Skip to content

Commit

Permalink
Improve load balancer by passing beta degeneracy to constraint distri…
Browse files Browse the repository at this point in the history 
…butor
  • Loading branch information
David Williams-Young committed Oct 20, 2023
1 parent 5fbc8f6 commit 43488f2
Show file tree
Hide file tree
Showing 4 changed files with 41 additions and 35 deletions.
10 changes: 10 additions & 0 deletions include/macis/asci/alpha_constraint.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -66,6 +66,16 @@ class alpha_constraint {
return overlap(state) == count_ and
constraint_traits::count(symmetric_difference(state) >> C_min_) == 0;
}

static alpha_constraint make_triplet(unsigned i, unsigned j, unsigned k) {
constraint_type C = 0;
C.flip(i).flip(j).flip(k);
constraint_type B = 1;
static_assert(B.size() <= 64, "ULLONG NOT POSSIBLE HERE");
B <<= k;
B = B.to_ullong() - 1;
return alpha_constraint(C, B, k);
}
};

} // namespace macis
2 changes: 1 addition & 1 deletion include/macis/asci/determinant_search.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -328,7 +328,7 @@ asci_contrib_container<wfn_t<N>> asci_contributions_constraint(

auto gen_c_st = clock_type::now();
auto constraints =
dist_constraint_general(asci_settings.constraint_level, norb, n_sing_beta,
dist_constraint_general<wfn_t<N>>(asci_settings.constraint_level, norb, n_sing_beta,
n_doub_beta, uniq_alpha_wfn, comm);
auto gen_c_en = clock_type::now();
duration_type gen_c_dur = gen_c_en - gen_c_st;
Expand Down
58 changes: 26 additions & 32 deletions include/macis/asci/mask_constraints.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -510,22 +510,22 @@ auto make_triplet(unsigned i, unsigned j, unsigned k) {
}

#ifdef MACIS_ENABLE_MPI
template <size_t N>
template <typename WfnType, typename ContainerType>
auto dist_constraint_general(size_t nlevels, size_t norb, size_t ns_othr,
size_t nd_othr,
const std::vector<wfn_t<N>>& unique_alpha,
const ContainerType& unique_alpha,
MPI_Comm comm) {
using wfn_type = wfn_t<N>;
using wfn_traits = wavefunction_traits<wfn_type>;

using wfn_traits = wavefunction_traits<WfnType>;
using constraint_type = alpha_constraint<wfn_traits>;
using string_type = typename constraint_type::constraint_type;
auto world_rank = comm_rank(comm);
auto world_size = comm_size(comm);

// wfn_t<N> O = full_mask<N>(norb);

// Global workloads
std::vector<size_t> workloads(world_size, 0);
constexpr bool flat_container = std::is_same_v<
std::decay_t<WfnType>,
std::decay_t<typename ContainerType::value_type>
>;

// Generate triplets + heuristic
std::vector<std::pair<constraint_type, size_t>> constraint_sizes;
Expand All @@ -534,18 +534,22 @@ auto dist_constraint_general(size_t nlevels, size_t norb, size_t ns_othr,
for(int t_i = 0; t_i < norb; ++t_i)
for(int t_j = 0; t_j < t_i; ++t_j)
for(int t_k = 0; t_k < t_j; ++t_k) {
auto constraint = make_triplet<N>(t_i, t_j, t_k);
auto constraint = constraint_type::make_triplet(t_i, t_j, t_k);

size_t nw = 0;
for(const auto& alpha : unique_alpha) {
nw += constraint_histogram(wfn_traits::alpha_string(alpha), ns_othr,
nd_othr, constraint);
if constexpr (flat_container)
nw += constraint_histogram(wfn_traits::alpha_string(alpha), ns_othr,
nd_othr, constraint);
else
nw += alpha.second *
constraint_histogram(alpha.first, ns_othr, nd_othr, constraint);
}
if(nw) constraint_sizes.emplace_back(constraint, nw);
total_work += nw;
}

size_t local_average = (0.6 * total_work) / world_size;
size_t local_average = (0.8 * total_work) / world_size;

for(size_t ilevel = 0; ilevel < nlevels; ++ilevel) {
// Select constraints larger than average to be broken apart
Expand Down Expand Up @@ -578,37 +582,27 @@ auto dist_constraint_general(size_t nlevels, size_t norb, size_t ns_othr,
size_t nw = 0;

for(const auto& alpha : unique_alpha) {
nw += constraint_histogram(wfn_traits::alpha_string(alpha), ns_othr,
nd_othr, c_next);
if constexpr (flat_container)
nw += constraint_histogram(wfn_traits::alpha_string(alpha), ns_othr,
nd_othr, c_next);
else
nw += alpha.second *
constraint_histogram(alpha.first, ns_othr, nd_othr, c_next);
}
if(nw) constraint_sizes.emplace_back(c_next, nw);
total_work += nw;
}
}
} // Recurse into constraints

// if(!world_rank) {
// const auto ntrip = std::count_if(constraint_sizes.begin(),
// constraint_sizes.end(), [](auto &c){ return c.first.C.count() == 3; });
// printf("[rank 0] NTRIP = %lu\n", ntrip);
// if(nlevels > 0) {
// const auto nquad = std::count_if(constraint_sizes.begin(),
// constraint_sizes.end(), [](auto &c){ return c.first.C.count() == 4;
// });
// printf("[rank 0] NQUAD = %lu\n", nquad);
// }
// if(nlevels > 1) {
// const auto nquint = std::count_if(constraint_sizes.begin(),
// constraint_sizes.end(), [](auto &c){ return c.first.C.count() == 5;
// });
// printf("[rank 0] NQINT = %lu\n", nquint);
// }
// }

// Sort to get optimal bucket partitioning
std::sort(constraint_sizes.begin(), constraint_sizes.end(),
[](const auto& a, const auto& b) { return a.second > b.second; });


// Global workloads
std::vector<size_t> workloads(world_size, 0);

// Assign work
std::vector<constraint_type> constraints;
constraints.reserve(constraint_sizes.size() / world_size);
Expand Down
6 changes: 4 additions & 2 deletions include/macis/asci/pt2.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -113,8 +113,10 @@ double asci_pt2_constraint(wavefunction_iterator_t<N> cdets_begin,
logger->info(" * NS = {} ND = {}", n_sing_alpha, n_doub_alpha);

auto gen_c_st = clock_type::now();
auto constraints = dist_constraint_general(
5, norb, n_sing_beta, n_doub_beta, uniq_alpha_wfn, comm);
//auto constraints = dist_constraint_general<wfn_t<N>>(
// 5, norb, n_sing_beta, n_doub_beta, uniq_alpha_wfn, comm);
auto constraints = dist_constraint_general<wfn_t<N>>(
5, norb, n_sing_beta, n_doub_beta, uniq_alpha, comm);
auto gen_c_en = clock_type::now();
duration_type gen_c_dur = gen_c_en - gen_c_st;
logger->info(" * GEN_DUR = {:.2e} ms", gen_c_dur.count());
Expand Down

0 comments on commit 43488f2

Please sign in to comment.