void
nbnxn_kernel_simd_2xnn(nbnxn_pairlist_set_t gmx_unused *nbl_list,
const nbnxn_atomdata_t gmx_unused *nbat,
const interaction_const_t gmx_unused *ic,
int gmx_unused ewald_excl,
rvec gmx_unused *shift_vec,
int gmx_unused force_flags,
int gmx_unused clearF,
real gmx_unused *fshift,
real gmx_unused *Vc,
real gmx_unused *Vvdw)
#ifdef GMX_NBNXN_SIMD_2XNN
{
int nnbl;
nbnxn_pairlist_t **nbl;
int coulkt, vdwkt = 0;
int nb;
int nthreads gmx_unused;
nnbl = nbl_list->nnbl;
nbl = nbl_list->nbl;
if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
{
coulkt = coulktRF;
}
else
{
if (ewald_excl == ewaldexclTable)
{
if (ic->rcoulomb == ic->rvdw)
{
coulkt = coulktTAB;
}
else
{
coulkt = coulktTAB_TWIN;
}
}
else
{
if (ic->rcoulomb == ic->rvdw)
{
coulkt = coulktEWALD;
}
else
{
coulkt = coulktEWALD_TWIN;
}
}
}
if (ic->vdwtype == evdwCUT)
{
switch (ic->vdw_modifier)
{
case eintmodNONE:
case eintmodPOTSHIFT:
switch (nbat->comb_rule)
{
case ljcrGEOM: vdwkt = vdwktLJCUT_COMBGEOM; break;
case ljcrLB: vdwkt = vdwktLJCUT_COMBLB; break;
case ljcrNONE: vdwkt = vdwktLJCUT_COMBNONE; break;
default: gmx_incons("Unknown combination rule");
}
break;
case eintmodFORCESWITCH:
vdwkt = vdwktLJFORCESWITCH;
break;
case eintmodPOTSWITCH:
vdwkt = vdwktLJPOTSWITCH;
break;
default:
gmx_incons("Unsupported VdW interaction modifier");
}
}
else if (ic->vdwtype == evdwPME)
{
if (ic->ljpme_comb_rule == eljpmeLB)
{
gmx_incons("The nbnxn SIMD kernels don't suport LJ-PME with LB");
}
vdwkt = vdwktLJEWALDCOMBGEOM;
}
else
{
gmx_incons("Unsupported VdW interaction type");
}
nthreads = gmx_omp_nthreads_get(emntNonbonded);
#pragma omp parallel for schedule(static) num_threads(nthreads)
for (nb = 0; nb < nnbl; nb++)
{
nbnxn_atomdata_output_t *out;
real *fshift_p;
out = &nbat->out[nb];
if (clearF == enbvClearFYes)
{
clear_f(nbat, nb, out->f);
}
if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
{
fshift_p = fshift;
}
else
{
fshift_p = out->fshift;
if (clearF == enbvClearFYes)
{
clear_fshift(fshift_p);
}
}
if (!(force_flags & GMX_FORCE_ENERGY))
{
/* Don't calculate energies */
p_nbk_noener[coulkt][vdwkt](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p);
}
else if (out->nV == 1)
{
/* No energy groups */
out->Vvdw[0] = 0;
out->Vc[0] = 0;
p_nbk_ener[coulkt][vdwkt](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p,
out->Vvdw,
out->Vc);
}
else
{
/* Calculate energy group contributions */
int i;
for (i = 0; i < out->nVS; i++)
{
out->VSvdw[i] = 0;
}
for (i = 0; i < out->nVS; i++)
{
out->VSc[i] = 0;
}
p_nbk_energrp[coulkt][vdwkt](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p,
out->VSvdw,
out->VSc);
reduce_group_energies(nbat->nenergrp, nbat->neg_2log,
out->VSvdw, out->VSc,
out->Vvdw, out->Vc);
}
}
if (force_flags & GMX_FORCE_ENERGY)
{
reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
}
}
void
nbnxn_kernel_ref(const nbnxn_pairlist_set_t *nbl_list,
const nbnxn_atomdata_t *nbat,
const interaction_const_t *ic,
rvec *shift_vec,
int force_flags,
int clearF,
real *fshift,
real *Vc,
real *Vvdw)
{
int nnbl;
nbnxn_pairlist_t **nbl;
int coult;
int nb;
nnbl = nbl_list->nnbl;
nbl = nbl_list->nbl;
if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
{
coult = coultRF;
}
else
{
if (ic->rcoulomb == ic->rvdw)
{
coult = coultTAB;
}
else
{
coult = coultTAB_TWIN;
}
}
#pragma omp parallel for schedule(static) num_threads(gmx_omp_nthreads_get(emntNonbonded))
for (nb = 0; nb < nnbl; nb++)
{
nbnxn_atomdata_output_t *out;
real *fshift_p;
out = &nbat->out[nb];
if (clearF == enbvClearFYes)
{
clear_f(nbat, nb, out->f);
}
if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
{
fshift_p = fshift;
}
else
{
fshift_p = out->fshift;
if (clearF == enbvClearFYes)
{
clear_fshift(fshift_p);
}
}
if (!(force_flags & GMX_FORCE_ENERGY))
{
/* Don't calculate energies */
p_nbk_c_noener[coult](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p);
}
else if (out->nV == 1)
{
/* No energy groups */
out->Vvdw[0] = 0;
out->Vc[0] = 0;
p_nbk_c_ener[coult](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p,
out->Vvdw,
out->Vc);
}
else
{
/* Calculate energy group contributions */
int i;
for (i = 0; i < out->nV; i++)
{
out->Vvdw[i] = 0;
}
for (i = 0; i < out->nV; i++)
{
out->Vc[i] = 0;
}
p_nbk_c_energrp[coult](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p,
out->Vvdw,
out->Vc);
}
}
if (force_flags & GMX_FORCE_ENERGY)
{
reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
}
}
void
nbnxn_kernel_ref(const nbnxn_pairlist_set_t *nbl_list,
const nbnxn_atomdata_t *nbat,
const interaction_const_t *ic,
rvec *shift_vec,
int force_flags,
int clearF,
real *fshift,
real *Vc,
real *Vvdw)
{
int nnbl;
nbnxn_pairlist_t **nbl;
int coult;
int vdwt;
int nb;
int nthreads gmx_unused;
nnbl = nbl_list->nnbl;
nbl = nbl_list->nbl;
if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
{
coult = coultRF;
}
else
{
if (ic->rcoulomb == ic->rvdw)
{
coult = coultTAB;
}
else
{
coult = coultTAB_TWIN;
}
}
if (ic->vdwtype == evdwCUT)
{
switch (ic->vdw_modifier)
{
case eintmodPOTSHIFT:
case eintmodNONE:
vdwt = vdwtCUT;
break;
case eintmodFORCESWITCH:
vdwt = vdwtFSWITCH;
break;
case eintmodPOTSWITCH:
vdwt = vdwtPSWITCH;
break;
default:
gmx_incons("Unsupported VdW modifier");
break;
}
}
else if (ic->vdwtype == evdwPME)
{
if (ic->ljpme_comb_rule == ljcrGEOM)
{
assert(nbat->comb_rule == ljcrGEOM);
vdwt = vdwtEWALDGEOM;
}
else
{
assert(nbat->comb_rule == ljcrLB);
vdwt = vdwtEWALDLB;
}
}
else
{
gmx_incons("Unsupported vdwtype in nbnxn reference kernel");
}
// cppcheck-suppress unreadVariable
nthreads = gmx_omp_nthreads_get(emntNonbonded);
#pragma omp parallel for schedule(static) num_threads(nthreads)
for (nb = 0; nb < nnbl; nb++)
{
// Presently, the kernels do not call C++ code that can throw, so
// no need for a try/catch pair in this OpenMP region.
nbnxn_atomdata_output_t *out;
real *fshift_p;
out = &nbat->out[nb];
if (clearF == enbvClearFYes)
{
clear_f(nbat, nb, out->f);
}
if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
{
fshift_p = fshift;
}
else
{
fshift_p = out->fshift;
if (clearF == enbvClearFYes)
{
clear_fshift(fshift_p);
}
}
if (!(force_flags & GMX_FORCE_ENERGY))
{
/* Don't calculate energies */
p_nbk_c_noener[coult][vdwt](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p);
}
else if (out->nV == 1)
{
/* No energy groups */
out->Vvdw[0] = 0;
out->Vc[0] = 0;
p_nbk_c_ener[coult][vdwt](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p,
out->Vvdw,
out->Vc);
}
else
{
/* Calculate energy group contributions */
int i;
for (i = 0; i < out->nV; i++)
{
out->Vvdw[i] = 0;
}
for (i = 0; i < out->nV; i++)
{
out->Vc[i] = 0;
}
p_nbk_c_energrp[coult][vdwt](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p,
out->Vvdw,
out->Vc);
}
}
if (force_flags & GMX_FORCE_ENERGY)
{
reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
}
}
void
nbnxn_kernel_simd_2xnn(nbnxn_pairlist_set_t *nbl_list,
const nbnxn_atomdata_t *nbat,
const interaction_const_t *ic,
int ewald_excl,
rvec *shift_vec,
int force_flags,
int clearF,
real *fshift,
real *Vc,
real *Vvdw)
#ifdef GMX_NBNXN_SIMD_2XNN
{
int nnbl;
nbnxn_pairlist_t **nbl;
int coult;
int nb;
nnbl = nbl_list->nnbl;
nbl = nbl_list->nbl;
if (EEL_RF(ic->eeltype) || ic->eeltype == eelCUT)
{
coult = coultRF;
}
else
{
if (ewald_excl == ewaldexclTable)
{
if (ic->rcoulomb == ic->rvdw)
{
coult = coultTAB;
}
else
{
coult = coultTAB_TWIN;
}
}
else
{
if (ic->rcoulomb == ic->rvdw)
{
coult = coultEWALD;
}
else
{
coult = coultEWALD_TWIN;
}
}
}
#pragma omp parallel for schedule(static) num_threads(gmx_omp_nthreads_get(emntNonbonded))
for (nb = 0; nb < nnbl; nb++)
{
nbnxn_atomdata_output_t *out;
real *fshift_p;
out = &nbat->out[nb];
if (clearF == enbvClearFYes)
{
clear_f(nbat, nb, out->f);
}
if ((force_flags & GMX_FORCE_VIRIAL) && nnbl == 1)
{
fshift_p = fshift;
}
else
{
fshift_p = out->fshift;
if (clearF == enbvClearFYes)
{
clear_fshift(fshift_p);
}
}
/* With Ewald type electrostatics we the forces for excluded atom pairs
* should not contribute to the virial sum. The exclusion forces
* are not calculate in the energy kernels, but are in _noener.
*/
if (!((force_flags & GMX_FORCE_ENERGY) ||
(EEL_FULL(ic->eeltype) && (force_flags & GMX_FORCE_VIRIAL))))
{
/* Don't calculate energies */
p_nbk_noener[coult][nbat->comb_rule](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p);
}
else if (out->nV == 1 || !(force_flags & GMX_FORCE_ENERGY))
{
/* No energy groups */
out->Vvdw[0] = 0;
out->Vc[0] = 0;
p_nbk_ener[coult][nbat->comb_rule](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p,
out->Vvdw,
out->Vc);
}
else
{
/* Calculate energy group contributions */
int i;
for (i = 0; i < out->nVS; i++)
{
out->VSvdw[i] = 0;
}
for (i = 0; i < out->nVS; i++)
{
out->VSc[i] = 0;
}
p_nbk_energrp[coult][nbat->comb_rule](nbl[nb], nbat,
ic,
shift_vec,
out->f,
fshift_p,
out->VSvdw,
out->VSc);
reduce_group_energies(nbat->nenergrp, nbat->neg_2log,
out->VSvdw, out->VSc,
out->Vvdw, out->Vc);
}
}
if (force_flags & GMX_FORCE_ENERGY)
{
reduce_energies_over_lists(nbat, nnbl, Vvdw, Vc);
}
}
