Merge branch 'add_init_value_to_reduce' into 'master'

Add init value to reduce See merge request npneq/inq!1153
LLNL · Oct 23, 2024 · c962fea · c962fea
2 parents 4649d03 + 8638b17
commit c962fea
Show file tree

Hide file tree

Showing 8 changed files with 84 additions and 84 deletions.
diff --git a/external_libs/gpurun/include/gpu/reduce.hpp b/external_libs/gpurun/include/gpu/reduce.hpp
diff --git a/src/ground_state/calculator.hpp b/src/ground_state/calculator.hpp
@@ -88,7 +88,7 @@ class calculator {
 			assert(el.occupations()[iphi].size() == normres[iphi].size());
 
 			auto func = state_conv_func<decltype(begin(el.occupations()[iphi])), decltype(begin(normres[iphi]))>{begin(el.occupations()[iphi]), begin(normres[iphi])};
-			state_conv += gpu::run(gpu::reduce(normres[iphi].size()), func);
+			state_conv += gpu::run(gpu::reduce(normres[iphi].size()), 0.0, func);
 		}
 
 		el.kpin_states_comm().all_reduce_n(&state_conv, 1);

diff --git a/src/hamiltonian/energy.hpp b/src/hamiltonian/energy.hpp
@@ -48,7 +48,7 @@ namespace hamiltonian {
 
 			assert(occupations.size() == array.size());
 			auto func = occ_sum_func<decltype(begin(occupations)), decltype(begin(array))>{begin(occupations), begin(array)};
-			return gpu::run(gpu::reduce(array.size()), func);
+			return gpu::run(gpu::reduce(array.size()), 0.0, func);
 		}
 
 public:

diff --git a/src/hamiltonian/ks_hamiltonian.hpp b/src/hamiltonian/ks_hamiltonian.hpp
@@ -70,7 +70,7 @@ class ks_hamiltonian {
 
 			assert(occupations.size() == array.size());
 			auto func = occ_sum_func<decltype(begin(occupations)), decltype(begin(array))>{begin(occupations), begin(array)};
-			return gpu::run(gpu::reduce(array.size()), func);
+			return gpu::run(gpu::reduce(array.size()), 0.0, func);
 		}
 
 public:

diff --git a/src/hamiltonian/projector_all.hpp b/src/hamiltonian/projector_all.hpp
@@ -327,7 +327,7 @@ class projector_all {
 			phi.basis().comm().all_reduce_in_place_n(raw_pointer_cast(projections_all.data_elements()), projections_all.num_elements(), std::plus<>{});
 		}
 
-		auto en = gpu::run(gpu::reduce(phi.local_set_size()), gpu::reduce(max_nlm_), gpu::reduce(nprojs_),
+		auto en = gpu::run(gpu::reduce(phi.local_set_size()), gpu::reduce(max_nlm_), gpu::reduce(nprojs_), 0.0,
 											 energy_reduction<decltype(begin(projections_all)), decltype(begin(coeff_)), decltype(begin(occupations))>
 											 {begin(projections_all), begin(coeff_), begin(occupations), phi.local_spinor_set_size()});
 
@@ -399,7 +399,7 @@ class projector_all {
 			if(locally_empty_[iproj]) continue;
 
 				CALI_CXX_MARK_SCOPE("projector_force_sum");
-				force[iproj] = gpu::run(gpu::reduce(phi.local_set_size()), gpu::reduce(max_sphere_size_),
+				force[iproj] = gpu::run(gpu::reduce(phi.local_set_size()), gpu::reduce(max_sphere_size_), zero<vector3<double, covariant>>(),
 																force_term<decltype(begin(occs)), decltype(begin(sphere_phi_all[iproj])), decltype(begin(sphere_gphi_all[iproj]))>{begin(occs), begin(sphere_phi_all[iproj]), begin(sphere_gphi_all[iproj])});
 		}
 

diff --git a/src/observables/forces_stress.hpp b/src/observables/forces_stress.hpp
@@ -86,7 +86,7 @@ struct forces_stress {
 				auto ionic_long_range = solvers::poisson::solve(electrons.atomic_pot().ionic_density(electrons.states_comm(), electrons.density_basis(), ions, iatom));
 				auto ionic_short_range = electrons.atomic_pot().local_potential(electrons.states_comm(), electrons.density_basis(), ions, iatom);
 
-				auto force_cov = -gpu::run(gpu::reduce(electrons.density_basis().local_size()),
+				auto force_cov = -gpu::run(gpu::reduce(electrons.density_basis().local_size()), zero<vector3<double, inq::covariant>>(),
 																	 loc_pot<decltype(begin(ionic_long_range.linear())), decltype(begin(ionic_short_range.linear())), decltype(begin(gdensity.linear()))>
 																	 {begin(ionic_long_range.linear()), begin(ionic_short_range.linear()), begin(gdensity.linear())});
 

diff --git a/src/operations/overlap_diagonal.hpp b/src/operations/overlap_diagonal.hpp
@@ -50,7 +50,7 @@ gpu::array<typename PhiMatrix::element_type, 1> overlap_diagonal_impl(Basis cons
 		overlap_vector[0] = blas::dot(blas::C(phi1_matrix.rotated()[0]), phi2_matrix.rotated()[0]);
 		overlap_vector[0] *= basis.volume_element();
 	} else {
-		overlap_vector = gpu::run(nn, gpu::reduce(phi1_matrix.size()),
+		overlap_vector = gpu::run(nn, gpu::reduce(phi1_matrix.size()), zero<type>(),
 															overlap_diagonal_mult<decltype(begin(phi1_matrix))>{basis.volume_element(), begin(phi1_matrix), begin(phi2_matrix)});
 	}
 
@@ -152,7 +152,7 @@ auto overlap_diagonal_normalized_impl(Basis const & basis, PhiMatrix const & phi
 
 	using type = typename PhiMatrix::element_type;
 
-	auto overlap_and_norm = gpu::run(nn, gpu::reduce(phi1_matrix.size()),
+	auto overlap_and_norm = gpu::run(nn, gpu::reduce(phi1_matrix.size()), zero<value_and_norm<type>>(),
 																	 overlap_diagonal_normalized_mult<decltype(begin(phi1_matrix))>{begin(phi1_matrix), begin(phi2_matrix)});
 
 	if(basis.comm().size() > 1){

diff --git a/src/operations/sum.hpp b/src/operations/sum.hpp
@@ -26,8 +26,9 @@ template <class array_type>
 typename array_type::element sum(const array_type & phi){
 
 	CALI_CXX_MARK_SCOPE("sum(1arg)");
-	if(phi.size() == 0) return zero<typename array_type::element>();
-	return gpu::run(gpu::reduce(phi.size()), gpu::array_access<decltype(begin(phi))>{begin(phi)});
+	auto init = zero<typename array_type::element>();
+	if(phi.size() == 0) return init;
+	return gpu::run(gpu::reduce(phi.size()), init, gpu::array_access<decltype(begin(phi))>{begin(phi)});
 }
 
 template <class ArrayType, class UnaryOp>