diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 4754e0e..95f4b16 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -1,4 +1,4 @@
-image: "matrim/cmake-examples:3.10.3"
+image: "flofritz/ci-benchmark-hpx-embb"
 
 stages:
   - build
diff --git a/lib/pls/src/internal/base/spin_lock.cpp b/lib/pls/src/internal/base/spin_lock.cpp
index 10f6978..99a2c1d 100644
--- a/lib/pls/src/internal/base/spin_lock.cpp
+++ b/lib/pls/src/internal/base/spin_lock.cpp
@@ -11,10 +11,12 @@ namespace pls {
                         this_thread::yield();
                     }
                 }
+                std::atomic_thread_fence(std::memory_order_seq_cst);
             }
 
             void spin_lock::unlock() {
                 flag_.clear(std::memory_order_release);
+                std::atomic_thread_fence(std::memory_order_seq_cst);
             }
         }
     }
diff --git a/test/base_tests.cpp b/test/base_tests.cpp
index 7c7b990..d64614a 100644
--- a/test/base_tests.cpp
+++ b/test/base_tests.cpp
@@ -33,23 +33,25 @@ TEST_CASE( "thread state", "[internal/base/thread.h]") {
     REQUIRE(base_tests_local_value_two == vector<int>{1, 2});
 }
 
+int base_tests_shared_counter;
+
 TEST_CASE( "spinlock protects concurrent counter", "[internal/base/spinlock.h]") {
-    constexpr int num_iterations = 100000;
-    int shared_counter = 0;
+    constexpr int num_iterations = 1000000;
+    base_tests_shared_counter = 0;
     spin_lock lock{};
 
     SECTION( "lock can be used by itself" ) {
         auto t1 = start_thread([&]() {
             for (int i = 0; i < num_iterations; i++) {
                 lock.lock();
-                shared_counter++;
+                base_tests_shared_counter++;
                 lock.unlock();
             }
         });
         auto t2 = start_thread([&]() {
             for (int i = 0; i < num_iterations; i++) {
                 lock.lock();
-                shared_counter--;
+                base_tests_shared_counter--;
                 lock.unlock();
             }
         });
@@ -57,26 +59,26 @@ TEST_CASE( "spinlock protects concurrent counter", "[internal/base/spinlock.h]")
         t1.join();
         t2.join();
 
-        REQUIRE(shared_counter == 0);
+        REQUIRE(base_tests_shared_counter == 0);
     }
 
     SECTION( "lock can be used with std::lock_guard" ) {
         auto t1 = start_thread([&]() {
             for (int i = 0; i < num_iterations; i++) {
                 std::lock_guard<spin_lock> my_lock{lock};
-                shared_counter++;
+                base_tests_shared_counter++;
             }
         });
         auto t2 = start_thread([&]() {
             for (int i = 0; i < num_iterations; i++) {
                 std::lock_guard<spin_lock> my_lock{lock};
-                shared_counter--;
+                base_tests_shared_counter--;
             }
         });
 
         t1.join();
         t2.join();
 
-        REQUIRE(shared_counter == 0);
+        REQUIRE(base_tests_shared_counter == 0);
     }
 }