tx258/lammps-gran-kokkos
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add NeedDup_v and AtomicDup_v helpers to KOKKOS

Browse Source
This commit is contained in:
Richard Berger
2022-01-31 13:42:03 -05:00
parent dd416f61b5
commit b2916339a4
13 changed files with 133 additions and 127 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/KOKKOS/fix_acks2_reaxff_kokkos.cpp
View File

@ -864,8 +864,8 @@ KOKKOS_INLINE_FUNCTION
void FixACKS2ReaxFFKokkos<DeviceType>::compute_x_item(int ii, int &m_fill, const bool &final) const
{
// The X_diag array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_X_diag = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_X_diag),decltype(ndup_X_diag)>::get(dup_X_diag,ndup_X_diag);
auto a_X_diag = v_X_diag.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_X_diag = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_X_diag),decltype(ndup_X_diag)>::get(dup_X_diag,ndup_X_diag);
auto a_X_diag = v_X_diag.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
int j,jj,jtype;
@ -942,8 +942,8 @@ void FixACKS2ReaxFFKokkos<DeviceType>::compute_x_team(
int atoms_per_team, int vector_length) const {
// The X_diag array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_X_diag = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_X_diag),decltype(ndup_X_diag)>::get(dup_X_diag,ndup_X_diag);
auto a_X_diag = v_X_diag.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_X_diag = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_X_diag),decltype(ndup_X_diag)>::get(dup_X_diag,ndup_X_diag);
auto a_X_diag = v_X_diag.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
// scratch space setup
Kokkos::View<int *, Kokkos::ScratchMemorySpace<DeviceType>,
@ -1460,8 +1460,8 @@ KOKKOS_INLINE_FUNCTION
void FixACKS2ReaxFFKokkos<DeviceType>::operator() (TagACKS2SparseMatvec3_Half<NEIGHFLAG>, const int &ii) const
{
// The bb array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_bb = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_bb),decltype(ndup_bb)>::get(dup_bb,ndup_bb);
auto a_bb = v_bb.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_bb = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_bb),decltype(ndup_bb)>::get(dup_bb,ndup_bb);
auto a_bb = v_bb.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
if (mask[i] & groupbit) {

12
src/KOKKOS/fix_qeq_reaxff_kokkos.cpp
View File

@ -1041,8 +1041,8 @@ KOKKOS_INLINE_FUNCTION
void FixQEqReaxFFKokkos<DeviceType>::sparse13_item(int ii) const
{
// The q array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_o = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);
auto a_o = v_o.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_o = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);
auto a_o = v_o.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
if (mask[i] & groupbit) {
@ -1094,8 +1094,8 @@ KOKKOS_INLINE_FUNCTION
void FixQEqReaxFFKokkos<DeviceType>::sparse23_item(int ii) const
{
// The q array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_o = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);
auto a_o = v_o.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_o = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);
auto a_o = v_o.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
if (mask[i] & groupbit) {
@ -1154,8 +1154,8 @@ KOKKOS_INLINE_FUNCTION
void FixQEqReaxFFKokkos<DeviceType>::sparse33_item(int ii) const
{
// The q array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_o = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);
auto a_o = v_o.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_o = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);
auto a_o = v_o.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
if (mask[i] & groupbit) {

20
src/KOKKOS/fix_shake_kokkos.cpp
View File

@ -590,8 +590,8 @@ void FixShakeKokkos<DeviceType>::shake(int m, EV_FLOAT& ev) const
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
int nlist,list[2];
double v[6];
@ -701,8 +701,8 @@ void FixShakeKokkos<DeviceType>::shake3(int m, EV_FLOAT& ev) const
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
int nlist,list[3];
double v[6];
@ -884,8 +884,8 @@ void FixShakeKokkos<DeviceType>::shake4(int m, EV_FLOAT& ev) const
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
int nlist,list[4];
double v[6];
@ -1146,8 +1146,8 @@ void FixShakeKokkos<DeviceType>::shake3angle(int m, EV_FLOAT& ev) const
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
int nlist,list[3];
double v[6];
@ -1744,8 +1744,8 @@ void FixShakeKokkos<DeviceType>::v_tally(EV_FLOAT &ev, int n, int *list, double
if (vflag_atom) {
double fraction = 1.0/total;
for (int i = 0; i < n; i++) {
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
m = list[i];
a_vatom(m,0) += fraction*v[0];
a_vatom(m,1) += fraction*v[1];

6
src/KOKKOS/kokkos_type.h
View File

@ -283,6 +283,9 @@ struct AtomicDup {
using value = Kokkos::Experimental::ScatterNonAtomic;
};
template<int NEIGHFLAG, class DeviceType>
using AtomicDup_v = typename AtomicDup<NEIGHFLAG, DeviceType>::value;
#ifdef KOKKOS_ENABLE_CUDA
template<>
struct AtomicDup<HALFTHREAD,Kokkos::Cuda> {
@ -331,6 +334,9 @@ struct NeedDup {
using value = Kokkos::Experimental::ScatterNonDuplicated;
};
template<int NEIGHFLAG, class DeviceType>
using NeedDup_v = typename NeedDup<NEIGHFLAG,DeviceType>::value;
#ifndef LMP_KOKKOS_USE_ATOMICS
#ifdef KOKKOS_ENABLE_OPENMP

16
src/KOKKOS/pair_eam_alloy_kokkos.cpp
View File

@ -553,8 +553,8 @@ void PairEAMAlloyKokkos<DeviceType>::operator()(TagPairEAMAlloyKernelA<NEIGHFLAG
// The rho array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_rho = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_rho),decltype(ndup_rho)>::get(dup_rho,ndup_rho);
auto a_rho = v_rho.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_rho = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_rho),decltype(ndup_rho)>::get(dup_rho,ndup_rho);
auto a_rho = v_rho.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const X_FLOAT xtmp = x(i,0);
@ -718,8 +718,8 @@ void PairEAMAlloyKokkos<DeviceType>::operator()(TagPairEAMAlloyKernelC<NEIGHFLAG
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const X_FLOAT xtmp = x(i,0);
@ -832,11 +832,11 @@ void PairEAMAlloyKokkos<DeviceType>::ev_tally(EV_FLOAT &ev, const int &i, const
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (EFLAG) {
if (eflag_atom) {

16
src/KOKKOS/pair_eam_fs_kokkos.cpp
View File

@ -553,8 +553,8 @@ void PairEAMFSKokkos<DeviceType>::operator()(TagPairEAMFSKernelA<NEIGHFLAG,NEWTO
// The rho array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_rho = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_rho),decltype(ndup_rho)>::get(dup_rho,ndup_rho);
auto a_rho = v_rho.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_rho = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_rho),decltype(ndup_rho)>::get(dup_rho,ndup_rho);
auto a_rho = v_rho.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const X_FLOAT xtmp = x(i,0);
@ -719,8 +719,8 @@ void PairEAMFSKokkos<DeviceType>::operator()(TagPairEAMFSKernelC<NEIGHFLAG,NEWTO
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const X_FLOAT xtmp = x(i,0);
@ -833,11 +833,11 @@ void PairEAMFSKokkos<DeviceType>::ev_tally(EV_FLOAT &ev, const int &i, const int
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (EFLAG) {
if (eflag_atom) {

16
src/KOKKOS/pair_eam_kokkos.cpp
View File

@ -548,8 +548,8 @@ void PairEAMKokkos<DeviceType>::operator()(TagPairEAMKernelA<NEIGHFLAG,NEWTON_PA
// The rho array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_rho = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_rho),decltype(ndup_rho)>::get(dup_rho,ndup_rho);
auto a_rho = v_rho.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_rho = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_rho),decltype(ndup_rho)>::get(dup_rho,ndup_rho);
auto a_rho = v_rho.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const X_FLOAT xtmp = x(i,0);
@ -713,8 +713,8 @@ void PairEAMKokkos<DeviceType>::operator()(TagPairEAMKernelC<NEIGHFLAG,NEWTON_PA
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const X_FLOAT xtmp = x(i,0);
@ -827,11 +827,11 @@ void PairEAMKokkos<DeviceType>::ev_tally(EV_FLOAT &ev, const int &i, const int &
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (EFLAG) {
if (eflag_atom) {

86
src/KOKKOS/pair_reaxff_kokkos.cpp
View File

@ -1122,8 +1122,8 @@ void PairReaxFFKokkos<DeviceType>::operator()(PairReaxFFComputeLJCoulomb<NEIGHFL
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
F_FLOAT powr_vdw, powgi_vdw, fn13, dfn13, exp1, exp2, etmp;
F_FLOAT evdwl, fvdwl;
@ -1313,8 +1313,8 @@ void PairReaxFFKokkos<DeviceType>::operator()(PairReaxFFComputeTabulatedLJCoulom
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const X_FLOAT xtmp = x(i,0);
@ -1714,11 +1714,11 @@ void PairReaxFFKokkos<DeviceType>::operator()(PairReaxBuildListsHalf<NEIGHFLAG>,
if (d_resize_bo() || d_resize_hb())
return;
auto v_dDeltap_self = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_dDeltap_self),decltype(ndup_dDeltap_self)>::get(dup_dDeltap_self,ndup_dDeltap_self);
auto a_dDeltap_self = v_dDeltap_self.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_dDeltap_self = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_dDeltap_self),decltype(ndup_dDeltap_self)>::get(dup_dDeltap_self,ndup_dDeltap_self);
auto a_dDeltap_self = v_dDeltap_self.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_total_bo = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_total_bo),decltype(ndup_total_bo)>::get(dup_total_bo,ndup_total_bo);
auto a_total_bo = v_total_bo.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_total_bo = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_total_bo),decltype(ndup_total_bo)>::get(dup_total_bo,ndup_total_bo);
auto a_total_bo = v_total_bo.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const X_FLOAT xtmp = x(i,0);
@ -2142,8 +2142,8 @@ template<int NEIGHFLAG, int EVFLAG>
KOKKOS_INLINE_FUNCTION
void PairReaxFFKokkos<DeviceType>::operator()(PairReaxFFComputeMulti2<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {
auto v_CdDelta = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);
auto a_CdDelta = v_CdDelta.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_CdDelta = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);
auto a_CdDelta = v_CdDelta.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const int itype = type(i);
@ -2294,12 +2294,12 @@ template<int NEIGHFLAG, int EVFLAG>
KOKKOS_INLINE_FUNCTION
void PairReaxFFKokkos<DeviceType>::operator()(PairReaxFFComputeAngular<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
Kokkos::View<F_FLOAT**, typename DAT::t_ffloat_2d_dl::array_layout,typename KKDevice<DeviceType>::value,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_Cdbo = d_Cdbo;
auto v_CdDelta = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);
auto a_CdDelta = v_CdDelta.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_CdDelta = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);
auto a_CdDelta = v_CdDelta.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const int itype = type(i);
@ -2606,13 +2606,13 @@ template<int NEIGHFLAG, int EVFLAG>
KOKKOS_INLINE_FUNCTION
void PairReaxFFKokkos<DeviceType>::operator()(PairReaxFFComputeTorsion<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_CdDelta = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);
auto a_CdDelta = v_CdDelta.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_CdDelta = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);
auto a_CdDelta = v_CdDelta.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
Kokkos::View<F_FLOAT**, typename DAT::t_ffloat_2d_dl::array_layout,typename KKDevice<DeviceType>::value,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_Cdbo = d_Cdbo;
//auto a_Cdbo = dup_Cdbo.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
//auto a_Cdbo = dup_Cdbo.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
// in reaxff_torsion_angles: j = i, k = j, i = k;
@ -2981,8 +2981,8 @@ template<int NEIGHFLAG, int EVFLAG>
KOKKOS_INLINE_FUNCTION
void PairReaxFFKokkos<DeviceType>::operator()(PairReaxFFComputeHydrogen<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
int hblist[MAX_BONDS];
F_FLOAT theta, cos_theta, sin_xhz4, cos_xhz1, sin_theta2;
@ -3131,9 +3131,9 @@ void PairReaxFFKokkos<DeviceType>::operator()(PairReaxUpdateBond<NEIGHFLAG>, con
Kokkos::View<F_FLOAT**, typename DAT::t_ffloat_2d_dl::array_layout,typename KKDevice<DeviceType>::value,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_Cdbo = d_Cdbo;
Kokkos::View<F_FLOAT**, typename DAT::t_ffloat_2d_dl::array_layout,typename KKDevice<DeviceType>::value,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_Cdbopi = d_Cdbopi;
Kokkos::View<F_FLOAT**, typename DAT::t_ffloat_2d_dl::array_layout,typename KKDevice<DeviceType>::value,Kokkos::MemoryTraits<AtomicF<NEIGHFLAG>::value> > a_Cdbopi2 = d_Cdbopi2;
//auto a_Cdbo = dup_Cdbo.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
//auto a_Cdbopi = dup_Cdbopi.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
//auto a_Cdbopi2 = dup_Cdbopi2.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
//auto a_Cdbo = dup_Cdbo.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
//auto a_Cdbopi = dup_Cdbopi.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
//auto a_Cdbopi2 = dup_Cdbopi2.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
const tagint itag = tag(i);
@ -3180,10 +3180,10 @@ template<int NEIGHFLAG, int EVFLAG>
KOKKOS_INLINE_FUNCTION
void PairReaxFFKokkos<DeviceType>::operator()(PairReaxFFComputeBond1<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto v_CdDelta = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);
auto a_CdDelta = v_CdDelta.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_CdDelta = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_CdDelta),decltype(ndup_CdDelta)>::get(dup_CdDelta,ndup_CdDelta);
auto a_CdDelta = v_CdDelta.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
F_FLOAT p_be1, p_be2, De_s, De_p, De_pp, pow_BOs_be2, exp_be12, CEbo, ebond;
@ -3298,8 +3298,8 @@ template<int NEIGHFLAG, int EVFLAG>
KOKKOS_INLINE_FUNCTION
void PairReaxFFKokkos<DeviceType>::operator()(PairReaxFFComputeBond2<NEIGHFLAG,EVFLAG>, const int &ii, EV_FLOAT_REAX& ev) const {
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
F_FLOAT delij[3], delik[3], deljk[3], tmpvec[3];
F_FLOAT dBOp_i[3], dBOp_k[3], dln_BOp_pi[3], dln_BOp_pi2[3];
@ -3505,11 +3505,11 @@ void PairReaxFFKokkos<DeviceType>::ev_tally(EV_FLOAT_REAX &ev, const int &i, con
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (eflag_atom) {
const E_FLOAT epairhalf = 0.5 * epair;
@ -3564,8 +3564,8 @@ void PairReaxFFKokkos<DeviceType>::e_tally(EV_FLOAT_REAX & /*ev*/, const int &i,
if (eflag_atom) {
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const E_FLOAT epairhalf = 0.5 * epair;
a_eatom[i] += epairhalf;
@ -3582,8 +3582,8 @@ void PairReaxFFKokkos<DeviceType>::e_tally_single(EV_FLOAT_REAX & /*ev*/, const
const F_FLOAT &epair) const
{
// The eatom array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
a_eatom[i] += epair;
}
@ -3616,8 +3616,8 @@ void PairReaxFFKokkos<DeviceType>::v_tally(EV_FLOAT_REAX &ev, const int &i,
}
if (vflag_atom) {
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
a_vatom(i,0) += v[0]; a_vatom(i,1) += v[1]; a_vatom(i,2) += v[2];
a_vatom(i,3) += v[3]; a_vatom(i,4) += v[4]; a_vatom(i,5) += v[5];
@ -3634,8 +3634,8 @@ void PairReaxFFKokkos<DeviceType>::v_tally3(EV_FLOAT_REAX &ev, const int &i, con
{
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
F_FLOAT v[6];
@ -3696,8 +3696,8 @@ void PairReaxFFKokkos<DeviceType>::v_tally4(EV_FLOAT_REAX &ev, const int &i, con
}
if (vflag_atom) {
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
a_vatom(i,0) += 0.25 * v[0]; a_vatom(i,1) += 0.25 * v[1]; a_vatom(i,2) += 0.25 * v[2];
a_vatom(i,3) += 0.25 * v[3]; a_vatom(i,4) += 0.25 * v[4]; a_vatom(i,5) += 0.25 * v[5];

8
src/KOKKOS/pair_snap_kokkos_impl.h
View File

@ -1257,8 +1257,8 @@ KOKKOS_INLINE_FUNCTION
void PairSNAPKokkos<DeviceType, real_type, vector_length>::operator() (TagPairSNAPComputeForce<NEIGHFLAG,EVFLAG>, const int& ii, EV_FLOAT& ev) const {
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii + chunk_offset];
@ -1358,8 +1358,8 @@ void PairSNAPKokkos<DeviceType, real_type, vector_length>::v_tally_xyz(EV_FLOAT
{
// The vatom array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const E_FLOAT v0 = delx*fx;
const E_FLOAT v1 = dely*fy;

20
src/KOKKOS/pair_sw_kokkos.cpp
View File

@ -246,8 +246,8 @@ void PairSWKokkos<DeviceType>::operator()(TagPairSWComputeHalf<NEIGHFLAG,EVFLAG>
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
F_FLOAT delr1[3],delr2[3],fj[3],fk[3];
F_FLOAT evdwl = 0.0;
@ -800,11 +800,11 @@ void PairSWKokkos<DeviceType>::ev_tally(EV_FLOAT &ev, const int &i, const int &j
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (eflag_atom) {
const E_FLOAT epairhalf = 0.5 * epair;
@ -878,11 +878,11 @@ void PairSWKokkos<DeviceType>::ev_tally3(EV_FLOAT &ev, const int &i, const int &
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (eflag_atom) {
epairthird = THIRD * (evdwl + ecoul);

16
src/KOKKOS/pair_tersoff_kokkos.cpp
View File

@ -328,8 +328,8 @@ void PairTersoffKokkos<DeviceType>::operator()(TagPairTersoffComputeHalf<NEIGHFL
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
if (i >= nlocal) return;
@ -1143,11 +1143,11 @@ void PairTersoffKokkos<DeviceType>::ev_tally(EV_FLOAT &ev, const int &i, const i
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (eflag_atom) {
const E_FLOAT epairhalf = 0.5 * epair;
@ -1211,8 +1211,8 @@ void PairTersoffKokkos<DeviceType>::v_tally3(EV_FLOAT &ev, const int &i, const i
{
// The vatom array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
F_FLOAT v[6];

16
src/KOKKOS/pair_tersoff_mod_kokkos.cpp
View File

@ -328,8 +328,8 @@ void PairTersoffMODKokkos<DeviceType>::operator()(TagPairTersoffMODComputeHalf<N
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
if (i >= nlocal) return;
@ -1146,11 +1146,11 @@ void PairTersoffMODKokkos<DeviceType>::ev_tally(EV_FLOAT &ev, const int &i, cons
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (eflag_atom) {
const E_FLOAT epairhalf = 0.5 * epair;
@ -1214,8 +1214,8 @@ void PairTersoffMODKokkos<DeviceType>::v_tally3(EV_FLOAT &ev, const int &i, cons
{
// The vatom array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
F_FLOAT v[6];

16
src/KOKKOS/pair_tersoff_zbl_kokkos.cpp
View File

@ -342,8 +342,8 @@ void PairTersoffZBLKokkos<DeviceType>::operator()(TagPairTersoffZBLComputeHalf<N
// The f array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_f = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_f = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_f),decltype(ndup_f)>::get(dup_f,ndup_f);
auto a_f = v_f.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
const int i = d_ilist[ii];
if (i >= nlocal) return;
@ -1240,11 +1240,11 @@ void PairTersoffZBLKokkos<DeviceType>::ev_tally(EV_FLOAT &ev, const int &i, cons
// The eatom and vatom arrays are duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_eatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_eatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_eatom),decltype(ndup_eatom)>::get(dup_eatom,ndup_eatom);
auto a_eatom = v_eatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (eflag_atom) {
const E_FLOAT epairhalf = 0.5 * epair;
@ -1308,8 +1308,8 @@ void PairTersoffZBLKokkos<DeviceType>::v_tally3(EV_FLOAT &ev, const int &i, cons
{
// The vatom array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_vatom = ScatterViewHelper<typename NeedDup<NEIGHFLAG,DeviceType>::value,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<typename AtomicDup<NEIGHFLAG,DeviceType>::value>();
auto v_vatom = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_vatom),decltype(ndup_vatom)>::get(dup_vatom,ndup_vatom);
auto a_vatom = v_vatom.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
F_FLOAT v[6];