[video_core] Avoid stale macro upload references

Skip FileSystemController factory recreation during game-list repopulation while emulation is powered on.

This avoids poking live FS/VFS state during homebrew self-update and in-place NextLoad flows.

src/android/app/build.gradle.kts

@@ -65,7 +65,7 @@ android {
 
     defaultConfig {
         applicationId = "dev.eden.eden_emulator"
-        minSdk = 33
+        minSdk = 24
         targetSdk = 36
         versionName = getGitVersion()
         versionCode = autoVersion

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

@@ -50,6 +50,7 @@ enum class IntSetting(override val key: String) : AbstractIntSetting {
     GPU_UNSWIZZLE_TEXTURE_SIZE("gpu_unswizzle_texture_size"),
     GPU_UNSWIZZLE_STREAM_SIZE("gpu_unswizzle_stream_size"),
     GPU_UNSWIZZLE_CHUNK_SIZE("gpu_unswizzle_chunk_size"),
+    HOMEBREW_NXLINK_SERVER_MODE("homebrew_nxlink_server_mode"),
     BAT_TEMPERATURE_UNIT("bat_temperature_unit"),
     CABINET_APPLET("cabinet_applet_mode"),
     CONTROLLER_APPLET("controller_applet_mode"),

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

@@ -132,6 +132,15 @@ abstract class SettingsItem(
                     descriptionId = R.string.program_args_description
                 )
             )
+            put(
+                SingleChoiceSetting(
+                    IntSetting.HOMEBREW_NXLINK_SERVER_MODE,
+                    titleId = R.string.nxlink_server_mode,
+                    descriptionId = R.string.nxlink_server_mode_description,
+                    choicesId = R.array.nxlinkServerModeEntries,
+                    valuesId = R.array.nxlinkServerModeValues
+                )
+            )
             put(
                 SwitchSetting(
                     BooleanSetting.RENDERER_USE_SPEED_LIMIT,
@@ -594,7 +603,7 @@ abstract class SettingsItem(
                     IntSetting.ANDROID_PIPELINE_WORKERS,
                     titleId = R.string.pipeline_worker_cores,
                     descriptionId = R.string.pipeline_worker_cores_description,
-                    min = 2,
+                    min = 4,
                     max = 8,
                     units = "cores"
                 )

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

@@ -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: 2024 yuzu Emulator Project
@@ -169,7 +169,7 @@ class InputDialogFragment : DialogFragment() {
         NativeInput.onGamePadButtonEvent(
             controllerData.getGUID(),
             controllerData.getPort(),
-            event.keyCode,
+            InputHandler.getButtonIdFromEvent(event),
             action
         )
         onInputReceived(event.device)

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

@@ -1288,6 +1288,7 @@ class SettingsFragmentPresenter(
 
             add(ShortSetting.DEBUG_KNOBS.key)
             add(StringSetting.PROGRAM_ARGS.key)
+            add(IntSetting.HOMEBREW_NXLINK_SERVER_MODE.key)
 
             if (!NativeConfig.isPerGameConfigLoaded()) {
                 add(HeaderSetting(R.string.gpu_logging_header))

src/android/app/src/main/java/org/yuzu/yuzu_emu/utils/InputHandler.kt

@@ -49,6 +49,12 @@ object InputHandler {
         MotionEvent.AXIS_RTRIGGER
     )
 
+    // Currently, Android doesn't support Joy-Con D-pad buttons. We fall back to the scan code
+    private const val LINUX_BUTTON_DPAD_UP = 0x220
+    private const val LINUX_BUTTON_DPAD_DOWN = 0x221
+    private const val LINUX_BUTTON_DPAD_LEFT = 0x222
+    private const val LINUX_BUTTON_DPAD_RIGHT = 0x223
+
     fun isPhysicalGameController(device: InputDevice?): Boolean {
         device ?: return false
 
@@ -87,12 +93,25 @@ object InputHandler {
         NativeInput.onGamePadButtonEvent(
             controllerData.getGUID(),
             controllerData.getPort(),
-            event.keyCode,
+            getButtonIdFromEvent(event),
             action
         )
         return true
     }
 
+    fun getButtonIdFromEvent(event: KeyEvent): Int {
+        if (event.keyCode == 0) {
+            return when (event.scanCode) {
+                LINUX_BUTTON_DPAD_UP -> KeyEvent.KEYCODE_DPAD_UP
+                LINUX_BUTTON_DPAD_DOWN -> KeyEvent.KEYCODE_DPAD_DOWN
+                LINUX_BUTTON_DPAD_LEFT -> KeyEvent.KEYCODE_DPAD_LEFT
+                LINUX_BUTTON_DPAD_RIGHT -> KeyEvent.KEYCODE_DPAD_RIGHT
+                else -> return 0
+            }
+        }
+        return event.keyCode
+    }
+
     fun dispatchGenericMotionEvent(event: MotionEvent): Boolean {
         val controllerData =
             androidControllers[event.device.controllerNumber] ?: return false

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

@@ -630,6 +630,16 @@
         <item>3</item>
     </integer-array>
 
+    <string-array name="nxlinkServerModeEntries">
+        <item>Disabled</item>
+        <item>Eden Log</item>
+    </string-array>
+
+    <integer-array name="nxlinkServerModeValues">
+        <item>0</item>
+        <item>1</item>
+    </integer-array>
+
     <string-array name="installKeysResults">
         <item>""</item>
         <item>""</item>

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

@@ -436,6 +436,8 @@
 
     <string name="program_args">Homebrew Args</string>
     <string name="program_args_description">Command-line arguments passed to homebrew at launch (e.g. -noglsl).</string>
+    <string name="nxlink_server_mode">nxlink Server</string>
+    <string name="nxlink_server_mode_description">Starts a local nxlink server for homebrew stdout/stderr streams.</string>
 
     <!-- System settings strings -->
     <string name="device_name">Device name</string>

src/common/address_space.inc

@@ -336,22 +336,22 @@ ALLOC_MEMBER(VaType)::Allocate(VaType size) {
         current_linear_alloc_end = alloc_start + size;
     } else { // If linear allocation overflows the AS then find a gap
         if (this->blocks.size() <= 2) {
-            ASSERT_MSG(false, "Unexpected allocator state!");
-        }
-
-        auto search_predecessor{std::next(this->blocks.begin())};
-        auto search_successor{std::next(search_predecessor)};
-
-        while (search_successor != this->blocks.end() &&
-               (search_successor->virt - search_predecessor->virt < size ||
-                search_predecessor->Mapped())) {
-            search_predecessor = search_successor++;
-        }
-
-        if (search_successor != this->blocks.end()) {
-            alloc_start = search_predecessor->virt;
+            alloc_start = virt_start;
         } else {
-            return {}; // AS is full
+            auto search_predecessor{std::next(this->blocks.begin())};
+            auto search_successor{std::next(search_predecessor)};
+
+            while (search_successor != this->blocks.end() &&
+                   (search_successor->virt - search_predecessor->virt < size ||
+                    search_predecessor->Mapped())) {
+                search_predecessor = search_successor++;
+            }
+
+            if (search_successor != this->blocks.end()) {
+                alloc_start = search_predecessor->virt;
+            } else {
+                return {}; // AS is full
+            }
         }
     }
 
@@ -364,6 +364,10 @@ ALLOC_MEMBER(void)::AllocateFixed(VaType virt, VaType size) {
 }
 
 ALLOC_MEMBER(void)::Free(VaType virt, VaType size) {
+    const VaType virt_end = virt + size;
     this->Unmap(virt, size);
+    if (virt_end >= virt && virt_end == current_linear_alloc_end) {
+        current_linear_alloc_end = virt < virt_start ? virt_start : virt;
+    }
 }
 } // namespace Common

src/common/fs/file.cpp

@@ -4,6 +4,8 @@
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <cerrno>
+#include <cstdint>
 #include <vector>
 
 #include "common/assert.h"
@@ -15,8 +17,10 @@
 #include "common/logging.h"
 
 #ifdef _WIN32
+#include <fcntl.h>
 #include <io.h>
 #include <share.h>
+#include <windows.h>
 #else
 #include <unistd.h>
 #endif
@@ -95,10 +99,65 @@ namespace {
     case FileShareFlag::ShareWriteOnly:
         return _SH_DENYRD;
     case FileShareFlag::ShareReadWrite:
+    case FileShareFlag::ShareReadWriteDelete:
         return _SH_DENYNO;
     }
 }
 
+[[nodiscard]] std::FILE* OpenWithWindowsShareDelete(const fs::path& path, FileAccessMode mode,
+                                                    FileType type) {
+    DWORD desired_access{};
+    DWORD creation_disposition{OPEN_EXISTING};
+    int open_flags = type == FileType::BinaryFile ? _O_BINARY : _O_TEXT;
+
+    switch (mode) {
+    case FileAccessMode::Read:
+        desired_access = GENERIC_READ;
+        open_flags |= _O_RDONLY;
+        break;
+    case FileAccessMode::Write:
+        desired_access = GENERIC_WRITE;
+        creation_disposition = CREATE_ALWAYS;
+        open_flags |= _O_WRONLY;
+        break;
+    case FileAccessMode::Append:
+        desired_access = GENERIC_WRITE;
+        creation_disposition = OPEN_ALWAYS;
+        open_flags |= _O_WRONLY | _O_APPEND;
+        break;
+    case FileAccessMode::ReadWrite:
+        desired_access = GENERIC_READ | GENERIC_WRITE;
+        open_flags |= _O_RDWR;
+        break;
+    case FileAccessMode::ReadAppend:
+        desired_access = GENERIC_READ | GENERIC_WRITE;
+        creation_disposition = OPEN_ALWAYS;
+        open_flags |= _O_RDWR | _O_APPEND;
+        break;
+    }
+
+    const auto handle =
+        CreateFileW(path.c_str(), desired_access,
+                    FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE, nullptr,
+                    creation_disposition, FILE_ATTRIBUTE_NORMAL, nullptr);
+    if (handle == INVALID_HANDLE_VALUE) {
+        errno = EACCES;
+        return nullptr;
+    }
+
+    const auto fd = _open_osfhandle(reinterpret_cast<intptr_t>(handle), open_flags);
+    if (fd == -1) {
+        CloseHandle(handle);
+        return nullptr;
+    }
+
+    auto* const file = _wfdopen(fd, AccessModeToWStr(mode, type));
+    if (file == nullptr) {
+        _close(fd);
+    }
+    return file;
+}
+
 #else
 
 /**
@@ -254,7 +313,9 @@ void IOFile::Open(const fs::path& path, FileAccessMode mode, FileType type, File
     errno = 0;
 
 #ifdef _WIN32
-    if (flag != FileShareFlag::ShareNone) {
+    if (flag == FileShareFlag::ShareReadWriteDelete) {
+        file = OpenWithWindowsShareDelete(path, mode, type);
+    } else if (flag != FileShareFlag::ShareNone) {
         file = _wfsopen(path.c_str(), AccessModeToWStr(mode, type), ToWindowsFileShareFlag(flag));
     } else {
         _wfopen_s(&file, path.c_str(), AccessModeToWStr(mode, type));

src/common/fs/fs_types.h

@@ -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
@@ -49,10 +49,11 @@ enum class FileType {
 };
 
 enum class FileShareFlag {
-    ShareNone,      // Provides exclusive access to the file.
-    ShareReadOnly,  // Provides read only shared access to the file.
-    ShareWriteOnly, // Provides write only shared access to the file.
-    ShareReadWrite, // Provides read and write shared access to the file.
+    ShareNone,            // Provides exclusive access to the file.
+    ShareReadOnly,        // Provides read only shared access to the file.
+    ShareWriteOnly,       // Provides write only shared access to the file.
+    ShareReadWrite,       // Provides read and write shared access to the file.
+    ShareReadWriteDelete, // Provides read, write, and delete shared access to the file.
 };
 
 enum class DirEntryFilter {

src/common/settings.h

@@ -702,7 +702,15 @@ struct Values {
 
     // Controls
     InputSetting<std::array<PlayerInput, 10>> players;
-
+    Setting<bool> disable_wgi_xinput{
+        linkage, false, "disable_wgi_xinput", Category::Controls, Specialization::Default,
+// Only read/write disable_wgi_xinput on Windows platforms
+#ifdef _WIN32
+        true
+#else
+        false
+#endif
+    };
     Setting<bool> enable_raw_input{
         linkage, false, "enable_raw_input", Category::Controls, Specialization::Default,
 // Only read/write enable_raw_input on Windows platforms
@@ -832,6 +840,9 @@ struct Values {
     Setting<bool> gpu_log_driver_debug{linkage, true, "gpu_log_driver_debug", Category::Debugging};
     Setting<s32> gpu_log_ring_buffer_size{linkage, 512, "gpu_log_ring_buffer_size",
                                            Category::Debugging};
+    Setting<HomebrewNxlinkServerMode> homebrew_nxlink_server_mode{
+        linkage, HomebrewNxlinkServerMode::Disabled, "homebrew_nxlink_server_mode",
+        Category::Debugging};
 
     SwitchableSetting<u16, true> debug_knobs{linkage,
                                            0,

src/common/settings_enums.h

@@ -159,6 +159,7 @@ ENUM(GpuUnswizzleChunk, VeryLow, Low, Normal, Medium, High)
 ENUM(TemperatureUnits, Celsius, Fahrenheit)
 ENUM(ExtendedDynamicState, Disabled, EDS1, EDS2, EDS3);
 ENUM(GpuLogLevel, Off, Errors, Standard, Verbose, All)
+ENUM(HomebrewNxlinkServerMode, Disabled, EdenLog, HostStdout, File)
 ENUM(GameListMode, TreeView, GridView, CarouselView);
 ENUM(SpeedMode, Standard, Turbo, Slow);
 

src/common/thread.cpp

@@ -1,6 +1,5 @@
 // SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: 2013 Dolphin Emulator Project
 // SPDX-FileCopyrightText: 2014 Citra Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
@@ -40,110 +39,6 @@
 #include <unistd.h>
 #endif
 
-#ifdef __ANDROID__
-#include <sys/resource.h>
-#include <algorithm>
-#include <fstream>
-#include <utility>
-#include <vector>
-
-namespace {
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_URGENT_AUDIO = -19;
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_AUDIO = -16;
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_URGENT_DISPLAY = -8;
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_DISPLAY = -4;
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_FOREGROUND = -2;
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_MORE_FAVORABLE = -1;
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_DEFAULT = 0;
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_LESS_FAVORABLE = 1;
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_BACKGROUND = 10;
-[[maybe_unused]] constexpr int ANDROID_THREAD_PRIORITY_LOWEST = 19;
-
-constexpr size_t ANDROID_MINIMUM_PERFORMANCE_CORES = 4;
-
-cpu_set_t ComputePerformanceCoreMask() {
-    cpu_set_t mask;
-    CPU_ZERO(&mask);
-
-    cpu_set_t allowed;
-    CPU_ZERO(&allowed);
-    if (sched_getaffinity(gettid(), sizeof(allowed), &allowed) != 0) {
-        return mask;
-    }
-
-    std::vector<std::pair<long, int>> cores;
-    const int total = static_cast<int>(std::thread::hardware_concurrency());
-    for (int cpu = 0; cpu < total; ++cpu) {
-        if (!CPU_ISSET(cpu, &allowed)) {
-            continue;
-        }
-        long max_frequency = 0;
-        std::ifstream file("/sys/devices/system/cpu/cpu" + std::to_string(cpu) +
-                           "/cpufreq/cpuinfo_max_freq");
-        if (!file || !(file >> max_frequency) || max_frequency <= 0) {
-            CPU_ZERO(&mask);
-            return mask;
-        }
-        cores.emplace_back(max_frequency, cpu);
-    }
-    if (cores.empty()) {
-        return mask;
-    }
-
-    std::sort(cores.begin(), cores.end(),
-              [](const auto& lhs, const auto& rhs) { return lhs.first > rhs.first; });
-
-    size_t taken = 0;
-    long cluster_frequency = cores.front().first;
-    for (const auto& [frequency, cpu] : cores) {
-        if (frequency != cluster_frequency) {
-            if (taken >= ANDROID_MINIMUM_PERFORMANCE_CORES) {
-                break;
-            }
-            cluster_frequency = frequency;
-        }
-        CPU_SET(cpu, &mask);
-        ++taken;
-    }
-    return mask;
-}
-
-const cpu_set_t& PerformanceCoreMask() {
-    static const cpu_set_t mask = ComputePerformanceCoreMask();
-    return mask;
-}
-
-cpu_set_t ComputeEfficiencyCoreMask() {
-    cpu_set_t mask;
-    CPU_ZERO(&mask);
-
-    const cpu_set_t& performance = PerformanceCoreMask();
-    if (CPU_COUNT(&performance) == 0) {
-        return mask;
-    }
-
-    cpu_set_t allowed;
-    CPU_ZERO(&allowed);
-    if (sched_getaffinity(gettid(), sizeof(allowed), &allowed) != 0) {
-        return mask;
-    }
-
-    const int total = static_cast<int>(std::thread::hardware_concurrency());
-    for (int cpu = 0; cpu < total; ++cpu) {
-        if (CPU_ISSET(cpu, &allowed) && !CPU_ISSET(cpu, &performance)) {
-            CPU_SET(cpu, &mask);
-        }
-    }
-    return mask;
-}
-
-const cpu_set_t& EfficiencyCoreMask() {
-    static const cpu_set_t mask = ComputeEfficiencyCoreMask();
-    return mask;
-}
-} // Anonymous namespace
-#endif
-
 #include "common/cpu_features.h"
 #ifdef ARCHITECTURE_x86_64
 #ifdef _MSC_VER
@@ -183,21 +78,6 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) {
         }
     }();
     set_thread_priority(find_thread(NULL), priority);
-#elif defined(__ANDROID__)
-    const int nice_value = [&]() {
-        switch (new_priority) {
-        case ThreadPriority::Low: return ANDROID_THREAD_PRIORITY_BACKGROUND;
-        case ThreadPriority::Normal: return ANDROID_THREAD_PRIORITY_DEFAULT;
-        case ThreadPriority::High: return ANDROID_THREAD_PRIORITY_DISPLAY;
-        case ThreadPriority::VeryHigh: return ANDROID_THREAD_PRIORITY_URGENT_DISPLAY;
-        case ThreadPriority::Critical: return ANDROID_THREAD_PRIORITY_AUDIO;
-        default: return ANDROID_THREAD_PRIORITY_DEFAULT;
-        }
-    }();
-    if (setpriority(PRIO_PROCESS, static_cast<id_t>(gettid()), nice_value) != 0) {
-        LOG_DEBUG(Common, "Could not set thread nice value to {}: {}", nice_value,
-                  GetLastErrorMsg());
-    }
 #else
     pthread_t this_thread = pthread_self();
     const auto scheduling_type = SCHED_OTHER;
@@ -252,28 +132,29 @@ void SetCurrentThreadName(const char* name) {
 #endif
 }
 
-void SetCurrentThreadToPerformanceCores() {
+void PinCurrentThreadToPerformanceCore(size_t core_id) {
+    ASSERT(core_id < 4);
+    // If we set a flag for a CPU that doesn't exist, the thread may not be allowed to
+    // run in ANY processor!
+    auto const total_cores = std::thread::hardware_concurrency();
+    if (core_id < total_cores) {
 #if defined(__ANDROID__)
-    const cpu_set_t& mask = PerformanceCoreMask();
-    if (CPU_COUNT(&mask) == 0) {
-        return;
-    }
-    if (sched_setaffinity(gettid(), sizeof(mask), &mask) != 0) {
-        LOG_DEBUG(Common, "Could not restrict thread to performance cores: {}", GetLastErrorMsg());
-    }
+        cpu_set_t set;
+        CPU_ZERO(&set);
+        CPU_SET(core_id, &set);
+        sched_setaffinity(pthread_self(), sizeof(set), &set);
+#elif defined(__linux__) || defined(__FreeBSD__)
+        cpu_set_t set;
+        CPU_ZERO(&set);
+        CPU_SET(core_id, &set);
+        pthread_setaffinity_np(pthread_self(), sizeof(set), &set);
+#elif defined(_WIN32)
+        DWORD set = 1UL << core_id;
+        SetThreadAffinityMask(GetCurrentThread(), set);
+#else
+        // No pin functionality implemented
 #endif
-}
-
-void SetCurrentThreadToEfficiencyCores() {
-#if defined(__ANDROID__)
-    const cpu_set_t& mask = EfficiencyCoreMask();
-    if (CPU_COUNT(&mask) == 0) {
-        return;
     }
-    if (sched_setaffinity(gettid(), sizeof(mask), &mask) != 0) {
-        LOG_DEBUG(Common, "Could not restrict thread to efficiency cores: {}", GetLastErrorMsg());
-    }
-#endif
 }
 
 #ifdef ARCHITECTURE_x86_64

src/common/thread.h

@@ -99,14 +99,8 @@ enum class ThreadPriority : u32 {
     Critical = 4,
 };
 
-enum class ThreadPlacement : u32 {
-    Default = 0,
-    Background = 1,
-};
-
 void SetCurrentThreadPriority(ThreadPriority new_priority);
 void SetCurrentThreadName(const char* name);
-void SetCurrentThreadToPerformanceCores();
-void SetCurrentThreadToEfficiencyCores();
+void PinCurrentThreadToPerformanceCore(size_t core_id);
 
 } // namespace Common

src/common/thread_worker.h

@@ -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
 
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
@@ -37,15 +37,10 @@ class StatefulThreadWorker {
     using StateMaker = std::conditional_t<with_state, std::function<StateType()>, DummyCallable>;
 
 public:
-    explicit StatefulThreadWorker(size_t num_workers, std::string name, StateMaker func = {},
-                                  ThreadPlacement placement = ThreadPlacement::Default)
+    explicit StatefulThreadWorker(size_t num_workers, std::string name, StateMaker func = {})
         : workers_queued{num_workers}, thread_name{std::move(name)} {
-        const auto lambda = [this, func, placement](std::stop_token stop_token) {
+        const auto lambda = [this, func](std::stop_token stop_token) {
             Common::SetCurrentThreadName(thread_name.c_str());
-            if (placement == ThreadPlacement::Background) {
-                Common::SetCurrentThreadPriority(ThreadPriority::Low);
-                Common::SetCurrentThreadToEfficiencyCores();
-            }
             {
                 [[maybe_unused]] std::conditional_t<with_state, StateType, int> state{func()};
                 while (!stop_token.stop_requested()) {

src/core/CMakeLists.txt

@@ -1139,6 +1139,8 @@ add_library(core STATIC
     launch_timestamp_cache.h
     loader/deconstructed_rom_directory.cpp
     loader/deconstructed_rom_directory.h
+    loader/homebrew_nxlink.cpp
+    loader/homebrew_nxlink.h
     loader/kip.cpp
     loader/kip.h
     loader/loader.cpp

src/core/core.cpp

@@ -52,6 +52,7 @@
 #include "core/hle/service/set/system_settings_server.h"
 #include "core/hle/service/sm/sm.h"
 #include "core/internal_network/network.h"
+#include "core/loader/homebrew_nxlink.h"
 #include "core/loader/loader.h"
 #include "core/memory.h"
 #include "core/memory/cheat_engine.h"
@@ -397,6 +398,7 @@ struct System::Impl {
 
         stop_event.request_stop();
         core_timing.SyncPause(false);
+        Loader::HomebrewNxlink::StopServer();
         Network::CancelPendingSocketOperations();
         kernel.SuspendEmulation(true);
         kernel.CloseServices();

src/core/cpu_manager.cpp

@@ -174,7 +174,12 @@ void CpuManager::RunThread(std::stop_token token, std::size_t core) {
     std::string name = is_multicore ? ("CPUCore_" + std::to_string(core)) : std::string{"CPUThread"};
     Common::SetCurrentThreadName(name.c_str());
     Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical);
-    Common::SetCurrentThreadToPerformanceCores();
+#ifdef __ANDROID__
+    // Aimed specifically for Snapdragon 8 Elite devices
+    // This kills performance on desktop, but boosts perf for UMA devices
+    // like the S8E. Mediatek and Mali likely won't suffer.
+    Common::PinCurrentThreadToPerformanceCore(core);
+#endif
     auto& data = core_data[core];
     data.host_context = Common::Fiber::ThreadToFiber();
 

src/core/file_sys/fs_filesystem.h

@@ -12,6 +12,7 @@
 namespace FileSys {
 
 enum class OpenMode : u32 {
+    Default = 0,
     Read = (1 << 0),
     Write = (1 << 1),
     AllowAppend = (1 << 2),

src/core/file_sys/vfs/vfs.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 2018 yuzu Emulator Project
@@ -40,7 +40,7 @@ VfsEntryType VfsFilesystem::GetEntryType(std::string_view path_) const {
 
 VirtualFile VfsFilesystem::OpenFile(std::string_view path_, OpenMode perms) {
     const auto path = Common::FS::SanitizePath(path_);
-    return root->GetFileRelative(path);
+    return root->GetFileRelative(path, perms);
 }
 
 VirtualFile VfsFilesystem::CreateFile(std::string_view path_, OpenMode perms) {
@@ -201,7 +201,7 @@ std::string VfsFile::GetFullPath() const {
     return GetContainingDirectory()->GetFullPath() + '/' + GetName();
 }
 
-VirtualFile VfsDirectory::GetFileRelative(std::string_view path) const {
+VirtualFile VfsDirectory::GetFileRelative(std::string_view path, OpenMode perms) const {
     auto vec = Common::FS::SplitPathComponents(path);
     if (vec.empty()) {
         return nullptr;
@@ -224,7 +224,10 @@ VirtualFile VfsDirectory::GetFileRelative(std::string_view path) const {
         return nullptr;
     }
 
-    return dir->GetFile(vec.back());
+    if (perms == OpenMode::Default) {
+        return dir->GetFile(vec.back());
+    }
+    return dir->GetFileRelative(vec.back(), perms);
 }
 
 VirtualFile VfsDirectory::GetFileAbsolute(std::string_view path) const {

src/core/file_sys/vfs/vfs.h

@@ -201,7 +201,8 @@ public:
 
     // Retrieves the file located at path as if the current directory was root. Returns nullptr if
     // not found.
-    virtual VirtualFile GetFileRelative(std::string_view path) const;
+    virtual VirtualFile GetFileRelative(std::string_view path,
+                                        OpenMode perms = OpenMode::Default) const;
     // Calls GetFileRelative(path) on the root of the current directory.
     virtual VirtualFile GetFileAbsolute(std::string_view path) const;
 

src/core/file_sys/vfs/vfs_layered.cpp

@@ -27,9 +27,9 @@ VirtualDir LayeredVfsDirectory::MakeLayeredDirectory(std::vector<VirtualDir> dir
     return VirtualDir(new LayeredVfsDirectory(std::move(dirs), std::move(name)));
 }
 
-VirtualFile LayeredVfsDirectory::GetFileRelative(std::string_view path) const {
+VirtualFile LayeredVfsDirectory::GetFileRelative(std::string_view path, OpenMode perms) const {
     for (const auto& layer : dirs) {
-        const auto file = layer->GetFileRelative(path);
+        const auto file = layer->GetFileRelative(path, perms);
         if (file != nullptr)
             return file;
     }

src/core/file_sys/vfs/vfs_layered.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -20,7 +23,8 @@ public:
     /// Wrapper function to allow for more efficient handling of dirs.size() == 0, 1 cases.
     static VirtualDir MakeLayeredDirectory(std::vector<VirtualDir> dirs, std::string name = "");
 
-    VirtualFile GetFileRelative(std::string_view path) const override;
+    VirtualFile GetFileRelative(std::string_view path,
+                                OpenMode perms = OpenMode::Default) const override;
     VirtualDir GetDirectoryRelative(std::string_view path) const override;
     VirtualFile GetFile(std::string_view file_name) const override;
     VirtualDir GetSubdirectory(std::string_view subdir_name) const override;

src/core/file_sys/vfs/vfs_real.cpp

@@ -12,7 +12,6 @@
 #include "common/fs/file.h"
 #include "common/fs/fs.h"
 #include "common/fs/path_util.h"
-#include "common/logging.h"
 #include "core/file_sys/vfs/vfs.h"
 #include "core/file_sys/vfs/vfs_real.h"
 
@@ -46,17 +45,11 @@ bool IsWithinRoot(std::string_view root, std::string_view full_path) {
 }
 
 constexpr FS::FileAccessMode ModeFlagsToFileAccessMode(OpenMode mode) {
-    switch (mode) {
-    case OpenMode::Read:
-        return FS::FileAccessMode::Read;
-    case OpenMode::Write:
-    case OpenMode::ReadWrite:
-    case OpenMode::AllowAppend:
-    case OpenMode::All:
+    if (True(mode & OpenMode::Write) || True(mode & OpenMode::AllowAppend)) {
         return FS::FileAccessMode::ReadWrite;
-    default:
-        return {};
     }
+
+    return FS::FileAccessMode::Read;
 }
 
 } // Anonymous namespace
@@ -94,9 +87,11 @@ VirtualFile RealVfsFilesystem::OpenFileFromEntry(std::string_view path_, std::op
                                                  std::optional<std::string> parent_path,
                                                  OpenMode perms) {
     const auto path = FS::SanitizePath(path_, FS::DirectorySeparator::PlatformDefault);
+    const auto open_perms = perms == OpenMode::Default ? OpenMode::Read : perms;
     std::scoped_lock lk{list_lock};
 
-    if (auto it = cache.find(path); it != cache.end()) {
+    const CacheKey cache_key{path, open_perms};
+    if (auto it = cache.find(cache_key); it != cache.end()) {
         if (auto file = it->second.lock(); file) {
             return file;
         }
@@ -110,8 +105,9 @@ VirtualFile RealVfsFilesystem::OpenFileFromEntry(std::string_view path_, std::op
     this->InsertReferenceIntoListLocked(*reference);
 
     auto file = std::shared_ptr<RealVfsFile>(
-        new RealVfsFile(*this, std::move(reference), path, perms, size, std::move(parent_path)));
-    cache[path] = file;
+        new RealVfsFile(*this, std::move(reference), path, open_perms, size,
+                        std::move(parent_path)));
+    cache[cache_key] = file;
 
     return file;
 }
@@ -124,7 +120,7 @@ VirtualFile RealVfsFilesystem::CreateFile(std::string_view path_, OpenMode perms
     const auto path = FS::SanitizePath(path_, FS::DirectorySeparator::PlatformDefault);
     {
         std::scoped_lock lk{list_lock};
-        cache.erase(path);
+        CloseCachedFileReferenceLocked(path);
     }
 
     // Current usages of CreateFile expect to delete the contents of an existing file.
@@ -157,8 +153,8 @@ VirtualFile RealVfsFilesystem::MoveFile(std::string_view old_path_, std::string_
     const auto new_path = FS::SanitizePath(new_path_, FS::DirectorySeparator::PlatformDefault);
     {
         std::scoped_lock lk{list_lock};
-        cache.erase(old_path);
-        cache.erase(new_path);
+        CloseCachedFileReferenceLocked(old_path);
+        CloseCachedFileReferenceLocked(new_path);
     }
     if (!FS::RenameFile(old_path, new_path)) {
         return nullptr;
@@ -170,14 +166,15 @@ bool RealVfsFilesystem::DeleteFile(std::string_view path_) {
     const auto path = FS::SanitizePath(path_, FS::DirectorySeparator::PlatformDefault);
     {
         std::scoped_lock lk{list_lock};
-        cache.erase(path);
+        CloseCachedFileReferenceLocked(path);
     }
     return FS::RemoveFile(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::shared_ptr<RealVfsDirectory>(
+        new RealVfsDirectory(*this, path, perms == OpenMode::Default ? OpenMode::Read : perms));
 }
 
 VirtualDir RealVfsFilesystem::CreateDirectory(std::string_view path_, OpenMode perms) {
@@ -222,8 +219,8 @@ std::unique_lock<std::mutex> RealVfsFilesystem::RefreshReference(const std::stri
     if (!reference.file) {
         this->EvictSingleReferenceLocked();
 
-        reference.file =
-            FS::FileOpen(path, ModeFlagsToFileAccessMode(perms), FS::FileType::BinaryFile);
+        reference.file = FS::FileOpen(path, ModeFlagsToFileAccessMode(perms),
+                                      FS::FileType::BinaryFile, FS::FileShareFlag::ShareReadWriteDelete);
         if (reference.file) {
             num_open_files++;
         }
@@ -297,15 +294,45 @@ RealVfsFile::~RealVfsFile() {
     base.DropReference(std::move(reference));
 }
 
+void RealVfsFilesystem::CloseCachedFileReferenceLocked(const std::string& path) {
+    for (auto it = cache.lower_bound(CacheKey{path, OpenMode::Default});
+         it != cache.end() && it->first.first == path;) {
+        const auto cached_file = it->second.lock();
+        if (cached_file) {
+            auto* const real_file = static_cast<RealVfsFile*>(cached_file.get());
+            auto& reference = real_file->reference;
+            if (reference && reference->file) {
+                RemoveReferenceFromListLocked(*reference);
+                reference->file.reset();
+                num_open_files--;
+                InsertReferenceIntoListLocked(*reference);
+            }
+        }
+        it = cache.erase(it);
+    }
+}
+
 std::string RealVfsFile::GetName() const {
 #ifdef __ANDROID__
-    if (path[0] != '/') {
+    if (!path.empty() && path[0] != '/') {
         return FS::Android::GetFilename(path);
     }
 #endif
     return path_components.empty() ? "" : std::string(path_components.back());
 }
 
+std::string RealVfsFile::GetFullPath() const {
+#ifdef __ANDROID__
+    if (!path.empty() && path[0] != '/') {
+        auto out = path;
+        std::replace(out.begin(), out.end(), '\\', '/');
+        return out;
+    }
+#endif
+
+    return VfsFile::GetFullPath();
+}
+
 std::size_t RealVfsFile::GetSize() const {
     if (size) {
         return *size;
@@ -411,13 +438,14 @@ RealVfsDirectory::RealVfsDirectory(RealVfsFilesystem& base_, const std::string&
 
 RealVfsDirectory::~RealVfsDirectory() = default;
 
-VirtualFile RealVfsDirectory::GetFileRelative(std::string_view relative_path) const {
+VirtualFile RealVfsDirectory::GetFileRelative(std::string_view relative_path,
+                                              OpenMode open_perms) const {
     const auto full_path = FS::SanitizePath(path + '/' + std::string(relative_path));
     if (!FS::Exists(full_path) || FS::IsDir(full_path)
         || !IsWithinRoot(FS::SanitizePath(path), full_path)) {
         return nullptr;
     }
-    return base.OpenFile(full_path, perms);
+    return base.OpenFile(full_path, open_perms == OpenMode::Default ? perms : open_perms);
 }
 
 VirtualDir RealVfsDirectory::GetDirectoryRelative(std::string_view relative_path) const {

src/core/file_sys/vfs/vfs_real.h

@@ -10,6 +10,7 @@
 #include <mutex>
 #include <optional>
 #include <string_view>
+#include <utility>
 #include "common/intrusive_list.h"
 #include "core/file_sys/fs_filesystem.h"
 #include "core/file_sys/vfs/vfs.h"
@@ -49,8 +50,9 @@ public:
     bool DeleteDirectory(std::string_view path) override;
 
 private:
+    using CacheKey = std::pair<std::string, OpenMode>;
     using ReferenceListType = Common::IntrusiveListBaseTraits<FileReference>::ListType;
-    std::map<std::string, std::weak_ptr<VfsFile>, std::less<>> cache;
+    std::map<CacheKey, std::weak_ptr<VfsFile>, std::less<>> cache;
     ReferenceListType open_references;
     ReferenceListType closed_references;
     std::mutex list_lock;
@@ -63,6 +65,7 @@ private:
     std::unique_lock<std::mutex> RefreshReference(const std::string& path, OpenMode perms,
                                                   FileReference& reference);
     void DropReference(std::unique_ptr<FileReference>&& reference);
+    void CloseCachedFileReferenceLocked(const std::string& path);
 
 private:
     friend class RealVfsDirectory;
@@ -85,6 +88,7 @@ public:
     ~RealVfsFile() override;
 
     std::string GetName() const override;
+    std::string GetFullPath() const override;
     std::size_t GetSize() const override;
     bool Resize(std::size_t new_size) override;
     VirtualDir GetContainingDirectory() const override;
@@ -115,7 +119,8 @@ class RealVfsDirectory : public VfsDirectory {
 public:
     ~RealVfsDirectory() override;
 
-    VirtualFile GetFileRelative(std::string_view relative_path) const override;
+    VirtualFile GetFileRelative(std::string_view relative_path,
+                                OpenMode perms = OpenMode::Default) const override;
     VirtualDir GetDirectoryRelative(std::string_view relative_path) const override;
     VirtualFile GetFile(std::string_view name) const override;
     VirtualDir GetSubdirectory(std::string_view name) const override;

src/core/hle/kernel/k_page_table_base.h

@@ -670,6 +670,11 @@ public:
     size_t GetHeapRegionSize() const {
         return m_heap_region_end - m_heap_region_start;
     }
+    size_t GetCurrentHeapSize() const {
+        KScopedLightLock lk(m_general_lock);
+
+        return m_current_heap_end - m_heap_region_start;
+    }
     size_t GetAliasRegionSize() const {
         return m_alias_region_end - m_alias_region_start;
     }

src/core/hle/kernel/k_process.cpp

@@ -210,6 +210,11 @@ Result KProcess::Initialize(KernelCore& kernel, const Svc::CreateProcessParamete
     m_arg_pointer = 0;
     m_arg_return_address = 0;
     m_main_thread_handle_addr = 0;
+    m_process_handle_addr = 0;
+    m_homebrew_next_load_path_addr = 0;
+    m_homebrew_next_load_argv_addr = 0;
+    m_is_homebrew_in_place_next_load = false;
+    m_has_homebrew_nxlink_argv_marker = false;
     m_code_size = params.code_num_pages * PageSize;
     m_is_application = True(params.flags & Svc::CreateProcessFlag::IsApplication);
 
@@ -947,6 +952,7 @@ Result KProcess::Run(KernelCore& kernel, s32 priority, size_t stack_size) {
 
         stack_top = stack_bottom + stack_size;
         m_main_thread_stack_size = stack_size;
+        m_main_thread_stack_top = stack_top;
     }
 
     // Ensure our stack is safe to clean up on exit.
@@ -1005,6 +1011,11 @@ Result KProcess::Run(KernelCore& kernel, s32 priority, size_t stack_size) {
         if (GetInteger(m_main_thread_handle_addr) != 0) {
             this->GetMemory().Write32(m_main_thread_handle_addr, thread_handle);
         }
+        if (GetInteger(m_process_handle_addr) != 0) {
+            Handle process_handle;
+            R_TRY(m_handle_table.Add(kernel, std::addressof(process_handle), this));
+            this->GetMemory().Write32(m_process_handle_addr, process_handle);
+        }
     } else {
         main_thread->GetContext().r[0] = 0;
         main_thread->GetContext().r[1] = thread_handle;

src/core/hle/kernel/k_process.h

@@ -6,7 +6,9 @@
 
 #pragma once
 
+#include <array>
 #include <map>
+#include <string_view>
 
 #include "core/arm/arm_interface.h"
 #include "core/file_sys/program_metadata.h"
@@ -87,6 +89,10 @@ private:
     KProcessAddress m_arg_pointer{};
     KProcessAddress m_arg_return_address{};
     KProcessAddress m_main_thread_handle_addr{};
+    KProcessAddress m_process_handle_addr{};
+    KProcessAddress m_homebrew_next_load_path_addr{};
+    KProcessAddress m_homebrew_next_load_argv_addr{};
+    std::array<char, 16> m_homebrew_nxlink_argv_marker{};
     KHandleTable m_handle_table;
     KProcessAddress m_plr_address{};
     ThreadList m_thread_list{};
@@ -112,6 +118,7 @@ private:
 
     size_t m_code_size{};
     size_t m_main_thread_stack_size{};
+    KProcessAddress m_main_thread_stack_top{};
     size_t m_max_process_memory{};
     size_t m_memory_release_hint{};
     s64 m_schedule_count{};
@@ -139,6 +146,8 @@ private:
     bool m_is_suspended : 1 = false;
     bool m_is_immortal : 1 = false;
     bool m_is_handle_table_initialized : 1 = false;
+    bool m_is_homebrew_in_place_next_load : 1 = false;
+    bool m_has_homebrew_nxlink_argv_marker : 1 = false;
 
 private:
     Result StartTermination(KernelCore& kernel);
@@ -231,6 +240,49 @@ public:
     void SetMainThreadHandleAddr(KProcessAddress addr) {
         m_main_thread_handle_addr = addr;
     }
+    void SetProcessHandleAddr(KProcessAddress addr) {
+        m_process_handle_addr = addr;
+    }
+    void SetHomebrewNextLoadBufferAddrs(KProcessAddress path_addr, KProcessAddress argv_addr) {
+        m_homebrew_next_load_path_addr = path_addr;
+        m_homebrew_next_load_argv_addr = argv_addr;
+    }
+    void SetHomebrewInPlaceNextLoad(bool enabled) {
+        m_is_homebrew_in_place_next_load = enabled;
+    }
+    void SetHomebrewNxlinkArgvMarker(std::string_view marker) {
+        if (marker.size() != m_homebrew_nxlink_argv_marker.size()) {
+            m_has_homebrew_nxlink_argv_marker = false;
+            return;
+        }
+
+        marker.copy(m_homebrew_nxlink_argv_marker.data(), m_homebrew_nxlink_argv_marker.size());
+        m_has_homebrew_nxlink_argv_marker = true;
+    }
+    void ClearHomebrewNxlinkArgvMarker() {
+        m_has_homebrew_nxlink_argv_marker = false;
+    }
+    std::string_view GetHomebrewNxlinkArgvMarker() const {
+        if (!m_has_homebrew_nxlink_argv_marker) {
+            return {};
+        }
+        return {m_homebrew_nxlink_argv_marker.data(), m_homebrew_nxlink_argv_marker.size()};
+    }
+    bool IsHomebrewInPlaceNextLoad() const {
+        return m_is_homebrew_in_place_next_load;
+    }
+    KProcessAddress GetHomebrewNextLoadPathAddr() const {
+        return m_homebrew_next_load_path_addr;
+    }
+    KProcessAddress GetHomebrewNextLoadArgvAddr() const {
+        return m_homebrew_next_load_argv_addr;
+    }
+    size_t GetCodeSize() const {
+        return m_code_size;
+    }
+    KProcessAddress GetMainThreadStackTop() const {
+        return m_main_thread_stack_top;
+    }
 
     size_t GetMainStackSize() const {
         return m_main_thread_stack_size;

src/core/hle/kernel/svc.cpp

@@ -1,7 +1,8 @@
-// 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 2023 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-late
+// SPDX-License-Identifier: GPL-2.0-or-later
 
 // This file is automatically generated using svc_generator.py.
 // DO NOT MODIFY IT MANUALLY
@@ -128,7 +129,7 @@ static void SvcWrap_QueryMemory64From32(Core::System& system, std::span<uint64_t
 }
 
 static void SvcWrap_ExitProcess64From32(Core::System& system, std::span<uint64_t, 8> args) {
-    ExitProcess64From32(system);
+    ExitProcess64From32(system, args);
 }
 
 static void SvcWrap_CreateThread64From32(Core::System& system, std::span<uint64_t, 8> args) {
@@ -1298,7 +1299,7 @@ static void SvcWrap_QueryMemory64(Core::System& system, std::span<uint64_t, 8> a
 }
 
 static void SvcWrap_ExitProcess64(Core::System& system, std::span<uint64_t, 8> args) {
-    ExitProcess64(system);
+    ExitProcess64(system, args);
 }
 
 static void SvcWrap_CreateThread64(Core::System& system, std::span<uint64_t, 8> args) {

src/core/hle/kernel/svc.h

@@ -1,7 +1,8 @@
-// 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 2023 yuzu Emulator Project
-// SPDX-License-Identifier: GPL-2.0-or-late
+// SPDX-License-Identifier: GPL-2.0-or-later
 
 // This file is automatically generated using svc_generator.py.
 // DO NOT MODIFY IT MANUALLY
@@ -25,7 +26,7 @@ Result SetMemoryAttribute(Core::System& system, uint64_t address, uint64_t size,
 Result MapMemory(Core::System& system, uint64_t dst_address, uint64_t src_address, uint64_t size);
 Result UnmapMemory(Core::System& system, uint64_t dst_address, uint64_t src_address, uint64_t size);
 Result QueryMemory(Core::System& system, uint64_t out_memory_info, PageInfo* out_page_info, uint64_t address);
-void ExitProcess(Core::System& system);
+void ExitProcess(Core::System& system, std::span<uint64_t, 8> args);
 Result CreateThread(Core::System& system, Handle* out_handle, uint64_t func, uint64_t arg, uint64_t stack_bottom, int32_t priority, int32_t core_id);
 Result StartThread(Core::System& system, Handle thread_handle);
 void ExitThread(Core::System& system);
@@ -146,7 +147,7 @@ Result SetMemoryAttribute64From32(Core::System& system, uint32_t address, uint32
 Result MapMemory64From32(Core::System& system, uint32_t dst_address, uint32_t src_address, uint32_t size);
 Result UnmapMemory64From32(Core::System& system, uint32_t dst_address, uint32_t src_address, uint32_t size);
 Result QueryMemory64From32(Core::System& system, uint32_t out_memory_info, PageInfo* out_page_info, uint32_t address);
-void ExitProcess64From32(Core::System& system);
+void ExitProcess64From32(Core::System& system, std::span<uint64_t, 8> args);
 Result CreateThread64From32(Core::System& system, Handle* out_handle, uint32_t func, uint32_t arg, uint32_t stack_bottom, int32_t priority, int32_t core_id);
 Result StartThread64From32(Core::System& system, Handle thread_handle);
 void ExitThread64From32(Core::System& system);
@@ -267,7 +268,7 @@ Result SetMemoryAttribute64(Core::System& system, uint64_t address, uint64_t siz
 Result MapMemory64(Core::System& system, uint64_t dst_address, uint64_t src_address, uint64_t size);
 Result UnmapMemory64(Core::System& system, uint64_t dst_address, uint64_t src_address, uint64_t size);
 Result QueryMemory64(Core::System& system, uint64_t out_memory_info, PageInfo* out_page_info, uint64_t address);
-void ExitProcess64(Core::System& system);
+void ExitProcess64(Core::System& system, std::span<uint64_t, 8> args);
 Result CreateThread64(Core::System& system, Handle* out_handle, uint64_t func, uint64_t arg, uint64_t stack_bottom, int32_t priority, int32_t core_id);
 Result StartThread64(Core::System& system, Handle thread_handle);
 void ExitThread64(Core::System& system);

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

@@ -4,6 +4,9 @@
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <algorithm>
+#include <vector>
+
 #include "core/core.h"
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/svc.h"
@@ -22,6 +25,179 @@ constexpr bool IsValidSetMemoryPermission(MemoryPermission perm) {
     }
 }
 
+bool IsHomebrewInPlaceNextLoadCodeRange(const KProcess& process, u64 address, u64 size) {
+    if (!process.IsHomebrewInPlaceNextLoad()) {
+        return false;
+    }
+
+    const u64 code_start = GetInteger(process.GetEntryPoint());
+    const size_t code_size = process.GetCodeSize();
+    if (code_start == 0 || code_size == 0 || size > code_size || address < code_start) {
+        return false;
+    }
+
+    return address - code_start <= code_size - size;
+}
+
+struct HomebrewInPlaceMemoryBlock {
+    u64 address;
+    u64 size;
+    KMemoryState state;
+    KMemoryPermission permission;
+    KMemoryAttribute attribute;
+    bool use_process_permission;
+};
+
+Result SetHomebrewInPlaceMemoryPermissionByBlocks(KProcess& process, u64 address, u64 size,
+                                                  MemoryPermission perm, Result original_result) {
+    auto& page_table = process.GetPageTable();
+    const u64 end = address + size;
+    const auto requested_permission = ConvertToKMemoryPermission(perm);
+    std::vector<HomebrewInPlaceMemoryBlock> blocks;
+
+    for (u64 cursor = address; cursor < end;) {
+        KMemoryInfo info;
+        PageInfo page_info;
+        const auto query_result = page_table.QueryInfo(std::addressof(info),
+                                                       std::addressof(page_info), cursor);
+        if (query_result.IsError()) {
+            LOG_WARNING(Kernel_SVC,
+                        "NextLoad in-place: split permission query failed "
+                        "address=0x{:016X}, result={:#X}, original={:#X}",
+                        cursor, query_result.raw, original_result.raw);
+            R_RETURN(original_result);
+        }
+
+        const u64 block_end = (std::min<u64>)(info.GetEndAddress(), end);
+        if (block_end <= cursor) {
+            LOG_WARNING(Kernel_SVC,
+                        "NextLoad in-place: split permission walk stalled "
+                        "cursor=0x{:016X}, block=0x{:016X}/0x{:X}, original={:#X}",
+                        cursor, info.GetAddress(), info.GetSize(), original_result.raw);
+            R_RETURN(original_result);
+        }
+
+        const bool can_reprotect = True(info.GetState() & KMemoryState::FlagCanReprotect);
+        const bool can_process_reprotect = True(info.GetState() & KMemoryState::FlagCode);
+        if (!can_reprotect && !can_process_reprotect) {
+            LOG_WARNING(Kernel_SVC,
+                        "NextLoad in-place: split permission unsupported block "
+                        "address=0x{:016X}, size=0x{:X}, state=0x{:08X}, svc_state={}, "
+                        "perm=0x{:08X}, attr=0x{:08X}, original={:#X}",
+                        cursor, block_end - cursor, static_cast<u32>(info.GetState()),
+                        static_cast<u32>(info.GetSvcState()),
+                        static_cast<u32>(info.GetPermission()),
+                        static_cast<u32>(info.GetAttribute()), original_result.raw);
+            R_RETURN(original_result);
+        }
+
+        blocks.push_back({
+            .address = cursor,
+            .size = block_end - cursor,
+            .state = info.GetState(),
+            .permission = info.GetPermission(),
+            .attribute = info.GetAttribute(),
+            .use_process_permission = !can_reprotect && can_process_reprotect,
+        });
+        cursor = block_end;
+    }
+
+    for (const auto& block : blocks) {
+        if (block.permission == requested_permission) {
+            continue;
+        }
+
+        const auto block_result =
+            block.use_process_permission
+                ? page_table.SetProcessMemoryPermission(block.address, block.size, perm)
+                : page_table.SetMemoryPermission(block.address, block.size, perm);
+        if (block_result.IsError()) {
+            LOG_WARNING(Kernel_SVC,
+                        "NextLoad in-place: split permission block failed "
+                        "address=0x{:016X}, size=0x{:X}, state=0x{:08X}, perm=0x{:08X}, "
+                        "attr=0x{:08X}, result={:#X}, original={:#X}",
+                        block.address, block.size, static_cast<u32>(block.state),
+                        static_cast<u32>(block.permission), static_cast<u32>(block.attribute),
+                        block_result.raw, original_result.raw);
+            R_RETURN(block_result);
+        }
+    }
+
+    R_SUCCEED();
+}
+
+struct HomebrewInPlaceDeviceSharedBlock {
+    u64 address;
+    u64 size;
+    u16 device_use_count;
+};
+
+Result UnlockHomebrewInPlaceDeviceSharedSource(KProcess& process, u64 address, u64 size,
+                                               Result original_result) {
+    auto& page_table = process.GetPageTable();
+    const u64 end = address + size;
+    std::vector<HomebrewInPlaceDeviceSharedBlock> blocks;
+
+    for (u64 cursor = address; cursor < end;) {
+        KMemoryInfo info;
+        PageInfo page_info;
+        const auto query_result = page_table.QueryInfo(std::addressof(info),
+                                                       std::addressof(page_info), cursor);
+        if (query_result.IsError()) {
+            R_RETURN(original_result);
+        }
+
+        const u64 block_end = (std::min<u64>)(info.GetEndAddress(), end);
+        if (block_end <= cursor) {
+            R_RETURN(original_result);
+        }
+
+        const bool can_device_map = True(info.GetState() & KMemoryState::FlagCanDeviceMap);
+        const bool is_clean_memory =
+            can_device_map && info.GetPermission() == KMemoryPermission::UserReadWrite &&
+            info.GetAttribute() == KMemoryAttribute::None && info.m_device_use_count == 0;
+        const bool is_stale_device_shared =
+            can_device_map && info.GetPermission() == KMemoryPermission::UserReadWrite &&
+            info.GetAttribute() == KMemoryAttribute::DeviceShared && info.m_device_use_count > 0;
+        if (is_clean_memory) {
+            cursor = block_end;
+            continue;
+        }
+        if (!is_stale_device_shared) {
+            R_RETURN(original_result);
+        }
+
+        blocks.push_back({
+            .address = cursor,
+            .size = block_end - cursor,
+            .device_use_count = info.m_device_use_count,
+        });
+        cursor = block_end;
+    }
+
+    if (blocks.empty()) {
+        R_RETURN(original_result);
+    }
+
+    for (const auto& block : blocks) {
+        for (u16 unlock = 0; unlock < block.device_use_count; unlock++) {
+            const auto unlock_result =
+                page_table.UnlockForDeviceAddressSpace(block.address, block.size);
+            if (unlock_result.IsError()) {
+                LOG_WARNING(Kernel_SVC,
+                            "NextLoad in-place: device-shared source unlock failed "
+                            "address=0x{:016X}, size=0x{:X}, remaining={}, result={:#X}, "
+                            "original={:#X}",
+                            block.address, block.size, block.device_use_count - unlock - 1,
+                            unlock_result.raw, original_result.raw);
+                R_RETURN(original_result);
+            }
+        }
+    }
+
+    R_SUCCEED();
+}
+
 // Checks if address + size is greater than the given address
 // This can return false if the size causes an overflow of a 64-bit type
 // or if the given size is zero.
@@ -92,11 +268,19 @@ Result SetMemoryPermission(Core::System& system, u64 address, u64 size, MemoryPe
     R_UNLESS(IsValidSetMemoryPermission(perm), ResultInvalidNewMemoryPermission);
 
     // Validate that the region is in range for the current process.
-    auto& page_table = GetCurrentProcess(system.Kernel()).GetPageTable();
+    auto& process = GetCurrentProcess(system.Kernel());
+    auto& page_table = process.GetPageTable();
     R_UNLESS(page_table.Contains(address, size), ResultInvalidCurrentMemory);
 
     // Set the memory attribute.
-    R_RETURN(page_table.SetMemoryPermission(address, size, perm));
+    const auto result = page_table.SetMemoryPermission(address, size, perm);
+    if (result.raw == ResultInvalidCurrentMemory.raw &&
+        IsHomebrewInPlaceNextLoadCodeRange(process, address, size)) {
+        R_RETURN(
+            SetHomebrewInPlaceMemoryPermissionByBlocks(process, address, size, perm, result));
+    }
+
+    R_RETURN(result);
 }
 
 Result SetMemoryAttribute(Core::System& system, u64 address, u64 size, u32 mask, u32 attr) {
@@ -135,14 +319,30 @@ Result MapMemory(Core::System& system, u64 dst_addr, u64 src_addr, u64 size) {
     LOG_TRACE(Kernel_SVC, "called, dst_addr={:#x}, src_addr={:#x}, size={:#x}", dst_addr,
               src_addr, size);
 
-    auto& page_table{GetCurrentProcess(system.Kernel()).GetPageTable()};
+    auto& process = GetCurrentProcess(system.Kernel());
+    auto& page_table = process.GetPageTable();
 
     if (const Result result{MapUnmapMemorySanityChecks(page_table, dst_addr, src_addr, size)};
         result.IsError()) {
         return result;
     }
 
-    R_RETURN(page_table.MapMemory(dst_addr, src_addr, size));
+    const auto result = page_table.MapMemory(dst_addr, src_addr, size);
+    if (result.raw == ResultInvalidCurrentMemory.raw && process.IsHomebrewInPlaceNextLoad()) {
+        if (UnlockHomebrewInPlaceDeviceSharedSource(process, src_addr, size, result).IsSuccess()) {
+            const auto retry_result = page_table.MapMemory(dst_addr, src_addr, size);
+            if (retry_result.IsError()) {
+                LOG_WARNING(Kernel_SVC,
+                            "NextLoad in-place: svcMapMemory retry failed after "
+                            "DeviceShared cleanup dst=0x{:016X}, src=0x{:016X}, size=0x{:X}, "
+                            "result={:#X}",
+                            dst_addr, src_addr, size, retry_result.raw);
+            }
+            R_RETURN(retry_result);
+        }
+    }
+
+    R_RETURN(result);
 }
 
 /// Unmaps a region that was previously mapped with svcMapMemory
@@ -150,14 +350,23 @@ Result UnmapMemory(Core::System& system, u64 dst_addr, u64 src_addr, u64 size) {
     LOG_TRACE(Kernel_SVC, "called, dst_addr={:#x}, src_addr={:#x}, size={:#x}", dst_addr,
               src_addr, size);
 
-    auto& page_table{GetCurrentProcess(system.Kernel()).GetPageTable()};
+    auto& process = GetCurrentProcess(system.Kernel());
+    auto& page_table = process.GetPageTable();
 
     if (const Result result{MapUnmapMemorySanityChecks(page_table, dst_addr, src_addr, size)};
         result.IsError()) {
         return result;
     }
 
-    R_RETURN(page_table.UnmapMemory(dst_addr, src_addr, size));
+    const auto result = page_table.UnmapMemory(dst_addr, src_addr, size);
+    if (result.raw == ResultInvalidCurrentMemory.raw && process.IsHomebrewInPlaceNextLoad()) {
+        LOG_WARNING(Kernel_SVC,
+                    "NextLoad in-place: svcUnmapMemory failed dst=0x{:016X}, "
+                    "src=0x{:016X}, size=0x{:X}, result={:#X}",
+                    dst_addr, src_addr, size, result.raw);
+    }
+
+    R_RETURN(result);
 }
 
 Result SetMemoryPermission64(Core::System& system, uint64_t address, uint64_t size,

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

@@ -4,17 +4,216 @@
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <algorithm>
+#include <filesystem>
+#include <string>
+#include <string_view>
+#include <vector>
+
+#include "common/fs/fs.h"
+#include "common/fs/path_util.h"
+#include "common/input.h"
 #include "core/core.h"
+#include "core/file_sys/vfs/vfs_types.h"
 #include "core/hle/kernel/k_process.h"
+#include "core/hle/kernel/k_thread.h"
+#include "core/hle/kernel/kernel.h"
+#include "core/hle/kernel/physical_core.h"
 #include "core/hle/kernel/svc.h"
+#include "core/hle/service/hid/hid_server.h"
+#include "core/hle/service/nvdrv/nvdrv_interface.h"
+#include "core/hle/service/sm/sm.h"
+#include "core/loader/nro.h"
+#include "hid_core/frontend/emulated_controller.h"
+#include "hid_core/hid_core.h"
+#include "hid_core/resource_manager.h"
 
 namespace Kernel::Svc {
 
+namespace {
+
+constexpr size_t HomebrewNextLoadPathSize = 0x200;
+constexpr size_t HomebrewNextLoadArgvSize = 0x800;
+
+std::string ReadHomebrewString(Core::Memory::Memory& memory, KProcessAddress address,
+                               size_t max_size) {
+    if (GetInteger(address) == 0) {
+        return {};
+    }
+    return memory.ReadCString(Common::ProcessAddress{GetInteger(address)}, max_size);
+}
+
+} // namespace
+
 /// Exits the current process
-void ExitProcess(Core::System& system) {
+void ExitProcess(Core::System& system, std::span<uint64_t, 8> args) {
     auto* current_process = GetCurrentProcessPointer(system.Kernel());
+    auto* current_thread = GetCurrentThreadPointer(system.Kernel());
 
     LOG_INFO(Kernel_SVC, "Process {} exiting", current_process->GetProcessId());
+    const auto next_load_path_addr = current_process->GetHomebrewNextLoadPathAddr();
+    const auto next_load_argv_addr = current_process->GetHomebrewNextLoadArgvAddr();
+    if (GetInteger(next_load_path_addr) != 0) {
+        auto& memory = current_process->GetMemory();
+        const auto next_load_path =
+            ReadHomebrewString(memory, next_load_path_addr, HomebrewNextLoadPathSize);
+        const auto next_load_argv =
+            ReadHomebrewString(memory, next_load_argv_addr, HomebrewNextLoadArgvSize);
+        if (!next_load_path.empty()) {
+            auto guest_path = Common::FS::SanitizePath(next_load_path);
+            constexpr std::string_view SdmcPrefix{"sdmc:"};
+            FileSys::VirtualFile file{};
+
+            const bool is_sdmc_path = guest_path.rfind(SdmcPrefix, 0) == 0;
+            const bool is_absolute_guest_path = !guest_path.empty() && guest_path.front() == '/';
+            if (is_sdmc_path || is_absolute_guest_path) {
+                auto relative_path =
+                    is_sdmc_path ? guest_path.substr(SdmcPrefix.size()) : guest_path;
+                while (!relative_path.empty() && relative_path.front() == '/') {
+                    relative_path.erase(relative_path.begin());
+                }
+
+                const auto host_path = Common::FS::GetEdenPath(Common::FS::EdenPath::SDMCDir) /
+                                       std::filesystem::path{Common::FS::ToU8String(relative_path)};
+                const auto host_path_string = Common::FS::PathToUTF8String(host_path);
+                file = Core::GetGameFileFromPath(system.GetFilesystem(), host_path_string);
+                if (!file) {
+                    LOG_WARNING(Kernel_SVC,
+                                "NextLoad: failed to open guest_path='{}', host_path='{}'",
+                                next_load_path, host_path_string);
+                }
+            } else {
+                file = Core::GetGameFileFromPath(system.GetFilesystem(), guest_path);
+                if (!file) {
+                    LOG_WARNING(Kernel_SVC, "NextLoad: failed to open guest_path='{}'",
+                                next_load_path);
+                }
+            }
+
+            if (file) {
+                const auto nvdrv =
+                    system.ServiceManager().GetService<Service::Nvidia::NVDRV>("nvdrv:s");
+                if (!nvdrv) {
+                    LOG_WARNING(Kernel_SVC, "NextLoad: NVDRV service unavailable for reset");
+                } else {
+                    nvdrv->GetModule()->ResetForProcess(current_process);
+                }
+
+                auto& page_table = current_process->GetPageTable();
+                const u64 heap_start = GetInteger(page_table.GetHeapRegionStart());
+                const u64 heap_size = page_table.GetHeapRegionSize();
+                const u64 heap_end = heap_start + heap_size;
+                if (heap_start != 0 && heap_size != 0 && heap_end > heap_start) {
+                    struct DeviceSharedBlock {
+                        u64 address;
+                        u64 size;
+                        u16 device_use_count;
+                    };
+
+                    std::vector<DeviceSharedBlock> blocks;
+                    for (u64 cursor = heap_start; cursor < heap_end;) {
+                        KMemoryInfo info;
+                        PageInfo page_info;
+                        const auto query_result =
+                            page_table.QueryInfo(std::addressof(info), std::addressof(page_info),
+                                                 cursor);
+                        if (query_result.IsError()) {
+                            LOG_WARNING(Kernel_SVC,
+                                        "NextLoad: DeviceShared heap cleanup query failed "
+                                        "address=0x{:016X}, result={:#X}",
+                                        cursor, query_result.raw);
+                            break;
+                        }
+
+                        const u64 block_end = (std::min<u64>)(info.GetEndAddress(), heap_end);
+                        if (block_end <= cursor) {
+                            LOG_WARNING(Kernel_SVC,
+                                        "NextLoad: DeviceShared heap cleanup walk stalled "
+                                        "cursor=0x{:016X}, block=0x{:016X}/0x{:X}",
+                                        cursor, info.GetAddress(), info.GetSize());
+                            break;
+                        }
+
+                        const bool is_device_shared =
+                            True(info.GetState() & KMemoryState::FlagCanDeviceMap) &&
+                            info.GetAttribute() == KMemoryAttribute::DeviceShared &&
+                            info.m_device_use_count > 0;
+                        if (is_device_shared) {
+                            blocks.push_back(DeviceSharedBlock{
+                                .address = cursor,
+                                .size = block_end - cursor,
+                                .device_use_count = info.m_device_use_count,
+                            });
+                        }
+                        cursor = block_end;
+                    }
+
+                    for (const auto& block : blocks) {
+                        for (u16 unlock = 0; unlock < block.device_use_count; unlock++) {
+                            const auto unlock_result =
+                                page_table.UnlockForDeviceAddressSpace(block.address, block.size);
+                            if (unlock_result.IsError()) {
+                                LOG_WARNING(Kernel_SVC,
+                                            "NextLoad: DeviceShared heap cleanup unlock failed "
+                                            "address=0x{:016X}, size=0x{:X}, remaining={}, "
+                                            "result={:#X}",
+                                            block.address, block.size,
+                                            block.device_use_count - unlock - 1, unlock_result.raw);
+                                break;
+                            }
+                        }
+                    }
+                }
+
+                if (Loader::LoadNroInPlace(system, *current_process, *current_thread, file,
+                                           next_load_path, next_load_argv)) {
+                    const auto aruid = current_process->GetProcessId();
+                    if (const auto hid =
+                            system.ServiceManager().GetService<Service::HID::IHidServer>("hid")) {
+                        const auto resource_manager = hid->GetResourceManager();
+                        resource_manager->UnregisterAppletResourceUserId(aruid);
+                        const auto register_result =
+                            resource_manager->RegisterAppletResourceUserId(aruid, true);
+                        if (register_result.IsError()) {
+                            LOG_WARNING(Kernel_SVC,
+                                        "NextLoad: failed to register HID applet resource "
+                                        "aruid={}, result={:#X}",
+                                        aruid, register_result.raw);
+                        }
+                    } else {
+                        LOG_WARNING(Kernel_SVC, "NextLoad: HID service unavailable for reset");
+                    }
+
+                    auto& hid_core = system.HIDCore();
+                    hid_core.DisableAllControllerConfiguration();
+                    hid_core.SetSupportedStyleTag({Core::HID::NpadStyleSet::All});
+                    hid_core.ReloadInputDevices();
+
+                    const auto activate_controller = [&](Core::HID::NpadIdType npad_id) {
+                        auto* controller = hid_core.GetEmulatedController(npad_id);
+                        if (controller == nullptr) {
+                            return;
+                        }
+
+                        (void)controller->SetPollingMode(Core::HID::EmulatedDeviceIndex::AllDevices,
+                                                         Common::Input::PollingMode::Active);
+                    };
+
+                    activate_controller(Core::HID::NpadIdType::Player1);
+                    activate_controller(Core::HID::NpadIdType::Handheld);
+
+                    system.Kernel().CurrentPhysicalCore().LoadContext(current_thread);
+
+                    const auto& context = current_thread->GetContext();
+                    for (size_t i = 0; i < args.size(); i++) {
+                        args[i] = context.r[i];
+                    }
+
+                    return;
+                }
+            }
+        }
+    }
     ASSERT_MSG(current_process->GetState() == KProcess::State::Running,
                "Process has already exited");
 
@@ -132,8 +331,8 @@ Result TerminateProcess(Core::System& system, Handle process_handle) {
     R_THROW(ResultNotImplemented);
 }
 
-void ExitProcess64(Core::System& system) {
-    ExitProcess(system);
+void ExitProcess64(Core::System& system, std::span<uint64_t, 8> args) {
+    ExitProcess(system, args);
 }
 
 Result GetProcessId64(Core::System& system, uint64_t* out_process_id, Handle process_handle) {
@@ -164,8 +363,8 @@ Result GetProcessInfo64(Core::System& system, int64_t* out_info, Handle process_
     R_RETURN(GetProcessInfo(system, out_info, process_handle, info_type));
 }
 
-void ExitProcess64From32(Core::System& system) {
-    ExitProcess(system);
+void ExitProcess64From32(Core::System& system, std::span<uint64_t, 8> args) {
+    ExitProcess(system, args);
 }
 
 Result GetProcessId64From32(Core::System& system, uint64_t* out_process_id, Handle process_handle) {

src/core/hle/service/filesystem/filesystem.cpp

@@ -4,6 +4,7 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <string_view>
 #include <utility>
 
 #include "common/assert.h"
@@ -257,7 +258,7 @@ Result VfsDirectoryServiceWrapper::OpenFile(FileSys::VirtualFile* out_file,
         npath.remove_prefix(1);
     }
 
-    auto file = backing->GetFileRelative(npath);
+    auto file = backing->GetFileRelative(npath, mode);
     if (file == nullptr) {
         return FileSys::ResultPathNotFound;
     }
@@ -333,14 +334,16 @@ FileSystemController::~FileSystemController() = default;
 
 Result FileSystemController::RegisterProcess(
     ProcessId process_id, ProgramId program_id,
-    std::shared_ptr<FileSys::RomFSFactory>&& romfs_factory) {
+    std::shared_ptr<FileSys::RomFSFactory>&& romfs_factory, std::string homebrew_initial_cwd) {
     std::scoped_lock lk{registration_lock};
 
-    registrations.emplace(process_id, Registration{
-                                          .program_id = program_id,
-                                          .romfs_factory = std::move(romfs_factory),
-                                          .save_data_factory = CreateSaveDataFactory(program_id),
-                                      });
+    registrations.insert_or_assign(process_id,
+                                   Registration{
+                                       .program_id = program_id,
+                                       .romfs_factory = std::move(romfs_factory),
+                                       .save_data_factory = CreateSaveDataFactory(program_id),
+                                       .homebrew_initial_cwd = std::move(homebrew_initial_cwd),
+                                   });
 
     LOG_DEBUG(Service_FS, "Registered for process {}", process_id);
     return ResultSuccess;
@@ -348,7 +351,8 @@ Result FileSystemController::RegisterProcess(
 
 Result FileSystemController::OpenProcess(
     ProgramId* out_program_id, std::shared_ptr<SaveDataController>* out_save_data_controller,
-    std::shared_ptr<RomFsController>* out_romfs_controller, ProcessId process_id) {
+    std::shared_ptr<RomFsController>* out_romfs_controller, ProcessId process_id,
+    std::string* out_homebrew_initial_cwd) {
     std::scoped_lock lk{registration_lock};
 
     const auto it = registrations.find(process_id);
@@ -361,6 +365,9 @@ Result FileSystemController::OpenProcess(
         std::make_shared<SaveDataController>(system, it->second.save_data_factory);
     *out_romfs_controller =
         std::make_shared<RomFsController>(it->second.romfs_factory, it->second.program_id);
+    if (out_homebrew_initial_cwd != nullptr) {
+        *out_homebrew_initial_cwd = it->second.homebrew_initial_cwd;
+    }
     return ResultSuccess;
 }
 

src/core/hle/service/filesystem/filesystem.h

@@ -8,6 +8,7 @@
 
 #include <memory>
 #include <mutex>
+#include <string>
 #include "common/common_types.h"
 #include "core/file_sys/fs_directory.h"
 #include "core/file_sys/fs_filesystem.h"
@@ -71,11 +72,12 @@ public:
     ~FileSystemController();
 
     Result RegisterProcess(ProcessId process_id, ProgramId program_id,
-                           std::shared_ptr<FileSys::RomFSFactory>&& factory);
+                           std::shared_ptr<FileSys::RomFSFactory>&& factory,
+                           std::string homebrew_initial_cwd = {});
     Result OpenProcess(ProgramId* out_program_id,
-                       std::shared_ptr<SaveDataController>* out_save_data_controller,
-                       std::shared_ptr<RomFsController>* out_romfs_controller,
-                       ProcessId process_id);
+                        std::shared_ptr<SaveDataController>* out_save_data_controller,
+                        std::shared_ptr<RomFsController>* out_romfs_controller,
+                        ProcessId process_id, std::string* out_homebrew_initial_cwd = nullptr);
     void SetPackedUpdate(ProcessId process_id, FileSys::VirtualFile update_raw);
 
     std::shared_ptr<SaveDataController> OpenSaveDataController();
@@ -136,6 +138,7 @@ private:
         ProgramId program_id;
         std::shared_ptr<FileSys::RomFSFactory> romfs_factory;
         std::shared_ptr<FileSys::SaveDataFactory> save_data_factory;
+        std::string homebrew_initial_cwd;
     };
 
     std::mutex registration_lock;

src/core/hle/service/filesystem/fsp/fs_i_filesystem.cpp

@@ -4,6 +4,9 @@
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <string_view>
+
+#include "common/fs/path_util.h"
 #include "common/string_util.h"
 #include "core/file_sys/fssrv/fssrv_sf_path.h"
 #include "core/hle/service/cmif_serialization.h"
@@ -13,10 +16,29 @@
 
 namespace Service::FileSystem {
 
-IFileSystem::IFileSystem(Core::System& system_, FileSys::VirtualDir dir_, SizeGetter size_getter_)
-    : ServiceFramework{system_, "IFileSystem"}, backend{std::make_unique<FileSys::Fsa::IFileSystem>(
-                                                    dir_)},
-      size_getter{std::move(size_getter_)} {
+static std::string ResolveHomebrewCwdRootAlias(std::string path,
+                                               std::string_view homebrew_initial_cwd) {
+    if (homebrew_initial_cwd.empty()) {
+        return path;
+    }
+
+    const std::string normalized_path = Common::FS::SanitizePath(path);
+    if (normalized_path.empty() || normalized_path == "/" ||
+        normalized_path != homebrew_initial_cwd ||
+        path.size() != normalized_path.size() + 2 ||
+        path.substr(0, normalized_path.size()) != normalized_path ||
+        path.substr(normalized_path.size()) != "//") {
+        return path;
+    }
+
+    return "/";
+}
+
+IFileSystem::IFileSystem(Core::System& system_, FileSys::VirtualDir dir_, SizeGetter size_getter_,
+                         std::string homebrew_initial_cwd_)
+    : ServiceFramework{system_, "IFileSystem"},
+      backend{std::make_unique<FileSys::Fsa::IFileSystem>(dir_)},
+      size_getter{std::move(size_getter_)}, homebrew_initial_cwd{std::move(homebrew_initial_cwd_)} {
     static const FunctionInfo functions[] = {
         {0, D<&IFileSystem::CreateFile>, "CreateFile"},
         {1, D<&IFileSystem::DeleteFile>, "DeleteFile"},
@@ -43,7 +65,8 @@ Result IFileSystem::CreateFile(const InLargeData<FileSys::Sf::Path, BufferAttr_H
                                s32 option, s64 size) {
     LOG_DEBUG(Service_FS, "called. file={}, option={:#x}, size={:#08x}", path->str, option, size);
 
-    R_RETURN(backend->CreateFile(FileSys::Path(path->str), size));
+    const auto fs_path = ResolveHomebrewCwdRootAlias(path->str, homebrew_initial_cwd);
+    R_RETURN(backend->CreateFile(FileSys::Path(fs_path.c_str()), size));
 }
 
 Result IFileSystem::DeleteFile(const InLargeData<FileSys::Sf::Path, BufferAttr_HipcPointer> path) {
@@ -94,7 +117,8 @@ Result IFileSystem::OpenFile(OutInterface<IFile> out_interface,
     LOG_DEBUG(Service_FS, "called. file={}, mode={}", path->str, mode);
 
     FileSys::VirtualFile vfs_file{};
-    R_TRY(backend->OpenFile(&vfs_file, FileSys::Path(path->str),
+    const auto fs_path = ResolveHomebrewCwdRootAlias(path->str, homebrew_initial_cwd);
+    R_TRY(backend->OpenFile(&vfs_file, FileSys::Path(fs_path.c_str()),
                             static_cast<FileSys::OpenMode>(mode)));
 
     *out_interface = std::make_shared<IFile>(system, vfs_file);
@@ -107,7 +131,8 @@ Result IFileSystem::OpenDirectory(OutInterface<IDirectory> out_interface,
     LOG_DEBUG(Service_FS, "called. directory={}, mode={}", path->str, mode);
 
     FileSys::VirtualDir vfs_dir{};
-    R_TRY(backend->OpenDirectory(&vfs_dir, FileSys::Path(path->str),
+    const auto fs_path = ResolveHomebrewCwdRootAlias(path->str, homebrew_initial_cwd);
+    R_TRY(backend->OpenDirectory(&vfs_dir, FileSys::Path(fs_path.c_str()),
                                  static_cast<FileSys::OpenDirectoryMode>(mode)));
 
     *out_interface = std::make_shared<IDirectory>(system, vfs_dir,
@@ -120,7 +145,8 @@ Result IFileSystem::GetEntryType(
     LOG_DEBUG(Service_FS, "called. file={}", path->str);
 
     FileSys::DirectoryEntryType vfs_entry_type{};
-    R_TRY(backend->GetEntryType(&vfs_entry_type, FileSys::Path(path->str)));
+    const auto fs_path = ResolveHomebrewCwdRootAlias(path->str, homebrew_initial_cwd);
+    R_TRY(backend->GetEntryType(&vfs_entry_type, FileSys::Path(fs_path.c_str())));
 
     *out_type = static_cast<u32>(vfs_entry_type);
     R_SUCCEED();

src/core/hle/service/filesystem/fsp/fs_i_filesystem.h

@@ -1,8 +1,13 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
+#include <string>
+
 #include "common/common_funcs.h"
 #include "core/file_sys/fs_filesystem.h"
 #include "core/file_sys/fsa/fs_i_filesystem.h"
@@ -23,7 +28,8 @@ class IDirectory;
 
 class IFileSystem final : public ServiceFramework<IFileSystem> {
 public:
-    explicit IFileSystem(Core::System& system_, FileSys::VirtualDir dir_, SizeGetter size_getter_);
+    explicit IFileSystem(Core::System& system_, FileSys::VirtualDir dir_, SizeGetter size_getter_,
+                         std::string homebrew_initial_cwd_ = {});
 
     Result CreateFile(const InLargeData<FileSys::Sf::Path, BufferAttr_HipcPointer> path, s32 option,
                       s64 size);
@@ -55,6 +61,7 @@ public:
 private:
     std::unique_ptr<FileSys::Fsa::IFileSystem> backend;
     SizeGetter size_getter;
+    std::string homebrew_initial_cwd;
 };
 
 } // namespace Service::FileSystem

src/core/hle/service/filesystem/fsp/fsp_srv.cpp

@@ -192,8 +192,9 @@ Result FSP_SRV::SetCurrentProcess(ClientProcessId pid) {
 
     LOG_DEBUG(Service_FS, "called. current_process_id={:#016x}", current_process_id);
 
-    R_RETURN(
-        fsc.OpenProcess(&program_id, &save_data_controller, &romfs_controller, current_process_id));
+    homebrew_initial_cwd.clear();
+    R_RETURN(fsc.OpenProcess(&program_id, &save_data_controller, &romfs_controller,
+                             current_process_id, &homebrew_initial_cwd));
 }
 
 Result FSP_SRV::OpenFileSystemWithPatch(OutInterface<IFileSystem> out_interface,
@@ -224,7 +225,8 @@ Result FSP_SRV::OpenSdCardFileSystem(OutInterface<IFileSystem> out_interface) {
     fsc.OpenSDMC(&sdmc_dir);
 
     *out_interface = std::make_shared<IFileSystem>(
-        system, sdmc_dir, SizeGetter::FromStorageId(fsc, FileSys::StorageId::SdCard));
+        system, sdmc_dir, SizeGetter::FromStorageId(fsc, FileSys::StorageId::SdCard),
+        homebrew_initial_cwd);
 
     R_SUCCEED();
 }

src/core/hle/service/filesystem/fsp/fsp_srv.h

@@ -7,6 +7,7 @@
 #pragma once
 
 #include <memory>
+#include <string>
 #include "core/file_sys/fs_save_data_types.h"
 #include "core/hle/service/cmif_types.h"
 #include "core/hle/service/filesystem/fsp/fsp_types.h"
@@ -123,6 +124,7 @@ private:
     u32 access_log_program_index = 0;
     AccessLogMode access_log_mode = AccessLogMode::None;
     u64 program_id = 0;
+    std::string homebrew_initial_cwd;
     std::shared_ptr<SaveDataController> save_data_controller;
     std::shared_ptr<RomFsController> romfs_controller;
 };

src/core/hle/service/nvdrv/core/container.cpp

@@ -1,10 +1,15 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: 2022 yuzu Emulator Project
 // SPDX-FileCopyrightText: 2022 Skyline Team and Contributors
 // SPDX-License-Identifier: GPL-3.0-or-later
 
+#include <algorithm>
 #include <atomic>
 #include <deque>
 #include <mutex>
+#include <vector>
 
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/service/nvdrv/core/container.h"
@@ -40,6 +45,16 @@ Container::Container(Tegra::Host1x::Host1x& host1x_) {
 
 Container::~Container() = default;
 
+static bool IsSameProcess(Kernel::KProcess* lhs, Kernel::KProcess* rhs) {
+    if (lhs == rhs) {
+        return true;
+    }
+    if (lhs == nullptr || rhs == nullptr) {
+        return false;
+    }
+    return lhs->GetProcessId() == rhs->GetProcessId();
+}
+
 SessionId Container::OpenSession(Kernel::KProcess* process) {
     using namespace Common::Literals;
 
@@ -48,7 +63,7 @@ SessionId Container::OpenSession(Kernel::KProcess* process) {
         if (!session.is_active) {
             continue;
         }
-        if (session.process == process) {
+        if (IsSameProcess(session.process, process)) {
             session.ref_count++;
             return session.id;
         }
@@ -116,7 +131,15 @@ SessionId Container::OpenSession(Kernel::KProcess* process) {
 
 void Container::CloseSession(SessionId session_id) {
     std::scoped_lock lk(impl->session_guard);
+    if (session_id.id >= impl->sessions.size()) {
+        return;
+    }
+
     auto& session = impl->sessions[session_id.id];
+    if (!session.is_active || session.ref_count <= 0) {
+        return;
+    }
+
     if (--session.ref_count > 0) {
         return;
     }
@@ -134,6 +157,73 @@ void Container::CloseSession(SessionId session_id) {
     impl->id_pool.emplace_front(session_id.id);
 }
 
+size_t Container::CloseSessions(std::span<const SessionId> session_ids) {
+    std::vector<SessionId> valid_session_ids;
+    valid_session_ids.reserve(session_ids.size());
+
+    {
+        std::scoped_lock lk(impl->session_guard);
+        for (const auto session_id : session_ids) {
+            if (session_id.id >= impl->sessions.size()) {
+                continue;
+            }
+
+            auto& session = impl->sessions[session_id.id];
+            if (!session.is_active) {
+                continue;
+            }
+
+            const auto duplicate = std::ranges::any_of(
+                valid_session_ids, [session_id](const auto candidate) {
+                    return candidate.id == session_id.id;
+                });
+            if (duplicate) {
+                continue;
+            }
+
+            session.ref_count = 1;
+            valid_session_ids.push_back(session_id);
+        }
+    }
+
+    for (const auto session_id : valid_session_ids) {
+        CloseSession(session_id);
+    }
+
+    return valid_session_ids.size();
+}
+
+std::vector<SessionId> Container::GetSessionIdsForProcess(Kernel::KProcess* process) {
+    std::vector<SessionId> session_ids;
+    std::scoped_lock lk(impl->session_guard);
+    for (const auto& session : impl->sessions) {
+        if (!session.is_active || !IsSameProcess(session.process, process)) {
+            continue;
+        }
+
+        session_ids.push_back(session.id);
+    }
+
+    return session_ids;
+}
+
+std::vector<SessionId> Container::GetActiveSessionIds() const {
+    std::vector<SessionId> session_ids;
+    std::scoped_lock lk(impl->session_guard);
+    for (const auto& session : impl->sessions) {
+        if (session.is_active) {
+            session_ids.push_back(session.id);
+        }
+    }
+
+    return session_ids;
+}
+
+bool Container::IsSessionActive(SessionId session_id) const {
+    std::scoped_lock lk(impl->session_guard);
+    return session_id.id < impl->sessions.size() && impl->sessions[session_id.id].is_active;
+}
+
 Session* Container::GetSession(SessionId session_id) {
     std::atomic_thread_fence(std::memory_order_acquire);
     return &impl->sessions[session_id.id];

src/core/hle/service/nvdrv/core/container.h

@@ -9,7 +9,10 @@
 
 #include <deque>
 #include <memory>
+#include <span>
+#include <cstddef>
 #include <ankerl/unordered_dense.h>
+#include <vector>
 
 #include "core/device_memory_manager.h"
 #include "core/hle/service/nvdrv/nvdata.h"
@@ -59,6 +62,10 @@ public:
 
     SessionId OpenSession(Kernel::KProcess* process);
     void CloseSession(SessionId id);
+    size_t CloseSessions(std::span<const SessionId> session_ids);
+    std::vector<SessionId> GetSessionIdsForProcess(Kernel::KProcess* process);
+    std::vector<SessionId> GetActiveSessionIds() const;
+    bool IsSessionActive(SessionId id) const;
 
     Session* GetSession(SessionId id);
 

src/core/hle/service/nvdrv/core/nvmap.cpp

@@ -6,10 +6,12 @@
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 #include <functional>
+#include <vector>
 
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging.h"
+#include "core/hle/kernel/k_process.h"
 #include "core/hle/service/nvdrv/core/container.h"
 #include "core/hle/service/nvdrv/core/heap_mapper.h"
 #include "core/hle/service/nvdrv/core/nvmap.h"
@@ -326,19 +328,65 @@ std::optional<NvMap::FreeInfo> NvMap::FreeHandle(Handle::Id handle, bool interna
 }
 
 void NvMap::UnmapAllHandles(NvCore::SessionId session_id) {
-    auto handles_copy = [&] {
+    auto* session = core.GetSession(session_id);
+    auto* process = session != nullptr ? session->process : nullptr;
+    auto handle_ids = [&] {
         std::scoped_lock lk{handles_lock};
-        return handles;
+        std::vector<Handle::Id> ids;
+        ids.reserve(handles.size());
+
+        for (const auto& entry : handles) {
+            ids.push_back(entry.first);
+        }
+
+        return ids;
     }();
 
-    for (auto& [id, handle] : handles_copy) {
-        {
-            std::scoped_lock lk{handle->mutex};
-            if (handle->session_id.id != session_id.id || handle->dupes <= 0) {
-                continue;
+    for (const auto id : handle_ids) {
+        bool unlocked_pages = false;
+        while (true) {
+            bool last_user_reference = false;
+            VAddr address = 0;
+            size_t size = 0;
+            {
+                const auto handle = GetHandle(id);
+                if (!handle) {
+                    break;
+                }
+
+                std::scoped_lock lk{handle->mutex};
+                if (handle->session_id.id != session_id.id || handle->dupes <= 0) {
+                    break;
+                }
+
+                last_user_reference = handle->dupes == 1;
+                address = handle->address;
+                size = handle->size;
+            }
+
+            const auto free_info = FreeHandle(id, false);
+            if (!free_info) {
+                break;
+            }
+
+            if (!unlocked_pages && process != nullptr && address != 0 && size != 0 &&
+                (free_info->can_unlock || last_user_reference)) {
+                const auto unlock_result =
+                    process->GetPageTable().UnlockForDeviceAddressSpace(address, size);
+                if (unlock_result.IsError()) {
+                    LOG_WARNING(Service_NVDRV,
+                                "NextLoad: nvmap session cleanup unlock failed, "
+                                "handle={}, session={}, address=0x{:016X}, size=0x{:X}, "
+                                "result={:#X}",
+                                id, session_id.id, address, size, unlock_result.raw);
+                }
+                unlocked_pages = true;
+            }
+
+            if (last_user_reference) {
+                break;
             }
         }
-        FreeHandle(id, false);
     }
 }
 

src/core/hle/service/nvdrv/nvdrv.cpp

@@ -1,12 +1,17 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: 2021 yuzu Emulator Project
 // SPDX-FileCopyrightText: 2021 Skyline Team and Contributors
 // SPDX-License-Identifier: GPL-3.0-or-later
 
+#include <algorithm>
 #include <utility>
+#include <vector>
 
-#include <fmt/ranges.h>
 #include "core/core.h"
 #include "core/hle/kernel/k_event.h"
+#include "core/hle/kernel/k_process.h"
 #include "core/hle/service/ipc_helpers.h"
 #include "core/hle/service/nvdrv/core/container.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
@@ -133,6 +138,9 @@ DeviceFD Module::Open(const std::string& device_name, NvCore::SessionId session_
     auto device = builder(fd)->second;
 
     device->OnOpen(session_id, fd);
+    if (container.IsSessionActive(session_id)) {
+        open_file_sessions.emplace(fd, session_id);
+    }
 
     return fd;
 }
@@ -204,6 +212,7 @@ NvResult Module::Close(DeviceFD fd) {
     itr->second->OnClose(fd);
 
     open_files.erase(itr);
+    open_file_sessions.erase(fd);
 
     return NvResult::Success;
 }
@@ -228,4 +237,98 @@ NvResult Module::QueryEvent(DeviceFD fd, u32 event_id, Kernel::KEvent*& event) {
     return NvResult::Success;
 }
 
+static bool ContainsSession(std::span<const NvCore::SessionId> session_ids,
+                            NvCore::SessionId session_id) {
+    return std::ranges::any_of(session_ids, [session_id](const auto candidate) {
+        return candidate.id == session_id.id;
+    });
+}
+
+static void AppendUniqueSession(std::vector<NvCore::SessionId>& session_ids,
+                                NvCore::SessionId session_id) {
+    if (!ContainsSession(session_ids, session_id)) {
+        session_ids.push_back(session_id);
+    }
+}
+
+size_t Module::CloseFilesForSessions(std::span<const NvCore::SessionId> session_ids) {
+    std::vector<DeviceFD> fds;
+    fds.reserve(open_file_sessions.size());
+
+    for (const auto& [fd, session_id] : open_file_sessions) {
+        if (ContainsSession(session_ids, session_id)) {
+            fds.push_back(fd);
+        }
+    }
+
+    for (const auto fd : fds) {
+        Close(fd);
+    }
+
+    return fds.size();
+}
+
+void Module::CloseSession(NvCore::SessionId session_id) {
+    container.CloseSession(session_id);
+}
+
+void Module::TrackSessionAruid(NvCore::SessionId session_id, u64 aruid) {
+    const bool active = container.IsSessionActive(session_id);
+    if (active) {
+        session_aruids[session_id.id] = aruid;
+    }
+}
+
+std::vector<NvCore::SessionId> Module::GetSessionIdsForAruid(u64 aruid) const {
+    std::vector<NvCore::SessionId> session_ids;
+    for (const auto& [session_id, session_aruid] : session_aruids) {
+        if (session_aruid == aruid) {
+            session_ids.push_back(NvCore::SessionId{session_id});
+        }
+    }
+
+    return session_ids;
+}
+
+size_t Module::ResetForProcess(Kernel::KProcess* process) {
+    const auto process_id = process != nullptr ? process->GetProcessId() : 0;
+    auto session_ids = container.GetSessionIdsForProcess(process);
+
+    if (process_id != 0) {
+        for (const auto session_id : GetSessionIdsForAruid(process_id)) {
+            AppendUniqueSession(session_ids, session_id);
+        }
+    }
+
+    const auto active_session_ids = container.GetActiveSessionIds();
+    const auto active_before = active_session_ids.size();
+    const bool has_active_candidate =
+        std::ranges::any_of(session_ids, [this](const auto session_id) {
+            return container.IsSessionActive(session_id);
+        });
+    bool used_active_sessions = false;
+    if (!has_active_candidate && !active_session_ids.empty()) {
+        for (const auto session_id : active_session_ids) {
+            AppendUniqueSession(session_ids, session_id);
+        }
+        used_active_sessions = true;
+    }
+
+    const auto closed_files = CloseFilesForSessions(session_ids);
+    const auto closed_sessions = container.CloseSessions(session_ids);
+    for (const auto session_id : session_ids) {
+        if (!container.IsSessionActive(session_id)) {
+            session_aruids.erase(session_id.id);
+        }
+    }
+
+    if (used_active_sessions) {
+        LOG_WARNING(Service_NVDRV,
+                    "NextLoad: NVDRV reset used active sessions because process-owned "
+                    "sessions were not found, process_id={}, sessions={}, files={}, active_before={}",
+                    process_id, closed_sessions, closed_files, active_before);
+    }
+    return closed_sessions;
+}
+
 } // namespace Service::Nvidia

src/core/hle/service/nvdrv/nvdrv.h

@@ -12,6 +12,7 @@
 #include <memory>
 #include <span>
 #include <string>
+#include <vector>
 #include <ankerl/unordered_dense.h>
 
 #include "common/common_types.h"
@@ -26,6 +27,7 @@ class System;
 
 namespace Kernel {
 class KEvent;
+class KProcess;
 }
 
 namespace Service::Nvidia {
@@ -89,6 +91,9 @@ public:
     NvResult Close(DeviceFD fd);
 
     NvResult QueryEvent(DeviceFD fd, u32 event_id, Kernel::KEvent*& event);
+    void CloseSession(NvCore::SessionId session_id);
+    void TrackSessionAruid(NvCore::SessionId session_id, u64 aruid);
+    size_t ResetForProcess(Kernel::KProcess* process);
 
     NvCore::Container& GetContainer() {
         return container;
@@ -106,12 +111,17 @@ private:
     using FilesContainerType = ankerl::unordered_dense::map<DeviceFD, std::shared_ptr<Devices::nvdevice>>;
     /// Mapping of file descriptors to the devices they reference.
     FilesContainerType open_files;
+    ankerl::unordered_dense::map<DeviceFD, NvCore::SessionId> open_file_sessions;
+    ankerl::unordered_dense::map<size_t, u64> session_aruids;
 
     KernelHelpers::ServiceContext service_context;
 
     EventInterface events_interface;
 
     ankerl::unordered_dense::map<std::string, std::function<FilesContainerType::iterator(DeviceFD)>> builders;
+
+    size_t CloseFilesForSessions(std::span<const NvCore::SessionId> session_ids);
+    std::vector<NvCore::SessionId> GetSessionIdsForAruid(u64 aruid) const;
 };
 
 void LoopProcess(Core::System& system);

src/core/hle/service/nvdrv/nvdrv_interface.cpp

@@ -212,7 +212,10 @@ void NVDRV::QueryEvent(HLERequestContext& ctx) {
 void NVDRV::SetAruid(HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     pid = rp.Pop<u64>();
-    LOG_WARNING(Service_NVDRV, "(STUBBED) called, pid={:#x}", pid);
+    LOG_WARNING(Service_NVDRV, "(STUBBED) called, pid={:#X}", pid);
+    if (is_initialized) {
+        nvdrv->TrackSessionAruid(session_id, pid);
+    }
 
     IPC::ResponseBuilder rb{ctx, 3};
     rb.Push(ResultSuccess);

src/core/internal_network/network.cpp

@@ -40,6 +40,7 @@ namespace Network {
 namespace {
 
 enum class CallType {
+    Connect,
     Send,
     Other,
 };
@@ -131,7 +132,7 @@ Errno TranslateNativeError(int e, CallType call_type = CallType::Other) {
     case WSAENOTCONN:
         return Errno::NOTCONN;
     case WSAEWOULDBLOCK:
-        return Errno::AGAIN;
+        return call_type == CallType::Connect ? Errno::INPROGRESS : Errno::AGAIN;
     case WSAECONNREFUSED:
         return Errno::CONNREFUSED;
     case WSAECONNABORTED:
@@ -563,6 +564,7 @@ int TranslateTypeToNative(Type type) {
     NETWORK_PROTOCOL_TRANSLATE_ELEM(UDPLITE)
 #elif defined(_WIN32)
 #define NETWORK_PROTOCOL_TRANSLATE_LIST \
+    NETWORK_PROTOCOL_TRANSLATE_ELEM(IP) \
     /*NETWORK_PROTOCOL_TRANSLATE_ELEM(HOPOPTS)*/ \
     NETWORK_PROTOCOL_TRANSLATE_ELEM(ICMP) \
     NETWORK_PROTOCOL_TRANSLATE_ELEM(IGMP) \
@@ -888,7 +890,7 @@ Errno Socket::Connect(SockAddrIn addr_in) {
         return Errno::SUCCESS;
     }
 
-    return GetAndLogLastError();
+    return GetAndLogLastError(CallType::Connect);
 }
 
 std::pair<SockAddrIn, Errno> Socket::GetPeerName() {

src/core/loader/homebrew_nxlink.cpp

@@ -0,0 +1,311 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#include "core/loader/homebrew_nxlink.h"
+
+#include <array>
+#include <cctype>
+#include <memory>
+#include <mutex>
+#include <optional>
+#include <stop_token>
+#include <utility>
+
+#include <boost/asio.hpp>
+
+#include "common/logging.h"
+#include "common/polyfill_thread.h"
+#include "common/thread.h"
+
+namespace Loader::HomebrewNxlink {
+namespace {
+
+using boost::asio::ip::tcp;
+
+constexpr u16 NxlinkClientPort = 28771;
+constexpr size_t ReceiveBufferSize = 1024;
+constexpr size_t MaxLogLineSize = 16 * 1024;
+
+struct LogConnection {
+    explicit LogConnection(boost::asio::io_context& io_context_) : socket{io_context_} {}
+
+    tcp::socket socket;
+    std::array<char, ReceiveBufferSize> buffer{};
+    std::string line;
+};
+
+struct LogServerState {
+    LogServerState() : acceptor{io_context} {}
+
+    boost::asio::io_context io_context;
+    tcp::acceptor acceptor;
+};
+
+std::mutex server_mutex;
+std::shared_ptr<LogServerState> server_state;
+std::optional<std::jthread> server_thread;
+
+std::optional<std::string> GetLastArgvToken(std::string_view argv_string) {
+    while (!argv_string.empty() && argv_string.back() == '\0') {
+        argv_string.remove_suffix(1);
+    }
+
+    std::optional<std::string_view> last_token;
+    bool in_token = false;
+    bool quoted = false;
+    size_t token_begin = 0;
+    size_t token_size = 0;
+
+    for (size_t i = 0; i <= argv_string.size(); i++) {
+        const char c = i < argv_string.size() ? argv_string[i] : '\0';
+
+        if (!in_token) {
+            if (c == '\0' || std::isspace(static_cast<unsigned char>(c))) {
+                continue;
+            }
+
+            in_token = true;
+            token_size = 0;
+            if (c == '"') {
+                quoted = true;
+                token_begin = i + 1;
+            } else {
+                quoted = false;
+                token_begin = i;
+                token_size = 1;
+            }
+            continue;
+        }
+
+        const bool token_end =
+            quoted ? c == '"' || c == '\0'
+                   : c == '\0' || std::isspace(static_cast<unsigned char>(c));
+        if (token_end) {
+            if (token_size != 0) {
+                last_token = argv_string.substr(token_begin, token_size);
+            }
+            in_token = false;
+            quoted = false;
+            token_size = 0;
+            continue;
+        }
+
+        token_size++;
+    }
+
+    if (!last_token) {
+        return std::nullopt;
+    }
+    return std::string{*last_token};
+}
+
+void AppendArgvToken(std::string& argv_string, std::string_view token) {
+    while (!argv_string.empty() && argv_string.back() == '\0') {
+        argv_string.pop_back();
+    }
+    if (!argv_string.empty()) {
+        argv_string.push_back(' ');
+    }
+    argv_string.append(token);
+}
+
+void FlushLogLine(std::string& line) {
+    if (!line.empty() && line.back() == '\r') {
+        line.pop_back();
+    }
+    if (!line.empty()) {
+        LOG_INFO(Loader, "{}", line);
+        line.clear();
+    }
+}
+
+void ConsumeLogBytes(LogConnection& connection, std::string_view bytes) {
+    for (const char byte : bytes) {
+        if (byte == '\n') {
+            FlushLogLine(connection.line);
+            continue;
+        }
+
+        connection.line.push_back(byte);
+        if (connection.line.size() >= MaxLogLineSize) {
+            FlushLogLine(connection.line);
+        }
+    }
+}
+
+void StartRead(std::shared_ptr<LogConnection> connection) {
+    connection->socket.async_read_some(
+        boost::asio::buffer(connection->buffer),
+        [connection](const boost::system::error_code& error, size_t bytes_read) {
+            if (error.failed()) {
+                FlushLogLine(connection->line);
+                return;
+            }
+
+            ConsumeLogBytes(*connection, {connection->buffer.data(), bytes_read});
+            StartRead(connection);
+        });
+}
+
+void StartAccept(std::shared_ptr<LogServerState> state) {
+    auto connection = std::make_shared<LogConnection>(state->io_context);
+    state->acceptor.async_accept(
+        connection->socket,
+        [state, connection](const boost::system::error_code& error) {
+            if (!error.failed()) {
+                LOG_INFO(Loader, "Homebrew nxlink log client connected");
+                StartRead(connection);
+            }
+            if (state->acceptor.is_open()) {
+                StartAccept(state);
+            }
+        });
+}
+
+void StartLogServer() {
+    std::scoped_lock lock{server_mutex};
+    if (server_thread) {
+        return;
+    }
+
+    auto state = std::make_shared<LogServerState>();
+    const tcp::endpoint endpoint{boost::asio::ip::address_v4::loopback(), NxlinkClientPort};
+
+    boost::system::error_code error;
+    state->acceptor.open(endpoint.protocol(), error);
+    if (!error.failed()) {
+        state->acceptor.set_option(tcp::acceptor::reuse_address(true), error);
+    }
+    if (!error.failed()) {
+        state->acceptor.bind(endpoint, error);
+    }
+    if (!error.failed()) {
+        state->acceptor.listen(boost::asio::socket_base::max_listen_connections, error);
+    }
+    if (error.failed()) {
+        LOG_WARNING(Loader, "Homebrew nxlink log server could not listen on 127.0.0.1:{}: {}",
+                    NxlinkClientPort, error.message());
+        return;
+    }
+
+    StartAccept(state);
+    server_state = state;
+    server_thread.emplace([state](std::stop_token stop_token) {
+        Common::SetCurrentThreadName("HomebrewNxlink");
+        std::stop_callback stop_callback{stop_token, [state] {
+            boost::system::error_code ignored;
+            state->acceptor.close(ignored);
+            state->io_context.stop();
+        }};
+
+        LOG_INFO(Loader, "Homebrew nxlink log server listening on 127.0.0.1:{}",
+                 NxlinkClientPort);
+        state->io_context.run();
+        LOG_INFO(Loader, "Homebrew nxlink log server stopped");
+    });
+}
+
+} // namespace
+
+bool IsArgvMarker(std::string_view token) {
+    if (token.size() != ArgvMarkerSize || token.substr(8) != ArgvMarkerSuffix) {
+        return false;
+    }
+
+    for (size_t i = 0; i < 8; i++) {
+        if (!std::isxdigit(static_cast<unsigned char>(token[i]))) {
+            return false;
+        }
+    }
+
+    return true;
+}
+
+bool IsLoopbackArgvMarker(std::string_view token) {
+    if (!IsArgvMarker(token)) {
+        return false;
+    }
+
+    for (size_t i = 0; i < 8; i++) {
+        if (std::tolower(static_cast<unsigned char>(token[i])) !=
+            std::tolower(static_cast<unsigned char>(LoopbackArgvMarker[i]))) {
+            return false;
+        }
+    }
+    return true;
+}
+
+std::optional<std::string> GetArgvMarker(std::string_view argv_string) {
+    auto last_token = GetLastArgvToken(argv_string);
+    if (!last_token || !IsArgvMarker(*last_token)) {
+        return std::nullopt;
+    }
+    return last_token;
+}
+
+std::optional<std::string> PrepareArgv(std::string& argv_string,
+                                       std::string_view inherited_marker,
+                                       bool append_loopback_marker) {
+    auto marker = GetArgvMarker(argv_string);
+    if (!marker && IsArgvMarker(inherited_marker)) {
+        AppendArgvToken(argv_string, inherited_marker);
+        marker = std::string{inherited_marker};
+    }
+    if (!marker && append_loopback_marker) {
+        AppendArgvToken(argv_string, LoopbackArgvMarker);
+        marker = std::string{LoopbackArgvMarker};
+    }
+    return marker;
+}
+
+void ApplyServerMode(Settings::HomebrewNxlinkServerMode mode,
+                     const std::optional<std::string>& active_marker) {
+    switch (mode) {
+    case Settings::HomebrewNxlinkServerMode::Disabled:
+        StopServer();
+        return;
+    case Settings::HomebrewNxlinkServerMode::EdenLog:
+        if (active_marker && IsLoopbackArgvMarker(*active_marker)) {
+            StartLogServer();
+            return;
+        }
+        StopServer();
+        if (active_marker) {
+            LOG_INFO(Loader,
+                     "Homebrew nxlink server mode overridden by argv marker '{}'; not starting "
+                     "local nxlink log server",
+                     *active_marker);
+        } else {
+            LOG_WARNING(Loader,
+                        "Homebrew nxlink log server requested, but no argv marker is active");
+        }
+        return;
+    case Settings::HomebrewNxlinkServerMode::HostStdout:
+    case Settings::HomebrewNxlinkServerMode::File:
+        StopServer();
+        LOG_WARNING(Loader, "Homebrew nxlink server mode {} is not implemented",
+                    static_cast<int>(mode));
+        return;
+    }
+}
+
+void StopServer() {
+    std::shared_ptr<LogServerState> state;
+    std::optional<std::jthread> thread;
+    {
+        std::scoped_lock lock{server_mutex};
+        state = std::move(server_state);
+        thread = std::move(server_thread);
+    }
+
+    if (state) {
+        boost::system::error_code ignored;
+        state->acceptor.close(ignored);
+        state->io_context.stop();
+    }
+    if (thread) {
+        thread->request_stop();
+    }
+}
+
+} // namespace Loader::HomebrewNxlink

src/core/loader/homebrew_nxlink.h

@@ -0,0 +1,30 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+#pragma once
+
+#include <cstddef>
+#include <optional>
+#include <string>
+#include <string_view>
+
+#include "common/settings_enums.h"
+
+namespace Loader::HomebrewNxlink {
+
+constexpr size_t ArgvMarkerSize = 16;
+constexpr std::string_view ArgvMarkerSuffix = "_NXLINK_";
+constexpr std::string_view LoopbackArgvMarker = "0100007F_NXLINK_";
+
+bool IsArgvMarker(std::string_view token);
+bool IsLoopbackArgvMarker(std::string_view token);
+std::optional<std::string> GetArgvMarker(std::string_view argv_string);
+std::optional<std::string> PrepareArgv(std::string& argv_string,
+                                       std::string_view inherited_marker,
+                                       bool append_loopback_marker = false);
+
+void ApplyServerMode(Settings::HomebrewNxlinkServerMode mode,
+                     const std::optional<std::string>& active_marker);
+void StopServer();
+
+} // namespace Loader::HomebrewNxlink

src/core/loader/nro.cpp

@@ -4,17 +4,20 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <algorithm>
 #include <array>
 #include <cstddef>
 #include <cstring>
 #include <optional>
 #include <string>
+#include <string_view>
 #include <utility>
 #include <vector>
 
 #include "common/alignment.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "common/fs/path_util.h"
 #include "common/logging.h"
 #include "common/settings.h"
 #include "common/random.h"
@@ -23,11 +26,14 @@
 #include "core/file_sys/control_metadata.h"
 #include "core/file_sys/romfs_factory.h"
 #include "core/file_sys/vfs/vfs_offset.h"
+#include "core/hardware_properties.h"
+#include "core/hle/api_version.h"
 #include "core/hle/kernel/code_set.h"
 #include "core/hle/kernel/k_page_table.h"
 #include "core/hle/kernel/k_process.h"
 #include "core/hle/kernel/k_thread.h"
 #include "core/hle/service/filesystem/filesystem.h"
+#include "core/loader/homebrew_nxlink.h"
 #include "core/loader/nro.h"
 #include "core/memory.h"
 
@@ -152,8 +158,432 @@ static constexpr u32 PageAlignSize(u32 size) {
     return static_cast<u32>((size + Core::Memory::YUZU_PAGEMASK) & ~Core::Memory::YUZU_PAGEMASK);
 }
 
+static std::string MakeHomebrewSdmcPath(std::string path) {
+    std::replace(path.begin(), path.end(), '\\', '/');
+    while (!path.empty() && path.front() == '/') {
+        path.erase(path.begin());
+    }
+    return "sdmc:/" + path;
+}
+
+static std::optional<std::string> TryMakeHomebrewSdmcPathFromHostPath(std::string path) {
+    std::replace(path.begin(), path.end(), '\\', '/');
+
+    std::string sdmc_root =
+        Common::FS::PathToUTF8String(Common::FS::GetEdenPath(Common::FS::EdenPath::SDMCDir));
+    std::replace(sdmc_root.begin(), sdmc_root.end(), '\\', '/');
+    while (!sdmc_root.empty() && sdmc_root.back() == '/') {
+        sdmc_root.pop_back();
+    }
+
+    if (!sdmc_root.empty() && path.size() > sdmc_root.size() &&
+        path.compare(0, sdmc_root.size(), sdmc_root) == 0 && path[sdmc_root.size()] == '/') {
+        return MakeHomebrewSdmcPath(path.substr(sdmc_root.size() + 1));
+    }
+
+    constexpr std::string_view SdmcMarker{"/sdmc/"};
+    if (const auto pos = path.rfind(SdmcMarker); pos != std::string::npos) {
+        return MakeHomebrewSdmcPath(path.substr(pos + SdmcMarker.size()));
+    }
+
+    return std::nullopt;
+}
+
+static std::string MakeHomebrewArgv0(std::string nro_path, std::string file_name) {
+    if (nro_path.empty()) {
+        nro_path = std::move(file_name);
+    }
+
+    std::replace(nro_path.begin(), nro_path.end(), '\\', '/');
+
+    if (nro_path.rfind("sdmc:/", 0) == 0) {
+        return nro_path;
+    }
+
+    if (auto sdmc_path = TryMakeHomebrewSdmcPathFromHostPath(nro_path)) {
+        return *sdmc_path;
+    }
+
+#ifdef __ANDROID__
+    if (nro_path.find('%') != std::string::npos) {
+        const auto percent_decode = [](std::string value) {
+            const auto hex_value = [](char c) -> int {
+                if (c >= '0' && c <= '9') {
+                    return c - '0';
+                }
+                if (c >= 'a' && c <= 'f') {
+                    return c - 'a' + 10;
+                }
+                if (c >= 'A' && c <= 'F') {
+                    return c - 'A' + 10;
+                }
+                return -1;
+            };
+
+            std::string decoded;
+            decoded.reserve(value.size());
+            for (size_t i = 0; i < value.size(); i++) {
+                if (value[i] == '%' && i + 2 < value.size()) {
+                    const int high = hex_value(value[i + 1]);
+                    const int low = hex_value(value[i + 2]);
+                    if (high >= 0 && low >= 0) {
+                        decoded.push_back(static_cast<char>((high << 4) | low));
+                        i += 2;
+                        continue;
+                    }
+                }
+                decoded.push_back(value[i]);
+            }
+            return decoded;
+        };
+
+        if (auto sdmc_path = TryMakeHomebrewSdmcPathFromHostPath(percent_decode(nro_path))) {
+            return *sdmc_path;
+        }
+    }
+#endif
+
+    if (!nro_path.empty() && nro_path.front() == '/') {
+        return "sdmc:" + nro_path;
+    }
+
+    return nro_path.empty() ? "homebrew" : nro_path;
+}
+
+static std::string QuoteHomebrewArgvComponent(const std::string& argument) {
+    if (argument.find_first_of(" \t\r\n") == std::string::npos) {
+        return argument;
+    }
+
+    std::string quoted{"\""};
+    quoted += argument;
+    quoted.push_back('"');
+    return quoted;
+}
+
+static std::string GetHomebrewInitialCwd(const std::string& argv0) {
+    constexpr std::string_view SdmcPrefix = "sdmc:";
+    if (argv0.substr(0, SdmcPrefix.size()) != SdmcPrefix) {
+        return {};
+    }
+
+    const auto last_slash = argv0.find_last_of('/');
+    if (last_slash == std::string_view::npos || last_slash < SdmcPrefix.size()) {
+        return {};
+    }
+
+    std::string cwd = argv0.substr(SdmcPrefix.size(), last_slash - SdmcPrefix.size());
+    if (cwd.empty()) {
+        cwd = "/";
+    }
+    while (cwd.size() > 1 && cwd.back() == '/') {
+        cwd.pop_back();
+    }
+    return cwd;
+}
+
+constexpr size_t HomebrewNextLoadPathSize = 0x200;
+constexpr size_t HomebrewNextLoadArgvSize = 0x800;
+constexpr u32 HomebrewSvcExitProcessInstruction = 0xD40000E1;
+constexpr u32 HomebrewEntryEndOfList = 0;
+constexpr u32 HomebrewEntryMainThreadHandle = 1;
+constexpr u32 HomebrewEntryNextLoadPath = 2;
+constexpr u32 HomebrewEntryOverrideHeap = 3;
+constexpr u32 HomebrewEntryArgv = 5;
+constexpr u32 HomebrewEntrySyscallAvailableHint = 6;
+constexpr u32 HomebrewEntryAppletType = 7;
+constexpr u32 HomebrewEntryProcessHandle = 10;
+constexpr u32 HomebrewEntryRandomSeed = 14;
+constexpr u32 HomebrewEntryHosVersion = 16;
+constexpr u32 HomebrewEntrySyscallAvailableHint2 = 17;
+constexpr u32 HomebrewAppletTypeApplication = 0;
+constexpr u64 HomebrewAllSvcHints = ~u64{0};
+constexpr u32 HomebrewHosVersion = (u32{HLE::ApiVersion::HOS_VERSION_MAJOR} << 16) |
+                                   (u32{HLE::ApiVersion::HOS_VERSION_MINOR} << 8) |
+                                   u32{HLE::ApiVersion::HOS_VERSION_MICRO};
+
+struct HomebrewConfigEntry {
+    u32_le key;
+    u32_le flags;
+    u64_le value[2];
+};
+static_assert(sizeof(HomebrewConfigEntry) == 0x18);
+constexpr size_t HomebrewBaseConfigEntryCount = 10;
+constexpr size_t HomebrewInPlaceConfigEntryCount = 11;
+constexpr size_t HomebrewBaseConfigTableSize =
+    HomebrewBaseConfigEntryCount * sizeof(HomebrewConfigEntry);
+constexpr size_t HomebrewInPlaceConfigTableSize =
+    HomebrewInPlaceConfigEntryCount * sizeof(HomebrewConfigEntry);
+
+struct HomebrewNroImage {
+    Kernel::CodeSet codeset;
+    size_t image_size{};
+    size_t args_offset{};
+    std::optional<size_t> exit_process_offset;
+    std::optional<std::string> nxlink_argv_marker;
+    std::string argv_string;
+};
+
+static void SetHomebrewConfigPointers(Kernel::KProcess& process, u64 config_addr,
+                                      u64 next_load_path_addr, u64 next_load_argv_addr) {
+    constexpr size_t MainThreadHandleEntryIndex = 0;
+    constexpr size_t ProcessHandleEntryIndex = 1;
+    constexpr size_t EntryValueOffset = offsetof(HomebrewConfigEntry, value);
+
+    process.SetArgPointer(Kernel::KProcessAddress{config_addr});
+    process.SetMainThreadHandleAddr(Kernel::KProcessAddress{
+        config_addr + MainThreadHandleEntryIndex * sizeof(HomebrewConfigEntry) + EntryValueOffset});
+    process.SetProcessHandleAddr(Kernel::KProcessAddress{
+        config_addr + ProcessHandleEntryIndex * sizeof(HomebrewConfigEntry) + EntryValueOffset});
+    process.SetHomebrewNextLoadBufferAddrs(Kernel::KProcessAddress{next_load_path_addr},
+                                           Kernel::KProcessAddress{next_load_argv_addr});
+}
+
+static std::optional<HomebrewNroImage> BuildHomebrewNroImage(const std::vector<u8>& data,
+                                                             std::string nro_path,
+                                                             std::string file_name,
+                                                             std::string launch_argv,
+                                                             std::string_view inherited_marker,
+                                                             bool append_loopback_nxlink_marker) {
+    if (data.size() < sizeof(NroHeader)) {
+        return std::nullopt;
+    }
+
+    NroHeader nro_header{};
+    std::memcpy(&nro_header, data.data(), sizeof(NroHeader));
+    if (nro_header.magic != Common::MakeMagic('N', 'R', 'O', '0')) {
+        return std::nullopt;
+    }
+    if (data.size() < nro_header.file_size ||
+        nro_header.module_header_offset + sizeof(ModHeader) > PageAlignSize(nro_header.file_size)) {
+        return std::nullopt;
+    }
+
+    std::vector<u8> program_image(PageAlignSize(nro_header.file_size));
+    std::memcpy(program_image.data(), data.data(), nro_header.file_size);
+
+    Kernel::CodeSet codeset;
+    for (std::size_t i = 0; i < nro_header.segments.size(); ++i) {
+        codeset.segments[i].addr = nro_header.segments[i].offset;
+        codeset.segments[i].offset = nro_header.segments[i].offset;
+        codeset.segments[i].size = PageAlignSize(nro_header.segments[i].size);
+    }
+
+    u32 bss_size{PageAlignSize(nro_header.bss_size)};
+    ModHeader mod_header{};
+    std::memcpy(&mod_header, program_image.data() + nro_header.module_header_offset,
+                sizeof(ModHeader));
+
+    if (mod_header.magic == Common::MakeMagic('M', 'O', 'D', '0')) {
+        bss_size = PageAlignSize(mod_header.bss_end_offset - mod_header.bss_start_offset);
+    }
+
+    codeset.DataSegment().size += bss_size;
+    program_image.resize(static_cast<u32>(program_image.size()) + bss_size);
+
+    HomebrewNroImage image{.codeset = std::move(codeset)};
+    const auto argv0 = MakeHomebrewArgv0(std::move(nro_path), std::move(file_name));
+
+    if (!launch_argv.empty()) {
+        image.argv_string = std::move(launch_argv);
+    } else {
+        image.argv_string = QuoteHomebrewArgvComponent(argv0);
+    }
+    image.nxlink_argv_marker = HomebrewNxlink::PrepareArgv(
+        image.argv_string, inherited_marker, append_loopback_nxlink_marker);
+    if (image.argv_string.empty() || image.argv_string.back() != '\0') {
+        image.argv_string.push_back('\0');
+    }
+
+    const auto& code = image.codeset.CodeSegment();
+    const size_t code_end = (std::min)(program_image.size(), code.offset + code.size);
+    for (size_t offset = code.offset; offset + sizeof(u32) <= code_end; offset += sizeof(u32)) {
+        u32 instruction{};
+        std::memcpy(&instruction, program_image.data() + offset, sizeof(instruction));
+        if (instruction == HomebrewSvcExitProcessInstruction) {
+            image.exit_process_offset = offset;
+            break;
+        }
+    }
+
+    const size_t entries_and_buffers =
+        Common::AlignUp(HomebrewInPlaceConfigTableSize + HomebrewNextLoadPathSize +
+                            HomebrewNextLoadArgvSize + image.argv_string.size(),
+                        Core::Memory::YUZU_PAGESIZE);
+
+    image.args_offset = program_image.size();
+    image.codeset.DataSegment().size += static_cast<u32>(entries_and_buffers);
+    program_image.resize(image.args_offset + entries_and_buffers);
+    image.image_size = program_image.size();
+    image.codeset.memory = std::move(program_image);
+
+    return image;
+}
+
+bool LoadNroInPlace(Core::System& system, Kernel::KProcess& process, Kernel::KThread& thread,
+                    const FileSys::VirtualFile& nro_file, const std::string& nro_path,
+                    const std::string& launch_argv) {
+    if (!nro_file) {
+        return false;
+    }
+
+#ifdef HAS_NCE
+    if (Settings::IsNceEnabled()) {
+        LOG_WARNING(Loader,
+                    "Homebrew next-load: in-place handoff unavailable because NCE is enabled");
+        return false;
+    }
+#endif
+
+    size_t live_threads = 0;
+    for (auto& candidate : process.GetThreadList()) {
+        if (candidate.GetState() != Kernel::ThreadState::Terminated) {
+            live_threads++;
+        }
+    }
+    if (live_threads != 1) {
+        LOG_WARNING(Loader, "NextLoad: in-place handoff failed because live_threads={}",
+                    live_threads);
+        return false;
+    }
+
+    const auto stack_top = process.GetMainThreadStackTop();
+    if (GetInteger(stack_top) == 0) {
+        LOG_WARNING(Loader, "NextLoad: in-place handoff failed because stack_top=0");
+        return false;
+    }
+
+    const auto nxlink_server_mode = Settings::values.homebrew_nxlink_server_mode.GetValue();
+    const bool append_loopback_nxlink_marker =
+        nxlink_server_mode != Settings::HomebrewNxlinkServerMode::Disabled;
+    auto image = BuildHomebrewNroImage(
+        nro_file->ReadAllBytes(), nro_path, nro_file->GetName(), launch_argv,
+        process.GetHomebrewNxlinkArgvMarker(), append_loopback_nxlink_marker);
+    if (!image) {
+        LOG_WARNING(Loader, "NextLoad: in-place handoff failed because '{}' is invalid",
+                    nro_path);
+        return false;
+    }
+
+    const auto capacity = process.GetCodeSize();
+    if (image->image_size > capacity) {
+        LOG_WARNING(Loader,
+                    "NextLoad: in-place handoff failed because image_size=0x{:X} exceeds "
+                    "capacity=0x{:X}",
+                    image->image_size, capacity);
+        return false;
+    }
+
+    const u64 base = GetInteger(process.GetEntryPoint());
+    const u64 heap_addr = GetInteger(process.GetPageTable().GetHeapRegionStart());
+    const size_t heap_size = process.GetPageTable().GetBasePageTable().GetCurrentHeapSize();
+    if (heap_addr == 0 || heap_size == 0) {
+        LOG_WARNING(Loader,
+                    "NextLoad: in-place handoff failed because heap override is "
+                    "unavailable (addr=0x{:016X}, size=0x{:X})",
+                    heap_addr, heap_size);
+        return false;
+    }
+
+    const u64 config_addr = base + image->args_offset;
+    const u64 next_load_path_addr = config_addr + HomebrewInPlaceConfigTableSize;
+    const u64 next_load_argv_addr = next_load_path_addr + HomebrewNextLoadPathSize;
+    const u64 argv_addr = next_load_argv_addr + HomebrewNextLoadArgvSize;
+    const u64 argv_entry_addr = image->argv_string.empty() ? 0 : argv_addr;
+    const std::string argv0 = MakeHomebrewArgv0(nro_path, nro_file->GetName());
+    const std::string homebrew_initial_cwd = GetHomebrewInitialCwd(argv0);
+    u64 program_id{};
+    AppLoader_NRO loader{nro_file};
+    if (loader.ReadProgramId(program_id) != Loader::ResultStatus::Success) {
+        LOG_WARNING(Loader, "NextLoad: in-place handoff could not read NRO program id");
+    }
+
+    Kernel::Handle main_thread_handle{};
+    if (process.GetHandleTable().Add(system.Kernel(), std::addressof(main_thread_handle), &thread)
+            .IsError()) {
+        LOG_WARNING(Loader,
+                    "NextLoad: in-place handoff failed because main thread handle failed");
+        return false;
+    }
+
+    Kernel::Handle process_handle{};
+    if (process.GetHandleTable().Add(system.Kernel(), std::addressof(process_handle), &process)
+            .IsError()) {
+        LOG_WARNING(Loader,
+                    "NextLoad: in-place handoff failed because process handle failed");
+        return false;
+    }
+
+    if (!process.GetMemory().ZeroBlock(Common::ProcessAddress{heap_addr}, heap_size)) {
+        LOG_WARNING(Loader,
+                    "NextLoad: in-place handoff failed because heap clear failed "
+                    "(addr=0x{:016X}, size=0x{:X})",
+                    heap_addr, heap_size);
+        return false;
+    }
+
+    process.LoadModule(system.Kernel(), std::move(image->codeset), process.GetEntryPoint());
+
+    const HomebrewConfigEntry entries[HomebrewInPlaceConfigEntryCount] = {
+        {HomebrewEntryMainThreadHandle, 0, {main_thread_handle, 0}},
+        {HomebrewEntryProcessHandle, 0, {process_handle, 0}},
+        {HomebrewEntryNextLoadPath, 0, {next_load_path_addr, next_load_argv_addr}},
+        {HomebrewEntryOverrideHeap, 0, {heap_addr, heap_size}},
+        {HomebrewEntryAppletType, 0, {HomebrewAppletTypeApplication, 0}},
+        {HomebrewEntryArgv, 0, {0, argv_entry_addr}},
+        {HomebrewEntrySyscallAvailableHint, 0, {HomebrewAllSvcHints, HomebrewAllSvcHints}},
+        {HomebrewEntryRandomSeed, 0,
+         {process.GetRandomEntropy(0), process.GetRandomEntropy(1)}},
+        {HomebrewEntryHosVersion, 0, {HomebrewHosVersion, 0}},
+        {HomebrewEntrySyscallAvailableHint2, 0, {HomebrewAllSvcHints, 0}},
+        {HomebrewEntryEndOfList, 0, {0, 0}},
+    };
+
+    process.GetMemory().WriteBlock(Common::ProcessAddress{config_addr}, entries, sizeof(entries));
+    process.GetMemory().WriteBlock(Common::ProcessAddress{argv_addr}, image->argv_string.data(),
+                                   image->argv_string.size());
+    process.GetMemory().Write32(Common::ProcessAddress{GetInteger(thread.GetTlsAddress()) + 0x110},
+                                main_thread_handle);
+
+    process.SetArgReturnAddress(Kernel::KProcessAddress{
+        image->exit_process_offset ? base + *image->exit_process_offset : 0});
+    SetHomebrewConfigPointers(process, config_addr, next_load_path_addr, next_load_argv_addr);
+    if (image->nxlink_argv_marker) {
+        process.SetHomebrewNxlinkArgvMarker(*image->nxlink_argv_marker);
+    } else {
+        process.ClearHomebrewNxlinkArgvMarker();
+    }
+    HomebrewNxlink::ApplyServerMode(nxlink_server_mode, image->nxlink_argv_marker);
+    system.GetFileSystemController().RegisterProcess(
+        process.GetProcessId(), program_id,
+        std::make_unique<FileSys::RomFSFactory>(loader, system.GetContentProvider(),
+                                                system.GetFileSystemController()),
+        homebrew_initial_cwd);
+    process.SetHomebrewInPlaceNextLoad(true);
+
+    auto& context = thread.GetContext();
+    context = {};
+    context.r[0] = config_addr;
+    context.r[1] = UINT64_MAX;
+    context.r[18] = Common::Random::Random64(0) | 1;
+    context.lr = image->exit_process_offset ? base + *image->exit_process_offset : 0;
+    context.pc = base;
+    context.sp = GetInteger(stack_top);
+    context.fpcr = 0;
+    context.fpsr = 0;
+
+    for (std::size_t core = 0; core < Core::Hardware::NUM_CPU_CORES; core++) {
+        if (auto* arm = process.GetArmInterface(core); arm != nullptr) {
+            arm->ClearInstructionCache();
+        }
+    }
+
+    return true;
+}
+
 static bool LoadNroImpl(Core::System& system, Kernel::KProcess& process,
-                        const std::vector<u8>& data) {
+                        const std::vector<u8>& data, std::string nro_path,
+                        std::string file_name) {
     if (data.size() < sizeof(NroHeader)) {
         return {};
     }
@@ -195,47 +625,49 @@ static bool LoadNroImpl(Core::System& system, Kernel::KProcess& process,
 
     codeset.DataSegment().size += bss_size;
     program_image.resize(static_cast<u32>(program_image.size()) + bss_size);
-    struct ConfigEntry {
-        u32_le key;
-        u32_le flags;
-        u64_le value[2];
-    };
-    static_assert(sizeof(ConfigEntry) == 0x18);
-    // AArch64 encoding for svc #0x7 (ExitProcess).
-    constexpr u32 kSvcExitProcessInstruction = 0xD40000E1;
-    constexpr size_t kNumEntries = 4; // MainThreadHandle, AppletType, Argv, EndOfList
-    constexpr size_t kConfigTableSize = kNumEntries * sizeof(ConfigEntry);
     std::string argv_string;
     size_t args_offset_in_image = 0;
     std::optional<size_t> exit_process_offset_in_image;
     const auto& program_args = Settings::values.program_args.GetValue();
+    const std::string argv0 = MakeHomebrewArgv0(std::move(nro_path), std::move(file_name));
+    argv_string = QuoteHomebrewArgvComponent(argv0);
     if (!program_args.empty()) {
-        argv_string = "homebrew ";
+        argv_string.push_back(' ');
         argv_string += program_args;
-        argv_string.push_back('\0');
-
-        const auto& code = codeset.CodeSegment();
-        const size_t code_end = (std::min)(program_image.size(), code.offset + code.size);
-        for (size_t offset = code.offset; offset + sizeof(u32) <= code_end; offset += sizeof(u32)) {
-            u32 instruction{};
-            std::memcpy(&instruction, program_image.data() + offset, sizeof(instruction));
-            if (instruction == kSvcExitProcessInstruction) {
-                exit_process_offset_in_image = offset;
-                break;
-            }
-        }
-        if (!exit_process_offset_in_image) {
-            LOG_WARNING(Loader,
-                        "Unable to find svcExitProcess in NRO; returning from main may fault");
-        }
-
-        const size_t entries_and_argv =
-            Common::AlignUp(kConfigTableSize + argv_string.size(), Core::Memory::YUZU_PAGESIZE);
-
-        args_offset_in_image = program_image.size();
-        codeset.DataSegment().size += static_cast<u32>(entries_and_argv);
-        program_image.resize(args_offset_in_image + entries_and_argv);
     }
+    const auto nxlink_server_mode = Settings::values.homebrew_nxlink_server_mode.GetValue();
+    const bool append_loopback_nxlink_marker =
+        nxlink_server_mode != Settings::HomebrewNxlinkServerMode::Disabled;
+    const auto nxlink_argv_marker =
+        HomebrewNxlink::PrepareArgv(argv_string, {}, append_loopback_nxlink_marker);
+    if (argv_string.empty() || argv_string.back() != '\0') {
+        argv_string.push_back('\0');
+    }
+
+    const auto& code_segment = codeset.CodeSegment();
+    const size_t code_end =
+        (std::min)(program_image.size(), code_segment.offset + code_segment.size);
+    for (size_t offset = code_segment.offset; offset + sizeof(u32) <= code_end;
+         offset += sizeof(u32)) {
+        u32 instruction{};
+        std::memcpy(&instruction, program_image.data() + offset, sizeof(instruction));
+        if (instruction == HomebrewSvcExitProcessInstruction) {
+            exit_process_offset_in_image = offset;
+            break;
+        }
+    }
+    if (!exit_process_offset_in_image) {
+        LOG_WARNING(Loader, "Unable to find svcExitProcess in NRO; returning from main may fault");
+    }
+
+    const size_t entries_and_buffers =
+        Common::AlignUp(HomebrewBaseConfigTableSize + HomebrewNextLoadPathSize +
+                            HomebrewNextLoadArgvSize + argv_string.size(),
+                        Core::Memory::YUZU_PAGESIZE);
+
+    args_offset_in_image = program_image.size();
+    codeset.DataSegment().size += static_cast<u32>(entries_and_buffers);
+    program_image.resize(args_offset_in_image + entries_and_buffers);
     size_t image_size = program_image.size();
 
 #ifdef HAS_NCE
@@ -263,6 +695,9 @@ static bool LoadNroImpl(Core::System& system, Kernel::KProcess& process,
         image_size += patch_segment.size;
     }
 #endif
+    // In-place NextLoad reuses the original code mapping; leave room for larger homebrew cores.
+    constexpr size_t HomebrewCodeArenaSize = 192 * 1024 * 1024;
+    image_size = (std::max)(image_size, HomebrewCodeArenaSize);
 
     // Enable direct memory mapping in case of NCE.
     const u64 fastmem_base = [&]() -> size_t {
@@ -297,31 +732,44 @@ static bool LoadNroImpl(Core::System& system, Kernel::KProcess& process,
     // Load codeset for current process
     codeset.memory = std::move(program_image);
     process.LoadModule(system.Kernel(), std::move(codeset), process.GetEntryPoint());
-    if (!argv_string.empty()) {
-        constexpr u32 kEntryEndOfList = 0;
-        constexpr u32 kEntryMainThreadHandle = 1;
-        constexpr u32 kEntryArgv = 5;
-        constexpr u32 kEntryAppletType = 7;
-        constexpr u32 kAppletTypeApplication = 0;
-
+    process.SetHomebrewInPlaceNextLoad(false);
+    if (nxlink_argv_marker) {
+        process.SetHomebrewNxlinkArgvMarker(*nxlink_argv_marker);
+    } else {
+        process.ClearHomebrewNxlinkArgvMarker();
+    }
+    HomebrewNxlink::ApplyServerMode(nxlink_server_mode, nxlink_argv_marker);
+    {
         const u64 base = GetInteger(process.GetEntryPoint());
         const u64 config_addr = base + args_offset_in_image;
-        const u64 argv_addr = config_addr + kConfigTableSize;
+        const u64 next_load_path_addr = config_addr + HomebrewBaseConfigTableSize;
+        const u64 next_load_argv_addr = next_load_path_addr + HomebrewNextLoadPathSize;
+        const u64 argv_addr = next_load_argv_addr + HomebrewNextLoadArgvSize;
+        const u64 argv_entry_addr = argv_string.empty() ? 0 : argv_addr;
 
-        const ConfigEntry entries[kNumEntries] = {
-            {kEntryMainThreadHandle, 0, {0, 0}}, // Value[0] patched in Run()
-            {kEntryAppletType,       0, {kAppletTypeApplication, 0}},
-            {kEntryArgv,             0, {0, argv_addr}},
-            {kEntryEndOfList,        0, {0, 0}},
+        const HomebrewConfigEntry entries[HomebrewBaseConfigEntryCount] = {
+            {HomebrewEntryMainThreadHandle, 0, {0, 0}}, // Value[0] patched in Run()
+            {HomebrewEntryProcessHandle, 0, {0, 0}},    // Value[0] patched in Run()
+            {HomebrewEntryNextLoadPath, 0, {next_load_path_addr, next_load_argv_addr}},
+            {HomebrewEntryAppletType, 0, {HomebrewAppletTypeApplication, 0}},
+            {HomebrewEntryArgv, 0, {0, argv_entry_addr}},
+            {HomebrewEntrySyscallAvailableHint, 0, {HomebrewAllSvcHints, HomebrewAllSvcHints}},
+            {HomebrewEntryRandomSeed, 0,
+             {process.GetRandomEntropy(0), process.GetRandomEntropy(1)}},
+            {HomebrewEntryHosVersion, 0, {HomebrewHosVersion, 0}},
+            {HomebrewEntrySyscallAvailableHint2, 0, {HomebrewAllSvcHints, 0}},
+            {HomebrewEntryEndOfList, 0, {0, 0}},
         };
-        process.GetMemory().WriteBlock(Common::ProcessAddress{config_addr}, entries, sizeof(entries));
-        process.GetMemory().WriteBlock(Common::ProcessAddress{argv_addr}, argv_string.data(), argv_string.size());
-        constexpr size_t kMainThreadHandleValueOffset = offsetof(ConfigEntry, value);
-        process.SetArgPointer(Kernel::KProcessAddress{config_addr});
+        process.GetMemory().WriteBlock(Common::ProcessAddress{config_addr}, entries,
+                                       sizeof(entries));
+        if (!argv_string.empty()) {
+            process.GetMemory().WriteBlock(Common::ProcessAddress{argv_addr}, argv_string.data(),
+                                           argv_string.size());
+        }
         if (exit_process_offset_in_image) {
             process.SetArgReturnAddress(Kernel::KProcessAddress{base + *exit_process_offset_in_image});
         }
-        process.SetMainThreadHandleAddr(Kernel::KProcessAddress{config_addr + kMainThreadHandleValueOffset});
+        SetHomebrewConfigPointers(process, config_addr, next_load_path_addr, next_load_argv_addr);
     }
 
     return true;
@@ -329,7 +777,8 @@ static bool LoadNroImpl(Core::System& system, Kernel::KProcess& process,
 
 bool AppLoader_NRO::LoadNro(Core::System& system, Kernel::KProcess& process,
                             const FileSys::VfsFile& nro_file) {
-    return LoadNroImpl(system, process, nro_file.ReadAllBytes());
+    return LoadNroImpl(system, process, nro_file.ReadAllBytes(), nro_file.GetFullPath(),
+                       nro_file.GetName());
 }
 
 AppLoader_NRO::LoadResult AppLoader_NRO::Load(Kernel::KProcess& process, Core::System& system) {
@@ -343,10 +792,13 @@ AppLoader_NRO::LoadResult AppLoader_NRO::Load(Kernel::KProcess& process, Core::S
 
     u64 program_id{};
     ReadProgramId(program_id);
+    const std::string argv0 = MakeHomebrewArgv0(file->GetFullPath(), file->GetName());
+    const std::string homebrew_initial_cwd = GetHomebrewInitialCwd(argv0);
     system.GetFileSystemController().RegisterProcess(
         process.GetProcessId(), program_id,
         std::make_unique<FileSys::RomFSFactory>(*this, system.GetContentProvider(),
-                                                system.GetFileSystemController()));
+                                                system.GetFileSystemController()),
+        homebrew_initial_cwd);
 
     is_loaded = true;
     return {ResultStatus::Success, LoadParameters{Kernel::KThread::DefaultThreadPriority,

src/core/loader/nro.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -19,10 +22,15 @@ class NACP;
 
 namespace Kernel {
 class KProcess;
+class KThread;
 }
 
 namespace Loader {
 
+[[nodiscard]] bool LoadNroInPlace(Core::System& system, Kernel::KProcess& process,
+                                  Kernel::KThread& thread, const FileSys::VirtualFile& nro_file,
+                                  const std::string& nro_path, const std::string& launch_argv);
+
 /// Loads an NRO file
 class AppLoader_NRO final : public AppLoader {
 public:

src/hid_core/frontend/emulated_console.cpp

@@ -100,6 +100,7 @@ void EmulatedConsole::ReloadInput() {
     motion.gyro = emulated_motion.GetGyroscope();
     motion.rotation = emulated_motion.GetRotations();
     motion.orientation = emulated_motion.GetOrientation();
+    motion.quaternion = emulated_motion.GetQuaternion();
     motion.is_at_rest = !emulated_motion.IsMoving(motion_sensitivity);
 
     // Unique index for identifying touch device source

src/input_common/drivers/sdl_driver.cpp

@@ -648,6 +648,13 @@ SDLDriver::SDLDriver(std::string input_engine_) : InputEngine(std::move(input_en
     // Disable raw input. When enabled this setting causes SDL to die when a web applet opens
     SDL_SetHint(SDL_HINT_JOYSTICK_RAWINPUT, Settings::values.enable_raw_input ? "1" : "0");
 
+#ifdef _WIN32
+    if (Settings::values.disable_wgi_xinput) {
+        SDL_SetHintWithPriority(SDL_HINT_JOYSTICK_RAWINPUT_CORRELATE_XINPUT, "0", SDL_HINT_OVERRIDE);
+        SDL_SetHintWithPriority(SDL_HINT_JOYSTICK_WGI, "0", SDL_HINT_OVERRIDE);
+    }
+#endif
+
     // SDL3 defaults Steam Controller Bluetooth HIDAPI support to off, which can disable gyro.
     SDL_SetHint(SDL_HINT_JOYSTICK_HIDAPI_STEAM, "1");
     SDL_SetHint(SDL_HINT_GAMECONTROLLER_SENSOR_FUSION, "1");

src/qt_common/config/shared_translation.cpp

@@ -302,6 +302,8 @@ std::unique_ptr<TranslationMap> InitializeTranslations(QObject* parent) {
     INSERT(Settings, device_name, tr("Device Name"), tr("The name of the console."));
     INSERT(Settings, program_args, tr("Homebrew Args"),
            tr("Command-line arguments passed to homebrew at launch (e.g. -noglsl)."));
+    INSERT(Settings, homebrew_nxlink_server_mode, tr("nxlink Server"),
+           tr("Starts a local nxlink server for homebrew stdout/stderr streams."));
     INSERT(Settings, custom_rtc, tr("Custom RTC Date:"),
            tr("This option allows to change the clock of the console.\n"
               "Can be used to manipulate time in games."));

src/qt_common/game_list/model.cpp

@@ -159,7 +159,9 @@ void GameListModel::RemoveFavorite(u64 program_id) {
 
 void GameListModel::Repopulate() {
     current_worker.reset();
-    QtCommon::system->GetFileSystemController().CreateFactories(*QtCommon::vfs);
+    if (!QtCommon::system->IsPoweredOn()) {
+        QtCommon::system->GetFileSystemController().CreateFactories(*QtCommon::vfs);
+    }
     PopulateAsync(UISettings::values.game_dirs);
 }
 

src/video_core/CMakeLists.txt

@@ -33,7 +33,7 @@ add_library(video_core STATIC
     control/channel_state_cache.h
     control/scheduler.cpp
     control/scheduler.h
-    deferred_destruction_queue.h
+    delayed_destruction_ring.h
     dirty_flags.cpp
     dirty_flags.h
     dma_pusher.cpp

src/video_core/buffer_cache/buffer_cache.h

@@ -31,77 +31,44 @@ BufferCache<P>::BufferCache(Tegra::MaxwellDeviceMemoryManager& device_memory_, R
     immediately_free = (Settings::values.vram_usage_mode.GetValue() == Settings::VramUsageMode::Aggressive);
 #endif
     if (!runtime.CanReportMemoryUsage()) {
-        memory_budget = FALLBACK_MEMORY_BUDGET;
+        minimum_memory = DEFAULT_EXPECTED_MEMORY;
+        critical_memory = DEFAULT_CRITICAL_MEMORY;
         return;
     }
 
-    memory_budget = runtime.GetDeviceLocalMemory();
+    const s64 device_local_memory = static_cast<s64>(runtime.GetDeviceLocalMemory());
+    const s64 min_spacing_expected = device_local_memory - 1_GiB;
+    const s64 min_spacing_critical = device_local_memory - 512_MiB;
+    const s64 mem_threshold = (std::min)(device_local_memory, TARGET_THRESHOLD);
+    const s64 min_vacancy_expected = (6 * mem_threshold) / 10;
+    const s64 min_vacancy_critical = (2 * mem_threshold) / 10;
+    minimum_memory = static_cast<u64>(
+        (std::max)((std::min)(device_local_memory - min_vacancy_expected, min_spacing_expected),
+                 DEFAULT_EXPECTED_MEMORY));
+    critical_memory = static_cast<u64>(
+        (std::max)((std::min)(device_local_memory - min_vacancy_critical, min_spacing_critical),
+                 DEFAULT_CRITICAL_MEMORY));
 }
 
 template <class P>
 BufferCache<P>::~BufferCache() = default;
 
 template <class P>
-u64 BufferCache<P>::DeviceUsage(bool force_refresh) {
-    if (!runtime.CanReportAllocationUsage()) {
-        return total_used_memory;
-    }
-    if (force_refresh || usage_refresh_countdown == 0) {
-        cached_device_usage = runtime.GetDeviceAllocationUsage();
-        usage_refresh_countdown = USAGE_REFRESH_INTERVAL;
-    } else {
-        --usage_refresh_countdown;
-    }
-    return cached_device_usage;
-}
-
-template <class P>
-u64 BufferCache<P>::ReclaimMemory(u64 target_bytes, bool allow_download) {
-    if (target_bytes == 0 || in_reclaim) {
-        return 0;
-    }
-    in_reclaim = true;
-    sentenced_buffers.Reclaim(runtime.CompletedSyncPoint());
-    u64 freed = 0;
-    const auto clean_up = [&](BufferId buffer_id) {
-        if (freed >= target_bytes) {
+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) {
+        if (num_iterations == 0) {
             return true;
         }
+        --num_iterations;
         auto& buffer = slot_buffers[buffer_id];
-        if (!allow_download && IsRegionGpuModified(buffer.CpuAddr(), buffer.SizeBytes())) {
-            return false;
-        }
-        const u64 buffer_bytes = Common::AlignUp(buffer.SizeBytes(), 1024);
         DownloadBufferMemory(buffer);
         DeleteBuffer(buffer_id);
-        freed += buffer_bytes;
         return false;
     };
-    const u64 cold_tick =
-        frame_tick > RECLAIM_GUARD_FRAMES ? frame_tick - RECLAIM_GUARD_FRAMES : 0;
-    lru_cache.ForEachItemBelow(cold_tick, clean_up);
-    if (freed < target_bytes) {
-        lru_cache.ForEachItemBelow(frame_tick > 0 ? frame_tick - 1 : 0, clean_up);
-    }
-    sentenced_buffers.Reclaim(runtime.CompletedSyncPoint());
-    in_reclaim = false;
-    usage_refresh_countdown = 0;
-    reclaim_stalled = freed == 0;
-    return freed;
-}
-
-template <class P>
-void BufferCache<P>::EnsureHeadroom(bool allow_download) {
-    if (reclaim_stalled) {
-        return;
-    }
-    const u64 limit = memory_budget > RECLAIM_HEADROOM ? memory_budget - RECLAIM_HEADROOM : 0;
-    const u64 usage = DeviceUsage(false);
-    if (usage <= limit) {
-        return;
-    }
-    const u64 target = (limit / 100) * RECLAIM_TARGET_PERCENT;
-    ReclaimMemory((std::min)(usage - target, total_used_memory), allow_download);
+    lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, clean_up);
 }
 
 template <class P>
@@ -129,11 +96,15 @@ void BufferCache<P>::TickFrame() {
     const bool skip_preferred = hits * 256 < shots * 251;
     channel_state->uniform_buffer_skip_cache_size = skip_preferred ? DEFAULT_SKIP_CACHE_SIZE : 0;
 
-    usage_refresh_countdown = 0;
-    reclaim_stalled = false;
-    EnsureHeadroom(true);
+    // If we can obtain the memory info, use it instead of the estimate.
+    if (runtime.CanReportMemoryUsage()) {
+        total_used_memory = runtime.GetDeviceMemoryUsage();
+    }
+    if (total_used_memory >= minimum_memory) {
+        RunGarbageCollector();
+    }
     ++frame_tick;
-    sentenced_buffers.Reclaim(runtime.CompletedSyncPoint());
+    delayed_destruction_ring.Tick();
 
     for (auto& buffer : async_buffers_death_ring) {
         runtime.FreeDeferredStagingBuffer(buffer);
@@ -1605,7 +1576,6 @@ void BufferCache<P>::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id,
 
 template <class P>
 BufferId BufferCache<P>::CreateBuffer(DAddr device_addr, u32 wanted_size) {
-    EnsureHeadroom(false);
     DAddr device_addr_end = Common::AlignUp(device_addr + wanted_size, CACHING_PAGESIZE);
     device_addr = Common::AlignDown(device_addr, CACHING_PAGESIZE);
     wanted_size = static_cast<u32>(device_addr_end - device_addr);
@@ -1643,7 +1613,7 @@ void BufferCache<P>::ChangeRegister(BufferId buffer_id) {
         total_used_memory += Common::AlignUp(size, 1024);
         buffer.setLRUID(lru_cache.Insert(buffer_id, frame_tick));
     } else {
-        total_used_memory -= std::min<u64>(total_used_memory, Common::AlignUp(size, 1024));
+        total_used_memory -= Common::AlignUp(size, 1024);
         lru_cache.Free(buffer.getLRUID());
     }
     const DAddr device_addr_begin = buffer.CpuAddr();
@@ -1902,7 +1872,7 @@ void BufferCache<P>::DeleteBuffer(BufferId buffer_id, bool do_not_mark) {
 #ifdef YUZU_LEGACY
     if (!do_not_mark || !immediately_free)
 #endif
-        sentenced_buffers.Push(std::move(slot_buffers[buffer_id]), runtime.CurrentSyncPoint());
+        delayed_destruction_ring.Push(std::move(slot_buffers[buffer_id]));
 
     slot_buffers.erase(buffer_id);
 

src/video_core/buffer_cache/buffer_cache_base.h

@@ -9,7 +9,6 @@
 #include <algorithm>
 #include <array>
 #include <bit>
-#include <deque>
 #include <functional>
 #include <memory>
 #include <mutex>
@@ -31,7 +30,7 @@
 #include "common/slot_vector.h"
 #include "video_core/buffer_cache/buffer_base.h"
 #include "video_core/control/channel_state_cache.h"
-#include "video_core/deferred_destruction_queue.h"
+#include "video_core/delayed_destruction_ring.h"
 #include "video_core/dirty_flags.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/kepler_compute.h"
@@ -183,15 +182,13 @@ class BufferCache : public VideoCommon::ChannelSetupCaches<BufferCacheChannelInf
     static constexpr bool USE_MEMORY_MAPS_FOR_UPLOADS = P::USE_MEMORY_MAPS_FOR_UPLOADS;
 
 #ifdef YUZU_LEGACY
-    static constexpr u64 RECLAIM_HEADROOM = 384_MiB;
+    static constexpr s64 TARGET_THRESHOLD = 3_GiB;
 #else
-    static constexpr u64 RECLAIM_HEADROOM = 512_MiB;
+    static constexpr s64 TARGET_THRESHOLD = 4_GiB;
 #endif
 
-    static constexpr u64 FALLBACK_MEMORY_BUDGET = 2_GiB;
-    static constexpr u32 USAGE_REFRESH_INTERVAL = 16;
-    static constexpr u64 RECLAIM_GUARD_FRAMES = 8;
-    static constexpr u64 RECLAIM_TARGET_PERCENT = 88;
+    static constexpr s64 DEFAULT_EXPECTED_MEMORY = 512_MiB;
+    static constexpr s64 DEFAULT_CRITICAL_MEMORY = 1_GiB;
 
     // Debug Flags.
 
@@ -218,8 +215,6 @@ public:
 
     void TickFrame();
 
-    u64 ReclaimMemory(u64 target_bytes, bool allow_download);
-
     void WriteMemory(DAddr device_addr, u64 size);
 
     void CachedWriteMemory(DAddr device_addr, u64 size);
@@ -363,9 +358,7 @@ private:
                ((device_addr + size) & ~Core::DEVICE_PAGEMASK);
     }
 
-    u64 DeviceUsage(bool force_refresh);
-
-    void EnsureHeadroom(bool allow_download);
+    void RunGarbageCollector();
 
     void BindHostIndexBuffer();
 
@@ -482,7 +475,12 @@ private:
     Tegra::MaxwellDeviceMemoryManager& device_memory;
 
     Common::SlotVector<Buffer> slot_buffers;
-    DeferredDestructionQueue<Buffer> sentenced_buffers;
+#ifdef YUZU_LEGACY
+    static constexpr size_t TICKS_TO_DESTROY = 6;
+#else
+    static constexpr size_t TICKS_TO_DESTROY = 8;
+#endif
+    DelayedDestructionRing<Buffer, TICKS_TO_DESTROY> delayed_destruction_ring;
 
     const Tegra::Engines::Maxwell3D::DrawManager::IndirectParams* current_draw_indirect{};
 
@@ -517,11 +515,8 @@ private:
     Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
     u64 frame_tick = 0;
     u64 total_used_memory = 0;
-    u64 memory_budget = 0;
-    u64 cached_device_usage = 0;
-    u32 usage_refresh_countdown = 0;
-    bool in_reclaim = false;
-    bool reclaim_stalled = false;
+    u64 minimum_memory = 0;
+    u64 critical_memory = 0;
     BufferId inline_buffer_id;
 #ifdef YUZU_LEGACY
     bool immediately_free = false;

src/video_core/deferred_destruction_queue.h

@@ -1,56 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
-#pragma once
-
-#include <cstddef>
-#include <utility>
-
-#include <boost/container/deque.hpp>
-#include <boost/container/options.hpp>
-
-#include "common/common_types.h"
-
-namespace VideoCommon {
-
-template <typename T>
-class DeferredDestructionQueue {
-public:
-    void Push(T&& object, u64 sync_point) {
-        entries.emplace_back(std::move(object), sync_point);
-    }
-
-    void Reclaim(u64 completed_sync_point) {
-        while (!entries.empty() && entries.front().sync_point <= completed_sync_point) {
-            entries.pop_front();
-        }
-    }
-
-    void Clear() {
-        entries.clear();
-    }
-
-    [[nodiscard]] size_t Size() const noexcept {
-        return entries.size();
-    }
-
-    [[nodiscard]] bool Empty() const noexcept {
-        return entries.empty();
-    }
-
-private:
-    struct Entry {
-        Entry(T&& object_, u64 sync_point_) noexcept
-            : object{std::move(object_)}, sync_point{sync_point_} {}
-
-        T object;
-        u64 sync_point;
-    };
-
-    using EntryDequeOptions =
-        boost::container::deque_options<boost::container::block_size<8u>>::type;
-
-    boost::container::deque<Entry, void, EntryDequeOptions> entries;
-};
-
-} // namespace VideoCommon

src/video_core/delayed_destruction_ring.h

@@ -0,0 +1,34 @@
+// SPDX-FileCopyrightText: Copyright 2025 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
+
+#pragma once
+
+#include <array>
+#include <cstddef>
+#include <utility>
+#include <vector>
+
+namespace VideoCommon {
+
+/// Container to push objects to be destroyed a few ticks in the future
+template <typename T, size_t TICKS_TO_DESTROY>
+class DelayedDestructionRing {
+public:
+    void Tick() {
+        index = (index + 1) % TICKS_TO_DESTROY;
+        elements[index].clear();
+    }
+
+    void Push(T&& object) {
+        elements[index].push_back(std::move(object));
+    }
+
+private:
+    size_t index = 0;
+    std::array<std::vector<T>, TICKS_TO_DESTROY> elements;
+};
+
+} // namespace VideoCommon

src/video_core/fence_manager.h

@@ -18,7 +18,7 @@
 #include "common/common_types.h"
 #include "common/settings.h"
 #include "common/thread.h"
-#include "video_core/deferred_destruction_queue.h"
+#include "video_core/delayed_destruction_ring.h"
 #include "video_core/gpu.h"
 #include "video_core/host1x/host1x.h"
 #include "video_core/host1x/syncpoint_manager.h"
@@ -50,8 +50,7 @@ public:
     /// Notify the fence manager about a new frame
     void TickFrame() {
         std::unique_lock lock(ring_guard);
-        ++retire_tick;
-        sentenced_fences.Reclaim(retire_tick > RETIRE_DELAY ? retire_tick - RETIRE_DELAY : 0);
+        delayed_destruction_ring.Tick();
     }
 
     // Unlike other fences, this one doesn't
@@ -187,7 +186,7 @@ private:
             }
             {
                 std::unique_lock lock(ring_guard);
-                sentenced_fences.Push(std::move(current_fence), retire_tick);
+                delayed_destruction_ring.Push(std::move(current_fence));
             }
             fences.pop();
         }
@@ -220,7 +219,7 @@ private:
             }
             {
                 std::unique_lock lock(ring_guard);
-                sentenced_fences.Push(std::move(current_fence), retire_tick);
+                delayed_destruction_ring.Push(std::move(current_fence));
             }
         }
     }
@@ -265,9 +264,7 @@ private:
 
     std::jthread fence_thread;
 
-    static constexpr u64 RETIRE_DELAY = 8;
-    u64 retire_tick = 1;
-    DeferredDestructionQueue<TFence> sentenced_fences;
+    DelayedDestructionRing<TFence, 8> delayed_destruction_ring;
 };
 
 } // namespace VideoCommon

src/video_core/gpu_thread.cpp

@@ -30,7 +30,6 @@ void ThreadManager::StartThread(VideoCore::RendererBase& renderer, Core::Fronten
     thread = std::jthread([&](std::stop_token stop_token) {
         Common::SetCurrentThreadName("GPU");
         Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical);
-        Common::SetCurrentThreadToPerformanceCores();
         system.RegisterHostThread();
 
         auto current_context = context.Acquire();

src/video_core/host_shaders/CMakeLists.txt

@@ -32,7 +32,6 @@ set(SHADER_FILES
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_msaa_to_non_msaa.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa.comp
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa.frag
-    ${CMAKE_CURRENT_SOURCE_DIR}/convert_non_msaa_to_msaa_depth.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/convert_s8d24_to_abgr8.frag
     ${CMAKE_CURRENT_SOURCE_DIR}/full_screen_triangle.vert
     ${CMAKE_CURRENT_SOURCE_DIR}/fxaa.frag

src/video_core/host_shaders/astc_decoder.comp

@@ -1384,7 +1384,11 @@ void DecompressBlock(ivec3 coord) {
                 p = Cf / 65535.0f;
             }
 
+#ifdef VULKAN
+            imageStore(dest_image, coord + ivec3(i, j, 0), p.gbar);
+#else
             imageStore(dest_image, coord + ivec3(i, j, 0), clamp(p, 0.0f, 1.0f).gbar);
+#endif
         }
     }
 }

src/video_core/host_shaders/convert_non_msaa_to_msaa_depth.frag

@@ -1,19 +0,0 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
-#version 450 core
-
-layout(binding = 0) uniform sampler2D img_in;
-
-layout(push_constant) uniform PushConstants {
-    ivec2 dst_offset;
-    ivec2 src_offset;
-    ivec2 scale;
-};
-
-void main() {
-    const ivec2 msaa_coord = ivec2(gl_FragCoord.xy) - dst_offset;
-    const ivec2 sample_offset = ivec2(gl_SampleID % scale.x, gl_SampleID / scale.x);
-    const ivec2 coord = msaa_coord * scale + sample_offset + src_offset;
-    gl_FragDepth = texelFetch(img_in, coord, 0).r;
-}

src/video_core/macro.cpp

@@ -417,14 +417,6 @@ void HLE_TransformFeedbackSetup::Execute(Core::System& system, Engines::Maxwell3
     default: return std::monostate{};
     }
 }
-[[nodiscard]] inline bool CanBeHLEProgram(u64 hash) noexcept {
-    switch (hash) {
-#define HLE_MACRO_ELEM(HASH, TY, VAL) case HASH: return true;
-    HLE_MACRO_LIST
-#undef HLE_MACRO_ELEM
-    default: return false;
-    }
-}
 
 void MacroInterpreterImpl::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, std::span<const u32> params, u32 method) {
     Reset();
@@ -790,13 +782,17 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
     const bool valid_operation = !is_a_zero && !is_b_zero;
     [[maybe_unused]] const bool is_move_operation = !is_a_zero && is_b_zero;
     const bool has_zero_register = is_a_zero || is_b_zero;
-    const bool no_zero_reg_skip = opcode.alu_operation == Macro::ALUOperation::AddWithCarry ||
-                                  opcode.alu_operation == Macro::ALUOperation::SubtractWithBorrow;
+    const bool zero_reg_skip =
+        optimizer.zero_reg_skip && !is_a_zero && is_b_zero &&
+        (opcode.alu_operation == Macro::ALUOperation::Xor ||
+         opcode.alu_operation == Macro::ALUOperation::Or ||
+         (optimizer.can_skip_carry && (opcode.alu_operation == Macro::ALUOperation::Add ||
+                                       opcode.alu_operation == Macro::ALUOperation::Subtract)));
 
     Xbyak::Reg32 src_a;
     Xbyak::Reg32 src_b;
 
-    if (!optimizer.zero_reg_skip || no_zero_reg_skip) {
+    if (!zero_reg_skip) {
         src_a = Compile_GetRegister(opcode.src_a, RESULT);
         src_b = Compile_GetRegister(opcode.src_b, eax);
     } else {
@@ -812,7 +808,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
 
     switch (opcode.alu_operation) {
     case Macro::ALUOperation::Add:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (valid_operation) {
                 add(src_a, src_b);
             }
@@ -830,7 +826,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         setc(byte[STATE + offsetof(JITState, carry_flag)]);
         break;
     case Macro::ALUOperation::Subtract:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (valid_operation) {
                 sub(src_a, src_b);
                 has_emitted = true;
@@ -849,7 +845,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         setc(byte[STATE + offsetof(JITState, carry_flag)]);
         break;
     case Macro::ALUOperation::Xor:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (valid_operation) {
                 xor_(src_a, src_b);
             }
@@ -858,7 +854,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         }
         break;
     case Macro::ALUOperation::Or:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (valid_operation) {
                 or_(src_a, src_b);
             }
@@ -867,7 +863,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         }
         break;
     case Macro::ALUOperation::And:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (!has_zero_register) {
                 and_(src_a, src_b);
             }
@@ -876,7 +872,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         }
         break;
     case Macro::ALUOperation::AndNot:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (!is_a_zero) {
                 not_(src_b);
                 and_(src_a, src_b);
@@ -887,7 +883,7 @@ void MacroJITx64Impl::Compile_ALU(Core::System& system, Macro::Opcode opcode) {
         }
         break;
     case Macro::ALUOperation::Nand:
-        if (optimizer.zero_reg_skip) {
+        if (zero_reg_skip) {
             if (!is_a_zero) {
                 and_(src_a, src_b);
                 not_(src_a);
@@ -1345,80 +1341,89 @@ static void Dump(u64 hash, std::span<const u32> code, bool decompiled = false) {
     macro_file.write(reinterpret_cast<const char*>(code.data()), code.size_bytes());
 }
 
-void MacroEngine::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, u32 method, std::span<const u32> parameters) {
-    auto const execute_variant = [&system, &maxwell3d, &parameters, method](AnyCachedMacro& acm) {
-        if (auto a = std::get_if<HLE_DrawArraysIndirect>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_DrawIndexedIndirect>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_MultiDrawIndexedIndirectCount>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_MultiLayerClear>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_C713C83D8F63CCF3>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_D7333D26E0A93EDE>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_BindShader>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_SetRasterBoundingBox>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_ClearConstBuffer>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_ClearMemory>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_TransformFeedbackSetup>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<HLE_DrawIndirectByteCount>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<MacroInterpreterImpl>(&acm))
-            return a->Execute(system, maxwell3d, parameters, method);
-        if (auto a = std::get_if<std::unique_ptr<DynamicCachedMacro>>(&acm))
-            return a->get()->Execute(system, maxwell3d, parameters, method);
-    };
-    if (auto const it = macro_cache.find(method); it != macro_cache.end()) {
-        auto& ci = it->second;
-        if (!CanBeHLEProgram(ci.hash) || Settings::values.disable_macro_hle)
-            maxwell3d.RefreshParameters(); //LLE must reload parameters
-        execute_variant(ci.program);
-    } else {
-        // Macro not compiled, check if it's uploaded and if so, compile it
-        std::optional<u32> mid_method;
-        const auto macro_code = uploaded_macro_code.find(method);
-        if (macro_code == uploaded_macro_code.end()) {
-            for (const auto& [method_base, code] : uploaded_macro_code) {
-                if (method >= method_base && (method - method_base) < code.size()) {
-                    mid_method = method_base;
-                    break;
-                }
-            }
-            if (!mid_method.has_value()) {
-                ASSERT_MSG(false, "Macro 0x{0:x} was not uploaded", method);
-                return;
-            }
-        }
-        auto& ci = macro_cache[method];
-        if (mid_method) {
-            const auto& macro_cached = uploaded_macro_code[mid_method.value()];
-            const auto rebased_method = method - mid_method.value();
-            auto& code = uploaded_macro_code[method];
-            code.resize(macro_cached.size() - rebased_method);
-            std::memcpy(code.data(), macro_cached.data() + rebased_method, code.size() * sizeof(u32));
-            ci.hash = Common::HashValue(code);
-            ci.program = Compile(system, maxwell3d, code);
-        } else {
-            ci.program = Compile(system, maxwell3d, macro_code->second);
-            ci.hash = Common::HashValue(macro_code->second);
-        }
-        if (CanBeHLEProgram(ci.hash) && !Settings::values.disable_macro_hle) {
-            ci.program = GetHLEProgram(ci.hash);
-        } else {
+void MacroEngine::Execute(Core::System& system, Engines::Maxwell3D& maxwell3d, u32 method,
+                          std::span<const u32> parameters) {
+    const auto execute_variant = [&system, &maxwell3d, &parameters,
+                                  method](AnyCachedMacro& cached) {
+        if (std::holds_alternative<MacroInterpreterImpl>(cached) ||
+            std::holds_alternative<std::unique_ptr<DynamicCachedMacro>>(cached) ||
+            Settings::values.disable_macro_hle) {
             maxwell3d.RefreshParameters();
         }
-        execute_variant(ci.program);
-        if (Settings::values.dump_macros) {
-            Dump(ci.hash, macro_code->second, !std::holds_alternative<std::monostate>(ci.program));
+
+        if (auto program = std::get_if<HLE_DrawArraysIndirect>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_DrawIndexedIndirect>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_MultiDrawIndexedIndirectCount>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_MultiLayerClear>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_C713C83D8F63CCF3>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_D7333D26E0A93EDE>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_BindShader>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_SetRasterBoundingBox>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_ClearConstBuffer>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_ClearMemory>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_TransformFeedbackSetup>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<HLE_DrawIndirectByteCount>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<MacroInterpreterImpl>(&cached))
+            return program->Execute(system, maxwell3d, parameters, method);
+        if (auto program = std::get_if<std::unique_ptr<DynamicCachedMacro>>(&cached))
+            return program->get()->Execute(system, maxwell3d, parameters, method);
+
+        UNREACHABLE();
+    };
+    if (auto const it = macro_cache.find(method); it != macro_cache.end()) {
+        execute_variant(it->second.program);
+        return;
+    }
+
+    // Macro not compiled, check if it's uploaded and if so, compile it
+    std::span<const u32> code;
+    auto macro_code = uploaded_macro_code.find(method);
+    if (macro_code == uploaded_macro_code.end()) {
+        for (auto it = uploaded_macro_code.begin(); it != uploaded_macro_code.end(); ++it) {
+            const auto& [method_base, uploaded_code] = *it;
+            if (method >= method_base && (method - method_base) < uploaded_code.size()) {
+                macro_code = it;
+                break;
+            }
         }
+        if (macro_code == uploaded_macro_code.end()) {
+            ASSERT_MSG(false, "Macro 0x{0:x} was not uploaded", method);
+            return;
+        }
+
+        const auto rebased_method = method - macro_code->first;
+        std::vector<u32> rebased_code(macro_code->second.begin() + rebased_method,
+                                      macro_code->second.end());
+        const auto [it, inserted] = uploaded_macro_code.emplace(method, std::move(rebased_code));
+        ASSERT(inserted);
+        macro_code = it;
+    }
+    code = macro_code->second;
+
+    auto& ci = macro_cache[method];
+    ci.hash = Common::HashRange(code.begin(), code.end());
+    if (!Settings::values.disable_macro_hle) {
+        ci.program = GetHLEProgram(ci.hash);
+    }
+    if (std::holds_alternative<std::monostate>(ci.program)) {
+        ci.program = Compile(system, maxwell3d, code);
+    }
+
+    execute_variant(ci.program);
+    if (Settings::values.dump_macros) {
+        Dump(ci.hash, code, !std::holds_alternative<std::monostate>(ci.program));
     }
 }
 

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -93,17 +93,7 @@ public:
     void PostCopyBarrier();
     void Finish();
 
-    void TickFrame(Common::SlotVector<Buffer>&) noexcept {
-        ++sync_point;
-    }
-
-    u64 CurrentSyncPoint() const noexcept {
-        return sync_point;
-    }
-
-    u64 CompletedSyncPoint() const noexcept {
-        return sync_point > SYNC_POINT_DELAY ? sync_point - SYNC_POINT_DELAY : 0;
-    }
+    void TickFrame(Common::SlotVector<Buffer>&) noexcept {}
 
     void ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value);
 
@@ -138,14 +128,6 @@ public:
 
     u64 GetDeviceMemoryUsage() const;
 
-    u64 GetDeviceAllocationUsage() const {
-        return GetDeviceMemoryUsage();
-    }
-
-    bool CanReportAllocationUsage() const {
-        return device.CanReportMemoryUsage();
-    }
-
     void BindFastUniformBuffer(size_t stage, u32 binding_index, u32 size) {
         const GLuint handle = fast_uniforms[stage][binding_index].handle;
         const GLsizeiptr gl_size = static_cast<GLsizeiptr>(size);
@@ -231,13 +213,9 @@ private:
         GL_FRAGMENT_PROGRAM_PARAMETER_BUFFER_NV,
     };
 
-    static constexpr u64 SYNC_POINT_DELAY = 8;
-
     const Device& device;
     StagingBufferPool& staging_buffer_pool;
 
-    u64 sync_point = 1;
-
     bool has_fast_buffer_sub_data = false;
     bool use_assembly_shaders = false;
     bool has_unified_vertex_buffers = false;

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -87,14 +87,6 @@ public:
 
     u64 GetDeviceMemoryUsage() const;
 
-    u64 GetDeviceAllocationUsage() const {
-        return GetDeviceMemoryUsage();
-    }
-
-    bool CanReportAllocationUsage() const {
-        return device.CanReportMemoryUsage();
-    }
-
     bool CanReportMemoryUsage() const {
         return device.CanReportMemoryUsage();
     }
@@ -147,19 +139,7 @@ public:
 
     bool HasNativeASTC() const noexcept;
 
-    void TickFrame() {
-        ++sync_point;
-    }
-
-    u64 CurrentSyncPoint() const noexcept {
-        return sync_point;
-    }
-
-    u64 CompletedSyncPoint() const noexcept {
-        return sync_point > SYNC_POINT_DELAY ? sync_point - SYNC_POINT_DELAY : 0;
-    }
-
-    void WaitSyncPoint(u64) {}
+    void TickFrame() {}
 
     StateTracker& GetStateTracker() {
         return state_tracker;
@@ -194,9 +174,6 @@ private:
     std::array<OGLFramebuffer, 4> rescale_read_fbos;
     const Settings::ResolutionScalingInfo& resolution;
     u64 device_access_memory;
-
-    static constexpr u64 SYNC_POINT_DELAY = 8;
-    u64 sync_point = 1;
 };
 
 class Image : public VideoCommon::ImageBase {
@@ -393,7 +370,6 @@ struct TextureCacheParams {
     static constexpr bool HAS_EMULATED_COPIES = true;
     static constexpr bool HAS_DEVICE_MEMORY_INFO = true;
     static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = true;
-    static constexpr bool HAS_TIMELINE_SYNC_POINTS = false;
 
     using Runtime = OpenGL::TextureCacheRuntime;
     using Image = OpenGL::Image;

src/video_core/renderer_vulkan/blit_image.cpp

@@ -21,7 +21,6 @@
 #include "video_core/host_shaders/convert_depth_to_float_frag_spv.h"
 #include "video_core/host_shaders/convert_float_to_depth_frag_spv.h"
 #include "video_core/host_shaders/convert_msaa_to_non_msaa_frag_spv.h"
-#include "video_core/host_shaders/convert_non_msaa_to_msaa_depth_frag_spv.h"
 #include "video_core/host_shaders/convert_non_msaa_to_msaa_frag_spv.h"
 #include "video_core/host_shaders/convert_s8d24_to_abgr8_frag_spv.h"
 #include "video_core/host_shaders/full_screen_triangle_vert_spv.h"
@@ -520,8 +519,7 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view)
 }
 
 [[nodiscard]] vk::ImageView MakeMSAACopyView(const vk::Device& device, VkImage image,
-                                             VkFormat format, u32 base_level,
-                                             VkImageAspectFlags aspect_mask) {
+                                             VkFormat format, u32 base_level) {
     return device.CreateImageView(VkImageViewCreateInfo{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
         .pNext = nullptr,
@@ -536,7 +534,7 @@ void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view)
             .a = VK_COMPONENT_SWIZZLE_IDENTITY,
         },
         .subresourceRange{
-            .aspectMask = aspect_mask,
+            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
             .baseMipLevel = base_level,
             .levelCount = 1,
             .baseArrayLayer = 0,
@@ -612,8 +610,6 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
       convert_s8d24_to_abgr8_frag(BuildShader(device, CONVERT_S8D24_TO_ABGR8_FRAG_SPV)),
       convert_msaa_to_non_msaa_frag(BuildShader(device, CONVERT_MSAA_TO_NON_MSAA_FRAG_SPV)),
       convert_non_msaa_to_msaa_frag(BuildShader(device, CONVERT_NON_MSAA_TO_MSAA_FRAG_SPV)),
-      convert_non_msaa_to_msaa_depth_frag(
-          BuildShader(device, CONVERT_NON_MSAA_TO_MSAA_DEPTH_FRAG_SPV)),
       linear_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_LINEAR>)),
       nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {}
 
@@ -899,34 +895,16 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i
     const s32 scale_y = 1 << samples_y;
     const VkSampleCountFlagBits samples =
         msaa_to_non_msaa ? VK_SAMPLE_COUNT_1_BIT : SampleCountFlag(num_samples);
-    const auto dst_surface_type = VideoCore::Surface::GetFormatType(dst_format);
-    const bool is_depth = dst_surface_type == VideoCore::Surface::SurfaceType::Depth ||
-                          dst_surface_type == VideoCore::Surface::SurfaceType::DepthStencil;
-    const bool has_stencil = dst_surface_type == VideoCore::Surface::SurfaceType::DepthStencil;
-    const VkImageAspectFlags view_aspect =
-        is_depth ? VK_IMAGE_ASPECT_DEPTH_BIT : VK_IMAGE_ASPECT_COLOR_BIT;
-    VkImageAspectFlags barrier_aspect = VK_IMAGE_ASPECT_COLOR_BIT;
-    if (is_depth) {
-        barrier_aspect = VK_IMAGE_ASPECT_DEPTH_BIT;
-        if (has_stencil) {
-            barrier_aspect |= VK_IMAGE_ASPECT_STENCIL_BIT;
-        }
-    }
     RenderPassKey renderpass_key{};
     renderpass_key.color_formats.fill(VideoCore::Surface::PixelFormat::Invalid);
-    if (is_depth) {
-        renderpass_key.depth_format = dst_format;
-    } else {
-        renderpass_key.color_formats[0] = dst_format;
-        renderpass_key.depth_format = VideoCore::Surface::PixelFormat::Invalid;
-    }
+    renderpass_key.color_formats[0] = dst_format;
+    renderpass_key.depth_format = VideoCore::Surface::PixelFormat::Invalid;
     renderpass_key.samples = samples;
     const VkRenderPass renderpass = render_pass_cache.Get(renderpass_key);
     const MSAACopyPipelineKey key{
         .renderpass = renderpass,
         .samples = samples,
         .msaa_to_non_msaa = msaa_to_non_msaa,
-        .is_depth = is_depth,
     };
     const VkPipeline pipeline = FindOrEmplaceMSAACopyPipeline(key);
     const VkPipelineLayout layout = *msaa_copy_pipeline_layout;
@@ -942,10 +920,10 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i
         ASSERT(copy.dst_subresource.num_layers == 1);
         vk::ImageView src_view =
             MakeMSAACopyView(device.GetLogical(), src_image, src_vk_format,
-                             static_cast<u32>(copy.src_subresource.base_level), view_aspect);
+                             static_cast<u32>(copy.src_subresource.base_level));
         vk::ImageView dst_view =
             MakeMSAACopyView(device.GetLogical(), dst_image, dst_vk_format,
-                             static_cast<u32>(copy.dst_subresource.base_level), view_aspect);
+                             static_cast<u32>(copy.dst_subresource.base_level));
         const VkOffset2D dst_offset{copy.dst_offset.x, copy.dst_offset.y};
         const VkExtent2D dst_extent{copy.extent.width, copy.extent.height};
         const VkRect2D render_area{
@@ -971,64 +949,50 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i
         scheduler.RequestOutsideRenderPassOperationContext();
         scheduler.Record([this, pipeline, layout, sampler, renderpass,
                           framebuffer_handle = *framebuffer, src_view_handle = *src_view,
-                          src = src_image, dst = dst_image, render_area, is_depth, barrier_aspect,
+                          src = src_image, dst = dst_image, render_area,
                           push_constants](vk::CommandBuffer cmdbuf) {
-            const VkImageSubresourceRange src_range{
-                .aspectMask = barrier_aspect,
+            constexpr VkImageSubresourceRange color_range{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
                 .baseMipLevel = 0,
                 .levelCount = VK_REMAINING_MIP_LEVELS,
                 .baseArrayLayer = 0,
                 .layerCount = VK_REMAINING_ARRAY_LAYERS,
             };
-            const VkImageSubresourceRange dst_range = src_range;
-            const VkAccessFlags attachment_read =
-                is_depth ? VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT
-                         : VK_ACCESS_COLOR_ATTACHMENT_READ_BIT;
-            const VkAccessFlags attachment_write =
-                is_depth ? VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT
-                         : VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
-            const VkPipelineStageFlags depth_stage =
-                VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
-            const VkPipelineStageFlags attachment_stage =
-                is_depth ? depth_stage
-                         : static_cast<VkPipelineStageFlags>(
-                               VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT);
             const std::array pre_barriers{
                 VkImageMemoryBarrier{
                     .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                     .pNext = nullptr,
-                    .srcAccessMask = attachment_write | VK_ACCESS_SHADER_WRITE_BIT |
-                                     VK_ACCESS_TRANSFER_WRITE_BIT,
+                    .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+                                     VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
                     .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
                     .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                     .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                     .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                     .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                     .image = src,
-                    .subresourceRange = src_range,
+                    .subresourceRange = color_range,
                 },
                 VkImageMemoryBarrier{
                     .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                     .pNext = nullptr,
-                    .srcAccessMask = attachment_write | VK_ACCESS_SHADER_WRITE_BIT |
-                                     VK_ACCESS_TRANSFER_WRITE_BIT,
-                    .dstAccessMask = attachment_read | attachment_write,
+                    .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+                                     VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
+                    .dstAccessMask = VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
+                                     VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
                     .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                     .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                     .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                     .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                     .image = dst,
-                    .subresourceRange = dst_range,
+                    .subresourceRange = color_range,
                 },
             };
             cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
-                                       VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
-                                       VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT |
                                        VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
                                        VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
                                        VK_PIPELINE_STAGE_TRANSFER_BIT,
-                                   VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | attachment_stage,
+                                   VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
+                                       VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
                                    0, nullptr, nullptr, pre_barriers);
             const VkRenderPassBeginInfo renderpass_bi{
                 .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
@@ -1061,16 +1025,16 @@ void BlitImageHelper::CopyMSAA(RenderPassCache& render_pass_cache, VkImage dst_i
             const VkImageMemoryBarrier post_barrier{
                 .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
                 .pNext = nullptr,
-                .srcAccessMask = attachment_write,
+                .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT,
                 .dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT,
                 .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
                 .newLayout = VK_IMAGE_LAYOUT_GENERAL,
                 .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                 .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
                 .image = dst,
-                .subresourceRange = dst_range,
+                .subresourceRange = color_range,
             };
-            cmdbuf.PipelineBarrier(attachment_stage,
+            cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
                                    VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
                                        VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
                                        VK_PIPELINE_STAGE_TRANSFER_BIT,
@@ -1459,36 +1423,9 @@ VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipeline
         return *msaa_copy_pipelines[std::distance(msaa_copy_keys.begin(), it)];
     }
     msaa_copy_keys.push_back(key);
-    const VkShaderModule frag_module =
-        key.msaa_to_non_msaa
-            ? *convert_msaa_to_non_msaa_frag
-            : (key.is_depth ? *convert_non_msaa_to_msaa_depth_frag
-                            : *convert_non_msaa_to_msaa_frag);
-    const std::array stages = MakeStages(*clear_color_vert, frag_module);
-    const VkPipelineDepthStencilStateCreateInfo depth_stencil_ci{
-        .sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .depthTestEnable = VK_TRUE,
-        .depthWriteEnable = VK_TRUE,
-        .depthCompareOp = VK_COMPARE_OP_ALWAYS,
-        .depthBoundsTestEnable = VK_FALSE,
-        .stencilTestEnable = VK_FALSE,
-        .front = {},
-        .back = {},
-        .minDepthBounds = 0.0f,
-        .maxDepthBounds = 0.0f,
-    };
-    static constexpr VkPipelineColorBlendStateCreateInfo no_color_blend_ci{
-        .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .logicOpEnable = VK_FALSE,
-        .logicOp = VK_LOGIC_OP_CLEAR,
-        .attachmentCount = 0,
-        .pAttachments = nullptr,
-        .blendConstants = {0.0f, 0.0f, 0.0f, 0.0f},
-    };
+    const std::array stages = MakeStages(*clear_color_vert, key.msaa_to_non_msaa
+                                                                ? *convert_msaa_to_non_msaa_frag
+                                                                : *convert_non_msaa_to_msaa_frag);
     const VkPipelineMultisampleStateCreateInfo multisample_ci{
         .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
         .pNext = nullptr,
@@ -1513,9 +1450,8 @@ VkPipeline BlitImageHelper::FindOrEmplaceMSAACopyPipeline(const MSAACopyPipeline
         .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
         .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
         .pMultisampleState = &multisample_ci,
-        .pDepthStencilState = key.is_depth ? &depth_stencil_ci : nullptr,
-        .pColorBlendState = key.is_depth ? &no_color_blend_ci
-                                         : &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
+        .pDepthStencilState = nullptr,
+        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_GENERIC_CREATE_INFO,
         .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
         .layout = *msaa_copy_pipeline_layout,
         .renderPass = key.renderpass,

src/video_core/renderer_vulkan/blit_image.h

@@ -51,7 +51,6 @@ struct MSAACopyPipelineKey {
     VkRenderPass renderpass;
     VkSampleCountFlagBits samples;
     bool msaa_to_non_msaa;
-    bool is_depth;
 };
 
 struct BlitMSAAPipelineKey {
@@ -181,7 +180,6 @@ private:
     vk::ShaderModule convert_s8d24_to_abgr8_frag;
     vk::ShaderModule convert_msaa_to_non_msaa_frag;
     vk::ShaderModule convert_non_msaa_to_msaa_frag;
-    vk::ShaderModule convert_non_msaa_to_msaa_depth_frag;
     vk::Sampler linear_sampler;
     vk::Sampler nearest_sampler;
 

src/video_core/renderer_vulkan/fixed_pipeline_state.cpp

@@ -164,9 +164,7 @@ void FixedPipelineState::Refresh(Tegra::Engines::Maxwell3D& maxwell3d, DynamicFe
     }
 
     provoking_vertex_last.Assign(use_last_provoking_vertex ? 1 : 0);
-    if (!features.has_dynamic_state3_conservative_raster_mode) {
-        conservative_raster_enable.Assign(regs.conservative_raster_enable != 0 ? 1 : 0);
-    }
+    conservative_raster_enable.Assign(regs.conservative_raster_enable != 0 ? 1 : 0);
     smooth_lines.Assign(regs.line_anti_alias_enable != 0 ? 1 : 0);
     alpha_to_coverage_enabled.Assign(regs.anti_alias_alpha_control.alpha_to_coverage != 0 ? 1 : 0);
     alpha_to_one_enabled.Assign(regs.anti_alias_alpha_control.alpha_to_one != 0 ? 1 : 0);
@@ -362,35 +360,18 @@ void FixedPipelineState::DynamicState::Refresh2(const Maxwell& regs,
     depth_bias_enable.Assign(enabled_lut[POLYGON_OFFSET_ENABLE_LUT[topology_index]] != 0 ? 1 : 0);
 }
 
-bool IsDepthClipEnabled(const Maxwell& regs) {
-    const auto clip = regs.viewport_clip_control.geometry_clip.Value();
-    return clip == Maxwell::ViewportClipControl::GeometryClip::Passthrough ||
-           clip == Maxwell::ViewportClipControl::GeometryClip::FrustumXYZ ||
-           clip == Maxwell::ViewportClipControl::GeometryClip::FrustumZ;
-}
-
-bool IsDepthClampEnabled(const Maxwell& regs, bool has_depth_clip_enable) {
-    if (!IsDepthClipEnabled(regs)) {
-        return true;
-    }
-    if (!has_depth_clip_enable) {
-        return false;
-    }
-    return regs.viewport_clip_control.pixel_min_z.Value() != 0 ||
-           regs.viewport_clip_control.pixel_max_z.Value() != 0;
-}
-
 void FixedPipelineState::DynamicState::Refresh3(const Maxwell& regs,
                                                 const DynamicFeatures& features) {
     if (!features.has_dynamic_state3_logic_op_enable) {
         logic_op_enable.Assign(regs.logic_op.enable != 0 ? 1 : 0);
     }
-    if (features.has_depth_clip_enable) {
-        depth_clip_disabled.Assign(IsDepthClipEnabled(regs) ? 0 : 1);
-    }
     if (!features.has_dynamic_state3_depth_clamp_enable) {
-        depth_clamp_disabled.Assign(
-            IsDepthClampEnabled(regs, features.has_depth_clip_enable) ? 0 : 1);
+        depth_clamp_disabled.Assign(regs.viewport_clip_control.geometry_clip ==
+                                        Maxwell::ViewportClipControl::GeometryClip::Passthrough ||
+                                    regs.viewport_clip_control.geometry_clip ==
+                                        Maxwell::ViewportClipControl::GeometryClip::FrustumXYZ ||
+                                    regs.viewport_clip_control.geometry_clip ==
+                                        Maxwell::ViewportClipControl::GeometryClip::FrustumZ);
     }
     if (!features.has_dynamic_state3_line_stipple_enable) {
         line_stipple_enable.Assign(regs.line_stipple_enable);

src/video_core/renderer_vulkan/fixed_pipeline_state.h

@@ -30,8 +30,6 @@ struct DynamicFeatures {
     bool has_extended_dynamic_state_3_blend;
     bool has_extended_dynamic_state_3_enables;
     bool has_dynamic_state3_depth_clamp_enable;
-    bool has_dynamic_state3_conservative_raster_mode;
-    bool has_depth_clip_enable;
     bool has_dynamic_state3_logic_op_enable;
     bool has_dynamic_state3_line_stipple_enable;
     bool has_dynamic_vertex_input;
@@ -167,7 +165,6 @@ struct FixedPipelineState {
             BitField<10, 1, u32> logic_op_enable;
             BitField<11, 1, u32> depth_clamp_disabled;
             BitField<12, 1, u32> line_stipple_enable;
-            BitField<13, 1, u32> depth_clip_disabled;
         };
         union {
             u32 raw2;
@@ -301,9 +298,6 @@ static_assert(std::has_unique_object_representations_v<FixedPipelineState>);
 static_assert(std::is_trivially_copyable_v<FixedPipelineState>);
 static_assert(std::is_trivially_constructible_v<FixedPipelineState>);
 
-bool IsDepthClipEnabled(const Maxwell& regs);
-bool IsDepthClampEnabled(const Maxwell& regs, bool has_depth_clip_enable);
-
 } // namespace Vulkan
 
 namespace std {

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -246,6 +246,7 @@ protected:
     StagingBufferPool& staging_pool;
 
     vk::Buffer buffer{};
+    MemoryCommit memory_commit{};
     VkIndexType index_type{};
     u32 num_indices = 0;
 };
@@ -375,10 +376,6 @@ u64 BufferCacheRuntime::GetDeviceMemoryUsage() const {
     return device.GetDeviceMemoryUsage();
 }
 
-u64 BufferCacheRuntime::GetDeviceAllocationUsage() const {
-    return device.GetMemoryBudgetInfo().allocation_bytes;
-}
-
 bool BufferCacheRuntime::CanReportMemoryUsage() const {
     return device.CanReportMemoryUsage();
 }
@@ -407,16 +404,6 @@ u64 BufferCacheRuntime::KnownGpuTick() {
     return scheduler.GetMasterSemaphore().KnownGpuTick();
 }
 
-u64 BufferCacheRuntime::CurrentSyncPoint() const noexcept {
-    return scheduler.GetMasterSemaphore().CurrentTick();
-}
-
-u64 BufferCacheRuntime::CompletedSyncPoint() const {
-    auto& master_semaphore = scheduler.GetMasterSemaphore();
-    master_semaphore.Refresh();
-    return master_semaphore.KnownGpuTick();
-}
-
 void BufferCacheRuntime::Wait(u64 buffer_tick) {
     scheduler.Wait(buffer_tick);
 }
@@ -654,7 +641,6 @@ void BufferCacheRuntime::BindTransformFeedbackBuffer(u32 index, VkBuffer buffer,
         offset = 0;
         size = 0;
     }
-    scheduler.MarkTransformFeedbackUsed();
     scheduler.Record([index, buffer, offset, size](vk::CommandBuffer cmdbuf) {
         const VkDeviceSize vk_offset = offset;
         const VkDeviceSize vk_size = size;
@@ -667,26 +653,19 @@ void BufferCacheRuntime::BindTransformFeedbackBuffers(VideoCommon::HostBindings<
         // Already logged in the rasterizer
         return;
     }
-    const u32 count = std::min<u32>(static_cast<u32>(bindings.buffers.size()),
-                                    VideoCommon::NUM_TRANSFORM_FEEDBACK_BUFFERS);
-    std::array<VkBuffer, VideoCommon::NUM_TRANSFORM_FEEDBACK_BUFFERS> handles{};
-    std::array<VkDeviceSize, VideoCommon::NUM_TRANSFORM_FEEDBACK_BUFFERS> offsets{};
-    std::array<VkDeviceSize, VideoCommon::NUM_TRANSFORM_FEEDBACK_BUFFERS> sizes{};
-    for (u32 i = 0; i < count; ++i) {
+    boost::container::static_vector<VkBuffer, VideoCommon::NUM_VERTEX_BUFFERS> buffer_handles(bindings.buffers.size());
+    for (u32 i = 0; i < bindings.buffers.size(); ++i) {
         auto handle = bindings.buffers[i]->Handle();
         if (handle == VK_NULL_HANDLE) {
             ReserveNullBuffer();
             handle = *null_buffer;
-        } else {
-            offsets[i] = bindings.offsets[i];
-            sizes[i] = bindings.sizes[i];
+            bindings.offsets[i] = 0;
+            bindings.sizes[i] = 0;
         }
-        handles[i] = handle;
+        buffer_handles[i] = handle;
     }
-    scheduler.MarkTransformFeedbackUsed();
-    scheduler.Record([count, handles, offsets, sizes](vk::CommandBuffer cmdbuf) {
-        cmdbuf.BindTransformFeedbackBuffersEXT(0, count, handles.data(), offsets.data(),
-                                               sizes.data());
+    scheduler.Record([bindings_ = std::move(bindings), buffer_handles_ = std::move(buffer_handles)](vk::CommandBuffer cmdbuf) {
+        cmdbuf.BindTransformFeedbackBuffersEXT(0, u32(buffer_handles_.size()), buffer_handles_.data(), bindings_.offsets.data(), bindings_.sizes.data());
     });
 }
 

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -100,20 +100,10 @@ public:
 
     void Finish();
 
-    u64 CurrentSyncPoint() const noexcept;
-
-    u64 CompletedSyncPoint() const;
-
     u64 GetDeviceLocalMemory() const;
 
     u64 GetDeviceMemoryUsage() const;
 
-    u64 GetDeviceAllocationUsage() const;
-
-    bool CanReportAllocationUsage() const noexcept {
-        return true;
-    }
-
     bool CanReportMemoryUsage() const;
 
     u32 GetUniformBufferAlignment() const;

src/video_core/renderer_vulkan/vk_fence_manager.cpp

@@ -37,7 +37,7 @@ void InnerFence::Wait() {
     if (is_stubbed) {
         return;
     }
-    scheduler.WaitSubmitted(wait_tick);
+    scheduler.Wait(wait_tick);
 }
 
 FenceManager::FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -757,13 +757,16 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
         .lineWidth = 1.0f,
         // TODO(alekpop): Transfer from regs
     };
-    const VkLineRasterizationModeEXT line_raster_mode =
-        device.GetLineRasterizationMode(key.state.smooth_lines != 0);
-    const bool stippled_lines_supported = device.SupportsStippleForMode(line_raster_mode);
+    const bool smooth_lines_supported =
+        device.IsExtLineRasterizationSupported() && device.SupportsSmoothLines();
+    const bool stippled_lines_supported =
+        device.IsExtLineRasterizationSupported() && device.SupportsStippledRectangularLines();
     VkPipelineRasterizationLineStateCreateInfoEXT line_state{
         .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_LINE_STATE_CREATE_INFO_EXT,
         .pNext = nullptr,
-        .lineRasterizationMode = line_raster_mode,
+        .lineRasterizationMode = key.state.smooth_lines != 0 && smooth_lines_supported
+                                     ? VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT
+                                     : VK_LINE_RASTERIZATION_MODE_RECTANGULAR_EXT,
         .stippledLineEnable =
             (dynamic.line_stipple_enable && stippled_lines_supported) ? VK_TRUE : VK_FALSE,
         .lineStippleFactor = key.state.line_stipple_factor,
@@ -802,16 +805,6 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
     if (device.IsExtProvokingVertexSupported()) {
         provoking_vertex.pNext = std::exchange(rasterization_ci.pNext, &provoking_vertex);
     }
-    VkPipelineRasterizationDepthClipStateCreateInfoEXT depth_clip_state{
-        .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_DEPTH_CLIP_STATE_CREATE_INFO_EXT,
-        .pNext = nullptr,
-        .flags = 0,
-        .depthClipEnable = static_cast<VkBool32>(dynamic.depth_clip_disabled == 0 ? VK_TRUE
-                                                                                  : VK_FALSE),
-    };
-    if (device.IsExtDepthClipEnableSupported()) {
-        depth_clip_state.pNext = std::exchange(rasterization_ci.pNext, &depth_clip_state);
-    }
 
     const bool supports_alpha_output = fragment_has_color0_output;
     const bool alpha_to_one_supported = device.SupportsAlphaToOne();

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -305,7 +305,7 @@ size_t GetTotalPipelineWorkers() {
         std::max<size_t>(static_cast<size_t>(std::thread::hardware_concurrency()), 2ULL) - 1ULL;
 #ifdef __ANDROID__
     const int configured = AndroidSettings::values.pipeline_worker_count.GetValue();
-    const int clamped = std::clamp(configured, 2, 8);
+    const int clamped = std::clamp(configured, 4, 8);
     const size_t desired = static_cast<size_t>(clamped);
     if (desired == 0) {
         return 1ULL;
@@ -349,9 +349,8 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
       use_asynchronous_shaders{Settings::values.use_asynchronous_shaders.GetValue()},
       use_vulkan_pipeline_cache{Settings::values.use_vulkan_driver_pipeline_cache.GetValue()},
       workers(device.HasBrokenParallelShaderCompiling() ? 1ULL : GetTotalPipelineWorkers(),
-              "VkPipelineBuilder", {}, Common::ThreadPlacement::Background),
-      serialization_thread(1, "VkPipelineSerialization", {},
-                           Common::ThreadPlacement::Background) {
+              "VkPipelineBuilder"),
+      serialization_thread(1, "VkPipelineSerialization") {
     const auto& float_control{device.FloatControlProperties()};
     const VkDriverId driver_id{device.GetDriverID()};
     const VkShaderStageFlags subgroup_stages{device.GetSubgroupSupportedStages()};
@@ -515,11 +514,6 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
     dynamic_features.has_dynamic_state3_depth_clamp_enable =
         dynamic_features.has_extended_dynamic_state_3_enables &&
         device.SupportsDynamicState3DepthClampEnable();
-    dynamic_features.has_dynamic_state3_conservative_raster_mode =
-        dynamic_features.has_extended_dynamic_state_3_enables &&
-        device.SupportsDynamicState3ConservativeRasterizationMode();
-    dynamic_features.has_depth_clip_enable = 
-        device.IsExtDepthClipEnableSupported();
     dynamic_features.has_dynamic_state3_logic_op_enable =
         dynamic_features.has_extended_dynamic_state_3_enables &&
         device.SupportsDynamicState3LogicOpEnable();
@@ -532,8 +526,7 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
         device.IsExtVertexInputDynamicStateSupported() &&
         Settings::values.vertex_input_dynamic_state.GetValue();
 
-    dynamic_features.has_provoking_vertex =
-        device.IsExtProvokingVertexSupported();
+    dynamic_features.has_provoking_vertex = device.IsExtProvokingVertexSupported();
     dynamic_features.has_provoking_vertex_first_mode =
         device.SupportsProvokingVertexFirstMode();
     dynamic_features.has_provoking_vertex_last_mode =

src/video_core/renderer_vulkan/vk_present_manager.cpp

@@ -296,6 +296,9 @@ void PresentManager::RecreateSwapchain(Frame* frame) {
 }
 
 void PresentManager::SetImageCount() {
+    // We cannot have more than 7 images in flight at any given time.
+    // FRAMES_IN_FLIGHT is 8, and the cache TICKS_TO_DESTROY is 8.
+    // Mali drivers will give us 6.
     image_count = std::min<size_t>(swapchain.GetImageCount(), 7);
 }
 

src/video_core/renderer_vulkan/vk_query_cache.cpp

@@ -919,7 +919,7 @@ private:
             return;
         }
         has_flushed_end_pending = true;
-        scheduler.MarkTransformFeedbackUsed();
+        // Refresh buffers state before beginning transform feedback so counters are up-to-date
         UpdateBuffers();
         if (!has_started || buffers_count == 0) {
             // No counter buffers available: begin without counters

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -6,7 +6,6 @@
 
 #include <algorithm>
 #include <array>
-#include <limits>
 #include <memory>
 #include <mutex>
 
@@ -204,8 +203,7 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
     : gpu{gpu_}, device_memory{device_memory_}, device{device_},
       memory_allocator{memory_allocator_}, state_tracker{state_tracker_}, scheduler{scheduler_},
       staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler),
-      guest_descriptor_queue(device, UpdateDescriptorQueue::GUEST_FRAME_PAYLOAD_SIZE),
-      compute_pass_descriptor_queue(device, UpdateDescriptorQueue::COMPUTE_FRAME_PAYLOAD_SIZE),
+      guest_descriptor_queue(device), compute_pass_descriptor_queue(device),
       blit_image(device, scheduler, state_tracker, descriptor_pool), render_pass_cache(device),
       texture_cache_runtime{
           device,     scheduler,         memory_allocator, staging_pool,
@@ -223,23 +221,9 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
       fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
       wfi_event(device.GetLogical().CreateEvent()) {
     scheduler.SetQueryCache(query_cache);
-    memory_allocator.SetReclaimCallback([this](u64 bytes) -> u64 {
-        u64 freed = staging_pool.ReclaimMemory(bytes);
-        if (freed < bytes) {
-            freed += texture_cache.ReclaimMemory(bytes - freed, false);
-        }
-        if (freed < bytes) {
-            freed += buffer_cache.ReclaimMemory(bytes - freed, false);
-        }
-        auto& master_semaphore = scheduler.GetMasterSemaphore();
-        master_semaphore.Refresh();
-        vk::TickDeletionQueue(master_semaphore.KnownGpuTick());
-        return freed;
-    });
 }
 
 RasterizerVulkan::~RasterizerVulkan() {
-    memory_allocator.SetReclaimCallback(nullptr);
     scheduler.WaitWorker();
     scheduler.Finish();
 }
@@ -896,9 +880,6 @@ void RasterizerVulkan::FlushCommands() {
 
 void RasterizerVulkan::TickFrame() {
     draw_counter = 0;
-    auto& master_semaphore = scheduler.GetMasterSemaphore();
-    master_semaphore.Refresh();
-    vk::TickDeletionQueue(master_semaphore.KnownGpuTick());
     guest_descriptor_queue.TickFrame();
     compute_pass_descriptor_queue.TickFrame();
     fence_manager.TickFrame();
@@ -1470,10 +1451,7 @@ void RasterizerVulkan::UpdateLineWidth(Tegra::Engines::Maxwell3D::Regs& regs) {
     }
     const float width =
         regs.line_anti_alias_enable ? regs.line_width_smooth : regs.line_width_aliased;
-    const float clamped_width = device.ClampLineWidth(width);
-    scheduler.Record([clamped_width](vk::CommandBuffer cmdbuf) {
-        cmdbuf.SetLineWidth(clamped_width);
-    });
+    scheduler.Record([width](vk::CommandBuffer cmdbuf) { cmdbuf.SetLineWidth(width); });
 }
 
 void RasterizerVulkan::UpdateCullMode(Tegra::Engines::Maxwell3D::Regs& regs) {
@@ -1570,10 +1548,7 @@ void RasterizerVulkan::UpdateLineStippleEnable(Tegra::Engines::Maxwell3D::Regs&
         return;
     }
 
-    const VkLineRasterizationModeEXT mode =
-        device.GetLineRasterizationMode(regs.line_anti_alias_enable != 0);
-    const bool enable = regs.line_stipple_enable != 0 && device.SupportsStippleForMode(mode);
-    scheduler.Record([enable](vk::CommandBuffer cmdbuf) {
+    scheduler.Record([enable = regs.line_stipple_enable](vk::CommandBuffer cmdbuf) {
         cmdbuf.SetLineStippleEnableEXT(enable);
     });
 }
@@ -1587,24 +1562,28 @@ void RasterizerVulkan::UpdateLineRasterizationMode(Tegra::Engines::Maxwell3D::Re
     }
 
     if (!device.SupportsDynamicState3LineRasterizationMode()) {
-        static std::once_flag warn_missing_dynamic_state;
-        std::call_once(warn_missing_dynamic_state, [] {
+        static std::once_flag warn_missing_rect;
+        std::call_once(warn_missing_rect, [] {
             LOG_WARNING(Render_Vulkan,
-                        "Driver lacks dynamic line rasterization mode; the pipeline static value "
-                        "is used instead");
+                        "Driver lacks rectangular line rasterization support; skipping dynamic "
+                        "line state updates");
         });
         return;
     }
 
     const bool wants_smooth = regs.line_anti_alias_enable != 0;
-    const VkLineRasterizationModeEXT mode = device.GetLineRasterizationMode(wants_smooth);
-    if (wants_smooth && mode != VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT) {
-        static std::once_flag warn_missing_smooth;
-        std::call_once(warn_missing_smooth, [] {
-            LOG_WARNING(Render_Vulkan,
-                        "Line anti-aliasing requested but smoothLines feature unavailable; "
-                        "falling back to the closest supported mode");
-        });
+    VkLineRasterizationModeEXT mode = VK_LINE_RASTERIZATION_MODE_RECTANGULAR_EXT;
+    if (wants_smooth) {
+        if (device.SupportsSmoothLines()) {
+            mode = VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT;
+        } else {
+            static std::once_flag warn_missing_smooth;
+            std::call_once(warn_missing_smooth, [] {
+                LOG_WARNING(Render_Vulkan,
+                            "Line anti-aliasing requested but smoothLines feature unavailable; "
+                            "using rectangular rasterization");
+            });
+        }
     }
     scheduler.Record([mode](vk::CommandBuffer cmdbuf) {
         cmdbuf.SetLineRasterizationModeEXT(mode);
@@ -1664,7 +1643,12 @@ void RasterizerVulkan::UpdateDepthClampEnable(Tegra::Engines::Maxwell3D::Regs& r
     if (!device.SupportsDynamicState3DepthClampEnable()) {
         return;
     }
-    const bool is_enabled = IsDepthClampEnabled(regs, device.IsExtDepthClipEnableSupported());
+    bool is_enabled = !(regs.viewport_clip_control.geometry_clip ==
+                            Maxwell::ViewportClipControl::GeometryClip::Passthrough ||
+                        regs.viewport_clip_control.geometry_clip ==
+                            Maxwell::ViewportClipControl::GeometryClip::FrustumXYZ ||
+                        regs.viewport_clip_control.geometry_clip ==
+                            Maxwell::ViewportClipControl::GeometryClip::FrustumZ);
     scheduler.Record(
         [is_enabled](vk::CommandBuffer cmdbuf) { cmdbuf.SetDepthClampEnableEXT(is_enabled); });
 }

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -179,6 +179,7 @@ private:
     void UpdateRasterizerDiscardEnable(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateConservativeRasterizationMode(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateLineStippleEnable(Tegra::Engines::Maxwell3D::Regs& regs);
+    void UpdateLineStipple(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateLineRasterizationMode(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateDepthBiasEnable(Tegra::Engines::Maxwell3D::Regs& regs);
     void UpdateLogicOpEnable(Tegra::Engines::Maxwell3D::Regs& regs);

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -47,7 +47,6 @@ Scheduler::Scheduler(const Device& device_, StateTracker& state_tracker_)
       master_semaphore{std::make_unique<MasterSemaphore>(device)},
       command_pool{std::make_unique<CommandPool>(*master_semaphore, device)} {
 
-    vk::SetDeletionTimeline(master_semaphore->CurrentTick());
     AcquireNewChunk();
     AllocateWorkerCommandBuffer();
     worker_thread = std::jthread([this](std::stop_token token) { WorkerThread(token); });
@@ -323,7 +322,6 @@ u64 Scheduler::SubmitExecution(VkSemaphore signal_semaphore, VkSemaphore wait_se
     InvalidateState();
 
     const u64 signal_value = master_semaphore->NextTick();
-    vk::SetDeletionTimeline(master_semaphore->CurrentTick());
     RecordWithUploadBuffer([signal_semaphore, wait_semaphore, signal_value,
                             this](vk::CommandBuffer cmdbuf, vk::CommandBuffer upload_cmdbuf) {
         static constexpr VkMemoryBarrier WRITE_BARRIER{
@@ -400,7 +398,7 @@ void Scheduler::EndRenderPass()
         Record([num_images = num_renderpass_images,
                        images = renderpass_images,
                        ranges = renderpass_image_ranges,
-                       has_transform_feedback = state.uses_transform_feedback](
+                       has_transform_feedback = device.IsExtTransformFeedbackSupported()](
                           vk::CommandBuffer cmdbuf) {
             std::array<VkImageMemoryBarrier, 9> barriers;
             for (size_t i = 0; i < num_images; ++i) {
@@ -455,7 +453,6 @@ void Scheduler::EndRenderPass()
         });
 
         state.renderpass = VkRenderPass{};
-        state.uses_transform_feedback = false;
         num_renderpass_images = 0;
     }
 

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -74,11 +74,6 @@ public:
         return state.renderpass != VK_NULL_HANDLE;
     }
 
-    /// Flags that transform feedback writes have been recorded since the last render pass end.
-    void MarkTransformFeedbackUsed() noexcept {
-        state.uses_transform_feedback = true;
-    }
-
     /// Update the pipeline to the current execution context.
     bool UpdateGraphicsPipeline(GraphicsPipeline* pipeline);
 
@@ -136,16 +131,33 @@ public:
             }
             master_semaphore->Wait(tick);
         }
-        ApplyFramePacing(target_fps);
-    }
-
-    void WaitSubmitted(u64 tick, double target_fps = 0.0) {
-        if (tick > 0 && tick < master_semaphore->CurrentTick()) {
-            master_semaphore->Wait(tick);
+        if (Settings::values.use_speed_limit.GetValue() && target_fps > 0.0) {
+            auto now = std::chrono::steady_clock::now();
+            if (last_target_fps != target_fps) {
+                frame_interval = std::chrono::duration_cast<std::chrono::steady_clock::duration>(std::chrono::duration<double>(1.0 / target_fps));
+                max_frame_count = static_cast<int>(0.1 * target_fps);
+                last_target_fps = target_fps;
+                frame_counter = 0;
+                start_time = now;
+            }
+            frame_counter++;
+            auto target_time = start_time + frame_interval * frame_counter;
+            if (target_time >= now) {
+                auto sleep_time = target_time - now;
+                if (sleep_time > std::chrono::milliseconds(15)) {
+                    std::this_thread::sleep_for(sleep_time - std::chrono::milliseconds(1));
+                }
+                while (std::chrono::steady_clock::now() < target_time) {
+                    std::this_thread::yield();
+                }
+            } else if (frame_counter > max_frame_count) {
+                frame_counter = 0;
+                start_time = now;
+            }
         }
-        ApplyFramePacing(target_fps);
     }
 
+    /// Returns the master timeline semaphore.
     [[nodiscard]] MasterSemaphore& GetMasterSemaphore() const noexcept {
         return *master_semaphore;
     }
@@ -153,35 +165,6 @@ public:
     std::mutex submit_mutex;
 
 private:
-    void ApplyFramePacing(double target_fps) {
-        if (!Settings::values.use_speed_limit.GetValue() || target_fps <= 0.0) {
-            return;
-        }
-        auto now = std::chrono::steady_clock::now();
-        if (last_target_fps != target_fps) {
-            frame_interval = std::chrono::duration_cast<std::chrono::steady_clock::duration>(
-                std::chrono::duration<double>(1.0 / target_fps));
-            max_frame_count = static_cast<int>(0.1 * target_fps);
-            last_target_fps = target_fps;
-            frame_counter = 0;
-            start_time = now;
-        }
-        frame_counter++;
-        auto target_time = start_time + frame_interval * frame_counter;
-        if (target_time >= now) {
-            auto sleep_time = target_time - now;
-            if (sleep_time > std::chrono::milliseconds(15)) {
-                std::this_thread::sleep_for(sleep_time - std::chrono::milliseconds(1));
-            }
-            while (std::chrono::steady_clock::now() < target_time) {
-                std::this_thread::yield();
-            }
-        } else if (frame_counter > max_frame_count) {
-            frame_counter = 0;
-            start_time = now;
-        }
-    }
-
     class Command {
     public:
         virtual ~Command() = default;
@@ -272,7 +255,6 @@ private:
         bool is_rescaling = false;
         bool rescaling_defined = false;
         bool needs_state_enable_refresh = false;
-        bool uses_transform_feedback = false;
     };
 
     struct DeferredClear {

src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp

@@ -252,62 +252,25 @@ void StagingBufferPool::ReleaseLevel(StagingBuffersCache& cache, size_t log2) {
     constexpr size_t deletions_per_tick = 16;
     auto& staging = cache[log2];
     auto& entries = staging.entries;
-    if (entries.empty()) {
-        staging.delete_index = 0;
-        staging.iterate_index = 0;
-        return;
-    }
+    const size_t old_size = entries.size();
 
     const auto is_deletable = [this](const StagingBuffer& entry) {
         return scheduler.IsFree(entry.tick);
     };
-    const size_t begin_offset = (std::min)(staging.delete_index, entries.size());
-    const size_t end_offset = (std::min)(begin_offset + deletions_per_tick, entries.size());
+    const size_t begin_offset = staging.delete_index;
+    const size_t end_offset = (std::min)(begin_offset + deletions_per_tick, old_size);
     const auto begin = entries.begin() + begin_offset;
     const auto end = entries.begin() + end_offset;
-    const auto surviving_end = std::remove_if(begin, end, is_deletable);
-    const size_t removed = static_cast<size_t>(std::distance(surviving_end, end));
-    entries.erase(surviving_end, end);
+    entries.erase(std::remove_if(begin, end, is_deletable), end);
 
-    staging.delete_index = end_offset - removed;
-    if (staging.delete_index >= entries.size()) {
+    const size_t new_size = entries.size();
+    staging.delete_index += deletions_per_tick;
+    if (staging.delete_index >= new_size) {
         staging.delete_index = 0;
     }
-    if (staging.iterate_index > entries.size()) {
+    if (staging.iterate_index > new_size) {
         staging.iterate_index = 0;
     }
 }
 
-u64 StagingBufferPool::ReclaimMemory(u64 target_bytes) {
-    u64 freed = 0;
-    const auto is_deletable = [this](const StagingBuffer& entry) {
-        return scheduler.IsFree(entry.tick);
-    };
-    const auto reclaim_cache = [&](StagingBuffersCache& cache) {
-        for (size_t level = NUM_LEVELS; level-- > 0 && freed < target_bytes;) {
-            auto& staging = cache[level];
-            auto& entries = staging.entries;
-            if (entries.empty()) {
-                continue;
-            }
-            const u64 entry_bytes = 1ULL << level;
-            auto it = entries.begin();
-            while (it != entries.end() && freed < target_bytes) {
-                if (is_deletable(*it)) {
-                    it = entries.erase(it);
-                    freed += entry_bytes;
-                } else {
-                    ++it;
-                }
-            }
-            staging.delete_index = 0;
-            staging.iterate_index = 0;
-        }
-    };
-    reclaim_cache(device_local_cache);
-    reclaim_cache(upload_cache);
-    reclaim_cache(download_cache);
-    return freed;
-}
-
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_staging_buffer_pool.h

@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 // SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
@@ -45,8 +42,6 @@ public:
 
     void TickFrame();
 
-    u64 ReclaimMemory(u64 target_bytes);
-
 private:
     struct StreamBufferCommit {
         size_t upper_bound;

src/video_core/renderer_vulkan/vk_swapchain.cpp

@@ -9,6 +9,10 @@
 #include <limits>
 #include <vector>
 
+#ifdef __ANDROID__
+#include <android/api-level.h>
+#endif
+
 #include "common/logging.h"
 #include "common/settings.h"
 #include "common/settings_enums.h"
@@ -172,26 +176,34 @@ bool Swapchain::AcquireNextImage() {
         break;
     }
 
-#ifdef __ANDROID__
-    scheduler.WaitSubmitted(resource_ticks[image_index]);
-#else
+    const auto wait_with_frame_pacing = [this] {
     switch (Settings::values.frame_pacing_mode.GetValue()) {
     case Settings::FramePacingMode::Target_Auto:
-        scheduler.WaitSubmitted(resource_ticks[image_index]);
+        scheduler.Wait(resource_ticks[image_index]);
         break;
     case Settings::FramePacingMode::Target_30:
-        scheduler.WaitSubmitted(resource_ticks[image_index], 30.0);
+        scheduler.Wait(resource_ticks[image_index], 30.0);
         break;
     case Settings::FramePacingMode::Target_60:
-        scheduler.WaitSubmitted(resource_ticks[image_index], 60.0);
+        scheduler.Wait(resource_ticks[image_index], 60.0);
         break;
     case Settings::FramePacingMode::Target_90:
-        scheduler.WaitSubmitted(resource_ticks[image_index], 90.0);
+        scheduler.Wait(resource_ticks[image_index], 90.0);
         break;
     case Settings::FramePacingMode::Target_120:
-        scheduler.WaitSubmitted(resource_ticks[image_index], 120.0);
+        scheduler.Wait(resource_ticks[image_index], 120.0);
         break;
     }
+    };
+
+#ifdef __ANDROID__
+    if (android_get_device_api_level() >= 30) {
+        scheduler.Wait(resource_ticks[image_index]);
+    } else {
+        wait_with_frame_pacing();
+    }
+#else
+    wait_with_frame_pacing();
 #endif
 
     resource_ticks[image_index] = scheduler.CurrentTick();

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -144,6 +144,11 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
           info.size.depth == 1;
 }
 
+[[nodiscard]] bool WillUseWidenedAstcFormat(const Device& device, const ImageInfo& info) {
+    return WillUseAcceleratedAstcDecode(device, info) &&
+           !VideoCore::Surface::IsPixelFormatSRGB(info.format);
+}
+
 [[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info,
                                                     std::optional<VkFormat> format_override = {}) {
     auto format_info =
@@ -207,11 +212,7 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
                 return device.IsFormatSupported(view_format, VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT,
                                                 FormatType::Optimal);
             });
-        const bool storage_allowed_for_samples =
-            image_ci.samples == VK_SAMPLE_COUNT_1_BIT ||
-            (device.GetStorageImageSampleCounts() &
-             static_cast<VkSampleCountFlags>(image_ci.samples)) != 0;
-        if (has_storage_compatible_view && storage_allowed_for_samples) {
+        if (has_storage_compatible_view) {
             image_ci.usage |= VK_IMAGE_USAGE_STORAGE_BIT;
         }
 
@@ -895,15 +896,6 @@ void BlitScale(Scheduler& scheduler, VkImage src_image, VkImage dst_image, const
                        0, nullptr, nullptr, write_barriers);
     });
 }
-
-[[nodiscard]] bool CanBlitNatively(const Device& device, PixelFormat format) {
-    static constexpr auto OPTIMAL_FORMAT = FormatType::Optimal;
-    static constexpr VkFormatFeatureFlags BLIT_USAGE =
-        VK_FORMAT_FEATURE_BLIT_SRC_BIT | VK_FORMAT_FEATURE_BLIT_DST_BIT;
-    const VkFormat vk_format =
-        MaxwellToVK::SurfaceFormat(device, OPTIMAL_FORMAT, false, format).format;
-    return device.IsFormatSupported(vk_format, BLIT_USAGE, OPTIMAL_FORMAT);
-}
 } // Anonymous namespace
 
 TextureCacheRuntime::TextureCacheRuntime(const Device& device_, Scheduler& scheduler_,
@@ -1236,19 +1228,27 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
         blit_image_helper.ResolveDepthStencil(dst_framebuffer, src, dst_region, src_region);
         return;
     }
-    static constexpr VkImageAspectFlags DEPTH_STENCIL_ASPECTS =
-        VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT;
-    if ((aspect_mask & DEPTH_STENCIL_ASPECTS) != 0 && !CanBlitNatively(device, src.format)) {
-        if (aspect_mask != DEPTH_STENCIL_ASPECTS) {
-            UNIMPLEMENTED_MSG("Host cannot blit format {} and no helper path exists for aspect "
-                              "mask 0x{:x}",
-                              src.format, aspect_mask);
+    if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
+        const auto format = src.format;
+        const auto can_blit_depth_stencil = [this, format] {
+            switch (format) {
+            case VideoCore::Surface::PixelFormat::D24_UNORM_S8_UINT:
+            case VideoCore::Surface::PixelFormat::S8_UINT_D24_UNORM:
+                return device.IsBlitDepth24Stencil8Supported();
+            case VideoCore::Surface::PixelFormat::D32_FLOAT_S8_UINT:
+                return device.IsBlitDepth32Stencil8Supported();
+            default:
+                UNREACHABLE();
+            }
+        }();
+        // Use shader-based depth/stencil blits if hardware doesn't support the format
+        // Note: MSAA resolves (MSAA->single) use vkCmdResolveImage which works fine
+        if (!can_blit_depth_stencil) {
+            UNIMPLEMENTED_IF(is_src_msaa || is_dst_msaa);
+            blit_image_helper.BlitDepthStencil(dst_framebuffer, src, dst_region, src_region,
+                                               filter, operation);
             return;
         }
-        UNIMPLEMENTED_IF(is_src_msaa || is_dst_msaa);
-        blit_image_helper.BlitDepthStencil(dst_framebuffer, src, dst_region, src_region, filter,
-                                           operation);
-        return;
     }
     ASSERT(!(is_dst_msaa && !is_src_msaa));
     ASSERT(operation == Fermi2D::Operation::SrcCopy);
@@ -1643,14 +1643,7 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
     const u32 num_samples = msaa_to_non_msaa ? src.info.num_samples : dst.info.num_samples;
     if (dst.AspectMask() != VK_IMAGE_ASPECT_COLOR_BIT ||
         VideoCore::Surface::IsPixelFormatInteger(dst.info.format)) {
-        const u64 key{(static_cast<u64>(dst.AspectMask()) << 32) |
-                      static_cast<u64>(dst.info.format)};
-        if (unsupported_msaa_resolves.insert(key).second) {
-            LOG_WARNING(Render_Vulkan,
-                        "MSAA resolve unsupported: format={}, aspect={:#x}, samples {}->{}",
-                        dst.info.format, dst.AspectMask(), src.info.num_samples,
-                        dst.info.num_samples);
-        }
+        UNIMPLEMENTED_MSG("Copying images with different samples is not supported.");
         return;
     }
     if (ENABLE_MSAA_RESOLVE_CONSUME && msaa_to_non_msaa && copies.size() == 1 &&
@@ -1760,20 +1753,6 @@ void TextureCacheRuntime::CopyImageMSAA(Image& dst, Image& src,
                                src.info.format, num_samples, copies, msaa_to_non_msaa);
 }
 
-u64 TextureCacheRuntime::CurrentSyncPoint() const noexcept {
-    return scheduler.CurrentTick();
-}
-
-u64 TextureCacheRuntime::CompletedSyncPoint() const {
-    auto& master_semaphore = scheduler.GetMasterSemaphore();
-    master_semaphore.Refresh();
-    return master_semaphore.KnownGpuTick();
-}
-
-void TextureCacheRuntime::WaitSyncPoint(u64 sync_point) {
-    scheduler.Wait(sync_point);
-}
-
 u64 TextureCacheRuntime::GetDeviceLocalMemory() const {
     return device.GetDeviceLocalMemory();
 }
@@ -1782,10 +1761,6 @@ u64 TextureCacheRuntime::GetDeviceMemoryUsage() const {
     return device.GetDeviceMemoryUsage();
 }
 
-u64 TextureCacheRuntime::GetDeviceAllocationUsage() const {
-    return device.GetMemoryBudgetInfo().allocation_bytes;
-}
-
 bool TextureCacheRuntime::CanReportMemoryUsage() const {
     return device.CanReportMemoryUsage();
 }
@@ -1795,7 +1770,6 @@ std::optional<size_t> TextureCacheRuntime::GetSamplerHeapBudget() const {
 }
 
 void TextureCacheRuntime::TickFrame() {
-    device.TickAllocatorFrame();
     std::erase_if(pending_msaa_images, [this](const auto& pending) {
         return scheduler.IsFree(pending.first);
     });
@@ -1806,7 +1780,12 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
     : VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime_.scheduler},
       runtime{&runtime_},
       original_image(MakeImage(runtime_.device, runtime_.memory_allocator, info,
-                               runtime->ViewFormats(info.format))),
+                               WillUseWidenedAstcFormat(runtime_.device, info)
+                                   ? std::span<const VkFormat>{}
+                                   : runtime->ViewFormats(info.format),
+                               WillUseWidenedAstcFormat(runtime_.device, info)
+                                   ? std::make_optional(VK_FORMAT_R32G32B32A32_SFLOAT)
+                                   : std::nullopt)),
       aspect_mask(ImageAspectMask(info.format)) {
     if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) {
         switch (Settings::values.accelerate_astc.GetValue()) {
@@ -1837,7 +1816,9 @@ Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu
         Settings::values.astc_recompression.GetValue() ==
             Settings::AstcRecompression::Uncompressed) {
         const auto& device = runtime->device.GetLogical();
-        const VkFormat storage_format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
+        const VkFormat storage_format = WillUseWidenedAstcFormat(runtime->device, info)
+                                            ? VK_FORMAT_R32G32B32A32_SFLOAT
+                                            : VK_FORMAT_A8B8G8R8_UNORM_PACK32;
         for (s32 level = 0; level < info.resources.levels; ++level) {
             storage_image_views[level] =
                 MakeStorageView(device, level, *original_image, storage_format);
@@ -1911,11 +1892,9 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
         ScaleDown(true);
     }
 
-    const bool is_color_upload = (aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0
+    const bool wants_msaa_upload = info.num_samples > 1
+        && (aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT) != 0
         && !VideoCore::Surface::IsPixelFormatInteger(info.format);
-    const bool is_depth_upload = (aspect_mask & VK_IMAGE_ASPECT_DEPTH_BIT) != 0;
-    const bool wants_msaa_upload =
-        info.num_samples > 1 && (is_color_upload || is_depth_upload);
 
     if (wants_msaa_upload) {
         ImageInfo temp_info = info;
@@ -1954,10 +1933,10 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
             image_copies.push_back(image_copy);
         }
 
-        runtime->TransitionImageLayout(*this);
         runtime->blit_image_helper.CopyMSAA(runtime->render_pass_cache, Handle(), info.format,
                                             temp_vk_image, info.format, info.num_samples,
                                             image_copies, false);
+        initialized = true;
         runtime->pending_msaa_images.emplace_back(scheduler->CurrentTick(), std::move(temp_image));
 
         if (is_rescaled) {
@@ -1968,9 +1947,6 @@ void Image::UploadMemory(VkBuffer buffer, VkDeviceSize offset,
 
     if (info.num_samples > 1) {
         LOG_WARNING(Render_Vulkan, "MSAA upload not implemented for format {}", info.format);
-        if (runtime != nullptr) {
-            runtime->TransitionImageLayout(*this);
-        }
         if (is_rescaled) {
             ScaleUp();
         }
@@ -2228,7 +2204,9 @@ VkImageView Image::StorageImageView(s32 level) noexcept {
         auto format_info =
             MaxwellToVK::SurfaceFormat(runtime->device, FormatType::Optimal, true, info.format);
         if (WillUseAcceleratedAstcDecode(runtime->device, info)) {
-            format_info.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32;
+            format_info.format = WillUseWidenedAstcFormat(runtime->device, info)
+                                     ? VK_FORMAT_R32G32B32A32_SFLOAT
+                                     : VK_FORMAT_A8B8G8R8_UNORM_PACK32;
         }
         view = MakeStorageView(runtime->device.GetLogical(), level, *(this->*current_image),
                                format_info.format);
@@ -2404,7 +2382,11 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
             SanitizeDepthStencilSwizzle(swizzle, device->SupportsDepthStencilSwizzleOne());
         }
     }
+    uses_widened_astc_format = WillUseWidenedAstcFormat(*device, image.info);
     auto format_info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format);
+    if (uses_widened_astc_format) {
+        format_info.format = VK_FORMAT_R32G32B32A32_SFLOAT;
+    }
     if (device->ApiVersion() >= VK_API_VERSION_1_3) {
         const VkFormatProperties3 properties3 =
             device->GetPhysical().GetFormatProperties3(format_info.format);
@@ -2547,12 +2529,14 @@ VkImageView ImageView::StorageView(Shader::TextureType texture_type,
                                    Shader::ImageFormat image_format) {
     if (image_handle) {
         if (image_format == Shader::ImageFormat::Typeless) {
-            auto& view{typeless_storage_views[static_cast<size_t>(texture_type)]};
-            if (!view) {
+            if (!typeless_storage_view) {
                 auto info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format);
-                view = MakeView(info.format, VK_IMAGE_ASPECT_COLOR_BIT, texture_type);
+                if (uses_widened_astc_format) {
+                    info.format = VK_FORMAT_R32G32B32A32_SFLOAT;
+                }
+                typeless_storage_view = MakeView(info.format, VK_IMAGE_ASPECT_COLOR_BIT, texture_type);
             }
-            return *view;
+            return *typeless_storage_view;
         }
         const bool is_signed = image_format == Shader::ImageFormat::R8_SINT
             || image_format == Shader::ImageFormat::R16_SINT;

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -60,22 +60,10 @@ public:
 
     void TickFrame();
 
-    u64 CurrentSyncPoint() const noexcept;
-
-    u64 CompletedSyncPoint() const;
-
-    void WaitSyncPoint(u64 sync_point);
-
     u64 GetDeviceLocalMemory() const;
 
     u64 GetDeviceMemoryUsage() const;
 
-    u64 GetDeviceAllocationUsage() const;
-
-    bool CanReportAllocationUsage() const noexcept {
-        return true;
-    }
-
     bool CanReportMemoryUsage() const;
 
     std::optional<size_t> GetSamplerHeapBudget() const;
@@ -168,7 +156,6 @@ public:
     std::array<vk::Buffer, indexing_slots> buffers{};
     std::vector<std::pair<u64, vk::Image>> pending_msaa_images;
     ankerl::unordered_dense::map<VkImage, ResolveShadow> resolve_shadows;
-    ankerl::unordered_dense::set<u64> unsupported_msaa_resolves;
 };
 
 class Framebuffer {
@@ -439,7 +426,7 @@ private:
 
     std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> image_views;
     std::optional<StorageViews> storage_views;
-    std::array<vk::ImageView, Shader::NUM_TEXTURE_TYPES> typeless_storage_views;
+    vk::ImageView typeless_storage_view;
     vk::ImageView depth_view;
     vk::ImageView stencil_view;
     vk::ImageView color_view;
@@ -449,6 +436,7 @@ private:
     VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT;
     u32 buffer_size = 0;
 
+    bool uses_widened_astc_format = false;
     bool supports_depth_comparison = false;
 };
 
@@ -499,7 +487,6 @@ struct TextureCacheParams {
     static constexpr bool HAS_EMULATED_COPIES = false;
     static constexpr bool HAS_DEVICE_MEMORY_INFO = true;
     static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = true;
-    static constexpr bool HAS_TIMELINE_SYNC_POINTS = true;
 
     using Runtime = Vulkan::TextureCacheRuntime;
     using Image = Vulkan::Image;

src/video_core/renderer_vulkan/vk_update_descriptor.cpp

@@ -16,9 +16,8 @@
 
 namespace Vulkan {
 
-UpdateDescriptorQueue::UpdateDescriptorQueue(const Device& device_, size_t frame_payload_size_)
-    : device{device_}, frame_payload_size{frame_payload_size_},
-      payload(frame_payload_size_ * FRAMES_IN_FLIGHT)
+UpdateDescriptorQueue::UpdateDescriptorQueue(const Device& device_)
+    : device{device_}
 {
     payload_start = payload.data();
     payload_cursor = payload.data();
@@ -30,19 +29,19 @@ void UpdateDescriptorQueue::TickFrame() {
     if (++frame_index >= FRAMES_IN_FLIGHT) {
         frame_index = 0;
     }
-    payload_start = payload.data() + frame_index * frame_payload_size;
+    payload_start = payload.data() + frame_index * FRAME_PAYLOAD_SIZE;
     payload_cursor = payload_start;
 }
 
 void UpdateDescriptorQueue::Acquire(Scheduler& scheduler, size_t required_entries) {
     static constexpr size_t DEFAULT_REQUIRED_ENTRIES = 0x400;
     const size_t reserve = required_entries > 0 ? required_entries : DEFAULT_REQUIRED_ENTRIES;
-    ASSERT_MSG(reserve < frame_payload_size, "Descriptor reservation {} >= frame capacity {}",
-               reserve, frame_payload_size);
+    ASSERT_MSG(reserve < FRAME_PAYLOAD_SIZE, "Descriptor reservation {} >= frame capacity {}",
+               reserve, FRAME_PAYLOAD_SIZE);
     const size_t used = static_cast<size_t>(std::distance(payload_start, payload_cursor));
-    if (used + reserve >= frame_payload_size) {
+    if (used + reserve >= FRAME_PAYLOAD_SIZE) {
         LOG_WARNING(Render_Vulkan, "Payload overflow (used={}, reserve={}, capacity={})",
-                    used, reserve, frame_payload_size);
+                    used, reserve, FRAME_PAYLOAD_SIZE);
         scheduler.WaitWorker();
         payload_cursor = payload_start;
     }

src/video_core/renderer_vulkan/vk_update_descriptor.h

@@ -6,8 +6,8 @@
 
 #pragma once
 
+#include <array>
 #include <variant>
-#include <vector>
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
 namespace Vulkan {
@@ -30,12 +30,11 @@ class UpdateDescriptorQueue final {
     // This should be plenty for the vast majority of cases. Most desktop platforms only
     // provide up to 3 swapchain images.
     static constexpr size_t FRAMES_IN_FLIGHT = 8;
+    static constexpr size_t FRAME_PAYLOAD_SIZE = 0x20000;
+    static constexpr size_t PAYLOAD_SIZE = FRAME_PAYLOAD_SIZE * FRAMES_IN_FLIGHT;
 
 public:
-    static constexpr size_t GUEST_FRAME_PAYLOAD_SIZE = 0x80000;
-    static constexpr size_t COMPUTE_FRAME_PAYLOAD_SIZE = 0x20000;
-
-    explicit UpdateDescriptorQueue(const Device& device_, size_t frame_payload_size_);
+    explicit UpdateDescriptorQueue(const Device& device_);
     ~UpdateDescriptorQueue();
 
     void TickFrame();
@@ -75,12 +74,11 @@ public:
 
 private:
     const Device& device;
-    const size_t frame_payload_size;
     size_t frame_index{0};
     DescriptorUpdateEntry* payload_cursor = nullptr;
     DescriptorUpdateEntry* payload_start = nullptr;
     const DescriptorUpdateEntry* upload_start = nullptr;
-    std::vector<DescriptorUpdateEntry> payload;
+    std::array<DescriptorUpdateEntry, PAYLOAD_SIZE> payload;
 };
 
 // TODO: should these be separate classes instead?

src/video_core/shader_environment.cpp

@@ -284,24 +284,7 @@ std::optional<u64> GenericEnvironment::TryFindSize() {
 Tegra::Texture::TICEntry GenericEnvironment::ReadTextureInfo(GPUVAddr tic_addr, u32 tic_limit,
                                                              bool via_header_index, u32 raw) {
     const auto handle{Tegra::Texture::TexturePair(raw, via_header_index)};
-    if (handle.first > tic_limit) {
-        LOG_CRITICAL(Shader,
-                    "TIC index out of range: raw=0x{:08x} tic_index={} tsc_index={} tic_limit={} "
-                    "tic_addr=0x{:x} via_header_index={} stage={} program_base=0x{:x} "
-                    "start_address=0x{:x}",
-                    raw, handle.first, handle.second, tic_limit, tic_addr, via_header_index,
-                    static_cast<u32>(stage), program_base, start_address);
-        ASSERT(handle.first <= tic_limit);
-        Tegra::Texture::TICEntry fallback{};
-        fallback.format.Assign(Tegra::Texture::TextureFormat::A8B8G8R8);
-        fallback.r_type.Assign(Tegra::Texture::ComponentType::UNORM);
-        fallback.g_type.Assign(Tegra::Texture::ComponentType::UNORM);
-        fallback.b_type.Assign(Tegra::Texture::ComponentType::UNORM);
-        fallback.a_type.Assign(Tegra::Texture::ComponentType::UNORM);
-        fallback.texture_type.Assign(Tegra::Texture::TextureType::Texture2D);
-        fallback.normalized_coords.Assign(1);
-        return fallback;
-    }
+    ASSERT(handle.first <= tic_limit);
     const GPUVAddr descriptor_addr{tic_addr + handle.first * sizeof(Tegra::Texture::TICEntry)};
     Tegra::Texture::TICEntry entry;
     gpu_memory->ReadBlock(descriptor_addr, &entry, sizeof(entry));

src/video_core/texture_cache/texture_cache.h

@@ -58,9 +58,23 @@ TextureCache<P>::TextureCache(Runtime& runtime_, Tegra::MaxwellDeviceMemoryManag
     void(slot_samplers.insert(runtime, sampler_descriptor));
 
     if constexpr (HAS_DEVICE_MEMORY_INFO) {
-        memory_budget = runtime.GetDeviceLocalMemory();
+        const s64 device_local_memory = static_cast<s64>(runtime.GetDeviceLocalMemory());
+        const s64 min_spacing_expected = device_local_memory - 1_GiB;
+        const s64 min_spacing_critical = device_local_memory - 512_MiB;
+        const s64 mem_threshold = (std::min)(device_local_memory, TARGET_THRESHOLD);
+        const s64 min_vacancy_expected = (6 * mem_threshold) / 10;
+        const s64 min_vacancy_critical = (2 * mem_threshold) / 10;
+        expected_memory = static_cast<u64>(
+            (std::max)((std::min)(device_local_memory - min_vacancy_expected, min_spacing_expected),
+                     DEFAULT_EXPECTED_MEMORY));
+        critical_memory = static_cast<u64>(
+            (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);
     } else {
-        memory_budget = FALLBACK_MEMORY_BUDGET;
+        expected_memory = DEFAULT_EXPECTED_MEMORY + 512_MiB;
+        critical_memory = DEFAULT_CRITICAL_MEMORY + 1_GiB;
+        minimum_memory = 0;
     }
 
     const bool gpu_unswizzle_enabled = Settings::values.gpu_unswizzle_enabled.GetValue();
@@ -100,154 +114,71 @@ TextureCache<P>::TextureCache(Runtime& runtime_, Tegra::MaxwellDeviceMemoryManag
 }
 
 template <class P>
-void TextureCache<P>::QueueEvictionDownload(Image& image) {
-    auto copies = FullDownloadCopies(image.info);
-    auto staging = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes, true);
-    image.DownloadMemory(staging, FixSmallVectorADL(copies));
-    pending_eviction_downloads.push_back(PendingEvictionDownload{
-        .staging = staging,
-        .gpu_memory = gpu_memory,
-        .copies = std::move(copies),
-        .info = image.info,
-        .gpu_addr = image.gpu_addr,
-        .sync_point = runtime.CurrentSyncPoint(),
-    });
-}
-
-template <class P>
-void TextureCache<P>::TickEvictionDownloads(u64 completed_sync_point) {
-    while (!pending_eviction_downloads.empty() &&
-           pending_eviction_downloads.front().sync_point <= completed_sync_point) {
-        auto& entry = pending_eviction_downloads.front();
-        SwizzleImage(*entry.gpu_memory, entry.gpu_addr, entry.info, FixSmallVectorADL(entry.copies),
-                     entry.staging.mapped_span.subspan(entry.staging.offset), swizzle_data_buffer);
-        runtime.FreeDeferredStagingBuffer(entry.staging);
-        pending_eviction_downloads.pop_front();
-    }
-}
-
-template <class P>
-void TextureCache<P>::FlushEvictionDownloads() {
-    if (pending_eviction_downloads.empty()) {
-        return;
-    }
-    const u64 last_sync_point = pending_eviction_downloads.back().sync_point;
-    runtime.WaitSyncPoint(last_sync_point);
-    TickEvictionDownloads(last_sync_point);
-}
-
-template <class P>
-u64 TextureCache<P>::ImageSizeBytes(const ImageBase& image) {
-    u64 tentative_size = (std::max)(image.guest_size_bytes, image.unswizzled_size_bytes);
-    if ((IsPixelFormatASTC(image.info.format) &&
-         True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
-        True(image.flags & ImageFlagBits::Converted)) {
-        tentative_size = TranscodedAstcSize(tentative_size, image.info.format);
-    }
-    u64 size = Common::AlignUp(tentative_size, 1024);
-    if (image.HasScaled()) {
-        size += GetScaledImageSizeBytes(image);
-    }
-    return size;
-}
-
-template <class P>
-u64 TextureCache<P>::DeviceUsage(bool force_refresh) {
-    if (!runtime.CanReportAllocationUsage()) {
-        return total_used_memory;
-    }
-    if (force_refresh || usage_refresh_countdown == 0) {
-        cached_device_usage = runtime.GetDeviceAllocationUsage();
-        usage_refresh_countdown = USAGE_REFRESH_INTERVAL;
-    } else {
-        --usage_refresh_countdown;
-    }
-    return cached_device_usage;
-}
-
-template <class P>
-u64 TextureCache<P>::ReclaimMemory(u64 target_bytes, bool allow_download) {
-    if (target_bytes == 0 || in_reclaim) {
-        return 0;
-    }
-    in_reclaim = true;
-    const u64 drain_point = runtime.CompletedSyncPoint();
-    TickEvictionDownloads(drain_point);
-    sentenced_images.Reclaim(drain_point);
-    sentenced_image_view.Reclaim(drain_point);
-    sentenced_framebuffers.Reclaim(drain_point);
-    u64 freed = 0;
-    const auto evict = [&](ImageId image_id) {
-        if (freed >= target_bytes) {
+void TextureCache<P>::RunGarbageCollector() {
+    bool high_priority_mode = false;
+    bool aggressive_mode = false;
+    u64 ticks_to_destroy = 0;
+    size_t num_iterations = 0;
+    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);
+    };
+    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];
         if (True(image.flags & ImageFlagBits::IsDecoding)) {
             return false;
         }
-        const bool must_download =
-            image.IsSafeDownload() && False(image.flags & ImageFlagBits::BadOverlap);
-        bool queued_download = false;
-        if (must_download) {
-            if constexpr (HAS_TIMELINE_SYNC_POINTS) {
-                if (!allow_download) {
-                    return false;
-                }
-                QueueEvictionDownload(image);
-                queued_download = true;
-            } else {
-                return false;
-            }
+        const bool must_download = image.IsSafeDownload() && False(image.flags & ImageFlagBits::BadOverlap);
+        if ((!aggressive_mode && True(image.flags & ImageFlagBits::CostlyLoad)) || (!high_priority_mode && must_download)) {
+            return false;
+        }
+        if (must_download) {
+            auto map = runtime.DownloadStagingBuffer(image.unswizzled_size_bytes);
+            const auto copies = FixSmallVectorADL(FullDownloadCopies(image.info));
+            image.DownloadMemory(map, copies);
+            runtime.Finish();
+            SwizzleImage(*gpu_memory, image.gpu_addr, image.info, copies, map.mapped_span, swizzle_data_buffer);
         }
-        const u64 image_bytes = ImageSizeBytes(image);
         if (True(image.flags & ImageFlagBits::Tracked)) {
             UntrackImage(image, image_id);
         }
         UnregisterImage(image_id);
-        DeleteImage(image_id, !queued_download && image.scale_tick > frame_tick + 5);
-        freed += image_bytes;
+        DeleteImage(image_id, image.scale_tick > frame_tick + 5);
+        if (aggressive_mode && total_used_memory < critical_memory) {
+            num_iterations >>= 2;
+            aggressive_mode = false;
+        } else if (high_priority_mode && total_used_memory < expected_memory) {
+            num_iterations >>= 1;
+            high_priority_mode = false;
+        }
         return false;
     };
-    const u64 cold_tick =
-        frame_tick > RECLAIM_GUARD_FRAMES ? frame_tick - RECLAIM_GUARD_FRAMES : 0;
-    lru_cache.ForEachItemBelow(cold_tick, evict);
-    if (freed < target_bytes) {
-        lru_cache.ForEachItemBelow(frame_tick > 0 ? frame_tick - 1 : 0, evict);
+    Configure(false);
+    lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, Cleanup);
+    if (total_used_memory >= critical_memory) {
+        Configure(true);
+        lru_cache.ForEachItemBelow(frame_tick - ticks_to_destroy, Cleanup);
     }
-    const u64 exit_point = runtime.CompletedSyncPoint();
-    sentenced_images.Reclaim(exit_point);
-    sentenced_image_view.Reclaim(exit_point);
-    sentenced_framebuffers.Reclaim(exit_point);
-    in_reclaim = false;
-    usage_refresh_countdown = 0;
-    reclaim_stalled = freed == 0;
-    return freed;
-}
-
-template <class P>
-void TextureCache<P>::EnsureHeadroom(bool allow_download) {
-    if (reclaim_stalled) {
-        return;
-    }
-    const u64 limit = memory_budget > RECLAIM_HEADROOM ? memory_budget - RECLAIM_HEADROOM : 0;
-    const u64 usage = DeviceUsage(false);
-    if (usage <= limit) {
-        return;
-    }
-    const u64 target = (limit / 100) * RECLAIM_TARGET_PERCENT;
-    ReclaimMemory((std::min)(usage - target, total_used_memory), allow_download);
 }
 
 template <class P>
 void TextureCache<P>::TickFrame() {
-    usage_refresh_countdown = 0;
-    reclaim_stalled = false;
-    EnsureHeadroom(true);
-    const u64 completed_sync_point = runtime.CompletedSyncPoint();
-    TickEvictionDownloads(completed_sync_point);
-    sentenced_images.Reclaim(completed_sync_point);
-    sentenced_framebuffers.Reclaim(completed_sync_point);
-    sentenced_image_view.Reclaim(completed_sync_point);
+    // If we can obtain the memory info, use it instead of the estimate.
+    if (runtime.CanReportMemoryUsage()) {
+        total_used_memory = runtime.GetDeviceMemoryUsage();
+    }
+    if (total_used_memory > minimum_memory) {
+        RunGarbageCollector();
+    }
+    sentenced_images.Tick();
+    sentenced_framebuffers.Tick();
+    sentenced_image_view.Tick();
     TickAsyncDecode();
     TickAsyncUnswizzle();
 
@@ -665,7 +596,6 @@ void TextureCache<P>::WriteMemory(DAddr cpu_addr, size_t size) {
 
 template <class P>
 void TextureCache<P>::DownloadMemory(DAddr cpu_addr, size_t size) {
-    FlushEvictionDownloads();
     boost::container::small_vector<ImageId, 16> images;
     ForEachImageInRegion(cpu_addr, size, [&images](ImageId image_id, ImageBase& image) {
         if (!image.IsSafeDownload()) {
@@ -964,7 +894,6 @@ void TextureCache<P>::CommitAsyncFlushes() {
 
 template <class P>
 void TextureCache<P>::PopAsyncFlushes() {
-    TickEvictionDownloads(runtime.CompletedSyncPoint());
     if (committed_downloads.empty()) {
         return;
     }
@@ -1365,9 +1294,8 @@ void TextureCache<P>::InvalidateScale(Image& image) {
     }
     RemoveImageViewReferences(image_view_ids);
     RemoveFramebuffers(image_view_ids);
-    const u64 sync_point = runtime.CurrentSyncPoint();
     for (const ImageViewId image_view_id : image_view_ids) {
-        sentenced_image_view.Push(std::move(slot_image_views[image_view_id]), sync_point);
+        sentenced_image_view.Push(std::move(slot_image_views[image_view_id]));
         slot_image_views.erase(image_view_id);
     }
     image.image_view_ids.clear();
@@ -1403,7 +1331,6 @@ void TextureCache<P>::QueueAsyncDecode(Image& image, ImageId image_id) {
     LOG_INFO(HW_GPU, "Queuing async texture decode");
 
     image.flags |= ImageFlagBits::IsDecoding;
-    runtime.TransitionImageLayout(image);
     auto decode = std::make_unique<AsyncDecodeContext>();
     auto* decode_ptr = decode.get();
     decode->image_id = image_id;
@@ -1436,7 +1363,6 @@ void TextureCache<P>::QueueAsyncUnswizzle(Image& image, ImageId image_id) {
     }
 
     image.flags |= ImageFlagBits::IsDecoding;
-    runtime.TransitionImageLayout(image);
 
     unswizzle_queue.push_back({
         .image_id = image_id,
@@ -1597,7 +1523,6 @@ ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
 
 template <class P>
 ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, DAddr cpu_addr) {
-    EnsureHeadroom(false);
     ImageInfo new_info = info;
     const size_t size_bytes = CalculateGuestSizeInBytes(new_info);
     const bool broken_views = runtime.HasBrokenTextureViewFormats();
@@ -1706,28 +1631,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, DA
     for (const ImageId overlap_id : join_ignore_textures) {
         Image& overlap = slot_images[overlap_id];
         if (True(overlap.flags & ImageFlagBits::GpuModified)) {
-            if (new_image.TryFindBase(overlap.gpu_addr) &&
-                (!can_rescale || ImageCanRescale(overlap))) {
-                if (can_rescale) {
-                    ScaleUp(overlap);
-                } else {
-                    ScaleDown(overlap);
-                }
-                join_copies_to_do.emplace_back(JoinCopy{false, overlap_id});
-                continue;
-            }
-            if (overlap.IsSafeDownload() && False(overlap.flags & ImageFlagBits::BadOverlap) &&
-                gpu_memory->GpuToCpuAddress(overlap.gpu_addr).has_value()) {
-                QueueEvictionDownload(overlap);
-            } else {
-                LOG_WARNING(HW_GPU,
-                            "Dropping GPU modified overlap, contents are not recoverable: "
-                            "gpu_addr=0x{:x} format={} size={}x{}x{} levels={} layers={}",
-                            overlap.gpu_addr, static_cast<int>(overlap.info.format),
-                            overlap.info.size.width, overlap.info.size.height,
-                            overlap.info.size.depth, overlap.info.resources.levels,
-                            overlap.info.resources.layers);
-            }
+            UNIMPLEMENTED();
         }
         if (True(overlap.flags & ImageFlagBits::Tracked)) {
             UntrackImage(overlap, overlap_id);
@@ -2281,7 +2185,13 @@ void TextureCache<P>::RegisterImage(ImageId image_id) {
     ASSERT_MSG(False(image.flags & ImageFlagBits::Registered),
                "Trying to register an already registered image");
     image.flags |= ImageFlagBits::Registered;
-    total_used_memory += ImageSizeBytes(image);
+    u64 tentative_size = (std::max)(image.guest_size_bytes, image.unswizzled_size_bytes);
+    if ((IsPixelFormatASTC(image.info.format) &&
+         True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
+        True(image.flags & ImageFlagBits::Converted)) {
+        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);
 
     ForEachGPUPage(image.gpu_addr, image.guest_size_bytes, [this, image_id](u64 page) {
@@ -2444,7 +2354,16 @@ void TextureCache<P>::UntrackImage(ImageBase& image, ImageId image_id) {
 template <class P>
 void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
     ImageBase& image = slot_images[image_id];
-    total_used_memory -= std::min<u64>(total_used_memory, ImageSizeBytes(image));
+    if (image.HasScaled()) {
+        total_used_memory -= GetScaledImageSizeBytes(image);
+    }
+    u64 tentative_size = (std::max)(image.guest_size_bytes, image.unswizzled_size_bytes);
+    if ((IsPixelFormatASTC(image.info.format) &&
+         True(image.flags & ImageFlagBits::AcceleratedUpload)) ||
+        True(image.flags & ImageFlagBits::Converted)) {
+        tentative_size = TranscodedAstcSize(tentative_size, image.info.format);
+    }
+    total_used_memory -= Common::AlignUp(tentative_size, 1024);
     const GPUVAddr gpu_addr = image.gpu_addr;
     const auto alloc_it = image_allocs_table.find(gpu_addr);
     if (alloc_it == image_allocs_table.end()) {
@@ -2498,15 +2417,14 @@ void TextureCache<P>::DeleteImage(ImageId image_id, bool immediate_delete) {
         ASSERT_MSG(num_removed_overlaps == 1, "Invalid number of removed overlapps: {}",
                    num_removed_overlaps);
     }
-    const u64 sync_point = runtime.CurrentSyncPoint();
     for (const ImageViewId image_view_id : image_view_ids) {
         if (!immediate_delete) {
-            sentenced_image_view.Push(std::move(slot_image_views[image_view_id]), sync_point);
+            sentenced_image_view.Push(std::move(slot_image_views[image_view_id]));
         }
         slot_image_views.erase(image_view_id);
     }
     if (!immediate_delete) {
-        sentenced_images.Push(std::move(slot_images[image_id]), sync_point);
+        sentenced_images.Push(std::move(slot_images[image_id]));
     }
     slot_images.erase(image_id);
 
@@ -2552,8 +2470,7 @@ void TextureCache<P>::RemoveFramebuffers(std::span<const ImageViewId> removed_vi
                 last_framebuffer_id = {};
                 last_framebuffer_serial = 0;
             }
-            sentenced_framebuffers.Push(std::move(slot_framebuffers[framebuffer_id]),
-                                        runtime.CurrentSyncPoint());
+            sentenced_framebuffers.Push(std::move(slot_framebuffers[framebuffer_id]));
             it = framebuffers.erase(it);
         } else {
             ++it;

src/video_core/texture_cache/texture_cache_base.h

@@ -30,7 +30,7 @@
 #include "common/thread_worker.h"
 #include "video_core/compatible_formats.h"
 #include "video_core/control/channel_state_cache.h"
-#include "video_core/deferred_destruction_queue.h"
+#include "video_core/delayed_destruction_ring.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/surface.h"
 #include "video_core/texture_cache/descriptor_table.h"
@@ -108,20 +108,18 @@ class TextureCache : public VideoCommon::ChannelSetupCaches<TextureCacheChannelI
     static constexpr bool HAS_DEVICE_MEMORY_INFO = P::HAS_DEVICE_MEMORY_INFO;
     /// True when the API can do asynchronous texture downloads.
     static constexpr bool IMPLEMENTS_ASYNC_DOWNLOADS = P::IMPLEMENTS_ASYNC_DOWNLOADS;
-    static constexpr bool HAS_TIMELINE_SYNC_POINTS = P::HAS_TIMELINE_SYNC_POINTS;
 
     static constexpr size_t UNSET_CHANNEL{(std::numeric_limits<size_t>::max)()};
 
 #ifdef YUZU_LEGACY
-    static constexpr u64 RECLAIM_HEADROOM = 384_MiB;
+    static constexpr s64 TARGET_THRESHOLD = 3_GiB;
 #else
-    static constexpr u64 RECLAIM_HEADROOM = 512_MiB;
+    static constexpr s64 TARGET_THRESHOLD = 4_GiB;
 #endif
 
-    static constexpr u64 FALLBACK_MEMORY_BUDGET = 2_GiB;
-    static constexpr u32 USAGE_REFRESH_INTERVAL = 16;
-    static constexpr u64 RECLAIM_GUARD_FRAMES = 8;
-    static constexpr u64 RECLAIM_TARGET_PERCENT = 88;
+    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;
 
     using Runtime = typename P::Runtime;
     using Image = typename P::Image;
@@ -156,8 +154,6 @@ public:
     /// Notify the cache that a new frame has been queued
     void TickFrame();
 
-    u64 ReclaimMemory(u64 target_bytes, bool allow_download);
-
     /// Return a constant reference to the given image view id
     [[nodiscard]] const ImageView& GetImageView(ImageViewId id) const noexcept;
 
@@ -297,17 +293,8 @@ private:
 
     void OnGPUASRegister(size_t map_id) final override;
 
-    u64 ImageSizeBytes(const ImageBase& image);
-
-    u64 DeviceUsage(bool force_refresh);
-
-    void EnsureHeadroom(bool allow_download);
-
-    void QueueEvictionDownload(Image& image);
-
-    void TickEvictionDownloads(u64 completed_sync_point);
-
-    void FlushEvictionDownloads();
+    /// Runs the Garbage Collector.
+    void RunGarbageCollector();
 
     /// Find or create an image view in the guest descriptor table
     ImageViewId VisitImageView(u32 index, bool compute);
@@ -464,11 +451,9 @@ private:
     bool has_deleted_images = false;
     bool is_rescaling = false;
     u64 total_used_memory = 0;
-    u64 memory_budget = 0;
-    u64 cached_device_usage = 0;
-    u32 usage_refresh_countdown = 0;
-    bool in_reclaim = false;
-    bool reclaim_stalled = false;
+    u64 minimum_memory;
+    u64 expected_memory;
+    u64 critical_memory;
     size_t gpu_unswizzle_maxsize = 0;
     size_t swizzle_chunk_size = 0;
     u32 swizzle_slices_per_batch = 0;
@@ -506,19 +491,14 @@ private:
     };
     Common::LeastRecentlyUsedCache<LRUItemParams> lru_cache;
 
-    DeferredDestructionQueue<Image> sentenced_images;
-    DeferredDestructionQueue<ImageView> sentenced_image_view;
-    DeferredDestructionQueue<Framebuffer> sentenced_framebuffers;
-
-    struct PendingEvictionDownload {
-        AsyncBuffer staging;
-        Tegra::MemoryManager* gpu_memory;
-        boost::container::small_vector<VideoCommon::BufferImageCopy, 16> copies;
-        VideoCommon::ImageInfo info;
-        GPUVAddr gpu_addr;
-        u64 sync_point;
-    };
-    std::deque<PendingEvictionDownload> pending_eviction_downloads;
+ #ifdef YUZU_LEGACY
+    static constexpr size_t TICKS_TO_DESTROY = 6;
+ #else
+    static constexpr size_t TICKS_TO_DESTROY = 8;
+#endif
+    DelayedDestructionRing<Image, TICKS_TO_DESTROY> sentenced_images;
+    DelayedDestructionRing<ImageView, TICKS_TO_DESTROY> sentenced_image_view;
+    DelayedDestructionRing<Framebuffer, TICKS_TO_DESTROY> sentenced_framebuffers;
 
     ankerl::unordered_dense::map<GPUVAddr, ImageAllocId> image_allocs_table;
 
@@ -529,8 +509,7 @@ private:
     u64 frame_tick = 0;
     u64 last_sampler_gc_frame = (std::numeric_limits<u64>::max)();
 
-    Common::ThreadWorker texture_decode_worker{1, "TextureDecoder", {},
-                                               Common::ThreadPlacement::Background};
+    Common::ThreadWorker texture_decode_worker{1, "TextureDecoder"};
     std::vector<std::unique_ptr<AsyncDecodeContext>> async_decodes;
 
     std::deque<PendingUnswizzle> unswizzle_queue;

src/video_core/texture_cache/util.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
 
 // SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
@@ -747,7 +747,7 @@ boost::container::small_vector<ImageCopy, 16> MakeShrinkImageCopies(const ImageI
 
     const bool is_dst_3d = dst.type == ImageType::e3D;
     if (is_dst_3d) {
-        ASSERT(src.type == ImageType::e3D || src.resources.layers == 1);
+        ASSERT(src.type == ImageType::e3D);
         ASSERT(src.resources.levels == 1);
     }
     const bool both_2d{src.type == ImageType::e2D && dst.type == ImageType::e2D};

src/video_core/textures/workers.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
 
 // SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
@@ -10,8 +10,7 @@ namespace Tegra::Texture {
 
 Common::ThreadWorker& GetThreadWorkers() {
     static Common::ThreadWorker workers{(std::max)(std::thread::hardware_concurrency(), 2U) / 2,
-                                        "ImageTranscode", {},
-                                        Common::ThreadPlacement::Background};
+                                        "ImageTranscode"};
 
     return workers;
 }

src/video_core/vulkan_common/vulkan_device.cpp

@@ -5,7 +5,6 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <algorithm>
-#include <array>
 #include <bitset>
 #include <chrono>
 #include <optional>
@@ -733,7 +732,7 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
             .device = *logical,
             .preferredLargeHeapBlockSize = is_integrated
                                            ? (64u * 1024u * 1024u)
-                                           : (128u * 1024u * 1024u),
+                                           : (256u * 1024u * 1024u),
             .pAllocationCallbacks = nullptr,
             .pDeviceMemoryCallbacks = nullptr,
             .pHeapSizeLimit = nullptr,
@@ -745,32 +744,12 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
 
     vk::Check(vmaCreateAllocator(&allocator_info, &allocator));
 
-    {
-        const auto& limits = properties.properties.limits;
-        LOG_INFO(Render_Vulkan, "MSAA sample count support:");
-        LOG_INFO(Render_Vulkan, "  framebufferColorSampleCounts:   {:#x}",
-                 limits.framebufferColorSampleCounts);
-        LOG_INFO(Render_Vulkan, "  framebufferDepthSampleCounts:   {:#x}",
-                 limits.framebufferDepthSampleCounts);
-        LOG_INFO(Render_Vulkan, "  framebufferStencilSampleCounts: {:#x}",
-                 limits.framebufferStencilSampleCounts);
-        LOG_INFO(Render_Vulkan, "  sampledImageColorSampleCounts:  {:#x}",
-                 limits.sampledImageColorSampleCounts);
-        LOG_INFO(Render_Vulkan, "  sampledImageDepthSampleCounts:  {:#x}",
-                 limits.sampledImageDepthSampleCounts);
-        LOG_INFO(Render_Vulkan, "  sampledImageIntegerSampleCounts:{:#x}",
-                 limits.sampledImageIntegerSampleCounts);
-        LOG_INFO(Render_Vulkan, "  storageImageSampleCounts:       {:#x}",
-                 limits.storageImageSampleCounts);
-    }
-
     // Initialize GPU logging if enabled
     InitializeGPULogging();
 }
 
 Device::~Device() {
     ShutdownGPULogging();
-    vk::FlushDeletionQueue();
     vmaDestroyAllocator(allocator);
 }
 
@@ -838,8 +817,7 @@ bool Device::ComputeIsOptimalAstcSupported() const {
         VK_FORMAT_ASTC_12x10_UNORM_BLOCK, VK_FORMAT_ASTC_12x10_SRGB_BLOCK,
         VK_FORMAT_ASTC_12x12_UNORM_BLOCK, VK_FORMAT_ASTC_12x12_SRGB_BLOCK,
     };
-    if (!features.features.textureCompressionASTC_LDR ||
-        !features.texture_compression_astc_hdr.textureCompressionASTC_HDR) {
+    if (!features.features.textureCompressionASTC_LDR) {
         return false;
     }
     const auto format_feature_usage{VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT |
@@ -1231,20 +1209,10 @@ void Device::RemoveUnsuitableExtensions() {
     RemoveExtensionFeatureIfUnsuitable(extensions.depth_bias_control, features.depth_bias_control,
                                        VK_EXT_DEPTH_BIAS_CONTROL_EXTENSION_NAME);
 
-    // VK_EXT_depth_clamp_zero_one
-    extensions.depth_clamp_zero_one = features.depth_clamp_zero_one.depthClampZeroOne;
-    RemoveExtensionFeatureIfUnsuitable(extensions.depth_clamp_zero_one,
-                                       features.depth_clamp_zero_one,
-                                       VK_EXT_DEPTH_CLAMP_ZERO_ONE_EXTENSION_NAME);
-
     // VK_EXT_depth_clip_control
     extensions.depth_clip_control = features.depth_clip_control.depthClipControl;
     RemoveExtensionFeatureIfUnsuitable(extensions.depth_clip_control, features.depth_clip_control,
                                        VK_EXT_DEPTH_CLIP_CONTROL_EXTENSION_NAME);
-    // VK_EXT_depth_clip_enable
-    extensions.depth_clip_enable = features.depth_clip_enable.depthClipEnable;
-    RemoveExtensionFeatureIfUnsuitable(extensions.depth_clip_enable, features.depth_clip_enable,
-                                       VK_EXT_DEPTH_CLIP_ENABLE_EXTENSION_NAME);
 
     // VK_EXT_extended_dynamic_state
     extensions.extended_dynamic_state = features.extended_dynamic_state.extendedDynamicState;
@@ -1476,23 +1444,6 @@ std::optional<size_t> Device::GetSamplerHeapBudget() const {
     return sampler_heap_budget;
 }
 
-Device::MemoryBudgetInfo Device::GetMemoryBudgetInfo() const {
-    std::array<VmaBudget, VK_MAX_MEMORY_HEAPS> budgets{};
-    vmaGetHeapBudgets(allocator, budgets.data());
-    MemoryBudgetInfo info{};
-    for (const size_t heap : valid_heap_memory) {
-        info.usage += budgets[heap].usage;
-        info.budget += budgets[heap].budget;
-        info.block_bytes += budgets[heap].statistics.blockBytes;
-        info.allocation_bytes += budgets[heap].statistics.allocationBytes;
-    }
-    return info;
-}
-
-void Device::TickAllocatorFrame() const {
-    vmaSetCurrentFrameIndex(allocator, ++allocator_frame_index);
-}
-
 u64 Device::GetDeviceMemoryUsage() const {
     VkPhysicalDeviceMemoryBudgetPropertiesEXT budget;
     budget.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT;
@@ -1542,12 +1493,9 @@ void Device::CollectPhysicalMemoryInfo() {
         device_access_memory -= reserve_memory;
         if (Settings::values.vram_usage_mode.GetValue() != Settings::VramUsageMode::Aggressive) {
             // Account for resolution scaling in memory limits
-            const u64 normal_memory = 6_GiB;
-            const u64 scaler_memory = 1_GiB * Settings::values.resolution_info.ScaleUp(1);
-            const u64 baseline = normal_memory + scaler_memory;
-            const u64 proportional = (device_access_memory / 4) * 3;
-            device_access_memory =
-                std::min<u64>(device_access_memory, std::max<u64>(baseline, proportional));
+            const size_t normal_memory = 6_GiB;
+            const size_t scaler_memory = 1_GiB * Settings::values.resolution_info.ScaleUp(1);
+            device_access_memory = std::min<u64>(device_access_memory, normal_memory + scaler_memory);
         }
     }
 }

src/video_core/vulkan_common/vulkan_device.h

@@ -54,9 +54,7 @@ VK_DEFINE_HANDLE(VmaAllocator)
     FEATURE(EXT, ColorWriteEnable, COLOR_WRITE_ENABLE, color_write_enable)                         \
     FEATURE(EXT, CustomBorderColor, CUSTOM_BORDER_COLOR, custom_border_color)                      \
     FEATURE(EXT, DepthBiasControl, DEPTH_BIAS_CONTROL, depth_bias_control)                         \
-    FEATURE(EXT, DepthClampZeroOne, DEPTH_CLAMP_ZERO_ONE, depth_clamp_zero_one)                   \
     FEATURE(EXT, DepthClipControl, DEPTH_CLIP_CONTROL, depth_clip_control)                         \
-    FEATURE(EXT, DepthClipEnable, DEPTH_CLIP_ENABLE, depth_clip_enable)                            \
     FEATURE(EXT, ExtendedDynamicState, EXTENDED_DYNAMIC_STATE, extended_dynamic_state)             \
     FEATURE(EXT, ExtendedDynamicState2, EXTENDED_DYNAMIC_STATE_2, extended_dynamic_state2)         \
     FEATURE(EXT, ExtendedDynamicState3, EXTENDED_DYNAMIC_STATE_3, extended_dynamic_state3)         \
@@ -257,17 +255,6 @@ public:
         return allocator;
     }
 
-    struct MemoryBudgetInfo {
-        u64 usage;
-        u64 budget;
-        u64 block_bytes;
-        u64 allocation_bytes;
-    };
-
-    MemoryBudgetInfo GetMemoryBudgetInfo() const;
-
-    void TickAllocatorFrame() const;
-
     /// Returns the logical device.
     const vk::Device& GetLogical() const {
         return logical;
@@ -382,7 +369,8 @@ FN_MAX_LIMIT_LIST
     }
 
     bool IsOptimalAstcSupported() const {
-        return is_optimal_astc_supported;
+        return features.features.textureCompressionASTC_LDR &&
+               features.texture_compression_astc_hdr.textureCompressionASTC_HDR;
     }
 
     /// Returns true if BCn is natively supported.
@@ -614,16 +602,6 @@ FN_MAX_LIMIT_LIST
         return extensions.depth_clip_control;
     }
 
-    /// Returns true if the device supports VK_EXT_depth_clamp_zero_one.
-    bool IsExtDepthClampZeroOneSupported() const {
-        return extensions.depth_clamp_zero_one;
-    }
-
-    /// Returns true if the device supports VK_EXT_depth_clip_enable.
-    bool IsExtDepthClipEnableSupported() const {
-        return extensions.depth_clip_enable;
-    }
-
     /// Returns true if the device supports VK_EXT_depth_bias_control.
     bool IsExtDepthBiasControlSupported() const {
         return extensions.depth_bias_control;
@@ -756,38 +734,6 @@ FN_MAX_LIMIT_LIST
         return features.line_rasterization.stippledRectangularLines != VK_FALSE;
     }
 
-    VkLineRasterizationModeEXT GetLineRasterizationMode(bool wants_smooth) const {
-        if (wants_smooth && SupportsSmoothLines()) {
-            return VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT;
-        }
-        if (SupportsRectangularLines()) {
-            return VK_LINE_RASTERIZATION_MODE_RECTANGULAR_EXT;
-        }
-        return VK_LINE_RASTERIZATION_MODE_DEFAULT_EXT;
-    }
-
-    bool SupportsStippleForMode(VkLineRasterizationModeEXT mode) const {
-        switch (mode) {
-        case VK_LINE_RASTERIZATION_MODE_RECTANGULAR_SMOOTH_EXT:
-            return features.line_rasterization.stippledSmoothLines != VK_FALSE;
-        case VK_LINE_RASTERIZATION_MODE_BRESENHAM_EXT:
-            return features.line_rasterization.stippledBresenhamLines != VK_FALSE;
-        default:
-            return features.line_rasterization.stippledRectangularLines != VK_FALSE;
-        }
-    }
-
-    float ClampLineWidth(float width) const {
-        if (!features.features.wideLines) {
-            return 1.0f;
-        }
-        const auto& range = properties.properties.limits.lineWidthRange;
-        if (!(width >= range[0])) {
-            return range[0];
-        }
-        return width > range[1] ? range[1] : width;
-    }
-
     bool SupportsAlphaToOne() const {
         return features.features.alphaToOne != VK_FALSE;
     }
@@ -930,10 +876,6 @@ FN_MAX_LIMIT_LIST
 
     u64 GetDeviceMemoryUsage() const;
 
-    VkSampleCountFlags GetStorageImageSampleCounts() const {
-        return properties.properties.limits.storageImageSampleCounts;
-    }
-
     u32 GetSetsPerPool() const {
         return sets_per_pool;
     }
@@ -1118,7 +1060,6 @@ private:
 private:
     VkInstance instance;         ///< Vulkan instance.
     VmaAllocator allocator;      ///< VMA allocator.
-    mutable u32 allocator_frame_index{};
     vk::DeviceDispatch dld;      ///< Device function pointers.
     vk::PhysicalDevice physical; ///< Physical device.
     vk::Device logical;          ///< Logical device.

src/video_core/vulkan_common/vulkan_memory_allocator.cpp

@@ -30,6 +30,26 @@ namespace Vulkan {
 
 // Helpers translating MemoryUsage to flags/usage
 
+        [[maybe_unused]] VkMemoryPropertyFlags MemoryUsagePropertyFlags(MemoryUsage usage) {
+            switch (usage) {
+                case MemoryUsage::DeviceLocal:
+                    return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+                case MemoryUsage::Upload:
+                    return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
+                case MemoryUsage::Download:
+                    return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
+                           VK_MEMORY_PROPERTY_HOST_CACHED_BIT;
+                case MemoryUsage::Stream:
+                    return VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
+                           VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
+            }
+            ASSERT_MSG(false, "Invalid memory usage={}", usage);
+            return VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
+        }
+
         [[nodiscard]] VkMemoryPropertyFlags MemoryUsagePreferredVmaFlags(MemoryUsage usage) {
             if (usage == MemoryUsage::Download) {
                 return VK_MEMORY_PROPERTY_HOST_CACHED_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
@@ -66,11 +86,125 @@ namespace Vulkan {
         }
 
 
+// This avoids calling vkGetBufferMemoryRequirements* directly.
+        template<typename T>
+        static VkBuffer GetVkHandleFromBuffer(const T &buf) {
+            if constexpr (requires { static_cast<VkBuffer>(buf); }) {
+                return static_cast<VkBuffer>(buf);
+            } else if constexpr (requires {{ buf.GetHandle() } -> std::convertible_to<VkBuffer>; }) {
+                return buf.GetHandle();
+            } else if constexpr (requires {{ buf.Handle() } -> std::convertible_to<VkBuffer>; }) {
+                return buf.Handle();
+            } else if constexpr (requires {{ buf.vk_handle() } -> std::convertible_to<VkBuffer>; }) {
+                return buf.vk_handle();
+            } else {
+                static_assert(sizeof(T) == 0, "Cannot extract VkBuffer handle from vk::Buffer");
+                return VK_NULL_HANDLE;
+            }
+        }
+
     } // namespace
 
+//MemoryCommit is now VMA-backed
+    MemoryCommit::MemoryCommit(VmaAllocator alloc, VmaAllocation a,
+                               const VmaAllocationInfo &info) noexcept
+            : allocator{alloc}, allocation{a}, memory{info.deviceMemory},
+              offset{info.offset}, size{info.size}, mapped_ptr{info.pMappedData} {
+        // Log GPU memory allocation
+        if (GPU::Logging::IsActive() &&
+            Settings::values.gpu_log_memory_tracking.GetValue()) {
+            GPU::Logging::GPULogger::GetInstance().LogMemoryAllocation(
+                reinterpret_cast<uintptr_t>(memory),
+                static_cast<u64>(size),
+                0  // Memory property flags (not easily available from VMA)
+            );
+        }
+    }
+
+    MemoryCommit::~MemoryCommit() { Release(); }
+
+    MemoryCommit::MemoryCommit(MemoryCommit &&rhs) noexcept
+            : allocator{std::exchange(rhs.allocator, nullptr)},
+              allocation{std::exchange(rhs.allocation, nullptr)},
+              memory{std::exchange(rhs.memory, VK_NULL_HANDLE)},
+              offset{std::exchange(rhs.offset, 0)},
+              size{std::exchange(rhs.size, 0)},
+              mapped_ptr{std::exchange(rhs.mapped_ptr, nullptr)} {}
+
+    MemoryCommit &MemoryCommit::operator=(MemoryCommit &&rhs) noexcept {
+        if (this != &rhs) {
+            Release();
+            allocator = std::exchange(rhs.allocator, nullptr);
+            allocation = std::exchange(rhs.allocation, nullptr);
+            memory = std::exchange(rhs.memory, VK_NULL_HANDLE);
+            offset = std::exchange(rhs.offset, 0);
+            size = std::exchange(rhs.size, 0);
+            mapped_ptr = std::exchange(rhs.mapped_ptr, nullptr);
+        }
+        return *this;
+    }
+
+    std::span<u8> MemoryCommit::Map()
+    {
+        if (!allocation) return {};
+        if (!mapped_ptr) {
+            if (vmaMapMemory(allocator, allocation, &mapped_ptr) != VK_SUCCESS) return {};
+        }
+        const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
+                                                                    (std::numeric_limits<size_t>::max)()));
+        return std::span<u8>{static_cast<u8 *>(mapped_ptr), n};
+    }
+
+    std::span<const u8> MemoryCommit::Map() const
+    {
+        if (!allocation) return {};
+        if (!mapped_ptr) {
+            void *p = nullptr;
+            if (vmaMapMemory(allocator, allocation, &p) != VK_SUCCESS) return {};
+            const_cast<MemoryCommit *>(this)->mapped_ptr = p;
+        }
+        const size_t n = static_cast<size_t>(std::min<VkDeviceSize>(size,
+                                                                    (std::numeric_limits<size_t>::max)()));
+        return std::span<const u8>{static_cast<const u8 *>(mapped_ptr), n};
+    }
+
+    void MemoryCommit::Unmap()
+    {
+        if (allocation && mapped_ptr) {
+            vmaUnmapMemory(allocator, allocation);
+            mapped_ptr = nullptr;
+        }
+    }
+
+    void MemoryCommit::Release() {
+        if (allocation && allocator) {
+            // Log GPU memory deallocation
+            if (GPU::Logging::IsActive() &&
+                Settings::values.gpu_log_memory_tracking.GetValue() &&
+                memory != VK_NULL_HANDLE) {
+                GPU::Logging::GPULogger::GetInstance().LogMemoryDeallocation(
+                    reinterpret_cast<uintptr_t>(memory)
+                );
+            }
+
+            if (mapped_ptr) {
+                vmaUnmapMemory(allocator, allocation);
+                mapped_ptr = nullptr;
+            }
+            vmaFreeMemory(allocator, allocation);
+        }
+        allocation = nullptr;
+        allocator = nullptr;
+        memory = VK_NULL_HANDLE;
+        offset = 0;
+        size = 0;
+    }
+
     MemoryAllocator::MemoryAllocator(const Device &device_)
             : device{device_}, allocator{device.GetAllocator()},
-              properties{device_.GetPhysical().GetMemoryProperties().memoryProperties} {
+              properties{device_.GetPhysical().GetMemoryProperties().memoryProperties},
+              buffer_image_granularity{
+                      device_.GetPhysical().GetProperties().limits.bufferImageGranularity} {
 
         // Preserve the previous "RenderDoc small heap" trimming behavior that we had in original vma minus the heap bug
         if (device.HasDebuggingToolAttached())
@@ -90,21 +224,6 @@ namespace Vulkan {
 
     MemoryAllocator::~MemoryAllocator() = default;
 
-    void MemoryAllocator::SetReclaimCallback(ReclaimCallback callback) {
-        reclaim_callback = std::move(callback);
-        vk::SetAllocatorOwnerThread();
-    }
-
-    bool MemoryAllocator::ReclaimAtLeast(u64 hint_bytes) const {
-        if (!reclaim_callback || in_reclaim) {
-            return false;
-        }
-        in_reclaim = true;
-        const u64 freed = reclaim_callback(hint_bytes);
-        in_reclaim = false;
-        return freed > 0;
-    }
-
     vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const
     {
         const VmaAllocationCreateInfo alloc_ci = {
@@ -121,26 +240,7 @@ namespace Vulkan {
         VkImage handle{};
         VmaAllocation allocation{};
         VmaAllocationInfo alloc_info{};
-        DEBUG_ASSERT(vk::OnAllocatorOwnerThread());
-
-        VkResult res = vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info);
-
-        if (res != VK_SUCCESS && ReclaimAtLeast(IMAGE_RECLAIM_HINT)) {
-            res = vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info);
-        }
-
-        if (res != VK_SUCCESS) {
-            auto relaxed_ci = alloc_ci;
-            relaxed_ci.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
-            res = vmaCreateImage(allocator, &ci, &relaxed_ci, &handle, &allocation, &alloc_info);
-
-            if (res != VK_SUCCESS) {
-                relaxed_ci.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
-                res = vmaCreateImage(allocator, &ci, &relaxed_ci, &handle, &allocation, &alloc_info);
-            }
-        }
-
-        vk::Check(res);
+        vk::Check(vmaCreateImage(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
 
         // Log GPU memory allocation for images
         if (GPU::Logging::IsActive() &&
@@ -177,28 +277,7 @@ namespace Vulkan {
         VmaAllocation allocation{};
         VkMemoryPropertyFlags property_flags{};
 
-        DEBUG_ASSERT(vk::OnAllocatorOwnerThread());
-
-        VkResult res = vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info);
-
-        if (res != VK_SUCCESS && ReclaimAtLeast(ci.size)) {
-            res = vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info);
-        }
-
-        if (res != VK_SUCCESS) {
-            auto relaxed_ci = alloc_ci;
-            relaxed_ci.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
-            res = vmaCreateBuffer(allocator, &ci, &relaxed_ci, &handle, &allocation, &alloc_info);
-
-            if (res != VK_SUCCESS &&
-                (relaxed_ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
-                relaxed_ci.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
-                res = vmaCreateBuffer(allocator, &ci, &relaxed_ci, &handle, &allocation,
-                                      &alloc_info);
-            }
-        }
-
-        vk::Check(res);
+        vk::Check(vmaCreateBuffer(allocator, &ci, &alloc_ci, &handle, &allocation, &alloc_info));
         vmaGetAllocationMemoryProperties(allocator, allocation, &property_flags);
 
         // Log GPU memory allocation for buffers
@@ -220,4 +299,77 @@ namespace Vulkan {
                           device.GetDispatchLoader());
     }
 
+    MemoryCommit MemoryAllocator::Commit(const VkMemoryRequirements &reqs, MemoryUsage usage)
+    {
+        const auto vma_usage = MemoryUsageVma(usage);
+        VmaAllocationCreateInfo ci{};
+        ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
+        ci.usage = vma_usage;
+        ci.memoryTypeBits = reqs.memoryTypeBits & valid_memory_types;
+        ci.requiredFlags = 0;
+        ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
+
+        VmaAllocation a{};
+        VmaAllocationInfo info{};
+
+        VkResult res = vmaAllocateMemory(allocator, &reqs, &ci, &a, &info);
+
+        if (res != VK_SUCCESS) {
+            // Relax 1: drop budget constraint
+            auto ci2 = ci;
+            ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
+            res = vmaAllocateMemory(allocator, &reqs, &ci2, &a, &info);
+
+            // Relax 2: if we preferred DEVICE_LOCAL, drop that preference
+            if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
+                auto ci3 = ci2;
+                ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+                res = vmaAllocateMemory(allocator, &reqs, &ci3, &a, &info);
+            }
+        }
+
+        vk::Check(res);
+        return MemoryCommit(allocator, a, info);
+    }
+
+    MemoryCommit MemoryAllocator::Commit(const vk::Buffer &buffer, MemoryUsage usage) {
+        // Allocate memory appropriate for this buffer automatically
+        const auto vma_usage = MemoryUsageVma(usage);
+
+        VmaAllocationCreateInfo ci{};
+        ci.flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage);
+        ci.usage = vma_usage;
+        ci.requiredFlags = 0;
+        ci.preferredFlags = MemoryUsagePreferredVmaFlags(usage);
+        ci.pool = VK_NULL_HANDLE;
+        ci.pUserData = nullptr;
+        ci.priority = 0.0f;
+
+        const VkBuffer raw = *buffer;
+
+        VmaAllocation a{};
+        VmaAllocationInfo info{};
+
+        // Let VMA infer memory requirements from the buffer
+        VkResult res = vmaAllocateMemoryForBuffer(allocator, raw, &ci, &a, &info);
+
+        if (res != VK_SUCCESS) {
+            auto ci2 = ci;
+            ci2.flags &= ~VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT;
+            res = vmaAllocateMemoryForBuffer(allocator, raw, &ci2, &a, &info);
+
+            if (res != VK_SUCCESS && (ci.preferredFlags & VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT)) {
+                auto ci3 = ci2;
+                ci3.preferredFlags &= ~VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+                res = vmaAllocateMemoryForBuffer(allocator, raw, &ci3, &a, &info);
+            }
+        }
+
+        vk::Check(res);
+        vk::Check(vmaBindBufferMemory2(allocator, a, 0, raw, nullptr));
+        return MemoryCommit(allocator, a, info);
+    }
+
+
+
 } // namespace Vulkan

src/video_core/vulkan_common/vulkan_memory_allocator.h

@@ -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
 
 // SPDX-FileCopyrightText: Copyright 2019 yuzu Emulator Project
@@ -6,7 +6,6 @@
 
 #pragma once
 
-#include <functional>
 #include <memory>
 #include <span>
 #include <vector>
@@ -40,6 +39,51 @@ namespace Vulkan {
         }
     }
 
+/// Ownership handle of a memory commitment (real VMA allocation).
+    class MemoryCommit {
+    public:
+        MemoryCommit() noexcept = default;
+
+        MemoryCommit(VmaAllocator allocator, VmaAllocation allocation,
+                     const VmaAllocationInfo &info) noexcept;
+
+        ~MemoryCommit();
+
+        MemoryCommit(const MemoryCommit &) = delete;
+
+        MemoryCommit &operator=(const MemoryCommit &) = delete;
+
+        MemoryCommit(MemoryCommit &&) noexcept;
+
+        MemoryCommit &operator=(MemoryCommit &&) noexcept;
+
+        [[nodiscard]] std::span<u8> Map();
+
+        [[nodiscard]] std::span<const u8> Map() const;
+
+        void Unmap();
+
+        explicit operator bool() const noexcept { return allocation != nullptr; }
+
+        VkDeviceMemory Memory() const noexcept { return memory; }
+
+        VkDeviceSize Offset() const noexcept { return offset; }
+
+        VkDeviceSize Size() const noexcept { return size; }
+
+        VmaAllocation Allocation() const noexcept { return allocation; }
+
+    private:
+        void Release();
+
+        VmaAllocator allocator{};   ///< VMA allocator
+        VmaAllocation allocation{};  ///< VMA allocation handle
+        VkDeviceMemory memory{};      ///< Underlying VkDeviceMemory chosen by VMA
+        VkDeviceSize offset{};      ///< Offset of this allocation inside VkDeviceMemory
+        VkDeviceSize size{};        ///< Size of the allocation
+        void *mapped_ptr{};  ///< Optional persistent mapped pointer
+    };
+
 /// Memory allocator container.
 /// Allocates and releases memory allocations on demand.
     class MemoryAllocator {
@@ -63,21 +107,36 @@ namespace Vulkan {
 
         vk::Buffer CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage) const;
 
-        using ReclaimCallback = std::function<u64(u64)>;
+        /**
+         * Commits a memory with the specified requirements.
+         *
+         * @param requirements Requirements returned from a Vulkan call.
+         * @param usage        Indicates how the memory will be used.
+         *
+         * @returns A memory commit.
+         */
+        MemoryCommit Commit(const VkMemoryRequirements &requirements, MemoryUsage usage);
 
-        void SetReclaimCallback(ReclaimCallback callback);
+        /// Commits memory required by the buffer and binds it (for buffers created outside VMA).
+        MemoryCommit Commit(const vk::Buffer &buffer, MemoryUsage usage);
 
     private:
-        bool ReclaimAtLeast(u64 hint_bytes) const;
-
-        static constexpr u64 IMAGE_RECLAIM_HINT = 64ULL * 1024 * 1024;
+        static bool IsAutoUsage(VmaMemoryUsage u) noexcept {
+            switch (u) {
+                case VMA_MEMORY_USAGE_AUTO:
+                case VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE:
+                case VMA_MEMORY_USAGE_AUTO_PREFER_HOST:
+                    return true;
+                default:
+                    return false;
+            }
+        }
 
         const Device &device;                              ///< Device handle.
         VmaAllocator allocator;                           ///< VMA allocator.
         const VkPhysicalDeviceMemoryProperties properties; ///< Physical device memory properties.
+        VkDeviceSize buffer_image_granularity;            ///< Adjacent buffer/image granularity
         u32 valid_memory_types{~0u};
-        ReclaimCallback reclaim_callback;
-        mutable bool in_reclaim{false};
     };
 
 } // namespace Vulkan

src/video_core/vulkan_common/vulkan_wrapper.cpp

@@ -5,16 +5,11 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <algorithm>
-#include <atomic>
-#include <limits>
 #include <memory>
-#include <mutex>
 #include <optional>
-#include <thread>
 #include <utility>
 #include <vector>
 
-#include "common/assert.h"
 #include "common/common_types.h"
 #include "common/logging.h"
 #include "video_core/vulkan_common/vk_enum_string_helper.h"
@@ -25,60 +20,6 @@ namespace Vulkan::vk {
 
 namespace {
 
-std::thread::id allocator_owner_thread;
-
-template <typename HandleType>
-struct PendingRelease {
-    VmaAllocator allocator;
-    HandleType handle;
-    VmaAllocation allocation;
-    u64 timeline;
-};
-
-std::mutex deletion_mutex;
-std::atomic<u64> deletion_timeline{1};
-std::vector<PendingRelease<VkImage>> pending_images;
-std::vector<PendingRelease<VkBuffer>> pending_buffers;
-
-template <typename HandleType>
-void PushPendingRelease(std::vector<PendingRelease<HandleType>>& pending, VmaAllocator allocator,
-                        HandleType handle, VmaAllocation allocation) noexcept {
-    std::scoped_lock lock{deletion_mutex};
-    pending.push_back(PendingRelease<HandleType>{
-        .allocator = allocator,
-        .handle = handle,
-        .allocation = allocation,
-        .timeline = deletion_timeline.load(std::memory_order_acquire),
-    });
-}
-
-template <typename HandleType>
-void ExtractReleased(std::vector<PendingRelease<HandleType>>& pending,
-                     std::vector<PendingRelease<HandleType>>& released, u64 completed_value) {
-    const auto split = std::partition(pending.begin(), pending.end(),
-                                      [completed_value](const PendingRelease<HandleType>& entry) {
-                                          return entry.timeline > completed_value;
-                                      });
-    released.assign(split, pending.end());
-    pending.erase(split, pending.end());
-}
-
-void DrainDeletionQueue(u64 completed_value) noexcept {
-    std::vector<PendingRelease<VkImage>> images;
-    std::vector<PendingRelease<VkBuffer>> buffers;
-    {
-        std::scoped_lock lock{deletion_mutex};
-        ExtractReleased(pending_images, images, completed_value);
-        ExtractReleased(pending_buffers, buffers, completed_value);
-    }
-    for (const auto& entry : images) {
-        vmaDestroyImage(entry.allocator, entry.handle, entry.allocation);
-    }
-    for (const auto& entry : buffers) {
-        vmaDestroyBuffer(entry.allocator, entry.handle, entry.allocation);
-    }
-}
-
 template <typename Func>
 void SortPhysicalDevices(std::vector<VkPhysicalDevice>& devices, const InstanceDispatch& dld,
                          Func&& func) {
@@ -561,35 +502,13 @@ DebugReportCallback Instance::CreateDebugReportCallback(
     return DebugReportCallback(object, handle, *dld);
 }
 
-void SetAllocatorOwnerThread() {
-    allocator_owner_thread = std::this_thread::get_id();
-}
-
-bool OnAllocatorOwnerThread() noexcept {
-    return allocator_owner_thread == std::thread::id{} ||
-           allocator_owner_thread == std::this_thread::get_id();
-}
-
-void SetDeletionTimeline(u64 value) noexcept {
-    deletion_timeline.store(value, std::memory_order_release);
-}
-
-void TickDeletionQueue(u64 completed_value) noexcept {
-    DEBUG_ASSERT(OnAllocatorOwnerThread());
-    DrainDeletionQueue(completed_value);
-}
-
-void FlushDeletionQueue() noexcept {
-    DrainDeletionQueue((std::numeric_limits<u64>::max)());
-}
-
 void Image::SetObjectNameEXT(const char* name) const {
     SetObjectName(dld, owner, handle, VK_OBJECT_TYPE_IMAGE, name);
 }
 
 void Image::Release() const noexcept {
     if (handle) {
-        PushPendingRelease(pending_images, allocator, handle, allocation);
+        vmaDestroyImage(allocator, handle, allocation);
     }
 }
 
@@ -611,7 +530,7 @@ void Buffer::SetObjectNameEXT(const char* name) const {
 
 void Buffer::Release() const noexcept {
     if (handle) {
-        PushPendingRelease(pending_buffers, allocator, handle, allocation);
+        vmaDestroyBuffer(allocator, handle, allocation);
     }
 }
 

src/video_core/vulkan_common/vulkan_wrapper.h

@@ -131,16 +131,6 @@ private:
     VkResult result;
 };
 
-void SetAllocatorOwnerThread();
-
-[[nodiscard]] bool OnAllocatorOwnerThread() noexcept;
-
-void SetDeletionTimeline(u64 value) noexcept;
-
-void TickDeletionQueue(u64 completed_value) noexcept;
-
-void FlushDeletionQueue() noexcept;
-
 /// Throws a Vulkan exception if result is not success.
 inline void Check(VkResult result) {
     if (result != VK_SUCCESS) {