externals: Update sdl2 to 2.0.16

texture_cache: Split out template definitions

.ci/scripts/linux/docker.sh

@@ -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)
 

CMakeLists.txt

@@ -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)
 
@@ -376,7 +376,7 @@ if (ENABLE_SDL2)
     if (YUZU_USE_BUNDLED_SDL2)
         # Detect toolchain and platform
         if ((MSVC_VERSION GREATER_EQUAL 1910 AND MSVC_VERSION LESS 1930) AND ARCHITECTURE_x86_64)
-            set(SDL2_VER "SDL2-2.0.15-prerelease")
+            set(SDL2_VER "SDL2-2.0.16")
         else()
             message(FATAL_ERROR "No bundled SDL2 binaries for your toolchain. Disable YUZU_USE_BUNDLED_SDL2 and provide your own.")
         endif()
@@ -396,7 +396,7 @@ if (ENABLE_SDL2)
     elseif (YUZU_USE_EXTERNAL_SDL2)
         message(STATUS "Using SDL2 from externals.")
     else()
-        find_package(SDL2 2.0.15 REQUIRED)
+        find_package(SDL2 2.0.16 REQUIRED)
 
         # Some installations don't set SDL2_LIBRARIES
         if("${SDL2_LIBRARIES}" STREQUAL "")
@@ -496,7 +496,7 @@ endif()
 # Ensure libusb is properly configured (based on dolphin libusb include)
 if(NOT APPLE AND NOT YUZU_USE_BUNDLED_LIBUSB)
     include(FindPkgConfig)
-    if (PKG_CONFIG_FOUND)
+    if (PKG_CONFIG_FOUND AND NOT CMAKE_SYSTEM_NAME MATCHES "DragonFly|FreeBSD")
         pkg_check_modules(LIBUSB QUIET libusb-1.0>=1.0.24)
     else()
         find_package(LibUSB)
@@ -583,8 +583,32 @@ if (YUZU_USE_BUNDLED_FFMPEG)
             "${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}")

dist/qt_themes/default/style.qss

@@ -51,11 +51,11 @@ QPushButton#GPUStatusBarButton:hover {
 }
 
 QPushButton#GPUStatusBarButton:checked {
-    color: #ff8040;
+    color: #b06020;
 }
 
 QPushButton#GPUStatusBarButton:!checked {
-    color: #40dd40;
+    color: #109010;
 }
 
 QPushButton#buttonRefreshDevices {

externals/libusb/CMakeLists.txt

@@ -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}

src/common/assert.h

@@ -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!")

src/common/fs/fs_util.cpp

@@ -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());
 }

src/common/fs/fs_util.h

@@ -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.
  *

src/common/hex_util.h

@@ -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);
 }
 

src/common/host_memory.cpp

@@ -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{};

src/common/settings.cpp

@@ -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);

src/common/settings.h

@@ -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"};

src/common/uuid.cpp

@@ -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 {

src/common/uuid.h

@@ -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

src/core/core.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 {

src/core/core.h

@@ -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;

src/core/hle/api_version.h

@@ -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 |

src/core/hle/service/acc/acc.cpp

@@ -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);
 }
 

src/core/hle/service/am/applets/applet_software_keyboard.cpp

@@ -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));

src/core/hle/service/friend/friend.cpp

@@ -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);

src/core/hle/service/ns/language.cpp

@@ -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:

src/core/hle/service/ns/ns_language.h

@@ -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

src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp

@@ -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();
 }
 

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

@@ -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

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

@@ -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

src/core/hle/service/nvflinger/nvflinger.cpp

@@ -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

src/core/hle/service/set/set.cpp

@@ -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) {

src/core/hle/service/set/set.h

@@ -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);
 

src/core/memory.cpp

@@ -4,8 +4,6 @@
 
 #include <algorithm>
 #include <cstring>
-#include <optional>
-#include <utility>
 
 #include "common/assert.h"
 #include "common/atomic_ops.h"
@@ -14,12 +12,10 @@
 #include "common/page_table.h"
 #include "common/settings.h"
 #include "common/swap.h"
-#include "core/arm/arm_interface.h"
 #include "core/core.h"
 #include "core/device_memory.h"
 #include "core/hle/kernel/k_page_table.h"
 #include "core/hle/kernel/k_process.h"
-#include "core/hle/kernel/physical_memory.h"
 #include "core/memory.h"
 #include "video_core/gpu.h"
 
@@ -62,17 +58,7 @@ struct Memory::Impl {
         }
     }
 
-    bool IsValidVirtualAddress(const Kernel::KProcess& process, const VAddr vaddr) const {
-        const auto& page_table = process.PageTable().PageTableImpl();
-        const auto [pointer, type] = page_table.pointers[vaddr >> PAGE_BITS].PointerType();
-        return pointer != nullptr || type == Common::PageType::RasterizerCachedMemory;
-    }
-
-    bool IsValidVirtualAddress(VAddr vaddr) const {
-        return IsValidVirtualAddress(*system.CurrentProcess(), vaddr);
-    }
-
-    u8* GetPointerFromRasterizerCachedMemory(VAddr vaddr) const {
+    [[nodiscard]] u8* GetPointerFromRasterizerCachedMemory(VAddr vaddr) const {
         const PAddr paddr{current_page_table->backing_addr[vaddr >> PAGE_BITS]};
 
         if (!paddr) {
@@ -82,18 +68,6 @@ struct Memory::Impl {
         return system.DeviceMemory().GetPointer(paddr) + vaddr;
     }
 
-    u8* GetPointer(const VAddr vaddr) const {
-        const uintptr_t raw_pointer = current_page_table->pointers[vaddr >> PAGE_BITS].Raw();
-        if (u8* const pointer = Common::PageTable::PageInfo::ExtractPointer(raw_pointer)) {
-            return pointer + vaddr;
-        }
-        const auto type = Common::PageTable::PageInfo::ExtractType(raw_pointer);
-        if (type == Common::PageType::RasterizerCachedMemory) {
-            return GetPointerFromRasterizerCachedMemory(vaddr);
-        }
-        return nullptr;
-    }
-
     u8 Read8(const VAddr addr) {
         return Read<u8>(addr);
     }
@@ -179,7 +153,7 @@ struct Memory::Impl {
         std::string string;
         string.reserve(max_length);
         for (std::size_t i = 0; i < max_length; ++i) {
-            const char c = Read8(vaddr);
+            const char c = Read<s8>(vaddr);
             if (c == '\0') {
                 break;
             }
@@ -190,15 +164,14 @@ struct Memory::Impl {
         return string;
     }
 
-    void ReadBlock(const Kernel::KProcess& process, const VAddr src_addr, void* dest_buffer,
-                   const std::size_t size) {
+    void WalkBlock(const Kernel::KProcess& process, const VAddr addr, const std::size_t size,
+                   auto on_unmapped, auto on_memory, auto on_rasterizer, auto increment) {
         const auto& page_table = process.PageTable().PageTableImpl();
-
         std::size_t remaining_size = size;
-        std::size_t page_index = src_addr >> PAGE_BITS;
-        std::size_t page_offset = src_addr & PAGE_MASK;
+        std::size_t page_index = addr >> PAGE_BITS;
+        std::size_t page_offset = addr & PAGE_MASK;
 
-        while (remaining_size > 0) {
+        while (remaining_size) {
             const std::size_t copy_amount =
                 std::min(static_cast<std::size_t>(PAGE_SIZE) - page_offset, remaining_size);
             const auto current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
@@ -206,22 +179,18 @@ struct Memory::Impl {
             const auto [pointer, type] = page_table.pointers[page_index].PointerType();
             switch (type) {
             case Common::PageType::Unmapped: {
-                LOG_ERROR(HW_Memory,
-                          "Unmapped ReadBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
-                          current_vaddr, src_addr, size);
-                std::memset(dest_buffer, 0, copy_amount);
+                on_unmapped(copy_amount, current_vaddr);
                 break;
             }
             case Common::PageType::Memory: {
                 DEBUG_ASSERT(pointer);
-                const u8* const src_ptr = pointer + page_offset + (page_index << PAGE_BITS);
-                std::memcpy(dest_buffer, src_ptr, copy_amount);
+                u8* mem_ptr = pointer + page_offset + (page_index << PAGE_BITS);
+                on_memory(copy_amount, mem_ptr);
                 break;
             }
             case Common::PageType::RasterizerCachedMemory: {
-                const u8* const host_ptr{GetPointerFromRasterizerCachedMemory(current_vaddr)};
-                system.GPU().FlushRegion(current_vaddr, copy_amount);
-                std::memcpy(dest_buffer, host_ptr, copy_amount);
+                u8* const host_ptr{GetPointerFromRasterizerCachedMemory(current_vaddr)};
+                on_rasterizer(current_vaddr, copy_amount, host_ptr);
                 break;
             }
             default:
@@ -230,248 +199,122 @@ struct Memory::Impl {
 
             page_index++;
             page_offset = 0;
-            dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
+            increment(copy_amount);
             remaining_size -= copy_amount;
         }
     }
 
-    void ReadBlockUnsafe(const Kernel::KProcess& process, const VAddr src_addr, void* dest_buffer,
-                         const std::size_t size) {
-        const auto& page_table = process.PageTable().PageTableImpl();
-
-        std::size_t remaining_size = size;
-        std::size_t page_index = src_addr >> PAGE_BITS;
-        std::size_t page_offset = src_addr & PAGE_MASK;
-
-        while (remaining_size > 0) {
-            const std::size_t copy_amount =
-                std::min(static_cast<std::size_t>(PAGE_SIZE) - page_offset, remaining_size);
-            const auto current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
-
-            const auto [pointer, type] = page_table.pointers[page_index].PointerType();
-            switch (type) {
-            case Common::PageType::Unmapped: {
+    template <bool UNSAFE>
+    void ReadBlockImpl(const Kernel::KProcess& process, const VAddr src_addr, void* dest_buffer,
+                       const std::size_t size) {
+        WalkBlock(
+            process, src_addr, size,
+            [src_addr, size, &dest_buffer](const std::size_t copy_amount,
+                                           const VAddr current_vaddr) {
                 LOG_ERROR(HW_Memory,
                           "Unmapped ReadBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
                           current_vaddr, src_addr, size);
                 std::memset(dest_buffer, 0, copy_amount);
-                break;
-            }
-            case Common::PageType::Memory: {
-                DEBUG_ASSERT(pointer);
-                const u8* const src_ptr = pointer + page_offset + (page_index << PAGE_BITS);
+            },
+            [&dest_buffer](const std::size_t copy_amount, const u8* const src_ptr) {
                 std::memcpy(dest_buffer, src_ptr, copy_amount);
-                break;
-            }
-            case Common::PageType::RasterizerCachedMemory: {
-                const u8* const host_ptr{GetPointerFromRasterizerCachedMemory(current_vaddr)};
+            },
+            [&system = system, &dest_buffer](const VAddr current_vaddr,
+                                             const std::size_t copy_amount,
+                                             const u8* const host_ptr) {
+                if constexpr (!UNSAFE) {
+                    system.GPU().FlushRegion(current_vaddr, copy_amount);
+                }
                 std::memcpy(dest_buffer, host_ptr, copy_amount);
-                break;
-            }
-            default:
-                UNREACHABLE();
-            }
-
-            page_index++;
-            page_offset = 0;
-            dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
-            remaining_size -= copy_amount;
-        }
+            },
+            [&dest_buffer](const std::size_t copy_amount) {
+                dest_buffer = static_cast<u8*>(dest_buffer) + copy_amount;
+            });
     }
 
     void ReadBlock(const VAddr src_addr, void* dest_buffer, const std::size_t size) {
-        ReadBlock(*system.CurrentProcess(), src_addr, dest_buffer, size);
+        ReadBlockImpl<false>(*system.CurrentProcess(), src_addr, dest_buffer, size);
     }
 
     void ReadBlockUnsafe(const VAddr src_addr, void* dest_buffer, const std::size_t size) {
-        ReadBlockUnsafe(*system.CurrentProcess(), src_addr, dest_buffer, size);
+        ReadBlockImpl<true>(*system.CurrentProcess(), src_addr, dest_buffer, size);
     }
 
-    void WriteBlock(const Kernel::KProcess& process, const VAddr dest_addr, const void* src_buffer,
-                    const std::size_t size) {
-        const auto& page_table = process.PageTable().PageTableImpl();
-        std::size_t remaining_size = size;
-        std::size_t page_index = dest_addr >> PAGE_BITS;
-        std::size_t page_offset = dest_addr & PAGE_MASK;
-
-        while (remaining_size > 0) {
-            const std::size_t copy_amount =
-                std::min(static_cast<std::size_t>(PAGE_SIZE) - page_offset, remaining_size);
-            const auto current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
-
-            const auto [pointer, type] = page_table.pointers[page_index].PointerType();
-            switch (type) {
-            case Common::PageType::Unmapped: {
+    template <bool UNSAFE>
+    void WriteBlockImpl(const Kernel::KProcess& process, const VAddr dest_addr,
+                        const void* src_buffer, const std::size_t size) {
+        WalkBlock(
+            process, dest_addr, size,
+            [dest_addr, size](const std::size_t copy_amount, const VAddr current_vaddr) {
                 LOG_ERROR(HW_Memory,
                           "Unmapped WriteBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
                           current_vaddr, dest_addr, size);
-                break;
-            }
-            case Common::PageType::Memory: {
-                DEBUG_ASSERT(pointer);
-                u8* const dest_ptr = pointer + page_offset + (page_index << PAGE_BITS);
+            },
+            [&src_buffer](const std::size_t copy_amount, u8* const dest_ptr) {
                 std::memcpy(dest_ptr, src_buffer, copy_amount);
-                break;
-            }
-            case Common::PageType::RasterizerCachedMemory: {
-                u8* const host_ptr{GetPointerFromRasterizerCachedMemory(current_vaddr)};
-                system.GPU().InvalidateRegion(current_vaddr, copy_amount);
+            },
+            [&system = system, &src_buffer](const VAddr current_vaddr,
+                                            const std::size_t copy_amount, u8* const host_ptr) {
+                if constexpr (!UNSAFE) {
+                    system.GPU().InvalidateRegion(current_vaddr, copy_amount);
+                }
                 std::memcpy(host_ptr, src_buffer, copy_amount);
-                break;
-            }
-            default:
-                UNREACHABLE();
-            }
-
-            page_index++;
-            page_offset = 0;
-            src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
-            remaining_size -= copy_amount;
-        }
-    }
-
-    void WriteBlockUnsafe(const Kernel::KProcess& process, const VAddr dest_addr,
-                          const void* src_buffer, const std::size_t size) {
-        const auto& page_table = process.PageTable().PageTableImpl();
-        std::size_t remaining_size = size;
-        std::size_t page_index = dest_addr >> PAGE_BITS;
-        std::size_t page_offset = dest_addr & PAGE_MASK;
-
-        while (remaining_size > 0) {
-            const std::size_t copy_amount =
-                std::min(static_cast<std::size_t>(PAGE_SIZE) - page_offset, remaining_size);
-            const auto current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
-
-            const auto [pointer, type] = page_table.pointers[page_index].PointerType();
-            switch (type) {
-            case Common::PageType::Unmapped: {
-                LOG_ERROR(HW_Memory,
-                          "Unmapped WriteBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
-                          current_vaddr, dest_addr, size);
-                break;
-            }
-            case Common::PageType::Memory: {
-                DEBUG_ASSERT(pointer);
-                u8* const dest_ptr = pointer + page_offset + (page_index << PAGE_BITS);
-                std::memcpy(dest_ptr, src_buffer, copy_amount);
-                break;
-            }
-            case Common::PageType::RasterizerCachedMemory: {
-                u8* const host_ptr{GetPointerFromRasterizerCachedMemory(current_vaddr)};
-                std::memcpy(host_ptr, src_buffer, copy_amount);
-                break;
-            }
-            default:
-                UNREACHABLE();
-            }
-
-            page_index++;
-            page_offset = 0;
-            src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
-            remaining_size -= copy_amount;
-        }
+            },
+            [&src_buffer](const std::size_t copy_amount) {
+                src_buffer = static_cast<const u8*>(src_buffer) + copy_amount;
+            });
     }
 
     void WriteBlock(const VAddr dest_addr, const void* src_buffer, const std::size_t size) {
-        WriteBlock(*system.CurrentProcess(), dest_addr, src_buffer, size);
+        WriteBlockImpl<false>(*system.CurrentProcess(), dest_addr, src_buffer, size);
     }
 
     void WriteBlockUnsafe(const VAddr dest_addr, const void* src_buffer, const std::size_t size) {
-        WriteBlockUnsafe(*system.CurrentProcess(), dest_addr, src_buffer, size);
+        WriteBlockImpl<true>(*system.CurrentProcess(), dest_addr, src_buffer, size);
     }
 
     void ZeroBlock(const Kernel::KProcess& process, const VAddr dest_addr, const std::size_t size) {
-        const auto& page_table = process.PageTable().PageTableImpl();
-        std::size_t remaining_size = size;
-        std::size_t page_index = dest_addr >> PAGE_BITS;
-        std::size_t page_offset = dest_addr & PAGE_MASK;
-
-        while (remaining_size > 0) {
-            const std::size_t copy_amount =
-                std::min(static_cast<std::size_t>(PAGE_SIZE) - page_offset, remaining_size);
-            const auto current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
-
-            const auto [pointer, type] = page_table.pointers[page_index].PointerType();
-            switch (type) {
-            case Common::PageType::Unmapped: {
+        WalkBlock(
+            process, dest_addr, size,
+            [dest_addr, size](const std::size_t copy_amount, const VAddr current_vaddr) {
                 LOG_ERROR(HW_Memory,
                           "Unmapped ZeroBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
                           current_vaddr, dest_addr, size);
-                break;
-            }
-            case Common::PageType::Memory: {
-                DEBUG_ASSERT(pointer);
-                u8* const dest_ptr = pointer + page_offset + (page_index << PAGE_BITS);
+            },
+            [](const std::size_t copy_amount, u8* const dest_ptr) {
                 std::memset(dest_ptr, 0, copy_amount);
-                break;
-            }
-            case Common::PageType::RasterizerCachedMemory: {
-                u8* const host_ptr{GetPointerFromRasterizerCachedMemory(current_vaddr)};
+            },
+            [&system = system](const VAddr current_vaddr, const std::size_t copy_amount,
+                               u8* const host_ptr) {
                 system.GPU().InvalidateRegion(current_vaddr, copy_amount);
                 std::memset(host_ptr, 0, copy_amount);
-                break;
-            }
-            default:
-                UNREACHABLE();
-            }
-
-            page_index++;
-            page_offset = 0;
-            remaining_size -= copy_amount;
-        }
-    }
-
-    void ZeroBlock(const VAddr dest_addr, const std::size_t size) {
-        ZeroBlock(*system.CurrentProcess(), dest_addr, size);
+            },
+            [](const std::size_t copy_amount) {});
     }
 
     void CopyBlock(const Kernel::KProcess& process, VAddr dest_addr, VAddr src_addr,
                    const std::size_t size) {
-        const auto& page_table = process.PageTable().PageTableImpl();
-        std::size_t remaining_size = size;
-        std::size_t page_index = src_addr >> PAGE_BITS;
-        std::size_t page_offset = src_addr & PAGE_MASK;
-
-        while (remaining_size > 0) {
-            const std::size_t copy_amount =
-                std::min(static_cast<std::size_t>(PAGE_SIZE) - page_offset, remaining_size);
-            const auto current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
-
-            const auto [pointer, type] = page_table.pointers[page_index].PointerType();
-            switch (type) {
-            case Common::PageType::Unmapped: {
+        WalkBlock(
+            process, dest_addr, size,
+            [this, &process, &dest_addr, &src_addr, size](const std::size_t copy_amount,
+                                                          const VAddr current_vaddr) {
                 LOG_ERROR(HW_Memory,
                           "Unmapped CopyBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
                           current_vaddr, src_addr, size);
                 ZeroBlock(process, dest_addr, copy_amount);
-                break;
-            }
-            case Common::PageType::Memory: {
-                DEBUG_ASSERT(pointer);
-                const u8* src_ptr = pointer + page_offset + (page_index << PAGE_BITS);
-                WriteBlock(process, dest_addr, src_ptr, copy_amount);
-                break;
-            }
-            case Common::PageType::RasterizerCachedMemory: {
-                const u8* const host_ptr{GetPointerFromRasterizerCachedMemory(current_vaddr)};
+            },
+            [this, &process, &dest_addr](const std::size_t copy_amount, const u8* const src_ptr) {
+                WriteBlockImpl<false>(process, dest_addr, src_ptr, copy_amount);
+            },
+            [this, &system = system, &process, &dest_addr](
+                const VAddr current_vaddr, const std::size_t copy_amount, u8* const host_ptr) {
                 system.GPU().FlushRegion(current_vaddr, copy_amount);
-                WriteBlock(process, dest_addr, host_ptr, copy_amount);
-                break;
-            }
-            default:
-                UNREACHABLE();
-            }
-
-            page_index++;
-            page_offset = 0;
-            dest_addr += static_cast<VAddr>(copy_amount);
-            src_addr += static_cast<VAddr>(copy_amount);
-            remaining_size -= copy_amount;
-        }
-    }
-
-    void CopyBlock(VAddr dest_addr, VAddr src_addr, std::size_t size) {
-        return CopyBlock(*system.CurrentProcess(), dest_addr, src_addr, size);
+                WriteBlockImpl<false>(process, dest_addr, host_ptr, copy_amount);
+            },
+            [&dest_addr, &src_addr](const std::size_t copy_amount) {
+                dest_addr += static_cast<VAddr>(copy_amount);
+                src_addr += static_cast<VAddr>(copy_amount);
+            });
     }
 
     void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) {
@@ -514,7 +357,7 @@ struct Memory::Impl {
             } else {
                 // Switch page type to uncached if now uncached
                 switch (page_type) {
-                case Common::PageType::Unmapped:
+                case Common::PageType::Unmapped: // NOLINT(bugprone-branch-clone)
                     // It is not necessary for a process to have this region mapped into its address
                     // space, for example, a system module need not have a VRAM mapping.
                     break;
@@ -597,6 +440,44 @@ struct Memory::Impl {
         }
     }
 
+    [[nodiscard]] u8* GetPointerImpl(VAddr vaddr, auto on_unmapped, auto on_rasterizer) const {
+        // AARCH64 masks the upper 16 bit of all memory accesses
+        vaddr &= 0xffffffffffffLL;
+
+        if (vaddr >= 1uLL << current_page_table->GetAddressSpaceBits()) {
+            on_unmapped();
+            return nullptr;
+        }
+
+        // Avoid adding any extra logic to this fast-path block
+        const uintptr_t raw_pointer = current_page_table->pointers[vaddr >> PAGE_BITS].Raw();
+        if (u8* const pointer = Common::PageTable::PageInfo::ExtractPointer(raw_pointer)) {
+            return &pointer[vaddr];
+        }
+        switch (Common::PageTable::PageInfo::ExtractType(raw_pointer)) {
+        case Common::PageType::Unmapped:
+            on_unmapped();
+            return nullptr;
+        case Common::PageType::Memory:
+            ASSERT_MSG(false, "Mapped memory page without a pointer @ 0x{:016X}", vaddr);
+            return nullptr;
+        case Common::PageType::RasterizerCachedMemory: {
+            u8* const host_ptr{GetPointerFromRasterizerCachedMemory(vaddr)};
+            on_rasterizer();
+            return host_ptr;
+        }
+        default:
+            UNREACHABLE();
+        }
+        return nullptr;
+    }
+
+    [[nodiscard]] u8* GetPointer(const VAddr vaddr) const {
+        return GetPointerImpl(
+            vaddr, [vaddr]() { LOG_ERROR(HW_Memory, "Unmapped GetPointer @ 0x{:016X}", vaddr); },
+            []() {});
+    }
+
     /**
      * Reads a particular data type out of memory at the given virtual address.
      *
@@ -610,39 +491,17 @@ struct Memory::Impl {
      */
     template <typename T>
     T Read(VAddr vaddr) {
-        // AARCH64 masks the upper 16 bit of all memory accesses
-        vaddr &= 0xffffffffffffLL;
-
-        if (vaddr >= 1uLL << current_page_table->GetAddressSpaceBits()) {
-            LOG_ERROR(HW_Memory, "Unmapped Read{} @ 0x{:08X}", sizeof(T) * 8, vaddr);
-            return 0;
+        T result = 0;
+        const u8* const ptr = GetPointerImpl(
+            vaddr,
+            [vaddr]() {
+                LOG_ERROR(HW_Memory, "Unmapped Read{} @ 0x{:016X}", sizeof(T) * 8, vaddr);
+            },
+            [&system = system, vaddr]() { system.GPU().FlushRegion(vaddr, sizeof(T)); });
+        if (ptr) {
+            std::memcpy(&result, ptr, sizeof(T));
         }
-
-        // Avoid adding any extra logic to this fast-path block
-        const uintptr_t raw_pointer = current_page_table->pointers[vaddr >> PAGE_BITS].Raw();
-        if (const u8* const pointer = Common::PageTable::PageInfo::ExtractPointer(raw_pointer)) {
-            T value;
-            std::memcpy(&value, &pointer[vaddr], sizeof(T));
-            return value;
-        }
-        switch (Common::PageTable::PageInfo::ExtractType(raw_pointer)) {
-        case Common::PageType::Unmapped:
-            LOG_ERROR(HW_Memory, "Unmapped Read{} @ 0x{:08X}", sizeof(T) * 8, vaddr);
-            return 0;
-        case Common::PageType::Memory:
-            ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr);
-            break;
-        case Common::PageType::RasterizerCachedMemory: {
-            const u8* const host_ptr{GetPointerFromRasterizerCachedMemory(vaddr)};
-            system.GPU().FlushRegion(vaddr, sizeof(T));
-            T value;
-            std::memcpy(&value, host_ptr, sizeof(T));
-            return value;
-        }
-        default:
-            UNREACHABLE();
-        }
-        return {};
+        return result;
     }
 
     /**
@@ -656,110 +515,46 @@ struct Memory::Impl {
      */
     template <typename T>
     void Write(VAddr vaddr, const T data) {
-        // AARCH64 masks the upper 16 bit of all memory accesses
-        vaddr &= 0xffffffffffffLL;
-
-        if (vaddr >= 1uLL << current_page_table->GetAddressSpaceBits()) {
-            LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8,
-                      static_cast<u32>(data), vaddr);
-            return;
-        }
-
-        // Avoid adding any extra logic to this fast-path block
-        const uintptr_t raw_pointer = current_page_table->pointers[vaddr >> PAGE_BITS].Raw();
-        if (u8* const pointer = Common::PageTable::PageInfo::ExtractPointer(raw_pointer)) {
-            std::memcpy(&pointer[vaddr], &data, sizeof(T));
-            return;
-        }
-        switch (Common::PageTable::PageInfo::ExtractType(raw_pointer)) {
-        case Common::PageType::Unmapped:
-            LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8,
-                      static_cast<u32>(data), vaddr);
-            return;
-        case Common::PageType::Memory:
-            ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr);
-            break;
-        case Common::PageType::RasterizerCachedMemory: {
-            u8* const host_ptr{GetPointerFromRasterizerCachedMemory(vaddr)};
-            system.GPU().InvalidateRegion(vaddr, sizeof(T));
-            std::memcpy(host_ptr, &data, sizeof(T));
-            break;
-        }
-        default:
-            UNREACHABLE();
+        u8* const ptr = GetPointerImpl(
+            vaddr,
+            [vaddr, data]() {
+                LOG_ERROR(HW_Memory, "Unmapped Write{} @ 0x{:016X} = 0x{:016X}", sizeof(T) * 8,
+                          vaddr, static_cast<u64>(data));
+            },
+            [&system = system, vaddr]() { system.GPU().InvalidateRegion(vaddr, sizeof(T)); });
+        if (ptr) {
+            std::memcpy(ptr, &data, sizeof(T));
         }
     }
 
     template <typename T>
     bool WriteExclusive(VAddr vaddr, const T data, const T expected) {
-        // AARCH64 masks the upper 16 bit of all memory accesses
-        vaddr &= 0xffffffffffffLL;
-
-        if (vaddr >= 1uLL << current_page_table->GetAddressSpaceBits()) {
-            LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8,
-                      static_cast<u32>(data), vaddr);
-            return true;
-        }
-
-        const uintptr_t raw_pointer = current_page_table->pointers[vaddr >> PAGE_BITS].Raw();
-        if (u8* const pointer = Common::PageTable::PageInfo::ExtractPointer(raw_pointer)) {
-            // NOTE: Avoid adding any extra logic to this fast-path block
-            const auto volatile_pointer = reinterpret_cast<volatile T*>(&pointer[vaddr]);
+        u8* const ptr = GetPointerImpl(
+            vaddr,
+            [vaddr, data]() {
+                LOG_ERROR(HW_Memory, "Unmapped WriteExclusive{} @ 0x{:016X} = 0x{:016X}",
+                          sizeof(T) * 8, vaddr, static_cast<u64>(data));
+            },
+            [&system = system, vaddr]() { system.GPU().InvalidateRegion(vaddr, sizeof(T)); });
+        if (ptr) {
+            const auto volatile_pointer = reinterpret_cast<volatile T*>(ptr);
             return Common::AtomicCompareAndSwap(volatile_pointer, data, expected);
         }
-        switch (Common::PageTable::PageInfo::ExtractType(raw_pointer)) {
-        case Common::PageType::Unmapped:
-            LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8,
-                      static_cast<u32>(data), vaddr);
-            return true;
-        case Common::PageType::Memory:
-            ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr);
-            break;
-        case Common::PageType::RasterizerCachedMemory: {
-            u8* host_ptr{GetPointerFromRasterizerCachedMemory(vaddr)};
-            system.GPU().InvalidateRegion(vaddr, sizeof(T));
-            auto* pointer = reinterpret_cast<volatile T*>(&host_ptr);
-            return Common::AtomicCompareAndSwap(pointer, data, expected);
-        }
-        default:
-            UNREACHABLE();
-        }
         return true;
     }
 
     bool WriteExclusive128(VAddr vaddr, const u128 data, const u128 expected) {
-        // AARCH64 masks the upper 16 bit of all memory accesses
-        vaddr &= 0xffffffffffffLL;
-
-        if (vaddr >= 1uLL << current_page_table->GetAddressSpaceBits()) {
-            LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8,
-                      static_cast<u32>(data[0]), vaddr);
-            return true;
-        }
-
-        const uintptr_t raw_pointer = current_page_table->pointers[vaddr >> PAGE_BITS].Raw();
-        if (u8* const pointer = Common::PageTable::PageInfo::ExtractPointer(raw_pointer)) {
-            // NOTE: Avoid adding any extra logic to this fast-path block
-            const auto volatile_pointer = reinterpret_cast<volatile u64*>(&pointer[vaddr]);
+        u8* const ptr = GetPointerImpl(
+            vaddr,
+            [vaddr, data]() {
+                LOG_ERROR(HW_Memory, "Unmapped WriteExclusive128 @ 0x{:016X} = 0x{:016X}{:016X}",
+                          vaddr, static_cast<u64>(data[1]), static_cast<u64>(data[0]));
+            },
+            [&system = system, vaddr]() { system.GPU().InvalidateRegion(vaddr, sizeof(u128)); });
+        if (ptr) {
+            const auto volatile_pointer = reinterpret_cast<volatile u64*>(ptr);
             return Common::AtomicCompareAndSwap(volatile_pointer, data, expected);
         }
-        switch (Common::PageTable::PageInfo::ExtractType(raw_pointer)) {
-        case Common::PageType::Unmapped:
-            LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}{:016X}", sizeof(data) * 8,
-                      static_cast<u64>(data[1]), static_cast<u64>(data[0]), vaddr);
-            return true;
-        case Common::PageType::Memory:
-            ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr);
-            break;
-        case Common::PageType::RasterizerCachedMemory: {
-            u8* host_ptr{GetPointerFromRasterizerCachedMemory(vaddr)};
-            system.GPU().InvalidateRegion(vaddr, sizeof(u128));
-            auto* pointer = reinterpret_cast<volatile u64*>(&host_ptr);
-            return Common::AtomicCompareAndSwap(pointer, data, expected);
-        }
-        default:
-            UNREACHABLE();
-        }
         return true;
     }
 
@@ -789,12 +584,11 @@ void Memory::UnmapRegion(Common::PageTable& page_table, VAddr base, u64 size) {
     impl->UnmapRegion(page_table, base, size);
 }
 
-bool Memory::IsValidVirtualAddress(const Kernel::KProcess& process, const VAddr vaddr) const {
-    return impl->IsValidVirtualAddress(process, vaddr);
-}
-
 bool Memory::IsValidVirtualAddress(const VAddr vaddr) const {
-    return impl->IsValidVirtualAddress(vaddr);
+    const Kernel::KProcess& process = *system.CurrentProcess();
+    const auto& page_table = process.PageTable().PageTableImpl();
+    const auto [pointer, type] = page_table.pointers[vaddr >> PAGE_BITS].PointerType();
+    return pointer != nullptr || type == Common::PageType::RasterizerCachedMemory;
 }
 
 u8* Memory::GetPointer(VAddr vaddr) {
@@ -863,64 +657,38 @@ std::string Memory::ReadCString(VAddr vaddr, std::size_t max_length) {
 
 void Memory::ReadBlock(const Kernel::KProcess& process, const VAddr src_addr, void* dest_buffer,
                        const std::size_t size) {
-    impl->ReadBlock(process, src_addr, dest_buffer, size);
+    impl->ReadBlockImpl<false>(process, src_addr, dest_buffer, size);
 }
 
 void Memory::ReadBlock(const VAddr src_addr, void* dest_buffer, const std::size_t size) {
     impl->ReadBlock(src_addr, dest_buffer, size);
 }
 
-void Memory::ReadBlockUnsafe(const Kernel::KProcess& process, const VAddr src_addr,
-                             void* dest_buffer, const std::size_t size) {
-    impl->ReadBlockUnsafe(process, src_addr, dest_buffer, size);
-}
-
 void Memory::ReadBlockUnsafe(const VAddr src_addr, void* dest_buffer, const std::size_t size) {
     impl->ReadBlockUnsafe(src_addr, dest_buffer, size);
 }
 
 void Memory::WriteBlock(const Kernel::KProcess& process, VAddr dest_addr, const void* src_buffer,
                         std::size_t size) {
-    impl->WriteBlock(process, dest_addr, src_buffer, size);
+    impl->WriteBlockImpl<false>(process, dest_addr, src_buffer, size);
 }
 
 void Memory::WriteBlock(const VAddr dest_addr, const void* src_buffer, const std::size_t size) {
     impl->WriteBlock(dest_addr, src_buffer, size);
 }
 
-void Memory::WriteBlockUnsafe(const Kernel::KProcess& process, VAddr dest_addr,
-                              const void* src_buffer, std::size_t size) {
-    impl->WriteBlockUnsafe(process, dest_addr, src_buffer, size);
-}
-
 void Memory::WriteBlockUnsafe(const VAddr dest_addr, const void* src_buffer,
                               const std::size_t size) {
     impl->WriteBlockUnsafe(dest_addr, src_buffer, size);
 }
 
-void Memory::ZeroBlock(const Kernel::KProcess& process, VAddr dest_addr, std::size_t size) {
-    impl->ZeroBlock(process, dest_addr, size);
-}
-
-void Memory::ZeroBlock(VAddr dest_addr, std::size_t size) {
-    impl->ZeroBlock(dest_addr, size);
-}
-
 void Memory::CopyBlock(const Kernel::KProcess& process, VAddr dest_addr, VAddr src_addr,
                        const std::size_t size) {
     impl->CopyBlock(process, dest_addr, src_addr, size);
 }
 
-void Memory::CopyBlock(VAddr dest_addr, VAddr src_addr, std::size_t size) {
-    impl->CopyBlock(dest_addr, src_addr, size);
-}
-
 void Memory::RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) {
     impl->RasterizerMarkRegionCached(vaddr, size, cached);
 }
 
-bool IsKernelVirtualAddress(const VAddr vaddr) {
-    return KERNEL_REGION_VADDR <= vaddr && vaddr < KERNEL_REGION_END;
-}
-
 } // namespace Core::Memory

src/core/memory.h

@@ -39,11 +39,6 @@ enum : VAddr {
 
     /// Application stack
     DEFAULT_STACK_SIZE = 0x100000,
-
-    /// Kernel Virtual Address Range
-    KERNEL_REGION_VADDR = 0xFFFFFF8000000000,
-    KERNEL_REGION_SIZE = 0x7FFFE00000,
-    KERNEL_REGION_END = KERNEL_REGION_VADDR + KERNEL_REGION_SIZE,
 };
 
 /// Central class that handles all memory operations and state.
@@ -56,7 +51,7 @@ public:
     Memory& operator=(const Memory&) = delete;
 
     Memory(Memory&&) = default;
-    Memory& operator=(Memory&&) = default;
+    Memory& operator=(Memory&&) = delete;
 
     /**
      * Resets the state of the Memory system.
@@ -90,17 +85,6 @@ public:
      */
     void UnmapRegion(Common::PageTable& page_table, VAddr base, u64 size);
 
-    /**
-     * Checks whether or not the supplied address is a valid virtual
-     * address for the given process.
-     *
-     * @param process The emulated process to check the address against.
-     * @param vaddr   The virtual address to check the validity of.
-     *
-     * @returns True if the given virtual address is valid, false otherwise.
-     */
-    bool IsValidVirtualAddress(const Kernel::KProcess& process, VAddr vaddr) const;
-
     /**
      * Checks whether or not the supplied address is a valid virtual
      * address for the current process.
@@ -109,7 +93,7 @@ public:
      *
      * @returns True if the given virtual address is valid, false otherwise.
      */
-    bool IsValidVirtualAddress(VAddr vaddr) const;
+    [[nodiscard]] bool IsValidVirtualAddress(VAddr vaddr) const;
 
     /**
      * Gets a pointer to the given address.
@@ -134,7 +118,7 @@ public:
      * @returns The pointer to the given address, if the address is valid.
      *          If the address is not valid, nullptr will be returned.
      */
-    const u8* GetPointer(VAddr vaddr) const;
+    [[nodiscard]] const u8* GetPointer(VAddr vaddr) const;
 
     template <typename T>
     const T* GetPointer(VAddr vaddr) const {
@@ -327,27 +311,6 @@ public:
     void ReadBlock(const Kernel::KProcess& process, VAddr src_addr, void* dest_buffer,
                    std::size_t size);
 
-    /**
-     * Reads a contiguous block of bytes from a specified process' address space.
-     * This unsafe version does not trigger GPU flushing.
-     *
-     * @param process     The process to read the data from.
-     * @param src_addr    The virtual address to begin reading from.
-     * @param dest_buffer The buffer to place the read bytes into.
-     * @param size        The amount of data to read, in bytes.
-     *
-     * @note If a size of 0 is specified, then this function reads nothing and
-     *       no attempts to access memory are made at all.
-     *
-     * @pre dest_buffer must be at least size bytes in length, otherwise a
-     *      buffer overrun will occur.
-     *
-     * @post The range [dest_buffer, size) contains the read bytes from the
-     *       process' address space.
-     */
-    void ReadBlockUnsafe(const Kernel::KProcess& process, VAddr src_addr, void* dest_buffer,
-                         std::size_t size);
-
     /**
      * Reads a contiguous block of bytes from the current process' address space.
      *
@@ -408,26 +371,6 @@ public:
     void WriteBlock(const Kernel::KProcess& process, VAddr dest_addr, const void* src_buffer,
                     std::size_t size);
 
-    /**
-     * Writes a range of bytes into a given process' address space at the specified
-     * virtual address.
-     * This unsafe version does not invalidate GPU Memory.
-     *
-     * @param process    The process to write data into the address space of.
-     * @param dest_addr  The destination virtual address to begin writing the data at.
-     * @param src_buffer The data to write into the process' address space.
-     * @param size       The size of the data to write, in bytes.
-     *
-     * @post The address range [dest_addr, size) in the process' address space
-     *       contains the data that was within src_buffer.
-     *
-     * @post If an attempt is made to write into an unmapped region of memory, the writes
-     *       will be ignored and an error will be logged.
-     *
-     */
-    void WriteBlockUnsafe(const Kernel::KProcess& process, VAddr dest_addr, const void* src_buffer,
-                          std::size_t size);
-
     /**
      * Writes a range of bytes into the current process' address space at the specified
      * virtual address.
@@ -467,29 +410,6 @@ public:
      */
     void WriteBlockUnsafe(VAddr dest_addr, const void* src_buffer, std::size_t size);
 
-    /**
-     * Fills the specified address range within a process' address space with zeroes.
-     *
-     * @param process   The process that will have a portion of its memory zeroed out.
-     * @param dest_addr The starting virtual address of the range to zero out.
-     * @param size      The size of the address range to zero out, in bytes.
-     *
-     * @post The range [dest_addr, size) within the process' address space is
-     *       filled with zeroes.
-     */
-    void ZeroBlock(const Kernel::KProcess& process, VAddr dest_addr, std::size_t size);
-
-    /**
-     * Fills the specified address range within the current process' address space with zeroes.
-     *
-     * @param dest_addr The starting virtual address of the range to zero out.
-     * @param size      The size of the address range to zero out, in bytes.
-     *
-     * @post The range [dest_addr, size) within the current process' address space is
-     *       filled with zeroes.
-     */
-    void ZeroBlock(VAddr dest_addr, std::size_t size);
-
     /**
      * Copies data within a process' address space to another location within the
      * same address space.
@@ -505,19 +425,6 @@ public:
     void CopyBlock(const Kernel::KProcess& process, VAddr dest_addr, VAddr src_addr,
                    std::size_t size);
 
-    /**
-     * Copies data within the current process' address space to another location within the
-     * same address space.
-     *
-     * @param dest_addr The destination virtual address to begin copying the data into.
-     * @param src_addr  The source virtual address to begin copying the data from.
-     * @param size      The size of the data to copy, in bytes.
-     *
-     * @post The range [dest_addr, size) within the current process' address space
-     *       contains the same data within the range [src_addr, size).
-     */
-    void CopyBlock(VAddr dest_addr, VAddr src_addr, std::size_t size);
-
     /**
      * Marks each page within the specified address range as cached or uncached.
      *
@@ -535,7 +442,4 @@ private:
     std::unique_ptr<Impl> impl;
 };
 
-/// Determines if the given VAddr is a kernel address
-bool IsKernelVirtualAddress(VAddr vaddr);
-
 } // namespace Core::Memory

src/core/network/network.cpp

@@ -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) {

src/core/perf_stats.cpp

@@ -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;
     }
 

src/core/perf_stats.h

@@ -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

src/core/telemetry_session.cpp

@@ -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",

src/input_common/main.cpp

@@ -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();
 }
 

src/input_common/mouse/mouse_poller.cpp

@@ -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;
         }
     }

src/input_common/sdl/sdl_impl.cpp

@@ -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;
             }

src/shader_recompiler/backend/glasm/emit_glasm.cpp

@@ -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:

src/shader_recompiler/exception.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)...)} {}
 };
 

src/shader_recompiler/frontend/ir/ir_emitter.h

@@ -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);

src/shader_recompiler/frontend/ir/value.h

@@ -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();

src/shader_recompiler/frontend/maxwell/control_flow.cpp

@@ -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;

src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp

@@ -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:

src/shader_recompiler/ir_opt/constant_propagation_pass.cpp

@@ -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:

src/shader_recompiler/object_pool.h

@@ -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

src/video_core/CMakeLists.txt

@@ -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
@@ -92,6 +97,7 @@ add_library(video_core STATIC
     renderer_opengl/gl_stream_buffer.h
     renderer_opengl/gl_texture_cache.cpp
     renderer_opengl/gl_texture_cache.h
+    renderer_opengl/gl_texture_cache_base.cpp
     renderer_opengl/gl_query_cache.cpp
     renderer_opengl/gl_query_cache.h
     renderer_opengl/maxwell_to_gl.h
@@ -106,6 +112,8 @@ 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
@@ -148,6 +156,7 @@ add_library(video_core STATIC
     renderer_vulkan/vk_swapchain.h
     renderer_vulkan/vk_texture_cache.cpp
     renderer_vulkan/vk_texture_cache.h
+    renderer_vulkan/vk_texture_cache_base.cpp
     renderer_vulkan/vk_update_descriptor.cpp
     renderer_vulkan/vk_update_descriptor.h
     shader_cache.cpp
@@ -179,6 +188,7 @@ add_library(video_core STATIC
     texture_cache/samples_helper.h
     texture_cache/slot_vector.h
     texture_cache/texture_cache.h
+    texture_cache/texture_cache_base.h
     texture_cache/types.h
     texture_cache/util.cpp
     texture_cache/util.h

src/video_core/buffer_cache/buffer_cache.h

@@ -817,7 +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();

src/video_core/command_classes/codecs/codec.cpp

@@ -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

src/video_core/command_classes/codecs/codec.h

@@ -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;

src/video_core/command_classes/codecs/vp9.cpp

@@ -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;
 }
 

src/video_core/command_classes/codecs/vp9.h

@@ -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

src/video_core/command_classes/codecs/vp9_types.h

@@ -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")

src/video_core/command_classes/nvdec.cpp

@@ -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;
     }
 }

src/video_core/command_classes/vic.cpp

@@ -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;
     }

src/video_core/command_classes/vic.h

@@ -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{};

src/video_core/engines/fermi_2d.h

@@ -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.

src/video_core/engines/maxwell_dma.h

@@ -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;

src/video_core/host_shaders/astc_decoder.comp

@@ -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);
 }

src/video_core/renderer_base.cpp

@@ -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");

src/video_core/renderer_base.h

@@ -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:

src/video_core/renderer_opengl/gl_graphics_pipeline.cpp

@@ -15,7 +15,7 @@
 #include "video_core/renderer_opengl/gl_shader_util.h"
 #include "video_core/renderer_opengl/gl_state_tracker.h"
 #include "video_core/shader_notify.h"
-#include "video_core/texture_cache/texture_cache.h"
+#include "video_core/texture_cache/texture_cache_base.h"
 
 #if defined(_MSC_VER) && defined(NDEBUG)
 #define LAMBDA_FORCEINLINE [[msvc::forceinline]]

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -32,7 +32,7 @@
 #include "video_core/renderer_opengl/maxwell_to_gl.h"
 #include "video_core/renderer_opengl/renderer_opengl.h"
 #include "video_core/shader_cache.h"
-#include "video_core/texture_cache/texture_cache.h"
+#include "video_core/texture_cache/texture_cache_base.h"
 
 namespace OpenGL {
 

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -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;

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -18,10 +18,8 @@
 #include "video_core/renderer_opengl/maxwell_to_gl.h"
 #include "video_core/renderer_opengl/util_shaders.h"
 #include "video_core/surface.h"
-#include "video_core/texture_cache/format_lookup_table.h"
+#include "video_core/texture_cache/formatter.h"
 #include "video_core/texture_cache/samples_helper.h"
-#include "video_core/texture_cache/texture_cache.h"
-#include "video_core/textures/decoders.h"
 
 namespace OpenGL {
 namespace {

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -12,7 +12,7 @@
 #include "shader_recompiler/shader_info.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/renderer_opengl/util_shaders.h"
-#include "video_core/texture_cache/texture_cache.h"
+#include "video_core/texture_cache/texture_cache_base.h"
 
 namespace OpenGL {
 

src/video_core/renderer_opengl/gl_texture_cache_base.cpp

@@ -0,0 +1,10 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "video_core/renderer_opengl/gl_texture_cache.h"
+#include "video_core/texture_cache/texture_cache.h"
+
+namespace VideoCommon {
+template class VideoCommon::TextureCache<OpenGL::TextureCacheParams>;
+}

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -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;
 }
 

src/video_core/renderer_opengl/util_shaders.cpp

@@ -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);
     }

src/video_core/renderer_opengl/util_shaders.h

@@ -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;

src/video_core/renderer_vulkan/pipeline_statistics.cpp

@@ -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

src/video_core/renderer_vulkan/pipeline_statistics.h

@@ -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

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -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

src/video_core/renderer_vulkan/renderer_vulkan.h

@@ -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;

src/video_core/renderer_vulkan/vk_blit_screen.cpp

@@ -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());

src/video_core/renderer_vulkan/vk_blit_screen.h

@@ -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,

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -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);
 

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -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,

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -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

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -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();

src/video_core/renderer_vulkan/vk_compute_pipeline.h

@@ -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);
 

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -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,

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -60,6 +60,7 @@ struct hash<Vulkan::GraphicsPipelineCacheKey> {
 namespace Vulkan {
 
 class Device;
+class PipelineStatistics;
 class RenderPassCache;
 class VKScheduler;
 class VKUpdateDescriptorQueue;
@@ -73,8 +74,9 @@ public:
         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,
+        Common::ThreadWorker* worker_thread, 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);
 
     GraphicsPipeline& operator=(GraphicsPipeline&&) noexcept = delete;

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -29,6 +29,7 @@
 #include "video_core/renderer_vulkan/fixed_pipeline_state.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_compute_pipeline.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
 #include "video_core/renderer_vulkan/vk_pipeline_cache.h"
@@ -389,15 +390,19 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
         size_t total{};
         size_t built{};
         bool has_loaded{};
+        std::unique_ptr<PipelineStatistics> statistics;
     } state;
 
+    if (device.IsKhrPipelineEexecutablePropertiesEnabled()) {
+        state.statistics = std::make_unique<PipelineStatistics>(device);
+    }
     const auto load_compute{[&](std::ifstream& file, FileEnvironment env) {
         ComputePipelineCacheKey key;
         file.read(reinterpret_cast<char*>(&key), sizeof(key));
 
         workers.QueueWork([this, key, env = std::move(env), &state, &callback]() mutable {
             ShaderPools pools;
-            auto pipeline{CreateComputePipeline(pools, key, env, false)};
+            auto pipeline{CreateComputePipeline(pools, key, env, state.statistics.get(), false)};
             std::lock_guard lock{state.mutex};
             if (pipeline) {
                 compute_cache.emplace(key, std::move(pipeline));
@@ -425,7 +430,8 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
             for (auto& env : envs) {
                 env_ptrs.push_back(&env);
             }
-            auto pipeline{CreateGraphicsPipeline(pools, key, MakeSpan(env_ptrs), false)};
+            auto pipeline{CreateGraphicsPipeline(pools, key, MakeSpan(env_ptrs),
+                                                 state.statistics.get(), false)};
 
             std::lock_guard lock{state.mutex};
             graphics_cache.emplace(key, std::move(pipeline));
@@ -445,6 +451,10 @@ void PipelineCache::LoadDiskResources(u64 title_id, std::stop_token stop_loading
     lock.unlock();
 
     workers.WaitForRequests();
+
+    if (state.statistics) {
+        state.statistics->Report();
+    }
 }
 
 GraphicsPipeline* PipelineCache::CurrentGraphicsPipelineSlowPath() {
@@ -486,7 +496,8 @@ GraphicsPipeline* PipelineCache::BuiltPipeline(GraphicsPipeline* pipeline) const
 
 std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
     ShaderPools& pools, const GraphicsPipelineCacheKey& key,
-    std::span<Shader::Environment* const> envs, bool build_in_parallel) try {
+    std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
+    bool build_in_parallel) try {
     LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
     size_t env_index{0};
     std::array<Shader::IR::Program, Maxwell::MaxShaderProgram> programs;
@@ -540,7 +551,7 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline(
     Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
     return std::make_unique<GraphicsPipeline>(
         maxwell3d, gpu_memory, scheduler, buffer_cache, texture_cache, &shader_notify, device,
-        descriptor_pool, update_descriptor_queue, thread_worker, render_pass_cache, key,
+        descriptor_pool, update_descriptor_queue, thread_worker, statistics, render_pass_cache, key,
         std::move(modules), infos);
 
 } catch (const Shader::Exception& exception) {
@@ -553,7 +564,8 @@ std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() {
     GetGraphicsEnvironments(environments, graphics_key.unique_hashes);
 
     main_pools.ReleaseContents();
-    auto pipeline{CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(), true)};
+    auto pipeline{
+        CreateGraphicsPipeline(main_pools, graphics_key, environments.Span(), nullptr, true)};
     if (!pipeline || pipeline_cache_filename.empty()) {
         return pipeline;
     }
@@ -578,7 +590,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
     env.SetCachedSize(shader->size_bytes);
 
     main_pools.ReleaseContents();
-    auto pipeline{CreateComputePipeline(main_pools, key, env, true)};
+    auto pipeline{CreateComputePipeline(main_pools, key, env, nullptr, true)};
     if (!pipeline || pipeline_cache_filename.empty()) {
         return pipeline;
     }
@@ -591,7 +603,7 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
 
 std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
     ShaderPools& pools, const ComputePipelineCacheKey& key, Shader::Environment& env,
-    bool build_in_parallel) try {
+    PipelineStatistics* statistics, bool build_in_parallel) try {
     LOG_INFO(Render_Vulkan, "0x{:016x}", key.Hash());
 
     Shader::Maxwell::Flow::CFG cfg{env, pools.flow_block, env.StartAddress()};
@@ -605,8 +617,8 @@ std::unique_ptr<ComputePipeline> PipelineCache::CreateComputePipeline(
     }
     Common::ThreadWorker* const thread_worker{build_in_parallel ? &workers : nullptr};
     return std::make_unique<ComputePipeline>(device, descriptor_pool, update_descriptor_queue,
-                                             thread_worker, &shader_notify, program.info,
-                                             std::move(spv_module));
+                                             thread_worker, statistics, &shader_notify,
+                                             program.info, std::move(spv_module));
 
 } catch (const Shader::Exception& exception) {
     LOG_ERROR(Render_Vulkan, "{}", exception.what());

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -80,8 +80,9 @@ struct hash<Vulkan::ComputePipelineCacheKey> {
 namespace Vulkan {
 
 class ComputePipeline;
-class Device;
 class DescriptorPool;
+class Device;
+class PipelineStatistics;
 class RasterizerVulkan;
 class RenderPassCache;
 class VKScheduler;
@@ -128,7 +129,8 @@ private:
 
     std::unique_ptr<GraphicsPipeline> CreateGraphicsPipeline(
         ShaderPools& pools, const GraphicsPipelineCacheKey& key,
-        std::span<Shader::Environment* const> envs, bool build_in_parallel);
+        std::span<Shader::Environment* const> envs, PipelineStatistics* statistics,
+        bool build_in_parallel);
 
     std::unique_ptr<ComputePipeline> CreateComputePipeline(const ComputePipelineCacheKey& key,
                                                            const ShaderInfo* shader);
@@ -136,6 +138,7 @@ private:
     std::unique_ptr<ComputePipeline> CreateComputePipeline(ShaderPools& pools,
                                                            const ComputePipelineCacheKey& key,
                                                            Shader::Environment& env,
+                                                           PipelineStatistics* statistics,
                                                            bool build_in_parallel);
 
     const Device& device;

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -32,7 +32,7 @@
 #include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "video_core/renderer_vulkan/vk_update_descriptor.h"
 #include "video_core/shader_cache.h"
-#include "video_core/texture_cache/texture_cache.h"
+#include "video_core/texture_cache/texture_cache_base.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 
@@ -61,11 +61,16 @@ struct DrawParams {
 VkViewport GetViewportState(const Device& device, const Maxwell& regs, size_t index) {
     const auto& src = regs.viewport_transform[index];
     const float width = src.scale_x * 2.0f;
-    const float height = src.scale_y * 2.0f;
+    float y = src.translate_y - src.scale_y;
+    float height = src.scale_y * 2.0f;
+    if (regs.screen_y_control.y_negate) {
+        y += height;
+        height = -height;
+    }
     const float reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne ? 1.0f : 0.0f;
     VkViewport viewport{
         .x = src.translate_x - src.scale_x,
-        .y = src.translate_y - src.scale_y,
+        .y = y,
         .width = width != 0.0f ? width : 1.0f,
         .height = height != 0.0f ? height : 1.0f,
         .minDepth = src.translate_z - src.scale_z * reduce_z,

src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp

@@ -61,11 +61,15 @@ std::optional<u32> FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& p
     return std::nullopt;
 }
 
-u32 FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_mask) {
-    // Try to find a DEVICE_LOCAL_BIT type, Nvidia and AMD have a dedicated heap for this
-    std::optional<u32> type = FindMemoryTypeIndex(props, type_mask, STREAM_FLAGS);
-    if (type) {
-        return *type;
+u32 FindMemoryTypeIndex(const VkPhysicalDeviceMemoryProperties& props, u32 type_mask,
+                        bool try_device_local) {
+    std::optional<u32> type;
+    if (try_device_local) {
+        // Try to find a DEVICE_LOCAL_BIT type, Nvidia and AMD have a dedicated heap for this
+        type = FindMemoryTypeIndex(props, type_mask, STREAM_FLAGS);
+        if (type) {
+            return *type;
+        }
     }
     // Otherwise try without the DEVICE_LOCAL_BIT
     type = FindMemoryTypeIndex(props, type_mask, HOST_FLAGS);
@@ -115,12 +119,21 @@ StagingBufferPool::StagingBufferPool(const Device& device_, MemoryAllocator& mem
         .buffer = *stream_buffer,
     };
     const auto memory_properties = device.GetPhysical().GetMemoryProperties();
-    stream_memory = dev.AllocateMemory(VkMemoryAllocateInfo{
+    VkMemoryAllocateInfo stream_memory_info{
         .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
         .pNext = make_dedicated ? &dedicated_info : nullptr,
         .allocationSize = requirements.size,
-        .memoryTypeIndex = FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits),
-    });
+        .memoryTypeIndex =
+            FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits, true),
+    };
+    stream_memory = dev.TryAllocateMemory(stream_memory_info);
+    if (!stream_memory) {
+        LOG_INFO(Render_Vulkan, "Dynamic memory allocation failed, trying with system memory");
+        stream_memory_info.memoryTypeIndex =
+            FindMemoryTypeIndex(memory_properties, requirements.memoryTypeBits, false);
+        stream_memory = dev.AllocateMemory(stream_memory_info);
+    }
+
     if (device.HasDebuggingToolAttached()) {
         stream_memory.SetObjectNameEXT("Stream Buffer Memory");
     }

src/video_core/renderer_vulkan/vk_stream_buffer.cpp

@@ -1,168 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <algorithm>
-#include <limits>
-#include <optional>
-#include <tuple>
-#include <vector>
-
-#include "common/alignment.h"
-#include "common/assert.h"
-#include "common/literals.h"
-#include "video_core/renderer_vulkan/vk_scheduler.h"
-#include "video_core/renderer_vulkan/vk_stream_buffer.h"
-#include "video_core/vulkan_common/vulkan_device.h"
-#include "video_core/vulkan_common/vulkan_wrapper.h"
-
-namespace Vulkan {
-
-namespace {
-
-using namespace Common::Literals;
-
-constexpr VkBufferUsageFlags BUFFER_USAGE =
-    VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
-    VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
-
-constexpr u64 WATCHES_INITIAL_RESERVE = 0x4000;
-constexpr u64 WATCHES_RESERVE_CHUNK = 0x1000;
-
-constexpr u64 PREFERRED_STREAM_BUFFER_SIZE = 256_MiB;
-
-/// Find a memory type with the passed requirements
-std::optional<u32> FindMemoryType(const VkPhysicalDeviceMemoryProperties& properties,
-                                  VkMemoryPropertyFlags wanted,
-                                  u32 filter = std::numeric_limits<u32>::max()) {
-    for (u32 i = 0; i < properties.memoryTypeCount; ++i) {
-        const auto flags = properties.memoryTypes[i].propertyFlags;
-        if ((flags & wanted) == wanted && (filter & (1U << i)) != 0) {
-            return i;
-        }
-    }
-    return std::nullopt;
-}
-
-/// Get the preferred host visible memory type.
-u32 GetMemoryType(const VkPhysicalDeviceMemoryProperties& properties,
-                  u32 filter = std::numeric_limits<u32>::max()) {
-    // Prefer device local host visible allocations. Both AMD and Nvidia now provide one.
-    // Otherwise search for a host visible allocation.
-    static constexpr auto HOST_MEMORY =
-        VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
-    static constexpr auto DYNAMIC_MEMORY = HOST_MEMORY | VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
-
-    std::optional preferred_type = FindMemoryType(properties, DYNAMIC_MEMORY);
-    if (!preferred_type) {
-        preferred_type = FindMemoryType(properties, HOST_MEMORY);
-        ASSERT_MSG(preferred_type, "No host visible and coherent memory type found");
-    }
-    return preferred_type.value_or(0);
-}
-
-} // Anonymous namespace
-
-VKStreamBuffer::VKStreamBuffer(const Device& device_, VKScheduler& scheduler_)
-    : device{device_}, scheduler{scheduler_} {
-    CreateBuffers();
-    ReserveWatches(current_watches, WATCHES_INITIAL_RESERVE);
-    ReserveWatches(previous_watches, WATCHES_INITIAL_RESERVE);
-}
-
-VKStreamBuffer::~VKStreamBuffer() = default;
-
-std::pair<u8*, u64> VKStreamBuffer::Map(u64 size, u64 alignment) {
-    ASSERT(size <= stream_buffer_size);
-    mapped_size = size;
-
-    if (alignment > 0) {
-        offset = Common::AlignUp(offset, alignment);
-    }
-
-    WaitPendingOperations(offset);
-
-    if (offset + size > stream_buffer_size) {
-        // The buffer would overflow, save the amount of used watches and reset the state.
-        invalidation_mark = current_watch_cursor;
-        current_watch_cursor = 0;
-        offset = 0;
-
-        // Swap watches and reset waiting cursors.
-        std::swap(previous_watches, current_watches);
-        wait_cursor = 0;
-        wait_bound = 0;
-
-        // Ensure that we don't wait for uncommitted fences.
-        scheduler.Flush();
-    }
-
-    return std::make_pair(memory.Map(offset, size), offset);
-}
-
-void VKStreamBuffer::Unmap(u64 size) {
-    ASSERT_MSG(size <= mapped_size, "Reserved size is too small");
-
-    memory.Unmap();
-
-    offset += size;
-
-    if (current_watch_cursor + 1 >= current_watches.size()) {
-        // Ensure that there are enough watches.
-        ReserveWatches(current_watches, WATCHES_RESERVE_CHUNK);
-    }
-    auto& watch = current_watches[current_watch_cursor++];
-    watch.upper_bound = offset;
-    watch.tick = scheduler.CurrentTick();
-}
-
-void VKStreamBuffer::CreateBuffers() {
-    const auto memory_properties = device.GetPhysical().GetMemoryProperties();
-    const u32 preferred_type = GetMemoryType(memory_properties);
-    const u32 preferred_heap = memory_properties.memoryTypes[preferred_type].heapIndex;
-
-    // Substract from the preferred heap size some bytes to avoid getting out of memory.
-    const VkDeviceSize heap_size = memory_properties.memoryHeaps[preferred_heap].size;
-    // As per DXVK's example, using `heap_size / 2`
-    const VkDeviceSize allocable_size = heap_size / 2;
-    buffer = device.GetLogical().CreateBuffer({
-        .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
-        .pNext = nullptr,
-        .flags = 0,
-        .size = std::min(PREFERRED_STREAM_BUFFER_SIZE, allocable_size),
-        .usage = BUFFER_USAGE,
-        .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
-        .queueFamilyIndexCount = 0,
-        .pQueueFamilyIndices = nullptr,
-    });
-
-    const auto requirements = device.GetLogical().GetBufferMemoryRequirements(*buffer);
-    const u32 required_flags = requirements.memoryTypeBits;
-    stream_buffer_size = static_cast<u64>(requirements.size);
-
-    memory = device.GetLogical().AllocateMemory({
-        .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
-        .pNext = nullptr,
-        .allocationSize = requirements.size,
-        .memoryTypeIndex = GetMemoryType(memory_properties, required_flags),
-    });
-    buffer.BindMemory(*memory, 0);
-}
-
-void VKStreamBuffer::ReserveWatches(std::vector<Watch>& watches, std::size_t grow_size) {
-    watches.resize(watches.size() + grow_size);
-}
-
-void VKStreamBuffer::WaitPendingOperations(u64 requested_upper_bound) {
-    if (!invalidation_mark) {
-        return;
-    }
-    while (requested_upper_bound < wait_bound && wait_cursor < *invalidation_mark) {
-        auto& watch = previous_watches[wait_cursor];
-        wait_bound = watch.upper_bound;
-        scheduler.Wait(watch.tick);
-        ++wait_cursor;
-    }
-}
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_stream_buffer.h

@@ -1,76 +0,0 @@
-// Copyright 2019 yuzu Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <optional>
-#include <utility>
-#include <vector>
-
-#include "common/common_types.h"
-#include "video_core/vulkan_common/vulkan_wrapper.h"
-
-namespace Vulkan {
-
-class Device;
-class VKFenceWatch;
-class VKScheduler;
-
-class VKStreamBuffer final {
-public:
-    explicit VKStreamBuffer(const Device& device, VKScheduler& scheduler);
-    ~VKStreamBuffer();
-
-    /**
-     * Reserves a region of memory from the stream buffer.
-     * @param size Size to reserve.
-     * @returns A pair of a raw memory pointer (with offset added), and the buffer offset
-     */
-    std::pair<u8*, u64> Map(u64 size, u64 alignment);
-
-    /// Ensures that "size" bytes of memory are available to the GPU, potentially recording a copy.
-    void Unmap(u64 size);
-
-    VkBuffer Handle() const noexcept {
-        return *buffer;
-    }
-
-    u64 Address() const noexcept {
-        return 0;
-    }
-
-private:
-    struct Watch {
-        u64 tick{};
-        u64 upper_bound{};
-    };
-
-    /// Creates Vulkan buffer handles committing the required the required memory.
-    void CreateBuffers();
-
-    /// Increases the amount of watches available.
-    void ReserveWatches(std::vector<Watch>& watches, std::size_t grow_size);
-
-    void WaitPendingOperations(u64 requested_upper_bound);
-
-    const Device& device;   ///< Vulkan device manager.
-    VKScheduler& scheduler; ///< Command scheduler.
-
-    vk::Buffer buffer;        ///< Mapped buffer.
-    vk::DeviceMemory memory;  ///< Memory allocation.
-    u64 stream_buffer_size{}; ///< Stream buffer size.
-
-    u64 offset{};      ///< Buffer iterator.
-    u64 mapped_size{}; ///< Size reserved for the current copy.
-
-    std::vector<Watch> current_watches;           ///< Watches recorded in the current iteration.
-    std::size_t current_watch_cursor{};           ///< Count of watches, reset on invalidation.
-    std::optional<std::size_t> invalidation_mark; ///< Number of watches used in the previous cycle.
-
-    std::vector<Watch> previous_watches; ///< Watches used in the previous iteration.
-    std::size_t wait_cursor{};           ///< Last watch being waited for completion.
-    u64 wait_bound{};                    ///< Highest offset being watched for completion.
-};
-
-} // namespace Vulkan

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -19,6 +19,8 @@
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
 #include "video_core/renderer_vulkan/vk_texture_cache.h"
+#include "video_core/texture_cache/formatter.h"
+#include "video_core/texture_cache/samples_helper.h"
 #include "video_core/vulkan_common/vulkan_device.h"
 #include "video_core/vulkan_common/vulkan_memory_allocator.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -9,7 +9,7 @@
 
 #include "shader_recompiler/shader_info.h"
 #include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
-#include "video_core/texture_cache/texture_cache.h"
+#include "video_core/texture_cache/texture_cache_base.h"
 #include "video_core/vulkan_common/vulkan_memory_allocator.h"
 #include "video_core/vulkan_common/vulkan_wrapper.h"
 

src/video_core/renderer_vulkan/vk_texture_cache_base.cpp

@@ -0,0 +1,10 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "video_core/renderer_vulkan/vk_texture_cache.h"
+#include "video_core/texture_cache/texture_cache.h"
+
+namespace VideoCommon {
+template class VideoCommon::TextureCache<Vulkan::TextureCacheParams>;
+}

src/video_core/texture_cache/image_view_info.cpp

@@ -6,7 +6,7 @@
 
 #include "common/assert.h"
 #include "video_core/texture_cache/image_view_info.h"
-#include "video_core/texture_cache/texture_cache.h"
+#include "video_core/texture_cache/texture_cache_base.h"
 #include "video_core/texture_cache/types.h"
 #include "video_core/textures/texture.h"
 
@@ -14,6 +14,8 @@ namespace VideoCommon {
 
 namespace {
 
+using Tegra::Texture::TextureType;
+
 constexpr u8 RENDER_TARGET_SWIZZLE = std::numeric_limits<u8>::max();
 
 [[nodiscard]] u8 CastSwizzle(SwizzleSource source) {

src/video_core/texture_cache/render_targets.h

@@ -24,10 +24,10 @@ struct RenderTargets {
         return std::ranges::any_of(color_buffer_ids, contains) || contains(depth_buffer_id);
     }
 
-    std::array<ImageViewId, NUM_RT> color_buffer_ids;
-    ImageViewId depth_buffer_id;
+    std::array<ImageViewId, NUM_RT> color_buffer_ids{};
+    ImageViewId depth_buffer_id{};
     std::array<u8, NUM_RT> draw_buffers{};
-    Extent2D size;
+    Extent2D size{};
 };
 
 } // namespace VideoCommon

src/video_core/texture_cache/texture_cache.h

@@ -4,48 +4,11 @@
 
 #pragma once
 
-#include <algorithm>
-#include <array>
-#include <bit>
-#include <memory>
-#include <mutex>
-#include <optional>
-#include <span>
-#include <type_traits>
-#include <unordered_map>
-#include <unordered_set>
-#include <utility>
-#include <vector>
-
-#include <boost/container/small_vector.hpp>
-
 #include "common/alignment.h"
-#include "common/common_types.h"
-#include "common/literals.h"
-#include "common/logging/log.h"
 #include "common/settings.h"
-#include "video_core/compatible_formats.h"
-#include "video_core/delayed_destruction_ring.h"
 #include "video_core/dirty_flags.h"
-#include "video_core/engines/fermi_2d.h"
-#include "video_core/engines/kepler_compute.h"
-#include "video_core/engines/maxwell_3d.h"
-#include "video_core/memory_manager.h"
-#include "video_core/rasterizer_interface.h"
-#include "video_core/surface.h"
-#include "video_core/texture_cache/descriptor_table.h"
-#include "video_core/texture_cache/format_lookup_table.h"
-#include "video_core/texture_cache/formatter.h"
-#include "video_core/texture_cache/image_base.h"
-#include "video_core/texture_cache/image_info.h"
-#include "video_core/texture_cache/image_view_base.h"
-#include "video_core/texture_cache/image_view_info.h"
-#include "video_core/texture_cache/render_targets.h"
 #include "video_core/texture_cache/samples_helper.h"
-#include "video_core/texture_cache/slot_vector.h"
-#include "video_core/texture_cache/types.h"
-#include "video_core/texture_cache/util.h"
-#include "video_core/textures/texture.h"
+#include "video_core/texture_cache/texture_cache_base.h"
 
 namespace VideoCommon {
 
@@ -61,352 +24,6 @@ using VideoCore::Surface::PixelFormatFromRenderTargetFormat;
 using VideoCore::Surface::SurfaceType;
 using namespace Common::Literals;
 
-template <class P>
-class TextureCache {
-    /// Address shift for caching images into a hash table
-    static constexpr u64 PAGE_BITS = 20;
-
-    /// Enables debugging features to the texture cache
-    static constexpr bool ENABLE_VALIDATION = P::ENABLE_VALIDATION;
-    /// Implement blits as copies between framebuffers
-    static constexpr bool FRAMEBUFFER_BLITS = P::FRAMEBUFFER_BLITS;
-    /// True when some copies have to be emulated
-    static constexpr bool HAS_EMULATED_COPIES = P::HAS_EMULATED_COPIES;
-    /// True when the API can provide info about the memory of the device.
-    static constexpr bool HAS_DEVICE_MEMORY_INFO = P::HAS_DEVICE_MEMORY_INFO;
-
-    /// Image view ID for null descriptors
-    static constexpr ImageViewId NULL_IMAGE_VIEW_ID{0};
-    /// Sampler ID for bugged sampler ids
-    static constexpr SamplerId NULL_SAMPLER_ID{0};
-
-    static constexpr u64 DEFAULT_EXPECTED_MEMORY = 1_GiB;
-    static constexpr u64 DEFAULT_CRITICAL_MEMORY = 2_GiB;
-
-    using Runtime = typename P::Runtime;
-    using Image = typename P::Image;
-    using ImageAlloc = typename P::ImageAlloc;
-    using ImageView = typename P::ImageView;
-    using Sampler = typename P::Sampler;
-    using Framebuffer = typename P::Framebuffer;
-
-    struct BlitImages {
-        ImageId dst_id;
-        ImageId src_id;
-        PixelFormat dst_format;
-        PixelFormat src_format;
-    };
-
-    template <typename T>
-    struct IdentityHash {
-        [[nodiscard]] size_t operator()(T value) const noexcept {
-            return static_cast<size_t>(value);
-        }
-    };
-
-public:
-    explicit TextureCache(Runtime&, VideoCore::RasterizerInterface&, Tegra::Engines::Maxwell3D&,
-                          Tegra::Engines::KeplerCompute&, Tegra::MemoryManager&);
-
-    /// Notify the cache that a new frame has been queued
-    void TickFrame();
-
-    /// Return a constant reference to the given image view id
-    [[nodiscard]] const ImageView& GetImageView(ImageViewId id) const noexcept;
-
-    /// Return a reference to the given image view id
-    [[nodiscard]] ImageView& GetImageView(ImageViewId id) noexcept;
-
-    /// Mark an image as modified from the GPU
-    void MarkModification(ImageId id) noexcept;
-
-    /// Fill image_view_ids with the graphics images in indices
-    void FillGraphicsImageViews(std::span<const u32> indices,
-                                std::span<ImageViewId> image_view_ids);
-
-    /// Fill image_view_ids with the compute images in indices
-    void FillComputeImageViews(std::span<const u32> indices, std::span<ImageViewId> image_view_ids);
-
-    /// Get the sampler from the graphics descriptor table in the specified index
-    Sampler* GetGraphicsSampler(u32 index);
-
-    /// Get the sampler from the compute descriptor table in the specified index
-    Sampler* GetComputeSampler(u32 index);
-
-    /// Refresh the state for graphics image view and sampler descriptors
-    void SynchronizeGraphicsDescriptors();
-
-    /// Refresh the state for compute image view and sampler descriptors
-    void SynchronizeComputeDescriptors();
-
-    /// Update bound render targets and upload memory if necessary
-    /// @param is_clear True when the render targets are being used for clears
-    void UpdateRenderTargets(bool is_clear);
-
-    /// Find a framebuffer with the currently bound render targets
-    /// UpdateRenderTargets should be called before this
-    Framebuffer* GetFramebuffer();
-
-    /// Mark images in a range as modified from the CPU
-    void WriteMemory(VAddr cpu_addr, size_t size);
-
-    /// Download contents of host images to guest memory in a region
-    void DownloadMemory(VAddr cpu_addr, size_t size);
-
-    /// Remove images in a region
-    void UnmapMemory(VAddr cpu_addr, size_t size);
-
-    /// Remove images in a region
-    void UnmapGPUMemory(GPUVAddr gpu_addr, size_t size);
-
-    /// Blit an image with the given parameters
-    void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
-                   const Tegra::Engines::Fermi2D::Surface& src,
-                   const Tegra::Engines::Fermi2D::Config& copy);
-
-    /// Invalidate the contents of the color buffer index
-    /// These contents become unspecified, the cache can assume aggressive optimizations.
-    void InvalidateColorBuffer(size_t index);
-
-    /// Invalidate the contents of the depth buffer
-    /// These contents become unspecified, the cache can assume aggressive optimizations.
-    void InvalidateDepthBuffer();
-
-    /// Try to find a cached image view in the given CPU address
-    [[nodiscard]] ImageView* TryFindFramebufferImageView(VAddr cpu_addr);
-
-    /// Return true when there are uncommitted images to be downloaded
-    [[nodiscard]] bool HasUncommittedFlushes() const noexcept;
-
-    /// Return true when the caller should wait for async downloads
-    [[nodiscard]] bool ShouldWaitAsyncFlushes() const noexcept;
-
-    /// Commit asynchronous downloads
-    void CommitAsyncFlushes();
-
-    /// Pop asynchronous downloads
-    void PopAsyncFlushes();
-
-    /// Return true when a CPU region is modified from the GPU
-    [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
-
-    std::mutex mutex;
-
-private:
-    /// Iterate over all page indices in a range
-    template <typename Func>
-    static void ForEachCPUPage(VAddr addr, size_t size, Func&& func) {
-        static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
-        const u64 page_end = (addr + size - 1) >> PAGE_BITS;
-        for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) {
-            if constexpr (RETURNS_BOOL) {
-                if (func(page)) {
-                    break;
-                }
-            } else {
-                func(page);
-            }
-        }
-    }
-
-    template <typename Func>
-    static void ForEachGPUPage(GPUVAddr addr, size_t size, Func&& func) {
-        static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
-        const u64 page_end = (addr + size - 1) >> PAGE_BITS;
-        for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) {
-            if constexpr (RETURNS_BOOL) {
-                if (func(page)) {
-                    break;
-                }
-            } else {
-                func(page);
-            }
-        }
-    }
-
-    /// Runs the Garbage Collector.
-    void RunGarbageCollector();
-
-    /// Fills image_view_ids in the image views in indices
-    void FillImageViews(DescriptorTable<TICEntry>& table,
-                        std::span<ImageViewId> cached_image_view_ids, std::span<const u32> indices,
-                        std::span<ImageViewId> image_view_ids);
-
-    /// Find or create an image view in the guest descriptor table
-    ImageViewId VisitImageView(DescriptorTable<TICEntry>& table,
-                               std::span<ImageViewId> cached_image_view_ids, u32 index);
-
-    /// Find or create a framebuffer with the given render target parameters
-    FramebufferId GetFramebufferId(const RenderTargets& key);
-
-    /// Refresh the contents (pixel data) of an image
-    void RefreshContents(Image& image, ImageId image_id);
-
-    /// Upload data from guest to an image
-    template <typename StagingBuffer>
-    void UploadImageContents(Image& image, StagingBuffer& staging_buffer);
-
-    /// Find or create an image view from a guest descriptor
-    [[nodiscard]] ImageViewId FindImageView(const TICEntry& config);
-
-    /// Create a new image view from a guest descriptor
-    [[nodiscard]] ImageViewId CreateImageView(const TICEntry& config);
-
-    /// Find or create an image from the given parameters
-    [[nodiscard]] ImageId FindOrInsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
-                                            RelaxedOptions options = RelaxedOptions{});
-
-    /// Find an image from the given parameters
-    [[nodiscard]] ImageId FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
-                                    RelaxedOptions options);
-
-    /// Create an image from the given parameters
-    [[nodiscard]] ImageId InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
-                                      RelaxedOptions options);
-
-    /// Create a new image and join perfectly matching existing images
-    /// Remove joined images from the cache
-    [[nodiscard]] ImageId JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr);
-
-    /// Return a blit image pair from the given guest blit parameters
-    [[nodiscard]] BlitImages GetBlitImages(const Tegra::Engines::Fermi2D::Surface& dst,
-                                           const Tegra::Engines::Fermi2D::Surface& src);
-
-    /// Find or create a sampler from a guest descriptor sampler
-    [[nodiscard]] SamplerId FindSampler(const TSCEntry& config);
-
-    /// Find or create an image view for the given color buffer index
-    [[nodiscard]] ImageViewId FindColorBuffer(size_t index, bool is_clear);
-
-    /// Find or create an image view for the depth buffer
-    [[nodiscard]] ImageViewId FindDepthBuffer(bool is_clear);
-
-    /// Find or create a view for a render target with the given image parameters
-    [[nodiscard]] ImageViewId FindRenderTargetView(const ImageInfo& info, GPUVAddr gpu_addr,
-                                                   bool is_clear);
-
-    /// Iterates over all the images in a region calling func
-    template <typename Func>
-    void ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& func);
-
-    template <typename Func>
-    void ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Func&& func);
-
-    template <typename Func>
-    void ForEachSparseImageInRegion(GPUVAddr gpu_addr, size_t size, Func&& func);
-
-    /// Iterates over all the images in a region calling func
-    template <typename Func>
-    void ForEachSparseSegment(ImageBase& image, Func&& func);
-
-    /// Find or create an image view in the given image with the passed parameters
-    [[nodiscard]] ImageViewId FindOrEmplaceImageView(ImageId image_id, const ImageViewInfo& info);
-
-    /// Register image in the page table
-    void RegisterImage(ImageId image);
-
-    /// Unregister image from the page table
-    void UnregisterImage(ImageId image);
-
-    /// Track CPU reads and writes for image
-    void TrackImage(ImageBase& image, ImageId image_id);
-
-    /// Stop tracking CPU reads and writes for image
-    void UntrackImage(ImageBase& image, ImageId image_id);
-
-    /// Delete image from the cache
-    void DeleteImage(ImageId image);
-
-    /// Remove image views references from the cache
-    void RemoveImageViewReferences(std::span<const ImageViewId> removed_views);
-
-    /// Remove framebuffers using the given image views from the cache
-    void RemoveFramebuffers(std::span<const ImageViewId> removed_views);
-
-    /// Mark an image as modified from the GPU
-    void MarkModification(ImageBase& image) noexcept;
-
-    /// Synchronize image aliases, copying data if needed
-    void SynchronizeAliases(ImageId image_id);
-
-    /// Prepare an image to be used
-    void PrepareImage(ImageId image_id, bool is_modification, bool invalidate);
-
-    /// Prepare an image view to be used
-    void PrepareImageView(ImageViewId image_view_id, bool is_modification, bool invalidate);
-
-    /// Execute copies from one image to the other, even if they are incompatible
-    void CopyImage(ImageId dst_id, ImageId src_id, std::span<const ImageCopy> copies);
-
-    /// Bind an image view as render target, downloading resources preemtively if needed
-    void BindRenderTarget(ImageViewId* old_id, ImageViewId new_id);
-
-    /// Create a render target from a given image and image view parameters
-    [[nodiscard]] std::pair<FramebufferId, ImageViewId> RenderTargetFromImage(
-        ImageId, const ImageViewInfo& view_info);
-
-    /// Returns true if the current clear parameters clear the whole image of a given image view
-    [[nodiscard]] bool IsFullClear(ImageViewId id);
-
-    Runtime& runtime;
-    VideoCore::RasterizerInterface& rasterizer;
-    Tegra::Engines::Maxwell3D& maxwell3d;
-    Tegra::Engines::KeplerCompute& kepler_compute;
-    Tegra::MemoryManager& gpu_memory;
-
-    DescriptorTable<TICEntry> graphics_image_table{gpu_memory};
-    DescriptorTable<TSCEntry> graphics_sampler_table{gpu_memory};
-    std::vector<SamplerId> graphics_sampler_ids;
-    std::vector<ImageViewId> graphics_image_view_ids;
-
-    DescriptorTable<TICEntry> compute_image_table{gpu_memory};
-    DescriptorTable<TSCEntry> compute_sampler_table{gpu_memory};
-    std::vector<SamplerId> compute_sampler_ids;
-    std::vector<ImageViewId> compute_image_view_ids;
-
-    RenderTargets render_targets;
-
-    std::unordered_map<TICEntry, ImageViewId> image_views;
-    std::unordered_map<TSCEntry, SamplerId> samplers;
-    std::unordered_map<RenderTargets, FramebufferId> framebuffers;
-
-    std::unordered_map<u64, std::vector<ImageMapId>, IdentityHash<u64>> page_table;
-    std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> gpu_page_table;
-    std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> sparse_page_table;
-
-    std::unordered_map<ImageId, std::vector<ImageViewId>> sparse_views;
-
-    VAddr virtual_invalid_space{};
-
-    bool has_deleted_images = false;
-    u64 total_used_memory = 0;
-    u64 minimum_memory;
-    u64 expected_memory;
-    u64 critical_memory;
-
-    SlotVector<Image> slot_images;
-    SlotVector<ImageMapView> slot_map_views;
-    SlotVector<ImageView> slot_image_views;
-    SlotVector<ImageAlloc> slot_image_allocs;
-    SlotVector<Sampler> slot_samplers;
-    SlotVector<Framebuffer> slot_framebuffers;
-
-    // TODO: This data structure is not optimal and it should be reworked
-    std::vector<ImageId> uncommitted_downloads;
-    std::queue<std::vector<ImageId>> committed_downloads;
-
-    static constexpr size_t TICKS_TO_DESTROY = 6;
-    DelayedDestructionRing<Image, TICKS_TO_DESTROY> sentenced_images;
-    DelayedDestructionRing<ImageView, TICKS_TO_DESTROY> sentenced_image_view;
-    DelayedDestructionRing<Framebuffer, TICKS_TO_DESTROY> sentenced_framebuffers;
-
-    std::unordered_map<GPUVAddr, ImageAllocId> image_allocs_table;
-
-    u64 modification_tick = 0;
-    u64 frame_tick = 0;
-    typename SlotVector<Image>::Iterator deletion_iterator;
-};
-
 template <class P>
 TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface& rasterizer_,
                               Tegra::Engines::Maxwell3D& maxwell3d_,
@@ -820,40 +437,6 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
     }
 }
 
-template <class P>
-void TextureCache<P>::InvalidateColorBuffer(size_t index) {
-    ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index];
-    color_buffer_id = FindColorBuffer(index, false);
-    if (!color_buffer_id) {
-        LOG_ERROR(HW_GPU, "Invalidating invalid color buffer in index={}", index);
-        return;
-    }
-    // When invalidating a color buffer, the old contents are no longer relevant
-    ImageView& color_buffer = slot_image_views[color_buffer_id];
-    Image& image = slot_images[color_buffer.image_id];
-    image.flags &= ~ImageFlagBits::CpuModified;
-    image.flags &= ~ImageFlagBits::GpuModified;
-
-    runtime.InvalidateColorBuffer(color_buffer, index);
-}
-
-template <class P>
-void TextureCache<P>::InvalidateDepthBuffer() {
-    ImageViewId& depth_buffer_id = render_targets.depth_buffer_id;
-    depth_buffer_id = FindDepthBuffer(false);
-    if (!depth_buffer_id) {
-        LOG_ERROR(HW_GPU, "Invalidating invalid depth buffer");
-        return;
-    }
-    // When invalidating the depth buffer, the old contents are no longer relevant
-    ImageBase& image = slot_images[slot_image_views[depth_buffer_id].image_id];
-    image.flags &= ~ImageFlagBits::CpuModified;
-    image.flags &= ~ImageFlagBits::GpuModified;
-
-    ImageView& depth_buffer = slot_image_views[depth_buffer_id];
-    runtime.InvalidateDepthBuffer(depth_buffer);
-}
-
 template <class P>
 typename P::ImageView* TextureCache<P>::TryFindFramebufferImageView(VAddr cpu_addr) {
     // TODO: Properly implement this

src/video_core/texture_cache/texture_cache_base.h

@@ -0,0 +1,385 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <mutex>
+#include <span>
+#include <type_traits>
+#include <unordered_map>
+#include <unordered_set>
+#include <vector>
+
+#include "common/common_types.h"
+#include "common/literals.h"
+#include "video_core/compatible_formats.h"
+#include "video_core/delayed_destruction_ring.h"
+#include "video_core/engines/fermi_2d.h"
+#include "video_core/engines/kepler_compute.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/memory_manager.h"
+#include "video_core/rasterizer_interface.h"
+#include "video_core/surface.h"
+#include "video_core/texture_cache/descriptor_table.h"
+#include "video_core/texture_cache/image_base.h"
+#include "video_core/texture_cache/image_info.h"
+#include "video_core/texture_cache/image_view_info.h"
+#include "video_core/texture_cache/render_targets.h"
+#include "video_core/texture_cache/slot_vector.h"
+#include "video_core/texture_cache/types.h"
+#include "video_core/texture_cache/util.h"
+#include "video_core/textures/texture.h"
+
+namespace VideoCommon {
+
+using Tegra::Texture::SwizzleSource;
+using Tegra::Texture::TICEntry;
+using Tegra::Texture::TSCEntry;
+using VideoCore::Surface::GetFormatType;
+using VideoCore::Surface::IsCopyCompatible;
+using VideoCore::Surface::PixelFormat;
+using VideoCore::Surface::PixelFormatFromDepthFormat;
+using VideoCore::Surface::PixelFormatFromRenderTargetFormat;
+using namespace Common::Literals;
+
+template <class P>
+class TextureCache {
+    /// Address shift for caching images into a hash table
+    static constexpr u64 PAGE_BITS = 20;
+
+    /// Enables debugging features to the texture cache
+    static constexpr bool ENABLE_VALIDATION = P::ENABLE_VALIDATION;
+    /// Implement blits as copies between framebuffers
+    static constexpr bool FRAMEBUFFER_BLITS = P::FRAMEBUFFER_BLITS;
+    /// True when some copies have to be emulated
+    static constexpr bool HAS_EMULATED_COPIES = P::HAS_EMULATED_COPIES;
+    /// True when the API can provide info about the memory of the device.
+    static constexpr bool HAS_DEVICE_MEMORY_INFO = P::HAS_DEVICE_MEMORY_INFO;
+
+    /// Image view ID for null descriptors
+    static constexpr ImageViewId NULL_IMAGE_VIEW_ID{0};
+    /// Sampler ID for bugged sampler ids
+    static constexpr SamplerId NULL_SAMPLER_ID{0};
+
+    static constexpr u64 DEFAULT_EXPECTED_MEMORY = 1_GiB;
+    static constexpr u64 DEFAULT_CRITICAL_MEMORY = 2_GiB;
+
+    using Runtime = typename P::Runtime;
+    using Image = typename P::Image;
+    using ImageAlloc = typename P::ImageAlloc;
+    using ImageView = typename P::ImageView;
+    using Sampler = typename P::Sampler;
+    using Framebuffer = typename P::Framebuffer;
+
+    struct BlitImages {
+        ImageId dst_id;
+        ImageId src_id;
+        PixelFormat dst_format;
+        PixelFormat src_format;
+    };
+
+    template <typename T>
+    struct IdentityHash {
+        [[nodiscard]] size_t operator()(T value) const noexcept {
+            return static_cast<size_t>(value);
+        }
+    };
+
+public:
+    explicit TextureCache(Runtime&, VideoCore::RasterizerInterface&, Tegra::Engines::Maxwell3D&,
+                          Tegra::Engines::KeplerCompute&, Tegra::MemoryManager&);
+
+    /// Notify the cache that a new frame has been queued
+    void TickFrame();
+
+    /// Return a constant reference to the given image view id
+    [[nodiscard]] const ImageView& GetImageView(ImageViewId id) const noexcept;
+
+    /// Return a reference to the given image view id
+    [[nodiscard]] ImageView& GetImageView(ImageViewId id) noexcept;
+
+    /// Mark an image as modified from the GPU
+    void MarkModification(ImageId id) noexcept;
+
+    /// Fill image_view_ids with the graphics images in indices
+    void FillGraphicsImageViews(std::span<const u32> indices,
+                                std::span<ImageViewId> image_view_ids);
+
+    /// Fill image_view_ids with the compute images in indices
+    void FillComputeImageViews(std::span<const u32> indices, std::span<ImageViewId> image_view_ids);
+
+    /// Get the sampler from the graphics descriptor table in the specified index
+    Sampler* GetGraphicsSampler(u32 index);
+
+    /// Get the sampler from the compute descriptor table in the specified index
+    Sampler* GetComputeSampler(u32 index);
+
+    /// Refresh the state for graphics image view and sampler descriptors
+    void SynchronizeGraphicsDescriptors();
+
+    /// Refresh the state for compute image view and sampler descriptors
+    void SynchronizeComputeDescriptors();
+
+    /// Update bound render targets and upload memory if necessary
+    /// @param is_clear True when the render targets are being used for clears
+    void UpdateRenderTargets(bool is_clear);
+
+    /// Find a framebuffer with the currently bound render targets
+    /// UpdateRenderTargets should be called before this
+    Framebuffer* GetFramebuffer();
+
+    /// Mark images in a range as modified from the CPU
+    void WriteMemory(VAddr cpu_addr, size_t size);
+
+    /// Download contents of host images to guest memory in a region
+    void DownloadMemory(VAddr cpu_addr, size_t size);
+
+    /// Remove images in a region
+    void UnmapMemory(VAddr cpu_addr, size_t size);
+
+    /// Remove images in a region
+    void UnmapGPUMemory(GPUVAddr gpu_addr, size_t size);
+
+    /// Blit an image with the given parameters
+    void BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
+                   const Tegra::Engines::Fermi2D::Surface& src,
+                   const Tegra::Engines::Fermi2D::Config& copy);
+
+    /// Try to find a cached image view in the given CPU address
+    [[nodiscard]] ImageView* TryFindFramebufferImageView(VAddr cpu_addr);
+
+    /// Return true when there are uncommitted images to be downloaded
+    [[nodiscard]] bool HasUncommittedFlushes() const noexcept;
+
+    /// Return true when the caller should wait for async downloads
+    [[nodiscard]] bool ShouldWaitAsyncFlushes() const noexcept;
+
+    /// Commit asynchronous downloads
+    void CommitAsyncFlushes();
+
+    /// Pop asynchronous downloads
+    void PopAsyncFlushes();
+
+    /// Return true when a CPU region is modified from the GPU
+    [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size);
+
+    std::mutex mutex;
+
+private:
+    /// Iterate over all page indices in a range
+    template <typename Func>
+    static void ForEachCPUPage(VAddr addr, size_t size, Func&& func) {
+        static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
+        const u64 page_end = (addr + size - 1) >> PAGE_BITS;
+        for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) {
+            if constexpr (RETURNS_BOOL) {
+                if (func(page)) {
+                    break;
+                }
+            } else {
+                func(page);
+            }
+        }
+    }
+
+    template <typename Func>
+    static void ForEachGPUPage(GPUVAddr addr, size_t size, Func&& func) {
+        static constexpr bool RETURNS_BOOL = std::is_same_v<std::invoke_result<Func, u64>, bool>;
+        const u64 page_end = (addr + size - 1) >> PAGE_BITS;
+        for (u64 page = addr >> PAGE_BITS; page <= page_end; ++page) {
+            if constexpr (RETURNS_BOOL) {
+                if (func(page)) {
+                    break;
+                }
+            } else {
+                func(page);
+            }
+        }
+    }
+
+    /// Runs the Garbage Collector.
+    void RunGarbageCollector();
+
+    /// Fills image_view_ids in the image views in indices
+    void FillImageViews(DescriptorTable<TICEntry>& table,
+                        std::span<ImageViewId> cached_image_view_ids, std::span<const u32> indices,
+                        std::span<ImageViewId> image_view_ids);
+
+    /// Find or create an image view in the guest descriptor table
+    ImageViewId VisitImageView(DescriptorTable<TICEntry>& table,
+                               std::span<ImageViewId> cached_image_view_ids, u32 index);
+
+    /// Find or create a framebuffer with the given render target parameters
+    FramebufferId GetFramebufferId(const RenderTargets& key);
+
+    /// Refresh the contents (pixel data) of an image
+    void RefreshContents(Image& image, ImageId image_id);
+
+    /// Upload data from guest to an image
+    template <typename StagingBuffer>
+    void UploadImageContents(Image& image, StagingBuffer& staging_buffer);
+
+    /// Find or create an image view from a guest descriptor
+    [[nodiscard]] ImageViewId FindImageView(const TICEntry& config);
+
+    /// Create a new image view from a guest descriptor
+    [[nodiscard]] ImageViewId CreateImageView(const TICEntry& config);
+
+    /// Find or create an image from the given parameters
+    [[nodiscard]] ImageId FindOrInsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
+                                            RelaxedOptions options = RelaxedOptions{});
+
+    /// Find an image from the given parameters
+    [[nodiscard]] ImageId FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
+                                    RelaxedOptions options);
+
+    /// Create an image from the given parameters
+    [[nodiscard]] ImageId InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
+                                      RelaxedOptions options);
+
+    /// Create a new image and join perfectly matching existing images
+    /// Remove joined images from the cache
+    [[nodiscard]] ImageId JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VAddr cpu_addr);
+
+    /// Return a blit image pair from the given guest blit parameters
+    [[nodiscard]] BlitImages GetBlitImages(const Tegra::Engines::Fermi2D::Surface& dst,
+                                           const Tegra::Engines::Fermi2D::Surface& src);
+
+    /// Find or create a sampler from a guest descriptor sampler
+    [[nodiscard]] SamplerId FindSampler(const TSCEntry& config);
+
+    /// Find or create an image view for the given color buffer index
+    [[nodiscard]] ImageViewId FindColorBuffer(size_t index, bool is_clear);
+
+    /// Find or create an image view for the depth buffer
+    [[nodiscard]] ImageViewId FindDepthBuffer(bool is_clear);
+
+    /// Find or create a view for a render target with the given image parameters
+    [[nodiscard]] ImageViewId FindRenderTargetView(const ImageInfo& info, GPUVAddr gpu_addr,
+                                                   bool is_clear);
+
+    /// Iterates over all the images in a region calling func
+    template <typename Func>
+    void ForEachImageInRegion(VAddr cpu_addr, size_t size, Func&& func);
+
+    template <typename Func>
+    void ForEachImageInRegionGPU(GPUVAddr gpu_addr, size_t size, Func&& func);
+
+    template <typename Func>
+    void ForEachSparseImageInRegion(GPUVAddr gpu_addr, size_t size, Func&& func);
+
+    /// Iterates over all the images in a region calling func
+    template <typename Func>
+    void ForEachSparseSegment(ImageBase& image, Func&& func);
+
+    /// Find or create an image view in the given image with the passed parameters
+    [[nodiscard]] ImageViewId FindOrEmplaceImageView(ImageId image_id, const ImageViewInfo& info);
+
+    /// Register image in the page table
+    void RegisterImage(ImageId image);
+
+    /// Unregister image from the page table
+    void UnregisterImage(ImageId image);
+
+    /// Track CPU reads and writes for image
+    void TrackImage(ImageBase& image, ImageId image_id);
+
+    /// Stop tracking CPU reads and writes for image
+    void UntrackImage(ImageBase& image, ImageId image_id);
+
+    /// Delete image from the cache
+    void DeleteImage(ImageId image);
+
+    /// Remove image views references from the cache
+    void RemoveImageViewReferences(std::span<const ImageViewId> removed_views);
+
+    /// Remove framebuffers using the given image views from the cache
+    void RemoveFramebuffers(std::span<const ImageViewId> removed_views);
+
+    /// Mark an image as modified from the GPU
+    void MarkModification(ImageBase& image) noexcept;
+
+    /// Synchronize image aliases, copying data if needed
+    void SynchronizeAliases(ImageId image_id);
+
+    /// Prepare an image to be used
+    void PrepareImage(ImageId image_id, bool is_modification, bool invalidate);
+
+    /// Prepare an image view to be used
+    void PrepareImageView(ImageViewId image_view_id, bool is_modification, bool invalidate);
+
+    /// Execute copies from one image to the other, even if they are incompatible
+    void CopyImage(ImageId dst_id, ImageId src_id, std::span<const ImageCopy> copies);
+
+    /// Bind an image view as render target, downloading resources preemtively if needed
+    void BindRenderTarget(ImageViewId* old_id, ImageViewId new_id);
+
+    /// Create a render target from a given image and image view parameters
+    [[nodiscard]] std::pair<FramebufferId, ImageViewId> RenderTargetFromImage(
+        ImageId, const ImageViewInfo& view_info);
+
+    /// Returns true if the current clear parameters clear the whole image of a given image view
+    [[nodiscard]] bool IsFullClear(ImageViewId id);
+
+    Runtime& runtime;
+    VideoCore::RasterizerInterface& rasterizer;
+    Tegra::Engines::Maxwell3D& maxwell3d;
+    Tegra::Engines::KeplerCompute& kepler_compute;
+    Tegra::MemoryManager& gpu_memory;
+
+    DescriptorTable<TICEntry> graphics_image_table{gpu_memory};
+    DescriptorTable<TSCEntry> graphics_sampler_table{gpu_memory};
+    std::vector<SamplerId> graphics_sampler_ids;
+    std::vector<ImageViewId> graphics_image_view_ids;
+
+    DescriptorTable<TICEntry> compute_image_table{gpu_memory};
+    DescriptorTable<TSCEntry> compute_sampler_table{gpu_memory};
+    std::vector<SamplerId> compute_sampler_ids;
+    std::vector<ImageViewId> compute_image_view_ids;
+
+    RenderTargets render_targets;
+
+    std::unordered_map<TICEntry, ImageViewId> image_views;
+    std::unordered_map<TSCEntry, SamplerId> samplers;
+    std::unordered_map<RenderTargets, FramebufferId> framebuffers;
+
+    std::unordered_map<u64, std::vector<ImageMapId>, IdentityHash<u64>> page_table;
+    std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> gpu_page_table;
+    std::unordered_map<u64, std::vector<ImageId>, IdentityHash<u64>> sparse_page_table;
+
+    std::unordered_map<ImageId, std::vector<ImageViewId>> sparse_views;
+
+    VAddr virtual_invalid_space{};
+
+    bool has_deleted_images = false;
+    u64 total_used_memory = 0;
+    u64 minimum_memory;
+    u64 expected_memory;
+    u64 critical_memory;
+
+    SlotVector<Image> slot_images;
+    SlotVector<ImageMapView> slot_map_views;
+    SlotVector<ImageView> slot_image_views;
+    SlotVector<ImageAlloc> slot_image_allocs;
+    SlotVector<Sampler> slot_samplers;
+    SlotVector<Framebuffer> slot_framebuffers;
+
+    // TODO: This data structure is not optimal and it should be reworked
+    std::vector<ImageId> uncommitted_downloads;
+    std::queue<std::vector<ImageId>> committed_downloads;
+
+    static constexpr size_t TICKS_TO_DESTROY = 6;
+    DelayedDestructionRing<Image, TICKS_TO_DESTROY> sentenced_images;
+    DelayedDestructionRing<ImageView, TICKS_TO_DESTROY> sentenced_image_view;
+    DelayedDestructionRing<Framebuffer, TICKS_TO_DESTROY> sentenced_framebuffers;
+
+    std::unordered_map<GPUVAddr, ImageAllocId> image_allocs_table;
+
+    u64 modification_tick = 0;
+    u64 frame_tick = 0;
+    typename SlotVector<Image>::Iterator deletion_iterator;
+};
+
+} // namespace VideoCommon

src/video_core/textures/astc.cpp

@@ -151,6 +151,76 @@ private:
     const IntType& m_Bits;
 };
 
+enum class IntegerEncoding { JustBits, Quint, Trit };
+
+struct IntegerEncodedValue {
+    constexpr IntegerEncodedValue() = default;
+
+    constexpr IntegerEncodedValue(IntegerEncoding encoding_, u32 num_bits_)
+        : encoding{encoding_}, num_bits{num_bits_} {}
+
+    constexpr bool MatchesEncoding(const IntegerEncodedValue& other) const {
+        return encoding == other.encoding && num_bits == other.num_bits;
+    }
+
+    // Returns the number of bits required to encode num_vals values.
+    u32 GetBitLength(u32 num_vals) const {
+        u32 total_bits = num_bits * num_vals;
+        if (encoding == IntegerEncoding::Trit) {
+            total_bits += (num_vals * 8 + 4) / 5;
+        } else if (encoding == IntegerEncoding::Quint) {
+            total_bits += (num_vals * 7 + 2) / 3;
+        }
+        return total_bits;
+    }
+
+    IntegerEncoding encoding{};
+    u32 num_bits = 0;
+    u32 bit_value = 0;
+    union {
+        u32 quint_value = 0;
+        u32 trit_value;
+    };
+};
+
+// Returns a new instance of this struct that corresponds to the
+// can take no more than mav_value values
+static constexpr IntegerEncodedValue CreateEncoding(u32 mav_value) {
+    while (mav_value > 0) {
+        u32 check = mav_value + 1;
+
+        // Is mav_value a power of two?
+        if (!(check & (check - 1))) {
+            return IntegerEncodedValue(IntegerEncoding::JustBits, std::popcount(mav_value));
+        }
+
+        // Is mav_value of the type 3*2^n - 1?
+        if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) {
+            return IntegerEncodedValue(IntegerEncoding::Trit, std::popcount(check / 3 - 1));
+        }
+
+        // Is mav_value of the type 5*2^n - 1?
+        if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) {
+            return IntegerEncodedValue(IntegerEncoding::Quint, std::popcount(check / 5 - 1));
+        }
+
+        // Apparently it can't be represented with a bounded integer sequence...
+        // just iterate.
+        mav_value--;
+    }
+    return IntegerEncodedValue(IntegerEncoding::JustBits, 0);
+}
+
+static constexpr std::array<IntegerEncodedValue, 256> MakeEncodedValues() {
+    std::array<IntegerEncodedValue, 256> encodings{};
+    for (std::size_t i = 0; i < encodings.size(); ++i) {
+        encodings[i] = CreateEncoding(static_cast<u32>(i));
+    }
+    return encodings;
+}
+
+static constexpr std::array<IntegerEncodedValue, 256> ASTC_ENCODINGS_VALUES = MakeEncodedValues();
+
 namespace Tegra::Texture::ASTC {
 using IntegerEncodedVector = boost::container::static_vector<
     IntegerEncodedValue, 256,
@@ -521,35 +591,41 @@ static TexelWeightParams DecodeBlockInfo(InputBitStream& strm) {
     return params;
 }
 
-static void FillVoidExtentLDR(InputBitStream& strm, std::span<u32> outBuf, u32 blockWidth,
-                              u32 blockHeight) {
-    // Don't actually care about the void extent, just read the bits...
-    for (s32 i = 0; i < 4; ++i) {
-        strm.ReadBits<13>();
+// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
+// is the same as [(num_bits - 1):0] and repeats all the way down.
+template <typename IntType>
+static constexpr IntType Replicate(IntType val, u32 num_bits, u32 to_bit) {
+    if (num_bits == 0 || to_bit == 0) {
+        return 0;
     }
-
-    // Decode the RGBA components and renormalize them to the range [0, 255]
-    u16 r = static_cast<u16>(strm.ReadBits<16>());
-    u16 g = static_cast<u16>(strm.ReadBits<16>());
-    u16 b = static_cast<u16>(strm.ReadBits<16>());
-    u16 a = static_cast<u16>(strm.ReadBits<16>());
-
-    u32 rgba = (r >> 8) | (g & 0xFF00) | (static_cast<u32>(b) & 0xFF00) << 8 |
-               (static_cast<u32>(a) & 0xFF00) << 16;
-
-    for (u32 j = 0; j < blockHeight; j++) {
-        for (u32 i = 0; i < blockWidth; i++) {
-            outBuf[j * blockWidth + i] = rgba;
+    const IntType v = val & static_cast<IntType>((1 << num_bits) - 1);
+    IntType res = v;
+    u32 reslen = num_bits;
+    while (reslen < to_bit) {
+        u32 comp = 0;
+        if (num_bits > to_bit - reslen) {
+            u32 newshift = to_bit - reslen;
+            comp = num_bits - newshift;
+            num_bits = newshift;
         }
+        res = static_cast<IntType>(res << num_bits);
+        res = static_cast<IntType>(res | (v >> comp));
+        reslen += num_bits;
     }
+    return res;
 }
 
-static void FillError(std::span<u32> outBuf, u32 blockWidth, u32 blockHeight) {
-    for (u32 j = 0; j < blockHeight; j++) {
-        for (u32 i = 0; i < blockWidth; i++) {
-            outBuf[j * blockWidth + i] = 0xFFFF00FF;
-        }
+static constexpr std::size_t NumReplicateEntries(u32 num_bits) {
+    return std::size_t(1) << num_bits;
+}
+
+template <typename IntType, u32 num_bits, u32 to_bit>
+static constexpr auto MakeReplicateTable() {
+    std::array<IntType, NumReplicateEntries(num_bits)> table{};
+    for (IntType value = 0; value < static_cast<IntType>(std::size(table)); ++value) {
+        table[value] = Replicate(value, num_bits, to_bit);
     }
+    return table;
 }
 
 static constexpr auto REPLICATE_BYTE_TO_16_TABLE = MakeReplicateTable<u32, 8, 16>();
@@ -572,6 +648,9 @@ static constexpr auto REPLICATE_2_BIT_TO_8_TABLE = MakeReplicateTable<u32, 2, 8>
 static constexpr auto REPLICATE_3_BIT_TO_8_TABLE = MakeReplicateTable<u32, 3, 8>();
 static constexpr auto REPLICATE_4_BIT_TO_8_TABLE = MakeReplicateTable<u32, 4, 8>();
 static constexpr auto REPLICATE_5_BIT_TO_8_TABLE = MakeReplicateTable<u32, 5, 8>();
+static constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
+static constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
+static constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
 /// Use a precompiled table with the most common usages, if it's not in the expected range, fallback
 /// to the runtime implementation
 static constexpr u32 FastReplicateTo8(u32 value, u32 num_bits) {
@@ -1316,6 +1395,37 @@ static void ComputeEndpoints(Pixel& ep1, Pixel& ep2, const u32*& colorValues,
 #undef READ_INT_VALUES
 }
 
+static void FillVoidExtentLDR(InputBitStream& strm, std::span<u32> outBuf, u32 blockWidth,
+                              u32 blockHeight) {
+    // Don't actually care about the void extent, just read the bits...
+    for (s32 i = 0; i < 4; ++i) {
+        strm.ReadBits<13>();
+    }
+
+    // Decode the RGBA components and renormalize them to the range [0, 255]
+    u16 r = static_cast<u16>(strm.ReadBits<16>());
+    u16 g = static_cast<u16>(strm.ReadBits<16>());
+    u16 b = static_cast<u16>(strm.ReadBits<16>());
+    u16 a = static_cast<u16>(strm.ReadBits<16>());
+
+    u32 rgba = (r >> 8) | (g & 0xFF00) | (static_cast<u32>(b) & 0xFF00) << 8 |
+               (static_cast<u32>(a) & 0xFF00) << 16;
+
+    for (u32 j = 0; j < blockHeight; j++) {
+        for (u32 i = 0; i < blockWidth; i++) {
+            outBuf[j * blockWidth + i] = rgba;
+        }
+    }
+}
+
+static void FillError(std::span<u32> outBuf, u32 blockWidth, u32 blockHeight) {
+    for (u32 j = 0; j < blockHeight; j++) {
+        for (u32 i = 0; i < blockWidth; i++) {
+            outBuf[j * blockWidth + i] = 0xFFFF00FF;
+        }
+    }
+}
+
 static void DecompressBlock(std::span<const u8, 16> inBuf, const u32 blockWidth,
                             const u32 blockHeight, std::span<u32, 12 * 12> outBuf) {
     InputBitStream strm(inBuf);

src/video_core/textures/astc.h

@@ -9,117 +9,6 @@
 
 namespace Tegra::Texture::ASTC {
 
-enum class IntegerEncoding { JustBits, Quint, Trit };
-
-struct IntegerEncodedValue {
-    constexpr IntegerEncodedValue() = default;
-
-    constexpr IntegerEncodedValue(IntegerEncoding encoding_, u32 num_bits_)
-        : encoding{encoding_}, num_bits{num_bits_} {}
-
-    constexpr bool MatchesEncoding(const IntegerEncodedValue& other) const {
-        return encoding == other.encoding && num_bits == other.num_bits;
-    }
-
-    // Returns the number of bits required to encode num_vals values.
-    u32 GetBitLength(u32 num_vals) const {
-        u32 total_bits = num_bits * num_vals;
-        if (encoding == IntegerEncoding::Trit) {
-            total_bits += (num_vals * 8 + 4) / 5;
-        } else if (encoding == IntegerEncoding::Quint) {
-            total_bits += (num_vals * 7 + 2) / 3;
-        }
-        return total_bits;
-    }
-
-    IntegerEncoding encoding{};
-    u32 num_bits = 0;
-    u32 bit_value = 0;
-    union {
-        u32 quint_value = 0;
-        u32 trit_value;
-    };
-};
-
-// Returns a new instance of this struct that corresponds to the
-// can take no more than mav_value values
-constexpr IntegerEncodedValue CreateEncoding(u32 mav_value) {
-    while (mav_value > 0) {
-        u32 check = mav_value + 1;
-
-        // Is mav_value a power of two?
-        if (!(check & (check - 1))) {
-            return IntegerEncodedValue(IntegerEncoding::JustBits, std::popcount(mav_value));
-        }
-
-        // Is mav_value of the type 3*2^n - 1?
-        if ((check % 3 == 0) && !((check / 3) & ((check / 3) - 1))) {
-            return IntegerEncodedValue(IntegerEncoding::Trit, std::popcount(check / 3 - 1));
-        }
-
-        // Is mav_value of the type 5*2^n - 1?
-        if ((check % 5 == 0) && !((check / 5) & ((check / 5) - 1))) {
-            return IntegerEncodedValue(IntegerEncoding::Quint, std::popcount(check / 5 - 1));
-        }
-
-        // Apparently it can't be represented with a bounded integer sequence...
-        // just iterate.
-        mav_value--;
-    }
-    return IntegerEncodedValue(IntegerEncoding::JustBits, 0);
-}
-
-constexpr std::array<IntegerEncodedValue, 256> MakeEncodedValues() {
-    std::array<IntegerEncodedValue, 256> encodings{};
-    for (std::size_t i = 0; i < encodings.size(); ++i) {
-        encodings[i] = CreateEncoding(static_cast<u32>(i));
-    }
-    return encodings;
-}
-
-constexpr std::array<IntegerEncodedValue, 256> ASTC_ENCODINGS_VALUES = MakeEncodedValues();
-
-// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
-// is the same as [(num_bits - 1):0] and repeats all the way down.
-template <typename IntType>
-constexpr IntType Replicate(IntType val, u32 num_bits, u32 to_bit) {
-    if (num_bits == 0 || to_bit == 0) {
-        return 0;
-    }
-    const IntType v = val & static_cast<IntType>((1 << num_bits) - 1);
-    IntType res = v;
-    u32 reslen = num_bits;
-    while (reslen < to_bit) {
-        u32 comp = 0;
-        if (num_bits > to_bit - reslen) {
-            u32 newshift = to_bit - reslen;
-            comp = num_bits - newshift;
-            num_bits = newshift;
-        }
-        res = static_cast<IntType>(res << num_bits);
-        res = static_cast<IntType>(res | (v >> comp));
-        reslen += num_bits;
-    }
-    return res;
-}
-
-constexpr std::size_t NumReplicateEntries(u32 num_bits) {
-    return std::size_t(1) << num_bits;
-}
-
-template <typename IntType, u32 num_bits, u32 to_bit>
-constexpr auto MakeReplicateTable() {
-    std::array<IntType, NumReplicateEntries(num_bits)> table{};
-    for (IntType value = 0; value < static_cast<IntType>(std::size(table)); ++value) {
-        table[value] = Replicate(value, num_bits, to_bit);
-    }
-    return table;
-}
-
-constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
-constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
-constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
-
 void Decompress(std::span<const uint8_t> data, uint32_t width, uint32_t height, uint32_t depth,
                 uint32_t block_width, uint32_t block_height, std::span<uint8_t> output);
 

src/video_core/textures/decoders.cpp

@@ -18,9 +18,9 @@
 
 namespace Tegra::Texture {
 namespace {
-template <bool TO_LINEAR>
-void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width,
-             u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) {
+template <bool TO_LINEAR, u32 BYTES_PER_PIXEL>
+void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height, u32 depth,
+                 u32 block_height, u32 block_depth, u32 stride_alignment) {
     // The origin of the transformation can be configured here, leave it as zero as the current API
     // doesn't expose it.
     static constexpr u32 origin_x = 0;
@@ -28,9 +28,9 @@ void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixe
     static constexpr u32 origin_z = 0;
 
     // We can configure here a custom pitch
-    // As it's not exposed 'width * bpp' will be the expected pitch.
-    const u32 pitch = width * bytes_per_pixel;
-    const u32 stride = Common::AlignUpLog2(width, stride_alignment) * bytes_per_pixel;
+    // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch.
+    const u32 pitch = width * BYTES_PER_PIXEL;
+    const u32 stride = Common::AlignUpLog2(width, stride_alignment) * BYTES_PER_PIXEL;
 
     const u32 gobs_in_x = Common::DivCeilLog2(stride, GOB_SIZE_X_SHIFT);
     const u32 block_size = gobs_in_x << (GOB_SIZE_SHIFT + block_height + block_depth);
@@ -54,14 +54,14 @@ void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixe
                                  ((block_y & block_height_mask) << GOB_SIZE_SHIFT);
 
             for (u32 column = 0; column < width; ++column) {
-                const u32 x = (column + origin_x) * bytes_per_pixel;
+                const u32 x = (column + origin_x) * BYTES_PER_PIXEL;
                 const u32 offset_x = (x >> GOB_SIZE_X_SHIFT) << x_shift;
 
                 const u32 base_swizzled_offset = offset_z + offset_y + offset_x;
                 const u32 swizzled_offset = base_swizzled_offset + table[x % GOB_SIZE_X];
 
                 const u32 unswizzled_offset =
-                    slice * pitch * height + line * pitch + column * bytes_per_pixel;
+                    slice * pitch * height + line * pitch + column * BYTES_PER_PIXEL;
 
                 if (const auto offset = (TO_LINEAR ? unswizzled_offset : swizzled_offset);
                     offset >= input.size()) {
@@ -73,11 +73,45 @@ void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixe
 
                 u8* const dst = &output[TO_LINEAR ? swizzled_offset : unswizzled_offset];
                 const u8* const src = &input[TO_LINEAR ? unswizzled_offset : swizzled_offset];
-                std::memcpy(dst, src, bytes_per_pixel);
+
+                std::memcpy(dst, src, BYTES_PER_PIXEL);
             }
         }
     }
 }
+
+template <bool TO_LINEAR>
+void Swizzle(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel, u32 width,
+             u32 height, u32 depth, u32 block_height, u32 block_depth, u32 stride_alignment) {
+    switch (bytes_per_pixel) {
+    case 1:
+        return SwizzleImpl<TO_LINEAR, 1>(output, input, width, height, depth, block_height,
+                                         block_depth, stride_alignment);
+    case 2:
+        return SwizzleImpl<TO_LINEAR, 2>(output, input, width, height, depth, block_height,
+                                         block_depth, stride_alignment);
+    case 3:
+        return SwizzleImpl<TO_LINEAR, 3>(output, input, width, height, depth, block_height,
+                                         block_depth, stride_alignment);
+    case 4:
+        return SwizzleImpl<TO_LINEAR, 4>(output, input, width, height, depth, block_height,
+                                         block_depth, stride_alignment);
+    case 6:
+        return SwizzleImpl<TO_LINEAR, 6>(output, input, width, height, depth, block_height,
+                                         block_depth, stride_alignment);
+    case 8:
+        return SwizzleImpl<TO_LINEAR, 8>(output, input, width, height, depth, block_height,
+                                         block_depth, stride_alignment);
+    case 12:
+        return SwizzleImpl<TO_LINEAR, 12>(output, input, width, height, depth, block_height,
+                                          block_depth, stride_alignment);
+    case 16:
+        return SwizzleImpl<TO_LINEAR, 16>(output, input, width, height, depth, block_height,
+                                          block_depth, stride_alignment);
+    default:
+        UNREACHABLE_MSG("Invalid bytes_per_pixel={}", bytes_per_pixel);
+    }
+}
 } // Anonymous namespace
 
 void UnswizzleTexture(std::span<u8> output, std::span<const u8> input, u32 bytes_per_pixel,

src/video_core/vulkan_common/vulkan_device.cpp

@@ -526,6 +526,17 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
         SetNext(next, workgroup_layout);
     }
 
+    VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR executable_properties;
+    if (khr_pipeline_executable_properties) {
+        LOG_INFO(Render_Vulkan, "Enabling shader feedback, expect slower shader build times");
+        executable_properties = {
+            .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PIPELINE_EXECUTABLE_PROPERTIES_FEATURES_KHR,
+            .pNext = nullptr,
+            .pipelineExecutableInfo = VK_TRUE,
+        };
+        SetNext(next, executable_properties);
+    }
+
     if (!ext_depth_range_unrestricted) {
         LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted");
     }
@@ -824,6 +835,7 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) {
 
     bool has_khr_shader_float16_int8{};
     bool has_khr_workgroup_memory_explicit_layout{};
+    bool has_khr_pipeline_executable_properties{};
     bool has_ext_subgroup_size_control{};
     bool has_ext_transform_feedback{};
     bool has_ext_custom_border_color{};
@@ -878,6 +890,10 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) {
             test(nv_device_diagnostics_config, VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME,
                  true);
         }
+        if (Settings::values.renderer_shader_feedback) {
+            test(has_khr_pipeline_executable_properties,
+                 VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME, false);
+        }
     }
     VkPhysicalDeviceFeatures2KHR features{};
     features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2_KHR;
@@ -1033,6 +1049,19 @@ std::vector<const char*> Device::LoadExtensions(bool requires_surface) {
             khr_workgroup_memory_explicit_layout = true;
         }
     }
+    if (has_khr_pipeline_executable_properties) {
+        VkPhysicalDevicePipelineExecutablePropertiesFeaturesKHR executable_properties;
+        executable_properties.sType =
+            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PIPELINE_EXECUTABLE_PROPERTIES_FEATURES_KHR;
+        executable_properties.pNext = nullptr;
+        features.pNext = &executable_properties;
+        physical.GetFeatures2KHR(features);
+
+        if (executable_properties.pipelineExecutableInfo) {
+            extensions.push_back(VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME);
+            khr_pipeline_executable_properties = true;
+        }
+    }
     if (khr_push_descriptor) {
         VkPhysicalDevicePushDescriptorPropertiesKHR push_descriptor;
         push_descriptor.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR;

src/video_core/vulkan_common/vulkan_device.h

@@ -214,6 +214,11 @@ public:
         return khr_push_descriptor;
     }
 
+    /// Returns true if VK_KHR_pipeline_executable_properties is enabled.
+    bool IsKhrPipelineEexecutablePropertiesEnabled() const {
+        return khr_pipeline_executable_properties;
+    }
+
     /// Returns true if the device supports VK_KHR_workgroup_memory_explicit_layout.
     bool IsKhrWorkgroupMemoryExplicitLayoutSupported() const {
         return khr_workgroup_memory_explicit_layout;
@@ -378,6 +383,7 @@ private:
     bool khr_spirv_1_4{};                       ///< Support for VK_KHR_spirv_1_4.
     bool khr_workgroup_memory_explicit_layout{}; ///< Support for explicit workgroup layouts.
     bool khr_push_descriptor{};                  ///< Support for VK_KHR_push_descritor.
+    bool khr_pipeline_executable_properties{};   ///< Support for executable properties.
     bool ext_index_type_uint8{};                 ///< Support for VK_EXT_index_type_uint8.
     bool ext_sampler_filter_minmax{};            ///< Support for VK_EXT_sampler_filter_minmax.
     bool ext_depth_range_unrestricted{};         ///< Support for VK_EXT_depth_range_unrestricted.