Small experiment

docs/Deps.md

@@ -79,7 +79,7 @@ Certain other dependencies will be fetched by CPM regardless. System packages *c
 
 On amd64:
 
-* [xbyak](https://github.com/herumi/xbyak) - 7.35.2+
+* [xbyak](https://github.com/herumi/xbyak) - 7.22 or earlier is recommended
 
 On aarch64 OR if `DYNARMIC_TESTS` is on:
 
@@ -161,12 +161,11 @@ sudo pacman -Syu --needed base-devel boost catch2 cmake enet ffmpeg fmt git glsl
 <summary>Ubuntu, Debian, Mint Linux</summary>
 
 ```sh
-sudo apt-get install autoconf cmake g++ gcc git glslang-tools libglu1-mesa-dev libhidapi-dev libpulse-dev libtool libudev-dev libxcb-icccm4 libxcb-image0 libxcb-keysyms1 libxcb-render-util0 libxcb-xinerama0 libxcb-xkb1 libxext-dev libxkbcommon-x11-0 mesa-common-dev nasm ninja-build qt6-base-private-dev catch2 libfmt-dev liblz4-dev nlohmann-json3-dev libzstd-dev libssl-dev libavfilter-dev libavcodec-dev libswscale-dev pkg-config zlib1g-dev libva-dev libvdpau-dev qt6-tools-dev qt6-charts-dev libvulkan-dev spirv-tools spirv-headers libusb-1.0-0-dev libboost-dev libboost-fiber-dev libboost-context-dev libsdl2-dev libopus-dev libasound2t64 vulkan-utility-libraries-dev
+sudo apt-get install autoconf cmake g++ gcc git glslang-tools libglu1-mesa-dev libhidapi-dev libpulse-dev libtool libudev-dev libxcb-icccm4 libxcb-image0 libxcb-keysyms1 libxcb-render-util0 libxcb-xinerama0 libxcb-xkb1 libxext-dev libxkbcommon-x11-0 mesa-common-dev nasm ninja-build qt6-base-private-dev catch2 libfmt-dev liblz4-dev nlohmann-json3-dev libzstd-dev libssl-dev libavfilter-dev libavcodec-dev libswscale-dev pkg-config zlib1g-dev libva-dev libvdpau-dev qt6-tools-dev qt6-charts-dev libvulkan-dev spirv-tools spirv-headers libusb-1.0-0-dev libxbyak-dev libboost-dev libboost-fiber-dev libboost-context-dev libsdl2-dev libopus-dev libasound2t64 vulkan-utility-libraries-dev
 ```
 
 * Ubuntu 22.04, Linux Mint 20, or Debian 12 or later is required.
 * To enable QT Web Engine, add `-DYUZU_USE_QT_WEB_ENGINE=ON` when running CMake.
-* `libxbyak-dev` is optional for x86_64 targets
 
 </details>
 

externals/cpmfile.json

@@ -58,13 +58,8 @@
         "package": "xbyak",
         "repo": "herumi/xbyak",
         "tag": "v%VERSION%",
-        "hash": "d93971cc8f17f20818e36099aa862d15d30b60f17c7dff0cc8b7ac89ad6dc6453256dd47a454904666a85cea6ca27f9867f782c7b2c6d0c16039af8e3e28e6cc",
-        "git_version": "7.35.4",
-        "bundled": true,
-        "skip_updates": true,
-        "patches": [
-            "0001-rvalue-optimize.patch"
-        ]
+        "hash": "b6475276b2faaeb315734ea8f4f8bd87ededcee768961b39679bee547e7f3e98884d8b7851e176d861dab30a80a76e6ea302f8c111483607dde969b4797ea95a",
+        "git_version": "7.35.2"
     },
     "oaknut": {
         "repo": "eden-emulator/oaknut",

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

@@ -287,7 +287,7 @@
     <string name="install_amiibo_keys">تثبيت مفاتيح أميبو</string>
     <string name="install_amiibo_keys_description">مطلوب لاستخدام أميبو في اللعبة</string>
     <string name="gpu_driver_fetcher">أداة جلب برامج تشغيل وحدة المعالجة الرسومية</string>
-    <string name="gpu_driver_manager">إدارة برامج تشغيل وحدة معالجة الرسومات</string>
+    <string name="gpu_driver_manager">مدير برامج تشغيل وحدة معالجة الرسومات</string>
     <string name="install_gpu_driver_description">تثبيت برامج تشغيل بديلة لأداء أو دقة أفضل</string>
     <string name="advanced_settings">الإعدادات المتقدمة</string>
     <string name="settings_description">ضبط إعدادات المحاكي</string>
@@ -486,10 +486,6 @@
     <string name="use_disk_shader_cache_description">يقلل من التأتأة عن طريق تخزين وتحميل التظليلات التي تم إنشاؤها محليًا.</string>
     <string name="renderer_force_max_clock">إجبار السرعة القصوى (لأجهزة Adreno فقط)</string>
     <string name="renderer_force_max_clock_description">يجبر وحدة معالجة الرسومات على العمل بأقصى سرعة ممكنة (سيظل يتم تطبيق القيود الحرارية).</string>
-    <string name="renderer_asynchronous_gpu_emulation">محاكاة غير متزامنة لوحدة معالجة الرسومات</string>
-    <string name="renderer_asynchronous_gpu_emulation_description">يُشغّل هذا الخيار محاكاة وحدة معالجة الرسومات بشكل غير متزامن لتقليل توقف وحدة المعالجة المركزية وتحسين الإنتاجية. عطّل هذا الخيار فقط في حال واجهت مشاكل متعلقة بالتوقيت.</string>
-    <string name="renderer_async_presentation">عرض غير متزامن</string>
-    <string name="renderer_async_presentation_description">يحسّن الأداء بشكل طفيف عن طريق نقل عملية العرض إلى خيط معالجة منفصل لوحدة المعالجة المركزية.</string>
     <string name="renderer_reactive_flushing">استخدم التنظيف التفاعلي</string>
     <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>
@@ -686,8 +682,8 @@
     <string name="import_success">تم الاستيراد بنجاح</string>
     <string name="export_success">تم التصدير بنجاح</string>
     <string name="start">بدء</string>
-    <string name="global">الإعدادات العامة</string>
-    <string name="custom">الإعدادات المخصصة للعبة</string>
+    <string name="global">المتغيرات العامة</string>
+    <string name="custom">مخصص</string>
     <string name="import_complete">اكتمل الاستيراد</string>
     <string name="use_global_setting">استخدام الإعداد العام</string>
     <string name="operation_completed_successfully">اكتملت العملية بنجاح</string>
@@ -766,7 +762,7 @@
     <string name="info_description">معرف العنوان، المطور، الإصدار</string>
     <string name="per_game_settings">الإعدادات الخاصة للعبة</string>
     <string name="per_game_settings_description">تعديل الإعدادات الخاصة بهذه اللعبة</string>
-    <string name="launch_options">إعدادات التشغيل</string>
+    <string name="launch_options">تشغيل الإعدادات</string>
     <string name="path">المسار</string>
     <string name="program_id">معرف العنوان</string>
     <string name="developer">المطور</string>
@@ -1126,7 +1122,7 @@
     <string name="enable_overlay_description">تمكين التطبيق الصغير المدمج في Horizon. اضغط مع الاستمرار على زر الشاشة الرئيسية لمدة 1 ثانية لإظهاره.</string>
 
     <!-- Profile Management -->
-    <string name="profile_manager">إدارة ملف التعريف</string>
+    <string name="profile_manager">مدير الملف الشخصي</string>
     <string name="profile_manager_description">إدارة ملفات تعريف المستخدمين</string>
     <string name="profile_add_user">إضافة مستخدم</string>
     <string name="profile_new_user">مستخدم جديد</string>

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

@@ -351,7 +351,7 @@
         <item quantity="other">%d archivos de guardado importados con éxito.</item>
     </plurals>
     <string name="no_save_data_found">No se encontraron datos de guardado</string>
-    <string name="verify_installed_content">Verificar los contenidos instalados</string>
+    <string name="verify_installed_content">Verificar contenido instalado</string>
     <string name="verify_installed_content_description">Comprueba todo el contenido instalado por si hubiese alguno corrupto</string>
 
     <string name="keys_missing">Faltan las claves de encriptación</string>
@@ -480,10 +480,6 @@
     <string name="use_disk_shader_cache_description">Reduce los tirones almacenando y cargando los sombreadores generados.</string>
     <string name="renderer_force_max_clock">Forzar velocidad al máximo (solo Adreno)</string>
     <string name="renderer_force_max_clock_description">Fuerza a la GPU a ejecutarse a la velocidad máxima de reloj posible (se seguirán aplicando restricciones térmicas).</string>
-    <string name="renderer_asynchronous_gpu_emulation">Emulación de GPU asíncrona</string>
-    <string name="renderer_asynchronous_gpu_emulation_description">Ejecuta la emulación de la GPU de forma asíncrona para reducir los bloqueos de la CPU y mejorar el rendimiento. Desactiva esta opción solo si experimentas problemas de sincronización.</string>
-    <string name="renderer_async_presentation">Presentación asíncrona</string>
-    <string name="renderer_async_presentation_description">Mejora ligeramente el rendimiento al mover la presentación a un hilo independiente de la CPU.</string>
     <string name="renderer_reactive_flushing">Usar limpieza reactiva</string>
     <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>

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

@@ -55,7 +55,7 @@
     <string name="stats_overlay_options_description">配置性能统计叠加层中显示的信息</string>
     <string name="show_fps">显示帧率</string>
     <string name="show_fps_description">显示当前帧率</string>
-    <string name="show_frametime">显示 Frametime</string>
+    <string name="show_frametime">显示帧时间</string>
     <string name="show_app_ram_usage">显示应用内存使用情况</string>
     <string name="show_app_ram_usage_description">显示模拟器内存用量</string>
     <string name="show_system_ram_usage">显示系统内存使用情况</string>

src/common/lru_cache.h

@@ -1,139 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-later
-
-#pragma once
-
-#include <deque>
-#include <memory>
-#include <type_traits>
-
-#include "common/common_types.h"
-
-namespace Common {
-
-template <class Traits>
-class LeastRecentlyUsedCache {
-    using ObjectType = typename Traits::ObjectType;
-    using TickType = typename Traits::TickType;
-
-    struct Item {
-        ObjectType obj;
-        TickType tick;
-        Item* next{};
-        Item* prev{};
-    };
-
-public:
-    LeastRecentlyUsedCache() : first_item{}, last_item{} {}
-    ~LeastRecentlyUsedCache() = default;
-
-    size_t Insert(ObjectType obj, TickType tick) {
-        const auto new_id = Build();
-        auto& item = item_pool[new_id];
-        item.obj = obj;
-        item.tick = tick;
-        Attach(item);
-        return new_id;
-    }
-
-    void Touch(size_t id, TickType tick) {
-        auto& item = item_pool[id];
-        if (item.tick >= tick) {
-            return;
-        }
-        item.tick = tick;
-        if (&item == last_item) {
-            return;
-        }
-        Detach(item);
-        Attach(item);
-    }
-
-    void Free(size_t id) {
-        auto& item = item_pool[id];
-        Detach(item);
-        item.prev = nullptr;
-        item.next = nullptr;
-        free_items.push_back(id);
-    }
-
-    template <typename Func>
-    void ForEachItemBelow(TickType tick, Func&& func) {
-        static constexpr bool RETURNS_BOOL =
-            std::is_same_v<std::invoke_result<Func, ObjectType>, bool>;
-        Item* iterator = first_item;
-        while (iterator) {
-            if (static_cast<s64>(tick) - static_cast<s64>(iterator->tick) < 0) {
-                return;
-            }
-            Item* next = iterator->next;
-            if constexpr (RETURNS_BOOL) {
-                if (func(iterator->obj)) {
-                    return;
-                }
-            } else {
-                func(iterator->obj);
-            }
-            iterator = next;
-        }
-    }
-
-private:
-    size_t Build() {
-        if (free_items.empty()) {
-            const size_t item_id = item_pool.size();
-            auto& item = item_pool.emplace_back();
-            item.next = nullptr;
-            item.prev = nullptr;
-            return item_id;
-        }
-        const size_t item_id = free_items.front();
-        free_items.pop_front();
-        auto& item = item_pool[item_id];
-        item.next = nullptr;
-        item.prev = nullptr;
-        return item_id;
-    }
-
-    void Attach(Item& item) {
-        if (!first_item) {
-            first_item = &item;
-        }
-        if (!last_item) {
-            last_item = &item;
-        } else {
-            item.prev = last_item;
-            last_item->next = &item;
-            item.next = nullptr;
-            last_item = &item;
-        }
-    }
-
-    void Detach(Item& item) {
-        if (item.prev) {
-            item.prev->next = item.next;
-        }
-        if (item.next) {
-            item.next->prev = item.prev;
-        }
-        if (&item == first_item) {
-            first_item = item.next;
-            if (first_item) {
-                first_item->prev = nullptr;
-            }
-        }
-        if (&item == last_item) {
-            last_item = item.prev;
-            if (last_item) {
-                last_item->next = nullptr;
-            }
-        }
-    }
-
-    std::deque<Item> item_pool;
-    std::deque<size_t> free_items;
-    Item* first_item{};
-    Item* last_item{};
-};
-
-} // namespace Common

src/core/hle/kernel/svc/svc_debug_string.cpp

@@ -6,8 +6,6 @@
 
 #include <condition_variable>
 #include <mutex>
-#include <optional>
-#include <stop_token>
 #include <thread>
 
 #include "core/core.h"
@@ -20,53 +18,59 @@ namespace Kernel::Svc {
 constexpr auto MAX_MSG_TIME = std::chrono::milliseconds(250);
 const auto MAX_MSG_SIZE = 0x1000;
 
+static std::string msg_buffer;
+static std::mutex msg_mutex;
+static std::condition_variable msg_cv;
+static std::chrono::steady_clock::time_point last_msg_time;
+static bool worker_running = true;
+static std::unique_ptr<std::thread> flush_thread;
+static std::once_flag start_flag;
+
+static void FlushDbgLoop() {
+    while (true) {
+        std::unique_lock lock(msg_mutex);
+
+        msg_cv.wait(lock, [] { return !msg_buffer.empty() || !worker_running; });
+        if (!worker_running && msg_buffer.empty()) break;
+
+        auto timeout = last_msg_time + MAX_MSG_TIME;
+        bool woke_early = msg_cv.wait_until(lock, timeout, [] {
+            return msg_buffer.size() >= MAX_MSG_SIZE || !worker_running;
+        });
+
+        if (!woke_early || msg_buffer.size() >= MAX_MSG_SIZE || !worker_running) {
+            if (!msg_buffer.empty()) {
+                // Remove trailing newline as LOG_INFO adds that anyways
+                if (msg_buffer.back() == '\n')
+                    msg_buffer.pop_back();
+
+                LOG_INFO(Debug_Emulated, "\n{}", msg_buffer);
+                msg_buffer.clear();
+            }
+            if (!worker_running) break;
+        }
+    }
+}
+
 /// Used to output a message on a debug hardware unit - does nothing on a retail unit
 Result OutputDebugString(Core::System& system, u64 address, u64 len) {
-    static struct DebugFlusher {
-        std::string msg_buffer;
-        std::mutex msg_mutex;
-        std::condition_variable msg_cv;
-        std::chrono::steady_clock::time_point last_msg_time;
-        std::optional<std::jthread> thread;
-    } flusher_data;
     R_SUCCEED_IF(len == 0);
-    // Only start the thread the very first time this function is called
-    if (!flusher_data.thread) {
-        flusher_data.thread.emplace([](std::stop_token stop_token) {
-            while (!stop_token.stop_requested()) {
-                std::unique_lock lock(flusher_data.msg_mutex);
-                flusher_data.msg_cv.wait(lock, [&stop_token] {
-                    return !flusher_data.msg_buffer.empty() || stop_token.stop_requested();
-                });
-                if (stop_token.stop_requested() && flusher_data.msg_buffer.empty())
-                    break;
-                auto timeout = flusher_data.last_msg_time + MAX_MSG_TIME;
-                bool woke_early = flusher_data.msg_cv.wait_until(lock, timeout, [&stop_token] {
-                    return flusher_data.msg_buffer.size() >= MAX_MSG_SIZE || stop_token.stop_requested();
-                });
-                if (!woke_early || flusher_data.msg_buffer.size() >= MAX_MSG_SIZE || stop_token.stop_requested()) {
-                    if (!flusher_data.msg_buffer.empty()) {
-                        // Remove trailing newline as LOG_INFO adds that anyways
-                        if (flusher_data.msg_buffer.back() == '\n')
-                            flusher_data.msg_buffer.pop_back();
 
-                        LOG_INFO(Debug_Emulated, "\n{}", flusher_data.msg_buffer);
-                        flusher_data.msg_buffer.clear();
-                    }
-                    if (stop_token.stop_requested()) break;
-                }
-            }
-            flusher_data.msg_cv.notify_all();
-        });
-    }
+    // Only start the thread the very first time this function is called
+    std::call_once(start_flag, [] {
+        flush_thread = std::make_unique<std::thread>(FlushDbgLoop);
+    });
+
     {
-        std::lock_guard lock(flusher_data.msg_mutex);
-        const auto old_size = flusher_data.msg_buffer.size();
-        flusher_data.msg_buffer.resize(old_size + len);
-        GetCurrentMemory(system.Kernel()).ReadBlock(address, flusher_data.msg_buffer.data() + old_size, len);
-        flusher_data.last_msg_time = std::chrono::steady_clock::now();
+        std::lock_guard lock(msg_mutex);
+        const auto old_size = msg_buffer.size();
+        msg_buffer.resize(old_size + len);
+        GetCurrentMemory(system.Kernel()).ReadBlock(address, msg_buffer.data() + old_size, len);
+
+        last_msg_time = std::chrono::steady_clock::now();
     }
-    flusher_data.msg_cv.notify_one();
+
+    msg_cv.notify_one();
     R_SUCCEED();
 }
 
@@ -78,4 +82,16 @@ Result OutputDebugString64From32(Core::System& system, uint32_t debug_str, uint3
     R_RETURN(OutputDebugString(system, debug_str, len));
 }
 
+struct BufferAutoFlush {
+    ~BufferAutoFlush() {
+        {
+            std::lock_guard lock(msg_mutex);
+            worker_running = false;
+        }
+        msg_cv.notify_all();
+        if (flush_thread && flush_thread->joinable()) flush_thread->join();
+    }
+};
+static BufferAutoFlush auto_flusher;
+
 } // namespace Kernel::Svc

src/dynarmic/src/dynarmic/CMakeLists.txt

@@ -367,6 +367,7 @@ if (BOOST_NO_HEADERS)
 else()
     target_link_libraries(dynarmic PRIVATE Boost::headers)
 endif()
+
 if (DYNARMIC_USE_LLVM)
     target_include_directories(dynarmic PRIVATE ${LLVM_INCLUDE_DIRS})
     target_compile_definitions(dynarmic PRIVATE DYNARMIC_USE_LLVM=1 ${LLVM_DEFINITIONS})
@@ -384,7 +385,4 @@ endif()
 if (CMAKE_SYSTEM_NAME STREQUAL "Windows")
     target_compile_definitions(dynarmic PRIVATE FMT_USE_WINDOWS_H=0)
 endif()
-if (NOT DEFINED xbyak_ADDED)
-    target_compile_definitions(dynarmic PRIVATE XBYAK_BUNDLED=1)
-endif()
 target_compile_definitions(dynarmic PRIVATE FMT_USE_USER_DEFINED_LITERALS=1)

src/dynarmic/src/dynarmic/backend/x64/block_of_code.cpp

@@ -519,6 +519,10 @@ void BlockOfCode::LoadRequiredFlagsForCondFromRax(IR::Cond cond) {
     }
 }
 
+Xbyak::Address BlockOfCode::Const(const Xbyak::AddressFrame& frame, u64 lower, u64 upper) {
+    return constant_pool.GetConstant(frame, lower, upper);
+}
+
 CodePtr BlockOfCode::GetCodeBegin() const {
     return code_begin;
 }

src/dynarmic/src/dynarmic/backend/x64/block_of_code.h

@@ -24,7 +24,6 @@
 #include "dynarmic/common/cast_util.h"
 #include "dynarmic/interface/halt_reason.h"
 #include "dynarmic/ir/cond.h"
-#include "xbyak/xbyak.h"
 
 namespace Dynarmic::Backend::X64 {
 
@@ -102,8 +101,8 @@ public:
     }
 
     /// Code emitter: Calls the lambda. Lambda must not have any captures.
-    template<typename F>
-    void CallLambda(F l) {
+    template<typename Lambda>
+    void CallLambda(Lambda l) {
         CallFunction(Common::FptrCast(l));
     }
 
@@ -125,35 +124,22 @@ public:
         }
     }
 
-    // Xbyak benefits slightly from not having & references on some of its ctors, however one main
-    // disadvantage is that this breaks ABI slightly because we need to hijack the main privated ctor, hence
-    // we define two paths, one for gentoo (non bundled) and one for our custom bundled set with patch
-#ifdef XBYAK_BUNDLED
-    [[nodiscard]] Xbyak::Address Const(const Xbyak::AddressFrame frame, u64 lower, u64 upper = 0) {
-        return constant_pool.GetConstant(Xbyak::AddressFrame(frame.bit_, frame.broadcast_), lower, upper);
-    }
+    Xbyak::Address Const(const Xbyak::AddressFrame& frame, u64 lower, u64 upper = 0);
+
     template<size_t esize>
-    [[nodiscard]] Xbyak::Address BConst(const Xbyak::AddressFrame frame, u64 value) {
-        return Const(Xbyak::AddressFrame(frame.bit_, frame.broadcast_), mcl::bit::replicate_element<u64>(esize, value), mcl::bit::replicate_element<u64>(esize, value));
+    Xbyak::Address BConst(const Xbyak::AddressFrame& frame, u64 value) {
+        return Const(frame, mcl::bit::replicate_element<u64>(esize, value),
+                     mcl::bit::replicate_element<u64>(esize, value));
     }
-#else
-    [[nodiscard]] Xbyak::Address Const(const Xbyak::AddressFrame& frame, u64 lower, u64 upper = 0) {
-        return constant_pool.GetConstant(frame, lower, upper);
-    }
-    template<size_t esize>
-    [[nodiscard]] Xbyak::Address BConst(const Xbyak::AddressFrame& frame, u64 value) {
-        return Const(frame, mcl::bit::replicate_element<u64>(esize, value), mcl::bit::replicate_element<u64>(esize, value));
-    }
-#endif
 
     CodePtr GetCodeBegin() const;
     size_t GetTotalCodeSize() const;
 
-    [[nodiscard]] const void* GetReturnFromRunCodeAddress() const {
+    const void* GetReturnFromRunCodeAddress() const {
         return return_from_run_code[0];
     }
 
-    [[nodiscard]] const void* GetForceReturnFromRunCodeAddress() const {
+    const void* GetForceReturnFromRunCodeAddress() const {
         return return_from_run_code[FORCE_RETURN];
     }
 

src/dynarmic/src/dynarmic/backend/x64/constant_pool.cpp

@@ -25,11 +25,7 @@ ConstantPool::ConstantPool(BlockOfCode& code, size_t size)
         reinterpret_cast<ConstantT*>(code.AllocateFromCodeSpace(size)), size / align_size);
 }
 
-#ifdef XBYAK_BUNDLED
-Xbyak::Address ConstantPool::GetConstant(const Xbyak::AddressFrame frame, u64 lower, u64 upper) {
-#else
 Xbyak::Address ConstantPool::GetConstant(const Xbyak::AddressFrame& frame, u64 lower, u64 upper) {
-#endif
     const auto constant = ConstantT(lower, upper);
     auto iter = constant_info.find(constant);
     if (iter == constant_info.end()) {

src/dynarmic/src/dynarmic/backend/x64/constant_pool.h

@@ -29,11 +29,7 @@ class ConstantPool final {
 public:
     ConstantPool(BlockOfCode& code, size_t size);
 
-#ifdef XBYAK_BUNDLED
-    Xbyak::Address GetConstant(const Xbyak::AddressFrame frame, u64 lower, u64 upper = 0);
-#else
     Xbyak::Address GetConstant(const Xbyak::AddressFrame& frame, u64 lower, u64 upper = 0);
-#endif
 
 private:
     static constexpr size_t align_size = 16;  // bytes

src/dynarmic/src/dynarmic/backend/x64/emit_x64_packed.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 /* This file is part of the dynarmic project.
@@ -595,8 +595,7 @@ void EmitX64::EmitPackedHalvingSubAddS16(EmitContext& ctx, IR::Inst* inst) {
     EmitPackedSubAdd(code, ctx, inst, false, true, true);
 }
 
-template<typename F>
-static void EmitPackedOperation(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, F&& fn) {
+static void EmitPackedOperation(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, void (Xbyak::CodeGenerator::*fn)(const Xbyak::Mmx& mmx, const Xbyak::Operand&)) {
     auto args = ctx.reg_alloc.GetArgumentInfo(inst);
 
     const Xbyak::Xmm xmm_a = ctx.reg_alloc.UseScratchXmm(code, args[0]);

src/dynarmic/src/dynarmic/backend/x64/emit_x64_vector_saturation.cpp

@@ -23,8 +23,7 @@ using namespace Xbyak::util;
 
 namespace {
 
-template<typename F1, typename F2, typename F3>
-static void EmitVectorSaturatedNative(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, F1&& saturated_fn, F2&& unsaturated_fn, F3&& sub_fn) {
+void EmitVectorSaturatedNative(BlockOfCode& code, EmitContext& ctx, IR::Inst* inst, void (Xbyak::CodeGenerator::*saturated_fn)(const Xbyak::Mmx& mmx, const Xbyak::Operand&), void (Xbyak::CodeGenerator::*unsaturated_fn)(const Xbyak::Mmx& mmx, const Xbyak::Operand&), void (Xbyak::CodeGenerator::*sub_fn)(const Xbyak::Mmx& mmx, const Xbyak::Operand&)) {
     auto args = ctx.reg_alloc.GetArgumentInfo(inst);
 
     const Xbyak::Xmm result = ctx.reg_alloc.UseScratchXmm(code, args[0]);

src/dynarmic/src/dynarmic/backend/x64/hostloc.h

@@ -8,6 +8,8 @@
 #pragma once
 
 #include <bitset>
+#include <xbyak/xbyak.h>
+
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "dynarmic/backend/x64/xbyak.h"

src/qt_common/config/shared_translation.cpp

@@ -169,8 +169,6 @@ std::unique_ptr<TranslationMap> InitializeTranslations(QObject* parent) {
            tr("Runs an additional optimization pass over generated SPIRV shaders.\n"
               "Will increase time required for shader compilation.\nMay slightly improve "
               "performance.\nThis feature is experimental."));
-    INSERT(Settings, use_asynchronous_gpu_emulation, tr("Use asynchronous GPU emulation"),
-           tr("Uses an extra CPU thread for rendering.\nThis option should always remain enabled."));
     INSERT(Settings, nvdec_emulation, tr("NVDEC emulation:"),
            tr("Specifies how videos should be decoded.\nIt can either use the CPU or the GPU for "
               "decoding, or perform no decoding at all (black screen on videos).\n"

src/video_core/buffer_cache/buffer_base.h

@@ -109,12 +109,12 @@ public:
         return static_cast<u32>(other_cpu_addr - cpu_addr);
     }
 
-    size_t getLRUID() const noexcept {
-        return lru_id;
+    u64 GetFrameTick() const noexcept {
+        return frame_tick;
     }
 
-    void setLRUID(size_t lru_id_) {
-        lru_id = lru_id_;
+    void SetFrameTick(u64 tick) noexcept {
+        frame_tick = tick;
     }
 
     size_t SizeBytes() const {
@@ -125,7 +125,7 @@ private:
     VAddr cpu_addr = 0;
     BufferFlagBits flags{};
     int stream_score = 0;
-    size_t lru_id = SIZE_MAX;
+    u64 frame_tick = 0;
     size_t size_bytes = 0;
 };
 

src/video_core/buffer_cache/buffer_cache.h

@@ -58,17 +58,22 @@ void BufferCache<P>::RunGarbageCollector() {
     const bool aggressive_gc = total_used_memory >= critical_memory;
     const u64 ticks_to_destroy = aggressive_gc ? 60 : 120;
     int num_iterations = aggressive_gc ? 64 : 32;
-    const auto clean_up = [this, &num_iterations](BufferId buffer_id) {
+    const u64 threshold = frame_tick - ticks_to_destroy;
+    boost::container::small_vector<BufferId, 64> expired;
+    for (auto [id, buffer] : slot_buffers) {
+        if (buffer->GetFrameTick() < threshold) {
+            expired.push_back(id);
+        }
+    }
+    for (const auto buffer_id : expired) {
         if (num_iterations == 0) {
-            return true;
+            break;
         }
         --num_iterations;
         auto& buffer = slot_buffers[buffer_id];
         DownloadBufferMemory(buffer);
         DeleteBuffer(buffer_id);
-        return false;
-    };
-    lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, clean_up);
+    }
 }
 
 template <class P>
@@ -1595,10 +1600,9 @@ void BufferCache<P>::ChangeRegister(BufferId buffer_id) {
     const auto size = buffer.SizeBytes();
     if (insert) {
         total_used_memory += Common::AlignUp(size, 1024);
-        buffer.setLRUID(lru_cache.Insert(buffer_id, frame_tick));
+        buffer.SetFrameTick(frame_tick);
     } else {
         total_used_memory -= Common::AlignUp(size, 1024);
-        lru_cache.Free(buffer.getLRUID());
     }
     const DAddr device_addr_begin = buffer.CpuAddr();
     const DAddr device_addr_end = device_addr_begin + size;
@@ -1616,7 +1620,7 @@ void BufferCache<P>::ChangeRegister(BufferId buffer_id) {
 template <class P>
 void BufferCache<P>::TouchBuffer(Buffer& buffer, BufferId buffer_id) noexcept {
     if (buffer_id != NULL_BUFFER_ID) {
-        lru_cache.Touch(buffer.getLRUID(), frame_tick);
+        buffer.SetFrameTick(frame_tick);
     }
 }
 

src/video_core/buffer_cache/buffer_cache_base.h

@@ -23,7 +23,6 @@
 #include "common/common_types.h"
 #include "common/div_ceil.h"
 #include "common/literals.h"
-#include "common/lru_cache.h"
 #include "common/range_sets.h"
 #include "common/scope_exit.h"
 #include "common/settings.h"
@@ -506,11 +505,6 @@ private:
     size_t immediate_buffer_capacity = 0;
     Common::ScratchBuffer<u8> immediate_buffer_alloc;
 
-    struct LRUItemParams {
-        using ObjectType = BufferId;
-        using TickType = u64;
-    };
-    Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
     u64 frame_tick = 0;
     u64 total_used_memory = 0;
     u64 minimum_memory = 0;

src/video_core/dma_pusher.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 2021 yuzu Emulator Project
@@ -122,7 +122,35 @@ void DmaPusher::ProcessCommands(std::span<const CommandHeader> commands) {
             dma_state.is_last_call = true;
             index += max_write;
         } else if (dma_state.method_count) {
-            auto const command_header = commands[index]; //can copy
+            if (!dma_state.non_incrementing && !dma_increment_once &&
+                dma_state.method >= non_puller_methods) {
+                auto subchannel = subchannels[dma_state.subchannel];
+                const u32 available = u32(std::min<size_t>(
+                    index + dma_state.method_count, commands.size()) - index);
+                u32 batch = 0;
+                u32 method = dma_state.method;
+                while (batch < available) {
+                    const bool needs_exec =
+                        (method < Engines::EngineInterface::EXECUTION_MASK_TABLE_SIZE)
+                            ? subchannel->execution_mask[method]
+                            : subchannel->execution_mask_default;
+                    if (needs_exec) break;
+                    batch++;
+                    method++;
+                }
+                if (batch > 0) {
+                    auto& sink = subchannel->method_sink;
+                    sink.reserve(sink.size() + batch);
+                    for (u32 j = 0; j < batch; j++) {
+                        sink.emplace_back(dma_state.method + j, commands[index + j].argument);
+                    }
+                    dma_state.method += batch;
+                    dma_state.method_count -= batch;
+                    index += batch;
+                    continue;
+                }
+            }
+            auto const command_header = commands[index];
             dma_state.dma_word_offset = u32(index * sizeof(u32));
             dma_state.is_last_call = dma_state.method_count <= 1;
             CallMethod(command_header.argument);
@@ -181,7 +209,11 @@ void DmaPusher::CallMethod(u32 argument) const {
         });
     } else {
         auto subchannel = subchannels[dma_state.subchannel];
-        if (!subchannel->execution_mask[dma_state.method]) {
+        const bool needs_execution =
+            (dma_state.method < Engines::EngineInterface::EXECUTION_MASK_TABLE_SIZE)
+                ? subchannel->execution_mask[dma_state.method]
+                : subchannel->execution_mask_default;
+        if (!needs_execution) {
             subchannel->method_sink.emplace_back(dma_state.method, argument);
         } else {
             subchannel->ConsumeSink();

src/video_core/engines/engine_interface.h

@@ -6,9 +6,9 @@
 
 #pragma once
 
-#include <bitset>
-#include <limits>
-#include <vector>
+#include <array>
+
+#include <boost/container/small_vector.hpp>
 
 #include "common/common_types.h"
 
@@ -41,8 +41,11 @@ public:
         ConsumeSinkImpl();
     }
 
-    std::bitset<(std::numeric_limits<u16>::max)()> execution_mask{};
-    std::vector<std::pair<u32, u32>> method_sink{};
+    static constexpr size_t EXECUTION_MASK_TABLE_SIZE = 0xE00;
+
+    std::array<u8, EXECUTION_MASK_TABLE_SIZE> execution_mask{};
+    bool execution_mask_default{};
+    boost::container::small_vector<std::pair<u32, u32>, 64> method_sink{};
     bool current_dirty{};
     GPUVAddr current_dma_segment;
 

src/video_core/engines/fermi_2d.cpp

@@ -26,7 +26,7 @@ Fermi2D::Fermi2D(MemoryManager& memory_manager_) : memory_manager{memory_manager
     regs.src.depth = 1;
     regs.dst.depth = 1;
 
-    execution_mask.reset();
+    execution_mask.fill(0);
     execution_mask[FERMI2D_REG_INDEX(pixels_from_memory.src_y0) + 1] = true;
 }
 

src/video_core/engines/kepler_compute.cpp

@@ -18,7 +18,7 @@ namespace Tegra::Engines {
 
 KeplerCompute::KeplerCompute(Core::System& system_, MemoryManager& memory_manager_)
     : system{system_}, memory_manager{memory_manager_}, upload_state{memory_manager, regs.upload} {
-    execution_mask.reset();
+    execution_mask.fill(0);
     execution_mask[KEPLER_COMPUTE_REG_INDEX(exec_upload)] = true;
     execution_mask[KEPLER_COMPUTE_REG_INDEX(data_upload)] = true;
     execution_mask[KEPLER_COMPUTE_REG_INDEX(launch)] = true;

src/video_core/engines/kepler_memory.cpp

@@ -22,7 +22,7 @@ KeplerMemory::~KeplerMemory() = default;
 void KeplerMemory::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_) {
     upload_state.BindRasterizer(rasterizer_);
 
-    execution_mask.reset();
+    execution_mask.fill(0);
     execution_mask[KEPLERMEMORY_REG_INDEX(exec)] = true;
     execution_mask[KEPLERMEMORY_REG_INDEX(data)] = true;
 }

src/video_core/engines/maxwell_3d.cpp

@@ -4,8 +4,10 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <algorithm>
 #include <cstring>
 #include <optional>
+
 #include "common/assert.h"
 #include "common/bit_util.h"
 #include "common/scope_exit.h"
@@ -37,9 +39,10 @@ Maxwell3D::Maxwell3D(Core::System& system_, MemoryManager& memory_manager_)
 {
     dirty.flags.flip();
     InitializeRegisterDefaults();
-    execution_mask.reset();
-    for (size_t i = 0; i < execution_mask.size(); i++)
+    execution_mask.fill(0);
+    for (size_t i = 0; i < EXECUTION_MASK_TABLE_SIZE; i++)
         execution_mask[i] = IsMethodExecutable(u32(i));
+    execution_mask_default = true;
 }
 
 Maxwell3D::~Maxwell3D() = default;
@@ -298,18 +301,27 @@ u32 Maxwell3D::ProcessShadowRam(u32 method, u32 argument) {
 }
 
 void Maxwell3D::ConsumeSinkImpl() {
+    std::stable_sort(method_sink.begin(), method_sink.end(),
+                     [](const auto& a, const auto& b) { return a.first < b.first; });
+
+    const auto sink_size = method_sink.size();
     const auto control = shadow_state.shadow_ram_control;
     if (control == Regs::ShadowRamControl::Track || control == Regs::ShadowRamControl::TrackWithFilter) {
-        for (auto [method, value] : method_sink) {
+        for (size_t i = 0; i < sink_size; ++i) {
+            const auto [method, value] = method_sink[i];
             shadow_state.reg_array[method] = value;
             ProcessDirtyRegisters(method, value);
         }
     } else if (control == Regs::ShadowRamControl::Replay) {
-        for (auto [method, value] : method_sink)
+        for (size_t i = 0; i < sink_size; ++i) {
+            const auto [method, value] = method_sink[i];
             ProcessDirtyRegisters(method, shadow_state.reg_array[method]);
+        }
     } else {
-        for (auto [method, value] : method_sink)
+        for (size_t i = 0; i < sink_size; ++i) {
+            const auto [method, value] = method_sink[i];
             ProcessDirtyRegisters(method, value);
+        }
     }
     method_sink.clear();
 }

src/video_core/engines/maxwell_dma.cpp

@@ -23,7 +23,7 @@ using namespace Texture;
 
 MaxwellDMA::MaxwellDMA(Core::System& system_, MemoryManager& memory_manager_)
     : system{system_}, memory_manager{memory_manager_} {
-    execution_mask.reset();
+    execution_mask.fill(0);
     execution_mask[offsetof(Regs, launch_dma) / sizeof(u32)] = true;
 }
 

src/video_core/renderer_vulkan/vk_swapchain.cpp

@@ -65,10 +65,8 @@ static VkPresentModeKHR ChooseSwapPresentMode(bool has_imm, bool has_mailbox,
             return mode;
         }
     }();
-    if (setting == Settings::VSyncMode::Immediate && !has_imm) {
-        setting = Settings::VSyncMode::Mailbox;
-    }
     if ((setting == Settings::VSyncMode::Mailbox && !has_mailbox) ||
+        (setting == Settings::VSyncMode::Immediate && !has_imm) ||
         (setting == Settings::VSyncMode::FifoRelaxed && !has_fifo_relaxed)) {
         setting = Settings::VSyncMode::Fifo;
     }

src/video_core/texture_cache/image_base.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -102,7 +105,7 @@ struct ImageBase {
     VAddr cpu_addr_end = 0;
 
     u64 modification_tick = 0;
-    size_t lru_index = SIZE_MAX;
+    u64 last_use_tick = 0;
 
     std::array<u32, MAX_MIP_LEVELS> mip_level_offsets{};
 

src/video_core/texture_cache/texture_cache.h

@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <algorithm>
 #include <limits>
 #include <optional>
 #include <bit>
@@ -70,14 +71,10 @@ TextureCache<P>::TextureCache(Runtime& runtime_, Tegra::MaxwellDeviceMemoryManag
             (std::max)((std::min)(device_local_memory - min_vacancy_critical, min_spacing_critical),
                      DEFAULT_CRITICAL_MEMORY));
         minimum_memory = static_cast<u64>((device_local_memory - mem_threshold) / 2);
-
-        lowmemorydevice = false;
     } else {
         expected_memory = DEFAULT_EXPECTED_MEMORY + 512_MiB;
         critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB;
         minimum_memory = 0;
-
-        lowmemorydevice = true;
     }
 
     const bool gpu_unswizzle_enabled = Settings::values.gpu_unswizzle_enabled.GetValue();
@@ -117,25 +114,66 @@ TextureCache<P>::TextureCache(Runtime& runtime_, Tegra::MaxwellDeviceMemoryManag
 }
 
 template <class P>
-void TextureCache<P>::RunGarbageCollector() {
-    bool high_priority_mode = false;
-    bool aggressive_mode = false;
-    u64 ticks_to_destroy = 0;
-    size_t num_iterations = 0;
+void TextureCache<P>::RunAllocationGarbageCollector(size_t requested_bytes) {
+    if (requested_bytes == 0) {
+        return;
+    }
 
-    const auto Configure = [&](bool allow_aggressive) {
-        high_priority_mode = total_used_memory >= expected_memory;
-        aggressive_mode = allow_aggressive && total_used_memory >= critical_memory;
-        ticks_to_destroy = aggressive_mode ? 10ULL : high_priority_mode ? 25ULL : 50ULL;
-        num_iterations = aggressive_mode ? 40 : (high_priority_mode ? 20 : 10);
-    };
+    if (allocation_gc_frame != frame_tick) {
+        allocation_gc_frame = frame_tick;
+        allocation_gc_passes = 0;
+    }
+    if (allocation_gc_passes >= MAX_ALLOCATION_GC_PASSES_PER_FRAME) {
+        return;
+    }
+
+    if (runtime.CanReportMemoryUsage()) {
+        total_used_memory = runtime.GetDeviceMemoryUsage();
+    }
+
+    const u64 request = static_cast<u64>(requested_bytes);
+    const u64 max_u64 = (std::numeric_limits<u64>::max)();
+    const u64 projected_usage = request > (max_u64 - total_used_memory)
+                                    ? max_u64
+                                    : total_used_memory + request;
+    if (projected_usage < expected_memory) {
+        return;
+    }
+
+    RunGarbageCollector();
+    ++allocation_gc_passes;
+
+    if (runtime.CanReportMemoryUsage()) {
+        total_used_memory = runtime.GetDeviceMemoryUsage();
+    }
+
+    const u64 projected_after_gc = request > (max_u64 - total_used_memory)
+                                       ? max_u64
+                                       : total_used_memory + request;
+    if (projected_after_gc >= critical_memory &&
+        allocation_gc_passes < MAX_ALLOCATION_GC_PASSES_PER_FRAME) {
+        RunGarbageCollector();
+        ++allocation_gc_passes;
+
+        if (runtime.CanReportMemoryUsage()) {
+            total_used_memory = runtime.GetDeviceMemoryUsage();
+        }
+    }
+}
+
+template <class P>
+void TextureCache<P>::RunGarbageCollector() {
+    bool high_priority_mode = total_used_memory >= expected_memory;
+    bool aggressive_mode = false;
+    u64 ticks_to_destroy = high_priority_mode ? 25ULL : 50ULL;
+    size_t num_iterations = high_priority_mode ? 20 : 10;
 
     const auto Cleanup = [this, &num_iterations, &high_priority_mode,
                           &aggressive_mode](ImageId image_id) {
         if (num_iterations == 0) {
             return true;
         }
-        --num_iterations;
+        
         auto& image = slot_images[image_id];
 
         // Never delete recently allocated sparse textures (within 3 frames)
@@ -145,28 +183,22 @@ void TextureCache<P>::RunGarbageCollector() {
         }
 
         if (True(image.flags & ImageFlagBits::IsDecoding)) {
-            // This image is still being decoded, deleting it will invalidate the slot
-            // used by the async decoder thread.
             return false;
         }
 
-        // Prioritize large sparse textures for cleanup
-        const bool is_large_sparse = lowmemorydevice &&
-                                     image.info.is_sparse &&
-                                     image.guest_size_bytes >= 256_MiB;
-
-        if (!aggressive_mode && !is_large_sparse &&
-            True(image.flags & ImageFlagBits::CostlyLoad)) {
+        if (!aggressive_mode && True(image.flags & ImageFlagBits::CostlyLoad)) {
             return false;
         }
 
         const bool must_download =
             image.IsSafeDownload() && False(image.flags & ImageFlagBits::BadOverlap);
-        if (!high_priority_mode && !is_large_sparse && must_download) {
+        if (!high_priority_mode && must_download) {
             return false;
         }
 
-        if (must_download && !is_large_sparse) {
+        --num_iterations;
+
+        if (must_download) {
             auto map = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes);
             const auto copies = FixSmallVectorADL(FullDownloadCopies(image.info));
             image.DownloadMemory(map, copies);
@@ -183,7 +215,6 @@ void TextureCache<P>::RunGarbageCollector() {
 
         if (total_used_memory < critical_memory) {
             if (aggressive_mode) {
-                // Sink the aggresiveness.
                 num_iterations >>= 2;
                 aggressive_mode = false;
                 return false;
@@ -196,31 +227,49 @@ void TextureCache<P>::RunGarbageCollector() {
         return false;
     };
 
-    // Aggressively clear massive sparse textures
-    if (total_used_memory >= expected_memory) {
-        lru_cache.ForEachItemBelow(frame_tick, [&](ImageId image_id) {
-            auto& image = slot_images[image_id];
-            // Only target sparse textures that are old enough
-            if (lowmemorydevice &&
-                image.info.is_sparse &&
-                image.guest_size_bytes >= 256_MiB &&
-                image.allocation_tick < frame_tick - 3) {
-                LOG_DEBUG(HW_GPU, "GC targeting old sparse texture at 0x{:X} ({} MiB, age: {} frames)",
-                         image.gpu_addr, image.guest_size_bytes / (1024 * 1024),
-                         frame_tick - image.allocation_tick);
-                return Cleanup(image_id);
-            }
-            return false;
+    const auto SortByAge = [this](auto& vec) {
+        std::sort(vec.begin(), vec.end(), [this](ImageId a, ImageId b) {
+            return slot_images[a].last_use_tick < slot_images[b].last_use_tick;
         });
+    };
+
+    // Single pass: collect all candidates, classified by tier
+    const u64 normal_threshold = frame_tick > ticks_to_destroy ? frame_tick - ticks_to_destroy : 0;
+    const u64 aggressive_threshold = frame_tick > 10 ? frame_tick - 10 : 0;
+    boost::container::small_vector<ImageId, 64> expired;
+    boost::container::small_vector<ImageId, 64> aggressive_expired;
+
+    for (auto [id, image] : slot_images) {
+        if (False(image->flags & ImageFlagBits::Registered)) {
+            continue;
+        }
+        const u64 tick = image->last_use_tick;
+        if (tick < normal_threshold) {
+            expired.push_back(id);
+        } else if (tick < aggressive_threshold) {
+            aggressive_expired.push_back(id);
+        }
     }
 
-    Configure(false);
-    lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, Cleanup);
+    SortByAge(expired);
+    SortByAge(aggressive_expired);
 
-    // If pressure is still too high, prune aggressively.
+    // Tier 1: normal expiration
+    for (const auto image_id : expired) {
+        if (Cleanup(image_id)) {
+            break;
+        }
+    }
+
+    // Tier 2: if still critical, use aggressive threshold with more iterations
     if (total_used_memory >= critical_memory) {
-        Configure(true);
-        lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, Cleanup);
+        aggressive_mode = true;
+        num_iterations = 40;
+        for (const auto image_id : aggressive_expired) {
+            if (Cleanup(image_id)) {
+                break;
+            }
+        }
     }
 }
 
@@ -1196,9 +1245,6 @@ void TextureCache<P>::RefreshContents(Image& image, ImageId image_id) {
     }
 
     image.flags &= ~ImageFlagBits::CpuModified;
-    if( lowmemorydevice && image.info.format == PixelFormat::BC1_RGBA_UNORM && MapSizeBytes(image) >= 256_MiB ) {
-        return;
-    }
 
     TrackImage(image, image_id);
 
@@ -1608,49 +1654,19 @@ bool TextureCache<P>::ScaleDown(Image& image) {
 template <class P>
 ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
                                      RelaxedOptions options) {
+    const size_t requested_size = CalculateGuestSizeInBytes(info);
     std::optional<DAddr> cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr);
     if (!cpu_addr) {
-        const auto size = CalculateGuestSizeInBytes(info);
-        cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr, size);
+        cpu_addr = gpu_memory->GpuToCpuAddress(gpu_addr, requested_size);
         if (!cpu_addr) {
             const DAddr fake_addr = ~(1ULL << 40ULL) + virtual_invalid_space;
-            virtual_invalid_space += Common::AlignUp(size, 32);
+            virtual_invalid_space += Common::AlignUp(requested_size, 32);
             cpu_addr = std::optional<DAddr>(fake_addr);
         }
     }
     ASSERT_MSG(cpu_addr, "Tried to insert an image to an invalid gpu_addr=0x{:x}", gpu_addr);
 
-    // For large sparse textures, aggressively clean up old allocations at same address
-    if (lowmemorydevice && info.is_sparse && CalculateGuestSizeInBytes(info) >= 256_MiB) {
-        const auto alloc_it = image_allocs_table.find(gpu_addr);
-        if (alloc_it != image_allocs_table.end()) {
-            const ImageAllocId alloc_id = alloc_it->second;
-            auto& alloc_images = slot_image_allocs[alloc_id].images;
-
-            // Collect old images at this address that were created more than 2 frames ago
-            boost::container::small_vector<ImageId, 4> to_delete;
-            for (ImageId old_image_id : alloc_images) {
-                Image& old_image = slot_images[old_image_id];
-                if (old_image.info.is_sparse &&
-                    old_image.gpu_addr == gpu_addr &&
-                    old_image.allocation_tick < frame_tick - 2) {  // Try not to delete fresh textures
-                    to_delete.push_back(old_image_id);
-                }
-            }
-
-            // Delete old images immediately
-            for (ImageId old_id : to_delete) {
-                Image& old_image = slot_images[old_id];
-                LOG_DEBUG(HW_GPU, "Immediately deleting old sparse texture at 0x{:X} ({} MiB)",
-                         gpu_addr, old_image.guest_size_bytes / (1024 * 1024));
-                if (True(old_image.flags & ImageFlagBits::Tracked)) {
-                    UntrackImage(old_image, old_id);
-                }
-                UnregisterImage(old_id);
-                DeleteImage(old_id, true);
-            }
-        }
-    }
+    RunAllocationGarbageCollector(requested_size);
 
     const ImageId image_id = JoinImages(info, gpu_addr, *cpu_addr);
     const Image& image = slot_images[image_id];
@@ -1668,25 +1684,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, DA
     ImageInfo new_info = info;
     const size_t size_bytes = CalculateGuestSizeInBytes(new_info);
 
-    // Proactive cleanup for large sparse texture allocations
-    if (lowmemorydevice && new_info.is_sparse && size_bytes >= 256_MiB) {
-        const u64 estimated_alloc_size = size_bytes;
-
-        if (total_used_memory + estimated_alloc_size >= critical_memory) {
-            LOG_DEBUG(HW_GPU, "Large sparse texture allocation ({} MiB) - running aggressive GC. "
-                       "Current memory: {} MiB, Critical: {} MiB",
-                       size_bytes / (1024 * 1024),
-                       total_used_memory / (1024 * 1024),
-                       critical_memory / (1024 * 1024));
-            RunGarbageCollector();
-
-            // If still over threshold after GC, try one more aggressive pass
-            if (total_used_memory + estimated_alloc_size >= critical_memory) {
-                LOG_DEBUG(HW_GPU, "Still critically low on memory, running second GC pass");
-                RunGarbageCollector();
-            }
-        }
-    }
+    RunAllocationGarbageCollector(size_bytes);
 
     const bool broken_views = runtime.HasBrokenTextureViewFormats();
     const bool native_bgr = runtime.HasNativeBgr();
@@ -2027,8 +2025,8 @@ std::pair<u32, u32> TextureCache<P>::PrepareDmaImage(ImageId dst_id, GPUVAddr ba
     const auto& image = slot_images[dst_id];
     const auto base = image.TryFindBase(base_addr);
     PrepareImage(dst_id, mark_as_modified, false);
-    const auto& new_image = slot_images[dst_id];
-    lru_cache.Touch(new_image.lru_index, frame_tick);
+    auto& new_image = slot_images[dst_id];
+    new_image.last_use_tick = frame_tick;
     return std::make_pair(base->level, base->layer);
 }
 
@@ -2377,7 +2375,7 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
         tentative_size = TranscodedAstcSize(tentative_size, image.info.format);
     }
     total_used_memory += Common::AlignUp(tentative_size, 1024);
-    image.lru_index = lru_cache.Insert(image_id, frame_tick);
+    image.last_use_tick = frame_tick;
 
     ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {
         (*channel_state->gpu_page_table)[page].push_back(image_id);
@@ -2411,7 +2409,7 @@ void TextureCache<P>::UnregisterImage(ImageId image_id) {
                "Trying to unregister an already registered image");
     image.flags &= ~ImageFlagBits::Registered;
     image.flags &= ~ImageFlagBits::BadOverlap;
-    lru_cache.Free(image.lru_index);
+
     const auto& clear_page_table =
         [image_id](u64 page, ankerl::unordered_dense::map<u64, std::vector<ImageId>, Common::IdentityHash<u64>>& selected_page_table) {
             const auto page_it = selected_page_table.find(page);
@@ -2740,7 +2738,7 @@ void TextureCache<P>::PrepareImage(ImageId image_id, bool is_modification, bool
     if (is_modification) {
         MarkModification(image);
     }
-    lru_cache.Touch(image.lru_index, frame_tick);
+    image.last_use_tick = frame_tick;
 }
 
 template <class P>

src/video_core/texture_cache/texture_cache_base.h

@@ -22,7 +22,7 @@
 #include "common/common_types.h"
 #include "common/hash.h"
 #include "common/literals.h"
-#include "common/lru_cache.h"
+
 #include <ranges>
 #include "common/scratch_buffer.h"
 #include "common/slot_vector.h"
@@ -120,7 +120,7 @@ class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelI
 
     static constexpr s64 DEFAULT_EXPECTED_MEMORY = 1_GiB + 125_MiB;
     static constexpr s64 DEFAULT_CRITICAL_MEMORY = 1_GiB + 625_MiB;
-    static constexpr size_t GC_EMERGENCY_COUNTS = 2;
+    static constexpr u32 MAX_ALLOCATION_GC_PASSES_PER_FRAME = 2;
 
     using Runtime = typename P::Runtime;
     using Image = typename P::Image;
@@ -310,6 +310,8 @@ private:
     /// Runs the Garbage Collector.
     void RunGarbageCollector();
 
+    void RunAllocationGarbageCollector(size_t requested_bytes);
+
     /// Fills image_view_ids in the image views in indices
     template <bool has_blacklists>
     void FillImageViews(DescriptorTable<TICEntry>& table,
@@ -478,7 +480,6 @@ private:
     u64 minimum_memory;
     u64 expected_memory;
     u64 critical_memory;
-    bool lowmemorydevice = false;
     size_t gpu_unswizzle_maxsize = 0;
     size_t swizzle_chunk_size = 0;
     u32 swizzle_slices_per_batch = 0;
@@ -510,11 +511,7 @@ private:
     std::deque<std::vector<AsyncBuffer>> async_buffers;
     std::deque<AsyncBuffer> async_buffers_death_ring;
 
-    struct LRUItemParams {
-        using ObjectType = ImageId;
-        using TickType = u64;
-    };
-    Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
+
 
  #ifdef YUZU_LEGACY
     static constexpr size_t TICKS_TO_DESTROY = 6;
@@ -532,6 +529,8 @@ private:
 
     u64 modification_tick = 0;
     u64 frame_tick = 0;
+    u64 allocation_gc_frame = (std::numeric_limits<u64>::max)();
+    u32 allocation_gc_passes = 0;
     u64 last_sampler_gc_frame = (std::numeric_limits<u64>::max)();
 
     Common::ThreadWorker texture_decode_worker{1, "TextureDecoder"};