copyright: Update year and names of many .cpp and .h files

yuzu-copyright: Update year and names of many .cpp and .h files
Merge pull request #6844 from ameerj/vp9-empty-frame
2021-08-10 13:07:49 -05:00 · 2021-08-10 13:00:50 -05:00 · 2021-08-08 19:02:39 -04:00 · 2021-08-08 11:31:47 -07:00 · 2021-08-08 20:24:53 +02:00 · 2021-08-08 13:49:00 -04:00
243 changed files with 2537 additions and 1703 deletions
@@ -18,7 +18,8 @@ cmake .. \
      -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON \
      -DENABLE_QT_TRANSLATION=ON \
      -DUSE_DISCORD_PRESENCE=ON \
-      -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${ENABLE_COMPATIBILITY_REPORTING:-"OFF"}
+      -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${ENABLE_COMPATIBILITY_REPORTING:-"OFF"} \
+      -DYUZU_USE_BUNDLED_FFMPEG=ON

 make -j$(nproc)

@@ -25,7 +25,7 @@ option(YUZU_USE_BUNDLED_BOOST "Download bundled Boost" OFF)

 option(YUZU_USE_BUNDLED_LIBUSB "Compile bundled libusb" OFF)

-CMAKE_DEPENDENT_OPTION(YUZU_USE_BUNDLED_FFMPEG "Download/Build bundled FFmpeg" ON "WIN32" OFF)
+option(YUZU_USE_BUNDLED_FFMPEG "Download/Build bundled FFmpeg" "${WIN32}")

 option(YUZU_USE_QT_WEB_ENGINE "Use QtWebEngine for web applet implementation" OFF)

@@ -496,7 +496,7 @@ endif()
 # Ensure libusb is properly configured (based on dolphin libusb include)
 if(NOT APPLE AND NOT YUZU_USE_BUNDLED_LIBUSB)
    include(FindPkgConfig)
-    if (PKG_CONFIG_FOUND)
+    if (PKG_CONFIG_FOUND AND NOT CMAKE_SYSTEM_NAME MATCHES "DragonFly|FreeBSD")
        pkg_check_modules(LIBUSB QUIET libusb-1.0>=1.0.24)
    else()
        find_package(LibUSB)
@@ -583,8 +583,32 @@ if (YUZU_USE_BUNDLED_FFMPEG)
            "${FFmpeg_PREFIX};${FFmpeg_BUILD_DIR}"
            CACHE PATH "Path to FFmpeg headers" FORCE)

+        if (${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
+            Include(FindPkgConfig REQUIRED)
+            pkg_check_modules(LIBVA libva)
+        endif()
+        if(LIBVA_FOUND)
+            pkg_check_modules(LIBDRM libdrm REQUIRED)
+            find_package(X11 REQUIRED)
+            pkg_check_modules(LIBVA-DRM libva-drm REQUIRED)
+            pkg_check_modules(LIBVA-X11 libva-x11 REQUIRED)
+            set(FFmpeg_LIBVA_LIBRARIES
+                ${LIBDRM_LIBRARIES}
+                ${X11_LIBRARIES}
+                ${LIBVA-DRM_LIBRARIES}
+                ${LIBVA-X11_LIBRARIES}
+                ${LIBVA_LIBRARIES})
+            set(FFmpeg_HWACCEL_FLAGS
+                --enable-hwaccel=h264_vaapi
+                --enable-hwaccel=vp9_vaapi
+                --enable-libdrm)
+            message(STATUS "VA-API found")
+        else()
+            set(FFmpeg_HWACCEL_FLAGS --disable-vaapi)
+        endif()
+
        # `configure` parameters builds only exactly what yuzu needs from FFmpeg
-        # `--disable-{vaapi,vdpau}` is needed to avoid linking issues
+        # `--disable-vdpau` is needed to avoid linking issues
        add_custom_command(
            OUTPUT
                ${FFmpeg_MAKEFILE}
@@ -600,13 +624,16 @@ if (YUZU_USE_BUNDLED_FFMPEG)
                    --disable-network
                    --disable-postproc
                    --disable-swresample
-                    --disable-vaapi
                    --disable-vdpau
                    --enable-decoder=h264
                    --enable-decoder=vp9
+                    --cc="${CMAKE_C_COMPILER}"
+                    --cxx="${CMAKE_CXX_COMPILER}"
+                    ${FFmpeg_HWACCEL_FLAGS}
            WORKING_DIRECTORY
                ${FFmpeg_BUILD_DIR}
        )
+        unset(FFmpeg_HWACCEL_FLAGS)

        # Workaround for Ubuntu 18.04's older version of make not being able to call make as a child
        # with context of the jobserver. Also helps ninja users.
@@ -616,9 +643,10 @@ if (YUZU_USE_BUNDLED_FFMPEG)
            OUTPUT_VARIABLE
                SYSTEM_THREADS)

+        set(FFmpeg_BUILD_LIBRARIES ${FFmpeg_LIBRARIES})
        add_custom_command(
            OUTPUT
-                ${FFmpeg_LIBRARIES}
+                ${FFmpeg_BUILD_LIBRARIES}
            COMMAND
                make -j${SYSTEM_THREADS}
            WORKING_DIRECTORY
@@ -628,7 +656,12 @@ if (YUZU_USE_BUNDLED_FFMPEG)
        # ALL makes this custom target build every time
        # but it won't actually build if the DEPENDS parameter is up to date
        add_custom_target(ffmpeg-configure ALL DEPENDS ${FFmpeg_MAKEFILE})
-        add_custom_target(ffmpeg-build ALL DEPENDS ${FFmpeg_LIBRARIES} ffmpeg-configure)
+        add_custom_target(ffmpeg-build ALL DEPENDS ${FFmpeg_BUILD_LIBRARIES} ffmpeg-configure)
+        link_libraries(${FFmpeg_LIBVA_LIBRARIES})
+        set(FFmpeg_LIBRARIES ${FFmpeg_LIBVA_LIBRARIES} ${FFmpeg_BUILD_LIBRARIES}
+            CACHE PATH "Paths to FFmpeg libraries" FORCE)
+        unset(FFmpeg_BUILD_LIBRARIES)
+        unset(FFmpeg_LIBVA_LIBRARIES)

        if (FFmpeg_FOUND)
            message(STATUS "Found FFmpeg version ${FFmpeg_VERSION}")
@@ -35,7 +35,7 @@ It is written in C++ with portability in mind, and we actively maintain builds f

 The emulator is capable of running most commercial games at full speed, provided you meet the [necessary hardware requirements](https://yuzu-emu.org/help/quickstart/#hardware-requirements).

-For a full list of games yuzu support, please visit our [Compatibility page](https://yuzu-emu.org/game/) 
+For a full list of games yuzu support, please visit our [Compatibility page](https://yuzu-emu.org/game/)

 Check out our [website](https://yuzu-emu.org/) for the latest news on exciting features, monthly progress reports, and more!

@@ -43,7 +43,7 @@ Check out our [website](https://yuzu-emu.org/) for the latest news on exciting f

 Most of the development happens on GitHub. It's also where [our central repository](https://github.com/yuzu-emu/yuzu) is hosted. For development discussion, please join us on [Discord](https://discord.com/invite/u77vRWY).

-If you want to contribute, please take a look at the [Contributor's Guide](https://github.com/yuzu-emu/yuzu/wiki/Contributing) and [Developer Information](https://github.com/yuzu-emu/yuzu/wiki/Developer-Information). 
+If you want to contribute, please take a look at the [Contributor's Guide](https://github.com/yuzu-emu/yuzu/wiki/Contributing) and [Developer Information](https://github.com/yuzu-emu/yuzu/wiki/Developer-Information).
 You can also contact any of the developers on Discord in order to know about the current state of the emulator.

 If you want to contribute to the user interface translation project, please check out the [yuzu project on transifex](https://www.transifex.com/yuzu-emulator/yuzu). We centralize translation work there, and periodically upstream translations.
@@ -78,3 +78,5 @@ If you wish to support us a different way, please join our [Discord](https://dis
 ## License

 yuzu is licensed under the GPLv2 (or any later version). Refer to the [license.txt](https://github.com/yuzu-emu/yuzu/blob/master/license.txt) file.
+
+The [Skyline-Emulator Team](https://github.com/skyline-emu/skyline) is exempt from GPLv2 for the contributions from all these contributors [FernandoS27](https://github.com/FernandoS27), [lioncash](https://github.com/lioncash), [bunnei](https://github.com/bunnei), [ReinUsesLisp](https://github.com/ReinUsesLisp), [Morph1984](https://github.com/Morph1984), [ogniK5377](https://github.com/ogniK5377), [german77](https://github.com/german77), [ameerj](https://github.com/ameerj), [Kelebek1](https://github.com/Kelebek1) and [lat9nq](https://github.com/lat9nq). They may only use the code from these contributors under Mozilla Public License, version 2.0.
@@ -38,6 +38,26 @@ QPushButton#RendererStatusBarButton:!checked {
    color: #0066ff;
 }

+QPushButton#GPUStatusBarButton {
+    color: #656565;
+    border: 1px solid transparent;
+    background-color: transparent;
+    padding: 0px 3px 0px 3px;
+    text-align: center;
+}
+
+QPushButton#GPUStatusBarButton:hover {
+    border: 1px solid #76797C;
+}
+
+QPushButton#GPUStatusBarButton:checked {
+    color: #b06020;
+}
+
+QPushButton#GPUStatusBarButton:!checked {
+    color: #109010;
+}
+
 QPushButton#buttonRefreshDevices {
    min-width: 21px;
    min-height: 21px;
@@ -1283,6 +1283,27 @@ QPushButton#RendererStatusBarButton:!checked {
    color: #00ccdd;
 }

+QPushButton#GPUStatusBarButton {
+    min-width: 0px;
+    color: #656565;
+    border: 1px solid transparent;
+    background-color: transparent;
+    padding: 0px 3px 0px 3px;
+    text-align: center;
+}
+
+QPushButton#GPUStatusBarButton:hover {
+    border: 1px solid #76797C;
+}
+
+QPushButton#GPUStatusBarButton:checked {
+    color: #ff8040;
+}
+
+QPushButton#GPUStatusBarButton:!checked {
+    color: #40dd40;
+}
+
 QPushButton#buttonRefreshDevices {
    min-width: 23px;
    min-height: 23px;
@@ -2186,6 +2186,27 @@ QPushButton#RendererStatusBarButton:!checked {
  color: #00ccdd;
 }

+QPushButton#GPUStatusBarButton {
+  min-width: 0px;
+  color: #656565;
+  border: 1px solid transparent;
+  background-color: transparent;
+  padding: 0px 3px 0px 3px;
+  text-align: center;
+}
+
+QPushButton#GPUStatusBarButton:hover {
+  border: 1px solid #76797C;
+}
+
+QPushButton#GPUStatusBarButton:checked {
+  color: #ff8040;
+}
+
+QPushButton#GPUStatusBarButton:!checked {
+  color: #40dd40;
+}
+
 QPushButton#buttonRefreshDevices {
  min-width: 19px;
  min-height: 19px;
@@ -67,6 +67,8 @@ if (MINGW OR (${CMAKE_SYSTEM_NAME} MATCHES "Linux") OR APPLE)
            "${LIBUSB_MAKEFILE}"
        COMMAND
            env
+                CC="${CMAKE_C_COMPILER}"
+                CXX="${CMAKE_CXX_COMPILER}"
                CFLAGS="${LIBUSB_CFLAGS}"
            sh "${LIBUSB_CONFIGURE}"
                ${LIBUSB_CONFIGURE_ARGS}
@@ -169,7 +169,7 @@ endif()

 create_target_directory_groups(common)

-target_link_libraries(common PUBLIC ${Boost_LIBRARIES} fmt::fmt microprofile)
+target_link_libraries(common PUBLIC ${Boost_LIBRARIES} fmt::fmt microprofile Threads::Threads)
 target_link_libraries(common PRIVATE lz4::lz4 xbyak)
 if (MSVC)
  target_link_libraries(common PRIVATE zstd::zstd)
@@ -52,8 +52,12 @@ assert_noinline_call(const Fn& fn) {
 #define DEBUG_ASSERT(_a_) ASSERT(_a_)
 #define DEBUG_ASSERT_MSG(_a_, ...) ASSERT_MSG(_a_, __VA_ARGS__)
 #else // not debug
-#define DEBUG_ASSERT(_a_)
-#define DEBUG_ASSERT_MSG(_a_, _desc_, ...)
+#define DEBUG_ASSERT(_a_)                                                                          \
+    do {                                                                                           \
+    } while (0)
+#define DEBUG_ASSERT_MSG(_a_, _desc_, ...)                                                         \
+    do {                                                                                           \
+    } while (0)
 #endif

 #define UNIMPLEMENTED() ASSERT_MSG(false, "Unimplemented code!")
@@ -20,6 +20,10 @@ std::string ToUTF8String(std::u8string_view u8_string) {
    return std::string{u8_string.begin(), u8_string.end()};
 }

+std::string BufferToUTF8String(std::span<const u8> buffer) {
+    return std::string{buffer.begin(), std::ranges::find(buffer, u8{0})};
+}
+
 std::string PathToUTF8String(const std::filesystem::path& path) {
    return ToUTF8String(path.u8string());
 }
@@ -46,6 +46,17 @@ concept IsChar = std::same_as<T, char>;
 */
 [[nodiscard]] std::string ToUTF8String(std::u8string_view u8_string);

+/**
+ * Converts a buffer of bytes to a UTF8-encoded std::string.
+ * This converts from the start of the buffer until the first encountered null-terminator.
+ * If no null-terminator is found, this converts the entire buffer instead.
+ *
+ * @param buffer Buffer of bytes
+ *
+ * @returns UTF-8 encoded std::string.
+ */
+[[nodiscard]] std::string BufferToUTF8String(std::span<const u8> buffer);
+
 /**
 * Converts a filesystem path to a UTF-8 encoded std::string.
 *
@@ -61,7 +61,7 @@ template <typename ContiguousContainer>
    return out;
 }

-[[nodiscard]] constexpr std::array<u8, 16> AsArray(const char (&data)[17]) {
+[[nodiscard]] constexpr std::array<u8, 16> AsArray(const char (&data)[33]) {
    return HexStringToArray<16>(data);
 }

@@ -6,7 +6,7 @@
 #include <windows.h>
 #include "common/dynamic_library.h"

-#elif defined(__linux__) // ^^^ Windows ^^^ vvv Linux vvv
+#elif defined(__linux__) || defined(__FreeBSD__) // ^^^ Windows ^^^ vvv Linux vvv

 #ifndef _GNU_SOURCE
 #define _GNU_SOURCE
@@ -343,7 +343,7 @@ private:
    std::unordered_map<size_t, size_t> placeholder_host_pointers; ///< Placeholder backing offset
 };

-#elif defined(__linux__) // ^^^ Windows ^^^ vvv Linux vvv
+#elif defined(__linux__) || defined(__FreeBSD__) // ^^^ Windows ^^^ vvv Linux vvv

 class HostMemory::Impl {
 public:
@@ -357,7 +357,12 @@ public:
        });

        // Backing memory initialization
+#if defined(__FreeBSD__) && __FreeBSD__ < 13
+        // XXX Drop after FreeBSD 12.* reaches EOL on 2024-06-30
+        fd = shm_open(SHM_ANON, O_RDWR, 0600);
+#else
        fd = memfd_create("HostMemory", 0);
+#endif
        if (fd == -1) {
            LOG_CRITICAL(HW_Memory, "memfd_create failed: {}", strerror(errno));
            throw std::bad_alloc{};
@@ -48,8 +48,8 @@ void LogSettings() {
    log_setting("Core_UseMultiCore", values.use_multi_core.GetValue());
    log_setting("CPU_Accuracy", values.cpu_accuracy.GetValue());
    log_setting("Renderer_UseResolutionFactor", values.resolution_factor.GetValue());
-    log_setting("Renderer_UseFrameLimit", values.use_frame_limit.GetValue());
-    log_setting("Renderer_FrameLimit", values.frame_limit.GetValue());
+    log_setting("Renderer_UseSpeedLimit", values.use_speed_limit.GetValue());
+    log_setting("Renderer_SpeedLimit", values.speed_limit.GetValue());
    log_setting("Renderer_UseDiskShaderCache", values.use_disk_shader_cache.GetValue());
    log_setting("Renderer_GPUAccuracyLevel", values.gpu_accuracy.GetValue());
    log_setting("Renderer_UseAsynchronousGpuEmulation",
@@ -132,8 +132,8 @@ void RestoreGlobalState(bool is_powered_on) {
    values.vulkan_device.SetGlobal(true);
    values.aspect_ratio.SetGlobal(true);
    values.max_anisotropy.SetGlobal(true);
-    values.use_frame_limit.SetGlobal(true);
-    values.frame_limit.SetGlobal(true);
+    values.use_speed_limit.SetGlobal(true);
+    values.speed_limit.SetGlobal(true);
    values.use_disk_shader_cache.SetGlobal(true);
    values.gpu_accuracy.SetGlobal(true);
    values.use_asynchronous_gpu_emulation.SetGlobal(true);
@@ -42,6 +42,11 @@ enum class CPUAccuracy : u32 {
    Unsafe = 2,
 };

+enum class FullscreenMode : u32 {
+    Borderless = 0,
+    Exclusive = 1,
+};
+
 /** The BasicSetting class is a simple resource manager. It defines a label and default value
 * alongside the actual value of the setting for simpler and less-error prone use with frontend
 * configurations. Setting a default value and label is required, though subclasses may deviate from
@@ -314,6 +319,7 @@ struct Values {
    // Renderer
    Setting<RendererBackend> renderer_backend{RendererBackend::OpenGL, "backend"};
    BasicSetting<bool> renderer_debug{false, "debug"};
+    BasicSetting<bool> renderer_shader_feedback{false, "shader_feedback"};
    BasicSetting<bool> enable_nsight_aftermath{false, "nsight_aftermath"};
    BasicSetting<bool> disable_shader_loop_safety_checks{false,
                                                         "disable_shader_loop_safety_checks"};
@@ -322,23 +328,24 @@ struct Values {
    Setting<u16> resolution_factor{1, "resolution_factor"};
    // *nix platforms may have issues with the borderless windowed fullscreen mode.
    // Default to exclusive fullscreen on these platforms for now.
-    Setting<int> fullscreen_mode{
+    Setting<FullscreenMode> fullscreen_mode{
 #ifdef _WIN32
-        0,
+        FullscreenMode::Borderless,
 #else
-        1,
+        FullscreenMode::Exclusive,
 #endif
        "fullscreen_mode"};
    Setting<int> aspect_ratio{0, "aspect_ratio"};
    Setting<int> max_anisotropy{0, "max_anisotropy"};
-    Setting<bool> use_frame_limit{true, "use_frame_limit"};
-    Setting<u16> frame_limit{100, "frame_limit"};
+    Setting<bool> use_speed_limit{true, "use_speed_limit"};
+    Setting<u16> speed_limit{100, "speed_limit"};
    Setting<bool> use_disk_shader_cache{true, "use_disk_shader_cache"};
    Setting<GPUAccuracy> gpu_accuracy{GPUAccuracy::High, "gpu_accuracy"};
    Setting<bool> use_asynchronous_gpu_emulation{true, "use_asynchronous_gpu_emulation"};
    Setting<bool> use_nvdec_emulation{true, "use_nvdec_emulation"};
    Setting<bool> accelerate_astc{true, "accelerate_astc"};
    Setting<bool> use_vsync{true, "use_vsync"};
+    BasicSetting<u16> fps_cap{1000, "fps_cap"};
    BasicSetting<bool> disable_fps_limit{false, "disable_fps_limit"};
    Setting<ShaderBackend> shader_backend{ShaderBackend::GLASM, "shader_backend"};
    Setting<bool> use_asynchronous_shaders{false, "use_asynchronous_shaders"};
@@ -6,10 +6,64 @@

 #include <fmt/format.h>

+#include "common/assert.h"
 #include "common/uuid.h"

 namespace Common {

+namespace {
+
+bool IsHexDigit(char c) {
+    return (c >= '0' && c <= '9') || (c >= 'a' && c <= 'f') || (c >= 'A' && c <= 'F');
+}
+
+u8 HexCharToByte(char c) {
+    if (c >= '0' && c <= '9') {
+        return static_cast<u8>(c - '0');
+    }
+    if (c >= 'a' && c <= 'f') {
+        return static_cast<u8>(c - 'a' + 10);
+    }
+    if (c >= 'A' && c <= 'F') {
+        return static_cast<u8>(c - 'A' + 10);
+    }
+    ASSERT_MSG(false, "{} is not a hexadecimal digit!", c);
+    return u8{0};
+}
+
+} // Anonymous namespace
+
+u128 HexStringToU128(std::string_view hex_string) {
+    const size_t length = hex_string.length();
+
+    // Detect "0x" prefix.
+    const bool has_0x_prefix = length > 2 && hex_string[0] == '0' && hex_string[1] == 'x';
+    const size_t offset = has_0x_prefix ? 2 : 0;
+
+    // Check length.
+    if (length > 32 + offset) {
+        ASSERT_MSG(false, "hex_string has more than 32 hexadecimal characters!");
+        return INVALID_UUID;
+    }
+
+    u64 lo = 0;
+    u64 hi = 0;
+    for (size_t i = 0; i < length - offset; ++i) {
+        const char c = hex_string[length - 1 - i];
+        if (!IsHexDigit(c)) {
+            ASSERT_MSG(false, "{} is not a hexadecimal digit!", c);
+            return INVALID_UUID;
+        }
+        if (i < 16) {
+            lo |= u64{HexCharToByte(c)} << (i * 4);
+        }
+        if (i >= 16) {
+            hi |= u64{HexCharToByte(c)} << ((i - 16) * 4);
+        }
+    }
+    return u128{lo, hi};
+}
+
 UUID UUID::Generate() {
    std::random_device device;
    std::mt19937 gen(device());
@@ -18,7 +72,7 @@ UUID UUID::Generate() {
 }

 std::string UUID::Format() const {
-    return fmt::format("0x{:016X}{:016X}", uuid[1], uuid[0]);
+    return fmt::format("{:016x}{:016x}", uuid[1], uuid[0]);
 }

 std::string UUID::FormatSwitch() const {
@@ -5,6 +5,7 @@
 #pragma once

 #include <string>
+#include <string_view>

 #include "common/common_types.h"

@@ -12,12 +13,30 @@ namespace Common {

 constexpr u128 INVALID_UUID{{0, 0}};

+/**
+ * Converts a hex string to a 128-bit unsigned integer.
+ *
+ * The hex string can be formatted in lowercase or uppercase, with or without the "0x" prefix.
+ *
+ * This function will assert and return INVALID_UUID under the following conditions:
+ * - If the hex string is more than 32 characters long
+ * - If the hex string contains non-hexadecimal characters
+ *
+ * @param hex_string Hexadecimal string
+ *
+ * @returns A 128-bit unsigned integer if successfully converted, INVALID_UUID otherwise.
+ */
+[[nodiscard]] u128 HexStringToU128(std::string_view hex_string);
+
 struct UUID {
    // UUIDs which are 0 are considered invalid!
    u128 uuid;
    UUID() = default;
    constexpr explicit UUID(const u128& id) : uuid{id} {}
    constexpr explicit UUID(const u64 lo, const u64 hi) : uuid{{lo, hi}} {}
+    explicit UUID(std::string_view hex_string) {
+        uuid = HexStringToU128(hex_string);
+    }

    [[nodiscard]] constexpr explicit operator bool() const {
        return uuid != INVALID_UUID;
@@ -50,3 +69,14 @@ struct UUID {
 static_assert(sizeof(UUID) == 16, "UUID is an invalid size!");

 } // namespace Common
+
+namespace std {
+
+template <>
+struct hash<Common::UUID> {
+    size_t operator()(const Common::UUID& uuid) const noexcept {
+        return uuid.uuid[1] ^ uuid.uuid[0];
+    }
+};
+
+} // namespace std
@@ -517,6 +517,8 @@ add_library(core STATIC
    hle/service/psc/psc.h
    hle/service/ptm/psm.cpp
    hle/service/ptm/psm.h
+    hle/service/kernel_helpers.cpp
+    hle/service/kernel_helpers.h
    hle/service/service.cpp
    hle/service/service.h
    hle/service/set/set.cpp
@@ -411,7 +411,7 @@ struct System::Impl {
    std::string status_details = "";

    std::unique_ptr<Core::PerfStats> perf_stats;
-    Core::FrameLimiter frame_limiter;
+    Core::SpeedLimiter speed_limiter;

    bool is_multicore{};
    bool is_async_gpu{};
@@ -606,12 +606,12 @@ const Core::PerfStats& System::GetPerfStats() const {
    return *impl->perf_stats;
 }

-Core::FrameLimiter& System::FrameLimiter() {
-    return impl->frame_limiter;
+Core::SpeedLimiter& System::SpeedLimiter() {
+    return impl->speed_limiter;
 }

-const Core::FrameLimiter& System::FrameLimiter() const {
-    return impl->frame_limiter;
+const Core::SpeedLimiter& System::SpeedLimiter() const {
+    return impl->speed_limiter;
 }

 Loader::ResultStatus System::GetGameName(std::string& out) const {
@@ -94,7 +94,7 @@ class ARM_Interface;
 class CpuManager;
 class DeviceMemory;
 class ExclusiveMonitor;
-class FrameLimiter;
+class SpeedLimiter;
 class PerfStats;
 class Reporter;
 class TelemetrySession;
@@ -292,11 +292,11 @@ public:
    /// Provides a constant reference to the internal PerfStats instance.
    [[nodiscard]] const Core::PerfStats& GetPerfStats() const;

-    /// Provides a reference to the frame limiter;
-    [[nodiscard]] Core::FrameLimiter& FrameLimiter();
+    /// Provides a reference to the speed limiter;
+    [[nodiscard]] Core::SpeedLimiter& SpeedLimiter();

-    /// Provides a constant referent to the frame limiter
-    [[nodiscard]] const Core::FrameLimiter& FrameLimiter() const;
+    /// Provides a constant reference to the speed limiter
+    [[nodiscard]] const Core::SpeedLimiter& SpeedLimiter() const;

    /// Gets the name of the current game
    [[nodiscard]] Loader::ResultStatus GetGameName(std::string& out) const;
@@ -12,9 +12,9 @@ namespace HLE::ApiVersion {

 // Horizon OS version constants.

-constexpr u8 HOS_VERSION_MAJOR = 11;
-constexpr u8 HOS_VERSION_MINOR = 0;
-constexpr u8 HOS_VERSION_MICRO = 1;
+constexpr u8 HOS_VERSION_MAJOR = 12;
+constexpr u8 HOS_VERSION_MINOR = 1;
+constexpr u8 HOS_VERSION_MICRO = 0;

 // NintendoSDK version constants.

@@ -22,15 +22,15 @@ constexpr u8 SDK_REVISION_MAJOR = 1;
 constexpr u8 SDK_REVISION_MINOR = 0;

 constexpr char PLATFORM_STRING[] = "NX";
-constexpr char VERSION_HASH[] = "69103fcb2004dace877094c2f8c29e6113be5dbf";
-constexpr char DISPLAY_VERSION[] = "11.0.1";
-constexpr char DISPLAY_TITLE[] = "NintendoSDK Firmware for NX 11.0.1-1.0";
+constexpr char VERSION_HASH[] = "76b10c2dab7d3aa73fc162f8dff1655e6a21caf4";
+constexpr char DISPLAY_VERSION[] = "12.1.0";
+constexpr char DISPLAY_TITLE[] = "NintendoSDK Firmware for NX 12.1.0-1.0";

 // Atmosphere version constants.

 constexpr u8 ATMOSPHERE_RELEASE_VERSION_MAJOR = 0;
 constexpr u8 ATMOSPHERE_RELEASE_VERSION_MINOR = 19;
-constexpr u8 ATMOSPHERE_RELEASE_VERSION_MICRO = 4;
+constexpr u8 ATMOSPHERE_RELEASE_VERSION_MICRO = 5;

 constexpr u32 GetTargetFirmware() {
    return u32{HOS_VERSION_MAJOR} << 24 | u32{HOS_VERSION_MINOR} << 16 |
@@ -58,6 +58,9 @@ bool SessionRequestManager::HasSessionRequestHandler(const HLERequestContext& co

 void SessionRequestHandler::ClientConnected(KServerSession* session) {
    session->ClientConnected(shared_from_this());
+
+    // Ensure our server session is tracked globally.
+    kernel.RegisterServerSession(session);
 }

 void SessionRequestHandler::ClientDisconnected(KServerSession* session) {
@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.

 #include "core/hle/kernel/k_auto_object.h"
+#include "core/hle/kernel/kernel.h"

 namespace Kernel {

@@ -11,4 +12,12 @@ KAutoObject* KAutoObject::Create(KAutoObject* obj) {
    return obj;
 }

+void KAutoObject::RegisterWithKernel() {
+    kernel.RegisterKernelObject(this);
+}
+
+void KAutoObject::UnregisterWithKernel() {
+    kernel.UnregisterKernelObject(this);
+}
+
 } // namespace Kernel
@@ -85,8 +85,12 @@ private:
    KERNEL_AUTOOBJECT_TRAITS(KAutoObject, KAutoObject);

 public:
-    explicit KAutoObject(KernelCore& kernel_) : kernel(kernel_) {}
-    virtual ~KAutoObject() = default;
+    explicit KAutoObject(KernelCore& kernel_) : kernel(kernel_) {
+        RegisterWithKernel();
+    }
+    virtual ~KAutoObject() {
+        UnregisterWithKernel();
+    }

    static KAutoObject* Create(KAutoObject* ptr);

@@ -166,6 +170,10 @@ public:
        }
    }

+private:
+    void RegisterWithKernel();
+    void UnregisterWithKernel();
+
 protected:
    KernelCore& kernel;
    std::string name;
@@ -10,6 +10,7 @@
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "common/scope_exit.h"
 #include "common/settings.h"
 #include "core/core.h"
 #include "core/device_memory.h"
@@ -43,6 +44,8 @@ void SetupMainThread(Core::System& system, KProcess& owner_process, u32 priority
    ASSERT(owner_process.GetResourceLimit()->Reserve(LimitableResource::Threads, 1));

    KThread* thread = KThread::Create(system.Kernel());
+    SCOPE_EXIT({ thread->Close(); });
+
    ASSERT(KThread::InitializeUserThread(system, thread, entry_point, 0, stack_top, priority,
                                         owner_process.GetIdealCoreId(), &owner_process)
               .IsSuccess());
@@ -162,7 +165,7 @@ void KProcess::DecrementThreadCount() {
    ASSERT(num_threads > 0);

    if (const auto count = --num_threads; count == 0) {
-        UNIMPLEMENTED_MSG("Process termination is not implemented!");
+        LOG_WARNING(Kernel, "Process termination is not fully implemented.");
    }
 }

@@ -406,6 +409,9 @@ void KProcess::Finalize() {
        resource_limit->Close();
    }

+    // Finalize the handle table and close any open handles.
+    handle_table.Finalize();
+
    // Perform inherited finalization.
    KAutoObjectWithSlabHeapAndContainer<KProcess, KSynchronizationObject>::Finalize();
 }
@@ -28,7 +28,10 @@ namespace Kernel {

 KServerSession::KServerSession(KernelCore& kernel_) : KSynchronizationObject{kernel_} {}

-KServerSession::~KServerSession() {}
+KServerSession::~KServerSession() {
+    // Ensure that the global list tracking server sessions does not hold on to a reference.
+    kernel.UnregisterServerSession(this);
+}

 void KServerSession::Initialize(KSession* parent_session_, std::string&& name_,
                                std::shared_ptr<SessionRequestManager> manager_) {
@@ -61,6 +61,7 @@ struct KernelCore::Impl {
    void Initialize(KernelCore& kernel) {
        global_scheduler_context = std::make_unique<Kernel::GlobalSchedulerContext>(kernel);
        global_handle_table = std::make_unique<Kernel::KHandleTable>(kernel);
+        global_handle_table->Initialize(KHandleTable::MaxTableSize);

        is_phantom_mode_for_singlecore = false;

@@ -90,9 +91,39 @@ struct KernelCore::Impl {
    }

    void Shutdown() {
+        // Shutdown all processes.
+        if (current_process) {
+            current_process->Finalize();
+            current_process->Close();
+            current_process = nullptr;
+        }
        process_list.clear();

-        // Ensures all service threads gracefully shutdown
+        // Close all open server ports.
+        std::unordered_set<KServerPort*> server_ports_;
+        {
+            std::lock_guard lk(server_ports_lock);
+            server_ports_ = server_ports;
+            server_ports.clear();
+        }
+        for (auto* server_port : server_ports_) {
+            server_port->Close();
+        }
+        // Close all open server sessions.
+        std::unordered_set<KServerSession*> server_sessions_;
+        {
+            std::lock_guard lk(server_sessions_lock);
+            server_sessions_ = server_sessions;
+            server_sessions.clear();
+        }
+        for (auto* server_session : server_sessions_) {
+            server_session->Close();
+        }
+
+        // Ensure that the object list container is finalized and properly shutdown.
+        object_list_container.Finalize();
+
+        // Ensures all service threads gracefully shutdown.
        service_threads.clear();

        next_object_id = 0;
@@ -111,11 +142,7 @@ struct KernelCore::Impl {

        cores.clear();

-        if (current_process) {
-            current_process->Close();
-            current_process = nullptr;
-        }
-
+        global_handle_table->Finalize();
        global_handle_table.reset();

        preemption_event = nullptr;
@@ -142,6 +169,16 @@ struct KernelCore::Impl {

        // Next host thead ID to use, 0-3 IDs represent core threads, >3 represent others
        next_host_thread_id = Core::Hardware::NUM_CPU_CORES;
+
+        // Track kernel objects that were not freed on shutdown
+        {
+            std::lock_guard lk(registered_objects_lock);
+            if (registered_objects.size()) {
+                LOG_WARNING(Kernel, "{} kernel objects were dangling on shutdown!",
+                            registered_objects.size());
+                registered_objects.clear();
+            }
+        }
    }

    void InitializePhysicalCores() {
@@ -630,6 +667,21 @@ struct KernelCore::Impl {
            user_slab_heap_size);
    }

+    KClientPort* CreateNamedServicePort(std::string name) {
+        auto search = service_interface_factory.find(name);
+        if (search == service_interface_factory.end()) {
+            UNIMPLEMENTED();
+            return {};
+        }
+
+        KClientPort* port = &search->second(system.ServiceManager(), system);
+        {
+            std::lock_guard lk(server_ports_lock);
+            server_ports.insert(&port->GetParent()->GetServerPort());
+        }
+        return port;
+    }
+
    std::atomic<u32> next_object_id{0};
    std::atomic<u64> next_kernel_process_id{KProcess::InitialKIPIDMin};
    std::atomic<u64> next_user_process_id{KProcess::ProcessIDMin};
@@ -656,6 +708,12 @@ struct KernelCore::Impl {
    /// the ConnectToPort SVC.
    std::unordered_map<std::string, ServiceInterfaceFactory> service_interface_factory;
    NamedPortTable named_ports;
+    std::unordered_set<KServerPort*> server_ports;
+    std::unordered_set<KServerSession*> server_sessions;
+    std::unordered_set<KAutoObject*> registered_objects;
+    std::mutex server_ports_lock;
+    std::mutex server_sessions_lock;
+    std::mutex registered_objects_lock;

    std::unique_ptr<Core::ExclusiveMonitor> exclusive_monitor;
    std::vector<Kernel::PhysicalCore> cores;
@@ -844,12 +902,27 @@ void KernelCore::RegisterNamedService(std::string name, ServiceInterfaceFactory&
 }

 KClientPort* KernelCore::CreateNamedServicePort(std::string name) {
-    auto search = impl->service_interface_factory.find(name);
-    if (search == impl->service_interface_factory.end()) {
-        UNIMPLEMENTED();
-        return {};
-    }
-    return &search->second(impl->system.ServiceManager(), impl->system);
+    return impl->CreateNamedServicePort(std::move(name));
+}
+
+void KernelCore::RegisterServerSession(KServerSession* server_session) {
+    std::lock_guard lk(impl->server_sessions_lock);
+    impl->server_sessions.insert(server_session);
+}
+
+void KernelCore::UnregisterServerSession(KServerSession* server_session) {
+    std::lock_guard lk(impl->server_sessions_lock);
+    impl->server_sessions.erase(server_session);
+}
+
+void KernelCore::RegisterKernelObject(KAutoObject* object) {
+    std::lock_guard lk(impl->registered_objects_lock);
+    impl->registered_objects.insert(object);
+}
+
+void KernelCore::UnregisterKernelObject(KAutoObject* object) {
+    std::lock_guard lk(impl->registered_objects_lock);
+    impl->registered_objects.erase(object);
 }

 bool KernelCore::IsValidNamedPort(NamedPortTable::const_iterator port) const {
@@ -45,6 +45,7 @@ class KPort;
 class KProcess;
 class KResourceLimit;
 class KScheduler;
+class KServerSession;
 class KSession;
 class KSharedMemory;
 class KThread;
@@ -185,6 +186,22 @@ public:
    /// Opens a port to a service previously registered with RegisterNamedService.
    KClientPort* CreateNamedServicePort(std::string name);

+    /// Registers a server session with the gobal emulation state, to be freed on shutdown. This is
+    /// necessary because we do not emulate processes for HLE sessions.
+    void RegisterServerSession(KServerSession* server_session);
+
+    /// Unregisters a server session previously registered with RegisterServerSession when it was
+    /// destroyed during the current emulation session.
+    void UnregisterServerSession(KServerSession* server_session);
+
+    /// Registers all kernel objects with the global emulation state, this is purely for tracking
+    /// leaks after emulation has been shutdown.
+    void RegisterKernelObject(KAutoObject* object);
+
+    /// Unregisters a kernel object previously registered with RegisterKernelObject when it was
+    /// destroyed during the current emulation session.
+    void UnregisterKernelObject(KAutoObject* object);
+
    /// Determines whether or not the given port is a valid named port.
    bool IsValidNamedPort(NamedPortTable::const_iterator port) const;

@@ -298,6 +298,7 @@ static ResultCode ConnectToNamedPort(Core::System& system, Handle* out, VAddr po
    // Create a session.
    KClientSession* session{};
    R_TRY(port->CreateSession(std::addressof(session)));
+    port->Close();

    // Register the session in the table, close the extra reference.
    handle_table.Register(*out, session);
@@ -1439,11 +1440,6 @@ static void ExitProcess(Core::System& system) {
    LOG_INFO(Kernel_SVC, "Process {} exiting", current_process->GetProcessID());
    ASSERT_MSG(current_process->GetStatus() == ProcessStatus::Running,
               "Process has already exited");
-
-    current_process->PrepareForTermination();
-
-    // Kill the current thread
-    system.Kernel().CurrentScheduler()->GetCurrentThread()->Exit();
 }

 static void ExitProcess32(Core::System& system) {
@@ -292,7 +292,7 @@ public:

 protected:
    void Get(Kernel::HLERequestContext& ctx) {
-        LOG_DEBUG(Service_ACC, "called user_id={}", user_id.Format());
+        LOG_DEBUG(Service_ACC, "called user_id=0x{}", user_id.Format());
        ProfileBase profile_base{};
        ProfileData data{};
        if (profile_manager.GetProfileBaseAndData(user_id, profile_base, data)) {
@@ -301,7 +301,7 @@ protected:
            rb.Push(ResultSuccess);
            rb.PushRaw(profile_base);
        } else {
-            LOG_ERROR(Service_ACC, "Failed to get profile base and data for user={}",
+            LOG_ERROR(Service_ACC, "Failed to get profile base and data for user=0x{}",
                      user_id.Format());
            IPC::ResponseBuilder rb{ctx, 2};
            rb.Push(ResultUnknown); // TODO(ogniK): Get actual error code
@@ -309,14 +309,14 @@ protected:
    }

    void GetBase(Kernel::HLERequestContext& ctx) {
-        LOG_DEBUG(Service_ACC, "called user_id={}", user_id.Format());
+        LOG_DEBUG(Service_ACC, "called user_id=0x{}", user_id.Format());
        ProfileBase profile_base{};
        if (profile_manager.GetProfileBase(user_id, profile_base)) {
            IPC::ResponseBuilder rb{ctx, 16};
            rb.Push(ResultSuccess);
            rb.PushRaw(profile_base);
        } else {
-            LOG_ERROR(Service_ACC, "Failed to get profile base for user={}", user_id.Format());
+            LOG_ERROR(Service_ACC, "Failed to get profile base for user=0x{}", user_id.Format());
            IPC::ResponseBuilder rb{ctx, 2};
            rb.Push(ResultUnknown); // TODO(ogniK): Get actual error code
        }
@@ -372,7 +372,7 @@ protected:

        const auto user_data = ctx.ReadBuffer();

-        LOG_DEBUG(Service_ACC, "called, username='{}', timestamp={:016X}, uuid={}",
+        LOG_DEBUG(Service_ACC, "called, username='{}', timestamp={:016X}, uuid=0x{}",
                  Common::StringFromFixedZeroTerminatedBuffer(
                      reinterpret_cast<const char*>(base.username.data()), base.username.size()),
                  base.timestamp, base.user_uuid.Format());
@@ -405,7 +405,7 @@ protected:
        const auto user_data = ctx.ReadBuffer();
        const auto image_data = ctx.ReadBuffer(1);

-        LOG_DEBUG(Service_ACC, "called, username='{}', timestamp={:016X}, uuid={}",
+        LOG_DEBUG(Service_ACC, "called, username='{}', timestamp={:016X}, uuid=0x{}",
                  Common::StringFromFixedZeroTerminatedBuffer(
                      reinterpret_cast<const char*>(base.username.data()), base.username.size()),
                  base.timestamp, base.user_uuid.Format());
@@ -662,7 +662,7 @@ void Module::Interface::GetUserCount(Kernel::HLERequestContext& ctx) {
 void Module::Interface::GetUserExistence(Kernel::HLERequestContext& ctx) {
    IPC::RequestParser rp{ctx};
    Common::UUID user_id = rp.PopRaw<Common::UUID>();
-    LOG_DEBUG(Service_ACC, "called user_id={}", user_id.Format());
+    LOG_DEBUG(Service_ACC, "called user_id=0x{}", user_id.Format());

    IPC::ResponseBuilder rb{ctx, 3};
    rb.Push(ResultSuccess);
@@ -693,7 +693,7 @@ void Module::Interface::GetLastOpenedUser(Kernel::HLERequestContext& ctx) {
 void Module::Interface::GetProfile(Kernel::HLERequestContext& ctx) {
    IPC::RequestParser rp{ctx};
    Common::UUID user_id = rp.PopRaw<Common::UUID>();
-    LOG_DEBUG(Service_ACC, "called user_id={}", user_id.Format());
+    LOG_DEBUG(Service_ACC, "called user_id=0x{}", user_id.Format());

    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
    rb.Push(ResultSuccess);
@@ -802,7 +802,7 @@ void Module::Interface::GetProfileEditor(Kernel::HLERequestContext& ctx) {
    IPC::RequestParser rp{ctx};
    Common::UUID user_id = rp.PopRaw<Common::UUID>();

-    LOG_DEBUG(Service_ACC, "called, user_id={}", user_id.Format());
+    LOG_DEBUG(Service_ACC, "called, user_id=0x{}", user_id.Format());

    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
    rb.Push(ResultSuccess);
@@ -844,7 +844,7 @@ void Module::Interface::StoreSaveDataThumbnailApplication(Kernel::HLERequestCont
    IPC::RequestParser rp{ctx};
    const auto uuid = rp.PopRaw<Common::UUID>();

-    LOG_WARNING(Service_ACC, "(STUBBED) called, uuid={}", uuid.Format());
+    LOG_WARNING(Service_ACC, "(STUBBED) called, uuid=0x{}", uuid.Format());

    // TODO(ogniK): Check if application ID is zero on acc initialize. As we don't have a reliable
    // way of confirming things like the TID, we're going to assume a non zero value for the time
@@ -858,7 +858,7 @@ void Module::Interface::StoreSaveDataThumbnailSystem(Kernel::HLERequestContext&
    const auto uuid = rp.PopRaw<Common::UUID>();
    const auto tid = rp.Pop<u64_le>();

-    LOG_WARNING(Service_ACC, "(STUBBED) called, uuid={}, tid={:016X}", uuid.Format(), tid);
+    LOG_WARNING(Service_ACC, "(STUBBED) called, uuid=0x{}, tid={:016X}", uuid.Format(), tid);
    StoreSaveDataThumbnail(ctx, uuid, tid);
 }

@@ -377,7 +377,8 @@ void SoftwareKeyboard::SubmitForTextCheck(std::u16string submitted_text) {

    if (swkbd_config_common.use_utf8) {
        std::string utf8_submitted_text = Common::UTF16ToUTF8(current_text);
-        const u64 buffer_size = sizeof(u64) + utf8_submitted_text.size();
+        // Include the null terminator in the buffer size.
+        const u64 buffer_size = utf8_submitted_text.size() + 1;

        LOG_DEBUG(Service_AM, "\nBuffer Size: {}\nUTF-8 Submitted Text: {}", buffer_size,
                  utf8_submitted_text);
@@ -386,7 +387,8 @@ void SoftwareKeyboard::SubmitForTextCheck(std::u16string submitted_text) {
        std::memcpy(out_data.data() + sizeof(u64), utf8_submitted_text.data(),
                    utf8_submitted_text.size());
    } else {
-        const u64 buffer_size = sizeof(u64) + current_text.size() * sizeof(char16_t);
+        // Include the null terminator in the buffer size.
+        const u64 buffer_size = (current_text.size() + 1) * sizeof(char16_t);

        LOG_DEBUG(Service_AM, "\nBuffer Size: {}\nUTF-16 Submitted Text: {}", buffer_size,
                  Common::UTF16ToUTF8(current_text));
@@ -158,7 +158,7 @@ private:
        const auto local_play = rp.Pop<bool>();
        const auto uuid = rp.PopRaw<Common::UUID>();

-        LOG_WARNING(Service_Friend, "(STUBBED) called local_play={} uuid={}", local_play,
+        LOG_WARNING(Service_Friend, "(STUBBED) called, local_play={}, uuid=0x{}", local_play,
                    uuid.Format());

        IPC::ResponseBuilder rb{ctx, 2};
@@ -171,7 +171,7 @@ private:
        const auto uuid = rp.PopRaw<Common::UUID>();
        [[maybe_unused]] const auto filter = rp.PopRaw<SizedFriendFilter>();
        const auto pid = rp.Pop<u64>();
-        LOG_WARNING(Service_Friend, "(STUBBED) called, offset={}, uuid={}, pid={}", friend_offset,
+        LOG_WARNING(Service_Friend, "(STUBBED) called, offset={}, uuid=0x{}, pid={}", friend_offset,
                    uuid.Format(), pid);

        IPC::ResponseBuilder rb{ctx, 3};
@@ -289,7 +289,7 @@ void Module::Interface::CreateNotificationService(Kernel::HLERequestContext& ctx
    IPC::RequestParser rp{ctx};
    auto uuid = rp.PopRaw<Common::UUID>();

-    LOG_DEBUG(Service_Friend, "called, uuid={}", uuid.Format());
+    LOG_DEBUG(Service_Friend, "called, uuid=0x{}", uuid.Format());

    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
    rb.Push(ResultSuccess);
@@ -18,6 +18,7 @@
 #include "core/hle/kernel/k_writable_event.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/service/hid/controllers/npad.h"
+#include "core/hle/service/kernel_helpers.h"

 namespace Service::HID {
 constexpr s32 HID_JOYSTICK_MAX = 0x7fff;
@@ -147,7 +148,9 @@ bool Controller_NPad::IsDeviceHandleValid(const DeviceHandle& device_handle) {
           device_handle.device_index < DeviceIndex::MaxDeviceIndex;
 }

-Controller_NPad::Controller_NPad(Core::System& system_) : ControllerBase{system_} {
+Controller_NPad::Controller_NPad(Core::System& system_,
+                                 KernelHelpers::ServiceContext& service_context_)
+    : ControllerBase{system_}, service_context{service_context_} {
    latest_vibration_values.fill({DEFAULT_VIBRATION_VALUE, DEFAULT_VIBRATION_VALUE});
 }

@@ -251,10 +254,9 @@ void Controller_NPad::InitNewlyAddedController(std::size_t controller_idx) {
 }

 void Controller_NPad::OnInit() {
-    auto& kernel = system.Kernel();
    for (std::size_t i = 0; i < styleset_changed_events.size(); ++i) {
-        styleset_changed_events[i] = Kernel::KEvent::Create(kernel);
-        styleset_changed_events[i]->Initialize(fmt::format("npad:NpadStyleSetChanged_{}", i));
+        styleset_changed_events[i] =
+            service_context.CreateEvent(fmt::format("npad:NpadStyleSetChanged_{}", i));
    }

    if (!IsControllerActivated()) {
@@ -344,8 +346,7 @@ void Controller_NPad::OnRelease() {
    }

    for (std::size_t i = 0; i < styleset_changed_events.size(); ++i) {
-        styleset_changed_events[i]->Close();
-        styleset_changed_events[i] = nullptr;
+        service_context.CloseEvent(styleset_changed_events[i]);
    }
 }

@@ -20,6 +20,10 @@ class KEvent;
 class KReadableEvent;
 } // namespace Kernel

+namespace Service::KernelHelpers {
+class ServiceContext;
+}
+
 namespace Service::HID {

 constexpr u32 NPAD_HANDHELD = 32;
@@ -27,7 +31,8 @@ constexpr u32 NPAD_UNKNOWN = 16; // TODO(ogniK): What is this?

 class Controller_NPad final : public ControllerBase {
 public:
-    explicit Controller_NPad(Core::System& system_);
+    explicit Controller_NPad(Core::System& system_,
+                             KernelHelpers::ServiceContext& service_context_);
    ~Controller_NPad() override;

    // Called when the controller is initialized
@@ -566,6 +571,7 @@ private:
        std::array<std::unique_ptr<Input::MotionDevice>, Settings::NativeMotion::NUM_MOTIONS_HID>,
        10>;

+    KernelHelpers::ServiceContext& service_context;
    std::mutex mutex;
    ButtonArray buttons;
    StickArray sticks;
@@ -46,8 +46,9 @@ constexpr auto pad_update_ns = std::chrono::nanoseconds{1000 * 1000};         //
 constexpr auto motion_update_ns = std::chrono::nanoseconds{15 * 1000 * 1000}; // (15ms, 66.666Hz)
 constexpr std::size_t SHARED_MEMORY_SIZE = 0x40000;

-IAppletResource::IAppletResource(Core::System& system_)
-    : ServiceFramework{system_, "IAppletResource"} {
+IAppletResource::IAppletResource(Core::System& system_,
+                                 KernelHelpers::ServiceContext& service_context_)
+    : ServiceFramework{system_, "IAppletResource"}, service_context{service_context_} {
    static const FunctionInfo functions[] = {
        {0, &IAppletResource::GetSharedMemoryHandle, "GetSharedMemoryHandle"},
    };
@@ -63,7 +64,7 @@ IAppletResource::IAppletResource(Core::System& system_)
    MakeController<Controller_Stubbed>(HidController::CaptureButton);
    MakeController<Controller_Stubbed>(HidController::InputDetector);
    MakeController<Controller_Stubbed>(HidController::UniquePad);
-    MakeController<Controller_NPad>(HidController::NPad);
+    MakeControllerWithServiceContext<Controller_NPad>(HidController::NPad);
    MakeController<Controller_Gesture>(HidController::Gesture);
    MakeController<Controller_ConsoleSixAxis>(HidController::ConsoleSixAxisSensor);

@@ -191,13 +192,14 @@ private:

 std::shared_ptr<IAppletResource> Hid::GetAppletResource() {
    if (applet_resource == nullptr) {
-        applet_resource = std::make_shared<IAppletResource>(system);
+        applet_resource = std::make_shared<IAppletResource>(system, service_context);
    }

    return applet_resource;
 }

-Hid::Hid(Core::System& system_) : ServiceFramework{system_, "hid"} {
+Hid::Hid(Core::System& system_)
+    : ServiceFramework{system_, "hid"}, service_context{system_, service_name} {
    // clang-format off
    static const FunctionInfo functions[] = {
        {0, &Hid::CreateAppletResource, "CreateAppletResource"},
@@ -347,7 +349,7 @@ void Hid::CreateAppletResource(Kernel::HLERequestContext& ctx) {
    LOG_DEBUG(Service_HID, "called, applet_resource_user_id={}", applet_resource_user_id);

    if (applet_resource == nullptr) {
-        applet_resource = std::make_shared<IAppletResource>(system);
+        applet_resource = std::make_shared<IAppletResource>(system, service_context);
    }

    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
@@ -7,6 +7,7 @@
 #include <chrono>

 #include "core/hle/service/hid/controllers/controller_base.h"
+#include "core/hle/service/kernel_helpers.h"
 #include "core/hle/service/service.h"

 namespace Core::Timing {
@@ -39,7 +40,8 @@ enum class HidController : std::size_t {

 class IAppletResource final : public ServiceFramework<IAppletResource> {
 public:
-    explicit IAppletResource(Core::System& system_);
+    explicit IAppletResource(Core::System& system_,
+                             KernelHelpers::ServiceContext& service_context_);
    ~IAppletResource() override;

    void ActivateController(HidController controller);
@@ -60,11 +62,18 @@ private:
    void MakeController(HidController controller) {
        controllers[static_cast<std::size_t>(controller)] = std::make_unique<T>(system);
    }
+    template <typename T>
+    void MakeControllerWithServiceContext(HidController controller) {
+        controllers[static_cast<std::size_t>(controller)] =
+            std::make_unique<T>(system, service_context);
+    }

    void GetSharedMemoryHandle(Kernel::HLERequestContext& ctx);
    void UpdateControllers(std::uintptr_t user_data, std::chrono::nanoseconds ns_late);
    void UpdateMotion(std::uintptr_t user_data, std::chrono::nanoseconds ns_late);

+    KernelHelpers::ServiceContext& service_context;
+
    std::shared_ptr<Core::Timing::EventType> pad_update_event;
    std::shared_ptr<Core::Timing::EventType> motion_update_event;

@@ -176,6 +185,8 @@ private:
    static_assert(sizeof(VibrationDeviceInfo) == 0x8, "VibrationDeviceInfo has incorrect size.");

    std::shared_ptr<IAppletResource> applet_resource;
+
+    KernelHelpers::ServiceContext service_context;
 };

 /// Reload input devices. Used when input configuration changed
@@ -0,0 +1,62 @@
+// Copyright 2021 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "core/core.h"
+#include "core/hle/kernel/k_event.h"
+#include "core/hle/kernel/k_process.h"
+#include "core/hle/kernel/k_readable_event.h"
+#include "core/hle/kernel/k_resource_limit.h"
+#include "core/hle/kernel/k_scoped_resource_reservation.h"
+#include "core/hle/kernel/k_writable_event.h"
+#include "core/hle/service/kernel_helpers.h"
+
+namespace Service::KernelHelpers {
+
+ServiceContext::ServiceContext(Core::System& system_, std::string name_)
+    : kernel(system_.Kernel()) {
+    process = Kernel::KProcess::Create(kernel);
+    ASSERT(Kernel::KProcess::Initialize(process, system_, std::move(name_),
+                                        Kernel::KProcess::ProcessType::Userland)
+               .IsSuccess());
+}
+
+ServiceContext::~ServiceContext() {
+    process->Close();
+    process = nullptr;
+}
+
+Kernel::KEvent* ServiceContext::CreateEvent(std::string&& name) {
+    // Reserve a new event from the process resource limit
+    Kernel::KScopedResourceReservation event_reservation(process,
+                                                         Kernel::LimitableResource::Events);
+    if (!event_reservation.Succeeded()) {
+        LOG_CRITICAL(Service, "Resource limit reached!");
+        return {};
+    }
+
+    // Create a new event.
+    auto* event = Kernel::KEvent::Create(kernel);
+    if (!event) {
+        LOG_CRITICAL(Service, "Unable to create event!");
+        return {};
+    }
+
+    // Initialize the event.
+    event->Initialize(std::move(name));
+
+    // Commit the thread reservation.
+    event_reservation.Commit();
+
+    // Register the event.
+    Kernel::KEvent::Register(kernel, event);
+
+    return event;
+}
+
+void ServiceContext::CloseEvent(Kernel::KEvent* event) {
+    event->GetReadableEvent().Close();
+    event->GetWritableEvent().Close();
+}
+
+} // namespace Service::KernelHelpers
@@ -0,0 +1,35 @@
+// Copyright 2021 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <string>
+
+namespace Core {
+class System;
+}
+
+namespace Kernel {
+class KernelCore;
+class KEvent;
+class KProcess;
+} // namespace Kernel
+
+namespace Service::KernelHelpers {
+
+class ServiceContext {
+public:
+    ServiceContext(Core::System& system_, std::string name_);
+    ~ServiceContext();
+
+    Kernel::KEvent* CreateEvent(std::string&& name);
+
+    void CloseEvent(Kernel::KEvent* event);
+
+private:
+    Kernel::KernelCore& kernel;
+    Kernel::KProcess* process{};
+};
+
+} // namespace Service::KernelHelpers
@@ -339,13 +339,16 @@ std::optional<ApplicationLanguage> ConvertToApplicationLanguage(
    case Set::LanguageCode::FR_CA:
        return ApplicationLanguage::CanadianFrench;
    case Set::LanguageCode::PT:
+    case Set::LanguageCode::PT_BR:
        return ApplicationLanguage::Portuguese;
    case Set::LanguageCode::RU:
        return ApplicationLanguage::Russian;
    case Set::LanguageCode::KO:
        return ApplicationLanguage::Korean;
+    case Set::LanguageCode::ZH_TW:
    case Set::LanguageCode::ZH_HANT:
        return ApplicationLanguage::TraditionalChinese;
+    case Set::LanguageCode::ZH_CN:
    case Set::LanguageCode::ZH_HANS:
        return ApplicationLanguage::SimplifiedChinese;
    default:
@@ -1,42 +0,0 @@
-// Copyright 2019 yuzu emulator team
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-#include <optional>
-#include <string>
-#include "common/common_types.h"
-#include "core/hle/service/set/set.h"
-
-namespace Service::NS {
-/// This is nn::ns::detail::ApplicationLanguage
-enum class ApplicationLanguage : u8 {
-    AmericanEnglish = 0,
-    BritishEnglish,
-    Japanese,
-    French,
-    German,
-    LatinAmericanSpanish,
-    Spanish,
-    Italian,
-    Dutch,
-    CanadianFrench,
-    Portuguese,
-    Russian,
-    Korean,
-    TraditionalChinese,
-    SimplifiedChinese,
-    Count
-};
-using ApplicationLanguagePriorityList =
-    const std::array<ApplicationLanguage, static_cast<std::size_t>(ApplicationLanguage::Count)>;
-
-constexpr u32 GetSupportedLanguageFlag(const ApplicationLanguage lang) {
-    return 1U << static_cast<u32>(lang);
-}
-
-const ApplicationLanguagePriorityList* GetApplicationLanguagePriorityList(ApplicationLanguage lang);
-std::optional<ApplicationLanguage> ConvertToApplicationLanguage(
-    Service::Set::LanguageCode language_code);
-std::optional<Service::Set::LanguageCode> ConvertToLanguageCode(ApplicationLanguage lang);
-} // namespace Service::NS
@@ -54,7 +54,7 @@ void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u3

    system.GetPerfStats().EndSystemFrame();
    system.GPU().SwapBuffers(&framebuffer);
-    system.FrameLimiter().DoFrameLimiting(system.CoreTiming().GetGlobalTimeUs());
+    system.SpeedLimiter().DoSpeedLimiting(system.CoreTiming().GetGlobalTimeUs());
    system.GetPerfStats().BeginSystemFrame();
 }

@@ -166,8 +166,6 @@ NvResult nvhost_nvdec_common::MapBuffer(const std::vector<u8>& input, std::vecto
            LOG_ERROR(Service_NVDRV, "failed to map size={}", object->size);
        } else {
            cmd_buffer.map_address = object->dma_map_addr;
-            AddBufferMap(object->dma_map_addr, object->size, object->addr,
-                         object->status == nvmap::Object::Status::Allocated);
        }
    }
    std::memcpy(output.data(), &params, sizeof(IoctlMapBuffer));
@@ -178,30 +176,11 @@ NvResult nvhost_nvdec_common::MapBuffer(const std::vector<u8>& input, std::vecto
 }

 NvResult nvhost_nvdec_common::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output) {
-    IoctlMapBuffer params{};
-    std::memcpy(&params, input.data(), sizeof(IoctlMapBuffer));
-    std::vector<MapBufferEntry> cmd_buffer_handles(params.num_entries);
-    SliceVectors(input, cmd_buffer_handles, params.num_entries, sizeof(IoctlMapBuffer));
-
-    auto& gpu = system.GPU();
-
-    for (auto& cmd_buffer : cmd_buffer_handles) {
-        const auto object{nvmap_dev->GetObject(cmd_buffer.map_handle)};
-        if (!object) {
-            LOG_ERROR(Service_NVDRV, "invalid cmd_buffer nvmap_handle={:X}", cmd_buffer.map_handle);
-            std::memcpy(output.data(), &params, output.size());
-            return NvResult::InvalidState;
-        }
-        if (const auto size{RemoveBufferMap(object->dma_map_addr)}; size) {
-            gpu.MemoryManager().Unmap(object->dma_map_addr, *size);
-        } else {
-            // This occurs quite frequently, however does not seem to impact functionality
-            LOG_DEBUG(Service_NVDRV, "invalid offset=0x{:X} dma=0x{:X}", object->addr,
-                      object->dma_map_addr);
-        }
-        object->dma_map_addr = 0;
-    }
+    // This is intntionally stubbed.
+    // Skip unmapping buffers here, as to not break the continuity of the VP9 reference frame
+    // addresses, and risk invalidating data before the async GPU thread is done with it
    std::memset(output.data(), 0, output.size());
+    LOG_DEBUG(Service_NVDRV, "(STUBBED) called");
    return NvResult::Success;
 }

@@ -212,33 +191,4 @@ NvResult nvhost_nvdec_common::SetSubmitTimeout(const std::vector<u8>& input,
    return NvResult::Success;
 }

-std::optional<nvhost_nvdec_common::BufferMap> nvhost_nvdec_common::FindBufferMap(
-    GPUVAddr gpu_addr) const {
-    const auto it = std::find_if(
-        buffer_mappings.begin(), buffer_mappings.upper_bound(gpu_addr), [&](const auto& entry) {
-            return (gpu_addr >= entry.second.StartAddr() && gpu_addr < entry.second.EndAddr());
-        });
-
-    ASSERT(it != buffer_mappings.end());
-    return it->second;
-}
-
-void nvhost_nvdec_common::AddBufferMap(GPUVAddr gpu_addr, std::size_t size, VAddr cpu_addr,
-                                       bool is_allocated) {
-    buffer_mappings.insert_or_assign(gpu_addr, BufferMap{gpu_addr, size, cpu_addr, is_allocated});
-}
-
-std::optional<std::size_t> nvhost_nvdec_common::RemoveBufferMap(GPUVAddr gpu_addr) {
-    const auto iter{buffer_mappings.find(gpu_addr)};
-    if (iter == buffer_mappings.end()) {
-        return std::nullopt;
-    }
-    std::size_t size = 0;
-    if (iter->second.IsAllocated()) {
-        size = iter->second.Size();
-    }
-    buffer_mappings.erase(iter);
-    return size;
-}
-
 } // namespace Service::Nvidia::Devices
@@ -23,45 +23,6 @@ public:
    ~nvhost_nvdec_common() override;

 protected:
-    class BufferMap final {
-    public:
-        constexpr BufferMap() = default;
-
-        constexpr BufferMap(GPUVAddr start_addr_, std::size_t size_)
-            : start_addr{start_addr_}, end_addr{start_addr_ + size_} {}
-
-        constexpr BufferMap(GPUVAddr start_addr_, std::size_t size_, VAddr cpu_addr_,
-                            bool is_allocated_)
-            : start_addr{start_addr_}, end_addr{start_addr_ + size_}, cpu_addr{cpu_addr_},
-              is_allocated{is_allocated_} {}
-
-        constexpr VAddr StartAddr() const {
-            return start_addr;
-        }
-
-        constexpr VAddr EndAddr() const {
-            return end_addr;
-        }
-
-        constexpr std::size_t Size() const {
-            return end_addr - start_addr;
-        }
-
-        constexpr VAddr CpuAddr() const {
-            return cpu_addr;
-        }
-
-        constexpr bool IsAllocated() const {
-            return is_allocated;
-        }
-
-    private:
-        GPUVAddr start_addr{};
-        GPUVAddr end_addr{};
-        VAddr cpu_addr{};
-        bool is_allocated{};
-    };
-
    struct IoctlSetNvmapFD {
        s32_le nvmap_fd{};
    };
@@ -154,17 +115,11 @@ protected:
    NvResult UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& output);
    NvResult SetSubmitTimeout(const std::vector<u8>& input, std::vector<u8>& output);

-    std::optional<BufferMap> FindBufferMap(GPUVAddr gpu_addr) const;
-    void AddBufferMap(GPUVAddr gpu_addr, std::size_t size, VAddr cpu_addr, bool is_allocated);
-    std::optional<std::size_t> RemoveBufferMap(GPUVAddr gpu_addr);
-
    s32_le nvmap_fd{};
    u32_le submit_timeout{};
    std::shared_ptr<nvmap> nvmap_dev;
    SyncpointManager& syncpoint_manager;
    std::array<u32, MaxSyncPoints> device_syncpoints{};
-    // This is expected to be ordered, therefore we must use a map, not unordered_map
-    std::map<GPUVAddr, BufferMap> buffer_mappings;
 };
 }; // namespace Devices
 } // namespace Service::Nvidia
@@ -39,11 +39,11 @@ void InstallInterfaces(SM::ServiceManager& service_manager, NVFlinger::NVFlinger
    nvflinger.SetNVDrvInstance(module_);
 }

-Module::Module(Core::System& system) : syncpoint_manager{system.GPU()} {
-    auto& kernel = system.Kernel();
+Module::Module(Core::System& system)
+    : syncpoint_manager{system.GPU()}, service_context{system, "nvdrv"} {
    for (u32 i = 0; i < MaxNvEvents; i++) {
-        events_interface.events[i].event = Kernel::KEvent::Create(kernel);
-        events_interface.events[i].event->Initialize(fmt::format("NVDRV::NvEvent_{}", i));
+        events_interface.events[i].event =
+            service_context.CreateEvent(fmt::format("NVDRV::NvEvent_{}", i));
        events_interface.status[i] = EventState::Free;
        events_interface.registered[i] = false;
    }
@@ -65,8 +65,7 @@ Module::Module(Core::System& system) : syncpoint_manager{system.GPU()} {

 Module::~Module() {
    for (u32 i = 0; i < MaxNvEvents; i++) {
-        events_interface.events[i].event->Close();
-        events_interface.events[i].event = nullptr;
+        service_context.CloseEvent(events_interface.events[i].event);
    }
 }

@@ -9,6 +9,7 @@
 #include <vector>

 #include "common/common_types.h"
+#include "core/hle/service/kernel_helpers.h"
 #include "core/hle/service/nvdrv/nvdata.h"
 #include "core/hle/service/nvdrv/syncpoint_manager.h"
 #include "core/hle/service/service.h"
@@ -154,6 +155,8 @@ private:
    std::unordered_map<std::string, std::shared_ptr<Devices::nvdevice>> devices;

    EventInterface events_interface;
+
+    KernelHelpers::ServiceContext service_context;
 };

 /// Registers all NVDRV services with the specified service manager.
@@ -307,11 +307,12 @@ void NVFlinger::Compose() {
 }

 s64 NVFlinger::GetNextTicks() const {
-    if (Settings::values.disable_fps_limit.GetValue()) {
-        return 0;
-    }
-    constexpr s64 max_hertz = 120LL;
-    return (1000000000 * (1LL << swap_interval)) / max_hertz;
+    static constexpr s64 max_hertz = 120LL;
+
+    const auto& settings = Settings::values;
+    const bool unlocked_fps = settings.disable_fps_limit.GetValue();
+    const s64 fps_cap = unlocked_fps ? static_cast<s64>(settings.fps_cap.GetValue()) : 1;
+    return (1000000000 * (1LL << swap_interval)) / (max_hertz * fps_cap);
 }

 } // namespace Service::NVFlinger
@@ -104,23 +104,22 @@ ServiceFrameworkBase::~ServiceFrameworkBase() {
 void ServiceFrameworkBase::InstallAsService(SM::ServiceManager& service_manager) {
    const auto guard = LockService();

-    ASSERT(!port_installed);
+    ASSERT(!service_registered);

-    auto port = service_manager.RegisterService(service_name, max_sessions).Unwrap();
-    port->SetSessionHandler(shared_from_this());
-    port_installed = true;
+    service_manager.RegisterService(service_name, max_sessions, shared_from_this());
+    service_registered = true;
 }

 Kernel::KClientPort& ServiceFrameworkBase::CreatePort() {
    const auto guard = LockService();

-    ASSERT(!port_installed);
+    ASSERT(!service_registered);

    auto* port = Kernel::KPort::Create(kernel);
    port->Initialize(max_sessions, false, service_name);
    port->GetServerPort().SetSessionHandler(shared_from_this());

-    port_installed = true;
+    service_registered = true;

    return port->GetClientPort();
 }
@@ -96,6 +96,9 @@ protected:
    /// System context that the service operates under.
    Core::System& system;

+    /// Identifier string used to connect to the service.
+    std::string service_name;
+
 private:
    template <typename T>
    friend class ServiceFramework;
@@ -117,14 +120,12 @@ private:
    void RegisterHandlersBaseTipc(const FunctionInfoBase* functions, std::size_t n);
    void ReportUnimplementedFunction(Kernel::HLERequestContext& ctx, const FunctionInfoBase* info);

-    /// Identifier string used to connect to the service.
-    std::string service_name;
    /// Maximum number of concurrent sessions that this service can handle.
    u32 max_sessions;

    /// Flag to store if a port was already create/installed to detect multiple install attempts,
    /// which is not supported.
-    bool port_installed = false;
+    bool service_registered = false;

    /// Function used to safely up-cast pointers to the derived class before invoking a handler.
    InvokerFn* handler_invoker;
@@ -12,7 +12,7 @@

 namespace Service::Set {
 namespace {
-constexpr std::array<LanguageCode, 17> available_language_codes = {{
+constexpr std::array<LanguageCode, 18> available_language_codes = {{
    LanguageCode::JA,
    LanguageCode::EN_US,
    LanguageCode::FR,
@@ -30,6 +30,7 @@ constexpr std::array<LanguageCode, 17> available_language_codes = {{
    LanguageCode::ES_419,
    LanguageCode::ZH_HANS,
    LanguageCode::ZH_HANT,
+    LanguageCode::PT_BR,
 }};

 enum class KeyboardLayout : u64 {
@@ -50,7 +51,7 @@ enum class KeyboardLayout : u64 {
    ChineseTraditional = 14,
 };

-constexpr std::array<std::pair<LanguageCode, KeyboardLayout>, 17> language_to_layout{{
+constexpr std::array<std::pair<LanguageCode, KeyboardLayout>, 18> language_to_layout{{
    {LanguageCode::JA, KeyboardLayout::Japanese},
    {LanguageCode::EN_US, KeyboardLayout::EnglishUs},
    {LanguageCode::FR, KeyboardLayout::French},
@@ -68,10 +69,11 @@ constexpr std::array<std::pair<LanguageCode, KeyboardLayout>, 17> language_to_la
    {LanguageCode::ES_419, KeyboardLayout::SpanishLatin},
    {LanguageCode::ZH_HANS, KeyboardLayout::ChineseSimplified},
    {LanguageCode::ZH_HANT, KeyboardLayout::ChineseTraditional},
+    {LanguageCode::PT_BR, KeyboardLayout::Portuguese},
 }};

-constexpr std::size_t pre4_0_0_max_entries = 15;
-constexpr std::size_t post4_0_0_max_entries = 17;
+constexpr std::size_t PRE_4_0_0_MAX_ENTRIES = 0xF;
+constexpr std::size_t POST_4_0_0_MAX_ENTRIES = 0x40;

 constexpr ResultCode ERR_INVALID_LANGUAGE{ErrorModule::Settings, 625};

@@ -81,9 +83,10 @@ void PushResponseLanguageCode(Kernel::HLERequestContext& ctx, std::size_t num_la
    rb.Push(static_cast<u32>(num_language_codes));
 }

-void GetAvailableLanguageCodesImpl(Kernel::HLERequestContext& ctx, std::size_t max_size) {
+void GetAvailableLanguageCodesImpl(Kernel::HLERequestContext& ctx, std::size_t max_entries) {
    const std::size_t requested_amount = ctx.GetWriteBufferSize() / sizeof(LanguageCode);
-    const std::size_t copy_amount = std::min(requested_amount, max_size);
+    const std::size_t max_amount = std::min(requested_amount, max_entries);
+    const std::size_t copy_amount = std::min(available_language_codes.size(), max_amount);
    const std::size_t copy_size = copy_amount * sizeof(LanguageCode);

    ctx.WriteBuffer(available_language_codes.data(), copy_size);
@@ -118,7 +121,7 @@ LanguageCode GetLanguageCodeFromIndex(std::size_t index) {
 void SET::GetAvailableLanguageCodes(Kernel::HLERequestContext& ctx) {
    LOG_DEBUG(Service_SET, "called");

-    GetAvailableLanguageCodesImpl(ctx, pre4_0_0_max_entries);
+    GetAvailableLanguageCodesImpl(ctx, PRE_4_0_0_MAX_ENTRIES);
 }

 void SET::MakeLanguageCode(Kernel::HLERequestContext& ctx) {
@@ -140,19 +143,19 @@ void SET::MakeLanguageCode(Kernel::HLERequestContext& ctx) {
 void SET::GetAvailableLanguageCodes2(Kernel::HLERequestContext& ctx) {
    LOG_DEBUG(Service_SET, "called");

-    GetAvailableLanguageCodesImpl(ctx, post4_0_0_max_entries);
+    GetAvailableLanguageCodesImpl(ctx, POST_4_0_0_MAX_ENTRIES);
 }

 void SET::GetAvailableLanguageCodeCount(Kernel::HLERequestContext& ctx) {
    LOG_DEBUG(Service_SET, "called");

-    PushResponseLanguageCode(ctx, pre4_0_0_max_entries);
+    PushResponseLanguageCode(ctx, PRE_4_0_0_MAX_ENTRIES);
 }

 void SET::GetAvailableLanguageCodeCount2(Kernel::HLERequestContext& ctx) {
    LOG_DEBUG(Service_SET, "called");

-    PushResponseLanguageCode(ctx, post4_0_0_max_entries);
+    PushResponseLanguageCode(ctx, POST_4_0_0_MAX_ENTRIES);
 }

 void SET::GetQuestFlag(Kernel::HLERequestContext& ctx) {
@@ -31,6 +31,7 @@ enum class LanguageCode : u64 {
    ES_419 = 0x00003931342D7365,
    ZH_HANS = 0x00736E61482D687A,
    ZH_HANT = 0x00746E61482D687A,
+    PT_BR = 0x00000052422D7470,
 };
 LanguageCode GetLanguageCodeFromIndex(std::size_t idx);

@@ -4,6 +4,7 @@

 #include <tuple>
 #include "common/assert.h"
+#include "common/scope_exit.h"
 #include "core/core.h"
 #include "core/hle/ipc_helpers.h"
 #include "core/hle/kernel/k_client_port.h"
@@ -40,17 +41,13 @@ static ResultCode ValidateServiceName(const std::string& name) {
 }

 Kernel::KClientPort& ServiceManager::InterfaceFactory(ServiceManager& self, Core::System& system) {
-    ASSERT(self.sm_interface.expired());
-
-    auto sm = std::make_shared<SM>(self, system);
-    self.sm_interface = sm;
+    self.sm_interface = std::make_shared<SM>(self, system);
    self.controller_interface = std::make_unique<Controller>(system);
-
-    return sm->CreatePort();
+    return self.sm_interface->CreatePort();
 }

-ResultVal<Kernel::KServerPort*> ServiceManager::RegisterService(std::string name,
-                                                                u32 max_sessions) {
+ResultCode ServiceManager::RegisterService(std::string name, u32 max_sessions,
+                                           Kernel::SessionRequestHandlerPtr handler) {

    CASCADE_CODE(ValidateServiceName(name));

@@ -59,12 +56,9 @@ ResultVal<Kernel::KServerPort*> ServiceManager::RegisterService(std::string name
        return ERR_ALREADY_REGISTERED;
    }

-    auto* port = Kernel::KPort::Create(kernel);
-    port->Initialize(max_sessions, false, name);
+    registered_services.emplace(std::move(name), handler);

-    registered_services.emplace(std::move(name), port);
-
-    return MakeResult(&port->GetServerPort());
+    return ResultSuccess;
 }

 ResultCode ServiceManager::UnregisterService(const std::string& name) {
@@ -76,14 +70,11 @@ ResultCode ServiceManager::UnregisterService(const std::string& name) {
        return ERR_SERVICE_NOT_REGISTERED;
    }

-    iter->second->Close();
-
    registered_services.erase(iter);
    return ResultSuccess;
 }

 ResultVal<Kernel::KPort*> ServiceManager::GetServicePort(const std::string& name) {
-
    CASCADE_CODE(ValidateServiceName(name));
    auto it = registered_services.find(name);
    if (it == registered_services.end()) {
@@ -91,10 +82,13 @@ ResultVal<Kernel::KPort*> ServiceManager::GetServicePort(const std::string& name
        return ERR_SERVICE_NOT_REGISTERED;
    }

-    return MakeResult(it->second);
-}
+    auto* port = Kernel::KPort::Create(kernel);
+    port->Initialize(ServerSessionCountMax, false, name);
+    auto handler = it->second;
+    port->GetServerPort().SetSessionHandler(std::move(handler));

-SM::~SM() = default;
+    return MakeResult(port);
+}

 /**
 * SM::Initialize service function
@@ -156,11 +150,15 @@ ResultVal<Kernel::KClientSession*> SM::GetServiceImpl(Kernel::HLERequestContext&
        LOG_ERROR(Service_SM, "called service={} -> error 0x{:08X}", name, port_result.Code().raw);
        return port_result.Code();
    }
-    auto& port = port_result.Unwrap()->GetClientPort();
+    auto& port = port_result.Unwrap();
+    SCOPE_EXIT({ port->GetClientPort().Close(); });
+
+    server_ports.emplace_back(&port->GetServerPort());

    // Create a new session.
    Kernel::KClientSession* session{};
-    if (const auto result = port.CreateSession(std::addressof(session)); result.IsError()) {
+    if (const auto result = port->GetClientPort().CreateSession(std::addressof(session));
+        result.IsError()) {
        LOG_ERROR(Service_SM, "called service={} -> error 0x{:08X}", name, result.raw);
        return result;
    }
@@ -180,20 +178,21 @@ void SM::RegisterService(Kernel::HLERequestContext& ctx) {
    LOG_DEBUG(Service_SM, "called with name={}, max_session_count={}, is_light={}", name,
              max_session_count, is_light);

-    auto handle = service_manager.RegisterService(name, max_session_count);
-    if (handle.Failed()) {
-        LOG_ERROR(Service_SM, "failed to register service with error_code={:08X}",
-                  handle.Code().raw);
+    if (const auto result = service_manager.RegisterService(name, max_session_count, nullptr);
+        result.IsError()) {
+        LOG_ERROR(Service_SM, "failed to register service with error_code={:08X}", result.raw);
        IPC::ResponseBuilder rb{ctx, 2};
-        rb.Push(handle.Code());
+        rb.Push(result);
        return;
    }

-    IPC::ResponseBuilder rb{ctx, 2, 0, 1, IPC::ResponseBuilder::Flags::AlwaysMoveHandles};
-    rb.Push(handle.Code());
+    auto* port = Kernel::KPort::Create(kernel);
+    port->Initialize(ServerSessionCountMax, is_light, name);
+    SCOPE_EXIT({ port->GetClientPort().Close(); });

-    auto server_port = handle.Unwrap();
-    rb.PushMoveObjects(server_port);
+    IPC::ResponseBuilder rb{ctx, 2, 0, 1, IPC::ResponseBuilder::Flags::AlwaysMoveHandles};
+    rb.Push(ResultSuccess);
+    rb.PushMoveObjects(port->GetServerPort());
 }

 void SM::UnregisterService(Kernel::HLERequestContext& ctx) {
@@ -225,4 +224,10 @@ SM::SM(ServiceManager& service_manager_, Core::System& system_)
    });
 }

+SM::~SM() {
+    for (auto& server_port : server_ports) {
+        server_port->Close();
+    }
+}
+
 } // namespace Service::SM
@@ -49,6 +49,7 @@ private:
    ServiceManager& service_manager;
    bool is_initialized{};
    Kernel::KernelCore& kernel;
+    std::vector<Kernel::KServerPort*> server_ports;
 };

 class ServiceManager {
@@ -58,7 +59,8 @@ public:
    explicit ServiceManager(Kernel::KernelCore& kernel_);
    ~ServiceManager();

-    ResultVal<Kernel::KServerPort*> RegisterService(std::string name, u32 max_sessions);
+    ResultCode RegisterService(std::string name, u32 max_sessions,
+                               Kernel::SessionRequestHandlerPtr handler);
    ResultCode UnregisterService(const std::string& name);
    ResultVal<Kernel::KPort*> GetServicePort(const std::string& name);

@@ -69,21 +71,17 @@ public:
            LOG_DEBUG(Service, "Can't find service: {}", service_name);
            return nullptr;
        }
-        auto* port = service->second;
-        if (port == nullptr) {
-            return nullptr;
-        }
-        return std::static_pointer_cast<T>(port->GetServerPort().GetSessionRequestHandler());
+        return std::static_pointer_cast<T>(service->second);
    }

    void InvokeControlRequest(Kernel::HLERequestContext& context);

 private:
-    std::weak_ptr<SM> sm_interface;
+    std::shared_ptr<SM> sm_interface;
    std::unique_ptr<Controller> controller_interface;

    /// Map of registered services, retrieved using GetServicePort.
-    std::unordered_map<std::string, Kernel::KPort*> registered_services;
+    std::unordered_map<std::string, Kernel::SessionRequestHandlerPtr> registered_services;

    /// Kernel context
    Kernel::KernelCore& kernel;
@@ -570,7 +570,7 @@ std::pair<s32, Errno> Socket::SendTo(u32 flags, const std::vector<u8>& message,
    ASSERT(flags == 0);

    const sockaddr* to = nullptr;
-    const int tolen = addr ? 0 : sizeof(sockaddr);
+    const int tolen = addr ? sizeof(sockaddr) : 0;
    sockaddr host_addr_in;

    if (addr) {
@@ -127,15 +127,15 @@ double PerfStats::GetLastFrameTimeScale() const {
    return duration_cast<DoubleSecs>(previous_frame_length).count() / FRAME_LENGTH;
 }

-void FrameLimiter::DoFrameLimiting(microseconds current_system_time_us) {
-    if (!Settings::values.use_frame_limit.GetValue() ||
+void SpeedLimiter::DoSpeedLimiting(microseconds current_system_time_us) {
+    if (!Settings::values.use_speed_limit.GetValue() ||
        Settings::values.use_multi_core.GetValue()) {
        return;
    }

    auto now = Clock::now();

-    const double sleep_scale = Settings::values.frame_limit.GetValue() / 100.0;
+    const double sleep_scale = Settings::values.speed_limit.GetValue() / 100.0;

    // Max lag caused by slow frames. Shouldn't be more than the length of a frame at the current
    // speed percent or it will clamp too much and prevent this from properly limiting to that
@@ -143,17 +143,17 @@ void FrameLimiter::DoFrameLimiting(microseconds current_system_time_us) {
    // limiting
    const microseconds max_lag_time_us = duration_cast<microseconds>(
        std::chrono::duration<double, std::chrono::microseconds::period>(25ms / sleep_scale));
-    frame_limiting_delta_err += duration_cast<microseconds>(
+    speed_limiting_delta_err += duration_cast<microseconds>(
        std::chrono::duration<double, std::chrono::microseconds::period>(
            (current_system_time_us - previous_system_time_us) / sleep_scale));
-    frame_limiting_delta_err -= duration_cast<microseconds>(now - previous_walltime);
-    frame_limiting_delta_err =
-        std::clamp(frame_limiting_delta_err, -max_lag_time_us, max_lag_time_us);
+    speed_limiting_delta_err -= duration_cast<microseconds>(now - previous_walltime);
+    speed_limiting_delta_err =
+        std::clamp(speed_limiting_delta_err, -max_lag_time_us, max_lag_time_us);

-    if (frame_limiting_delta_err > microseconds::zero()) {
-        std::this_thread::sleep_for(frame_limiting_delta_err);
+    if (speed_limiting_delta_err > microseconds::zero()) {
+        std::this_thread::sleep_for(speed_limiting_delta_err);
        auto now_after_sleep = Clock::now();
-        frame_limiting_delta_err -= duration_cast<microseconds>(now_after_sleep - now);
+        speed_limiting_delta_err -= duration_cast<microseconds>(now_after_sleep - now);
        now = now_after_sleep;
    }

@@ -85,11 +85,11 @@ private:
    double previous_fps = 0;
 };

-class FrameLimiter {
+class SpeedLimiter {
 public:
    using Clock = std::chrono::high_resolution_clock;

-    void DoFrameLimiting(std::chrono::microseconds current_system_time_us);
+    void DoSpeedLimiting(std::chrono::microseconds current_system_time_us);

 private:
    /// Emulated system time (in microseconds) at the last limiter invocation
@@ -98,7 +98,7 @@ private:
    Clock::time_point previous_walltime = Clock::now();

    /// Accumulated difference between walltime and emulated time
-    std::chrono::microseconds frame_limiting_delta_err{0};
+    std::chrono::microseconds speed_limiting_delta_err{0};
 };

 } // namespace Core
@@ -221,8 +221,8 @@ void TelemetrySession::AddInitialInfo(Loader::AppLoader& app_loader,
             TranslateRenderer(Settings::values.renderer_backend.GetValue()));
    AddField(field_type, "Renderer_ResolutionFactor",
             Settings::values.resolution_factor.GetValue());
-    AddField(field_type, "Renderer_UseFrameLimit", Settings::values.use_frame_limit.GetValue());
-    AddField(field_type, "Renderer_FrameLimit", Settings::values.frame_limit.GetValue());
+    AddField(field_type, "Renderer_UseSpeedLimit", Settings::values.use_speed_limit.GetValue());
+    AddField(field_type, "Renderer_SpeedLimit", Settings::values.speed_limit.GetValue());
    AddField(field_type, "Renderer_UseDiskShaderCache",
             Settings::values.use_disk_shader_cache.GetValue());
    AddField(field_type, "Renderer_GPUAccuracyLevel",
@@ -304,10 +304,10 @@ std::vector<std::unique_ptr<Polling::DevicePoller>> InputSubsystem::GetPollers([
 }

 std::string GenerateKeyboardParam(int key_code) {
-    Common::ParamPackage param{
-        {"engine", "keyboard"},
-        {"code", std::to_string(key_code)},
-    };
+    Common::ParamPackage param;
+    param.Set("engine", "keyboard");
+    param.Set("code", key_code);
+    param.Set("toggle", false);
    return param.Serialize();
 }

@@ -57,6 +57,7 @@ Common::ParamPackage MouseButtonFactory::GetNextInput() const {
        if (pad.button != MouseInput::MouseButton::Undefined) {
            params.Set("engine", "mouse");
            params.Set("button", static_cast<u16>(pad.button));
+            params.Set("toggle", false);
            return params;
        }
    }
@@ -82,6 +82,12 @@ public:
        state.buttons.insert_or_assign(button, value);
    }

+    void PreSetButton(int button) {
+        if (!state.buttons.contains(button)) {
+            SetButton(button, false);
+        }
+    }
+
    void SetMotion(SDL_ControllerSensorEvent event) {
        constexpr float gravity_constant = 9.80665f;
        std::lock_guard lock{mutex};
@@ -155,9 +161,16 @@ public:
        state.axes.insert_or_assign(axis, value);
    }

-    float GetAxis(int axis, float range) const {
+    void PreSetAxis(int axis) {
+        if (!state.axes.contains(axis)) {
+            SetAxis(axis, 0);
+        }
+    }
+
+    float GetAxis(int axis, float range, float offset) const {
        std::lock_guard lock{mutex};
-        return static_cast<float>(state.axes.at(axis)) / (32767.0f * range);
+        const float value = static_cast<float>(state.axes.at(axis)) / 32767.0f;
+        return (value + offset) / range;
    }

    bool RumblePlay(u16 amp_low, u16 amp_high) {
@@ -174,9 +187,10 @@ public:
        return false;
    }

-    std::tuple<float, float> GetAnalog(int axis_x, int axis_y, float range) const {
-        float x = GetAxis(axis_x, range);
-        float y = GetAxis(axis_y, range);
+    std::tuple<float, float> GetAnalog(int axis_x, int axis_y, float range, float offset_x,
+                                       float offset_y) const {
+        float x = GetAxis(axis_x, range, offset_x);
+        float y = GetAxis(axis_y, range, offset_y);
        y = -y; // 3DS uses an y-axis inverse from SDL

        // Make sure the coordinates are in the unit circle,
@@ -483,7 +497,7 @@ public:
          trigger_if_greater(trigger_if_greater_) {}

    bool GetStatus() const override {
-        const float axis_value = joystick->GetAxis(axis, 1.0f);
+        const float axis_value = joystick->GetAxis(axis, 1.0f, 0.0f);
        if (trigger_if_greater) {
            return axis_value > threshold;
        }
@@ -500,12 +514,14 @@ private:
 class SDLAnalog final : public Input::AnalogDevice {
 public:
    explicit SDLAnalog(std::shared_ptr<SDLJoystick> joystick_, int axis_x_, int axis_y_,
-                       bool invert_x_, bool invert_y_, float deadzone_, float range_)
+                       bool invert_x_, bool invert_y_, float deadzone_, float range_,
+                       float offset_x_, float offset_y_)
        : joystick(std::move(joystick_)), axis_x(axis_x_), axis_y(axis_y_), invert_x(invert_x_),
-          invert_y(invert_y_), deadzone(deadzone_), range(range_) {}
+          invert_y(invert_y_), deadzone(deadzone_), range(range_), offset_x(offset_x_),
+          offset_y(offset_y_) {}

    std::tuple<float, float> GetStatus() const override {
-        auto [x, y] = joystick->GetAnalog(axis_x, axis_y, range);
+        auto [x, y] = joystick->GetAnalog(axis_x, axis_y, range, offset_x, offset_y);
        const float r = std::sqrt((x * x) + (y * y));
        if (invert_x) {
            x = -x;
@@ -522,8 +538,8 @@ public:
    }

    std::tuple<float, float> GetRawStatus() const override {
-        const float x = joystick->GetAxis(axis_x, range);
-        const float y = joystick->GetAxis(axis_y, range);
+        const float x = joystick->GetAxis(axis_x, range, offset_x);
+        const float y = joystick->GetAxis(axis_y, range, offset_y);
        return {x, -y};
    }

@@ -555,6 +571,8 @@ private:
    const bool invert_y;
    const float deadzone;
    const float range;
+    const float offset_x;
+    const float offset_y;
 };

 class SDLVibration final : public Input::VibrationDevice {
@@ -621,7 +639,7 @@ public:
          trigger_if_greater(trigger_if_greater_) {}

    Input::MotionStatus GetStatus() const override {
-        const float axis_value = joystick->GetAxis(axis, 1.0f);
+        const float axis_value = joystick->GetAxis(axis, 1.0f, 0.0f);
        bool trigger = axis_value < threshold;
        if (trigger_if_greater) {
            trigger = axis_value > threshold;
@@ -720,13 +738,13 @@ public:
                LOG_ERROR(Input, "Unknown direction {}", direction_name);
            }
            // This is necessary so accessing GetAxis with axis won't crash
-            joystick->SetAxis(axis, 0);
+            joystick->PreSetAxis(axis);
            return std::make_unique<SDLAxisButton>(joystick, axis, threshold, trigger_if_greater);
        }

        const int button = params.Get("button", 0);
        // This is necessary so accessing GetButton with button won't crash
-        joystick->SetButton(button, false);
+        joystick->PreSetButton(button);
        return std::make_unique<SDLButton>(joystick, button, toggle);
    }

@@ -757,13 +775,15 @@ public:
        const std::string invert_y_value = params.Get("invert_y", "+");
        const bool invert_x = invert_x_value == "-";
        const bool invert_y = invert_y_value == "-";
+        const float offset_x = params.Get("offset_x", 0.0f);
+        const float offset_y = params.Get("offset_y", 0.0f);
        auto joystick = state.GetSDLJoystickByGUID(guid, port);

        // This is necessary so accessing GetAxis with axis_x and axis_y won't crash
-        joystick->SetAxis(axis_x, 0);
-        joystick->SetAxis(axis_y, 0);
+        joystick->PreSetAxis(axis_x);
+        joystick->PreSetAxis(axis_y);
        return std::make_unique<SDLAnalog>(joystick, axis_x, axis_y, invert_x, invert_y, deadzone,
-                                           range);
+                                           range, offset_x, offset_y);
    }

 private:
@@ -844,13 +864,13 @@ public:
                LOG_ERROR(Input, "Unknown direction {}", direction_name);
            }
            // This is necessary so accessing GetAxis with axis won't crash
-            joystick->SetAxis(axis, 0);
+            joystick->PreSetAxis(axis);
            return std::make_unique<SDLAxisMotion>(joystick, axis, threshold, trigger_if_greater);
        }

        const int button = params.Get("button", 0);
        // This is necessary so accessing GetButton with button won't crash
-        joystick->SetButton(button, false);
+        joystick->PreSetButton(button);
        return std::make_unique<SDLButtonMotion>(joystick, button);
    }

@@ -995,6 +1015,7 @@ Common::ParamPackage BuildButtonParamPackageForButton(int port, std::string guid
    params.Set("port", port);
    params.Set("guid", std::move(guid));
    params.Set("button", button);
+    params.Set("toggle", false);
    return params;
 }

@@ -1134,13 +1155,15 @@ Common::ParamPackage BuildParamPackageForBinding(int port, const std::string& gu
 }

 Common::ParamPackage BuildParamPackageForAnalog(int port, const std::string& guid, int axis_x,
-                                                int axis_y) {
+                                                int axis_y, float offset_x, float offset_y) {
    Common::ParamPackage params;
    params.Set("engine", "sdl");
    params.Set("port", port);
    params.Set("guid", guid);
    params.Set("axis_x", axis_x);
    params.Set("axis_y", axis_y);
+    params.Set("offset_x", offset_x);
+    params.Set("offset_y", offset_y);
    params.Set("invert_x", "+");
    params.Set("invert_y", "+");
    return params;
@@ -1342,24 +1365,39 @@ AnalogMapping SDLState::GetAnalogMappingForDevice(const Common::ParamPackage& pa
    const auto& binding_left_y =
        SDL_GameControllerGetBindForAxis(controller, SDL_CONTROLLER_AXIS_LEFTY);
    if (params.Has("guid2")) {
+        joystick2->PreSetAxis(binding_left_x.value.axis);
+        joystick2->PreSetAxis(binding_left_y.value.axis);
+        const auto left_offset_x = -joystick2->GetAxis(binding_left_x.value.axis, 1.0f, 0);
+        const auto left_offset_y = -joystick2->GetAxis(binding_left_y.value.axis, 1.0f, 0);
        mapping.insert_or_assign(
            Settings::NativeAnalog::LStick,
            BuildParamPackageForAnalog(joystick2->GetPort(), joystick2->GetGUID(),
-                                       binding_left_x.value.axis, binding_left_y.value.axis));
+                                       binding_left_x.value.axis, binding_left_y.value.axis,
+                                       left_offset_x, left_offset_y));
    } else {
+        joystick->PreSetAxis(binding_left_x.value.axis);
+        joystick->PreSetAxis(binding_left_y.value.axis);
+        const auto left_offset_x = -joystick->GetAxis(binding_left_x.value.axis, 1.0f, 0);
+        const auto left_offset_y = -joystick->GetAxis(binding_left_y.value.axis, 1.0f, 0);
        mapping.insert_or_assign(
            Settings::NativeAnalog::LStick,
            BuildParamPackageForAnalog(joystick->GetPort(), joystick->GetGUID(),
-                                       binding_left_x.value.axis, binding_left_y.value.axis));
+                                       binding_left_x.value.axis, binding_left_y.value.axis,
+                                       left_offset_x, left_offset_y));
    }
    const auto& binding_right_x =
        SDL_GameControllerGetBindForAxis(controller, SDL_CONTROLLER_AXIS_RIGHTX);
    const auto& binding_right_y =
        SDL_GameControllerGetBindForAxis(controller, SDL_CONTROLLER_AXIS_RIGHTY);
+    joystick->PreSetAxis(binding_right_x.value.axis);
+    joystick->PreSetAxis(binding_right_y.value.axis);
+    const auto right_offset_x = -joystick->GetAxis(binding_right_x.value.axis, 1.0f, 0);
+    const auto right_offset_y = -joystick->GetAxis(binding_right_y.value.axis, 1.0f, 0);
    mapping.insert_or_assign(Settings::NativeAnalog::RStick,
                             BuildParamPackageForAnalog(joystick->GetPort(), joystick->GetGUID(),
                                                        binding_right_x.value.axis,
-                                                        binding_right_y.value.axis));
+                                                        binding_right_y.value.axis, right_offset_x,
+                                                        right_offset_y));
    return mapping;
 }

@@ -1563,8 +1601,9 @@ public:
            }

            if (const auto joystick = state.GetSDLJoystickBySDLID(event.jaxis.which)) {
+                // Set offset to zero since the joystick is not on center
                auto params = BuildParamPackageForAnalog(joystick->GetPort(), joystick->GetGUID(),
-                                                         first_axis, axis);
+                                                         first_axis, axis, 0, 0);
                first_axis = -1;
                return params;
            }
@@ -350,7 +350,7 @@ std::string_view InputPrimitive(InputTopology topology) {
    case InputTopology::Lines:
        return "LINES";
    case InputTopology::LinesAdjacency:
-        return "LINESS_ADJACENCY";
+        return "LINES_ADJACENCY";
    case InputTopology::Triangles:
        return "TRIANGLES";
    case InputTopology::TrianglesAdjacency:
@@ -2,6 +2,8 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.

+#pragma once
+
 #include <sirit/sirit.h>

 #include "common/common_types.h"
@@ -4,7 +4,7 @@

 #pragma once

-#include <stdexcept>
+#include <exception>
 #include <string>
 #include <string_view>
 #include <utility>
@@ -17,7 +17,7 @@ class Exception : public std::exception {
 public:
    explicit Exception(std::string message) noexcept : err_message{std::move(message)} {}

-    const char* what() const noexcept override {
+    [[nodiscard]] const char* what() const noexcept override {
        return err_message.c_str();
    }

@@ -36,21 +36,21 @@ private:
 class LogicError : public Exception {
 public:
    template <typename... Args>
-    LogicError(const char* message, Args&&... args)
+    explicit LogicError(const char* message, Args&&... args)
        : Exception{fmt::format(fmt::runtime(message), std::forward<Args>(args)...)} {}
 };

 class RuntimeError : public Exception {
 public:
    template <typename... Args>
-    RuntimeError(const char* message, Args&&... args)
+    explicit RuntimeError(const char* message, Args&&... args)
        : Exception{fmt::format(fmt::runtime(message), std::forward<Args>(args)...)} {}
 };

 class NotImplementedException : public Exception {
 public:
    template <typename... Args>
-    NotImplementedException(const char* message, Args&&... args)
+    explicit NotImplementedException(const char* message, Args&&... args)
        : Exception{fmt::format(fmt::runtime(message), std::forward<Args>(args)...)} {
        Append(" is not implemented");
    }
@@ -59,7 +59,7 @@ public:
 class InvalidArgument : public Exception {
 public:
    template <typename... Args>
-    InvalidArgument(const char* message, Args&&... args)
+    explicit InvalidArgument(const char* message, Args&&... args)
        : Exception{fmt::format(fmt::runtime(message), std::forward<Args>(args)...)} {}
 };

@@ -327,8 +327,8 @@ public:
                                      const Value& derivates, const Value& offset,
                                      const F32& lod_clamp, TextureInstInfo info);
    [[nodiscard]] Value ImageRead(const Value& handle, const Value& coords, TextureInstInfo info);
-    [[nodiscard]] void ImageWrite(const Value& handle, const Value& coords, const Value& color,
-                                  TextureInstInfo info);
+    void ImageWrite(const Value& handle, const Value& coords, const Value& color,
+                    TextureInstInfo info);

    [[nodiscard]] Value ImageAtomicIAdd(const Value& handle, const Value& coords,
                                        const Value& value, TextureInstInfo info);
@@ -57,6 +57,7 @@ public:

    [[nodiscard]] IR::Inst* Inst() const;
    [[nodiscard]] IR::Inst* InstRecursive() const;
+    [[nodiscard]] IR::Inst* TryInstRecursive() const;
    [[nodiscard]] IR::Value Resolve() const;
    [[nodiscard]] IR::Reg Reg() const;
    [[nodiscard]] IR::Pred Pred() const;
@@ -197,8 +198,8 @@ public:
    }

    template <typename FlagsType>
-    requires(sizeof(FlagsType) <= sizeof(u32) && std::is_trivially_copyable_v<FlagsType>)
-        [[nodiscard]] void SetFlags(FlagsType value) noexcept {
+    requires(sizeof(FlagsType) <= sizeof(u32) &&
+             std::is_trivially_copyable_v<FlagsType>) void SetFlags(FlagsType value) noexcept {
        std::memcpy(&flags, &value, sizeof(value));
    }

@@ -308,6 +309,13 @@ inline IR::Inst* Value::InstRecursive() const {
    return inst;
 }

+inline IR::Inst* Value::TryInstRecursive() const {
+    if (IsIdentity()) {
+        return inst->Arg(0).TryInstRecursive();
+    }
+    return type == Type::Opaque ? inst : nullptr;
+}
+
 inline IR::Value Value::Resolve() const {
    if (IsIdentity()) {
        return inst->Arg(0).Resolve();
@@ -73,7 +73,7 @@ Token OpcodeToken(Opcode opcode) {
        return Token::PBK;
    case Opcode::PCNT:
    case Opcode::CONT:
-        return Token::PBK;
+        return Token::PCNT;
    case Opcode::PEXIT:
    case Opcode::EXIT:
        return Token::PEXIT;
@@ -111,6 +111,8 @@ void VisitUsages(Info& info, IR::Inst& inst) {
    case IR::Opcode::ConvertF16U16:
    case IR::Opcode::ConvertF16U32:
    case IR::Opcode::ConvertF16U64:
+    case IR::Opcode::ConvertF16F32:
+    case IR::Opcode::ConvertF32F16:
    case IR::Opcode::FPAbs16:
    case IR::Opcode::FPAdd16:
    case IR::Opcode::FPCeil16:
@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.

 #include <algorithm>
+#include <functional>
 #include <tuple>
 #include <type_traits>

@@ -88,6 +89,26 @@ bool FoldWhenAllImmediates(IR::Inst& inst, Func&& func) {
    return true;
 }

+/// Return true when all values in a range are equal
+template <typename Range>
+bool AreEqual(const Range& range) {
+    auto resolver{[](const auto& value) { return value.Resolve(); }};
+    auto equal{[](const IR::Value& lhs, const IR::Value& rhs) {
+        if (lhs == rhs) {
+            return true;
+        }
+        // Not equal, but try to match if they read the same constant buffer
+        if (!lhs.IsImmediate() && !rhs.IsImmediate() &&
+            lhs.Inst()->GetOpcode() == IR::Opcode::GetCbufU32 &&
+            rhs.Inst()->GetOpcode() == IR::Opcode::GetCbufU32 &&
+            lhs.Inst()->Arg(0) == rhs.Inst()->Arg(0) && lhs.Inst()->Arg(1) == rhs.Inst()->Arg(1)) {
+            return true;
+        }
+        return false;
+    }};
+    return std::ranges::adjacent_find(range, std::not_fn(equal), resolver) == std::end(range);
+}
+
 void FoldGetRegister(IR::Inst& inst) {
    if (inst.Arg(0).Reg() == IR::Reg::RZ) {
        inst.ReplaceUsesWith(IR::Value{u32{0}});
@@ -100,6 +121,157 @@ void FoldGetPred(IR::Inst& inst) {
    }
 }

+/// Replaces the XMAD pattern generated by an integer FMA
+bool FoldXmadMultiplyAdd(IR::Block& block, IR::Inst& inst) {
+    /*
+     * We are looking for this specific pattern:
+     *   %6      = BitFieldUExtract %op_b, #0, #16
+     *   %7      = BitFieldUExtract %op_a', #16, #16
+     *   %8      = IMul32 %6, %7
+     *   %10     = BitFieldUExtract %op_a', #0, #16
+     *   %11     = BitFieldInsert %8, %10, #16, #16
+     *   %15     = BitFieldUExtract %op_b, #0, #16
+     *   %16     = BitFieldUExtract %op_a, #0, #16
+     *   %17     = IMul32 %15, %16
+     *   %18     = IAdd32 %17, %op_c
+     *   %22     = BitFieldUExtract %op_b, #16, #16
+     *   %23     = BitFieldUExtract %11, #16, #16
+     *   %24     = IMul32 %22, %23
+     *   %25     = ShiftLeftLogical32 %24, #16
+     *   %26     = ShiftLeftLogical32 %11, #16
+     *   %27     = IAdd32 %26, %18
+     *   %result = IAdd32 %25, %27
+     *
+     * And replace it with:
+     *  %temp   = IMul32 %op_a, %op_b
+     *  %result = IAdd32 %temp, %op_c
+     *
+     * This optimization has been proven safe by Nvidia's compiler logic being reversed.
+     * (If Nvidia generates this code from 'fma(a, b, c)', we can do the same in the reverse order.)
+     */
+    const IR::Value zero{0u};
+    const IR::Value sixteen{16u};
+    IR::Inst* const _25{inst.Arg(0).TryInstRecursive()};
+    IR::Inst* const _27{inst.Arg(1).TryInstRecursive()};
+    if (!_25 || !_27) {
+        return false;
+    }
+    if (_27->GetOpcode() != IR::Opcode::IAdd32) {
+        return false;
+    }
+    if (_25->GetOpcode() != IR::Opcode::ShiftLeftLogical32 || _25->Arg(1) != sixteen) {
+        return false;
+    }
+    IR::Inst* const _24{_25->Arg(0).TryInstRecursive()};
+    if (!_24 || _24->GetOpcode() != IR::Opcode::IMul32) {
+        return false;
+    }
+    IR::Inst* const _22{_24->Arg(0).TryInstRecursive()};
+    IR::Inst* const _23{_24->Arg(1).TryInstRecursive()};
+    if (!_22 || !_23) {
+        return false;
+    }
+    if (_22->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_23->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_22->Arg(1) != sixteen || _22->Arg(2) != sixteen) {
+        return false;
+    }
+    if (_23->Arg(1) != sixteen || _23->Arg(2) != sixteen) {
+        return false;
+    }
+    IR::Inst* const _11{_23->Arg(0).TryInstRecursive()};
+    if (!_11 || _11->GetOpcode() != IR::Opcode::BitFieldInsert) {
+        return false;
+    }
+    if (_11->Arg(2) != sixteen || _11->Arg(3) != sixteen) {
+        return false;
+    }
+    IR::Inst* const _8{_11->Arg(0).TryInstRecursive()};
+    IR::Inst* const _10{_11->Arg(1).TryInstRecursive()};
+    if (!_8 || !_10) {
+        return false;
+    }
+    if (_8->GetOpcode() != IR::Opcode::IMul32) {
+        return false;
+    }
+    if (_10->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    IR::Inst* const _6{_8->Arg(0).TryInstRecursive()};
+    IR::Inst* const _7{_8->Arg(1).TryInstRecursive()};
+    if (!_6 || !_7) {
+        return false;
+    }
+    if (_6->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_7->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_6->Arg(1) != zero || _6->Arg(2) != sixteen) {
+        return false;
+    }
+    if (_7->Arg(1) != sixteen || _7->Arg(2) != sixteen) {
+        return false;
+    }
+    IR::Inst* const _26{_27->Arg(0).TryInstRecursive()};
+    IR::Inst* const _18{_27->Arg(1).TryInstRecursive()};
+    if (!_26 || !_18) {
+        return false;
+    }
+    if (_26->GetOpcode() != IR::Opcode::ShiftLeftLogical32 || _26->Arg(1) != sixteen) {
+        return false;
+    }
+    if (_26->Arg(0).InstRecursive() != _11) {
+        return false;
+    }
+    if (_18->GetOpcode() != IR::Opcode::IAdd32) {
+        return false;
+    }
+    IR::Inst* const _17{_18->Arg(0).TryInstRecursive()};
+    if (!_17 || _17->GetOpcode() != IR::Opcode::IMul32) {
+        return false;
+    }
+    IR::Inst* const _15{_17->Arg(0).TryInstRecursive()};
+    IR::Inst* const _16{_17->Arg(1).TryInstRecursive()};
+    if (!_15 || !_16) {
+        return false;
+    }
+    if (_15->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_16->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return false;
+    }
+    if (_15->Arg(1) != zero || _16->Arg(1) != zero || _10->Arg(1) != zero) {
+        return false;
+    }
+    if (_15->Arg(2) != sixteen || _16->Arg(2) != sixteen || _10->Arg(2) != sixteen) {
+        return false;
+    }
+    const std::array<IR::Value, 3> op_as{
+        _7->Arg(0).Resolve(),
+        _16->Arg(0).Resolve(),
+        _10->Arg(0).Resolve(),
+    };
+    const std::array<IR::Value, 3> op_bs{
+        _22->Arg(0).Resolve(),
+        _6->Arg(0).Resolve(),
+        _15->Arg(0).Resolve(),
+    };
+    const IR::U32 op_c{_18->Arg(1)};
+    if (!AreEqual(op_as) || !AreEqual(op_bs)) {
+        return false;
+    }
+    IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
+    inst.ReplaceUsesWith(ir.IAdd(ir.IMul(IR::U32{op_as[0]}, IR::U32{op_bs[1]}), op_c));
+    return true;
+}
+
 /// Replaces the pattern generated by two XMAD multiplications
 bool FoldXmadMultiply(IR::Block& block, IR::Inst& inst) {
    /*
@@ -116,33 +288,31 @@ bool FoldXmadMultiply(IR::Block& block, IR::Inst& inst) {
     *
     * This optimization has been proven safe by LLVM and MSVC.
     */
-    const IR::Value lhs_arg{inst.Arg(0)};
-    const IR::Value rhs_arg{inst.Arg(1)};
-    if (lhs_arg.IsImmediate() || rhs_arg.IsImmediate()) {
+    IR::Inst* const lhs_shl{inst.Arg(0).TryInstRecursive()};
+    IR::Inst* const rhs_mul{inst.Arg(1).TryInstRecursive()};
+    if (!lhs_shl || !rhs_mul) {
        return false;
    }
-    IR::Inst* const lhs_shl{lhs_arg.InstRecursive()};
    if (lhs_shl->GetOpcode() != IR::Opcode::ShiftLeftLogical32 ||
        lhs_shl->Arg(1) != IR::Value{16U}) {
        return false;
    }
-    if (lhs_shl->Arg(0).IsImmediate()) {
+    IR::Inst* const lhs_mul{lhs_shl->Arg(0).TryInstRecursive()};
+    if (!lhs_mul) {
        return false;
    }
-    IR::Inst* const lhs_mul{lhs_shl->Arg(0).InstRecursive()};
-    IR::Inst* const rhs_mul{rhs_arg.InstRecursive()};
    if (lhs_mul->GetOpcode() != IR::Opcode::IMul32 || rhs_mul->GetOpcode() != IR::Opcode::IMul32) {
        return false;
    }
-    if (lhs_mul->Arg(1).Resolve() != rhs_mul->Arg(1).Resolve()) {
-        return false;
-    }
    const IR::U32 factor_b{lhs_mul->Arg(1)};
-    if (lhs_mul->Arg(0).IsImmediate() || rhs_mul->Arg(0).IsImmediate()) {
+    if (factor_b.Resolve() != rhs_mul->Arg(1).Resolve()) {
+        return false;
+    }
+    IR::Inst* const lhs_bfe{lhs_mul->Arg(0).TryInstRecursive()};
+    IR::Inst* const rhs_bfe{rhs_mul->Arg(0).TryInstRecursive()};
+    if (!lhs_bfe || !rhs_bfe) {
        return false;
    }
-    IR::Inst* const lhs_bfe{lhs_mul->Arg(0).InstRecursive()};
-    IR::Inst* const rhs_bfe{rhs_mul->Arg(0).InstRecursive()};
    if (lhs_bfe->GetOpcode() != IR::Opcode::BitFieldUExtract) {
        return false;
    }
@@ -155,10 +325,10 @@ bool FoldXmadMultiply(IR::Block& block, IR::Inst& inst) {
    if (rhs_bfe->Arg(1) != IR::Value{0U} || rhs_bfe->Arg(2) != IR::Value{16U}) {
        return false;
    }
-    if (lhs_bfe->Arg(0).Resolve() != rhs_bfe->Arg(0).Resolve()) {
+    const IR::U32 factor_a{lhs_bfe->Arg(0)};
+    if (factor_a.Resolve() != rhs_bfe->Arg(0).Resolve()) {
        return false;
    }
-    const IR::U32 factor_a{lhs_bfe->Arg(0)};
    IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
    inst.ReplaceUsesWith(ir.IMul(factor_a, factor_b));
    return true;
@@ -181,6 +351,9 @@ void FoldAdd(IR::Block& block, IR::Inst& inst) {
        if (FoldXmadMultiply(block, inst)) {
            return;
        }
+        if (FoldXmadMultiplyAdd(block, inst)) {
+            return;
+        }
    }
 }

@@ -476,6 +649,10 @@ void ConstantPropagation(IR::Block& block, IR::Inst& inst) {
        return FoldInverseFunc(inst, IR::Opcode::UnpackHalf2x16);
    case IR::Opcode::UnpackHalf2x16:
        return FoldInverseFunc(inst, IR::Opcode::PackHalf2x16);
+    case IR::Opcode::PackFloat2x16:
+        return FoldInverseFunc(inst, IR::Opcode::UnpackFloat2x16);
+    case IR::Opcode::UnpackFloat2x16:
+        return FoldInverseFunc(inst, IR::Opcode::PackFloat2x16);
    case IR::Opcode::SelectU1:
    case IR::Opcode::SelectU8:
    case IR::Opcode::SelectU16:
@@ -63,6 +63,7 @@ private:
            used_objects = std::exchange(rhs.used_objects, 0);
            num_objects = std::exchange(rhs.num_objects, 0);
            storage = std::move(rhs.storage);
+            return *this;
        }

        Chunk(Chunk&& rhs) noexcept
@@ -1,5 +1,10 @@
 add_subdirectory(host_shaders)

+if(LIBVA_FOUND)
+    set_source_files_properties(command_classes/codecs/codec.cpp
+        PROPERTIES COMPILE_DEFINITIONS LIBVA_FOUND=1)
+endif()
+
 add_library(video_core STATIC
    buffer_cache/buffer_base.h
    buffer_cache/buffer_cache.cpp
@@ -106,6 +111,8 @@ add_library(video_core STATIC
    renderer_vulkan/maxwell_to_vk.cpp
    renderer_vulkan/maxwell_to_vk.h
    renderer_vulkan/pipeline_helper.h
+    renderer_vulkan/pipeline_statistics.cpp
+    renderer_vulkan/pipeline_statistics.h
    renderer_vulkan/renderer_vulkan.h
    renderer_vulkan/renderer_vulkan.cpp
    renderer_vulkan/vk_blit_screen.cpp
@@ -600,13 +600,9 @@ bool BufferCache<P>::DMAClear(GPUVAddr dst_address, u64 amount, u32 value) {
    ClearDownload(subtract_interval);
    common_ranges.subtract(subtract_interval);

-    BufferId buffer;
-    do {
-        has_deleted_buffers = false;
-        buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size));
-    } while (has_deleted_buffers);
+    const BufferId buffer = FindBuffer(*cpu_dst_address, static_cast<u32>(size));
    auto& dest_buffer = slot_buffers[buffer];
-    const u32 offset = static_cast<u32>(*cpu_dst_address - dest_buffer.CpuAddr());
+    const u32 offset = dest_buffer.Offset(*cpu_dst_address);
    runtime.ClearBuffer(dest_buffer, offset, size, value);
    return true;
 }
@@ -821,7 +817,6 @@ void BufferCache<P>::CommitAsyncFlushesHigh() {
            const std::size_t size = interval.upper() - interval.lower();
            const VAddr cpu_addr = interval.lower();
            ForEachBufferInRange(cpu_addr, size, [&](BufferId buffer_id, Buffer& buffer) {
-                boost::container::small_vector<BufferCopy, 1> copies;
                buffer.ForEachDownloadRangeAndClear(
                    cpu_addr, size, [&](u64 range_offset, u64 range_size) {
                        const VAddr buffer_addr = buffer.CpuAddr();
@@ -2,7 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.

-#include <cstring>
 #include <fstream>
 #include <vector>
 #include "common/assert.h"
@@ -17,10 +16,47 @@ extern "C" {
 }

 namespace Tegra {
+#if defined(LIBVA_FOUND)
+// Hardware acceleration code from FFmpeg/doc/examples/hw_decode.c originally under MIT license
+namespace {
+constexpr std::array<const char*, 2> VAAPI_DRIVERS = {
+    "i915",
+    "amdgpu",
+};
+
+AVPixelFormat GetHwFormat(AVCodecContext*, const AVPixelFormat* pix_fmts) {
+    for (const AVPixelFormat* p = pix_fmts; *p != AV_PIX_FMT_NONE; ++p) {
+        if (*p == AV_PIX_FMT_VAAPI) {
+            return AV_PIX_FMT_VAAPI;
+        }
+    }
+    LOG_INFO(Service_NVDRV, "Could not find compatible GPU AV format, falling back to CPU");
+    return *pix_fmts;
+}
+
+bool CreateVaapiHwdevice(AVBufferRef** av_hw_device) {
+    AVDictionary* hwdevice_options = nullptr;
+    av_dict_set(&hwdevice_options, "connection_type", "drm", 0);
+    for (const auto& driver : VAAPI_DRIVERS) {
+        av_dict_set(&hwdevice_options, "kernel_driver", driver, 0);
+        const int hwdevice_error = av_hwdevice_ctx_create(av_hw_device, AV_HWDEVICE_TYPE_VAAPI,
+                                                          nullptr, hwdevice_options, 0);
+        if (hwdevice_error >= 0) {
+            LOG_INFO(Service_NVDRV, "Using VA-API with {}", driver);
+            av_dict_free(&hwdevice_options);
+            return true;
+        }
+        LOG_DEBUG(Service_NVDRV, "VA-API av_hwdevice_ctx_create failed {}", hwdevice_error);
+    }
+    LOG_DEBUG(Service_NVDRV, "VA-API av_hwdevice_ctx_create failed for all drivers");
+    av_dict_free(&hwdevice_options);
+    return false;
+}
+} // namespace
+#endif

 void AVFrameDeleter(AVFrame* ptr) {
-    av_frame_unref(ptr);
-    av_free(ptr);
+    av_frame_free(&ptr);
 }

 Codec::Codec(GPU& gpu_, const NvdecCommon::NvdecRegisters& regs)
@@ -32,19 +68,31 @@ Codec::~Codec() {
        return;
    }
    // Free libav memory
-    AVFrame* av_frame{nullptr};
    avcodec_send_packet(av_codec_ctx, nullptr);
-    av_frame = av_frame_alloc();
+    AVFrame* av_frame = av_frame_alloc();
    avcodec_receive_frame(av_codec_ctx, av_frame);
    avcodec_flush_buffers(av_codec_ctx);
-
-    av_frame_unref(av_frame);
-    av_free(av_frame);
+    av_frame_free(&av_frame);
    avcodec_close(av_codec_ctx);
+    av_buffer_unref(&av_hw_device);
+}
+
+void Codec::InitializeHwdec() {
+    // Prioritize integrated GPU to mitigate bandwidth bottlenecks
+#if defined(LIBVA_FOUND)
+    if (CreateVaapiHwdevice(&av_hw_device)) {
+        const auto hw_device_ctx = av_buffer_ref(av_hw_device);
+        ASSERT_MSG(hw_device_ctx, "av_buffer_ref failed");
+        av_codec_ctx->hw_device_ctx = hw_device_ctx;
+        av_codec_ctx->get_format = GetHwFormat;
+        return;
+    }
+#endif
+    // TODO more GPU accelerated decoders
 }

 void Codec::Initialize() {
-    AVCodecID codec{AV_CODEC_ID_NONE};
+    AVCodecID codec;
    switch (current_codec) {
    case NvdecCommon::VideoCodec::H264:
        codec = AV_CODEC_ID_H264;
@@ -53,22 +101,24 @@ void Codec::Initialize() {
        codec = AV_CODEC_ID_VP9;
        break;
    default:
+        UNIMPLEMENTED_MSG("Unknown codec {}", current_codec);
        return;
    }
    av_codec = avcodec_find_decoder(codec);
    av_codec_ctx = avcodec_alloc_context3(av_codec);
    av_opt_set(av_codec_ctx->priv_data, "tune", "zerolatency", 0);
-
-    // TODO(ameerj): libavcodec gpu hw acceleration
-
+    InitializeHwdec();
+    if (!av_codec_ctx->hw_device_ctx) {
+        LOG_INFO(Service_NVDRV, "Using FFmpeg software decoding");
+    }
    const auto av_error = avcodec_open2(av_codec_ctx, av_codec, nullptr);
    if (av_error < 0) {
        LOG_ERROR(Service_NVDRV, "avcodec_open2() Failed.");
        avcodec_close(av_codec_ctx);
+        av_buffer_unref(&av_hw_device);
        return;
    }
    initialized = true;
-    return;
 }

 void Codec::SetTargetCodec(NvdecCommon::VideoCodec codec) {
@@ -80,36 +130,64 @@ void Codec::SetTargetCodec(NvdecCommon::VideoCodec codec) {

 void Codec::Decode() {
    const bool is_first_frame = !initialized;
-    if (!initialized) {
+    if (is_first_frame) {
        Initialize();
    }
-
    bool vp9_hidden_frame = false;
-    AVPacket packet{};
-    av_init_packet(&packet);
    std::vector<u8> frame_data;
-
    if (current_codec == NvdecCommon::VideoCodec::H264) {
        frame_data = h264_decoder->ComposeFrameHeader(state, is_first_frame);
    } else if (current_codec == NvdecCommon::VideoCodec::Vp9) {
        frame_data = vp9_decoder->ComposeFrameHeader(state);
        vp9_hidden_frame = vp9_decoder->WasFrameHidden();
    }
-
+    AVPacket packet{};
+    av_init_packet(&packet);
    packet.data = frame_data.data();
    packet.size = static_cast<s32>(frame_data.size());
-
-    avcodec_send_packet(av_codec_ctx, &packet);
-
-    if (!vp9_hidden_frame) {
-        // Only receive/store visible frames
-        AVFramePtr frame = AVFramePtr{av_frame_alloc(), AVFrameDeleter};
-        avcodec_receive_frame(av_codec_ctx, frame.get());
-        av_frames.push(std::move(frame));
-        // Limit queue to 10 frames. Workaround for ZLA decode and queue spam
-        if (av_frames.size() > 10) {
-            av_frames.pop();
-        }
+    if (const int ret = avcodec_send_packet(av_codec_ctx, &packet); ret) {
+        LOG_DEBUG(Service_NVDRV, "avcodec_send_packet error {}", ret);
+        return;
+    }
+    // Only receive/store visible frames
+    if (vp9_hidden_frame) {
+        return;
+    }
+    AVFrame* hw_frame = av_frame_alloc();
+    AVFrame* sw_frame = hw_frame;
+    ASSERT_MSG(hw_frame, "av_frame_alloc hw_frame failed");
+    if (const int ret = avcodec_receive_frame(av_codec_ctx, hw_frame); ret) {
+        LOG_DEBUG(Service_NVDRV, "avcodec_receive_frame error {}", ret);
+        av_frame_free(&hw_frame);
+        return;
+    }
+    if (!hw_frame->width || !hw_frame->height) {
+        LOG_WARNING(Service_NVDRV, "Zero width or height in frame");
+        av_frame_free(&hw_frame);
+        return;
+    }
+#if defined(LIBVA_FOUND)
+    // Hardware acceleration code from FFmpeg/doc/examples/hw_decode.c under MIT license
+    if (hw_frame->format == AV_PIX_FMT_VAAPI) {
+        sw_frame = av_frame_alloc();
+        ASSERT_MSG(sw_frame, "av_frame_alloc sw_frame failed");
+        // Can't use AV_PIX_FMT_YUV420P and share code with software decoding in vic.cpp
+        // because Intel drivers crash unless using AV_PIX_FMT_NV12
+        sw_frame->format = AV_PIX_FMT_NV12;
+        const int transfer_data_ret = av_hwframe_transfer_data(sw_frame, hw_frame, 0);
+        ASSERT_MSG(!transfer_data_ret, "av_hwframe_transfer_data error {}", transfer_data_ret);
+        av_frame_free(&hw_frame);
+    }
+#endif
+    if (sw_frame->format != AV_PIX_FMT_YUV420P && sw_frame->format != AV_PIX_FMT_NV12) {
+        UNIMPLEMENTED_MSG("Unexpected video format from host graphics: {}", sw_frame->format);
+        av_frame_free(&sw_frame);
+        return;
+    }
+    av_frames.push(AVFramePtr{sw_frame, AVFrameDeleter});
+    if (av_frames.size() > 10) {
+        LOG_TRACE(Service_NVDRV, "av_frames.push overflow dropped frame");
+        av_frames.pop();
    }
 }

@@ -119,7 +197,6 @@ AVFramePtr Codec::GetCurrentFrame() {
    if (av_frames.empty()) {
        return AVFramePtr{nullptr, AVFrameDeleter};
    }
-
    AVFramePtr frame = std::move(av_frames.front());
    av_frames.pop();
    return frame;
@@ -144,6 +221,5 @@ std::string_view Codec::GetCurrentCodecName() const {
    default:
        return "Unknown";
    }
-};
-
+}
 } // namespace Tegra
@@ -22,7 +22,6 @@ extern "C" {

 namespace Tegra {
 class GPU;
-struct VicRegisters;

 void AVFrameDeleter(AVFrame* ptr);
 using AVFramePtr = std::unique_ptr<AVFrame, decltype(&AVFrameDeleter)>;
@@ -55,10 +54,13 @@ public:
    [[nodiscard]] std::string_view GetCurrentCodecName() const;

 private:
+    void InitializeHwdec();
+
    bool initialized{};
    NvdecCommon::VideoCodec current_codec{NvdecCommon::VideoCodec::None};

    AVCodec* av_codec{nullptr};
+    AVBufferRef* av_hw_device{nullptr};
    AVCodecContext* av_codec_ctx{nullptr};

    GPU& gpu;
@@ -11,6 +11,9 @@

 namespace Tegra::Decoder {
 namespace {
+constexpr u32 diff_update_probability = 252;
+constexpr u32 frame_sync_code = 0x498342;
+
 // Default compressed header probabilities once frame context resets
 constexpr Vp9EntropyProbs default_probs{
    .y_mode_prob{
@@ -361,8 +364,7 @@ Vp9PictureInfo VP9::GetVp9PictureInfo(const NvdecCommon::NvdecRegisters& state)
    InsertEntropy(state.vp9_entropy_probs_offset, vp9_info.entropy);

    // surface_luma_offset[0:3] contains the address of the reference frame offsets in the following
-    // order: last, golden, altref, current. It may be worthwhile to track the updates done here
-    // to avoid buffering frame data needed for reference frame updating in the header composition.
+    // order: last, golden, altref, current.
    std::copy(state.surface_luma_offset.begin(), state.surface_luma_offset.begin() + 4,
              vp9_info.frame_offsets.begin());

@@ -384,40 +386,25 @@ Vp9FrameContainer VP9::GetCurrentFrame(const NvdecCommon::NvdecRegisters& state)
        gpu.MemoryManager().ReadBlock(state.frame_bitstream_offset, current_frame.bit_stream.data(),
                                      current_frame.info.bitstream_size);
    }
-    // Buffer two frames, saving the last show frame info
-    if (!next_next_frame.bit_stream.empty()) {
+    if (!next_frame.bit_stream.empty()) {
        Vp9FrameContainer temp{
            .info = current_frame.info,
            .bit_stream = std::move(current_frame.bit_stream),
        };
-        next_next_frame.info.show_frame = current_frame.info.last_frame_shown;
-        current_frame.info = next_next_frame.info;
-        current_frame.bit_stream = std::move(next_next_frame.bit_stream);
-        next_next_frame = std::move(temp);
-
-        if (!next_frame.bit_stream.empty()) {
-            Vp9FrameContainer temp2{
-                .info = current_frame.info,
-                .bit_stream = std::move(current_frame.bit_stream),
-            };
-            next_frame.info.show_frame = current_frame.info.last_frame_shown;
-            current_frame.info = next_frame.info;
-            current_frame.bit_stream = std::move(next_frame.bit_stream);
-            next_frame = std::move(temp2);
-        } else {
-            next_frame.info = current_frame.info;
-            next_frame.bit_stream = std::move(current_frame.bit_stream);
-        }
+        next_frame.info.show_frame = current_frame.info.last_frame_shown;
+        current_frame.info = next_frame.info;
+        current_frame.bit_stream = std::move(next_frame.bit_stream);
+        next_frame = std::move(temp);
    } else {
-        next_next_frame.info = current_frame.info;
-        next_next_frame.bit_stream = std::move(current_frame.bit_stream);
+        next_frame.info = current_frame.info;
+        next_frame.bit_stream = current_frame.bit_stream;
    }
    return current_frame;
 }

 std::vector<u8> VP9::ComposeCompressedHeader() {
    VpxRangeEncoder writer{};
-    const bool update_probs = current_frame_info.show_frame && !current_frame_info.is_key_frame;
+    const bool update_probs = !current_frame_info.is_key_frame && current_frame_info.show_frame;
    if (!current_frame_info.lossless) {
        if (static_cast<u32>(current_frame_info.transform_mode) >= 3) {
            writer.Write(3, 2);
@@ -613,86 +600,64 @@ VpxBitStreamWriter VP9::ComposeUncompressedHeader() {

        // Reset context
        prev_frame_probs = default_probs;
-        swap_next_golden = false;
+        swap_ref_indices = false;
        loop_filter_ref_deltas.fill(0);
        loop_filter_mode_deltas.fill(0);
-
-        // allow frames offsets to stabilize before checking for golden frames
-        grace_period = 4;
-
-        // On key frames, all frame slots are set to the current frame,
-        // so the value of the selected slot doesn't really matter.
-        frame_ctxs.fill({current_frame_number, false, default_probs});
+        frame_ctxs.fill(default_probs);

        // intra only, meaning the frame can be recreated with no other references
        current_frame_info.intra_only = true;
-
    } else {
-
        if (!current_frame_info.show_frame) {
            uncomp_writer.WriteBit(current_frame_info.intra_only);
-            if (!current_frame_info.last_frame_was_key) {
-                swap_next_golden = !swap_next_golden;
-            }
        } else {
            current_frame_info.intra_only = false;
        }
        if (!current_frame_info.error_resilient_mode) {
            uncomp_writer.WriteU(0, 2); // Reset frame context.
        }
-
-        // Last, Golden, Altref frames
-        std::array<s32, 3> ref_frame_index{0, 1, 2};
-
-        // Set when next frame is hidden
-        // altref and golden references are swapped
-        if (swap_next_golden) {
-            ref_frame_index = std::array<s32, 3>{0, 2, 1};
+        const auto& curr_offsets = current_frame_info.frame_offsets;
+        const auto& next_offsets = next_frame.info.frame_offsets;
+        const bool ref_frames_different = curr_offsets[1] != curr_offsets[2];
+        const bool next_references_swap =
+            (next_offsets[1] == curr_offsets[2]) || (next_offsets[2] == curr_offsets[1]);
+        const bool needs_ref_swap = ref_frames_different && next_references_swap;
+        if (needs_ref_swap) {
+            swap_ref_indices = !swap_ref_indices;
        }
+        union {
+            u32 raw;
+            BitField<0, 1, u32> refresh_last;
+            BitField<1, 2, u32> refresh_golden;
+            BitField<2, 1, u32> refresh_alt;
+        } refresh_frame_flags;

-        // update Last Frame
-        u64 refresh_frame_flags = 1;
-
-        // golden frame may refresh, determined if the next golden frame offset is changed
-        bool golden_refresh = false;
-        if (grace_period <= 0) {
-            for (s32 index = 1; index < 3; ++index) {
-                if (current_frame_info.frame_offsets[index] !=
-                    next_frame.info.frame_offsets[index]) {
-                    current_frame_info.refresh_frame[index] = true;
-                    golden_refresh = true;
-                    grace_period = 3;
-                }
+        refresh_frame_flags.raw = 0;
+        for (u32 index = 0; index < 3; ++index) {
+            // Refresh indices that use the current frame as an index
+            if (curr_offsets[3] == next_offsets[index]) {
+                refresh_frame_flags.raw |= 1u << index;
            }
        }
-
-        if (current_frame_info.show_frame &&
-            (!next_frame.info.show_frame || next_frame.info.is_key_frame)) {
-            // Update golden frame
-            refresh_frame_flags = swap_next_golden ? 2 : 4;
+        if (swap_ref_indices) {
+            const u32 temp = refresh_frame_flags.refresh_golden;
+            refresh_frame_flags.refresh_golden.Assign(refresh_frame_flags.refresh_alt.Value());
+            refresh_frame_flags.refresh_alt.Assign(temp);
        }
-
-        if (!current_frame_info.show_frame) {
-            // Update altref
-            refresh_frame_flags = swap_next_golden ? 2 : 4;
-        } else if (golden_refresh) {
-            refresh_frame_flags = 3;
-        }
-
        if (current_frame_info.intra_only) {
            uncomp_writer.WriteU(frame_sync_code, 24);
-            uncomp_writer.WriteU(static_cast<s32>(refresh_frame_flags), 8);
+            uncomp_writer.WriteU(refresh_frame_flags.raw, 8);
            uncomp_writer.WriteU(current_frame_info.frame_size.width - 1, 16);
            uncomp_writer.WriteU(current_frame_info.frame_size.height - 1, 16);
            uncomp_writer.WriteBit(false); // Render and frame size different.
        } else {
-            uncomp_writer.WriteU(static_cast<s32>(refresh_frame_flags), 8);
-
-            for (s32 index = 1; index < 4; index++) {
+            const bool swap_indices = needs_ref_swap ^ swap_ref_indices;
+            const auto ref_frame_index = swap_indices ? std::array{0, 2, 1} : std::array{0, 1, 2};
+            uncomp_writer.WriteU(refresh_frame_flags.raw, 8);
+            for (size_t index = 1; index < 4; index++) {
                uncomp_writer.WriteU(ref_frame_index[index - 1], 3);
                uncomp_writer.WriteU(current_frame_info.ref_frame_sign_bias[index], 1);
            }
-
            uncomp_writer.WriteBit(true);  // Frame size with refs.
            uncomp_writer.WriteBit(false); // Render and frame size different.
            uncomp_writer.WriteBit(current_frame_info.allow_high_precision_mv);
@@ -714,10 +679,9 @@ VpxBitStreamWriter VP9::ComposeUncompressedHeader() {
        frame_ctx_idx = 1;
    }

-    uncomp_writer.WriteU(frame_ctx_idx, 2); // Frame context index.
-    prev_frame_probs =
-        frame_ctxs[frame_ctx_idx].probs; // reference probabilities for compressed header
-    frame_ctxs[frame_ctx_idx] = {current_frame_number, false, current_frame_info.entropy};
+    uncomp_writer.WriteU(frame_ctx_idx, 2);       // Frame context index.
+    prev_frame_probs = frame_ctxs[frame_ctx_idx]; // reference probabilities for compressed header
+    frame_ctxs[frame_ctx_idx] = current_frame_info.entropy;

    uncomp_writer.WriteU(current_frame_info.first_level, 6);
    uncomp_writer.WriteU(current_frame_info.sharpness_level, 3);
@@ -812,7 +776,6 @@ const std::vector<u8>& VP9::ComposeFrameHeader(const NvdecCommon::NvdecRegisters
        current_frame_info = curr_frame.info;
        bitstream = std::move(curr_frame.bit_stream);
    }
-
    // The uncompressed header routine sets PrevProb parameters needed for the compressed header
    auto uncomp_writer = ComposeUncompressedHeader();
    std::vector<u8> compressed_header = ComposeCompressedHeader();
@@ -828,13 +791,6 @@ const std::vector<u8>& VP9::ComposeFrameHeader(const NvdecCommon::NvdecRegisters
              frame.begin() + uncompressed_header.size());
    std::copy(bitstream.begin(), bitstream.end(),
              frame.begin() + uncompressed_header.size() + compressed_header.size());
-
-    // keep track of frame number
-    current_frame_number++;
-    grace_period--;
-
-    // don't display hidden frames
-    hidden = !current_frame_info.show_frame;
    return frame;
 }

@@ -14,7 +14,6 @@

 namespace Tegra {
 class GPU;
-enum class FrameType { KeyFrame = 0, InterFrame = 1 };
 namespace Decoder {

 /// The VpxRangeEncoder, and VpxBitStreamWriter classes are used to compose the
@@ -124,7 +123,7 @@ public:

    /// Returns true if the most recent frame was a hidden frame.
    [[nodiscard]] bool WasFrameHidden() const {
-        return hidden;
+        return !current_frame_info.show_frame;
    }

 private:
@@ -178,19 +177,12 @@ private:
    std::array<s8, 4> loop_filter_ref_deltas{};
    std::array<s8, 2> loop_filter_mode_deltas{};

-    bool hidden = false;
-    s64 current_frame_number = -2; // since we buffer 2 frames
-    s32 grace_period = 6;          // frame offsets need to stabilize
-    std::array<FrameContexts, 4> frame_ctxs{};
    Vp9FrameContainer next_frame{};
-    Vp9FrameContainer next_next_frame{};
-    bool swap_next_golden{};
+    std::array<Vp9EntropyProbs, 4> frame_ctxs{};
+    bool swap_ref_indices{};

    Vp9PictureInfo current_frame_info{};
    Vp9EntropyProbs prev_frame_probs{};
-
-    s32 diff_update_probability = 252;
-    s32 frame_sync_code = 0x498342;
 };

 } // namespace Decoder
@@ -176,7 +176,7 @@ struct PictureInfo {
            .frame_size_changed = (vp9_flags & FrameFlags::FrameSizeChanged) != 0,
            .error_resilient_mode = (vp9_flags & FrameFlags::ErrorResilientMode) != 0,
            .last_frame_shown = (vp9_flags & FrameFlags::LastShowFrame) != 0,
-            .show_frame = false,
+            .show_frame = true,
            .ref_frame_sign_bias = ref_frame_sign_bias,
            .base_q_index = base_q_index,
            .y_dc_delta_q = y_dc_delta_q,
@@ -296,12 +296,6 @@ struct RefPoolElement {
    bool refresh{};
 };

-struct FrameContexts {
-    s64 from;
-    bool adapted;
-    Vp9EntropyProbs probs;
-};
-
 #define ASSERT_POSITION(field_name, position)                                                      \
    static_assert(offsetof(Vp9EntropyProbs, field_name) == position,                               \
                  "Field " #field_name " has invalid position")
@@ -39,7 +39,7 @@ void Nvdec::Execute() {
        codec->Decode();
        break;
    default:
-        UNIMPLEMENTED_MSG("Unknown codec {}", static_cast<u32>(codec->GetCurrentCodec()));
+        UNIMPLEMENTED_MSG("Codec {}", codec->GetCurrentCodecName());
        break;
    }
 }
@@ -46,11 +46,8 @@ void Vic::ProcessMethod(Method method, u32 argument) {
    case Method::SetOutputSurfaceLumaOffset:
        output_surface_luma_address = arg;
        break;
-    case Method::SetOutputSurfaceChromaUOffset:
-        output_surface_chroma_u_address = arg;
-        break;
-    case Method::SetOutputSurfaceChromaVOffset:
-        output_surface_chroma_v_address = arg;
+    case Method::SetOutputSurfaceChromaOffset:
+        output_surface_chroma_address = arg;
        break;
    default:
        break;
@@ -65,11 +62,10 @@ void Vic::Execute() {
    const VicConfig config{gpu.MemoryManager().Read<u64>(config_struct_address + 0x20)};
    const AVFramePtr frame_ptr = nvdec_processor->GetFrame();
    const auto* frame = frame_ptr.get();
-    if (!frame || frame->width == 0 || frame->height == 0) {
+    if (!frame) {
        return;
    }
-    const VideoPixelFormat pixel_format =
-        static_cast<VideoPixelFormat>(config.pixel_format.Value());
+    const auto pixel_format = static_cast<VideoPixelFormat>(config.pixel_format.Value());
    switch (pixel_format) {
    case VideoPixelFormat::BGRA8:
    case VideoPixelFormat::RGBA8: {
@@ -83,16 +79,18 @@ void Vic::Execute() {
            sws_freeContext(scaler_ctx);
            scaler_ctx = nullptr;

-            // FFmpeg returns all frames in YUV420, convert it into expected format
-            scaler_ctx =
-                sws_getContext(frame->width, frame->height, AV_PIX_FMT_YUV420P, frame->width,
-                               frame->height, target_format, 0, nullptr, nullptr, nullptr);
+            // Frames are decoded into either YUV420 or NV12 formats. Convert to desired format
+            scaler_ctx = sws_getContext(frame->width, frame->height,
+                                        static_cast<AVPixelFormat>(frame->format), frame->width,
+                                        frame->height, target_format, 0, nullptr, nullptr, nullptr);

            scaler_width = frame->width;
            scaler_height = frame->height;
        }
        // Get Converted frame
-        const std::size_t linear_size = frame->width * frame->height * 4;
+        const u32 width = static_cast<u32>(frame->width);
+        const u32 height = static_cast<u32>(frame->height);
+        const std::size_t linear_size = width * height * 4;

        // Only allocate frame_buffer once per stream, as the size is not expected to change
        if (!converted_frame_buffer) {
@@ -109,11 +107,10 @@ void Vic::Execute() {
        if (blk_kind != 0) {
            // swizzle pitch linear to block linear
            const u32 block_height = static_cast<u32>(config.block_linear_height_log2);
-            const auto size = Tegra::Texture::CalculateSize(true, 4, frame->width, frame->height, 1,
-                                                            block_height, 0);
+            const auto size =
+                Tegra::Texture::CalculateSize(true, 4, width, height, 1, block_height, 0);
            luma_buffer.resize(size);
-            Tegra::Texture::SwizzleSubrect(frame->width, frame->height, frame->width * 4,
-                                           frame->width, 4, luma_buffer.data(),
+            Tegra::Texture::SwizzleSubrect(width, height, width * 4, width, 4, luma_buffer.data(),
                                           converted_frame_buffer.get(), block_height, 0, 0);

            gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(), size);
@@ -131,41 +128,65 @@ void Vic::Execute() {
        const std::size_t surface_height = config.surface_height_minus1 + 1;
        const auto frame_width = std::min(surface_width, static_cast<size_t>(frame->width));
        const auto frame_height = std::min(surface_height, static_cast<size_t>(frame->height));
-        const std::size_t half_width = frame_width / 2;
-        const std::size_t half_height = frame_height / 2;
-        const std::size_t aligned_width = (surface_width + 0xff) & ~0xff;
+        const std::size_t aligned_width = (surface_width + 0xff) & ~0xffUL;

-        const auto* luma_ptr = frame->data[0];
-        const auto* chroma_b_ptr = frame->data[1];
-        const auto* chroma_r_ptr = frame->data[2];
        const auto stride = static_cast<size_t>(frame->linesize[0]);
-        const auto half_stride = static_cast<size_t>(frame->linesize[1]);

        luma_buffer.resize(aligned_width * surface_height);
        chroma_buffer.resize(aligned_width * surface_height / 2);

        // Populate luma buffer
+        const u8* luma_src = frame->data[0];
        for (std::size_t y = 0; y < frame_height; ++y) {
            const std::size_t src = y * stride;
            const std::size_t dst = y * aligned_width;
            for (std::size_t x = 0; x < frame_width; ++x) {
-                luma_buffer[dst + x] = luma_ptr[src + x];
+                luma_buffer[dst + x] = luma_src[src + x];
            }
        }
        gpu.MemoryManager().WriteBlock(output_surface_luma_address, luma_buffer.data(),
                                       luma_buffer.size());

-        // Populate chroma buffer from both channels with interleaving.
-        for (std::size_t y = 0; y < half_height; ++y) {
-            const std::size_t src = y * half_stride;
-            const std::size_t dst = y * aligned_width;
+        // Chroma
+        const std::size_t half_height = frame_height / 2;
+        const auto half_stride = static_cast<size_t>(frame->linesize[1]);

-            for (std::size_t x = 0; x < half_width; ++x) {
-                chroma_buffer[dst + x * 2] = chroma_b_ptr[src + x];
-                chroma_buffer[dst + x * 2 + 1] = chroma_r_ptr[src + x];
+        switch (frame->format) {
+        case AV_PIX_FMT_YUV420P: {
+            // Frame from FFmpeg software
+            // Populate chroma buffer from both channels with interleaving.
+            const std::size_t half_width = frame_width / 2;
+            const u8* chroma_b_src = frame->data[1];
+            const u8* chroma_r_src = frame->data[2];
+            for (std::size_t y = 0; y < half_height; ++y) {
+                const std::size_t src = y * half_stride;
+                const std::size_t dst = y * aligned_width;
+
+                for (std::size_t x = 0; x < half_width; ++x) {
+                    chroma_buffer[dst + x * 2] = chroma_b_src[src + x];
+                    chroma_buffer[dst + x * 2 + 1] = chroma_r_src[src + x];
+                }
            }
+            break;
        }
-        gpu.MemoryManager().WriteBlock(output_surface_chroma_u_address, chroma_buffer.data(),
+        case AV_PIX_FMT_NV12: {
+            // Frame from VA-API hardware
+            // This is already interleaved so just copy
+            const u8* chroma_src = frame->data[1];
+            for (std::size_t y = 0; y < half_height; ++y) {
+                const std::size_t src = y * stride;
+                const std::size_t dst = y * aligned_width;
+                for (std::size_t x = 0; x < frame_width; ++x) {
+                    chroma_buffer[dst + x] = chroma_src[src + x];
+                }
+            }
+            break;
+        }
+        default:
+            UNREACHABLE();
+            break;
+        }
+        gpu.MemoryManager().WriteBlock(output_surface_chroma_address, chroma_buffer.data(),
                                       chroma_buffer.size());
        break;
    }
@@ -22,8 +22,8 @@ public:
        SetControlParams = 0x1c1,
        SetConfigStructOffset = 0x1c2,
        SetOutputSurfaceLumaOffset = 0x1c8,
-        SetOutputSurfaceChromaUOffset = 0x1c9,
-        SetOutputSurfaceChromaVOffset = 0x1ca
+        SetOutputSurfaceChromaOffset = 0x1c9,
+        SetOutputSurfaceChromaUnusedOffset = 0x1ca
    };

    explicit Vic(GPU& gpu, std::shared_ptr<Nvdec> nvdec_processor);
@@ -64,8 +64,7 @@ private:

    GPUVAddr config_struct_address{};
    GPUVAddr output_surface_luma_address{};
-    GPUVAddr output_surface_chroma_u_address{};
-    GPUVAddr output_surface_chroma_v_address{};
+    GPUVAddr output_surface_chroma_address{};

    SwsContext* scaler_ctx{};
    s32 scaler_width{};
@@ -299,7 +299,7 @@ public:
    };

 private:
-    VideoCore::RasterizerInterface* rasterizer;
+    VideoCore::RasterizerInterface* rasterizer = nullptr;

    /// Performs the copy from the source surface to the destination surface as configured in the
    /// registers.
@@ -227,7 +227,7 @@ private:
    Core::System& system;

    MemoryManager& memory_manager;
-    VideoCore::RasterizerInterface* rasterizer;
+    VideoCore::RasterizerInterface* rasterizer = nullptr;

    std::vector<u8> read_buffer;
    std::vector<u8> write_buffer;
@@ -10,33 +10,27 @@
 #define END_PUSH_CONSTANTS };
 #define UNIFORM(n)
 #define BINDING_INPUT_BUFFER 0
-#define BINDING_ENC_BUFFER 1
-#define BINDING_SWIZZLE_BUFFER 2
-#define BINDING_OUTPUT_IMAGE 3
+#define BINDING_OUTPUT_IMAGE 1

 #else // ^^^ Vulkan ^^^ // vvv OpenGL vvv

 #define BEGIN_PUSH_CONSTANTS
 #define END_PUSH_CONSTANTS
 #define UNIFORM(n) layout(location = n) uniform
-#define BINDING_SWIZZLE_BUFFER 0
-#define BINDING_INPUT_BUFFER 1
-#define BINDING_ENC_BUFFER 2
+#define BINDING_INPUT_BUFFER 0
 #define BINDING_OUTPUT_IMAGE 0

 #endif

-layout(local_size_x = 32, local_size_y = 32, local_size_z = 1) in;
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;

 BEGIN_PUSH_CONSTANTS
 UNIFORM(1) uvec2 block_dims;
-
-UNIFORM(2) uint bytes_per_block_log2;
-UNIFORM(3) uint layer_stride;
-UNIFORM(4) uint block_size;
-UNIFORM(5) uint x_shift;
-UNIFORM(6) uint block_height;
-UNIFORM(7) uint block_height_mask;
+UNIFORM(2) uint layer_stride;
+UNIFORM(3) uint block_size;
+UNIFORM(4) uint x_shift;
+UNIFORM(5) uint block_height;
+UNIFORM(6) uint block_height_mask;
 END_PUSH_CONSTANTS

 struct EncodingData {
@@ -55,45 +49,35 @@ struct TexelWeightParams {
    bool void_extent_hdr;
 };

-// Swizzle data
-layout(binding = BINDING_SWIZZLE_BUFFER, std430) readonly buffer SwizzleTable {
-    uint swizzle_table[];
-};
-
 layout(binding = BINDING_INPUT_BUFFER, std430) readonly buffer InputBufferU32 {
-    uint astc_data[];
-};
-
-// ASTC Encodings data
-layout(binding = BINDING_ENC_BUFFER, std430) readonly buffer EncodingsValues {
-    EncodingData encoding_values[];
+    uvec4 astc_data[];
 };

 layout(binding = BINDING_OUTPUT_IMAGE, rgba8) uniform writeonly image2DArray dest_image;

-const uint GOB_SIZE_X = 64;
-const uint GOB_SIZE_Y = 8;
-const uint GOB_SIZE_Z = 1;
-const uint GOB_SIZE = GOB_SIZE_X * GOB_SIZE_Y * GOB_SIZE_Z;
-
 const uint GOB_SIZE_X_SHIFT = 6;
 const uint GOB_SIZE_Y_SHIFT = 3;
-const uint GOB_SIZE_Z_SHIFT = 0;
-const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT + GOB_SIZE_Z_SHIFT;
+const uint GOB_SIZE_SHIFT = GOB_SIZE_X_SHIFT + GOB_SIZE_Y_SHIFT;

-const uvec2 SWIZZLE_MASK = uvec2(GOB_SIZE_X - 1, GOB_SIZE_Y - 1);
-
-const int BLOCK_SIZE_IN_BYTES = 16;
-
-const int BLOCK_INFO_ERROR = 0;
-const int BLOCK_INFO_VOID_EXTENT_HDR = 1;
-const int BLOCK_INFO_VOID_EXTENT_LDR = 2;
-const int BLOCK_INFO_NORMAL = 3;
+const uint BYTES_PER_BLOCK_LOG2 = 4;

 const int JUST_BITS = 0;
 const int QUINT = 1;
 const int TRIT = 2;

+// ASTC Encodings data, sorted in ascending order based on their BitLength value
+// (see GetBitLength() function)
+EncodingData encoding_values[22] = EncodingData[](
+    EncodingData(JUST_BITS, 0, 0, 0), EncodingData(JUST_BITS, 1, 0, 0), EncodingData(TRIT, 0, 0, 0),
+    EncodingData(JUST_BITS, 2, 0, 0), EncodingData(QUINT, 0, 0, 0), EncodingData(TRIT, 1, 0, 0),
+    EncodingData(JUST_BITS, 3, 0, 0), EncodingData(QUINT, 1, 0, 0), EncodingData(TRIT, 2, 0, 0),
+    EncodingData(JUST_BITS, 4, 0, 0), EncodingData(QUINT, 2, 0, 0), EncodingData(TRIT, 3, 0, 0),
+    EncodingData(JUST_BITS, 5, 0, 0), EncodingData(QUINT, 3, 0, 0), EncodingData(TRIT, 4, 0, 0),
+    EncodingData(JUST_BITS, 6, 0, 0), EncodingData(QUINT, 4, 0, 0), EncodingData(TRIT, 5, 0, 0),
+    EncodingData(JUST_BITS, 7, 0, 0), EncodingData(QUINT, 5, 0, 0), EncodingData(TRIT, 6, 0, 0),
+    EncodingData(JUST_BITS, 8, 0, 0)
+);
+
 // The following constants are expanded variants of the Replicate()
 // function calls corresponding to the following arguments:
 // value: index into the generated table
@@ -135,44 +119,37 @@ const uint REPLICATE_7_BIT_TO_8_TABLE[128] =
 // Input ASTC texture globals
 uint current_index = 0;
 int bitsread = 0;
-uint total_bitsread = 0;
-uint local_buff[16];
+int total_bitsread = 0;
+uvec4 local_buff;

 // Color data globals
-uint color_endpoint_data[16];
+uvec4 color_endpoint_data;
 int color_bitsread = 0;
-uint total_color_bitsread = 0;
-int color_index = 0;

 // Four values, two endpoints, four maximum paritions
 uint color_values[32];
 int colvals_index = 0;

 // Weight data globals
-uint texel_weight_data[16];
+uvec4 texel_weight_data;
 int texel_bitsread = 0;
-uint total_texel_bitsread = 0;
-int texel_index = 0;

 bool texel_flag = false;

 // Global "vectors" to be pushed into when decoding
-EncodingData result_vector[100];
+EncodingData result_vector[144];
 int result_index = 0;

-EncodingData texel_vector[100];
+EncodingData texel_vector[144];
 int texel_vector_index = 0;

 uint unquantized_texel_weights[2][144];

 uint SwizzleOffset(uvec2 pos) {
-    pos = pos & SWIZZLE_MASK;
-    return swizzle_table[pos.y * 64 + pos.x];
-}
-
-uint ReadTexel(uint offset) {
-    // extract the 8-bit value from the 32-bit packed data.
-    return bitfieldExtract(astc_data[offset / 4], int((offset * 8) & 24), 8);
+    uint x = pos.x;
+    uint y = pos.y;
+    return ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
+                          (y % 2) * 16 + (x % 16);
 }

 // Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
@@ -278,14 +255,10 @@ uint Hash52(uint p) {
    return p;
 }

-uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bool small_block) {
-    if (partition_count == 1) {
-        return 0;
-    }
+uint Select2DPartition(uint seed, uint x, uint y, uint partition_count, bool small_block) {
    if (small_block) {
        x <<= 1;
        y <<= 1;
-        z <<= 1;
    }

    seed += (partition_count - 1) * 1024;
@@ -299,10 +272,6 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
    uint seed6 = uint((rnum >> 20) & 0xF);
    uint seed7 = uint((rnum >> 24) & 0xF);
    uint seed8 = uint((rnum >> 28) & 0xF);
-    uint seed9 = uint((rnum >> 18) & 0xF);
-    uint seed10 = uint((rnum >> 22) & 0xF);
-    uint seed11 = uint((rnum >> 26) & 0xF);
-    uint seed12 = uint(((rnum >> 30) | (rnum << 2)) & 0xF);

    seed1 = (seed1 * seed1);
    seed2 = (seed2 * seed2);
@@ -312,12 +281,8 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
    seed6 = (seed6 * seed6);
    seed7 = (seed7 * seed7);
    seed8 = (seed8 * seed8);
-    seed9 = (seed9 * seed9);
-    seed10 = (seed10 * seed10);
-    seed11 = (seed11 * seed11);
-    seed12 = (seed12 * seed12);

-    int sh1, sh2, sh3;
+    uint sh1, sh2;
    if ((seed & 1) > 0) {
        sh1 = (seed & 2) > 0 ? 4 : 5;
        sh2 = (partition_count == 3) ? 6 : 5;
@@ -325,25 +290,19 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
        sh1 = (partition_count == 3) ? 6 : 5;
        sh2 = (seed & 2) > 0 ? 4 : 5;
    }
-    sh3 = (seed & 0x10) > 0 ? sh1 : sh2;
+    seed1 >>= sh1;
+    seed2 >>= sh2;
+    seed3 >>= sh1;
+    seed4 >>= sh2;
+    seed5 >>= sh1;
+    seed6 >>= sh2;
+    seed7 >>= sh1;
+    seed8 >>= sh2;

-    seed1 = (seed1 >> sh1);
-    seed2 = (seed2 >> sh2);
-    seed3 = (seed3 >> sh1);
-    seed4 = (seed4 >> sh2);
-    seed5 = (seed5 >> sh1);
-    seed6 = (seed6 >> sh2);
-    seed7 = (seed7 >> sh1);
-    seed8 = (seed8 >> sh2);
-    seed9 = (seed9 >> sh3);
-    seed10 = (seed10 >> sh3);
-    seed11 = (seed11 >> sh3);
-    seed12 = (seed12 >> sh3);
-
-    uint a = seed1 * x + seed2 * y + seed11 * z + (rnum >> 14);
-    uint b = seed3 * x + seed4 * y + seed12 * z + (rnum >> 10);
-    uint c = seed5 * x + seed6 * y + seed9 * z + (rnum >> 6);
-    uint d = seed7 * x + seed8 * y + seed10 * z + (rnum >> 2);
+    uint a = seed1 * x + seed2 * y + (rnum >> 14);
+    uint b = seed3 * x + seed4 * y + (rnum >> 10);
+    uint c = seed5 * x + seed6 * y + (rnum >> 6);
+    uint d = seed7 * x + seed8 * y + (rnum >> 2);

    a &= 0x3F;
    b &= 0x3F;
@@ -368,58 +327,37 @@ uint SelectPartition(uint seed, uint x, uint y, uint z, uint partition_count, bo
    }
 }

-uint Select2DPartition(uint seed, uint x, uint y, uint partition_count, bool small_block) {
-    return SelectPartition(seed, x, y, 0, partition_count, small_block);
-}
-
-uint ReadBit() {
-    if (current_index >= local_buff.length()) {
+uint ExtractBits(uvec4 payload, int offset, int bits) {
+    if (bits <= 0) {
        return 0;
    }
-    uint bit = bitfieldExtract(local_buff[current_index], bitsread, 1);
-    ++bitsread;
-    ++total_bitsread;
-    if (bitsread == 8) {
-        ++current_index;
-        bitsread = 0;
+    int last_offset = offset + bits - 1;
+    int shifted_offset = offset >> 5;
+    if ((last_offset >> 5) == shifted_offset) {
+        return bitfieldExtract(payload[shifted_offset], offset & 31, bits);
    }
-    return bit;
+    int first_bits = 32 - (offset & 31);
+    int result_first = int(bitfieldExtract(payload[shifted_offset], offset & 31, first_bits));
+    int result_second = int(bitfieldExtract(payload[shifted_offset + 1], 0, bits - first_bits));
+    return result_first | (result_second << first_bits);
 }

 uint StreamBits(uint num_bits) {
-    uint ret = 0;
-    for (uint i = 0; i < num_bits; i++) {
-        ret |= ((ReadBit() & 1) << i);
-    }
+    int int_bits = int(num_bits);
+    uint ret = ExtractBits(local_buff, total_bitsread, int_bits);
+    total_bitsread += int_bits;
    return ret;
 }

-uint ReadColorBit() {
-    uint bit = 0;
-    if (texel_flag) {
-        bit = bitfieldExtract(texel_weight_data[texel_index], texel_bitsread, 1);
-        ++texel_bitsread;
-        ++total_texel_bitsread;
-        if (texel_bitsread == 8) {
-            ++texel_index;
-            texel_bitsread = 0;
-        }
-    } else {
-        bit = bitfieldExtract(color_endpoint_data[color_index], color_bitsread, 1);
-        ++color_bitsread;
-        ++total_color_bitsread;
-        if (color_bitsread == 8) {
-            ++color_index;
-            color_bitsread = 0;
-        }
-    }
-    return bit;
-}
-
 uint StreamColorBits(uint num_bits) {
    uint ret = 0;
-    for (uint i = 0; i < num_bits; i++) {
-        ret |= ((ReadColorBit() & 1) << i);
+    int int_bits = int(num_bits);
+    if (texel_flag) {
+        ret = ExtractBits(texel_weight_data, texel_bitsread, int_bits);
+        texel_bitsread += int_bits;
+    } else {
+        ret = ExtractBits(color_endpoint_data, color_bitsread, int_bits);
+        color_bitsread += int_bits;
    }
    return ret;
 }
@@ -596,22 +534,16 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits) {
    for (uint i = 0; i < num_partitions; i++) {
        num_values += ((modes[i] >> 2) + 1) << 1;
    }
-    int range = 256;
-    while (--range > 0) {
-        EncodingData val = encoding_values[range];
+    // Find the largest encoding that's within color_data_bits
+    // TODO(ameerj): profile with binary search
+    int range = 0;
+    while (++range < encoding_values.length()) {
        uint bit_length = GetBitLength(num_values, range);
-        if (bit_length <= color_data_bits) {
-            while (--range > 0) {
-                EncodingData newval = encoding_values[range];
-                if (newval.encoding != val.encoding && newval.num_bits != val.num_bits) {
-                    break;
-                }
-            }
-            ++range;
+        if (bit_length > color_data_bits) {
            break;
        }
    }
-    DecodeIntegerSequence(range, num_values);
+    DecodeIntegerSequence(range - 1, num_values);
    uint out_index = 0;
    for (int itr = 0; itr < result_index; ++itr) {
        if (out_index >= num_values) {
@@ -1028,7 +960,7 @@ int FindLayout(uint mode) {
    return 5;
 }

-TexelWeightParams DecodeBlockInfo(uint block_index) {
+TexelWeightParams DecodeBlockInfo() {
    TexelWeightParams params = TexelWeightParams(uvec2(0), 0, false, false, false, false);
    uint mode = StreamBits(11);
    if ((mode & 0x1ff) == 0x1fc) {
@@ -1110,10 +1042,10 @@ TexelWeightParams DecodeBlockInfo(uint block_index) {
    }
    weight_index -= 2;
    if ((mode_layout != 9) && ((mode & 0x200) != 0)) {
-        const int max_weights[6] = int[6](9, 11, 15, 19, 23, 31);
+        const int max_weights[6] = int[6](7, 8, 9, 10, 11, 12);
        params.max_weight = max_weights[weight_index];
    } else {
-        const int max_weights[6] = int[6](1, 2, 3, 4, 5, 7);
+        const int max_weights[6] = int[6](1, 2, 3, 4, 5, 6);
        params.max_weight = max_weights[weight_index];
    }
    return params;
@@ -1144,8 +1076,8 @@ void FillVoidExtentLDR(ivec3 coord) {
    }
 }

-void DecompressBlock(ivec3 coord, uint block_index) {
-    TexelWeightParams params = DecodeBlockInfo(block_index);
+void DecompressBlock(ivec3 coord) {
+    TexelWeightParams params = DecodeBlockInfo();
    if (params.error_state) {
        FillError(coord);
        return;
@@ -1212,7 +1144,7 @@ void DecompressBlock(ivec3 coord, uint block_index) {
    // Read color data...
    uint color_data_bits = remaining_bits;
    while (remaining_bits > 0) {
-        int nb = int(min(remaining_bits, 8U));
+        int nb = int(min(remaining_bits, 32U));
        uint b = StreamBits(nb);
        color_endpoint_data[ced_pointer] = uint(bitfieldExtract(b, 0, nb));
        ++ced_pointer;
@@ -1254,25 +1186,20 @@ void DecompressBlock(ivec3 coord, uint block_index) {
        ComputeEndpoints(endpoints[i][0], endpoints[i][1], color_endpoint_mode[i]);
    }

-    for (uint i = 0; i < 16; i++) {
-        texel_weight_data[i] = local_buff[i];
-    }
-    for (uint i = 0; i < 8; i++) {
-#define REVERSE_BYTE(b) ((b * 0x0802U & 0x22110U) | (b * 0x8020U & 0x88440U)) * 0x10101U >> 16
-        uint a = REVERSE_BYTE(texel_weight_data[i]);
-        uint b = REVERSE_BYTE(texel_weight_data[15 - i]);
-#undef REVERSE_BYTE
-        texel_weight_data[i] = uint(bitfieldExtract(b, 0, 8));
-        texel_weight_data[15 - i] = uint(bitfieldExtract(a, 0, 8));
-    }
+    texel_weight_data = local_buff;
+    texel_weight_data = bitfieldReverse(texel_weight_data).wzyx;
    uint clear_byte_start =
        (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) >> 3) + 1;
-    texel_weight_data[clear_byte_start - 1] =
-        texel_weight_data[clear_byte_start - 1] &
+
+    uint byte_insert = ExtractBits(texel_weight_data, int(clear_byte_start - 1) * 8, 8) &
        uint(
            ((1 << (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) % 8)) - 1));
-    for (uint i = 0; i < 16 - clear_byte_start; i++) {
-        texel_weight_data[clear_byte_start + i] = 0U;
+    uint vec_index = (clear_byte_start - 1) >> 2;
+    texel_weight_data[vec_index] =
+        bitfieldInsert(texel_weight_data[vec_index], byte_insert, int((clear_byte_start - 1) % 4) * 8, 8);
+    for (uint i = clear_byte_start; i < 16; ++i) {
+        uint idx = i >> 2;
+        texel_weight_data[idx] = bitfieldInsert(texel_weight_data[idx], 0, int(i % 4) * 8, 8);
    }
    texel_flag = true; // use texel "vector" and bit stream in integer decoding
    DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane));
@@ -1281,8 +1208,11 @@ void DecompressBlock(ivec3 coord, uint block_index) {

    for (uint j = 0; j < block_dims.y; j++) {
        for (uint i = 0; i < block_dims.x; i++) {
-            uint local_partition = Select2DPartition(partition_index, i, j, num_partitions,
+            uint local_partition = 0;
+            if (num_partitions > 1) {
+                local_partition = Select2DPartition(partition_index, i, j, num_partitions,
                                                     (block_dims.y * block_dims.x) < 32);
+            }
            vec4 p;
            uvec4 C0 = ReplicateByteTo16(endpoints[local_partition][0]);
            uvec4 C1 = ReplicateByteTo16(endpoints[local_partition][1]);
@@ -1303,7 +1233,7 @@ void DecompressBlock(ivec3 coord, uint block_index) {

 void main() {
    uvec3 pos = gl_GlobalInvocationID;
-    pos.x <<= bytes_per_block_log2;
+    pos.x <<= BYTES_PER_BLOCK_LOG2;

    // Read as soon as possible due to its latency
    const uint swizzle = SwizzleOffset(pos.xy);
@@ -1321,13 +1251,8 @@ void main() {
    if (any(greaterThanEqual(coord, imageSize(dest_image)))) {
        return;
    }
-    uint block_index =
-        pos.z * gl_WorkGroupSize.x * gl_WorkGroupSize.y + pos.y * gl_WorkGroupSize.x + pos.x;
-
    current_index = 0;
    bitsread = 0;
-    for (int i = 0; i < 16; i++) {
-        local_buff[i] = ReadTexel(offset + i);
-    }
-    DecompressBlock(coord, block_index);
+    local_buff = astc_data[offset / 16];
+    DecompressBlock(coord);
 }
@@ -19,9 +19,6 @@ RendererBase::~RendererBase() = default;

 void RendererBase::RefreshBaseSettings() {
    UpdateCurrentFramebufferLayout();
-
-    renderer_settings.use_framelimiter = Settings::values.use_frame_limit.GetValue();
-    renderer_settings.set_background_color = true;
 }

 void RendererBase::UpdateCurrentFramebufferLayout() {
@@ -30,7 +27,7 @@ void RendererBase::UpdateCurrentFramebufferLayout() {
    render_window.UpdateCurrentFramebufferLayout(layout.width, layout.height);
 }

-void RendererBase::RequestScreenshot(void* data, std::function<void()> callback,
+void RendererBase::RequestScreenshot(void* data, std::function<void(bool)> callback,
                                     const Layout::FramebufferLayout& layout) {
    if (renderer_settings.screenshot_requested) {
        LOG_ERROR(Render, "A screenshot is already requested or in progress, ignoring the request");
@@ -21,13 +21,10 @@ class GraphicsContext;
 namespace VideoCore {

 struct RendererSettings {
-    std::atomic_bool use_framelimiter{false};
-    std::atomic_bool set_background_color{false};
-
    // Screenshot
    std::atomic<bool> screenshot_requested{false};
    void* screenshot_bits{};
-    std::function<void()> screenshot_complete_callback;
+    std::function<void(bool)> screenshot_complete_callback;
    Layout::FramebufferLayout screenshot_framebuffer_layout;
 };

@@ -83,7 +80,7 @@ public:
    void RefreshBaseSettings();

    /// Request a screenshot of the next frame
-    void RequestScreenshot(void* data, std::function<void()> callback,
+    void RequestScreenshot(void* data, std::function<void(bool)> callback,
                           const Layout::FramebufferLayout& layout);

 protected:
@@ -147,7 +147,7 @@ void BufferCacheRuntime::CopyBuffer(Buffer& dst_buffer, Buffer& src_buffer,

 void BufferCacheRuntime::ClearBuffer(Buffer& dest_buffer, u32 offset, size_t size, u32 value) {
    glClearNamedBufferSubData(dest_buffer.Handle(), GL_R32UI, static_cast<GLintptr>(offset),
-                              static_cast<GLsizeiptr>(size / sizeof(u32)), GL_RGBA, GL_UNSIGNED_INT,
+                              static_cast<GLsizeiptr>(size / sizeof(u32)), GL_RED, GL_UNSIGNED_INT,
                              &value);
 }

@@ -441,7 +441,6 @@ std::unique_ptr<GraphicsPipeline> ShaderCache::CreateGraphicsPipeline(

    std::array<const Shader::Info*, Maxwell::MaxShaderStage> infos{};

-    OGLProgram source_program;
    std::array<std::string, 5> sources;
    std::array<std::vector<u32>, 5> sources_spirv;
    Shader::Backend::Bindings binding;
@@ -328,12 +328,10 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture,
 }

 void RendererOpenGL::DrawScreen(const Layout::FramebufferLayout& layout) {
-    if (renderer_settings.set_background_color) {
-        // Update background color before drawing
-        glClearColor(Settings::values.bg_red.GetValue() / 255.0f,
-                     Settings::values.bg_green.GetValue() / 255.0f,
-                     Settings::values.bg_blue.GetValue() / 255.0f, 1.0f);
-    }
+    // Update background color before drawing
+    glClearColor(Settings::values.bg_red.GetValue() / 255.0f,
+                 Settings::values.bg_green.GetValue() / 255.0f,
+                 Settings::values.bg_blue.GetValue() / 255.0f, 1.0f);

    // Set projection matrix
    const std::array ortho_matrix =
@@ -488,7 +486,7 @@ void RendererOpenGL::RenderScreenshot() {
    glBindFramebuffer(GL_READ_FRAMEBUFFER, old_read_fb);
    glBindFramebuffer(GL_DRAW_FRAMEBUFFER, old_draw_fb);

-    renderer_settings.screenshot_complete_callback();
+    renderer_settings.screenshot_complete_callback(true);
    renderer_settings.screenshot_requested = false;
 }

@@ -60,19 +60,14 @@ UtilShaders::UtilShaders(ProgramManager& program_manager_)
      copy_bc4_program(MakeProgram(OPENGL_COPY_BC4_COMP)) {
    const auto swizzle_table = Tegra::Texture::MakeSwizzleTable();
    swizzle_table_buffer.Create();
-    astc_buffer.Create();
    glNamedBufferStorage(swizzle_table_buffer.handle, sizeof(swizzle_table), &swizzle_table, 0);
-    glNamedBufferStorage(astc_buffer.handle, sizeof(ASTC_ENCODINGS_VALUES), &ASTC_ENCODINGS_VALUES,
-                         0);
 }

 UtilShaders::~UtilShaders() = default;

 void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
                             std::span<const VideoCommon::SwizzleParameters> swizzles) {
-    static constexpr GLuint BINDING_SWIZZLE_BUFFER = 0;
-    static constexpr GLuint BINDING_INPUT_BUFFER = 1;
-    static constexpr GLuint BINDING_ENC_BUFFER = 2;
+    static constexpr GLuint BINDING_INPUT_BUFFER = 0;
    static constexpr GLuint BINDING_OUTPUT_IMAGE = 0;

    const Extent2D tile_size{
@@ -80,34 +75,32 @@ void UtilShaders::ASTCDecode(Image& image, const ImageBufferMap& map,
        .height = VideoCore::Surface::DefaultBlockHeight(image.info.format),
    };
    program_manager.BindComputeProgram(astc_decoder_program.handle);
-    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_SWIZZLE_BUFFER, swizzle_table_buffer.handle);
-    glBindBufferBase(GL_SHADER_STORAGE_BUFFER, BINDING_ENC_BUFFER, astc_buffer.handle);
-
    glFlushMappedNamedBufferRange(map.buffer, map.offset, image.guest_size_bytes);
    glUniform2ui(1, tile_size.width, tile_size.height);
+
    // Ensure buffer data is valid before dispatching
    glFlush();
    for (const SwizzleParameters& swizzle : swizzles) {
        const size_t input_offset = swizzle.buffer_offset + map.offset;
-        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
-        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
+        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 8U);
+        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U);

        const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
        ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
        ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
+        ASSERT(params.bytes_per_block_log2 == 4);

-        glUniform1ui(2, params.bytes_per_block_log2);
-        glUniform1ui(3, params.layer_stride);
-        glUniform1ui(4, params.block_size);
-        glUniform1ui(5, params.x_shift);
-        glUniform1ui(6, params.block_height);
-        glUniform1ui(7, params.block_height_mask);
+        glUniform1ui(2, params.layer_stride);
+        glUniform1ui(3, params.block_size);
+        glUniform1ui(4, params.x_shift);
+        glUniform1ui(5, params.block_height);
+        glUniform1ui(6, params.block_height_mask);

-        glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), swizzle.level, GL_TRUE, 0,
-                           GL_WRITE_ONLY, GL_RGBA8);
        // ASTC texture data
        glBindBufferRange(GL_SHADER_STORAGE_BUFFER, BINDING_INPUT_BUFFER, map.buffer, input_offset,
                          image.guest_size_bytes - swizzle.buffer_offset);
+        glBindImageTexture(BINDING_OUTPUT_IMAGE, image.StorageHandle(), swizzle.level, GL_TRUE, 0,
+                           GL_WRITE_ONLY, GL_RGBA8);

        glDispatchCompute(num_dispatches_x, num_dispatches_y, image.info.resources.layers);
    }
@@ -62,7 +62,6 @@ private:
    ProgramManager& program_manager;

    OGLBuffer swizzle_table_buffer;
-    OGLBuffer astc_buffer;

    OGLProgram astc_decoder_program;
    OGLProgram block_linear_unswizzle_2d_program;
@@ -0,0 +1,100 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <string_view>
+
+#include <fmt/format.h>
+
+#include "common/common_types.h"
+#include "common/logging/log.h"
+#include "video_core/renderer_vulkan/pipeline_statistics.h"
+#include "video_core/vulkan_common/vulkan_device.h"
+#include "video_core/vulkan_common/vulkan_wrapper.h"
+
+namespace Vulkan {
+
+using namespace std::string_view_literals;
+
+static u64 GetUint64(const VkPipelineExecutableStatisticKHR& statistic) {
+    switch (statistic.format) {
+    case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_INT64_KHR:
+        return static_cast<u64>(statistic.value.i64);
+    case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_UINT64_KHR:
+        return statistic.value.u64;
+    case VK_PIPELINE_EXECUTABLE_STATISTIC_FORMAT_FLOAT64_KHR:
+        return static_cast<u64>(statistic.value.f64);
+    default:
+        return 0;
+    }
+}
+
+PipelineStatistics::PipelineStatistics(const Device& device_) : device{device_} {}
+
+void PipelineStatistics::Collect(VkPipeline pipeline) {
+    const auto& dev{device.GetLogical()};
+    const std::vector properties{dev.GetPipelineExecutablePropertiesKHR(pipeline)};
+    const u32 num_executables{static_cast<u32>(properties.size())};
+    for (u32 executable = 0; executable < num_executables; ++executable) {
+        const auto statistics{dev.GetPipelineExecutableStatisticsKHR(pipeline, executable)};
+        if (statistics.empty()) {
+            continue;
+        }
+        Stats stage_stats;
+        for (const auto& statistic : statistics) {
+            const char* const name{statistic.name};
+            if (name == "Binary Size"sv || name == "Code size"sv || name == "Instruction Count"sv) {
+                stage_stats.code_size = GetUint64(statistic);
+            } else if (name == "Register Count"sv) {
+                stage_stats.register_count = GetUint64(statistic);
+            } else if (name == "SGPRs"sv || name == "numUsedSgprs"sv) {
+                stage_stats.sgpr_count = GetUint64(statistic);
+            } else if (name == "VGPRs"sv || name == "numUsedVgprs"sv) {
+                stage_stats.vgpr_count = GetUint64(statistic);
+            } else if (name == "Branches"sv) {
+                stage_stats.branches_count = GetUint64(statistic);
+            } else if (name == "Basic Block Count"sv) {
+                stage_stats.basic_block_count = GetUint64(statistic);
+            }
+        }
+        std::lock_guard lock{mutex};
+        collected_stats.push_back(stage_stats);
+    }
+}
+
+void PipelineStatistics::Report() const {
+    double num{};
+    Stats total;
+    {
+        std::lock_guard lock{mutex};
+        for (const Stats& stats : collected_stats) {
+            total.code_size += stats.code_size;
+            total.register_count += stats.register_count;
+            total.sgpr_count += stats.sgpr_count;
+            total.vgpr_count += stats.vgpr_count;
+            total.branches_count += stats.branches_count;
+            total.basic_block_count += stats.basic_block_count;
+        }
+        num = static_cast<double>(collected_stats.size());
+    }
+    std::string report;
+    const auto add = [&](const char* fmt, u64 value) {
+        if (value > 0) {
+            report += fmt::format(fmt::runtime(fmt), static_cast<double>(value) / num);
+        }
+    };
+    add("Code size:      {:9.03f}\n", total.code_size);
+    add("Register count: {:9.03f}\n", total.register_count);
+    add("SGPRs:          {:9.03f}\n", total.sgpr_count);
+    add("VGPRs:          {:9.03f}\n", total.vgpr_count);
+    add("Branches count: {:9.03f}\n", total.branches_count);
+    add("Basic blocks:   {:9.03f}\n", total.basic_block_count);
+
+    LOG_INFO(Render_Vulkan,
+             "\nAverage pipeline statistics\n"
+             "==========================================\n"
+             "{}\n",
+             report);
+}
+
+} // namespace Vulkan
@@ -0,0 +1,40 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <mutex>
+#include <vector>
+
+#include "common/common_types.h"
+#include "video_core/vulkan_common/vulkan_wrapper.h"
+
+namespace Vulkan {
+
+class Device;
+
+class PipelineStatistics {
+public:
+    explicit PipelineStatistics(const Device& device_);
+
+    void Collect(VkPipeline pipeline);
+
+    void Report() const;
+
+private:
+    struct Stats {
+        u64 code_size{};
+        u64 register_count{};
+        u64 sgpr_count{};
+        u64 vgpr_count{};
+        u64 branches_count{};
+        u64 basic_block_count{};
+    };
+
+    const Device& device;
+    mutable std::mutex mutex;
+    std::vector<Stats> collected_stats;
+};
+
+} // namespace Vulkan
@@ -138,6 +138,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
    const bool use_accelerated =
        rasterizer.AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride);
    const bool is_srgb = use_accelerated && screen_info.is_srgb;
+    RenderScreenshot(*framebuffer, use_accelerated);

    bool has_been_recreated = false;
    const auto recreate_swapchain = [&] {
@@ -162,7 +163,7 @@ void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
    if (has_been_recreated) {
        blit_screen.Recreate();
    }
-    const VkSemaphore render_semaphore = blit_screen.Draw(*framebuffer, use_accelerated);
+    const VkSemaphore render_semaphore = blit_screen.DrawToSwapchain(*framebuffer, use_accelerated);
    scheduler.Flush(render_semaphore);
    scheduler.WaitWorker();
    swapchain.Present(render_semaphore);
@@ -193,4 +194,153 @@ void RendererVulkan::Report() const {
    telemetry_session.AddField(field, "GPU_Vulkan_Extensions", extensions);
 }

+void Vulkan::RendererVulkan::RenderScreenshot(const Tegra::FramebufferConfig& framebuffer,
+                                              bool use_accelerated) {
+    if (!renderer_settings.screenshot_requested) {
+        return;
+    }
+    const Layout::FramebufferLayout layout{renderer_settings.screenshot_framebuffer_layout};
+    vk::Image staging_image = device.GetLogical().CreateImage(VkImageCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT,
+        .imageType = VK_IMAGE_TYPE_2D,
+        .format = VK_FORMAT_B8G8R8A8_UNORM,
+        .extent =
+            {
+                .width = layout.width,
+                .height = layout.height,
+                .depth = 1,
+            },
+        .mipLevels = 1,
+        .arrayLayers = 1,
+        .samples = VK_SAMPLE_COUNT_1_BIT,
+        .tiling = VK_IMAGE_TILING_OPTIMAL,
+        .usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT |
+                 VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+        .initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
+    });
+    const auto image_commit = memory_allocator.Commit(staging_image, MemoryUsage::DeviceLocal);
+
+    const vk::ImageView dst_view = device.GetLogical().CreateImageView(VkImageViewCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .image = *staging_image,
+        .viewType = VK_IMAGE_VIEW_TYPE_2D,
+        .format = screen_info.is_srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM,
+        .components{
+            .r = VK_COMPONENT_SWIZZLE_IDENTITY,
+            .g = VK_COMPONENT_SWIZZLE_IDENTITY,
+            .b = VK_COMPONENT_SWIZZLE_IDENTITY,
+            .a = VK_COMPONENT_SWIZZLE_IDENTITY,
+        },
+        .subresourceRange{
+            .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+            .baseMipLevel = 0,
+            .levelCount = 1,
+            .baseArrayLayer = 0,
+            .layerCount = VK_REMAINING_ARRAY_LAYERS,
+        },
+    });
+    const VkExtent2D render_area{.width = layout.width, .height = layout.height};
+    const vk::Framebuffer screenshot_fb = blit_screen.CreateFramebuffer(*dst_view, render_area);
+    // Since we're not rendering to the screen, ignore the render semaphore.
+    void(blit_screen.Draw(framebuffer, *screenshot_fb, layout, render_area, use_accelerated));
+
+    const auto buffer_size = static_cast<VkDeviceSize>(layout.width * layout.height * 4);
+    const VkBufferCreateInfo dst_buffer_info{
+        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .size = buffer_size,
+        .usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+        .queueFamilyIndexCount = 0,
+        .pQueueFamilyIndices = nullptr,
+    };
+    const vk::Buffer dst_buffer = device.GetLogical().CreateBuffer(dst_buffer_info);
+    MemoryCommit dst_buffer_memory = memory_allocator.Commit(dst_buffer, MemoryUsage::Download);
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([&](vk::CommandBuffer cmdbuf) {
+        const VkImageMemoryBarrier read_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR,
+            .newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = *staging_image,
+            .subresourceRange{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        const VkImageMemoryBarrier image_write_barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = 0,
+            .dstAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = *staging_image,
+            .subresourceRange{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .baseMipLevel = 0,
+                .levelCount = VK_REMAINING_MIP_LEVELS,
+                .baseArrayLayer = 0,
+                .layerCount = VK_REMAINING_ARRAY_LAYERS,
+            },
+        };
+        static constexpr VkMemoryBarrier memory_write_barrier{
+            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_MEMORY_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT,
+        };
+        const VkBufferImageCopy copy{
+            .bufferOffset = 0,
+            .bufferRowLength = 0,
+            .bufferImageHeight = 0,
+            .imageSubresource{
+                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+                .mipLevel = 0,
+                .baseArrayLayer = 0,
+                .layerCount = 1,
+            },
+            .imageOffset{.x = 0, .y = 0, .z = 0},
+            .imageExtent{
+                .width = layout.width,
+                .height = layout.height,
+                .depth = 1,
+            },
+        };
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
+                               0, read_barrier);
+        cmdbuf.CopyImageToBuffer(*staging_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, *dst_buffer,
+                                 copy);
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
+                               0, memory_write_barrier, nullptr, image_write_barrier);
+    });
+    // Ensure the copy is fully completed before saving the screenshot
+    scheduler.Finish();
+
+    // Copy backing image data to the QImage screenshot buffer
+    const auto dst_memory_map = dst_buffer_memory.Map();
+    std::memcpy(renderer_settings.screenshot_bits, dst_memory_map.data(), dst_memory_map.size());
+    renderer_settings.screenshot_complete_callback(false);
+    renderer_settings.screenshot_requested = false;
+}
+
 } // namespace Vulkan
@@ -54,6 +54,8 @@ public:
 private:
    void Report() const;

+    void RenderScreenshot(const Tegra::FramebufferConfig& framebuffer, bool use_accelerated);
+
    Core::TelemetrySession& telemetry_session;
    Core::Memory::Memory& cpu_memory;
    Tegra::GPU& gpu;
@@ -130,7 +130,10 @@ void VKBlitScreen::Recreate() {
    CreateDynamicResources();
 }

-VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool use_accelerated) {
+VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
+                               const VkFramebuffer& host_framebuffer,
+                               const Layout::FramebufferLayout layout, VkExtent2D render_area,
+                               bool use_accelerated) {
    RefreshResources(framebuffer);

    // Finish any pending renderpass
@@ -145,8 +148,8 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
                        use_accelerated ? screen_info.image_view : *raw_image_views[image_index]);

    BufferData data;
-    SetUniformData(data, framebuffer);
-    SetVertexData(data, framebuffer);
+    SetUniformData(data, layout);
+    SetVertexData(data, framebuffer, layout);

    const std::span<u8> mapped_span = buffer_commit.Map();
    std::memcpy(mapped_span.data(), &data, sizeof(data));
@@ -220,52 +223,75 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer, bool
                                   VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier);
        });
    }
-    scheduler.Record([this, image_index, size = swapchain.GetSize()](vk::CommandBuffer cmdbuf) {
-        const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f;
-        const f32 bg_green = Settings::values.bg_green.GetValue() / 255.0f;
-        const f32 bg_blue = Settings::values.bg_blue.GetValue() / 255.0f;
-        const VkClearValue clear_color{
-            .color = {.float32 = {bg_red, bg_green, bg_blue, 1.0f}},
-        };
-        const VkRenderPassBeginInfo renderpass_bi{
-            .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
-            .pNext = nullptr,
-            .renderPass = *renderpass,
-            .framebuffer = *framebuffers[image_index],
-            .renderArea =
-                {
-                    .offset = {0, 0},
-                    .extent = size,
-                },
-            .clearValueCount = 1,
-            .pClearValues = &clear_color,
-        };
-        const VkViewport viewport{
-            .x = 0.0f,
-            .y = 0.0f,
-            .width = static_cast<float>(size.width),
-            .height = static_cast<float>(size.height),
-            .minDepth = 0.0f,
-            .maxDepth = 1.0f,
-        };
-        const VkRect2D scissor{
-            .offset = {0, 0},
-            .extent = size,
-        };
-        cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
-        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
-        cmdbuf.SetViewport(0, viewport);
-        cmdbuf.SetScissor(0, scissor);
+    scheduler.Record(
+        [this, host_framebuffer, image_index, size = render_area](vk::CommandBuffer cmdbuf) {
+            const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f;
+            const f32 bg_green = Settings::values.bg_green.GetValue() / 255.0f;
+            const f32 bg_blue = Settings::values.bg_blue.GetValue() / 255.0f;
+            const VkClearValue clear_color{
+                .color = {.float32 = {bg_red, bg_green, bg_blue, 1.0f}},
+            };
+            const VkRenderPassBeginInfo renderpass_bi{
+                .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
+                .pNext = nullptr,
+                .renderPass = *renderpass,
+                .framebuffer = host_framebuffer,
+                .renderArea =
+                    {
+                        .offset = {0, 0},
+                        .extent = size,
+                    },
+                .clearValueCount = 1,
+                .pClearValues = &clear_color,
+            };
+            const VkViewport viewport{
+                .x = 0.0f,
+                .y = 0.0f,
+                .width = static_cast<float>(size.width),
+                .height = static_cast<float>(size.height),
+                .minDepth = 0.0f,
+                .maxDepth = 1.0f,
+            };
+            const VkRect2D scissor{
+                .offset = {0, 0},
+                .extent = size,
+            };
+            cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
+            cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline);
+            cmdbuf.SetViewport(0, viewport);
+            cmdbuf.SetScissor(0, scissor);

-        cmdbuf.BindVertexBuffer(0, *buffer, offsetof(BufferData, vertices));
-        cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline_layout, 0,
-                                  descriptor_sets[image_index], {});
-        cmdbuf.Draw(4, 1, 0, 0);
-        cmdbuf.EndRenderPass();
-    });
+            cmdbuf.BindVertexBuffer(0, *buffer, offsetof(BufferData, vertices));
+            cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, *pipeline_layout, 0,
+                                      descriptor_sets[image_index], {});
+            cmdbuf.Draw(4, 1, 0, 0);
+            cmdbuf.EndRenderPass();
+        });
    return *semaphores[image_index];
 }

+VkSemaphore VKBlitScreen::DrawToSwapchain(const Tegra::FramebufferConfig& framebuffer,
+                                          bool use_accelerated) {
+    const std::size_t image_index = swapchain.GetImageIndex();
+    const VkExtent2D render_area = swapchain.GetSize();
+    const Layout::FramebufferLayout layout = render_window.GetFramebufferLayout();
+    return Draw(framebuffer, *framebuffers[image_index], layout, render_area, use_accelerated);
+}
+
+vk::Framebuffer VKBlitScreen::CreateFramebuffer(const VkImageView& image_view, VkExtent2D extent) {
+    return device.GetLogical().CreateFramebuffer(VkFramebufferCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .renderPass = *renderpass,
+        .attachmentCount = 1,
+        .pAttachments = &image_view,
+        .width = extent.width,
+        .height = extent.height,
+        .layers = 1,
+    });
+}
+
 void VKBlitScreen::CreateStaticResources() {
    CreateShaders();
    CreateSemaphores();
@@ -609,22 +635,9 @@ void VKBlitScreen::CreateFramebuffers() {
    const VkExtent2D size{swapchain.GetSize()};
    framebuffers.resize(image_count);

-    VkFramebufferCreateInfo ci{
-        .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .renderPass = *renderpass,
-        .attachmentCount = 1,
-        .pAttachments = nullptr,
-        .width = size.width,
-        .height = size.height,
-        .layers = 1,
-    };
-
    for (std::size_t i = 0; i < image_count; ++i) {
        const VkImageView image_view{swapchain.GetImageViewIndex(i)};
-        ci.pAttachments = &image_view;
-        framebuffers[i] = device.GetLogical().CreateFramebuffer(ci);
+        framebuffers[i] = CreateFramebuffer(image_view, size);
    }
 }

@@ -752,15 +765,13 @@ void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView imag
    device.GetLogical().UpdateDescriptorSets(std::array{ubo_write, sampler_write}, {});
 }

-void VKBlitScreen::SetUniformData(BufferData& data,
-                                  const Tegra::FramebufferConfig& framebuffer) const {
-    const auto& layout = render_window.GetFramebufferLayout();
+void VKBlitScreen::SetUniformData(BufferData& data, const Layout::FramebufferLayout layout) const {
    data.uniform.modelview_matrix =
        MakeOrthographicMatrix(static_cast<f32>(layout.width), static_cast<f32>(layout.height));
 }

-void VKBlitScreen::SetVertexData(BufferData& data,
-                                 const Tegra::FramebufferConfig& framebuffer) const {
+void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer,
+                                 const Layout::FramebufferLayout layout) const {
    const auto& framebuffer_transform_flags = framebuffer.transform_flags;
    const auto& framebuffer_crop_rect = framebuffer.crop_rect;

@@ -798,7 +809,7 @@ void VKBlitScreen::SetVertexData(BufferData& data,
                  static_cast<f32>(screen_info.height);
    }

-    const auto& screen = render_window.GetFramebufferLayout().screen;
+    const auto& screen = layout.screen;
    const auto x = static_cast<f32>(screen.left);
    const auto y = static_cast<f32>(screen.top);
    const auto w = static_cast<f32>(screen.GetWidth());
@@ -56,8 +56,16 @@ public:
    void Recreate();

    [[nodiscard]] VkSemaphore Draw(const Tegra::FramebufferConfig& framebuffer,
+                                   const VkFramebuffer& host_framebuffer,
+                                   const Layout::FramebufferLayout layout, VkExtent2D render_area,
                                   bool use_accelerated);

+    [[nodiscard]] VkSemaphore DrawToSwapchain(const Tegra::FramebufferConfig& framebuffer,
+                                              bool use_accelerated);
+
+    [[nodiscard]] vk::Framebuffer CreateFramebuffer(const VkImageView& image_view,
+                                                    VkExtent2D extent);
+
 private:
    struct BufferData;

@@ -81,8 +89,9 @@ private:
    void CreateRawImages(const Tegra::FramebufferConfig& framebuffer);

    void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const;
-    void SetUniformData(BufferData& data, const Tegra::FramebufferConfig& framebuffer) const;
-    void SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer) const;
+    void SetUniformData(BufferData& data, const Layout::FramebufferLayout layout) const;
+    void SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer,
+                       const Layout::FramebufferLayout layout) const;

    u64 CalculateBufferSize(const Tegra::FramebufferConfig& framebuffer) const;
    u64 GetRawImageOffset(const Tegra::FramebufferConfig& framebuffer,
@@ -358,7 +358,7 @@ void BufferCacheRuntime::ReserveNullBuffer() {
    if (null_buffer) {
        return;
    }
-    null_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
+    VkBufferCreateInfo create_info{
        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
        .pNext = nullptr,
        .flags = 0,
@@ -367,9 +367,13 @@ void BufferCacheRuntime::ReserveNullBuffer() {
        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
        .queueFamilyIndexCount = 0,
        .pQueueFamilyIndices = nullptr,
-    });
+    };
+    if (device.IsExtTransformFeedbackSupported()) {
+        create_info.usage |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT;
+    }
+    null_buffer = device.GetLogical().CreateBuffer(create_info);
    if (device.HasDebuggingToolAttached()) {
-        null_buffer.SetObjectNameEXT("Null index buffer");
+        null_buffer.SetObjectNameEXT("Null buffer");
    }
    null_buffer_commit = memory_allocator.Commit(null_buffer, MemoryUsage::DeviceLocal);

@@ -30,16 +30,12 @@
 namespace Vulkan {

 using Tegra::Texture::SWIZZLE_TABLE;
-using Tegra::Texture::ASTC::ASTC_ENCODINGS_VALUES;
-using namespace Tegra::Texture::ASTC;

 namespace {

 constexpr u32 ASTC_BINDING_INPUT_BUFFER = 0;
-constexpr u32 ASTC_BINDING_ENC_BUFFER = 1;
-constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 2;
-constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 3;
-constexpr size_t ASTC_NUM_BINDINGS = 4;
+constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 1;
+constexpr size_t ASTC_NUM_BINDINGS = 2;

 template <size_t size>
 inline constexpr VkPushConstantRange COMPUTE_PUSH_CONSTANT_RANGE{
@@ -75,7 +71,7 @@ constexpr DescriptorBankInfo INPUT_OUTPUT_BANK_INFO{
    .score = 2,
 };

-constexpr std::array<VkDescriptorSetLayoutBinding, 4> ASTC_DESCRIPTOR_SET_BINDINGS{{
+constexpr std::array<VkDescriptorSetLayoutBinding, ASTC_NUM_BINDINGS> ASTC_DESCRIPTOR_SET_BINDINGS{{
    {
        .binding = ASTC_BINDING_INPUT_BUFFER,
        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
@@ -83,20 +79,6 @@ constexpr std::array<VkDescriptorSetLayoutBinding, 4> ASTC_DESCRIPTOR_SET_BINDIN
        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
        .pImmutableSamplers = nullptr,
    },
-    {
-        .binding = ASTC_BINDING_ENC_BUFFER,
-        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-        .descriptorCount = 1,
-        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-        .pImmutableSamplers = nullptr,
-    },
-    {
-        .binding = ASTC_BINDING_SWIZZLE_BUFFER,
-        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-        .descriptorCount = 1,
-        .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
-        .pImmutableSamplers = nullptr,
-    },
    {
        .binding = ASTC_BINDING_OUTPUT_IMAGE,
        .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
@@ -108,12 +90,12 @@ constexpr std::array<VkDescriptorSetLayoutBinding, 4> ASTC_DESCRIPTOR_SET_BINDIN

 constexpr DescriptorBankInfo ASTC_BANK_INFO{
    .uniform_buffers = 0,
-    .storage_buffers = 3,
+    .storage_buffers = 1,
    .texture_buffers = 0,
    .image_buffers = 0,
    .textures = 0,
    .images = 1,
-    .score = 4,
+    .score = 2,
 };

 constexpr VkDescriptorUpdateTemplateEntryKHR INPUT_OUTPUT_DESCRIPTOR_UPDATE_TEMPLATE{
@@ -135,22 +117,6 @@ constexpr std::array<VkDescriptorUpdateTemplateEntryKHR, ASTC_NUM_BINDINGS>
            .offset = ASTC_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
            .stride = sizeof(DescriptorUpdateEntry),
        },
-        {
-            .dstBinding = ASTC_BINDING_ENC_BUFFER,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-            .offset = ASTC_BINDING_ENC_BUFFER * sizeof(DescriptorUpdateEntry),
-            .stride = sizeof(DescriptorUpdateEntry),
-        },
-        {
-            .dstBinding = ASTC_BINDING_SWIZZLE_BUFFER,
-            .dstArrayElement = 0,
-            .descriptorCount = 1,
-            .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
-            .offset = ASTC_BINDING_SWIZZLE_BUFFER * sizeof(DescriptorUpdateEntry),
-            .stride = sizeof(DescriptorUpdateEntry),
-        },
        {
            .dstBinding = ASTC_BINDING_OUTPUT_IMAGE,
            .dstArrayElement = 0,
@@ -163,7 +129,6 @@ constexpr std::array<VkDescriptorUpdateTemplateEntryKHR, ASTC_NUM_BINDINGS>

 struct AstcPushConstants {
    std::array<u32, 2> blocks_dims;
-    u32 bytes_per_block_log2;
    u32 layer_stride;
    u32 block_size;
    u32 x_shift;
@@ -258,10 +223,9 @@ std::pair<VkBuffer, VkDeviceSize> Uint8Pass::Assemble(u32 num_vertices, VkBuffer
    update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices);
    update_descriptor_queue.AddBuffer(staging.buffer, staging.offset, staging_size);
    const void* const descriptor_data{update_descriptor_queue.UpdateData()};
-    const VkBuffer buffer{staging.buffer};

    scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([this, buffer, descriptor_data, num_vertices](vk::CommandBuffer cmdbuf) {
+    scheduler.Record([this, descriptor_data, num_vertices](vk::CommandBuffer cmdbuf) {
        static constexpr u32 DISPATCH_SIZE = 1024;
        static constexpr VkMemoryBarrier WRITE_BARRIER{
            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
@@ -319,14 +283,14 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
    const void* const descriptor_data{update_descriptor_queue.UpdateData()};

    scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([this, buffer = staging.buffer, descriptor_data, num_tri_vertices, base_vertex,
+    scheduler.Record([this, descriptor_data, num_tri_vertices, base_vertex,
                      index_shift](vk::CommandBuffer cmdbuf) {
        static constexpr u32 DISPATCH_SIZE = 1024;
        static constexpr VkMemoryBarrier WRITE_BARRIER{
            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
            .pNext = nullptr,
            .srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT,
+            .dstAccessMask = VK_ACCESS_INDEX_READ_BIT,
        };
        const std::array push_constants{base_vertex, index_shift};
        const VkDescriptorSet set = descriptor_allocator.Commit();
@@ -355,46 +319,6 @@ ASTCDecoderPass::ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,

 ASTCDecoderPass::~ASTCDecoderPass() = default;

-void ASTCDecoderPass::MakeDataBuffer() {
-    constexpr size_t TOTAL_BUFFER_SIZE = sizeof(ASTC_ENCODINGS_VALUES) + sizeof(SWIZZLE_TABLE);
-    data_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
-        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .size = TOTAL_BUFFER_SIZE,
-        .usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
-        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
-        .queueFamilyIndexCount = 0,
-        .pQueueFamilyIndices = nullptr,
-    });
-    data_buffer_commit = memory_allocator.Commit(data_buffer, MemoryUsage::Upload);
-
-    const auto staging_ref = staging_buffer_pool.Request(TOTAL_BUFFER_SIZE, MemoryUsage::Upload);
-    std::memcpy(staging_ref.mapped_span.data(), &ASTC_ENCODINGS_VALUES,
-                sizeof(ASTC_ENCODINGS_VALUES));
-    // Tack on the swizzle table at the end of the buffer
-    std::memcpy(staging_ref.mapped_span.data() + sizeof(ASTC_ENCODINGS_VALUES), &SWIZZLE_TABLE,
-                sizeof(SWIZZLE_TABLE));
-
-    scheduler.Record([src = staging_ref.buffer, offset = staging_ref.offset, dst = *data_buffer,
-                      TOTAL_BUFFER_SIZE](vk::CommandBuffer cmdbuf) {
-        static constexpr VkMemoryBarrier write_barrier{
-            .sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
-            .pNext = nullptr,
-            .srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT,
-            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
-        };
-        const VkBufferCopy copy{
-            .srcOffset = offset,
-            .dstOffset = 0,
-            .size = TOTAL_BUFFER_SIZE,
-        };
-        cmdbuf.CopyBuffer(src, dst, copy);
-        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
-                               0, write_barrier);
-    });
-}
-
 void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
                               std::span<const VideoCommon::SwizzleParameters> swizzles) {
    using namespace VideoCommon::Accelerated;
@@ -403,9 +327,6 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
        VideoCore::Surface::DefaultBlockHeight(image.info.format),
    };
    scheduler.RequestOutsideRenderPassOperationContext();
-    if (!data_buffer) {
-        MakeDataBuffer();
-    }
    const VkPipeline vk_pipeline = *pipeline;
    const VkImageAspectFlags aspect_mask = image.AspectMask();
    const VkImage vk_image = image.Handle();
@@ -437,16 +358,13 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
        });
    for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
        const size_t input_offset = swizzle.buffer_offset + map.offset;
-        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
-        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
+        const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 8U);
+        const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 8U);
        const u32 num_dispatches_z = image.info.resources.layers;

        update_descriptor_queue.Acquire();
        update_descriptor_queue.AddBuffer(map.buffer, input_offset,
                                          image.guest_size_bytes - swizzle.buffer_offset);
-        update_descriptor_queue.AddBuffer(*data_buffer, 0, sizeof(ASTC_ENCODINGS_VALUES));
-        update_descriptor_queue.AddBuffer(*data_buffer, sizeof(ASTC_ENCODINGS_VALUES),
-                                          sizeof(SWIZZLE_TABLE));
        update_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
        const void* const descriptor_data{update_descriptor_queue.UpdateData()};

@@ -454,11 +372,11 @@ void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
        const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
        ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
        ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
+        ASSERT(params.bytes_per_block_log2 == 4);
        scheduler.Record([this, num_dispatches_x, num_dispatches_y, num_dispatches_z, block_dims,
                          params, descriptor_data](vk::CommandBuffer cmdbuf) {
            const AstcPushConstants uniforms{
                .blocks_dims = block_dims,
-                .bytes_per_block_log2 = params.bytes_per_block_log2,
                .layer_stride = params.layer_stride,
                .block_size = params.block_size,
                .x_shift = params.x_shift,
@@ -96,15 +96,10 @@ public:
                  std::span<const VideoCommon::SwizzleParameters> swizzles);

 private:
-    void MakeDataBuffer();
-
    VKScheduler& scheduler;
    StagingBufferPool& staging_buffer_pool;
    VKUpdateDescriptorQueue& update_descriptor_queue;
    MemoryAllocator& memory_allocator;
-
-    vk::Buffer data_buffer;
-    MemoryCommit data_buffer_commit;
 };

 } // namespace Vulkan
@@ -8,6 +8,7 @@
 #include <boost/container/small_vector.hpp>

 #include "video_core/renderer_vulkan/pipeline_helper.h"
+#include "video_core/renderer_vulkan/pipeline_statistics.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
 #include "video_core/renderer_vulkan/vk_compute_pipeline.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
@@ -26,6 +27,7 @@ using Tegra::Texture::TexturePair;
 ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descriptor_pool,
                                 VKUpdateDescriptorQueue& update_descriptor_queue_,
                                 Common::ThreadWorker* thread_worker,
+                                 PipelineStatistics* pipeline_statistics,
                                 VideoCore::ShaderNotify* shader_notify, const Shader::Info& info_,
                                 vk::ShaderModule spv_module_)
    : device{device_}, update_descriptor_queue{update_descriptor_queue_}, info{info_},
@@ -36,7 +38,7 @@ ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descript
    std::copy_n(info.constant_buffer_used_sizes.begin(), uniform_buffer_sizes.size(),
                uniform_buffer_sizes.begin());

-    auto func{[this, &descriptor_pool, shader_notify] {
+    auto func{[this, &descriptor_pool, shader_notify, pipeline_statistics] {
        DescriptorLayoutBuilder builder{device};
        builder.Add(info, VK_SHADER_STAGE_COMPUTE_BIT);

@@ -50,10 +52,14 @@ ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descript
            .pNext = nullptr,
            .requiredSubgroupSize = GuestWarpSize,
        };
+        VkPipelineCreateFlags flags{};
+        if (device.IsKhrPipelineEexecutablePropertiesEnabled()) {
+            flags |= VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR;
+        }
        pipeline = device.GetLogical().CreateComputePipeline({
            .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
            .pNext = nullptr,
-            .flags = 0,
+            .flags = flags,
            .stage{
                .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
                .pNext = device.IsExtSubgroupSizeControlSupported() ? &subgroup_size_ci : nullptr,
@@ -67,6 +73,9 @@ ComputePipeline::ComputePipeline(const Device& device_, DescriptorPool& descript
            .basePipelineHandle = 0,
            .basePipelineIndex = 0,
        });
+        if (pipeline_statistics) {
+            pipeline_statistics->Collect(*pipeline);
+        }
        std::lock_guard lock{build_mutex};
        is_built = true;
        build_condvar.notify_one();
@@ -25,6 +25,7 @@ class ShaderNotify;
 namespace Vulkan {

 class Device;
+class PipelineStatistics;
 class VKScheduler;

 class ComputePipeline {
@@ -32,6 +33,7 @@ public:
    explicit ComputePipeline(const Device& device, DescriptorPool& descriptor_pool,
                             VKUpdateDescriptorQueue& update_descriptor_queue,
                             Common::ThreadWorker* thread_worker,
+                             PipelineStatistics* pipeline_statistics,
                             VideoCore::ShaderNotify* shader_notify, const Shader::Info& info,
                             vk::ShaderModule spv_module);

@@ -11,6 +11,7 @@
 #include "common/bit_field.h"
 #include "video_core/renderer_vulkan/maxwell_to_vk.h"
 #include "video_core/renderer_vulkan/pipeline_helper.h"
+#include "video_core/renderer_vulkan/pipeline_statistics.h"
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
 #include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
 #include "video_core/renderer_vulkan/vk_render_pass_cache.h"
@@ -217,8 +218,8 @@ GraphicsPipeline::GraphicsPipeline(
    VKScheduler& scheduler_, BufferCache& buffer_cache_, TextureCache& texture_cache_,
    VideoCore::ShaderNotify* shader_notify, const Device& device_, DescriptorPool& descriptor_pool,
    VKUpdateDescriptorQueue& update_descriptor_queue_, Common::ThreadWorker* worker_thread,
-    RenderPassCache& render_pass_cache, const GraphicsPipelineCacheKey& key_,
-    std::array<vk::ShaderModule, NUM_STAGES> stages,
+    PipelineStatistics* pipeline_statistics, RenderPassCache& render_pass_cache,
+    const GraphicsPipelineCacheKey& key_, std::array<vk::ShaderModule, NUM_STAGES> stages,
    const std::array<const Shader::Info*, NUM_STAGES>& infos)
    : key{key_}, maxwell3d{maxwell3d_}, gpu_memory{gpu_memory_}, device{device_},
      texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, scheduler{scheduler_},
@@ -235,7 +236,7 @@ GraphicsPipeline::GraphicsPipeline(
        enabled_uniform_buffer_masks[stage] = info->constant_buffer_mask;
        std::ranges::copy(info->constant_buffer_used_sizes, uniform_buffer_sizes[stage].begin());
    }
-    auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool] {
+    auto func{[this, shader_notify, &render_pass_cache, &descriptor_pool, pipeline_statistics] {
        DescriptorLayoutBuilder builder{MakeBuilder(device, stage_infos)};
        uses_push_descriptor = builder.CanUsePushDescriptor();
        descriptor_set_layout = builder.CreateDescriptorSetLayout(uses_push_descriptor);
@@ -250,6 +251,9 @@ GraphicsPipeline::GraphicsPipeline(
        const VkRenderPass render_pass{render_pass_cache.Get(MakeRenderPassKey(key.state))};
        Validate();
        MakePipeline(render_pass);
+        if (pipeline_statistics) {
+            pipeline_statistics->Collect(*pipeline);
+        }

        std::lock_guard lock{build_mutex};
        is_built = true;
@@ -782,10 +786,14 @@ void GraphicsPipeline::MakePipeline(VkRenderPass render_pass) {
        }
        */
    }
+    VkPipelineCreateFlags flags{};
+    if (device.IsKhrPipelineEexecutablePropertiesEnabled()) {
+        flags |= VK_PIPELINE_CREATE_CAPTURE_STATISTICS_BIT_KHR;
+    }
    pipeline = device.GetLogical().CreateGraphicsPipeline({
        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
        .pNext = nullptr,
-        .flags = 0,
+        .flags = flags,
        .stageCount = static_cast<u32>(shader_stages.size()),
        .pStages = shader_stages.data(),
        .pVertexInputState = &vertex_input_ci,
--- a/Show More
+++ b/Show More