diff --git a/Makefile.common b/Makefile.common
index 5cbed8d5f9..04f80c36b0 100644
--- a/Makefile.common
+++ b/Makefile.common
@@ -1636,7 +1636,6 @@ DOLPHIN_SOURCES_CXX += \
 	$(CORE_DIR)/Source/Core/Core/HW/GBACore.cpp \
 	$(CORE_DIR)/Source/Core/Core/HW/SI/SI_DeviceGBAEmu.cpp \
 	$(CORE_DIR)/Source/Core/UICommon/GameFile.cpp \
-	$(CORE_DIR)/Source/Core/UICommon/GameFileCache.cpp \
 	$(CORE_DIR)/Source/Core/UICommon/ResourcePack/Manager.cpp \
 	$(CORE_DIR)/Source/Core/UICommon/ResourcePack/Manifest.cpp \
 	$(CORE_DIR)/Source/Core/UICommon/ResourcePack/ResourcePack.cpp \
diff --git a/Source/Core/Core/HW/MMIO.h b/Source/Core/Core/HW/MMIO.h
index cee439ab43..1f0258cf3f 100644
--- a/Source/Core/Core/HW/MMIO.h
+++ b/Source/Core/Core/HW/MMIO.h
@@ -4,12 +4,13 @@
 #pragma once
 
 #include <array>
-#include <atomic>
 #include <bit>
 #include <string>
 #include <tuple>
 #include <type_traits>
 
+#include <retro_atomic.h>
+
 #include "Common/Assert.h"
 #include "Common/CommonTypes.h"
 #include "Core/HW/GPFifo.h"
@@ -85,19 +86,20 @@ inline u16* LowPart(u32* ptr)
 {
   return (u16*)ptr;
 }
-inline u16* LowPart(std::atomic<u32>* ptr)
+/* A register shared between threads: the halves are its storage, as the CPU sees it. */
+inline u16* LowPart(retro_atomic_int_t* ptr)
 {
-  static_assert(std::atomic<u32>::is_always_lock_free && sizeof(std::atomic<u32>) == sizeof(u32));
-  return LowPart(std::bit_cast<u32*>(ptr));
+  static_assert(sizeof(retro_atomic_int_t) == sizeof(u32));
+  return LowPart(reinterpret_cast<u32*>(ptr));
 }
 inline u16* HighPart(u32* ptr)
 {
   return LowPart(ptr) + 1;
 }
-inline u16* HighPart(std::atomic<u32>* ptr)
+inline u16* HighPart(retro_atomic_int_t* ptr)
 {
-  static_assert(std::atomic<u32>::is_always_lock_free && sizeof(std::atomic<u32>) == sizeof(u32));
-  return HighPart(std::bit_cast<u32*>(ptr));
+  static_assert(sizeof(retro_atomic_int_t) == sizeof(u32));
+  return HighPart(reinterpret_cast<u32*>(ptr));
 }
 }  // namespace Utils
 
diff --git a/Source/Core/DiscIO/Blob.h b/Source/Core/DiscIO/Blob.h
index 8119694961..e38b45b5ee 100644
--- a/Source/Core/DiscIO/Blob.h
+++ b/Source/Core/DiscIO/Blob.h
@@ -27,7 +27,6 @@ namespace DiscIO
 {
 enum class WIARVZCompressionType : u32;
 
-// Increment CACHE_REVISION (GameFileCache.cpp) if the enum below is modified
 enum class BlobType
 {
   PLAIN,
diff --git a/Source/Core/DiscIO/CachedBlob.cpp b/Source/Core/DiscIO/CachedBlob.cpp
index 9c0cd23a08..f3b6f7a434 100644
--- a/Source/Core/DiscIO/CachedBlob.cpp
+++ b/Source/Core/DiscIO/CachedBlob.cpp
@@ -3,9 +3,9 @@
 
 #include "DiscIO/CachedBlob.h"
 
-#include <atomic>
 #include <chrono>
 
+#include <retro_atomic.h>
 #include <rthreads/rthreads.h>
 
 #include "Common/CommonTypes.h"
@@ -29,7 +29,7 @@ public:
 
   ~CacheFiller()
   {
-    m_stop_thread.store(true, std::memory_order_relaxed);
+    retro_atomic_store_relaxed_int(&m_stop_thread, 1);
     Common::JoinThread(m_thread);
   }
 
@@ -69,7 +69,7 @@ private:
 
   CacheState GetCacheState(u64 offset, u64 size)
   {
-    const auto cache_pos = m_cache_filled_pos.load(std::memory_order_acquire);
+    const u64 cache_pos = static_cast<u64>(retro_atomic_load_acquire_64(&m_cache_filled_pos));
 
     const auto end_pos = offset + size;
     if (end_pos > cache_pos)
@@ -131,7 +131,7 @@ private:
 
     while (true)
     {
-      if (m_stop_thread.load(std::memory_order_relaxed))
+      if (retro_atomic_load_relaxed_int(&m_stop_thread))
       {
         INFO_LOG_FMT(DISCIO, "CachedBlobReader: Stopped");
         break;
@@ -164,7 +164,7 @@ private:
       }
 
       read_offset += read_size;
-      m_cache_filled_pos.store(read_offset, std::memory_order_release);
+      retro_atomic_store_release_64(&m_cache_filled_pos, static_cast<s64>(read_offset));
 
       if (const auto now = Clock::now(); now >= next_log_time)
       {
@@ -176,14 +176,14 @@ private:
   }
 
   // The thread has read non-scrubbed bytes into memory up to this point.
-  std::atomic<u64> m_cache_filled_pos{};
+  retro_atomic_64_t m_cache_filled_pos{};
 
   Common::LazyMemoryRegion m_memory_region;
   u8* m_memory_region_data{};
 
   DiscScrubber m_scrubber;
 
-  std::atomic_bool m_stop_thread{};
+  retro_atomic_int_t m_stop_thread{};
   std::unique_ptr<BlobReader> m_thread_reader;  // handed to the thread
   bool m_attempt_to_scrub;
   sthread_t* m_thread = nullptr;
diff --git a/Source/Core/DiscIO/Enums.h b/Source/Core/DiscIO/Enums.h
index 62ad9e846a..6eafca957b 100644
--- a/Source/Core/DiscIO/Enums.h
+++ b/Source/Core/DiscIO/Enums.h
@@ -10,7 +10,6 @@
 
 namespace DiscIO
 {
-// Increment CACHE_REVISION (GameFileCache.cpp) if these enums are modified
 
 enum class Platform
 {
diff --git a/Source/Core/InputCommon/ControllerEmu/Setting/NumericSetting.h b/Source/Core/InputCommon/ControllerEmu/Setting/NumericSetting.h
index 839b3f7949..7c1ab8ec45 100644
--- a/Source/Core/InputCommon/ControllerEmu/Setting/NumericSetting.h
+++ b/Source/Core/InputCommon/ControllerEmu/Setting/NumericSetting.h
@@ -4,7 +4,7 @@
 #pragma once
 
 #include <algorithm>
-#include <atomic>
+#include <cstring>
 #include <functional>
 #include <string>
 #include <utility>
@@ -15,8 +15,38 @@
 #include "InputCommon/ControlReference/ControlReference.h"
 #include "InputCommon/ControllerInterface/CoreDevice.h"
 
+#include <retro_atomic.h>
+
 namespace ControllerEmu
 {
+/* A setting's value lives in one atomic 64-bit word, as its bits. */
+inline s64 SettingToBits(double value)
+{
+  s64 bits;
+  std::memcpy(&bits, &value, sizeof(bits));
+  return bits;
+}
+inline s64 SettingToBits(int value)
+{
+  return value;
+}
+inline s64 SettingToBits(bool value)
+{
+  return value ? 1 : 0;
+}
+inline void SettingFromBits(s64 bits, double* value)
+{
+  std::memcpy(value, &bits, sizeof(*value));
+}
+inline void SettingFromBits(s64 bits, int* value)
+{
+  *value = static_cast<int>(bits);
+}
+inline void SettingFromBits(s64 bits, bool* value)
+{
+  *value = bits != 0;
+}
+
 enum class SettingType
 {
   Int,
@@ -185,26 +215,36 @@ public:
     // Otherwise settings will all change to 0 when window focus is lost.
     // This is very undesirable for things like battery level or attached extension.
     if (!IsSimpleValue() && ControlReference::GetInputGate())
-      m_value = m_input.GetState<ValueType>();
+      StoreValue(m_input.GetState<ValueType>());
 
-    return m_value;
+    return GetCachedValue();
   }
 
-  ValueType GetCachedValue() const { return m_value; }
+  ValueType GetCachedValue() const
+  {
+    ValueType value;
+    SettingFromBits(retro_atomic_load_relaxed_64(&m_value), &value);
+    return value;
+  }
 
   bool IsSimpleValue() const { return !m_input.HasExpression(); }
 
   virtual void SetValue(const ValueType value)
   {
-    m_value = value;
+    StoreValue(value);
 
     // Clear the expression to use our new "simple" value.
     m_input.SetExpression("");
   }
 
 private:
+  void StoreValue(ValueType value) const
+  {
+    retro_atomic_store_relaxed_64(&m_value, SettingToBits(value));
+  }
+
   // Values are R/W by both UI and CPU threads.
-  mutable std::atomic<ValueType> m_value = {};
+  mutable retro_atomic_64_t m_value{};
 
   // Unfortunately InputReference's state grabbing is non-const requiring mutable here.
   mutable InputReference m_input;
diff --git a/Source/Core/InputCommon/ControllerInterface/Android/Android.cpp b/Source/Core/InputCommon/ControllerInterface/Android/Android.cpp
index 467cb11000..9e0dd0f4fa 100644
--- a/Source/Core/InputCommon/ControllerInterface/Android/Android.cpp
+++ b/Source/Core/InputCommon/ControllerInterface/Android/Android.cpp
@@ -4,8 +4,8 @@
 #include "InputCommon/ControllerInterface/Android/Android.h"
 
 #include <algorithm>
-#include <atomic>
 #include <chrono>
+#include <cstring>
 #include <memory>
 #include <string>
 #include <unordered_map>
@@ -17,6 +17,8 @@
 #include <android/keycodes.h>
 #include <jni.h>
 
+#include <retro_atomic.h>
+
 #include "Common/Assert.h"
 #include "Common/Contains.h"
 #include "Common/Logging/Log.h"
@@ -31,6 +33,21 @@
 
 namespace
 {
+/* A ControlState shared between threads lives in an atomic 64-bit word, as its bits. */
+s64 StateToBits(ControlState state)
+{
+  s64 bits;
+  std::memcpy(&bits, &state, sizeof(bits));
+  return bits;
+}
+
+ControlState StateFromBits(s64 bits)
+{
+  ControlState state;
+  std::memcpy(&state, &bits, sizeof(state));
+  return state;
+}
+
 std::string SOURCE = "Android";
 
 jclass s_list_class;
@@ -496,18 +513,21 @@ public:
 
   ControlState GetState() const override
   {
-    m_last_polled.store(Clock::now(), std::memory_order_relaxed);
-    return m_state.load(std::memory_order_relaxed);
+    retro_atomic_store_relaxed_64(&m_last_polled, Clock::now().time_since_epoch().count());
+    return StateFromBits(retro_atomic_load_relaxed_64(&m_state));
   }
 
-  void SetState(ControlState state) { m_state.store(state, std::memory_order_relaxed); }
+  void SetState(ControlState state) { retro_atomic_store_relaxed_64(&m_state, StateToBits(state)); }
 
-  Clock::time_point GetLastPolled() const { return m_last_polled.load(std::memory_order_relaxed); }
+  Clock::time_point GetLastPolled() const
+  {
+    return Clock::time_point(Clock::duration(retro_atomic_load_relaxed_64(&m_last_polled)));
+  }
 
 private:
   std::string m_name;
-  std::atomic<ControlState> m_state = 0;
-  mutable std::atomic<Clock::time_point> m_last_polled{};
+  retro_atomic_64_t m_state{};
+  mutable retro_atomic_64_t m_last_polled{};
 };
 
 class AndroidKey final : public AndroidInput
@@ -556,7 +576,7 @@ public:
 
   ControlState GetState() const override
   {
-    if (m_is_suspended.load(std::memory_order_relaxed))
+    if (retro_atomic_load_relaxed_int(&m_is_suspended))
     {
       IDCache::GetEnvForThread()->CallVoidMethod(
           m_sensor_event_listener, s_sensor_event_listener_request_unsuspend_sensor, m_j_name);
@@ -572,13 +592,13 @@ public:
 
   void NotifyIsSuspended(bool is_suspended)
   {
-    m_is_suspended.store(is_suspended, std::memory_order_relaxed);
+    retro_atomic_store_relaxed_int(&m_is_suspended, is_suspended ? 1 : 0);
   }
 
 private:
   const jobject m_sensor_event_listener;
   const jstring m_j_name;
-  std::atomic<bool> m_is_suspended = true;
+  retro_atomic_int_t m_is_suspended{1};
 };
 
 class AndroidMotor : public Core::Device::Output
@@ -595,7 +615,7 @@ public:
 
   void SetState(ControlState state) override
   {
-    ControlState old_state = m_state.exchange(state, std::memory_order_relaxed);
+    ControlState old_state = StateFromBits(retro_atomic_exchange_64(&m_state, StateToBits(state)));
 
     if (old_state < 0.5 && state >= 0.5)
     {
@@ -612,7 +632,7 @@ public:
 private:
   const jobject m_vibrator;
   const jint m_index;
-  std::atomic<ControlState> m_state = 0;
+  retro_atomic_64_t m_state{};
 };
 
 class AndroidDevice final : public Core::Device
diff --git a/Source/Core/InputCommon/ControllerInterface/ControllerInterface.cpp b/Source/Core/InputCommon/ControllerInterface/ControllerInterface.cpp
index d5dc313f52..c584d4b4c2 100644
--- a/Source/Core/InputCommon/ControllerInterface/ControllerInterface.cpp
+++ b/Source/Core/InputCommon/ControllerInterface/ControllerInterface.cpp
@@ -4,6 +4,7 @@
 #include "InputCommon/ControllerInterface/ControllerInterface.h"
 
 #include <algorithm>
+#include <cstring>
 
 #include "Common/Assert.h"
 #include "Common/Logging/Log.h"
@@ -31,12 +32,12 @@ static thread_local bool tls_is_updating_devices = false;
 
 void ControllerInterface::Initialize(const WindowSystemInfo& wsi)
 {
-  if (m_is_init)
+  if (IsInit())
     return;
 
   m_wsi = wsi;
 
-  m_populating_devices_counter = 1;
+  retro_atomic_store_release_int(&m_populating_devices_counter, 1);
 
 #if defined(CIFACE_USE_ANDROID) && !defined(__LIBRETRO__)
   m_input_backends.emplace_back(ciface::Android::CreateInputBackend(this));
@@ -50,7 +51,7 @@ void ControllerInterface::Initialize(const WindowSystemInfo& wsi)
 
   // Don't allow backends to add devices before the first RefreshDevices() as they will be cleaned
   // there. Or they'd end up waiting on the devices mutex if populated from another thread.
-  m_is_init = true;
+  retro_atomic_store_release_int(&m_is_init, 1);
 
   RefreshDevices();
 
@@ -61,13 +62,13 @@ void ControllerInterface::Initialize(const WindowSystemInfo& wsi)
     devices_empty = !list || list->devices.empty();
   }
 
-  if (m_populating_devices_counter.fetch_sub(1) == 1 && !devices_empty)
+  if (retro_atomic_fetch_sub_int(&m_populating_devices_counter, 1) == 1 && !devices_empty)
     InvokeDevicesChangedCallbacks();
 }
 
 void ControllerInterface::ChangeWindow(void* hwnd, WindowChangeReason reason)
 {
-  if (!m_is_init)
+  if (!IsInit())
     return;
 
   if (m_wsi.render_window == hwnd)
@@ -85,7 +86,7 @@ void ControllerInterface::ChangeWindow(void* hwnd, WindowChangeReason reason)
 
 void ControllerInterface::RefreshDevices(RefreshReason reason)
 {
-  if (!m_is_init)
+  if (!IsInit())
     return;
 
   // Backends with a hotplug thread may add or remove devices meanwhile; every change is
@@ -94,18 +95,18 @@ void ControllerInterface::RefreshDevices(RefreshReason reason)
   // Instead only refresh devices that require the window handle.
   if (reason == RefreshReason::WindowChangeOnly)
   {
-    m_populating_devices_counter.fetch_add(1);
+    retro_atomic_fetch_add_int(&m_populating_devices_counter, 1);
 
     for (auto& backend : m_input_backends)
       backend->HandleWindowChange();
 
-    if (m_populating_devices_counter.fetch_sub(1) == 1)
+    if (retro_atomic_fetch_sub_int(&m_populating_devices_counter, 1) == 1)
       InvokeDevicesChangedCallbacks();
 
     return;
   }
 
-  m_populating_devices_counter.fetch_add(1);
+  retro_atomic_fetch_add_int(&m_populating_devices_counter, 1);
 
   // Make sure shared_ptr<Device> objects are released before repopulating.
   ClearDevices();
@@ -123,33 +124,33 @@ void ControllerInterface::RefreshDevices(RefreshReason reason)
 
   WiimoteReal::PopulateDevices();
 
-  if (m_populating_devices_counter.fetch_sub(1) == 1)
+  if (retro_atomic_fetch_sub_int(&m_populating_devices_counter, 1) == 1)
     InvokeDevicesChangedCallbacks();
 }
 
 void ControllerInterface::PlatformPopulateDevices(const std::function<void()>& callback)
 {
-  if (!m_is_init)
+  if (!IsInit())
     return;
 
-  m_populating_devices_counter.fetch_add(1);
+  retro_atomic_fetch_add_int(&m_populating_devices_counter, 1);
 
   callback();
 
-  if (m_populating_devices_counter.fetch_sub(1) == 1)
+  if (retro_atomic_fetch_sub_int(&m_populating_devices_counter, 1) == 1)
     InvokeDevicesChangedCallbacks();
 }
 
 // Remove all devices and call library cleanup functions
 void ControllerInterface::Shutdown()
 {
-  if (!m_is_init)
+  if (!IsInit())
     return;
 
   // Prevent additional devices from being added during shutdown.
-  m_is_init = false;
+  retro_atomic_store_release_int(&m_is_init, 0);
   // Additional safety measure to avoid InvokeDevicesChangedCallbacks()
-  m_populating_devices_counter = 1;
+  retro_atomic_store_release_int(&m_populating_devices_counter, 1);
 
   // Update control references so shared_ptr<Device>s are freed up BEFORE we shutdown the backends.
   ClearDevices();
@@ -191,7 +192,7 @@ void ControllerInterface::ClearDevices()
 bool ControllerInterface::AddDevice(std::shared_ptr<ciface::Core::Device> device)
 {
   // If we are shutdown (or in process of shutting down) ignore this request:
-  if (!m_is_init)
+  if (!IsInit())
     return false;
 
   ASSERT_MSG(CONTROLLERINTERFACE, !tls_is_updating_devices,
@@ -236,7 +237,7 @@ bool ControllerInterface::AddDevice(std::shared_ptr<ciface::Core::Device> device
   });
   NOTICE_LOG_FMT(CONTROLLERINTERFACE, "Added device: {}", device->GetQualifiedName());
 
-  if (!m_populating_devices_counter)
+  if (!retro_atomic_load_acquire_int(&m_populating_devices_counter))
     InvokeDevicesChangedCallbacks();
   return true;
 }
@@ -245,7 +246,7 @@ void ControllerInterface::RemoveDevice(std::function<bool(const ciface::Core::De
                                        bool force_devices_release)
 {
   // If we are shutdown (or in process of shutting down) ignore this request:
-  if (!m_is_init)
+  if (!IsInit())
     return;
 
   ASSERT_MSG(CONTROLLERINTERFACE, !tls_is_updating_devices,
@@ -265,7 +266,8 @@ void ControllerInterface::RemoveDevice(std::function<bool(const ciface::Core::De
                }) != 0;
       });
 
-  if (any_removed && (!m_populating_devices_counter || force_devices_release))
+  if (any_removed &&
+      (!retro_atomic_load_acquire_int(&m_populating_devices_counter) || force_devices_release))
     InvokeDevicesChangedCallbacks();
 }
 
@@ -273,8 +275,8 @@ void ControllerInterface::RemoveDevice(std::function<bool(const ciface::Core::De
 void ControllerInterface::UpdateInput()
 {
   // This should never happen
-  ASSERT(m_is_init);
-  if (!m_is_init)
+  ASSERT(IsInit());
+  if (!IsInit())
     return;
 
   // The devices to remove are collected as weak pointers: they are not kept alive by this, and
@@ -328,12 +330,16 @@ WindowSystemInfo ControllerInterface::GetWindowSystemInfo() const
 
 void ControllerInterface::SetAspectRatioAdjustment(float value)
 {
-  m_aspect_ratio_adjustment = value;
+  int bits;
+  std::memcpy(&bits, &value, sizeof(bits));
+  retro_atomic_store_release_int(&m_aspect_ratio_adjustment, bits);
 }
 
 Common::Vec2 ControllerInterface::GetWindowInputScale() const
 {
-  const auto ar = m_aspect_ratio_adjustment.load();
+  const int bits = retro_atomic_load_acquire_int(&m_aspect_ratio_adjustment);
+  float ar;
+  std::memcpy(&ar, &bits, sizeof(ar));
 
   if (ar > 1)
     return {1.f, ar};
@@ -343,12 +349,12 @@ Common::Vec2 ControllerInterface::GetWindowInputScale() const
 
 void ControllerInterface::SetMouseCenteringRequested(bool center)
 {
-  m_requested_mouse_centering = center;
+  retro_atomic_store_release_int(&m_requested_mouse_centering, center ? 1 : 0);
 }
 
 bool ControllerInterface::IsMouseCenteringRequested() const
 {
-  return m_requested_mouse_centering.load();
+  return retro_atomic_load_acquire_int(&m_requested_mouse_centering) != 0;
 }
 
 // Register a callback to be called when a device is added or removed (as from the input backends'
diff --git a/Source/Core/InputCommon/ControllerInterface/ControllerInterface.h b/Source/Core/InputCommon/ControllerInterface/ControllerInterface.h
index 5fa6440c3b..d13bcb73b6 100644
--- a/Source/Core/InputCommon/ControllerInterface/ControllerInterface.h
+++ b/Source/Core/InputCommon/ControllerInterface/ControllerInterface.h
@@ -3,7 +3,6 @@
 
 #pragma once
 
-#include <atomic>
 #include <functional>
 #include <memory>
 
@@ -14,6 +13,8 @@
 #include "InputCommon/ControllerInterface/CoreDevice.h"
 #include "InputCommon/ControllerInterface/InputBackend.h"
 
+#include <retro_atomic.h>
+
 // enable disable sources
 #define CIFACE_USE_DUALSHOCKUDPCLIENT
 #if defined(HAVE_HIDAPI)
@@ -62,7 +63,7 @@ public:
     Other
   };
 
-  ControllerInterface() : m_is_init(false) {}
+  ControllerInterface() {}
   void Initialize(const WindowSystemInfo& wsi);
   // Only call from one thread at a time.
   void ChangeWindow(void* hwnd, WindowChangeReason reason = WindowChangeReason::Other);
@@ -78,7 +79,7 @@ public:
   // Runs a backend's population, holding back the devices-changed callbacks until it is
   // done so they run once for the batch.
   void PlatformPopulateDevices(const std::function<void()>& callback);
-  bool IsInit() const { return m_is_init; }
+  bool IsInit() const { return retro_atomic_load_acquire_int(&m_is_init) != 0; }
   void UpdateInput();
 
   // Set adjustment from the full render window aspect-ratio to the drawn aspect-ratio.
@@ -109,12 +110,12 @@ private:
 
   Common::HookableEvent<> m_devices_changed_event;
 
-  std::atomic<bool> m_is_init;
+  retro_atomic_int_t m_is_init{};
   // Populations in progress on any thread; the last to finish runs the callbacks.
-  std::atomic<int> m_populating_devices_counter;
+  retro_atomic_int_t m_populating_devices_counter{};
   WindowSystemInfo m_wsi;
-  std::atomic<float> m_aspect_ratio_adjustment = 1;
-  std::atomic<bool> m_requested_mouse_centering = false;
+  retro_atomic_int_t m_aspect_ratio_adjustment{0x3f800000}; /* the bits of a float, 1.0f */
+  retro_atomic_int_t m_requested_mouse_centering{};
 
   std::vector<std::unique_ptr<ciface::InputBackend>> m_input_backends;
 };
diff --git a/Source/Core/InputCommon/ControllerInterface/Wiimote/WiimoteController.cpp b/Source/Core/InputCommon/ControllerInterface/Wiimote/WiimoteController.cpp
index e7724c1840..7af96bdbef 100644
--- a/Source/Core/InputCommon/ControllerInterface/Wiimote/WiimoteController.cpp
+++ b/Source/Core/InputCommon/ControllerInterface/Wiimote/WiimoteController.cpp
@@ -3,6 +3,8 @@
 
 #include "InputCommon/ControllerInterface/Wiimote/WiimoteController.h"
 
+#include <cstring>
+
 #include "Common/BitUtils.h"
 #include "Common/Logging/Log.h"
 #include "Common/MathUtil.h"
@@ -81,14 +83,19 @@ using UndetectableSignedAnalogInput = SignedInput<false>;
 class Motor final : public Core::Device::Output
 {
 public:
-  Motor(std::atomic<ControlState>* value) : m_value(*value) {}
+  Motor(retro_atomic_64_t* value) : m_value(value) {}
 
   std::string GetName() const override { return "Motor"; }
 
-  void SetState(ControlState state) override { m_value = state; }
+  void SetState(ControlState state) override
+  {
+    s64 bits;
+    std::memcpy(&bits, &state, sizeof(bits));
+    retro_atomic_store_release_64(m_value, bits);
+  }
 
 private:
-  std::atomic<ControlState>& m_value;
+  retro_atomic_64_t* m_value;
 };
 
 template <typename T>
@@ -1422,8 +1429,11 @@ void Device::UpdateRumble()
 {
   static constexpr auto rumble_period = std::chrono::milliseconds(100);
 
-  const auto on_time =
-      std::chrono::duration_cast<Clock::duration>(rumble_period * m_rumble_level.load());
+  const s64 bits = retro_atomic_load_acquire_64(&m_rumble_level);
+  ControlState rumble_level;
+  std::memcpy(&rumble_level, &bits, sizeof(rumble_level));
+
+  const auto on_time = std::chrono::duration_cast<Clock::duration>(rumble_period * rumble_level);
   const auto off_time = rumble_period - on_time;
 
   const auto now = Clock::now();
diff --git a/Source/Core/InputCommon/ControllerInterface/Wiimote/WiimoteController.h b/Source/Core/InputCommon/ControllerInterface/Wiimote/WiimoteController.h
index db0dcb138b..e992073cf8 100644
--- a/Source/Core/InputCommon/ControllerInterface/Wiimote/WiimoteController.h
+++ b/Source/Core/InputCommon/ControllerInterface/Wiimote/WiimoteController.h
@@ -4,7 +4,6 @@
 #pragma once
 
 #include <array>
-#include <atomic>
 #include <chrono>
 #include <memory>
 #include <vector>
@@ -18,6 +17,8 @@
 #include "Core/HW/WiimoteReal/WiimoteReal.h"
 #include "InputCommon/ControllerInterface/CoreDevice.h"
 
+#include <retro_atomic.h>
+
 namespace ciface::WiimoteController
 {
 using namespace WiimoteCommon;
@@ -264,7 +265,7 @@ private:
   bool m_rumble = false;
 
   // For pulse of rumble motor to simulate multiple levels.
-  std::atomic<ControlState> m_rumble_level;
+  retro_atomic_64_t m_rumble_level{}; /* the bits of a ControlState */
   Clock::time_point m_last_rumble_change = Clock::now();
 
   // Assume mode is disabled so one gets set.
diff --git a/Source/Core/UICommon/GameFileCache.cpp b/Source/Core/UICommon/GameFileCache.cpp
deleted file mode 100644
index 898e818852..0000000000
--- a/Source/Core/UICommon/GameFileCache.cpp
+++ /dev/null
@@ -1,281 +0,0 @@
-// Copyright 2017 Dolphin Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#include "UICommon/GameFileCache.h"
-
-#include <algorithm>
-#include <cstddef>
-#include <memory>
-#include <string>
-#include <unordered_set>
-#include <utility>
-#include <vector>
-
-#include "Common/ChunkFile.h"
-#include "Common/CommonTypes.h"
-#include "Common/FileSearch.h"
-#include "Common/FileUtil.h"
-#include "Common/IOFile.h"
-
-#include "DiscIO/DirectoryBlob.h"
-
-#include "UICommon/GameFile.h"
-
-namespace UICommon
-{
-static constexpr u32 CACHE_REVISION = 27;  // Last changed in PR 13844
-
-std::vector<std::string> FindAllGamePaths(std::span<const std::string_view> directories_to_scan,
-                                          bool recursive_scan)
-{
-  constexpr auto search_extensions =
-      std::to_array<std::string_view>({".gcm", ".tgc", ".bin", ".iso", ".ciso", ".gcz", ".wbfs",
-                                       ".wia", ".rvz", ".nfs", ".wad", ".dol", ".elf", ".json"});
-
-  // TODO: We could process paths iteratively as they are found
-  return Common::DoFileSearch(directories_to_scan, search_extensions, recursive_scan);
-}
-
-GameFileCache::GameFileCache() : m_path(File::GetUserPath(D_CACHE_IDX) + "gamelist.cache")
-{
-}
-
-void GameFileCache::ForEach(const ForEachFn& f) const
-{
-  for (const std::shared_ptr<GameFile>& item : m_cached_files)
-    f(item);
-}
-
-size_t GameFileCache::GetSize() const
-{
-  return m_cached_files.size();
-}
-
-void GameFileCache::Clear(DeleteOnDisk delete_on_disk)
-{
-  if (delete_on_disk != DeleteOnDisk::No)
-    File::Delete(m_path);
-
-  m_cached_files.clear();
-}
-
-std::shared_ptr<const GameFile> GameFileCache::AddOrGet(const std::string& path,
-                                                        bool* cache_changed)
-{
-  auto it = std::ranges::find(m_cached_files, path, &GameFile::GetFilePath);
-  const bool found = it != m_cached_files.cend();
-  if (!found)
-  {
-    std::shared_ptr<UICommon::GameFile> game = std::make_shared<GameFile>(path);
-    if (!game->IsValid())
-      return nullptr;
-    m_cached_files.emplace_back(std::move(game));
-  }
-  std::shared_ptr<GameFile>& result = found ? *it : m_cached_files.back();
-  if (UpdateAdditionalMetadata(&result) || !found)
-    *cache_changed = true;
-
-  return result;
-}
-
-bool GameFileCache::Update(std::span<const std::string> all_game_paths,
-                           const GameAddedToCacheFn& game_added_to_cache,
-                           const GameRemovedFromCacheFn& game_removed_from_cache,
-                           const std::atomic_bool& processing_halted)
-{
-  // Copy game paths into a set, except ones that match DiscIO::ShouldHideFromGameList.
-  // TODO: Prevent DoFileSearch from looking inside /files/ directories of DirectoryBlobs at all?
-  // TODO: Make DoFileSearch support filter predicates so we don't have remove things afterwards?
-  std::unordered_set<std::string> game_paths;
-  game_paths.reserve(all_game_paths.size());
-  for (const std::string& path : all_game_paths)
-  {
-    if (!DiscIO::ShouldHideFromGameList(path))
-      game_paths.insert(path);
-  }
-
-  bool cache_changed = false;
-
-  // Delete paths that aren't in game_paths from m_cached_files,
-  // while simultaneously deleting paths that are in m_cached_files from game_paths.
-  // For the sake of speed, we don't care about maintaining the order of m_cached_files.
-  {
-    auto it = m_cached_files.begin();
-    auto end = m_cached_files.end();
-    while (it != end)
-    {
-      if (processing_halted)
-        break;
-
-      if (game_paths.erase((*it)->GetFilePath()))
-      {
-        ++it;
-      }
-      else
-      {
-        if (game_removed_from_cache)
-          game_removed_from_cache((*it)->GetFilePath());
-
-        cache_changed = true;
-        --end;
-        *it = std::move(*end);
-      }
-    }
-    m_cached_files.erase(it, m_cached_files.end());
-  }
-
-  // Now that the previous loop has run, game_paths only contains paths that
-  // aren't in m_cached_files, so we simply add all of them to m_cached_files.
-  for (const std::string& path : game_paths)
-  {
-    if (processing_halted)
-      break;
-
-    auto file = std::make_shared<GameFile>(path);
-    if (file->IsValid())
-    {
-      if (game_added_to_cache)
-        game_added_to_cache(file);
-
-      cache_changed = true;
-      m_cached_files.push_back(std::move(file));
-    }
-  }
-
-  return cache_changed;
-}
-
-bool GameFileCache::UpdateAdditionalMetadata(const GameUpdatedFn& game_updated,
-                                             const std::atomic_bool& processing_halted)
-{
-  bool cache_changed = false;
-
-  for (std::shared_ptr<GameFile>& file : m_cached_files)
-  {
-    if (processing_halted)
-      break;
-
-    const bool updated = UpdateAdditionalMetadata(&file);
-    cache_changed |= updated;
-    if (game_updated && updated)
-      game_updated(file);
-  }
-
-  return cache_changed;
-}
-
-bool GameFileCache::UpdateAdditionalMetadata(std::shared_ptr<GameFile>* game_file)
-{
-  const bool xml_metadata_changed = (*game_file)->XMLMetadataChanged();
-  const bool wii_banner_changed = (*game_file)->WiiBannerChanged();
-  const bool custom_banner_changed = (*game_file)->CustomBannerChanged();
-
-  (*game_file)->DownloadDefaultCover();
-
-  const bool default_cover_changed = (*game_file)->DefaultCoverChanged();
-  const bool custom_cover_changed = (*game_file)->CustomCoverChanged();
-
-  if (!xml_metadata_changed && !wii_banner_changed && !custom_banner_changed &&
-      !default_cover_changed && !custom_cover_changed)
-  {
-    return false;
-  }
-
-  // If a cached file needs an update, apply the updates to a copy and delete the original.
-  // This makes the usage of cached files in other threads safe.
-
-  std::shared_ptr<GameFile> copy = std::make_shared<GameFile>(**game_file);
-  if (xml_metadata_changed)
-    copy->XMLMetadataCommit();
-  if (wii_banner_changed)
-    copy->WiiBannerCommit();
-  if (custom_banner_changed)
-    copy->CustomBannerCommit();
-  if (default_cover_changed)
-    copy->DefaultCoverCommit();
-  if (custom_cover_changed)
-    copy->CustomCoverCommit();
-
-  *game_file = std::move(copy);
-
-  return true;
-}
-
-bool GameFileCache::Load()
-{
-  return SyncCacheFile(false);
-}
-
-bool GameFileCache::Save()
-{
-  return SyncCacheFile(true);
-}
-
-bool GameFileCache::SyncCacheFile(bool save)
-{
-  const char* open_mode = save ? "wb" : "rb";
-  File::IOFile f(m_path, open_mode);
-  if (!f)
-    return false;
-  bool success = false;
-  if (save)
-  {
-    // Measure the size of the buffer.
-    u8* ptr = nullptr;
-    PointerWrap p_measure(&ptr, 0, PointerWrap::Mode::Measure);
-    DoState(&p_measure);
-    const size_t buffer_size = reinterpret_cast<size_t>(ptr);
-
-    // Then actually do the write.
-    std::vector<u8> buffer(buffer_size);
-    ptr = buffer.data();
-    PointerWrap p(&ptr, buffer_size, PointerWrap::Mode::Write);
-    DoState(&p, buffer_size);
-    if (f.WriteBytes(buffer.data(), buffer.size()))
-      success = true;
-  }
-  else
-  {
-    std::vector<u8> buffer(f.GetSize());
-    if (!buffer.empty() && f.ReadBytes(buffer.data(), buffer.size()))
-    {
-      u8* ptr = buffer.data();
-      PointerWrap p(&ptr, buffer.size(), PointerWrap::Mode::Read);
-      DoState(&p, buffer.size());
-      if (p.IsReadMode())
-        success = true;
-    }
-  }
-  if (!success)
-  {
-    // If some file operation failed, try to delete the probably-corrupted cache
-    f.Close();
-    File::Delete(m_path);
-  }
-  return success;
-}
-
-void GameFileCache::DoState(PointerWrap* p, u64 size)
-{
-  struct
-  {
-    u32 revision;
-    u64 expected_size;
-  } header = {CACHE_REVISION, size};
-  p->Do(header);
-  if (p->IsReadMode())
-  {
-    if (header.revision != CACHE_REVISION || header.expected_size != size)
-    {
-      p->SetMeasureMode();
-      return;
-    }
-  }
-  p->DoEachElement(m_cached_files, [](PointerWrap& state, std::shared_ptr<GameFile>& elem) {
-    if (state.IsReadMode())
-      elem = std::make_shared<GameFile>();
-    elem->DoState(state);
-  });
-}
-
-}  // namespace UICommon
diff --git a/Source/Core/UICommon/GameFileCache.h b/Source/Core/UICommon/GameFileCache.h
deleted file mode 100644
index 4c4ab15557..0000000000
--- a/Source/Core/UICommon/GameFileCache.h
+++ /dev/null
@@ -1,70 +0,0 @@
-// Copyright 2017 Dolphin Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#pragma once
-
-#include <atomic>
-#include <cstddef>
-#include <functional>
-#include <memory>
-#include <span>
-#include <string>
-#include <vector>
-
-#include "Common/CommonTypes.h"
-
-class PointerWrap;
-
-namespace UICommon
-{
-class GameFile;
-
-std::vector<std::string> FindAllGamePaths(std::span<const std::string_view> directories_to_scan,
-                                          bool recursive_scan);
-
-class GameFileCache
-{
-public:
-  enum class DeleteOnDisk
-  {
-    No = 0,
-    Yes = 1,
-  };
-
-  using ForEachFn = std::function<void(const std::shared_ptr<const GameFile>&)>;
-  using GameAddedToCacheFn = std::function<void(const std::shared_ptr<const GameFile>&)>;
-  using GameRemovedFromCacheFn = std::function<void(const std::string&)>;
-  using GameUpdatedFn = std::function<void(const std::shared_ptr<const GameFile>&)>;
-
-  GameFileCache();
-
-  void ForEach(const ForEachFn& f) const;
-
-  size_t GetSize() const;
-  void Clear(DeleteOnDisk delete_on_disk);
-
-  // Returns nullptr if the file is invalid.
-  std::shared_ptr<const GameFile> AddOrGet(const std::string& path, bool* cache_changed);
-
-  // These functions return true if the call modified the cache.
-  bool Update(std::span<const std::string> all_game_paths,
-              const GameAddedToCacheFn& game_added_to_cache = {},
-              const GameRemovedFromCacheFn& game_removed_from_cache = {},
-              const std::atomic_bool& processing_halted = false);
-  bool UpdateAdditionalMetadata(const GameUpdatedFn& game_updated = {},
-                                const std::atomic_bool& processing_halted = false);
-
-  bool Load();
-  bool Save();
-
-private:
-  bool UpdateAdditionalMetadata(std::shared_ptr<GameFile>* game_file);
-
-  bool SyncCacheFile(bool save);
-  void DoState(PointerWrap* p, u64 size = 0);
-
-  std::string m_path;
-  std::vector<std::shared_ptr<GameFile>> m_cached_files;
-};
-
-}  // namespace UICommon
diff --git a/Source/Core/VideoBackends/D3D/D3DNativeVertexFormat.cpp b/Source/Core/VideoBackends/D3D/D3DNativeVertexFormat.cpp
index 631d164847..146cff1317 100644
--- a/Source/Core/VideoBackends/D3D/D3DNativeVertexFormat.cpp
+++ b/Source/Core/VideoBackends/D3D/D3DNativeVertexFormat.cpp
@@ -142,7 +142,8 @@ D3DVertexFormat::D3DVertexFormat(const PortableVertexDeclaration& vtx_decl)
 
 D3DVertexFormat::~D3DVertexFormat()
 {
-  ID3D11InputLayout* layout = m_layout.load();
+  ID3D11InputLayout* layout =
+      static_cast<ID3D11InputLayout*>(retro_atomic_load_acquire_ptr(&m_layout));
   if (layout)
     layout->Release();
 }
@@ -151,7 +152,8 @@ ID3D11InputLayout* D3DVertexFormat::GetInputLayout(const void* vs_bytecode, size
 {
   // CreateInputLayout requires a shader input, but it only looks at the signature of the shader,
   // so we don't need to recompute it if the shader changes.
-  ID3D11InputLayout* layout = m_layout.load();
+  ID3D11InputLayout* layout =
+      static_cast<ID3D11InputLayout*>(retro_atomic_load_acquire_ptr(&m_layout));
   if (layout)
     return layout;
 
@@ -161,13 +163,12 @@ ID3D11InputLayout* D3DVertexFormat::GetInputLayout(const void* vs_bytecode, size
 
   // This method can be called from multiple threads, so ensure that only one thread sets the
   // cached input layout pointer. If another thread beats this thread, use the existing layout.
-  ID3D11InputLayout* expected = nullptr;
-  if (!m_layout.compare_exchange_strong(expected, layout))
+  if (!retro_atomic_cas_ptr(&m_layout, nullptr, layout))
   {
     if (layout)
       layout->Release();
 
-    layout = expected;
+    layout = static_cast<ID3D11InputLayout*>(retro_atomic_load_acquire_ptr(&m_layout));
   }
 
   return layout;
diff --git a/Source/Core/VideoBackends/D3D/D3DPerfQuery.cpp b/Source/Core/VideoBackends/D3D/D3DPerfQuery.cpp
index 1da7f2683d..014b5a2ce2 100644
--- a/Source/Core/VideoBackends/D3D/D3DPerfQuery.cpp
+++ b/Source/Core/VideoBackends/D3D/D3DPerfQuery.cpp
@@ -27,20 +27,20 @@ PerfQuery::~PerfQuery() = default;
 
 void PerfQuery::EnableQuery(PerfQueryGroup group)
 {
-  u32 query_count = m_query_count.load(std::memory_order_relaxed);
+  u32 query_count = QueryCount();
 
   // Is this sane?
   if (query_count > m_query_buffer.size() / 2)
   {
     WeakFlush();
-    query_count = m_query_count.load(std::memory_order_relaxed);
+    query_count = QueryCount();
   }
 
   if (m_query_buffer.size() == query_count)
   {
     // TODO
     FlushOne();
-    query_count = m_query_count.load(std::memory_order_relaxed);
+    query_count = QueryCount();
     ERROR_LOG_FMT(VIDEO, "Flushed query buffer early!");
   }
 
@@ -52,7 +52,7 @@ void PerfQuery::EnableQuery(PerfQueryGroup group)
     D3D::context->Begin(entry.query.Get());
     entry.query_group = group;
 
-    m_query_count.fetch_add(1, std::memory_order_relaxed);
+    AddQueryCount(1);
   }
 }
 
@@ -61,8 +61,7 @@ void PerfQuery::DisableQuery(PerfQueryGroup group)
   // stop query
   if (group == PQG_ZCOMP_ZCOMPLOC || group == PQG_ZCOMP)
   {
-    auto& entry = m_query_buffer[(m_query_read_pos + m_query_count.load(std::memory_order_relaxed) +
-                                  m_query_buffer.size() - 1) %
+    auto& entry = m_query_buffer[(m_query_read_pos + QueryCount() + m_query_buffer.size() - 1) %
                                  m_query_buffer.size()];
     D3D::context->End(entry.query.Get());
   }
@@ -70,9 +69,7 @@ void PerfQuery::DisableQuery(PerfQueryGroup group)
 
 void PerfQuery::ResetQuery()
 {
-  m_query_count.store(0, std::memory_order_relaxed);
-  for (auto& result : m_results)
-    result.store(0, std::memory_order_relaxed);
+  ResetQueryState();
 }
 
 u32 PerfQuery::GetQueryResult(PerfQueryType type)
@@ -81,20 +78,19 @@ u32 PerfQuery::GetQueryResult(PerfQueryType type)
 
   if (type == PQ_ZCOMP_INPUT_ZCOMPLOC || type == PQ_ZCOMP_OUTPUT_ZCOMPLOC)
   {
-    result = m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_ZCOMP_INPUT || type == PQ_ZCOMP_OUTPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP);
   }
   else if (type == PQ_BLEND_INPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed) +
-             m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP) + QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_EFB_COPY_CLOCKS)
   {
-    result = m_results[PQG_EFB_COPY_CLOCKS].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_EFB_COPY_CLOCKS);
   }
 
   return result / 4;
@@ -119,11 +115,10 @@ void PerfQuery::FlushOne()
                           g_framebuffer_manager->GetEFBHeight();
   if (g_ActiveConfig.iMultisamples > 1)
     native_res_result /= g_ActiveConfig.iMultisamples;
-  m_results[entry.query_group].fetch_add(static_cast<u32>(native_res_result),
-                                         std::memory_order_relaxed);
+  AddQueryResult(entry.query_group, static_cast<u32>(native_res_result));
 
   m_query_read_pos = (m_query_read_pos + 1) % m_query_buffer.size();
-  m_query_count.fetch_sub(1, std::memory_order_relaxed);
+  SubQueryCount(1);
 }
 
 // TODO: could selectively flush things, but I don't think that will do much
@@ -148,11 +143,10 @@ void PerfQuery::WeakFlush()
       // NOTE: Reported pixel metrics should be referenced to native resolution
       const u64 native_res_result = result * EFB_WIDTH / g_framebuffer_manager->GetEFBWidth() *
                                     EFB_HEIGHT / g_framebuffer_manager->GetEFBHeight();
-      m_results[entry.query_group].store(static_cast<u32>(native_res_result),
-                                         std::memory_order_relaxed);
+      SetQueryResult(entry.query_group, static_cast<u32>(native_res_result));
 
       m_query_read_pos = (m_query_read_pos + 1) % m_query_buffer.size();
-      m_query_count.fetch_sub(1, std::memory_order_relaxed);
+      SubQueryCount(1);
     }
     else
     {
@@ -163,7 +157,7 @@ void PerfQuery::WeakFlush()
 
 bool PerfQuery::IsFlushed() const
 {
-  return m_query_count.load(std::memory_order_relaxed) == 0;
+  return QueryCount() == 0;
 }
 
 }  // namespace DX11
diff --git a/Source/Core/VideoBackends/D3D/D3DVertexManager.h b/Source/Core/VideoBackends/D3D/D3DVertexManager.h
index 18a9194ece..dcbbedb43d 100644
--- a/Source/Core/VideoBackends/D3D/D3DVertexManager.h
+++ b/Source/Core/VideoBackends/D3D/D3DVertexManager.h
@@ -4,10 +4,11 @@
 #pragma once
 
 #include <array>
-#include <atomic>
 #include <memory>
 #include <vector>
 
+#include <retro_atomic.h>
+
 #include "VideoBackends/D3D/D3DBase.h"
 #include "VideoCommon/NativeVertexFormat.h"
 #include "VideoCommon/VertexManagerBase.h"
@@ -29,7 +30,7 @@ private:
   std::array<D3D11_INPUT_ELEMENT_DESC, 32> m_elems{};
   UINT m_num_elems = 0;
 
-  std::atomic<ID3D11InputLayout*> m_layout{nullptr};
+  retro_atomic_ptr_t m_layout{};
 };
 
 class VertexManager : public VertexManagerBase
diff --git a/Source/Core/VideoBackends/D3D12/D3D12PerfQuery.cpp b/Source/Core/VideoBackends/D3D12/D3D12PerfQuery.cpp
index a27633c344..df7d6677cb 100644
--- a/Source/Core/VideoBackends/D3D12/D3D12PerfQuery.cpp
+++ b/Source/Core/VideoBackends/D3D12/D3D12PerfQuery.cpp
@@ -55,7 +55,7 @@ void PerfQuery::EnableQuery(PerfQueryGroup group)
 {
   // Block if there are no free slots.
   // Otherwise, try to keep half of them available.
-  const u32 query_count = m_query_count.load(std::memory_order_relaxed);
+  const u32 query_count = QueryCount();
   if (query_count > m_query_buffer.size() / 2)
   {
     const bool do_resolve = m_unresolved_queries > m_query_buffer.size() / 2;
@@ -88,20 +88,19 @@ void PerfQuery::DisableQuery(PerfQueryGroup group)
     g_dx_context->GetCommandList()->EndQuery(m_query_heap.Get(), D3D12_QUERY_TYPE_OCCLUSION,
                                              m_query_next_pos);
     m_query_next_pos = (m_query_next_pos + 1) % PERF_QUERY_BUFFER_SIZE;
-    m_query_count.fetch_add(1, std::memory_order_relaxed);
+    AddQueryCount(1);
     m_unresolved_queries++;
   }
 }
 
 void PerfQuery::ResetQuery()
 {
-  m_query_count.store(0, std::memory_order_relaxed);
+  retro_atomic_store_release_int(&m_query_count, 0);
   m_unresolved_queries = 0;
   m_query_resolve_pos = 0;
   m_query_readback_pos = 0;
   m_query_next_pos = 0;
-  for (size_t i = 0; i < m_results.size(); ++i)
-    m_results[i].store(0, std::memory_order_relaxed);
+  ResetQueryResults();
   for (auto& entry : m_query_buffer)
   {
     entry.fence_value = 0;
@@ -115,20 +114,19 @@ u32 PerfQuery::GetQueryResult(PerfQueryType type)
   u32 result = 0;
   if (type == PQ_ZCOMP_INPUT_ZCOMPLOC || type == PQ_ZCOMP_OUTPUT_ZCOMPLOC)
   {
-    result = m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_ZCOMP_INPUT || type == PQ_ZCOMP_OUTPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP);
   }
   else if (type == PQ_BLEND_INPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed) +
-             m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP) + QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_EFB_COPY_CLOCKS)
   {
-    result = m_results[PQG_EFB_COPY_CLOCKS].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_EFB_COPY_CLOCKS);
   }
 
   return result / 4;
@@ -142,7 +140,7 @@ void PerfQuery::FlushResults()
 
 bool PerfQuery::IsFlushed() const
 {
-  return m_query_count.load(std::memory_order_relaxed) == 0;
+  return QueryCount() == 0;
 }
 
 void PerfQuery::ResolveQueries()
@@ -180,7 +178,7 @@ void PerfQuery::ReadbackQueries(bool blocking)
   u64 completed_fence_counter = g_dx_context->GetCompletedFenceValue();
 
   // Need to save these since ProcessResults will modify them.
-  const u32 outstanding_queries = m_query_count.load(std::memory_order_relaxed);
+  const u32 outstanding_queries = QueryCount();
   u32 readback_count = 0;
   for (u32 i = 0; i < outstanding_queries; i++)
   {
@@ -218,7 +216,7 @@ void PerfQuery::ReadbackQueries(bool blocking)
 void PerfQuery::AccumulateQueriesFromBuffer(u32 query_count)
 {
   // Should be at maximum query_count queries pending.
-  ASSERT(query_count <= m_query_count.load(std::memory_order_relaxed) &&
+  ASSERT(query_count <= QueryCount() &&
          (m_query_readback_pos + query_count) <= PERF_QUERY_BUFFER_SIZE);
 
   const D3D12_RANGE read_range = {m_query_readback_pos * sizeof(PerfQueryDataType),
@@ -251,15 +249,14 @@ void PerfQuery::AccumulateQueriesFromBuffer(u32 query_count)
                             g_framebuffer_manager->GetEFBHeight();
     if (g_ActiveConfig.iMultisamples > 1)
       native_res_result /= g_ActiveConfig.iMultisamples;
-    m_results[entry.query_group].fetch_add(static_cast<u32>(native_res_result),
-                                           std::memory_order_relaxed);
+    AddQueryResult(entry.query_group, static_cast<u32>(native_res_result));
   }
 
   constexpr D3D12_RANGE write_range = {0, 0};
   m_query_readback_buffer->Unmap(0, &write_range);
 
   m_query_readback_pos = (m_query_readback_pos + query_count) % PERF_QUERY_BUFFER_SIZE;
-  m_query_count.fetch_sub(query_count, std::memory_order_relaxed);
+  SubQueryCount(query_count);
 }
 
 void PerfQuery::PartialFlush(bool resolve, bool blocking)
diff --git a/Source/Core/VideoBackends/Metal/MTLPerfQuery.h b/Source/Core/VideoBackends/Metal/MTLPerfQuery.h
index 993b92e93c..63711bcfc4 100644
--- a/Source/Core/VideoBackends/Metal/MTLPerfQuery.h
+++ b/Source/Core/VideoBackends/Metal/MTLPerfQuery.h
@@ -22,7 +22,7 @@ public:
 
   /// Notify PerfQuery of a new pending encoder
   /// One call to ReturnResults should be made for every call to IncCount
-  void IncCount() { m_query_count.fetch_add(1, std::memory_order_relaxed); }
+  void IncCount() { AddQueryCount(1); }
   /// May be called from any thread
   void ReturnResults(const u64* data, const PerfQueryGroup* groups, size_t count, u32 query_id);
 
diff --git a/Source/Core/VideoBackends/Metal/MTLPerfQuery.mm b/Source/Core/VideoBackends/Metal/MTLPerfQuery.mm
index 37036f28aa..c60888fa2e 100644
--- a/Source/Core/VideoBackends/Metal/MTLPerfQuery.mm
+++ b/Source/Core/VideoBackends/Metal/MTLPerfQuery.mm
@@ -21,8 +21,7 @@ void Metal::PerfQuery::ResetQuery()
 {
   std::lock_guard<std::mutex> lock(m_results_mtx);
   m_current_query++;
-  for (std::atomic<u32>& result : m_results)
-    result.store(0, std::memory_order_relaxed);
+  ResetQueryResults();
 }
 
 u32 Metal::PerfQuery::GetQueryResult(PerfQueryType type)
@@ -30,20 +29,19 @@ u32 Metal::PerfQuery::GetQueryResult(PerfQueryType type)
   u32 result = 0;
   if (type == PQ_ZCOMP_INPUT_ZCOMPLOC || type == PQ_ZCOMP_OUTPUT_ZCOMPLOC)
   {
-    result = m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_ZCOMP_INPUT || type == PQ_ZCOMP_OUTPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP);
   }
   else if (type == PQ_BLEND_INPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed) +
-             m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP) + QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_EFB_COPY_CLOCKS)
   {
-    result = m_results[PQG_EFB_COPY_CLOCKS].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_EFB_COPY_CLOCKS);
   }
 
   return result / 4;
@@ -65,7 +63,7 @@ void Metal::PerfQuery::FlushResults()
 
 bool Metal::PerfQuery::IsFlushed() const
 {
-  return m_query_count.load(std::memory_order_acquire) == 0;
+  return QueryCount() == 0;
 }
 
 void Metal::PerfQuery::ReturnResults(const u64* data, const PerfQueryGroup* groups, size_t count,
@@ -83,10 +81,10 @@ void Metal::PerfQuery::ReturnResults(const u64* data, const PerfQueryGroup* grou
 
         native_res_result /= g_ActiveConfig.iMultisamples;
 
-        m_results[groups[i]].fetch_add(native_res_result, std::memory_order_relaxed);
+        AddQueryResult(groups[i], static_cast<u32>(native_res_result));
       }
     }
-    m_query_count.fetch_sub(1, std::memory_order_release);
+    SubQueryCount(1);
   }
   m_cv.notify_one();
 }
diff --git a/Source/Core/VideoBackends/OGL/OGLPerfQuery.cpp b/Source/Core/VideoBackends/OGL/OGLPerfQuery.cpp
index 022b4de70e..b8f269c8af 100644
--- a/Source/Core/VideoBackends/OGL/OGLPerfQuery.cpp
+++ b/Source/Core/VideoBackends/OGL/OGLPerfQuery.cpp
@@ -41,7 +41,7 @@ void PerfQuery::DisableQuery(PerfQueryGroup group)
 
 bool PerfQuery::IsFlushed() const
 {
-  return m_query_count.load(std::memory_order_relaxed) == 0;
+  return QueryCount() == 0;
 }
 
 // TODO: could selectively flush things, but I don't think that will do much
@@ -52,9 +52,7 @@ void PerfQuery::FlushResults()
 
 void PerfQuery::ResetQuery()
 {
-  m_query_count.store(0, std::memory_order_relaxed);
-  for (auto& result : m_results)
-    result.store(0, std::memory_order_relaxed);
+  ResetQueryState();
 }
 
 u32 PerfQuery::GetQueryResult(PerfQueryType type)
@@ -63,20 +61,19 @@ u32 PerfQuery::GetQueryResult(PerfQueryType type)
 
   if (type == PQ_ZCOMP_INPUT_ZCOMPLOC || type == PQ_ZCOMP_OUTPUT_ZCOMPLOC)
   {
-    result = m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_ZCOMP_INPUT || type == PQ_ZCOMP_OUTPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP);
   }
   else if (type == PQ_BLEND_INPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed) +
-             m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP) + QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_EFB_COPY_CLOCKS)
   {
-    result = m_results[PQG_EFB_COPY_CLOCKS].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_EFB_COPY_CLOCKS);
   }
 
   return result / 4;
@@ -97,19 +94,19 @@ PerfQueryGL::~PerfQueryGL()
 
 void PerfQueryGL::EnableQuery(PerfQueryGroup group)
 {
-  u32 query_count = m_query_count.load(std::memory_order_relaxed);
+  u32 query_count = QueryCount();
 
   // Is this sane?
   if (query_count > m_query_buffer.size() / 2)
   {
     WeakFlush();
-    query_count = m_query_count.load(std::memory_order_relaxed);
+    query_count = QueryCount();
   }
 
   if (m_query_buffer.size() == query_count)
   {
     FlushOne();
-    query_count = m_query_count.load(std::memory_order_relaxed);
+    query_count = QueryCount();
     // ERROR_LOG_FMT(VIDEO, "Flushed query buffer early!");
   }
 
@@ -121,7 +118,7 @@ void PerfQueryGL::EnableQuery(PerfQueryGroup group)
     glBeginQuery(m_query_type, entry.query_id);
     entry.query_group = group;
 
-    m_query_count.fetch_add(1, std::memory_order_relaxed);
+    AddQueryCount(1);
   }
 }
 void PerfQueryGL::DisableQuery(PerfQueryGroup group)
@@ -170,10 +167,10 @@ void PerfQueryGL::FlushOne()
   if (g_ActiveConfig.iMultisamples > 1)
     result /= g_ActiveConfig.iMultisamples;
 
-  m_results[entry.query_group].fetch_add(result, std::memory_order_relaxed);
+  AddQueryResult(entry.query_group, result);
 
   m_query_read_pos = (m_query_read_pos + 1) % m_query_buffer.size();
-  m_query_count.fetch_sub(1, std::memory_order_relaxed);
+  SubQueryCount(1);
 }
 
 // TODO: could selectively flush things, but I don't think that will do much
@@ -197,19 +194,19 @@ PerfQueryGLESNV::~PerfQueryGLESNV()
 
 void PerfQueryGLESNV::EnableQuery(PerfQueryGroup group)
 {
-  u32 query_count = m_query_count.load(std::memory_order_relaxed);
+  u32 query_count = QueryCount();
 
   // Is this sane?
   if (query_count > m_query_buffer.size() / 2)
   {
     WeakFlush();
-    query_count = m_query_count.load(std::memory_order_relaxed);
+    query_count = QueryCount();
   }
 
   if (m_query_buffer.size() == query_count)
   {
     FlushOne();
-    query_count = m_query_count.load(std::memory_order_relaxed);
+    query_count = QueryCount();
     // ERROR_LOG_FMT(VIDEO, "Flushed query buffer early!");
   }
 
@@ -221,7 +218,7 @@ void PerfQueryGLESNV::EnableQuery(PerfQueryGroup group)
     glBeginOcclusionQueryNV(entry.query_id);
     entry.query_group = group;
 
-    m_query_count.fetch_add(1, std::memory_order_relaxed);
+    AddQueryCount(1);
   }
 }
 void PerfQueryGLESNV::DisableQuery(PerfQueryGroup group)
@@ -266,11 +263,10 @@ void PerfQueryGLESNV::FlushOne()
   const u64 native_res_result =
       static_cast<u64>(result) * EFB_WIDTH * EFB_HEIGHT /
       (g_framebuffer_manager->GetEFBWidth() * g_framebuffer_manager->GetEFBHeight());
-  m_results[entry.query_group].fetch_add(static_cast<u32>(native_res_result),
-                                         std::memory_order_relaxed);
+  AddQueryResult(entry.query_group, static_cast<u32>(native_res_result));
 
   m_query_read_pos = (m_query_read_pos + 1) % m_query_buffer.size();
-  m_query_count.fetch_sub(1, std::memory_order_relaxed);
+  SubQueryCount(1);
 }
 
 // TODO: could selectively flush things, but I don't think that will do much
diff --git a/Source/Core/VideoBackends/Vulkan/CommandBufferManager.cpp b/Source/Core/VideoBackends/Vulkan/CommandBufferManager.cpp
index c82417eb8c..b9206c5d5a 100644
--- a/Source/Core/VideoBackends/Vulkan/CommandBufferManager.cpp
+++ b/Source/Core/VideoBackends/Vulkan/CommandBufferManager.cpp
@@ -234,7 +234,7 @@ bool CommandBufferManager::CreateSubmitThread()
     SubmitCommandBuffer(submit.command_buffer_index, submit.present_swap_chain,
                         submit.present_image_index);
     CmdBufferResources& resources = m_command_buffers[submit.command_buffer_index];
-    resources.waiting_for_submit.store(false, std::memory_order_release);
+    retro_atomic_store_release_int(&resources.waiting_for_submit, 0);
   });
 
   return true;
@@ -272,10 +272,10 @@ void CommandBufferManager::WaitForCommandBufferCompletion(u32 index)
   CmdBufferResources& resources = m_command_buffers[index];
 
   // Ensure this command buffer has been submitted.
-  if (resources.waiting_for_submit.load(std::memory_order_acquire))
+  if (retro_atomic_load_acquire_int(&resources.waiting_for_submit))
   {
     WaitForWorkerThreadIdle();
-    ASSERT_MSG(VIDEO, !resources.waiting_for_submit.load(std::memory_order_relaxed),
+    ASSERT_MSG(VIDEO, !retro_atomic_load_relaxed_int(&resources.waiting_for_submit),
                "Submit thread is idle but command buffer is still waiting for submission!");
   }
 
@@ -329,7 +329,7 @@ void CommandBufferManager::SubmitCommandBuffer(bool submit_on_worker_thread,
   // Submitting off-thread?
   if (m_use_threaded_submission && submit_on_worker_thread && !wait_for_completion)
   {
-    resources.waiting_for_submit.store(true, std::memory_order_relaxed);
+    retro_atomic_store_relaxed_int(&resources.waiting_for_submit, 1);
     // Push to the pending submit queue.
     m_submit_thread.Push({present_swap_chain, present_image_index, m_current_cmd_buffer});
   }
diff --git a/Source/Core/VideoBackends/Vulkan/CommandBufferManager.h b/Source/Core/VideoBackends/Vulkan/CommandBufferManager.h
index 45a322c270..e4abb98d07 100644
--- a/Source/Core/VideoBackends/Vulkan/CommandBufferManager.h
+++ b/Source/Core/VideoBackends/Vulkan/CommandBufferManager.h
@@ -8,6 +8,8 @@
 #include <memory>
 #include <vector>
 
+#include <retro_atomic.h>
+
 #include <Common/WorkQueueThread.h>
 #include "Common/Flag.h"
 
@@ -118,7 +120,7 @@ private:
     u64 fence_counter = 0;
     bool init_command_buffer_used = false;
     bool semaphore_used = false;
-    std::atomic<bool> waiting_for_submit{false};
+    retro_atomic_int_t waiting_for_submit{};
     u32 frame_index = 0;
 
     std::vector<std::function<void()>> cleanup_resources;
diff --git a/Source/Core/VideoBackends/Vulkan/VKPerfQuery.cpp b/Source/Core/VideoBackends/Vulkan/VKPerfQuery.cpp
index 9949535d88..c07900b660 100644
--- a/Source/Core/VideoBackends/Vulkan/VKPerfQuery.cpp
+++ b/Source/Core/VideoBackends/Vulkan/VKPerfQuery.cpp
@@ -44,7 +44,7 @@ void PerfQuery::EnableQuery(PerfQueryGroup group)
 {
   // Block if there are no free slots.
   // Otherwise, try to keep half of them available.
-  const u32 query_count = m_query_count.load(std::memory_order_relaxed);
+  const u32 query_count = QueryCount();
   if (query_count > m_query_buffer.size() / 2)
     PartialFlush(query_count == PERF_QUERY_BUFFER_SIZE);
 
@@ -79,17 +79,16 @@ void PerfQuery::DisableQuery(PerfQueryGroup group)
     entry.fence_counter = g_command_buffer_mgr->GetCurrentFenceCounter();
 
     m_query_next_pos = (m_query_next_pos + 1) % PERF_QUERY_BUFFER_SIZE;
-    m_query_count.fetch_add(1, std::memory_order_relaxed);
+    AddQueryCount(1);
   }
 }
 
 void PerfQuery::ResetQuery()
 {
-  m_query_count.store(0, std::memory_order_relaxed);
+  retro_atomic_store_release_int(&m_query_count, 0);
   m_query_readback_pos = 0;
   m_query_next_pos = 0;
-  for (size_t i = 0; i < m_results.size(); ++i)
-    m_results[i].store(0, std::memory_order_relaxed);
+  ResetQueryResults();
 
   // Reset entire query pool, ensuring all queries are ready to write to.
   StateTracker::GetInstance()->EndRenderPass();
@@ -104,20 +103,19 @@ u32 PerfQuery::GetQueryResult(PerfQueryType type)
   u32 result = 0;
   if (type == PQ_ZCOMP_INPUT_ZCOMPLOC || type == PQ_ZCOMP_OUTPUT_ZCOMPLOC)
   {
-    result = m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_ZCOMP_INPUT || type == PQ_ZCOMP_OUTPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP);
   }
   else if (type == PQ_BLEND_INPUT)
   {
-    result = m_results[PQG_ZCOMP].load(std::memory_order_relaxed) +
-             m_results[PQG_ZCOMP_ZCOMPLOC].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_ZCOMP) + QueryResult(PQG_ZCOMP_ZCOMPLOC);
   }
   else if (type == PQ_EFB_COPY_CLOCKS)
   {
-    result = m_results[PQG_EFB_COPY_CLOCKS].load(std::memory_order_relaxed);
+    result = QueryResult(PQG_EFB_COPY_CLOCKS);
   }
 
   return result / 4;
@@ -133,7 +131,7 @@ void PerfQuery::FlushResults()
 
 bool PerfQuery::IsFlushed() const
 {
-  return m_query_count.load(std::memory_order_relaxed) == 0;
+  return QueryCount() == 0;
 }
 
 bool PerfQuery::CreateQueryPool()
@@ -162,7 +160,7 @@ void PerfQuery::ReadbackQueries()
   const u64 completed_fence_counter = g_command_buffer_mgr->GetCompletedFenceCounter();
 
   // Need to save these since ProcessResults will modify them.
-  const u32 outstanding_queries = m_query_count.load(std::memory_order_relaxed);
+  const u32 outstanding_queries = QueryCount();
   u32 readback_count = 0;
   for (u32 i = 0; i < outstanding_queries; i++)
   {
@@ -189,7 +187,7 @@ void PerfQuery::ReadbackQueries()
 void PerfQuery::ReadbackQueries(u32 query_count)
 {
   // Should be at maximum query_count queries pending.
-  ASSERT(query_count <= m_query_count.load(std::memory_order_relaxed) &&
+  ASSERT(query_count <= QueryCount() &&
          (m_query_readback_pos + query_count) <= PERF_QUERY_BUFFER_SIZE);
 
   // Read back from the GPU.
@@ -221,12 +219,11 @@ void PerfQuery::ReadbackQueries(u32 query_count)
                             g_framebuffer_manager->GetEFBHeight();
     if (g_ActiveConfig.iMultisamples > 1)
       native_res_result /= g_ActiveConfig.iMultisamples;
-    m_results[entry.query_group].fetch_add(static_cast<u32>(native_res_result),
-                                           std::memory_order_relaxed);
+    AddQueryResult(entry.query_group, static_cast<u32>(native_res_result));
   }
 
   m_query_readback_pos = (m_query_readback_pos + query_count) % PERF_QUERY_BUFFER_SIZE;
-  m_query_count.fetch_sub(query_count, std::memory_order_relaxed);
+  SubQueryCount(query_count);
 }
 
 void PerfQuery::PartialFlush(bool blocking)
diff --git a/Source/Core/VideoCommon/Assets/CustomAsset.cpp b/Source/Core/VideoCommon/Assets/CustomAsset.cpp
index 3105575d80..2d41eb86a6 100644
--- a/Source/Core/VideoCommon/Assets/CustomAsset.cpp
+++ b/Source/Core/VideoCommon/Assets/CustomAsset.cpp
@@ -32,7 +32,7 @@ std::size_t CustomAsset::Load()
   if (load_information.bytes_loaded > 0)
   {
     m_bytes_loaded = load_information.bytes_loaded;
-    m_last_loaded_time = load_time;
+    retro_atomic_store_release_64(&m_last_loaded_time, load_time.time_since_epoch().count());
     retro_atomic_store_release_int(&m_busy, 0);
     return load_information.bytes_loaded;
   }
@@ -52,7 +52,7 @@ std::size_t CustomAsset::Unload()
 
 CustomAsset::TimeType CustomAsset::GetLastLoadedTime() const
 {
-  return m_last_loaded_time;
+  return TimeType(ClockType::duration(retro_atomic_load_acquire_64(&m_last_loaded_time)));
 }
 
 std::size_t CustomAsset::GetHandle() const
diff --git a/Source/Core/VideoCommon/Assets/CustomAsset.h b/Source/Core/VideoCommon/Assets/CustomAsset.h
index e7bc9c64af..f3ff014b6e 100644
--- a/Source/Core/VideoCommon/Assets/CustomAsset.h
+++ b/Source/Core/VideoCommon/Assets/CustomAsset.h
@@ -9,7 +9,6 @@
 #include "Common/CommonTypes.h"
 #include "VideoCommon/Assets/CustomAssetLibrary.h"
 
-#include <atomic>
 #include <memory>
 
 namespace VideoCommon
@@ -64,7 +63,7 @@ private:
   // Held (with a CAS) by the one thread loading or unloading; anyone else backs off.
   retro_atomic_int_t m_busy{};
   std::size_t m_bytes_loaded = 0;  // only touched while busy
-  std::atomic<TimeType> m_last_loaded_time = {};
+  retro_atomic_64_t m_last_loaded_time{}; /* TimeType ticks */
 };
 
 // An abstract class that is expected to
diff --git a/Source/Core/VideoCommon/Assets/CustomAssetLoader.cpp b/Source/Core/VideoCommon/Assets/CustomAssetLoader.cpp
index 0a1e1a486c..60863c7366 100644
--- a/Source/Core/VideoCommon/Assets/CustomAssetLoader.cpp
+++ b/Source/Core/VideoCommon/Assets/CustomAssetLoader.cpp
@@ -97,7 +97,7 @@ void CustomAssetLoader::WorkerThreadRun(u32 thread_index)
     {
       // If more memory than allowed has already been loaded, we will load nothing more
       //  until the next ScheduleAssetsToLoad from Manager.
-      if (m_change_in_memory > batch->allowed_memory)
+      if (retro_atomic_load_acquire_64(&m_change_in_memory) > batch->allowed_memory)
         retro_atomic_store_release_size(&batch->cursor, batch->assets.size());
 
       const size_t index = retro_atomic_fetch_add_size(&batch->cursor, 1);
@@ -126,10 +126,10 @@ void CustomAssetLoader::WorkerThreadRun(u32 thread_index)
 
     // Unload previously loaded asset. An asset another thread is busy with comes back 0
     // and is skipped; the Manager schedules it again.
-    m_change_in_memory -= item->Unload();
+    retro_atomic_fetch_sub_64(&m_change_in_memory, static_cast<s64>(item->Unload()));
 
     const std::size_t bytes_loaded = item->Load();
-    m_change_in_memory += s64(bytes_loaded);
+    retro_atomic_fetch_add_64(&m_change_in_memory, static_cast<s64>(bytes_loaded));
 
     INFO_LOG_FMT(VIDEO, "CustomAssetLoader thread {} loaded: {} ({})", thread_index,
                  item->GetAssetId(), UICommon::FormatSize(bytes_loaded));
@@ -149,7 +149,7 @@ auto CustomAssetLoader::TakeLoadResults() -> LoadResults
     results.asset_handles.push_back(loaded->handle);
     delete loaded;
   }
-  results.change_in_memory = m_change_in_memory.exchange(0);
+  results.change_in_memory = retro_atomic_exchange_64(&m_change_in_memory, 0);
   return results;
 }
 
@@ -177,7 +177,7 @@ void CustomAssetLoader::Reset(bool restart_worker_threads)
   delete static_cast<Batch*>(retro_atomic_exchange_ptr(&m_batch, nullptr));
   quiesce_deinit(&m_batch_quiesce);
   TakeLoadResults();
-  m_change_in_memory = 0;
+  retro_atomic_store_release_64(&m_change_in_memory, 0);
 
   if (restart_worker_threads)
   {
diff --git a/Source/Core/VideoCommon/Assets/CustomAssetLoader.h b/Source/Core/VideoCommon/Assets/CustomAssetLoader.h
index 703bf0f03e..702e0c62fe 100644
--- a/Source/Core/VideoCommon/Assets/CustomAssetLoader.h
+++ b/Source/Core/VideoCommon/Assets/CustomAssetLoader.h
@@ -3,7 +3,6 @@
 
 #pragma once
 
-#include <atomic>
 #include <list>
 #include <vector>
 
@@ -96,6 +95,6 @@ private:
   mpsc_stack_t m_loaded{};
 
   // Change in memory from just-loaded/unloaded asset results yet to be taken by the Manager.
-  std::atomic<s64> m_change_in_memory = 0;
+  retro_atomic_64_t m_change_in_memory{};
 };
 }  // namespace VideoCommon
diff --git a/Source/Core/VideoCommon/CommandProcessor.cpp b/Source/Core/VideoCommon/CommandProcessor.cpp
index 1dbf466f1c..eb2cd58e3e 100644
--- a/Source/Core/VideoCommon/CommandProcessor.cpp
+++ b/Source/Core/VideoCommon/CommandProcessor.cpp
@@ -3,7 +3,6 @@
 
 #include "VideoCommon/CommandProcessor.h"
 
-#include <atomic>
 #include <fmt/format.h>
 
 #include "Common/Assert.h"
@@ -34,26 +33,26 @@ static void UpdateInterrupts_Wrapper(Core::System& system, u64 userdata, s64 cyc
 
 void SCPFifoStruct::Init()
 {
-  CPBase = 0;
-  CPEnd = 0;
+  StoreReg(&CPBase, 0);
+  StoreReg(&CPEnd, 0);
   CPHiWatermark = 0;
   CPLoWatermark = 0;
-  CPReadWriteDistance = 0;
-  CPWritePointer = 0;
-  CPReadPointer = 0;
-  CPBreakpoint = 0;
-  SafeCPReadPointer = 0;
+  StoreReg(&CPReadWriteDistance, 0);
+  StoreReg(&CPWritePointer, 0);
+  StoreReg(&CPReadPointer, 0);
+  StoreReg(&CPBreakpoint, 0);
+  StoreReg(&SafeCPReadPointer, 0);
 
-  bFF_GPLinkEnable = 0;
-  bFF_GPReadEnable = 0;
-  bFF_BPEnable = 0;
-  bFF_BPInt = 0;
+  StoreReg(&bFF_GPLinkEnable, 0);
+  StoreReg(&bFF_GPReadEnable, 0);
+  StoreReg(&bFF_BPEnable, 0);
+  StoreReg(&bFF_BPInt, 0);
 
-  bFF_Breakpoint.store(0, std::memory_order_relaxed);
-  bFF_HiWatermark.store(0, std::memory_order_relaxed);
-  bFF_HiWatermarkInt.store(0, std::memory_order_relaxed);
-  bFF_LoWatermark.store(0, std::memory_order_relaxed);
-  bFF_LoWatermarkInt.store(0, std::memory_order_relaxed);
+  StoreReg(&bFF_Breakpoint, 0);
+  StoreReg(&bFF_HiWatermark, 0);
+  StoreReg(&bFF_HiWatermarkInt, 0);
+  StoreReg(&bFF_LoWatermark, 0);
+  StoreReg(&bFF_LoWatermarkInt, 0);
 }
 
 void SCPFifoStruct::DoState(PointerWrap& p)
@@ -94,10 +93,9 @@ void CommandProcessorManager::DoState(PointerWrap& p)
   p.Do(m_interrupt_waiting);
 }
 
-static inline void WriteHigh(std::atomic<u32>& reg, u16 highbits)
+static inline void WriteHigh(retro_atomic_int_t* reg, u16 highbits)
 {
-  reg.store((reg.load(std::memory_order_relaxed) & 0x0000FFFF) | (static_cast<u32>(highbits) << 16),
-            std::memory_order_relaxed);
+  StoreReg(reg, (LoadReg(reg) & 0x0000FFFF) | (static_cast<u32>(highbits) << 16));
 }
 
 CommandProcessorManager::CommandProcessorManager(Core::System& system) : m_system{system}
@@ -232,18 +230,14 @@ void CommandProcessorManager::RegisterMMIO(MMIO::Mapping* mmio, u32 base)
   {
     fifo_rw_distance_lo_r = MMIO::ComplexRead<u16>([](Core::System& system_, u32) {
       const auto& fifo_ = system_.GetCommandProcessor().GetFifo();
-      if (fifo_.CPWritePointer.load(std::memory_order_relaxed) >=
-          fifo_.SafeCPReadPointer.load(std::memory_order_relaxed))
+      if (LoadReg(&fifo_.CPWritePointer) >= LoadReg(&fifo_.SafeCPReadPointer))
       {
-        return static_cast<u16>(fifo_.CPWritePointer.load(std::memory_order_relaxed) -
-                                fifo_.SafeCPReadPointer.load(std::memory_order_relaxed));
+        return static_cast<u16>(LoadReg(&fifo_.CPWritePointer) - LoadReg(&fifo_.SafeCPReadPointer));
       }
       else
       {
-        return static_cast<u16>(fifo_.CPEnd.load(std::memory_order_relaxed) -
-                                fifo_.SafeCPReadPointer.load(std::memory_order_relaxed) +
-                                fifo_.CPWritePointer.load(std::memory_order_relaxed) -
-                                fifo_.CPBase.load(std::memory_order_relaxed) + 32);
+        return static_cast<u16>(LoadReg(&fifo_.CPEnd) - LoadReg(&fifo_.SafeCPReadPointer) +
+                                LoadReg(&fifo_.CPWritePointer) - LoadReg(&fifo_.CPBase) + 32);
       }
     });
   }
@@ -262,19 +256,14 @@ void CommandProcessorManager::RegisterMMIO(MMIO::Mapping* mmio, u32 base)
     fifo_rw_distance_hi_r = MMIO::ComplexRead<u16>([](Core::System& system_, u32) {
       const auto& fifo_ = system_.GetCommandProcessor().GetFifo();
       system_.GetFifo().SyncGPUForRegisterAccess();
-      if (fifo_.CPWritePointer.load(std::memory_order_relaxed) >=
-          fifo_.SafeCPReadPointer.load(std::memory_order_relaxed))
+      if (LoadReg(&fifo_.CPWritePointer) >= LoadReg(&fifo_.SafeCPReadPointer))
       {
-        return (fifo_.CPWritePointer.load(std::memory_order_relaxed) -
-                fifo_.SafeCPReadPointer.load(std::memory_order_relaxed)) >>
-               16;
+        return (LoadReg(&fifo_.CPWritePointer) - LoadReg(&fifo_.SafeCPReadPointer)) >> 16;
       }
       else
       {
-        return (fifo_.CPEnd.load(std::memory_order_relaxed) -
-                fifo_.SafeCPReadPointer.load(std::memory_order_relaxed) +
-                fifo_.CPWritePointer.load(std::memory_order_relaxed) -
-                fifo_.CPBase.load(std::memory_order_relaxed) + 32) >>
+        return (LoadReg(&fifo_.CPEnd) - LoadReg(&fifo_.SafeCPReadPointer) +
+                LoadReg(&fifo_.CPWritePointer) - LoadReg(&fifo_.CPBase) + 32) >>
                16;
       }
     });
@@ -284,14 +273,14 @@ void CommandProcessorManager::RegisterMMIO(MMIO::Mapping* mmio, u32 base)
     fifo_rw_distance_hi_r = MMIO::ComplexRead<u16>([](Core::System& system_, u32) {
       const auto& fifo_ = system_.GetCommandProcessor().GetFifo();
       system_.GetFifo().SyncGPUForRegisterAccess();
-      return fifo_.CPReadWriteDistance.load(std::memory_order_relaxed) >> 16;
+      return LoadReg(&fifo_.CPReadWriteDistance) >> 16;
     });
   }
   mmio->Register(base | FIFO_RW_DISTANCE_HI, fifo_rw_distance_hi_r,
                  MMIO::ComplexWrite<u16>([WMASK_HI_RESTRICT](Core::System& system_, u32, u16 val) {
                    auto& fifo_ = system_.GetCommandProcessor().GetFifo();
                    system_.GetFifo().SyncGPUForRegisterAccess();
-                   WriteHigh(fifo_.CPReadWriteDistance, val & WMASK_HI_RESTRICT);
+                   WriteHigh(&fifo_.CPReadWriteDistance, val & WMASK_HI_RESTRICT);
                    system_.GetFifo().RunGpu();
                  }));
 
@@ -308,15 +297,14 @@ void CommandProcessorManager::RegisterMMIO(MMIO::Mapping* mmio, u32 base)
     fifo_read_hi_r = MMIO::ComplexRead<u16>([](Core::System& system_, u32) {
       auto& fifo_ = system_.GetCommandProcessor().GetFifo();
       system_.GetFifo().SyncGPUForRegisterAccess();
-      return fifo_.SafeCPReadPointer.load(std::memory_order_relaxed) >> 16;
+      return LoadReg(&fifo_.SafeCPReadPointer) >> 16;
     });
     fifo_read_hi_w =
         MMIO::ComplexWrite<u16>([WMASK_HI_RESTRICT](Core::System& system_, u32, u16 val) {
           auto& fifo_ = system_.GetCommandProcessor().GetFifo();
           system_.GetFifo().SyncGPUForRegisterAccess();
-          WriteHigh(fifo_.CPReadPointer, val & WMASK_HI_RESTRICT);
-          fifo_.SafeCPReadPointer.store(fifo_.CPReadPointer.load(std::memory_order_relaxed),
-                                        std::memory_order_relaxed);
+          WriteHigh(&fifo_.CPReadPointer, val & WMASK_HI_RESTRICT);
+          StoreReg(&fifo_.SafeCPReadPointer, LoadReg(&fifo_.CPReadPointer));
         });
   }
   else
@@ -324,13 +312,13 @@ void CommandProcessorManager::RegisterMMIO(MMIO::Mapping* mmio, u32 base)
     fifo_read_hi_r = MMIO::ComplexRead<u16>([](Core::System& system_, u32) {
       const auto& fifo_ = system_.GetCommandProcessor().GetFifo();
       system_.GetFifo().SyncGPUForRegisterAccess();
-      return fifo_.CPReadPointer.load(std::memory_order_relaxed) >> 16;
+      return LoadReg(&fifo_.CPReadPointer) >> 16;
     });
     fifo_read_hi_w =
         MMIO::ComplexWrite<u16>([WMASK_HI_RESTRICT](Core::System& system_, u32, u16 val) {
           auto& fifo_ = system_.GetCommandProcessor().GetFifo();
           system_.GetFifo().SyncGPUForRegisterAccess();
-          WriteHigh(fifo_.CPReadPointer, val & WMASK_HI_RESTRICT);
+          WriteHigh(&fifo_.CPReadPointer, val & WMASK_HI_RESTRICT);
         });
   }
   mmio->Register(base | FIFO_READ_POINTER_HI, fifo_read_hi_r, fifo_read_hi_w);
@@ -349,9 +337,9 @@ void CommandProcessorManager::GatherPipeBursted()
     {
       // In multibuffer mode is not allowed write in the same FIFO attached to the GPU.
       // Fix Pokemon XD in DC mode.
-      if ((processor_interface.m_fifo_cpu_end == m_fifo.CPEnd.load(std::memory_order_relaxed)) &&
-          (processor_interface.m_fifo_cpu_base == m_fifo.CPBase.load(std::memory_order_relaxed)) &&
-          m_fifo.CPReadWriteDistance.load(std::memory_order_relaxed) > 0)
+      if ((processor_interface.m_fifo_cpu_end == LoadReg(&m_fifo.CPEnd)) &&
+          (processor_interface.m_fifo_cpu_base == LoadReg(&m_fifo.CPBase)) &&
+          LoadReg(&m_fifo.CPReadWriteDistance) > 0)
       {
         m_system.GetFifo().FlushGpu();
       }
@@ -361,48 +349,42 @@ void CommandProcessorManager::GatherPipeBursted()
   }
 
   // update the fifo pointer
-  if (m_fifo.CPWritePointer.load(std::memory_order_relaxed) ==
-      m_fifo.CPEnd.load(std::memory_order_relaxed))
+  if (LoadReg(&m_fifo.CPWritePointer) == LoadReg(&m_fifo.CPEnd))
   {
-    m_fifo.CPWritePointer.store(m_fifo.CPBase, std::memory_order_relaxed);
+    StoreReg(&m_fifo.CPWritePointer, LoadReg(&m_fifo.CPBase));
   }
   else
   {
-    m_fifo.CPWritePointer.fetch_add(GPFifo::GATHER_PIPE_SIZE, std::memory_order_relaxed);
+    retro_atomic_fetch_add_int(&m_fifo.CPWritePointer, GPFifo::GATHER_PIPE_SIZE);
   }
 
   if (m_cp_ctrl_reg.GPReadEnable && m_cp_ctrl_reg.GPLinkEnable)
   {
-    processor_interface.m_fifo_cpu_write_pointer =
-        m_fifo.CPWritePointer.load(std::memory_order_relaxed);
-    processor_interface.m_fifo_cpu_base = m_fifo.CPBase.load(std::memory_order_relaxed);
-    processor_interface.m_fifo_cpu_end = m_fifo.CPEnd.load(std::memory_order_relaxed);
+    processor_interface.m_fifo_cpu_write_pointer = LoadReg(&m_fifo.CPWritePointer);
+    processor_interface.m_fifo_cpu_base = LoadReg(&m_fifo.CPBase);
+    processor_interface.m_fifo_cpu_end = LoadReg(&m_fifo.CPEnd);
   }
 
   // If the game is running close to overflowing, make the exception checking more frequent.
-  if (m_fifo.bFF_HiWatermark.load(std::memory_order_relaxed) != 0)
+  if (LoadReg(&m_fifo.bFF_HiWatermark) != 0)
     m_system.GetCoreTiming().ForceExceptionCheck(0);
 
-  m_fifo.CPReadWriteDistance.fetch_add(GPFifo::GATHER_PIPE_SIZE, std::memory_order_seq_cst);
+  retro_atomic_fetch_add_seq_cst_int(&m_fifo.CPReadWriteDistance, GPFifo::GATHER_PIPE_SIZE);
 
   m_system.GetFifo().RunGpu();
 
   ASSERT_MSG(COMMANDPROCESSOR,
-             m_fifo.CPReadWriteDistance.load(std::memory_order_relaxed) <=
-                 m_fifo.CPEnd.load(std::memory_order_relaxed) -
-                     m_fifo.CPBase.load(std::memory_order_relaxed),
+             LoadReg(&m_fifo.CPReadWriteDistance) <=
+                 LoadReg(&m_fifo.CPEnd) - LoadReg(&m_fifo.CPBase),
              "FIFO is overflowed by GatherPipe !\nCPU thread is too fast!");
 
   // check if we are in sync
   ASSERT_MSG(COMMANDPROCESSOR,
-             m_fifo.CPWritePointer.load(std::memory_order_relaxed) ==
-                 processor_interface.m_fifo_cpu_write_pointer,
+             LoadReg(&m_fifo.CPWritePointer) == processor_interface.m_fifo_cpu_write_pointer,
              "FIFOs linked but out of sync");
-  ASSERT_MSG(COMMANDPROCESSOR,
-             m_fifo.CPBase.load(std::memory_order_relaxed) == processor_interface.m_fifo_cpu_base,
+  ASSERT_MSG(COMMANDPROCESSOR, LoadReg(&m_fifo.CPBase) == processor_interface.m_fifo_cpu_base,
              "FIFOs linked but out of sync");
-  ASSERT_MSG(COMMANDPROCESSOR,
-             m_fifo.CPEnd.load(std::memory_order_relaxed) == processor_interface.m_fifo_cpu_end,
+  ASSERT_MSG(COMMANDPROCESSOR, LoadReg(&m_fifo.CPEnd) == processor_interface.m_fifo_cpu_end,
              "FIFOs linked but out of sync");
 }
 
@@ -442,26 +424,23 @@ bool CommandProcessorManager::IsInterruptWaiting() const
 void CommandProcessorManager::SetCPStatusFromGPU()
 {
   // breakpoint
-  const bool breakpoint = m_fifo.bFF_Breakpoint.load(std::memory_order_relaxed);
-  if (m_fifo.bFF_BPEnable.load(std::memory_order_relaxed) != 0)
+  const bool breakpoint = LoadReg(&m_fifo.bFF_Breakpoint);
+  if (LoadReg(&m_fifo.bFF_BPEnable) != 0)
   {
-    if (m_fifo.CPBreakpoint.load(std::memory_order_relaxed) ==
-        m_fifo.CPReadPointer.load(std::memory_order_relaxed))
+    if (LoadReg(&m_fifo.CPBreakpoint) == LoadReg(&m_fifo.CPReadPointer))
     {
       if (!breakpoint)
       {
-        DEBUG_LOG_FMT(COMMANDPROCESSOR, "Hit breakpoint at {}",
-                      m_fifo.CPReadPointer.load(std::memory_order_relaxed));
-        m_fifo.bFF_Breakpoint.store(1, std::memory_order_relaxed);
+        DEBUG_LOG_FMT(COMMANDPROCESSOR, "Hit breakpoint at {}", LoadReg(&m_fifo.CPReadPointer));
+        StoreReg(&m_fifo.bFF_Breakpoint, 1);
       }
     }
     else
     {
       if (breakpoint)
       {
-        DEBUG_LOG_FMT(COMMANDPROCESSOR, "Cleared breakpoint at {}",
-                      m_fifo.CPReadPointer.load(std::memory_order_relaxed));
-        m_fifo.bFF_Breakpoint.store(0, std::memory_order_relaxed);
+        DEBUG_LOG_FMT(COMMANDPROCESSOR, "Cleared breakpoint at {}", LoadReg(&m_fifo.CPReadPointer));
+        StoreReg(&m_fifo.bFF_Breakpoint, 0);
       }
     }
   }
@@ -469,26 +448,18 @@ void CommandProcessorManager::SetCPStatusFromGPU()
   {
     if (breakpoint)
     {
-      DEBUG_LOG_FMT(COMMANDPROCESSOR, "Cleared breakpoint at {}",
-                    m_fifo.CPReadPointer.load(std::memory_order_relaxed));
-      m_fifo.bFF_Breakpoint = false;
+      DEBUG_LOG_FMT(COMMANDPROCESSOR, "Cleared breakpoint at {}", LoadReg(&m_fifo.CPReadPointer));
+      StoreReg(&m_fifo.bFF_Breakpoint, 0);
     }
   }
 
   // overflow & underflow check
-  m_fifo.bFF_HiWatermark.store(
-      (m_fifo.CPReadWriteDistance.load(std::memory_order_relaxed) > m_fifo.CPHiWatermark),
-      std::memory_order_relaxed);
-  m_fifo.bFF_LoWatermark.store(
-      (m_fifo.CPReadWriteDistance.load(std::memory_order_relaxed) < m_fifo.CPLoWatermark),
-      std::memory_order_relaxed);
+  StoreReg(&m_fifo.bFF_HiWatermark, (LoadReg(&m_fifo.CPReadWriteDistance) > m_fifo.CPHiWatermark));
+  StoreReg(&m_fifo.bFF_LoWatermark, (LoadReg(&m_fifo.CPReadWriteDistance) < m_fifo.CPLoWatermark));
 
-  bool bpInt = m_fifo.bFF_Breakpoint.load(std::memory_order_relaxed) &&
-               m_fifo.bFF_BPInt.load(std::memory_order_relaxed);
-  bool ovfInt = m_fifo.bFF_HiWatermark.load(std::memory_order_relaxed) &&
-                m_fifo.bFF_HiWatermarkInt.load(std::memory_order_relaxed);
-  bool undfInt = m_fifo.bFF_LoWatermark.load(std::memory_order_relaxed) &&
-                 m_fifo.bFF_LoWatermarkInt.load(std::memory_order_relaxed);
+  bool bpInt = LoadReg(&m_fifo.bFF_Breakpoint) && LoadReg(&m_fifo.bFF_BPInt);
+  bool ovfInt = LoadReg(&m_fifo.bFF_HiWatermark) && LoadReg(&m_fifo.bFF_HiWatermarkInt);
+  bool undfInt = LoadReg(&m_fifo.bFF_LoWatermark) && LoadReg(&m_fifo.bFF_LoWatermarkInt);
 
   bool interrupt = (bpInt || ovfInt || undfInt) && m_cp_ctrl_reg.GPReadEnable;
 
@@ -514,19 +485,12 @@ void CommandProcessorManager::SetCPStatusFromGPU()
 void CommandProcessorManager::SetCPStatusFromCPU()
 {
   // overflow & underflow check
-  m_fifo.bFF_HiWatermark.store(
-      (m_fifo.CPReadWriteDistance.load(std::memory_order_relaxed) > m_fifo.CPHiWatermark),
-      std::memory_order_relaxed);
-  m_fifo.bFF_LoWatermark.store(
-      (m_fifo.CPReadWriteDistance.load(std::memory_order_relaxed) < m_fifo.CPLoWatermark),
-      std::memory_order_relaxed);
+  StoreReg(&m_fifo.bFF_HiWatermark, (LoadReg(&m_fifo.CPReadWriteDistance) > m_fifo.CPHiWatermark));
+  StoreReg(&m_fifo.bFF_LoWatermark, (LoadReg(&m_fifo.CPReadWriteDistance) < m_fifo.CPLoWatermark));
 
-  bool bpInt = m_fifo.bFF_Breakpoint.load(std::memory_order_relaxed) &&
-               m_fifo.bFF_BPInt.load(std::memory_order_relaxed);
-  bool ovfInt = m_fifo.bFF_HiWatermark.load(std::memory_order_relaxed) &&
-                m_fifo.bFF_HiWatermarkInt.load(std::memory_order_relaxed);
-  bool undfInt = m_fifo.bFF_LoWatermark.load(std::memory_order_relaxed) &&
-                 m_fifo.bFF_LoWatermarkInt.load(std::memory_order_relaxed);
+  bool bpInt = LoadReg(&m_fifo.bFF_Breakpoint) && LoadReg(&m_fifo.bFF_BPInt);
+  bool ovfInt = LoadReg(&m_fifo.bFF_HiWatermark) && LoadReg(&m_fifo.bFF_HiWatermarkInt);
+  bool undfInt = LoadReg(&m_fifo.bFF_LoWatermark) && LoadReg(&m_fifo.bFF_LoWatermarkInt);
 
   bool interrupt = (bpInt || ovfInt || undfInt) && m_cp_ctrl_reg.GPReadEnable;
 
@@ -552,15 +516,13 @@ void CommandProcessorManager::SetCPStatusFromCPU()
 void CommandProcessorManager::SetCpStatusRegister()
 {
   // Here always there is one fifo attached to the GPU
-  m_cp_status_reg.Breakpoint = m_fifo.bFF_Breakpoint.load(std::memory_order_relaxed);
-  m_cp_status_reg.ReadIdle = !m_fifo.CPReadWriteDistance.load(std::memory_order_relaxed) ||
-                             (m_fifo.CPReadPointer.load(std::memory_order_relaxed) ==
-                              m_fifo.CPWritePointer.load(std::memory_order_relaxed));
-  m_cp_status_reg.CommandIdle = !m_fifo.CPReadWriteDistance.load(std::memory_order_relaxed) ||
-                                Fifo::AtBreakpoint(m_system) ||
-                                !m_fifo.bFF_GPReadEnable.load(std::memory_order_relaxed);
-  m_cp_status_reg.UnderflowLoWatermark = m_fifo.bFF_LoWatermark.load(std::memory_order_relaxed);
-  m_cp_status_reg.OverflowHiWatermark = m_fifo.bFF_HiWatermark.load(std::memory_order_relaxed);
+  m_cp_status_reg.Breakpoint = LoadReg(&m_fifo.bFF_Breakpoint);
+  m_cp_status_reg.ReadIdle = !LoadReg(&m_fifo.CPReadWriteDistance) ||
+                             (LoadReg(&m_fifo.CPReadPointer) == LoadReg(&m_fifo.CPWritePointer));
+  m_cp_status_reg.CommandIdle = !LoadReg(&m_fifo.CPReadWriteDistance) ||
+                                Fifo::AtBreakpoint(m_system) || !LoadReg(&m_fifo.bFF_GPReadEnable);
+  m_cp_status_reg.UnderflowLoWatermark = LoadReg(&m_fifo.bFF_LoWatermark);
+  m_cp_status_reg.OverflowHiWatermark = LoadReg(&m_fifo.bFF_HiWatermark);
 
   DEBUG_LOG_FMT(COMMANDPROCESSOR, "\t Read from STATUS_REGISTER : {:04x}", m_cp_status_reg.Hex);
   DEBUG_LOG_FMT(COMMANDPROCESSOR,
@@ -574,21 +536,21 @@ void CommandProcessorManager::SetCpStatusRegister()
 void CommandProcessorManager::SetCpControlRegister()
 {
   // Just before disabling reads, give the GPU a chance to catch up.
-  if (m_fifo.bFF_GPReadEnable.load(std::memory_order_relaxed) && !m_cp_ctrl_reg.GPReadEnable)
+  if (LoadReg(&m_fifo.bFF_GPReadEnable) && !m_cp_ctrl_reg.GPReadEnable)
   {
     m_system.GetFifo().SyncGPUForRegisterAccess();
   }
-  m_fifo.bFF_GPReadEnable.store(m_cp_ctrl_reg.GPReadEnable, std::memory_order_relaxed);
-  m_fifo.bFF_BPInt.store(m_cp_ctrl_reg.BPInt, std::memory_order_relaxed);
-  m_fifo.bFF_BPEnable.store(m_cp_ctrl_reg.BPEnable, std::memory_order_relaxed);
-  m_fifo.bFF_HiWatermarkInt.store(m_cp_ctrl_reg.FifoOverflowIntEnable, std::memory_order_relaxed);
-  m_fifo.bFF_LoWatermarkInt.store(m_cp_ctrl_reg.FifoUnderflowIntEnable, std::memory_order_relaxed);
-  m_fifo.bFF_GPLinkEnable.store(m_cp_ctrl_reg.GPLinkEnable, std::memory_order_relaxed);
+  StoreReg(&m_fifo.bFF_GPReadEnable, m_cp_ctrl_reg.GPReadEnable);
+  StoreReg(&m_fifo.bFF_BPInt, m_cp_ctrl_reg.BPInt);
+  StoreReg(&m_fifo.bFF_BPEnable, m_cp_ctrl_reg.BPEnable);
+  StoreReg(&m_fifo.bFF_HiWatermarkInt, m_cp_ctrl_reg.FifoOverflowIntEnable);
+  StoreReg(&m_fifo.bFF_LoWatermarkInt, m_cp_ctrl_reg.FifoUnderflowIntEnable);
+  StoreReg(&m_fifo.bFF_GPLinkEnable, m_cp_ctrl_reg.GPLinkEnable);
 
   DEBUG_LOG_FMT(COMMANDPROCESSOR, "\t GPREAD {} | BP {} | Int {} | OvF {} | UndF {} | LINK {}",
-                m_fifo.bFF_GPReadEnable.load(std::memory_order_relaxed) ? "ON" : "OFF",
-                m_fifo.bFF_BPEnable.load(std::memory_order_relaxed) ? "ON" : "OFF",
-                m_fifo.bFF_BPInt.load(std::memory_order_relaxed) ? "ON" : "OFF",
+                LoadReg(&m_fifo.bFF_GPReadEnable) ? "ON" : "OFF",
+                LoadReg(&m_fifo.bFF_BPEnable) ? "ON" : "OFF",
+                LoadReg(&m_fifo.bFF_BPInt) ? "ON" : "OFF",
                 m_cp_ctrl_reg.FifoOverflowIntEnable ? "ON" : "OFF",
                 m_cp_ctrl_reg.FifoUnderflowIntEnable ? "ON" : "OFF",
                 m_cp_ctrl_reg.GPLinkEnable ? "ON" : "OFF");
@@ -647,21 +609,17 @@ void CommandProcessorManager::HandleUnknownOpcode(u8 cmd_byte, const u8* buffer,
       "{:#010x}, bFF_GPReadEnable: {}, bFF_BPEnable: {}, bFF_BPInt: {}, bFF_Breakpoint: "
       "{}, bFF_GPLinkEnable: {}, bFF_HiWatermarkInt: {}, bFF_LoWatermarkInt: {}, "
       "approximate PC: {:08x}, approximate LR: {:08x}",
-      cmd_byte, fmt::ptr(buffer), preprocess ? "yes" : "no",
-      m_fifo.CPBase.load(std::memory_order_relaxed), m_fifo.CPEnd.load(std::memory_order_relaxed),
-      m_fifo.CPHiWatermark, m_fifo.CPLoWatermark,
-      m_fifo.CPReadWriteDistance.load(std::memory_order_relaxed),
-      m_fifo.CPWritePointer.load(std::memory_order_relaxed),
-      m_fifo.CPReadPointer.load(std::memory_order_relaxed),
-      m_fifo.CPBreakpoint.load(std::memory_order_relaxed),
-      m_fifo.bFF_GPReadEnable.load(std::memory_order_relaxed) ? "true" : "false",
-      m_fifo.bFF_BPEnable.load(std::memory_order_relaxed) ? "true" : "false",
-      m_fifo.bFF_BPInt.load(std::memory_order_relaxed) ? "true" : "false",
-      m_fifo.bFF_Breakpoint.load(std::memory_order_relaxed) ? "true" : "false",
-      m_fifo.bFF_GPLinkEnable.load(std::memory_order_relaxed) ? "true" : "false",
-      m_fifo.bFF_HiWatermarkInt.load(std::memory_order_relaxed) ? "true" : "false",
-      m_fifo.bFF_LoWatermarkInt.load(std::memory_order_relaxed) ? "true" : "false", ppc_state.pc,
-      LR(ppc_state));
+      cmd_byte, fmt::ptr(buffer), preprocess ? "yes" : "no", LoadReg(&m_fifo.CPBase),
+      LoadReg(&m_fifo.CPEnd), m_fifo.CPHiWatermark, m_fifo.CPLoWatermark,
+      LoadReg(&m_fifo.CPReadWriteDistance), LoadReg(&m_fifo.CPWritePointer),
+      LoadReg(&m_fifo.CPReadPointer), LoadReg(&m_fifo.CPBreakpoint),
+      LoadReg(&m_fifo.bFF_GPReadEnable) ? "true" : "false",
+      LoadReg(&m_fifo.bFF_BPEnable) ? "true" : "false",
+      LoadReg(&m_fifo.bFF_BPInt) ? "true" : "false",
+      LoadReg(&m_fifo.bFF_Breakpoint) ? "true" : "false",
+      LoadReg(&m_fifo.bFF_GPLinkEnable) ? "true" : "false",
+      LoadReg(&m_fifo.bFF_HiWatermarkInt) ? "true" : "false",
+      LoadReg(&m_fifo.bFF_LoWatermarkInt) ? "true" : "false", ppc_state.pc, LR(ppc_state));
 
   if (!m_is_fifo_error_seen && !suppress_panic_alert)
   {
diff --git a/Source/Core/VideoCommon/CommandProcessor.h b/Source/Core/VideoCommon/CommandProcessor.h
index 976592c829..f23f6852ba 100644
--- a/Source/Core/VideoCommon/CommandProcessor.h
+++ b/Source/Core/VideoCommon/CommandProcessor.h
@@ -3,11 +3,11 @@
 
 #pragma once
 
-#include <atomic>
-
 #include "Common/CommonTypes.h"
 #include "Common/Flag.h"
 
+#include <retro_atomic.h>
+
 class PointerWrap;
 namespace MMIO
 {
@@ -24,30 +24,41 @@ struct EventType;
 
 namespace CommandProcessor
 {
+/* The FIFO registers the CPU and GPU threads share: u32 values in atomic words. */
+inline u32 LoadReg(const retro_atomic_int_t* reg)
+{
+  return static_cast<u32>(retro_atomic_load_relaxed_int(reg));
+}
+
+inline void StoreReg(retro_atomic_int_t* reg, u32 value)
+{
+  retro_atomic_store_relaxed_int(reg, static_cast<int>(value));
+}
+
 struct SCPFifoStruct
 {
   // fifo registers
-  std::atomic<u32> CPBase = 0;
-  std::atomic<u32> CPEnd = 0;
+  retro_atomic_int_t CPBase{};
+  retro_atomic_int_t CPEnd{};
   u32 CPHiWatermark = 0;
   u32 CPLoWatermark = 0;
-  std::atomic<u32> CPReadWriteDistance = 0;
-  std::atomic<u32> CPWritePointer = 0;
-  std::atomic<u32> CPReadPointer = 0;
-  std::atomic<u32> CPBreakpoint = 0;
-  std::atomic<u32> SafeCPReadPointer = 0;
+  retro_atomic_int_t CPReadWriteDistance{};
+  retro_atomic_int_t CPWritePointer{};
+  retro_atomic_int_t CPReadPointer{};
+  retro_atomic_int_t CPBreakpoint{};
+  retro_atomic_int_t SafeCPReadPointer{};
 
-  std::atomic<u32> bFF_GPLinkEnable = 0;
-  std::atomic<u32> bFF_GPReadEnable = 0;
-  std::atomic<u32> bFF_BPEnable = 0;
-  std::atomic<u32> bFF_BPInt = 0;
-  std::atomic<u32> bFF_Breakpoint = 0;
+  retro_atomic_int_t bFF_GPLinkEnable{};
+  retro_atomic_int_t bFF_GPReadEnable{};
+  retro_atomic_int_t bFF_BPEnable{};
+  retro_atomic_int_t bFF_BPInt{};
+  retro_atomic_int_t bFF_Breakpoint{};
 
-  std::atomic<u32> bFF_LoWatermarkInt = 0;
-  std::atomic<u32> bFF_HiWatermarkInt = 0;
+  retro_atomic_int_t bFF_LoWatermarkInt{};
+  retro_atomic_int_t bFF_HiWatermarkInt{};
 
-  std::atomic<u32> bFF_LoWatermark = 0;
-  std::atomic<u32> bFF_HiWatermark = 0;
+  retro_atomic_int_t bFF_LoWatermark{};
+  retro_atomic_int_t bFF_HiWatermark{};
 
   void Init();
   void DoState(PointerWrap& p);
diff --git a/Source/Core/VideoCommon/Fifo.cpp b/Source/Core/VideoCommon/Fifo.cpp
index 75a219c7c2..08df32e8e1 100644
--- a/Source/Core/VideoCommon/Fifo.cpp
+++ b/Source/Core/VideoCommon/Fifo.cpp
@@ -3,7 +3,6 @@
 
 #include "VideoCommon/Fifo.h"
 
-#include <atomic>
 #include <cstring>
 
 #include "Common/Assert.h"
@@ -34,6 +33,9 @@
 
 namespace Fifo
 {
+using CommandProcessor::LoadReg;
+using CommandProcessor::StoreReg;
+
 static constexpr int GPU_TIME_SLOT_SIZE = 1000;
 
 FifoManager::FifoManager(Core::System& system) : m_system{system}
@@ -53,14 +55,15 @@ void FifoManager::RefreshConfig()
 void FifoManager::DoState(PointerWrap& p)
 {
   p.DoArray(m_video_buffer, FIFO_SIZE);
-  u8* write_ptr = m_video_buffer_write_ptr;
+  u8* write_ptr = GetWritePtr();
   p.DoPointer(write_ptr, m_video_buffer);
-  m_video_buffer_write_ptr = write_ptr;
+  SetWritePtr(write_ptr);
   p.DoPointer(m_video_buffer_read_ptr, m_video_buffer);
   if (p.IsReadMode() && m_use_deterministic_gpu_thread)
   {
     // We're good and paused, right?
-    m_video_buffer_seen_ptr = m_video_buffer_pp_read_ptr = m_video_buffer_read_ptr;
+    m_video_buffer_pp_read_ptr = m_video_buffer_read_ptr;
+    SetSeenPtr(m_video_buffer_read_ptr);
   }
 
   p.Do(m_sync_ticks);
@@ -95,7 +98,7 @@ void FifoManager::Init()
   ResetVideoBuffer();
   if (m_system.IsDualCoreMode())
     m_gpu_mainloop.Prepare();
-  m_sync_ticks.store(0);
+  retro_atomic_store_release_int(&m_sync_ticks, 0);
 }
 
 void FifoManager::Shutdown()
@@ -105,10 +108,10 @@ void FifoManager::Shutdown()
 
   Common::FreeMemoryPages(m_video_buffer, FIFO_SIZE + 4);
   m_video_buffer = nullptr;
-  m_video_buffer_write_ptr = nullptr;
+  SetWritePtr(nullptr);
   m_video_buffer_pp_read_ptr = nullptr;
   m_video_buffer_read_ptr = nullptr;
-  m_video_buffer_seen_ptr = nullptr;
+  SetSeenPtr(nullptr);
   m_fifo_aux_write_ptr = nullptr;
   m_fifo_aux_read_ptr = nullptr;
 
@@ -127,7 +130,7 @@ void FifoManager::ExitGpuLoop()
   auto& fifo = command_processor.GetFifo();
 
   // This should break the wait loop in CPU thread
-  fifo.bFF_GPReadEnable.store(0, std::memory_order_relaxed);
+  StoreReg(&fifo.bFF_GPReadEnable, 0);
   FlushGpu();
 
   // Terminate GPU thread loop
@@ -141,7 +144,7 @@ void FifoManager::ExitGpuLoop()
 // and this Stop() does nothing, so the request is kept for RunGpuLoop() to act on.
 void FifoManager::StopGpuLoop()
 {
-  m_stop_gpu_loop_requested.store(true);
+  retro_atomic_store_release_int(&m_stop_gpu_loop_requested, 1);
   m_gpu_mainloop.Stop(Common::BlockingLoop::StopMode::NonBlock);
 }
 #endif
@@ -176,7 +179,7 @@ void FifoManager::SyncGPU(SyncGPUReason reason, bool may_move_read_ptr)
 
     if (may_move_read_ptr)
     {
-      u8* write_ptr = m_video_buffer_write_ptr;
+      u8* write_ptr = GetWritePtr();
 
       // what's left over in the buffer
       size_t size = write_ptr - m_video_buffer_pp_read_ptr;
@@ -185,10 +188,11 @@ void FifoManager::SyncGPU(SyncGPUReason reason, bool may_move_read_ptr)
       // This change always decreases the pointers.  We write seen_ptr
       // after write_ptr here, and read it before in RunGpuLoop, so
       // 'write_ptr > seen_ptr' there cannot become spuriously true.
-      m_video_buffer_write_ptr = write_ptr = m_video_buffer + size;
+      write_ptr = m_video_buffer + size;
+      SetWritePtr(write_ptr);
       m_video_buffer_pp_read_ptr = m_video_buffer;
       m_video_buffer_read_ptr = m_video_buffer;
-      m_video_buffer_seen_ptr = write_ptr;
+      SetSeenPtr(write_ptr);
     }
   }
 }
@@ -225,10 +229,10 @@ void* FifoManager::PopFifoAuxBuffer(size_t size)
 // Description: RunGpuLoop() sends data through this function.
 void FifoManager::ReadDataFromFifo(u32 read_ptr)
 {
-  if (GPFifo::GATHER_PIPE_SIZE >
-      static_cast<size_t>(m_video_buffer + FIFO_SIZE - m_video_buffer_write_ptr))
+  u8* write_ptr = GetWritePtr();
+  if (GPFifo::GATHER_PIPE_SIZE > static_cast<size_t>(m_video_buffer + FIFO_SIZE - write_ptr))
   {
-    const size_t existing_len = m_video_buffer_write_ptr - m_video_buffer_read_ptr;
+    const size_t existing_len = write_ptr - m_video_buffer_read_ptr;
     if (GPFifo::GATHER_PIPE_SIZE > static_cast<size_t>(FIFO_SIZE - existing_len))
     {
       PanicAlertFmt("FIFO out of bounds (existing {} + new {} > {})", existing_len,
@@ -236,19 +240,19 @@ void FifoManager::ReadDataFromFifo(u32 read_ptr)
       return;
     }
     memmove(m_video_buffer, m_video_buffer_read_ptr, existing_len);
-    m_video_buffer_write_ptr = m_video_buffer + existing_len;
+    write_ptr = m_video_buffer + existing_len;
     m_video_buffer_read_ptr = m_video_buffer;
   }
   // Copy new video instructions to m_video_buffer for future use in rendering the new picture
   auto& memory = m_system.GetMemory();
-  memory.CopyFromEmu(m_video_buffer_write_ptr, read_ptr, GPFifo::GATHER_PIPE_SIZE);
-  m_video_buffer_write_ptr += GPFifo::GATHER_PIPE_SIZE;
+  memory.CopyFromEmu(write_ptr, read_ptr, GPFifo::GATHER_PIPE_SIZE);
+  SetWritePtr(write_ptr + GPFifo::GATHER_PIPE_SIZE);
 }
 
 // The deterministic_gpu_thread version.
 void FifoManager::ReadDataFromFifoOnCPU(u32 read_ptr)
 {
-  u8* write_ptr = m_video_buffer_write_ptr;
+  u8* write_ptr = GetWritePtr();
   if (GPFifo::GATHER_PIPE_SIZE > static_cast<size_t>(m_video_buffer + FIFO_SIZE - write_ptr))
   {
     // We can't wrap around while the GPU is working on the data.
@@ -265,7 +269,7 @@ void FifoManager::ReadDataFromFifoOnCPU(u32 read_ptr)
       PanicAlertFmt("Desynced read pointers");
       return;
     }
-    write_ptr = m_video_buffer_write_ptr;
+    write_ptr = GetWritePtr();
     const size_t existing_len = write_ptr - m_video_buffer_pp_read_ptr;
     if (GPFifo::GATHER_PIPE_SIZE > static_cast<size_t>(FIFO_SIZE - existing_len))
     {
@@ -275,18 +279,18 @@ void FifoManager::ReadDataFromFifoOnCPU(u32 read_ptr)
     }
   }
   auto& memory = m_system.GetMemory();
-  memory.CopyFromEmu(m_video_buffer_write_ptr, read_ptr, GPFifo::GATHER_PIPE_SIZE);
+  memory.CopyFromEmu(write_ptr, read_ptr, GPFifo::GATHER_PIPE_SIZE);
   m_video_buffer_pp_read_ptr = OpcodeDecoder::RunFifo<true>(
       DataReader(m_video_buffer_pp_read_ptr, write_ptr + GPFifo::GATHER_PIPE_SIZE), nullptr);
   // This would have to be locked if the GPU thread didn't spin.
-  m_video_buffer_write_ptr = write_ptr + GPFifo::GATHER_PIPE_SIZE;
+  SetWritePtr(write_ptr + GPFifo::GATHER_PIPE_SIZE);
 }
 
 void FifoManager::ResetVideoBuffer()
 {
   m_video_buffer_read_ptr = m_video_buffer;
-  m_video_buffer_write_ptr = m_video_buffer;
-  m_video_buffer_seen_ptr = m_video_buffer;
+  SetWritePtr(m_video_buffer);
+  SetSeenPtr(m_video_buffer);
   m_video_buffer_pp_read_ptr = m_video_buffer;
   m_fifo_aux_write_ptr = m_fifo_aux_data;
   m_fifo_aux_read_ptr = m_fifo_aux_data;
@@ -299,7 +303,7 @@ void FifoManager::RunGpuLoop()
   m_gpu_mainloop.Run(
       [this] {
 #ifdef __LIBRETRO__
-        if (m_stop_gpu_loop_requested.load())
+        if (retro_atomic_load_acquire_int(&m_stop_gpu_loop_requested))
           m_gpu_mainloop.Stop(Common::BlockingLoop::StopMode::NonBlock);
 #endif
 
@@ -313,14 +317,14 @@ void FifoManager::RunGpuLoop()
         if (m_use_deterministic_gpu_thread)
         {
           // All the fifo/CP stuff is on the CPU.  We just need to run the opcode decoder.
-          u8* seen_ptr = m_video_buffer_seen_ptr;
-          u8* write_ptr = m_video_buffer_write_ptr;
+          u8* seen_ptr = GetSeenPtr();
+          u8* write_ptr = GetWritePtr();
           // See comment in SyncGPU
           if (write_ptr > seen_ptr)
           {
             m_video_buffer_read_ptr =
                 OpcodeDecoder::RunFifo(DataReader(m_video_buffer_read_ptr, write_ptr), nullptr);
-            m_video_buffer_seen_ptr = write_ptr;
+            SetSeenPtr(write_ptr);
           }
         }
         else
@@ -330,41 +334,39 @@ void FifoManager::RunGpuLoop()
           command_processor.SetCPStatusFromGPU();
 
           // check if we are able to run this buffer
-          while (!command_processor.IsInterruptWaiting() &&
-                 fifo.bFF_GPReadEnable.load(std::memory_order_relaxed) &&
-                 fifo.CPReadWriteDistance.load(std::memory_order_relaxed) &&
-                 !AtBreakpoint(m_system))
+          while (!command_processor.IsInterruptWaiting() && LoadReg(&fifo.bFF_GPReadEnable) &&
+                 LoadReg(&fifo.CPReadWriteDistance) && !AtBreakpoint(m_system))
           {
-            if (m_config_sync_gpu && m_sync_ticks.load() < m_config_sync_gpu_min_distance)
+            if (m_config_sync_gpu &&
+                retro_atomic_load_seq_cst_int(&m_sync_ticks) < m_config_sync_gpu_min_distance)
               break;
 
             u32 cyclesExecuted = 0;
-            u32 readPtr = fifo.CPReadPointer.load(std::memory_order_relaxed);
+            u32 readPtr = LoadReg(&fifo.CPReadPointer);
             ReadDataFromFifo(readPtr);
 
-            if (readPtr == fifo.CPEnd.load(std::memory_order_relaxed))
-              readPtr = fifo.CPBase.load(std::memory_order_relaxed);
+            if (readPtr == LoadReg(&fifo.CPEnd))
+              readPtr = LoadReg(&fifo.CPBase);
             else
               readPtr += GPFifo::GATHER_PIPE_SIZE;
 
             const s32 distance =
-                static_cast<s32>(fifo.CPReadWriteDistance.load(std::memory_order_relaxed)) -
-                GPFifo::GATHER_PIPE_SIZE;
+                static_cast<s32>(LoadReg(&fifo.CPReadWriteDistance)) - GPFifo::GATHER_PIPE_SIZE;
             ASSERT_MSG(COMMANDPROCESSOR, distance >= 0,
                        "Negative fifo.CPReadWriteDistance = {} in FIFO Loop !\nThat can produce "
                        "instability in the game. Please report it.",
                        distance);
 
-            u8* write_ptr = m_video_buffer_write_ptr;
+            u8* write_ptr = GetWritePtr();
             m_video_buffer_read_ptr = OpcodeDecoder::RunFifo(
                 DataReader(m_video_buffer_read_ptr, write_ptr), &cyclesExecuted);
 
-            fifo.CPReadPointer.store(readPtr, std::memory_order_relaxed);
-            fifo.CPReadWriteDistance.fetch_sub(GPFifo::GATHER_PIPE_SIZE, std::memory_order_seq_cst);
+            StoreReg(&fifo.CPReadPointer, readPtr);
+            retro_atomic_fetch_add_seq_cst_int(&fifo.CPReadWriteDistance,
+                                               -static_cast<int>(GPFifo::GATHER_PIPE_SIZE));
             if ((write_ptr - m_video_buffer_read_ptr) == 0)
             {
-              fifo.SafeCPReadPointer.store(fifo.CPReadPointer.load(std::memory_order_relaxed),
-                                           std::memory_order_relaxed);
+              StoreReg(&fifo.SafeCPReadPointer, LoadReg(&fifo.CPReadPointer));
             }
 
             command_processor.SetCPStatusFromGPU();
@@ -372,7 +374,8 @@ void FifoManager::RunGpuLoop()
             if (m_config_sync_gpu)
             {
               cyclesExecuted = (int)(cyclesExecuted / m_config_sync_gpu_overclock);
-              int old = m_sync_ticks.fetch_sub(cyclesExecuted);
+              int old = retro_atomic_fetch_add_seq_cst_int(&m_sync_ticks,
+                                                           -static_cast<int>(cyclesExecuted));
               if (old >= m_config_sync_gpu_max_distance &&
                   old - (int)cyclesExecuted < m_config_sync_gpu_max_distance)
               {
@@ -388,9 +391,9 @@ void FifoManager::RunGpuLoop()
           }
 
           // fast skip remaining GPU time if fifo is empty
-          if (m_sync_ticks.load() > 0)
+          if (retro_atomic_load_seq_cst_int(&m_sync_ticks) > 0)
           {
-            int old = m_sync_ticks.exchange(0);
+            int old = retro_atomic_exchange_int(&m_sync_ticks, 0);
             if (old >= m_config_sync_gpu_max_distance)
               m_sync_wakeup_event.Set();
           }
@@ -402,7 +405,7 @@ void FifoManager::RunGpuLoop()
         }
       });
 #ifdef __LIBRETRO__
-  m_stop_gpu_loop_requested.store(false);
+  retro_atomic_store_release_int(&m_stop_gpu_loop_requested, 0);
 #endif
 }
 
@@ -423,9 +426,8 @@ bool AtBreakpoint(Core::System& system)
 {
   auto& command_processor = system.GetCommandProcessor();
   const auto& fifo = command_processor.GetFifo();
-  return fifo.bFF_BPEnable.load(std::memory_order_relaxed) &&
-         (fifo.CPReadPointer.load(std::memory_order_relaxed) ==
-          fifo.CPBreakpoint.load(std::memory_order_relaxed));
+  return LoadReg(&fifo.bFF_BPEnable) &&
+         (LoadReg(&fifo.CPReadPointer) == LoadReg(&fifo.CPBreakpoint));
 }
 
 void FifoManager::RunGpu()
@@ -455,14 +457,14 @@ int FifoManager::RunGpuOnCpu(int ticks)
   auto& command_processor = m_system.GetCommandProcessor();
   auto& fifo = command_processor.GetFifo();
   bool reset_simd_state = false;
-  int available_ticks = int(ticks * m_config_sync_gpu_overclock) + m_sync_ticks.load();
-  while (fifo.bFF_GPReadEnable.load(std::memory_order_relaxed) &&
-         fifo.CPReadWriteDistance.load(std::memory_order_relaxed) && !AtBreakpoint(m_system) &&
-         available_ticks >= 0)
+  int available_ticks =
+      int(ticks * m_config_sync_gpu_overclock) + retro_atomic_load_seq_cst_int(&m_sync_ticks);
+  while (LoadReg(&fifo.bFF_GPReadEnable) && LoadReg(&fifo.CPReadWriteDistance) &&
+         !AtBreakpoint(m_system) && available_ticks >= 0)
   {
     if (m_use_deterministic_gpu_thread)
     {
-      ReadDataFromFifoOnCPU(fifo.CPReadPointer.load(std::memory_order_relaxed));
+      ReadDataFromFifoOnCPU(LoadReg(&fifo.CPReadPointer));
       m_gpu_mainloop.Wakeup();
     }
     else
@@ -473,25 +475,23 @@ int FifoManager::RunGpuOnCpu(int ticks)
         Common::FPU::LoadDefaultSIMDState();
         reset_simd_state = true;
       }
-      ReadDataFromFifo(fifo.CPReadPointer.load(std::memory_order_relaxed));
+      ReadDataFromFifo(LoadReg(&fifo.CPReadPointer));
       u32 cycles = 0;
-      m_video_buffer_read_ptr = OpcodeDecoder::RunFifo(
-          DataReader(m_video_buffer_read_ptr, m_video_buffer_write_ptr), &cycles);
+      m_video_buffer_read_ptr =
+          OpcodeDecoder::RunFifo(DataReader(m_video_buffer_read_ptr, GetWritePtr()), &cycles);
       available_ticks -= cycles;
     }
 
-    if (fifo.CPReadPointer.load(std::memory_order_relaxed) ==
-        fifo.CPEnd.load(std::memory_order_relaxed))
+    if (LoadReg(&fifo.CPReadPointer) == LoadReg(&fifo.CPEnd))
     {
-      fifo.CPReadPointer.store(fifo.CPBase.load(std::memory_order_relaxed),
-                               std::memory_order_relaxed);
+      StoreReg(&fifo.CPReadPointer, LoadReg(&fifo.CPBase));
     }
     else
     {
-      fifo.CPReadPointer.fetch_add(GPFifo::GATHER_PIPE_SIZE, std::memory_order_relaxed);
+      retro_atomic_fetch_add_int(&fifo.CPReadPointer, GPFifo::GATHER_PIPE_SIZE);
     }
 
-    fifo.CPReadWriteDistance.fetch_sub(GPFifo::GATHER_PIPE_SIZE, std::memory_order_relaxed);
+    retro_atomic_fetch_sub_int(&fifo.CPReadWriteDistance, GPFifo::GATHER_PIPE_SIZE);
   }
 
   command_processor.SetCPStatusFromGPU();
@@ -502,7 +502,7 @@ int FifoManager::RunGpuOnCpu(int ticks)
   }
 
   // Discard all available ticks as there is nothing to do any more.
-  m_sync_ticks.store(std::min(available_ticks, 0));
+  retro_atomic_store_release_int(&m_sync_ticks, std::min(available_ticks, 0));
 
   // If the GPU is idle, drop the handler.
   if (available_ticks >= 0)
@@ -538,7 +538,8 @@ void FifoManager::UpdateWantDeterminism(bool want)
     if (gpu_thread)
     {
       // These haven't been updated in non-deterministic mode.
-      m_video_buffer_seen_ptr = m_video_buffer_pp_read_ptr = m_video_buffer_read_ptr;
+      m_video_buffer_pp_read_ptr = m_video_buffer_read_ptr;
+      SetSeenPtr(m_video_buffer_read_ptr);
       CopyPreprocessCPStateFromMain();
       VertexLoaderManager::MarkAllDirty();
     }
@@ -552,7 +553,7 @@ void FifoManager::UpdateWantDeterminism(bool want)
  */
 int FifoManager::WaitForGpuThread(int ticks)
 {
-  int old = m_sync_ticks.fetch_add(ticks);
+  int old = retro_atomic_fetch_add_seq_cst_int(&m_sync_ticks, ticks);
   int now = old + ticks;
 
   // GPU is idle, so stop polling.
diff --git a/Source/Core/VideoCommon/Fifo.h b/Source/Core/VideoCommon/Fifo.h
index cff4e7980b..7064660a69 100644
--- a/Source/Core/VideoCommon/Fifo.h
+++ b/Source/Core/VideoCommon/Fifo.h
@@ -3,9 +3,9 @@
 
 #pragma once
 
-#include <atomic>
 #include <cstddef>
 #include <optional>
+#include <retro_atomic.h>
 
 #include "Common/BlockingLoop.h"
 #include "Common/CommonTypes.h"
@@ -92,7 +92,7 @@ private:
   Common::BlockingLoop m_gpu_mainloop;
 #ifdef __LIBRETRO__
   // Set by StopGpuLoop(), so that a stop requested before RunGpuLoop() got going still ends it.
-  std::atomic<bool> m_stop_gpu_loop_requested{false};
+  retro_atomic_int_t m_stop_gpu_loop_requested{};
 #endif
 
   Common::Flag m_emu_running_state;
@@ -111,8 +111,8 @@ private:
   // STATE_TO_SAVE
   u8* m_video_buffer = nullptr;
   u8* m_video_buffer_read_ptr = nullptr;
-  std::atomic<u8*> m_video_buffer_write_ptr = nullptr;
-  std::atomic<u8*> m_video_buffer_seen_ptr = nullptr;
+  retro_atomic_ptr_t m_video_buffer_write_ptr{};
+  retro_atomic_ptr_t m_video_buffer_seen_ptr{};
   u8* m_video_buffer_pp_read_ptr = nullptr;
   // The read_ptr is always owned by the GPU thread.  In normal mode, so is the
   // write_ptr, despite it being atomic.  In deterministic GPU thread mode,
@@ -126,7 +126,18 @@ private:
   // polls, it's just atomic.
   // - The pp_read_ptr is the CPU preprocessing version of the read_ptr.
 
-  std::atomic<int> m_sync_ticks = 0;
+  retro_atomic_int_t m_sync_ticks{};
+
+  u8* GetWritePtr() const
+  {
+    return static_cast<u8*>(retro_atomic_load_acquire_ptr(&m_video_buffer_write_ptr));
+  }
+  void SetWritePtr(u8* ptr) { retro_atomic_store_release_ptr(&m_video_buffer_write_ptr, ptr); }
+  u8* GetSeenPtr() const
+  {
+    return static_cast<u8*>(retro_atomic_load_acquire_ptr(&m_video_buffer_seen_ptr));
+  }
+  void SetSeenPtr(u8* ptr) { retro_atomic_store_release_ptr(&m_video_buffer_seen_ptr, ptr); }
   bool m_syncing_suspended = false;
   Common::Event m_sync_wakeup_event;
 
diff --git a/Source/Core/VideoCommon/PerfQueryBase.h b/Source/Core/VideoCommon/PerfQueryBase.h
index 183a27c80c..9afd6b6666 100644
--- a/Source/Core/VideoCommon/PerfQueryBase.h
+++ b/Source/Core/VideoCommon/PerfQueryBase.h
@@ -3,10 +3,10 @@
 
 #pragma once
 
-#include <array>
-#include <atomic>
 #include <memory>
 
+#include <retro_atomic.h>
+
 #include "Common/CommonTypes.h"
 
 enum PerfQueryType
@@ -31,7 +31,7 @@ enum PerfQueryGroup
 class PerfQueryBase
 {
 public:
-  PerfQueryBase() : m_query_count(0) {}
+  PerfQueryBase() {}
   virtual ~PerfQueryBase() {}
 
   virtual bool Initialize() { return true; }
@@ -65,8 +65,45 @@ public:
   virtual bool IsFlushed() const { return true; }
 
 protected:
-  std::atomic<u32> m_query_count;
-  std::array<std::atomic<u32>, PQG_NUM_MEMBERS> m_results;
+  /* The GPU thread retires queries while the CPU thread reads results: u32s in atomic words. */
+  u32 QueryCount() const { return static_cast<u32>(retro_atomic_load_acquire_int(&m_query_count)); }
+  void AddQueryCount(u32 count)
+  {
+    retro_atomic_fetch_add_int(&m_query_count, static_cast<int>(count));
+  }
+  void SubQueryCount(u32 count)
+  {
+    retro_atomic_fetch_sub_int(&m_query_count, static_cast<int>(count));
+  }
+  u32 QueryResult(PerfQueryGroup group) const
+  {
+    return static_cast<u32>(retro_atomic_load_acquire_int(&m_results[group]));
+  }
+  void AddQueryResult(PerfQueryGroup group, u32 value)
+  {
+    retro_atomic_fetch_add_int(&m_results[group], static_cast<int>(value));
+  }
+  void SetQueryResult(PerfQueryGroup group, u32 value)
+  {
+    retro_atomic_store_release_int(&m_results[group], static_cast<int>(value));
+  }
+  /* Drops the pending queries and zeroes the results. */
+  void ResetQueryState()
+  {
+    int i;
+    retro_atomic_store_release_int(&m_query_count, 0);
+    for (i = 0; i < PQG_NUM_MEMBERS; i++)
+      retro_atomic_store_release_int(&m_results[i], 0);
+  }
+  void ResetQueryResults()
+  {
+    int i;
+    for (i = 0; i < PQG_NUM_MEMBERS; i++)
+      retro_atomic_store_release_int(&m_results[i], 0);
+  }
+
+  retro_atomic_int_t m_query_count{};
+  retro_atomic_int_t m_results[PQG_NUM_MEMBERS]{};
 };
 
 extern std::unique_ptr<PerfQueryBase> g_perf_query;
diff --git a/Source/Core/VideoCommon/Present.cpp b/Source/Core/VideoCommon/Present.cpp
index 13ff604077..b41e778204 100644
--- a/Source/Core/VideoCommon/Present.cpp
+++ b/Source/Core/VideoCommon/Present.cpp
@@ -107,7 +107,7 @@ Presenter::Presenter()
       video_events.config_changed_event.Register([this](u32 bits) { ConfigChanged(bits); });
 
   m_end_field_hook = video_events.vi_end_field_event.Register(
-      [this] { m_immediate_swap_happened_this_field.store(false, std::memory_order_relaxed); });
+      [this] { retro_atomic_store_relaxed_int(&m_immediate_swap_happened_this_field, 0); });
 }
 
 Presenter::~Presenter()
@@ -120,7 +120,7 @@ bool Presenter::Initialize()
 {
   UpdateDrawRectangle();
 
-  m_immediate_swap_happened_this_field.store(false, std::memory_order_relaxed);
+  retro_atomic_store_relaxed_int(&m_immediate_swap_happened_this_field, 0);
 
   if (!g_gfx->IsHeadless())
   {
@@ -237,7 +237,7 @@ void Presenter::ViSwap(u32 xfb_addr, u32 fb_width, u32 fb_stride, u32 fb_height,
 
 void Presenter::ImmediateSwap(u32 xfb_addr, u32 fb_width, u32 fb_stride, u32 fb_height)
 {
-  if (m_immediate_swap_happened_this_field.exchange(true, std::memory_order_relaxed) &&
+  if (retro_atomic_exchange_int(&m_immediate_swap_happened_this_field, 1) &&
       Config::Get(Config::GFX_HACK_CAP_IMMEDIATE_XFB))
   {
     return;
@@ -959,7 +959,7 @@ void Presenter::DoState(PointerWrap& p)
 
     m_next_swap_estimated_ticks = m_last_xfb_ticks;
 
-    m_immediate_swap_happened_this_field.store(false, std::memory_order_relaxed);
+    retro_atomic_store_relaxed_int(&m_immediate_swap_happened_this_field, 0);
 
 #ifdef __LIBRETRO__
     if (g_video_backend && g_video_backend->GetConfigName() != "OGL")
diff --git a/Source/Core/VideoCommon/Present.h b/Source/Core/VideoCommon/Present.h
index 0b92a9acdf..a4d13be095 100644
--- a/Source/Core/VideoCommon/Present.h
+++ b/Source/Core/VideoCommon/Present.h
@@ -13,6 +13,8 @@
 #include <memory>
 #include <tuple>
 
+#include <retro_atomic.h>
+
 class AbstractTexture;
 struct SurfaceInfo;
 enum class AspectMode : int;
@@ -165,7 +167,7 @@ private:
   // Can be used for presentation of ImmediateXFB swaps which don't have timing information.
   u64 m_next_swap_estimated_ticks = 0;
 
-  std::atomic_bool m_immediate_swap_happened_this_field{};
+  retro_atomic_int_t m_immediate_swap_happened_this_field{};
 };
 
 }  // namespace VideoCommon