diff --git a/CMakeLists.txt b/CMakeLists.txt
index 0a66e63..bc061e3 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -35,6 +35,7 @@ add_subdirectory(app/test_for_new)
 add_subdirectory(app/invoke_parallel)
 add_subdirectory(app/benchmark_fft)
 add_subdirectory(app/benchmark_unbalanced)
+add_subdirectory(app/benchmark_matrix)
 
 # Add optional tests
 option(PACKAGE_TESTS "Build the tests" ON)
diff --git a/lib/pls/include/pls/algorithms/parallel_for.h b/lib/pls/include/pls/algorithms/parallel_for.h
new file mode 100644
index 0000000..4df1104
--- /dev/null
+++ b/lib/pls/include/pls/algorithms/parallel_for.h
@@ -0,0 +1,15 @@
+
+#ifndef PLS_PARALLEL_FOR_H
+#define PLS_PARALLEL_FOR_H
+
+namespace pls {
+namespace algorithm {
+
+template<typename RandomIt, typename Function>
+void parallel_for(RandomIt first, RandomIt last, const Function &function);
+
+}
+}
+#include "parallel_for_impl.h"
+
+#endif //PLS_PARALLEL_FOR_H
diff --git a/lib/pls/include/pls/algorithms/parallel_for_impl.h b/lib/pls/include/pls/algorithms/parallel_for_impl.h
new file mode 100644
index 0000000..afd7baa
--- /dev/null
+++ b/lib/pls/include/pls/algorithms/parallel_for_impl.h
@@ -0,0 +1,57 @@
+
+#ifndef PLS_PARALLEL_FOR_IMPL_H
+#define PLS_PARALLEL_FOR_IMPL_H
+
+#include "pls/internal/scheduling/fork_join_task.h"
+#include "pls/internal/scheduling/scheduler.h"
+#include "pls/internal/scheduling/scheduler.h"
+
+#include "pls/internal/helpers/unique_id.h"
+
+namespace pls {
+namespace algorithm {
+namespace internal {
+template<typename RandomIt, typename Function>
+void parallel_for(RandomIt first, RandomIt last, const Function &function) {
+  using namespace ::pls::internal::scheduling;
+  using namespace ::pls::internal::helpers;
+  using namespace ::pls::internal::base;
+  constexpr long min_elements = 4;
+
+  long num_elements = std::distance(first, last);
+  if (num_elements <= min_elements) {
+    // calculate last elements in loop to avoid overhead
+    for (auto current = first; current != last; current++) {
+      function(*current);
+    }
+  } else {
+    // Cut in half recursively
+    long middle_index = num_elements / 2;
+
+    auto body = [=] { internal::parallel_for(first + middle_index, last, function); };
+    fork_join_lambda_by_reference<decltype(body)> second_half_task(body);
+    fork_join_sub_task::current()->spawn_child(second_half_task);
+
+    parallel_for(first, first + middle_index, function);
+    fork_join_sub_task::current()->wait_for_all();
+  }
+}
+}
+
+template<typename RandomIt, typename Function>
+void parallel_for(RandomIt first, RandomIt last, const Function &function) {
+  using namespace ::pls::internal::scheduling;
+  using namespace ::pls::internal::helpers;
+  using namespace ::pls::internal::base;
+  static abstract_task::id id = unique_id::create<RandomIt, Function>();
+
+  auto body = [=] { internal::parallel_for(first, last, function); };
+  fork_join_lambda_by_reference<decltype(body)> root_body(body);
+  fork_join_task root_task{&root_body, id};
+  scheduler::execute_task(root_task);
+}
+
+}
+}
+
+#endif //PLS_INVOKE_PARALLEL_IMPL_H
diff --git a/lib/pls/include/pls/internal/scheduling/parallel_iterator_task.h b/lib/pls/include/pls/internal/scheduling/parallel_iterator_task.h
index c57b1ab..2013110 100644
--- a/lib/pls/include/pls/internal/scheduling/parallel_iterator_task.h
+++ b/lib/pls/include/pls/internal/scheduling/parallel_iterator_task.h
@@ -5,23 +5,28 @@
 #include "abstract_task.h"
 
 namespace pls {
-    namespace internal {
-        namespace scheduling {
-            template<typename RandomIt, typename Function>
-            class parallel_iterator_task: public abstract_task {
-                RandomIt first_, last_;
-                Function function_;
+namespace internal {
+namespace scheduling {
+template<typename RandomIt, typename Function>
+class parallel_iterator_task : public abstract_task {
+  RandomIt first_, last_;
+  Function function_;
 
-            protected:
-                bool internal_stealing(abstract_task* other_task) override;
-                bool split_task(base::spin_lock* /*lock*/) override;
+  // External stealing
+  size_t first_index_ava, last_index_ava;
+  // My internal state
+  size_t current_index, max_index;
 
-            public:
-                explicit parallel_iterator_task(RandomIt first, RandomIt last, Function function, const abstract_task::id& id);
-                void execute() override;
-            };
-        }
-    }
+ protected:
+  bool internal_stealing(abstract_task *other_task) override;
+  bool split_task(base::swmr_spin_lock * /*lock*/) override;
+
+ public:
+  explicit parallel_iterator_task(RandomIt first, RandomIt last, Function function, const abstract_task::id &id);
+  void execute() override;
+};
+}
+}
 }
 #include "parallel_iterator_task_impl.h"
 
diff --git a/lib/pls/include/pls/internal/scheduling/parallel_iterator_task_impl.h b/lib/pls/include/pls/internal/scheduling/parallel_iterator_task_impl.h
index a0ab2dd..1883cde 100644
--- a/lib/pls/include/pls/internal/scheduling/parallel_iterator_task_impl.h
+++ b/lib/pls/include/pls/internal/scheduling/parallel_iterator_task_impl.h
@@ -2,34 +2,33 @@
 #ifndef PLS_PARALLEL_ITERATOR_TASK_IMPL_H
 #define PLS_PARALLEL_ITERATOR_TASK_IMPL_H
 
-
 namespace pls {
-    namespace internal {
-        namespace scheduling {
-            template<typename RandomIt, typename Function>
-            parallel_iterator_task<RandomIt, Function>::parallel_iterator_task
-                (RandomIt first, RandomIt last, Function function, const abstract_task::id& id):
-                    abstract_task(0, id),
-                    first_{first},
-                    last_{last},
-                    function_{function} {}
-
-            template<typename RandomIt, typename Function>
-            void parallel_iterator_task<RandomIt, Function>::execute() {
+namespace internal {
+namespace scheduling {
+template<typename RandomIt, typename Function>
+parallel_iterator_task<RandomIt, Function>::parallel_iterator_task
+    (RandomIt first, RandomIt last, Function function, const abstract_task::id &id):
+    abstract_task(0, id),
+    first_{first},
+    last_{last},
+    function_{function} {}
+
+template<typename RandomIt, typename Function>
+void parallel_iterator_task<RandomIt, Function>::execute() {
 
-            }
+}
 
-            template<typename RandomIt, typename Function>
-            bool parallel_iterator_task<RandomIt, Function>::split_task(base::spin_lock* /*lock*/) {
-                return false;
-            }
+template<typename RandomIt, typename Function>
+bool parallel_iterator_task<RandomIt, Function>::split_task(base::swmr_spin_lock * /*lock*/) {
+  return false;
+}
 
-            template<typename RandomIt, typename Function>
-            bool parallel_iterator_task<RandomIt, Function>::internal_stealing(abstract_task* other_task) {
-                return false;
-            }
-        }
-    }
+template<typename RandomIt, typename Function>
+bool parallel_iterator_task<RandomIt, Function>::internal_stealing(abstract_task *other_task) {
+  return false;
+}
+}
+}
 }
 
 #endif //PLS_PARALLEL_ITERATOR_TASK_IMPL_H
diff --git a/lib/pls/include/pls/pls.h b/lib/pls/include/pls/pls.h
index b7a218b..0b0e366 100644
--- a/lib/pls/include/pls/pls.h
+++ b/lib/pls/include/pls/pls.h
@@ -2,6 +2,7 @@
 #define PLS_LIBRARY_H
 
 #include "pls/algorithms/invoke_parallel.h"
+#include "pls/algorithms/parallel_for.h"
 #include "pls/internal/scheduling/abstract_task.h"
 #include "pls/internal/scheduling/fork_join_task.h"
 #include "pls/internal/scheduling/scheduler.h"
@@ -23,6 +24,7 @@ using internal::scheduling::fork_join_lambda_by_value;
 using internal::scheduling::fork_join_task;
 
 using algorithm::invoke_parallel;
+using algorithm::parallel_for;
 
 }