void DihedralCharmmIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
const int tp1 = atom->ntypes + 1;
const int bp1 = atom->ndihedraltypes + 1;
fc.set_ntypes(tp1,bp1,memory);
buffers->set_ntypes(tp1);
for (int i = 0; i < tp1; i++) {
for (int j = 0; j < tp1; j++) {
fc.ljp[i][j].lj1 = lj14_1[i][j];
fc.ljp[i][j].lj2 = lj14_2[i][j];
fc.ljp[i][j].lj3 = lj14_3[i][j];
fc.ljp[i][j].lj4 = lj14_4[i][j];
}
}
for (int i = 0; i < bp1; i++) {
fc.bp[i].multiplicity = multiplicity[i];
fc.bp[i].cos_shift = cos_shift[i];
fc.bp[i].sin_shift = sin_shift[i];
fc.bp[i].k = k[i];
fc.weight[i] = weight[i];
}
}
void PairBuckCoulCutIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
int tp1 = atom->ntypes + 1;
int ntable = 1;
if (ncoultablebits)
for (int i = 0; i < ncoultablebits; i++) ntable *= 2;
fc.set_ntypes(tp1, ntable, memory, _cop);
// Repeat cutsq calculation because done after call to init_style
for (int i = 1; i <= atom->ntypes; i++) {
for (int j = i; j <= atom->ntypes; j++) {
double cut;
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0))
cut = init_one(i, j);
else
cut = 0.0;
cutsq[i][j] = cutsq[j][i] = cut*cut;
}
}
for (int i = 0; i < 4; i++) {
fc.special_lj[i] = force->special_lj[i];
fc.special_coul[i] = force->special_coul[i];
fc.special_coul[0] = 1.0;
fc.special_lj[0] = 1.0;
}
for (int i = 1; i < tp1; i++) {
for (int j = 1; j < tp1; j++) {
fc.c_cut[i][j].cutsq = cutsq[i][j];
fc.c_cut[i][j].cut_ljsq = cut_ljsq[i][j];
fc.c_cut[i][j].cut_coulsq = cut_coulsq[i][j];
fc.c_force[i][j].buck1 = buck1[i][j];
fc.c_force[i][j].buck2 = buck2[i][j];
fc.c_force[i][j].rhoinv = rhoinv[i][j];
fc.c_energy[i][j].a = a[i][j];
fc.c_energy[i][j].c = c[i][j];
fc.c_energy[i][j].offset = offset[i][j];
}
}
#ifdef _LMP_INTEL_OFFLOAD
if (_cop < 0) return;
flt_t * special_lj = fc.special_lj;
flt_t * special_coul = fc.special_coul;
C_FORCE_T * c_force = fc.c_force[0];
C_ENERGY_T * c_energy = fc.c_energy[0];
C_CUT_T * c_cut = fc.c_cut[0];
int tp1sq = tp1 * tp1;
#pragma offload_transfer target(mic:_cop) \
in(special_lj, special_coul: length(4) alloc_if(0) free_if(0)) \
in(c_force, c_energy, c_cut: length(tp1sq) alloc_if(0) free_if(0))
#endif
}
void ImproperHarmonicIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
const int bp1 = atom->nimpropertypes + 1;
fc.set_ntypes(bp1,memory);
for (int i = 0; i < bp1; i++) {
fc.fc[i].k = k[i];
fc.fc[i].chi = chi[i];
}
}
void AngleHarmonicIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
const int bp1 = atom->nangletypes + 1;
fc.set_ntypes(bp1,memory);
for (int i = 0; i < bp1; i++) {
fc.fc[i].k = k[i];
fc.fc[i].theta0 = theta0[i];
}
}
void BondHarmonicIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
const int bp1 = atom->nbondtypes + 1;
fc.set_ntypes(bp1,memory);
for (int i = 1; i < bp1; i++) {
fc.fc[i].k = k[i];
fc.fc[i].r0 = r0[i];
}
}
void PairBuckIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
int tp1 = atom->ntypes + 1;
fc.set_ntypes(tp1, memory, _cop);
buffers->set_ntypes(tp1);
flt_t **cutneighsq = buffers->get_cutneighsq();
// Repeat cutsq calculation because done after call to init_style
double cut, cutneigh;
for (int i = 1; i <= atom->ntypes; i++) {
for (int j = i; j <= atom->ntypes; j++) {
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
cut = init_one(i, j);
cutneigh = cut + neighbor->skin;
cutsq[i][j] = cutsq[j][i] = cut*cut;
cutneighsq[i][j] = cutneighsq[j][i] = cutneigh * cutneigh;
}
}
}
for (int i = 0; i < 4; i++) {
fc.special_lj[i] = force->special_lj[i];
fc.special_lj[0] = 1.0;
}
for (int i = 0; i < tp1; i++) {
for (int j = 0; j < tp1; j++) {
fc.c_force[i][j].buck1 = buck1[i][j];
fc.c_force[i][j].buck2 = buck2[i][j];
fc.c_force[i][j].rhoinv = rhoinv[i][j];
fc.c_force[i][j].cutsq = cutsq[i][j];
fc.c_energy[i][j].a = a[i][j];
fc.c_energy[i][j].c = c[i][j];
fc.c_energy[i][j].offset = offset[i][j];
}
}
#ifdef _LMP_INTEL_OFFLOAD
if (_cop < 0) return;
flt_t * special_lj = fc.special_lj;
C_FORCE_T * c_force = fc.c_force[0];
C_ENERGY_T * c_energy = fc.c_energy[0];
flt_t * ocutneighsq = cutneighsq[0];
int tp1sq = tp1 * tp1;
#pragma offload_transfer target(mic:_cop) \
in(special_lj: length(4) alloc_if(0) free_if(0)) \
in(c_force, c_energy: length(tp1sq) alloc_if(0) free_if(0)) \
in(ocutneighsq: length(tp1sq) alloc_if(0) free_if(0))
#endif
}
void DihedralOPLSIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
const int bp1 = atom->ndihedraltypes + 1;
fc.set_ntypes(bp1,memory);
for (int i = 0; i < bp1; i++) {
fc.bp[i].k1 = k1[i];
fc.bp[i].k2 = k2[i];
fc.bp[i].k3 = k3[i];
fc.bp[i].k4 = k4[i];
}
}
void DihedralHarmonicIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> * /*buffers*/)
{
const int bp1 = atom->ndihedraltypes + 1;
fc.set_ntypes(bp1,memory);
for (int i = 1; i < bp1; i++) {
fc.bp[i].multiplicity = multiplicity[i];
fc.bp[i].cos_shift = cos_shift[i];
fc.bp[i].sin_shift = sin_shift[i];
fc.bp[i].k = k[i];
}
}
void AngleCharmmIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
const int bp1 = atom->ndihedraltypes + 1;
fc.set_ntypes(bp1,memory);
for (int i = 0; i < bp1; i++) {
fc.fc[i].k = k[i];
fc.fc[i].theta0 = theta0[i];
fc.fc[i].k_ub = k_ub[i];
fc.fc[i].r_ub = r_ub[i];
}
}
void BondFENEIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> * /*buffers*/)
{
const int bp1 = atom->nbondtypes + 1;
fc.set_ntypes(bp1,memory);
for (int i = 1; i < bp1; i++) {
fc.fc[i].k = k[i];
fc.fc[i].ir0sq = 1.0 / (r0[i] * r0[i]);
fc.fc[i].sigma = sigma[i];
fc.fc[i].epsilon = epsilon[i];
}
}
void PairLJCutCoulLongIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
int tp1 = atom->ntypes + 1;
int ntable = 1;
if (ncoultablebits)
for (int i = 0; i < ncoultablebits; i++) ntable *= 2;
fc.set_ntypes(tp1, ntable, memory, _cop);
buffers->set_ntypes(tp1);
flt_t **cutneighsq = buffers->get_cutneighsq();
// Repeat cutsq calculation because done after call to init_style
double cut, cutneigh;
for (int i = 1; i <= atom->ntypes; i++) {
for (int j = i; j <= atom->ntypes; j++) {
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
cut = init_one(i, j);
cutneigh = cut + neighbor->skin;
cutsq[i][j] = cutsq[j][i] = cut*cut;
cutneighsq[i][j] = cutneighsq[j][i] = cutneigh * cutneigh;
}
}
}
fc.g_ewald = force->kspace->g_ewald;
fc.tabinnersq = tabinnersq;
for (int i = 0; i < 4; i++) {
fc.special_lj[i] = force->special_lj[i];
fc.special_coul[i] = force->special_coul[i];
fc.special_coul[0] = 1.0;
fc.special_lj[0] = 1.0;
}
for (int i = 0; i < tp1; i++) {
for (int j = 0; j < tp1; j++) {
fc.c_force[i][j].cutsq = cutsq[i][j];
fc.c_force[i][j].cut_ljsq = cut_ljsq[i][j];
fc.c_force[i][j].lj1 = lj1[i][j];
fc.c_force[i][j].lj2 = lj2[i][j];
fc.c_energy[i][j].lj3 = lj3[i][j];
fc.c_energy[i][j].lj4 = lj4[i][j];
fc.c_energy[i][j].offset = offset[i][j];
}
}
if (ncoultablebits) {
for (int i = 0; i < ntable; i++) {
fc.table[i].r = rtable[i];
fc.table[i].dr = drtable[i];
fc.table[i].f = ftable[i];
fc.table[i].df = dftable[i];
fc.etable[i] = etable[i];
fc.detable[i] = detable[i];
fc.ctable[i] = ctable[i];
fc.dctable[i] = dctable[i];
}
}
#ifdef _LMP_INTEL_OFFLOAD
if (_cop < 0) return;
flt_t * special_lj = fc.special_lj;
flt_t * special_coul = fc.special_coul;
C_FORCE_T * c_force = fc.c_force[0];
C_ENERGY_T * c_energy = fc.c_energy[0];
TABLE_T * table = fc.table;
flt_t * etable = fc.etable;
flt_t * detable = fc.detable;
flt_t * ctable = fc.ctable;
flt_t * dctable = fc.dctable;
flt_t * ocutneighsq = cutneighsq[0];
int tp1sq = tp1 * tp1;
#pragma offload_transfer target(mic:_cop) \
in(special_lj, special_coul: length(4) alloc_if(0) free_if(0)) \
in(c_force, c_energy: length(tp1sq) alloc_if(0) free_if(0)) \
in(table: length(ntable) alloc_if(0) free_if(0)) \
in(etable,detable,ctable,dctable: length(ntable) alloc_if(0) free_if(0)) \
in(ocutneighsq: length(tp1sq) alloc_if(0) free_if(0))
#endif
}
void PairSWIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
int off_ccache = 0;
#ifdef _LMP_INTEL_OFFLOAD
if (_cop >= 0) off_ccache = 1;
#endif
buffers->grow_ccache(off_ccache, comm->nthreads);
_ccache_stride = buffers->ccache_stride();
int tp1 = atom->ntypes + 1;
fc.set_ntypes(tp1,memory,_cop);
buffers->set_ntypes(tp1);
flt_t **cutneighsq = buffers->get_cutneighsq();
// Repeat cutsq calculation because done after call to init_style
double cut, cutneigh;
for (int i = 1; i <= atom->ntypes; i++) {
for (int j = i; j <= atom->ntypes; j++) {
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
cut = init_one(i,j);
cutneigh = cut + neighbor->skin;
cutsq[i][j] = cutsq[j][i] = cut*cut;
cutneighsq[i][j] = cutneighsq[j][i] = cutneigh * cutneigh;
}
}
}
_spq = 1;
for (int ii = 0; ii < tp1; ii++) {
int i = map[ii];
for (int jj = 0; jj < tp1; jj++) {
int j = map[jj];
if (i < 0 || j < 0 || ii == 0 || jj == 0) {
fc.p2[ii][jj].cutsq = 0;
fc.p2[ii][jj].cut = 0;
fc.p2[ii][jj].sigma_gamma = 0;
fc.p2f[ii][jj].cut = 0;
fc.p2f[ii][jj].powerp = 0;
fc.p2f[ii][jj].powerq = 0;
fc.p2f[ii][jj].sigma = 0;
fc.p2f[ii][jj].c1 = 0;
fc.p2f[ii][jj].c2 = 0;
fc.p2f[ii][jj].c3 = 0;
fc.p2f[ii][jj].c4 = 0;
fc.p2e[ii][jj].c5 = 0;
fc.p2e[ii][jj].c6 = 0;
} else {
int ijparam = elem2param[i][j][j];
fc.p2[ii][jj].cutsq = params[ijparam].cutsq;
fc.p2[ii][jj].cut = params[ijparam].cut;
fc.p2[ii][jj].sigma_gamma = params[ijparam].sigma_gamma;
fc.p2f[ii][jj].cut = params[ijparam].cut;
fc.p2f[ii][jj].powerp = params[ijparam].powerp;
fc.p2f[ii][jj].powerq = params[ijparam].powerq;
fc.p2f[ii][jj].sigma = params[ijparam].sigma;
fc.p2f[ii][jj].c1 = params[ijparam].c1;
fc.p2f[ii][jj].c2 = params[ijparam].c2;
fc.p2f[ii][jj].c3 = params[ijparam].c3;
fc.p2f[ii][jj].c4 = params[ijparam].c4;
fc.p2e[ii][jj].c5 = params[ijparam].c5;
fc.p2e[ii][jj].c6 = params[ijparam].c6;
double cutcut = params[ijparam].cut * params[ijparam].cut;
if (params[ijparam].cutsq >= cutcut)
fc.p2[ii][jj].cutsq *= 0.98;
if (params[ijparam].powerp != 4.0 || params[ijparam].powerq != 0.0)
_spq = 0;
}
for (int kk = 0; kk < tp1; kk++) {
int k = map[kk];
if (i < 0 || j < 0 || k < 0 || ii == 0 || jj == 0 || kk == 0) {
fc.p3[ii][jj][kk].costheta = 0;
fc.p3[ii][jj][kk].lambda_epsilon = 0;
fc.p3[ii][jj][kk].lambda_epsilon2 = 0;
} else {
int ijkparam = elem2param[i][j][k];
fc.p3[ii][jj][kk].costheta = params[ijkparam].costheta;
fc.p3[ii][jj][kk].lambda_epsilon = params[ijkparam].lambda_epsilon;
fc.p3[ii][jj][kk].lambda_epsilon2 = params[ijkparam].lambda_epsilon2;
}
}
}
}
_host_pad = 1;
_offload_pad = 1;
if (INTEL_NBOR_PAD > 1)
_host_pad = INTEL_NBOR_PAD * sizeof(float) / sizeof(flt_t);
#ifdef _LMP_INTEL_OFFLOAD
if (_cop < 0) return;
FC_PACKED0_T *op2 = fc.p2[0];
FC_PACKED1_T *op2f = fc.p2f[0];
FC_PACKED2_T *op2e = fc.p2e[0];
FC_PACKED3_T *op3 = fc.p3[0][0];
flt_t * ocutneighsq = cutneighsq[0];
int tp1sq = tp1 * tp1;
int tp1cu = tp1sq * tp1;
if (op2 != NULL && op2f != NULL && op2e != NULL && op3 != NULL &&
ocutneighsq != NULL) {
#pragma offload_transfer target(mic:_cop) \
in(op2,op2f,op2e: length(tp1sq) alloc_if(0) free_if(0)) \
in(op3: length(tp1cu) alloc_if(0) free_if(0)) \
in(ocutneighsq: length(tp1sq))
}
#endif
}
void PairLJCharmmCoulLongIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
int tp1 = atom->ntypes + 1;
int ntable = 1;
if (ncoultablebits)
for (int i = 0; i < ncoultablebits; i++) ntable *= 2;
fc.set_ntypes(tp1, ntable, memory, _cop);
buffers->set_ntypes(tp1);
flt_t **cutneighsq = buffers->get_cutneighsq();
// Repeat cutsq calculation because done after call to init_style
double cut, cutneigh;
if (cut_lj > cut_coul)
error->all(FLERR,
"Intel varient of lj/charmm/coul/long expects lj cutoff<=coulombic");
for (int i = 1; i <= atom->ntypes; i++) {
for (int j = i; j <= atom->ntypes; j++) {
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
cut = init_one(i, j);
cutneigh = cut + neighbor->skin;
cutsq[i][j] = cutsq[j][i] = cut*cut;
cutneighsq[i][j] = cutneighsq[j][i] = cutneigh * cutneigh;
}
}
}
cut_lj_innersq = cut_lj_inner * cut_lj_inner;
cut_ljsq = cut_lj * cut_lj;
cut_coulsq = cut_coul * cut_coul;
cut_bothsq = MAX(cut_ljsq, cut_coulsq);
fc.g_ewald = force->kspace->g_ewald;
fc.tabinnersq = tabinnersq;
fc.cut_coulsq = cut_coulsq;
fc.cut_ljsq = cut_ljsq;
fc.cut_lj_innersq = cut_lj_innersq;
for (int i = 0; i < 4; i++) {
fc.special_lj[i] = force->special_lj[i];
fc.special_coul[i] = force->special_coul[i];
fc.special_coul[0] = 1.0;
fc.special_lj[0] = 1.0;
}
for (int i = 0; i < tp1; i++) {
for (int j = 0; j < tp1; j++) {
fc.lj[i][j].x = lj1[i][j];
fc.lj[i][j].y = lj2[i][j];
fc.lj[i][j].z = lj3[i][j];
fc.lj[i][j].w = lj4[i][j];
fc.cutsq[i][j] = cutsq[i][j];
}
}
if (ncoultablebits) {
for (int i = 0; i < ntable; i++) {
fc.table[i].r = rtable[i];
fc.table[i].dr = drtable[i];
fc.table[i].f = ftable[i];
fc.table[i].df = dftable[i];
fc.etable[i] = etable[i];
fc.detable[i] = detable[i];
fc.ctable[i] = ctable[i];
fc.dctable[i] = dctable[i];
}
}
#ifdef _LMP_INTEL_OFFLOAD
if (_cop < 0) return;
flt_t * special_lj = fc.special_lj;
flt_t * special_coul = fc.special_coul;
flt_t * cutsq = fc.cutsq[0];
LJ_T * lj = fc.lj[0];
TABLE_T * table = fc.table;
flt_t * etable = fc.etable;
flt_t * detable = fc.detable;
flt_t * ctable = fc.ctable;
flt_t * dctable = fc.dctable;
flt_t * ocutneighsq = cutneighsq[0];
int tp1sq = tp1 * tp1;
#pragma offload_transfer target(mic:_cop) \
in(special_lj, special_coul: length(4) alloc_if(0) free_if(0)) \
in(cutsq,lj: length(tp1sq) alloc_if(0) free_if(0)) \
in(table: length(ntable) alloc_if(0) free_if(0)) \
in(etable,detable,ctable,dctable: length(ntable) alloc_if(0) free_if(0)) \
in(ocutneighsq: length(tp1sq) alloc_if(0) free_if(0))
#endif
}
void PairBuckCoulLongIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
int off_ccache = 0;
#ifdef _LMP_INTEL_OFFLOAD
if (_cop >= 0) off_ccache = 1;
#endif
buffers->grow_ccache(off_ccache, comm->nthreads, 1);
_ccache_stride = buffers->ccache_stride();
int tp1 = atom->ntypes + 1;
int ntable = 1;
if (ncoultablebits)
for (int i = 0; i < ncoultablebits; i++) ntable *= 2;
fc.set_ntypes(tp1, ntable, memory, _cop);
buffers->set_ntypes(tp1);
flt_t **cutneighsq = buffers->get_cutneighsq();
// Repeat cutsq calculation because done after call to init_style
double cut, cutneigh;
for (int i = 1; i <= atom->ntypes; i++) {
for (int j = i; j <= atom->ntypes; j++) {
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
cut = init_one(i, j);
cutneigh = cut + neighbor->skin;
cutsq[i][j] = cutsq[j][i] = cut*cut;
cutneighsq[i][j] = cutneighsq[j][i] = cutneigh * cutneigh;
}
}
}
fc.g_ewald = force->kspace->g_ewald;
fc.tabinnersq = tabinnersq;
for (int i = 0; i < 4; i++) {
fc.special_lj[i] = force->special_lj[i];
fc.special_coul[i] = force->special_coul[i];
fc.special_coul[0] = 1.0;
fc.special_lj[0] = 1.0;
}
for (int i = 0; i < tp1; i++) {
for (int j = 0; j < tp1; j++) {
if (cutsq[i][j] < cut_ljsq[i][j])
error->all(FLERR,
"Intel variant of lj/buck/coul/long expects lj cutoff<=coulombic");
fc.c_force[i][j].cutsq = cutsq[i][j];
fc.c_force[i][j].cut_ljsq = cut_ljsq[i][j];
fc.c_force[i][j].buck1 = buck1[i][j];
fc.c_force[i][j].buck2 = buck2[i][j];
fc.rho_inv[i][j] = rhoinv[i][j];
fc.c_energy[i][j].a = a[i][j];
fc.c_energy[i][j].c = c[i][j];
fc.c_energy[i][j].offset = offset[i][j];
fc.c_energy[i][j].pad = rhoinv[i][j];
}
}
if (ncoultablebits) {
for (int i = 0; i < ntable; i++) {
fc.table[i].r = rtable[i];
fc.table[i].dr = drtable[i];
fc.table[i].f = ftable[i];
fc.table[i].df = dftable[i];
fc.etable[i] = etable[i];
fc.detable[i] = detable[i];
fc.ctable[i] = ctable[i];
fc.dctable[i] = dctable[i];
}
}
#ifdef _LMP_INTEL_OFFLOAD
if (_cop < 0) return;
flt_t * special_lj = fc.special_lj;
flt_t * special_coul = fc.special_coul;
C_FORCE_T * c_force = fc.c_force[0];
C_ENERGY_T * c_energy = fc.c_energy[0];
TABLE_T * table = fc.table;
flt_t * rho_inv = fc.rho_inv[0];
flt_t * etable = fc.etable;
flt_t * detable = fc.detable;
flt_t * ctable = fc.ctable;
flt_t * dctable = fc.dctable;
flt_t * ocutneighsq = cutneighsq[0];
int tp1sq = tp1 * tp1;
#pragma offload_transfer target(mic:_cop) \
in(special_lj, special_coul: length(4) alloc_if(0) free_if(0)) \
in(c_force, c_energy: length(tp1sq) alloc_if(0) free_if(0)) \
in(rho_inv: length(tp1sq) alloc_if(0) free_if(0)) \
in(table: length(ntable) alloc_if(0) free_if(0)) \
in(etable,detable,ctable,dctable: length(ntable) alloc_if(0) free_if(0)) \
in(ocutneighsq: length(tp1sq) alloc_if(0) free_if(0))
#endif
}
void PairLJCharmmCoulCharmmIntel::pack_force_const(ForceConst<flt_t> &fc,
IntelBuffers<flt_t,acc_t> *buffers)
{
int off_ccache = 0;
#ifdef _LMP_INTEL_OFFLOAD
if (_cop >= 0) off_ccache = 1;
#endif
buffers->grow_ccache(off_ccache, comm->nthreads, 1);
_ccache_stride = buffers->ccache_stride();
int tp1 = atom->ntypes + 1;
fc.set_ntypes(tp1, memory, _cop);
buffers->set_ntypes(tp1);
flt_t **cutneighsq = buffers->get_cutneighsq();
// Repeat cutsq calculation because done after call to init_style
double cut, cutneigh;
if (cut_lj > cut_coul)
error->all(FLERR,
"Intel varient of lj/charmm/coul/long expects lj cutoff<=coulombic");
for (int i = 1; i <= atom->ntypes; i++) {
for (int j = i; j <= atom->ntypes; j++) {
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0)) {
cut = init_one(i, j);
cutneigh = cut + neighbor->skin;
cutsq[i][j] = cutsq[j][i] = cut*cut;
cutneighsq[i][j] = cutneighsq[j][i] = cutneigh * cutneigh;
}
}
}
cut_coul_innersq = cut_coul_inner * cut_coul_inner;
cut_lj_innersq = cut_lj_inner * cut_lj_inner;
cut_ljsq = cut_lj * cut_lj;
cut_coulsq = cut_coul * cut_coul;
cut_bothsq = MAX(cut_ljsq, cut_coulsq);
fc.cut_coulsq = cut_coulsq;
fc.cut_ljsq = cut_ljsq;
fc.cut_coul_innersq = cut_coul_innersq;
fc.cut_lj_innersq = cut_lj_innersq;
for (int i = 0; i < 4; i++) {
fc.special_lj[i] = force->special_lj[i];
fc.special_coul[i] = force->special_coul[i];
fc.special_coul[0] = 1.0;
fc.special_lj[0] = 1.0;
}
for (int i = 1; i < tp1; i++) {
for (int j = 1; j < tp1; j++) {
fc.lj[i][j].x = lj1[i][j];
fc.lj[i][j].y = lj2[i][j];
fc.lj[i][j].z = lj3[i][j];
fc.lj[i][j].w = lj4[i][j];
fc.cutsq[i][j] = cutsq[i][j];
}
}
#ifdef _LMP_INTEL_OFFLOAD
if (_cop < 0) return;
flt_t * special_lj = fc.special_lj;
flt_t * special_coul = fc.special_coul;
flt_t * cutsq = fc.cutsq[0];
LJ_T * lj = fc.lj[0];
flt_t * ocutneighsq = cutneighsq[0];
int tp1sq = tp1 * tp1;
#pragma offload_transfer target(mic:_cop) \
in(special_lj, special_coul: length(4) alloc_if(0) free_if(0)) \
in(cutsq,lj: length(tp1sq) alloc_if(0) free_if(0)) \
in(ocutneighsq: length(tp1sq) alloc_if(0) free_if(0))
#endif
}