static void lo_fcv(int i0,int i1,int g0,
real x[],real f[],tensor vir,
int is[],real box[], bool bTriclinic)
{
int i,i3,gg,g3,tx,ty,tz;
real xx,yy,zz;
real dvxx=0,dvxy=0,dvxz=0,dvyx=0,dvyy=0,dvyz=0,dvzx=0,dvzy=0,dvzz=0;
if(bTriclinic) {
for(i=i0,gg=g0; (i<i1); i++,gg++) {
i3=DIM*i;
g3=DIM*gg;
tx=is[g3+XX];
ty=is[g3+YY];
tz=is[g3+ZZ];
xx=x[i3+XX]-tx*box[XXXX]-ty*box[YYXX]-tz*box[ZZXX];
dvxx+=xx*f[i3+XX];
dvxy+=xx*f[i3+YY];
dvxz+=xx*f[i3+ZZ];
yy=x[i3+YY]-ty*box[YYYY]-tz*box[ZZYY];
dvyx+=yy*f[i3+XX];
dvyy+=yy*f[i3+YY];
dvyz+=yy*f[i3+ZZ];
zz=x[i3+ZZ]-tz*box[ZZZZ];
dvzx+=zz*f[i3+XX];
dvzy+=zz*f[i3+YY];
dvzz+=zz*f[i3+ZZ];
}
} else {
for(i=i0,gg=g0; (i<i1); i++,gg++) {
i3=DIM*i;
g3=DIM*gg;
tx=is[g3+XX];
ty=is[g3+YY];
tz=is[g3+ZZ];
xx=x[i3+XX]-tx*box[XXXX];
dvxx+=xx*f[i3+XX];
dvxy+=xx*f[i3+YY];
dvxz+=xx*f[i3+ZZ];
yy=x[i3+YY]-ty*box[YYYY];
dvyx+=yy*f[i3+XX];
dvyy+=yy*f[i3+YY];
dvyz+=yy*f[i3+ZZ];
zz=x[i3+ZZ]-tz*box[ZZZZ];
dvzx+=zz*f[i3+XX];
dvzy+=zz*f[i3+YY];
dvzz+=zz*f[i3+ZZ];
}
}
upd_vir(vir[XX],dvxx,dvxy,dvxz);
upd_vir(vir[YY],dvyx,dvyy,dvyz);
upd_vir(vir[ZZ],dvzx,dvzy,dvzz);
}
static void lo_fcv2(int i0,int i1,
rvec x[],rvec f[],tensor vir,
ivec is[],matrix box, bool bTriclinic)
{
int i,gg,tx,ty,tz;
real xx,yy,zz;
real dvxx=0,dvxy=0,dvxz=0,dvyx=0,dvyy=0,dvyz=0,dvzx=0,dvzy=0,dvzz=0;
if(bTriclinic) {
for(i=i0,gg=0; (i<i1); i++,gg++) {
tx=is[gg][XX];
ty=is[gg][YY];
tz=is[gg][ZZ];
xx=x[i][XX]-tx*box[XX][XX]-ty*box[YY][XX]-tz*box[ZZ][XX];
dvxx+=xx*f[i][XX];
dvxy+=xx*f[i][YY];
dvxz+=xx*f[i][ZZ];
yy=x[i][YY]-ty*box[YY][YY]-tz*box[ZZ][YY];
dvyx+=yy*f[i][XX];
dvyy+=yy*f[i][YY];
dvyz+=yy*f[i][ZZ];
zz=x[i][ZZ]-tz*box[ZZ][ZZ];
dvzx+=zz*f[i][XX];
dvzy+=zz*f[i][YY];
dvzz+=zz*f[i][ZZ];
}
} else {
for(i=i0,gg=0; (i<i1); i++,gg++) {
tx=is[gg][XX];
ty=is[gg][YY];
tz=is[gg][ZZ];
xx=x[i][XX]-tx*box[XX][XX];
dvxx+=xx*f[i][XX];
dvxy+=xx*f[i][YY];
dvxz+=xx*f[i][ZZ];
yy=x[i][YY]-ty*box[YY][YY];
dvyx+=yy*f[i][XX];
dvyy+=yy*f[i][YY];
dvyz+=yy*f[i][ZZ];
zz=x[i][ZZ]-tz*box[ZZ][ZZ];
dvzx+=zz*f[i][XX];
dvzy+=zz*f[i][YY];
dvzz+=zz*f[i][ZZ];
}
}
upd_vir(vir[XX],dvxx,dvxy,dvxz);
upd_vir(vir[YY],dvyx,dvyy,dvyz);
upd_vir(vir[ZZ],dvzx,dvzy,dvzz);
}
void calc_vir(int nxf, rvec x[], rvec f[], tensor vir,
gmx_bool bScrewPBC, matrix box)
{
int i;
double dvxx = 0, dvxy = 0, dvxz = 0, dvyx = 0, dvyy = 0, dvyz = 0, dvzx = 0, dvzy = 0, dvzz = 0;
#pragma omp parallel for num_threads(gmx_omp_nthreads_get(emntDefault)) \
schedule(static) \
reduction(+: dvxx, dvxy, dvxz, dvyx, dvyy, dvyz, dvzx, dvzy, dvzz)
for (i = 0; i < nxf; i++)
{
dvxx += x[i][XX]*f[i][XX];
dvxy += x[i][XX]*f[i][YY];
dvxz += x[i][XX]*f[i][ZZ];
dvyx += x[i][YY]*f[i][XX];
dvyy += x[i][YY]*f[i][YY];
dvyz += x[i][YY]*f[i][ZZ];
dvzx += x[i][ZZ]*f[i][XX];
dvzy += x[i][ZZ]*f[i][YY];
dvzz += x[i][ZZ]*f[i][ZZ];
if (bScrewPBC)
{
int isx = IS2X(i);
/* We should correct all odd x-shifts, but the range of isx is -2 to 2 */
if (isx == 1 || isx == -1)
{
dvyy += box[YY][YY]*f[i][YY];
dvyz += box[YY][YY]*f[i][ZZ];
dvzy += box[ZZ][ZZ]*f[i][YY];
dvzz += box[ZZ][ZZ]*f[i][ZZ];
}
}
}
upd_vir(vir[XX], dvxx, dvxy, dvxz);
upd_vir(vir[YY], dvyx, dvyy, dvyz);
upd_vir(vir[ZZ], dvzx, dvzy, dvzz);
}
void calc_vir(int nxf, rvec x[], rvec f[], tensor vir,
gmx_bool bScrewPBC, matrix box)
{
int i, isx;
double dvxx = 0, dvxy = 0, dvxz = 0, dvyx = 0, dvyy = 0, dvyz = 0, dvzx = 0, dvzy = 0, dvzz = 0;
for (i = 0; (i < nxf); i++)
{
dvxx += x[i][XX]*f[i][XX];
dvxy += x[i][XX]*f[i][YY];
dvxz += x[i][XX]*f[i][ZZ];
dvyx += x[i][YY]*f[i][XX];
dvyy += x[i][YY]*f[i][YY];
dvyz += x[i][YY]*f[i][ZZ];
dvzx += x[i][ZZ]*f[i][XX];
dvzy += x[i][ZZ]*f[i][YY];
dvzz += x[i][ZZ]*f[i][ZZ];
if (bScrewPBC)
{
isx = IS2X(i);
/* We should correct all odd x-shifts, but the range of isx is -2 to 2 */
if (isx == 1 || isx == -1)
{
dvyy += box[YY][YY]*f[i][YY];
dvyz += box[YY][YY]*f[i][ZZ];
dvzy += box[ZZ][ZZ]*f[i][YY];
dvzz += box[ZZ][ZZ]*f[i][ZZ];
}
}
}
upd_vir(vir[XX], dvxx, dvxy, dvxz);
upd_vir(vir[YY], dvyx, dvyy, dvyz);
upd_vir(vir[ZZ], dvzx, dvzy, dvzz);
}
void calc_vir(FILE *log,int nxf,rvec x[],rvec f[],tensor vir)
{
int i;
real dvxx=0,dvxy=0,dvxz=0,dvyx=0,dvyy=0,dvyz=0,dvzx=0,dvzy=0,dvzz=0;
for(i=0; (i<nxf); i++) {
dvxx+=x[i][XX]*f[i][XX];
dvxy+=x[i][XX]*f[i][YY];
dvxz+=x[i][XX]*f[i][ZZ];
dvyx+=x[i][YY]*f[i][XX];
dvyy+=x[i][YY]*f[i][YY];
dvyz+=x[i][YY]*f[i][ZZ];
dvzx+=x[i][ZZ]*f[i][XX];
dvzy+=x[i][ZZ]*f[i][YY];
dvzz+=x[i][ZZ]*f[i][ZZ];
}
upd_vir(vir[XX],dvxx,dvxy,dvxz);
upd_vir(vir[YY],dvyx,dvyy,dvyz);
upd_vir(vir[ZZ],dvzx,dvzy,dvzz);
}
void calc_vir(int nxf, const rvec x[], const rvec f[], tensor vir,
bool bScrewPBC, const matrix box)
{
matrix x_times_f;
int nthreads = gmx_omp_nthreads_get_simple_rvec_task(emntDefault, nxf*9);
GMX_ASSERT(nthreads >= 1, "Avoids uninitialized x_times_f (warning)");
if (nthreads == 1)
{
calc_x_times_f(nxf, x, f, bScrewPBC, box, x_times_f);
}
else
{
/* Use a buffer on the stack for storing thread-local results.
* We use 2 extra elements (=18 reals) per thread to separate thread
* local data by at least a cache line. Element 0 is not used.
*/
matrix xf_buf[GMX_OPENMP_MAX_THREADS*3];
#pragma omp parallel for num_threads(nthreads) schedule(static)
for (int thread = 0; thread < nthreads; thread++)
{
int start = (nxf*thread)/nthreads;
int end = std::min(nxf*(thread + 1)/nthreads, nxf);
calc_x_times_f(end - start, x + start, f + start, bScrewPBC, box,
thread == 0 ? x_times_f : xf_buf[thread*3]);
}
for (int thread = 1; thread < nthreads; thread++)
{
m_add(x_times_f, xf_buf[thread*3], x_times_f);
}
}
for (int d = 0; d < DIM; d++)
{
upd_vir(vir[d], x_times_f[d][XX], x_times_f[d][YY], x_times_f[d][ZZ]);
}
}
