Merge pull request #45617 from RandomShaper/modernize_atomics

Modernize atomics (and fix `volatile`)

17 files changed, 302 insertions, 337 deletions

diff --git a/core/core_bind.cpp b/core/core_bind.cpp
index dd99c32fa8..fe49899aa2 100644
--- a/core/core_bind.cpp
+++ b/core/core_bind.cpp
@@ -1990,7 +1990,7 @@ void _Thread::_start_func(void *ud) {
 }
 
 Error _Thread::start(Object *p_instance, const StringName &p_method, const Variant &p_userdata, Priority p_priority) {
-	ERR_FAIL_COND_V_MSG(active, ERR_ALREADY_IN_USE, "Thread already started.");
+	ERR_FAIL_COND_V_MSG(active.is_set(), ERR_ALREADY_IN_USE, "Thread already started.");
 	ERR_FAIL_COND_V(!p_instance, ERR_INVALID_PARAMETER);
 	ERR_FAIL_COND_V(p_method == StringName(), ERR_INVALID_PARAMETER);
 	ERR_FAIL_INDEX_V(p_priority, PRIORITY_MAX, ERR_INVALID_PARAMETER);
@@ -1999,7 +1999,7 @@ Error _Thread::start(Object *p_instance, const StringName &p_method, const Varia
 	target_method = p_method;
 	target_instance = p_instance;
 	userdata = p_userdata;
-	active = true;
+	active.set();
 
 	Ref<_Thread> *ud = memnew(Ref<_Thread>(this));
 
@@ -2015,14 +2015,14 @@ String _Thread::get_id() const {
 }
 
 bool _Thread::is_active() const {
-	return active;
+	return active.is_set();
 }
 
 Variant _Thread::wait_to_finish() {
-	ERR_FAIL_COND_V_MSG(!active, Variant(), "Thread must be active to wait for its completion.");
+	ERR_FAIL_COND_V_MSG(!active.is_set(), Variant(), "Thread must be active to wait for its completion.");
 	thread.wait_to_finish();
 	Variant r = ret;
-	active = false;
+	active.clear();
 	target_method = StringName();
 	target_instance = nullptr;
 	userdata = Variant();
diff --git a/core/core_bind.h b/core/core_bind.h
index 7f945a9314..46a7a94f2d 100644
--- a/core/core_bind.h
+++ b/core/core_bind.h
@@ -40,6 +40,7 @@
 #include "core/os/os.h"
 #include "core/os/semaphore.h"
 #include "core/os/thread.h"
+#include "core/templates/safe_refcount.h"
 
 class _ResourceLoader : public Object {
 	GDCLASS(_ResourceLoader, Object);
@@ -559,7 +560,7 @@ class _Thread : public Reference {
 protected:
 	Variant ret;
 	Variant userdata;
-	volatile bool active = false;
+	SafeFlag active;
 	Object *target_instance = nullptr;
 	StringName target_method;
 	Thread thread;
diff --git a/core/error/error_macros.h b/core/error/error_macros.h
index 8eb6217ce8..f909a67d55 100644
--- a/core/error/error_macros.h
+++ b/core/error/error_macros.h
@@ -33,6 +33,8 @@
 
 #include "core/typedefs.h"
 
+#include "core/templates/safe_refcount.h"
+
 class String;
 
 enum ErrorHandlerType {
@@ -577,10 +579,10 @@ void _err_print_index_error(const char *p_function, const char *p_file, int p_li
  */
 #define WARN_DEPRECATED                                                                                                                                    \
 	if (true) {                                                                                                                                            \
-		static volatile bool warning_shown = false;                                                                                                        \
-		if (!warning_shown) {                                                                                                                              \
+		static SafeFlag warning_shown;                                                                                                                     \
+		if (!warning_shown.is_set()) {                                                                                                                     \
 			_err_print_error(FUNCTION_STR, __FILE__, __LINE__, "This method has been deprecated and will be removed in the future.", ERR_HANDLER_WARNING); \
-			warning_shown = true;                                                                                                                          \
+			warning_shown.set();                                                                                                                           \
 		}                                                                                                                                                  \
 	} else                                                                                                                                                 \
 		((void)0)
@@ -590,10 +592,10 @@ void _err_print_index_error(const char *p_function, const char *p_file, int p_li
  */
 #define WARN_DEPRECATED_MSG(m_msg)                                                                                                                                           \
 	if (true) {                                                                                                                                                              \
-		static volatile bool warning_shown = false;                                                                                                                          \
-		if (!warning_shown) {                                                                                                                                                \
+		static SafeFlag warning_shown;                                                                                                                                       \
+		if (!warning_shown.is_set()) {                                                                                                                                       \
 			_err_print_error(FUNCTION_STR, __FILE__, __LINE__, "This method has been deprecated and will be removed in the future.", DEBUG_STR(m_msg), ERR_HANDLER_WARNING); \
-			warning_shown = true;                                                                                                                                            \
+			warning_shown.set();                                                                                                                                             \
 		}                                                                                                                                                                    \
 	} else                                                                                                                                                                   \
 		((void)0)
diff --git a/core/io/ip.cpp b/core/io/ip.cpp
index df95785000..e1d9c19f10 100644
--- a/core/io/ip.cpp
+++ b/core/io/ip.cpp
@@ -40,13 +40,13 @@ VARIANT_ENUM_CAST(IP::ResolverStatus);
 
 struct _IP_ResolverPrivate {
 	struct QueueItem {
-		volatile IP::ResolverStatus status;
+		SafeNumeric<IP::ResolverStatus> status;
 		IP_Address response;
 		String hostname;
 		IP::Type type;
 
 		void clear() {
-			status = IP::RESOLVER_STATUS_NONE;
+			status.set(IP::RESOLVER_STATUS_NONE);
 			response = IP_Address();
 			type = IP::TYPE_NONE;
 			hostname = "";
@@ -61,7 +61,7 @@ struct _IP_ResolverPrivate {
 
 	IP::ResolverID find_empty_id() const {
 		for (int i = 0; i < IP::RESOLVER_MAX_QUERIES; i++) {
-			if (queue[i].status == IP::RESOLVER_STATUS_NONE) {
+			if (queue[i].status.get() == IP::RESOLVER_STATUS_NONE) {
 				return i;
 			}
 		}
@@ -77,15 +77,15 @@ struct _IP_ResolverPrivate {
 
 	void resolve_queues() {
 		for (int i = 0; i < IP::RESOLVER_MAX_QUERIES; i++) {
-			if (queue[i].status != IP::RESOLVER_STATUS_WAITING) {
+			if (queue[i].status.get() != IP::RESOLVER_STATUS_WAITING) {
 				continue;
 			}
 			queue[i].response = IP::get_singleton()->resolve_hostname(queue[i].hostname, queue[i].type);
 
 			if (!queue[i].response.is_valid()) {
-				queue[i].status = IP::RESOLVER_STATUS_ERROR;
+				queue[i].status.set(IP::RESOLVER_STATUS_ERROR);
 			} else {
-				queue[i].status = IP::RESOLVER_STATUS_DONE;
+				queue[i].status.set(IP::RESOLVER_STATUS_DONE);
 			}
 		}
 	}
@@ -137,10 +137,10 @@ IP::ResolverID IP::resolve_hostname_queue_item(const String &p_hostname, IP::Typ
 	resolver->queue[id].type = p_type;
 	if (resolver->cache.has(key) && resolver->cache[key].is_valid()) {
 		resolver->queue[id].response = resolver->cache[key];
-		resolver->queue[id].status = IP::RESOLVER_STATUS_DONE;
+		resolver->queue[id].status.set(IP::RESOLVER_STATUS_DONE);
 	} else {
 		resolver->queue[id].response = IP_Address();
-		resolver->queue[id].status = IP::RESOLVER_STATUS_WAITING;
+		resolver->queue[id].status.set(IP::RESOLVER_STATUS_WAITING);
 		if (resolver->thread.is_started()) {
 			resolver->sem.post();
 		} else {
@@ -156,12 +156,12 @@ IP::ResolverStatus IP::get_resolve_item_status(ResolverID p_id) const {
 
 	MutexLock lock(resolver->mutex);
 
-	if (resolver->queue[p_id].status == IP::RESOLVER_STATUS_NONE) {
+	if (resolver->queue[p_id].status.get() == IP::RESOLVER_STATUS_NONE) {
 		ERR_PRINT("Condition status == IP::RESOLVER_STATUS_NONE");
 		resolver->mutex.unlock();
 		return IP::RESOLVER_STATUS_NONE;
 	}
-	return resolver->queue[p_id].status;
+	return resolver->queue[p_id].status.get();
 }
 
 IP_Address IP::get_resolve_item_address(ResolverID p_id) const {
@@ -169,7 +169,7 @@ IP_Address IP::get_resolve_item_address(ResolverID p_id) const {
 
 	MutexLock lock(resolver->mutex);
 
-	if (resolver->queue[p_id].status != IP::RESOLVER_STATUS_DONE) {
+	if (resolver->queue[p_id].status.get() != IP::RESOLVER_STATUS_DONE) {
 		ERR_PRINT("Resolve of '" + resolver->queue[p_id].hostname + "'' didn't complete yet.");
 		resolver->mutex.unlock();
 		return IP_Address();
@@ -183,7 +183,7 @@ void IP::erase_resolve_item(ResolverID p_id) {
 
 	MutexLock lock(resolver->mutex);
 
-	resolver->queue[p_id].status = IP::RESOLVER_STATUS_NONE;
+	resolver->queue[p_id].status.set(IP::RESOLVER_STATUS_NONE);
 }
 
 void IP::clear_cache(const String &p_hostname) {
diff --git a/core/object/object.cpp b/core/object/object.cpp
index 8f2eed3200..1a9cce49d8 100644
--- a/core/object/object.cpp
+++ b/core/object/object.cpp
@@ -1727,7 +1727,7 @@ void *Object::get_script_instance_binding(int p_script_language_index) {
 	if (!_script_instance_bindings[p_script_language_index]) {
 		void *script_data = ScriptServer::get_language(p_script_language_index)->alloc_instance_binding_data(this);
 		if (script_data) {
-			atomic_increment(&instance_binding_count);
+			instance_binding_count.increment();
 			_script_instance_bindings[p_script_language_index] = script_data;
 		}
 	}
diff --git a/core/object/object.h b/core/object/object.h
index 5021fa47a1..029478873d 100644
--- a/core/object/object.h
+++ b/core/object/object.h
@@ -37,6 +37,7 @@
 #include "core/templates/hash_map.h"
 #include "core/templates/list.h"
 #include "core/templates/map.h"
+#include "core/templates/safe_refcount.h"
 #include "core/templates/set.h"
 #include "core/templates/vmap.h"
 #include "core/variant/callable_bind.h"
@@ -486,7 +487,7 @@ private:
 
 	friend class Reference;
 	bool type_is_reference = false;
-	uint32_t instance_binding_count = 0;
+	SafeNumeric<uint32_t> instance_binding_count;
 	void *_script_instance_bindings[MAX_SCRIPT_INSTANCE_BINDINGS];
 
 	Object(bool p_reference);
diff --git a/core/object/reference.cpp b/core/object/reference.cpp
index 71a52a9ba5..22e4e8a336 100644
--- a/core/object/reference.cpp
+++ b/core/object/reference.cpp
@@ -62,7 +62,7 @@ bool Reference::reference() {
 		if (get_script_instance()) {
 			get_script_instance()->refcount_incremented();
 		}
-		if (instance_binding_count > 0 && !ScriptServer::are_languages_finished()) {
+		if (instance_binding_count.get() > 0 && !ScriptServer::are_languages_finished()) {
 			for (int i = 0; i < MAX_SCRIPT_INSTANCE_BINDINGS; i++) {
 				if (_script_instance_bindings[i]) {
 					ScriptServer::get_language(i)->refcount_incremented_instance_binding(this);
@@ -83,7 +83,7 @@ bool Reference::unreference() {
 			bool script_ret = get_script_instance()->refcount_decremented();
 			die = die && script_ret;
 		}
-		if (instance_binding_count > 0 && !ScriptServer::are_languages_finished()) {
+		if (instance_binding_count.get() > 0 && !ScriptServer::are_languages_finished()) {
 			for (int i = 0; i < MAX_SCRIPT_INSTANCE_BINDINGS; i++) {
 				if (_script_instance_bindings[i]) {
 					bool script_ret = ScriptServer::get_language(i)->refcount_decremented_instance_binding(this);
diff --git a/core/os/memory.cpp b/core/os/memory.cpp
index 14808c2ce6..5910cb0e7b 100644
--- a/core/os/memory.cpp
+++ b/core/os/memory.cpp
@@ -60,11 +60,11 @@ void operator delete(void *p_mem, void *p_pointer, size_t check, const char *p_d
 #endif
 
 #ifdef DEBUG_ENABLED
-uint64_t Memory::mem_usage = 0;
-uint64_t Memory::max_usage = 0;
+SafeNumeric<uint64_t> Memory::mem_usage;
+SafeNumeric<uint64_t> Memory::max_usage;
 #endif
 
-uint64_t Memory::alloc_count = 0;
+SafeNumeric<uint64_t> Memory::alloc_count;
 
 void *Memory::alloc_static(size_t p_bytes, bool p_pad_align) {
 #ifdef DEBUG_ENABLED
@@ -77,7 +77,7 @@ void *Memory::alloc_static(size_t p_bytes, bool p_pad_align) {
 
 	ERR_FAIL_COND_V(!mem, nullptr);
 
-	atomic_increment(&alloc_count);
+	alloc_count.increment();
 
 	if (prepad) {
 		uint64_t *s = (uint64_t *)mem;
@@ -86,8 +86,8 @@ void *Memory::alloc_static(size_t p_bytes, bool p_pad_align) {
 		uint8_t *s8 = (uint8_t *)mem;
 
 #ifdef DEBUG_ENABLED
-		atomic_add(&mem_usage, p_bytes);
-		atomic_exchange_if_greater(&max_usage, mem_usage);
+		uint64_t new_mem_usage = mem_usage.add(p_bytes);
+		max_usage.exchange_if_greater(new_mem_usage);
 #endif
 		return s8 + PAD_ALIGN;
 	} else {
@@ -114,10 +114,10 @@ void *Memory::realloc_static(void *p_memory, size_t p_bytes, bool p_pad_align) {
 
 #ifdef DEBUG_ENABLED
 		if (p_bytes > *s) {
-			atomic_add(&mem_usage, p_bytes - *s);
-			atomic_exchange_if_greater(&max_usage, mem_usage);
+			uint64_t new_mem_usage = mem_usage.add(p_bytes - *s);
+			max_usage.exchange_if_greater(new_mem_usage);
 		} else {
-			atomic_sub(&mem_usage, *s - p_bytes);
+			mem_usage.sub(*s - p_bytes);
 		}
 #endif
 
@@ -156,14 +156,14 @@ void Memory::free_static(void *p_ptr, bool p_pad_align) {
 	bool prepad = p_pad_align;
 #endif
 
-	atomic_decrement(&alloc_count);
+	alloc_count.decrement();
 
 	if (prepad) {
 		mem -= PAD_ALIGN;
 
 #ifdef DEBUG_ENABLED
 		uint64_t *s = (uint64_t *)mem;
-		atomic_sub(&mem_usage, *s);
+		mem_usage.sub(*s);
 #endif
 
 		free(mem);
@@ -178,7 +178,7 @@ uint64_t Memory::get_mem_available() {
 
 uint64_t Memory::get_mem_usage() {
 #ifdef DEBUG_ENABLED
-	return mem_usage;
+	return mem_usage.get();
 #else
 	return 0;
 #endif
@@ -186,7 +186,7 @@ uint64_t Memory::get_mem_usage() {
 
 uint64_t Memory::get_mem_max_usage() {
 #ifdef DEBUG_ENABLED
-	return max_usage;
+	return max_usage.get();
 #else
 	return 0;
 #endif
diff --git a/core/os/memory.h b/core/os/memory.h
index c2ae3f4ae7..10e678103d 100644
--- a/core/os/memory.h
+++ b/core/os/memory.h
@@ -43,11 +43,11 @@
 class Memory {
 	Memory();
 #ifdef DEBUG_ENABLED
-	static uint64_t mem_usage;
-	static uint64_t max_usage;
+	static SafeNumeric<uint64_t> mem_usage;
+	static SafeNumeric<uint64_t> max_usage;
 #endif
 
-	static uint64_t alloc_count;
+	static SafeNumeric<uint64_t> alloc_count;
 
 public:
 	static void *alloc_static(size_t p_bytes, bool p_pad_align = false);
diff --git a/core/os/thread.cpp b/core/os/thread.cpp
index 936e5d5500..88744eed63 100644
--- a/core/os/thread.cpp
+++ b/core/os/thread.cpp
@@ -34,13 +34,15 @@
 
 #if !defined(NO_THREADS)
 
+#include "core/templates/safe_refcount.h"
+
 Error (*Thread::set_name_func)(const String &) = nullptr;
 void (*Thread::set_priority_func)(Thread::Priority) = nullptr;
 void (*Thread::init_func)() = nullptr;
 void (*Thread::term_func)() = nullptr;
 
 Thread::ID Thread::main_thread_id = 1;
-Thread::ID Thread::last_thread_id = 1;
+SafeNumeric<Thread::ID> Thread::last_thread_id{ 1 };
 thread_local Thread::ID Thread::caller_id = 1;
 
 void Thread::_set_platform_funcs(
@@ -79,7 +81,7 @@ void Thread::start(Thread::Callback p_callback, void *p_user, const Settings &p_
 		std::thread empty_thread;
 		thread.swap(empty_thread);
 	}
-	id = atomic_increment(&last_thread_id);
+	id = last_thread_id.increment();
 	std::thread new_thread(&Thread::callback, this, p_settings, p_callback, p_user);
 	thread.swap(new_thread);
 }
diff --git a/core/os/thread.h b/core/os/thread.h
index b5449d2ed6..76f5be182e 100644
--- a/core/os/thread.h
+++ b/core/os/thread.h
@@ -34,6 +34,7 @@
 #include "core/typedefs.h"
 
 #if !defined(NO_THREADS)
+#include "core/templates/safe_refcount.h"
 #include <thread>
 #endif
 
@@ -61,7 +62,7 @@ private:
 	friend class Main;
 
 	static ID main_thread_id;
-	static ID last_thread_id;
+	static SafeNumeric<Thread::ID> last_thread_id;
 
 	ID id = 0;
 	static thread_local ID caller_id;
diff --git a/core/os/threaded_array_processor.h b/core/os/threaded_array_processor.h
index 9538ac5957..4f270001d3 100644
--- a/core/os/threaded_array_processor.h
+++ b/core/os/threaded_array_processor.h
@@ -40,7 +40,7 @@
 template <class C, class U>
 struct ThreadArrayProcessData {
 	uint32_t elements;
-	uint32_t index;
+	SafeNumeric<uint32_t> index;
 	C *instance;
 	U userdata;
 	void (C::*method)(uint32_t, U);
@@ -56,7 +56,7 @@ template <class T>
 void process_array_thread(void *ud) {
 	T &data = *(T *)ud;
 	while (true) {
-		uint32_t index = atomic_increment(&data.index);
+		uint32_t index = data.index.increment();
 		if (index >= data.elements) {
 			break;
 		}
@@ -70,9 +70,9 @@ void thread_process_array(uint32_t p_elements, C *p_instance, M p_method, U p_us
 	data.method = p_method;
 	data.instance = p_instance;
 	data.userdata = p_userdata;
-	data.index = 0;
+	data.index.set(0);
 	data.elements = p_elements;
-	data.process(data.index); //process first, let threads increment for next
+	data.process(0); //process first, let threads increment for next
 
 	int thread_count = OS::get_singleton()->get_processor_count();
 	Thread *threads = memnew_arr(Thread, thread_count);
diff --git a/core/templates/cowdata.h b/core/templates/cowdata.h
index 4becb1be59..6077f5f107 100644
--- a/core/templates/cowdata.h
+++ b/core/templates/cowdata.h
@@ -45,6 +45,9 @@ class CharString;
 template <class T, class V>
 class VMap;
 
+// CowData is relying on this to be true
+static_assert(sizeof(SafeNumeric<uint32_t>) == sizeof(uint32_t));
+
 template <class T>
 class CowData {
 	template <class TV>
@@ -60,12 +63,12 @@ private:
 
 	// internal helpers
 
-	_FORCE_INLINE_ uint32_t *_get_refcount() const {
+	_FORCE_INLINE_ SafeNumeric<uint32_t> *_get_refcount() const {
 		if (!_ptr) {
 			return nullptr;
 		}
 
-		return reinterpret_cast<uint32_t *>(_ptr) - 2;
+		return reinterpret_cast<SafeNumeric<uint32_t> *>(_ptr) - 2;
 	}
 
 	_FORCE_INLINE_ uint32_t *_get_size() const {
@@ -192,9 +195,9 @@ void CowData<T>::_unref(void *p_data) {
 		return;
 	}
 
-	uint32_t *refc = _get_refcount();
+	SafeNumeric<uint32_t> *refc = _get_refcount();
 
-	if (atomic_decrement(refc) > 0) {
+	if (refc->decrement() > 0) {
 		return; // still in use
 	}
 	// clean up
@@ -219,15 +222,15 @@ void CowData<T>::_copy_on_write() {
 		return;
 	}
 
-	uint32_t *refc = _get_refcount();
+	SafeNumeric<uint32_t> *refc = _get_refcount();
 
-	if (unlikely(*refc > 1)) {
+	if (unlikely(refc->get() > 1)) {
 		/* in use by more than me */
 		uint32_t current_size = *_get_size();
 
 		uint32_t *mem_new = (uint32_t *)Memory::alloc_static(_get_alloc_size(current_size), true);
 
-		*(mem_new - 2) = 1; //refcount
+		reinterpret_cast<SafeNumeric<uint32_t> *>(mem_new - 2)->set(1); //refcount
 		*(mem_new - 1) = current_size; //size
 
 		T *_data = (T *)(mem_new);
@@ -278,7 +281,7 @@ Error CowData<T>::resize(int p_size) {
 				uint32_t *ptr = (uint32_t *)Memory::alloc_static(alloc_size, true);
 				ERR_FAIL_COND_V(!ptr, ERR_OUT_OF_MEMORY);
 				*(ptr - 1) = 0; //size, currently none
-				*(ptr - 2) = 1; //refcount
+				reinterpret_cast<SafeNumeric<uint32_t> *>(ptr - 2)->set(1); //refcount
 
 				_ptr = (T *)ptr;
 
@@ -359,7 +362,7 @@ void CowData<T>::_ref(const CowData &p_from) {
 		return; //nothing to do
 	}
 
-	if (atomic_conditional_increment(p_from._get_refcount()) > 0) { // could reference
+	if (p_from._get_refcount()->increment() > 0) { // could reference
 		_ptr = p_from._ptr;
 	}
 }
diff --git a/core/templates/rid_owner.cpp b/core/templates/rid_owner.cpp
index f75a2eb9df..56f39ab779 100644
--- a/core/templates/rid_owner.cpp
+++ b/core/templates/rid_owner.cpp
@@ -30,4 +30,4 @@
 
 #include "rid_owner.h"
 
-volatile uint64_t RID_AllocBase::base_id = 1;
+SafeNumeric<uint64_t> RID_AllocBase::base_id{ 1 };
diff --git a/core/templates/rid_owner.h b/core/templates/rid_owner.h
index 3edc73b1a9..c4aa93c394 100644
--- a/core/templates/rid_owner.h
+++ b/core/templates/rid_owner.h
@@ -44,7 +44,7 @@
 #include <typeinfo>
 
 class RID_AllocBase {
-	static volatile uint64_t base_id;
+	static SafeNumeric<uint64_t> base_id;
 
 protected:
 	static RID _make_from_id(uint64_t p_id) {
@@ -54,7 +54,7 @@ protected:
 	}
 
 	static uint64_t _gen_id() {
-		return atomic_increment(&base_id);
+		return base_id.increment();
 	}
 
 	static RID _gen_rid() {
diff --git a/core/templates/safe_refcount.cpp b/core/templates/safe_refcount.cpp
deleted file mode 100644
index a915ee662f..0000000000
--- a/core/templates/safe_refcount.cpp
+++ /dev/null
@@ -1,161 +0,0 @@
-/*************************************************************************/
-/*  safe_refcount.cpp                                                    */
-/*************************************************************************/
-/*                       This file is part of:                           */
-/*                           GODOT ENGINE                                */
-/*                      https://godotengine.org                          */
-/*************************************************************************/
-/* Copyright (c) 2007-2021 Juan Linietsky, Ariel Manzur.                 */
-/* Copyright (c) 2014-2021 Godot Engine contributors (cf. AUTHORS.md).   */
-/*                                                                       */
-/* Permission is hereby granted, free of charge, to any person obtaining */
-/* a copy of this software and associated documentation files (the       */
-/* "Software"), to deal in the Software without restriction, including   */
-/* without limitation the rights to use, copy, modify, merge, publish,   */
-/* distribute, sublicense, and/or sell copies of the Software, and to    */
-/* permit persons to whom the Software is furnished to do so, subject to */
-/* the following conditions:                                             */
-/*                                                                       */
-/* The above copyright notice and this permission notice shall be        */
-/* included in all copies or substantial portions of the Software.       */
-/*                                                                       */
-/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,       */
-/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF    */
-/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/
-/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY  */
-/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,  */
-/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE     */
-/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.                */
-/*************************************************************************/
-
-#include "safe_refcount.h"
-
-#if defined(_MSC_VER)
-
-/* Implementation for MSVC-Windows */
-
-// don't pollute my namespace!
-#include <windows.h>
-
-#define ATOMIC_CONDITIONAL_INCREMENT_BODY(m_pw, m_win_type, m_win_cmpxchg, m_cpp_type) \
-	/* try to increment until it actually works */                                     \
-	/* taken from boost */                                                             \
-	while (true) {                                                                     \
-		m_cpp_type tmp = static_cast<m_cpp_type const volatile &>(*(m_pw));            \
-		if (tmp == 0) {                                                                \
-			return 0; /* if zero, can't add to it anymore */                           \
-		}                                                                              \
-		if (m_win_cmpxchg((m_win_type volatile *)(m_pw), tmp + 1, tmp) == tmp) {       \
-			return tmp + 1;                                                            \
-		}                                                                              \
-	}
-
-#define ATOMIC_EXCHANGE_IF_GREATER_BODY(m_pw, m_val, m_win_type, m_win_cmpxchg, m_cpp_type) \
-	while (true) {                                                                          \
-		m_cpp_type tmp = static_cast<m_cpp_type const volatile &>(*(m_pw));                 \
-		if (tmp >= m_val) {                                                                 \
-			return tmp; /* already greater, or equal */                                     \
-		}                                                                                   \
-		if (m_win_cmpxchg((m_win_type volatile *)(m_pw), m_val, tmp) == tmp) {              \
-			return m_val;                                                                   \
-		}                                                                                   \
-	}
-
-_ALWAYS_INLINE_ uint32_t _atomic_conditional_increment_impl(volatile uint32_t *pw) {
-	ATOMIC_CONDITIONAL_INCREMENT_BODY(pw, LONG, InterlockedCompareExchange, uint32_t);
-}
-
-_ALWAYS_INLINE_ uint32_t _atomic_decrement_impl(volatile uint32_t *pw) {
-	return InterlockedDecrement((LONG volatile *)pw);
-}
-
-_ALWAYS_INLINE_ uint32_t _atomic_increment_impl(volatile uint32_t *pw) {
-	return InterlockedIncrement((LONG volatile *)pw);
-}
-
-_ALWAYS_INLINE_ uint32_t _atomic_sub_impl(volatile uint32_t *pw, volatile uint32_t val) {
-	return InterlockedExchangeAdd((LONG volatile *)pw, -(int32_t)val) - val;
-}
-
-_ALWAYS_INLINE_ uint32_t _atomic_add_impl(volatile uint32_t *pw, volatile uint32_t val) {
-	return InterlockedAdd((LONG volatile *)pw, val);
-}
-
-_ALWAYS_INLINE_ uint32_t _atomic_exchange_if_greater_impl(volatile uint32_t *pw, volatile uint32_t val) {
-	ATOMIC_EXCHANGE_IF_GREATER_BODY(pw, val, LONG, InterlockedCompareExchange, uint32_t);
-}
-
-_ALWAYS_INLINE_ uint64_t _atomic_conditional_increment_impl(volatile uint64_t *pw) {
-	ATOMIC_CONDITIONAL_INCREMENT_BODY(pw, LONGLONG, InterlockedCompareExchange64, uint64_t);
-}
-
-_ALWAYS_INLINE_ uint64_t _atomic_decrement_impl(volatile uint64_t *pw) {
-	return InterlockedDecrement64((LONGLONG volatile *)pw);
-}
-
-_ALWAYS_INLINE_ uint64_t _atomic_increment_impl(volatile uint64_t *pw) {
-	return InterlockedIncrement64((LONGLONG volatile *)pw);
-}
-
-_ALWAYS_INLINE_ uint64_t _atomic_sub_impl(volatile uint64_t *pw, volatile uint64_t val) {
-	return InterlockedExchangeAdd64((LONGLONG volatile *)pw, -(int64_t)val) - val;
-}
-
-_ALWAYS_INLINE_ uint64_t _atomic_add_impl(volatile uint64_t *pw, volatile uint64_t val) {
-	return InterlockedAdd64((LONGLONG volatile *)pw, val);
-}
-
-_ALWAYS_INLINE_ uint64_t _atomic_exchange_if_greater_impl(volatile uint64_t *pw, volatile uint64_t val) {
-	ATOMIC_EXCHANGE_IF_GREATER_BODY(pw, val, LONGLONG, InterlockedCompareExchange64, uint64_t);
-}
-
-// The actual advertised functions; they'll call the right implementation
-
-uint32_t atomic_conditional_increment(volatile uint32_t *pw) {
-	return _atomic_conditional_increment_impl(pw);
-}
-
-uint32_t atomic_decrement(volatile uint32_t *pw) {
-	return _atomic_decrement_impl(pw);
-}
-
-uint32_t atomic_increment(volatile uint32_t *pw) {
-	return _atomic_increment_impl(pw);
-}
-
-uint32_t atomic_sub(volatile uint32_t *pw, volatile uint32_t val) {
-	return _atomic_sub_impl(pw, val);
-}
-
-uint32_t atomic_add(volatile uint32_t *pw, volatile uint32_t val) {
-	return _atomic_add_impl(pw, val);
-}
-
-uint32_t atomic_exchange_if_greater(volatile uint32_t *pw, volatile uint32_t val) {
-	return _atomic_exchange_if_greater_impl(pw, val);
-}
-
-uint64_t atomic_conditional_increment(volatile uint64_t *pw) {
-	return _atomic_conditional_increment_impl(pw);
-}
-
-uint64_t atomic_decrement(volatile uint64_t *pw) {
-	return _atomic_decrement_impl(pw);
-}
-
-uint64_t atomic_increment(volatile uint64_t *pw) {
-	return _atomic_increment_impl(pw);
-}
-
-uint64_t atomic_sub(volatile uint64_t *pw, volatile uint64_t val) {
-	return _atomic_sub_impl(pw, val);
-}
-
-uint64_t atomic_add(volatile uint64_t *pw, volatile uint64_t val) {
-	return _atomic_add_impl(pw, val);
-}
-
-uint64_t atomic_exchange_if_greater(volatile uint64_t *pw, volatile uint64_t val) {
-	return _atomic_exchange_if_greater_impl(pw, val);
-}
-#endif
diff --git a/core/templates/safe_refcount.h b/core/templates/safe_refcount.h
index d94444fad6..6aebc24ec3 100644
--- a/core/templates/safe_refcount.h
+++ b/core/templates/safe_refcount.h
@@ -31,167 +31,281 @@
 #ifndef SAFE_REFCOUNT_H
 #define SAFE_REFCOUNT_H
 
-#include "core/os/mutex.h"
 #include "core/typedefs.h"
-#include "platform_config.h"
 
-// Atomic functions, these are used for multithread safe reference counters!
+#if !defined(NO_THREADS)
 
-#ifdef NO_THREADS
+#include <atomic>
 
-/* Bogus implementation unaware of multiprocessing */
+// Design goals for these classes:
+// - No automatic conversions or arithmetic operators,
+//   to keep explicit the use of atomics everywhere.
+// - Using acquire-release semantics, even to set the first value.
+//   The first value may be set relaxedly in many cases, but adding the distinction
+//   between relaxed and unrelaxed operation to the interface would make it needlessly
+//   flexible. There's negligible waste in having release semantics for the initial
+//   value and, as an important benefit, you can be sure the value is properly synchronized
+//   even with threads that are already running.
 
 template <class T>
-static _ALWAYS_INLINE_ T atomic_conditional_increment(volatile T *pw) {
-	if (*pw == 0) {
-		return 0;
+class SafeNumeric {
+	std::atomic<T> value;
+
+public:
+	_ALWAYS_INLINE_ void set(T p_value) {
+		value.store(p_value, std::memory_order_release);
 	}
 
-	(*pw)++;
+	_ALWAYS_INLINE_ T get() const {
+		return value.load(std::memory_order_acquire);
+	}
 
-	return *pw;
-}
+	_ALWAYS_INLINE_ T increment() {
+		return value.fetch_add(1, std::memory_order_acq_rel) + 1;
+	}
 
-template <class T>
-static _ALWAYS_INLINE_ T atomic_decrement(volatile T *pw) {
-	(*pw)--;
+	// Returns the original value instead of the new one
+	_ALWAYS_INLINE_ T postincrement() {
+		return value.fetch_add(1, std::memory_order_acq_rel);
+	}
 
-	return *pw;
-}
+	_ALWAYS_INLINE_ T decrement() {
+		return value.fetch_sub(1, std::memory_order_acq_rel) - 1;
+	}
 
-template <class T>
-static _ALWAYS_INLINE_ T atomic_increment(volatile T *pw) {
-	(*pw)++;
+	// Returns the original value instead of the new one
+	_ALWAYS_INLINE_ T postdecrement() {
+		return value.fetch_sub(1, std::memory_order_acq_rel);
+	}
 
-	return *pw;
-}
+	_ALWAYS_INLINE_ T add(T p_value) {
+		return value.fetch_add(p_value, std::memory_order_acq_rel) + p_value;
+	}
 
-template <class T, class V>
-static _ALWAYS_INLINE_ T atomic_sub(volatile T *pw, volatile V val) {
-	(*pw) -= val;
+	// Returns the original value instead of the new one
+	_ALWAYS_INLINE_ T postadd(T p_value) {
+		return value.fetch_add(p_value, std::memory_order_acq_rel);
+	}
 
-	return *pw;
-}
+	_ALWAYS_INLINE_ T sub(T p_value) {
+		return value.fetch_sub(p_value, std::memory_order_acq_rel) - p_value;
+	}
 
-template <class T, class V>
-static _ALWAYS_INLINE_ T atomic_add(volatile T *pw, volatile V val) {
-	(*pw) += val;
+	// Returns the original value instead of the new one
+	_ALWAYS_INLINE_ T postsub(T p_value) {
+		return value.fetch_sub(p_value, std::memory_order_acq_rel);
+	}
 
-	return *pw;
-}
+	_ALWAYS_INLINE_ T exchange_if_greater(T p_value) {
+		while (true) {
+			T tmp = value.load(std::memory_order_acquire);
+			if (tmp >= p_value) {
+				return tmp; // already greater, or equal
+			}
+			if (value.compare_exchange_weak(tmp, p_value, std::memory_order_release)) {
+				return p_value;
+			}
+		}
+	}
 
-template <class T, class V>
-static _ALWAYS_INLINE_ T atomic_exchange_if_greater(volatile T *pw, volatile V val) {
-	if (val > *pw) {
-		*pw = val;
+	_ALWAYS_INLINE_ T conditional_increment() {
+		while (true) {
+			T c = value.load(std::memory_order_acquire);
+			if (c == 0) {
+				return 0;
+			}
+			if (value.compare_exchange_weak(c, c + 1, std::memory_order_release)) {
+				return c + 1;
+			}
+		}
 	}
 
-	return *pw;
-}
+	_ALWAYS_INLINE_ explicit SafeNumeric<T>(T p_value = static_cast<T>(0)) {
+		set(p_value);
+	}
+};
 
-#elif defined(__GNUC__)
+class SafeFlag {
+	std::atomic_bool flag;
 
-/* Implementation for GCC & Clang */
+public:
+	_ALWAYS_INLINE_ bool is_set() const {
+		return flag.load(std::memory_order_acquire);
+	}
 
-// GCC guarantees atomic intrinsics for sizes of 1, 2, 4 and 8 bytes.
-// Clang states it supports GCC atomic builtins.
+	_ALWAYS_INLINE_ void set() {
+		flag.store(true, std::memory_order_release);
+	}
 
-template <class T>
-static _ALWAYS_INLINE_ T atomic_conditional_increment(volatile T *pw) {
-	while (true) {
-		T tmp = static_cast<T const volatile &>(*pw);
-		if (tmp == 0) {
-			return 0; // if zero, can't add to it anymore
-		}
-		if (__sync_val_compare_and_swap(pw, tmp, tmp + 1) == tmp) {
-			return tmp + 1;
-		}
+	_ALWAYS_INLINE_ void clear() {
+		flag.store(false, std::memory_order_release);
 	}
-}
 
-template <class T>
-static _ALWAYS_INLINE_ T atomic_decrement(volatile T *pw) {
-	return __sync_sub_and_fetch(pw, 1);
-}
+	_ALWAYS_INLINE_ void set_to(bool p_value) {
+		flag.store(p_value, std::memory_order_release);
+	}
+
+	_ALWAYS_INLINE_ explicit SafeFlag(bool p_value = false) {
+		set_to(p_value);
+	}
+};
+
+class SafeRefCount {
+	SafeNumeric<uint32_t> count;
+
+public:
+	_ALWAYS_INLINE_ bool ref() { // true on success
+		return count.conditional_increment() != 0;
+	}
+
+	_ALWAYS_INLINE_ uint32_t refval() { // none-zero on success
+		return count.conditional_increment();
+	}
+
+	_ALWAYS_INLINE_ bool unref() { // true if must be disposed of
+		return count.decrement() == 0;
+	}
+
+	_ALWAYS_INLINE_ uint32_t unrefval() { // 0 if must be disposed of
+		return count.decrement();
+	}
+
+	_ALWAYS_INLINE_ uint32_t get() const {
+		return count.get();
+	}
+
+	_ALWAYS_INLINE_ void init(uint32_t p_value = 1) {
+		count.set(p_value);
+	}
+};
+
+#else
 
 template <class T>
-static _ALWAYS_INLINE_ T atomic_increment(volatile T *pw) {
-	return __sync_add_and_fetch(pw, 1);
-}
-
-template <class T, class V>
-static _ALWAYS_INLINE_ T atomic_sub(volatile T *pw, volatile V val) {
-	return __sync_sub_and_fetch(pw, val);
-}
-
-template <class T, class V>
-static _ALWAYS_INLINE_ T atomic_add(volatile T *pw, volatile V val) {
-	return __sync_add_and_fetch(pw, val);
-}
-
-template <class T, class V>
-static _ALWAYS_INLINE_ T atomic_exchange_if_greater(volatile T *pw, volatile V val) {
-	while (true) {
-		T tmp = static_cast<T const volatile &>(*pw);
-		if (tmp >= val) {
-			return tmp; // already greater, or equal
+class SafeNumeric {
+protected:
+	T value;
+
+public:
+	_ALWAYS_INLINE_ void set(T p_value) {
+		value = p_value;
+	}
+
+	_ALWAYS_INLINE_ T get() const {
+		return value;
+	}
+
+	_ALWAYS_INLINE_ T increment() {
+		return ++value;
+	}
+
+	_ALWAYS_INLINE_ T postincrement() {
+		return value++;
+	}
+
+	_ALWAYS_INLINE_ T decrement() {
+		return --value;
+	}
+
+	_ALWAYS_INLINE_ T postdecrement() {
+		return value--;
+	}
+
+	_ALWAYS_INLINE_ T add(T p_value) {
+		return value += p_value;
+	}
+
+	_ALWAYS_INLINE_ T postadd(T p_value) {
+		T old = value;
+		value += p_value;
+		return old;
+	}
+
+	_ALWAYS_INLINE_ T sub(T p_value) {
+		return value -= p_value;
+	}
+
+	_ALWAYS_INLINE_ T postsub(T p_value) {
+		T old = value;
+		value -= p_value;
+		return old;
+	}
+
+	_ALWAYS_INLINE_ T exchange_if_greater(T p_value) {
+		if (value < p_value) {
+			value = p_value;
 		}
-		if (__sync_val_compare_and_swap(pw, tmp, val) == tmp) {
-			return val;
+		return value;
+	}
+
+	_ALWAYS_INLINE_ T conditional_increment() {
+		if (value != 0) {
+			return 0;
+		} else {
+			return ++value;
 		}
 	}
-}
-
-#elif defined(_MSC_VER)
-// For MSVC use a separate compilation unit to prevent windows.h from polluting
-// the global namespace.
-uint32_t atomic_conditional_increment(volatile uint32_t *pw);
-uint32_t atomic_decrement(volatile uint32_t *pw);
-uint32_t atomic_increment(volatile uint32_t *pw);
-uint32_t atomic_sub(volatile uint32_t *pw, volatile uint32_t val);
-uint32_t atomic_add(volatile uint32_t *pw, volatile uint32_t val);
-uint32_t atomic_exchange_if_greater(volatile uint32_t *pw, volatile uint32_t val);
-
-uint64_t atomic_conditional_increment(volatile uint64_t *pw);
-uint64_t atomic_decrement(volatile uint64_t *pw);
-uint64_t atomic_increment(volatile uint64_t *pw);
-uint64_t atomic_sub(volatile uint64_t *pw, volatile uint64_t val);
-uint64_t atomic_add(volatile uint64_t *pw, volatile uint64_t val);
-uint64_t atomic_exchange_if_greater(volatile uint64_t *pw, volatile uint64_t val);
 
-#else
-//no threads supported?
-#error Must provide atomic functions for this platform or compiler!
-#endif
+	_ALWAYS_INLINE_ explicit SafeNumeric<T>(T p_value = static_cast<T>(0)) :
+			value(p_value) {
+	}
+};
 
-struct SafeRefCount {
-	uint32_t count = 0;
+class SafeFlag {
+protected:
+	bool flag;
 
 public:
-	// destroy() is called when weak_count_ drops to zero.
-
-	_ALWAYS_INLINE_ bool ref() { // true on success
+	_ALWAYS_INLINE_ bool is_set() const {
+		return flag;
+	}
 
-		return atomic_conditional_increment(&count) != 0;
+	_ALWAYS_INLINE_ void set() {
+		flag = true;
 	}
 
-	_ALWAYS_INLINE_ uint32_t refval() { // none-zero on success
+	_ALWAYS_INLINE_ void clear() {
+		flag = false;
+	}
 
-		return atomic_conditional_increment(&count);
+	_ALWAYS_INLINE_ void set_to(bool p_value) {
+		flag = p_value;
 	}
 
-	_ALWAYS_INLINE_ bool unref() { // true if must be disposed of
+	_ALWAYS_INLINE_ explicit SafeFlag(bool p_value = false) :
+			flag(p_value) {}
+};
+
+class SafeRefCount {
+	uint32_t count = 0;
 
-		return atomic_decrement(&count) == 0;
+public:
+	_ALWAYS_INLINE_ bool ref() { // true on success
+		if (count != 0) {
+			++count;
+			return true;
+		} else {
+			return false;
+		}
 	}
 
-	_ALWAYS_INLINE_ uint32_t unrefval() { // 0 if must be disposed of
+	_ALWAYS_INLINE_ uint32_t refval() { // none-zero on success
+		if (count != 0) {
+			return ++count;
+		} else {
+			return 0;
+		}
+	}
 
-		return atomic_decrement(&count);
+	_ALWAYS_INLINE_ bool unref() { // true if must be disposed of
+		return --count == 0;
 	}
 
-	_ALWAYS_INLINE_ uint32_t get() const { // nothrow
+	_ALWAYS_INLINE_ uint32_t unrefval() { // 0 if must be disposed of
+		return --count;
+	}
 
+	_ALWAYS_INLINE_ uint32_t get() const {
 		return count;
 	}
 
@@ -200,4 +314,6 @@ public:
 	}
 };
 
+#endif
+
 #endif // SAFE_REFCOUNT_H