void list_xtc(const char *fn)
{
t_fileio *xd;
int indent;
char buf[256];
rvec *x;
matrix box;
int nframe, natoms, step;
real prec, time;
gmx_bool bOK;
xd = open_xtc(fn, "r");
read_first_xtc(xd, &natoms, &step, &time, box, &x, &prec, &bOK);
nframe = 0;
do
{
sprintf(buf, "%s frame %d", fn, nframe);
indent = 0;
indent = pr_title(stdout, indent, buf);
pr_indent(stdout, indent);
fprintf(stdout, "natoms=%10d step=%10d time=%12.7e prec=%10g\n",
natoms, step, time, prec);
pr_rvecs(stdout, indent, "box", box, DIM);
pr_rvecs(stdout, indent, "x", x, natoms);
nframe++;
}
while (read_next_xtc(xd, natoms, &step, &time, box, x, &prec, &bOK));
if (!bOK)
{
fprintf(stderr, "\nWARNING: Incomplete frame at time %g\n", time);
}
sfree(x);
close_xtc(xd);
}
/* Check whether the box is unchanged */
static void check_box_unchanged(matrix f_box, matrix box, char fn[], warninp_t wi)
{
int i, ii;
gmx_bool bSame = TRUE;
char warn_buf[STRLEN];
for (i = 0; i < DIM; i++)
{
for (ii = 0; ii < DIM; ii++)
{
if (f_box[i][ii] != box[i][ii])
{
bSame = FALSE;
}
}
}
if (!bSame)
{
sprintf(warn_buf, "%s Box size in reference file %s differs from actual box size!",
RotStr, fn);
warning(wi, warn_buf);
pr_rvecs(stderr, 0, "Your box is:", box, 3);
pr_rvecs(stderr, 0, "Box in file:", f_box, 3);
}
}
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]);
}
static void list_trn(char *fn)
{
int fpread,fpwrite,nframe,indent;
char buf[256];
rvec *x,*v,*f;
matrix box;
t_trnheader trn;
bool bOK;
fpread = open_trn(fn,"r");
fpwrite = open_tpx(NULL,"w");
gmx_fio_setdebug(fpwrite,TRUE);
nframe = 0;
while (fread_trnheader(fpread,&trn,&bOK)) {
snew(x,trn.natoms);
snew(v,trn.natoms);
snew(f,trn.natoms);
if (fread_htrn(fpread,&trn,
trn.box_size ? box : NULL,
trn.x_size ? x : NULL,
trn.v_size ? v : NULL,
trn.f_size ? f : NULL)) {
sprintf(buf,"%s frame %d",fn,nframe);
indent=0;
indent=pr_title(stdout,indent,buf);
pr_indent(stdout,indent);
fprintf(stdout,"natoms=%10d step=%10d time=%12.7e lambda=%10g\n",
trn.natoms,trn.step,trn.t,trn.lambda);
if (trn.box_size)
pr_rvecs(stdout,indent,"box",box,DIM);
if (trn.x_size)
pr_rvecs(stdout,indent,"x",x,trn.natoms);
if (trn.v_size)
pr_rvecs(stdout,indent,"v",v,trn.natoms);
if (trn.f_size)
pr_rvecs(stdout,indent,"f",f,trn.natoms);
}
else
fprintf(stderr,"\nWARNING: Incomplete frame: nr %d, t=%g\n",
nframe,trn.t);
sfree(x);
sfree(v);
sfree(f);
nframe++;
}
if (!bOK)
fprintf(stderr,"\nWARNING: Incomplete frame header: nr %d, t=%g\n",
nframe,trn.t);
close_tpx(fpwrite);
close_trn(fpread);
}
static void dump_it_all(FILE *fp,char *title,
int natoms,rvec x[],rvec xp[],rvec v[],rvec f[])
{
#ifdef DEBUG
if (fp) {
fprintf(fp,"%s\n",title);
pr_rvecs(fp,0,"x",x,natoms);
pr_rvecs(fp,0,"xp",xp,natoms);
pr_rvecs(fp,0,"v",v,natoms);
pr_rvecs(fp,0,"f",f,natoms);
}
#endif
}
static int correct_box_elem(FILE *fplog, int step, tensor box, int v, int d)
{
int shift, maxshift = 10;
shift = 0;
/* correct elem d of vector v with vector d */
while (box[v][d] > BOX_MARGIN_CORRECT*0.5*box[d][d])
{
if (fplog)
{
fprintf(fplog, "Step %d: correcting invalid box:\n", step);
pr_rvecs(fplog, 0, "old box", box, DIM);
}
rvec_dec(box[v], box[d]);
shift--;
if (fplog)
{
pr_rvecs(fplog, 0, "new box", box, DIM);
}
if (shift <= -maxshift)
{
gmx_fatal(FARGS,
"Box was shifted at least %d times. Please see log-file.",
maxshift);
}
}
while (box[v][d] < -BOX_MARGIN_CORRECT*0.5*box[d][d])
{
if (fplog)
{
fprintf(fplog, "Step %d: correcting invalid box:\n", step);
pr_rvecs(fplog, 0, "old box", box, DIM);
}
rvec_inc(box[v], box[d]);
shift++;
if (fplog)
{
pr_rvecs(fplog, 0, "new box", box, DIM);
}
if (shift >= maxshift)
{
gmx_fatal(FARGS,
"Box was shifted at least %d times. Please see log-file.",
maxshift);
}
}
return shift;
}
void calc_force(int natom,rvec f[],rvec fff[])
{
int i,j,m;
int jindex[] = { 0, 5, 10};
rvec dx,df;
real msf1,msf2;
for(j=0; (j<2); j++) {
clear_rvec(fff[j]);
for(i=jindex[j]; (i<jindex[j+1]); i++) {
for(m=0; (m<DIM); m++) {
fff[j][m] += f[i][m];
}
}
}
msf1 = iprod(fff[0],fff[0]);
msf2 = iprod(fff[1],fff[1]);
if (debug) {
pr_rvecs(debug,0,"force",f,natom);
fprintf(debug,"FMOL: %10.3f %10.3f %10.3f %10.3f %10.3f %10.3f\n",
fff[0][XX],fff[0][YY],fff[0][ZZ],fff[1][XX],fff[1][YY],fff[1][ZZ]);
fprintf(debug,"RMSF: %10.3e %10.3e\n",msf1,msf2);
}
}
static void calc_virial(FILE *fplog,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(fplog,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(fplog,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 pr_matrix(FILE *fp,int indent,const char *title,rvec *m,
gmx_bool bMDPformat)
{
if (bMDPformat)
fprintf(fp,"%-10s = %g %g %g %g %g %g\n",title,
m[XX][XX],m[YY][YY],m[ZZ][ZZ],m[XX][YY],m[XX][ZZ],m[YY][ZZ]);
else
pr_rvecs(fp,indent,title,m,DIM);
}
void init_parallel(FILE *log,char *tpxfile,t_commrec *cr,
t_inputrec *inputrec,gmx_mtop_t *mtop,
t_state *state,
int list)
{
int step;
real t;
char buf[256];
if (MASTER(cr)) {
init_inputrec(inputrec);
read_tpx_state(tpxfile,&step,&t,inputrec,state,NULL,mtop);
/* When we will be doing domain decomposition with separate PME nodes
* the rng entries will be too large, we correct for this later.
*/
set_state_entries(state,inputrec,cr->nnodes);
}
bcast_ir_mtop(cr,inputrec,mtop);
if (inputrec->eI == eiBD || EI_SD(inputrec->eI)) {
/* Make sure the random seeds are different on each node */
inputrec->ld_seed += cr->nodeid;
}
/* Printing */
if (list!=0 && log!=NULL)
{
if (list&LIST_INPUTREC)
pr_inputrec(log,0,"parameters of the run",inputrec,FALSE);
if (list&LIST_X)
pr_rvecs(log,0,"box",state->box,DIM);
if (list&LIST_X)
pr_rvecs(log,0,"box_rel",state->box_rel,DIM);
if (list&LIST_V)
pr_rvecs(log,0,"boxv",state->boxv,DIM);
if (list&LIST_X)
pr_rvecs(log,0,int_title("x",0,buf,255),state->x,state->natoms);
if (list&LIST_V)
pr_rvecs(log,0,int_title("v",0,buf,255),state->v,state->natoms);
if (list&LIST_TOP)
pr_mtop(log,0,int_title("topology",cr->nodeid,buf,255),mtop,TRUE);
fflush(log);
}
}
static void pr_molblock(FILE *fp,int indent,const char *title,
gmx_molblock_t *molb,int n,
gmx_moltype_t *molt,
gmx_bool bShowNumbers)
{
indent = pr_title_n(fp,indent,title,n);
(void) pr_indent(fp,indent);
(void) fprintf(fp,"%-20s = %d \"%s\"\n",
"moltype",molb->type,*(molt[molb->type].name));
pr_int(fp,indent,"#molecules",molb->nmol);
pr_int(fp,indent,"#atoms_mol",molb->natoms_mol);
pr_int(fp,indent,"#posres_xA",molb->nposres_xA);
if (molb->nposres_xA > 0) {
pr_rvecs(fp,indent,"posres_xA",molb->posres_xA,molb->nposres_xA);
}
pr_int(fp,indent,"#posres_xB",molb->nposres_xB);
if (molb->nposres_xB > 0) {
pr_rvecs(fp,indent,"posres_xB",molb->posres_xB,molb->nposres_xB);
}
}
static void check_pbc(FILE *fp,rvec x[],int shell)
{
int m,now;
now = shell-4;
for(m=0; (m<DIM); m++)
if (fabs(x[shell][m]-x[now][m]) > 0.3) {
pr_rvecs(fp,0,"SHELL-X",x+now,5);
break;
}
}
real calc_pres(int ePBC,int nwall,matrix box,tensor ekin,tensor vir,
tensor pres,real Elr)
{
int n,m;
real fac,Plr;
if (ePBC==epbcNONE || (ePBC==epbcXY && nwall!=2))
clear_mat(pres);
else {
/* Uitzoeken welke ekin hier van toepassing is, zie Evans & Morris - E.
* Wrs. moet de druktensor gecorrigeerd worden voor de netto stroom in
* het systeem...
*/
/* Long range correction for periodic systems, see
* Neumann et al. JCP
* divide by 6 because it is multiplied by fac later on.
* If Elr = 0, no correction is made.
*/
/* This formula should not be used with Ewald or PME,
* where the full long-range virial is calculated. EL 990823
*/
Plr = Elr/6.0;
fac=PRESFAC*2.0/det(box);
for(n=0; (n<DIM); n++)
for(m=0; (m<DIM); m++)
pres[n][m]=(ekin[n][m]-vir[n][m]+Plr)*fac;
if (debug) {
pr_rvecs(debug,0,"PC: pres",pres,DIM);
pr_rvecs(debug,0,"PC: ekin",ekin,DIM);
pr_rvecs(debug,0,"PC: vir ",vir, DIM);
pr_rvecs(debug,0,"PC: box ",box, DIM);
}
}
return trace(pres)/3.0;
}
void list_xtc(char *fn, bool bXVG)
{
int xd,indent;
char buf[256];
rvec *x;
matrix box;
int nframe,natoms,step;
real prec,time;
bool bOK;
xd = open_xtc(fn,"r");
read_first_xtc(xd,&natoms,&step,&time,box,&x,&prec,&bOK);
nframe=0;
do {
if (bXVG) {
int i,d;
fprintf(stdout,"%g",time);
for(i=0; i<natoms; i++)
for(d=0; d<DIM; d++)
fprintf(stdout," %g",x[i][d]);
fprintf(stdout,"\n");
} else {
sprintf(buf,"%s frame %d",fn,nframe);
indent=0;
indent=pr_title(stdout,indent,buf);
pr_indent(stdout,indent);
fprintf(stdout,"natoms=%10d step=%10d time=%12.7e prec=%10g\n",
natoms,step,time,prec);
pr_rvecs(stdout,indent,"box",box,DIM);
pr_rvecs(stdout,indent,"x",x,natoms);
}
nframe++;
} while (read_next_xtc(xd,natoms,&step,&time,box,x,&prec,&bOK));
if (!bOK)
fprintf(stderr,"\nWARNING: Incomplete frame at time %g\n",time);
close_xtc(xd);
}
real calc_gyro(rvec x[], int gnx, atom_id index[], t_atom atom[], real tm,
rvec gvec, rvec d, gmx_bool bQ, gmx_bool bRot, gmx_bool bMOI, matrix trans)
{
int i, ii, m;
real gyro, dx2, m0, Itot;
rvec comp;
if (bRot)
{
principal_comp(gnx, index, atom, x, trans, d);
Itot = norm(d);
if (bMOI)
{
return Itot;
}
for (m = 0; (m < DIM); m++)
{
d[m] = std::sqrt(d[m]/tm);
}
#ifdef DEBUG
pr_rvecs(stderr, 0, "trans", trans, DIM);
#endif
/* rotate_atoms(gnx,index,x,trans); */
}
clear_rvec(comp);
for (i = 0; (i < gnx); i++)
{
ii = index[i];
if (bQ)
{
m0 = std::abs(atom[ii].q);
}
else
{
m0 = atom[ii].m;
}
for (m = 0; (m < DIM); m++)
{
dx2 = x[ii][m]*x[ii][m];
comp[m] += dx2*m0;
}
}
gyro = comp[XX]+comp[YY]+comp[ZZ];
for (m = 0; (m < DIM); m++)
{
gvec[m] = std::sqrt((gyro-comp[m])/tm);
}
return std::sqrt(gyro/tm);
}
void calc_virial(FILE *log,int start,int homenr,rvec x[],rvec f[],
tensor vir_part,tensor pme_vir,
t_graph *graph,matrix box,
t_nrnb *nrnb,t_forcerec *fr,bool bTweak)
{
int i,j;
tensor virtest;
/* Now it is time for the short range virial. At this timepoint vir_part
* already contains the virial from surrounding boxes.
* 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(log,start,start+homenr,x,f,vir_part,graph,box);
inc_nrnb(nrnb,eNR_VIRIAL,homenr);
/* Add up the long range forces if necessary */
/* if (!bTweak) {
sum_lrforces(f,fr,start,homenr);
}*/
/* Add up virial if necessary */
if (EEL_LR(fr->eeltype) && (fr->eeltype != eelPPPM)) {
if (debug && bTweak) {
clear_mat(virtest);
f_calc_vir(log,start,start+homenr,x,fr->f_pme,virtest,graph,box);
pr_rvecs(debug,0,"virtest",virtest,DIM);
pr_rvecs(debug,0,"pme_vir",pme_vir,DIM);
}
/* PPPM virial sucks */
if (!bTweak)
for(i=0; (i<DIM); i++)
for(j=0; (j<DIM); j++)
vir_part[i][j]+=pme_vir[i][j];
}
if (debug)
pr_rvecs(debug,0,"vir_part",vir_part,DIM);
}
void calc_angles_dihs(t_params *ang, t_params *dih, rvec x[], gmx_bool bPBC,
matrix box)
{
int i, ai, aj, ak, al, t1, t2, t3;
rvec r_ij, r_kj, r_kl, m, n;
real sign, th, costh, ph;
t_pbc pbc;
if (bPBC)
{
set_pbc(&pbc, epbcXYZ, box);
}
if (debug)
{
pr_rvecs(debug, 0, "X2TOP", box, DIM);
}
for (i = 0; (i < ang->nr); i++)
{
ai = ang->param[i].AI;
aj = ang->param[i].AJ;
ak = ang->param[i].AK;
th = RAD2DEG*bond_angle(x[ai], x[aj], x[ak], bPBC ? &pbc : NULL,
r_ij, r_kj, &costh, &t1, &t2);
if (debug)
{
fprintf(debug, "X2TOP: ai=%3d aj=%3d ak=%3d r_ij=%8.3f r_kj=%8.3f th=%8.3f\n",
ai, aj, ak, norm(r_ij), norm(r_kj), th);
}
ang->param[i].C0 = th;
}
for (i = 0; (i < dih->nr); i++)
{
ai = dih->param[i].AI;
aj = dih->param[i].AJ;
ak = dih->param[i].AK;
al = dih->param[i].AL;
ph = RAD2DEG*dih_angle(x[ai], x[aj], x[ak], x[al], bPBC ? &pbc : NULL,
r_ij, r_kj, r_kl, m, n, &sign, &t1, &t2, &t3);
if (debug)
{
fprintf(debug, "X2TOP: ai=%3d aj=%3d ak=%3d al=%3d r_ij=%8.3f r_kj=%8.3f r_kl=%8.3f ph=%8.3f\n",
ai, aj, ak, al, norm(r_ij), norm(r_kj), norm(r_kl), ph);
}
dih->param[i].C0 = ph;
}
}
static void pr_rotgrp(FILE *fp,int indent,int g,t_rotgrp *rotg)
{
pr_indent(fp,indent);
fprintf(fp,"rotation_group %d:\n",g);
indent += 2;
PS("type",EROTGEOM(rotg->eType));
PS("massw",BOOL(rotg->bMassW));
pr_ivec_block(fp,indent,"atom",rotg->ind,rotg->nat,TRUE);
pr_rvecs(fp,indent,"x_ref",rotg->x_ref,rotg->nat);
pr_rvec(fp,indent,"vec",rotg->vec,DIM,TRUE);
pr_rvec(fp,indent,"pivot",rotg->pivot,DIM,TRUE);
PR("rate",rotg->rate);
PR("k",rotg->k);
PR("slab_dist",rotg->slab_dist);
PR("min_gaussian",rotg->min_gaussian);
PR("epsilon",rotg->eps);
PS("fit_method",EROTFIT(rotg->eFittype));
PI("potfitangle_nstep",rotg->PotAngle_nstep);
PR("potfitangle_step",rotg->PotAngle_step);
}
void dump_pbc(FILE *fp, t_pbc *pbc)
{
rvec sum_box;
fprintf(fp, "ePBCDX = %d\n", pbc->ePBCDX);
pr_rvecs(fp, 0, "box", pbc->box, DIM);
pr_rvecs(fp, 0, "fbox_diag", &pbc->fbox_diag, 1);
pr_rvecs(fp, 0, "hbox_diag", &pbc->hbox_diag, 1);
pr_rvecs(fp, 0, "mhbox_diag", &pbc->mhbox_diag, 1);
rvec_add(pbc->hbox_diag, pbc->mhbox_diag, sum_box);
pr_rvecs(fp, 0, "sum of the above two", &sum_box, 1);
fprintf(fp, "max_cutoff2 = %g\n", pbc->max_cutoff2);
fprintf(fp, "ntric_vec = %d\n", pbc->ntric_vec);
if (pbc->ntric_vec > 0)
{
pr_ivecs(fp, 0, "tric_shift", pbc->tric_shift, pbc->ntric_vec, FALSE);
pr_rvecs(fp, 0, "tric_vec", pbc->tric_vec, pbc->ntric_vec);
}
}
static void list_trr(const char *fn)
{
t_fileio *fpread;
int nframe, indent;
char buf[256];
rvec *x, *v, *f;
matrix box;
gmx_trr_header_t trrheader;
gmx_bool bOK;
fpread = gmx_trr_open(fn, "r");
nframe = 0;
while (gmx_trr_read_frame_header(fpread, &trrheader, &bOK))
{
snew(x, trrheader.natoms);
snew(v, trrheader.natoms);
snew(f, trrheader.natoms);
if (gmx_trr_read_frame_data(fpread, &trrheader,
trrheader.box_size ? box : NULL,
trrheader.x_size ? x : NULL,
trrheader.v_size ? v : NULL,
trrheader.f_size ? f : NULL))
{
sprintf(buf, "%s frame %d", fn, nframe);
indent = 0;
indent = pr_title(stdout, indent, buf);
pr_indent(stdout, indent);
fprintf(stdout, "natoms=%10d step=%10d time=%12.7e lambda=%10g\n",
trrheader.natoms, trrheader.step, trrheader.t, trrheader.lambda);
if (trrheader.box_size)
{
pr_rvecs(stdout, indent, "box", box, DIM);
}
if (trrheader.x_size)
{
pr_rvecs(stdout, indent, "x", x, trrheader.natoms);
}
if (trrheader.v_size)
{
pr_rvecs(stdout, indent, "v", v, trrheader.natoms);
}
if (trrheader.f_size)
{
pr_rvecs(stdout, indent, "f", f, trrheader.natoms);
}
}
else
{
fprintf(stderr, "\nWARNING: Incomplete frame: nr %d, t=%g\n",
nframe, trrheader.t);
}
sfree(x);
sfree(v);
sfree(f);
nframe++;
}
if (!bOK)
{
fprintf(stderr, "\nWARNING: Incomplete frame header: nr %d, t=%g\n",
nframe, trrheader.t);
}
gmx_trr_close(fpread);
}
void check_cm_grp(FILE *fp, t_vcm *vcm, t_inputrec *ir, real Temp_Max)
{
int m, g;
real ekcm, ekrot, tm, tm_1, Temp_cm;
rvec jcm;
tensor Icm;
/* First analyse the total results */
if (vcm->mode != ecmNO)
{
for (g = 0; (g < vcm->nr); g++)
{
tm = vcm->group_mass[g];
if (tm != 0)
{
tm_1 = 1.0/tm;
svmul(tm_1, vcm->group_p[g], vcm->group_v[g]);
}
/* Else it's zero anyway! */
}
if (vcm->mode == ecmANGULAR)
{
for (g = 0; (g < vcm->nr); g++)
{
tm = vcm->group_mass[g];
if (tm != 0)
{
tm_1 = 1.0/tm;
/* Compute center of mass for this group */
for (m = 0; (m < DIM); m++)
{
vcm->group_x[g][m] *= tm_1;
}
/* Subtract the center of mass contribution to the
* angular momentum
*/
cprod(vcm->group_x[g], vcm->group_v[g], jcm);
for (m = 0; (m < DIM); m++)
{
vcm->group_j[g][m] -= tm*jcm[m];
}
/* Subtract the center of mass contribution from the inertia
* tensor (this is not as trivial as it seems, but due to
* some cancellation we can still do it, even in parallel).
*/
clear_mat(Icm);
update_tensor(vcm->group_x[g], tm, Icm);
m_sub(vcm->group_i[g], Icm, vcm->group_i[g]);
/* Compute angular velocity, using matrix operation
* Since J = I w
* we have
* w = I^-1 J
*/
get_minv(vcm->group_i[g], Icm);
mvmul(Icm, vcm->group_j[g], vcm->group_w[g]);
}
/* Else it's zero anyway! */
}
}
}
for (g = 0; (g < vcm->nr); g++)
{
ekcm = 0;
if (vcm->group_mass[g] != 0 && vcm->group_ndf[g] > 0)
{
for (m = 0; m < vcm->ndim; m++)
{
ekcm += sqr(vcm->group_v[g][m]);
}
ekcm *= 0.5*vcm->group_mass[g];
Temp_cm = 2*ekcm/vcm->group_ndf[g];
if ((Temp_cm > Temp_Max) && fp)
{
fprintf(fp, "Large VCM(group %s): %12.5f, %12.5f, %12.5f, Temp-cm: %12.5e\n",
vcm->group_name[g], vcm->group_v[g][XX],
vcm->group_v[g][YY], vcm->group_v[g][ZZ], Temp_cm);
}
if (vcm->mode == ecmANGULAR)
{
ekrot = 0.5*iprod(vcm->group_j[g], vcm->group_w[g]);
if ((ekrot > 1) && fp && !EI_RANDOM(ir->eI))
{
/* if we have an integrator that may not conserve momenta, skip */
tm = vcm->group_mass[g];
fprintf(fp, "Group %s with mass %12.5e, Ekrot %12.5e Det(I) = %12.5e\n",
vcm->group_name[g], tm, ekrot, det(vcm->group_i[g]));
fprintf(fp, " COM: %12.5f %12.5f %12.5f\n",
vcm->group_x[g][XX], vcm->group_x[g][YY], vcm->group_x[g][ZZ]);
fprintf(fp, " P: %12.5f %12.5f %12.5f\n",
vcm->group_p[g][XX], vcm->group_p[g][YY], vcm->group_p[g][ZZ]);
fprintf(fp, " V: %12.5f %12.5f %12.5f\n",
vcm->group_v[g][XX], vcm->group_v[g][YY], vcm->group_v[g][ZZ]);
fprintf(fp, " J: %12.5f %12.5f %12.5f\n",
vcm->group_j[g][XX], vcm->group_j[g][YY], vcm->group_j[g][ZZ]);
fprintf(fp, " w: %12.5f %12.5f %12.5f\n",
vcm->group_w[g][XX], vcm->group_w[g][YY], vcm->group_w[g][ZZ]);
pr_rvecs(fp, 0, "Inertia tensor", vcm->group_i[g], DIM);
}
}
}
}
}
static void do_my_pme(FILE *fp,real tm,gmx_bool bVerbose,t_inputrec *ir,
rvec x[],rvec xbuf[],rvec f[],
real charge[],real qbuf[],real qqbuf[],
matrix box,gmx_bool bSort,
t_commrec *cr,t_nsborder *nsb,t_nrnb *nrnb,
t_block *excl,real qtot,
t_forcerec *fr,int index[],FILE *fp_xvg,
int ngroups,unsigned short cENER[])
{
real ener,vcorr,q,xx,dvdl=0,vdip,vcharge;
tensor vir,vir_corr,vir_tot;
rvec mu_tot[2];
int i,m,ii,ig,jg;
real **epme,*qptr;
/* Initiate local variables */
fr->f_el_recip = f;
clear_mat(vir);
clear_mat(vir_corr);
if (ngroups > 1) {
fprintf(fp,"There are %d energy groups\n",ngroups);
snew(epme,ngroups);
for(i=0; (i<ngroups); i++)
snew(epme[i],ngroups);
}
/* Put x is in the box, this part needs to be parallellized properly */
/*put_atoms_in_box(box,nsb->natoms,x);*/
/* Here sorting of X (and q) is done.
* Alternatively, one could just put the atoms in one of the
* cr->nnodes slabs. That is much cheaper than sorting.
*/
for(i=0; (i<nsb->natoms); i++)
index[i] = i;
if (bSort) {
xptr = x;
qsort(index,nsb->natoms,sizeof(index[0]),comp_xptr);
xptr = NULL; /* To trap unintentional use of the ptr */
}
/* After sorting we only need the part that is to be computed on
* this processor. We also compute the mu_tot here (system dipole)
*/
clear_rvec(mu_tot[0]);
for(i=START(nsb); (i<START(nsb)+HOMENR(nsb)); i++) {
ii = index[i];
q = charge[ii];
qbuf[i] = q;
for(m=0; (m<DIM); m++) {
xx = x[ii][m];
xbuf[i][m] = xx;
mu_tot[0][m] += q*xx;
}
clear_rvec(f[ii]);
}
copy_rvec(mu_tot[0],mu_tot[1]);
if (debug) {
pr_rvec(debug,0,"qbuf",qbuf,nsb->natoms,TRUE);
pr_rvecs(debug,0,"xbuf",xbuf,nsb->natoms);
pr_rvecs(debug,0,"box",box,DIM);
}
for(ig=0; (ig<ngroups); ig++) {
for(jg=ig; (jg<ngroups); jg++) {
if (ngroups > 1) {
for(i=START(nsb); (i<START(nsb)+HOMENR(nsb)); i++) {
if ((cENER[i] == ig) || (cENER[i] == jg))
qqbuf[i] = qbuf[i];
else
qqbuf[i] = 0;
}
qptr = qqbuf;
}
else
qptr = qbuf;
ener = do_pme(fp,bVerbose,ir,xbuf,f,qptr,qptr,box,cr,
nsb,nrnb,vir,fr->ewaldcoeff,FALSE,0,&dvdl,FALSE);
vcorr = ewald_LRcorrection(fp,nsb,cr,fr,qptr,qptr,excl,xbuf,box,mu_tot,
ir->ewald_geometry,ir->epsilon_surface,
0,&dvdl,&vdip,&vcharge);
gmx_sum(1,&ener,cr);
gmx_sum(1,&vcorr,cr);
if (ngroups > 1)
epme[ig][jg] = ener+vcorr;
}
}
if (ngroups > 1) {
if (fp_xvg)
fprintf(fp_xvg,"%10.3f",tm);
for(ig=0; (ig<ngroups); ig++) {
for(jg=ig; (jg<ngroups); jg++) {
if (ig != jg)
epme[ig][jg] -= epme[ig][ig]+epme[jg][jg];
if (fp_xvg)
fprintf(fp_xvg," %12.5e",epme[ig][jg]);
}
}
if (fp_xvg)
fprintf(fp_xvg,"\n");
}
else {
fprintf(fp,"Time: %10.3f Energy: %12.5e Correction: %12.5e Total: %12.5e\n",
tm,ener,vcorr,ener+vcorr);
if (fp_xvg)
fprintf(fp_xvg,"%10.3f %12.5e %12.5e %12.5e\n",tm,ener+vcorr,vdip,vcharge);
if (bVerbose) {
m_add(vir,vir_corr,vir_tot);
gmx_sum(9,vir_tot[0],cr);
pr_rvecs(fp,0,"virial",vir_tot,DIM);
}
fflush(fp);
}
}
void do_force_lowlevel(t_forcerec *fr, t_inputrec *ir,
t_idef *idef, t_commrec *cr,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_mdatoms *md,
rvec x[], history_t *hist,
rvec f[],
rvec f_longrange[],
gmx_enerdata_t *enerd,
t_fcdata *fcd,
gmx_localtop_t *top,
gmx_genborn_t *born,
gmx_bool bBornRadii,
matrix box,
t_lambda *fepvals,
real *lambda,
t_graph *graph,
t_blocka *excl,
rvec mu_tot[],
int flags,
float *cycles_pme)
{
int i, j;
int donb_flags;
gmx_bool bSB;
int pme_flags;
matrix boxs;
rvec box_size;
t_pbc pbc;
real dvdl_dum[efptNR], dvdl_nb[efptNR];
#ifdef GMX_MPI
double t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
#endif
set_pbc(&pbc, fr->ePBC, box);
/* reset free energy components */
for (i = 0; i < efptNR; i++)
{
dvdl_nb[i] = 0;
dvdl_dum[i] = 0;
}
/* Reset box */
for (i = 0; (i < DIM); i++)
{
box_size[i] = box[i][i];
}
debug_gmx();
/* do QMMM first if requested */
if (fr->bQMMM)
{
enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr);
}
/* Call the short range functions all in one go. */
#ifdef GMX_MPI
/*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
#define TAKETIME FALSE
if (TAKETIME)
{
MPI_Barrier(cr->mpi_comm_mygroup);
t0 = MPI_Wtime();
}
#endif
if (ir->nwall)
{
/* foreign lambda component for walls */
real dvdl_walls = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
enerd->grpp.ener[egLJSR], nrnb);
enerd->dvdl_lin[efptVDW] += dvdl_walls;
}
/* If doing GB, reset dvda and calculate the Born radii */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsNONBONDED);
for (i = 0; i < born->nr; i++)
{
fr->dvda[i] = 0;
}
if (bBornRadii)
{
calc_gb_rad(cr, fr, ir, top, x, &(fr->gblist), born, md, nrnb);
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
}
where();
/* We only do non-bonded calculation with group scheme here, the verlet
* calls are done from do_force_cutsVERLET(). */
if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
{
donb_flags = 0;
/* Add short-range interactions */
donb_flags |= GMX_NONBONDED_DO_SR;
/* Currently all group scheme kernels always calculate (shift-)forces */
if (flags & GMX_FORCE_FORCES)
{
donb_flags |= GMX_NONBONDED_DO_FORCE;
}
if (flags & GMX_FORCE_VIRIAL)
{
donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
}
if (flags & GMX_FORCE_ENERGY)
{
donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
}
if (flags & GMX_FORCE_DO_LR)
{
donb_flags |= GMX_NONBONDED_DO_LR;
}
wallcycle_sub_start(wcycle, ewcsNONBONDED);
do_nonbonded(fr, x, f, f_longrange, md, excl,
&enerd->grpp, nrnb,
lambda, dvdl_nb, -1, -1, donb_flags);
/* If we do foreign lambda and we have soft-core interactions
* we have to recalculate the (non-linear) energies contributions.
*/
if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
{
for (i = 0; i < enerd->n_lambda; i++)
{
real lam_i[efptNR];
for (j = 0; j < efptNR; j++)
{
lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
}
reset_foreign_enerdata(enerd);
do_nonbonded(fr, x, f, f_longrange, md, excl,
&(enerd->foreign_grpp), nrnb,
lam_i, dvdl_dum, -1, -1,
(donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
}
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
where();
}
/* If we are doing GB, calculate bonded forces and apply corrections
* to the solvation forces */
/* MRS: Eventually, many need to include free energy contribution here! */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsLISTED);
calc_gb_forces(cr, md, born, top, x, f, fr, idef,
ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
wallcycle_sub_stop(wcycle, ewcsLISTED);
}
#ifdef GMX_MPI
if (TAKETIME)
{
t1 = MPI_Wtime();
fr->t_fnbf += t1-t0;
}
#endif
if (fepvals->sc_alpha != 0)
{
enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
}
else
{
enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
}
if (fepvals->sc_alpha != 0)
/* even though coulomb part is linear, we already added it, beacuse we
need to go through the vdw calculation anyway */
{
enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
}
else
{
enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
}
debug_gmx();
if (debug)
{
pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
}
/* Shift the coordinates. Must be done before listed forces and PPPM,
* but is also necessary for SHAKE and update, therefore it can NOT
* go when no listed forces have to be evaluated.
*
* The shifting and PBC code is deliberately not timed, since with
* the Verlet scheme it only takes non-zero time with triclinic
* boxes, and even then the time is around a factor of 100 less
* than the next smallest counter.
*/
/* Here sometimes we would not need to shift with NBFonly,
* but we do so anyhow for consistency of the returned coordinates.
*/
if (graph)
{
shift_self(graph, box, x);
if (TRICLINIC(box))
{
inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
}
else
{
inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
}
}
/* Check whether we need to do listed interactions or correct for exclusions */
if (fr->bMolPBC &&
((flags & GMX_FORCE_LISTED)
|| EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
{
/* TODO There are no electrostatics methods that require this
transformation, when using the Verlet scheme, so update the
above conditional. */
/* Since all atoms are in the rectangular or triclinic unit-cell,
* only single box vector shifts (2 in x) are required.
*/
set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
}
debug_gmx();
do_force_listed(wcycle, box, ir->fepvals, cr->ms,
idef, (const rvec *) x, hist, f, fr,
&pbc, graph, enerd, nrnb, lambda, md, fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL,
flags);
where();
*cycles_pme = 0;
clear_mat(fr->vir_el_recip);
clear_mat(fr->vir_lj_recip);
/* Do long-range electrostatics and/or LJ-PME, including related short-range
* corrections.
*/
if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
{
int status = 0;
real Vlr_q = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
bSB = (ir->nwall == 2);
if (bSB)
{
copy_mat(box, boxs);
svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
box_size[ZZ] *= ir->wall_ewald_zfac;
}
if (EEL_PME_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
{
real dvdl_long_range_correction_q = 0;
real dvdl_long_range_correction_lj = 0;
/* With the Verlet scheme exclusion forces are calculated
* in the non-bonded kernel.
*/
/* The TPI molecule does not have exclusions with the rest
* of the system and no intra-molecular PME grid
* contributions will be calculated in
* gmx_pme_calc_energy.
*/
if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
ir->ewald_geometry != eewg3D ||
ir->epsilon_surface != 0)
{
int nthreads, t;
wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
if (fr->n_tpi > 0)
{
gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
}
nthreads = gmx_omp_nthreads_get(emntBonded);
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (t = 0; t < nthreads; t++)
{
int i;
rvec *fnv;
tensor *vir_q, *vir_lj;
real *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
if (t == 0)
{
fnv = fr->f_novirsum;
vir_q = &fr->vir_el_recip;
vir_lj = &fr->vir_lj_recip;
Vcorrt_q = &Vcorr_q;
Vcorrt_lj = &Vcorr_lj;
dvdlt_q = &dvdl_long_range_correction_q;
dvdlt_lj = &dvdl_long_range_correction_lj;
}
else
{
fnv = fr->f_t[t].f;
vir_q = &fr->f_t[t].vir_q;
vir_lj = &fr->f_t[t].vir_lj;
Vcorrt_q = &fr->f_t[t].Vcorr_q;
Vcorrt_lj = &fr->f_t[t].Vcorr_lj;
dvdlt_q = &fr->f_t[t].dvdl[efptCOUL];
dvdlt_lj = &fr->f_t[t].dvdl[efptVDW];
for (i = 0; i < fr->natoms_force; i++)
{
clear_rvec(fnv[i]);
}
clear_mat(*vir_q);
clear_mat(*vir_lj);
}
*dvdlt_q = 0;
*dvdlt_lj = 0;
ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
cr, t, fr,
md->chargeA, md->chargeB,
md->sqrt_c6A, md->sqrt_c6B,
md->sigmaA, md->sigmaB,
md->sigma3A, md->sigma3B,
md->nChargePerturbed || md->nTypePerturbed,
ir->cutoff_scheme != ecutsVERLET,
excl, x, bSB ? boxs : box, mu_tot,
ir->ewald_geometry,
ir->epsilon_surface,
fnv, *vir_q, *vir_lj,
Vcorrt_q, Vcorrt_lj,
lambda[efptCOUL], lambda[efptVDW],
dvdlt_q, dvdlt_lj);
}
if (nthreads > 1)
{
reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
fr->vir_el_recip, fr->vir_lj_recip,
&Vcorr_q, &Vcorr_lj,
&dvdl_long_range_correction_q,
&dvdl_long_range_correction_lj,
nthreads, fr->f_t);
}
wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
}
if (EEL_PME_EWALD(fr->eeltype) && fr->n_tpi == 0)
{
/* This is not in a subcounter because it takes a
negligible and constant-sized amount of time */
Vcorr_q += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
&dvdl_long_range_correction_q,
fr->vir_el_recip);
}
enerd->dvdl_lin[efptCOUL] += dvdl_long_range_correction_q;
enerd->dvdl_lin[efptVDW] += dvdl_long_range_correction_lj;
if ((EEL_PME(fr->eeltype) || EVDW_PME(fr->vdwtype)) && (cr->duty & DUTY_PME))
{
/* Do reciprocal PME for Coulomb and/or LJ. */
assert(fr->n_tpi >= 0);
if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
{
pme_flags = GMX_PME_SPREAD | GMX_PME_SOLVE;
if (EEL_PME(fr->eeltype))
{
pme_flags |= GMX_PME_DO_COULOMB;
}
if (EVDW_PME(fr->vdwtype))
{
pme_flags |= GMX_PME_DO_LJ;
}
if (flags & GMX_FORCE_FORCES)
{
pme_flags |= GMX_PME_CALC_F;
}
if (flags & GMX_FORCE_VIRIAL)
{
pme_flags |= GMX_PME_CALC_ENER_VIR;
}
if (fr->n_tpi > 0)
{
/* We don't calculate f, but we do want the potential */
pme_flags |= GMX_PME_CALC_POT;
}
wallcycle_start(wcycle, ewcPMEMESH);
status = gmx_pme_do(fr->pmedata,
0, md->homenr - fr->n_tpi,
x, fr->f_novirsum,
md->chargeA, md->chargeB,
md->sqrt_c6A, md->sqrt_c6B,
md->sigmaA, md->sigmaB,
bSB ? boxs : box, cr,
DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
nrnb, wcycle,
fr->vir_el_recip, fr->ewaldcoeff_q,
fr->vir_lj_recip, fr->ewaldcoeff_lj,
&Vlr_q, &Vlr_lj,
lambda[efptCOUL], lambda[efptVDW],
&dvdl_long_range_q, &dvdl_long_range_lj, pme_flags);
*cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
if (status != 0)
{
gmx_fatal(FARGS, "Error %d in reciprocal PME routine", status);
}
/* We should try to do as little computation after
* this as possible, because parallel PME synchronizes
* the nodes, so we want all load imbalance of the
* rest of the force calculation to be before the PME
* call. DD load balancing is done on the whole time
* of the force call (without PME).
*/
}
if (fr->n_tpi > 0)
{
if (EVDW_PME(ir->vdwtype))
{
gmx_fatal(FARGS, "Test particle insertion not implemented with LJ-PME");
}
/* Determine the PME grid energy of the test molecule
* with the PME grid potential of the other charges.
*/
gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
x + md->homenr - fr->n_tpi,
md->chargeA + md->homenr - fr->n_tpi,
&Vlr_q);
}
}
}
if (!EEL_PME(fr->eeltype) && EEL_PME_EWALD(fr->eeltype))
{
Vlr_q = do_ewald(ir, x, fr->f_novirsum,
md->chargeA, md->chargeB,
box_size, cr, md->homenr,
fr->vir_el_recip, fr->ewaldcoeff_q,
lambda[efptCOUL], &dvdl_long_range_q, fr->ewald_table);
}
/* Note that with separate PME nodes we get the real energies later */
enerd->dvdl_lin[efptCOUL] += dvdl_long_range_q;
enerd->dvdl_lin[efptVDW] += dvdl_long_range_lj;
enerd->term[F_COUL_RECIP] = Vlr_q + Vcorr_q;
enerd->term[F_LJ_RECIP] = Vlr_lj + Vcorr_lj;
if (debug)
{
fprintf(debug, "Vlr_q = %g, Vcorr_q = %g, Vlr_corr_q = %g\n",
Vlr_q, Vcorr_q, enerd->term[F_COUL_RECIP]);
pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
fprintf(debug, "Vlr_lj: %g, Vcorr_lj = %g, Vlr_corr_lj = %g\n",
Vlr_lj, Vcorr_lj, enerd->term[F_LJ_RECIP]);
pr_rvecs(debug, 0, "vir_lj_recip after corr", fr->vir_lj_recip, DIM);
}
}
else
{
/* Is there a reaction-field exclusion correction needed? */
if (EEL_RF(fr->eeltype) && eelRF_NEC != fr->eeltype)
{
/* With the Verlet scheme, exclusion forces are calculated
* in the non-bonded kernel.
*/
if (ir->cutoff_scheme != ecutsVERLET)
{
real dvdl_rf_excl = 0;
enerd->term[F_RF_EXCL] =
RF_excl_correction(fr, graph, md, excl, x, f,
fr->fshift, &pbc, lambda[efptCOUL], &dvdl_rf_excl);
enerd->dvdl_lin[efptCOUL] += dvdl_rf_excl;
}
}
}
where();
debug_gmx();
if (debug)
{
print_nrnb(debug, nrnb);
}
debug_gmx();
#ifdef GMX_MPI
if (TAKETIME)
{
t2 = MPI_Wtime();
MPI_Barrier(cr->mpi_comm_mygroup);
t3 = MPI_Wtime();
fr->t_wait += t3-t2;
if (fr->timesteps == 11)
{
char buf[22];
fprintf(stderr, "* PP load balancing info: rank %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
cr->nodeid, gmx_step_str(fr->timesteps, buf),
100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
(fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
}
fr->timesteps++;
}
#endif
if (debug)
{
pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
}
}
int relax_shell_flexcon(FILE *fplog, t_commrec *cr, gmx_bool bVerbose,
gmx_int64_t mdstep, t_inputrec *inputrec,
gmx_bool bDoNS, int force_flags,
gmx_localtop_t *top,
gmx_constr_t constr,
gmx_enerdata_t *enerd, t_fcdata *fcd,
t_state *state, rvec f[],
tensor force_vir,
t_mdatoms *md,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_graph *graph,
gmx_groups_t *groups,
struct gmx_shellfc *shfc,
t_forcerec *fr,
gmx_bool bBornRadii,
double t, rvec mu_tot,
gmx_bool *bConverged,
gmx_vsite_t *vsite,
FILE *fp_field)
{
int nshell;
t_shell *shell;
t_idef *idef;
rvec *pos[2], *force[2], *acc_dir = NULL, *x_old = NULL;
real Epot[2], df[2];
rvec dx;
real sf_dir, invdt;
real ftol, xiH, xiS, dum = 0;
char sbuf[22];
gmx_bool bCont, bInit;
int nat, dd_ac0, dd_ac1 = 0, i;
int start = 0, homenr = md->homenr, end = start+homenr, cg0, cg1;
int nflexcon, g, number_steps, d, Min = 0, count = 0;
#define Try (1-Min) /* At start Try = 1 */
bCont = (mdstep == inputrec->init_step) && inputrec->bContinuation;
bInit = (mdstep == inputrec->init_step) || shfc->bRequireInit;
ftol = inputrec->em_tol;
number_steps = inputrec->niter;
nshell = shfc->nshell;
shell = shfc->shell;
nflexcon = shfc->nflexcon;
idef = &top->idef;
if (DOMAINDECOMP(cr))
{
nat = dd_natoms_vsite(cr->dd);
if (nflexcon > 0)
{
dd_get_constraint_range(cr->dd, &dd_ac0, &dd_ac1);
nat = max(nat, dd_ac1);
}
}
else
{
nat = state->natoms;
}
if (nat > shfc->x_nalloc)
{
/* Allocate local arrays */
shfc->x_nalloc = over_alloc_dd(nat);
for (i = 0; (i < 2); i++)
{
srenew(shfc->x[i], shfc->x_nalloc);
srenew(shfc->f[i], shfc->x_nalloc);
}
}
for (i = 0; (i < 2); i++)
{
pos[i] = shfc->x[i];
force[i] = shfc->f[i];
}
/* When we had particle decomposition, this code only worked with
* PD when all particles involved with each shell were in the same
* charge group. Not sure if this is still relevant. */
if (bDoNS && inputrec->ePBC != epbcNONE && !DOMAINDECOMP(cr))
{
/* This is the only time where the coordinates are used
* before do_force is called, which normally puts all
* charge groups in the box.
*/
cg0 = 0;
cg1 = top->cgs.nr;
put_charge_groups_in_box(fplog, cg0, cg1, fr->ePBC, state->box,
&(top->cgs), state->x, fr->cg_cm);
if (graph)
{
mk_mshift(fplog, graph, fr->ePBC, state->box, state->x);
}
}
/* After this all coordinate arrays will contain whole molecules */
if (graph)
{
shift_self(graph, state->box, state->x);
}
if (nflexcon)
{
if (nat > shfc->flex_nalloc)
{
shfc->flex_nalloc = over_alloc_dd(nat);
srenew(shfc->acc_dir, shfc->flex_nalloc);
srenew(shfc->x_old, shfc->flex_nalloc);
}
acc_dir = shfc->acc_dir;
x_old = shfc->x_old;
for (i = 0; i < homenr; i++)
{
for (d = 0; d < DIM; d++)
{
shfc->x_old[i][d] =
state->x[start+i][d] - state->v[start+i][d]*inputrec->delta_t;
}
}
}
/* Do a prediction of the shell positions */
if (shfc->bPredict && !bCont)
{
predict_shells(fplog, state->x, state->v, inputrec->delta_t, nshell, shell,
md->massT, NULL, bInit);
}
/* do_force expected the charge groups to be in the box */
if (graph)
{
unshift_self(graph, state->box, state->x);
}
/* Calculate the forces first time around */
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "x b4 do_force", state->x + start, homenr);
}
do_force(fplog, cr, inputrec, mdstep, nrnb, wcycle, top, groups,
state->box, state->x, &state->hist,
force[Min], force_vir, md, enerd, fcd,
state->lambda, graph,
fr, vsite, mu_tot, t, fp_field, NULL, bBornRadii,
(bDoNS ? GMX_FORCE_NS : 0) | force_flags);
sf_dir = 0;
if (nflexcon)
{
init_adir(fplog, shfc,
constr, idef, inputrec, cr, dd_ac1, mdstep, md, start, end,
shfc->x_old-start, state->x, state->x, force[Min],
shfc->acc_dir-start,
fr->bMolPBC, state->box, state->lambda, &dum, nrnb);
for (i = start; i < end; i++)
{
sf_dir += md->massT[i]*norm2(shfc->acc_dir[i-start]);
}
}
Epot[Min] = enerd->term[F_EPOT];
df[Min] = rms_force(cr, shfc->f[Min], nshell, shell, nflexcon, &sf_dir, &Epot[Min]);
df[Try] = 0;
if (debug)
{
fprintf(debug, "df = %g %g\n", df[Min], df[Try]);
}
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "force0", force[Min], md->nr);
}
if (nshell+nflexcon > 0)
{
/* Copy x to pos[Min] & pos[Try]: during minimization only the
* shell positions are updated, therefore the other particles must
* be set here.
*/
memcpy(pos[Min], state->x, nat*sizeof(state->x[0]));
memcpy(pos[Try], state->x, nat*sizeof(state->x[0]));
}
if (bVerbose && MASTER(cr))
{
print_epot(stdout, mdstep, 0, Epot[Min], df[Min], nflexcon, sf_dir);
}
if (debug)
{
fprintf(debug, "%17s: %14.10e\n",
interaction_function[F_EKIN].longname, enerd->term[F_EKIN]);
fprintf(debug, "%17s: %14.10e\n",
interaction_function[F_EPOT].longname, enerd->term[F_EPOT]);
fprintf(debug, "%17s: %14.10e\n",
interaction_function[F_ETOT].longname, enerd->term[F_ETOT]);
fprintf(debug, "SHELLSTEP %s\n", gmx_step_str(mdstep, sbuf));
}
/* First check whether we should do shells, or whether the force is
* low enough even without minimization.
*/
*bConverged = (df[Min] < ftol);
for (count = 1; (!(*bConverged) && (count < number_steps)); count++)
{
if (vsite)
{
construct_vsites(vsite, pos[Min], inputrec->delta_t, state->v,
idef->iparams, idef->il,
fr->ePBC, fr->bMolPBC, cr, state->box);
}
if (nflexcon)
{
init_adir(fplog, shfc,
constr, idef, inputrec, cr, dd_ac1, mdstep, md, start, end,
x_old-start, state->x, pos[Min], force[Min], acc_dir-start,
fr->bMolPBC, state->box, state->lambda, &dum, nrnb);
directional_sd(pos[Min], pos[Try], acc_dir-start, start, end,
fr->fc_stepsize);
}
/* New positions, Steepest descent */
shell_pos_sd(pos[Min], pos[Try], force[Min], nshell, shell, count);
/* do_force expected the charge groups to be in the box */
if (graph)
{
unshift_self(graph, state->box, pos[Try]);
}
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "RELAX: pos[Min] ", pos[Min] + start, homenr);
pr_rvecs(debug, 0, "RELAX: pos[Try] ", pos[Try] + start, homenr);
}
/* Try the new positions */
do_force(fplog, cr, inputrec, 1, nrnb, wcycle,
top, groups, state->box, pos[Try], &state->hist,
force[Try], force_vir,
md, enerd, fcd, state->lambda, graph,
fr, vsite, mu_tot, t, fp_field, NULL, bBornRadii,
force_flags);
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "RELAX: force[Min]", force[Min] + start, homenr);
pr_rvecs(debug, 0, "RELAX: force[Try]", force[Try] + start, homenr);
}
sf_dir = 0;
if (nflexcon)
{
init_adir(fplog, shfc,
constr, idef, inputrec, cr, dd_ac1, mdstep, md, start, end,
x_old-start, state->x, pos[Try], force[Try], acc_dir-start,
fr->bMolPBC, state->box, state->lambda, &dum, nrnb);
for (i = start; i < end; i++)
{
sf_dir += md->massT[i]*norm2(acc_dir[i-start]);
}
}
Epot[Try] = enerd->term[F_EPOT];
df[Try] = rms_force(cr, force[Try], nshell, shell, nflexcon, &sf_dir, &Epot[Try]);
if (debug)
{
fprintf(debug, "df = %g %g\n", df[Min], df[Try]);
}
if (debug)
{
if (gmx_debug_at)
{
pr_rvecs(debug, 0, "F na do_force", force[Try] + start, homenr);
}
if (gmx_debug_at)
{
fprintf(debug, "SHELL ITER %d\n", count);
dump_shells(debug, pos[Try], force[Try], ftol, nshell, shell);
}
}
if (bVerbose && MASTER(cr))
{
print_epot(stdout, mdstep, count, Epot[Try], df[Try], nflexcon, sf_dir);
}
*bConverged = (df[Try] < ftol);
if ((df[Try] < df[Min]))
{
if (debug)
{
fprintf(debug, "Swapping Min and Try\n");
}
if (nflexcon)
{
/* Correct the velocities for the flexible constraints */
invdt = 1/inputrec->delta_t;
for (i = start; i < end; i++)
{
for (d = 0; d < DIM; d++)
{
state->v[i][d] += (pos[Try][i][d] - pos[Min][i][d])*invdt;
}
}
}
Min = Try;
}
else
{
decrease_step_size(nshell, shell);
}
}
if (MASTER(cr) && !(*bConverged))
{
/* Note that the energies and virial are incorrect when not converged */
if (fplog)
{
fprintf(fplog,
"step %s: EM did not converge in %d iterations, RMS force %.3f\n",
gmx_step_str(mdstep, sbuf), number_steps, df[Min]);
}
fprintf(stderr,
"step %s: EM did not converge in %d iterations, RMS force %.3f\n",
gmx_step_str(mdstep, sbuf), number_steps, df[Min]);
}
/* Copy back the coordinates and the forces */
memcpy(state->x, pos[Min], nat*sizeof(state->x[0]));
memcpy(f, force[Min], nat*sizeof(f[0]));
return count;
}
void do_force_lowlevel(FILE *fplog, gmx_large_int_t step,
t_forcerec *fr, t_inputrec *ir,
t_idef *idef, t_commrec *cr,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_mdatoms *md,
t_grpopts *opts,
rvec x[], history_t *hist,
rvec f[],
rvec f_longrange[],
gmx_enerdata_t *enerd,
t_fcdata *fcd,
gmx_mtop_t *mtop,
gmx_localtop_t *top,
gmx_genborn_t *born,
t_atomtypes *atype,
gmx_bool bBornRadii,
matrix box,
t_lambda *fepvals,
real *lambda,
t_graph *graph,
t_blocka *excl,
rvec mu_tot[],
int flags,
float *cycles_pme)
{
int i, j, status;
int donb_flags;
gmx_bool bDoEpot, bSepDVDL, bSB;
int pme_flags;
matrix boxs;
rvec box_size;
real Vsr, Vlr, Vcorr = 0;
t_pbc pbc;
real dvdgb;
char buf[22];
double clam_i, vlam_i;
real dvdl_dum[efptNR], dvdl, dvdl_nb[efptNR], lam_i[efptNR];
real dvdlsum;
#ifdef GMX_MPI
double t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
#endif
#define PRINT_SEPDVDL(s, v, dvdlambda) if (bSepDVDL) {fprintf(fplog, sepdvdlformat, s, v, dvdlambda); }
GMX_MPE_LOG(ev_force_start);
set_pbc(&pbc, fr->ePBC, box);
/* reset free energy components */
for (i = 0; i < efptNR; i++)
{
dvdl_nb[i] = 0;
dvdl_dum[i] = 0;
}
/* Reset box */
for (i = 0; (i < DIM); i++)
{
box_size[i] = box[i][i];
}
bSepDVDL = (fr->bSepDVDL && do_per_step(step, ir->nstlog));
debug_gmx();
/* do QMMM first if requested */
if (fr->bQMMM)
{
enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr, md);
}
if (bSepDVDL)
{
fprintf(fplog, "Step %s: non-bonded V and dVdl for node %d:\n",
gmx_step_str(step, buf), cr->nodeid);
}
/* Call the short range functions all in one go. */
GMX_MPE_LOG(ev_do_fnbf_start);
#ifdef GMX_MPI
/*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
#define TAKETIME FALSE
if (TAKETIME)
{
MPI_Barrier(cr->mpi_comm_mygroup);
t0 = MPI_Wtime();
}
#endif
if (ir->nwall)
{
/* foreign lambda component for walls */
dvdl = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
enerd->grpp.ener[egLJSR], nrnb);
PRINT_SEPDVDL("Walls", 0.0, dvdl);
enerd->dvdl_lin[efptVDW] += dvdl;
}
/* If doing GB, reset dvda and calculate the Born radii */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsNONBONDED);
for (i = 0; i < born->nr; i++)
{
fr->dvda[i] = 0;
}
if (bBornRadii)
{
calc_gb_rad(cr, fr, ir, top, atype, x, &(fr->gblist), born, md, nrnb);
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
}
where();
/* We only do non-bonded calculation with group scheme here, the verlet
* calls are done from do_force_cutsVERLET(). */
if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
{
donb_flags = 0;
/* Add short-range interactions */
donb_flags |= GMX_NONBONDED_DO_SR;
if (flags & GMX_FORCE_FORCES)
{
donb_flags |= GMX_NONBONDED_DO_FORCE;
}
if (flags & GMX_FORCE_ENERGY)
{
donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
}
if (flags & GMX_FORCE_DO_LR)
{
donb_flags |= GMX_NONBONDED_DO_LR;
}
wallcycle_sub_start(wcycle, ewcsNONBONDED);
do_nonbonded(cr, fr, x, f, f_longrange, md, excl,
&enerd->grpp, box_size, nrnb,
lambda, dvdl_nb, -1, -1, donb_flags);
/* If we do foreign lambda and we have soft-core interactions
* we have to recalculate the (non-linear) energies contributions.
*/
if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
{
for (i = 0; i < enerd->n_lambda; i++)
{
for (j = 0; j < efptNR; j++)
{
lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
}
reset_foreign_enerdata(enerd);
do_nonbonded(cr, fr, x, f, f_longrange, md, excl,
&(enerd->foreign_grpp), box_size, nrnb,
lam_i, dvdl_dum, -1, -1,
(donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
sum_epot(&ir->opts, &(enerd->foreign_grpp), enerd->foreign_term);
enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
}
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
where();
}
/* If we are doing GB, calculate bonded forces and apply corrections
* to the solvation forces */
/* MRS: Eventually, many need to include free energy contribution here! */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsBONDED);
calc_gb_forces(cr, md, born, top, atype, x, f, fr, idef,
ir->gb_algorithm, ir->sa_algorithm, nrnb, bBornRadii, &pbc, graph, enerd);
wallcycle_sub_stop(wcycle, ewcsBONDED);
}
#ifdef GMX_MPI
if (TAKETIME)
{
t1 = MPI_Wtime();
fr->t_fnbf += t1-t0;
}
#endif
if (fepvals->sc_alpha != 0)
{
enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
}
else
{
enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
}
if (fepvals->sc_alpha != 0)
/* even though coulomb part is linear, we already added it, beacuse we
need to go through the vdw calculation anyway */
{
enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
}
else
{
enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
}
Vsr = 0;
if (bSepDVDL)
{
for (i = 0; i < enerd->grpp.nener; i++)
{
Vsr +=
(fr->bBHAM ?
enerd->grpp.ener[egBHAMSR][i] :
enerd->grpp.ener[egLJSR][i])
+ enerd->grpp.ener[egCOULSR][i] + enerd->grpp.ener[egGB][i];
}
dvdlsum = dvdl_nb[efptVDW] + dvdl_nb[efptCOUL];
PRINT_SEPDVDL("VdW and Coulomb SR particle-p.", Vsr, dvdlsum);
}
debug_gmx();
GMX_MPE_LOG(ev_do_fnbf_finish);
if (debug)
{
pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
}
/* Shift the coordinates. Must be done before bonded forces and PPPM,
* but is also necessary for SHAKE and update, therefore it can NOT
* go when no bonded forces have to be evaluated.
*/
/* Here sometimes we would not need to shift with NBFonly,
* but we do so anyhow for consistency of the returned coordinates.
*/
if (graph)
{
shift_self(graph, box, x);
if (TRICLINIC(box))
{
inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
}
else
{
inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
}
}
/* Check whether we need to do bondeds or correct for exclusions */
if (fr->bMolPBC &&
((flags & GMX_FORCE_BONDED)
|| EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype)))
{
/* Since all atoms are in the rectangular or triclinic unit-cell,
* only single box vector shifts (2 in x) are required.
*/
set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
}
debug_gmx();
if (flags & GMX_FORCE_BONDED)
{
GMX_MPE_LOG(ev_calc_bonds_start);
wallcycle_sub_start(wcycle, ewcsBONDED);
calc_bonds(fplog, cr->ms,
idef, x, hist, f, fr, &pbc, graph, enerd, nrnb, lambda, md, fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL, atype, born,
flags,
fr->bSepDVDL && do_per_step(step, ir->nstlog), step);
/* Check if we have to determine energy differences
* at foreign lambda's.
*/
if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) &&
idef->ilsort != ilsortNO_FE)
{
if (idef->ilsort != ilsortFE_SORTED)
{
gmx_incons("The bonded interactions are not sorted for free energy");
}
for (i = 0; i < enerd->n_lambda; i++)
{
reset_foreign_enerdata(enerd);
for (j = 0; j < efptNR; j++)
{
lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
}
calc_bonds_lambda(fplog, idef, x, fr, &pbc, graph, &(enerd->foreign_grpp), enerd->foreign_term, nrnb, lam_i, md,
fcd, DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
sum_epot(&ir->opts, &(enerd->foreign_grpp), enerd->foreign_term);
enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
}
}
debug_gmx();
GMX_MPE_LOG(ev_calc_bonds_finish);
wallcycle_sub_stop(wcycle, ewcsBONDED);
}
where();
*cycles_pme = 0;
if (EEL_FULL(fr->eeltype))
{
bSB = (ir->nwall == 2);
if (bSB)
{
copy_mat(box, boxs);
svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
box_size[ZZ] *= ir->wall_ewald_zfac;
}
clear_mat(fr->vir_el_recip);
if (fr->bEwald)
{
Vcorr = 0;
dvdl = 0;
/* With the Verlet scheme exclusion forces are calculated
* in the non-bonded kernel.
*/
/* The TPI molecule does not have exclusions with the rest
* of the system and no intra-molecular PME grid contributions
* will be calculated in gmx_pme_calc_energy.
*/
if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
ir->ewald_geometry != eewg3D ||
ir->epsilon_surface != 0)
{
int nthreads, t;
wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
if (fr->n_tpi > 0)
{
gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
}
nthreads = gmx_omp_nthreads_get(emntBonded);
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (t = 0; t < nthreads; t++)
{
int s, e, i;
rvec *fnv;
tensor *vir;
real *Vcorrt, *dvdlt;
if (t == 0)
{
fnv = fr->f_novirsum;
vir = &fr->vir_el_recip;
Vcorrt = &Vcorr;
dvdlt = &dvdl;
}
else
{
fnv = fr->f_t[t].f;
vir = &fr->f_t[t].vir;
Vcorrt = &fr->f_t[t].Vcorr;
dvdlt = &fr->f_t[t].dvdl[efptCOUL];
for (i = 0; i < fr->natoms_force; i++)
{
clear_rvec(fnv[i]);
}
clear_mat(*vir);
}
*dvdlt = 0;
*Vcorrt =
ewald_LRcorrection(fplog,
fr->excl_load[t], fr->excl_load[t+1],
cr, t, fr,
md->chargeA,
md->nChargePerturbed ? md->chargeB : NULL,
ir->cutoff_scheme != ecutsVERLET,
excl, x, bSB ? boxs : box, mu_tot,
ir->ewald_geometry,
ir->epsilon_surface,
fnv, *vir,
lambda[efptCOUL], dvdlt);
}
if (nthreads > 1)
{
reduce_thread_forces(fr->natoms_force, fr->f_novirsum,
fr->vir_el_recip,
&Vcorr, efptCOUL, &dvdl,
nthreads, fr->f_t);
}
wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
}
if (fr->n_tpi == 0)
{
Vcorr += ewald_charge_correction(cr, fr, lambda[efptCOUL], box,
&dvdl, fr->vir_el_recip);
}
PRINT_SEPDVDL("Ewald excl./charge/dip. corr.", Vcorr, dvdl);
enerd->dvdl_lin[efptCOUL] += dvdl;
}
status = 0;
Vlr = 0;
dvdl = 0;
switch (fr->eeltype)
{
case eelPME:
case eelPMESWITCH:
case eelPMEUSER:
case eelPMEUSERSWITCH:
case eelP3M_AD:
if (cr->duty & DUTY_PME)
{
assert(fr->n_tpi >= 0);
if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
{
pme_flags = GMX_PME_SPREAD_Q | GMX_PME_SOLVE;
if (flags & GMX_FORCE_FORCES)
{
pme_flags |= GMX_PME_CALC_F;
}
if (flags & (GMX_FORCE_VIRIAL | GMX_FORCE_ENERGY))
{
pme_flags |= GMX_PME_CALC_ENER_VIR;
}
if (fr->n_tpi > 0)
{
/* We don't calculate f, but we do want the potential */
pme_flags |= GMX_PME_CALC_POT;
}
wallcycle_start(wcycle, ewcPMEMESH);
status = gmx_pme_do(fr->pmedata,
md->start, md->homenr - fr->n_tpi,
x, fr->f_novirsum,
md->chargeA, md->chargeB,
bSB ? boxs : box, cr,
DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
nrnb, wcycle,
fr->vir_el_recip, fr->ewaldcoeff,
&Vlr, lambda[efptCOUL], &dvdl,
pme_flags);
*cycles_pme = wallcycle_stop(wcycle, ewcPMEMESH);
/* We should try to do as little computation after
* this as possible, because parallel PME synchronizes
* the nodes, so we want all load imbalance of the rest
* of the force calculation to be before the PME call.
* DD load balancing is done on the whole time of
* the force call (without PME).
*/
}
if (fr->n_tpi > 0)
{
/* Determine the PME grid energy of the test molecule
* with the PME grid potential of the other charges.
*/
gmx_pme_calc_energy(fr->pmedata, fr->n_tpi,
x + md->homenr - fr->n_tpi,
md->chargeA + md->homenr - fr->n_tpi,
&Vlr);
}
PRINT_SEPDVDL("PME mesh", Vlr, dvdl);
}
break;
case eelEWALD:
Vlr = do_ewald(fplog, FALSE, ir, x, fr->f_novirsum,
md->chargeA, md->chargeB,
box_size, cr, md->homenr,
fr->vir_el_recip, fr->ewaldcoeff,
lambda[efptCOUL], &dvdl, fr->ewald_table);
PRINT_SEPDVDL("Ewald long-range", Vlr, dvdl);
break;
default:
gmx_fatal(FARGS, "No such electrostatics method implemented %s",
eel_names[fr->eeltype]);
}
if (status != 0)
{
gmx_fatal(FARGS, "Error %d in long range electrostatics routine %s",
status, EELTYPE(fr->eeltype));
}
/* Note that with separate PME nodes we get the real energies later */
enerd->dvdl_lin[efptCOUL] += dvdl;
enerd->term[F_COUL_RECIP] = Vlr + Vcorr;
if (debug)
{
fprintf(debug, "Vlr = %g, Vcorr = %g, Vlr_corr = %g\n",
Vlr, Vcorr, enerd->term[F_COUL_RECIP]);
pr_rvecs(debug, 0, "vir_el_recip after corr", fr->vir_el_recip, DIM);
pr_rvecs(debug, 0, "fshift after LR Corrections", fr->fshift, SHIFTS);
}
}
else
{
if (EEL_RF(fr->eeltype))
{
/* With the Verlet scheme exclusion forces are calculated
* in the non-bonded kernel.
*/
if (ir->cutoff_scheme != ecutsVERLET && fr->eeltype != eelRF_NEC)
{
dvdl = 0;
enerd->term[F_RF_EXCL] =
RF_excl_correction(fplog, fr, graph, md, excl, x, f,
fr->fshift, &pbc, lambda[efptCOUL], &dvdl);
}
enerd->dvdl_lin[efptCOUL] += dvdl;
PRINT_SEPDVDL("RF exclusion correction",
enerd->term[F_RF_EXCL], dvdl);
}
}
where();
debug_gmx();
if (debug)
{
print_nrnb(debug, nrnb);
}
debug_gmx();
#ifdef GMX_MPI
if (TAKETIME)
{
t2 = MPI_Wtime();
MPI_Barrier(cr->mpi_comm_mygroup);
t3 = MPI_Wtime();
fr->t_wait += t3-t2;
if (fr->timesteps == 11)
{
fprintf(stderr, "* PP load balancing info: node %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
cr->nodeid, gmx_step_str(fr->timesteps, buf),
100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
(fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
}
fr->timesteps++;
}
#endif
if (debug)
{
pr_rvecs(debug, 0, "fshift after bondeds", fr->fshift, SHIFTS);
}
GMX_MPE_LOG(ev_force_finish);
}
void do_force_lowlevel(t_forcerec *fr, t_inputrec *ir,
t_idef *idef, t_commrec *cr,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_mdatoms *md,
rvec x[], history_t *hist,
rvec f[],
rvec f_longrange[],
gmx_enerdata_t *enerd,
t_fcdata *fcd,
gmx_localtop_t *top,
gmx_genborn_t *born,
gmx_bool bBornRadii,
matrix box,
t_lambda *fepvals,
real *lambda,
t_graph *graph,
t_blocka *excl,
rvec mu_tot[],
int flags,
float *cycles_pme)
{
int i, j;
int donb_flags;
gmx_bool bSB;
int pme_flags;
matrix boxs;
rvec box_size;
t_pbc pbc;
real dvdl_dum[efptNR], dvdl_nb[efptNR];
#ifdef GMX_MPI
double t0 = 0.0, t1, t2, t3; /* time measurement for coarse load balancing */
#endif
set_pbc(&pbc, fr->ePBC, box);
/* reset free energy components */
for (i = 0; i < efptNR; i++)
{
dvdl_nb[i] = 0;
dvdl_dum[i] = 0;
}
/* Reset box */
for (i = 0; (i < DIM); i++)
{
box_size[i] = box[i][i];
}
debug_gmx();
/* do QMMM first if requested */
if (fr->bQMMM)
{
enerd->term[F_EQM] = calculate_QMMM(cr, x, f, fr);
}
/* Call the short range functions all in one go. */
#ifdef GMX_MPI
/*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
#define TAKETIME FALSE
if (TAKETIME)
{
MPI_Barrier(cr->mpi_comm_mygroup);
t0 = MPI_Wtime();
}
#endif
if (ir->nwall)
{
/* foreign lambda component for walls */
real dvdl_walls = do_walls(ir, fr, box, md, x, f, lambda[efptVDW],
enerd->grpp.ener[egLJSR], nrnb);
enerd->dvdl_lin[efptVDW] += dvdl_walls;
}
/* If doing GB, reset dvda and calculate the Born radii */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsNONBONDED);
for (i = 0; i < born->nr; i++)
{
fr->dvda[i] = 0;
}
if (bBornRadii)
{
calc_gb_rad(cr, fr, ir, top, x, &(fr->gblist), born, md, nrnb);
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
}
where();
/* We only do non-bonded calculation with group scheme here, the verlet
* calls are done from do_force_cutsVERLET(). */
if (fr->cutoff_scheme == ecutsGROUP && (flags & GMX_FORCE_NONBONDED))
{
donb_flags = 0;
/* Add short-range interactions */
donb_flags |= GMX_NONBONDED_DO_SR;
/* Currently all group scheme kernels always calculate (shift-)forces */
if (flags & GMX_FORCE_FORCES)
{
donb_flags |= GMX_NONBONDED_DO_FORCE;
}
if (flags & GMX_FORCE_VIRIAL)
{
donb_flags |= GMX_NONBONDED_DO_SHIFTFORCE;
}
if (flags & GMX_FORCE_ENERGY)
{
donb_flags |= GMX_NONBONDED_DO_POTENTIAL;
}
if (flags & GMX_FORCE_DO_LR)
{
donb_flags |= GMX_NONBONDED_DO_LR;
}
wallcycle_sub_start(wcycle, ewcsNONBONDED);
do_nonbonded(fr, x, f, f_longrange, md, excl,
&enerd->grpp, nrnb,
lambda, dvdl_nb, -1, -1, donb_flags);
/* If we do foreign lambda and we have soft-core interactions
* we have to recalculate the (non-linear) energies contributions.
*/
if (fepvals->n_lambda > 0 && (flags & GMX_FORCE_DHDL) && fepvals->sc_alpha != 0)
{
for (i = 0; i < enerd->n_lambda; i++)
{
real lam_i[efptNR];
for (j = 0; j < efptNR; j++)
{
lam_i[j] = (i == 0 ? lambda[j] : fepvals->all_lambda[j][i-1]);
}
reset_foreign_enerdata(enerd);
do_nonbonded(fr, x, f, f_longrange, md, excl,
&(enerd->foreign_grpp), nrnb,
lam_i, dvdl_dum, -1, -1,
(donb_flags & ~GMX_NONBONDED_DO_FORCE) | GMX_NONBONDED_DO_FOREIGNLAMBDA);
sum_epot(&(enerd->foreign_grpp), enerd->foreign_term);
enerd->enerpart_lambda[i] += enerd->foreign_term[F_EPOT];
}
}
wallcycle_sub_stop(wcycle, ewcsNONBONDED);
where();
}
/* If we are doing GB, calculate bonded forces and apply corrections
* to the solvation forces */
/* MRS: Eventually, many need to include free energy contribution here! */
if (ir->implicit_solvent)
{
wallcycle_sub_start(wcycle, ewcsLISTED);
calc_gb_forces(cr, md, born, top, x, f, fr, idef,
ir->gb_algorithm, ir->sa_algorithm, nrnb, &pbc, graph, enerd);
wallcycle_sub_stop(wcycle, ewcsLISTED);
}
#ifdef GMX_MPI
if (TAKETIME)
{
t1 = MPI_Wtime();
fr->t_fnbf += t1-t0;
}
#endif
if (fepvals->sc_alpha != 0)
{
enerd->dvdl_nonlin[efptVDW] += dvdl_nb[efptVDW];
}
else
{
enerd->dvdl_lin[efptVDW] += dvdl_nb[efptVDW];
}
if (fepvals->sc_alpha != 0)
/* even though coulomb part is linear, we already added it, beacuse we
need to go through the vdw calculation anyway */
{
enerd->dvdl_nonlin[efptCOUL] += dvdl_nb[efptCOUL];
}
else
{
enerd->dvdl_lin[efptCOUL] += dvdl_nb[efptCOUL];
}
debug_gmx();
if (debug)
{
pr_rvecs(debug, 0, "fshift after SR", fr->fshift, SHIFTS);
}
/* Shift the coordinates. Must be done before listed forces and PPPM,
* but is also necessary for SHAKE and update, therefore it can NOT
* go when no listed forces have to be evaluated.
*
* The shifting and PBC code is deliberately not timed, since with
* the Verlet scheme it only takes non-zero time with triclinic
* boxes, and even then the time is around a factor of 100 less
* than the next smallest counter.
*/
/* Here sometimes we would not need to shift with NBFonly,
* but we do so anyhow for consistency of the returned coordinates.
*/
if (graph)
{
shift_self(graph, box, x);
if (TRICLINIC(box))
{
inc_nrnb(nrnb, eNR_SHIFTX, 2*graph->nnodes);
}
else
{
inc_nrnb(nrnb, eNR_SHIFTX, graph->nnodes);
}
}
/* Check whether we need to do listed interactions or correct for exclusions */
if (fr->bMolPBC &&
((flags & GMX_FORCE_LISTED)
|| EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype)))
{
/* TODO There are no electrostatics methods that require this
transformation, when using the Verlet scheme, so update the
above conditional. */
/* Since all atoms are in the rectangular or triclinic unit-cell,
* only single box vector shifts (2 in x) are required.
*/
set_pbc_dd(&pbc, fr->ePBC, cr->dd, TRUE, box);
}
debug_gmx();
do_force_listed(wcycle, box, ir->fepvals, cr->ms,
idef, (const rvec *) x, hist, f, fr,
&pbc, graph, enerd, nrnb, lambda, md, fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL,
flags);
where();
*cycles_pme = 0;
clear_mat(fr->vir_el_recip);
clear_mat(fr->vir_lj_recip);
/* Do long-range electrostatics and/or LJ-PME, including related short-range
* corrections.
*/
if (EEL_FULL(fr->eeltype) || EVDW_PME(fr->vdwtype))
{
int status = 0;
real Vlr_q = 0, Vlr_lj = 0, Vcorr_q = 0, Vcorr_lj = 0;
real dvdl_long_range_q = 0, dvdl_long_range_lj = 0;
bSB = (ir->nwall == 2);
if (bSB)
{
copy_mat(box, boxs);
svmul(ir->wall_ewald_zfac, boxs[ZZ], boxs[ZZ]);
box_size[ZZ] *= ir->wall_ewald_zfac;
}
if (EEL_PME_EWALD(fr->eeltype) || EVDW_PME(fr->vdwtype))
{
real dvdl_long_range_correction_q = 0;
real dvdl_long_range_correction_lj = 0;
/* With the Verlet scheme exclusion forces are calculated
* in the non-bonded kernel.
*/
/* The TPI molecule does not have exclusions with the rest
* of the system and no intra-molecular PME grid
* contributions will be calculated in
* gmx_pme_calc_energy.
*/
if ((ir->cutoff_scheme == ecutsGROUP && fr->n_tpi == 0) ||
ir->ewald_geometry != eewg3D ||
ir->epsilon_surface != 0)
{
int nthreads, t;
wallcycle_sub_start(wcycle, ewcsEWALD_CORRECTION);
if (fr->n_tpi > 0)
{
gmx_fatal(FARGS, "TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
}
nthreads = fr->nthread_ewc;
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (t = 0; t < nthreads; t++)
{
try
{
tensor *vir_q, *vir_lj;
real *Vcorrt_q, *Vcorrt_lj, *dvdlt_q, *dvdlt_lj;
if (t == 0)
{
vir_q = &fr->vir_el_recip;
vir_lj = &fr->vir_lj_recip;
Vcorrt_q = &Vcorr_q;
Vcorrt_lj = &Vcorr_lj;
dvdlt_q = &dvdl_long_range_correction_q;
dvdlt_lj = &dvdl_long_range_correction_lj;
}
else
{
vir_q = &fr->ewc_t[t].vir_q;
vir_lj = &fr->ewc_t[t].vir_lj;
Vcorrt_q = &fr->ewc_t[t].Vcorr_q;
Vcorrt_lj = &fr->ewc_t[t].Vcorr_lj;
dvdlt_q = &fr->ewc_t[t].dvdl[efptCOUL];
dvdlt_lj = &fr->ewc_t[t].dvdl[efptVDW];
clear_mat(*vir_q);
clear_mat(*vir_lj);
}
*dvdlt_q = 0;
*dvdlt_lj = 0;
/* Threading is only supported with the Verlet cut-off
* scheme and then only single particle forces (no
* exclusion forces) are calculated, so we can store
* the forces in the normal, single fr->f_novirsum array.
*/
ewald_LRcorrection(fr->excl_load[t], fr->excl_load[t+1],
cr, t, fr,
md->chargeA, md->chargeB,
md->sqrt_c6A, md->sqrt_c6B,
md->sigmaA, md->sigmaB,
md->sigma3A, md->sigma3B,
md->nChargePerturbed || md->nTypePerturbed,
ir->cutoff_scheme != ecutsVERLET,
excl, x, bSB ? boxs : box, mu_tot,
ir->ewald_geometry,
ir->epsilon_surface,
fr->f_novirsum, *vir_q, *vir_lj,
Vcorrt_q, Vcorrt_lj,
lambda[efptCOUL], lambda[efptVDW],
dvdlt_q, dvdlt_lj);
}
GMX_CATCH_ALL_AND_EXIT_WITH_FATAL_ERROR;
}
if (nthreads > 1)
{
reduce_thread_energies(fr->vir_el_recip, fr->vir_lj_recip,
&Vcorr_q, &Vcorr_lj,
&dvdl_long_range_correction_q,
&dvdl_long_range_correction_lj,
nthreads, fr->ewc_t);
}
wallcycle_sub_stop(wcycle, ewcsEWALD_CORRECTION);
}
void virial(FILE *fp,gmx_bool bFull,int nmol,rvec x[],matrix box,real rcut,
gmx_bool bYaw,real q[],gmx_bool bLJ)
{
int i,j,im,jm,natmol,ik,jk,m,ninter;
rvec dx,f,ftot,dvir,vir,pres,xcmi,xcmj,*force;
real dx6,dx2,dx1,fscal,c6,c12,vcoul,v12,v6,vctot,v12tot,v6tot;
t_pbc pbc;
set_pbc(&pbc,box);
fprintf(fp,"%3s - %3s: %6s %6s %6s %6s %8s %8s %8s\n",
"ai","aj","dx","dy","dz","|d|","virx","viry","virz");
clear_rvec(ftot);
clear_rvec(vir);
ninter = 0;
vctot = 0;
v12tot = 0;
v6tot = 0;
natmol = bYaw ? 5 : 3;
snew(force,nmol*natmol);
for(i=0; (i<nmol); i++) {
im = natmol*i;
/* Center of geometry */
clear_rvec(xcmi);
for(ik=0; (ik<natmol); ik++)
rvec_inc(xcmi,x[im+ik]);
for(m=0; (m<DIM); m++)
xcmi[m] /= natmol;
for(j=i+1; (j<nmol); j++) {
jm = natmol*j;
/* Center of geometry */
clear_rvec(xcmj);
for(jk=0; (jk<natmol); jk++)
rvec_inc(xcmj,x[jm+jk]);
for(m=0; (m<DIM); m++)
xcmj[m] /= natmol;
/* First check COM-COM distance */
pbc_dx(&pbc,xcmi,xcmj,dx);
if (norm(dx) < rcut) {
ninter++;
/* Neirest neighbour molecules! */
clear_rvec(dvir);
for(ik=0; (ik<natmol); ik++) {
for(jk=0; (jk<natmol); jk++) {
pbc_dx(&pbc,x[im+ik],x[jm+jk],dx);
dx2 = iprod(dx,dx);
dx1 = sqrt(dx2);
vcoul = q[ik]*q[jk]*ONE_4PI_EPS0/dx1;
vctot += vcoul;
if (bLJ) {
if (bYaw) {
c6 = yaw_lj[ik][2*jk];
c12 = yaw_lj[ik][2*jk+1];
}
else {
c6 = spc_lj[ik][2*jk];
c12 = spc_lj[ik][2*jk+1];
}
dx6 = dx2*dx2*dx2;
v6 = c6/dx6;
v12 = c12/(dx6*dx6);
v6tot -= v6;
v12tot+= v12;
fscal = (vcoul+12*v12-6*v6)/dx2;
}
else
fscal = vcoul/dx2;
for(m=0; (m<DIM); m++) {
f[m] = dx[m]*fscal;
dvir[m] -= 0.5*dx[m]*f[m];
}
rvec_inc(force[ik+im],f);
rvec_dec(force[jk+jm],f);
/*if (bFull)
fprintf(fp,"%3s%4d-%3s%4d: %6.3f %6.3f %6.3f %6.3f"
" %8.3f %8.3f %8.3f\n",
watname[ik],im+ik,watname[jk],jm+jk,
dx[XX],dx[YY],dx[ZZ],norm(dx),
dvir[XX],dvir[YY],dvir[ZZ]);*/
}
}
if (bFull)
fprintf(fp,"%3s%4d-%3s%4d: "
" %8.3f %8.3f %8.3f\n",
"SOL",i,"SOL",j,dvir[XX],dvir[YY],dvir[ZZ]);
rvec_inc(vir,dvir);
}
}
}
fprintf(fp,"There were %d interactions between the %d molecules (%.2f %%)\n",
ninter,nmol,(real)ninter/(0.5*nmol*(nmol-1)));
fprintf(fp,"Vcoul: %10.4e V12: %10.4e V6: %10.4e Vtot: %10.4e (kJ/mol)\n",
vctot/nmol,v12tot/nmol,v6tot/nmol,(vctot+v12tot+v6tot)/nmol);
pr_rvec(fp,0,"vir ",vir,DIM,TRUE);
for(m=0; (m<DIM); m++)
pres[m] = -2*PRESFAC/(det(box))*vir[m];
pr_rvec(fp,0,"pres",pres,DIM,TRUE);
pr_rvecs(fp,0,"force",force,natmol*nmol);
sfree(force);
}
static void list_trn(char *fn)
{
static real mass[5] = { 15.9994, 1.008, 1.008, 0.0, 0.0 };
int i,j=0,m,fpread,fpwrite,nframe;
rvec *x,*v,*f,fmol[2],xcm[2],torque[j],dx;
real mmm,len;
matrix box;
t_trnheader trn;
gmx_bool bOK;
printf("Going to open %s\n",fn);
fpread = open_trn(fn,"r");
fpwrite = open_tpx(NULL,"w");
gmx_fio_setdebug(fpwrite,TRUE);
mmm=mass[0]+2*mass[1];
for(i=0; (i<5); i++)
mass[i] /= mmm;
nframe = 0;
while (fread_trnheader(fpread,&trn,&bOK)) {
snew(x,trn.natoms);
snew(v,trn.natoms);
snew(f,trn.natoms);
if (fread_htrn(fpread,&trn,
trn.box_size ? box : NULL,
trn.x_size ? x : NULL,
trn.v_size ? v : NULL,
trn.f_size ? f : NULL)) {
if (trn.x_size && trn.f_size) {
printf("There are %d atoms\n",trn.natoms);
for(j=0; (j<2); j++) {
clear_rvec(xcm[j]);
clear_rvec(fmol[j]);
clear_rvec(torque[j]);
for(i=5*j; (i<5*j+5); i++) {
rvec_inc(fmol[j],f[i]);
for(m=0; (m<DIM); m++)
xcm[j][m] += mass[i%5]*x[i][m];
}
for(i=5*j; (i<5*j+5); i++) {
rvec_dec(x[i],xcm[j]);
cprod(x[i],f[i],dx);
rvec_inc(torque[j],dx);
rvec_inc(x[i],xcm[j]);
}
}
pr_rvecs(stdout,0,"FMOL ",fmol,2);
pr_rvecs(stdout,0,"TORQUE",torque,2);
printf("Distance matrix Water1-Water2\n%5s","");
for(j=0; (j<5); j++)
printf(" %10s",nm[j]);
printf("\n");
for(j=0; (j<5); j++) {
printf("%5s",nm[j]);
for(i=5; (i<10); i++) {
rvec_sub(x[i],x[j],dx);
len = sqrt(iprod(dx,dx));
printf(" %10.7f",len);
}
printf("\n");
}
}
}
sfree(x);
sfree(v);
sfree(f);
nframe++;
}
if (!bOK)
fprintf(stderr,"\nWARNING: Incomplete frame header: nr %d, t=%g\n",
nframe,trn.t);
close_tpx(fpwrite);
close_trn(fpread);
}
static void list_tpx(const char *fn, gmx_bool bShowNumbers, const char *mdpfn,
gmx_bool bSysTop)
{
FILE *gp;
int fp, indent, i, j, **gcount, atot;
t_state state;
rvec *f = NULL;
t_inputrec ir;
t_tpxheader tpx;
gmx_mtop_t mtop;
gmx_groups_t *groups;
t_topology top;
read_tpxheader(fn, &tpx, TRUE, NULL, NULL);
read_tpx_state(fn,
tpx.bIr ? &ir : NULL,
&state, tpx.bF ? f : NULL,
tpx.bTop ? &mtop : NULL);
if (mdpfn && tpx.bIr)
{
gp = gmx_fio_fopen(mdpfn, "w");
pr_inputrec(gp, 0, NULL, &(ir), TRUE);
gmx_fio_fclose(gp);
}
if (!mdpfn)
{
if (bSysTop)
{
top = gmx_mtop_t_to_t_topology(&mtop);
}
if (available(stdout, &tpx, 0, fn))
{
indent = 0;
indent = pr_title(stdout, indent, fn);
pr_inputrec(stdout, 0, "inputrec", tpx.bIr ? &(ir) : NULL, FALSE);
indent = 0;
pr_header(stdout, indent, "header", &(tpx));
if (!bSysTop)
{
pr_mtop(stdout, indent, "topology", &(mtop), bShowNumbers);
}
else
{
pr_top(stdout, indent, "topology", &(top), bShowNumbers);
}
pr_rvecs(stdout, indent, "box", tpx.bBox ? state.box : NULL, DIM);
pr_rvecs(stdout, indent, "box_rel", tpx.bBox ? state.box_rel : NULL, DIM);
pr_rvecs(stdout, indent, "boxv", tpx.bBox ? state.boxv : NULL, DIM);
pr_rvecs(stdout, indent, "pres_prev", tpx.bBox ? state.pres_prev : NULL, DIM);
pr_rvecs(stdout, indent, "svir_prev", tpx.bBox ? state.svir_prev : NULL, DIM);
pr_rvecs(stdout, indent, "fvir_prev", tpx.bBox ? state.fvir_prev : NULL, DIM);
/* leave nosehoover_xi in for now to match the tpr version */
pr_doubles(stdout, indent, "nosehoover_xi", state.nosehoover_xi, state.ngtc);
/*pr_doubles(stdout,indent,"nosehoover_vxi",state.nosehoover_vxi,state.ngtc);*/
/*pr_doubles(stdout,indent,"therm_integral",state.therm_integral,state.ngtc);*/
pr_rvecs(stdout, indent, "x", tpx.bX ? state.x : NULL, state.natoms);
pr_rvecs(stdout, indent, "v", tpx.bV ? state.v : NULL, state.natoms);
if (tpx.bF)
{
pr_rvecs(stdout, indent, "f", f, state.natoms);
}
}
groups = &mtop.groups;
snew(gcount, egcNR);
for (i = 0; (i < egcNR); i++)
{
snew(gcount[i], groups->grps[i].nr);
}
for (i = 0; (i < mtop.natoms); i++)
{
for (j = 0; (j < egcNR); j++)
{
gcount[j][ggrpnr(groups, j, i)]++;
}
}
printf("Group statistics\n");
for (i = 0; (i < egcNR); i++)
{
atot = 0;
printf("%-12s: ", gtypes[i]);
for (j = 0; (j < groups->grps[i].nr); j++)
{
printf(" %5d", gcount[i][j]);
atot += gcount[i][j];
}
printf(" (total %d atoms)\n", atot);
sfree(gcount[i]);
}
sfree(gcount);
}
done_state(&state);
sfree(f);
}
void do_force_lowlevel(FILE *fplog, gmx_large_int_t step,
t_forcerec *fr, t_inputrec *ir,
t_idef *idef, t_commrec *cr,
t_nrnb *nrnb, gmx_wallcycle_t wcycle,
t_mdatoms *md,
t_grpopts *opts,
rvec x[], history_t *hist,
rvec f[],
gmx_enerdata_t *enerd,
t_fcdata *fcd,
gmx_mtop_t *mtop,
gmx_localtop_t *top,
gmx_genborn_t *born,
t_atomtypes *atype,
gmx_bool bBornRadii,
matrix box,
real lambda,
t_graph *graph,
t_blocka *excl,
rvec mu_tot[],
int flags,
float *cycles_pme)
{
int i,status;
int donb_flags;
gmx_bool bDoEpot,bSepDVDL,bSB;
int pme_flags;
matrix boxs;
rvec box_size;
real dvdlambda,Vsr,Vlr,Vcorr=0,vdip,vcharge;
t_pbc pbc;
real dvdgb;
char buf[22];
gmx_enerdata_t ed_lam;
double lam_i;
real dvdl_dum;
#ifdef GMX_MPI
double t0=0.0,t1,t2,t3; /* time measurement for coarse load balancing */
#endif
#define PRINT_SEPDVDL(s,v,dvdl) if (bSepDVDL) fprintf(fplog,sepdvdlformat,s,v,dvdl);
GMX_MPE_LOG(ev_force_start);
set_pbc(&pbc,fr->ePBC,box);
/* Reset box */
for(i=0; (i<DIM); i++)
{
box_size[i]=box[i][i];
}
bSepDVDL=(fr->bSepDVDL && do_per_step(step,ir->nstlog));
debug_gmx();
/* do QMMM first if requested */
if(fr->bQMMM)
{
enerd->term[F_EQM] = calculate_QMMM(cr,x,f,fr,md);
}
if (bSepDVDL)
{
fprintf(fplog,"Step %s: non-bonded V and dVdl for node %d:\n",
gmx_step_str(step,buf),cr->nodeid);
}
/* Call the short range functions all in one go. */
GMX_MPE_LOG(ev_do_fnbf_start);
dvdlambda = 0;
#ifdef GMX_MPI
/*#define TAKETIME ((cr->npmenodes) && (fr->timesteps < 12))*/
#define TAKETIME FALSE
if (TAKETIME)
{
MPI_Barrier(cr->mpi_comm_mygroup);
t0=MPI_Wtime();
}
#endif
if (ir->nwall)
{
dvdlambda = do_walls(ir,fr,box,md,x,f,lambda,
enerd->grpp.ener[egLJSR],nrnb);
PRINT_SEPDVDL("Walls",0.0,dvdlambda);
enerd->dvdl_lin += dvdlambda;
}
/* If doing GB, reset dvda and calculate the Born radii */
if (ir->implicit_solvent)
{
/* wallcycle_start(wcycle,ewcGB); */
for(i=0; i<born->nr; i++)
{
fr->dvda[i]=0;
}
if(bBornRadii)
{
calc_gb_rad(cr,fr,ir,top,atype,x,&(fr->gblist),born,md,nrnb);
}
/* wallcycle_stop(wcycle, ewcGB); */
}
where();
donb_flags = 0;
if (flags & GMX_FORCE_FORCES)
{
donb_flags |= GMX_DONB_FORCES;
}
do_nonbonded(cr,fr,x,f,md,excl,
fr->bBHAM ?
enerd->grpp.ener[egBHAMSR] :
enerd->grpp.ener[egLJSR],
enerd->grpp.ener[egCOULSR],
enerd->grpp.ener[egGB],box_size,nrnb,
lambda,&dvdlambda,-1,-1,donb_flags);
/* If we do foreign lambda and we have soft-core interactions
* we have to recalculate the (non-linear) energies contributions.
*/
if (ir->n_flambda > 0 && (flags & GMX_FORCE_DHDL) && ir->sc_alpha != 0)
{
init_enerdata(mtop->groups.grps[egcENER].nr,ir->n_flambda,&ed_lam);
for(i=0; i<enerd->n_lambda; i++)
{
lam_i = (i==0 ? lambda : ir->flambda[i-1]);
dvdl_dum = 0;
reset_enerdata(&ir->opts,fr,TRUE,&ed_lam,FALSE);
do_nonbonded(cr,fr,x,f,md,excl,
fr->bBHAM ?
ed_lam.grpp.ener[egBHAMSR] :
ed_lam.grpp.ener[egLJSR],
ed_lam.grpp.ener[egCOULSR],
enerd->grpp.ener[egGB], box_size,nrnb,
lam_i,&dvdl_dum,-1,-1,
GMX_DONB_FOREIGNLAMBDA);
sum_epot(&ir->opts,&ed_lam);
enerd->enerpart_lambda[i] += ed_lam.term[F_EPOT];
}
destroy_enerdata(&ed_lam);
}
where();
/* If we are doing GB, calculate bonded forces and apply corrections
* to the solvation forces */
if (ir->implicit_solvent) {
calc_gb_forces(cr,md,born,top,atype,x,f,fr,idef,
ir->gb_algorithm,ir->sa_algorithm,nrnb,bBornRadii,&pbc,graph,enerd);
}
#ifdef GMX_MPI
if (TAKETIME)
{
t1=MPI_Wtime();
fr->t_fnbf += t1-t0;
}
#endif
if (ir->sc_alpha != 0)
{
enerd->dvdl_nonlin += dvdlambda;
}
else
{
enerd->dvdl_lin += dvdlambda;
}
Vsr = 0;
if (bSepDVDL)
{
for(i=0; i<enerd->grpp.nener; i++)
{
Vsr +=
(fr->bBHAM ?
enerd->grpp.ener[egBHAMSR][i] :
enerd->grpp.ener[egLJSR][i])
+ enerd->grpp.ener[egCOULSR][i] + enerd->grpp.ener[egGB][i];
}
}
PRINT_SEPDVDL("VdW and Coulomb SR particle-p.",Vsr,dvdlambda);
debug_gmx();
GMX_MPE_LOG(ev_do_fnbf_finish);
if (debug)
{
pr_rvecs(debug,0,"fshift after SR",fr->fshift,SHIFTS);
}
/* Shift the coordinates. Must be done before bonded forces and PPPM,
* but is also necessary for SHAKE and update, therefore it can NOT
* go when no bonded forces have to be evaluated.
*/
/* Here sometimes we would not need to shift with NBFonly,
* but we do so anyhow for consistency of the returned coordinates.
*/
if (graph)
{
shift_self(graph,box,x);
if (TRICLINIC(box))
{
inc_nrnb(nrnb,eNR_SHIFTX,2*graph->nnodes);
}
else
{
inc_nrnb(nrnb,eNR_SHIFTX,graph->nnodes);
}
}
/* Check whether we need to do bondeds or correct for exclusions */
if (fr->bMolPBC &&
((flags & GMX_FORCE_BONDED)
|| EEL_RF(fr->eeltype) || EEL_FULL(fr->eeltype)))
{
/* Since all atoms are in the rectangular or triclinic unit-cell,
* only single box vector shifts (2 in x) are required.
*/
set_pbc_dd(&pbc,fr->ePBC,cr->dd,TRUE,box);
}
debug_gmx();
if (flags & GMX_FORCE_BONDED)
{
GMX_MPE_LOG(ev_calc_bonds_start);
calc_bonds(fplog,cr->ms,
idef,x,hist,f,fr,&pbc,graph,enerd,nrnb,lambda,md,fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL, atype, born,
fr->bSepDVDL && do_per_step(step,ir->nstlog),step);
/* Check if we have to determine energy differences
* at foreign lambda's.
*/
if (ir->n_flambda > 0 && (flags & GMX_FORCE_DHDL) &&
idef->ilsort != ilsortNO_FE)
{
if (idef->ilsort != ilsortFE_SORTED)
{
gmx_incons("The bonded interactions are not sorted for free energy");
}
init_enerdata(mtop->groups.grps[egcENER].nr,ir->n_flambda,&ed_lam);
for(i=0; i<enerd->n_lambda; i++)
{
lam_i = (i==0 ? lambda : ir->flambda[i-1]);
dvdl_dum = 0;
reset_enerdata(&ir->opts,fr,TRUE,&ed_lam,FALSE);
calc_bonds_lambda(fplog,
idef,x,fr,&pbc,graph,&ed_lam,nrnb,lam_i,md,
fcd,
DOMAINDECOMP(cr) ? cr->dd->gatindex : NULL);
sum_epot(&ir->opts,&ed_lam);
enerd->enerpart_lambda[i] += ed_lam.term[F_EPOT];
}
destroy_enerdata(&ed_lam);
}
debug_gmx();
GMX_MPE_LOG(ev_calc_bonds_finish);
}
where();
*cycles_pme = 0;
if (EEL_FULL(fr->eeltype))
{
bSB = (ir->nwall == 2);
if (bSB)
{
copy_mat(box,boxs);
svmul(ir->wall_ewald_zfac,boxs[ZZ],boxs[ZZ]);
box_size[ZZ] *= ir->wall_ewald_zfac;
}
clear_mat(fr->vir_el_recip);
if (fr->bEwald)
{
if (fr->n_tpi == 0)
{
dvdlambda = 0;
Vcorr = ewald_LRcorrection(fplog,md->start,md->start+md->homenr,
cr,fr,
md->chargeA,
md->nChargePerturbed ? md->chargeB : NULL,
excl,x,bSB ? boxs : box,mu_tot,
ir->ewald_geometry,
ir->epsilon_surface,
lambda,&dvdlambda,&vdip,&vcharge);
PRINT_SEPDVDL("Ewald excl./charge/dip. corr.",Vcorr,dvdlambda);
enerd->dvdl_lin += dvdlambda;
}
else
{
if (ir->ewald_geometry != eewg3D || ir->epsilon_surface != 0)
{
gmx_fatal(FARGS,"TPI with PME currently only works in a 3D geometry with tin-foil boundary conditions");
}
/* The TPI molecule does not have exclusions with the rest
* of the system and no intra-molecular PME grid contributions
* will be calculated in gmx_pme_calc_energy.
*/
Vcorr = 0;
}
}
else
{
Vcorr = shift_LRcorrection(fplog,md->start,md->homenr,cr,fr,
md->chargeA,excl,x,TRUE,box,
fr->vir_el_recip);
}
dvdlambda = 0;
status = 0;
switch (fr->eeltype)
{
case eelPPPM:
status = gmx_pppm_do(fplog,fr->pmedata,FALSE,x,fr->f_novirsum,
md->chargeA,
box_size,fr->phi,cr,md->start,md->homenr,
nrnb,ir->pme_order,&Vlr);
break;
case eelPME:
case eelPMESWITCH:
case eelPMEUSER:
case eelPMEUSERSWITCH:
if (cr->duty & DUTY_PME)
{
if (fr->n_tpi == 0 || (flags & GMX_FORCE_STATECHANGED))
{
pme_flags = GMX_PME_SPREAD_Q | GMX_PME_SOLVE;
if (flags & GMX_FORCE_FORCES)
{
pme_flags |= GMX_PME_CALC_F;
}
if (flags & GMX_FORCE_VIRIAL)
{
pme_flags |= GMX_PME_CALC_ENER_VIR;
}
if (fr->n_tpi > 0)
{
/* We don't calculate f, but we do want the potential */
pme_flags |= GMX_PME_CALC_POT;
}
wallcycle_start(wcycle,ewcPMEMESH);
status = gmx_pme_do(fr->pmedata,
md->start,md->homenr - fr->n_tpi,
x,fr->f_novirsum,
md->chargeA,md->chargeB,
bSB ? boxs : box,cr,
DOMAINDECOMP(cr) ? dd_pme_maxshift_x(cr->dd) : 0,
DOMAINDECOMP(cr) ? dd_pme_maxshift_y(cr->dd) : 0,
nrnb,wcycle,
fr->vir_el_recip,fr->ewaldcoeff,
&Vlr,lambda,&dvdlambda,
pme_flags);
*cycles_pme = wallcycle_stop(wcycle,ewcPMEMESH);
/* We should try to do as little computation after
* this as possible, because parallel PME synchronizes
* the nodes, so we want all load imbalance of the rest
* of the force calculation to be before the PME call.
* DD load balancing is done on the whole time of
* the force call (without PME).
*/
}
if (fr->n_tpi > 0)
{
/* Determine the PME grid energy of the test molecule
* with the PME grid potential of the other charges.
*/
gmx_pme_calc_energy(fr->pmedata,fr->n_tpi,
x + md->homenr - fr->n_tpi,
md->chargeA + md->homenr - fr->n_tpi,
&Vlr);
}
PRINT_SEPDVDL("PME mesh",Vlr,dvdlambda);
}
else
{
/* Energies and virial are obtained later from the PME nodes */
/* but values have to be zeroed out here */
Vlr=0.0;
}
break;
case eelEWALD:
Vlr = do_ewald(fplog,FALSE,ir,x,fr->f_novirsum,
md->chargeA,md->chargeB,
box_size,cr,md->homenr,
fr->vir_el_recip,fr->ewaldcoeff,
lambda,&dvdlambda,fr->ewald_table);
PRINT_SEPDVDL("Ewald long-range",Vlr,dvdlambda);
break;
default:
Vlr = 0;
gmx_fatal(FARGS,"No such electrostatics method implemented %s",
eel_names[fr->eeltype]);
}
if (status != 0)
{
gmx_fatal(FARGS,"Error %d in long range electrostatics routine %s",
status,EELTYPE(fr->eeltype));
}
enerd->dvdl_lin += dvdlambda;
enerd->term[F_COUL_RECIP] = Vlr + Vcorr;
if (debug)
{
fprintf(debug,"Vlr = %g, Vcorr = %g, Vlr_corr = %g\n",
Vlr,Vcorr,enerd->term[F_COUL_RECIP]);
pr_rvecs(debug,0,"vir_el_recip after corr",fr->vir_el_recip,DIM);
pr_rvecs(debug,0,"fshift after LR Corrections",fr->fshift,SHIFTS);
}
}
else
{
if (EEL_RF(fr->eeltype))
{
dvdlambda = 0;
if (fr->eeltype != eelRF_NEC)
{
enerd->term[F_RF_EXCL] =
RF_excl_correction(fplog,fr,graph,md,excl,x,f,
fr->fshift,&pbc,lambda,&dvdlambda);
}
enerd->dvdl_lin += dvdlambda;
PRINT_SEPDVDL("RF exclusion correction",
enerd->term[F_RF_EXCL],dvdlambda);
}
}
where();
debug_gmx();
if (debug)
{
print_nrnb(debug,nrnb);
}
debug_gmx();
#ifdef GMX_MPI
if (TAKETIME)
{
t2=MPI_Wtime();
MPI_Barrier(cr->mpi_comm_mygroup);
t3=MPI_Wtime();
fr->t_wait += t3-t2;
if (fr->timesteps == 11)
{
fprintf(stderr,"* PP load balancing info: node %d, step %s, rel wait time=%3.0f%% , load string value: %7.2f\n",
cr->nodeid, gmx_step_str(fr->timesteps,buf),
100*fr->t_wait/(fr->t_wait+fr->t_fnbf),
(fr->t_fnbf+fr->t_wait)/fr->t_fnbf);
}
fr->timesteps++;
}
#endif
if (debug)
{
pr_rvecs(debug,0,"fshift after bondeds",fr->fshift,SHIFTS);
}
GMX_MPE_LOG(ev_force_finish);
}
