ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Switch to vector parallelism for half list in QEq

Browse Source
This commit is contained in:
Stan Gerald Moore
2022-03-03 08:09:25 -07:00
parent b6b7846c50
commit ee2b9f28cb
2 changed files with 49 additions and 45 deletions
Download Patch File Download Diff File Expand all files Collapse all files

92
src/KOKKOS/fix_qeq_reaxff_kokkos.cpp
View File

@ -858,6 +858,19 @@ void FixQEqReaxFFKokkos<DeviceType>::sparse_matvec_kokkos(typename AT::t_ffloat2
Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagQEqSparseMatvec1>(0,nn),*this); Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagQEqSparseMatvec1>(0,nn),*this);
int teamsize;
int vectorsize;
int leaguesize;
if (execution_space == Host) {
teamsize = 1;
vectorsize = 1;
leaguesize = nn;
} else {
teamsize = FixQEqReaxFFKokkos<DeviceType>::spmv_teamsize;
vectorsize = FixQEqReaxFFKokkos<DeviceType>::vectorsize;
leaguesize = (nn + teamsize - 1) / (teamsize);
}
if (neighflag != FULL) { if (neighflag != FULL) {
Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagQEqZeroQGhosts>(nn,NN),*this); Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagQEqZeroQGhosts>(nn,NN),*this);
@ -865,28 +878,14 @@ void FixQEqReaxFFKokkos<DeviceType>::sparse_matvec_kokkos(typename AT::t_ffloat2
dup_o.reset_except(d_o); dup_o.reset_except(d_o);
if (neighflag == HALF) if (neighflag == HALF)
Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagQEqSparseMatvec2_Half<HALF> >(0,nn),*this); Kokkos::parallel_for(Kokkos::TeamPolicy<DeviceType, TagQEqSparseMatvec2_Half<HALF>>(leaguesize, teamsize, vectorsize), *this);
else if (neighflag == HALFTHREAD) else if (neighflag == HALFTHREAD)
Kokkos::parallel_for(Kokkos::RangePolicy<DeviceType,TagQEqSparseMatvec2_Half<HALFTHREAD> >(0,nn),*this); Kokkos::parallel_for(Kokkos::TeamPolicy<DeviceType, TagQEqSparseMatvec2_Half<HALFTHREAD>>(leaguesize, teamsize, vectorsize), *this);
if (need_dup) if (need_dup)
Kokkos::Experimental::contribute(d_o, dup_o); Kokkos::Experimental::contribute(d_o, dup_o);
} else { // FULL } else // FULL
int teamsize;
int vectorsize;
int leaguesize;
if (execution_space == Host) {
teamsize = 1;
vectorsize = 1;
leaguesize = nn;
} else {
teamsize = FixQEqReaxFFKokkos<DeviceType>::spmv_teamsize;
vectorsize = FixQEqReaxFFKokkos<DeviceType>::vectorsize;
leaguesize = (nn + teamsize - 1) / (teamsize);
}
Kokkos::parallel_for(Kokkos::TeamPolicy <DeviceType, TagQEqSparseMatvec2_Full>(leaguesize, teamsize, vectorsize), *this); Kokkos::parallel_for(Kokkos::TeamPolicy <DeviceType, TagQEqSparseMatvec2_Full>(leaguesize, teamsize, vectorsize), *this);
}
} }
/* ---------------------------------------------------------------------- */ /* ---------------------------------------------------------------------- */
@ -925,34 +924,39 @@ void FixQEqReaxFFKokkos<DeviceType>::operator()(TagQEqZeroQGhosts, const int &i)
template<class DeviceType> template<class DeviceType>
template<int NEIGHFLAG> template<int NEIGHFLAG>
KOKKOS_INLINE_FUNCTION KOKKOS_INLINE_FUNCTION
void FixQEqReaxFFKokkos<DeviceType>::operator()(TagQEqSparseMatvec2_Half<NEIGHFLAG>, const int &ii) const void FixQEqReaxFFKokkos<DeviceType>::operator()(TagQEqSparseMatvec2_Half<NEIGHFLAG>, const typename Kokkos::TeamPolicy<DeviceType, TagQEqSparseMatvec2_Half<NEIGHFLAG>>::member_type &team) const
{ {
// The q array is duplicated for OpenMP, atomic for CUDA, and neither for Serial int k = team.league_rank() * team.team_size() + team.team_rank();
auto v_o = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o); if (k < nn) {
auto a_o = v_o.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>(); // The q array is duplicated for OpenMP, atomic for CUDA, and neither for Serial
auto v_o = ScatterViewHelper<NeedDup_v<NEIGHFLAG,DeviceType>,decltype(dup_o),decltype(ndup_o)>::get(dup_o,ndup_o);
const int i = d_ilist[ii]; auto a_o = v_o.template access<AtomicDup_v<NEIGHFLAG,DeviceType>>();
if (mask[i] & groupbit) {
F_FLOAT2 tmp; const int i = d_ilist[k];
const auto d_xx_i0 = d_xx(i,0); if (mask[i] & groupbit) {
const auto d_xx_i1 = d_xx(i,1); F_FLOAT2 tmp;
const double d_xx_i0 = d_xx(i,0);
for (int jj = d_firstnbr[i]; jj < d_firstnbr[i] + d_numnbrs[i]; jj++) { const double d_xx_i1 = d_xx(i,1);
const int j = d_jlist(jj);
const auto d_val_jj = d_val(jj); Kokkos::parallel_reduce(Kokkos::ThreadVectorRange(team, d_firstnbr[i], d_firstnbr[i] + d_numnbrs[i]), [&] (const int &jj, F_FLOAT2& tmp) {
if (!(converged & 1)) { const int j = d_jlist(jj);
tmp.v[0] += d_val_jj * d_xx(j,0); const auto d_val_jj = d_val(jj);
a_o(j,0) += d_val_jj * d_xx_i0; if (!(converged & 1)) {
} tmp.v[0] += d_val_jj * d_xx(j,0);
if (!(converged & 2)) { a_o(j,0) += d_val_jj * d_xx_i0;
tmp.v[1] += d_val_jj * d_xx(j,1); }
a_o(j,1) += d_val_jj * d_xx_i1; if (!(converged & 2)) {
} tmp.v[1] += d_val_jj * d_xx(j,1);
a_o(j,1) += d_val_jj * d_xx_i1;
}
}, tmp);
Kokkos::single(Kokkos::PerThread(team), [&] () {
if (!(converged & 1))
a_o(i,0) += tmp.v[0];
if (!(converged & 2))
a_o(i,1) += tmp.v[1];
});
} }
if (!(converged & 1))
a_o(i,0) += tmp.v[0];
if (!(converged & 2))
a_o(i,1) += tmp.v[1];
} }
} }

2
src/KOKKOS/fix_qeq_reaxff_kokkos.h
View File

@ -92,7 +92,7 @@ class FixQEqReaxFFKokkos : public FixQEqReaxFF, public KokkosBase {
template<int NEIGHFLAG> template<int NEIGHFLAG>
KOKKOS_INLINE_FUNCTION KOKKOS_INLINE_FUNCTION
void operator()(TagQEqSparseMatvec2_Half<NEIGHFLAG>, const int&) const; void operator()(TagQEqSparseMatvec2_Half<NEIGHFLAG>, const typename Kokkos::TeamPolicy<DeviceType, TagQEqSparseMatvec2_Half<NEIGHFLAG>>::member_type &team) const;
typedef typename Kokkos::TeamPolicy<DeviceType, TagQEqSparseMatvec2_Full>::member_type membertype_vec; typedef typename Kokkos::TeamPolicy<DeviceType, TagQEqSparseMatvec2_Full>::member_type membertype_vec;
KOKKOS_INLINE_FUNCTION KOKKOS_INLINE_FUNCTION