diff --git a/ddprof-lib/src/main/cpp/arch_dd.h b/ddprof-lib/src/main/cpp/arch_dd.h
index 8378fc6f2..8bddcdd2c 100644
--- a/ddprof-lib/src/main/cpp/arch_dd.h
+++ b/ddprof-lib/src/main/cpp/arch_dd.h
@@ -14,19 +14,32 @@ static inline long long atomicInc(volatile long long &var,
   return __sync_fetch_and_add(&var, increment);
 }
 
-static inline u64 loadAcquire(volatile u64 &var) {
-  return __atomic_load_n(&var, __ATOMIC_ACQUIRE);
+template <typename T>
+static inline long long atomicIncRelaxed(volatile T &var,
+                                         T increment = 1) {
+  return __atomic_fetch_add(&var, increment, __ATOMIC_RELAXED);
 }
 
-static inline size_t loadAcquire(volatile size_t &var) {
-  return __atomic_load_n(&var, __ATOMIC_ACQUIRE);
+// Atomic load/store (unordered)
+template <typename T>
+static inline T load(volatile T& var) {
+    return __atomic_load_n(&var, __ATOMIC_RELAXED);
 }
 
-static inline void storeRelease(volatile long long &var, long long value) {
-  return __atomic_store_n(&var, value, __ATOMIC_RELEASE);
+template <typename T>
+static inline void store(volatile T& var, T value) {
+  return __atomic_store_n(&var, value, __ATOMIC_RELAXED);
+}
+
+
+// Atomic load-acquire/release-store
+template <typename T>
+static inline T loadAcquire(volatile T& var) {
+  return __atomic_load_n(&var, __ATOMIC_ACQUIRE);
 }
 
-static inline void storeRelease(volatile size_t &var, size_t value) {
+template <typename T>
+static inline void storeRelease(volatile T& var, T value) {
   return __atomic_store_n(&var, value, __ATOMIC_RELEASE);
 }
 
diff --git a/ddprof-lib/src/main/cpp/callTraceHashTable.cpp b/ddprof-lib/src/main/cpp/callTraceHashTable.cpp
index 1c32150bc..c10f4fd98 100644
--- a/ddprof-lib/src/main/cpp/callTraceHashTable.cpp
+++ b/ddprof-lib/src/main/cpp/callTraceHashTable.cpp
@@ -301,7 +301,7 @@ u64 CallTraceHashTable::put(int num_frames, ASGCT_CallFrame *frames,
 
     if (++step >= capacity) {
       // Very unlikely case of a table overflow
-      atomicInc(_overflow);
+      atomicIncRelaxed(_overflow);
       return OVERFLOW_TRACE_ID;
     }
     // Improved version of linear probing
@@ -359,7 +359,7 @@ void CallTraceHashTable::collectAndCopySelective(std::unordered_set<CallTrace *>
     traces.insert(&_overflow_trace);
     if (trace_ids_to_preserve.find(OVERFLOW_TRACE_ID) != trace_ids_to_preserve.end()) {
       // Copy overflow trace to target - it's a static trace so just increment overflow counter
-      atomicInc(target->_overflow);
+      atomicIncRelaxed(target->_overflow);
     }
   }
 }
diff --git a/ddprof-lib/src/main/cpp/callTraceHashTable.h b/ddprof-lib/src/main/cpp/callTraceHashTable.h
index 5dceb65b4..2541d12fb 100644
--- a/ddprof-lib/src/main/cpp/callTraceHashTable.h
+++ b/ddprof-lib/src/main/cpp/callTraceHashTable.h
@@ -52,7 +52,7 @@ class CallTraceHashTable {
 
   LinearAllocator _allocator;
   LongHashTable *_current_table;
-  u64 _overflow;
+  volatile u64 _overflow;
 
   u64 calcHash(int num_frames, ASGCT_CallFrame *frames, bool truncated);
   CallTrace *storeCallTrace(int num_frames, ASGCT_CallFrame *frames,
diff --git a/ddprof-lib/src/main/cpp/counters.h b/ddprof-lib/src/main/cpp/counters.h
index 1df8fe3f5..b276f029d 100644
--- a/ddprof-lib/src/main/cpp/counters.h
+++ b/ddprof-lib/src/main/cpp/counters.h
@@ -124,9 +124,9 @@ class Counters {
   static void increment(CounterId counter, long long delta = 1,
                         int offset = 0) {
 #ifdef COUNTERS
-    atomicInc(Counters::instance()
-                  ._counters[address(static_cast<int>(counter) + offset)],
-              delta);
+    atomicIncRelaxed(Counters::instance()
+                         ._counters[address(static_cast<int>(counter) + offset)],
+                     delta);
 #endif // COUNTERS
   }
 
diff --git a/ddprof-lib/src/main/cpp/livenessTracker.cpp b/ddprof-lib/src/main/cpp/livenessTracker.cpp
index e085f3cdd..1923c991f 100644
--- a/ddprof-lib/src/main/cpp/livenessTracker.cpp
+++ b/ddprof-lib/src/main/cpp/livenessTracker.cpp
@@ -26,12 +26,12 @@ constexpr int LivenessTracker::MAX_TRACKING_TABLE_SIZE;
 constexpr int LivenessTracker::MIN_SAMPLING_INTERVAL;
 
 void LivenessTracker::cleanup_table(bool forced) {
-  u64 current = loadAcquire(_last_gc_epoch);
-  u64 target_gc_epoch = loadAcquire(_gc_epoch);
+  u64 current = load(_last_gc_epoch);
+  u64 target_gc_epoch = load(_gc_epoch);
 
   if ((target_gc_epoch == _last_gc_epoch ||
-       !__sync_bool_compare_and_swap(&_last_gc_epoch, current,
-                                     target_gc_epoch)) &&
+       !__atomic_compare_exchange_n(&_last_gc_epoch, &current,
+                                     target_gc_epoch, false, __ATOMIC_RELAXED, __ATOMIC_RELAXED)) &&
       !forced) {
     // if the last processed GC epoch hasn't changed, or if we failed to update
     // it, there's nothing to do
@@ -383,10 +383,10 @@ void LivenessTracker::onGC() {
   }
 
   // just increment the epoch
-  atomicInc(_gc_epoch, 1);
+  atomicIncRelaxed(_gc_epoch,u64(1));
 
   if (!ddprof::HeapUsage::isLastGCUsageSupported()) {
-    storeRelease(_used_after_last_gc, ddprof::HeapUsage::get(false)._used);
+    store(_used_after_last_gc, ddprof::HeapUsage::get(false)._used);
   }
 }
 
diff --git a/ddprof-lib/src/main/cpp/profiler.cpp b/ddprof-lib/src/main/cpp/profiler.cpp
index e51b3dc70..727746367 100644
--- a/ddprof-lib/src/main/cpp/profiler.cpp
+++ b/ddprof-lib/src/main/cpp/profiler.cpp
@@ -559,7 +559,7 @@ int Profiler::getJavaTraceAsync(void *ucontext, ASGCT_CallFrame *frames,
     return 0;
   }
 
-  atomicInc(_failures[-trace.num_frames]);
+  atomicIncRelaxed(_failures[-trace.num_frames]);
   trace.frames->bci = BCI_ERROR;
   trace.frames->method_id = (jmethodID)err_string;
   return trace.frames - frames + 1;
@@ -607,14 +607,14 @@ void Profiler::fillFrameTypes(ASGCT_CallFrame *frames, int num_frames,
 }
 
 u64 Profiler::recordJVMTISample(u64 counter, int tid, jthread thread, jint event_type, Event *event, bool deferred) {
-  atomicInc(_total_samples);
+  atomicIncRelaxed(_total_samples);
 
   u32 lock_index = getLockIndex(tid);
   if (!_locks[lock_index].tryLock() &&
       !_locks[lock_index = (lock_index + 1) % CONCURRENCY_LEVEL].tryLock() &&
       !_locks[lock_index = (lock_index + 2) % CONCURRENCY_LEVEL].tryLock()) {
     // Too many concurrent signals already
-    atomicInc(_failures[-ticks_skipped]);
+    atomicIncRelaxed(_failures[-ticks_skipped]);
 
     return 0;
   }
@@ -654,14 +654,14 @@ u64 Profiler::recordJVMTISample(u64 counter, int tid, jthread thread, jint event
 }
 
 void Profiler::recordDeferredSample(int tid, u64 call_trace_id, jint event_type, Event *event) {
-  atomicInc(_total_samples);
+  atomicIncRelaxed(_total_samples);
 
   u32 lock_index = getLockIndex(tid);
   if (!_locks[lock_index].tryLock() &&
       !_locks[lock_index = (lock_index + 1) % CONCURRENCY_LEVEL].tryLock() &&
       !_locks[lock_index = (lock_index + 2) % CONCURRENCY_LEVEL].tryLock()) {
     // Too many concurrent signals already
-    atomicInc(_failures[-ticks_skipped]);
+    atomicIncRelaxed(_failures[-ticks_skipped]);
     return;
   }
 
@@ -672,14 +672,14 @@ void Profiler::recordDeferredSample(int tid, u64 call_trace_id, jint event_type,
 
 void Profiler::recordSample(void *ucontext, u64 counter, int tid,
                             jint event_type, u64 call_trace_id, Event *event) {
-  atomicInc(_total_samples);
+  atomicIncRelaxed(_total_samples);
 
   u32 lock_index = getLockIndex(tid);
   if (!_locks[lock_index].tryLock() &&
       !_locks[lock_index = (lock_index + 1) % CONCURRENCY_LEVEL].tryLock() &&
       !_locks[lock_index = (lock_index + 2) % CONCURRENCY_LEVEL].tryLock()) {
     // Too many concurrent signals already
-    atomicInc(_failures[-ticks_skipped]);
+    atomicIncRelaxed(_failures[-ticks_skipped]);
 
     if (event_type == BCI_CPU && _cpu_engine == &perf_events) {
       // Need to reset PerfEvents ring buffer, even though we discard the
@@ -788,7 +788,7 @@ void Profiler::recordQueueTime(int tid, QueueTimeEvent *event) {
 void Profiler::recordExternalSample(u64 weight, int tid, int num_frames,
                                     ASGCT_CallFrame *frames, bool truncated,
                                     jint event_type, Event *event) {
-  atomicInc(_total_samples);
+  atomicIncRelaxed(_total_samples);
 
   u64 call_trace_id =
       _call_trace_storage.put(num_frames, frames, truncated, weight);
@@ -798,7 +798,7 @@ void Profiler::recordExternalSample(u64 weight, int tid, int num_frames,
       !_locks[lock_index = (lock_index + 1) % CONCURRENCY_LEVEL].tryLock() &&
       !_locks[lock_index = (lock_index + 2) % CONCURRENCY_LEVEL].tryLock()) {
     // Too many concurrent signals already
-    atomicInc(_failures[-ticks_skipped]);
+    atomicIncRelaxed(_failures[-ticks_skipped]);
     return;
   }
 
diff --git a/ddprof-lib/src/main/cpp/profiler.h b/ddprof-lib/src/main/cpp/profiler.h
index 1e8e9902f..09229d845 100644
--- a/ddprof-lib/src/main/cpp/profiler.h
+++ b/ddprof-lib/src/main/cpp/profiler.h
@@ -93,7 +93,7 @@ class Profiler {
   WaitableMutex _timer_lock;
   void *_timer_id;
 
-  u64 _total_samples;
+  volatile u64 _total_samples;
   u64 _failures[ASGCT_FAILURE_TYPES];
 
   SpinLock _class_map_lock;