2026-09-25 09:06:19

Signed-off-by: lizzie <lizzie@eden-emu.dev>

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt

@@ -30,6 +30,7 @@ enum class BooleanSetting(override val key: String) : AbstractBooleanSetting {
     RENDERER_REACTIVE_FLUSHING("use_reactive_flushing"),
     ENABLE_BUFFER_HISTORY("enable_buffer_history"),
     USE_OPTIMIZED_VERTEX_BUFFERS("use_optimized_vertex_buffers"),
+    ENABLE_GPU_BUFFER_READBACK("enable_gpu_buffer_readback"),
     SYNC_MEMORY_OPERATIONS("sync_memory_operations"),
     BUFFER_REORDER_DISABLE("disable_buffer_reorder"),
     RENDERER_DEBUG("debug"),

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt

@@ -908,6 +908,13 @@ abstract class SettingsItem(
                     descriptionId = R.string.enable_buffer_history_description
                 )
             )
+            put(
+                SwitchSetting(
+                    BooleanSetting.ENABLE_GPU_BUFFER_READBACK,
+                    titleId = R.string.enable_gpu_buffer_readback,
+                    descriptionId = R.string.enable_gpu_buffer_readback_description
+                )
+            )
             put(
                 SwitchSetting(
                     BooleanSetting.USE_OPTIMIZED_VERTEX_BUFFERS,

src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt

@@ -549,6 +549,7 @@ class SettingsFragmentPresenter(
             add(BooleanSetting.RENDERER_FORCE_MAX_CLOCK.key)
             add(BooleanSetting.RENDERER_REACTIVE_FLUSHING.key)
             add(BooleanSetting.ENABLE_BUFFER_HISTORY.key)
+            add(BooleanSetting.ENABLE_GPU_BUFFER_READBACK.key)
             add(BooleanSetting.USE_OPTIMIZED_VERTEX_BUFFERS.key)
 
             add(HeaderSetting(R.string.hacks))

src/android/app/src/main/res/values-ar/strings.xml

@@ -580,6 +580,8 @@
     <string name="renderer_reactive_flushing_description">يحسن دقة العرض في بعض الألعاب على حساب الأداء.</string>
     <string name="enable_buffer_history">تمكين سجل التخزين المؤقت</string>
     <string name="enable_buffer_history_description">يُتيح هذا الخيار الوصول إلى حالات التخزين المؤقت السابقة. وقد يُحسّن جودة العرض وثبات الأداء في بعض الألعاب.</string>
+    <string name="enable_gpu_buffer_readback">تفعيل قراءة مخزن وحدة معالجة الرسومات</string>
+    <string name="enable_gpu_buffer_readback_description">يحافظ هذا النظام على بيانات المخزن المؤقت المُعدّلة بواسطة وحدة معالجة الرسومات عن طريق قراءتها مرة أخرى قبل التحميل. تتطلب بعض الألعاب ذلك لعرض بعض التأثيرات بشكل صحيح. قد يُسبب ذلك مشاكل إذا لم يتمكن الجهاز من التعامل مع عبء العمل الإضافي.</string>
     <string name="use_optimized_vertex_buffers">مخازن الرؤوس المُحسّنة</string>
     <string name="use_optimized_vertex_buffers_description">يُتيح ربطًا مُحسَّنًا لمخازن الرؤوس لتحسين الأداء. يتطلب برامج تشغيل Mesa 26.0+ Turnip/ برامج تشغيل QCOM. قد يتعطل على برامج تشغيل Turnip القديمة (25.3 وما دون).</string>
 
@@ -1097,6 +1099,7 @@
     <string name="gpu_fence_behavior_immediate">فوري</string>
     <string name="gpu_fence_behavior_balanced">متوازن</string>
     <string name="gpu_fence_behavior_accurate">دقيق</string>
+    <string name="gpu_fence_behavior_strict">صارم</string>
 
     <string name="vram_usage_conservative">محافظ</string>
     <string name="vram_usage_aggressive">عدواني</string>

src/android/app/src/main/res/values-de/strings.xml

@@ -967,6 +967,7 @@ Wirklich fortfahren?</string>
     <string name="gpu_fence_behavior_immediate">Direkt</string>
     <string name="gpu_fence_behavior_balanced">Ausgewogen</string>
     <string name="gpu_fence_behavior_accurate">Genau</string>
+    <string name="gpu_fence_behavior_strict">Strikt</string>
 
     <string name="vram_usage_conservative">Konservativ</string>
     <string name="vram_usage_aggressive">Aggressiv</string>

src/android/app/src/main/res/values-es/strings.xml

@@ -524,6 +524,8 @@
     <string name="renderer_reactive_flushing_description">Mejora la precisión de renderizado en algunos juegos, pero reduce el rendimiento.</string>
     <string name="enable_buffer_history">Activar el historial del búfer</string>
     <string name="enable_buffer_history_description">Permite el acceso al estado del búfer anterior. Esta opción puede mejorar la calidad de renderizado y la consistencia en el rendimiento de algunos juegos.</string>
+    <string name="enable_gpu_buffer_readback">Activar la lectura del buffer de la GPU</string>
+    <string name="enable_gpu_buffer_readback_description">Conserva los datos del búfer modificados por la GPU leyéndolos antes de subirlos.\nAlgunos juegos requieren esto para renderizar correctamente ciertos efectos.\nPuede causar problemas si el hardware no puede soportar la carga de trabajo adicional.</string>
     <string name="use_optimized_vertex_buffers">Búferes de vértices optimizados</string>
     <string name="use_optimized_vertex_buffers_description">Permite la optimización del enlace del búfer de vértices para un mejor rendimiento. Requiere controladores Mesa 26.0+ Turnip/ controladores QCOM. Fallará con controladores Turnip más antiguos (versión 25.3 o inferior).</string>
 
@@ -1036,6 +1038,7 @@
     <string name="gpu_fence_behavior_immediate">Inmediato</string>
     <string name="gpu_fence_behavior_balanced">Equilibrado</string>
     <string name="gpu_fence_behavior_accurate">Preciso</string>
+    <string name="gpu_fence_behavior_strict">Estricto</string>
 
     <string name="vram_usage_conservative">Conservador</string>
     <string name="vram_usage_aggressive">Agresivo</string>

src/android/app/src/main/res/values-ru/strings.xml

@@ -569,6 +569,8 @@
     <string name="renderer_reactive_flushing_description">Повышение точности рендеринга в некоторых играх за счет снижения производительности.</string>
     <string name="enable_buffer_history">Включить историю буфера</string>
     <string name="enable_buffer_history_description">Позволяет обращаться к предыдущим состояниям буфера. Эта опция может повысить качество рендеринга и стабильность производительности в некоторых играх.</string>
+    <string name="enable_gpu_buffer_readback">Включить обратное чтение буфера ГПУ</string>
+    <string name="enable_gpu_buffer_readback_description">Сохраняет измененные ГПУ данные буфера путем чтения их обратно перед выгрузками. Некоторые игры требуют этого, чтобы рендерить определенные эффекты правильно. Может вызывать проблемы если оборудование не может обработать дополнительную рабочую нагрузку.</string>
     <string name="use_optimized_vertex_buffers">Оптимизированные вершинные буферы</string>
     <string name="use_optimized_vertex_buffers_description">Включает оптимизированную привязку вершинного буфера для повышения производительности. Требует Mesa Turnip 26.0+ / QCOM. Приводит к вылету на старых версиях драйверов Turnip (25.3 и ниже).</string>
 
@@ -1086,6 +1088,7 @@
     <string name="gpu_fence_behavior_immediate">Мгновенный</string>
     <string name="gpu_fence_behavior_balanced">Сбалансированный</string>
     <string name="gpu_fence_behavior_accurate">Точный</string>
+    <string name="gpu_fence_behavior_strict">Строгий</string>
 
     <string name="vram_usage_conservative">Консервативный</string>
     <string name="vram_usage_aggressive">Агрессивный</string>

src/android/app/src/main/res/values-zh-rCN/strings.xml

@@ -570,6 +570,8 @@
     <string name="renderer_reactive_flushing_description">通过牺牲性能来提升某些游戏的渲染精度。</string>
     <string name="enable_buffer_history">启用缓冲区历史</string>
     <string name="enable_buffer_history_description">启用对先前缓冲区状态的访问。此选项可在某些游戏中提升渲染质量并保持性能的一致性。</string>
+    <string name="enable_gpu_buffer_readback">启用 GPU 缓冲区回读</string>
+    <string name="enable_gpu_buffer_readback_description">在上传前回读经由 GPU 修改过的缓冲区数据，以将其保留。一些游戏会用到这项设定以正确渲染某些效果。如果硬件无法处理额外的工作负载，则可能会导致问题。</string>
     <string name="use_optimized_vertex_buffers">优化顶点缓冲区</string>
     <string name="use_optimized_vertex_buffers_description">启用经过优化的顶点缓冲区绑定以提升性能。需要 Mesa 26.0 及以上版本的 Turnip 或 QCOM 驱动程序。若使用较旧版本的 Turnip 驱动 (25.3 及以下版本) 则会导致崩溃。</string>
 
@@ -1087,6 +1089,7 @@
     <string name="gpu_fence_behavior_immediate">即时</string>
     <string name="gpu_fence_behavior_balanced">均衡</string>
     <string name="gpu_fence_behavior_accurate">精确</string>
+    <string name="gpu_fence_behavior_strict">严格</string>
 
     <string name="vram_usage_conservative">保守式</string>
     <string name="vram_usage_aggressive">主动式</string>

src/android/app/src/main/res/values-zh-rTW/strings.xml

@@ -561,6 +561,8 @@
     <string name="renderer_reactive_flushing_description">犧牲效能，以改善部分遊戲的轉譯準確度</string>
     <string name="enable_buffer_history">啟用緩衝區歷史</string>
     <string name="enable_buffer_history_description">允許存取先前的緩衝區狀態。此選項可能會改善部分遊戲的渲染品質與效能穩定性</string>
+    <string name="enable_gpu_buffer_readback">啟用 GPU 緩衝區讀回</string>
+    <string name="enable_gpu_buffer_readback_description">透過在上傳之前先將 GPU 修改過的緩衝區資料讀回來保存資料，部分遊戲需要啟用此功能才能正常渲染遊戲特效。如果硬體無法負荷可能會導致錯誤</string>
     <string name="use_optimized_vertex_buffers">最佳化頂點緩衝區</string>
     <string name="use_optimized_vertex_buffers_description">啟用最佳化的頂點緩衝區綁定。需要安裝 Mesa 26.0+ Turnip drivers/Qualcomm drivers。使用舊版 Turnip drivers 會導致當機 (25.3版和更低的版本)</string>
 

src/android/app/src/main/res/values/arrays.xml

@@ -558,6 +558,7 @@
         <item>@string/gpu_fence_behavior_immediate</item>
         <item>@string/gpu_fence_behavior_balanced</item>
         <item>@string/gpu_fence_behavior_accurate</item>
+        <item>@string/gpu_fence_behavior_strict</item>
     </string-array>
     <integer-array name="gpuFenceBehaviorValues">
         <item>0</item>

src/android/app/src/main/res/values/strings.xml

@@ -586,6 +586,8 @@
     <string name="renderer_reactive_flushing_description">Improves rendering accuracy in some games at the cost of performance.</string>
     <string name="enable_buffer_history">Enable buffer history</string>
     <string name="enable_buffer_history_description">Enables access to previous buffer states. This option may improve rendering quality and performance consistency in some games.</string>
+    <string name="enable_gpu_buffer_readback">Enable GPU Buffer Readback</string>
+    <string name="enable_gpu_buffer_readback_description">Preserves GPU-modified buffer data by reading it back before uploads. Some games require this to render certain effects properly. May cause issues if the hardware cannot handle the additional workload.</string>
     <string name="use_optimized_vertex_buffers">Optimized Vertex Buffers</string>
     <string name="use_optimized_vertex_buffers_description">Enables optimized vertex buffer binding for improved performance. Requires Mesa 26.0+ Turnip drivers/ QCOM drivers. Will crash on older Turnip drivers (25.3 and below).</string>
 
@@ -1136,6 +1138,7 @@
     <string name="gpu_fence_behavior_immediate">Immediate</string>
     <string name="gpu_fence_behavior_balanced">Balanced</string>
     <string name="gpu_fence_behavior_accurate">Accurate</string>
+    <string name="gpu_fence_behavior_strict">Strict</string>
 
     <!-- ASTC Decoding Method Choices -->
     <string name="accelerate_astc_cpu" translatable="false">CPU</string>

src/common/CMakeLists.txt

@@ -76,7 +76,6 @@ add_library(
     logging.h
     lz4_compression.cpp
     lz4_compression.h
-    make_unique_for_overwrite.h
     math_util.h
     memory_detect.cpp
     memory_detect.h

src/common/make_unique_for_overwrite.h

@@ -1,27 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#pragma once
-
-#include <memory>
-#include <type_traits>
-
-namespace Common {
-
-template <class T>
-    requires(!std::is_array_v<T>)
-std::unique_ptr<T> make_unique_for_overwrite() {
-    return std::unique_ptr<T>(new T);
-}
-
-template <class T>
-    requires std::is_unbounded_array_v<T>
-std::unique_ptr<T> make_unique_for_overwrite(std::size_t n) {
-    return std::unique_ptr<T>(new std::remove_extent_t<T>[n]);
-}
-
-template <class T, class... Args>
-    requires std::is_bounded_array_v<T>
-void make_unique_for_overwrite(Args&&...) = delete;
-
-} // namespace Common

src/common/scratch_buffer.h

@@ -8,8 +8,7 @@
 
 #include <iterator>
 #include <cstring>
-
-#include "common/make_unique_for_overwrite.h"
+#include <memory>
 
 namespace Common {
 
@@ -38,8 +37,9 @@ public:
     ScratchBuffer() = default;
 
     explicit ScratchBuffer(size_type initial_capacity)
-        : last_requested_size{initial_capacity}, buffer_capacity{initial_capacity},
-          buffer{Common::make_unique_for_overwrite<T[]>(initial_capacity)} {}
+        : last_requested_size{initial_capacity}
+        , buffer_capacity{initial_capacity}
+        , buffer{std::make_unique_for_overwrite<T[]>(initial_capacity)} {}
 
     ~ScratchBuffer() = default;
     ScratchBuffer(const ScratchBuffer&) = delete;
@@ -64,7 +64,7 @@ public:
     /// The previously held data will remain intact.
     void resize(size_type size) {
         if (size > buffer_capacity) {
-            auto new_buffer = Common::make_unique_for_overwrite<T[]>(size);
+            auto new_buffer = std::make_unique_for_overwrite<T[]>(size);
             std::memcpy(new_buffer.get(), buffer.get(), buffer_capacity * sizeof(T));
             buffer = std::move(new_buffer);
             buffer_capacity = size;
@@ -77,7 +77,7 @@ public:
     void resize_destructive(size_type size) {
         if (size > buffer_capacity) {
             buffer_capacity = size;
-            buffer = Common::make_unique_for_overwrite<T[]>(buffer_capacity);
+            buffer = std::make_unique_for_overwrite<T[]>(buffer_capacity);
         }
         last_requested_size = size;
     }

src/common/settings.cpp

@@ -178,6 +178,10 @@ bool IsGPUFenceBehaviorAccurate() {
     return values.gpu_fence_behavior.GetValue() == GpuFenceBehavior::Accurate;
 }
 
+bool IsGPUFenceBehaviorStrict() {
+    return values.gpu_fence_behavior.GetValue() == GpuFenceBehavior::Strict;
+}
+
 bool IsFastmemEnabled() {
     if (values.cpu_accuracy.GetValue() == Settings::CpuAccuracy::Debugging)
         return bool(values.cpuopt_fastmem);

src/common/settings.h

@@ -525,7 +525,7 @@ struct Values {
     SwitchableSetting<GpuFenceBehavior, true> gpu_fence_behavior{linkage,
                                                                  GpuFenceBehavior::Default,
                                                                  GpuFenceBehavior::Default,
-                                                                 GpuFenceBehavior::Accurate,
+                                                                 GpuFenceBehavior::Strict,
                                                                  "gpu_fence_behavior",
                                                                  Category::RendererAdvanced,
                                                                  Specialization::Default,
@@ -655,6 +655,13 @@ struct Values {
 
     SwitchableSetting<bool> rescale_hack{linkage, false, "rescale_hack",
                                          Category::RendererHacks};
+    SwitchableSetting<bool> enable_gpu_buffer_readback{linkage,
+                                                       false,
+                                                       "enable_gpu_buffer_readback",
+                                                       Category::RendererAdvanced,
+                                                       Specialization::Default,
+                                                       true,
+                                                       true};
 
     SwitchableSetting<bool> use_asynchronous_shaders{linkage, false, "use_asynchronous_shaders",
                                                      Category::RendererHacks};
@@ -972,6 +979,7 @@ bool IsDMALevelSafe();
 bool IsGPUFenceBehaviorDefault();
 bool IsGPUFenceBehaviorBalanced();
 bool IsGPUFenceBehaviorAccurate();
+bool IsGPUFenceBehaviorStrict();
 
 bool IsFastmemEnabled();
 void SetNceEnabled(bool is_64bit);

src/common/settings_enums.h

@@ -137,7 +137,7 @@ ENUM(VramUsageMode, Conservative, Aggressive);
 ENUM(RendererBackend, OpenGL_GLSL, Vulkan, Null, OpenGL_GLASM, OpenGL_SPIRV);
 ENUM(GpuAccuracy, Low, High);
 ENUM(DmaAccuracy, Default, Unsafe, Safe);
-ENUM(GpuFenceBehavior, Default, Immediate, Balanced, Accurate);
+ENUM(GpuFenceBehavior, Default, Immediate, Balanced, Accurate, Strict);
 ENUM(CpuBackend, Dynarmic, Nce);
 ENUM(CpuAccuracy, Auto, Accurate, Unsafe, Paranoid, Debugging);
 ENUM(CpuClock, Normal, Boost, Overclock)

src/core/file_sys/vfs/vfs_real.cpp

@@ -109,8 +109,7 @@ VirtualFile RealVfsFilesystem::OpenFileFromEntry(std::string_view path_, std::op
     auto reference = std::make_unique<FileReference>();
     this->InsertReferenceIntoListLocked(*reference);
 
-    auto file = std::shared_ptr<RealVfsFile>(
-        new RealVfsFile(*this, std::move(reference), path, perms, size, std::move(parent_path)));
+    auto file = std::make_shared<RealVfsFile>(*this, std::move(reference), path, perms, size, std::move(parent_path));
     cache[path] = file;
 
     return file;
@@ -177,7 +176,7 @@ bool RealVfsFilesystem::DeleteFile(std::string_view path_) {
 
 VirtualDir RealVfsFilesystem::OpenDirectory(std::string_view path_, OpenMode perms) {
     const auto path = FS::SanitizePath(path_, FS::DirectorySeparator::PlatformDefault);
-    return std::shared_ptr<RealVfsDirectory>(new RealVfsDirectory(*this, path, perms));
+    return std::make_shared<RealVfsDirectory>(*this, path, perms);
 }
 
 VirtualDir RealVfsFilesystem::CreateDirectory(std::string_view path_, OpenMode perms) {
@@ -185,7 +184,7 @@ VirtualDir RealVfsFilesystem::CreateDirectory(std::string_view path_, OpenMode p
     if (!FS::CreateDirs(path)) {
         return nullptr;
     }
-    return std::shared_ptr<RealVfsDirectory>(new RealVfsDirectory(*this, path, perms));
+    return std::make_shared<RealVfsDirectory>(*this, path, perms);
 }
 
 VirtualDir RealVfsFilesystem::CopyDirectory(std::string_view old_path_,

src/core/file_sys/vfs/vfs_real.h

@@ -82,6 +82,9 @@ class RealVfsFile : public VfsFile {
     friend class RealVfsFilesystem;
 
 public:
+    RealVfsFile(RealVfsFilesystem& base, std::unique_ptr<FileReference> reference,
+                const std::string& path, OpenMode perms = OpenMode::Read,
+                std::optional<u64> size = {}, std::optional<std::string> parent_path = {});
     ~RealVfsFile() override;
 
     std::string GetName() const override;
@@ -95,9 +98,6 @@ public:
     bool Rename(std::string_view name) override;
 
 private:
-    RealVfsFile(RealVfsFilesystem& base, std::unique_ptr<FileReference> reference,
-                const std::string& path, OpenMode perms = OpenMode::Read,
-                std::optional<u64> size = {}, std::optional<std::string> parent_path = {});
 
     RealVfsFilesystem& base;
     std::unique_ptr<FileReference> reference;
@@ -113,6 +113,8 @@ class RealVfsDirectory : public VfsDirectory {
     friend class RealVfsFilesystem;
 
 public:
+    RealVfsDirectory(RealVfsFilesystem& base, const std::string& path,
+                     OpenMode perms = OpenMode::Read);
     ~RealVfsDirectory() override;
 
     VirtualFile GetFileRelative(std::string_view relative_path) const override;
@@ -138,9 +140,6 @@ public:
     std::map<std::string, VfsEntryType, std::less<>> GetEntries() const override;
 
 private:
-    RealVfsDirectory(RealVfsFilesystem& base, const std::string& path,
-                     OpenMode perms = OpenMode::Read);
-
     template <typename T, typename R>
     std::vector<std::shared_ptr<R>> IterateEntries() const;
 

src/core/hle/service/am/window_system.cpp

@@ -517,7 +517,7 @@ void WindowSystem::UpdateAppletStateLocked(Applet* applet, bool is_foreground, b
     // Layer ordering. Composition sorts back-to-front. Now with enums for calrity.
     s32 z_index = Background;
     if (is_overlay) {
-        z_index = Overlay;
+        z_index = this->IsOverlayOpenLocked(*applet) ? Overlay : Background;
     } else if (inherited_foreground) {
         z_index = is_obscured ? Foreground : ForegroundVisible;
     }

src/core/hle/service/psc/ovln/receiver_service.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2024 yuzu Emulator Project
@@ -24,7 +24,7 @@ IReceiverService::~IReceiverService() = default;
 
 Result IReceiverService::OpenReceiver(Out<SharedPointer<IReceiver>> out_receiver) {
     LOG_DEBUG(Service_PSC, "called");
-    *out_receiver = std::shared_ptr<IReceiver>(new IReceiver(system));
+    *out_receiver = std::make_shared<IReceiver>(system);
     R_SUCCEED();
 }
 

src/hid_core/resources/npad/npad.cpp

@@ -45,6 +45,19 @@ NPad::NPad(Core::HID::HIDCore& hid_core_, KernelHelpers::ServiceContext& service
         AbstractPad{hid_core_.kernel},
     }}
 {
+    for (std::size_t aruid_index = 0; aruid_index < AruidIndexMax; ++aruid_index) {
+        for (std::size_t i = 0; i < controller_data[aruid_index].size(); ++i) {
+            auto& controller = controller_data[aruid_index][i];
+            controller.device = hid_core.GetEmulatedControllerByIndex(i);
+            Core::HID::ControllerUpdateCallback engine_callback{
+                .on_change = [this, i, kernel = &hid_core.kernel](Core::HID::ControllerTriggerType type) {
+                    ControllerUpdate(*kernel, type, i);
+                },
+                .is_npad_service = true,
+            };
+            controller.callback_key = controller.device->SetCallback(engine_callback);
+        }
+    }
     for (std::size_t i = 0; i < abstracted_pads.size(); ++i) {
         abstracted_pads[i].SetNpadId(IndexToNpadIdType(i));
     }
@@ -93,16 +106,6 @@ Result NPad::Activate(u64 aruid) {
     for (std::size_t i = 0; i < controller_data[aruid_index].size(); ++i) {
         auto& controller = controller_data[aruid_index][i];
         controller.shared_memory = &data->shared_memory_format->npad.npad_entry[i].internal_state;
-        controller.device = hid_core.GetEmulatedControllerByIndex(i);
-        if (!controller.callback_key) {
-            Core::HID::ControllerUpdateCallback engine_callback{
-                .on_change = [this, i](Core::HID::ControllerTriggerType type) {
-                    ControllerUpdate(hid_core.kernel, type, i);
-                },
-                .is_npad_service = true,
-            };
-            controller.callback_key = controller.device->SetCallback(engine_callback);
-        }
     }
 
     // Prefill controller buffers

src/qt_common/config/shared_translation.cpp

@@ -226,7 +226,9 @@ std::unique_ptr<TranslationMap> InitializeTranslations(QObject* parent) {
     INSERT(Settings, dma_accuracy, tr("DMA Accuracy:"),
            tr("Controls the DMA read mode.\nUnsafe is faster, while Safe is more stable and can fix issues in some games.\nDefault follows the GPU Accuracy setting."));
     INSERT(Settings, gpu_fence_behavior, tr("GPU Fence Behavior:"),
-           tr("Controls the GPU fence synchronization behavior.\nImmediate is the fastest option, but can introduce some issues.\nBalanced offers better compatibility and may fix issues in some games.\nAccurate further improves compatibility at the cost of some performance.\nDefault follows the GPU Mode setting."));
+           tr("Controls the GPU fence synchronization behavior.\nImmediate is the fastest option, but can introduce some issues.\nBalanced offers better compatibility and may fix issues in some games.\nAccurate further improves compatibility at the cost of some performance.\nStrict is the slowest option, but can fix issues that require stricter synchronization.\nDefault follows the GPU Accuracy setting."));
+    INSERT(Settings, enable_gpu_buffer_readback, tr("Enable GPU buffer readback"),
+           tr("Preserves GPU-modified data by reading it back before uploading.\nSome games require this to render certain effects properly."));
     INSERT(Settings, use_asynchronous_shaders, tr("Enable asynchronous shader compilation"),
            tr("May reduce shader stutter."));
     INSERT(Settings, gpu_clock, tr("GPU Clocks"),
@@ -440,6 +442,7 @@ std::unique_ptr<ComboboxTranslationMap> ComboboxEnumeration(QObject* parent) {
                               PAIR(GpuFenceBehavior, Immediate, tr("Immediate")),
                               PAIR(GpuFenceBehavior, Balanced, tr("Balanced")),
                               PAIR(GpuFenceBehavior, Accurate, tr("Accurate")),
+                              PAIR(GpuFenceBehavior, Strict, tr("Strict")),
                           }});
     translations->insert(
         {Settings::EnumMetadata<Settings::CpuAccuracy>::Index(),

src/video_core/buffer_cache/buffer_cache.h

@@ -249,10 +249,6 @@ bool BufferCache<P>::DMACopy(GPUVAddr src_address, GPUVAddr dest_address, u64 am
     runtime.CopyBuffer(dest_buffer, src_buffer, copies, true);
     if (has_new_downloads) {
         memory_tracker.MarkRegionAsGpuModified(*cpu_dest_address, amount);
-        const bool should_sync = Settings::IsGPUFenceBehaviorBalanced() || Settings::IsGPUFenceBehaviorAccurate();
-        if (should_sync) {
-            runtime.Finish();
-        }
     }
 
     Tegra::Memory::DeviceGuestMemoryScoped<u8, Tegra::Memory::GuestMemoryFlags::UnsafeReadWrite>
@@ -1234,7 +1230,7 @@ void BufferCache<P>::BindHostComputeStorageBuffers() {
         buffer.MarkUsage(offset, size);
 
         if (is_written) {
-            MarkWrittenBuffer(binding.buffer_id, binding.device_addr, size, true);
+            MarkWrittenBuffer(binding.buffer_id, binding.device_addr, size);
         }
 
         if constexpr (NEEDS_BIND_STORAGE_INDEX) {
@@ -1520,13 +1516,11 @@ void BufferCache<P>::UpdateComputeTextureBuffers() {
 }
 
 template <class P>
-void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u32 size, bool needs_sync) {
+void BufferCache<P>::MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u32 size) {
     if constexpr (!IS_OPENGL) {
-        if (needs_sync) {
         Buffer& buffer = slot_buffers[buffer_id];
         buffer.setWriteTick(runtime.CurrentTick());
     }
-    }
     memory_tracker.MarkRegionAsGpuModified(device_addr, size);
     gpu_modified_ranges.Add(device_addr, size);
     uncommitted_gpu_modified_ranges.Add(device_addr, size);
@@ -1541,11 +1535,8 @@ BufferId BufferCache<P>::FindBuffer(DAddr device_addr, u32 size, bool sparse_com
     const BufferId buffer_id = page_table[page];
     if (buffer_id) {
         Buffer& buffer = slot_buffers[buffer_id];
+        WaitForGpuFenceIfNeeded(buffer);
         if (buffer.IsInBounds(device_addr, size)) {
-            const bool should_sync = Settings::IsGPUFenceBehaviorAccurate();
-            if (should_sync) {
-                SynchronizeBufferWrites(buffer);
-            }
             bool usable = true;
             if constexpr (requires { buffer.IsSparseCompatible(); }) {
                 if (sparse_compatible && !buffer.IsSparseCompatible()) {
@@ -1561,14 +1552,20 @@ BufferId BufferCache<P>::FindBuffer(DAddr device_addr, u32 size, bool sparse_com
 }
 
 template <class P>
-void BufferCache<P>::SynchronizeBufferWrites(Buffer& buffer) {
+void BufferCache<P>::WaitForGpuFenceIfNeeded(Buffer& buffer) {
     if constexpr (!IS_OPENGL) {
+        const bool gpu_fence_accurate = Settings::IsGPUFenceBehaviorAccurate();
+        const bool gpu_fence_strict = Settings::IsGPUFenceBehaviorStrict();
+        if (gpu_fence_accurate || gpu_fence_strict) {
+            const u64 gpu_tick_delay = gpu_fence_strict ? 0 : 3;
             const u64 buffer_tick = buffer.getWriteTick();
-        if (!runtime.IsFree(buffer_tick)) {
+            const u64 gpu_tick = runtime.KnownGpuTick();
+            if (buffer_tick > gpu_tick + gpu_tick_delay) {
                 runtime.Wait(buffer_tick);
             }
         }
     }
+}
 
 template <class P>
 typename BufferCache<P>::OverlapResult BufferCache<P>::ResolveOverlaps(DAddr device_addr,
@@ -1798,6 +1795,9 @@ void BufferCache<P>::ImmediateUploadMemory([[maybe_unused]] Buffer& buffer,
                 if (immediate_buffer.empty()) {
                     immediate_buffer = ImmediateBuffer(largest_copy);
                 }
+                if (Settings::values.enable_gpu_buffer_readback.GetValue()) {
+                    DownloadBufferMemory(buffer, device_addr, copy.size);
+                }
                 device_memory.ReadBlockUnsafe(device_addr, immediate_buffer.data(), copy.size);
                 upload_span = immediate_buffer.subspan(0, copy.size);
             }
@@ -1816,6 +1816,9 @@ void BufferCache<P>::MappedUploadMemory([[maybe_unused]] Buffer& buffer,
         for (BufferCopy& copy : copies) {
             u8* const src_pointer = staging_pointer.data() + copy.src_offset;
             const DAddr device_addr = buffer.CpuAddr() + copy.dst_offset;
+            if (Settings::values.enable_gpu_buffer_readback.GetValue()) {
+                DownloadBufferMemory(buffer, device_addr, copy.size);
+            }
             device_memory.ReadBlockUnsafe(device_addr, src_pointer, copy.size);
             // Apply the staging offset
             copy.src_offset += upload_staging.offset;

src/video_core/buffer_cache/buffer_cache_base.h

@@ -430,11 +430,11 @@ private:
 
     void UpdateComputeTextureBuffers();
 
-    void MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u32 size, bool needs_sync = false);
+    void MarkWrittenBuffer(BufferId buffer_id, DAddr device_addr, u32 size);
 
     [[nodiscard]] BufferId FindBuffer(DAddr device_addr, u32 size, bool sparse_compatible);
 
-    void SynchronizeBufferWrites(Buffer& buffer);
+    void WaitForGpuFenceIfNeeded(Buffer& buffer);
 
     [[nodiscard]] OverlapResult ResolveOverlaps(DAddr device_addr, u32 wanted_size);
 

src/video_core/fence_manager.h

@@ -72,7 +72,7 @@ public:
     }
 
     void SignalFence(std::function<void()>&& func) {
-        const bool delay_fence = Settings::IsGPUFenceBehaviorDefault() ? Settings::IsGPULevelHigh() : Settings::IsGPUFenceBehaviorBalanced() || Settings::IsGPUFenceBehaviorAccurate();
+        const bool delay_fence = Settings::IsGPUFenceBehaviorDefault() ? Settings::IsGPULevelHigh() : Settings::IsGPUFenceBehaviorBalanced() || Settings::IsGPUFenceBehaviorAccurate() || Settings::IsGPUFenceBehaviorStrict();
         const bool should_flush = ShouldFlush();
         if constexpr (!can_async_check) {
             TryReleasePendingFences<false>();

src/video_core/query_cache/query_cache.h

@@ -260,7 +260,7 @@ void QueryCacheBase<Traits>::CounterReport(GPUVAddr addr, QueryType counter_type
     };
     u8* pointer = impl->device_memory.template GetPointer<u8>(cpu_addr);
     u8* pointer_timestamp = impl->device_memory.template GetPointer<u8>(cpu_addr + 8);
-    bool is_synced = (Settings::IsGPUFenceBehaviorDefault() ? !Settings::IsGPULevelHigh() : !Settings::IsGPUFenceBehaviorBalanced() && !Settings::IsGPUFenceBehaviorAccurate()) && is_fence;
+    bool is_synced = (Settings::IsGPUFenceBehaviorDefault() ? !Settings::IsGPULevelHigh() : !Settings::IsGPUFenceBehaviorBalanced() && !Settings::IsGPUFenceBehaviorAccurate() && !Settings::IsGPUFenceBehaviorStrict()) && is_fence;
     std::function<void()> operation([this, is_synced, streamer, query_base = query, query_location,
                                      pointer, pointer_timestamp] {
         if (True(query_base->flags & QueryFlagBits::IsInvalidated)) {

src/video_core/renderer_vulkan/present/layer.cpp

@@ -94,13 +94,13 @@ void Layer::ConfigureDraw(const Device& device, PresentPushConstants* out_push_c
     const u32 scaled_width = texture_info ? texture_info->scaled_width : texture_width;
     const u32 scaled_height = texture_info ? texture_info->scaled_height : texture_height;
     const bool use_accelerated = texture_info.has_value();
+    const bool is_applet =
+        (framebuffer.layer_stack_mask & Service::Nvnflinger::LayerStackBit(
+                                            Service::Nvnflinger::LayerStackId::Recording)) == 0;
 
     RefreshResources(device, framebuffer);
     SetAntiAliasPass(device);
 #ifdef HAS_RESHADE
-    const bool is_applet =
-        (framebuffer.layer_stack_mask & Service::Nvnflinger::LayerStackBit(
-                                            Service::Nvnflinger::LayerStackId::Recording)) == 0;
     SetPostProcessPass(device, is_applet);
 #endif
 
@@ -141,9 +141,12 @@ void Layer::ConfigureDraw(const Device& device, PresentPushConstants* out_push_c
         source_image_view = fsr->Draw(device, scheduler, image_index, source_image, source_image_view, render_extent, crop_rect);
         crop_rect = {0, 0, 1, 1};
     } else if (auto* sgsr = std::get_if<SGSR>(&sr_filter)) {
-        source_image_view = sgsr->Draw(device, scheduler, image_index, source_image, source_image_view, render_extent, crop_rect);
+        if (!is_applet) {
+            source_image_view = sgsr->Draw(device, scheduler, image_index, source_image,
+                                           source_image_view, render_extent, crop_rect);
             crop_rect = {0, 0, 1, 1};
         }
+    }
 
     SetMatrixData(device, *out_push_constants, layout);
     SetVertexData(device, *out_push_constants, layout, crop_rect);

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -200,7 +200,7 @@ public:
             if (host_visible) {
                 return StagingBufferRef{};
             }
-            return staging_pool.Request(size_bytes, MemoryUsage::Upload);
+            return staging_pool.Request(device, size_bytes, MemoryUsage::Upload);
         }();
 
         u8* staging_data = host_visible ? buffer.Mapped().data() : staging.mapped_span.data();
@@ -375,11 +375,11 @@ BufferCacheRuntime::BufferCacheRuntime(const Device& device_, MemoryAllocator& m
 }
 
 StagingBufferRef BufferCacheRuntime::UploadStagingBuffer(size_t size) {
-    return staging_pool.Request(size, MemoryUsage::Upload);
+    return staging_pool.Request(device, size, MemoryUsage::Upload);
 }
 
 StagingBufferRef BufferCacheRuntime::DownloadStagingBuffer(size_t size, bool deferred) {
-    return staging_pool.Request(size, MemoryUsage::Download, deferred);
+    return staging_pool.Request(device, size, MemoryUsage::Download, deferred);
 }
 
 VkFormat BufferCacheRuntime::TexelBufferFormat(VideoCore::Surface::PixelFormat format) const {
@@ -419,15 +419,15 @@ void BufferCacheRuntime::TickFrame(Common::SlotVector<Buffer>& slot_buffers) noe
 }
 
 u64 BufferCacheRuntime::CurrentTick() {
-    return scheduler.CurrentTick();
+    return scheduler.GetMasterSemaphore().CurrentTick();
 }
 
-bool BufferCacheRuntime::IsFree(u64 tick) {
-    return scheduler.IsFree(tick);
+u64 BufferCacheRuntime::KnownGpuTick() {
+    return scheduler.GetMasterSemaphore().KnownGpuTick();
 }
 
-void BufferCacheRuntime::Wait(u64 tick) {
-    scheduler.Wait(tick);
+void BufferCacheRuntime::Wait(u64 buffer_tick) {
+    scheduler.Wait(buffer_tick);
 }
 
 void BufferCacheRuntime::Finish() {

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -112,9 +112,9 @@ public:
 
     u64 CurrentTick();
 
-    bool IsFree(u64 tick);
+    u64 KnownGpuTick();
 
-    void Wait(u64 tick);
+    void Wait(u64 buffer_tick);
 
     void Finish();
 
@@ -162,7 +162,7 @@ public:
     std::span<u8> BindMappedUniformBuffer([[maybe_unused]] size_t stage,
                                           [[maybe_unused]] u32 binding_index,
                                           u32 size) {
-        const StagingBufferRef ref = staging_pool.Request(size, MemoryUsage::Upload);
+        const StagingBufferRef ref = staging_pool.Request(device, size, MemoryUsage::Upload);
         guest_descriptor_queue.AddBuffer(ref.buffer, ref.device_address,
                                          static_cast<u32>(ref.offset), size);
         return ref.mapped_span;

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -287,7 +287,7 @@ Uint8Pass::~Uint8Pass() = default;
 std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer,
                                                       u32 src_offset) {
     const u32 staging_size = static_cast<u32>(num_vertices * sizeof(u16));
-    const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
+    const auto staging = staging_buffer_pool.Request(device, staging_size, MemoryUsage::DeviceLocal);
 
     compute_pass_descriptor_queue.Acquire(scheduler, 2);
     compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices);
@@ -345,7 +345,7 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
     const u32 num_tri_vertices = (is_strip ? (num_vertices - 2) / 2 : num_vertices / 4) * 6;
 
     const std::size_t staging_size = num_tri_vertices * sizeof(u32);
-    const auto staging = staging_buffer_pool.Request(staging_size, MemoryUsage::DeviceLocal);
+    const auto staging = staging_buffer_pool.Request(device, staging_size, MemoryUsage::DeviceLocal);
 
     compute_pass_descriptor_queue.Acquire(scheduler, 2);
     compute_pass_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size);

src/video_core/renderer_vulkan/vk_query_cache.cpp

@@ -852,7 +852,7 @@ public:
 
     void PushUnsyncedQueries() override {
         CloseCounter();
-        auto staging_ref = staging_pool.Request(
+        auto staging_ref = staging_pool.Request(device,
             pending_flush_queries.size() * TFBQueryBank::QUERY_SIZE, MemoryUsage::Download, true);
         size_t offset_base = staging_ref.offset;
         for (auto q : pending_flush_queries) {
@@ -1657,7 +1657,7 @@ void QueryCacheRuntime::SyncValues(std::span<SyncValuesType> values, VkBuffer ba
     impl->copies_setup.clear();
     impl->copies_setup.resize(impl->little_cache.size());
     if constexpr (SyncValuesType::GeneratesBaseBuffer) {
-        ref = impl->staging_pool.Request(total_size, MemoryUsage::Upload);
+        ref = impl->staging_pool.Request(impl->device, total_size, MemoryUsage::Upload);
         size_t current_offset = ref.offset;
         size_t accumulated_size = 0;
         for (size_t i = 0; i < values.size(); i++) {

src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp

@@ -28,55 +28,42 @@ using namespace Common::Literals;
 // Maximum potential alignment of a Vulkan buffer
 constexpr VkDeviceSize MAX_ALIGNMENT = 256;
 
-// Stream buffer size in bytes
-// *NIX drivers are more sensitive to increased buffers for streaming.
-// Windows ones however, can intake bigger buffers and generally do not OOM.
-// - GTX 960 on Windows will not OOM with 256mib
-// - GT 1030 on ^NIX will OOM with 256mib
-#if defined(__FreeBSD__)
-constexpr VkDeviceSize MAX_STREAM_BUFFER_SIZE = 128_MiB;
-#else
-constexpr VkDeviceSize MAX_STREAM_BUFFER_SIZE = 256_MiB;
-#endif
-
-size_t GetStreamBufferSize(const Device& device) {
+size_t GetStreamBufferSize(const Device& device, size_t max_stream_buffer_size, size_t max_alignment) {
     if (!device.HasDebuggingToolAttached()) {
-        return MAX_STREAM_BUFFER_SIZE;
+        return max_stream_buffer_size;
     }
 
     VkDeviceSize size{0};
     bool has_device_local_host_visible_heap{};
-    ForEachDeviceLocalHostVisibleHeap(device, [&size, &has_device_local_host_visible_heap](
-                                                  size_t index, VkMemoryHeap& heap) {
+    ForEachDeviceLocalHostVisibleHeap(device, [&size, &has_device_local_host_visible_heap](size_t index, VkMemoryHeap& heap) {
         has_device_local_host_visible_heap = true;
-        size = (std::max)(size, heap.size);
+        size = std::max<size_t>(size, heap.size);
     });
     if (has_device_local_host_visible_heap) {
         // If rebar is not supported, cut the max heap size to 40%. This will allow 2 captures to be
         // loaded at the same time in RenderDoc. If rebar is supported, this shouldn't be an issue
         // as the heap will be much larger.
-        if (size <= MAX_STREAM_BUFFER_SIZE) {
+        if (size <= max_stream_buffer_size) {
             size = size * 40 / 100;
         }
     } else {
-        size = MAX_STREAM_BUFFER_SIZE;
+        size = max_stream_buffer_size;
     }
-    return (std::min)(Common::AlignUp(size, MAX_ALIGNMENT), MAX_STREAM_BUFFER_SIZE);
+    return std::min<size_t>(Common::AlignUp(size, max_alignment), max_stream_buffer_size);
 }
 } // Anonymous namespace
 
-StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& memory_allocator_,
-                                     Scheduler& scheduler_)
-    : device{device_}, memory_allocator{memory_allocator_}, scheduler{scheduler_},
-      stream_buffer_size{GetStreamBufferSize(device)}, region_size{stream_buffer_size /
-                                                                   StagingBufferPool::NUM_SYNCS} {
+StagingBufferPool::StagingBufferPool(const Device& device, MemoryAllocator& memory_allocator_, Scheduler& scheduler_)
+    : memory_allocator{memory_allocator_}, scheduler{scheduler_}
+    , stream_buffer_size{GetStreamBufferSize(device, 256_MiB, MAX_ALIGNMENT)}
+{
     VkBufferCreateInfo stream_ci = {
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
         .pNext = nullptr,
         .flags = 0,
         .size = stream_buffer_size,
-        .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
-                 VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
+        .usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT
+            | VK_BUFFER_USAGE_INDEX_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT,
         .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
         .queueFamilyIndexCount = 0,
         .pQueueFamilyIndices = nullptr,
@@ -87,7 +74,16 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem
     if (device.IsBufferDeviceAddressSupported()) {
         stream_ci.usage |= VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT;
     }
+    // Some drivers are more sensitive to increased buffer sizes
+    try {
         stream_buffer = memory_allocator.CreateBuffer(stream_ci, MemoryUsage::Stream);
+    } catch (vk::Exception& e) {
+        LOG_ERROR(Render_Vulkan, "Can't fit {} bytes buffer, halving", stream_ci.size);
+        stream_buffer_size = GetStreamBufferSize(device, 128_MiB, MAX_ALIGNMENT);
+        stream_ci.size = stream_buffer_size;
+        stream_buffer = memory_allocator.CreateBuffer(stream_ci, MemoryUsage::Stream);
+    }
+    region_size = stream_buffer_size / StagingBufferPool::NUM_SYNCS;
     if (device.HasDebuggingToolAttached()) {
         stream_buffer.SetObjectNameEXT("Stream Buffer");
     }
@@ -100,11 +96,10 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem
 
 StagingBufferPool::~StagingBufferPool() = default;
 
-StagingBufferRef StagingBufferPool::Request(size_t size, MemoryUsage usage, bool deferred) {
-    if (!deferred && usage == MemoryUsage::Upload && size <= region_size) {
-        return GetStreamBuffer(size);
-    }
-    return GetStagingBuffer(size, usage, deferred);
+StagingBufferRef StagingBufferPool::Request(const Device& device, size_t size, MemoryUsage usage, bool deferred) {
+    return (!deferred && usage == MemoryUsage::Upload && size <= region_size)
+        ? GetStreamBuffer(device, size)
+        : GetStagingBuffer(device, size, usage, deferred);
 }
 
 void StagingBufferPool::FreeDeferred(StagingBufferRef& ref) {
@@ -127,11 +122,10 @@ void StagingBufferPool::TickFrame() {
     ReleaseCache(MemoryUsage::Download);
 }
 
-StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) {
-    if (AreRegionsActive(Region(free_iterator) + 1,
-                         (std::min)(Region(iterator + size) + 1, NUM_SYNCS))) {
+StagingBufferRef StagingBufferPool::GetStreamBuffer(const Device& device, size_t size) {
+    if (AreRegionsActive(Region(free_iterator) + 1, (std::min)(Region(iterator + size) + 1, NUM_SYNCS))) {
         // Avoid waiting for the previous usages to be free
-        return GetStagingBuffer(size, MemoryUsage::Upload);
+        return GetStagingBuffer(device, size, MemoryUsage::Upload);
     }
     const u64 current_tick = scheduler.CurrentTick();
     std::fill(sync_ticks.begin() + Region(used_iterator), sync_ticks.begin() + Region(iterator),
@@ -140,15 +134,14 @@ StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) {
     free_iterator = (std::max)(free_iterator, iterator + size);
 
     if (iterator + size >= stream_buffer_size) {
-        std::fill(sync_ticks.begin() + Region(used_iterator), sync_ticks.begin() + NUM_SYNCS,
-                  current_tick);
+        std::fill(sync_ticks.begin() + Region(used_iterator), sync_ticks.begin() + NUM_SYNCS, current_tick);
         used_iterator = 0;
         iterator = 0;
         free_iterator = size;
 
         if (AreRegionsActive(0, Region(size) + 1)) {
             // Avoid waiting for the previous usages to be free
-            return GetStagingBuffer(size, MemoryUsage::Upload);
+            return GetStagingBuffer(device, size, MemoryUsage::Upload);
         }
     }
     const size_t offset = iterator;
@@ -156,7 +149,7 @@ StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) {
     return StagingBufferRef{
         .buffer = *stream_buffer,
         .device_address = stream_buffer_address,
-        .offset = static_cast<VkDeviceSize>(offset),
+        .offset = VkDeviceSize(offset),
         .mapped_span = stream_pointer.subspan(offset, size),
         .usage{},
         .log2_level{},
@@ -166,21 +159,18 @@ StagingBufferRef StagingBufferPool::GetStreamBuffer(size_t size) {
 
 bool StagingBufferPool::AreRegionsActive(size_t region_begin, size_t region_end) const {
     const u64 gpu_tick = scheduler.GetMasterSemaphore().KnownGpuTick();
-    return std::any_of(sync_ticks.begin() + region_begin, sync_ticks.begin() + region_end,
-                       [gpu_tick](u64 sync_tick) { return gpu_tick < sync_tick; });
+    return std::any_of(sync_ticks.begin() + region_begin, sync_ticks.begin() + region_end, [gpu_tick](u64 sync_tick) {
+        return gpu_tick < sync_tick;
+    });
 };
 
-StagingBufferRef StagingBufferPool::GetStagingBuffer(size_t size, MemoryUsage usage,
-                                                     bool deferred) {
-    if (const std::optional<StagingBufferRef> ref = TryGetReservedBuffer(size, usage, deferred)) {
+StagingBufferRef StagingBufferPool::GetStagingBuffer(const Device& device, size_t size, MemoryUsage usage, bool deferred) {
+    if (const std::optional<StagingBufferRef> ref = TryGetReservedBuffer(size, usage, deferred))
         return *ref;
-    }
-    return CreateStagingBuffer(size, usage, deferred);
+    return CreateStagingBuffer(device, size, usage, deferred);
 }
 
-std::optional<StagingBufferRef> StagingBufferPool::TryGetReservedBuffer(size_t size,
-                                                                        MemoryUsage usage,
-                                                                        bool deferred) {
+std::optional<StagingBufferRef> StagingBufferPool::TryGetReservedBuffer(size_t size, MemoryUsage usage, bool deferred) {
     StagingBuffers& cache_level = GetCache(usage)[Common::Log2Ceil(size)];
 
     const auto is_free = [this](const StagingBuffer& entry) {
@@ -202,7 +192,7 @@ std::optional<StagingBufferRef> StagingBufferPool::TryGetReservedBuffer(size_t s
     return it->Ref();
 }
 
-StagingBufferRef StagingBufferPool::CreateStagingBuffer(size_t size, MemoryUsage usage, bool deferred) {
+StagingBufferRef StagingBufferPool::CreateStagingBuffer(const Device& device, size_t size, MemoryUsage usage, bool deferred) {
     auto const log2_size = Common::Log2Ceil<u32>(u32(size));
     VkBufferCreateInfo buffer_ci = {
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,

src/video_core/renderer_vulkan/vk_staging_buffer_pool.h

@@ -33,11 +33,10 @@ class StagingBufferPool {
 public:
     static constexpr size_t NUM_SYNCS = 16;
 
-    explicit StagingBufferPool(const Device& device, MemoryAllocator& memory_allocator,
-                               Scheduler& scheduler);
+    explicit StagingBufferPool(const Device& device, MemoryAllocator& memory_allocator, Scheduler& scheduler);
     ~StagingBufferPool();
 
-    StagingBufferRef Request(size_t size, MemoryUsage usage, bool deferred = false);
+    StagingBufferRef Request(const Device& device, size_t size, MemoryUsage usage, bool deferred = false);
     void FreeDeferred(StagingBufferRef& ref);
 
     [[nodiscard]] VkBuffer StreamBuf() const noexcept {
@@ -84,27 +83,18 @@ private:
     static constexpr size_t NUM_LEVELS = sizeof(size_t) * CHAR_BIT;
     using StagingBuffersCache = std::array<StagingBuffers, NUM_LEVELS>;
 
-    StagingBufferRef GetStreamBuffer(size_t size);
-
+    StagingBufferRef GetStreamBuffer(const Device& device, size_t size);
     bool AreRegionsActive(size_t region_begin, size_t region_end) const;
-
-    StagingBufferRef GetStagingBuffer(size_t size, MemoryUsage usage, bool deferred = false);
-
-    std::optional<StagingBufferRef> TryGetReservedBuffer(size_t size, MemoryUsage usage,
-                                                         bool deferred);
-
-    StagingBufferRef CreateStagingBuffer(size_t size, MemoryUsage usage, bool deferred);
-
+    StagingBufferRef GetStagingBuffer(const Device& device, size_t size, MemoryUsage usage, bool deferred = false);
+    std::optional<StagingBufferRef> TryGetReservedBuffer(size_t size, MemoryUsage usage, bool deferred);
+    StagingBufferRef CreateStagingBuffer(const Device& device, size_t size, MemoryUsage usage, bool deferred);
     StagingBuffersCache& GetCache(MemoryUsage usage);
-
     void ReleaseCache(MemoryUsage usage);
-
     void ReleaseLevel(StagingBuffersCache& cache, size_t log2);
     size_t Region(size_t iter) const noexcept {
         return iter / region_size;
     }
 
-    const Device& device;
     MemoryAllocator& memory_allocator;
     Scheduler& scheduler;
 

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -975,11 +975,11 @@ void TextureCacheRuntime::Finish() {
 }
 
 StagingBufferRef TextureCacheRuntime::UploadStagingBuffer(size_t size, bool deferred) {
-    return staging_buffer_pool.Request(size, MemoryUsage::Upload, deferred);
+    return staging_buffer_pool.Request(device, size, MemoryUsage::Upload, deferred);
 }
 
 StagingBufferRef TextureCacheRuntime::DownloadStagingBuffer(size_t size, bool deferred) {
-    return staging_buffer_pool.Request(size, MemoryUsage::Download, deferred);
+    return staging_buffer_pool.Request(device, size, MemoryUsage::Download, deferred);
 }
 
 void TextureCacheRuntime::FreeDeferredStagingBuffer(StagingBufferRef& ref) {