const nbnxn_atomdata_t::Params& nbatParams,
const DeviceContext& deviceContext)
{
- int ntypes;
-
- ntypes = nbatParams.numTypes;
+ const int ntypes = nbatParams.numTypes;
set_cutoff_parameters(nbp, ic, listParams);
* combination is rarely used. LJ force-switch with LB rule is more common,
* but gives only 1% speed-up.
*/
- if (ic->vdwtype == evdwCUT)
- {
- switch (ic->vdw_modifier)
- {
- case eintmodNONE:
- case eintmodPOTSHIFT:
- switch (nbatParams.comb_rule)
- {
- case ljcrNONE: nbp->vdwtype = evdwTypeCUT; break;
- case ljcrGEOM: nbp->vdwtype = evdwTypeCUTCOMBGEOM; break;
- case ljcrLB: nbp->vdwtype = evdwTypeCUTCOMBLB; break;
- default:
- gmx_incons(
- "The requested LJ combination rule is not implemented in the CUDA "
- "GPU accelerated kernels!");
- }
- break;
- case eintmodFORCESWITCH: nbp->vdwtype = evdwTypeFSWITCH; break;
- case eintmodPOTSWITCH: nbp->vdwtype = evdwTypePSWITCH; break;
- default:
- gmx_incons(
- "The requested VdW interaction modifier is not implemented in the CUDA GPU "
- "accelerated kernels!");
- }
- }
- else if (ic->vdwtype == evdwPME)
- {
- if (ic->ljpme_comb_rule == ljcrGEOM)
- {
- assert(nbatParams.comb_rule == ljcrGEOM);
- nbp->vdwtype = evdwTypeEWALDGEOM;
- }
- else
- {
- assert(nbatParams.comb_rule == ljcrLB);
- nbp->vdwtype = evdwTypeEWALDLB;
- }
- }
- else
- {
- gmx_incons(
- "The requested VdW type is not implemented in the CUDA GPU accelerated kernels!");
- }
-
- if (ic->eeltype == eelCUT)
- {
- nbp->eeltype = eelTypeCUT;
- }
- else if (EEL_RF(ic->eeltype))
- {
- nbp->eeltype = eelTypeRF;
- }
- else if ((EEL_PME(ic->eeltype) || ic->eeltype == eelEWALD))
- {
- nbp->eeltype = nbnxn_gpu_pick_ewald_kernel_type(*ic, deviceContext.deviceInfo());
- }
- else
- {
- /* Shouldn't happen, as this is checked when choosing Verlet-scheme */
- gmx_incons(
- "The requested electrostatics type is not implemented in the CUDA GPU accelerated "
- "kernels!");
- }
+ nbp->vdwType = nbnxmGpuPickVdwKernelType(ic, nbatParams.comb_rule);
+ nbp->elecType = nbnxmGpuPickElectrostaticsKernelType(ic, deviceContext.deviceInfo());
/* generate table for PME */
nbp->coulomb_tab = nullptr;
- if (nbp->eeltype == eelTypeEWALD_TAB || nbp->eeltype == eelTypeEWALD_TAB_TWIN)
+ if (nbp->elecType == ElecType::EwaldTab || nbp->elecType == ElecType::EwaldTabTwin)
{
GMX_RELEASE_ASSERT(ic->coulombEwaldTables, "Need valid Coulomb Ewald correction tables");
init_ewald_coulomb_force_table(*ic->coulombEwaldTables, nbp, deviceContext);
}
/* set up LJ parameter lookup table */
- if (!useLjCombRule(nbp->vdwtype))
+ if (!useLjCombRule(nbp->vdwType))
{
- initParamLookupTable(&nbp->nbfp, &nbp->nbfp_texobj, nbatParams.nbfp.data(),
- 2 * ntypes * ntypes, deviceContext);
+ initParamLookupTable(
+ &nbp->nbfp, &nbp->nbfp_texobj, nbatParams.nbfp.data(), 2 * ntypes * ntypes, deviceContext);
}
/* set up LJ-PME parameter lookup table */
if (ic->vdwtype == evdwPME)
{
- initParamLookupTable(&nbp->nbfp_comb, &nbp->nbfp_comb_texobj, nbatParams.nbfp_comb.data(),
- 2 * ntypes, deviceContext);
+ initParamLookupTable(
+ &nbp->nbfp_comb, &nbp->nbfp_comb_texobj, nbatParams.nbfp_comb.data(), 2 * ntypes, deviceContext);
}
}
{
static_assert(sizeof(adat->shift_vec[0]) == sizeof(nbatom->shift_vec[0]),
"Sizes of host- and device-side shift vectors should be the same.");
- copyToDeviceBuffer(&adat->shift_vec, reinterpret_cast<const float3*>(nbatom->shift_vec.data()),
- 0, SHIFTS, localStream, GpuApiCallBehavior::Async, nullptr);
+ copyToDeviceBuffer(&adat->shift_vec,
+ reinterpret_cast<const float3*>(nbatom->shift_vec.data()),
+ 0,
+ SHIFTS,
+ localStream,
+ GpuApiCallBehavior::Async,
+ nullptr);
adat->bShiftVecUploaded = true;
}
}
allocateDeviceBuffer(&d_atdat->f, nalloc, deviceContext);
allocateDeviceBuffer(&d_atdat->xq, nalloc, deviceContext);
- if (useLjCombRule(nb->nbparam->vdwtype))
+ if (useLjCombRule(nb->nbparam->vdwType))
{
allocateDeviceBuffer(&d_atdat->lj_comb, nalloc, deviceContext);
}
nbnxn_cuda_clear_f(nb, nalloc);
}
- if (useLjCombRule(nb->nbparam->vdwtype))
+ if (useLjCombRule(nb->nbparam->vdwType))
{
static_assert(sizeof(d_atdat->lj_comb[0]) == sizeof(float2),
"Size of the LJ parameters element should be equal to the size of float2.");
copyToDeviceBuffer(&d_atdat->lj_comb,
- reinterpret_cast<const float2*>(nbat->params().lj_comb.data()), 0,
- natoms, localStream, GpuApiCallBehavior::Async, nullptr);
+ reinterpret_cast<const float2*>(nbat->params().lj_comb.data()),
+ 0,
+ natoms,
+ localStream,
+ GpuApiCallBehavior::Async,
+ nullptr);
}
else
{
static_assert(sizeof(d_atdat->atom_types[0]) == sizeof(nbat->params().type[0]),
"Sizes of host- and device-side atom types should be the same.");
- copyToDeviceBuffer(&d_atdat->atom_types, nbat->params().type.data(), 0, natoms, localStream,
- GpuApiCallBehavior::Async, nullptr);
+ copyToDeviceBuffer(&d_atdat->atom_types,
+ nbat->params().type.data(),
+ 0,
+ natoms,
+ localStream,
+ GpuApiCallBehavior::Async,
+ nullptr);
}
if (bDoTime)
nbparam = nb->nbparam;
if ((!nbparam->coulomb_tab)
- && (nbparam->eeltype == eelTypeEWALD_TAB || nbparam->eeltype == eelTypeEWALD_TAB_TWIN))
+ && (nbparam->elecType == ElecType::EwaldTab || nbparam->elecType == ElecType::EwaldTabTwin))
{
destroyParamLookupTable(&nbparam->coulomb_tab, nbparam->coulomb_tab_texobj);
}
delete nb->timers;
- if (!useLjCombRule(nb->nbparam->vdwtype))
+ if (!useLjCombRule(nb->nbparam->vdwType))
{
destroyParamLookupTable(&nbparam->nbfp, nbparam->nbfp_texobj);
}
- if (nbparam->vdwtype == evdwTypeEWALDGEOM || nbparam->vdwtype == evdwTypeEWALDLB)
+ if (nbparam->vdwType == VdwType::EwaldGeom || nbparam->vdwType == VdwType::EwaldLB)
{
destroyParamLookupTable(&nbparam->nbfp_comb, nbparam->nbfp_comb_texobj);
}
bool bDoTime = gpu_nbv->bDoTime;
const int maxNumColumns = gridSet.numColumnsMax();
- reallocateDeviceBuffer(&gpu_nbv->cxy_na, maxNumColumns * gridSet.grids().size(),
- &gpu_nbv->ncxy_na, &gpu_nbv->ncxy_na_alloc, *gpu_nbv->deviceContext_);
- reallocateDeviceBuffer(&gpu_nbv->cxy_ind, maxNumColumns * gridSet.grids().size(),
- &gpu_nbv->ncxy_ind, &gpu_nbv->ncxy_ind_alloc, *gpu_nbv->deviceContext_);
+ reallocateDeviceBuffer(&gpu_nbv->cxy_na,
+ maxNumColumns * gridSet.grids().size(),
+ &gpu_nbv->ncxy_na,
+ &gpu_nbv->ncxy_na_alloc,
+ *gpu_nbv->deviceContext_);
+ reallocateDeviceBuffer(&gpu_nbv->cxy_ind,
+ maxNumColumns * gridSet.grids().size(),
+ &gpu_nbv->ncxy_ind,
+ &gpu_nbv->ncxy_ind_alloc,
+ *gpu_nbv->deviceContext_);
for (unsigned int g = 0; g < gridSet.grids().size(); g++)
{
const int* cxy_na = grid.cxy_na().data();
const int* cxy_ind = grid.cxy_ind().data();
- reallocateDeviceBuffer(&gpu_nbv->atomIndices, atomIndicesSize, &gpu_nbv->atomIndicesSize,
- &gpu_nbv->atomIndicesSize_alloc, *gpu_nbv->deviceContext_);
+ reallocateDeviceBuffer(&gpu_nbv->atomIndices,
+ atomIndicesSize,
+ &gpu_nbv->atomIndicesSize,
+ &gpu_nbv->atomIndicesSize_alloc,
+ *gpu_nbv->deviceContext_);
if (atomIndicesSize > 0)
{
gpu_nbv->timers->xf[AtomLocality::Local].nb_h2d.openTimingRegion(deviceStream);
}
- copyToDeviceBuffer(&gpu_nbv->atomIndices, atomIndices, 0, atomIndicesSize, deviceStream,
- GpuApiCallBehavior::Async, nullptr);
+ copyToDeviceBuffer(&gpu_nbv->atomIndices,
+ atomIndices,
+ 0,
+ atomIndicesSize,
+ deviceStream,
+ GpuApiCallBehavior::Async,
+ nullptr);
if (bDoTime)
{
}
int* destPtr = &gpu_nbv->cxy_na[maxNumColumns * g];
- copyToDeviceBuffer(&destPtr, cxy_na, 0, numColumns, deviceStream,
- GpuApiCallBehavior::Async, nullptr);
+ copyToDeviceBuffer(
+ &destPtr, cxy_na, 0, numColumns, deviceStream, GpuApiCallBehavior::Async, nullptr);
if (bDoTime)
{
}
destPtr = &gpu_nbv->cxy_ind[maxNumColumns * g];
- copyToDeviceBuffer(&destPtr, cxy_ind, 0, numColumns, deviceStream,
- GpuApiCallBehavior::Async, nullptr);
+ copyToDeviceBuffer(
+ &destPtr, cxy_ind, 0, numColumns, deviceStream, GpuApiCallBehavior::Async, nullptr);
if (bDoTime)
{
biod.pnpi.spb.ru / alexxy / gromacs.git / blobdiff