diff --git a/src/algorithm/nd_sort.cpp b/src/algorithm/nd_sort.cpp
index 3e61677c..a401bb58 100644
--- a/src/algorithm/nd_sort.cpp
+++ b/src/algorithm/nd_sort.cpp
@@ -188,6 +188,7 @@ namespace gapp::algorithm::dtl
         const size_t popsize = std::distance(first, last);
         DominanceMatrix dmat(popsize, popsize /*, MAXIMAL */);
 
+        GAPP_BARRIER();
         std::for_each(GAPP_EXEC, detail::iota_iterator(0_sz), detail::iota_iterator(first->size()), [&](size_t obj)
         {
             FitnessVector fvec(popsize);
@@ -211,6 +212,7 @@ namespace gapp::algorithm::dtl
                 });
             });
         });
+        GAPP_BARRIER();
 
         std::for_each(GAPP_EXEC, detail::iota_iterator(0_sz), detail::iota_iterator(popsize), [&](size_t row) noexcept
         {
@@ -225,6 +227,7 @@ namespace gapp::algorithm::dtl
                 }
             }
         });
+        GAPP_BARRIER();
 
         return dmat;
     }
diff --git a/src/algorithm/nsga3.cpp b/src/algorithm/nsga3.cpp
index ed1b6f55..a051abaa 100644
--- a/src/algorithm/nsga3.cpp
+++ b/src/algorithm/nsga3.cpp
@@ -239,6 +239,7 @@ namespace gapp::algorithm
 
         sol_info_.resize(last - first);
 
+        GAPP_BARRIER();
         std::for_each(GAPP_EXEC, pfirst, plast, [&](const FrontInfo& sol)
         {
             const FitnessVector fnorm = normalizeFitnessVec(first[sol.idx], ideal_point_, nadir_point_);
@@ -250,6 +251,7 @@ namespace gapp::algorithm
             sol_info_[sol.idx].ref_idx  = std::distance(ref_lines_.begin(), closest);
             sol_info_[sol.idx].ref_dist = math::perpendicularDistanceSq(*closest, fnorm);
         });
+        GAPP_BARRIER();
     }
 
     inline bool NSGA3::Impl::nichedCompare(size_t lhs, size_t rhs) const noexcept
diff --git a/src/algorithm/reference_lines.cpp b/src/algorithm/reference_lines.cpp
index f09c556c..ce39bdb0 100644
--- a/src/algorithm/reference_lines.cpp
+++ b/src/algorithm/reference_lines.cpp
@@ -182,6 +182,7 @@ namespace gapp::algorithm::reflines
             std::swap(min_distances[idx], min_distances.back());
             min_distances.pop_back();
 
+            GAPP_BARRIER();
             /* Calc the distance of each candidate to the closest ref point. */
             std::transform(GAPP_EXEC, candidate_points.begin(), candidate_points.end(), min_distances.begin(), min_distances.begin(),
             [&](const Point& candidate, double current_min) noexcept
@@ -189,6 +190,7 @@ namespace gapp::algorithm::reflines
                 const double dist = math::euclideanDistanceSq(candidate, points.back());
                 return std::min(current_min, dist);
             });
+            GAPP_BARRIER();
         }
 
         return points;
diff --git a/src/core/ga_base.impl.hpp b/src/core/ga_base.impl.hpp
index b3eff970..a105fba8 100644
--- a/src/core/ga_base.impl.hpp
+++ b/src/core/ga_base.impl.hpp
@@ -339,7 +339,9 @@ namespace gapp
         /* Create and evaluate the initial population of the algorithm. */
         num_objectives_ = findNumberOfObjectives();
         population_ = generatePopulation(population_size_, std::move(initial_population));
+        GAPP_BARRIER();
         std::for_each(GAPP_EXEC, population_.begin(), population_.end(), [this](Candidate<T>& sol) { evaluate(sol); });
+        GAPP_BARRIER();
         fitness_matrix_ = detail::toFitnessMatrix(population_);
         if (keep_all_optimal_sols_) solutions_ = detail::findParetoFront(population_);
 
@@ -504,6 +506,7 @@ namespace gapp
         std::vector<CandidatePair<T>> child_pairs(num_children / 2);
 
         prepareSelections();
+        GAPP_BARRIER();
         std::generate(GAPP_EXEC, child_pairs.begin(), child_pairs.end(),
         [this]
         {
@@ -512,12 +515,14 @@ namespace gapp
 
             return crossover(parent1, parent2);
         });
+        GAPP_BARRIER();
 
         auto children = detail::flatten(std::move(child_pairs));
 
         /* If the population size is odd, one too many child candidates were generated by the crossovers. */
         if (children.size() > population_size_) children.pop_back();
 
+        GAPP_BARRIER();
         std::for_each(GAPP_EXEC, children.begin(), children.end(),
         [this](Candidate<T>& child)
         {
@@ -525,6 +530,7 @@ namespace gapp
             repair(child);
             evaluate(child);
         });
+        GAPP_BARRIER();
 
         updatePopulation(std::move(children));
         if (keep_all_optimal_sols_) updateOptimalSolutions(solutions_, population_);
diff --git a/src/metrics/pop_stats.cpp b/src/metrics/pop_stats.cpp
index e96ea1e1..8e802a5f 100644
--- a/src/metrics/pop_stats.cpp
+++ b/src/metrics/pop_stats.cpp
@@ -154,6 +154,7 @@ namespace gapp::detail
         const FitnessMatrix front = uniqueSortedParetoFront(fmat);
 
         std::atomic<double> hypervolume = 0.0;
+        GAPP_BARRIER();
         std::for_each(GAPP_EXEC, detail::iota_iterator(0_sz), detail::iota_iterator(front.size()), [&](size_t idx)
         {
             const auto point = front[idx];
@@ -163,6 +164,7 @@ namespace gapp::detail
 
             hypervolume.fetch_add(exclusive_hypervolume, std::memory_order_acq_rel);
         });
+        GAPP_BARRIER();
 
         return hypervolume.load(std::memory_order_acquire);
     }
diff --git a/src/population/population.hpp b/src/population/population.hpp
index 1f6cd533..4080e120 100644
--- a/src/population/population.hpp
+++ b/src/population/population.hpp
@@ -106,6 +106,7 @@ namespace gapp::detail
         std::vector<Dominance> lhs_state(lhs.size());
         std::vector<std::atomic<Dominance>> rhs_state(rhs.size());
 
+        GAPP_BARRIER();
         std::for_each(GAPP_EXEC, iota_iterator(0_sz), iota_iterator(lhs.size()), [&](size_t i) noexcept
         {
             for (size_t j = 0; j < rhs.size(); j++)
@@ -146,6 +147,7 @@ namespace gapp::detail
                 // comp == 0 --> both are OPTIMAL or DOMINATED, can't know
             }
         });
+        GAPP_BARRIER();
 
         Candidates<T> optimal_solutions;
         optimal_solutions.reserve(lhs.size() + rhs.size());
diff --git a/src/utility/utility.hpp b/src/utility/utility.hpp
index 252b86ba..765a37c3 100644
--- a/src/utility/utility.hpp
+++ b/src/utility/utility.hpp
@@ -111,6 +111,11 @@
 #endif
 
 
+#include <atomic>
+inline std::atomic_bool global_barrier;
+#define GAPP_BARRIER() std::ignore = global_barrier.exchange(true, std::memory_order_acq_rel)
+
+
 namespace gapp
 {
     constexpr std::size_t operator ""_sz(unsigned long long arg) noexcept