Change Kokkos SNAP energy shared arrays from thread to team (60693a5f) · Commits · 郑智淋 / lammps

src/KOKKOS/pair_snap_kokkos_impl.h

+5 −1

Original line number	Diff line number	Diff line
		@@ -460,7 +460,9 @@ void PairSNAPKokkos<DeviceType>::operator() (TagPairSNAP<NEIGHFLAG,EVFLAG>,const

		if (quadraticflag) {
		my_sna.compute_bi(team);
		team.team_barrier();
		my_sna.copy_bi2bvec(team);
		team.team_barrier();
		}

		// for neighbors of I within cutoff:
		@@ -564,7 +566,9 @@ void PairSNAPKokkos<DeviceType>::operator() (TagPairSNAP<NEIGHFLAG,EVFLAG>,const

		if (!quadraticflag) {
		my_sna.compute_bi(team);
		team.team_barrier();
		my_sna.copy_bi2bvec(team);
		team.team_barrier();
		}

		// E = beta.B + 0.5*B^t.alpha.B
		@@ -572,7 +576,7 @@ void PairSNAPKokkos<DeviceType>::operator() (TagPairSNAP<NEIGHFLAG,EVFLAG>,const
		// coeff[k] = alpha_ii or
		// coeff[k] = alpha_ij = alpha_ji, j != i

		Kokkos::single(Kokkos::PerThread(team), [&] () {
		Kokkos::single(Kokkos::PerTeam(team), [&] () {

		// evdwl = energy of atom I, sum over coeffs_k * Bi_k

src/KOKKOS/sna_kokkos_impl.h

+10 −8

Original line number	Diff line number	Diff line
		@@ -368,11 +368,13 @@ void SNAKokkos<DeviceType>::compute_bi(const typename Kokkos::TeamPolicy<DeviceT
		uarraytot_i(j,ma,mb) * zarray_i(j1,j2,j,mb,ma))*0.5;
		}

		Kokkos::single(Kokkos::PerThread(team), [&] () {
		b_j1_j2_j *= 2.0;
		if (bzero_flag)
		b_j1_j2_j -= bzero[j];

		barray(j1,j2,j) = b_j1_j2_j;
		});
		});
		//} // end loop over j
		//} // end loop over j1, j2
		@@ -1039,6 +1041,8 @@ void SNAKokkos<DeviceType>::create_team_scratch_arrays(const typename Kokkos::Te
		uarraytot_i_a = uarraytot_i = t_sna_3d(team.team_scratch(1),jdim,jdim,jdim);
		zarray_r = t_sna_5d(team.team_scratch(1),jdim,jdim,jdim,jdim,jdim);
		zarray_i = t_sna_5d(team.team_scratch(1),jdim,jdim,jdim,jdim,jdim);
		bvec = Kokkos::View<double*, Kokkos::LayoutRight, DeviceType>(team.team_scratch(1),ncoeff);
		barray = t_sna_3d(team.team_scratch(1),jdim,jdim,jdim);

		rij = t_sna_2d(team.team_scratch(1),nmax,3);
		rcutij = t_sna_1d(team.team_scratch(1),nmax);
		@@ -1057,6 +1061,8 @@ T_INT SNAKokkos<DeviceType>::size_team_scratch_arrays() {
		size += t_sna_3d::shmem_size(jdim,jdim,jdim); // uarraytot_i_a
		size += t_sna_5d::shmem_size(jdim,jdim,jdim,jdim,jdim); // zarray_r
		size += t_sna_5d::shmem_size(jdim,jdim,jdim,jdim,jdim); // zarray_i
		size += Kokkos::View<double*, Kokkos::LayoutRight, DeviceType>::shmem_size(ncoeff); // bvec
		size += t_sna_3d::shmem_size(jdim,jdim,jdim); // barray

		size += t_sna_2d::shmem_size(nmax,3); // rij
		size += t_sna_1d::shmem_size(nmax); // rcutij
		@@ -1073,8 +1079,6 @@ KOKKOS_INLINE_FUNCTION
		void SNAKokkos<DeviceType>::create_thread_scratch_arrays(const typename Kokkos::TeamPolicy<DeviceType>::member_type& team)
		{
		int jdim = twojmax + 1;
		bvec = Kokkos::View<double*, Kokkos::LayoutRight, DeviceType>(team.thread_scratch(1),ncoeff);
		barray = t_sna_3d(team.thread_scratch(1),jdim,jdim,jdim);

		dbvec = Kokkos::View<double*[3], Kokkos::LayoutRight, DeviceType>(team.thread_scratch(1),ncoeff);
		dbarray = t_sna_4d(team.thread_scratch(1),jdim,jdim,jdim);
		@@ -1090,8 +1094,6 @@ inline
		T_INT SNAKokkos<DeviceType>::size_thread_scratch_arrays() {
		T_INT size = 0;
		int jdim = twojmax + 1;
		size += Kokkos::View<double*, Kokkos::LayoutRight, DeviceType>::shmem_size(ncoeff); // bvec
		size += t_sna_3d::shmem_size(jdim,jdim,jdim); // barray

		size += Kokkos::View<double*[3], Kokkos::LayoutRight, DeviceType>::shmem_size(ncoeff); // dbvec
		size += t_sna_4d::shmem_size(jdim,jdim,jdim); // dbarray

Admin message