
+ pme_flags, step);
+
+ wallcycle_stop(wcycle, ewcPP_PMESENDX);
+ }
+#endif /* GMX_MPI */
+
+ /* Communicate coordinates and sum dipole if necessary */
+ if (PAR(cr))
+ {
+ wallcycle_start(wcycle, ewcMOVEX);
+ if (DOMAINDECOMP(cr))
+ {
+ dd_move_x(cr->dd, box, x);
+ }
+ else
+ {
+ move_x(cr, x, nrnb);
+ }
+ wallcycle_stop(wcycle, ewcMOVEX);
+ }
+
+ /* update adress weight beforehand */
+ if (bStateChanged && bDoAdressWF)
+ {
+ /* need pbc for adress weight calculation with pbc_dx */
+ set_pbc(&pbc, inputrec->ePBC, box);
+ if (fr->adress_site == eAdressSITEcog)
+ {
+ update_adress_weights_cog(top->idef.iparams, top->idef.il, x, fr, mdatoms,
+ inputrec->ePBC == epbcNONE ? NULL : &pbc);
+ }
+ else if (fr->adress_site == eAdressSITEcom)
+ {
+ update_adress_weights_com(fplog, cg0, cg1, &(top->cgs), x, fr, mdatoms,
+ inputrec->ePBC == epbcNONE ? NULL : &pbc);
+ }
+ else if (fr->adress_site == eAdressSITEatomatom)
+ {
+ update_adress_weights_atom_per_atom(cg0, cg1, &(top->cgs), x, fr, mdatoms,
+ inputrec->ePBC == epbcNONE ? NULL : &pbc);
+ }
+ else
+ {
+ update_adress_weights_atom(cg0, cg1, &(top->cgs), x, fr, mdatoms,
+ inputrec->ePBC == epbcNONE ? NULL : &pbc);
+ }
+ }
+
+ if (NEED_MUTOT(*inputrec))
+ {
+
+ if (bStateChanged)
+ {
+ if (PAR(cr))
+ {
+ gmx_sumd(2*DIM, mu, cr);
+ }
+ for (i = 0; i < 2; i++)
+ {
+ for (j = 0; j < DIM; j++)
+ {
+ fr->mu_tot[i][j] = mu[i*DIM + j];
+ }
+ }
+ }
+ if (fr->efep == efepNO)
+ {
+ copy_rvec(fr->mu_tot[0], mu_tot);
+ }
+ else
+ {
+ for (j = 0; j < DIM; j++)
+ {
+ mu_tot[j] =
+ (1.0 - lambda[efptCOUL])*fr->mu_tot[0][j] + lambda[efptCOUL]*fr->mu_tot[1][j];
+ }
+ }
+ }
+
+ /* Reset energies */
+ reset_enerdata(fr, bNS, enerd, MASTER(cr));
+ clear_rvecs(SHIFTS, fr->fshift);
+
+ if (bNS)
+ {
+ wallcycle_start(wcycle, ewcNS);
+
+ if (graph && bStateChanged)
+ {
+ /* Calculate intramolecular shift vectors to make molecules whole */
+ mk_mshift(fplog, graph, fr->ePBC, box, x);
+ }
+
+ /* Do the actual neighbour searching */
+ ns(fplog, fr, box,
+ groups, top, mdatoms,
+ cr, nrnb, bFillGrid,
+ bDoLongRangeNS);
+
+ wallcycle_stop(wcycle, ewcNS);
+ }
+
+ if (inputrec->implicit_solvent && bNS)
+ {
+ make_gb_nblist(cr, inputrec->gb_algorithm,
+ x, box, fr, &top->idef, graph, fr->born);
+ }
+
+ if (DOMAINDECOMP(cr))
+ {
+ if (!(cr->duty & DUTY_PME))
+ {
+ wallcycle_start(wcycle, ewcPPDURINGPME);
+ dd_force_flop_start(cr->dd, nrnb);
+ }
+ }
+
+ if (inputrec->bRot)
+ {
+ /* Enforced rotation has its own cycle counter that starts after the collective
+ * coordinates have been communicated. It is added to ddCyclF to allow
+ * for proper load-balancing */
+ wallcycle_start(wcycle, ewcROT);
+ do_rotation(cr, inputrec, box, x, t, step, wcycle, bNS);
+ wallcycle_stop(wcycle, ewcROT);
+ }
+
+ /* Start the force cycle counter.
+ * This counter is stopped in do_forcelow_level.
+ * No parallel communication should occur while this counter is running,
+ * since that will interfere with the dynamic load balancing.
+ */
+ wallcycle_start(wcycle, ewcFORCE);
+
+ if (bDoForces)
+ {
+ /* Reset forces for which the virial is calculated separately:
+ * PME/Ewald forces if necessary */
+ if (fr->bF_NoVirSum)
+ {
+ if (flags & GMX_FORCE_VIRIAL)
+ {
+ fr->f_novirsum = fr->f_novirsum_alloc;
+ if (fr->bDomDec)
+ {
+ clear_rvecs(fr->f_novirsum_n, fr->f_novirsum);
+ }
+ else
+ {
+ clear_rvecs(homenr, fr->f_novirsum+start);
+ }
+ }
+ else
+ {
+ /* We are not calculating the pressure so we do not need
+ * a separate array for forces that do not contribute
+ * to the pressure.
+ */
+ fr->f_novirsum = f;
+ }
+ }
+
+ /* Clear the short- and long-range forces */
+ clear_rvecs(fr->natoms_force_constr, f);
+ if (bSepLRF && do_per_step(step, inputrec->nstcalclr))
+ {
+ clear_rvecs(fr->natoms_force_constr, fr->f_twin);
+ }
+
+ clear_rvec(fr->vir_diag_posres);
+ }
+ if (inputrec->ePull == epullCONSTRAINT)
+ {
+ clear_pull_forces(inputrec->pull);
+ }
+
+ /* update QMMMrec, if necessary */
+ if (fr->bQMMM)
+ {
+ update_QMMMrec(cr, fr, x, mdatoms, box, top);
+ }
+
+ if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_POSRES].nr > 0)
+ {
+ posres_wrapper(fplog, flags, bSepDVDL, inputrec, nrnb, top, box, x,
+ enerd, lambda, fr);
+ }
+
+ if ((flags & GMX_FORCE_BONDED) && top->idef.il[F_FBPOSRES].nr > 0)
+ {
+ fbposres_wrapper(inputrec, nrnb, top, box, x, enerd, fr);
+ }
+
+ /* Compute the bonded and non-bonded energies and optionally forces */
+ do_force_lowlevel(fplog, step, fr, inputrec, &(top->idef),
+ cr, nrnb, wcycle, mdatoms,
+ x, hist, f, bSepLRF ? fr->f_twin : f, enerd, fcd, top, fr->born,
+ &(top->atomtypes), bBornRadii, box,
+ inputrec->fepvals, lambda,
+ graph, &(top->excls), fr->mu_tot,
+ flags,
+ &cycles_pme);
+
+ if (bSepLRF)
+ {
+ if (do_per_step(step, inputrec->nstcalclr))
+ {
+ /* Add the long range forces to the short range forces */
+ for (i = 0; i < fr->natoms_force_constr; i++)
+ {
+ rvec_add(fr->f_twin[i], f[i], f[i]);
+ }
+ }
+ }
+
+ cycles_force = wallcycle_stop(wcycle, ewcFORCE);
+
+ if (ed)
+ {
+ do_flood(cr, inputrec, x, f, ed, box, step, bNS);
+ }
+
+ if (DOMAINDECOMP(cr))
+ {
+ dd_force_flop_stop(cr->dd, nrnb);
+ if (wcycle)
+ {
+ dd_cycles_add(cr->dd, cycles_force-cycles_pme, ddCyclF);
+ }
+ }
+
+ if (bDoForces)
+ {
+ if (IR_ELEC_FIELD(*inputrec))
+ {
+ /* Compute forces due to electric field */
+ calc_f_el(MASTER(cr) ? field : NULL,
+ start, homenr, mdatoms->chargeA, fr->f_novirsum,
+ inputrec->ex, inputrec->et, t);
+ }
+
+ if (bDoAdressWF && fr->adress_icor == eAdressICThermoForce)
+ {
+ /* Compute thermodynamic force in hybrid AdResS region */
+ adress_thermo_force(start, homenr, &(top->cgs), x, fr->f_novirsum, fr, mdatoms,
+ inputrec->ePBC == epbcNONE ? NULL : &pbc);
+ }
+
+ /* Communicate the forces */
+ if (PAR(cr))
+ {
+ wallcycle_start(wcycle, ewcMOVEF);
+ if (DOMAINDECOMP(cr))
+ {
+ dd_move_f(cr->dd, f, fr->fshift);
+ /* Do we need to communicate the separate force array
+ * for terms that do not contribute to the single sum virial?
+ * Position restraints and electric fields do not introduce
+ * inter-cg forces, only full electrostatics methods do.
+ * When we do not calculate the virial, fr->f_novirsum = f,
+ * so we have already communicated these forces.
+ */
+ if (EEL_FULL(fr->eeltype) && cr->dd->n_intercg_excl &&
+ (flags & GMX_FORCE_VIRIAL))
+ {
+ dd_move_f(cr->dd, fr->f_novirsum, NULL);
+ }
+ if (bSepLRF)
+ {
+ /* We should not update the shift forces here,
+ * since f_twin is already included in f.
+ */
+ dd_move_f(cr->dd, fr->f_twin, NULL);
+ }
+ }
+ else
+ {
+ pd_move_f(cr, f, nrnb);
+ if (bSepLRF)
+ {
+ pd_move_f(cr, fr->f_twin, nrnb);
+ }
+ }
+ wallcycle_stop(wcycle, ewcMOVEF);
+ }
+
+ /* If we have NoVirSum forces, but we do not calculate the virial,
+ * we sum fr->f_novirum=f later.
+ */
+ if (vsite && !(fr->bF_NoVirSum && !(flags & GMX_FORCE_VIRIAL)))
+ {
+ wallcycle_start(wcycle, ewcVSITESPREAD);
+ spread_vsite_f(vsite, x, f, fr->fshift, FALSE, NULL, nrnb,
+ &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
+ wallcycle_stop(wcycle, ewcVSITESPREAD);
+
+ if (bSepLRF)
+ {
+ wallcycle_start(wcycle, ewcVSITESPREAD);
+ spread_vsite_f(vsite, x, fr->f_twin, NULL, FALSE, NULL,
+ nrnb,
+ &top->idef, fr->ePBC, fr->bMolPBC, graph, box, cr);
+ wallcycle_stop(wcycle, ewcVSITESPREAD);
+ }
+ }
+
+ if (flags & GMX_FORCE_VIRIAL)
+ {
+ /* Calculation of the virial must be done after vsites! */
+ calc_virial(mdatoms->start, mdatoms->homenr, x, f,
+ vir_force, graph, box, nrnb, fr, inputrec->ePBC);
+ }
+ }
+
+ if (inputrec->ePull == epullUMBRELLA || inputrec->ePull == epullCONST_F)
+ {
+ pull_potential_wrapper(fplog, bSepDVDL, cr, inputrec, box, x,
+ f, vir_force, mdatoms, enerd, lambda, t);
+ }
+
+ /* Add the forces from enforced rotation potentials (if any) */
+ if (inputrec->bRot)
+ {
+ wallcycle_start(wcycle, ewcROTadd);
+ enerd->term[F_COM_PULL] += add_rot_forces(inputrec->rot, f, cr, step, t);
+ wallcycle_stop(wcycle, ewcROTadd);
+ }
+
+ if (PAR(cr) && !(cr->duty & DUTY_PME))
+ {
+ /* In case of node-splitting, the PP nodes receive the long-range
+ * forces, virial and energy from the PME nodes here.
+ */
+ pme_receive_force_ener(fplog, bSepDVDL, cr, wcycle, enerd, fr);
+ }
+
+ if (bDoForces)
+ {
+ post_process_forces(cr, step, nrnb, wcycle,
+ top, box, x, f, vir_force, mdatoms, graph, fr, vsite,
+ flags);
+ }
+
+ /* Sum the potential energy terms from group contributions */
+ sum_epot(&(enerd->grpp), enerd->term);
+}
+
+void do_force(FILE *fplog, t_commrec *cr,
+ t_inputrec *inputrec,
+ gmx_int64_t step, t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_localtop_t *top,
+ gmx_groups_t *groups,
+ matrix box, rvec x[], history_t *hist,
+ rvec f[],
+ tensor vir_force,
+ t_mdatoms *mdatoms,
+ gmx_enerdata_t *enerd, t_fcdata *fcd,
+ real *lambda, t_graph *graph,
+ t_forcerec *fr,
+ gmx_vsite_t *vsite, rvec mu_tot,
+ double t, FILE *field, gmx_edsam_t ed,
+ gmx_bool bBornRadii,
+ int flags)
+{
+ /* modify force flag if not doing nonbonded */
+ if (!fr->bNonbonded)
+ {
+ flags &= ~GMX_FORCE_NONBONDED;
+ }
+
+ switch (inputrec->cutoff_scheme)
+ {
+ case ecutsVERLET:
+ do_force_cutsVERLET(fplog, cr, inputrec,
+ step, nrnb, wcycle,
+ top,
+ groups,
+ box, x, hist,
+ f, vir_force,
+ mdatoms,
+ enerd, fcd,
+ lambda, graph,
+ fr, fr->ic,
+ vsite, mu_tot,
+ t, field, ed,
+ bBornRadii,
+ flags);
+ break;
+ case ecutsGROUP:
+ do_force_cutsGROUP(fplog, cr, inputrec,
+ step, nrnb, wcycle,
+ top,
+ groups,
+ box, x, hist,
+ f, vir_force,
+ mdatoms,
+ enerd, fcd,
+ lambda, graph,
+ fr, vsite, mu_tot,
+ t, field, ed,
+ bBornRadii,
+ flags);
+ break;
+ default:
+ gmx_incons("Invalid cut-off scheme passed!");
+ }
+}
+
+
+void do_constrain_first(FILE *fplog, gmx_constr_t constr,
+ t_inputrec *ir, t_mdatoms *md,
+ t_state *state, t_commrec *cr, t_nrnb *nrnb,
+ t_forcerec *fr, gmx_localtop_t *top)
+{
+ int i, m, start, end;
+ gmx_int64_t step;
+ real dt = ir->delta_t;
+ real dvdl_dum;
+ rvec *savex;
+
+ snew(savex, state->natoms);
+
+ start = md->start;
+ end = md->homenr + start;
+
+ if (debug)
+ {
+ fprintf(debug, "vcm: start=%d, homenr=%d, end=%d\n",
+ start, md->homenr, end);
+ }
+ /* Do a first constrain to reset particles... */
+ step = ir->init_step;
+ if (fplog)
+ {
+ char buf[STEPSTRSIZE];
+ fprintf(fplog, "\nConstraining the starting coordinates (step %s)\n",
+ gmx_step_str(step, buf));
+ }
+ dvdl_dum = 0;
+
+ /* constrain the current position */
+ constrain(NULL, TRUE, FALSE, constr, &(top->idef),
+ ir, NULL, cr, step, 0, md,
+ state->x, state->x, NULL,
+ fr->bMolPBC, state->box,
+ state->lambda[efptBONDED], &dvdl_dum,
+ NULL, NULL, nrnb, econqCoord,
+ ir->epc == epcMTTK, state->veta, state->veta);
+ if (EI_VV(ir->eI))
+ {
+ /* constrain the inital velocity, and save it */
+ /* also may be useful if we need the ekin from the halfstep for velocity verlet */
+ /* might not yet treat veta correctly */
+ constrain(NULL, TRUE, FALSE, constr, &(top->idef),
+ ir, NULL, cr, step, 0, md,
+ state->x, state->v, state->v,
+ fr->bMolPBC, state->box,
+ state->lambda[efptBONDED], &dvdl_dum,
+ NULL, NULL, nrnb, econqVeloc,
+ ir->epc == epcMTTK, state->veta, state->veta);
+ }
+ /* constrain the inital velocities at t-dt/2 */
+ if (EI_STATE_VELOCITY(ir->eI) && ir->eI != eiVV)
+ {
+ for (i = start; (i < end); i++)
+ {
+ for (m = 0; (m < DIM); m++)
+ {
+ /* Reverse the velocity */
+ state->v[i][m] = -state->v[i][m];
+ /* Store the position at t-dt in buf */
+ savex[i][m] = state->x[i][m] + dt*state->v[i][m];
+ }
+ }
+ /* Shake the positions at t=-dt with the positions at t=0
+ * as reference coordinates.
+ */
+ if (fplog)
+ {
+ char buf[STEPSTRSIZE];
+ fprintf(fplog, "\nConstraining the coordinates at t0-dt (step %s)\n",
+ gmx_step_str(step, buf));
+ }
+ dvdl_dum = 0;
+ constrain(NULL, TRUE, FALSE, constr, &(top->idef),
+ ir, NULL, cr, step, -1, md,
+ state->x, savex, NULL,
+ fr->bMolPBC, state->box,
+ state->lambda[efptBONDED], &dvdl_dum,
+ state->v, NULL, nrnb, econqCoord,
+ ir->epc == epcMTTK, state->veta, state->veta);
+
+ for (i = start; i < end; i++)
+ {
+ for (m = 0; m < DIM; m++)
+ {
+ /* Re-reverse the velocities */
+ state->v[i][m] = -state->v[i][m];
+ }
+ }
+ }
+ sfree(savex);
+}
+
+
+static void
+integrate_table(real vdwtab[], real scale, int offstart, int rstart, int rend,
+ double *enerout, double *virout)
+{
+ double enersum, virsum;
+ double invscale, invscale2, invscale3;
+ double r, ea, eb, ec, pa, pb, pc, pd;
+ double y0, f, g, h;
+ int ri, offset, tabfactor;
+
+ invscale = 1.0/scale;
+ invscale2 = invscale*invscale;
+ invscale3 = invscale*invscale2;
+
+ /* Following summation derived from cubic spline definition,
+ * Numerical Recipies in C, second edition, p. 113-116. Exact for
+ * the cubic spline. We first calculate the negative of the
+ * energy from rvdw to rvdw_switch, assuming that g(r)=1, and then
+ * add the more standard, abrupt cutoff correction to that result,
+ * yielding the long-range correction for a switched function. We
+ * perform both the pressure and energy loops at the same time for
+ * simplicity, as the computational cost is low. */
+
+ if (offstart == 0)
+ {
+ /* Since the dispersion table has been scaled down a factor
+ * 6.0 and the repulsion a factor 12.0 to compensate for the
+ * c6/c12 parameters inside nbfp[] being scaled up (to save
+ * flops in kernels), we need to correct for this.
+ */
+ tabfactor = 6.0;
+ }
+ else
+ {
+ tabfactor = 12.0;
+ }
+
+ enersum = 0.0;
+ virsum = 0.0;
+ for (ri = rstart; ri < rend; ++ri)
+ {
+ r = ri*invscale;
+ ea = invscale3;
+ eb = 2.0*invscale2*r;
+ ec = invscale*r*r;
+
+ pa = invscale3;
+ pb = 3.0*invscale2*r;
+ pc = 3.0*invscale*r*r;
+ pd = r*r*r;
+
+ /* this "8" is from the packing in the vdwtab array - perhaps
+ should be defined? */
+
+ offset = 8*ri + offstart;
+ y0 = vdwtab[offset];
+ f = vdwtab[offset+1];
+ g = vdwtab[offset+2];
+ h = vdwtab[offset+3];
+
+ enersum += y0*(ea/3 + eb/2 + ec) + f*(ea/4 + eb/3 + ec/2) + g*(ea/5 + eb/4 + ec/3) + h*(ea/6 + eb/5 + ec/4);
+ virsum += f*(pa/4 + pb/3 + pc/2 + pd) + 2*g*(pa/5 + pb/4 + pc/3 + pd/2) + 3*h*(pa/6 + pb/5 + pc/4 + pd/3);
+ }
+ *enerout = 4.0*M_PI*enersum*tabfactor;
+ *virout = 4.0*M_PI*virsum*tabfactor;
+}
+
+void calc_enervirdiff(FILE *fplog, int eDispCorr, t_forcerec *fr)
+{
+ double eners[2], virs[2], enersum, virsum, y0, f, g, h;
+ double r0, r1, r, rc3, rc9, ea, eb, ec, pa, pb, pc, pd;
+ double invscale, invscale2, invscale3;
+ int ri0, ri1, ri, i, offstart, offset;
+ real scale, *vdwtab, tabfactor, tmp;
+
+ fr->enershiftsix = 0;
+ fr->enershifttwelve = 0;
+ fr->enerdiffsix = 0;
+ fr->enerdifftwelve = 0;
+ fr->virdiffsix = 0;
+ fr->virdifftwelve = 0;
+
+ if (eDispCorr != edispcNO)
+ {
+ for (i = 0; i < 2; i++)
+ {
+ eners[i] = 0;
+ virs[i] = 0;
+ }
+ if ((fr->vdwtype == evdwSWITCH) || (fr->vdwtype == evdwSHIFT))
+ {
+ if (fr->rvdw_switch == 0)
+ {
+ gmx_fatal(FARGS,
+ "With dispersion correction rvdw-switch can not be zero "
+ "for vdw-type = %s", evdw_names[fr->vdwtype]);
+ }
+
+ scale = fr->nblists[0].table_elec_vdw.scale;
+ vdwtab = fr->nblists[0].table_vdw.data;
+
+ /* Round the cut-offs to exact table values for precision */
+ ri0 = floor(fr->rvdw_switch*scale);
+ ri1 = ceil(fr->rvdw*scale);
+ r0 = ri0/scale;
+ r1 = ri1/scale;
+ rc3 = r0*r0*r0;
+ rc9 = rc3*rc3*rc3;
+
+ if (fr->vdwtype == evdwSHIFT)
+ {
+ /* Determine the constant energy shift below rvdw_switch.
+ * Table has a scale factor since we have scaled it down to compensate
+ * for scaling-up c6/c12 with the derivative factors to save flops in analytical kernels.
+ */
+ fr->enershiftsix = (real)(-1.0/(rc3*rc3)) - 6.0*vdwtab[8*ri0];
+ fr->enershifttwelve = (real)( 1.0/(rc9*rc3)) - 12.0*vdwtab[8*ri0 + 4];
+ }
+ /* Add the constant part from 0 to rvdw_switch.
+ * This integration from 0 to rvdw_switch overcounts the number
+ * of interactions by 1, as it also counts the self interaction.
+ * We will correct for this later.
+ */
+ eners[0] += 4.0*M_PI*fr->enershiftsix*rc3/3.0;
+ eners[1] += 4.0*M_PI*fr->enershifttwelve*rc3/3.0;
+ for (i = 0; i < 2; i++)
+ {
+ enersum = 0;
+ virsum = 0;
+ integrate_table(vdwtab, scale, (i == 0 ? 0 : 4), ri0, ri1, &enersum, &virsum);
+ eners[i] -= enersum;
+ virs[i] -= virsum;
+ }
+
+ /* now add the correction for rvdw_switch to infinity */
+ eners[0] += -4.0*M_PI/(3.0*rc3);
+ eners[1] += 4.0*M_PI/(9.0*rc9);
+ virs[0] += 8.0*M_PI/rc3;
+ virs[1] += -16.0*M_PI/(3.0*rc9);
+ }
+ else if (EVDW_PME(fr->vdwtype))
+ {
+ if (EVDW_SWITCHED(fr->vdwtype) && fr->rvdw_switch == 0)
+ {
+ gmx_fatal(FARGS,
+ "With dispersion correction rvdw-switch can not be zero "
+ "for vdw-type = %s", evdw_names[fr->vdwtype]);
+ }
+
+ scale = fr->nblists[0].table_vdw.scale;
+ vdwtab = fr->nblists[0].table_vdw.data;
+
+ ri0 = floor(fr->rvdw_switch*scale);
+ ri1 = ceil(fr->rvdw*scale);
+ r0 = ri0/scale;
+ r1 = ri1/scale;
+ rc3 = r0*r0*r0;
+ rc9 = rc3*rc3*rc3;
+
+ /* Calculate self-interaction coefficient (assuming that
+ * the reciprocal-space contribution is constant in the
+ * region that contributes to the self-interaction).
+ */
+ fr->enershiftsix = pow(fr->ewaldcoeff_lj, 6) / 6.0;
+
+ /* Calculate C12 values as without PME. */
+ if (EVDW_SWITCHED(fr->vdwtype))
+ {
+ enersum = 0;
+ virsum = 0;
+ integrate_table(vdwtab, scale, 4, ri0, ri1, &enersum, &virsum);
+ eners[1] -= enersum;
+ virs[1] -= virsum;
+ }
+ /* Add analytical corrections, C6 for the whole range, C12
+ * from rvdw_switch to infinity.
+ */
+
+ eners[0] += -pow(sqrt(M_PI)*fr->ewaldcoeff_lj, 3)/3.0;
+ eners[1] += 4.0*M_PI/(9.0*rc9);
+ virs[0] += pow(sqrt(M_PI)*fr->ewaldcoeff_lj, 3);
+ virs[1] += -16.0*M_PI/(3.0*rc9);
+ }
+ else if ((fr->vdwtype == evdwCUT) || (fr->vdwtype == evdwUSER))
+ {
+ if (fr->vdwtype == evdwUSER && fplog)
+ {
+ fprintf(fplog,
+ "WARNING: using dispersion correction with user tables\n");
+ }
+ rc3 = fr->rvdw*fr->rvdw*fr->rvdw;
+ rc9 = rc3*rc3*rc3;
+ /* Contribution beyond the cut-off */
+ eners[0] += -4.0*M_PI/(3.0*rc3);
+ eners[1] += 4.0*M_PI/(9.0*rc9);
+ if (fr->vdw_modifier == eintmodPOTSHIFT)
+ {
+ /* Contribution within the cut-off */
+ eners[0] += -4.0*M_PI/(3.0*rc3);
+ eners[1] += 4.0*M_PI/(3.0*rc9);
+ }
+ /* Contribution beyond the cut-off */
+ virs[0] += 8.0*M_PI/rc3;
+ virs[1] += -16.0*M_PI/(3.0*rc9);
+ }
+ else
+ {
+ gmx_fatal(FARGS,
+ "Dispersion correction is not implemented for vdw-type = %s",
+ evdw_names[fr->vdwtype]);
+ }
+ fr->enerdiffsix = eners[0];
+ fr->enerdifftwelve = eners[1];
+ /* The 0.5 is due to the Gromacs definition of the virial */
+ fr->virdiffsix = 0.5*virs[0];
+ fr->virdifftwelve = 0.5*virs[1];
+ }
+}
+
+void calc_dispcorr(FILE *fplog, t_inputrec *ir, t_forcerec *fr,
+ gmx_int64_t step, int natoms,
+ matrix box, real lambda, tensor pres, tensor virial,
+ real *prescorr, real *enercorr, real *dvdlcorr)
+{
+ gmx_bool bCorrAll, bCorrPres;
+ real dvdlambda, invvol, dens, ninter, avcsix, avctwelve, enerdiff, svir = 0, spres = 0;
+ int m;
+
+ *prescorr = 0;
+ *enercorr = 0;
+ *dvdlcorr = 0;
+
+ clear_mat(virial);
+ clear_mat(pres);
+
+ if (ir->eDispCorr != edispcNO)
+ {
+ bCorrAll = (ir->eDispCorr == edispcAllEner ||
+ ir->eDispCorr == edispcAllEnerPres);
+ bCorrPres = (ir->eDispCorr == edispcEnerPres ||
+ ir->eDispCorr == edispcAllEnerPres);
+
+ invvol = 1/det(box);
+ if (fr->n_tpi)
+ {
+ /* Only correct for the interactions with the inserted molecule */
+ dens = (natoms - fr->n_tpi)*invvol;
+ ninter = fr->n_tpi;
+ }
+ else
+ {
+ dens = natoms*invvol;
+ ninter = 0.5*natoms;
+ }
+
+ if (ir->efep == efepNO)
+ {
+ avcsix = fr->avcsix[0];
+ avctwelve = fr->avctwelve[0];
+ }
+ else
+ {
+ avcsix = (1 - lambda)*fr->avcsix[0] + lambda*fr->avcsix[1];
+ avctwelve = (1 - lambda)*fr->avctwelve[0] + lambda*fr->avctwelve[1];
+ }
+
+ enerdiff = ninter*(dens*fr->enerdiffsix - fr->enershiftsix);
+ *enercorr += avcsix*enerdiff;
+ dvdlambda = 0.0;
+ if (ir->efep != efepNO)
+ {
+ dvdlambda += (fr->avcsix[1] - fr->avcsix[0])*enerdiff;
+ }
+ if (bCorrAll)
+ {
+ enerdiff = ninter*(dens*fr->enerdifftwelve - fr->enershifttwelve);
+ *enercorr += avctwelve*enerdiff;
+ if (fr->efep != efepNO)
+ {
+ dvdlambda += (fr->avctwelve[1] - fr->avctwelve[0])*enerdiff;
+ }
+ }
+
+ if (bCorrPres)
+ {
+ svir = ninter*dens*avcsix*fr->virdiffsix/3.0;
+ if (ir->eDispCorr == edispcAllEnerPres)
+ {
+ svir += ninter*dens*avctwelve*fr->virdifftwelve/3.0;
+ }
+ /* The factor 2 is because of the Gromacs virial definition */
+ spres = -2.0*invvol*svir*PRESFAC;
+
+ for (m = 0; m < DIM; m++)
+ {
+ virial[m][m] += svir;
+ pres[m][m] += spres;
+ }
+ *prescorr += spres;
+ }
+
+ /* Can't currently control when it prints, for now, just print when degugging */
+ if (debug)
+ {
+ if (bCorrAll)
+ {
+ fprintf(debug, "Long Range LJ corr.: <C6> %10.4e, <C12> %10.4e\n",
+ avcsix, avctwelve);
+ }
+ if (bCorrPres)
+ {
+ fprintf(debug,
+ "Long Range LJ corr.: Epot %10g, Pres: %10g, Vir: %10g\n",
+ *enercorr, spres, svir);
+ }
+ else
+ {
+ fprintf(debug, "Long Range LJ corr.: Epot %10g\n", *enercorr);
+ }
+ }
+
+ if (fr->bSepDVDL && do_per_step(step, ir->nstlog))
+ {
+ gmx_print_sepdvdl(fplog, "Dispersion correction", *enercorr, dvdlambda);
+ }
+ if (fr->efep != efepNO)
+ {
+ *dvdlcorr += dvdlambda;
+ }
+ }
+}
+
+void do_pbc_first(FILE *fplog, matrix box, t_forcerec *fr,
+ t_graph *graph, rvec x[])
+{
+ if (fplog)
+ {
+ fprintf(fplog, "Removing pbc first time\n");
+ }
+ calc_shifts(box, fr->shift_vec);
+ if (graph)
+ {
+ mk_mshift(fplog, graph, fr->ePBC, box, x);
+ if (gmx_debug_at)
+ {
+ p_graph(debug, "do_pbc_first 1", graph);
+ }
+ shift_self(graph, box, x);
+ /* By doing an extra mk_mshift the molecules that are broken
+ * because they were e.g. imported from another software
+ * will be made whole again. Such are the healing powers
+ * of GROMACS.
+ */
+ mk_mshift(fplog, graph, fr->ePBC, box, x);
+ if (gmx_debug_at)
+ {
+ p_graph(debug, "do_pbc_first 2", graph);
+ }
+ }
+ if (fplog)
+ {
+ fprintf(fplog, "Done rmpbc\n");
+ }
+}
+
+static void low_do_pbc_mtop(FILE *fplog, int ePBC, matrix box,
+ gmx_mtop_t *mtop, rvec x[],
+ gmx_bool bFirst)
+{
+ t_graph *graph;
+ int mb, as, mol;
+ gmx_molblock_t *molb;
+
+ if (bFirst && fplog)
+ {
+ fprintf(fplog, "Removing pbc first time\n");
+ }
+
+ snew(graph, 1);
+ as = 0;
+ for (mb = 0; mb < mtop->nmolblock; mb++)
+ {
+ molb = &mtop->molblock[mb];
+ if (molb->natoms_mol == 1 ||
+ (!bFirst && mtop->moltype[molb->type].cgs.nr == 1))
+ {
+ /* Just one atom or charge group in the molecule, no PBC required */
+ as += molb->nmol*molb->natoms_mol;
+ }
+ else
+ {
+ /* Pass NULL iso fplog to avoid graph prints for each molecule type */
+ mk_graph_ilist(NULL, mtop->moltype[molb->type].ilist,
+ 0, molb->natoms_mol, FALSE, FALSE, graph);
+
+ for (mol = 0; mol < molb->nmol; mol++)
+ {
+ mk_mshift(fplog, graph, ePBC, box, x+as);
+
+ shift_self(graph, box, x+as);
+ /* The molecule is whole now.
+ * We don't need the second mk_mshift call as in do_pbc_first,
+ * since we no longer need this graph.
+ */
+
+ as += molb->natoms_mol;
+ }
+ done_graph(graph);
+ }
+ }
+ sfree(graph);
+}
+
+void do_pbc_first_mtop(FILE *fplog, int ePBC, matrix box,
+ gmx_mtop_t *mtop, rvec x[])
+{
+ low_do_pbc_mtop(fplog, ePBC, box, mtop, x, TRUE);
+}
+
+void do_pbc_mtop(FILE *fplog, int ePBC, matrix box,
+ gmx_mtop_t *mtop, rvec x[])
+{
+ low_do_pbc_mtop(fplog, ePBC, box, mtop, x, FALSE);
+}
+
+void finish_run(FILE *fplog, t_commrec *cr,
+ t_inputrec *inputrec,
+ t_nrnb nrnb[], gmx_wallcycle_t wcycle,
+ gmx_walltime_accounting_t walltime_accounting,
+ wallclock_gpu_t *gputimes,
+ gmx_bool bWriteStat)
+{
+ int i, j;
+ t_nrnb *nrnb_tot = NULL;
+ real delta_t;
+ double nbfs, mflop;
+ double elapsed_time,
+ elapsed_time_over_all_ranks,
+ elapsed_time_over_all_threads,
+ elapsed_time_over_all_threads_over_all_ranks;
+ wallcycle_sum(cr, wcycle);
+
+ if (cr->nnodes > 1)
+ {
+ snew(nrnb_tot, 1);
+#ifdef GMX_MPI
+ MPI_Allreduce(nrnb->n, nrnb_tot->n, eNRNB, MPI_DOUBLE, MPI_SUM,
+ cr->mpi_comm_mysim);
+#endif
+ }
+ else
+ {
+ nrnb_tot = nrnb;
+ }
+
+ elapsed_time = walltime_accounting_get_elapsed_time(walltime_accounting);
+ elapsed_time_over_all_ranks = elapsed_time;
+ elapsed_time_over_all_threads = walltime_accounting_get_elapsed_time_over_all_threads(walltime_accounting);
+ elapsed_time_over_all_threads_over_all_ranks = elapsed_time_over_all_threads;
+#ifdef GMX_MPI
+ if (cr->nnodes > 1)
+ {
+ /* reduce elapsed_time over all MPI ranks in the current simulation */
+ MPI_Allreduce(&elapsed_time,
+ &elapsed_time_over_all_ranks,
+ 1, MPI_DOUBLE, MPI_SUM,
+ cr->mpi_comm_mysim);
+ elapsed_time_over_all_ranks /= cr->nnodes;
+ /* Reduce elapsed_time_over_all_threads over all MPI ranks in the
+ * current simulation. */
+ MPI_Allreduce(&elapsed_time_over_all_threads,
+ &elapsed_time_over_all_threads_over_all_ranks,
+ 1, MPI_DOUBLE, MPI_SUM,
+ cr->mpi_comm_mysim);
+ }
+#endif
+
+ if (SIMMASTER(cr))
+ {
+ print_flop(fplog, nrnb_tot, &nbfs, &mflop);
+ }
+ if (cr->nnodes > 1)
+ {
+ sfree(nrnb_tot);
+ }
+
+ if ((cr->duty & DUTY_PP) && DOMAINDECOMP(cr))
+ {
+ print_dd_statistics(cr, inputrec, fplog);
+ }
+
+#ifdef GMX_MPI
+ if (PARTDECOMP(cr))
+ {
+ if (MASTER(cr))
+ {
+ t_nrnb *nrnb_all;
+ int s;
+ MPI_Status stat;
+
+ snew(nrnb_all, cr->nnodes);
+ nrnb_all[0] = *nrnb;
+ for (s = 1; s < cr->nnodes; s++)
+ {
+ MPI_Recv(nrnb_all[s].n, eNRNB, MPI_DOUBLE, s, 0,
+ cr->mpi_comm_mysim, &stat);
+ }
+ pr_load(fplog, cr, nrnb_all);
+ sfree(nrnb_all);
+ }
+ else
+ {
+ MPI_Send(nrnb->n, eNRNB, MPI_DOUBLE, MASTERRANK(cr), 0,
+ cr->mpi_comm_mysim);
+ }
+ }
+#endif
+
+ if (SIMMASTER(cr))
+ {
+ wallcycle_print(fplog, cr->nnodes, cr->npmenodes,
+ elapsed_time_over_all_ranks,
+ wcycle, gputimes);
+
+ if (EI_DYNAMICS(inputrec->eI))
+ {
+ delta_t = inputrec->delta_t;
+ }
+ else
+ {
+ delta_t = 0;
+ }
+
+ if (fplog)
+ {
+ print_perf(fplog, elapsed_time_over_all_threads_over_all_ranks,
+ elapsed_time_over_all_ranks,
+ walltime_accounting_get_nsteps_done(walltime_accounting),
+ delta_t, nbfs, mflop);
+ }
+ if (bWriteStat)
+ {
+ print_perf(stderr, elapsed_time_over_all_threads_over_all_ranks,
+ elapsed_time_over_all_ranks,
+ walltime_accounting_get_nsteps_done(walltime_accounting),
+ delta_t, nbfs, mflop);
+ }
+ }
+}
+
+extern void initialize_lambdas(FILE *fplog, t_inputrec *ir, int *fep_state, real *lambda, double *lam0)
+{
+ /* this function works, but could probably use a logic rewrite to keep all the different
+ types of efep straight. */
+
+ int i;
+ t_lambda *fep = ir->fepvals;
+
+ if ((ir->efep == efepNO) && (ir->bSimTemp == FALSE))
+ {
+ for (i = 0; i < efptNR; i++)
+ {
+ lambda[i] = 0.0;
+ if (lam0)
+ {
+ lam0[i] = 0.0;
+ }
+ }
+ return;
+ }
+ else
+ {
+ *fep_state = fep->init_fep_state; /* this might overwrite the checkpoint
+ if checkpoint is set -- a kludge is in for now
+ to prevent this.*/
+ for (i = 0; i < efptNR; i++)
+ {
+ /* overwrite lambda state with init_lambda for now for backwards compatibility */
+ if (fep->init_lambda >= 0) /* if it's -1, it was never initializd */
+ {
+ lambda[i] = fep->init_lambda;
+ if (lam0)
+ {
+ lam0[i] = lambda[i];
+ }
+ }
+ else
+ {
+ lambda[i] = fep->all_lambda[i][*fep_state];
+ if (lam0)
+ {
+ lam0[i] = lambda[i];
+ }
+ }
+ }
+ if (ir->bSimTemp)
+ {
+ /* need to rescale control temperatures to match current state */
+ for (i = 0; i < ir->opts.ngtc; i++)
+ {
+ if (ir->opts.ref_t[i] > 0)
+ {
+ ir->opts.ref_t[i] = ir->simtempvals->temperatures[*fep_state];
+ }
+ }
+ }
+ }
+
+ /* Send to the log the information on the current lambdas */
+ if (fplog != NULL)
+ {
+ fprintf(fplog, "Initial vector of lambda components:[ ");
+ for (i = 0; i < efptNR; i++)
+ {
+ fprintf(fplog, "%10.4f ", lambda[i]);
+ }
+ fprintf(fplog, "]\n");
+ }
+ return;
+}
+
+
+void init_md(FILE *fplog,
+ t_commrec *cr, t_inputrec *ir, const output_env_t oenv,
+ double *t, double *t0,
+ real *lambda, int *fep_state, double *lam0,
+ t_nrnb *nrnb, gmx_mtop_t *mtop,
+ gmx_update_t *upd,
+ int nfile, const t_filenm fnm[],
+ gmx_mdoutf_t *outf, t_mdebin **mdebin,
+ tensor force_vir, tensor shake_vir, rvec mu_tot,
+ gmx_bool *bSimAnn, t_vcm **vcm, unsigned long Flags)
+{
+ int i, j, n;
+ real tmpt, mod;
+
+ /* Initial values */
+ *t = *t0 = ir->init_t;
+
+ *bSimAnn = FALSE;
+ for (i = 0; i < ir->opts.ngtc; i++)
+ {
+ /* set bSimAnn if any group is being annealed */
+ if (ir->opts.annealing[i] != eannNO)
+ {
+ *bSimAnn = TRUE;
+ }
+ }
+ if (*bSimAnn)
+ {
+ update_annealing_target_temp(&(ir->opts), ir->init_t);
+ }
+
+ /* Initialize lambda variables */
+ initialize_lambdas(fplog, ir, fep_state, lambda, lam0);
+
+ if (upd)
+ {
+ *upd = init_update(ir);
+ }
+
+
+ if (vcm != NULL)
+ {
+ *vcm = init_vcm(fplog, &mtop->groups, ir);
+ }
+
+ if (EI_DYNAMICS(ir->eI) && !(Flags & MD_APPENDFILES))
+ {
+ if (ir->etc == etcBERENDSEN)
+ {
+ please_cite(fplog, "Berendsen84a");
+ }
+ if (ir->etc == etcVRESCALE)
+ {
+ please_cite(fplog, "Bussi2007a");
+ }
+ }
+
+ init_nrnb(nrnb);
+
+ if (nfile != -1)
+ {
+ *outf = init_mdoutf(nfile, fnm, Flags, cr, ir, mtop, oenv);
+
+ *mdebin = init_mdebin((Flags & MD_APPENDFILES) ? NULL : mdoutf_get_fp_ene(*outf),
+ mtop, ir, mdoutf_get_fp_dhdl(*outf));
+ }
+
+ if (ir->bAdress)
+ {
+ please_cite(fplog, "Fritsch12");
+ please_cite(fplog, "Junghans10");
+ }
+ /* Initiate variables */
+ clear_mat(force_vir);
+ clear_mat(shake_vir);
+ clear_rvec(mu_tot);
+
+ debug_gmx();
+}
--- /dev/null
- * Copyright (c) 2013, by the GROMACS development team, led by
+/*
+ * This file is part of the GROMACS molecular simulation package.
+ *
+ * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
+ * Copyright (c) 2001-2004, The GROMACS development team.
- print_date_and_time(fplog, cr->nodeid,
- "Started Test Particle Insertion",
- walltime_accounting);
++ * Copyright (c) 2013,2014, by the GROMACS development team, led by
+ * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
+ * and including many others, as listed in the AUTHORS file in the
+ * top-level source directory and at http://www.gromacs.org.
+ *
+ * GROMACS is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public License
+ * as published by the Free Software Foundation; either version 2.1
+ * of the License, or (at your option) any later version.
+ *
+ * GROMACS is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with GROMACS; if not, see
+ * http://www.gnu.org/licenses, or write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * If you want to redistribute modifications to GROMACS, please
+ * consider that scientific software is very special. Version
+ * control is crucial - bugs must be traceable. We will be happy to
+ * consider code for inclusion in the official distribution, but
+ * derived work must not be called official GROMACS. Details are found
+ * in the README & COPYING files - if they are missing, get the
+ * official version at http://www.gromacs.org.
+ *
+ * To help us fund GROMACS development, we humbly ask that you cite
+ * the research papers on the package. Check out http://www.gromacs.org.
+ */
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <string.h>
+#include <time.h>
+#include <math.h>
+#include "sysstuff.h"
+#include "string2.h"
+#include "network.h"
+#include "smalloc.h"
+#include "nrnb.h"
+#include "main.h"
+#include "chargegroup.h"
+#include "force.h"
+#include "macros.h"
+#include "random.h"
+#include "names.h"
+#include "gmx_fatal.h"
+#include "txtdump.h"
+#include "typedefs.h"
+#include "update.h"
+#include "random.h"
+#include "constr.h"
+#include "vec.h"
+#include "tgroup.h"
+#include "mdebin.h"
+#include "vsite.h"
+#include "force.h"
+#include "mdrun.h"
+#include "domdec.h"
+#include "partdec.h"
+#include "gmx_random.h"
+#include "physics.h"
+#include "xvgr.h"
+#include "mdatoms.h"
+#include "ns.h"
+#include "mtop_util.h"
+#include "pme.h"
+#include "gbutil.h"
+
+#include "gromacs/fileio/confio.h"
+#include "gromacs/fileio/gmxfio.h"
+#include "gromacs/fileio/trxio.h"
+#include "gromacs/timing/wallcycle.h"
+#include "gromacs/timing/walltime_accounting.h"
+
+#ifdef GMX_X86_SSE2
+#include "gromacs/simd/general_x86_sse2.h"
+#endif
+
+
+static void global_max(t_commrec *cr, int *n)
+{
+ int *sum, i;
+
+ snew(sum, cr->nnodes);
+ sum[cr->nodeid] = *n;
+ gmx_sumi(cr->nnodes, sum, cr);
+ for (i = 0; i < cr->nnodes; i++)
+ {
+ *n = max(*n, sum[i]);
+ }
+
+ sfree(sum);
+}
+
+static void realloc_bins(double **bin, int *nbin, int nbin_new)
+{
+ int i;
+
+ if (nbin_new != *nbin)
+ {
+ srenew(*bin, nbin_new);
+ for (i = *nbin; i < nbin_new; i++)
+ {
+ (*bin)[i] = 0;
+ }
+ *nbin = nbin_new;
+ }
+}
+
+double do_tpi(FILE *fplog, t_commrec *cr,
+ int nfile, const t_filenm fnm[],
+ const output_env_t oenv, gmx_bool bVerbose, gmx_bool gmx_unused bCompact,
+ int gmx_unused nstglobalcomm,
+ gmx_vsite_t gmx_unused *vsite, gmx_constr_t gmx_unused constr,
+ int gmx_unused stepout,
+ t_inputrec *inputrec,
+ gmx_mtop_t *top_global, t_fcdata *fcd,
+ t_state *state,
+ t_mdatoms *mdatoms,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_edsam_t gmx_unused ed,
+ t_forcerec *fr,
+ int gmx_unused repl_ex_nst, int gmx_unused repl_ex_nex, int gmx_unused repl_ex_seed,
+ gmx_membed_t gmx_unused membed,
+ real gmx_unused cpt_period, real gmx_unused max_hours,
+ const char gmx_unused *deviceOptions,
+ unsigned long gmx_unused Flags,
+ gmx_walltime_accounting_t walltime_accounting)
+{
+ const char *TPI = "Test Particle Insertion";
+ gmx_localtop_t *top;
+ gmx_groups_t *groups;
+ gmx_enerdata_t *enerd;
+ rvec *f;
+ real lambda, t, temp, beta, drmax, epot;
+ double embU, sum_embU, *sum_UgembU, V, V_all, VembU_all;
+ t_trxstatus *status;
+ t_trxframe rerun_fr;
+ gmx_bool bDispCorr, bCharge, bRFExcl, bNotLastFrame, bStateChanged, bNS, bOurStep;
+ tensor force_vir, shake_vir, vir, pres;
+ int cg_tp, a_tp0, a_tp1, ngid, gid_tp, nener, e;
+ rvec *x_mol;
+ rvec mu_tot, x_init, dx, x_tp;
+ int nnodes, frame, nsteps, step;
+ int i, start, end;
+ gmx_rng_t tpi_rand;
+ FILE *fp_tpi = NULL;
+ char *ptr, *dump_pdb, **leg, str[STRLEN], str2[STRLEN];
+ double dbl, dump_ener;
+ gmx_bool bCavity;
+ int nat_cavity = 0, d;
+ real *mass_cavity = NULL, mass_tot;
+ int nbin;
+ double invbinw, *bin, refvolshift, logV, bUlogV;
+ real dvdl, prescorr, enercorr, dvdlcorr;
+ gmx_bool bEnergyOutOfBounds;
+ const char *tpid_leg[2] = {"direct", "reweighted"};
+
+ /* Since there is no upper limit to the insertion energies,
+ * we need to set an upper limit for the distribution output.
+ */
+ real bU_bin_limit = 50;
+ real bU_logV_bin_limit = bU_bin_limit + 10;
+
+ nnodes = cr->nnodes;
+
+ top = gmx_mtop_generate_local_top(top_global, inputrec);
+
+ groups = &top_global->groups;
+
+ bCavity = (inputrec->eI == eiTPIC);
+ if (bCavity)
+ {
+ ptr = getenv("GMX_TPIC_MASSES");
+ if (ptr == NULL)
+ {
+ nat_cavity = 1;
+ }
+ else
+ {
+ /* Read (multiple) masses from env var GMX_TPIC_MASSES,
+ * The center of mass of the last atoms is then used for TPIC.
+ */
+ nat_cavity = 0;
+ while (sscanf(ptr, "%lf%n", &dbl, &i) > 0)
+ {
+ srenew(mass_cavity, nat_cavity+1);
+ mass_cavity[nat_cavity] = dbl;
+ fprintf(fplog, "mass[%d] = %f\n",
+ nat_cavity+1, mass_cavity[nat_cavity]);
+ nat_cavity++;
+ ptr += i;
+ }
+ if (nat_cavity == 0)
+ {
+ gmx_fatal(FARGS, "Found %d masses in GMX_TPIC_MASSES", nat_cavity);
+ }
+ }
+ }
+
+ /*
+ init_em(fplog,TPI,inputrec,&lambda,nrnb,mu_tot,
+ state->box,fr,mdatoms,top,cr,nfile,fnm,NULL,NULL);*/
+ /* We never need full pbc for TPI */
+ fr->ePBC = epbcXYZ;
+ /* Determine the temperature for the Boltzmann weighting */
+ temp = inputrec->opts.ref_t[0];
+ if (fplog)
+ {
+ for (i = 1; (i < inputrec->opts.ngtc); i++)
+ {
+ if (inputrec->opts.ref_t[i] != temp)
+ {
+ fprintf(fplog, "\nWARNING: The temperatures of the different temperature coupling groups are not identical\n\n");
+ fprintf(stderr, "\nWARNING: The temperatures of the different temperature coupling groups are not identical\n\n");
+ }
+ }
+ fprintf(fplog,
+ "\n The temperature for test particle insertion is %.3f K\n\n",
+ temp);
+ }
+ beta = 1.0/(BOLTZ*temp);
+
+ /* Number of insertions per frame */
+ nsteps = inputrec->nsteps;
+
+ /* Use the same neighborlist with more insertions points
+ * in a sphere of radius drmax around the initial point
+ */
+ /* This should be a proper mdp parameter */
+ drmax = inputrec->rtpi;
+
+ /* An environment variable can be set to dump all configurations
+ * to pdb with an insertion energy <= this value.
+ */
+ dump_pdb = getenv("GMX_TPI_DUMP");
+ dump_ener = 0;
+ if (dump_pdb)
+ {
+ sscanf(dump_pdb, "%lf", &dump_ener);
+ }
+
+ atoms2md(top_global, inputrec, 0, NULL, 0, top_global->natoms, mdatoms);
+ update_mdatoms(mdatoms, inputrec->fepvals->init_lambda);
+
+ snew(enerd, 1);
+ init_enerdata(groups->grps[egcENER].nr, inputrec->fepvals->n_lambda, enerd);
+ snew(f, top_global->natoms);
+
+ /* Print to log file */
+ walltime_accounting_start(walltime_accounting);
+ wallcycle_start(wcycle, ewcRUN);
++ print_start(fplog, cr, walltime_accounting, "Test Particle Insertion");
+
+ /* The last charge group is the group to be inserted */
+ cg_tp = top->cgs.nr - 1;
+ a_tp0 = top->cgs.index[cg_tp];
+ a_tp1 = top->cgs.index[cg_tp+1];
+ if (debug)
+ {
+ fprintf(debug, "TPI cg %d, atoms %d-%d\n", cg_tp, a_tp0, a_tp1);
+ }
+ if (a_tp1 - a_tp0 > 1 &&
+ (inputrec->rlist < inputrec->rcoulomb ||
+ inputrec->rlist < inputrec->rvdw))
+ {
+ gmx_fatal(FARGS, "Can not do TPI for multi-atom molecule with a twin-range cut-off");
+ }
+ snew(x_mol, a_tp1-a_tp0);
+
+ bDispCorr = (inputrec->eDispCorr != edispcNO);
+ bCharge = FALSE;
+ for (i = a_tp0; i < a_tp1; i++)
+ {
+ /* Copy the coordinates of the molecule to be insterted */
+ copy_rvec(state->x[i], x_mol[i-a_tp0]);
+ /* Check if we need to print electrostatic energies */
+ bCharge |= (mdatoms->chargeA[i] != 0 ||
+ (mdatoms->chargeB && mdatoms->chargeB[i] != 0));
+ }
+ bRFExcl = (bCharge && EEL_RF(fr->eeltype) && fr->eeltype != eelRF_NEC);
+
+ calc_cgcm(fplog, cg_tp, cg_tp+1, &(top->cgs), state->x, fr->cg_cm);
+ if (bCavity)
+ {
+ if (norm(fr->cg_cm[cg_tp]) > 0.5*inputrec->rlist && fplog)
+ {
+ fprintf(fplog, "WARNING: Your TPI molecule is not centered at 0,0,0\n");
+ fprintf(stderr, "WARNING: Your TPI molecule is not centered at 0,0,0\n");
+ }
+ }
+ else
+ {
+ /* Center the molecule to be inserted at zero */
+ for (i = 0; i < a_tp1-a_tp0; i++)
+ {
+ rvec_dec(x_mol[i], fr->cg_cm[cg_tp]);
+ }
+ }
+
+ if (fplog)
+ {
+ fprintf(fplog, "\nWill insert %d atoms %s partial charges\n",
+ a_tp1-a_tp0, bCharge ? "with" : "without");
+
+ fprintf(fplog, "\nWill insert %d times in each frame of %s\n",
+ nsteps, opt2fn("-rerun", nfile, fnm));
+ }
+
+ if (!bCavity)
+ {
+ if (inputrec->nstlist > 1)
+ {
+ if (drmax == 0 && a_tp1-a_tp0 == 1)
+ {
+ gmx_fatal(FARGS, "Re-using the neighborlist %d times for insertions of a single atom in a sphere of radius %f does not make sense", inputrec->nstlist, drmax);
+ }
+ if (fplog)
+ {
+ fprintf(fplog, "Will use the same neighborlist for %d insertions in a sphere of radius %f\n", inputrec->nstlist, drmax);
+ }
+ }
+ }
+ else
+ {
+ if (fplog)
+ {
+ fprintf(fplog, "Will insert randomly in a sphere of radius %f around the center of the cavity\n", drmax);
+ }
+ }
+
+ ngid = groups->grps[egcENER].nr;
+ gid_tp = GET_CGINFO_GID(fr->cginfo[cg_tp]);
+ nener = 1 + ngid;
+ if (bDispCorr)
+ {
+ nener += 1;
+ }
+ if (bCharge)
+ {
+ nener += ngid;
+ if (bRFExcl)
+ {
+ nener += 1;
+ }
+ if (EEL_FULL(fr->eeltype))
+ {
+ nener += 1;
+ }
+ }
+ snew(sum_UgembU, nener);
+
+ /* Initialize random generator */
+ tpi_rand = gmx_rng_init(inputrec->ld_seed);
+
+ if (MASTER(cr))
+ {
+ fp_tpi = xvgropen(opt2fn("-tpi", nfile, fnm),
+ "TPI energies", "Time (ps)",
+ "(kJ mol\\S-1\\N) / (nm\\S3\\N)", oenv);
+ xvgr_subtitle(fp_tpi, "f. are averages over one frame", oenv);
+ snew(leg, 4+nener);
+ e = 0;
+ sprintf(str, "-kT log(<Ve\\S-\\betaU\\N>/<V>)");
+ leg[e++] = strdup(str);
+ sprintf(str, "f. -kT log<e\\S-\\betaU\\N>");
+ leg[e++] = strdup(str);
+ sprintf(str, "f. <e\\S-\\betaU\\N>");
+ leg[e++] = strdup(str);
+ sprintf(str, "f. V");
+ leg[e++] = strdup(str);
+ sprintf(str, "f. <Ue\\S-\\betaU\\N>");
+ leg[e++] = strdup(str);
+ for (i = 0; i < ngid; i++)
+ {
+ sprintf(str, "f. <U\\sVdW %s\\Ne\\S-\\betaU\\N>",
+ *(groups->grpname[groups->grps[egcENER].nm_ind[i]]));
+ leg[e++] = strdup(str);
+ }
+ if (bDispCorr)
+ {
+ sprintf(str, "f. <U\\sdisp c\\Ne\\S-\\betaU\\N>");
+ leg[e++] = strdup(str);
+ }
+ if (bCharge)
+ {
+ for (i = 0; i < ngid; i++)
+ {
+ sprintf(str, "f. <U\\sCoul %s\\Ne\\S-\\betaU\\N>",
+ *(groups->grpname[groups->grps[egcENER].nm_ind[i]]));
+ leg[e++] = strdup(str);
+ }
+ if (bRFExcl)
+ {
+ sprintf(str, "f. <U\\sRF excl\\Ne\\S-\\betaU\\N>");
+ leg[e++] = strdup(str);
+ }
+ if (EEL_FULL(fr->eeltype))
+ {
+ sprintf(str, "f. <U\\sCoul recip\\Ne\\S-\\betaU\\N>");
+ leg[e++] = strdup(str);
+ }
+ }
+ xvgr_legend(fp_tpi, 4+nener, (const char**)leg, oenv);
+ for (i = 0; i < 4+nener; i++)
+ {
+ sfree(leg[i]);
+ }
+ sfree(leg);
+ }
+ clear_rvec(x_init);
+ V_all = 0;
+ VembU_all = 0;
+
+ invbinw = 10;
+ nbin = 10;
+ snew(bin, nbin);
+
+ bNotLastFrame = read_first_frame(oenv, &status, opt2fn("-rerun", nfile, fnm),
+ &rerun_fr, TRX_NEED_X);
+ frame = 0;
+
+ if (rerun_fr.natoms - (bCavity ? nat_cavity : 0) !=
+ mdatoms->nr - (a_tp1 - a_tp0))
+ {
+ gmx_fatal(FARGS, "Number of atoms in trajectory (%d)%s "
+ "is not equal the number in the run input file (%d) "
+ "minus the number of atoms to insert (%d)\n",
+ rerun_fr.natoms, bCavity ? " minus one" : "",
+ mdatoms->nr, a_tp1-a_tp0);
+ }
+
+ refvolshift = log(det(rerun_fr.box));
+
+#ifdef GMX_X86_SSE2
+ /* Make sure we don't detect SSE overflow generated before this point */
+ gmx_mm_check_and_reset_overflow();
+#endif
+
+ while (bNotLastFrame)
+ {
+ lambda = rerun_fr.lambda;
+ t = rerun_fr.time;
+
+ sum_embU = 0;
+ for (e = 0; e < nener; e++)
+ {
+ sum_UgembU[e] = 0;
+ }
+
+ /* Copy the coordinates from the input trajectory */
+ for (i = 0; i < rerun_fr.natoms; i++)
+ {
+ copy_rvec(rerun_fr.x[i], state->x[i]);
+ }
+ copy_mat(rerun_fr.box, state->box);
+
+ V = det(state->box);
+ logV = log(V);
+
+ bStateChanged = TRUE;
+ bNS = TRUE;
+ for (step = 0; step < nsteps; step++)
+ {
+ /* In parallel all nodes generate all random configurations.
+ * In that way the result is identical to a single cpu tpi run.
+ */
+ if (!bCavity)
+ {
+ /* Random insertion in the whole volume */
+ bNS = (step % inputrec->nstlist == 0);
+ if (bNS)
+ {
+ /* Generate a random position in the box */
+ x_init[XX] = gmx_rng_uniform_real(tpi_rand)*state->box[XX][XX];
+ x_init[YY] = gmx_rng_uniform_real(tpi_rand)*state->box[YY][YY];
+ x_init[ZZ] = gmx_rng_uniform_real(tpi_rand)*state->box[ZZ][ZZ];
+ }
+ if (inputrec->nstlist == 1)
+ {
+ copy_rvec(x_init, x_tp);
+ }
+ else
+ {
+ /* Generate coordinates within |dx|=drmax of x_init */
+ do
+ {
+ dx[XX] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
+ dx[YY] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
+ dx[ZZ] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
+ }
+ while (norm2(dx) > drmax*drmax);
+ rvec_add(x_init, dx, x_tp);
+ }
+ }
+ else
+ {
+ /* Random insertion around a cavity location
+ * given by the last coordinate of the trajectory.
+ */
+ if (step == 0)
+ {
+ if (nat_cavity == 1)
+ {
+ /* Copy the location of the cavity */
+ copy_rvec(rerun_fr.x[rerun_fr.natoms-1], x_init);
+ }
+ else
+ {
+ /* Determine the center of mass of the last molecule */
+ clear_rvec(x_init);
+ mass_tot = 0;
+ for (i = 0; i < nat_cavity; i++)
+ {
+ for (d = 0; d < DIM; d++)
+ {
+ x_init[d] +=
+ mass_cavity[i]*rerun_fr.x[rerun_fr.natoms-nat_cavity+i][d];
+ }
+ mass_tot += mass_cavity[i];
+ }
+ for (d = 0; d < DIM; d++)
+ {
+ x_init[d] /= mass_tot;
+ }
+ }
+ }
+ /* Generate coordinates within |dx|=drmax of x_init */
+ do
+ {
+ dx[XX] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
+ dx[YY] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
+ dx[ZZ] = (2*gmx_rng_uniform_real(tpi_rand) - 1)*drmax;
+ }
+ while (norm2(dx) > drmax*drmax);
+ rvec_add(x_init, dx, x_tp);
+ }
+
+ if (a_tp1 - a_tp0 == 1)
+ {
+ /* Insert a single atom, just copy the insertion location */
+ copy_rvec(x_tp, state->x[a_tp0]);
+ }
+ else
+ {
+ /* Copy the coordinates from the top file */
+ for (i = a_tp0; i < a_tp1; i++)
+ {
+ copy_rvec(x_mol[i-a_tp0], state->x[i]);
+ }
+ /* Rotate the molecule randomly */
+ rotate_conf(a_tp1-a_tp0, state->x+a_tp0, NULL,
+ 2*M_PI*gmx_rng_uniform_real(tpi_rand),
+ 2*M_PI*gmx_rng_uniform_real(tpi_rand),
+ 2*M_PI*gmx_rng_uniform_real(tpi_rand));
+ /* Shift to the insertion location */
+ for (i = a_tp0; i < a_tp1; i++)
+ {
+ rvec_inc(state->x[i], x_tp);
+ }
+ }
+
+ /* Check if this insertion belongs to this node */
+ bOurStep = TRUE;
+ if (PAR(cr))
+ {
+ switch (inputrec->eI)
+ {
+ case eiTPI:
+ bOurStep = ((step / inputrec->nstlist) % nnodes == cr->nodeid);
+ break;
+ case eiTPIC:
+ bOurStep = (step % nnodes == cr->nodeid);
+ break;
+ default:
+ gmx_fatal(FARGS, "Unknown integrator %s", ei_names[inputrec->eI]);
+ }
+ }
+ if (bOurStep)
+ {
+ /* Clear some matrix variables */
+ clear_mat(force_vir);
+ clear_mat(shake_vir);
+ clear_mat(vir);
+ clear_mat(pres);
+
+ /* Set the charge group center of mass of the test particle */
+ copy_rvec(x_init, fr->cg_cm[top->cgs.nr-1]);
+
+ /* Calc energy (no forces) on new positions.
+ * Since we only need the intermolecular energy
+ * and the RF exclusion terms of the inserted molecule occur
+ * within a single charge group we can pass NULL for the graph.
+ * This also avoids shifts that would move charge groups
+ * out of the box.
+ *
+ * Some checks above ensure than we can not have
+ * twin-range interactions together with nstlist > 1,
+ * therefore we do not need to remember the LR energies.
+ */
+ /* Make do_force do a single node force calculation */
+ cr->nnodes = 1;
+ do_force(fplog, cr, inputrec,
+ step, nrnb, wcycle, top, &top_global->groups,
+ state->box, state->x, &state->hist,
+ f, force_vir, mdatoms, enerd, fcd,
+ state->lambda,
+ NULL, fr, NULL, mu_tot, t, NULL, NULL, FALSE,
+ GMX_FORCE_NONBONDED | GMX_FORCE_ENERGY |
+ (bNS ? GMX_FORCE_DYNAMICBOX | GMX_FORCE_NS | GMX_FORCE_DO_LR : 0) |
+ (bStateChanged ? GMX_FORCE_STATECHANGED : 0));
+ cr->nnodes = nnodes;
+ bStateChanged = FALSE;
+ bNS = FALSE;
+
+ /* Calculate long range corrections to pressure and energy */
+ calc_dispcorr(fplog, inputrec, fr, step, top_global->natoms, state->box,
+ lambda, pres, vir, &prescorr, &enercorr, &dvdlcorr);
+ /* figure out how to rearrange the next 4 lines MRS 8/4/2009 */
+ enerd->term[F_DISPCORR] = enercorr;
+ enerd->term[F_EPOT] += enercorr;
+ enerd->term[F_PRES] += prescorr;
+ enerd->term[F_DVDL_VDW] += dvdlcorr;
+
+ epot = enerd->term[F_EPOT];
+ bEnergyOutOfBounds = FALSE;
+#ifdef GMX_X86_SSE2
+ /* With SSE the energy can overflow, check for this */
+ if (gmx_mm_check_and_reset_overflow())
+ {
+ if (debug)
+ {
+ fprintf(debug, "Found an SSE overflow, assuming the energy is out of bounds\n");
+ }
+ bEnergyOutOfBounds = TRUE;
+ }
+#endif
+ /* If the compiler doesn't optimize this check away
+ * we catch the NAN energies.
+ * The epot>GMX_REAL_MAX check catches inf values,
+ * which should nicely result in embU=0 through the exp below,
+ * but it does not hurt to check anyhow.
+ */
+ /* Non-bonded Interaction usually diverge at r=0.
+ * With tabulated interaction functions the first few entries
+ * should be capped in a consistent fashion between
+ * repulsion, dispersion and Coulomb to avoid accidental
+ * negative values in the total energy.
+ * The table generation code in tables.c does this.
+ * With user tbales the user should take care of this.
+ */
+ if (epot != epot || epot > GMX_REAL_MAX)
+ {
+ bEnergyOutOfBounds = TRUE;
+ }
+ if (bEnergyOutOfBounds)
+ {
+ if (debug)
+ {
+ fprintf(debug, "\n time %.3f, step %d: non-finite energy %f, using exp(-bU)=0\n", t, step, epot);
+ }
+ embU = 0;
+ }
+ else
+ {
+ embU = exp(-beta*epot);
+ sum_embU += embU;
+ /* Determine the weighted energy contributions of each energy group */
+ e = 0;
+ sum_UgembU[e++] += epot*embU;
+ if (fr->bBHAM)
+ {
+ for (i = 0; i < ngid; i++)
+ {
+ sum_UgembU[e++] +=
+ (enerd->grpp.ener[egBHAMSR][GID(i, gid_tp, ngid)] +
+ enerd->grpp.ener[egBHAMLR][GID(i, gid_tp, ngid)])*embU;
+ }
+ }
+ else
+ {
+ for (i = 0; i < ngid; i++)
+ {
+ sum_UgembU[e++] +=
+ (enerd->grpp.ener[egLJSR][GID(i, gid_tp, ngid)] +
+ enerd->grpp.ener[egLJLR][GID(i, gid_tp, ngid)])*embU;
+ }
+ }
+ if (bDispCorr)
+ {
+ sum_UgembU[e++] += enerd->term[F_DISPCORR]*embU;
+ }
+ if (bCharge)
+ {
+ for (i = 0; i < ngid; i++)
+ {
+ sum_UgembU[e++] +=
+ (enerd->grpp.ener[egCOULSR][GID(i, gid_tp, ngid)] +
+ enerd->grpp.ener[egCOULLR][GID(i, gid_tp, ngid)])*embU;
+ }
+ if (bRFExcl)
+ {
+ sum_UgembU[e++] += enerd->term[F_RF_EXCL]*embU;
+ }
+ if (EEL_FULL(fr->eeltype))
+ {
+ sum_UgembU[e++] += enerd->term[F_COUL_RECIP]*embU;
+ }
+ }
+ }
+
+ if (embU == 0 || beta*epot > bU_bin_limit)
+ {
+ bin[0]++;
+ }
+ else
+ {
+ i = (int)((bU_logV_bin_limit
+ - (beta*epot - logV + refvolshift))*invbinw
+ + 0.5);
+ if (i < 0)
+ {
+ i = 0;
+ }
+ if (i >= nbin)
+ {
+ realloc_bins(&bin, &nbin, i+10);
+ }
+ bin[i]++;
+ }
+
+ if (debug)
+ {
+ fprintf(debug, "TPI %7d %12.5e %12.5f %12.5f %12.5f\n",
+ step, epot, x_tp[XX], x_tp[YY], x_tp[ZZ]);
+ }
+
+ if (dump_pdb && epot <= dump_ener)
+ {
+ sprintf(str, "t%g_step%d.pdb", t, step);
+ sprintf(str2, "t: %f step %d ener: %f", t, step, epot);
+ write_sto_conf_mtop(str, str2, top_global, state->x, state->v,
+ inputrec->ePBC, state->box);
+ }
+ }
+ }
+
+ if (PAR(cr))
+ {
+ /* When running in parallel sum the energies over the processes */
+ gmx_sumd(1, &sum_embU, cr);
+ gmx_sumd(nener, sum_UgembU, cr);
+ }
+
+ frame++;
+ V_all += V;
+ VembU_all += V*sum_embU/nsteps;
+
+ if (fp_tpi)
+ {
+ if (bVerbose || frame%10 == 0 || frame < 10)
+ {
+ fprintf(stderr, "mu %10.3e <mu> %10.3e\n",
+ -log(sum_embU/nsteps)/beta, -log(VembU_all/V_all)/beta);
+ }
+
+ fprintf(fp_tpi, "%10.3f %12.5e %12.5e %12.5e %12.5e",
+ t,
+ VembU_all == 0 ? 20/beta : -log(VembU_all/V_all)/beta,
+ sum_embU == 0 ? 20/beta : -log(sum_embU/nsteps)/beta,
+ sum_embU/nsteps, V);
+ for (e = 0; e < nener; e++)
+ {
+ fprintf(fp_tpi, " %12.5e", sum_UgembU[e]/nsteps);
+ }
+ fprintf(fp_tpi, "\n");
+ fflush(fp_tpi);
+ }
+
+ bNotLastFrame = read_next_frame(oenv, status, &rerun_fr);
+ } /* End of the loop */
+ walltime_accounting_end(walltime_accounting);
+
+ close_trj(status);
+
+ if (fp_tpi != NULL)
+ {
+ gmx_fio_fclose(fp_tpi);
+ }
+
+ if (fplog != NULL)
+ {
+ fprintf(fplog, "\n");
+ fprintf(fplog, " <V> = %12.5e nm^3\n", V_all/frame);
+ fprintf(fplog, " <mu> = %12.5e kJ/mol\n", -log(VembU_all/V_all)/beta);
+ }
+
+ /* Write the Boltzmann factor histogram */
+ if (PAR(cr))
+ {
+ /* When running in parallel sum the bins over the processes */
+ i = nbin;
+ global_max(cr, &i);
+ realloc_bins(&bin, &nbin, i);
+ gmx_sumd(nbin, bin, cr);
+ }
+ if (MASTER(cr))
+ {
+ fp_tpi = xvgropen(opt2fn("-tpid", nfile, fnm),
+ "TPI energy distribution",
+ "\\betaU - log(V/<V>)", "count", oenv);
+ sprintf(str, "number \\betaU > %g: %9.3e", bU_bin_limit, bin[0]);
+ xvgr_subtitle(fp_tpi, str, oenv);
+ xvgr_legend(fp_tpi, 2, (const char **)tpid_leg, oenv);
+ for (i = nbin-1; i > 0; i--)
+ {
+ bUlogV = -i/invbinw + bU_logV_bin_limit - refvolshift + log(V_all/frame);
+ fprintf(fp_tpi, "%6.2f %10d %12.5e\n",
+ bUlogV,
+ (int)(bin[i]+0.5),
+ bin[i]*exp(-bUlogV)*V_all/VembU_all);
+ }
+ gmx_fio_fclose(fp_tpi);
+ }
+ sfree(bin);
+
+ sfree(sum_UgembU);
+
+ walltime_accounting_set_nsteps_done(walltime_accounting, frame*inputrec->nsteps);
+
+ return 0;
+}
--- /dev/null
- /* Set and write start time */
+/*
+ * This file is part of the GROMACS molecular simulation package.
+ *
+ * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
+ * Copyright (c) 2001-2004, The GROMACS development team.
+ * Copyright (c) 2011,2012,2013,2014, by the GROMACS development team, led by
+ * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
+ * and including many others, as listed in the AUTHORS file in the
+ * top-level source directory and at http://www.gromacs.org.
+ *
+ * GROMACS is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public License
+ * as published by the Free Software Foundation; either version 2.1
+ * of the License, or (at your option) any later version.
+ *
+ * GROMACS is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with GROMACS; if not, see
+ * http://www.gnu.org/licenses, or write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * If you want to redistribute modifications to GROMACS, please
+ * consider that scientific software is very special. Version
+ * control is crucial - bugs must be traceable. We will be happy to
+ * consider code for inclusion in the official distribution, but
+ * derived work must not be called official GROMACS. Details are found
+ * in the README & COPYING files - if they are missing, get the
+ * official version at http://www.gromacs.org.
+ *
+ * To help us fund GROMACS development, we humbly ask that you cite
+ * the research papers on the package. Check out http://www.gromacs.org.
+ */
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include "typedefs.h"
+#include "smalloc.h"
+#include "sysstuff.h"
+#include "vec.h"
+#include "vcm.h"
+#include "mdebin.h"
+#include "nrnb.h"
+#include "calcmu.h"
+#include "index.h"
+#include "vsite.h"
+#include "update.h"
+#include "ns.h"
+#include "mdrun.h"
+#include "md_support.h"
+#include "md_logging.h"
+#include "network.h"
+#include "xvgr.h"
+#include "physics.h"
+#include "names.h"
+#include "force.h"
+#include "disre.h"
+#include "orires.h"
+#include "pme.h"
+#include "mdatoms.h"
+#include "repl_ex.h"
+#include "qmmm.h"
+#include "domdec.h"
+#include "domdec_network.h"
+#include "partdec.h"
+#include "coulomb.h"
+#include "constr.h"
+#include "shellfc.h"
+#include "compute_io.h"
+#include "mvdata.h"
+#include "checkpoint.h"
+#include "mtop_util.h"
+#include "sighandler.h"
+#include "txtdump.h"
+#include "string2.h"
+#include "pme_loadbal.h"
+#include "bondf.h"
+#include "membed.h"
+#include "types/nlistheuristics.h"
+#include "types/iteratedconstraints.h"
+#include "nbnxn_cuda_data_mgmt.h"
+
+#include "gromacs/utility/gmxmpi.h"
+#include "gromacs/fileio/confio.h"
+#include "gromacs/fileio/trajectory_writing.h"
+#include "gromacs/fileio/trnio.h"
+#include "gromacs/fileio/trxio.h"
+#include "gromacs/fileio/xtcio.h"
+#include "gromacs/timing/wallcycle.h"
+#include "gromacs/timing/walltime_accounting.h"
+#include "gromacs/pulling/pull.h"
+#include "gromacs/swap/swapcoords.h"
+
+#ifdef GMX_FAHCORE
+#include "corewrap.h"
+#endif
+
+static void reset_all_counters(FILE *fplog, t_commrec *cr,
+ gmx_int64_t step,
+ gmx_int64_t *step_rel, t_inputrec *ir,
+ gmx_wallcycle_t wcycle, t_nrnb *nrnb,
+ gmx_walltime_accounting_t walltime_accounting,
+ nbnxn_cuda_ptr_t cu_nbv)
+{
+ char sbuf[STEPSTRSIZE];
+
+ /* Reset all the counters related to performance over the run */
+ md_print_warn(cr, fplog, "step %s: resetting all time and cycle counters\n",
+ gmx_step_str(step, sbuf));
+
+ if (cu_nbv)
+ {
+ nbnxn_cuda_reset_timings(cu_nbv);
+ }
+
+ wallcycle_stop(wcycle, ewcRUN);
+ wallcycle_reset_all(wcycle);
+ if (DOMAINDECOMP(cr))
+ {
+ reset_dd_statistics_counters(cr->dd);
+ }
+ init_nrnb(nrnb);
+ ir->init_step += *step_rel;
+ ir->nsteps -= *step_rel;
+ *step_rel = 0;
+ wallcycle_start(wcycle, ewcRUN);
+ walltime_accounting_start(walltime_accounting);
+ print_date_and_time(fplog, cr->nodeid, "Restarted time", walltime_accounting);
+}
+
+double do_md(FILE *fplog, t_commrec *cr, int nfile, const t_filenm fnm[],
+ const output_env_t oenv, gmx_bool bVerbose, gmx_bool bCompact,
+ int nstglobalcomm,
+ gmx_vsite_t *vsite, gmx_constr_t constr,
+ int stepout, t_inputrec *ir,
+ gmx_mtop_t *top_global,
+ t_fcdata *fcd,
+ t_state *state_global,
+ t_mdatoms *mdatoms,
+ t_nrnb *nrnb, gmx_wallcycle_t wcycle,
+ gmx_edsam_t ed, t_forcerec *fr,
+ int repl_ex_nst, int repl_ex_nex, int repl_ex_seed, gmx_membed_t membed,
+ real cpt_period, real max_hours,
+ const char gmx_unused *deviceOptions,
+ unsigned long Flags,
+ gmx_walltime_accounting_t walltime_accounting)
+{
+ gmx_mdoutf_t outf = NULL;
+ gmx_int64_t step, step_rel;
+ double elapsed_time;
+ double t, t0, lam0[efptNR];
+ gmx_bool bGStatEveryStep, bGStat, bCalcVir, bCalcEner;
+ gmx_bool bNS, bNStList, bSimAnn, bStopCM, bRerunMD, bNotLastFrame = FALSE,
+ bFirstStep, bStateFromCP, bStateFromTPX, bInitStep, bLastStep,
+ bBornRadii, bStartingFromCpt;
+ gmx_bool bDoDHDL = FALSE, bDoFEP = FALSE, bDoExpanded = FALSE;
+ gmx_bool do_ene, do_log, do_verbose, bRerunWarnNoV = TRUE,
+ bForceUpdate = FALSE, bCPT;
+ gmx_bool bMasterState;
+ int force_flags, cglo_flags;
+ tensor force_vir, shake_vir, total_vir, tmp_vir, pres;
+ int i, m;
+ t_trxstatus *status;
+ rvec mu_tot;
+ t_vcm *vcm;
+ t_state *bufstate = NULL;
+ matrix *scale_tot, pcoupl_mu, M, ebox;
+ gmx_nlheur_t nlh;
+ t_trxframe rerun_fr;
+ gmx_repl_ex_t repl_ex = NULL;
+ int nchkpt = 1;
+ gmx_localtop_t *top;
+ t_mdebin *mdebin = NULL;
+ t_state *state = NULL;
+ rvec *f_global = NULL;
+ gmx_enerdata_t *enerd;
+ rvec *f = NULL;
+ gmx_global_stat_t gstat;
+ gmx_update_t upd = NULL;
+ t_graph *graph = NULL;
+ globsig_t gs;
+ gmx_rng_t mcrng = NULL;
+ gmx_groups_t *groups;
+ gmx_ekindata_t *ekind, *ekind_save;
+ gmx_shellfc_t shellfc;
+ int count, nconverged = 0;
+ real timestep = 0;
+ double tcount = 0;
+ gmx_bool bConverged = TRUE, bOK, bSumEkinhOld, bExchanged, bNeedRepartition;
+ gmx_bool bAppend;
+ gmx_bool bResetCountersHalfMaxH = FALSE;
+ gmx_bool bVV, bIterativeCase, bFirstIterate, bTemp, bPres, bTrotter;
+ gmx_bool bUpdateDoLR;
+ real dvdl_constr;
+ int a0, a1;
+ rvec *cbuf = NULL;
+ matrix lastbox;
+ real veta_save, scalevir, tracevir;
+ real vetanew = 0;
+ int lamnew = 0;
+ /* for FEP */
+ int nstfep;
+ double cycles;
+ real saved_conserved_quantity = 0;
+ real last_ekin = 0;
+ int iter_i;
+ t_extmass MassQ;
+ int **trotter_seq;
+ char sbuf[STEPSTRSIZE], sbuf2[STEPSTRSIZE];
+ int handled_stop_condition = gmx_stop_cond_none; /* compare to get_stop_condition*/
+ gmx_iterate_t iterate;
+ gmx_int64_t multisim_nsteps = -1; /* number of steps to do before first multisim
+ simulation stops. If equal to zero, don't
+ communicate any more between multisims.*/
+ /* PME load balancing data for GPU kernels */
+ pme_load_balancing_t pme_loadbal = NULL;
+ double cycles_pmes;
+ gmx_bool bPMETuneTry = FALSE, bPMETuneRunning = FALSE;
+
+#ifdef GMX_FAHCORE
+ /* Temporary addition for FAHCORE checkpointing */
+ int chkpt_ret;
+#endif
+
+ /* Check for special mdrun options */
+ bRerunMD = (Flags & MD_RERUN);
+ bAppend = (Flags & MD_APPENDFILES);
+ if (Flags & MD_RESETCOUNTERSHALFWAY)
+ {
+ if (ir->nsteps > 0)
+ {
+ /* Signal to reset the counters half the simulation steps. */
+ wcycle_set_reset_counters(wcycle, ir->nsteps/2);
+ }
+ /* Signal to reset the counters halfway the simulation time. */
+ bResetCountersHalfMaxH = (max_hours > 0);
+ }
+
+ /* md-vv uses averaged full step velocities for T-control
+ md-vv-avek uses averaged half step velocities for T-control (but full step ekin for P control)
+ md uses averaged half step kinetic energies to determine temperature unless defined otherwise by GMX_EKIN_AVE_VEL; */
+ bVV = EI_VV(ir->eI);
+ if (bVV) /* to store the initial velocities while computing virial */
+ {
+ snew(cbuf, top_global->natoms);
+ }
+ /* all the iteratative cases - only if there are constraints */
+ bIterativeCase = ((IR_NPH_TROTTER(ir) || IR_NPT_TROTTER(ir)) && (constr) && (!bRerunMD));
+ gmx_iterate_init(&iterate, FALSE); /* The default value of iterate->bIterationActive is set to
+ false in this step. The correct value, true or false,
+ is set at each step, as it depends on the frequency of temperature
+ and pressure control.*/
+ bTrotter = (bVV && (IR_NPT_TROTTER(ir) || IR_NPH_TROTTER(ir) || IR_NVT_TROTTER(ir)));
+
+ if (bRerunMD)
+ {
+ /* Since we don't know if the frames read are related in any way,
+ * rebuild the neighborlist at every step.
+ */
+ ir->nstlist = 1;
+ ir->nstcalcenergy = 1;
+ nstglobalcomm = 1;
+ }
+
+ check_ir_old_tpx_versions(cr, fplog, ir, top_global);
+
+ nstglobalcomm = check_nstglobalcomm(fplog, cr, nstglobalcomm, ir);
+ bGStatEveryStep = (nstglobalcomm == 1);
+
+ if (!bGStatEveryStep && ir->nstlist == -1 && fplog != NULL)
+ {
+ fprintf(fplog,
+ "To reduce the energy communication with nstlist = -1\n"
+ "the neighbor list validity should not be checked at every step,\n"
+ "this means that exact integration is not guaranteed.\n"
+ "The neighbor list validity is checked after:\n"
+ " <n.list life time> - 2*std.dev.(n.list life time) steps.\n"
+ "In most cases this will result in exact integration.\n"
+ "This reduces the energy communication by a factor of 2 to 3.\n"
+ "If you want less energy communication, set nstlist > 3.\n\n");
+ }
+
+ if (bRerunMD)
+ {
+ ir->nstxout_compressed = 0;
+ }
+ groups = &top_global->groups;
+
+ /* Initial values */
+ init_md(fplog, cr, ir, oenv, &t, &t0, state_global->lambda,
+ &(state_global->fep_state), lam0,
+ nrnb, top_global, &upd,
+ nfile, fnm, &outf, &mdebin,
+ force_vir, shake_vir, mu_tot, &bSimAnn, &vcm, Flags);
+
+ clear_mat(total_vir);
+ clear_mat(pres);
+ /* Energy terms and groups */
+ snew(enerd, 1);
+ init_enerdata(top_global->groups.grps[egcENER].nr, ir->fepvals->n_lambda,
+ enerd);
+ if (DOMAINDECOMP(cr))
+ {
+ f = NULL;
+ }
+ else
+ {
+ snew(f, top_global->natoms);
+ }
+
+ /* Kinetic energy data */
+ snew(ekind, 1);
+ init_ekindata(fplog, top_global, &(ir->opts), ekind);
+ /* needed for iteration of constraints */
+ snew(ekind_save, 1);
+ init_ekindata(fplog, top_global, &(ir->opts), ekind_save);
+ /* Copy the cos acceleration to the groups struct */
+ ekind->cosacc.cos_accel = ir->cos_accel;
+
+ gstat = global_stat_init(ir);
+ debug_gmx();
+
+ /* Check for polarizable models and flexible constraints */
+ shellfc = init_shell_flexcon(fplog,
+ top_global, n_flexible_constraints(constr),
+ (ir->bContinuation ||
+ (DOMAINDECOMP(cr) && !MASTER(cr))) ?
+ NULL : state_global->x);
+
+ if (DEFORM(*ir))
+ {
+ tMPI_Thread_mutex_lock(&deform_init_box_mutex);
+ set_deform_reference_box(upd,
+ deform_init_init_step_tpx,
+ deform_init_box_tpx);
+ tMPI_Thread_mutex_unlock(&deform_init_box_mutex);
+ }
+
+ {
+ double io = compute_io(ir, top_global->natoms, groups, mdebin->ebin->nener, 1);
+ if ((io > 2000) && MASTER(cr))
+ {
+ fprintf(stderr,
+ "\nWARNING: This run will generate roughly %.0f Mb of data\n\n",
+ io);
+ }
+ }
+
+ if (DOMAINDECOMP(cr))
+ {
+ top = dd_init_local_top(top_global);
+
+ snew(state, 1);
+ dd_init_local_state(cr->dd, state_global, state);
+
+ if (DDMASTER(cr->dd) && ir->nstfout)
+ {
+ snew(f_global, state_global->natoms);
+ }
+ }
+ else
+ {
+ if (PAR(cr))
+ {
+ /* Initialize the particle decomposition and split the topology */
+ top = split_system(fplog, top_global, ir, cr);
+
+ pd_cg_range(cr, &fr->cg0, &fr->hcg);
+ pd_at_range(cr, &a0, &a1);
+ }
+ else
+ {
+ top = gmx_mtop_generate_local_top(top_global, ir);
+
+ a0 = 0;
+ a1 = top_global->natoms;
+ }
+
+ forcerec_set_excl_load(fr, top, cr);
+
+ state = partdec_init_local_state(cr, state_global);
+ f_global = f;
+
+ atoms2md(top_global, ir, 0, NULL, a0, a1-a0, mdatoms);
+
+ if (vsite)
+ {
+ set_vsite_top(vsite, top, mdatoms, cr);
+ }
+
+ if (ir->ePBC != epbcNONE && !fr->bMolPBC)
+ {
+ graph = mk_graph(fplog, &(top->idef), 0, top_global->natoms, FALSE, FALSE);
+ }
+
+ if (shellfc)
+ {
+ make_local_shells(cr, mdatoms, shellfc);
+ }
+
+ setup_bonded_threading(fr, &top->idef);
+
+ if (ir->pull && PAR(cr))
+ {
+ dd_make_local_pull_groups(NULL, ir->pull, mdatoms);
+ }
+ }
+
+ if (DOMAINDECOMP(cr))
+ {
+ /* Distribute the charge groups over the nodes from the master node */
+ dd_partition_system(fplog, ir->init_step, cr, TRUE, 1,
+ state_global, top_global, ir,
+ state, &f, mdatoms, top, fr,
+ vsite, shellfc, constr,
+ nrnb, wcycle, FALSE);
+
+ }
+
+ update_mdatoms(mdatoms, state->lambda[efptMASS]);
+
+ if (opt2bSet("-cpi", nfile, fnm))
+ {
+ bStateFromCP = gmx_fexist_master(opt2fn_master("-cpi", nfile, fnm, cr), cr);
+ }
+ else
+ {
+ bStateFromCP = FALSE;
+ }
+
+ if (ir->bExpanded)
+ {
+ init_expanded_ensemble(bStateFromCP, ir, &mcrng, &state->dfhist);
+ }
+
+ if (MASTER(cr))
+ {
+ if (bStateFromCP)
+ {
+ /* Update mdebin with energy history if appending to output files */
+ if (Flags & MD_APPENDFILES)
+ {
+ restore_energyhistory_from_state(mdebin, &state_global->enerhist);
+ }
+ else
+ {
+ /* We might have read an energy history from checkpoint,
+ * free the allocated memory and reset the counts.
+ */
+ done_energyhistory(&state_global->enerhist);
+ init_energyhistory(&state_global->enerhist);
+ }
+ }
+ /* Set the initial energy history in state by updating once */
+ update_energyhistory(&state_global->enerhist, mdebin);
+ }
+
+ if ((state->flags & (1<<estLD_RNG)) && (Flags & MD_READ_RNG))
+ {
+ /* Set the random state if we read a checkpoint file */
+ set_stochd_state(upd, state);
+ }
+
+ if (state->flags & (1<<estMC_RNG))
+ {
+ set_mc_state(mcrng, state);
+ }
+
+ /* Initialize constraints */
+ if (constr)
+ {
+ if (!DOMAINDECOMP(cr))
+ {
+ set_constraints(constr, top, ir, mdatoms, cr);
+ }
+ }
+
+ if (repl_ex_nst > 0)
+ {
+ /* We need to be sure replica exchange can only occur
+ * when the energies are current */
+ check_nst_param(fplog, cr, "nstcalcenergy", ir->nstcalcenergy,
+ "repl_ex_nst", &repl_ex_nst);
+ /* This check needs to happen before inter-simulation
+ * signals are initialized, too */
+ }
+ if (repl_ex_nst > 0 && MASTER(cr))
+ {
+ repl_ex = init_replica_exchange(fplog, cr->ms, state_global, ir,
+ repl_ex_nst, repl_ex_nex, repl_ex_seed);
+ }
+
+ /* PME tuning is only supported with GPUs or PME nodes and not with rerun or LJ-PME.
+ * With perturbed charges with soft-core we should not change the cut-off.
+ */
+ if ((Flags & MD_TUNEPME) &&
+ EEL_PME(fr->eeltype) &&
+ ( (fr->cutoff_scheme == ecutsVERLET && fr->nbv->bUseGPU) || !(cr->duty & DUTY_PME)) &&
+ !(ir->efep != efepNO && mdatoms->nChargePerturbed > 0 && ir->fepvals->bScCoul) &&
+ !bRerunMD && !EVDW_PME(fr->vdwtype))
+ {
+ pme_loadbal_init(&pme_loadbal, ir, state->box, fr->ic, fr->pmedata);
+ cycles_pmes = 0;
+ if (cr->duty & DUTY_PME)
+ {
+ /* Start tuning right away, as we can't measure the load */
+ bPMETuneRunning = TRUE;
+ }
+ else
+ {
+ /* Separate PME nodes, we can measure the PP/PME load balance */
+ bPMETuneTry = TRUE;
+ }
+ }
+
+ if (!ir->bContinuation && !bRerunMD)
+ {
+ if (mdatoms->cFREEZE && (state->flags & (1<<estV)))
+ {
+ /* Set the velocities of frozen particles to zero */
+ for (i = mdatoms->start; i < mdatoms->start+mdatoms->homenr; i++)
+ {
+ for (m = 0; m < DIM; m++)
+ {
+ if (ir->opts.nFreeze[mdatoms->cFREEZE[i]][m])
+ {
+ state->v[i][m] = 0;
+ }
+ }
+ }
+ }
+
+ if (constr)
+ {
+ /* Constrain the initial coordinates and velocities */
+ do_constrain_first(fplog, constr, ir, mdatoms, state,
+ cr, nrnb, fr, top);
+ }
+ if (vsite)
+ {
+ /* Construct the virtual sites for the initial configuration */
+ construct_vsites(vsite, state->x, ir->delta_t, NULL,
+ top->idef.iparams, top->idef.il,
+ fr->ePBC, fr->bMolPBC, graph, cr, state->box);
+ }
+ }
+
+ debug_gmx();
+
+ /* set free energy calculation frequency as the minimum
+ greatest common denominator of nstdhdl, nstexpanded, and repl_ex_nst*/
+ nstfep = ir->fepvals->nstdhdl;
+ if (ir->bExpanded)
+ {
+ nstfep = gmx_greatest_common_divisor(ir->fepvals->nstdhdl, nstfep);
+ }
+ if (repl_ex_nst > 0)
+ {
+ nstfep = gmx_greatest_common_divisor(repl_ex_nst, nstfep);
+ }
+
+ /* I'm assuming we need global communication the first time! MRS */
+ cglo_flags = (CGLO_TEMPERATURE | CGLO_GSTAT
+ | ((ir->comm_mode != ecmNO) ? CGLO_STOPCM : 0)
+ | (bVV ? CGLO_PRESSURE : 0)
+ | (bVV ? CGLO_CONSTRAINT : 0)
+ | (bRerunMD ? CGLO_RERUNMD : 0)
+ | ((Flags & MD_READ_EKIN) ? CGLO_READEKIN : 0));
+
+ bSumEkinhOld = FALSE;
+ compute_globals(fplog, gstat, cr, ir, fr, ekind, state, state_global, mdatoms, nrnb, vcm,
+ NULL, enerd, force_vir, shake_vir, total_vir, pres, mu_tot,
+ constr, NULL, FALSE, state->box,
+ top_global, &bSumEkinhOld, cglo_flags);
+ if (ir->eI == eiVVAK)
+ {
+ /* a second call to get the half step temperature initialized as well */
+ /* we do the same call as above, but turn the pressure off -- internally to
+ compute_globals, this is recognized as a velocity verlet half-step
+ kinetic energy calculation. This minimized excess variables, but
+ perhaps loses some logic?*/
+
+ compute_globals(fplog, gstat, cr, ir, fr, ekind, state, state_global, mdatoms, nrnb, vcm,
+ NULL, enerd, force_vir, shake_vir, total_vir, pres, mu_tot,
+ constr, NULL, FALSE, state->box,
+ top_global, &bSumEkinhOld,
+ cglo_flags &~(CGLO_STOPCM | CGLO_PRESSURE));
+ }
+
+ /* Calculate the initial half step temperature, and save the ekinh_old */
+ if (!(Flags & MD_STARTFROMCPT))
+ {
+ for (i = 0; (i < ir->opts.ngtc); i++)
+ {
+ copy_mat(ekind->tcstat[i].ekinh, ekind->tcstat[i].ekinh_old);
+ }
+ }
+ if (ir->eI != eiVV)
+ {
+ enerd->term[F_TEMP] *= 2; /* result of averages being done over previous and current step,
+ and there is no previous step */
+ }
+
+ /* if using an iterative algorithm, we need to create a working directory for the state. */
+ if (bIterativeCase)
+ {
+ bufstate = init_bufstate(state);
+ }
+
+ /* need to make an initiation call to get the Trotter variables set, as well as other constants for non-trotter
+ temperature control */
+ trotter_seq = init_npt_vars(ir, state, &MassQ, bTrotter);
+
+ if (MASTER(cr))
+ {
+ if (constr && !ir->bContinuation && ir->eConstrAlg == econtLINCS)
+ {
+ fprintf(fplog,
+ "RMS relative constraint deviation after constraining: %.2e\n",
+ constr_rmsd(constr, FALSE));
+ }
+ if (EI_STATE_VELOCITY(ir->eI))
+ {
+ fprintf(fplog, "Initial temperature: %g K\n", enerd->term[F_TEMP]);
+ }
+ if (bRerunMD)
+ {
+ fprintf(stderr, "starting md rerun '%s', reading coordinates from"
+ " input trajectory '%s'\n\n",
+ *(top_global->name), opt2fn("-rerun", nfile, fnm));
+ if (bVerbose)
+ {
+ fprintf(stderr, "Calculated time to finish depends on nsteps from "
+ "run input file,\nwhich may not correspond to the time "
+ "needed to process input trajectory.\n\n");
+ }
+ }
+ else
+ {
+ char tbuf[20];
+ fprintf(stderr, "starting mdrun '%s'\n",
+ *(top_global->name));
+ if (ir->nsteps >= 0)
+ {
+ sprintf(tbuf, "%8.1f", (ir->init_step+ir->nsteps)*ir->delta_t);
+ }
+ else
+ {
+ sprintf(tbuf, "%s", "infinite");
+ }
+ if (ir->init_step > 0)
+ {
+ fprintf(stderr, "%s steps, %s ps (continuing from step %s, %8.1f ps).\n",
+ gmx_step_str(ir->init_step+ir->nsteps, sbuf), tbuf,
+ gmx_step_str(ir->init_step, sbuf2),
+ ir->init_step*ir->delta_t);
+ }
+ else
+ {
+ fprintf(stderr, "%s steps, %s ps.\n",
+ gmx_step_str(ir->nsteps, sbuf), tbuf);
+ }
+ }
+ fprintf(fplog, "\n");
+ }
+
- print_date_and_time(fplog, cr->nodeid, "Started mdrun", walltime_accounting);
+ walltime_accounting_start(walltime_accounting);
- if (fplog)
- {
- fprintf(fplog, "\n");
- }
+ wallcycle_start(wcycle, ewcRUN);
++ print_start(fplog, cr, walltime_accounting, "mdrun");
+
+ /* safest point to do file checkpointing is here. More general point would be immediately before integrator call */
+#ifdef GMX_FAHCORE
+ chkpt_ret = fcCheckPointParallel( cr->nodeid,
+ NULL, 0);
+ if (chkpt_ret == 0)
+ {
+ gmx_fatal( 3, __FILE__, __LINE__, "Checkpoint error on step %d\n", 0 );
+ }
+#endif
+
+ debug_gmx();
+ /***********************************************************
+ *
+ * Loop over MD steps
+ *
+ ************************************************************/
+
+ /* if rerunMD then read coordinates and velocities from input trajectory */
+ if (bRerunMD)
+ {
+ if (getenv("GMX_FORCE_UPDATE"))
+ {
+ bForceUpdate = TRUE;
+ }
+
+ rerun_fr.natoms = 0;
+ if (MASTER(cr))
+ {
+ bNotLastFrame = read_first_frame(oenv, &status,
+ opt2fn("-rerun", nfile, fnm),
+ &rerun_fr, TRX_NEED_X | TRX_READ_V);
+ if (rerun_fr.natoms != top_global->natoms)
+ {
+ gmx_fatal(FARGS,
+ "Number of atoms in trajectory (%d) does not match the "
+ "run input file (%d)\n",
+ rerun_fr.natoms, top_global->natoms);
+ }
+ if (ir->ePBC != epbcNONE)
+ {
+ if (!rerun_fr.bBox)
+ {
+ gmx_fatal(FARGS, "Rerun trajectory frame step %d time %f does not contain a box, while pbc is used", rerun_fr.step, rerun_fr.time);
+ }
+ if (max_cutoff2(ir->ePBC, rerun_fr.box) < sqr(fr->rlistlong))
+ {
+ gmx_fatal(FARGS, "Rerun trajectory frame step %d time %f has too small box dimensions", rerun_fr.step, rerun_fr.time);
+ }
+ }
+ }
+
+ if (PAR(cr))
+ {
+ rerun_parallel_comm(cr, &rerun_fr, &bNotLastFrame);
+ }
+
+ if (ir->ePBC != epbcNONE)
+ {
+ /* Set the shift vectors.
+ * Necessary here when have a static box different from the tpr box.
+ */
+ calc_shifts(rerun_fr.box, fr->shift_vec);
+ }
+ }
+
+ /* loop over MD steps or if rerunMD to end of input trajectory */
+ bFirstStep = TRUE;
+ /* Skip the first Nose-Hoover integration when we get the state from tpx */
+ bStateFromTPX = !bStateFromCP;
+ bInitStep = bFirstStep && (bStateFromTPX || bVV);
+ bStartingFromCpt = (Flags & MD_STARTFROMCPT) && bInitStep;
+ bLastStep = FALSE;
+ bSumEkinhOld = FALSE;
+ bExchanged = FALSE;
+ bNeedRepartition = FALSE;
+
+ init_global_signals(&gs, cr, ir, repl_ex_nst);
+
+ step = ir->init_step;
+ step_rel = 0;
+
+ if (ir->nstlist == -1)
+ {
+ init_nlistheuristics(&nlh, bGStatEveryStep, step);
+ }
+
+ if (MULTISIM(cr) && (repl_ex_nst <= 0 ))
+ {
+ /* check how many steps are left in other sims */
+ multisim_nsteps = get_multisim_nsteps(cr, ir->nsteps);
+ }
+
+
+ /* and stop now if we should */
+ bLastStep = (bRerunMD || (ir->nsteps >= 0 && step_rel > ir->nsteps) ||
+ ((multisim_nsteps >= 0) && (step_rel >= multisim_nsteps )));
+ while (!bLastStep || (bRerunMD && bNotLastFrame))
+ {
+
+ wallcycle_start(wcycle, ewcSTEP);
+
+ if (bRerunMD)
+ {
+ if (rerun_fr.bStep)
+ {
+ step = rerun_fr.step;
+ step_rel = step - ir->init_step;
+ }
+ if (rerun_fr.bTime)
+ {
+ t = rerun_fr.time;
+ }
+ else
+ {
+ t = step;
+ }
+ }
+ else
+ {
+ bLastStep = (step_rel == ir->nsteps);
+ t = t0 + step*ir->delta_t;
+ }
+
+ if (ir->efep != efepNO || ir->bSimTemp)
+ {
+ /* find and set the current lambdas. If rerunning, we either read in a state, or a lambda value,
+ requiring different logic. */
+
+ set_current_lambdas(step, ir->fepvals, bRerunMD, &rerun_fr, state_global, state, lam0);
+ bDoDHDL = do_per_step(step, ir->fepvals->nstdhdl);
+ bDoFEP = (do_per_step(step, nstfep) && (ir->efep != efepNO));
+ bDoExpanded = (do_per_step(step, ir->expandedvals->nstexpanded)
+ && (ir->bExpanded) && (step > 0) && (!bStartingFromCpt));
+ }
+
+ if (bSimAnn)
+ {
+ update_annealing_target_temp(&(ir->opts), t);
+ }
+
+ if (bRerunMD)
+ {
+ if (!(DOMAINDECOMP(cr) && !MASTER(cr)))
+ {
+ for (i = 0; i < state_global->natoms; i++)
+ {
+ copy_rvec(rerun_fr.x[i], state_global->x[i]);
+ }
+ if (rerun_fr.bV)
+ {
+ for (i = 0; i < state_global->natoms; i++)
+ {
+ copy_rvec(rerun_fr.v[i], state_global->v[i]);
+ }
+ }
+ else
+ {
+ for (i = 0; i < state_global->natoms; i++)
+ {
+ clear_rvec(state_global->v[i]);
+ }
+ if (bRerunWarnNoV)
+ {
+ fprintf(stderr, "\nWARNING: Some frames do not contain velocities.\n"
+ " Ekin, temperature and pressure are incorrect,\n"
+ " the virial will be incorrect when constraints are present.\n"
+ "\n");
+ bRerunWarnNoV = FALSE;
+ }
+ }
+ }
+ copy_mat(rerun_fr.box, state_global->box);
+ copy_mat(state_global->box, state->box);
+
+ if (vsite && (Flags & MD_RERUN_VSITE))
+ {
+ if (DOMAINDECOMP(cr))
+ {
+ gmx_fatal(FARGS, "Vsite recalculation with -rerun is not implemented for domain decomposition, use particle decomposition");
+ }
+ if (graph)
+ {
+ /* Following is necessary because the graph may get out of sync
+ * with the coordinates if we only have every N'th coordinate set
+ */
+ mk_mshift(fplog, graph, fr->ePBC, state->box, state->x);
+ shift_self(graph, state->box, state->x);
+ }
+ construct_vsites(vsite, state->x, ir->delta_t, state->v,
+ top->idef.iparams, top->idef.il,
+ fr->ePBC, fr->bMolPBC, graph, cr, state->box);
+ if (graph)
+ {
+ unshift_self(graph, state->box, state->x);
+ }
+ }
+ }
+
+ /* Stop Center of Mass motion */
+ bStopCM = (ir->comm_mode != ecmNO && do_per_step(step, ir->nstcomm));
+
+ if (bRerunMD)
+ {
+ /* for rerun MD always do Neighbour Searching */
+ bNS = (bFirstStep || ir->nstlist != 0);
+ bNStList = bNS;
+ }
+ else
+ {
+ /* Determine whether or not to do Neighbour Searching and LR */
+ bNStList = (ir->nstlist > 0 && step % ir->nstlist == 0);
+
+ bNS = (bFirstStep || bExchanged || bNeedRepartition || bNStList || bDoFEP ||
+ (ir->nstlist == -1 && nlh.nabnsb > 0));
+
+ if (bNS && ir->nstlist == -1)
+ {
+ set_nlistheuristics(&nlh, bFirstStep || bExchanged || bNeedRepartition || bDoFEP, step);
+ }
+ }
+
+ /* check whether we should stop because another simulation has
+ stopped. */
+ if (MULTISIM(cr))
+ {
+ if ( (multisim_nsteps >= 0) && (step_rel >= multisim_nsteps) &&
+ (multisim_nsteps != ir->nsteps) )
+ {
+ if (bNS)
+ {
+ if (MASTER(cr))
+ {
+ fprintf(stderr,
+ "Stopping simulation %d because another one has finished\n",
+ cr->ms->sim);
+ }
+ bLastStep = TRUE;
+ gs.sig[eglsCHKPT] = 1;
+ }
+ }
+ }
+
+ /* < 0 means stop at next step, > 0 means stop at next NS step */
+ if ( (gs.set[eglsSTOPCOND] < 0) ||
+ ( (gs.set[eglsSTOPCOND] > 0) && (bNStList || ir->nstlist == 0) ) )
+ {
+ bLastStep = TRUE;
+ }
+
+ /* Determine whether or not to update the Born radii if doing GB */
+ bBornRadii = bFirstStep;
+ if (ir->implicit_solvent && (step % ir->nstgbradii == 0))
+ {
+ bBornRadii = TRUE;
+ }
+
+ do_log = do_per_step(step, ir->nstlog) || bFirstStep || bLastStep;
+ do_verbose = bVerbose &&
+ (step % stepout == 0 || bFirstStep || bLastStep);
+
+ if (bNS && !(bFirstStep && ir->bContinuation && !bRerunMD))
+ {
+ if (bRerunMD)
+ {
+ bMasterState = TRUE;
+ }
+ else
+ {
+ bMasterState = FALSE;
+ /* Correct the new box if it is too skewed */
+ if (DYNAMIC_BOX(*ir))
+ {
+ if (correct_box(fplog, step, state->box, graph))
+ {
+ bMasterState = TRUE;
+ }
+ }
+ if (DOMAINDECOMP(cr) && bMasterState)
+ {
+ dd_collect_state(cr->dd, state, state_global);
+ }
+ }
+
+ if (DOMAINDECOMP(cr))
+ {
+ /* Repartition the domain decomposition */
+ wallcycle_start(wcycle, ewcDOMDEC);
+ dd_partition_system(fplog, step, cr,
+ bMasterState, nstglobalcomm,
+ state_global, top_global, ir,
+ state, &f, mdatoms, top, fr,
+ vsite, shellfc, constr,
+ nrnb, wcycle,
+ do_verbose && !bPMETuneRunning);
+ wallcycle_stop(wcycle, ewcDOMDEC);
+ /* If using an iterative integrator, reallocate space to match the decomposition */
+ }
+ }
+
+ if (MASTER(cr) && do_log)
+ {
+ print_ebin_header(fplog, step, t, state->lambda[efptFEP]); /* can we improve the information printed here? */
+ }
+
+ if (ir->efep != efepNO)
+ {
+ update_mdatoms(mdatoms, state->lambda[efptMASS]);
+ }
+
+ if ((bRerunMD && rerun_fr.bV) || bExchanged)
+ {
+
+ /* We need the kinetic energy at minus the half step for determining
+ * the full step kinetic energy and possibly for T-coupling.*/
+ /* This may not be quite working correctly yet . . . . */
+ compute_globals(fplog, gstat, cr, ir, fr, ekind, state, state_global, mdatoms, nrnb, vcm,
+ wcycle, enerd, NULL, NULL, NULL, NULL, mu_tot,
+ constr, NULL, FALSE, state->box,
+ top_global, &bSumEkinhOld,
+ CGLO_RERUNMD | CGLO_GSTAT | CGLO_TEMPERATURE);
+ }
+ clear_mat(force_vir);
+
+ /* We write a checkpoint at this MD step when:
+ * either at an NS step when we signalled through gs,
+ * or at the last step (but not when we do not want confout),
+ * but never at the first step or with rerun.
+ */
+ bCPT = (((gs.set[eglsCHKPT] && (bNS || ir->nstlist == 0)) ||
+ (bLastStep && (Flags & MD_CONFOUT))) &&
+ step > ir->init_step && !bRerunMD);
+ if (bCPT)
+ {
+ gs.set[eglsCHKPT] = 0;
+ }
+
+ /* Determine the energy and pressure:
+ * at nstcalcenergy steps and at energy output steps (set below).
+ */
+ if (EI_VV(ir->eI) && (!bInitStep))
+ {
+ /* for vv, the first half of the integration actually corresponds
+ to the previous step. bCalcEner is only required to be evaluated on the 'next' step,
+ but the virial needs to be calculated on both the current step and the 'next' step. Future
+ reorganization may be able to get rid of one of the bCalcVir=TRUE steps. */
+
+ bCalcEner = do_per_step(step-1, ir->nstcalcenergy);
+ bCalcVir = bCalcEner ||
+ (ir->epc != epcNO && (do_per_step(step, ir->nstpcouple) || do_per_step(step-1, ir->nstpcouple)));
+ }
+ else
+ {
+ bCalcEner = do_per_step(step, ir->nstcalcenergy);
+ bCalcVir = bCalcEner ||
+ (ir->epc != epcNO && do_per_step(step, ir->nstpcouple));
+ }
+
+ /* Do we need global communication ? */
+ bGStat = (bCalcVir || bCalcEner || bStopCM ||
+ do_per_step(step, nstglobalcomm) || (bVV && IR_NVT_TROTTER(ir) && do_per_step(step-1, nstglobalcomm)) ||
+ (ir->nstlist == -1 && !bRerunMD && step >= nlh.step_nscheck));
+
+ do_ene = (do_per_step(step, ir->nstenergy) || bLastStep);
+
+ if (do_ene || do_log)
+ {
+ bCalcVir = TRUE;
+ bCalcEner = TRUE;
+ bGStat = TRUE;
+ }
+
+ /* these CGLO_ options remain the same throughout the iteration */
+ cglo_flags = ((bRerunMD ? CGLO_RERUNMD : 0) |
+ (bGStat ? CGLO_GSTAT : 0)
+ );
+
+ force_flags = (GMX_FORCE_STATECHANGED |
+ ((DYNAMIC_BOX(*ir) || bRerunMD) ? GMX_FORCE_DYNAMICBOX : 0) |
+ GMX_FORCE_ALLFORCES |
+ GMX_FORCE_SEPLRF |
+ (bCalcVir ? GMX_FORCE_VIRIAL : 0) |
+ (bCalcEner ? GMX_FORCE_ENERGY : 0) |
+ (bDoFEP ? GMX_FORCE_DHDL : 0)
+ );
+
+ if (fr->bTwinRange)
+ {
+ if (do_per_step(step, ir->nstcalclr))
+ {
+ force_flags |= GMX_FORCE_DO_LR;
+ }
+ }
+
+ if (shellfc)
+ {
+ /* Now is the time to relax the shells */
+ count = relax_shell_flexcon(fplog, cr, bVerbose, step,
+ ir, bNS, force_flags,
+ top,
+ constr, enerd, fcd,
+ state, f, force_vir, mdatoms,
+ nrnb, wcycle, graph, groups,
+ shellfc, fr, bBornRadii, t, mu_tot,
+ &bConverged, vsite,
+ mdoutf_get_fp_field(outf));
+ tcount += count;
+
+ if (bConverged)
+ {
+ nconverged++;
+ }
+ }
+ else
+ {
+ /* The coordinates (x) are shifted (to get whole molecules)
+ * in do_force.
+ * This is parallellized as well, and does communication too.
+ * Check comments in sim_util.c
+ */
+ do_force(fplog, cr, ir, step, nrnb, wcycle, top, groups,
+ state->box, state->x, &state->hist,
+ f, force_vir, mdatoms, enerd, fcd,
+ state->lambda, graph,
+ fr, vsite, mu_tot, t, mdoutf_get_fp_field(outf), ed, bBornRadii,
+ (bNS ? GMX_FORCE_NS : 0) | force_flags);
+ }
+
+ if (bVV && !bStartingFromCpt && !bRerunMD)
+ /* ############### START FIRST UPDATE HALF-STEP FOR VV METHODS############### */
+ {
+ if (ir->eI == eiVV && bInitStep)
+ {
+ /* if using velocity verlet with full time step Ekin,
+ * take the first half step only to compute the
+ * virial for the first step. From there,
+ * revert back to the initial coordinates
+ * so that the input is actually the initial step.
+ */
+ copy_rvecn(state->v, cbuf, 0, state->natoms); /* should make this better for parallelizing? */
+ }
+ else
+ {
+ /* this is for NHC in the Ekin(t+dt/2) version of vv */
+ trotter_update(ir, step, ekind, enerd, state, total_vir, mdatoms, &MassQ, trotter_seq, ettTSEQ1);
+ }
+
+ /* If we are using twin-range interactions where the long-range component
+ * is only evaluated every nstcalclr>1 steps, we should do a special update
+ * step to combine the long-range forces on these steps.
+ * For nstcalclr=1 this is not done, since the forces would have been added
+ * directly to the short-range forces already.
+ */
+ bUpdateDoLR = (fr->bTwinRange && do_per_step(step, ir->nstcalclr));
+
+ update_coords(fplog, step, ir, mdatoms, state, fr->bMolPBC,
+ f, bUpdateDoLR, fr->f_twin, fcd,
+ ekind, M, upd, bInitStep, etrtVELOCITY1,
+ cr, nrnb, constr, &top->idef);
+
+ if (bIterativeCase && do_per_step(step-1, ir->nstpcouple) && !bInitStep)
+ {
+ gmx_iterate_init(&iterate, TRUE);
+ }
+ /* for iterations, we save these vectors, as we will be self-consistently iterating
+ the calculations */
+
+ /*#### UPDATE EXTENDED VARIABLES IN TROTTER FORMULATION */
+
+ /* save the state */
+ if (iterate.bIterationActive)
+ {
+ copy_coupling_state(state, bufstate, ekind, ekind_save, &(ir->opts));
+ }
+
+ bFirstIterate = TRUE;
+ while (bFirstIterate || iterate.bIterationActive)
+ {
+ if (iterate.bIterationActive)
+ {
+ copy_coupling_state(bufstate, state, ekind_save, ekind, &(ir->opts));
+ if (bFirstIterate && bTrotter)
+ {
+ /* The first time through, we need a decent first estimate
+ of veta(t+dt) to compute the constraints. Do
+ this by computing the box volume part of the
+ trotter integration at this time. Nothing else
+ should be changed by this routine here. If
+ !(first time), we start with the previous value
+ of veta. */
+
+ veta_save = state->veta;
+ trotter_update(ir, step, ekind, enerd, state, total_vir, mdatoms, &MassQ, trotter_seq, ettTSEQ0);
+ vetanew = state->veta;
+ state->veta = veta_save;
+ }
+ }
+
+ bOK = TRUE;
+ if (!bRerunMD || rerun_fr.bV || bForceUpdate) /* Why is rerun_fr.bV here? Unclear. */
+ {
+ update_constraints(fplog, step, NULL, ir, ekind, mdatoms,
+ state, fr->bMolPBC, graph, f,
+ &top->idef, shake_vir,
+ cr, nrnb, wcycle, upd, constr,
+ TRUE, bCalcVir, vetanew);
+
+ if (!bOK)
+ {
+ gmx_fatal(FARGS, "Constraint error: Shake, Lincs or Settle could not solve the constrains");
+ }
+
+ }
+ else if (graph)
+ {
+ /* Need to unshift here if a do_force has been
+ called in the previous step */
+ unshift_self(graph, state->box, state->x);
+ }
+
+ /* if VV, compute the pressure and constraints */
+ /* For VV2, we strictly only need this if using pressure
+ * control, but we really would like to have accurate pressures
+ * printed out.
+ * Think about ways around this in the future?
+ * For now, keep this choice in comments.
+ */
+ /*bPres = (ir->eI==eiVV || IR_NPT_TROTTER(ir)); */
+ /*bTemp = ((ir->eI==eiVV &&(!bInitStep)) || (ir->eI==eiVVAK && IR_NPT_TROTTER(ir)));*/
+ bPres = TRUE;
+ bTemp = ((ir->eI == eiVV && (!bInitStep)) || (ir->eI == eiVVAK));
+ if (bCalcEner && ir->eI == eiVVAK) /*MRS: 7/9/2010 -- this still doesn't fix it?*/
+ {
+ bSumEkinhOld = TRUE;
+ }
+ /* for vv, the first half of the integration actually corresponds to the previous step.
+ So we need information from the last step in the first half of the integration */
+ if (bGStat || do_per_step(step-1, nstglobalcomm))
+ {
+ compute_globals(fplog, gstat, cr, ir, fr, ekind, state, state_global, mdatoms, nrnb, vcm,
+ wcycle, enerd, force_vir, shake_vir, total_vir, pres, mu_tot,
+ constr, NULL, FALSE, state->box,
+ top_global, &bSumEkinhOld,
+ cglo_flags
+ | CGLO_ENERGY
+ | (bTemp ? CGLO_TEMPERATURE : 0)
+ | (bPres ? CGLO_PRESSURE : 0)
+ | (bPres ? CGLO_CONSTRAINT : 0)
+ | ((iterate.bIterationActive) ? CGLO_ITERATE : 0)
+ | (bFirstIterate ? CGLO_FIRSTITERATE : 0)
+ | CGLO_SCALEEKIN
+ );
+ /* explanation of above:
+ a) We compute Ekin at the full time step
+ if 1) we are using the AveVel Ekin, and it's not the
+ initial step, or 2) if we are using AveEkin, but need the full
+ time step kinetic energy for the pressure (always true now, since we want accurate statistics).
+ b) If we are using EkinAveEkin for the kinetic energy for the temperature control, we still feed in
+ EkinAveVel because it's needed for the pressure */
+ }
+ /* temperature scaling and pressure scaling to produce the extended variables at t+dt */
+ if (!bInitStep)
+ {
+ if (bTrotter)
+ {
+ m_add(force_vir, shake_vir, total_vir); /* we need the un-dispersion corrected total vir here */
+ trotter_update(ir, step, ekind, enerd, state, total_vir, mdatoms, &MassQ, trotter_seq, ettTSEQ2);
+ }
+ else
+ {
+ if (bExchanged)
+ {
+
+ /* We need the kinetic energy at minus the half step for determining
+ * the full step kinetic energy and possibly for T-coupling.*/
+ /* This may not be quite working correctly yet . . . . */
+ compute_globals(fplog, gstat, cr, ir, fr, ekind, state, state_global, mdatoms, nrnb, vcm,
+ wcycle, enerd, NULL, NULL, NULL, NULL, mu_tot,
+ constr, NULL, FALSE, state->box,
+ top_global, &bSumEkinhOld,
+ CGLO_RERUNMD | CGLO_GSTAT | CGLO_TEMPERATURE);
+ }
+ }
+ }
+
+ if (iterate.bIterationActive &&
+ done_iterating(cr, fplog, step, &iterate, bFirstIterate,
+ state->veta, &vetanew))
+ {
+ break;
+ }
+ bFirstIterate = FALSE;
+ }
+
+ if (bTrotter && !bInitStep)
+ {
+ copy_mat(shake_vir, state->svir_prev);
+ copy_mat(force_vir, state->fvir_prev);
+ if (IR_NVT_TROTTER(ir) && ir->eI == eiVV)
+ {
+ /* update temperature and kinetic energy now that step is over - this is the v(t+dt) point */
+ enerd->term[F_TEMP] = sum_ekin(&(ir->opts), ekind, NULL, (ir->eI == eiVV), FALSE);
+ enerd->term[F_EKIN] = trace(ekind->ekin);
+ }
+ }
+ /* if it's the initial step, we performed this first step just to get the constraint virial */
+ if (bInitStep && ir->eI == eiVV)
+ {
+ copy_rvecn(cbuf, state->v, 0, state->natoms);
+ }
+ }
+
+ /* MRS -- now done iterating -- compute the conserved quantity */
+ if (bVV)
+ {
+ saved_conserved_quantity = compute_conserved_from_auxiliary(ir, state, &MassQ);
+ if (ir->eI == eiVV)
+ {
+ last_ekin = enerd->term[F_EKIN];
+ }
+ if ((ir->eDispCorr != edispcEnerPres) && (ir->eDispCorr != edispcAllEnerPres))
+ {
+ saved_conserved_quantity -= enerd->term[F_DISPCORR];
+ }
+ /* sum up the foreign energy and dhdl terms for vv. currently done every step so that dhdl is correct in the .edr */
+ if (!bRerunMD)
+ {
+ sum_dhdl(enerd, state->lambda, ir->fepvals);
+ }
+ }
+
+ /* ######## END FIRST UPDATE STEP ############## */
+ /* ######## If doing VV, we now have v(dt) ###### */
+ if (bDoExpanded)
+ {
+ /* perform extended ensemble sampling in lambda - we don't
+ actually move to the new state before outputting
+ statistics, but if performing simulated tempering, we
+ do update the velocities and the tau_t. */
+
+ lamnew = ExpandedEnsembleDynamics(fplog, ir, enerd, state, &MassQ, state->fep_state, &state->dfhist, step, mcrng, state->v, mdatoms);
+ /* history is maintained in state->dfhist, but state_global is what is sent to trajectory and log output */
+ copy_df_history(&state_global->dfhist, &state->dfhist);
+ }
+
+ /* Now we have the energies and forces corresponding to the
+ * coordinates at time t. We must output all of this before
+ * the update.
+ */
+ do_md_trajectory_writing(fplog, cr, nfile, fnm, step, step_rel, t,
+ ir, state, state_global, top_global, fr, upd,
+ outf, mdebin, ekind, f, f_global,
+ wcycle, mcrng, &nchkpt,
+ bCPT, bRerunMD, bLastStep, (Flags & MD_CONFOUT),
+ bSumEkinhOld);
+
+ /* kludge -- virial is lost with restart for NPT control. Must restart */
+ if (bStartingFromCpt && bVV)
+ {
+ copy_mat(state->svir_prev, shake_vir);
+ copy_mat(state->fvir_prev, force_vir);
+ }
+
+ elapsed_time = walltime_accounting_get_current_elapsed_time(walltime_accounting);
+
+ /* Check whether everything is still allright */
+ if (((int)gmx_get_stop_condition() > handled_stop_condition)
+#ifdef GMX_THREAD_MPI
+ && MASTER(cr)
+#endif
+ )
+ {
+ /* this is just make gs.sig compatible with the hack
+ of sending signals around by MPI_Reduce with together with
+ other floats */
+ if (gmx_get_stop_condition() == gmx_stop_cond_next_ns)
+ {
+ gs.sig[eglsSTOPCOND] = 1;
+ }
+ if (gmx_get_stop_condition() == gmx_stop_cond_next)
+ {
+ gs.sig[eglsSTOPCOND] = -1;
+ }
+ /* < 0 means stop at next step, > 0 means stop at next NS step */
+ if (fplog)
+ {
+ fprintf(fplog,
+ "\n\nReceived the %s signal, stopping at the next %sstep\n\n",
+ gmx_get_signal_name(),
+ gs.sig[eglsSTOPCOND] == 1 ? "NS " : "");
+ fflush(fplog);
+ }
+ fprintf(stderr,
+ "\n\nReceived the %s signal, stopping at the next %sstep\n\n",
+ gmx_get_signal_name(),
+ gs.sig[eglsSTOPCOND] == 1 ? "NS " : "");
+ fflush(stderr);
+ handled_stop_condition = (int)gmx_get_stop_condition();
+ }
+ else if (MASTER(cr) && (bNS || ir->nstlist <= 0) &&
+ (max_hours > 0 && elapsed_time > max_hours*60.0*60.0*0.99) &&
+ gs.sig[eglsSTOPCOND] == 0 && gs.set[eglsSTOPCOND] == 0)
+ {
+ /* Signal to terminate the run */
+ gs.sig[eglsSTOPCOND] = 1;
+ if (fplog)
+ {
+ fprintf(fplog, "\nStep %s: Run time exceeded %.3f hours, will terminate the run\n", gmx_step_str(step, sbuf), max_hours*0.99);
+ }
+ fprintf(stderr, "\nStep %s: Run time exceeded %.3f hours, will terminate the run\n", gmx_step_str(step, sbuf), max_hours*0.99);
+ }
+
+ if (bResetCountersHalfMaxH && MASTER(cr) &&
+ elapsed_time > max_hours*60.0*60.0*0.495)
+ {
+ gs.sig[eglsRESETCOUNTERS] = 1;
+ }
+
+ if (ir->nstlist == -1 && !bRerunMD)
+ {
+ /* When bGStatEveryStep=FALSE, global_stat is only called
+ * when we check the atom displacements, not at NS steps.
+ * This means that also the bonded interaction count check is not
+ * performed immediately after NS. Therefore a few MD steps could
+ * be performed with missing interactions.
+ * But wrong energies are never written to file,
+ * since energies are only written after global_stat
+ * has been called.
+ */
+ if (step >= nlh.step_nscheck)
+ {
+ nlh.nabnsb = natoms_beyond_ns_buffer(ir, fr, &top->cgs,
+ nlh.scale_tot, state->x);
+ }
+ else
+ {
+ /* This is not necessarily true,
+ * but step_nscheck is determined quite conservatively.
+ */
+ nlh.nabnsb = 0;
+ }
+ }
+
+ /* In parallel we only have to check for checkpointing in steps
+ * where we do global communication,
+ * otherwise the other nodes don't know.
+ */
+ if (MASTER(cr) && ((bGStat || !PAR(cr)) &&
+ cpt_period >= 0 &&
+ (cpt_period == 0 ||
+ elapsed_time >= nchkpt*cpt_period*60.0)) &&
+ gs.set[eglsCHKPT] == 0)
+ {
+ gs.sig[eglsCHKPT] = 1;
+ }
+
+ /* at the start of step, randomize or scale the velocities (trotter done elsewhere) */
+ if (EI_VV(ir->eI))
+ {
+ if (!bInitStep)
+ {
+ update_tcouple(step, ir, state, ekind, upd, &MassQ, mdatoms);
+ }
+ if (ETC_ANDERSEN(ir->etc)) /* keep this outside of update_tcouple because of the extra info required to pass */
+ {
+ gmx_bool bIfRandomize;
+ bIfRandomize = update_randomize_velocities(ir, step, mdatoms, state, upd, &top->idef, constr);
+ /* if we have constraints, we have to remove the kinetic energy parallel to the bonds */
+ if (constr && bIfRandomize)
+ {
+ update_constraints(fplog, step, NULL, ir, ekind, mdatoms,
+ state, fr->bMolPBC, graph, f,
+ &top->idef, tmp_vir,
+ cr, nrnb, wcycle, upd, constr,
+ TRUE, bCalcVir, vetanew);
+ }
+ }
+ }
+
+ if (bIterativeCase && do_per_step(step, ir->nstpcouple))
+ {
+ gmx_iterate_init(&iterate, TRUE);
+ /* for iterations, we save these vectors, as we will be redoing the calculations */
+ copy_coupling_state(state, bufstate, ekind, ekind_save, &(ir->opts));
+ }
+
+ bFirstIterate = TRUE;
+ while (bFirstIterate || iterate.bIterationActive)
+ {
+ /* We now restore these vectors to redo the calculation with improved extended variables */
+ if (iterate.bIterationActive)
+ {
+ copy_coupling_state(bufstate, state, ekind_save, ekind, &(ir->opts));
+ }
+
+ /* We make the decision to break or not -after- the calculation of Ekin and Pressure,
+ so scroll down for that logic */
+
+ /* ######### START SECOND UPDATE STEP ################# */
+ /* Box is changed in update() when we do pressure coupling,
+ * but we should still use the old box for energy corrections and when
+ * writing it to the energy file, so it matches the trajectory files for
+ * the same timestep above. Make a copy in a separate array.
+ */
+ copy_mat(state->box, lastbox);
+
+ bOK = TRUE;
+ dvdl_constr = 0;
+
+ if (!(bRerunMD && !rerun_fr.bV && !bForceUpdate))
+ {
+ wallcycle_start(wcycle, ewcUPDATE);
+ /* UPDATE PRESSURE VARIABLES IN TROTTER FORMULATION WITH CONSTRAINTS */
+ if (bTrotter)
+ {
+ if (iterate.bIterationActive)
+ {
+ if (bFirstIterate)
+ {
+ scalevir = 1;
+ }
+ else
+ {
+ /* we use a new value of scalevir to converge the iterations faster */
+ scalevir = tracevir/trace(shake_vir);
+ }
+ msmul(shake_vir, scalevir, shake_vir);
+ m_add(force_vir, shake_vir, total_vir);
+ clear_mat(shake_vir);
+ }
+ trotter_update(ir, step, ekind, enerd, state, total_vir, mdatoms, &MassQ, trotter_seq, ettTSEQ3);
+ /* We can only do Berendsen coupling after we have summed
+ * the kinetic energy or virial. Since the happens
+ * in global_state after update, we should only do it at
+ * step % nstlist = 1 with bGStatEveryStep=FALSE.
+ */
+ }
+ else
+ {
+ update_tcouple(step, ir, state, ekind, upd, &MassQ, mdatoms);
+ update_pcouple(fplog, step, ir, state, pcoupl_mu, M, bInitStep);
+ }
+
+ if (bVV)
+ {
+ bUpdateDoLR = (fr->bTwinRange && do_per_step(step, ir->nstcalclr));
+
+ /* velocity half-step update */
+ update_coords(fplog, step, ir, mdatoms, state, fr->bMolPBC, f,
+ bUpdateDoLR, fr->f_twin, fcd,
+ ekind, M, upd, FALSE, etrtVELOCITY2,
+ cr, nrnb, constr, &top->idef);
+ }
+
+ /* Above, initialize just copies ekinh into ekin,
+ * it doesn't copy position (for VV),
+ * and entire integrator for MD.
+ */
+
+ if (ir->eI == eiVVAK)
+ {
+ copy_rvecn(state->x, cbuf, 0, state->natoms);
+ }
+ bUpdateDoLR = (fr->bTwinRange && do_per_step(step, ir->nstcalclr));
+
+ update_coords(fplog, step, ir, mdatoms, state, fr->bMolPBC, f,
+ bUpdateDoLR, fr->f_twin, fcd,
+ ekind, M, upd, bInitStep, etrtPOSITION, cr, nrnb, constr, &top->idef);
+ wallcycle_stop(wcycle, ewcUPDATE);
+
+ update_constraints(fplog, step, &dvdl_constr, ir, ekind, mdatoms, state,
+ fr->bMolPBC, graph, f,
+ &top->idef, shake_vir,
+ cr, nrnb, wcycle, upd, constr,
+ FALSE, bCalcVir, state->veta);
+
+ if (ir->eI == eiVVAK)
+ {
+ /* erase F_EKIN and F_TEMP here? */
+ /* just compute the kinetic energy at the half step to perform a trotter step */
+ compute_globals(fplog, gstat, cr, ir, fr, ekind, state, state_global, mdatoms, nrnb, vcm,
+ wcycle, enerd, force_vir, shake_vir, total_vir, pres, mu_tot,
+ constr, NULL, FALSE, lastbox,
+ top_global, &bSumEkinhOld,
+ cglo_flags | CGLO_TEMPERATURE
+ );
+ wallcycle_start(wcycle, ewcUPDATE);
+ trotter_update(ir, step, ekind, enerd, state, total_vir, mdatoms, &MassQ, trotter_seq, ettTSEQ4);
+ /* now we know the scaling, we can compute the positions again again */
+ copy_rvecn(cbuf, state->x, 0, state->natoms);
+
+ bUpdateDoLR = (fr->bTwinRange && do_per_step(step, ir->nstcalclr));
+
+ update_coords(fplog, step, ir, mdatoms, state, fr->bMolPBC, f,
+ bUpdateDoLR, fr->f_twin, fcd,
+ ekind, M, upd, bInitStep, etrtPOSITION, cr, nrnb, constr, &top->idef);
+ wallcycle_stop(wcycle, ewcUPDATE);
+
+ /* do we need an extra constraint here? just need to copy out of state->v to upd->xp? */
+ /* are the small terms in the shake_vir here due
+ * to numerical errors, or are they important
+ * physically? I'm thinking they are just errors, but not completely sure.
+ * For now, will call without actually constraining, constr=NULL*/
+ update_constraints(fplog, step, NULL, ir, ekind, mdatoms,
+ state, fr->bMolPBC, graph, f,
+ &top->idef, tmp_vir,
+ cr, nrnb, wcycle, upd, NULL,
+ FALSE, bCalcVir,
+ state->veta);
+ }
+ if (!bOK)
+ {
+ gmx_fatal(FARGS, "Constraint error: Shake, Lincs or Settle could not solve the constrains");
+ }
+
+ if (fr->bSepDVDL && fplog && do_log)
+ {
+ gmx_print_sepdvdl(fplog, "Constraint dV/dl", 0.0, dvdl_constr);
+ }
+ if (bVV)
+ {
+ /* this factor or 2 correction is necessary
+ because half of the constraint force is removed
+ in the vv step, so we have to double it. See
+ the Redmine issue #1255. It is not yet clear
+ if the factor of 2 is exact, or just a very
+ good approximation, and this will be
+ investigated. The next step is to see if this
+ can be done adding a dhdl contribution from the
+ rattle step, but this is somewhat more
+ complicated with the current code. Will be
+ investigated, hopefully for 4.6.3. However,
+ this current solution is much better than
+ having it completely wrong.
+ */
+ enerd->term[F_DVDL_CONSTR] += 2*dvdl_constr;
+ }
+ else
+ {
+ enerd->term[F_DVDL_CONSTR] += dvdl_constr;
+ }
+ }
+ else if (graph)
+ {
+ /* Need to unshift here */
+ unshift_self(graph, state->box, state->x);
+ }
+
+ if (vsite != NULL)
+ {
+ wallcycle_start(wcycle, ewcVSITECONSTR);
+ if (graph != NULL)
+ {
+ shift_self(graph, state->box, state->x);
+ }
+ construct_vsites(vsite, state->x, ir->delta_t, state->v,
+ top->idef.iparams, top->idef.il,
+ fr->ePBC, fr->bMolPBC, graph, cr, state->box);
+
+ if (graph != NULL)
+ {
+ unshift_self(graph, state->box, state->x);
+ }
+ wallcycle_stop(wcycle, ewcVSITECONSTR);
+ }
+
+ /* ############## IF NOT VV, Calculate globals HERE, also iterate constraints ############ */
+ /* With Leap-Frog we can skip compute_globals at
+ * non-communication steps, but we need to calculate
+ * the kinetic energy one step before communication.
+ */
+ if (bGStat || (!EI_VV(ir->eI) && do_per_step(step+1, nstglobalcomm)))
+ {
+ if (ir->nstlist == -1 && bFirstIterate)
+ {
+ gs.sig[eglsNABNSB] = nlh.nabnsb;
+ }
+ compute_globals(fplog, gstat, cr, ir, fr, ekind, state, state_global, mdatoms, nrnb, vcm,
+ wcycle, enerd, force_vir, shake_vir, total_vir, pres, mu_tot,
+ constr,
+ bFirstIterate ? &gs : NULL,
+ (step_rel % gs.nstms == 0) &&
+ (multisim_nsteps < 0 || (step_rel < multisim_nsteps)),
+ lastbox,
+ top_global, &bSumEkinhOld,
+ cglo_flags
+ | (!EI_VV(ir->eI) || bRerunMD ? CGLO_ENERGY : 0)
+ | (!EI_VV(ir->eI) && bStopCM ? CGLO_STOPCM : 0)
+ | (!EI_VV(ir->eI) ? CGLO_TEMPERATURE : 0)
+ | (!EI_VV(ir->eI) || bRerunMD ? CGLO_PRESSURE : 0)
+ | (iterate.bIterationActive ? CGLO_ITERATE : 0)
+ | (bFirstIterate ? CGLO_FIRSTITERATE : 0)
+ | CGLO_CONSTRAINT
+ );
+ if (ir->nstlist == -1 && bFirstIterate)
+ {
+ nlh.nabnsb = gs.set[eglsNABNSB];
+ gs.set[eglsNABNSB] = 0;
+ }
+ }
+ /* bIterate is set to keep it from eliminating the old ekin kinetic energy terms */
+ /* ############# END CALC EKIN AND PRESSURE ################# */
+
+ /* Note: this is OK, but there are some numerical precision issues with using the convergence of
+ the virial that should probably be addressed eventually. state->veta has better properies,
+ but what we actually need entering the new cycle is the new shake_vir value. Ideally, we could
+ generate the new shake_vir, but test the veta value for convergence. This will take some thought. */
+
+ if (iterate.bIterationActive &&
+ done_iterating(cr, fplog, step, &iterate, bFirstIterate,
+ trace(shake_vir), &tracevir))
+ {
+ break;
+ }
+ bFirstIterate = FALSE;
+ }
+
+ if (!bVV || bRerunMD)
+ {
+ /* sum up the foreign energy and dhdl terms for md and sd. currently done every step so that dhdl is correct in the .edr */
+ sum_dhdl(enerd, state->lambda, ir->fepvals);
+ }
+ update_box(fplog, step, ir, mdatoms, state, f,
+ ir->nstlist == -1 ? &nlh.scale_tot : NULL, pcoupl_mu, nrnb, upd);
+
+ /* ################# END UPDATE STEP 2 ################# */
+ /* #### We now have r(t+dt) and v(t+dt/2) ############# */
+
+ /* The coordinates (x) were unshifted in update */
+ if (!bGStat)
+ {
+ /* We will not sum ekinh_old,
+ * so signal that we still have to do it.
+ */
+ bSumEkinhOld = TRUE;
+ }
+
+ /* ######### BEGIN PREPARING EDR OUTPUT ########### */
+
+ /* use the directly determined last velocity, not actually the averaged half steps */
+ if (bTrotter && ir->eI == eiVV)
+ {
+ enerd->term[F_EKIN] = last_ekin;
+ }
+ enerd->term[F_ETOT] = enerd->term[F_EPOT] + enerd->term[F_EKIN];
+
+ if (bVV)
+ {
+ enerd->term[F_ECONSERVED] = enerd->term[F_ETOT] + saved_conserved_quantity;
+ }
+ else
+ {
+ enerd->term[F_ECONSERVED] = enerd->term[F_ETOT] + compute_conserved_from_auxiliary(ir, state, &MassQ);
+ }
+ /* ######### END PREPARING EDR OUTPUT ########### */
+
+ /* Output stuff */
+ if (MASTER(cr))
+ {
+ gmx_bool do_dr, do_or;
+
+ if (fplog && do_log && bDoExpanded)
+ {
+ /* only needed if doing expanded ensemble */
+ PrintFreeEnergyInfoToFile(fplog, ir->fepvals, ir->expandedvals, ir->bSimTemp ? ir->simtempvals : NULL,
+ &state_global->dfhist, state->fep_state, ir->nstlog, step);
+ }
+ if (!(bStartingFromCpt && (EI_VV(ir->eI))))
+ {
+ if (bCalcEner)
+ {
+ upd_mdebin(mdebin, bDoDHDL, TRUE,
+ t, mdatoms->tmass, enerd, state,
+ ir->fepvals, ir->expandedvals, lastbox,
+ shake_vir, force_vir, total_vir, pres,
+ ekind, mu_tot, constr);
+ }
+ else
+ {
+ upd_mdebin_step(mdebin);
+ }
+
+ do_dr = do_per_step(step, ir->nstdisreout);
+ do_or = do_per_step(step, ir->nstorireout);
+
+ print_ebin(mdoutf_get_fp_ene(outf), do_ene, do_dr, do_or, do_log ? fplog : NULL,
+ step, t,
+ eprNORMAL, bCompact, mdebin, fcd, groups, &(ir->opts));
+ }
+ if (ir->ePull != epullNO)
+ {
+ pull_print_output(ir->pull, step, t);
+ }
+
+ if (do_per_step(step, ir->nstlog))
+ {
+ if (fflush(fplog) != 0)
+ {
+ gmx_fatal(FARGS, "Cannot flush logfile - maybe you are out of disk space?");
+ }
+ }
+ }
+ if (bDoExpanded)
+ {
+ /* Have to do this part _after_ outputting the logfile and the edr file */
+ /* Gets written into the state at the beginning of next loop*/
+ state->fep_state = lamnew;
+ }
+ /* Print the remaining wall clock time for the run */
+ if (MULTIMASTER(cr) && (do_verbose || gmx_got_usr_signal()) && !bPMETuneRunning)
+ {
+ if (shellfc)
+ {
+ fprintf(stderr, "\n");
+ }
+ print_time(stderr, walltime_accounting, step, ir, cr);
+ }
+
+ /* Ion/water position swapping.
+ * Not done in last step since trajectory writing happens before this call
+ * in the MD loop and exchanges would be lost anyway. */
+ bNeedRepartition = FALSE;
+ if ((ir->eSwapCoords != eswapNO) && (step > 0) && !bLastStep &&
+ do_per_step(step, ir->swap->nstswap))
+ {
+ bNeedRepartition = do_swapcoords(cr, step, t, ir, wcycle,
+ bRerunMD ? rerun_fr.x : state->x,
+ bRerunMD ? rerun_fr.box : state->box,
+ top_global, MASTER(cr) && bVerbose, bRerunMD);
+
+ if (bNeedRepartition && DOMAINDECOMP(cr))
+ {
+ dd_collect_state(cr->dd, state, state_global);
+ }
+ }
+
+ /* Replica exchange */
+ bExchanged = FALSE;
+ if ((repl_ex_nst > 0) && (step > 0) && !bLastStep &&
+ do_per_step(step, repl_ex_nst))
+ {
+ bExchanged = replica_exchange(fplog, cr, repl_ex,
+ state_global, enerd,
+ state, step, t);
+ }
+
+ if ( (bExchanged || bNeedRepartition) && DOMAINDECOMP(cr) )
+ {
+ dd_partition_system(fplog, step, cr, TRUE, 1,
+ state_global, top_global, ir,
+ state, &f, mdatoms, top, fr,
+ vsite, shellfc, constr,
+ nrnb, wcycle, FALSE);
+ }
+
+ bFirstStep = FALSE;
+ bInitStep = FALSE;
+ bStartingFromCpt = FALSE;
+
+ /* ####### SET VARIABLES FOR NEXT ITERATION IF THEY STILL NEED IT ###### */
+ /* With all integrators, except VV, we need to retain the pressure
+ * at the current step for coupling at the next step.
+ */
+ if ((state->flags & (1<<estPRES_PREV)) &&
+ (bGStatEveryStep ||
+ (ir->nstpcouple > 0 && step % ir->nstpcouple == 0)))
+ {
+ /* Store the pressure in t_state for pressure coupling
+ * at the next MD step.
+ */
+ copy_mat(pres, state->pres_prev);
+ }
+
+ /* ####### END SET VARIABLES FOR NEXT ITERATION ###### */
+
+ if ( (membed != NULL) && (!bLastStep) )
+ {
+ rescale_membed(step_rel, membed, state_global->x);
+ }
+
+ if (bRerunMD)
+ {
+ if (MASTER(cr))
+ {
+ /* read next frame from input trajectory */
+ bNotLastFrame = read_next_frame(oenv, status, &rerun_fr);
+ }
+
+ if (PAR(cr))
+ {
+ rerun_parallel_comm(cr, &rerun_fr, &bNotLastFrame);
+ }
+ }
+
+ if (!bRerunMD || !rerun_fr.bStep)
+ {
+ /* increase the MD step number */
+ step++;
+ step_rel++;
+ }
+
+ cycles = wallcycle_stop(wcycle, ewcSTEP);
+ if (DOMAINDECOMP(cr) && wcycle)
+ {
+ dd_cycles_add(cr->dd, cycles, ddCyclStep);
+ }
+
+ if (bPMETuneRunning || bPMETuneTry)
+ {
+ /* PME grid + cut-off optimization with GPUs or PME nodes */
+
+ /* Count the total cycles over the last steps */
+ cycles_pmes += cycles;
+
+ /* We can only switch cut-off at NS steps */
+ if (step % ir->nstlist == 0)
+ {
+ /* PME grid + cut-off optimization with GPUs or PME nodes */
+ if (bPMETuneTry)
+ {
+ if (DDMASTER(cr->dd))
+ {
+ /* PME node load is too high, start tuning */
+ bPMETuneRunning = (dd_pme_f_ratio(cr->dd) >= 1.05);
+ }
+ dd_bcast(cr->dd, sizeof(gmx_bool), &bPMETuneRunning);
+
+ if (bPMETuneRunning || step_rel > ir->nstlist*50)
+ {
+ bPMETuneTry = FALSE;
+ }
+ }
+ if (bPMETuneRunning)
+ {
+ /* init_step might not be a multiple of nstlist,
+ * but the first cycle is always skipped anyhow.
+ */
+ bPMETuneRunning =
+ pme_load_balance(pme_loadbal, cr,
+ (bVerbose && MASTER(cr)) ? stderr : NULL,
+ fplog,
+ ir, state, cycles_pmes,
+ fr->ic, fr->nbv, &fr->pmedata,
+ step);
+
+ /* Update constants in forcerec/inputrec to keep them in sync with fr->ic */
+ fr->ewaldcoeff_q = fr->ic->ewaldcoeff_q;
+ fr->rlist = fr->ic->rlist;
+ fr->rlistlong = fr->ic->rlistlong;
+ fr->rcoulomb = fr->ic->rcoulomb;
+ fr->rvdw = fr->ic->rvdw;
+ }
+ cycles_pmes = 0;
+ }
+ }
+
+ if (step_rel == wcycle_get_reset_counters(wcycle) ||
+ gs.set[eglsRESETCOUNTERS] != 0)
+ {
+ /* Reset all the counters related to performance over the run */
+ reset_all_counters(fplog, cr, step, &step_rel, ir, wcycle, nrnb, walltime_accounting,
+ fr->nbv != NULL && fr->nbv->bUseGPU ? fr->nbv->cu_nbv : NULL);
+ wcycle_set_reset_counters(wcycle, -1);
+ if (!(cr->duty & DUTY_PME))
+ {
+ /* Tell our PME node to reset its counters */
+ gmx_pme_send_resetcounters(cr, step);
+ }
+ /* Correct max_hours for the elapsed time */
+ max_hours -= elapsed_time/(60.0*60.0);
+ bResetCountersHalfMaxH = FALSE;
+ gs.set[eglsRESETCOUNTERS] = 0;
+ }
+
+ }
+ /* End of main MD loop */
+ debug_gmx();
+
+ /* Stop measuring walltime */
+ walltime_accounting_end(walltime_accounting);
+
+ if (bRerunMD && MASTER(cr))
+ {
+ close_trj(status);
+ }
+
+ if (!(cr->duty & DUTY_PME))
+ {
+ /* Tell the PME only node to finish */
+ gmx_pme_send_finish(cr);
+ }
+
+ if (MASTER(cr))
+ {
+ if (ir->nstcalcenergy > 0 && !bRerunMD)
+ {
+ print_ebin(mdoutf_get_fp_ene(outf), FALSE, FALSE, FALSE, fplog, step, t,
+ eprAVER, FALSE, mdebin, fcd, groups, &(ir->opts));
+ }
+ }
+
+ done_mdoutf(outf);
+ debug_gmx();
+
+ if (ir->nstlist == -1 && nlh.nns > 0 && fplog)
+ {
+ fprintf(fplog, "Average neighborlist lifetime: %.1f steps, std.dev.: %.1f steps\n", nlh.s1/nlh.nns, sqrt(nlh.s2/nlh.nns - sqr(nlh.s1/nlh.nns)));
+ fprintf(fplog, "Average number of atoms that crossed the half buffer length: %.1f\n\n", nlh.ab/nlh.nns);
+ }
+
+ if (pme_loadbal != NULL)
+ {
+ pme_loadbal_done(pme_loadbal, cr, fplog,
+ fr->nbv != NULL && fr->nbv->bUseGPU);
+ }
+
+ if (shellfc && fplog)
+ {
+ fprintf(fplog, "Fraction of iterations that converged: %.2f %%\n",
+ (nconverged*100.0)/step_rel);
+ fprintf(fplog, "Average number of force evaluations per MD step: %.2f\n\n",
+ tcount/step_rel);
+ }
+
+ if (repl_ex_nst > 0 && MASTER(cr))
+ {
+ print_replica_exchange_statistics(fplog, repl_ex);
+ }
+
+ walltime_accounting_set_nsteps_done(walltime_accounting, step_rel);
+
+ return 0;
+}
--- /dev/null
- "Note that using combined MPI+OpenMP parallelization is almost always",
- "slower than single parallelization, except at the scaling limit, where",
- "especially OpenMP parallelization of PME reduces the communication cost.",
- "OpenMP-only parallelization is much faster than MPI-only parallelization",
+/*
+ * This file is part of the GROMACS molecular simulation package.
+ *
+ * Copyright (c) 1991-2000, University of Groningen, The Netherlands.
+ * Copyright (c) 2001-2004, The GROMACS development team.
+ * Copyright (c) 2011,2012,2013,2014, by the GROMACS development team, led by
+ * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
+ * and including many others, as listed in the AUTHORS file in the
+ * top-level source directory and at http://www.gromacs.org.
+ *
+ * GROMACS is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public License
+ * as published by the Free Software Foundation; either version 2.1
+ * of the License, or (at your option) any later version.
+ *
+ * GROMACS is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with GROMACS; if not, see
+ * http://www.gnu.org/licenses, or write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * If you want to redistribute modifications to GROMACS, please
+ * consider that scientific software is very special. Version
+ * control is crucial - bugs must be traceable. We will be happy to
+ * consider code for inclusion in the official distribution, but
+ * derived work must not be called official GROMACS. Details are found
+ * in the README & COPYING files - if they are missing, get the
+ * official version at http://www.gromacs.org.
+ *
+ * To help us fund GROMACS development, we humbly ask that you cite
+ * the research papers on the package. Check out http://www.gromacs.org.
+ */
+#include "mdrun_main.h"
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <stdio.h>
+
+#include "gromacs/legacyheaders/checkpoint.h"
+#include "gromacs/legacyheaders/copyrite.h"
+#include "gromacs/legacyheaders/gmx_fatal.h"
+#include "gromacs/legacyheaders/macros.h"
+#include "gromacs/legacyheaders/main.h"
+#include "gromacs/legacyheaders/mdrun.h"
+#include "gromacs/legacyheaders/network.h"
+#include "gromacs/legacyheaders/readinp.h"
+#include "gromacs/legacyheaders/typedefs.h"
+
+#include "gromacs/commandline/pargs.h"
+#include "gromacs/fileio/filenm.h"
+
+int gmx_mdrun(int argc, char *argv[])
+{
+ const char *desc[] = {
+ "[THISMODULE] is the main computational chemistry engine",
+ "within GROMACS. Obviously, it performs Molecular Dynamics simulations,",
+ "but it can also perform Stochastic Dynamics, Energy Minimization,",
+ "test particle insertion or (re)calculation of energies.",
+ "Normal mode analysis is another option. In this case [TT]mdrun[tt]",
+ "builds a Hessian matrix from single conformation.",
+ "For usual Normal Modes-like calculations, make sure that",
+ "the structure provided is properly energy-minimized.",
+ "The generated matrix can be diagonalized by [gmx-nmeig].[PAR]",
+ "The [TT]mdrun[tt] program reads the run input file ([TT]-s[tt])",
+ "and distributes the topology over nodes if needed.",
+ "[TT]mdrun[tt] produces at least four output files.",
+ "A single log file ([TT]-g[tt]) is written, unless the option",
+ "[TT]-seppot[tt] is used, in which case each node writes a log file.",
+ "The trajectory file ([TT]-o[tt]), contains coordinates, velocities and",
+ "optionally forces.",
+ "The structure file ([TT]-c[tt]) contains the coordinates and",
+ "velocities of the last step.",
+ "The energy file ([TT]-e[tt]) contains energies, the temperature,",
+ "pressure, etc, a lot of these things are also printed in the log file.",
+ "Optionally coordinates can be written to a compressed trajectory file",
+ "([TT]-x[tt]).[PAR]",
+ "The option [TT]-dhdl[tt] is only used when free energy calculation is",
+ "turned on.[PAR]",
+ "A simulation can be run in parallel using two different parallelization",
+ "schemes: MPI parallelization and/or OpenMP thread parallelization.",
+ "The MPI parallelization uses multiple processes when [TT]mdrun[tt] is",
+ "compiled with a normal MPI library or threads when [TT]mdrun[tt] is",
+ "compiled with the GROMACS built-in thread-MPI library. OpenMP threads",
+ "are supported when [TT]mdrun[tt] is compiled with OpenMP. Full OpenMP support",
+ "is only available with the Verlet cut-off scheme, with the (older)",
+ "group scheme only PME-only processes can use OpenMP parallelization.",
+ "In all cases [TT]mdrun[tt] will by default try to use all the available",
+ "hardware resources. With a normal MPI library only the options",
+ "[TT]-ntomp[tt] (with the Verlet cut-off scheme) and [TT]-ntomp_pme[tt],",
+ "for PME-only processes, can be used to control the number of threads.",
+ "With thread-MPI there are additional options [TT]-nt[tt], which sets",
+ "the total number of threads, and [TT]-ntmpi[tt], which sets the number",
+ "of thread-MPI threads.",
++ "The number of OpenMP threads used by [TT]mdrun[tt] can also be set with",
++ "the standard environment variable, [TT]OMP_NUM_THREADS[tt].",
++ "The [TT]GMX_PME_NUM_THREADS[tt] environment variable can be used to specify",
++ "the number of threads used by the PME-only processes.[PAR]",
++ "Note that combined MPI+OpenMP parallelization is in many cases",
++ "slower than either on its own. However, at high parallelization, using the",
++ "combination is often beneficial as it reduces the number of domains and/or",
++ "the number of MPI ranks. (Less and larger domains can improve scaling,",
++ "with separate PME processes fewer MPI ranks reduces communication cost.)",
++ "OpenMP-only parallelization is typically faster than MPI-only parallelization",
+ "on a single CPU(-die). Since we currently don't have proper hardware",
+ "topology detection, [TT]mdrun[tt] compiled with thread-MPI will only",
+ "automatically use OpenMP-only parallelization when you use up to 4",
+ "threads, up to 12 threads with Intel Nehalem/Westmere, or up to 16",
+ "threads with Intel Sandy Bridge or newer CPUs. Otherwise MPI-only",
+ "parallelization is used (except with GPUs, see below).",
+ "[PAR]",
+ "To quickly test the performance of the new Verlet cut-off scheme",
+ "with old [TT].tpr[tt] files, either on CPUs or CPUs+GPUs, you can use",
+ "the [TT]-testverlet[tt] option. This should not be used for production,",
+ "since it can slightly modify potentials and it will remove charge groups",
+ "making analysis difficult, as the [TT].tpr[tt] file will still contain",
+ "charge groups. For production simulations it is highly recommended",
+ "to specify [TT]cutoff-scheme = Verlet[tt] in the [TT].mdp[tt] file.",
+ "[PAR]",
+ "With GPUs (only supported with the Verlet cut-off scheme), the number",
+ "of GPUs should match the number of MPI processes or MPI threads,",
+ "excluding PME-only processes/threads. With thread-MPI, unless set on the command line, the number",
+ "of MPI threads will automatically be set to the number of GPUs detected.",
+ "To use a subset of the available GPUs, or to manually provide a mapping of",
+ "GPUs to PP ranks, you can use the [TT]-gpu_id[tt] option. The argument of [TT]-gpu_id[tt] is",
+ "a string of digits (without delimiter) representing device id-s of the GPUs to be used.",
+ "For example, \"[TT]02[tt]\" specifies using GPUs 0 and 2 in the first and second PP ranks per compute node",
+ "respectively. To select different sets of GPU-s",
+ "on different nodes of a compute cluster, use the [TT]GMX_GPU_ID[tt] environment",
+ "variable instead. The format for [TT]GMX_GPU_ID[tt] is identical to ",
+ "[TT]-gpu_id[tt], with the difference that an environment variable can have",
+ "different values on different compute nodes. Multiple MPI ranks on each node",
+ "can share GPUs. This is accomplished by specifying the id(s) of the GPU(s)",
+ "multiple times, e.g. \"[TT]0011[tt]\" for four ranks sharing two GPUs in this node.",
+ "This works within a single simulation, or a multi-simulation, with any form of MPI.",
+ "[PAR]",
+ "With the Verlet cut-off scheme and verlet-buffer-tolerance set,",
+ "the pair-list update interval nstlist can be chosen freely with",
+ "the option [TT]-nstlist[tt]. [TT]mdrun[tt] will then adjust",
+ "the pair-list cut-off to maintain accuracy.",
+ "By default [TT]mdrun[tt] will try to increase nstlist to improve",
+ "the performance. For CPU runs nstlist might increase to 20, for GPU",
+ "runs up till 40. But for medium to high parallelization or with",
+ "fast GPUs, a (user supplied) larger nstlist value can give much",
+ "better performance.",
+ "[PAR]",
+ "When using PME with separate PME nodes or with a GPU, the two major",
+ "compute tasks, the non-bonded force calculation and the PME calculation",
+ "run on different compute resources. If this load is not balanced,",
+ "some of the resources will be idle part of time. With the Verlet",
+ "cut-off scheme this load is automatically balanced when the PME load",
+ "is too high (but not when it is too low). This is done by scaling",
+ "the Coulomb cut-off and PME grid spacing by the same amount. In the first",
+ "few hundred steps different settings are tried and the fastest is chosen",
+ "for the rest of the simulation. This does not affect the accuracy of",
+ "the results, but it does affect the decomposition of the Coulomb energy",
+ "into particle and mesh contributions. The auto-tuning can be turned off",
+ "with the option [TT]-notunepme[tt].",
+ "[PAR]",
+ "[TT]mdrun[tt] pins (sets affinity of) threads to specific cores,",
+ "when all (logical) cores on a compute node are used by [TT]mdrun[tt],",
+ "even when no multi-threading is used,",
+ "as this usually results in significantly better performance.",
+ "If the queuing systems or the OpenMP library pinned threads, we honor",
+ "this and don't pin again, even though the layout may be sub-optimal.",
+ "If you want to have [TT]mdrun[tt] override an already set thread affinity",
+ "or pin threads when using less cores, use [TT]-pin on[tt].",
+ "With SMT (simultaneous multithreading), e.g. Intel Hyper-Threading,",
+ "there are multiple logical cores per physical core.",
+ "The option [TT]-pinstride[tt] sets the stride in logical cores for",
+ "pinning consecutive threads. Without SMT, 1 is usually the best choice.",
+ "With Intel Hyper-Threading 2 is best when using half or less of the",
+ "logical cores, 1 otherwise. The default value of 0 do exactly that:",
+ "it minimizes the threads per logical core, to optimize performance.",
+ "If you want to run multiple [TT]mdrun[tt] jobs on the same physical node,"
+ "you should set [TT]-pinstride[tt] to 1 when using all logical cores.",
+ "When running multiple [TT]mdrun[tt] (or other) simulations on the same physical",
+ "node, some simulations need to start pinning from a non-zero core",
+ "to avoid overloading cores; with [TT]-pinoffset[tt] you can specify",
+ "the offset in logical cores for pinning.",
+ "[PAR]",
+ "When [TT]mdrun[tt] is started using MPI with more than 1 process",
+ "or with thread-MPI with more than 1 thread, MPI parallelization is used.",
+ "By default domain decomposition is used, unless the [TT]-pd[tt]",
+ "option is set, which selects particle decomposition.",
+ "[PAR]",
+ "With domain decomposition, the spatial decomposition can be set",
+ "with option [TT]-dd[tt]. By default [TT]mdrun[tt] selects a good decomposition.",
+ "The user only needs to change this when the system is very inhomogeneous.",
+ "Dynamic load balancing is set with the option [TT]-dlb[tt],",
+ "which can give a significant performance improvement,",
+ "especially for inhomogeneous systems. The only disadvantage of",
+ "dynamic load balancing is that runs are no longer binary reproducible,",
+ "but in most cases this is not important.",
+ "By default the dynamic load balancing is automatically turned on",
+ "when the measured performance loss due to load imbalance is 5% or more.",
+ "At low parallelization these are the only important options",
+ "for domain decomposition.",
+ "At high parallelization the options in the next two sections",
+ "could be important for increasing the performace.",
+ "[PAR]",
+ "When PME is used with domain decomposition, separate nodes can",
+ "be assigned to do only the PME mesh calculation;",
+ "this is computationally more efficient starting at about 12 nodes.",
+ "The number of PME nodes is set with option [TT]-npme[tt],",
+ "this can not be more than half of the nodes.",
+ "By default [TT]mdrun[tt] makes a guess for the number of PME",
+ "nodes when the number of nodes is larger than 11 or performance wise",
+ "not compatible with the PME grid x dimension.",
+ "But the user should optimize npme. Performance statistics on this issue",
+ "are written at the end of the log file.",
+ "For good load balancing at high parallelization, the PME grid x and y",
+ "dimensions should be divisible by the number of PME nodes",
+ "(the simulation will run correctly also when this is not the case).",
+ "[PAR]",
+ "This section lists all options that affect the domain decomposition.",
+ "[PAR]",
+ "Option [TT]-rdd[tt] can be used to set the required maximum distance",
+ "for inter charge-group bonded interactions.",
+ "Communication for two-body bonded interactions below the non-bonded",
+ "cut-off distance always comes for free with the non-bonded communication.",
+ "Atoms beyond the non-bonded cut-off are only communicated when they have",
+ "missing bonded interactions; this means that the extra cost is minor",
+ "and nearly indepedent of the value of [TT]-rdd[tt].",
+ "With dynamic load balancing option [TT]-rdd[tt] also sets",
+ "the lower limit for the domain decomposition cell sizes.",
+ "By default [TT]-rdd[tt] is determined by [TT]mdrun[tt] based on",
+ "the initial coordinates. The chosen value will be a balance",
+ "between interaction range and communication cost.",
+ "[PAR]",
+ "When inter charge-group bonded interactions are beyond",
+ "the bonded cut-off distance, [TT]mdrun[tt] terminates with an error message.",
+ "For pair interactions and tabulated bonds",
+ "that do not generate exclusions, this check can be turned off",
+ "with the option [TT]-noddcheck[tt].",
+ "[PAR]",
+ "When constraints are present, option [TT]-rcon[tt] influences",
+ "the cell size limit as well.",
+ "Atoms connected by NC constraints, where NC is the LINCS order plus 1,",
+ "should not be beyond the smallest cell size. A error message is",
+ "generated when this happens and the user should change the decomposition",
+ "or decrease the LINCS order and increase the number of LINCS iterations.",
+ "By default [TT]mdrun[tt] estimates the minimum cell size required for P-LINCS",
+ "in a conservative fashion. For high parallelization it can be useful",
+ "to set the distance required for P-LINCS with the option [TT]-rcon[tt].",
+ "[PAR]",
+ "The [TT]-dds[tt] option sets the minimum allowed x, y and/or z scaling",
+ "of the cells with dynamic load balancing. [TT]mdrun[tt] will ensure that",
+ "the cells can scale down by at least this factor. This option is used",
+ "for the automated spatial decomposition (when not using [TT]-dd[tt])",
+ "as well as for determining the number of grid pulses, which in turn",
+ "sets the minimum allowed cell size. Under certain circumstances",
+ "the value of [TT]-dds[tt] might need to be adjusted to account for",
+ "high or low spatial inhomogeneity of the system.",
+ "[PAR]",
+ "The option [TT]-gcom[tt] can be used to only do global communication",
+ "every n steps.",
+ "This can improve performance for highly parallel simulations",
+ "where this global communication step becomes the bottleneck.",
+ "For a global thermostat and/or barostat the temperature",
+ "and/or pressure will also only be updated every [TT]-gcom[tt] steps.",
+ "By default it is set to the minimum of nstcalcenergy and nstlist.[PAR]",
+ "With [TT]-rerun[tt] an input trajectory can be given for which ",
+ "forces and energies will be (re)calculated. Neighbor searching will be",
+ "performed for every frame, unless [TT]nstlist[tt] is zero",
+ "(see the [TT].mdp[tt] file).[PAR]",
+ "ED (essential dynamics) sampling and/or additional flooding potentials",
+ "are switched on by using the [TT]-ei[tt] flag followed by an [TT].edi[tt]",
+ "file. The [TT].edi[tt] file can be produced with the [TT]make_edi[tt] tool",
+ "or by using options in the essdyn menu of the WHAT IF program.",
+ "[TT]mdrun[tt] produces a [TT].xvg[tt] output file that",
+ "contains projections of positions, velocities and forces onto selected",
+ "eigenvectors.[PAR]",
+ "When user-defined potential functions have been selected in the",
+ "[TT].mdp[tt] file the [TT]-table[tt] option is used to pass [TT]mdrun[tt]",
+ "a formatted table with potential functions. The file is read from",
+ "either the current directory or from the [TT]GMXLIB[tt] directory.",
+ "A number of pre-formatted tables are presented in the [TT]GMXLIB[tt] dir,",
+ "for 6-8, 6-9, 6-10, 6-11, 6-12 Lennard-Jones potentials with",
+ "normal Coulomb.",
+ "When pair interactions are present, a separate table for pair interaction",
+ "functions is read using the [TT]-tablep[tt] option.[PAR]",
+ "When tabulated bonded functions are present in the topology,",
+ "interaction functions are read using the [TT]-tableb[tt] option.",
+ "For each different tabulated interaction type the table file name is",
+ "modified in a different way: before the file extension an underscore is",
+ "appended, then a 'b' for bonds, an 'a' for angles or a 'd' for dihedrals",
+ "and finally the table number of the interaction type.[PAR]",
+ "The options [TT]-px[tt] and [TT]-pf[tt] are used for writing pull COM",
+ "coordinates and forces when pulling is selected",
+ "in the [TT].mdp[tt] file.[PAR]",
+ "With [TT]-multi[tt] or [TT]-multidir[tt], multiple systems can be ",
+ "simulated in parallel.",
+ "As many input files/directories are required as the number of systems. ",
+ "The [TT]-multidir[tt] option takes a list of directories (one for each ",
+ "system) and runs in each of them, using the input/output file names, ",
+ "such as specified by e.g. the [TT]-s[tt] option, relative to these ",
+ "directories.",
+ "With [TT]-multi[tt], the system number is appended to the run input ",
+ "and each output filename, for instance [TT]topol.tpr[tt] becomes",
+ "[TT]topol0.tpr[tt], [TT]topol1.tpr[tt] etc.",
+ "The number of nodes per system is the total number of nodes",
+ "divided by the number of systems.",
+ "One use of this option is for NMR refinement: when distance",
+ "or orientation restraints are present these can be ensemble averaged",
+ "over all the systems.[PAR]",
+ "With [TT]-replex[tt] replica exchange is attempted every given number",
+ "of steps. The number of replicas is set with the [TT]-multi[tt] or ",
+ "[TT]-multidir[tt] option, described above.",
+ "All run input files should use a different coupling temperature,",
+ "the order of the files is not important. The random seed is set with",
+ "[TT]-reseed[tt]. The velocities are scaled and neighbor searching",
+ "is performed after every exchange.[PAR]",
+ "Finally some experimental algorithms can be tested when the",
+ "appropriate options have been given. Currently under",
+ "investigation are: polarizability.",
+ "[PAR]",
+ "The option [TT]-membed[tt] does what used to be g_membed, i.e. embed",
+ "a protein into a membrane. The data file should contain the options",
+ "that where passed to g_membed before. The [TT]-mn[tt] and [TT]-mp[tt]",
+ "both apply to this as well.",
+ "[PAR]",
+ "The option [TT]-pforce[tt] is useful when you suspect a simulation",
+ "crashes due to too large forces. With this option coordinates and",
+ "forces of atoms with a force larger than a certain value will",
+ "be printed to stderr.",
+ "[PAR]",
+ "Checkpoints containing the complete state of the system are written",
+ "at regular intervals (option [TT]-cpt[tt]) to the file [TT]-cpo[tt],",
+ "unless option [TT]-cpt[tt] is set to -1.",
+ "The previous checkpoint is backed up to [TT]state_prev.cpt[tt] to",
+ "make sure that a recent state of the system is always available,",
+ "even when the simulation is terminated while writing a checkpoint.",
+ "With [TT]-cpnum[tt] all checkpoint files are kept and appended",
+ "with the step number.",
+ "A simulation can be continued by reading the full state from file",
+ "with option [TT]-cpi[tt]. This option is intelligent in the way that",
+ "if no checkpoint file is found, Gromacs just assumes a normal run and",
+ "starts from the first step of the [TT].tpr[tt] file. By default the output",
+ "will be appending to the existing output files. The checkpoint file",
+ "contains checksums of all output files, such that you will never",
+ "loose data when some output files are modified, corrupt or removed.",
+ "There are three scenarios with [TT]-cpi[tt]:[PAR]",
+ "[TT]*[tt] no files with matching names are present: new output files are written[PAR]",
+ "[TT]*[tt] all files are present with names and checksums matching those stored",
+ "in the checkpoint file: files are appended[PAR]",
+ "[TT]*[tt] otherwise no files are modified and a fatal error is generated[PAR]",
+ "With [TT]-noappend[tt] new output files are opened and the simulation",
+ "part number is added to all output file names.",
+ "Note that in all cases the checkpoint file itself is not renamed",
+ "and will be overwritten, unless its name does not match",
+ "the [TT]-cpo[tt] option.",
+ "[PAR]",
+ "With checkpointing the output is appended to previously written",
+ "output files, unless [TT]-noappend[tt] is used or none of the previous",
+ "output files are present (except for the checkpoint file).",
+ "The integrity of the files to be appended is verified using checksums",
+ "which are stored in the checkpoint file. This ensures that output can",
+ "not be mixed up or corrupted due to file appending. When only some",
+ "of the previous output files are present, a fatal error is generated",
+ "and no old output files are modified and no new output files are opened.",
+ "The result with appending will be the same as from a single run.",
+ "The contents will be binary identical, unless you use a different number",
+ "of nodes or dynamic load balancing or the FFT library uses optimizations",
+ "through timing.",
+ "[PAR]",
+ "With option [TT]-maxh[tt] a simulation is terminated and a checkpoint",
+ "file is written at the first neighbor search step where the run time",
+ "exceeds [TT]-maxh[tt]*0.99 hours.",
+ "[PAR]",
+ "When [TT]mdrun[tt] receives a TERM signal, it will set nsteps to the current",
+ "step plus one. When [TT]mdrun[tt] receives an INT signal (e.g. when ctrl+C is",
+ "pressed), it will stop after the next neighbor search step ",
+ "(with nstlist=0 at the next step).",
+ "In both cases all the usual output will be written to file.",
+ "When running with MPI, a signal to one of the [TT]mdrun[tt] processes",
+ "is sufficient, this signal should not be sent to mpirun or",
+ "the [TT]mdrun[tt] process that is the parent of the others.",
+ "[PAR]",
+ "When [TT]mdrun[tt] is started with MPI, it does not run niced by default."
+ };
+ t_commrec *cr;
+ t_filenm fnm[] = {
+ { efTPX, NULL, NULL, ffREAD },
+ { efTRN, "-o", NULL, ffWRITE },
+ { efCOMPRESSED, "-x", NULL, ffOPTWR },
+ { efCPT, "-cpi", NULL, ffOPTRD },
+ { efCPT, "-cpo", NULL, ffOPTWR },
+ { efSTO, "-c", "confout", ffWRITE },
+ { efEDR, "-e", "ener", ffWRITE },
+ { efLOG, "-g", "md", ffWRITE },
+ { efXVG, "-dhdl", "dhdl", ffOPTWR },
+ { efXVG, "-field", "field", ffOPTWR },
+ { efXVG, "-table", "table", ffOPTRD },
+ { efXVG, "-tabletf", "tabletf", ffOPTRD },
+ { efXVG, "-tablep", "tablep", ffOPTRD },
+ { efXVG, "-tableb", "table", ffOPTRD },
+ { efTRX, "-rerun", "rerun", ffOPTRD },
+ { efXVG, "-tpi", "tpi", ffOPTWR },
+ { efXVG, "-tpid", "tpidist", ffOPTWR },
+ { efEDI, "-ei", "sam", ffOPTRD },
+ { efXVG, "-eo", "edsam", ffOPTWR },
+ { efXVG, "-devout", "deviatie", ffOPTWR },
+ { efXVG, "-runav", "runaver", ffOPTWR },
+ { efXVG, "-px", "pullx", ffOPTWR },
+ { efXVG, "-pf", "pullf", ffOPTWR },
+ { efXVG, "-ro", "rotation", ffOPTWR },
+ { efLOG, "-ra", "rotangles", ffOPTWR },
+ { efLOG, "-rs", "rotslabs", ffOPTWR },
+ { efLOG, "-rt", "rottorque", ffOPTWR },
+ { efMTX, "-mtx", "nm", ffOPTWR },
+ { efNDX, "-dn", "dipole", ffOPTWR },
+ { efRND, "-multidir", NULL, ffOPTRDMULT},
+ { efDAT, "-membed", "membed", ffOPTRD },
+ { efTOP, "-mp", "membed", ffOPTRD },
+ { efNDX, "-mn", "membed", ffOPTRD },
+ { efXVG, "-swap", "swapions", ffOPTWR }
+ };
+#define NFILE asize(fnm)
+
+ /* Command line options ! */
+ gmx_bool bPartDec = FALSE;
+ gmx_bool bDDBondCheck = TRUE;
+ gmx_bool bDDBondComm = TRUE;
+ gmx_bool bTunePME = TRUE;
+ gmx_bool bTestVerlet = FALSE;
+ gmx_bool bVerbose = FALSE;
+ gmx_bool bCompact = TRUE;
+ gmx_bool bSepPot = FALSE;
+ gmx_bool bRerunVSite = FALSE;
+ gmx_bool bConfout = TRUE;
+ gmx_bool bReproducible = FALSE;
+
+ int npme = -1;
+ int nstlist = 0;
+ int nmultisim = 0;
+ int nstglobalcomm = -1;
+ int repl_ex_nst = 0;
+ int repl_ex_seed = -1;
+ int repl_ex_nex = 0;
+ int nstepout = 100;
+ int resetstep = -1;
+ gmx_int64_t nsteps = -2; /* the value -2 means that the mdp option will be used */
+
+ rvec realddxyz = {0, 0, 0};
+ const char *ddno_opt[ddnoNR+1] =
+ { NULL, "interleave", "pp_pme", "cartesian", NULL };
+ const char *dddlb_opt[] =
+ { NULL, "auto", "no", "yes", NULL };
+ const char *thread_aff_opt[threadaffNR+1] =
+ { NULL, "auto", "on", "off", NULL };
+ const char *nbpu_opt[] =
+ { NULL, "auto", "cpu", "gpu", "gpu_cpu", NULL };
+ real rdd = 0.0, rconstr = 0.0, dlb_scale = 0.8, pforce = -1;
+ char *ddcsx = NULL, *ddcsy = NULL, *ddcsz = NULL;
+ real cpt_period = 15.0, max_hours = -1;
+ gmx_bool bAppendFiles = TRUE;
+ gmx_bool bKeepAndNumCPT = FALSE;
+ gmx_bool bResetCountersHalfWay = FALSE;
+ output_env_t oenv = NULL;
+ const char *deviceOptions = "";
+
+ /* Non transparent initialization of a complex gmx_hw_opt_t struct.
+ * But unfortunately we are not allowed to call a function here,
+ * since declarations follow below.
+ */
+ gmx_hw_opt_t hw_opt = {
+ 0, 0, 0, 0, threadaffSEL, 0, 0,
+ { NULL, FALSE, 0, NULL }
+ };
+
+ t_pargs pa[] = {
+
+ { "-pd", FALSE, etBOOL, {&bPartDec},
+ "Use particle decompostion" },
+ { "-dd", FALSE, etRVEC, {&realddxyz},
+ "Domain decomposition grid, 0 is optimize" },
+ { "-ddorder", FALSE, etENUM, {ddno_opt},
+ "DD node order" },
+ { "-npme", FALSE, etINT, {&npme},
+ "Number of separate nodes to be used for PME, -1 is guess" },
+ { "-nt", FALSE, etINT, {&hw_opt.nthreads_tot},
+ "Total number of threads to start (0 is guess)" },
+ { "-ntmpi", FALSE, etINT, {&hw_opt.nthreads_tmpi},
+ "Number of thread-MPI threads to start (0 is guess)" },
+ { "-ntomp", FALSE, etINT, {&hw_opt.nthreads_omp},
+ "Number of OpenMP threads per MPI process/thread to start (0 is guess)" },
+ { "-ntomp_pme", FALSE, etINT, {&hw_opt.nthreads_omp_pme},
+ "Number of OpenMP threads per MPI process/thread to start (0 is -ntomp)" },
+ { "-pin", FALSE, etENUM, {thread_aff_opt},
+ "Fix threads (or processes) to specific cores" },
+ { "-pinoffset", FALSE, etINT, {&hw_opt.core_pinning_offset},
+ "The starting logical core number for pinning to cores; used to avoid pinning threads from different mdrun instances to the same core" },
+ { "-pinstride", FALSE, etINT, {&hw_opt.core_pinning_stride},
+ "Pinning distance in logical cores for threads, use 0 to minimize the number of threads per physical core" },
+ { "-gpu_id", FALSE, etSTR, {&hw_opt.gpu_opt.gpu_id},
+ "List of GPU device id-s to use, specifies the per-node PP rank to GPU mapping" },
+ { "-ddcheck", FALSE, etBOOL, {&bDDBondCheck},
+ "Check for all bonded interactions with DD" },
+ { "-ddbondcomm", FALSE, etBOOL, {&bDDBondComm},
+ "HIDDENUse special bonded atom communication when [TT]-rdd[tt] > cut-off" },
+ { "-rdd", FALSE, etREAL, {&rdd},
+ "The maximum distance for bonded interactions with DD (nm), 0 is determine from initial coordinates" },
+ { "-rcon", FALSE, etREAL, {&rconstr},
+ "Maximum distance for P-LINCS (nm), 0 is estimate" },
+ { "-dlb", FALSE, etENUM, {dddlb_opt},
+ "Dynamic load balancing (with DD)" },
+ { "-dds", FALSE, etREAL, {&dlb_scale},
+ "Minimum allowed dlb scaling of the DD cell size" },
+ { "-ddcsx", FALSE, etSTR, {&ddcsx},
+ "HIDDENThe DD cell sizes in x" },
+ { "-ddcsy", FALSE, etSTR, {&ddcsy},
+ "HIDDENThe DD cell sizes in y" },
+ { "-ddcsz", FALSE, etSTR, {&ddcsz},
+ "HIDDENThe DD cell sizes in z" },
+ { "-gcom", FALSE, etINT, {&nstglobalcomm},
+ "Global communication frequency" },
+ { "-nb", FALSE, etENUM, {&nbpu_opt},
+ "Calculate non-bonded interactions on" },
+ { "-nstlist", FALSE, etINT, {&nstlist},
+ "Set nstlist when using a Verlet buffer tolerance (0 is guess)" },
+ { "-tunepme", FALSE, etBOOL, {&bTunePME},
+ "Optimize PME load between PP/PME nodes or GPU/CPU" },
+ { "-testverlet", FALSE, etBOOL, {&bTestVerlet},
+ "Test the Verlet non-bonded scheme" },
+ { "-v", FALSE, etBOOL, {&bVerbose},
+ "Be loud and noisy" },
+ { "-compact", FALSE, etBOOL, {&bCompact},
+ "Write a compact log file" },
+ { "-seppot", FALSE, etBOOL, {&bSepPot},
+ "Write separate V and dVdl terms for each interaction type and node to the log file(s)" },
+ { "-pforce", FALSE, etREAL, {&pforce},
+ "Print all forces larger than this (kJ/mol nm)" },
+ { "-reprod", FALSE, etBOOL, {&bReproducible},
+ "Try to avoid optimizations that affect binary reproducibility" },
+ { "-cpt", FALSE, etREAL, {&cpt_period},
+ "Checkpoint interval (minutes)" },
+ { "-cpnum", FALSE, etBOOL, {&bKeepAndNumCPT},
+ "Keep and number checkpoint files" },
+ { "-append", FALSE, etBOOL, {&bAppendFiles},
+ "Append to previous output files when continuing from checkpoint instead of adding the simulation part number to all file names" },
+ { "-nsteps", FALSE, etINT64, {&nsteps},
+ "Run this number of steps, overrides .mdp file option" },
+ { "-maxh", FALSE, etREAL, {&max_hours},
+ "Terminate after 0.99 times this time (hours)" },
+ { "-multi", FALSE, etINT, {&nmultisim},
+ "Do multiple simulations in parallel" },
+ { "-replex", FALSE, etINT, {&repl_ex_nst},
+ "Attempt replica exchange periodically with this period (steps)" },
+ { "-nex", FALSE, etINT, {&repl_ex_nex},
+ "Number of random exchanges to carry out each exchange interval (N^3 is one suggestion). -nex zero or not specified gives neighbor replica exchange." },
+ { "-reseed", FALSE, etINT, {&repl_ex_seed},
+ "Seed for replica exchange, -1 is generate a seed" },
+ { "-rerunvsite", FALSE, etBOOL, {&bRerunVSite},
+ "HIDDENRecalculate virtual site coordinates with [TT]-rerun[tt]" },
+ { "-confout", FALSE, etBOOL, {&bConfout},
+ "HIDDENWrite the last configuration with [TT]-c[tt] and force checkpointing at the last step" },
+ { "-stepout", FALSE, etINT, {&nstepout},
+ "HIDDENFrequency of writing the remaining wall clock time for the run" },
+ { "-resetstep", FALSE, etINT, {&resetstep},
+ "HIDDENReset cycle counters after these many time steps" },
+ { "-resethway", FALSE, etBOOL, {&bResetCountersHalfWay},
+ "HIDDENReset the cycle counters after half the number of steps or halfway [TT]-maxh[tt]" }
+ };
+ unsigned long Flags, PCA_Flags;
+ ivec ddxyz;
+ int dd_node_order;
+ gmx_bool bAddPart;
+ FILE *fplog, *fpmulti;
+ int sim_part, sim_part_fn;
+ const char *part_suffix = ".part";
+ char suffix[STRLEN];
+ int rc;
+ char **multidir = NULL;
+
+
+ cr = init_commrec();
+
+ PCA_Flags = (PCA_CAN_SET_DEFFNM | (MASTER(cr) ? 0 : PCA_QUIET));
+
+ /* Comment this in to do fexist calls only on master
+ * works not with rerun or tables at the moment
+ * also comment out the version of init_forcerec in md.c
+ * with NULL instead of opt2fn
+ */
+ /*
+ if (!MASTER(cr))
+ {
+ PCA_Flags |= PCA_NOT_READ_NODE;
+ }
+ */
+
+ if (!parse_common_args(&argc, argv, PCA_Flags, NFILE, fnm, asize(pa), pa,
+ asize(desc), desc, 0, NULL, &oenv))
+ {
+ return 0;
+ }
+
+
+ /* we set these early because they might be used in init_multisystem()
+ Note that there is the potential for npme>nnodes until the number of
+ threads is set later on, if there's thread parallelization. That shouldn't
+ lead to problems. */
+ dd_node_order = nenum(ddno_opt);
+ cr->npmenodes = npme;
+
+ hw_opt.thread_affinity = nenum(thread_aff_opt);
+
+ /* now check the -multi and -multidir option */
+ if (opt2bSet("-multidir", NFILE, fnm))
+ {
+ if (nmultisim > 0)
+ {
+ gmx_fatal(FARGS, "mdrun -multi and -multidir options are mutually exclusive.");
+ }
+ nmultisim = opt2fns(&multidir, "-multidir", NFILE, fnm);
+ }
+
+
+ if (repl_ex_nst != 0 && nmultisim < 2)
+ {
+ gmx_fatal(FARGS, "Need at least two replicas for replica exchange (option -multi)");
+ }
+
+ if (repl_ex_nex < 0)
+ {
+ gmx_fatal(FARGS, "Replica exchange number of exchanges needs to be positive");
+ }
+
+ if (nmultisim > 1)
+ {
+#ifndef GMX_THREAD_MPI
+ gmx_bool bParFn = (multidir == NULL);
+ init_multisystem(cr, nmultisim, multidir, NFILE, fnm, bParFn);
+#else
+ gmx_fatal(FARGS, "mdrun -multi is not supported with the thread library. "
+ "Please compile GROMACS with MPI support");
+#endif
+ }
+
+ bAddPart = !bAppendFiles;
+
+ /* Check if there is ANY checkpoint file available */
+ sim_part = 1;
+ sim_part_fn = sim_part;
+ if (opt2bSet("-cpi", NFILE, fnm))
+ {
+ if (bSepPot && bAppendFiles)
+ {
+ gmx_fatal(FARGS, "Output file appending is not supported with -seppot");
+ }
+
+ bAppendFiles =
+ read_checkpoint_simulation_part(opt2fn_master("-cpi", NFILE,
+ fnm, cr),
+ &sim_part_fn, NULL, cr,
+ bAppendFiles, NFILE, fnm,
+ part_suffix, &bAddPart);
+ if (sim_part_fn == 0 && MULTIMASTER(cr))
+ {
+ fprintf(stdout, "No previous checkpoint file present, assuming this is a new run.\n");
+ }
+ else
+ {
+ sim_part = sim_part_fn + 1;
+ }
+
+ if (MULTISIM(cr) && MASTER(cr))
+ {
+ if (MULTIMASTER(cr))
+ {
+ /* Log file is not yet available, so if there's a
+ * problem we can only write to stderr. */
+ fpmulti = stderr;
+ }
+ else
+ {
+ fpmulti = NULL;
+ }
+ check_multi_int(fpmulti, cr->ms, sim_part, "simulation part", TRUE);
+ }
+ }
+ else
+ {
+ bAppendFiles = FALSE;
+ }
+
+ if (!bAppendFiles)
+ {
+ sim_part_fn = sim_part;
+ }
+
+ if (bAddPart)
+ {
+ /* Rename all output files (except checkpoint files) */
+ /* create new part name first (zero-filled) */
+ sprintf(suffix, "%s%04d", part_suffix, sim_part_fn);
+
+ add_suffix_to_output_names(fnm, NFILE, suffix);
+ if (MULTIMASTER(cr))
+ {
+ fprintf(stdout, "Checkpoint file is from part %d, new output files will be suffixed '%s'.\n", sim_part-1, suffix);
+ }
+ }
+
+ Flags = opt2bSet("-rerun", NFILE, fnm) ? MD_RERUN : 0;
+ Flags = Flags | (bSepPot ? MD_SEPPOT : 0);
+ Flags = Flags | (bPartDec ? MD_PARTDEC : 0);
+ Flags = Flags | (bDDBondCheck ? MD_DDBONDCHECK : 0);
+ Flags = Flags | (bDDBondComm ? MD_DDBONDCOMM : 0);
+ Flags = Flags | (bTunePME ? MD_TUNEPME : 0);
+ Flags = Flags | (bTestVerlet ? MD_TESTVERLET : 0);
+ Flags = Flags | (bConfout ? MD_CONFOUT : 0);
+ Flags = Flags | (bRerunVSite ? MD_RERUN_VSITE : 0);
+ Flags = Flags | (bReproducible ? MD_REPRODUCIBLE : 0);
+ Flags = Flags | (bAppendFiles ? MD_APPENDFILES : 0);
+ Flags = Flags | (opt2parg_bSet("-append", asize(pa), pa) ? MD_APPENDFILESSET : 0);
+ Flags = Flags | (bKeepAndNumCPT ? MD_KEEPANDNUMCPT : 0);
+ Flags = Flags | (sim_part > 1 ? MD_STARTFROMCPT : 0);
+ Flags = Flags | (bResetCountersHalfWay ? MD_RESETCOUNTERSHALFWAY : 0);
+
+
+ /* We postpone opening the log file if we are appending, so we can
+ first truncate the old log file and append to the correct position
+ there instead. */
+ if ((MASTER(cr) || bSepPot) && !bAppendFiles)
+ {
+ gmx_log_open(ftp2fn(efLOG, NFILE, fnm), cr,
+ !bSepPot, Flags & MD_APPENDFILES, &fplog);
+ please_cite(fplog, "Hess2008b");
+ please_cite(fplog, "Spoel2005a");
+ please_cite(fplog, "Lindahl2001a");
+ please_cite(fplog, "Berendsen95a");
+ }
+ else if (!MASTER(cr) && bSepPot)
+ {
+ gmx_log_open(ftp2fn(efLOG, NFILE, fnm), cr, !bSepPot, Flags, &fplog);
+ }
+ else
+ {
+ fplog = NULL;
+ }
+
+ ddxyz[XX] = (int)(realddxyz[XX] + 0.5);
+ ddxyz[YY] = (int)(realddxyz[YY] + 0.5);
+ ddxyz[ZZ] = (int)(realddxyz[ZZ] + 0.5);
+
+ rc = mdrunner(&hw_opt, fplog, cr, NFILE, fnm, oenv, bVerbose, bCompact,
+ nstglobalcomm, ddxyz, dd_node_order, rdd, rconstr,
+ dddlb_opt[0], dlb_scale, ddcsx, ddcsy, ddcsz,
+ nbpu_opt[0], nstlist,
+ nsteps, nstepout, resetstep,
+ nmultisim, repl_ex_nst, repl_ex_nex, repl_ex_seed,
+ pforce, cpt_period, max_hours, deviceOptions, Flags);
+
+ /* Log file has to be closed in mdrunner if we are appending to it
+ (fplog not set here) */
+ if (MASTER(cr) && !bAppendFiles)
+ {
+ gmx_log_close(fplog);
+ }
+
+ return rc;
+}