static void
nma_full_hessian(real * hess,
int ndim,
gmx_bool bM,
t_topology * top,
int begin,
int end,
real * eigenvalues,
real * eigenvectors)
{
int i, j, k, l;
real mass_fac, rdum;
int natoms;
natoms = top->atoms.nr;
/* divide elements hess[i][j] by sqrt(mas[i])*sqrt(mas[j]) when required */
if (bM)
{
for (i = 0; (i < natoms); i++)
{
for (j = 0; (j < DIM); j++)
{
for (k = 0; (k < natoms); k++)
{
mass_fac = gmx_invsqrt(top->atoms.atom[i].m*top->atoms.atom[k].m);
for (l = 0; (l < DIM); l++)
{
hess[(i*DIM+j)*ndim+k*DIM+l] *= mass_fac;
}
}
}
}
}
/* call diagonalization routine. */
fprintf(stderr, "\nDiagonalizing to find vectors %d through %d...\n", begin, end);
fflush(stderr);
eigensolver(hess, ndim, begin-1, end-1, eigenvalues, eigenvectors);
/* And scale the output eigenvectors */
if (bM && eigenvectors != NULL)
{
for (i = 0; i < (end-begin+1); i++)
{
for (j = 0; j < natoms; j++)
{
mass_fac = gmx_invsqrt(top->atoms.atom[j].m);
for (k = 0; (k < DIM); k++)
{
eigenvectors[i*ndim+j*DIM+k] *= mass_fac;
}
}
}
}
}
static void
nma_sparse_hessian(gmx_sparsematrix_t * sparse_hessian,
gmx_bool bM,
t_topology * top,
int neig,
real * eigenvalues,
real * eigenvectors)
{
int i, j, k;
int row, col;
real mass_fac;
int iatom, katom;
int natoms;
int ndim;
natoms = top->atoms.nr;
ndim = DIM*natoms;
/* Cannot check symmetry since we only store half matrix */
/* divide elements hess[i][j] by sqrt(mas[i])*sqrt(mas[j]) when required */
if (bM)
{
for (iatom = 0; (iatom < natoms); iatom++)
{
for (j = 0; (j < DIM); j++)
{
row = DIM*iatom+j;
for (k = 0; k < sparse_hessian->ndata[row]; k++)
{
col = sparse_hessian->data[row][k].col;
katom = col/3;
mass_fac = gmx_invsqrt(top->atoms.atom[iatom].m*top->atoms.atom[katom].m);
sparse_hessian->data[row][k].value *= mass_fac;
}
}
}
}
fprintf(stderr, "\nDiagonalizing to find eigenvectors 1 through %d...\n", neig);
fflush(stderr);
sparse_eigensolver(sparse_hessian, neig, eigenvalues, eigenvectors, 10000000);
/* Scale output eigenvectors */
if (bM && eigenvectors != NULL)
{
for (i = 0; i < neig; i++)
{
for (j = 0; j < natoms; j++)
{
mass_fac = gmx_invsqrt(top->atoms.atom[j].m);
for (k = 0; (k < DIM); k++)
{
eigenvectors[i*ndim+j*DIM+k] *= mass_fac;
}
}
}
}
}
void compute_factors_restangles(int type, const t_iparams forceparams[],
rvec delta_ante, rvec delta_post,
real *prefactor, real *ratio_ante, real *ratio_post, real *v)
{
real theta_equil, k_bending;
real cosine_theta_equil;
real c_ante, c_cros, c_post;
real norm;
real delta_cosine, cosine_theta;
real sine_theta_sq;
real term_theta_theta_equil;
k_bending = forceparams[type].harmonic.krA;
theta_equil = forceparams[type].harmonic.rA*DEG2RAD;
theta_equil = M_PI - theta_equil;
cosine_theta_equil = cos(theta_equil);
c_ante = iprod(delta_ante, delta_ante);
c_cros = iprod(delta_ante, delta_post);
c_post = iprod(delta_post, delta_post);
norm = gmx_invsqrt(c_ante * c_post);
cosine_theta = c_cros * norm;
sine_theta_sq = 1 - cosine_theta * cosine_theta;
*ratio_ante = c_cros / c_ante;
*ratio_post = c_cros / c_post;
delta_cosine = cosine_theta - cosine_theta_equil;
term_theta_theta_equil = 1 - cosine_theta * cosine_theta_equil;
*prefactor = -(k_bending) * delta_cosine * norm * term_theta_theta_equil / (sine_theta_sq * sine_theta_sq);
*v = k_bending * 0.5 * delta_cosine * delta_cosine / sine_theta_sq;
}
static void cconerr(gmx_domdec_t *dd,
int ncons,int *bla,real *bllen,rvec *x,t_pbc *pbc,
real *ncons_loc,real *ssd,real *max,int *imax)
{
real len,d,ma,ssd2,r2;
int *nlocat,count,b,im;
rvec dx;
if (dd && dd->constraints)
{
nlocat = dd_constraints_nlocalatoms(dd);
}
else
{
nlocat = 0;
}
ma = 0;
ssd2 = 0;
im = 0;
count = 0;
for(b=0;b<ncons;b++)
{
if (pbc)
{
pbc_dx_aiuc(pbc,x[bla[2*b]],x[bla[2*b+1]],dx);
}
else {
rvec_sub(x[bla[2*b]],x[bla[2*b+1]],dx);
}
r2 = norm2(dx);
len = r2*gmx_invsqrt(r2);
d = fabs(len/bllen[b]-1);
if (d > ma && (nlocat==NULL || nlocat[b]))
{
ma = d;
im = b;
}
if (nlocat == NULL)
{
ssd2 += d*d;
count++;
}
else
{
ssd2 += nlocat[b]*d*d;
count += nlocat[b];
}
}
*ncons_loc = (nlocat ? 0.5 : 1)*count;
*ssd = (nlocat ? 0.5 : 1)*ssd2;
*max = ma;
*imax = im;
}
int main(int argc,char *argv[])
{
real x,y,z,diff,av;
int i;
printf("%12s %12s %12s %12s %12s\n","X","invsqrt(X)","1/sqrt(X)","Abs. Diff.","Rel. Diff.");
for(i=1; (i<1000); i++) {
x = i*1.0;
y = gmx_invsqrt(x);
z = 1.0/sqrt(x);
diff = y-z;
av = 0.5*(y+z);
printf("%12.5e %12.5e %12.5e %12.5e %12.5e\n",x,y,z,diff,diff/z);
}
return 0;
}
/* This might logically belong better in the nb_generic.c module, but it is only
* used in do_nonbonded_listed(), and we want it to be inlined there to avoid an
* extra functional call for every single pair listed in the topology.
*/
static real
nb_evaluate_single(real r2, real tabscale, real *vftab,
real qq, real c6, real c12, real *velec, real *vvdw)
{
real rinv, r, rtab, eps, eps2, Y, F, Geps, Heps2, Fp, VVe, FFe, VVd, FFd, VVr, FFr, fscal;
int ntab;
/* Do the tabulated interactions - first table lookup */
rinv = gmx_invsqrt(r2);
r = r2*rinv;
rtab = r*tabscale;
ntab = rtab;
eps = rtab-ntab;
eps2 = eps*eps;
ntab = 12*ntab;
/* Electrostatics */
Y = vftab[ntab];
F = vftab[ntab+1];
Geps = eps*vftab[ntab+2];
Heps2 = eps2*vftab[ntab+3];
Fp = F+Geps+Heps2;
VVe = Y+eps*Fp;
FFe = Fp+Geps+2.0*Heps2;
/* Dispersion */
Y = vftab[ntab+4];
F = vftab[ntab+5];
Geps = eps*vftab[ntab+6];
Heps2 = eps2*vftab[ntab+7];
Fp = F+Geps+Heps2;
VVd = Y+eps*Fp;
FFd = Fp+Geps+2.0*Heps2;
/* Repulsion */
Y = vftab[ntab+8];
F = vftab[ntab+9];
Geps = eps*vftab[ntab+10];
Heps2 = eps2*vftab[ntab+11];
Fp = F+Geps+Heps2;
VVr = Y+eps*Fp;
FFr = Fp+Geps+2.0*Heps2;
*velec = qq*VVe;
*vvdw = c6*VVd+c12*VVr;
fscal = -(qq*FFe+c6*FFd+c12*FFr)*tabscale*rinv;
return fscal;
}
/* Estimate the direct space part error of the SPME Ewald sum */
static real estimate_direct(
t_inputinfo *info
)
{
real e_dir=0; /* Error estimate */
real beta=0; /* Splitting parameter (1/nm) */
real r_coulomb=0; /* Cut-off in direct space */
beta = info->ewald_beta[0];
r_coulomb = info->rcoulomb[0];
e_dir = 2.0 * info->q2all * gmx_invsqrt( info->q2allnr * r_coulomb * info->volume );
e_dir *= exp (-beta*beta*r_coulomb*r_coulomb);
return ONE_4PI_EPS0*e_dir;
}
void compute_factors_cbtdihs(int type, const t_iparams forceparams[],
rvec delta_ante, rvec delta_crnt, rvec delta_post,
rvec f_phi_ai, rvec f_phi_aj, rvec f_phi_ak, rvec f_phi_al,
rvec f_theta_ante_ai, rvec f_theta_ante_aj, rvec f_theta_ante_ak,
rvec f_theta_post_aj, rvec f_theta_post_ak, rvec f_theta_post_al,
real * v)
{
int j, d;
real torsion_coef[NR_CBTDIHS];
real c_self_ante, c_self_crnt, c_self_post;
real c_cros_ante, c_cros_acrs, c_cros_post;
real c_prod, d_ante, d_post;
real norm_phi, norm_theta_ante, norm_theta_post;
real cosine_phi, cosine_theta_ante, cosine_theta_post;
real sine_theta_ante_sq, sine_theta_post_sq;
real sine_theta_ante, sine_theta_post;
real prefactor_phi;
real ratio_phi_ante, ratio_phi_post;
real r1, r2;
real factor_phi_ai_ante, factor_phi_ai_crnt, factor_phi_ai_post;
real factor_phi_aj_ante, factor_phi_aj_crnt, factor_phi_aj_post;
real factor_phi_ak_ante, factor_phi_ak_crnt, factor_phi_ak_post;
real factor_phi_al_ante, factor_phi_al_crnt, factor_phi_al_post;
real prefactor_theta_ante, ratio_theta_ante_ante, ratio_theta_ante_crnt;
real prefactor_theta_post, ratio_theta_post_crnt, ratio_theta_post_post;
/* The formula for combined bending-torsion potential (see file "restcbt.h") contains
* in its expression not only the dihedral angle \f[\phi\f] but also \f[\theta_{i-1}\f]
* (theta_ante bellow) and \f[\theta_{i}\f] (theta_post bellow)--- the adjacent bending
* angles. The forces for the particles ai, aj, ak, al have components coming from the
* derivatives of the potential with respect to all three angles.
* This function is organised in 4 parts
* PART 1 - Computes force factors common to all the derivatives for the four particles
* PART 2 - Computes the force components due to the derivatives of dihedral angle Phi
* PART 3 - Computes the force components due to the derivatives of bending angle Theta_Ante
* PART 4 - Computes the force components due to the derivatives of bending angle Theta_Post
* Bellow we will respct thuis structure */
/* PART 1 - COMPUTES FORCE FACTORS COMMON TO ALL DERIVATIVES FOR THE FOUR PARTICLES */
for (j = 0; (j < NR_CBTDIHS); j++)
{
torsion_coef[j] = forceparams[type].cbtdihs.cbtcA[j];
}
/* Computation of the cosine of the dihedral angle. The scalar ("dot") product method
* is used. c_*_* cummulate the scalar products of the differences of particles
* positions while c_prod, d_ante and d_post are differences of products of scalar
* terms that are parts of the derivatives of forces */
c_self_ante = iprod(delta_ante, delta_ante);
c_self_crnt = iprod(delta_crnt, delta_crnt);
c_self_post = iprod(delta_post, delta_post);
c_cros_ante = iprod(delta_ante, delta_crnt);
c_cros_acrs = iprod(delta_ante, delta_post);
c_cros_post = iprod(delta_crnt, delta_post);
c_prod = c_cros_ante * c_cros_post - c_self_crnt * c_cros_acrs;
d_ante = c_self_ante * c_self_crnt - c_cros_ante * c_cros_ante;
d_post = c_self_post * c_self_crnt - c_cros_post * c_cros_post;
/* When three consecutive beads align, we obtain values close to zero.
Here we avoid small values to prevent round-off errors. */
if (d_ante < GMX_REAL_EPS)
{
d_ante = GMX_REAL_EPS;
}
if (d_post < GMX_REAL_EPS)
{
d_post = GMX_REAL_EPS;
}
/* Computations of cosines */
norm_phi = gmx_invsqrt(d_ante * d_post);
norm_theta_ante = gmx_invsqrt(c_self_ante * c_self_crnt);
norm_theta_post = gmx_invsqrt(c_self_crnt * c_self_post);
cosine_phi = c_prod * norm_phi;
cosine_theta_ante = c_cros_ante * norm_theta_ante;
cosine_theta_post = c_cros_post * norm_theta_post;
sine_theta_ante_sq = 1 - cosine_theta_ante * cosine_theta_ante;
sine_theta_post_sq = 1 - cosine_theta_post * cosine_theta_post;
/* It is possible that cosine_theta is slightly bigger than 1.0 due to round-off errors. */
if (sine_theta_ante_sq < 0.0)
{
sine_theta_ante_sq = 0.0;
}
if (sine_theta_post_sq < 0.0)
{
sine_theta_post_sq = 0.0;
}
sine_theta_ante = sqrt(sine_theta_ante_sq);
sine_theta_post = sqrt(sine_theta_post_sq);
/* PART 2 - COMPUTES FORCE COMPONENTS DUE TO DERIVATIVES TO DIHEDRAL ANGLE PHI */
/* Computation of ratios */
ratio_phi_ante = c_prod / d_ante;
ratio_phi_post = c_prod / d_post;
/* Computation of the prefactor */
/* Computing 2nd power */
r1 = cosine_phi;
prefactor_phi = -torsion_coef[0] * norm_phi * (torsion_coef[2] + torsion_coef[3] * 2.0 * cosine_phi + torsion_coef[4] * 3.0 * (r1 * r1) + 4*torsion_coef[5]*r1*r1*r1) *
sine_theta_ante_sq * sine_theta_ante * sine_theta_post_sq * sine_theta_post;
/* Computation of factors (important for gaining speed). Factors factor_phi_* are coming from the
* derivatives of the torsion angle (phi) with respect to the beads ai, aj, al, ak,
* (four) coordinates and they are multiplied in the force computations with the
* differences of the particles positions stored in parameters delta_ante,
* delta_crnt, delta_post. For formulas see file "restcbt.h" */
factor_phi_ai_ante = ratio_phi_ante * c_self_crnt;
factor_phi_ai_crnt = -c_cros_post - ratio_phi_ante * c_cros_ante;
factor_phi_ai_post = c_self_crnt;
factor_phi_aj_ante = -c_cros_post - ratio_phi_ante * (c_self_crnt + c_cros_ante);
factor_phi_aj_crnt = c_cros_post + c_cros_acrs * 2.0 + ratio_phi_ante * (c_self_ante + c_cros_ante) + ratio_phi_post * c_self_post;
factor_phi_aj_post = -(c_cros_ante + c_self_crnt) - ratio_phi_post * c_cros_post;
factor_phi_ak_ante = c_cros_post + c_self_crnt + ratio_phi_ante * c_cros_ante;
factor_phi_ak_crnt = -(c_cros_ante + c_cros_acrs * 2.0) - ratio_phi_ante * c_self_ante - ratio_phi_post * (c_self_post + c_cros_post);
factor_phi_ak_post = c_cros_ante + ratio_phi_post * (c_self_crnt + c_cros_post);
factor_phi_al_ante = -c_self_crnt;
factor_phi_al_crnt = c_cros_ante + ratio_phi_post * c_cros_post;
factor_phi_al_post = -ratio_phi_post * c_self_crnt;
/* Computation of forces due to the derivatives of dihedral angle phi*/
for (d = 0; d < DIM; d++)
{
f_phi_ai[d] = prefactor_phi * (factor_phi_ai_ante * delta_ante[d] + factor_phi_ai_crnt * delta_crnt[d] + factor_phi_ai_post * delta_post[d]);
f_phi_aj[d] = prefactor_phi * (factor_phi_aj_ante * delta_ante[d] + factor_phi_aj_crnt * delta_crnt[d] + factor_phi_aj_post * delta_post[d]);
f_phi_ak[d] = prefactor_phi * (factor_phi_ak_ante * delta_ante[d] + factor_phi_ak_crnt * delta_crnt[d] + factor_phi_ak_post * delta_post[d]);
f_phi_al[d] = prefactor_phi * (factor_phi_al_ante * delta_ante[d] + factor_phi_al_crnt * delta_crnt[d] + factor_phi_al_post * delta_post[d]);
}
/* PART 3 - COMPUTES THE FORCE COMPONENTS DUE TO THE DERIVATIVES OF BENDING ANGLE THETHA_ANTHE */
/* Computation of ratios */
ratio_theta_ante_ante = c_cros_ante / c_self_ante;
ratio_theta_ante_crnt = c_cros_ante / c_self_crnt;
/* Computation of the prefactor */
/* Computing 2nd power */
r1 = cosine_phi;
/* Computing 3rd power */
r2 = cosine_phi;
prefactor_theta_ante = -torsion_coef[0] * norm_theta_ante * ( torsion_coef[1] + torsion_coef[2] * cosine_phi + torsion_coef[3] * (r1 * r1) +
torsion_coef[4] * (r2 * (r2 * r2))+ torsion_coef[5] * (r2 * (r2 * (r2 * r2)))) * (-3.0) * cosine_theta_ante * sine_theta_ante * sine_theta_post_sq * sine_theta_post;
/* Computation of forces due to the derivatives of bending angle theta_ante */
for (d = 0; d < DIM; d++)
{
f_theta_ante_ai[d] = prefactor_theta_ante * (ratio_theta_ante_ante * delta_ante[d] - delta_crnt[d]);
f_theta_ante_aj[d] = prefactor_theta_ante * ((ratio_theta_ante_crnt + 1.0) * delta_crnt[d] - (ratio_theta_ante_ante + 1.0) * delta_ante[d]);
f_theta_ante_ak[d] = prefactor_theta_ante * (delta_ante[d] - ratio_theta_ante_crnt * delta_crnt[d]);
}
/* PART 4 - COMPUTES THE FORCE COMPONENTS DUE TO THE DERIVATIVES OF THE BENDING ANGLE THETA_POST */
/* Computation of ratios */
ratio_theta_post_crnt = c_cros_post / c_self_crnt;
ratio_theta_post_post = c_cros_post / c_self_post;
/* Computation of the prefactor */
/* Computing 2nd power */
r1 = cosine_phi;
/* Computing 3rd power */
r2 = cosine_phi;
prefactor_theta_post = -torsion_coef[0] * norm_theta_post * (torsion_coef[1] + torsion_coef[2] * cosine_phi + torsion_coef[3] * (r1 * r1) +
torsion_coef[4] * (r2 * (r2 * r2)) + torsion_coef[5] * (r2 * (r2 * (r2 * r2)))) * sine_theta_ante_sq * sine_theta_ante * (-3.0) * cosine_theta_post * sine_theta_post;
/* Computation of forces due to the derivatives of bending angle Theta_Post */
for (d = 0; d < DIM; d++)
{
f_theta_post_aj[d] = prefactor_theta_post * (ratio_theta_post_crnt * delta_crnt[d] - delta_post[d]);
f_theta_post_ak[d] = prefactor_theta_post * ((ratio_theta_post_post + 1.0) * delta_post[d] - (ratio_theta_post_crnt + 1.0) * delta_crnt[d]);
f_theta_post_al[d] = prefactor_theta_post * (delta_crnt[d] - ratio_theta_post_post * delta_post[d]);
}
r1 = cosine_phi;
r2 = cosine_phi;
/* Contribution to energy - for formula see file "restcbt.h" */
*v = torsion_coef[0] * (torsion_coef[1] + torsion_coef[2] * cosine_phi + torsion_coef[3] * (r1 * r1) +
torsion_coef[4] * (r2 * (r2 * r2)) + torsion_coef[5] * (r2 * (r2 * (r2 * r2)))) * sine_theta_ante_sq *
sine_theta_ante * sine_theta_post_sq * sine_theta_post;
}
void
nb_kernel_allvsallgb(t_forcerec * fr,
t_mdatoms * mdatoms,
t_blocka * excl,
real * x,
real * f,
real * Vc,
real * Vvdw,
real * vpol,
int * outeriter,
int * inneriter,
void * work)
{
gmx_allvsall_data_t *aadata;
int natoms;
int ni0,ni1;
int nj0,nj1,nj2;
int i,j,k;
real * charge;
int * type;
real facel;
real * pvdw;
int ggid;
int * mask;
real * GBtab;
real gbfactor;
real * invsqrta;
real * dvda;
real vgbtot,dvdasum;
int nnn,n0;
real ix,iy,iz,iq;
real fix,fiy,fiz;
real jx,jy,jz,qq;
real dx,dy,dz;
real tx,ty,tz;
real rsq,rinv,rinvsq,rinvsix;
real vcoul,vctot;
real c6,c12,Vvdw6,Vvdw12,Vvdwtot;
real fscal,dvdatmp,fijC,vgb;
real Y,F,Fp,Geps,Heps2,VV,FF,eps,eps2,r,rt;
real dvdaj,gbscale,isaprod,isai,isaj,gbtabscale;
charge = mdatoms->chargeA;
type = mdatoms->typeA;
gbfactor = ((1.0/fr->epsilon_r) - (1.0/fr->gb_epsilon_solvent));
facel = fr->epsfac;
GBtab = fr->gbtab.tab;
gbtabscale = fr->gbtab.scale;
invsqrta = fr->invsqrta;
dvda = fr->dvda;
natoms = mdatoms->nr;
ni0 = mdatoms->start;
ni1 = mdatoms->start+mdatoms->homenr;
aadata = *((gmx_allvsall_data_t **)work);
if(aadata==NULL)
{
setup_aadata(&aadata,excl,natoms,type,fr->ntype,fr->nbfp);
*((gmx_allvsall_data_t **)work) = aadata;
}
for(i=ni0; i<ni1; i++)
{
/* We assume shifts are NOT used for all-vs-all interactions */
/* Load i atom data */
ix = x[3*i];
iy = x[3*i+1];
iz = x[3*i+2];
iq = facel*charge[i];
isai = invsqrta[i];
pvdw = aadata->pvdwparam[type[i]];
/* Zero the potential energy for this list */
Vvdwtot = 0.0;
vctot = 0.0;
vgbtot = 0.0;
dvdasum = 0.0;
/* Clear i atom forces */
fix = 0.0;
fiy = 0.0;
fiz = 0.0;
/* Load limits for loop over neighbors */
nj0 = aadata->jindex[3*i];
nj1 = aadata->jindex[3*i+1];
nj2 = aadata->jindex[3*i+2];
mask = aadata->exclusion_mask[i];
/* Prologue part, including exclusion mask */
for(j=nj0; j<nj1; j++,mask++)
{
if(*mask!=0)
{
k = j%natoms;
/* load j atom coordinates */
jx = x[3*k];
jy = x[3*k+1];
jz = x[3*k+2];
/* Calculate distance */
dx = ix - jx;
dy = iy - jy;
dz = iz - jz;
rsq = dx*dx+dy*dy+dz*dz;
/* Calculate 1/r and 1/r2 */
rinv = gmx_invsqrt(rsq);
/* Load parameters for j atom */
isaj = invsqrta[k];
isaprod = isai*isaj;
qq = iq*charge[k];
vcoul = qq*rinv;
fscal = vcoul*rinv;
qq = isaprod*(-qq)*gbfactor;
gbscale = isaprod*gbtabscale;
c6 = pvdw[2*k];
c12 = pvdw[2*k+1];
rinvsq = rinv*rinv;
/* Tabulated Generalized-Born interaction */
dvdaj = dvda[k];
r = rsq*rinv;
/* Calculate table index */
rt = r*gbscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
Y = GBtab[nnn];
F = GBtab[nnn+1];
Geps = eps*GBtab[nnn+2];
Heps2 = eps2*GBtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vgb = qq*VV;
fijC = qq*FF*gbscale;
dvdatmp = -0.5*(vgb+fijC*r);
dvdasum = dvdasum + dvdatmp;
dvda[k] = dvdaj+dvdatmp*isaj*isaj;
vctot = vctot + vcoul;
vgbtot = vgbtot + vgb;
/* Lennard-Jones interaction */
rinvsix = rinvsq*rinvsq*rinvsq;
Vvdw6 = c6*rinvsix;
Vvdw12 = c12*rinvsix*rinvsix;
Vvdwtot = Vvdwtot+Vvdw12-Vvdw6;
fscal = (12.0*Vvdw12-6.0*Vvdw6)*rinvsq-(fijC-fscal)*rinv;
/* Calculate temporary vectorial force */
tx = fscal*dx;
ty = fscal*dy;
tz = fscal*dz;
/* Increment i atom force */
fix = fix + tx;
fiy = fiy + ty;
fiz = fiz + tz;
/* Decrement j atom force */
f[3*k] = f[3*k] - tx;
f[3*k+1] = f[3*k+1] - ty;
f[3*k+2] = f[3*k+2] - tz;
}
/* Inner loop uses 38 flops/iteration */
}
/* Main part, no exclusions */
for(j=nj1; j<nj2; j++)
{
k = j%natoms;
/* load j atom coordinates */
jx = x[3*k];
jy = x[3*k+1];
jz = x[3*k+2];
/* Calculate distance */
dx = ix - jx;
dy = iy - jy;
dz = iz - jz;
rsq = dx*dx+dy*dy+dz*dz;
/* Calculate 1/r and 1/r2 */
rinv = gmx_invsqrt(rsq);
/* Load parameters for j atom */
isaj = invsqrta[k];
isaprod = isai*isaj;
qq = iq*charge[k];
vcoul = qq*rinv;
fscal = vcoul*rinv;
qq = isaprod*(-qq)*gbfactor;
gbscale = isaprod*gbtabscale;
c6 = pvdw[2*k];
c12 = pvdw[2*k+1];
rinvsq = rinv*rinv;
/* Tabulated Generalized-Born interaction */
dvdaj = dvda[k];
r = rsq*rinv;
/* Calculate table index */
rt = r*gbscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
Y = GBtab[nnn];
F = GBtab[nnn+1];
Geps = eps*GBtab[nnn+2];
Heps2 = eps2*GBtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vgb = qq*VV;
fijC = qq*FF*gbscale;
dvdatmp = -0.5*(vgb+fijC*r);
dvdasum = dvdasum + dvdatmp;
dvda[k] = dvdaj+dvdatmp*isaj*isaj;
vctot = vctot + vcoul;
vgbtot = vgbtot + vgb;
/* Lennard-Jones interaction */
rinvsix = rinvsq*rinvsq*rinvsq;
Vvdw6 = c6*rinvsix;
Vvdw12 = c12*rinvsix*rinvsix;
Vvdwtot = Vvdwtot+Vvdw12-Vvdw6;
fscal = (12.0*Vvdw12-6.0*Vvdw6)*rinvsq-(fijC-fscal)*rinv;
/* Calculate temporary vectorial force */
tx = fscal*dx;
ty = fscal*dy;
tz = fscal*dz;
/* Increment i atom force */
fix = fix + tx;
fiy = fiy + ty;
fiz = fiz + tz;
/* Decrement j atom force */
f[3*k] = f[3*k] - tx;
f[3*k+1] = f[3*k+1] - ty;
f[3*k+2] = f[3*k+2] - tz;
/* Inner loop uses 38 flops/iteration */
}
f[3*i] += fix;
f[3*i+1] += fiy;
f[3*i+2] += fiz;
/* Add potential energies to the group for this list */
ggid = 0;
Vc[ggid] = Vc[ggid] + vctot;
Vvdw[ggid] = Vvdw[ggid] + Vvdwtot;
vpol[ggid] = vpol[ggid] + vgbtot;
dvda[i] = dvda[i] + dvdasum*isai*isai;
/* Outer loop uses 6 flops/iteration */
}
/* Write outer/inner iteration count to pointers */
*outeriter = ni1-ni0;
*inneriter = (ni1-ni0)*natoms/2;
}
int gmx_nmens(int argc, char *argv[])
{
const char *desc[] = {
"[THISMODULE] generates an ensemble around an average structure",
"in a subspace that is defined by a set of normal modes (eigenvectors).",
"The eigenvectors are assumed to be mass-weighted.",
"The position along each eigenvector is randomly taken from a Gaussian",
"distribution with variance kT/eigenvalue.[PAR]",
"By default the starting eigenvector is set to 7, since the first six",
"normal modes are the translational and rotational degrees of freedom."
};
static int nstruct = 100, first = 7, last = -1, seed = -1;
static real temp = 300.0;
t_pargs pa[] = {
{ "-temp", FALSE, etREAL, {&temp},
"Temperature in Kelvin" },
{ "-seed", FALSE, etINT, {&seed},
"Random seed, -1 generates a seed from time and pid" },
{ "-num", FALSE, etINT, {&nstruct},
"Number of structures to generate" },
{ "-first", FALSE, etINT, {&first},
"First eigenvector to use (-1 is select)" },
{ "-last", FALSE, etINT, {&last},
"Last eigenvector to use (-1 is till the last)" }
};
#define NPA asize(pa)
t_trxstatus *out;
int status, trjout;
t_topology top;
int ePBC;
t_atoms *atoms;
rvec *xtop, *xref, *xav, *xout1, *xout2;
gmx_bool bDMR, bDMA, bFit;
int nvec, *eignr = NULL;
rvec **eigvec = NULL;
matrix box;
real *eigval, totmass, *invsqrtm, t, disp;
int natoms, neigval;
char *grpname, title[STRLEN];
const char *indexfile;
int i, j, d, s, v;
int nout, *iout, noutvec, *outvec;
atom_id *index;
real rfac, invfr, rhalf, jr;
int * eigvalnr;
output_env_t oenv;
gmx_rng_t rng;
unsigned long jran;
const unsigned long im = 0xffff;
const unsigned long ia = 1093;
const unsigned long ic = 18257;
t_filenm fnm[] = {
{ efTRN, "-v", "eigenvec", ffREAD },
{ efXVG, "-e", "eigenval", ffREAD },
{ efTPS, NULL, NULL, ffREAD },
{ efNDX, NULL, NULL, ffOPTRD },
{ efTRO, "-o", "ensemble", ffWRITE }
};
#define NFILE asize(fnm)
if (!parse_common_args(&argc, argv, PCA_BE_NICE,
NFILE, fnm, NPA, pa, asize(desc), desc, 0, NULL, &oenv))
{
return 0;
}
indexfile = ftp2fn_null(efNDX, NFILE, fnm);
read_eigenvectors(opt2fn("-v", NFILE, fnm), &natoms, &bFit,
&xref, &bDMR, &xav, &bDMA, &nvec, &eignr, &eigvec, &eigval);
read_tps_conf(ftp2fn(efTPS, NFILE, fnm), title, &top, &ePBC, &xtop, NULL, box, bDMA);
atoms = &top.atoms;
printf("\nSelect an index group of %d elements that corresponds to the eigenvectors\n", natoms);
get_index(atoms, indexfile, 1, &i, &index, &grpname);
if (i != natoms)
{
gmx_fatal(FARGS, "you selected a group with %d elements instead of %d",
i, natoms);
}
printf("\n");
snew(invsqrtm, natoms);
if (bDMA)
{
for (i = 0; (i < natoms); i++)
{
invsqrtm[i] = gmx_invsqrt(atoms->atom[index[i]].m);
}
}
else
{
for (i = 0; (i < natoms); i++)
{
invsqrtm[i] = 1.0;
}
}
if (last == -1)
{
last = natoms*DIM;
}
if (first > -1)
{
/* make an index from first to last */
nout = last-first+1;
snew(iout, nout);
for (i = 0; i < nout; i++)
{
iout[i] = first-1+i;
}
}
else
{
printf("Select eigenvectors for output, end your selection with 0\n");
nout = -1;
iout = NULL;
do
{
nout++;
srenew(iout, nout+1);
if (1 != scanf("%d", &iout[nout]))
{
gmx_fatal(FARGS, "Error reading user input");
}
iout[nout]--;
}
while (iout[nout] >= 0);
printf("\n");
}
/* make an index of the eigenvectors which are present */
snew(outvec, nout);
noutvec = 0;
for (i = 0; i < nout; i++)
{
j = 0;
while ((j < nvec) && (eignr[j] != iout[i]))
{
j++;
}
if ((j < nvec) && (eignr[j] == iout[i]))
{
outvec[noutvec] = j;
iout[noutvec] = iout[i];
noutvec++;
}
}
fprintf(stderr, "%d eigenvectors selected for output\n", noutvec);
if (seed == -1)
{
seed = (int)gmx_rng_make_seed();
rng = gmx_rng_init(seed);
}
else
{
rng = gmx_rng_init(seed);
}
fprintf(stderr, "Using seed %d and a temperature of %g K\n", seed, temp);
snew(xout1, natoms);
snew(xout2, atoms->nr);
out = open_trx(ftp2fn(efTRO, NFILE, fnm), "w");
jran = (unsigned long)((real)im*gmx_rng_uniform_real(rng));
gmx_rng_destroy(rng);
for (s = 0; s < nstruct; s++)
{
for (i = 0; i < natoms; i++)
{
copy_rvec(xav[i], xout1[i]);
}
for (j = 0; j < noutvec; j++)
{
v = outvec[j];
/* (r-0.5) n times: var_n = n * var_1 = n/12
n=4: var_n = 1/3, so multiply with 3 */
rfac = sqrt(3.0 * BOLTZ*temp/eigval[iout[j]]);
rhalf = 2.0*rfac;
rfac = rfac/(real)im;
jran = (jran*ia+ic) & im;
jr = (real)jran;
jran = (jran*ia+ic) & im;
jr += (real)jran;
jran = (jran*ia+ic) & im;
jr += (real)jran;
jran = (jran*ia+ic) & im;
jr += (real)jran;
disp = rfac * jr - rhalf;
for (i = 0; i < natoms; i++)
{
for (d = 0; d < DIM; d++)
{
xout1[i][d] += disp*eigvec[v][i][d]*invsqrtm[i];
}
}
}
for (i = 0; i < natoms; i++)
{
copy_rvec(xout1[i], xout2[index[i]]);
}
t = s+1;
write_trx(out, natoms, index, atoms, 0, t, box, xout2, NULL, NULL);
fprintf(stderr, "\rGenerated %d structures", s+1);
}
fprintf(stderr, "\n");
close_trx(out);
return 0;
}
/*
* Gromacs nonbonded kernel pf_nb_kernel320
* Coulomb interaction: Tabulated
* VdW interaction: Buckingham
* water optimization: No
* Calculate forces: yes
*/
void pf_nb_kernel320(
int * p_nri,
int * iinr,
int * jindex,
int * jjnr,
int * shift,
real * shiftvec,
real * fshift,
int * gid,
real * pos,
real * faction,
real * charge,
real * p_facel,
real * p_krf,
real * p_crf,
real * Vc,
int * type,
int * p_ntype,
real * vdwparam,
real * Vvdw,
real * p_tabscale,
real * VFtab,
real * invsqrta,
real * dvda,
real * p_gbtabscale,
real * GBtab,
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
real * work,
t_pf_global * pf_global)
{
int nri,ntype,nthreads;
real facel,krf,crf,tabscale,gbtabscale;
int n,ii,is3,ii3,k,nj0,nj1,jnr,j3,ggid;
int nn0,nn1,nouter,ninner;
real shX,shY,shZ;
real fscal,tx,ty,tz;
real rinvsq;
real iq;
real qq,vcoul,vctot;
int nti;
int tj;
real rinvsix;
real Vvdw6,Vvdwtot;
real r,rt,eps,eps2;
int n0,nnn;
real Y,F,Geps,Heps2,Fp,VV;
real FF;
real fijC;
real Vvdwexp,br;
real ix1,iy1,iz1,fix1,fiy1,fiz1;
real jx1,jy1,jz1;
real dx11,dy11,dz11,rsq11,rinv11;
real c6,cexp1,cexp2;
real pf_coul, pf_lj;
nri = *p_nri;
ntype = *p_ntype;
nthreads = *p_nthreads;
facel = *p_facel;
krf = *p_krf;
crf = *p_crf;
tabscale = *p_tabscale;
/* Reset outer and inner iteration counters */
nouter = 0;
ninner = 0;
/* Loop over thread workunits */
do
{
#ifdef GMX_THREAD_SHM_FDECOMP
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
/* Take successively smaller chunks (at least 10 lists) */
nn1 = nn0+(nri-nn0)/(2*nthreads)+10;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
#else
nn0 = 0;
nn1 = nri;
#endif
/* Start outer loop over neighborlists */
for(n=nn0; (n<nn1); n++)
{
/* Load shift vector for this list */
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
/* Load limits for loop over neighbors */
nj0 = jindex[n];
nj1 = jindex[n+1];
/* Get outer coordinate index */
ii = iinr[n];
ii3 = 3*ii;
/* Load i atom data, add shift vector */
ix1 = shX + pos[ii3+0];
iy1 = shY + pos[ii3+1];
iz1 = shZ + pos[ii3+2];
/* Load parameters for i atom */
iq = facel*charge[ii];
nti = 3*ntype*type[ii];
/* Zero the potential energy for this list */
vctot = 0;
Vvdwtot = 0;
/* Clear i atom forces */
fix1 = 0;
fiy1 = 0;
fiz1 = 0;
for(k=nj0; (k<nj1); k++)
{
/* Get j neighbor index, and coordinate index */
jnr = jjnr[k];
j3 = 3*jnr;
/* load j atom coordinates */
jx1 = pos[j3+0];
jy1 = pos[j3+1];
jz1 = pos[j3+2];
/* Calculate distance */
dx11 = ix1 - jx1;
dy11 = iy1 - jy1;
dz11 = iz1 - jz1;
rsq11 = dx11*dx11+dy11*dy11+dz11*dz11;
/* Calculate 1/r and 1/r2 */
rinv11 = gmx_invsqrt(rsq11);
/* Load parameters for j atom */
qq = iq*charge[jnr];
tj = nti+3*type[jnr];
c6 = vdwparam[tj];
cexp1 = vdwparam[tj+1];
cexp2 = vdwparam[tj+2];
rinvsq = rinv11*rinv11;
/* Calculate table index */
r = rsq11*rinv11;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
/* Buckingham interaction */
rinvsix = rinvsq*rinvsq*rinvsq;
Vvdw6 = c6*rinvsix;
br = cexp2*rsq11*rinv11;
Vvdwexp = cexp1*exp(-br);
Vvdwtot = Vvdwtot+Vvdwexp-Vvdw6;
pf_coul = -((fijC)*tabscale)*rinv11;
pf_lj = (br*Vvdwexp-6.0*Vvdw6)*rinvsq;
fscal = pf_lj+pf_coul;
/* Calculate temporary vectorial force */
tx = fscal*dx11;
ty = fscal*dy11;
tz = fscal*dz11;
/* Increment i atom force */
fix1 = fix1 + tx;
fiy1 = fiy1 + ty;
fiz1 = fiz1 + tz;
/* Decrement j atom force */
faction[j3+0] = faction[j3+0] - tx;
faction[j3+1] = faction[j3+1] - ty;
faction[j3+2] = faction[j3+2] - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded(pf_global, ii, jnr, pf_coul, pf_lj, dx11, dy11, dz11);
/* Inner loop uses 81 flops/iteration */
}
/* Add i forces to mem and shifted force list */
faction[ii3+0] = faction[ii3+0] + fix1;
faction[ii3+1] = faction[ii3+1] + fiy1;
faction[ii3+2] = faction[ii3+2] + fiz1;
fshift[is3] = fshift[is3]+fix1;
fshift[is3+1] = fshift[is3+1]+fiy1;
fshift[is3+2] = fshift[is3+2]+fiz1;
/* Add potential energies to the group for this list */
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
Vvdw[ggid] = Vvdw[ggid] + Vvdwtot;
/* Increment number of inner iterations */
ninner = ninner + nj1 - nj0;
/* Outer loop uses 12 flops/iteration */
}
/* Increment number of outer iterations */
nouter = nouter + nn1 - nn0;
}
while (nn1<nri);
/* Write outer/inner iteration count to pointers */
*outeriter = nouter;
*inneriter = ninner;
}
/*
* Gromacs nonbonded kernel nb_kernel231
* Coulomb interaction: Reaction field
* VdW interaction: Tabulated
* water optimization: SPC/TIP3P - other atoms
* Calculate forces: yes
*/
void nb_kernel231(
int * p_nri,
int * iinr,
int * jindex,
int * jjnr,
int * shift,
real * shiftvec,
real * fshift,
int * gid,
real * pos,
real * faction,
real * charge,
real * p_facel,
real * p_krf,
real * p_crf,
real * Vc,
int * type,
int * p_ntype,
real * vdwparam,
real * Vvdw,
real * p_tabscale,
real * VFtab,
real * invsqrta,
real * dvda,
real * p_gbtabscale,
real * GBtab,
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
real * work)
{
int nri,ntype,nthreads;
real facel,krf,crf,tabscale,gbtabscale;
int n,ii,is3,ii3,k,nj0,nj1,jnr,j3,ggid;
int nn0,nn1,nouter,ninner;
real shX,shY,shZ;
real fscal,tx,ty,tz;
real rinvsq;
real jq;
real qq,vcoul,vctot;
int nti;
int tj;
real Vvdw6,Vvdwtot;
real Vvdw12;
real r,rt,eps,eps2;
int n0,nnn;
real Y,F,Geps,Heps2,Fp,VV;
real FF;
real fijD,fijR;
real krsq;
real ix1,iy1,iz1,fix1,fiy1,fiz1;
real ix2,iy2,iz2,fix2,fiy2,fiz2;
real ix3,iy3,iz3,fix3,fiy3,fiz3;
real jx1,jy1,jz1,fjx1,fjy1,fjz1;
real dx11,dy11,dz11,rsq11,rinv11;
real dx21,dy21,dz21,rsq21,rinv21;
real dx31,dy31,dz31,rsq31,rinv31;
real qO,qH;
real c6,c12;
nri = *p_nri;
ntype = *p_ntype;
nthreads = *p_nthreads;
facel = *p_facel;
krf = *p_krf;
crf = *p_crf;
tabscale = *p_tabscale;
/* Initialize water data */
ii = iinr[0];
qO = facel*charge[ii];
qH = facel*charge[ii+1];
nti = 2*ntype*type[ii];
/* Reset outer and inner iteration counters */
nouter = 0;
ninner = 0;
/* Loop over thread workunits */
do
{
#ifdef GMX_THREAD_SHM_FDECOMP
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
/* Take successively smaller chunks (at least 10 lists) */
nn1 = nn0+(nri-nn0)/(2*nthreads)+10;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
#else
nn0 = 0;
nn1 = nri;
#endif
/* Start outer loop over neighborlists */
for(n=nn0; (n<nn1); n++)
{
/* Load shift vector for this list */
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
/* Load limits for loop over neighbors */
nj0 = jindex[n];
nj1 = jindex[n+1];
/* Get outer coordinate index */
ii = iinr[n];
ii3 = 3*ii;
/* Load i atom data, add shift vector */
ix1 = shX + pos[ii3+0];
iy1 = shY + pos[ii3+1];
iz1 = shZ + pos[ii3+2];
ix2 = shX + pos[ii3+3];
iy2 = shY + pos[ii3+4];
iz2 = shZ + pos[ii3+5];
ix3 = shX + pos[ii3+6];
iy3 = shY + pos[ii3+7];
iz3 = shZ + pos[ii3+8];
/* Zero the potential energy for this list */
vctot = 0;
Vvdwtot = 0;
/* Clear i atom forces */
fix1 = 0;
fiy1 = 0;
fiz1 = 0;
fix2 = 0;
fiy2 = 0;
fiz2 = 0;
fix3 = 0;
fiy3 = 0;
fiz3 = 0;
for(k=nj0; (k<nj1); k++)
{
/* Get j neighbor index, and coordinate index */
jnr = jjnr[k];
j3 = 3*jnr;
/* load j atom coordinates */
jx1 = pos[j3+0];
jy1 = pos[j3+1];
jz1 = pos[j3+2];
/* Calculate distance */
dx11 = ix1 - jx1;
dy11 = iy1 - jy1;
dz11 = iz1 - jz1;
rsq11 = dx11*dx11+dy11*dy11+dz11*dz11;
dx21 = ix2 - jx1;
dy21 = iy2 - jy1;
dz21 = iz2 - jz1;
rsq21 = dx21*dx21+dy21*dy21+dz21*dz21;
dx31 = ix3 - jx1;
dy31 = iy3 - jy1;
dz31 = iz3 - jz1;
rsq31 = dx31*dx31+dy31*dy31+dz31*dz31;
/* Calculate 1/r and 1/r2 */
rinv11 = gmx_invsqrt(rsq11);
rinv21 = gmx_invsqrt(rsq21);
rinv31 = gmx_invsqrt(rsq31);
/* Load parameters for j atom */
jq = charge[jnr+0];
qq = qO*jq;
tj = nti+2*type[jnr];
c6 = vdwparam[tj];
c12 = vdwparam[tj+1];
rinvsq = rinv11*rinv11;
/* Coulomb reaction-field interaction */
krsq = krf*rsq11;
vcoul = qq*(rinv11+krsq-crf);
vctot = vctot+vcoul;
/* Calculate table index */
r = rsq11*rinv11;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 8*n0;
/* Tabulated VdW interaction - dispersion */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
Vvdw6 = c6*VV;
fijD = c6*FF;
/* Tabulated VdW interaction - repulsion */
nnn = nnn+4;
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
Vvdw12 = c12*VV;
fijR = c12*FF;
Vvdwtot = Vvdwtot+ Vvdw6 + Vvdw12;
fscal = (qq*(rinv11-2.0*krsq))*rinvsq-((fijD+fijR)*tabscale)*rinv11;
/* Calculate temporary vectorial force */
tx = fscal*dx11;
ty = fscal*dy11;
tz = fscal*dz11;
/* Increment i atom force */
fix1 = fix1 + tx;
fiy1 = fiy1 + ty;
fiz1 = fiz1 + tz;
/* Decrement j atom force */
fjx1 = faction[j3+0] - tx;
fjy1 = faction[j3+1] - ty;
fjz1 = faction[j3+2] - tz;
/* Load parameters for j atom */
qq = qH*jq;
rinvsq = rinv21*rinv21;
/* Coulomb reaction-field interaction */
krsq = krf*rsq21;
vcoul = qq*(rinv21+krsq-crf);
vctot = vctot+vcoul;
fscal = (qq*(rinv21-2.0*krsq))*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx21;
ty = fscal*dy21;
tz = fscal*dz21;
/* Increment i atom force */
fix2 = fix2 + tx;
fiy2 = fiy2 + ty;
fiz2 = fiz2 + tz;
/* Decrement j atom force */
fjx1 = fjx1 - tx;
fjy1 = fjy1 - ty;
fjz1 = fjz1 - tz;
/* Load parameters for j atom */
rinvsq = rinv31*rinv31;
/* Coulomb reaction-field interaction */
krsq = krf*rsq31;
vcoul = qq*(rinv31+krsq-crf);
vctot = vctot+vcoul;
fscal = (qq*(rinv31-2.0*krsq))*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx31;
ty = fscal*dy31;
tz = fscal*dz31;
/* Increment i atom force */
fix3 = fix3 + tx;
fiy3 = fiy3 + ty;
fiz3 = fiz3 + tz;
/* Decrement j atom force */
faction[j3+0] = fjx1 - tx;
faction[j3+1] = fjy1 - ty;
faction[j3+2] = fjz1 - tz;
/* Inner loop uses 130 flops/iteration */
}
/* Add i forces to mem and shifted force list */
faction[ii3+0] = faction[ii3+0] + fix1;
faction[ii3+1] = faction[ii3+1] + fiy1;
faction[ii3+2] = faction[ii3+2] + fiz1;
faction[ii3+3] = faction[ii3+3] + fix2;
faction[ii3+4] = faction[ii3+4] + fiy2;
faction[ii3+5] = faction[ii3+5] + fiz2;
faction[ii3+6] = faction[ii3+6] + fix3;
faction[ii3+7] = faction[ii3+7] + fiy3;
faction[ii3+8] = faction[ii3+8] + fiz3;
fshift[is3] = fshift[is3]+fix1+fix2+fix3;
fshift[is3+1] = fshift[is3+1]+fiy1+fiy2+fiy3;
fshift[is3+2] = fshift[is3+2]+fiz1+fiz2+fiz3;
/* Add potential energies to the group for this list */
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
Vvdw[ggid] = Vvdw[ggid] + Vvdwtot;
/* Increment number of inner iterations */
ninner = ninner + nj1 - nj0;
/* Outer loop uses 29 flops/iteration */
}
/* Increment number of outer iterations */
nouter = nouter + nn1 - nn0;
}
while (nn1<nri);
/* Write outer/inner iteration count to pointers */
*outeriter = nouter;
*inneriter = ninner;
}
static void gmx_print_version_info(FILE *fp)
{
fprintf(fp, "Gromacs version: %s\n", gmx_version());
const char *const git_hash = gmx_version_git_full_hash();
if (git_hash[0] != '\0')
{
fprintf(fp, "GIT SHA1 hash: %s\n", git_hash);
}
const char *const base_hash = gmx_version_git_central_base_hash();
if (base_hash[0] != '\0')
{
fprintf(fp, "Branched from: %s\n", base_hash);
}
#ifdef GMX_DOUBLE
fprintf(fp, "Precision: double\n");
#else
fprintf(fp, "Precision: single\n");
#endif
fprintf(fp, "Memory model: %u bit\n", (unsigned)(8*sizeof(void *)));
#ifdef GMX_THREAD_MPI
fprintf(fp, "MPI library: thread_mpi\n");
#elif defined(GMX_MPI)
fprintf(fp, "MPI library: MPI\n");
#else
fprintf(fp, "MPI library: none\n");
#endif
#ifdef GMX_OPENMP
fprintf(fp, "OpenMP support: enabled (GMX_OPENMP_MAX_THREADS = %d)\n", GMX_OPENMP_MAX_THREADS);
#else
fprintf(fp, "OpenMP support: disabled\n");
#endif
#ifdef GMX_GPU
fprintf(fp, "GPU support: enabled\n");
#else
fprintf(fp, "GPU support: disabled\n");
#endif
/* A preprocessor trick to avoid duplicating logic from vec.h */
#define gmx_stringify2(x) #x
#define gmx_stringify(x) gmx_stringify2(x)
fprintf(fp, "invsqrt routine: %s\n", gmx_stringify(gmx_invsqrt(x)));
fprintf(fp, "SIMD instructions: %s\n", GMX_SIMD_STRING);
fprintf(fp, "FFT library: %s\n", gmx_fft_get_version_info());
#ifdef HAVE_RDTSCP
fprintf(fp, "RDTSCP usage: enabled\n");
#else
fprintf(fp, "RDTSCP usage: disabled\n");
#endif
#ifdef GMX_CXX11
fprintf(fp, "C++11 compilation: enabled\n");
#else
fprintf(fp, "C++11 compilation: disabled\n");
#endif
#ifdef GMX_USE_TNG
fprintf(fp, "TNG support: enabled\n");
#else
fprintf(fp, "TNG support: disabled\n");
#endif
#ifdef HAVE_EXTRAE
unsigned major, minor, revision;
Extrae_get_version(&major, &minor, &revision);
fprintf(fp, "Tracing support: enabled. Using Extrae-%d.%d.%d\n", major, minor, revision);
#else
fprintf(fp, "Tracing support: disabled\n");
#endif
fprintf(fp, "Built on: %s\n", BUILD_TIME);
fprintf(fp, "Built by: %s\n", BUILD_USER);
fprintf(fp, "Build OS/arch: %s\n", BUILD_HOST);
fprintf(fp, "Build CPU vendor: %s\n", BUILD_CPU_VENDOR);
fprintf(fp, "Build CPU brand: %s\n", BUILD_CPU_BRAND);
fprintf(fp, "Build CPU family: %d Model: %d Stepping: %d\n",
BUILD_CPU_FAMILY, BUILD_CPU_MODEL, BUILD_CPU_STEPPING);
/* TODO: The below strings can be quite long, so it would be nice to wrap
* them. Can wait for later, as the master branch has ready code to do all
* that. */
fprintf(fp, "Build CPU features: %s\n", BUILD_CPU_FEATURES);
fprintf(fp, "C compiler: %s\n", BUILD_C_COMPILER);
fprintf(fp, "C compiler flags: %s\n", BUILD_CFLAGS);
fprintf(fp, "C++ compiler: %s\n", BUILD_CXX_COMPILER);
fprintf(fp, "C++ compiler flags: %s\n", BUILD_CXXFLAGS);
#ifdef HAVE_LIBMKL
/* MKL might be used for LAPACK/BLAS even if FFTs use FFTW, so keep it separate */
fprintf(fp, "Linked with Intel MKL version %d.%d.%d.\n",
__INTEL_MKL__, __INTEL_MKL_MINOR__, __INTEL_MKL_UPDATE__);
#endif
#ifdef GMX_EXTERNAL_BOOST
const bool bExternalBoost = true;
#else
const bool bExternalBoost = false;
#endif
fprintf(fp, "Boost version: %d.%d.%d%s\n", BOOST_VERSION / 100000,
BOOST_VERSION / 100 % 1000, BOOST_VERSION % 100,
bExternalBoost ? " (external)" : " (internal)");
#ifdef GMX_GPU
gmx_print_version_info_gpu(fp);
#endif
}
void
adress_thermo_force(int start,
int homenr,
t_block * cgs,
rvec x[],
rvec f[],
t_forcerec * fr,
t_mdatoms * mdatoms,
t_pbc * pbc)
{
int iatom, n0, nnn, nrcg, i;
int adresstype;
real adressw, adressr;
atom_id * cgindex;
unsigned short * ptype;
rvec * ref;
real * wf;
real tabscale;
real * ATFtab;
rvec dr;
real w, wsq, wmin1, wmin1sq, wp, wt, rinv, sqr_dl, dl;
real eps, eps2, F, Geps, Heps2, Fp, dmu_dwp, dwp_dr, fscal;
adresstype = fr->adress_type;
adressw = fr->adress_hy_width;
adressr = fr->adress_ex_width;
cgindex = cgs->index;
ptype = mdatoms->ptype;
ref = &(fr->adress_refs);
wf = mdatoms->wf;
for (iatom = start; (iatom < start+homenr); iatom++)
{
if (egp_coarsegrained(fr, mdatoms->cENER[iatom]))
{
if (ptype[iatom] == eptVSite)
{
w = wf[iatom];
/* is it hybrid or apply the thermodynamics force everywhere?*/
if (mdatoms->tf_table_index[iatom] != NO_TF_TABLE)
{
if (fr->n_adress_tf_grps > 0)
{
/* multi component tf is on, select the right table */
ATFtab = fr->atf_tabs[mdatoms->tf_table_index[iatom]].data;
tabscale = fr->atf_tabs[mdatoms->tf_table_index[iatom]].scale;
}
else
{
/* just on component*/
ATFtab = fr->atf_tabs[DEFAULT_TF_TABLE].data;
tabscale = fr->atf_tabs[DEFAULT_TF_TABLE].scale;
}
fscal = 0;
if (pbc)
{
pbc_dx(pbc, (*ref), x[iatom], dr);
}
else
{
rvec_sub((*ref), x[iatom], dr);
}
/* calculate distace to adress center again */
sqr_dl = 0.0;
switch (adresstype)
{
case eAdressXSplit:
/* plane through center of ref, varies in x direction */
sqr_dl = dr[0]*dr[0];
rinv = gmx_invsqrt(dr[0]*dr[0]);
break;
case eAdressSphere:
/* point at center of ref, assuming cubic geometry */
for (i = 0; i < 3; i++)
{
sqr_dl += dr[i]*dr[i];
}
rinv = gmx_invsqrt(sqr_dl);
break;
default:
/* This case should not happen */
rinv = 0.0;
}
dl = sqrt(sqr_dl);
/* table origin is adress center */
wt = dl*tabscale;
n0 = wt;
eps = wt-n0;
eps2 = eps*eps;
nnn = 4*n0;
F = ATFtab[nnn+1];
Geps = eps*ATFtab[nnn+2];
Heps2 = eps2*ATFtab[nnn+3];
Fp = F+Geps+Heps2;
F = (Fp+Geps+2.0*Heps2)*tabscale;
fscal = F*rinv;
f[iatom][0] += fscal*dr[0];
if (adresstype != eAdressXSplit)
{
f[iatom][1] += fscal*dr[1];
f[iatom][2] += fscal*dr[2];
}
}
}
}
}
}
/*
* Gromacs nonbonded kernel nb_kernel103
* Coulomb interaction: Normal Coulomb
* VdW interaction: Not calculated
* water optimization: TIP4P - other atoms
* Calculate forces: yes
*/
void nb_kernel103(
int * p_nri,
int * iinr,
int * jindex,
int * jjnr,
int * shift,
real * shiftvec,
real * fshift,
int * gid,
real * pos,
real * faction,
real * charge,
real * p_facel,
real * p_krf,
real * p_crf,
real * Vc,
int * type,
int * p_ntype,
real * vdwparam,
real * Vvdw,
real * p_tabscale,
real * VFtab,
real * invsqrta,
real * dvda,
real * p_gbtabscale,
real * GBtab,
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
real * work)
{
int nri,ntype,nthreads;
real facel,krf,crf,tabscale,gbtabscale;
int n,ii,is3,ii3,k,nj0,nj1,jnr,j3,ggid;
int nn0,nn1,nouter,ninner;
real shX,shY,shZ;
real fscal,tx,ty,tz;
real rinvsq;
real jq;
real qq,vcoul,vctot;
real ix2,iy2,iz2,fix2,fiy2,fiz2;
real ix3,iy3,iz3,fix3,fiy3,fiz3;
real ix4,iy4,iz4,fix4,fiy4,fiz4;
real jx1,jy1,jz1,fjx1,fjy1,fjz1;
real dx21,dy21,dz21,rsq21,rinv21;
real dx31,dy31,dz31,rsq31,rinv31;
real dx41,dy41,dz41,rsq41,rinv41;
real qH,qM;
nri = *p_nri;
ntype = *p_ntype;
nthreads = *p_nthreads;
facel = *p_facel;
krf = *p_krf;
crf = *p_crf;
tabscale = *p_tabscale;
/* Initialize water data */
ii = iinr[0];
qH = facel*charge[ii+1];
qM = facel*charge[ii+3];
/* Reset outer and inner iteration counters */
nouter = 0;
ninner = 0;
/* Loop over thread workunits */
do
{
#ifdef GMX_THREAD_SHM_FDECOMP
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
/* Take successively smaller chunks (at least 10 lists) */
nn1 = nn0+(nri-nn0)/(2*nthreads)+10;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
#else
nn0 = 0;
nn1 = nri;
#endif
/* Start outer loop over neighborlists */
for(n=nn0; (n<nn1); n++)
{
/* Load shift vector for this list */
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
/* Load limits for loop over neighbors */
nj0 = jindex[n];
nj1 = jindex[n+1];
/* Get outer coordinate index */
ii = iinr[n];
ii3 = 3*ii;
/* Load i atom data, add shift vector */
ix2 = shX + pos[ii3+3];
iy2 = shY + pos[ii3+4];
iz2 = shZ + pos[ii3+5];
ix3 = shX + pos[ii3+6];
iy3 = shY + pos[ii3+7];
iz3 = shZ + pos[ii3+8];
ix4 = shX + pos[ii3+9];
iy4 = shY + pos[ii3+10];
iz4 = shZ + pos[ii3+11];
/* Zero the potential energy for this list */
vctot = 0;
/* Clear i atom forces */
fix2 = 0;
fiy2 = 0;
fiz2 = 0;
fix3 = 0;
fiy3 = 0;
fiz3 = 0;
fix4 = 0;
fiy4 = 0;
fiz4 = 0;
for(k=nj0; (k<nj1); k++)
{
/* Get j neighbor index, and coordinate index */
jnr = jjnr[k];
j3 = 3*jnr;
/* load j atom coordinates */
jx1 = pos[j3+0];
jy1 = pos[j3+1];
jz1 = pos[j3+2];
/* Calculate distance */
dx21 = ix2 - jx1;
dy21 = iy2 - jy1;
dz21 = iz2 - jz1;
rsq21 = dx21*dx21+dy21*dy21+dz21*dz21;
dx31 = ix3 - jx1;
dy31 = iy3 - jy1;
dz31 = iz3 - jz1;
rsq31 = dx31*dx31+dy31*dy31+dz31*dz31;
dx41 = ix4 - jx1;
dy41 = iy4 - jy1;
dz41 = iz4 - jz1;
rsq41 = dx41*dx41+dy41*dy41+dz41*dz41;
/* Calculate 1/r and 1/r2 */
rinv21 = gmx_invsqrt(rsq21);
rinv31 = gmx_invsqrt(rsq31);
rinv41 = gmx_invsqrt(rsq41);
/* Load parameters for j atom */
jq = charge[jnr+0];
qq = qH*jq;
rinvsq = rinv21*rinv21;
/* Coulomb interaction */
vcoul = qq*rinv21;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx21;
ty = fscal*dy21;
tz = fscal*dz21;
/* Increment i atom force */
fix2 = fix2 + tx;
fiy2 = fiy2 + ty;
fiz2 = fiz2 + tz;
/* Decrement j atom force */
fjx1 = faction[j3+0] - tx;
fjy1 = faction[j3+1] - ty;
fjz1 = faction[j3+2] - tz;
/* Load parameters for j atom */
rinvsq = rinv31*rinv31;
/* Coulomb interaction */
vcoul = qq*rinv31;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx31;
ty = fscal*dy31;
tz = fscal*dz31;
/* Increment i atom force */
fix3 = fix3 + tx;
fiy3 = fiy3 + ty;
fiz3 = fiz3 + tz;
/* Decrement j atom force */
fjx1 = fjx1 - tx;
fjy1 = fjy1 - ty;
fjz1 = fjz1 - tz;
/* Load parameters for j atom */
qq = qM*jq;
rinvsq = rinv41*rinv41;
/* Coulomb interaction */
vcoul = qq*rinv41;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx41;
ty = fscal*dy41;
tz = fscal*dz41;
/* Increment i atom force */
fix4 = fix4 + tx;
fiy4 = fiy4 + ty;
fiz4 = fiz4 + tz;
/* Decrement j atom force */
faction[j3+0] = fjx1 - tx;
faction[j3+1] = fjy1 - ty;
faction[j3+2] = fjz1 - tz;
/* Inner loop uses 80 flops/iteration */
}
/* Add i forces to mem and shifted force list */
faction[ii3+3] = faction[ii3+3] + fix2;
faction[ii3+4] = faction[ii3+4] + fiy2;
faction[ii3+5] = faction[ii3+5] + fiz2;
faction[ii3+6] = faction[ii3+6] + fix3;
faction[ii3+7] = faction[ii3+7] + fiy3;
faction[ii3+8] = faction[ii3+8] + fiz3;
faction[ii3+9] = faction[ii3+9] + fix4;
faction[ii3+10] = faction[ii3+10] + fiy4;
faction[ii3+11] = faction[ii3+11] + fiz4;
fshift[is3] = fshift[is3]+fix2+fix3+fix4;
fshift[is3+1] = fshift[is3+1]+fiy2+fiy3+fiy4;
fshift[is3+2] = fshift[is3+2]+fiz2+fiz3+fiz4;
/* Add potential energies to the group for this list */
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
/* Increment number of inner iterations */
ninner = ninner + nj1 - nj0;
/* Outer loop uses 28 flops/iteration */
}
/* Increment number of outer iterations */
nouter = nouter + nn1 - nn0;
}
while (nn1<nri);
/* Write outer/inner iteration count to pointers */
*outeriter = nouter;
*inneriter = ninner;
}
void
gmx_nb_free_energy_kernel(t_nblist * nlist,
rvec * xx,
rvec * ff,
t_forcerec * fr,
t_mdatoms * mdatoms,
nb_kernel_data_t * kernel_data,
t_nrnb * nrnb)
{
#define STATE_A 0
#define STATE_B 1
#define NSTATES 2
int i, j, n, ii, is3, ii3, k, nj0, nj1, jnr, j3, ggid;
real shX, shY, shZ;
real Fscal, FscalC[NSTATES], FscalV[NSTATES], tx, ty, tz;
real Vcoul[NSTATES], Vvdw[NSTATES];
real rinv6, r, rt, rtC, rtV;
real iqA, iqB;
real qq[NSTATES], vctot, krsq;
int ntiA, ntiB, tj[NSTATES];
real Vvdw6, Vvdw12, vvtot;
real ix, iy, iz, fix, fiy, fiz;
real dx, dy, dz, rsq, rinv;
real c6[NSTATES], c12[NSTATES];
real LFC[NSTATES], LFV[NSTATES], DLF[NSTATES];
double dvdl_coul, dvdl_vdw;
real lfac_coul[NSTATES], dlfac_coul[NSTATES], lfac_vdw[NSTATES], dlfac_vdw[NSTATES];
real sigma6[NSTATES], alpha_vdw_eff, alpha_coul_eff, sigma2_def, sigma2_min;
real rp, rpm2, rC, rV, rinvC, rpinvC, rinvV, rpinvV;
real sigma2[NSTATES], sigma_pow[NSTATES], sigma_powm2[NSTATES], rs, rs2;
int do_coultab, do_vdwtab, do_tab, tab_elemsize;
int n0, n1C, n1V, nnn;
real Y, F, G, H, Fp, Geps, Heps2, epsC, eps2C, epsV, eps2V, VV, FF;
int icoul, ivdw;
int nri;
int * iinr;
int * jindex;
int * jjnr;
int * shift;
int * gid;
int * typeA;
int * typeB;
int ntype;
real * shiftvec;
real dvdl_part;
real * fshift;
real tabscale;
real * VFtab;
real * x;
real * f;
real facel, krf, crf;
real * chargeA;
real * chargeB;
real sigma6_min, sigma6_def, lam_power, sc_power, sc_r_power;
real alpha_coul, alpha_vdw, lambda_coul, lambda_vdw, ewc;
real * nbfp;
real * dvdl;
real * Vv;
real * Vc;
gmx_bool bDoForces;
real rcoulomb, rvdw, sh_invrc6;
gmx_bool bExactElecCutoff, bExactVdwCutoff;
real rcutoff, rcutoff2, rswitch, d, d2, swV3, swV4, swV5, swF2, swF3, swF4, sw, dsw, rinvcorr;
x = xx[0];
f = ff[0];
fshift = fr->fshift[0];
Vc = kernel_data->energygrp_elec;
Vv = kernel_data->energygrp_vdw;
tabscale = kernel_data->table_elec_vdw->scale;
VFtab = kernel_data->table_elec_vdw->data;
nri = nlist->nri;
iinr = nlist->iinr;
jindex = nlist->jindex;
jjnr = nlist->jjnr;
icoul = nlist->ielec;
ivdw = nlist->ivdw;
shift = nlist->shift;
gid = nlist->gid;
shiftvec = fr->shift_vec[0];
chargeA = mdatoms->chargeA;
chargeB = mdatoms->chargeB;
facel = fr->epsfac;
krf = fr->k_rf;
crf = fr->c_rf;
ewc = fr->ewaldcoeff;
Vc = kernel_data->energygrp_elec;
typeA = mdatoms->typeA;
typeB = mdatoms->typeB;
ntype = fr->ntype;
nbfp = fr->nbfp;
Vv = kernel_data->energygrp_vdw;
tabscale = kernel_data->table_elec_vdw->scale;
VFtab = kernel_data->table_elec_vdw->data;
lambda_coul = kernel_data->lambda[efptCOUL];
lambda_vdw = kernel_data->lambda[efptVDW];
dvdl = kernel_data->dvdl;
alpha_coul = fr->sc_alphacoul;
alpha_vdw = fr->sc_alphavdw;
lam_power = fr->sc_power;
sc_r_power = fr->sc_r_power;
sigma6_def = fr->sc_sigma6_def;
sigma6_min = fr->sc_sigma6_min;
bDoForces = kernel_data->flags & GMX_NONBONDED_DO_FORCE;
rcoulomb = fr->rcoulomb;
rvdw = fr->rvdw;
sh_invrc6 = fr->ic->sh_invrc6;
if (fr->coulomb_modifier == eintmodPOTSWITCH || fr->vdw_modifier == eintmodPOTSWITCH)
{
rcutoff = (fr->coulomb_modifier == eintmodPOTSWITCH) ? fr->rcoulomb : fr->rvdw;
rcutoff2 = rcutoff*rcutoff;
rswitch = (fr->coulomb_modifier == eintmodPOTSWITCH) ? fr->rcoulomb_switch : fr->rvdw_switch;
d = rcutoff-rswitch;
swV3 = -10.0/(d*d*d);
swV4 = 15.0/(d*d*d*d);
swV5 = -6.0/(d*d*d*d*d);
swF2 = -30.0/(d*d*d);
swF3 = 60.0/(d*d*d*d);
swF4 = -30.0/(d*d*d*d*d);
}
else
{
/* Stupid compilers dont realize these variables will not be used */
rswitch = 0.0;
swV3 = 0.0;
swV4 = 0.0;
swV5 = 0.0;
swF2 = 0.0;
swF3 = 0.0;
swF4 = 0.0;
}
bExactElecCutoff = (fr->coulomb_modifier != eintmodNONE) || fr->eeltype == eelRF_ZERO;
bExactVdwCutoff = (fr->vdw_modifier != eintmodNONE);
/* fix compiler warnings */
nj1 = 0;
n1C = n1V = 0;
epsC = epsV = 0;
eps2C = eps2V = 0;
dvdl_coul = 0;
dvdl_vdw = 0;
/* Lambda factor for state A, 1-lambda*/
LFC[STATE_A] = 1.0 - lambda_coul;
LFV[STATE_A] = 1.0 - lambda_vdw;
/* Lambda factor for state B, lambda*/
LFC[STATE_B] = lambda_coul;
LFV[STATE_B] = lambda_vdw;
/*derivative of the lambda factor for state A and B */
DLF[STATE_A] = -1;
DLF[STATE_B] = 1;
for (i = 0; i < NSTATES; i++)
{
lfac_coul[i] = (lam_power == 2 ? (1-LFC[i])*(1-LFC[i]) : (1-LFC[i]));
dlfac_coul[i] = DLF[i]*lam_power/sc_r_power*(lam_power == 2 ? (1-LFC[i]) : 1);
lfac_vdw[i] = (lam_power == 2 ? (1-LFV[i])*(1-LFV[i]) : (1-LFV[i]));
dlfac_vdw[i] = DLF[i]*lam_power/sc_r_power*(lam_power == 2 ? (1-LFV[i]) : 1);
}
/* precalculate */
sigma2_def = pow(sigma6_def, 1.0/3.0);
sigma2_min = pow(sigma6_min, 1.0/3.0);
/* Ewald (not PME) table is special (icoul==enbcoulFEWALD) */
do_coultab = (icoul == GMX_NBKERNEL_ELEC_CUBICSPLINETABLE);
do_vdwtab = (ivdw == GMX_NBKERNEL_VDW_CUBICSPLINETABLE);
do_tab = do_coultab || do_vdwtab;
/* we always use the combined table here */
tab_elemsize = 12;
for (n = 0; (n < nri); n++)
{
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
nj0 = jindex[n];
nj1 = jindex[n+1];
ii = iinr[n];
ii3 = 3*ii;
ix = shX + x[ii3+0];
iy = shY + x[ii3+1];
iz = shZ + x[ii3+2];
iqA = facel*chargeA[ii];
iqB = facel*chargeB[ii];
ntiA = 2*ntype*typeA[ii];
ntiB = 2*ntype*typeB[ii];
vctot = 0;
vvtot = 0;
fix = 0;
fiy = 0;
fiz = 0;
for (k = nj0; (k < nj1); k++)
{
jnr = jjnr[k];
j3 = 3*jnr;
dx = ix - x[j3];
dy = iy - x[j3+1];
dz = iz - x[j3+2];
rsq = dx*dx+dy*dy+dz*dz;
rinv = gmx_invsqrt(rsq);
r = rsq*rinv;
if (sc_r_power == 6.0)
{
rpm2 = rsq*rsq; /* r4 */
rp = rpm2*rsq; /* r6 */
}
else if (sc_r_power == 48.0)
{
rp = rsq*rsq*rsq; /* r6 */
rp = rp*rp; /* r12 */
rp = rp*rp; /* r24 */
rp = rp*rp; /* r48 */
rpm2 = rp/rsq; /* r46 */
}
else
{
rp = pow(r, sc_r_power); /* not currently supported as input, but can handle it */
rpm2 = rp/rsq;
}
tj[STATE_A] = ntiA+2*typeA[jnr];
tj[STATE_B] = ntiB+2*typeB[jnr];
qq[STATE_A] = iqA*chargeA[jnr];
qq[STATE_B] = iqB*chargeB[jnr];
for (i = 0; i < NSTATES; i++)
{
c6[i] = nbfp[tj[i]];
c12[i] = nbfp[tj[i]+1];
if ((c6[i] > 0) && (c12[i] > 0))
{
/* c12 is stored scaled with 12.0 and c6 is scaled with 6.0 - correct for this */
sigma6[i] = 0.5*c12[i]/c6[i];
sigma2[i] = pow(sigma6[i], 1.0/3.0);
/* should be able to get rid of this ^^^ internal pow call eventually. Will require agreement on
what data to store externally. Can't be fixed without larger scale changes, so not 4.6 */
if (sigma6[i] < sigma6_min) /* for disappearing coul and vdw with soft core at the same time */
{
sigma6[i] = sigma6_min;
sigma2[i] = sigma2_min;
}
}
else
{
sigma6[i] = sigma6_def;
sigma2[i] = sigma2_def;
}
if (sc_r_power == 6.0)
{
sigma_pow[i] = sigma6[i];
sigma_powm2[i] = sigma6[i]/sigma2[i];
}
else if (sc_r_power == 48.0)
{
sigma_pow[i] = sigma6[i]*sigma6[i]; /* sigma^12 */
sigma_pow[i] = sigma_pow[i]*sigma_pow[i]; /* sigma^24 */
sigma_pow[i] = sigma_pow[i]*sigma_pow[i]; /* sigma^48 */
sigma_powm2[i] = sigma_pow[i]/sigma2[i];
}
else
{ /* not really supported as input, but in here for testing the general case*/
sigma_pow[i] = pow(sigma2[i], sc_r_power/2);
sigma_powm2[i] = sigma_pow[i]/(sigma2[i]);
}
}
/* only use softcore if one of the states has a zero endstate - softcore is for avoiding infinities!*/
if ((c12[STATE_A] > 0) && (c12[STATE_B] > 0))
{
alpha_vdw_eff = 0;
alpha_coul_eff = 0;
}
else
{
alpha_vdw_eff = alpha_vdw;
alpha_coul_eff = alpha_coul;
}
for (i = 0; i < NSTATES; i++)
{
FscalC[i] = 0;
FscalV[i] = 0;
Vcoul[i] = 0;
Vvdw[i] = 0;
/* Only spend time on A or B state if it is non-zero */
if ( (qq[i] != 0) || (c6[i] != 0) || (c12[i] != 0) )
{
/* this section has to be inside the loop becaue of the dependence on sigma_pow */
rpinvC = 1.0/(alpha_coul_eff*lfac_coul[i]*sigma_pow[i]+rp);
rinvC = pow(rpinvC, 1.0/sc_r_power);
rC = 1.0/rinvC;
rpinvV = 1.0/(alpha_vdw_eff*lfac_vdw[i]*sigma_pow[i]+rp);
rinvV = pow(rpinvV, 1.0/sc_r_power);
rV = 1.0/rinvV;
if (do_tab)
{
rtC = rC*tabscale;
n0 = rtC;
epsC = rtC-n0;
eps2C = epsC*epsC;
n1C = tab_elemsize*n0;
rtV = rV*tabscale;
n0 = rtV;
epsV = rtV-n0;
eps2V = epsV*epsV;
n1V = tab_elemsize*n0;
}
/* With Ewald and soft-core we should put the cut-off on r,
* not on the soft-cored rC, as the real-space and
* reciprocal space contributions should (almost) cancel.
*/
if (qq[i] != 0 &&
!(bExactElecCutoff &&
((icoul != GMX_NBKERNEL_ELEC_EWALD && rC >= rcoulomb) ||
(icoul == GMX_NBKERNEL_ELEC_EWALD && r >= rcoulomb))))
{
switch (icoul)
{
case GMX_NBKERNEL_ELEC_COULOMB:
case GMX_NBKERNEL_ELEC_EWALD:
/* simple cutoff (yes, ewald is done all on direct space for free energy) */
Vcoul[i] = qq[i]*rinvC;
FscalC[i] = Vcoul[i]*rpinvC;
break;
case GMX_NBKERNEL_ELEC_REACTIONFIELD:
/* reaction-field */
Vcoul[i] = qq[i]*(rinvC+krf*rC*rC-crf);
FscalC[i] = qq[i]*(rinvC*rpinvC-2.0*krf);
break;
case GMX_NBKERNEL_ELEC_CUBICSPLINETABLE:
/* non-Ewald tabulated coulomb */
nnn = n1C;
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = epsC*VFtab[nnn+2];
Heps2 = eps2C*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+epsC*Fp;
FF = Fp+Geps+2.0*Heps2;
Vcoul[i] = qq[i]*VV;
FscalC[i] = -qq[i]*tabscale*FF*rC*rpinvC;
break;
default:
FscalC[i] = 0.0;
Vcoul[i] = 0.0;
break;
}
if (fr->coulomb_modifier == eintmodPOTSWITCH)
{
d = rC-rswitch;
d = (d > 0.0) ? d : 0.0;
d2 = d*d;
sw = 1.0+d2*d*(swV3+d*(swV4+d*swV5));
dsw = d2*(swF2+d*(swF3+d*swF4));
Vcoul[i] *= sw;
FscalC[i] = FscalC[i]*sw + Vcoul[i]*dsw;
}
}
if ((c6[i] != 0 || c12[i] != 0) &&
!(bExactVdwCutoff && rV >= rvdw))
{
switch (ivdw)
{
case GMX_NBKERNEL_VDW_LENNARDJONES:
/* cutoff LJ */
if (sc_r_power == 6.0)
{
rinv6 = rpinvV;
}
else
{
rinv6 = pow(rinvV, 6.0);
}
Vvdw6 = c6[i]*rinv6;
Vvdw12 = c12[i]*rinv6*rinv6;
if (fr->vdw_modifier == eintmodPOTSHIFT)
{
Vvdw[i] = ( (Vvdw12-c12[i]*sh_invrc6*sh_invrc6)*(1.0/12.0)
-(Vvdw6-c6[i]*sh_invrc6)*(1.0/6.0));
}
else
{
Vvdw[i] = Vvdw12*(1.0/12.0)-Vvdw6*(1.0/6.0);
}
FscalV[i] = (Vvdw12-Vvdw6)*rpinvV;
break;
case GMX_NBKERNEL_VDW_BUCKINGHAM:
gmx_fatal(FARGS, "Buckingham free energy not supported.");
break;
case GMX_NBKERNEL_VDW_CUBICSPLINETABLE:
/* Table LJ */
nnn = n1V+4;
/* dispersion */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = epsV*VFtab[nnn+2];
Heps2 = eps2V*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+epsV*Fp;
FF = Fp+Geps+2.0*Heps2;
Vvdw[i] += c6[i]*VV;
FscalV[i] -= c6[i]*tabscale*FF*rV*rpinvV;
/* repulsion */
Y = VFtab[nnn+4];
F = VFtab[nnn+5];
Geps = epsV*VFtab[nnn+6];
Heps2 = eps2V*VFtab[nnn+7];
Fp = F+Geps+Heps2;
VV = Y+epsV*Fp;
FF = Fp+Geps+2.0*Heps2;
Vvdw[i] += c12[i]*VV;
FscalV[i] -= c12[i]*tabscale*FF*rV*rpinvV;
break;
default:
Vvdw[i] = 0.0;
FscalV[i] = 0.0;
break;
}
if (fr->vdw_modifier == eintmodPOTSWITCH)
{
d = rV-rswitch;
d = (d > 0.0) ? d : 0.0;
d2 = d*d;
sw = 1.0+d2*d*(swV3+d*(swV4+d*swV5));
dsw = d2*(swF2+d*(swF3+d*swF4));
Vvdw[i] *= sw;
FscalV[i] = FscalV[i]*sw + Vvdw[i]*dsw;
FscalV[i] = (rV < rvdw) ? FscalV[i] : 0.0;
Vvdw[i] = (rV < rvdw) ? Vvdw[i] : 0.0;
}
}
}
}
Fscal = 0;
if (icoul == GMX_NBKERNEL_ELEC_EWALD &&
!(bExactElecCutoff && r >= rcoulomb))
{
/* because we compute the softcore normally,
we have to remove the ewald short range portion. Done outside of
the states loop because this part doesn't depend on the scaled R */
#ifdef GMX_DOUBLE
/* Relative accuracy at R_ERF_R_INACC of 3e-10 */
#define R_ERF_R_INACC 0.006
#else
/* Relative accuracy at R_ERF_R_INACC of 2e-5 */
#define R_ERF_R_INACC 0.1
#endif
if (ewc*r > R_ERF_R_INACC)
{
VV = gmx_erf(ewc*r)*rinv;
FF = rinv*rinv*(VV - ewc*M_2_SQRTPI*exp(-ewc*ewc*rsq));
}
else
{
VV = ewc*M_2_SQRTPI;
FF = ewc*ewc*ewc*M_2_SQRTPI*(2.0/3.0 - 0.4*ewc*ewc*rsq);
}
for (i = 0; i < NSTATES; i++)
{
vctot -= LFC[i]*qq[i]*VV;
Fscal -= LFC[i]*qq[i]*FF;
dvdl_coul -= (DLF[i]*qq[i])*VV;
}
}
/* Assemble A and B states */
for (i = 0; i < NSTATES; i++)
{
vctot += LFC[i]*Vcoul[i];
vvtot += LFV[i]*Vvdw[i];
Fscal += LFC[i]*FscalC[i]*rpm2;
Fscal += LFV[i]*FscalV[i]*rpm2;
dvdl_coul += Vcoul[i]*DLF[i] + LFC[i]*alpha_coul_eff*dlfac_coul[i]*FscalC[i]*sigma_pow[i];
dvdl_vdw += Vvdw[i]*DLF[i] + LFV[i]*alpha_vdw_eff*dlfac_vdw[i]*FscalV[i]*sigma_pow[i];
}
if (bDoForces)
{
tx = Fscal*dx;
ty = Fscal*dy;
tz = Fscal*dz;
fix = fix + tx;
fiy = fiy + ty;
fiz = fiz + tz;
f[j3] = f[j3] - tx;
f[j3+1] = f[j3+1] - ty;
f[j3+2] = f[j3+2] - tz;
}
}
if (bDoForces)
{
f[ii3] = f[ii3] + fix;
f[ii3+1] = f[ii3+1] + fiy;
f[ii3+2] = f[ii3+2] + fiz;
fshift[is3] = fshift[is3] + fix;
fshift[is3+1] = fshift[is3+1] + fiy;
fshift[is3+2] = fshift[is3+2] + fiz;
}
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
Vv[ggid] = Vv[ggid] + vvtot;
}
dvdl[efptCOUL] += dvdl_coul;
dvdl[efptVDW] += dvdl_vdw;
/* Estimate flops, average for free energy stuff:
* 12 flops per outer iteration
* 150 flops per inner iteration
*/
inc_nrnb(nrnb, eNR_NBKERNEL_FREE_ENERGY, nlist->nri*12 + nlist->jindex[n]*150);
}
real ewald_LRcorrection(FILE *fplog,
int start, int end,
t_commrec *cr, int thread, t_forcerec *fr,
real *chargeA, real *chargeB,
gmx_bool calc_excl_corr,
t_blocka *excl, rvec x[],
matrix box, rvec mu_tot[],
int ewald_geometry, real epsilon_surface,
rvec *f, tensor vir,
real lambda, real *dvdlambda)
{
int i, i1, i2, j, k, m, iv, jv, q;
atom_id *AA;
double q2sumA, q2sumB, Vexcl, dvdl_excl; /* Necessary for precision */
real one_4pi_eps;
real v, vc, qiA, qiB, dr, dr2, rinv, fscal, enercorr;
real Vself[2], Vdipole[2], rinv2, ewc = fr->ewaldcoeff, ewcdr;
rvec df, dx, mutot[2], dipcorrA, dipcorrB;
tensor dxdf;
real vol = box[XX][XX]*box[YY][YY]*box[ZZ][ZZ];
real L1, dipole_coeff, qqA, qqB, qqL, vr0;
/*#define TABLES*/
#ifdef TABLES
real tabscale = fr->tabscale;
real eps, eps2, VV, FF, F, Y, Geps, Heps2, Fp, fijC, r1t;
real *VFtab = fr->coulvdwtab;
int n0, n1, nnn;
#endif
gmx_bool bFreeEnergy = (chargeB != NULL);
gmx_bool bMolPBC = fr->bMolPBC;
one_4pi_eps = ONE_4PI_EPS0/fr->epsilon_r;
vr0 = ewc*M_2_SQRTPI;
AA = excl->a;
Vexcl = 0;
dvdl_excl = 0;
q2sumA = 0;
q2sumB = 0;
Vdipole[0] = 0;
Vdipole[1] = 0;
L1 = 1.0-lambda;
/* Note that we have to transform back to gromacs units, since
* mu_tot contains the dipole in debye units (for output).
*/
for (i = 0; (i < DIM); i++)
{
mutot[0][i] = mu_tot[0][i]*DEBYE2ENM;
mutot[1][i] = mu_tot[1][i]*DEBYE2ENM;
dipcorrA[i] = 0;
dipcorrB[i] = 0;
}
dipole_coeff = 0;
switch (ewald_geometry)
{
case eewg3D:
if (epsilon_surface != 0)
{
dipole_coeff =
2*M_PI*ONE_4PI_EPS0/((2*epsilon_surface + fr->epsilon_r)*vol);
for (i = 0; (i < DIM); i++)
{
dipcorrA[i] = 2*dipole_coeff*mutot[0][i];
dipcorrB[i] = 2*dipole_coeff*mutot[1][i];
}
}
break;
case eewg3DC:
dipole_coeff = 2*M_PI*one_4pi_eps/vol;
dipcorrA[ZZ] = 2*dipole_coeff*mutot[0][ZZ];
dipcorrB[ZZ] = 2*dipole_coeff*mutot[1][ZZ];
break;
default:
gmx_incons("Unsupported Ewald geometry");
break;
}
if (debug)
{
fprintf(debug, "dipcorr = %8.3f %8.3f %8.3f\n",
dipcorrA[XX], dipcorrA[YY], dipcorrA[ZZ]);
fprintf(debug, "mutot = %8.3f %8.3f %8.3f\n",
mutot[0][XX], mutot[0][YY], mutot[0][ZZ]);
}
clear_mat(dxdf);
if ((calc_excl_corr || dipole_coeff != 0) && !bFreeEnergy)
{
for (i = start; (i < end); i++)
{
/* Initiate local variables (for this i-particle) to 0 */
qiA = chargeA[i]*one_4pi_eps;
if (calc_excl_corr)
{
i1 = excl->index[i];
i2 = excl->index[i+1];
/* Loop over excluded neighbours */
for (j = i1; (j < i2); j++)
{
k = AA[j];
/*
* First we must test whether k <> i, and then, because the
* exclusions are all listed twice i->k and k->i we must select
* just one of the two.
* As a minor optimization we only compute forces when the charges
* are non-zero.
*/
if (k > i)
{
qqA = qiA*chargeA[k];
if (qqA != 0.0)
{
rvec_sub(x[i], x[k], dx);
if (bMolPBC)
{
/* Cheap pbc_dx, assume excluded pairs are at short distance. */
for (m = DIM-1; (m >= 0); m--)
{
if (dx[m] > 0.5*box[m][m])
{
rvec_dec(dx, box[m]);
}
else if (dx[m] < -0.5*box[m][m])
{
rvec_inc(dx, box[m]);
}
}
}
dr2 = norm2(dx);
/* Distance between two excluded particles may be zero in the
* case of shells
*/
if (dr2 != 0)
{
rinv = gmx_invsqrt(dr2);
rinv2 = rinv*rinv;
dr = 1.0/rinv;
#ifdef TABLES
r1t = tabscale*dr;
n0 = r1t;
assert(n0 >= 3);
n1 = 12*n0;
eps = r1t-n0;
eps2 = eps*eps;
nnn = n1;
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vc = qqA*(rinv-VV);
fijC = qqA*FF;
Vexcl += vc;
fscal = vc*rinv2+fijC*tabscale*rinv;
/* End of tabulated interaction part */
#else
/* This is the code you would want instead if not using
* tables:
*/
ewcdr = ewc*dr;
vc = qqA*gmx_erf(ewcdr)*rinv;
Vexcl += vc;
#ifdef GMX_DOUBLE
/* Relative accuracy at R_ERF_R_INACC of 3e-10 */
#define R_ERF_R_INACC 0.006
#else
/* Relative accuracy at R_ERF_R_INACC of 2e-5 */
#define R_ERF_R_INACC 0.1
#endif
if (ewcdr > R_ERF_R_INACC)
{
fscal = rinv2*(vc - qqA*ewc*M_2_SQRTPI*exp(-ewcdr*ewcdr));
}
else
{
/* Use a fourth order series expansion for small ewcdr */
fscal = ewc*ewc*qqA*vr0*(2.0/3.0 - 0.4*ewcdr*ewcdr);
}
#endif
/* The force vector is obtained by multiplication with the
* distance vector
*/
svmul(fscal, dx, df);
rvec_inc(f[k], df);
rvec_dec(f[i], df);
for (iv = 0; (iv < DIM); iv++)
{
for (jv = 0; (jv < DIM); jv++)
{
dxdf[iv][jv] += dx[iv]*df[jv];
}
}
}
else
{
Vexcl += qqA*vr0;
}
}
}
}
}
/* Dipole correction on force */
if (dipole_coeff != 0)
{
for (j = 0; (j < DIM); j++)
{
f[i][j] -= dipcorrA[j]*chargeA[i];
}
}
}
}
else if (calc_excl_corr || dipole_coeff != 0)
{
for (i = start; (i < end); i++)
{
/* Initiate local variables (for this i-particle) to 0 */
qiA = chargeA[i]*one_4pi_eps;
qiB = chargeB[i]*one_4pi_eps;
if (calc_excl_corr)
{
i1 = excl->index[i];
i2 = excl->index[i+1];
/* Loop over excluded neighbours */
for (j = i1; (j < i2); j++)
{
k = AA[j];
if (k > i)
{
qqA = qiA*chargeA[k];
qqB = qiB*chargeB[k];
if (qqA != 0.0 || qqB != 0.0)
{
qqL = L1*qqA + lambda*qqB;
rvec_sub(x[i], x[k], dx);
if (bMolPBC)
{
/* Cheap pbc_dx, assume excluded pairs are at short distance. */
for (m = DIM-1; (m >= 0); m--)
{
if (dx[m] > 0.5*box[m][m])
{
rvec_dec(dx, box[m]);
}
else if (dx[m] < -0.5*box[m][m])
{
rvec_inc(dx, box[m]);
}
}
}
dr2 = norm2(dx);
if (dr2 != 0)
{
rinv = gmx_invsqrt(dr2);
rinv2 = rinv*rinv;
dr = 1.0/rinv;
v = gmx_erf(ewc*dr)*rinv;
vc = qqL*v;
Vexcl += vc;
fscal = rinv2*(vc-qqL*ewc*M_2_SQRTPI*exp(-ewc*ewc*dr2));
svmul(fscal, dx, df);
rvec_inc(f[k], df);
rvec_dec(f[i], df);
for (iv = 0; (iv < DIM); iv++)
{
for (jv = 0; (jv < DIM); jv++)
{
dxdf[iv][jv] += dx[iv]*df[jv];
}
}
dvdl_excl += (qqB - qqA)*v;
}
else
{
Vexcl += qqL*vr0;
dvdl_excl += (qqB - qqA)*vr0;
}
}
}
}
}
/* Dipole correction on force */
if (dipole_coeff != 0)
{
for (j = 0; (j < DIM); j++)
{
f[i][j] -= L1*dipcorrA[j]*chargeA[i]
+ lambda*dipcorrB[j]*chargeB[i];
}
}
}
}
for (iv = 0; (iv < DIM); iv++)
{
for (jv = 0; (jv < DIM); jv++)
{
vir[iv][jv] += 0.5*dxdf[iv][jv];
}
}
Vself[0] = 0;
Vself[1] = 0;
/* Global corrections only on master process */
if (MASTER(cr) && thread == 0)
{
for (q = 0; q < (bFreeEnergy ? 2 : 1); q++)
{
if (calc_excl_corr)
{
/* Self-energy correction */
Vself[q] = ewc*one_4pi_eps*fr->q2sum[q]*M_1_SQRTPI;
}
/* Apply surface dipole correction:
* correction = dipole_coeff * (dipole)^2
*/
if (dipole_coeff != 0)
{
if (ewald_geometry == eewg3D)
{
Vdipole[q] = dipole_coeff*iprod(mutot[q], mutot[q]);
}
else if (ewald_geometry == eewg3DC)
{
Vdipole[q] = dipole_coeff*mutot[q][ZZ]*mutot[q][ZZ];
}
}
}
}
if (!bFreeEnergy)
{
enercorr = Vdipole[0] - Vself[0] - Vexcl;
}
else
{
enercorr = L1*(Vdipole[0] - Vself[0])
+ lambda*(Vdipole[1] - Vself[1])
- Vexcl;
*dvdlambda += Vdipole[1] - Vself[1]
- (Vdipole[0] - Vself[0]) - dvdl_excl;
}
if (debug)
{
fprintf(debug, "Long Range corrections for Ewald interactions:\n");
fprintf(debug, "start=%d,natoms=%d\n", start, end-start);
fprintf(debug, "q2sum = %g, Vself=%g\n",
L1*q2sumA+lambda*q2sumB, L1*Vself[0]+lambda*Vself[1]);
fprintf(debug, "Long Range correction: Vexcl=%g\n", Vexcl);
if (MASTER(cr) && thread == 0)
{
if (epsilon_surface > 0 || ewald_geometry == eewg3DC)
{
fprintf(debug, "Total dipole correction: Vdipole=%g\n",
L1*Vdipole[0]+lambda*Vdipole[1]);
}
}
}
/* Return the correction to the energy */
return enercorr;
}
void set_lincs_matrix(struct gmx_lincsdata *li,real *invmass,real lambda)
{
int i,a1,a2,n,k,sign,center;
int end,nk,kk;
const real invsqrt2=0.7071067811865475244;
for(i=0; (i<li->nc); i++)
{
a1 = li->bla[2*i];
a2 = li->bla[2*i+1];
li->blc[i] = gmx_invsqrt(invmass[a1] + invmass[a2]);
li->blc1[i] = invsqrt2;
}
/* Construct the coupling coefficient matrix blmf */
li->ntriangle = 0;
li->ncc_triangle = 0;
for(i=0; (i<li->nc); i++)
{
a1 = li->bla[2*i];
a2 = li->bla[2*i+1];
for(n=li->blnr[i]; (n<li->blnr[i+1]); n++)
{
k = li->blbnb[n];
if (a1 == li->bla[2*k] || a2 == li->bla[2*k+1])
{
sign = -1;
}
else
{
sign = 1;
}
if (a1 == li->bla[2*k] || a1 == li->bla[2*k+1])
{
center = a1;
end = a2;
}
else
{
center = a2;
end = a1;
}
li->blmf[n] = sign*invmass[center]*li->blc[i]*li->blc[k];
li->blmf1[n] = sign*0.5;
if (li->ncg_triangle > 0)
{
/* Look for constraint triangles */
for(nk=li->blnr[k]; (nk<li->blnr[k+1]); nk++)
{
kk = li->blbnb[nk];
if (kk != i && kk != k &&
(li->bla[2*kk] == end || li->bla[2*kk+1] == end))
{
if (li->ntriangle == 0 ||
li->triangle[li->ntriangle-1] < i)
{
/* Add this constraint to the triangle list */
li->triangle[li->ntriangle] = i;
li->tri_bits[li->ntriangle] = 0;
li->ntriangle++;
if (li->blnr[i+1] - li->blnr[i] > sizeof(li->tri_bits[0])*8 - 1)
{
gmx_fatal(FARGS,"A constraint is connected to %d constraints, this is more than the %d allowed for constraints participating in triangles",
li->blnr[i+1] - li->blnr[i],
sizeof(li->tri_bits[0])*8-1);
}
}
li->tri_bits[li->ntriangle-1] |= (1<<(n-li->blnr[i]));
li->ncc_triangle++;
}
}
}
}
}
if (debug)
{
fprintf(debug,"Of the %d constraints %d participate in triangles\n",
li->nc,li->ntriangle);
fprintf(debug,"There are %d couplings of which %d in triangles\n",
li->ncc,li->ncc_triangle);
}
/* Set matlam,
* so we know with which lambda value the masses have been set.
*/
li->matlam = lambda;
}
static void do_lincs(rvec *x,rvec *xp,matrix box,t_pbc *pbc,
struct gmx_lincsdata *lincsd,real *invmass,
t_commrec *cr,
real wangle,int *warn,
real invdt,rvec *v,
gmx_bool bCalcVir,tensor rmdr)
{
int b,i,j,k,n,iter;
real tmp0,tmp1,tmp2,im1,im2,mvb,rlen,len,len2,dlen2,wfac,lam;
rvec dx;
int ncons,*bla,*blnr,*blbnb;
rvec *r;
real *blc,*blmf,*bllen,*blcc,*rhs1,*rhs2,*sol,*lambda;
int *nlocat;
ncons = lincsd->nc;
bla = lincsd->bla;
r = lincsd->tmpv;
blnr = lincsd->blnr;
blbnb = lincsd->blbnb;
blc = lincsd->blc;
blmf = lincsd->blmf;
bllen = lincsd->bllen;
blcc = lincsd->tmpncc;
rhs1 = lincsd->tmp1;
rhs2 = lincsd->tmp2;
sol = lincsd->tmp3;
lambda = lincsd->lambda;
if (DOMAINDECOMP(cr) && cr->dd->constraints)
{
nlocat = dd_constraints_nlocalatoms(cr->dd);
}
else if (PARTDECOMP(cr))
{
nlocat = pd_constraints_nlocalatoms(cr->pd);
}
else
{
nlocat = NULL;
}
*warn = 0;
if (pbc)
{
/* Compute normalized i-j vectors */
for(b=0; b<ncons; b++)
{
pbc_dx_aiuc(pbc,x[bla[2*b]],x[bla[2*b+1]],dx);
unitv(dx,r[b]);
}
for(b=0; b<ncons; b++)
{
for(n=blnr[b]; n<blnr[b+1]; n++)
{
blcc[n] = blmf[n]*iprod(r[b],r[blbnb[n]]);
}
pbc_dx_aiuc(pbc,xp[bla[2*b]],xp[bla[2*b+1]],dx);
mvb = blc[b]*(iprod(r[b],dx) - bllen[b]);
rhs1[b] = mvb;
sol[b] = mvb;
}
}
else
{
/* Compute normalized i-j vectors */
for(b=0; b<ncons; b++)
{
i = bla[2*b];
j = bla[2*b+1];
tmp0 = x[i][0] - x[j][0];
tmp1 = x[i][1] - x[j][1];
tmp2 = x[i][2] - x[j][2];
rlen = gmx_invsqrt(tmp0*tmp0+tmp1*tmp1+tmp2*tmp2);
r[b][0] = rlen*tmp0;
r[b][1] = rlen*tmp1;
r[b][2] = rlen*tmp2;
} /* 16 ncons flops */
for(b=0; b<ncons; b++)
{
tmp0 = r[b][0];
tmp1 = r[b][1];
tmp2 = r[b][2];
len = bllen[b];
i = bla[2*b];
j = bla[2*b+1];
for(n=blnr[b]; n<blnr[b+1]; n++)
{
k = blbnb[n];
blcc[n] = blmf[n]*(tmp0*r[k][0] + tmp1*r[k][1] + tmp2*r[k][2]);
} /* 6 nr flops */
mvb = blc[b]*(tmp0*(xp[i][0] - xp[j][0]) +
tmp1*(xp[i][1] - xp[j][1]) +
tmp2*(xp[i][2] - xp[j][2]) - len);
rhs1[b] = mvb;
sol[b] = mvb;
/* 10 flops */
}
/* Together: 26*ncons + 6*nrtot flops */
}
lincs_matrix_expand(lincsd,blcc,rhs1,rhs2,sol);
/* nrec*(ncons+2*nrtot) flops */
for(b=0; b<ncons; b++)
{
i = bla[2*b];
j = bla[2*b+1];
mvb = blc[b]*sol[b];
lambda[b] = -mvb;
im1 = invmass[i];
im2 = invmass[j];
tmp0 = r[b][0]*mvb;
tmp1 = r[b][1]*mvb;
tmp2 = r[b][2]*mvb;
xp[i][0] -= tmp0*im1;
xp[i][1] -= tmp1*im1;
xp[i][2] -= tmp2*im1;
xp[j][0] += tmp0*im2;
xp[j][1] += tmp1*im2;
xp[j][2] += tmp2*im2;
} /* 16 ncons flops */
/*
******** Correction for centripetal effects ********
*/
wfac = cos(DEG2RAD*wangle);
wfac = wfac*wfac;
for(iter=0; iter<lincsd->nIter; iter++)
{
if (DOMAINDECOMP(cr) && cr->dd->constraints)
{
/* Communicate the corrected non-local coordinates */
dd_move_x_constraints(cr->dd,box,xp,NULL);
}
else if (PARTDECOMP(cr))
{
pd_move_x_constraints(cr,xp,NULL);
}
for(b=0; b<ncons; b++)
{
len = bllen[b];
if (pbc)
{
pbc_dx_aiuc(pbc,xp[bla[2*b]],xp[bla[2*b+1]],dx);
}
else
{
rvec_sub(xp[bla[2*b]],xp[bla[2*b+1]],dx);
}
len2 = len*len;
dlen2 = 2*len2 - norm2(dx);
if (dlen2 < wfac*len2 && (nlocat==NULL || nlocat[b]))
{
*warn = b;
}
if (dlen2 > 0)
{
mvb = blc[b]*(len - dlen2*gmx_invsqrt(dlen2));
}
else
{
mvb = blc[b]*len;
}
rhs1[b] = mvb;
sol[b] = mvb;
} /* 20*ncons flops */
lincs_matrix_expand(lincsd,blcc,rhs1,rhs2,sol);
/* nrec*(ncons+2*nrtot) flops */
for(b=0; b<ncons; b++)
{
i = bla[2*b];
j = bla[2*b+1];
lam = lambda[b];
mvb = blc[b]*sol[b];
lambda[b] = lam - mvb;
im1 = invmass[i];
im2 = invmass[j];
tmp0 = r[b][0]*mvb;
tmp1 = r[b][1]*mvb;
tmp2 = r[b][2]*mvb;
xp[i][0] -= tmp0*im1;
xp[i][1] -= tmp1*im1;
xp[i][2] -= tmp2*im1;
xp[j][0] += tmp0*im2;
xp[j][1] += tmp1*im2;
xp[j][2] += tmp2*im2;
} /* 17 ncons flops */
} /* nit*ncons*(37+9*nrec) flops */
if (v)
{
/* Correct the velocities */
for(b=0; b<ncons; b++)
{
i = bla[2*b];
j = bla[2*b+1];
im1 = invmass[i]*lambda[b]*invdt;
im2 = invmass[j]*lambda[b]*invdt;
v[i][0] += im1*r[b][0];
v[i][1] += im1*r[b][1];
v[i][2] += im1*r[b][2];
v[j][0] -= im2*r[b][0];
v[j][1] -= im2*r[b][1];
v[j][2] -= im2*r[b][2];
} /* 16 ncons flops */
}
if (nlocat)
{
/* Only account for local atoms */
for(b=0; b<ncons; b++)
{
lambda[b] *= 0.5*nlocat[b];
}
}
if (bCalcVir)
{
/* Constraint virial */
for(b=0; b<ncons; b++)
{
tmp0 = bllen[b]*lambda[b];
for(i=0; i<DIM; i++)
{
tmp1 = tmp0*r[b][i];
for(j=0; j<DIM; j++)
{
rmdr[i][j] -= tmp1*r[b][j];
}
}
} /* 22 ncons flops */
}
/* Total:
* 26*ncons + 6*nrtot + nrec*(ncons+2*nrtot)
* + nit * (20*ncons + nrec*(ncons+2*nrtot) + 17 ncons)
*
* (26+nrec)*ncons + (6+2*nrec)*nrtot
* + nit * ((37+nrec)*ncons + 2*nrec*nrtot)
* if nit=1
* (63+nrec)*ncons + (6+4*nrec)*nrtot
*/
}
/*
* Gromacs nonbonded kernel pf_nb_kernel304
* Coulomb interaction: Tabulated
* VdW interaction: Not calculated
* water optimization: pairs of TIP4P interactions
* Calculate forces: yes
*/
void pf_nb_kernel304(
int * p_nri,
int * iinr,
int * jindex,
int * jjnr,
int * shift,
real * shiftvec,
real * fshift,
int * gid,
real * pos,
real * faction,
real * charge,
real * p_facel,
real * p_krf,
real * p_crf,
real * Vc,
int * type,
int * p_ntype,
real * vdwparam,
real * Vvdw,
real * p_tabscale,
real * VFtab,
real * invsqrta,
real * dvda,
real * p_gbtabscale,
real * GBtab,
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
real * work,
t_pf_global * pf_global)
{
int nri,ntype,nthreads;
real facel,krf,crf,tabscale,gbtabscale;
int n,ii,is3,ii3,k,nj0,nj1,jnr,j3,ggid;
int nn0,nn1,nouter,ninner;
real shX,shY,shZ;
real fscal,tx,ty,tz;
real qq,vcoul,vctot;
real r,rt,eps,eps2;
int n0,nnn;
real Y,F,Geps,Heps2,Fp,VV;
real FF;
real fijC;
real ix2,iy2,iz2,fix2,fiy2,fiz2;
real ix3,iy3,iz3,fix3,fiy3,fiz3;
real ix4,iy4,iz4,fix4,fiy4,fiz4;
real jx2,jy2,jz2,fjx2,fjy2,fjz2;
real jx3,jy3,jz3,fjx3,fjy3,fjz3;
real jx4,jy4,jz4,fjx4,fjy4,fjz4;
real dx22,dy22,dz22,rsq22,rinv22;
real dx23,dy23,dz23,rsq23,rinv23;
real dx24,dy24,dz24,rsq24,rinv24;
real dx32,dy32,dz32,rsq32,rinv32;
real dx33,dy33,dz33,rsq33,rinv33;
real dx34,dy34,dz34,rsq34,rinv34;
real dx42,dy42,dz42,rsq42,rinv42;
real dx43,dy43,dz43,rsq43,rinv43;
real dx44,dy44,dz44,rsq44,rinv44;
real qH,qM,qqMM,qqMH,qqHH;
nri = *p_nri;
ntype = *p_ntype;
nthreads = *p_nthreads;
facel = *p_facel;
krf = *p_krf;
crf = *p_crf;
tabscale = *p_tabscale;
/* Initialize water data */
ii = iinr[0];
qH = charge[ii+1];
qM = charge[ii+3];
qqMM = facel*qM*qM;
qqMH = facel*qM*qH;
qqHH = facel*qH*qH;
/* Reset outer and inner iteration counters */
nouter = 0;
ninner = 0;
/* Loop over thread workunits */
do
{
#ifdef GMX_THREAD_SHM_FDECOMP
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
/* Take successively smaller chunks (at least 10 lists) */
nn1 = nn0+(nri-nn0)/(2*nthreads)+10;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
#else
nn0 = 0;
nn1 = nri;
#endif
/* Start outer loop over neighborlists */
for(n=nn0; (n<nn1); n++)
{
/* Load shift vector for this list */
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
/* Load limits for loop over neighbors */
nj0 = jindex[n];
nj1 = jindex[n+1];
/* Get outer coordinate index */
ii = iinr[n];
ii3 = 3*ii;
/* Load i atom data, add shift vector */
ix2 = shX + pos[ii3+3];
iy2 = shY + pos[ii3+4];
iz2 = shZ + pos[ii3+5];
ix3 = shX + pos[ii3+6];
iy3 = shY + pos[ii3+7];
iz3 = shZ + pos[ii3+8];
ix4 = shX + pos[ii3+9];
iy4 = shY + pos[ii3+10];
iz4 = shZ + pos[ii3+11];
/* Zero the potential energy for this list */
vctot = 0;
/* Clear i atom forces */
fix2 = 0;
fiy2 = 0;
fiz2 = 0;
fix3 = 0;
fiy3 = 0;
fiz3 = 0;
fix4 = 0;
fiy4 = 0;
fiz4 = 0;
for(k=nj0; (k<nj1); k++)
{
/* Get j neighbor index, and coordinate index */
jnr = jjnr[k];
j3 = 3*jnr;
/* load j atom coordinates */
jx2 = pos[j3+3];
jy2 = pos[j3+4];
jz2 = pos[j3+5];
jx3 = pos[j3+6];
jy3 = pos[j3+7];
jz3 = pos[j3+8];
jx4 = pos[j3+9];
jy4 = pos[j3+10];
jz4 = pos[j3+11];
/* Calculate distance */
dx22 = ix2 - jx2;
dy22 = iy2 - jy2;
dz22 = iz2 - jz2;
rsq22 = dx22*dx22+dy22*dy22+dz22*dz22;
dx23 = ix2 - jx3;
dy23 = iy2 - jy3;
dz23 = iz2 - jz3;
rsq23 = dx23*dx23+dy23*dy23+dz23*dz23;
dx24 = ix2 - jx4;
dy24 = iy2 - jy4;
dz24 = iz2 - jz4;
rsq24 = dx24*dx24+dy24*dy24+dz24*dz24;
dx32 = ix3 - jx2;
dy32 = iy3 - jy2;
dz32 = iz3 - jz2;
rsq32 = dx32*dx32+dy32*dy32+dz32*dz32;
dx33 = ix3 - jx3;
dy33 = iy3 - jy3;
dz33 = iz3 - jz3;
rsq33 = dx33*dx33+dy33*dy33+dz33*dz33;
dx34 = ix3 - jx4;
dy34 = iy3 - jy4;
dz34 = iz3 - jz4;
rsq34 = dx34*dx34+dy34*dy34+dz34*dz34;
dx42 = ix4 - jx2;
dy42 = iy4 - jy2;
dz42 = iz4 - jz2;
rsq42 = dx42*dx42+dy42*dy42+dz42*dz42;
dx43 = ix4 - jx3;
dy43 = iy4 - jy3;
dz43 = iz4 - jz3;
rsq43 = dx43*dx43+dy43*dy43+dz43*dz43;
dx44 = ix4 - jx4;
dy44 = iy4 - jy4;
dz44 = iz4 - jz4;
rsq44 = dx44*dx44+dy44*dy44+dz44*dz44;
/* Calculate 1/r and 1/r2 */
rinv22 = gmx_invsqrt(rsq22);
rinv23 = gmx_invsqrt(rsq23);
rinv24 = gmx_invsqrt(rsq24);
rinv32 = gmx_invsqrt(rsq32);
rinv33 = gmx_invsqrt(rsq33);
rinv34 = gmx_invsqrt(rsq34);
rinv42 = gmx_invsqrt(rsq42);
rinv43 = gmx_invsqrt(rsq43);
rinv44 = gmx_invsqrt(rsq44);
/* Load parameters for j atom */
qq = qqHH;
/* Calculate table index */
r = rsq22*rinv22;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
fscal = -((fijC)*tabscale)*rinv22;
/* Calculate temporary vectorial force */
tx = fscal*dx22;
ty = fscal*dy22;
tz = fscal*dz22;
/* Increment i atom force */
fix2 = fix2 + tx;
fiy2 = fiy2 + ty;
fiz2 = fiz2 + tz;
/* Decrement j atom force */
fjx2 = faction[j3+3] - tx;
fjy2 = faction[j3+4] - ty;
fjz2 = faction[j3+5] - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+1, jnr+1, PF_INTER_COULOMB, fscal, dx22, dy22, dz22);
/* Load parameters for j atom */
qq = qqHH;
/* Calculate table index */
r = rsq23*rinv23;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
fscal = -((fijC)*tabscale)*rinv23;
/* Calculate temporary vectorial force */
tx = fscal*dx23;
ty = fscal*dy23;
tz = fscal*dz23;
/* Increment i atom force */
fix2 = fix2 + tx;
fiy2 = fiy2 + ty;
fiz2 = fiz2 + tz;
/* Decrement j atom force */
fjx3 = faction[j3+6] - tx;
fjy3 = faction[j3+7] - ty;
fjz3 = faction[j3+8] - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+1, jnr+2, PF_INTER_COULOMB, fscal, dx23, dy23, dz23);
/* Load parameters for j atom */
qq = qqMH;
/* Calculate table index */
r = rsq24*rinv24;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
fscal = -((fijC)*tabscale)*rinv24;
/* Calculate temporary vectorial force */
tx = fscal*dx24;
ty = fscal*dy24;
tz = fscal*dz24;
/* Increment i atom force */
fix2 = fix2 + tx;
fiy2 = fiy2 + ty;
fiz2 = fiz2 + tz;
/* Decrement j atom force */
fjx4 = faction[j3+9] - tx;
fjy4 = faction[j3+10] - ty;
fjz4 = faction[j3+11] - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+1, jnr+3, PF_INTER_COULOMB, fscal, dx24, dy24, dz24);
/* Load parameters for j atom */
qq = qqHH;
/* Calculate table index */
r = rsq32*rinv32;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
fscal = -((fijC)*tabscale)*rinv32;
/* Calculate temporary vectorial force */
tx = fscal*dx32;
ty = fscal*dy32;
tz = fscal*dz32;
/* Increment i atom force */
fix3 = fix3 + tx;
fiy3 = fiy3 + ty;
fiz3 = fiz3 + tz;
/* Decrement j atom force */
fjx2 = fjx2 - tx;
fjy2 = fjy2 - ty;
fjz2 = fjz2 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+2, jnr+1, PF_INTER_COULOMB, fscal, dx32, dy32, dz32);
/* Load parameters for j atom */
qq = qqHH;
/* Calculate table index */
r = rsq33*rinv33;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
fscal = -((fijC)*tabscale)*rinv33;
/* Calculate temporary vectorial force */
tx = fscal*dx33;
ty = fscal*dy33;
tz = fscal*dz33;
/* Increment i atom force */
fix3 = fix3 + tx;
fiy3 = fiy3 + ty;
fiz3 = fiz3 + tz;
/* Decrement j atom force */
fjx3 = fjx3 - tx;
fjy3 = fjy3 - ty;
fjz3 = fjz3 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+2, jnr+2, PF_INTER_COULOMB, fscal, dx33, dy33, dz33);
/* Load parameters for j atom */
qq = qqMH;
/* Calculate table index */
r = rsq34*rinv34;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
fscal = -((fijC)*tabscale)*rinv34;
/* Calculate temporary vectorial force */
tx = fscal*dx34;
ty = fscal*dy34;
tz = fscal*dz34;
/* Increment i atom force */
fix3 = fix3 + tx;
fiy3 = fiy3 + ty;
fiz3 = fiz3 + tz;
/* Decrement j atom force */
fjx4 = fjx4 - tx;
fjy4 = fjy4 - ty;
fjz4 = fjz4 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+2, jnr+3, PF_INTER_COULOMB, fscal, dx34, dy34, dz34);
/* Load parameters for j atom */
qq = qqMH;
/* Calculate table index */
r = rsq42*rinv42;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
fscal = -((fijC)*tabscale)*rinv42;
/* Calculate temporary vectorial force */
tx = fscal*dx42;
ty = fscal*dy42;
tz = fscal*dz42;
/* Increment i atom force */
fix4 = fix4 + tx;
fiy4 = fiy4 + ty;
fiz4 = fiz4 + tz;
/* Decrement j atom force */
faction[j3+3] = fjx2 - tx;
faction[j3+4] = fjy2 - ty;
faction[j3+5] = fjz2 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+3, jnr+1, PF_INTER_COULOMB, fscal, dx42, dy42, dz42);
/* Load parameters for j atom */
qq = qqMH;
/* Calculate table index */
r = rsq43*rinv43;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
fscal = -((fijC)*tabscale)*rinv43;
/* Calculate temporary vectorial force */
tx = fscal*dx43;
ty = fscal*dy43;
tz = fscal*dz43;
/* Increment i atom force */
fix4 = fix4 + tx;
fiy4 = fiy4 + ty;
fiz4 = fiz4 + tz;
/* Decrement j atom force */
faction[j3+6] = fjx3 - tx;
faction[j3+7] = fjy3 - ty;
faction[j3+8] = fjz3 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+3, jnr+2, PF_INTER_COULOMB, fscal, dx43, dy43, dz43);
/* Load parameters for j atom */
qq = qqMM;
/* Calculate table index */
r = rsq44*rinv44;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vcoul = qq*VV;
fijC = qq*FF;
vctot = vctot + vcoul;
fscal = -((fijC)*tabscale)*rinv44;
/* Calculate temporary vectorial force */
tx = fscal*dx44;
ty = fscal*dy44;
tz = fscal*dz44;
/* Increment i atom force */
fix4 = fix4 + tx;
fiy4 = fiy4 + ty;
fiz4 = fiz4 + tz;
/* Decrement j atom force */
faction[j3+9] = fjx4 - tx;
faction[j3+10] = fjy4 - ty;
faction[j3+11] = fjz4 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+3, jnr+3, PF_INTER_COULOMB, fscal, dx44, dy44, dz44);
/* Inner loop uses 369 flops/iteration */
}
/* Add i forces to mem and shifted force list */
faction[ii3+3] = faction[ii3+3] + fix2;
faction[ii3+4] = faction[ii3+4] + fiy2;
faction[ii3+5] = faction[ii3+5] + fiz2;
faction[ii3+6] = faction[ii3+6] + fix3;
faction[ii3+7] = faction[ii3+7] + fiy3;
faction[ii3+8] = faction[ii3+8] + fiz3;
faction[ii3+9] = faction[ii3+9] + fix4;
faction[ii3+10] = faction[ii3+10] + fiy4;
faction[ii3+11] = faction[ii3+11] + fiz4;
fshift[is3] = fshift[is3]+fix2+fix3+fix4;
fshift[is3+1] = fshift[is3+1]+fiy2+fiy3+fiy4;
fshift[is3+2] = fshift[is3+2]+fiz2+fiz3+fiz4;
/* Add potential energies to the group for this list */
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
/* Increment number of inner iterations */
ninner = ninner + nj1 - nj0;
/* Outer loop uses 28 flops/iteration */
}
/* Increment number of outer iterations */
nouter = nouter + nn1 - nn0;
}
while (nn1<nri);
/* Write outer/inner iteration count to pointers */
*outeriter = nouter;
*inneriter = ninner;
}
void
nb_kernel_allvsall(t_forcerec * fr,
t_mdatoms * mdatoms,
t_blocka * excl,
real * x,
real * f,
real * Vc,
real * Vvdw,
int * outeriter,
int * inneriter,
void * work)
{
gmx_allvsall_data_t *aadata;
int natoms;
int ni0,ni1;
int nj0,nj1,nj2;
int i,j,k;
real * charge;
int * type;
real facel;
real * pvdw;
int ggid;
int * mask;
real ix,iy,iz,iq;
real fix,fiy,fiz;
real jx,jy,jz,qq;
real dx,dy,dz;
real tx,ty,tz;
real rsq,rinv,rinvsq,rinvsix;
real vcoul,vctot;
real c6,c12,Vvdw6,Vvdw12,Vvdwtot;
real fscal;
charge = mdatoms->chargeA;
type = mdatoms->typeA;
facel = fr->epsfac;
natoms = mdatoms->nr;
ni0 = mdatoms->start;
ni1 = mdatoms->start+mdatoms->homenr;
aadata = *((gmx_allvsall_data_t **)work);
if(aadata==NULL)
{
setup_aadata(&aadata,excl,natoms,type,fr->ntype,fr->nbfp);
*((gmx_allvsall_data_t **)work) = aadata;
}
for(i=ni0; i<ni1; i++)
{
/* We assume shifts are NOT used for all-vs-all interactions */
/* Load i atom data */
ix = x[3*i];
iy = x[3*i+1];
iz = x[3*i+2];
iq = facel*charge[i];
pvdw = aadata->pvdwparam[type[i]];
/* Zero the potential energy for this list */
Vvdwtot = 0.0;
vctot = 0.0;
/* Clear i atom forces */
fix = 0.0;
fiy = 0.0;
fiz = 0.0;
/* Load limits for loop over neighbors */
nj0 = aadata->jindex[3*i];
nj1 = aadata->jindex[3*i+1];
nj2 = aadata->jindex[3*i+2];
mask = aadata->exclusion_mask[i];
/* Prologue part, including exclusion mask */
for(j=nj0; j<nj1; j++,mask++)
{
if(*mask!=0)
{
k = j%natoms;
/* load j atom coordinates */
jx = x[3*k];
jy = x[3*k+1];
jz = x[3*k+2];
/* Calculate distance */
dx = ix - jx;
dy = iy - jy;
dz = iz - jz;
rsq = dx*dx+dy*dy+dz*dz;
/* Calculate 1/r and 1/r2 */
rinv = gmx_invsqrt(rsq);
rinvsq = rinv*rinv;
/* Load parameters for j atom */
qq = iq*charge[k];
c6 = pvdw[2*k];
c12 = pvdw[2*k+1];
/* Coulomb interaction */
vcoul = qq*rinv;
vctot = vctot+vcoul;
/* Lennard-Jones interaction */
rinvsix = rinvsq*rinvsq*rinvsq;
Vvdw6 = c6*rinvsix;
Vvdw12 = c12*rinvsix*rinvsix;
Vvdwtot = Vvdwtot+Vvdw12-Vvdw6;
fscal = (vcoul+12.0*Vvdw12-6.0*Vvdw6)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx;
ty = fscal*dy;
tz = fscal*dz;
/* Increment i atom force */
fix = fix + tx;
fiy = fiy + ty;
fiz = fiz + tz;
/* Decrement j atom force */
f[3*k] = f[3*k] - tx;
f[3*k+1] = f[3*k+1] - ty;
f[3*k+2] = f[3*k+2] - tz;
}
/* Inner loop uses 38 flops/iteration */
}
/* Main part, no exclusions */
for(j=nj1; j<nj2; j++)
{
k = j%natoms;
/* load j atom coordinates */
jx = x[3*k];
jy = x[3*k+1];
jz = x[3*k+2];
/* Calculate distance */
dx = ix - jx;
dy = iy - jy;
dz = iz - jz;
rsq = dx*dx+dy*dy+dz*dz;
/* Calculate 1/r and 1/r2 */
rinv = gmx_invsqrt(rsq);
rinvsq = rinv*rinv;
/* Load parameters for j atom */
qq = iq*charge[k];
c6 = pvdw[2*k];
c12 = pvdw[2*k+1];
/* Coulomb interaction */
vcoul = qq*rinv;
vctot = vctot+vcoul;
/* Lennard-Jones interaction */
rinvsix = rinvsq*rinvsq*rinvsq;
Vvdw6 = c6*rinvsix;
Vvdw12 = c12*rinvsix*rinvsix;
Vvdwtot = Vvdwtot+Vvdw12-Vvdw6;
fscal = (vcoul+12.0*Vvdw12-6.0*Vvdw6)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx;
ty = fscal*dy;
tz = fscal*dz;
/* Increment i atom force */
fix = fix + tx;
fiy = fiy + ty;
fiz = fiz + tz;
/* Decrement j atom force */
f[3*k] = f[3*k] - tx;
f[3*k+1] = f[3*k+1] - ty;
f[3*k+2] = f[3*k+2] - tz;
/* Inner loop uses 38 flops/iteration */
}
f[3*i] += fix;
f[3*i+1] += fiy;
f[3*i+2] += fiz;
/* Add potential energies to the group for this list */
ggid = 0;
Vc[ggid] = Vc[ggid] + vctot;
Vvdw[ggid] = Vvdw[ggid] + Vvdwtot;
/* Outer loop uses 6 flops/iteration */
}
/* Write outer/inner iteration count to pointers */
*outeriter = ni1-ni0;
*inneriter = (ni1-ni0)*natoms/2;
}
real ta_disres(int nfa,const t_iatom forceatoms[],const t_iparams ip[],
const rvec x[],rvec f[],rvec fshift[],
const t_pbc *pbc,const t_graph *g,
real lambda,real *dvdlambda,
const t_mdatoms *md,t_fcdata *fcd,
int *global_atom_index)
{
const real sixth=1.0/6.0;
const real seven_three=7.0/3.0;
atom_id ai,aj;
int fa,res,npair,p,pair,ki=CENTRAL,m;
int type;
rvec dx;
real weight_rt_1;
real smooth_fc,Rt,Rtav,rt2,*Rtl_6,*Rt_6,*Rtav_6;
real k0,f_scal=0,fmax_scal,fk_scal,fij;
real tav_viol,instant_viol,mixed_viol,violtot,vtot;
real tav_viol_Rtav7,instant_viol_Rtav7;
real up1,up2,low;
gmx_bool bConservative,bMixed,bViolation;
ivec it,jt,dt;
t_disresdata *dd;
int dr_weighting;
gmx_bool dr_bMixed;
dd = &(fcd->disres);
dr_weighting = dd->dr_weighting;
dr_bMixed = dd->dr_bMixed;
Rtl_6 = dd->Rtl_6;
Rt_6 = dd->Rt_6;
Rtav_6 = dd->Rtav_6;
tav_viol=instant_viol=mixed_viol=tav_viol_Rtav7=instant_viol_Rtav7=0;
smooth_fc = dd->dr_fc;
if (dd->dr_tau != 0)
{
/* scaling factor to smoothly turn on the restraint forces *
* when using time averaging */
smooth_fc *= (1.0 - dd->exp_min_t_tau);
}
violtot = 0;
vtot = 0;
/* 'loop' over all atom pairs (pair_nr=fa/3) involved in restraints, *
* the total number of atoms pairs is nfa/3 */
res = 0;
fa = 0;
while (fa < nfa)
{
type = forceatoms[fa];
/* Take action depending on restraint, calculate scalar force */
npair = ip[type].disres.npair;
up1 = ip[type].disres.up1;
up2 = ip[type].disres.up2;
low = ip[type].disres.low;
k0 = smooth_fc*ip[type].disres.kfac;
/* save some flops when there is only one pair */
if (ip[type].disres.type != 2)
{
bConservative = (dr_weighting == edrwConservative) && (npair > 1);
bMixed = dr_bMixed;
Rt = pow(Rt_6[res],-sixth);
Rtav = pow(Rtav_6[res],-sixth);
}
else
{
/* When rtype=2 use instantaneous not ensemble avereged distance */
bConservative = (npair > 1);
bMixed = FALSE;
Rt = pow(Rtl_6[res],-sixth);
Rtav = Rt;
}
if (Rtav > up1)
{
bViolation = TRUE;
tav_viol = Rtav - up1;
}
else if (Rtav < low)
{
bViolation = TRUE;
tav_viol = Rtav - low;
}
else
{
bViolation = FALSE;
}
if (bViolation)
{
/* NOTE:
* there is no real potential when time averaging is applied
*/
vtot += 0.5*k0*sqr(tav_viol);
if (1/vtot == 0)
{
printf("vtot is inf: %f\n",vtot);
}
if (!bMixed)
{
f_scal = -k0*tav_viol;
violtot += fabs(tav_viol);
}
else
{
if (Rt > up1)
{
if (tav_viol > 0)
{
instant_viol = Rt - up1;
}
else
{
bViolation = FALSE;
}
}
else if (Rt < low)
{
if (tav_viol < 0)
{
instant_viol = Rt - low;
}
else
{
bViolation = FALSE;
}
}
else
{
bViolation = FALSE;
}
if (bViolation)
{
mixed_viol = sqrt(tav_viol*instant_viol);
f_scal = -k0*mixed_viol;
violtot += mixed_viol;
}
}
}
if (bViolation)
{
fmax_scal = -k0*(up2-up1);
/* Correct the force for the number of restraints */
if (bConservative)
{
f_scal = max(f_scal,fmax_scal);
if (!bMixed)
{
f_scal *= Rtav/Rtav_6[res];
}
else
{
f_scal /= 2*mixed_viol;
tav_viol_Rtav7 = tav_viol*Rtav/Rtav_6[res];
instant_viol_Rtav7 = instant_viol*Rt/Rt_6[res];
}
}
else
{
f_scal /= (real)npair;
f_scal = max(f_scal,fmax_scal);
}
/* Exert the force ... */
/* Loop over the atom pairs of 'this' restraint */
for(p=0; p<npair; p++)
{
pair = fa/3;
ai = forceatoms[fa+1];
aj = forceatoms[fa+2];
if (pbc)
{
ki = pbc_dx_aiuc(pbc,x[ai],x[aj],dx);
}
else
{
rvec_sub(x[ai],x[aj],dx);
}
rt2 = iprod(dx,dx);
weight_rt_1 = gmx_invsqrt(rt2);
if (bConservative)
{
if (!dr_bMixed)
{
weight_rt_1 *= pow(dd->rm3tav[pair],seven_three);
}
else
{
weight_rt_1 *= tav_viol_Rtav7*pow(dd->rm3tav[pair],seven_three)+
instant_viol_Rtav7*pow(dd->rt[pair],-7);
}
}
fk_scal = f_scal*weight_rt_1;
if (g)
{
ivec_sub(SHIFT_IVEC(g,ai),SHIFT_IVEC(g,aj),dt);
ki=IVEC2IS(dt);
}
for(m=0; m<DIM; m++)
{
fij = fk_scal*dx[m];
f[ai][m] += fij;
f[aj][m] -= fij;
fshift[ki][m] += fij;
fshift[CENTRAL][m] -= fij;
}
fa += 3;
}
}
else
{
/* No violation so force and potential contributions */
fa += 3*npair;
}
res++;
}
dd->sumviol = violtot;
/* Return energy */
return vtot;
}
/* Compute factors for restricted dihedral potential
* For explanations on formula used see file "restcbt.h" */
void compute_factors_restrdihs(int type, const t_iparams forceparams[],
rvec delta_ante, rvec delta_crnt, rvec delta_post,
real *factor_phi_ai_ante, real *factor_phi_ai_crnt, real *factor_phi_ai_post,
real *factor_phi_aj_ante, real *factor_phi_aj_crnt, real *factor_phi_aj_post,
real *factor_phi_ak_ante, real *factor_phi_ak_crnt, real *factor_phi_ak_post,
real *factor_phi_al_ante, real *factor_phi_al_crnt, real *factor_phi_al_post,
real *prefactor_phi, real *v)
{
real phi0, cosine_phi0;
real k_torsion;
real c_self_ante, c_self_crnt, c_self_post;
real c_cros_ante, c_cros_acrs, c_cros_post;
real c_prod, d_post, d_ante;
real sine_phi_sq, cosine_phi;
real delta_cosine, term_phi_phi0;
real ratio_phi_ante, ratio_phi_post;
real norm_phi;
/* Read parameters phi0 and k_torsion */
phi0 = forceparams[type].pdihs.phiA * DEG2RAD;
cosine_phi0 = cos(phi0);
k_torsion = forceparams[type].pdihs.cpA;
/* Computation of the cosine of the dihedral angle. The scalar ("dot") product method
* is used. c_*_* cummulate the scalar products of the differences of particles
* positions while c_prod, d_ante and d_post are differences of products of scalar
* terms that are parts of the derivatives of forces */
c_self_ante = iprod(delta_ante, delta_ante);
c_self_crnt = iprod(delta_crnt, delta_crnt);
c_self_post = iprod(delta_post, delta_post);
c_cros_ante = iprod(delta_ante, delta_crnt);
c_cros_acrs = iprod(delta_ante, delta_post);
c_cros_post = iprod(delta_crnt, delta_post);
c_prod = c_cros_ante * c_cros_post - c_self_crnt * c_cros_acrs;
d_ante = c_self_ante * c_self_crnt - c_cros_ante * c_cros_ante;
d_post = c_self_post * c_self_crnt - c_cros_post * c_cros_post;
/* When three consecutive beads align, we obtain values close to zero.
* Here we avoid small values to prevent round-off errors. */
if (d_ante < GMX_REAL_EPS)
{
d_ante = GMX_REAL_EPS;
}
if (d_post < GMX_REAL_EPS)
{
d_post = GMX_REAL_EPS;
}
/* Computes the square of the sinus of phi in sine_phi_sq */
norm_phi = gmx_invsqrt(d_ante * d_post);
cosine_phi = c_prod * norm_phi;
sine_phi_sq = 1.0 - cosine_phi * cosine_phi;
/* It is possible that cosine_phi is slightly bigger than 1.0 due to round-off errors. */
if (sine_phi_sq < 0.0)
{
sine_phi_sq = 0.0;
}
/* Computation of the differences of cosines (delta_cosine) and a term (term_phi_phi0)
* that is part of the common prefactor_phi */
delta_cosine = cosine_phi - cosine_phi0;
term_phi_phi0 = 1 - cosine_phi * cosine_phi0;
/* Computation of ratios */
ratio_phi_ante = c_prod / d_ante;
ratio_phi_post = c_prod / d_post;
/* Computation of the prefactor - common term for all forces */
*prefactor_phi = -(k_torsion) * delta_cosine * norm_phi * term_phi_phi0 / (sine_phi_sq * sine_phi_sq);
/* Computation of force factors. Factors factor_phi_* are coming from the
* derivatives of the torsion angle (phi) with respect to the beads ai, aj, al, ak,
* (four) coordinates and they are multiplied in the force computations with the
* differences of the particles positions stored in parameters delta_ante,
* delta_crnt, delta_post. For formulas see file "restcbt.h" */
*factor_phi_ai_ante = ratio_phi_ante * c_self_crnt;
*factor_phi_ai_crnt = -c_cros_post - ratio_phi_ante * c_cros_ante;
*factor_phi_ai_post = c_self_crnt;
*factor_phi_aj_ante = -c_cros_post - ratio_phi_ante * (c_self_crnt + c_cros_ante);
*factor_phi_aj_crnt = c_cros_post + c_cros_acrs * 2.0 + ratio_phi_ante * (c_self_ante + c_cros_ante) + ratio_phi_post * c_self_post;
*factor_phi_aj_post = -(c_cros_ante + c_self_crnt) - ratio_phi_post * c_cros_post;
*factor_phi_ak_ante = c_cros_post + c_self_crnt + ratio_phi_ante * c_cros_ante;
*factor_phi_ak_crnt = -(c_cros_ante + c_cros_acrs * 2.0)- ratio_phi_ante * c_self_ante - ratio_phi_post * (c_self_post + c_cros_post);
*factor_phi_ak_post = c_cros_ante + ratio_phi_post * (c_self_crnt + c_cros_post);
*factor_phi_al_ante = -c_self_crnt;
*factor_phi_al_crnt = c_cros_ante + ratio_phi_post * c_cros_post;
*factor_phi_al_post = -ratio_phi_post * c_self_crnt;
/* Contribution to energy - see formula in file "restcbt.h"*/
*v = k_torsion * 0.5 * delta_cosine * delta_cosine / sine_phi_sq;
}
void calc_disres_R_6(const gmx_multisim_t *ms,
int nfa,const t_iatom forceatoms[],const t_iparams ip[],
const rvec x[],const t_pbc *pbc,
t_fcdata *fcd,history_t *hist)
{
atom_id ai,aj;
int fa,res,i,pair,ki,kj,m;
int type,npair,np;
rvec dx;
real *rt,*rm3tav,*Rtl_6,*Rt_6,*Rtav_6;
real rt_1,rt_3,rt2;
ivec it,jt,dt;
t_disresdata *dd;
real ETerm,ETerm1,cf1=0,cf2=0,invn=0;
gmx_bool bTav;
dd = &(fcd->disres);
bTav = (dd->dr_tau != 0);
ETerm = dd->ETerm;
ETerm1 = dd->ETerm1;
rt = dd->rt;
rm3tav = dd->rm3tav;
Rtl_6 = dd->Rtl_6;
Rt_6 = dd->Rt_6;
Rtav_6 = dd->Rtav_6;
if (bTav)
{
/* scaling factor to smoothly turn on the restraint forces *
* when using time averaging */
dd->exp_min_t_tau = hist->disre_initf*ETerm;
cf1 = dd->exp_min_t_tau;
cf2 = 1.0/(1.0 - dd->exp_min_t_tau);
}
if (dd->nsystems > 1)
{
invn = 1.0/dd->nsystems;
}
/* 'loop' over all atom pairs (pair_nr=fa/3) involved in restraints, *
* the total number of atoms pairs is nfa/3 */
res = 0;
fa = 0;
while (fa < nfa)
{
type = forceatoms[fa];
npair = ip[type].disres.npair;
Rtav_6[res] = 0.0;
Rt_6[res] = 0.0;
/* Loop over the atom pairs of 'this' restraint */
np = 0;
while (fa < nfa && np < npair)
{
pair = fa/3;
ai = forceatoms[fa+1];
aj = forceatoms[fa+2];
if (pbc)
{
pbc_dx_aiuc(pbc,x[ai],x[aj],dx);
}
else
{
rvec_sub(x[ai],x[aj],dx);
}
rt2 = iprod(dx,dx);
rt_1 = gmx_invsqrt(rt2);
rt_3 = rt_1*rt_1*rt_1;
rt[pair] = sqrt(rt2);
if (bTav)
{
/* Here we update rm3tav in t_fcdata using the data
* in history_t.
* Thus the results stay correct when this routine
* is called multiple times.
*/
rm3tav[pair] = cf2*((ETerm - cf1)*hist->disre_rm3tav[pair] +
ETerm1*rt_3);
}
else
{
rm3tav[pair] = rt_3;
}
Rt_6[res] += rt_3*rt_3;
Rtav_6[res] += rm3tav[pair]*rm3tav[pair];
fa += 3;
np++;
}
if (dd->nsystems > 1)
{
Rtl_6[res] = Rt_6[res];
Rt_6[res] *= invn;
Rtav_6[res] *= invn;
}
res++;
}
#ifdef GMX_MPI
if (dd->nsystems > 1)
{
gmx_sum_comm(2*dd->nres,Rt_6,dd->mpi_comm_ensemble);
}
#endif
}
/*
* Gromacs nonbonded kernel nb_kernel410
* Coulomb interaction: Generalized-Born
* VdW interaction: Lennard-Jones
* water optimization: No
* Calculate forces: yes
*/
void nb_kernel410(
int * p_nri,
int * iinr,
int * jindex,
int * jjnr,
int * shift,
real * shiftvec,
real * fshift,
int * gid,
real * pos,
real * faction,
real * charge,
real * p_facel,
real * p_krf,
real * p_crf,
real * Vc,
int * type,
int * p_ntype,
real * vdwparam,
real * Vvdw,
real * p_tabscale,
real * VFtab,
real * invsqrta,
real * dvda,
real * p_gbtabscale,
real * GBtab,
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
real * work)
{
int nri,ntype,nthreads;
real facel,krf,crf,tabscale,gbtabscale;
int n,ii,is3,ii3,k,nj0,nj1,jnr,j3,ggid;
int nn0,nn1,nouter,ninner;
real shX,shY,shZ;
real fscal,tx,ty,tz;
real rinvsq;
real iq;
real qq,vcoul,vctot;
int nti;
int tj;
real rinvsix;
real Vvdw6,Vvdwtot;
real Vvdw12;
real r,rt,eps,eps2;
int n0,nnn;
real Y,F,Geps,Heps2,Fp,VV;
real FF;
real fijC;
real isai,isaj,isaprod,gbscale,vgb,vgbtot;
real dvdasum,dvdatmp,dvdaj,fgb;
real ix1,iy1,iz1,fix1,fiy1,fiz1;
real jx1,jy1,jz1;
real dx11,dy11,dz11,rsq11,rinv11;
real c6,c12;
gmx_gbdata_t *gbdata;
real * gpol;
real scale_gb;
gbdata = (gmx_gbdata_t *)work;
gpol = gbdata->gpol;
nri = *p_nri;
ntype = *p_ntype;
nthreads = *p_nthreads;
facel = *p_facel;
scale_gb = (1.0/gbdata->epsilon_r) - (1.0/gbdata->gb_epsilon_solvent);
krf = *p_krf;
crf = *p_crf;
tabscale = *p_tabscale;
gbtabscale = *p_gbtabscale;
/* Reset outer and inner iteration counters */
nouter = 0;
ninner = 0;
/* Loop over thread workunits */
do
{
#ifdef GMX_THREAD_SHM_FDECOMP
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
/* Take successively smaller chunks (at least 10 lists) */
nn1 = nn0+(nri-nn0)/(2*nthreads)+10;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
#else
nn0 = 0;
nn1 = nri;
#endif
/* Start outer loop over neighborlists */
for(n=nn0; (n<nn1); n++)
{
/* Load shift vector for this list */
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
/* Load limits for loop over neighbors */
nj0 = jindex[n];
nj1 = jindex[n+1];
/* Get outer coordinate index */
ii = iinr[n];
ii3 = 3*ii;
/* Load i atom data, add shift vector */
ix1 = shX + pos[ii3+0];
iy1 = shY + pos[ii3+1];
iz1 = shZ + pos[ii3+2];
/* Load parameters for i atom */
iq = facel*charge[ii];
isai = invsqrta[ii];
nti = 2*ntype*type[ii];
/* Zero the potential energy for this list */
vctot = 0;
Vvdwtot = 0;
vgbtot = 0;
dvdasum = 0;
/* Clear i atom forces */
fix1 = 0;
fiy1 = 0;
fiz1 = 0;
for(k=nj0; (k<nj1); k++)
{
/* Get j neighbor index, and coordinate index */
jnr = jjnr[k];
j3 = 3*jnr;
/* load j atom coordinates */
jx1 = pos[j3+0];
jy1 = pos[j3+1];
jz1 = pos[j3+2];
/* Calculate distance */
dx11 = ix1 - jx1;
dy11 = iy1 - jy1;
dz11 = iz1 - jz1;
rsq11 = dx11*dx11+dy11*dy11+dz11*dz11;
/* Calculate 1/r and 1/r2 */
rinv11 = gmx_invsqrt(rsq11);
/* Load parameters for j atom */
isaj = invsqrta[jnr];
isaprod = isai*isaj;
qq = iq*charge[jnr];
vcoul = qq*rinv11;
fscal = vcoul*rinv11;
qq = isaprod*(-qq)*scale_gb;
gbscale = isaprod*gbtabscale;
tj = nti+2*type[jnr];
c6 = vdwparam[tj];
c12 = vdwparam[tj+1];
rinvsq = rinv11*rinv11;
/* Tabulated Generalized-Born interaction */
dvdaj = dvda[jnr];
r = rsq11*rinv11;
/* Calculate table index */
rt = r*gbscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
Y = GBtab[nnn];
F = GBtab[nnn+1];
Geps = eps*GBtab[nnn+2];
Heps2 = eps2*GBtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vgb = qq*VV;
fijC = qq*FF*gbscale;
dvdatmp = -0.5*(vgb+fijC*r);
dvdasum = dvdasum + dvdatmp;
dvda[jnr] = dvdaj+dvdatmp*isaj*isaj;
vctot = vctot + vcoul;
vgbtot = vgbtot + vgb;
/* Lennard-Jones interaction */
rinvsix = rinvsq*rinvsq*rinvsq;
Vvdw6 = c6*rinvsix;
Vvdw12 = c12*rinvsix*rinvsix;
Vvdwtot = Vvdwtot+Vvdw12-Vvdw6;
fscal = (12.0*Vvdw12-6.0*Vvdw6)*rinvsq-(fijC-fscal)*rinv11;
/* Calculate temporary vectorial force */
tx = fscal*dx11;
ty = fscal*dy11;
tz = fscal*dz11;
/* Increment i atom force */
fix1 = fix1 + tx;
fiy1 = fiy1 + ty;
fiz1 = fiz1 + tz;
/* Decrement j atom force */
faction[j3+0] = faction[j3+0] - tx;
faction[j3+1] = faction[j3+1] - ty;
faction[j3+2] = faction[j3+2] - tz;
/* Inner loop uses 62 flops/iteration */
}
/* Add i forces to mem and shifted force list */
faction[ii3+0] = faction[ii3+0] + fix1;
faction[ii3+1] = faction[ii3+1] + fiy1;
faction[ii3+2] = faction[ii3+2] + fiz1;
fshift[is3] = fshift[is3]+fix1;
fshift[is3+1] = fshift[is3+1]+fiy1;
fshift[is3+2] = fshift[is3+2]+fiz1;
/* Add potential energies to the group for this list */
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
gpol[ggid] = gpol[ggid] + vgbtot;
Vvdw[ggid] = Vvdw[ggid] + Vvdwtot;
dvda[ii] = dvda[ii] + dvdasum*isai*isai;
/* Increment number of inner iterations */
ninner = ninner + nj1 - nj0;
/* Outer loop uses 13 flops/iteration */
}
/* Increment number of outer iterations */
nouter = nouter + nn1 - nn0;
}
while (nn1<nri);
/* Write outer/inner iteration count to pointers */
*outeriter = nouter;
*inneriter = ninner;
}
/*
* Gromacs nonbonded kernel pf_nb_kernel302nf
* Coulomb interaction: Tabulated
* VdW interaction: Not calculated
* water optimization: pairs of SPC/TIP3P interactions
* Calculate forces: no
*/
void pf_nb_kernel302nf(
int * p_nri,
int * iinr,
int * jindex,
int * jjnr,
int * shift,
real * shiftvec,
real * fshift,
int * gid,
real * pos,
real * faction,
real * charge,
real * p_facel,
real * p_krf,
real * p_crf,
real * Vc,
int * type,
int * p_ntype,
real * vdwparam,
real * Vvdw,
real * p_tabscale,
real * VFtab,
real * invsqrta,
real * dvda,
real * p_gbtabscale,
real * GBtab,
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
real * work,
t_pf_global * pf_global)
{
int nri,ntype,nthreads;
real facel,krf,crf,tabscale,gbtabscale;
int n,ii,is3,ii3,k,nj0,nj1,jnr,j3,ggid;
int nn0,nn1,nouter,ninner;
real shX,shY,shZ;
real qq,vcoul,vctot;
real r,rt,eps,eps2;
int n0,nnn;
real Y,F,Geps,Heps2,Fp,VV;
real ix1,iy1,iz1;
real ix2,iy2,iz2;
real ix3,iy3,iz3;
real jx1,jy1,jz1;
real jx2,jy2,jz2;
real jx3,jy3,jz3;
real dx11,dy11,dz11,rsq11,rinv11;
real dx12,dy12,dz12,rsq12,rinv12;
real dx13,dy13,dz13,rsq13,rinv13;
real dx21,dy21,dz21,rsq21,rinv21;
real dx22,dy22,dz22,rsq22,rinv22;
real dx23,dy23,dz23,rsq23,rinv23;
real dx31,dy31,dz31,rsq31,rinv31;
real dx32,dy32,dz32,rsq32,rinv32;
real dx33,dy33,dz33,rsq33,rinv33;
real qO,qH,qqOO,qqOH,qqHH;
nri = *p_nri;
ntype = *p_ntype;
nthreads = *p_nthreads;
facel = *p_facel;
krf = *p_krf;
crf = *p_crf;
tabscale = *p_tabscale;
/* Initialize water data */
ii = iinr[0];
qO = charge[ii];
qH = charge[ii+1];
qqOO = facel*qO*qO;
qqOH = facel*qO*qH;
qqHH = facel*qH*qH;
/* Reset outer and inner iteration counters */
nouter = 0;
ninner = 0;
/* Loop over thread workunits */
do
{
#ifdef GMX_THREAD_SHM_FDECOMP
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
/* Take successively smaller chunks (at least 10 lists) */
nn1 = nn0+(nri-nn0)/(2*nthreads)+10;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
#else
nn0 = 0;
nn1 = nri;
#endif
/* Start outer loop over neighborlists */
for(n=nn0; (n<nn1); n++)
{
/* Load shift vector for this list */
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
/* Load limits for loop over neighbors */
nj0 = jindex[n];
nj1 = jindex[n+1];
/* Get outer coordinate index */
ii = iinr[n];
ii3 = 3*ii;
/* Load i atom data, add shift vector */
ix1 = shX + pos[ii3+0];
iy1 = shY + pos[ii3+1];
iz1 = shZ + pos[ii3+2];
ix2 = shX + pos[ii3+3];
iy2 = shY + pos[ii3+4];
iz2 = shZ + pos[ii3+5];
ix3 = shX + pos[ii3+6];
iy3 = shY + pos[ii3+7];
iz3 = shZ + pos[ii3+8];
/* Zero the potential energy for this list */
vctot = 0;
/* Clear i atom forces */
for(k=nj0; (k<nj1); k++)
{
/* Get j neighbor index, and coordinate index */
jnr = jjnr[k];
j3 = 3*jnr;
/* load j atom coordinates */
jx1 = pos[j3+0];
jy1 = pos[j3+1];
jz1 = pos[j3+2];
jx2 = pos[j3+3];
jy2 = pos[j3+4];
jz2 = pos[j3+5];
jx3 = pos[j3+6];
jy3 = pos[j3+7];
jz3 = pos[j3+8];
/* Calculate distance */
dx11 = ix1 - jx1;
dy11 = iy1 - jy1;
dz11 = iz1 - jz1;
rsq11 = dx11*dx11+dy11*dy11+dz11*dz11;
dx12 = ix1 - jx2;
dy12 = iy1 - jy2;
dz12 = iz1 - jz2;
rsq12 = dx12*dx12+dy12*dy12+dz12*dz12;
dx13 = ix1 - jx3;
dy13 = iy1 - jy3;
dz13 = iz1 - jz3;
rsq13 = dx13*dx13+dy13*dy13+dz13*dz13;
dx21 = ix2 - jx1;
dy21 = iy2 - jy1;
dz21 = iz2 - jz1;
rsq21 = dx21*dx21+dy21*dy21+dz21*dz21;
dx22 = ix2 - jx2;
dy22 = iy2 - jy2;
dz22 = iz2 - jz2;
rsq22 = dx22*dx22+dy22*dy22+dz22*dz22;
dx23 = ix2 - jx3;
dy23 = iy2 - jy3;
dz23 = iz2 - jz3;
rsq23 = dx23*dx23+dy23*dy23+dz23*dz23;
dx31 = ix3 - jx1;
dy31 = iy3 - jy1;
dz31 = iz3 - jz1;
rsq31 = dx31*dx31+dy31*dy31+dz31*dz31;
dx32 = ix3 - jx2;
dy32 = iy3 - jy2;
dz32 = iz3 - jz2;
rsq32 = dx32*dx32+dy32*dy32+dz32*dz32;
dx33 = ix3 - jx3;
dy33 = iy3 - jy3;
dz33 = iz3 - jz3;
rsq33 = dx33*dx33+dy33*dy33+dz33*dz33;
/* Calculate 1/r and 1/r2 */
rinv11 = gmx_invsqrt(rsq11);
rinv12 = gmx_invsqrt(rsq12);
rinv13 = gmx_invsqrt(rsq13);
rinv21 = gmx_invsqrt(rsq21);
rinv22 = gmx_invsqrt(rsq22);
rinv23 = gmx_invsqrt(rsq23);
rinv31 = gmx_invsqrt(rsq31);
rinv32 = gmx_invsqrt(rsq32);
rinv33 = gmx_invsqrt(rsq33);
/* Load parameters for j atom */
qq = qqOO;
/* Calculate table index */
r = rsq11*rinv11;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
vcoul = qq*VV;
vctot = vctot + vcoul;
/* Load parameters for j atom */
qq = qqOH;
/* Calculate table index */
r = rsq12*rinv12;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
vcoul = qq*VV;
vctot = vctot + vcoul;
/* Load parameters for j atom */
qq = qqOH;
/* Calculate table index */
r = rsq13*rinv13;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
vcoul = qq*VV;
vctot = vctot + vcoul;
/* Load parameters for j atom */
qq = qqOH;
/* Calculate table index */
r = rsq21*rinv21;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
vcoul = qq*VV;
vctot = vctot + vcoul;
/* Load parameters for j atom */
qq = qqHH;
/* Calculate table index */
r = rsq22*rinv22;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
vcoul = qq*VV;
vctot = vctot + vcoul;
/* Load parameters for j atom */
qq = qqHH;
/* Calculate table index */
r = rsq23*rinv23;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
vcoul = qq*VV;
vctot = vctot + vcoul;
/* Load parameters for j atom */
qq = qqOH;
/* Calculate table index */
r = rsq31*rinv31;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
vcoul = qq*VV;
vctot = vctot + vcoul;
/* Load parameters for j atom */
qq = qqHH;
/* Calculate table index */
r = rsq32*rinv32;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
vcoul = qq*VV;
vctot = vctot + vcoul;
/* Load parameters for j atom */
qq = qqHH;
/* Calculate table index */
r = rsq33*rinv33;
/* Calculate table index */
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = 4*n0;
/* Tabulated coulomb interaction */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
vcoul = qq*VV;
vctot = vctot + vcoul;
/* Inner loop uses 225 flops/iteration */
}
/* Add i forces to mem and shifted force list */
/* Add potential energies to the group for this list */
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
/* Increment number of inner iterations */
ninner = ninner + nj1 - nj0;
/* Outer loop uses 10 flops/iteration */
}
/* Increment number of outer iterations */
nouter = nouter + nn1 - nn0;
}
while (nn1<nri);
/* Write outer/inner iteration count to pointers */
*outeriter = nouter;
*inneriter = ninner;
}
/*
* Gromacs nonbonded kernel nb_kernel213nf
* Coulomb interaction: Reaction field
* VdW interaction: Lennard-Jones
* water optimization: TIP4P - other atoms
* Calculate forces: no
*/
void nb_kernel213nf(
int * p_nri,
int * iinr,
int * jindex,
int * jjnr,
int * shift,
real * shiftvec,
real * fshift,
int * gid,
real * pos,
real * faction,
real * charge,
real * p_facel,
real * p_krf,
real * p_crf,
real * Vc,
int * type,
int * p_ntype,
real * vdwparam,
real * Vvdw,
real * p_tabscale,
real * VFtab,
real * invsqrta,
real * dvda,
real * p_gbtabscale,
real * GBtab,
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
real * work)
{
int nri,ntype,nthreads;
real facel,krf,crf,tabscale,gbtabscale;
int n,ii,is3,ii3,k,nj0,nj1,jnr,j3,ggid;
int nn0,nn1,nouter,ninner;
real shX,shY,shZ;
real rinvsq;
real jq;
real qq,vcoul,vctot;
int nti;
int tj;
real rinvsix;
real Vvdw6,Vvdwtot;
real Vvdw12;
real krsq;
real ix1,iy1,iz1;
real ix2,iy2,iz2;
real ix3,iy3,iz3;
real ix4,iy4,iz4;
real jx1,jy1,jz1;
real dx11,dy11,dz11,rsq11;
real dx21,dy21,dz21,rsq21,rinv21;
real dx31,dy31,dz31,rsq31,rinv31;
real dx41,dy41,dz41,rsq41,rinv41;
real qH,qM;
real c6,c12;
nri = *p_nri;
ntype = *p_ntype;
nthreads = *p_nthreads;
facel = *p_facel;
krf = *p_krf;
crf = *p_crf;
tabscale = *p_tabscale;
/* Initialize water data */
ii = iinr[0];
qH = facel*charge[ii+1];
qM = facel*charge[ii+3];
nti = 2*ntype*type[ii];
/* Reset outer and inner iteration counters */
nouter = 0;
ninner = 0;
/* Loop over thread workunits */
do
{
#ifdef GMX_THREAD_SHM_FDECOMP
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
/* Take successively smaller chunks (at least 10 lists) */
nn1 = nn0+(nri-nn0)/(2*nthreads)+10;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
#else
nn0 = 0;
nn1 = nri;
#endif
/* Start outer loop over neighborlists */
for(n=nn0; (n<nn1); n++)
{
/* Load shift vector for this list */
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
/* Load limits for loop over neighbors */
nj0 = jindex[n];
nj1 = jindex[n+1];
/* Get outer coordinate index */
ii = iinr[n];
ii3 = 3*ii;
/* Load i atom data, add shift vector */
ix1 = shX + pos[ii3+0];
iy1 = shY + pos[ii3+1];
iz1 = shZ + pos[ii3+2];
ix2 = shX + pos[ii3+3];
iy2 = shY + pos[ii3+4];
iz2 = shZ + pos[ii3+5];
ix3 = shX + pos[ii3+6];
iy3 = shY + pos[ii3+7];
iz3 = shZ + pos[ii3+8];
ix4 = shX + pos[ii3+9];
iy4 = shY + pos[ii3+10];
iz4 = shZ + pos[ii3+11];
/* Zero the potential energy for this list */
vctot = 0;
Vvdwtot = 0;
/* Clear i atom forces */
for(k=nj0; (k<nj1); k++)
{
/* Get j neighbor index, and coordinate index */
jnr = jjnr[k];
j3 = 3*jnr;
/* load j atom coordinates */
jx1 = pos[j3+0];
jy1 = pos[j3+1];
jz1 = pos[j3+2];
/* Calculate distance */
dx11 = ix1 - jx1;
dy11 = iy1 - jy1;
dz11 = iz1 - jz1;
rsq11 = dx11*dx11+dy11*dy11+dz11*dz11;
dx21 = ix2 - jx1;
dy21 = iy2 - jy1;
dz21 = iz2 - jz1;
rsq21 = dx21*dx21+dy21*dy21+dz21*dz21;
dx31 = ix3 - jx1;
dy31 = iy3 - jy1;
dz31 = iz3 - jz1;
rsq31 = dx31*dx31+dy31*dy31+dz31*dz31;
dx41 = ix4 - jx1;
dy41 = iy4 - jy1;
dz41 = iz4 - jz1;
rsq41 = dx41*dx41+dy41*dy41+dz41*dz41;
/* Calculate 1/r and 1/r2 */
rinvsq = 1.0/rsq11;
rinv21 = gmx_invsqrt(rsq21);
rinv31 = gmx_invsqrt(rsq31);
rinv41 = gmx_invsqrt(rsq41);
/* Load parameters for j atom */
tj = nti+2*type[jnr];
c6 = vdwparam[tj];
c12 = vdwparam[tj+1];
/* Lennard-Jones interaction */
rinvsix = rinvsq*rinvsq*rinvsq;
Vvdw6 = c6*rinvsix;
Vvdw12 = c12*rinvsix*rinvsix;
Vvdwtot = Vvdwtot+Vvdw12-Vvdw6;
/* Load parameters for j atom */
jq = charge[jnr+0];
qq = qH*jq;
/* Coulomb reaction-field interaction */
krsq = krf*rsq21;
vcoul = qq*(rinv21+krsq-crf);
vctot = vctot+vcoul;
/* Load parameters for j atom */
/* Coulomb reaction-field interaction */
krsq = krf*rsq31;
vcoul = qq*(rinv31+krsq-crf);
vctot = vctot+vcoul;
/* Load parameters for j atom */
qq = qM*jq;
/* Coulomb reaction-field interaction */
krsq = krf*rsq41;
vcoul = qq*(rinv41+krsq-crf);
vctot = vctot+vcoul;
/* Inner loop uses 75 flops/iteration */
}
/* Add i forces to mem and shifted force list */
/* Add potential energies to the group for this list */
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
Vvdw[ggid] = Vvdw[ggid] + Vvdwtot;
/* Increment number of inner iterations */
ninner = ninner + nj1 - nj0;
/* Outer loop uses 14 flops/iteration */
}
/* Increment number of outer iterations */
nouter = nouter + nn1 - nn0;
}
while (nn1<nri);
/* Write outer/inner iteration count to pointers */
*outeriter = nouter;
*inneriter = ninner;
}
/*
* Gromacs nonbonded kernel pf_nb_kernel104nf
* Coulomb interaction: Normal Coulomb
* VdW interaction: Not calculated
* water optimization: pairs of TIP4P interactions
* Calculate forces: no
*/
void pf_nb_kernel104nf(
int * p_nri,
int * iinr,
int * jindex,
int * jjnr,
int * shift,
real * shiftvec,
real * fshift,
int * gid,
real * pos,
real * faction,
real * charge,
real * p_facel,
real * p_krf,
real * p_crf,
real * Vc,
int * type,
int * p_ntype,
real * vdwparam,
real * Vvdw,
real * p_tabscale,
real * VFtab,
real * invsqrta,
real * dvda,
real * p_gbtabscale,
real * GBtab,
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
real * work,
t_pf_global * pf_global)
{
int nri,ntype,nthreads;
real facel,krf,crf,tabscale,gbtabscale;
int n,ii,is3,ii3,k,nj0,nj1,jnr,j3,ggid;
int nn0,nn1,nouter,ninner;
real shX,shY,shZ;
real qq,vcoul,vctot;
real ix2,iy2,iz2;
real ix3,iy3,iz3;
real ix4,iy4,iz4;
real jx2,jy2,jz2;
real jx3,jy3,jz3;
real jx4,jy4,jz4;
real dx22,dy22,dz22,rsq22,rinv22;
real dx23,dy23,dz23,rsq23,rinv23;
real dx24,dy24,dz24,rsq24,rinv24;
real dx32,dy32,dz32,rsq32,rinv32;
real dx33,dy33,dz33,rsq33,rinv33;
real dx34,dy34,dz34,rsq34,rinv34;
real dx42,dy42,dz42,rsq42,rinv42;
real dx43,dy43,dz43,rsq43,rinv43;
real dx44,dy44,dz44,rsq44,rinv44;
real qH,qM,qqMM,qqMH,qqHH;
nri = *p_nri;
ntype = *p_ntype;
nthreads = *p_nthreads;
facel = *p_facel;
krf = *p_krf;
crf = *p_crf;
tabscale = *p_tabscale;
/* Initialize water data */
ii = iinr[0];
qH = charge[ii+1];
qM = charge[ii+3];
qqMM = facel*qM*qM;
qqMH = facel*qM*qH;
qqHH = facel*qH*qH;
/* Reset outer and inner iteration counters */
nouter = 0;
ninner = 0;
/* Loop over thread workunits */
do
{
#ifdef GMX_THREAD_SHM_FDECOMP
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
/* Take successively smaller chunks (at least 10 lists) */
nn1 = nn0+(nri-nn0)/(2*nthreads)+10;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
#else
nn0 = 0;
nn1 = nri;
#endif
/* Start outer loop over neighborlists */
for(n=nn0; (n<nn1); n++)
{
/* Load shift vector for this list */
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
/* Load limits for loop over neighbors */
nj0 = jindex[n];
nj1 = jindex[n+1];
/* Get outer coordinate index */
ii = iinr[n];
ii3 = 3*ii;
/* Load i atom data, add shift vector */
ix2 = shX + pos[ii3+3];
iy2 = shY + pos[ii3+4];
iz2 = shZ + pos[ii3+5];
ix3 = shX + pos[ii3+6];
iy3 = shY + pos[ii3+7];
iz3 = shZ + pos[ii3+8];
ix4 = shX + pos[ii3+9];
iy4 = shY + pos[ii3+10];
iz4 = shZ + pos[ii3+11];
/* Zero the potential energy for this list */
vctot = 0;
/* Clear i atom forces */
for(k=nj0; (k<nj1); k++)
{
/* Get j neighbor index, and coordinate index */
jnr = jjnr[k];
j3 = 3*jnr;
/* load j atom coordinates */
jx2 = pos[j3+3];
jy2 = pos[j3+4];
jz2 = pos[j3+5];
jx3 = pos[j3+6];
jy3 = pos[j3+7];
jz3 = pos[j3+8];
jx4 = pos[j3+9];
jy4 = pos[j3+10];
jz4 = pos[j3+11];
/* Calculate distance */
dx22 = ix2 - jx2;
dy22 = iy2 - jy2;
dz22 = iz2 - jz2;
rsq22 = dx22*dx22+dy22*dy22+dz22*dz22;
dx23 = ix2 - jx3;
dy23 = iy2 - jy3;
dz23 = iz2 - jz3;
rsq23 = dx23*dx23+dy23*dy23+dz23*dz23;
dx24 = ix2 - jx4;
dy24 = iy2 - jy4;
dz24 = iz2 - jz4;
rsq24 = dx24*dx24+dy24*dy24+dz24*dz24;
dx32 = ix3 - jx2;
dy32 = iy3 - jy2;
dz32 = iz3 - jz2;
rsq32 = dx32*dx32+dy32*dy32+dz32*dz32;
dx33 = ix3 - jx3;
dy33 = iy3 - jy3;
dz33 = iz3 - jz3;
rsq33 = dx33*dx33+dy33*dy33+dz33*dz33;
dx34 = ix3 - jx4;
dy34 = iy3 - jy4;
dz34 = iz3 - jz4;
rsq34 = dx34*dx34+dy34*dy34+dz34*dz34;
dx42 = ix4 - jx2;
dy42 = iy4 - jy2;
dz42 = iz4 - jz2;
rsq42 = dx42*dx42+dy42*dy42+dz42*dz42;
dx43 = ix4 - jx3;
dy43 = iy4 - jy3;
dz43 = iz4 - jz3;
rsq43 = dx43*dx43+dy43*dy43+dz43*dz43;
dx44 = ix4 - jx4;
dy44 = iy4 - jy4;
dz44 = iz4 - jz4;
rsq44 = dx44*dx44+dy44*dy44+dz44*dz44;
/* Calculate 1/r and 1/r2 */
rinv22 = gmx_invsqrt(rsq22);
rinv23 = gmx_invsqrt(rsq23);
rinv24 = gmx_invsqrt(rsq24);
rinv32 = gmx_invsqrt(rsq32);
rinv33 = gmx_invsqrt(rsq33);
rinv34 = gmx_invsqrt(rsq34);
rinv42 = gmx_invsqrt(rsq42);
rinv43 = gmx_invsqrt(rsq43);
rinv44 = gmx_invsqrt(rsq44);
/* Load parameters for j atom */
qq = qqHH;
/* Coulomb interaction */
vcoul = qq*rinv22;
vctot = vctot+vcoul;
/* Load parameters for j atom */
qq = qqHH;
/* Coulomb interaction */
vcoul = qq*rinv23;
vctot = vctot+vcoul;
/* Load parameters for j atom */
qq = qqMH;
/* Coulomb interaction */
vcoul = qq*rinv24;
vctot = vctot+vcoul;
/* Load parameters for j atom */
qq = qqHH;
/* Coulomb interaction */
vcoul = qq*rinv32;
vctot = vctot+vcoul;
/* Load parameters for j atom */
qq = qqHH;
/* Coulomb interaction */
vcoul = qq*rinv33;
vctot = vctot+vcoul;
/* Load parameters for j atom */
qq = qqMH;
/* Coulomb interaction */
vcoul = qq*rinv34;
vctot = vctot+vcoul;
/* Load parameters for j atom */
qq = qqMH;
/* Coulomb interaction */
vcoul = qq*rinv42;
vctot = vctot+vcoul;
/* Load parameters for j atom */
qq = qqMH;
/* Coulomb interaction */
vcoul = qq*rinv43;
vctot = vctot+vcoul;
/* Load parameters for j atom */
qq = qqMM;
/* Coulomb interaction */
vcoul = qq*rinv44;
vctot = vctot+vcoul;
/* Inner loop uses 135 flops/iteration */
}
/* Add i forces to mem and shifted force list */
/* Add potential energies to the group for this list */
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
/* Increment number of inner iterations */
ninner = ninner + nj1 - nj0;
/* Outer loop uses 10 flops/iteration */
}
/* Increment number of outer iterations */
nouter = nouter + nn1 - nn0;
}
while (nn1<nri);
/* Write outer/inner iteration count to pointers */
*outeriter = nouter;
*inneriter = ninner;
}
void
gmx_nb_generic_adress_kernel(t_nblist * nlist,
rvec * xx,
rvec * ff,
t_forcerec * fr,
t_mdatoms * mdatoms,
nb_kernel_data_t * kernel_data,
t_nrnb * nrnb)
{
int nri, ntype, table_nelements, ielec, ivdw;
real facel, gbtabscale;
int n, ii, is3, ii3, k, nj0, nj1, jnr, j3, ggid, nnn, n0;
real shX, shY, shZ;
real fscal, felec, fvdw, velec, vvdw, tx, ty, tz;
real rinvsq;
real iq;
real qq, vctot;
int nti, nvdwparam;
int tj;
real rt, r, eps, eps2, Y, F, Geps, Heps2, VV, FF, Fp, fijD, fijR;
real rinvsix;
real vvdwtot;
real vvdw_rep, vvdw_disp;
real ix, iy, iz, fix, fiy, fiz;
real jx, jy, jz;
real dx, dy, dz, rsq, rinv;
real c6, c12, cexp1, cexp2, br;
real * charge;
real * shiftvec;
real * vdwparam;
int * shift;
int * type;
real * fshift;
real * velecgrp;
real * vvdwgrp;
real tabscale;
real * VFtab;
real * x;
real * f;
int ewitab;
real ewtabscale, eweps, sh_ewald, ewrt, ewtabhalfspace;
real * ewtab;
real rcoulomb2, rvdw, rvdw2, sh_dispersion, sh_repulsion;
real rcutoff, rcutoff2;
real rswitch_elec, rswitch_vdw, d, d2, sw, dsw, rinvcorr;
real elec_swV3, elec_swV4, elec_swV5, elec_swF2, elec_swF3, elec_swF4;
real vdw_swV3, vdw_swV4, vdw_swV5, vdw_swF2, vdw_swF3, vdw_swF4;
gmx_bool bExactElecCutoff, bExactVdwCutoff, bExactCutoff;
real * wf;
real weight_cg1;
real weight_cg2;
real weight_product;
real hybscal; /* the multiplicator to the force for hybrid interactions*/
real force_cap;
gmx_bool bCG;
int egp_nr;
wf = mdatoms->wf;
force_cap = fr->adress_ex_forcecap;
x = xx[0];
f = ff[0];
ielec = nlist->ielec;
ivdw = nlist->ivdw;
fshift = fr->fshift[0];
velecgrp = kernel_data->energygrp_elec;
vvdwgrp = kernel_data->energygrp_vdw;
tabscale = kernel_data->table_elec_vdw->scale;
VFtab = kernel_data->table_elec_vdw->data;
sh_ewald = fr->ic->sh_ewald;
ewtab = fr->ic->tabq_coul_FDV0;
ewtabscale = fr->ic->tabq_scale;
ewtabhalfspace = 0.5/ewtabscale;
rcoulomb2 = fr->rcoulomb*fr->rcoulomb;
rvdw = fr->rvdw;
rvdw2 = rvdw*rvdw;
sh_dispersion = fr->ic->dispersion_shift.cpot;
sh_repulsion = fr->ic->repulsion_shift.cpot;
if (fr->coulomb_modifier == eintmodPOTSWITCH)
{
d = fr->rcoulomb-fr->rcoulomb_switch;
elec_swV3 = -10.0/(d*d*d);
elec_swV4 = 15.0/(d*d*d*d);
elec_swV5 = -6.0/(d*d*d*d*d);
elec_swF2 = -30.0/(d*d*d);
elec_swF3 = 60.0/(d*d*d*d);
elec_swF4 = -30.0/(d*d*d*d*d);
}
else
{
/* Avoid warnings from stupid compilers (looking at you, Clang!) */
elec_swV3 = elec_swV4 = elec_swV5 = elec_swF2 = elec_swF3 = elec_swF4 = 0.0;
}
if (fr->vdw_modifier == eintmodPOTSWITCH)
{
d = fr->rvdw-fr->rvdw_switch;
vdw_swV3 = -10.0/(d*d*d);
vdw_swV4 = 15.0/(d*d*d*d);
vdw_swV5 = -6.0/(d*d*d*d*d);
vdw_swF2 = -30.0/(d*d*d);
vdw_swF3 = 60.0/(d*d*d*d);
vdw_swF4 = -30.0/(d*d*d*d*d);
}
else
{
/* Avoid warnings from stupid compilers (looking at you, Clang!) */
vdw_swV3 = vdw_swV4 = vdw_swV5 = vdw_swF2 = vdw_swF3 = vdw_swF4 = 0.0;
}
bExactElecCutoff = (fr->coulomb_modifier != eintmodNONE) || fr->eeltype == eelRF_ZERO;
bExactVdwCutoff = (fr->vdw_modifier != eintmodNONE);
bExactCutoff = bExactElecCutoff || bExactVdwCutoff;
if (bExactCutoff)
{
rcutoff = ( fr->rcoulomb > fr->rvdw ) ? fr->rcoulomb : fr->rvdw;
rcutoff2 = rcutoff*rcutoff;
}
else
{
/* Fix warnings for stupid compilers */
rcutoff = rcutoff2 = 1e30;
}
/* avoid compiler warnings for cases that cannot happen */
nnn = 0;
eps = 0.0;
eps2 = 0.0;
/* 3 VdW parameters for buckingham, otherwise 2 */
nvdwparam = (ivdw == GMX_NBKERNEL_VDW_BUCKINGHAM) ? 3 : 2;
table_nelements = 12;
charge = mdatoms->chargeA;
type = mdatoms->typeA;
facel = fr->epsfac;
shiftvec = fr->shift_vec[0];
vdwparam = fr->nbfp;
ntype = fr->ntype;
for (n = 0; (n < nlist->nri); n++)
{
is3 = 3*nlist->shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
nj0 = nlist->jindex[n];
nj1 = nlist->jindex[n+1];
ii = nlist->iinr[n];
ii3 = 3*ii;
ix = shX + x[ii3+0];
iy = shY + x[ii3+1];
iz = shZ + x[ii3+2];
iq = facel*charge[ii];
nti = nvdwparam*ntype*type[ii];
vctot = 0;
vvdwtot = 0;
fix = 0;
fiy = 0;
fiz = 0;
/* We need to find out if this i atom is part of an
all-atom or CG energy group */
egp_nr = mdatoms->cENER[ii];
bCG = !fr->adress_group_explicit[egp_nr];
weight_cg1 = wf[ii];
if ((!bCG) && weight_cg1 < ALMOST_ZERO)
{
continue;
}
for (k = nj0; (k < nj1); k++)
{
jnr = nlist->jjnr[k];
weight_cg2 = wf[jnr];
weight_product = weight_cg1*weight_cg2;
if (weight_product < ALMOST_ZERO)
{
/* if it's a explicit loop, skip this atom */
if (!bCG)
{
continue;
}
else /* if it's a coarse grained loop, include this atom */
{
hybscal = 1.0;
}
}
else if (weight_product >= ALMOST_ONE)
{
/* if it's a explicit loop, include this atom */
if (!bCG)
{
hybscal = 1.0;
}
else /* if it's a coarse grained loop, skip this atom */
{
continue;
}
}
/* both have double identity, get hybrid scaling factor */
else
{
hybscal = weight_product;
if (bCG)
{
hybscal = 1.0 - hybscal;
}
}
j3 = 3*jnr;
jx = x[j3+0];
jy = x[j3+1];
jz = x[j3+2];
dx = ix - jx;
dy = iy - jy;
dz = iz - jz;
rsq = dx*dx+dy*dy+dz*dz;
rinv = gmx_invsqrt(rsq);
rinvsq = rinv*rinv;
felec = 0;
fvdw = 0;
velec = 0;
vvdw = 0;
if (bExactCutoff && rsq > rcutoff2)
{
continue;
}
if (ielec == GMX_NBKERNEL_ELEC_CUBICSPLINETABLE || ivdw == GMX_NBKERNEL_VDW_CUBICSPLINETABLE)
{
r = rsq*rinv;
rt = r*tabscale;
n0 = rt;
eps = rt-n0;
eps2 = eps*eps;
nnn = table_nelements*n0;
}
/* Coulomb interaction. ielec==0 means no interaction */
if (ielec != GMX_NBKERNEL_ELEC_NONE)
{
qq = iq*charge[jnr];
switch (ielec)
{
case GMX_NBKERNEL_ELEC_NONE:
break;
case GMX_NBKERNEL_ELEC_COULOMB:
/* Vanilla cutoff coulomb */
velec = qq*rinv;
felec = velec*rinvsq;
break;
case GMX_NBKERNEL_ELEC_REACTIONFIELD:
/* Reaction-field */
velec = qq*(rinv+fr->k_rf*rsq-fr->c_rf);
felec = qq*(rinv*rinvsq-2.0*fr->k_rf);
break;
case GMX_NBKERNEL_ELEC_CUBICSPLINETABLE:
/* Tabulated coulomb */
Y = VFtab[nnn];
F = VFtab[nnn+1];
Geps = eps*VFtab[nnn+2];
Heps2 = eps2*VFtab[nnn+3];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
velec = qq*VV;
felec = -qq*FF*tabscale*rinv;
break;
case GMX_NBKERNEL_ELEC_GENERALIZEDBORN:
/* GB */
gmx_fatal(FARGS, "Death & horror! GB generic interaction not implemented.\n");
break;
case GMX_NBKERNEL_ELEC_EWALD:
ewrt = rsq*rinv*ewtabscale;
ewitab = ewrt;
eweps = ewrt-ewitab;
ewitab = 4*ewitab;
felec = ewtab[ewitab]+eweps*ewtab[ewitab+1];
rinvcorr = (fr->coulomb_modifier == eintmodPOTSHIFT) ? rinv-fr->ic->sh_ewald : rinv;
velec = qq*(rinvcorr-(ewtab[ewitab+2]-ewtabhalfspace*eweps*(ewtab[ewitab]+felec)));
felec = qq*rinv*(rinvsq-felec);
break;
default:
gmx_fatal(FARGS, "Death & horror! No generic coulomb interaction for ielec=%d.\n", ielec);
break;
}
if (fr->coulomb_modifier == eintmodPOTSWITCH)
{
d = rsq*rinv-fr->rcoulomb_switch;
d = (d > 0.0) ? d : 0.0;
d2 = d*d;
sw = 1.0+d2*d*(elec_swV3+d*(elec_swV4+d*elec_swV5));
dsw = d2*(elec_swF2+d*(elec_swF3+d*elec_swF4));
/* Apply switch function. Note that felec=f/r since it will be multiplied
* by the i-j displacement vector. This means felec'=f'/r=-(v*sw)'/r=
* -(v'*sw+v*dsw)/r=-v'*sw/r-v*dsw/r=felec*sw-v*dsw/r
*/
felec = felec*sw - rinv*velec*dsw;
/* Once we have used velec to update felec we can modify velec too */
velec *= sw;
}
if (bExactElecCutoff)
{
felec = (rsq <= rcoulomb2) ? felec : 0.0;
velec = (rsq <= rcoulomb2) ? velec : 0.0;
}
vctot += velec;
} /* End of coulomb interactions */
/* VdW interaction. ivdw==0 means no interaction */
if (ivdw != GMX_NBKERNEL_VDW_NONE)
{
tj = nti+nvdwparam*type[jnr];
switch (ivdw)
{
case GMX_NBKERNEL_VDW_NONE:
break;
case GMX_NBKERNEL_VDW_LENNARDJONES:
/* Vanilla Lennard-Jones cutoff */
c6 = vdwparam[tj];
c12 = vdwparam[tj+1];
rinvsix = rinvsq*rinvsq*rinvsq;
vvdw_disp = c6*rinvsix;
vvdw_rep = c12*rinvsix*rinvsix;
fvdw = (vvdw_rep-vvdw_disp)*rinvsq;
if (fr->vdw_modifier == eintmodPOTSHIFT)
{
vvdw = (vvdw_rep + c12*sh_repulsion)/12.0 - (vvdw_disp + c6*sh_dispersion)/6.0;
}
else
{
vvdw = vvdw_rep/12.0-vvdw_disp/6.0;
}
break;
case GMX_NBKERNEL_VDW_BUCKINGHAM:
/* Buckingham */
c6 = vdwparam[tj];
cexp1 = vdwparam[tj+1];
cexp2 = vdwparam[tj+2];
rinvsix = rinvsq*rinvsq*rinvsq;
vvdw_disp = c6*rinvsix;
br = cexp2*rsq*rinv;
vvdw_rep = cexp1*exp(-br);
fvdw = (br*vvdw_rep-vvdw_disp)*rinvsq;
if (fr->vdw_modifier == eintmodPOTSHIFT)
{
vvdw = (vvdw_rep-cexp1*exp(-cexp2*rvdw)) - (vvdw_disp + c6*sh_dispersion)/6.0;
}
else
{
vvdw = vvdw_rep-vvdw_disp/6.0;
}
break;
case GMX_NBKERNEL_VDW_CUBICSPLINETABLE:
/* Tabulated VdW */
c6 = vdwparam[tj];
c12 = vdwparam[tj+1];
Y = VFtab[nnn+4];
F = VFtab[nnn+5];
Geps = eps*VFtab[nnn+6];
Heps2 = eps2*VFtab[nnn+7];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vvdw_disp = c6*VV;
fijD = c6*FF;
Y = VFtab[nnn+8];
F = VFtab[nnn+9];
Geps = eps*VFtab[nnn+10];
Heps2 = eps2*VFtab[nnn+11];
Fp = F+Geps+Heps2;
VV = Y+eps*Fp;
FF = Fp+Geps+2.0*Heps2;
vvdw_rep = c12*VV;
fijR = c12*FF;
fvdw = -(fijD+fijR)*tabscale*rinv;
vvdw = vvdw_disp + vvdw_rep;
break;
default:
gmx_fatal(FARGS, "Death & horror! No generic VdW interaction for ivdw=%d.\n", ivdw);
break;
}
if (fr->vdw_modifier == eintmodPOTSWITCH)
{
d = rsq*rinv-fr->rvdw_switch;
d = (d > 0.0) ? d : 0.0;
d2 = d*d;
sw = 1.0+d2*d*(vdw_swV3+d*(vdw_swV4+d*vdw_swV5));
dsw = d2*(vdw_swF2+d*(vdw_swF3+d*vdw_swF4));
/* See coulomb interaction for the force-switch formula */
fvdw = fvdw*sw - rinv*vvdw*dsw;
vvdw *= sw;
}
if (bExactVdwCutoff)
{
fvdw = (rsq <= rvdw2) ? fvdw : 0.0;
vvdw = (rsq <= rvdw2) ? vvdw : 0.0;
}
vvdwtot += vvdw;
} /* end VdW interactions */
fscal = felec+fvdw;
if (!bCG && force_cap > 0 && (fabs(fscal) > force_cap))
{
fscal = force_cap*fscal/fabs(fscal);
}
fscal *= hybscal;
tx = fscal*dx;
ty = fscal*dy;
tz = fscal*dz;
fix = fix + tx;
fiy = fiy + ty;
fiz = fiz + tz;
f[j3+0] = f[j3+0] - tx;
f[j3+1] = f[j3+1] - ty;
f[j3+2] = f[j3+2] - tz;
}
f[ii3+0] = f[ii3+0] + fix;
f[ii3+1] = f[ii3+1] + fiy;
f[ii3+2] = f[ii3+2] + fiz;
fshift[is3] = fshift[is3]+fix;
fshift[is3+1] = fshift[is3+1]+fiy;
fshift[is3+2] = fshift[is3+2]+fiz;
ggid = nlist->gid[n];
velecgrp[ggid] += vctot;
vvdwgrp[ggid] += vvdwtot;
}
/* Estimate flops, average for generic adress kernel:
* 14 flops per outer iteration
* 54 flops per inner iteration
*/
inc_nrnb(nrnb, eNR_NBKERNEL_GENERIC_ADRESS, nlist->nri*14 + nlist->jindex[n]*54);
}
/*
* Gromacs nonbonded kernel pf_nb_kernel112
* Coulomb interaction: Normal Coulomb
* VdW interaction: Lennard-Jones
* water optimization: pairs of SPC/TIP3P interactions
* Calculate forces: yes
*/
void pf_nb_kernel112(
int * p_nri,
int * iinr,
int * jindex,
int * jjnr,
int * shift,
real * shiftvec,
real * fshift,
int * gid,
real * pos,
real * faction,
real * charge,
real * p_facel,
real * p_krf,
real * p_crf,
real * Vc,
int * type,
int * p_ntype,
real * vdwparam,
real * Vvdw,
real * p_tabscale,
real * VFtab,
real * invsqrta,
real * dvda,
real * p_gbtabscale,
real * GBtab,
int * p_nthreads,
int * count,
void * mtx,
int * outeriter,
int * inneriter,
real * work,
t_pf_global * pf_global)
{
int nri,ntype,nthreads;
real facel,krf,crf,tabscale,gbtabscale;
int n,ii,is3,ii3,k,nj0,nj1,jnr,j3,ggid;
int nn0,nn1,nouter,ninner;
real shX,shY,shZ;
real fscal,tx,ty,tz;
real rinvsq;
real qq,vcoul,vctot;
int tj;
real rinvsix;
real Vvdw6,Vvdwtot;
real Vvdw12;
real ix1,iy1,iz1,fix1,fiy1,fiz1;
real ix2,iy2,iz2,fix2,fiy2,fiz2;
real ix3,iy3,iz3,fix3,fiy3,fiz3;
real jx1,jy1,jz1,fjx1,fjy1,fjz1;
real jx2,jy2,jz2,fjx2,fjy2,fjz2;
real jx3,jy3,jz3,fjx3,fjy3,fjz3;
real dx11,dy11,dz11,rsq11,rinv11;
real dx12,dy12,dz12,rsq12,rinv12;
real dx13,dy13,dz13,rsq13,rinv13;
real dx21,dy21,dz21,rsq21,rinv21;
real dx22,dy22,dz22,rsq22,rinv22;
real dx23,dy23,dz23,rsq23,rinv23;
real dx31,dy31,dz31,rsq31,rinv31;
real dx32,dy32,dz32,rsq32,rinv32;
real dx33,dy33,dz33,rsq33,rinv33;
real qO,qH,qqOO,qqOH,qqHH;
real c6,c12;
real pf_coul, pf_lj;
nri = *p_nri;
ntype = *p_ntype;
nthreads = *p_nthreads;
facel = *p_facel;
krf = *p_krf;
crf = *p_crf;
tabscale = *p_tabscale;
/* Initialize water data */
ii = iinr[0];
qO = charge[ii];
qH = charge[ii+1];
qqOO = facel*qO*qO;
qqOH = facel*qO*qH;
qqHH = facel*qH*qH;
tj = 2*(ntype+1)*type[ii];
c6 = vdwparam[tj];
c12 = vdwparam[tj+1];
/* Reset outer and inner iteration counters */
nouter = 0;
ninner = 0;
/* Loop over thread workunits */
do
{
#ifdef GMX_THREAD_SHM_FDECOMP
tMPI_Thread_mutex_lock((tMPI_Thread_mutex_t *)mtx);
nn0 = *count;
/* Take successively smaller chunks (at least 10 lists) */
nn1 = nn0+(nri-nn0)/(2*nthreads)+10;
*count = nn1;
tMPI_Thread_mutex_unlock((tMPI_Thread_mutex_t *)mtx);
if(nn1>nri) nn1=nri;
#else
nn0 = 0;
nn1 = nri;
#endif
/* Start outer loop over neighborlists */
for(n=nn0; (n<nn1); n++)
{
/* Load shift vector for this list */
is3 = 3*shift[n];
shX = shiftvec[is3];
shY = shiftvec[is3+1];
shZ = shiftvec[is3+2];
/* Load limits for loop over neighbors */
nj0 = jindex[n];
nj1 = jindex[n+1];
/* Get outer coordinate index */
ii = iinr[n];
ii3 = 3*ii;
/* Load i atom data, add shift vector */
ix1 = shX + pos[ii3+0];
iy1 = shY + pos[ii3+1];
iz1 = shZ + pos[ii3+2];
ix2 = shX + pos[ii3+3];
iy2 = shY + pos[ii3+4];
iz2 = shZ + pos[ii3+5];
ix3 = shX + pos[ii3+6];
iy3 = shY + pos[ii3+7];
iz3 = shZ + pos[ii3+8];
/* Zero the potential energy for this list */
vctot = 0;
Vvdwtot = 0;
/* Clear i atom forces */
fix1 = 0;
fiy1 = 0;
fiz1 = 0;
fix2 = 0;
fiy2 = 0;
fiz2 = 0;
fix3 = 0;
fiy3 = 0;
fiz3 = 0;
for(k=nj0; (k<nj1); k++)
{
/* Get j neighbor index, and coordinate index */
jnr = jjnr[k];
j3 = 3*jnr;
/* load j atom coordinates */
jx1 = pos[j3+0];
jy1 = pos[j3+1];
jz1 = pos[j3+2];
jx2 = pos[j3+3];
jy2 = pos[j3+4];
jz2 = pos[j3+5];
jx3 = pos[j3+6];
jy3 = pos[j3+7];
jz3 = pos[j3+8];
/* Calculate distance */
dx11 = ix1 - jx1;
dy11 = iy1 - jy1;
dz11 = iz1 - jz1;
rsq11 = dx11*dx11+dy11*dy11+dz11*dz11;
dx12 = ix1 - jx2;
dy12 = iy1 - jy2;
dz12 = iz1 - jz2;
rsq12 = dx12*dx12+dy12*dy12+dz12*dz12;
dx13 = ix1 - jx3;
dy13 = iy1 - jy3;
dz13 = iz1 - jz3;
rsq13 = dx13*dx13+dy13*dy13+dz13*dz13;
dx21 = ix2 - jx1;
dy21 = iy2 - jy1;
dz21 = iz2 - jz1;
rsq21 = dx21*dx21+dy21*dy21+dz21*dz21;
dx22 = ix2 - jx2;
dy22 = iy2 - jy2;
dz22 = iz2 - jz2;
rsq22 = dx22*dx22+dy22*dy22+dz22*dz22;
dx23 = ix2 - jx3;
dy23 = iy2 - jy3;
dz23 = iz2 - jz3;
rsq23 = dx23*dx23+dy23*dy23+dz23*dz23;
dx31 = ix3 - jx1;
dy31 = iy3 - jy1;
dz31 = iz3 - jz1;
rsq31 = dx31*dx31+dy31*dy31+dz31*dz31;
dx32 = ix3 - jx2;
dy32 = iy3 - jy2;
dz32 = iz3 - jz2;
rsq32 = dx32*dx32+dy32*dy32+dz32*dz32;
dx33 = ix3 - jx3;
dy33 = iy3 - jy3;
dz33 = iz3 - jz3;
rsq33 = dx33*dx33+dy33*dy33+dz33*dz33;
/* Calculate 1/r and 1/r2 */
rinv11 = gmx_invsqrt(rsq11);
rinv12 = gmx_invsqrt(rsq12);
rinv13 = gmx_invsqrt(rsq13);
rinv21 = gmx_invsqrt(rsq21);
rinv22 = gmx_invsqrt(rsq22);
rinv23 = gmx_invsqrt(rsq23);
rinv31 = gmx_invsqrt(rsq31);
rinv32 = gmx_invsqrt(rsq32);
rinv33 = gmx_invsqrt(rsq33);
/* Load parameters for j atom */
qq = qqOO;
rinvsq = rinv11*rinv11;
/* Coulomb interaction */
vcoul = qq*rinv11;
vctot = vctot+vcoul;
pf_coul = vcoul*rinvsq;
/* Lennard-Jones interaction */
rinvsix = rinvsq*rinvsq*rinvsq;
Vvdw6 = c6*rinvsix;
Vvdw12 = c12*rinvsix*rinvsix;
Vvdwtot = Vvdwtot+Vvdw12-Vvdw6;
pf_lj = (12.0*Vvdw12-6.0*Vvdw6)*rinvsq;
fscal = (vcoul+12.0*Vvdw12-6.0*Vvdw6)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx11;
ty = fscal*dy11;
tz = fscal*dz11;
/* Increment i atom force */
fix1 = fix1 + tx;
fiy1 = fiy1 + ty;
fiz1 = fiz1 + tz;
/* Decrement j atom force */
fjx1 = faction[j3+0] - tx;
fjy1 = faction[j3+1] - ty;
fjz1 = faction[j3+2] - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded(pf_global, ii, jnr, pf_coul, pf_lj, dx11, dy11, dz11);
/* Load parameters for j atom */
qq = qqOH;
rinvsq = rinv12*rinv12;
/* Coulomb interaction */
vcoul = qq*rinv12;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx12;
ty = fscal*dy12;
tz = fscal*dz12;
/* Increment i atom force */
fix1 = fix1 + tx;
fiy1 = fiy1 + ty;
fiz1 = fiz1 + tz;
/* Decrement j atom force */
fjx2 = faction[j3+3] - tx;
fjy2 = faction[j3+4] - ty;
fjz2 = faction[j3+5] - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii, jnr+1, PF_INTER_COULOMB, fscal, dx12, dy12, dz12);
/* Load parameters for j atom */
qq = qqOH;
rinvsq = rinv13*rinv13;
/* Coulomb interaction */
vcoul = qq*rinv13;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx13;
ty = fscal*dy13;
tz = fscal*dz13;
/* Increment i atom force */
fix1 = fix1 + tx;
fiy1 = fiy1 + ty;
fiz1 = fiz1 + tz;
/* Decrement j atom force */
fjx3 = faction[j3+6] - tx;
fjy3 = faction[j3+7] - ty;
fjz3 = faction[j3+8] - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii, jnr+2, PF_INTER_COULOMB, fscal, dx13, dy13, dz13);
/* Load parameters for j atom */
qq = qqOH;
rinvsq = rinv21*rinv21;
/* Coulomb interaction */
vcoul = qq*rinv21;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx21;
ty = fscal*dy21;
tz = fscal*dz21;
/* Increment i atom force */
fix2 = fix2 + tx;
fiy2 = fiy2 + ty;
fiz2 = fiz2 + tz;
/* Decrement j atom force */
fjx1 = fjx1 - tx;
fjy1 = fjy1 - ty;
fjz1 = fjz1 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+1, jnr, PF_INTER_COULOMB, fscal, dx21, dy21, dz21);
/* Load parameters for j atom */
qq = qqHH;
rinvsq = rinv22*rinv22;
/* Coulomb interaction */
vcoul = qq*rinv22;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx22;
ty = fscal*dy22;
tz = fscal*dz22;
/* Increment i atom force */
fix2 = fix2 + tx;
fiy2 = fiy2 + ty;
fiz2 = fiz2 + tz;
/* Decrement j atom force */
fjx2 = fjx2 - tx;
fjy2 = fjy2 - ty;
fjz2 = fjz2 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+1, jnr+1, PF_INTER_COULOMB, fscal, dx22, dy22, dz22);
/* Load parameters for j atom */
qq = qqHH;
rinvsq = rinv23*rinv23;
/* Coulomb interaction */
vcoul = qq*rinv23;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx23;
ty = fscal*dy23;
tz = fscal*dz23;
/* Increment i atom force */
fix2 = fix2 + tx;
fiy2 = fiy2 + ty;
fiz2 = fiz2 + tz;
/* Decrement j atom force */
fjx3 = fjx3 - tx;
fjy3 = fjy3 - ty;
fjz3 = fjz3 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+1, jnr+2, PF_INTER_COULOMB, fscal, dx23, dy23, dz23);
/* Load parameters for j atom */
qq = qqOH;
rinvsq = rinv31*rinv31;
/* Coulomb interaction */
vcoul = qq*rinv31;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx31;
ty = fscal*dy31;
tz = fscal*dz31;
/* Increment i atom force */
fix3 = fix3 + tx;
fiy3 = fiy3 + ty;
fiz3 = fiz3 + tz;
/* Decrement j atom force */
faction[j3+0] = fjx1 - tx;
faction[j3+1] = fjy1 - ty;
faction[j3+2] = fjz1 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+2, jnr, PF_INTER_COULOMB, fscal, dx31, dy31, dz31);
/* Load parameters for j atom */
qq = qqHH;
rinvsq = rinv32*rinv32;
/* Coulomb interaction */
vcoul = qq*rinv32;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx32;
ty = fscal*dy32;
tz = fscal*dz32;
/* Increment i atom force */
fix3 = fix3 + tx;
fiy3 = fiy3 + ty;
fiz3 = fiz3 + tz;
/* Decrement j atom force */
faction[j3+3] = fjx2 - tx;
faction[j3+4] = fjy2 - ty;
faction[j3+5] = fjz2 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+2, jnr+1, PF_INTER_COULOMB, fscal, dx32, dy32, dz32);
/* Load parameters for j atom */
qq = qqHH;
rinvsq = rinv33*rinv33;
/* Coulomb interaction */
vcoul = qq*rinv33;
vctot = vctot+vcoul;
fscal = (vcoul)*rinvsq;
/* Calculate temporary vectorial force */
tx = fscal*dx33;
ty = fscal*dy33;
tz = fscal*dz33;
/* Increment i atom force */
fix3 = fix3 + tx;
fiy3 = fiy3 + ty;
fiz3 = fiz3 + tz;
/* Decrement j atom force */
faction[j3+6] = fjx3 - tx;
faction[j3+7] = fjy3 - ty;
faction[j3+8] = fjz3 - tz;
/* pairwise forces */
if (pf_global->bInitialized)
pf_atom_add_nonbonded_single(pf_global, ii+2, jnr+2, PF_INTER_COULOMB, fscal, dx33, dy33, dz33);
/* Inner loop uses 245 flops/iteration */
}
/* Add i forces to mem and shifted force list */
faction[ii3+0] = faction[ii3+0] + fix1;
faction[ii3+1] = faction[ii3+1] + fiy1;
faction[ii3+2] = faction[ii3+2] + fiz1;
faction[ii3+3] = faction[ii3+3] + fix2;
faction[ii3+4] = faction[ii3+4] + fiy2;
faction[ii3+5] = faction[ii3+5] + fiz2;
faction[ii3+6] = faction[ii3+6] + fix3;
faction[ii3+7] = faction[ii3+7] + fiy3;
faction[ii3+8] = faction[ii3+8] + fiz3;
fshift[is3] = fshift[is3]+fix1+fix2+fix3;
fshift[is3+1] = fshift[is3+1]+fiy1+fiy2+fiy3;
fshift[is3+2] = fshift[is3+2]+fiz1+fiz2+fiz3;
/* Add potential energies to the group for this list */
ggid = gid[n];
Vc[ggid] = Vc[ggid] + vctot;
Vvdw[ggid] = Vvdw[ggid] + Vvdwtot;
/* Increment number of inner iterations */
ninner = ninner + nj1 - nj0;
/* Outer loop uses 29 flops/iteration */
}
/* Increment number of outer iterations */
nouter = nouter + nn1 - nn0;
}
while (nn1<nri);
/* Write outer/inner iteration count to pointers */
*outeriter = nouter;
*inneriter = ninner;
}
