[debug] debug knobs adjustments

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

- [x] I have read and followed the [Contribution Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/CONTRIBUTING.md#code-contributions).
- [x] I have read and followed the [AI Policy](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/AI.md)
- [x] I have read and followed the [Coding Guidelines](https://git.eden-emu.dev/eden-emu/eden/src/branch/master/docs/policies/Coding.md) to the best of my ability.

-------------------

issue was first reported here
https://github.com/eden-emulator/Issue-Reports/issues/607

tl;dr wrong boolean, now boxarts should work

Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4275
Reviewed-by: Maufeat <sahyno1996@gmail.com>
Reviewed-by: MaranBr <maranbr@eden-emu.dev>

cpmfile.json

@@ -340,10 +340,10 @@
         "version": "vulkan-sdk-%NUMERIC_VERSION%"
     },
     "xbyak": {
-        "hash": "e0aa0a603dd3ac1a39d82213df1e73c042831aec2d6b2fe382c899651eaae4ed7e8aeb6be9c41ea0097087c9d091961ab18f313cd6a9e10d621715ffd49bfe36",
+        "hash": "b6475276b2faaeb315734ea8f4f8bd87ededcee768961b39679bee547e7f3e98884d8b7851e176d861dab30a80a76e6ea302f8c111483607dde969b4797ea95a",
         "package": "xbyak",
         "repo": "herumi/xbyak",
-        "version": "v7.40.1"
+        "version": "v7.35.2"
     },
     "zlib": {
         "hash": "16fea4df307a68cf0035858abe2fd550250618a97590e202037acd18a666f57afc10f8836cbbd472d54a0e76539d0e558cb26f059d53de52ff90634bbf4f47d4",

docs/Settings.md

@@ -43,6 +43,7 @@ This guide will walk you through adding a new boolean toggle setting to Eden's c
 Firstly add your desired toggle:
 
 Example: `src/common/setting.h`
+
 ```cpp
 SwitchableSetting<bool> your_setting_name{linkage, false, "your_setting_name", Category::RendererExtensions};
 ```
@@ -67,6 +68,7 @@ Common Categories:
 Add the toggle to the Qt UI, where you wish for it to appear and place it there.
 
 Example: `src/qt_common/config/shared_translation.cpp`
+
 ```cpp
 INSERT(Settings,
        your_setting_name,
@@ -91,6 +93,7 @@ INSERT(Settings,
 Add where it should be in the settings.
 
 Example: `src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/BooleanSetting.kt`
+
 ```kts
 RENDERER_YOUR_SETTING_NAME("your_setting_name"),
 ```
@@ -106,6 +109,7 @@ RENDERER_YOUR_SETTING_NAME("your_setting_name"),
 Add the toggle to the Kotlin (Android) UI
 
 Example: `src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/model/view/SettingsItem.kt`
+
 ```kts
 put(
     SwitchSetting(
@@ -123,6 +127,7 @@ put(
 Add your setting within the right category.
 
 Example: `src/android/app/src/main/java/org/yuzu/yuzu_emu/features/settings/ui/SettingsFragmentPresenter.kt`
+
 ```kts
 add(BooleanSetting.RENDERER_YOUR_SETTING_NAME.key)
 ```
@@ -137,6 +142,7 @@ add(BooleanSetting.RENDERER_YOUR_SETTING_NAME.key)
 Add your setting and description in the appropriate place.
 
 Example: `src/android/app/src/main/res/values/strings.xml`
+
 ```xml
 <string name="your_setting_name">Your Setting Display Name</string>
 <string name="your_setting_name_description">Detailed description of what this setting does. Explain any caveats, requirements, or warnings here.</string>
@@ -150,6 +156,7 @@ Now the UI part is done find a place in the code for the toggle,
 And use it to your heart's desire!
 
 Example:
+
 ```cpp
 const bool your_value = Settings::values.your_setting_name.GetValue();
 
@@ -196,25 +203,31 @@ Common advantages recap:
 
 #### Accessing Debug Knobs (dev side)
 
-Use the `Settings::getDebugKnobAt(u8 i)` function to check if a specific bit is set:
+Use the `Settings::GetDebugKnobAt(u8 i)` function to check if a specific bit is set:
 
 ```cpp
 //cpp side
 #include "common/settings.h"
 
+//To use it as a general purpose uint var:
+unsigned int debug_knobs = Settings::values.debug_knobs.GetValue();
+
 // Check if bit 0 is set
-bool feature_enabled = Settings::getDebugKnobAt(0);
+bool feature_enabled = Settings::GetDebugKnobAt(0);
 
 // Check if bit 15 is set
-bool another_feature = Settings::getDebugKnobAt(15);
+bool another_feature = Settings::GetDebugKnobAt(15);
 ```
 
 ```kts
 //kotlin side
 import org.yuzu.yuzu_emu.features.settings.model.Settings
 
+//To use it as a general purpose uint var
+val debug_knobs: Int = UShortSetting.DEBUG_KNOBS.getInt()
+
 // Check if bit x is set
-bool feature_enabled = Settings.getDebugKnobAt(x); //x as integer from 0 to 15
+bool feature_enabled = Settings.GetDebugKnobAt(x); //x as integer from 0 to 15
 ```
 
 The function returns `true` if the specified bit (0-15) is set in the `debug_knobs` value, `false` otherwise.
@@ -247,6 +260,7 @@ There are two main confusions when talking about knobs:
 Sometimes when an user reports: knobs 1 and 2 gets better performance, dev may get confuse whether he means the knobs 1 and 2 literally, or the 1st and 2nd knobs (knobs 0 and 1).
 
 Debug knobs are **zero-based**, which means:
+
 * The first knob is the knob(0) (or knob0 henceforth), and the last one is the 15 (knob15, likewise)
 * You can talk: "knob0 is enabled/disabled", "In this video i was using only knobs 0 and 2", etc.
 
@@ -259,6 +273,7 @@ Whenever you're instructing tests or reporting results, be precise about whether
 
 ALWAYS use the word in PLURAL (knobs), without mentioning which one, to refer to the setting, aka multiple knobs at once:
 Examples:
+
 - **knobs=0**: no knobs enabled
 - **knobs=1**: knob0 enabled, others disabled
 - **knobs=2**: knob1 enabled, others disabled
@@ -270,6 +285,7 @@ Examples:
 
 Use the word in SINGULAR (knob), or in plural but referring which ones, when meaning multiple knobs at once:
 Examples:
+
 - **knob0**: knob 0 enabled, others disabled
 - **knob1**: knob 1 enabled, others disabled
 - **knobs 0 and 1**: knobs 0 and 1 enabled, others disabled
@@ -282,12 +298,12 @@ Examples:
 
 ```cpp
 void SomeFunction() {
-    if (Settings::getDebugKnobAt(0)) {
+    if (Settings::GetDebugKnobAt(0)) {
         LOG_DEBUG(Common, "Debug feature 0 is enabled");
         // Additional debug code here
     }
   
-    if (Settings::getDebugKnobAt(1)) {
+    if (Settings::GetDebugKnobAt(1)) {
         LOG_DEBUG(Common, "Debug feature 1 is enabled");
         // Different debug behavior
     }
@@ -299,7 +315,7 @@ void SomeFunction() {
 ```cpp
 bool UseOptimizedPath() {
     // Skip optimization if debug bit 2 is set for testing
-    return !Settings::getDebugKnobAt(2);
+    return !Settings::GetDebugKnobAt(2);
 }
 ```
 
@@ -309,7 +325,7 @@ bool UseOptimizedPath() {
 void ExperimentalFeature() {
     static constexpr u8 EXPERIMENTAL_FEATURE_BIT = 3;
   
-    if (!Settings::getDebugKnobAt(EXPERIMENTAL_FEATURE_BIT)) {
+    if (!Settings::GetDebugKnobAt(EXPERIMENTAL_FEATURE_BIT)) {
         // Fallback to stable implementation
         StableImplementation();
         return;

src/android/app/src/main/java/org/yuzu/yuzu_emu/NativeLibrary.kt

@@ -220,7 +220,7 @@ object NativeLibrary {
 
     external fun refreshThreadPolicies()
 
-    external fun getDebugKnobAt(index: Int): Boolean
+    external fun GetDebugKnobAt(index: Int): Boolean
 
     /**
      * Set the current speed limit to the configured turbo speed.

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

@@ -35,8 +35,8 @@ object Settings {
     fun getPlayerString(player: Int): String =
         YuzuApplication.appContext.getString(R.string.preferences_player, player)
 
-    fun getDebugKnobAt(index: Int): Boolean {
+    fun GetDebugKnobAt(index: Int): Boolean {
-        return org.yuzu.yuzu_emu.NativeLibrary.getDebugKnobAt(index)
+        return org.yuzu.yuzu_emu.NativeLibrary.GetDebugKnobAt(index)
     }
 
     const val PREF_FIRST_APP_LAUNCH = "FirstApplicationLaunch"

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

@@ -11,8 +11,7 @@ import org.yuzu.yuzu_emu.utils.NativeConfig
 enum class ShortSetting(override val key: String) : AbstractShortSetting {
     RENDERER_SPEED_LIMIT("speed_limit"),
     RENDERER_TURBO_SPEED_LIMIT("turbo_speed_limit"),
-    RENDERER_SLOW_SPEED_LIMIT("slow_speed_limit"),
+    RENDERER_SLOW_SPEED_LIMIT("slow_speed_limit")
-    DEBUG_KNOBS("debug_knobs")
     ;
 
     override fun getShort(needsGlobal: Boolean): Short = NativeConfig.getShort(key, needsGlobal)

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

@@ -0,0 +1,30 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+// SPDX-FileCopyrightText: 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+package org.yuzu.yuzu_emu.features.settings.model
+
+import org.yuzu.yuzu_emu.utils.NativeConfig
+
+enum class UShortSetting(override val key: String) : AbstractIntSetting {
+    DEBUG_KNOBS("debug_knobs")
+    ;
+
+    override fun getInt(needsGlobal: Boolean): Int =
+        NativeConfig.getUnsignedShort(key, needsGlobal)
+
+    override fun setInt(value: Int) {
+        if (NativeConfig.isPerGameConfigLoaded()) {
+            global = false
+        }
+        NativeConfig.setUnsignedShort(key, value)
+    }
+
+    override val defaultValue: Int by lazy { NativeConfig.getDefaultToString(key).toInt() }
+
+    override fun getValueAsString(needsGlobal: Boolean): String = getInt(needsGlobal).toString()
+
+    override fun reset() = NativeConfig.setUnsignedShort(key, defaultValue)
+}

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

@@ -20,6 +20,7 @@ import org.yuzu.yuzu_emu.features.settings.model.IntSetting
 import org.yuzu.yuzu_emu.features.settings.model.LongSetting
 import org.yuzu.yuzu_emu.features.settings.model.ShortSetting
 import org.yuzu.yuzu_emu.features.settings.model.StringSetting
+import org.yuzu.yuzu_emu.features.settings.model.UShortSetting
 import org.yuzu.yuzu_emu.network.NetDataValidators
 import org.yuzu.yuzu_emu.utils.LosslessScalingHelper
 import org.yuzu.yuzu_emu.utils.NativeConfig
@@ -1034,7 +1035,7 @@ abstract class SettingsItem(
             )
             put(
                 SpinBoxSetting(
-                    ShortSetting.DEBUG_KNOBS,
+                    UShortSetting.DEBUG_KNOBS,
                     titleId = R.string.debug_knobs,
                     descriptionId = R.string.debug_knobs_description,
                     valueHint = R.string.debug_knobs_hint,

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

@@ -25,6 +25,7 @@ import org.yuzu.yuzu_emu.features.settings.model.Settings
 import org.yuzu.yuzu_emu.features.settings.model.Settings.MenuTag
 import org.yuzu.yuzu_emu.features.settings.model.ShortSetting
 import org.yuzu.yuzu_emu.features.settings.model.StringSetting
+import org.yuzu.yuzu_emu.features.settings.model.UShortSetting
 import org.yuzu.yuzu_emu.features.settings.model.view.*
 import org.yuzu.yuzu_emu.utils.InputHandler
 import org.yuzu.yuzu_emu.utils.LosslessScalingHelper
@@ -1326,7 +1327,7 @@ class SettingsFragmentPresenter(
 
             add(HeaderSetting(R.string.general))
 
-            add(ShortSetting.DEBUG_KNOBS.key)
+            add(UShortSetting.DEBUG_KNOBS.key)
             add(StringSetting.PROGRAM_ARGS.key)
 
             if (!NativeConfig.isPerGameConfigLoaded()) {

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

@@ -80,6 +80,12 @@ object NativeConfig {
     @Synchronized
     external fun setShort(key: String, value: Short)
 
+    @Synchronized
+    external fun getUnsignedShort(key: String, needsGlobal: Boolean): Int
+
+    @Synchronized
+    external fun setUnsignedShort(key: String, value: Int)
+
     @Synchronized
     external fun getInt(key: String, needsGlobal: Boolean): Int
 

src/android/app/src/main/jni/native.cpp

@@ -1245,8 +1245,8 @@ void Java_org_yuzu_yuzu_1emu_NativeLibrary_refreshThreadPolicies(JNIEnv* env, jo
     Common::RefreshThreadPolicies();
 }
 
-jboolean Java_org_yuzu_yuzu_1emu_NativeLibrary_getDebugKnobAt(JNIEnv* env, jobject jobj, jint index) {
+jboolean Java_org_yuzu_yuzu_1emu_NativeLibrary_GetDebugKnobAt(JNIEnv* env, jobject jobj, jint index) {
-    return static_cast<jboolean>(Settings::getDebugKnobAt(static_cast<u8>(index)));
+    return static_cast<jboolean>(Settings::GetDebugKnobAt(static_cast<u8>(index)));
 }
 
 void Java_org_yuzu_yuzu_1emu_NativeLibrary_setTurboSpeedLimit(JNIEnv *env, jobject jobj, jboolean enabled) {

src/android/app/src/main/jni/native_config.cpp

@@ -130,6 +130,25 @@ void Java_org_yuzu_yuzu_1emu_utils_NativeConfig_setShort(JNIEnv* env, jobject ob
     setting->SetValue(value);
 }
 
+jint Java_org_yuzu_yuzu_1emu_utils_NativeConfig_getUnsignedShort(JNIEnv* env, jobject obj,
+                                                                 jstring jkey,
+                                                                 jboolean needGlobal) {
+    auto setting = getSetting<u16>(env, jkey);
+    if (setting == nullptr) {
+        return -1;
+    }
+    return static_cast<jint>(setting->GetValue(static_cast<bool>(needGlobal)));
+}
+
+void Java_org_yuzu_yuzu_1emu_utils_NativeConfig_setUnsignedShort(JNIEnv* env, jobject obj,
+                                                                  jstring jkey, jint value) {
+    auto setting = getSetting<u16>(env, jkey);
+    if (setting == nullptr) {
+        return;
+    }
+    setting->SetValue(static_cast<u16>(value));
+}
+
 jint Java_org_yuzu_yuzu_1emu_utils_NativeConfig_getInt(JNIEnv* env, jobject obj, jstring jkey,
                                                        jboolean needGlobal) {
     auto setting = getSetting<int>(env, jkey);

src/audio_core/sink/sdl3_sink.cpp

@@ -19,6 +19,23 @@
 namespace AudioCore::Sink {
 
 namespace {
+
+[[nodiscard]] bool InitializeAudio() {
+    if (!SDL_WasInit(SDL_INIT_AUDIO)) {
+        // See https://github.com/PCSX2/pcsx2/pull/12312
+        // "SDL and cubeb backends previously resulted in different names for the output which
+        // caused them be identified as different applications by the OS."
+        //
+        // Keep in sync with cubeb_sink.cpp name.
+        SDL_SetHint("SDL_AUDIO_DEVICE_APP_NAME", "yuzu Latency Getter");
+        if (!SDL_InitSubSystem(SDL_INIT_AUDIO)) {
+            LOG_CRITICAL(Audio_Sink, "SDL_InitSubSystem audio failed: {}", SDL_GetError());
+            return false;
+        }
+    }
+    return true;
+}
+
 SDL_AudioDeviceID FindAudioDeviceByName(const std::string& device_name, bool capture) {
     int device_count = 0;
     SDL_AudioDeviceID* devices = capture ? SDL_GetAudioRecordingDevices(&device_count)
@@ -204,20 +221,14 @@ private:
 };
 
 SDLSink::SDLSink(std::string_view target_device_name) {
-    if (!SDL_WasInit(SDL_INIT_AUDIO)) {
+    if (InitializeAudio()) {
-        if (!SDL_InitSubSystem(SDL_INIT_AUDIO)) {
-            LOG_CRITICAL(Audio_Sink, "SDL_InitSubSystem audio failed: {}", SDL_GetError());
-            return;
-        }
-    }
-
         if (target_device_name != auto_device_name && !target_device_name.empty()) {
             output_device = target_device_name;
         } else {
             output_device.clear();
         }
-
         device_channels = 2;
+    }
 }
 
 SDLSink::~SDLSink() = default;
@@ -265,15 +276,10 @@ void SDLSink::SetSystemVolume(f32 volume) {
 }
 
 std::vector<std::string> ListSDLSinkDevices(bool capture) {
+    if (!InitializeAudio())
+        return {}; //no devices
+
     std::vector<std::string> device_list;
-
-    if (!SDL_WasInit(SDL_INIT_AUDIO)) {
-        if (!SDL_InitSubSystem(SDL_INIT_AUDIO)) {
-            LOG_CRITICAL(Audio_Sink, "SDL_InitSubSystem audio failed: {}", SDL_GetError());
-            return {};
-        }
-    }
-
     int device_count = 0;
     SDL_AudioDeviceID* devices =
         capture ? SDL_GetAudioRecordingDevices(&device_count)
@@ -304,13 +310,8 @@ bool IsSDLSuitable() {
     return false;
 #else
     // Check SDL can init
-    if (!SDL_WasInit(SDL_INIT_AUDIO)) {
+    if (!InitializeAudio()!
-        if (SDL_InitSubSystem(SDL_INIT_AUDIO) < 0) {
-            LOG_ERROR(Audio_Sink, "SDL failed to init, it is not suitable. Error: {}",
-                      SDL_GetError());
         return false;
-        }
-    }
 
     // We can set any latency frequency we want with SDL, so no need to check that.
 

src/common/logging.cpp

@@ -224,7 +224,7 @@ struct ColorConsoleBackend final : public Backend {
             auto const df = GetDirectFormatArgs(entry);
             // more restrictive, because take for example this simple prelude:
             // [  50.872256] Config <Info> common/settings.cpp:142:LogSettings:
-            char buffer[128];
+            char buffer[256];
             auto result = fmt::format_to_n(buffer, sizeof(buffer) - 1, "\x1b{}[{:4d}.{:06d}] {} <{}> {}:{}:{}: ", color_str, df.time_seconds, df.time_fractional, df.class_name, df.level_name, entry.filename, entry.line_num, entry.function, entry.message);
             std::fwrite(buffer, 1, (std::min)(sizeof(buffer) - 1, result.size), stdout);
             std::fwrite(entry.message, 1, entry.message_len, stdout);
@@ -425,14 +425,14 @@ void SetColorConsoleBackendEnabled(bool enabled) {
 
 void FmtLogMessageImpl(Class log_class, Level log_level, const char* filename, unsigned int line_num, const char* function, fmt::string_view format, const fmt::format_args& args) {
     if (logging_instance && logging_instance->filter.CheckMessage(log_class, log_level)) {
+        auto const flush = ::Settings::values.log_flush_line.GetValue();
         char buffer[BUFSIZ];
         auto result = fmt::vformat_to_n(buffer, sizeof(buffer) - 1, format, args);
-        buffer[result.size] = '\0';
+        buffer[(std::min)(result.size, sizeof(buffer) - 1)] = '\0';
-        auto const flush = ::Settings::values.log_flush_line.GetValue();
         logging_instance->ForEachBackend([=](Backend& backend) {
             backend.Write(Entry{
                 .message = buffer,
-                .message_len = (std::min)(sizeof(buffer) - 1, result.size),
+                .message_len = (std::min)(result.size, sizeof(buffer) - 1),
                 .timestamp = std::chrono::duration_cast<std::chrono::microseconds>(std::chrono::steady_clock::now() - logging_instance->time_origin),
                 .log_class = log_class,
                 .log_level = log_level,

src/common/settings.cpp

@@ -132,11 +132,9 @@ void LogSettings() {
             }
         }
     }
-
+    LOG_INFO(Config, "Eden Configuration:");
-    std::string settings_str{};
     for (auto const& e : settings_list)
-        settings_str += e;
+        LOG_INFO(Config, "{}", e);
-    LOG_INFO(Config, "Eden Configuration:\n{}", settings_str);
 #define LOG_PATH(NAME) \
     LOG_INFO(Config, #NAME ": {}", Common::FS::PathToUTF8String(Common::FS::GetEdenPath(Common::FS::EdenPath::NAME)))
     LOG_PATH(CacheDir);
@@ -148,7 +146,7 @@ void LogSettings() {
 #undef LOG_PATH
 }
 
-bool getDebugKnobAt(u8 i) {
+bool GetDebugKnobAt(u8 i) {
     return (values.debug_knobs.GetValue() & (1 << (i & 0xF))) != 0;
 }
 

src/common/settings.h

@@ -904,7 +904,7 @@ struct Values {
                                            0,
                                            65535,
                                            "debug_knobs",
-                                           Category::Debugging,
+                                           Category::System,
                                            Specialization::Countable,
                                            true,
                                            true};

src/common/stb.cpp

@@ -7,6 +7,4 @@
 #define STB_IMAGE_IMPLEMENTATION 1
 #define STB_IMAGE_RESIZE_IMPLEMENTATION 1
 #define STB_IMAGE_WRITE_IMPLEMENTATION 1
-#define STBI_ONLY_JPEG 1
-
 #include "common/stb.h"

src/common/stb.h

@@ -7,6 +7,7 @@
 #pragma once
 
 #define STBI_ONLY_JPEG 1
+#define STBI_WRITE_NO_STDIO 1
 #include <stb_image.h>
 #include <stb_image_resize.h>
 #include <stb_image_write.h>

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

@@ -324,6 +324,7 @@ Result IApplicationFunctions::NotifyRunning(Out<bool> out_became_running) {
 
 Result IApplicationFunctions::GetPseudoDeviceId(Out<Common::UUID> out_pseudo_device_id) {
     LOG_WARNING(Service_AM, "(stubbed)");
+    R_UNLESS(out_pseudo_device_id, ResultUnknown);
 
     // This should be hashed with the device specific hash
     // for now this will do

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

@@ -9,11 +9,7 @@
 #include <optional>
 #include <string>
 
-#define STBI_ONLY_JPEG 1
+#include "common/stb.h"
-#include <stb_image.h>
-#include <stb_image_resize.h>
-#include <stb_image_write.h>
-
 #include "common/settings.h"
 #include "core/file_sys/control_metadata.h"
 #include "core/file_sys/patch_manager.h"

src/dynarmic/src/dynarmic/backend/exception_handler_posix.cpp

@@ -22,6 +22,7 @@
 #include <sys/mman.h>
 
 #include "common/assert.h"
+#include "common/logging.h"
 #include "common/common_types.h"
 #include "dynarmic/backend/exception_handler.h"
 #include "dynarmic/common/context.h"
@@ -53,23 +54,21 @@ class SigHandler {
             return e.first <= offset && e.first + e.second.size > offset;
         });
     }
-    static void SigAction(int sig, siginfo_t* info, void* raw_context);
 
-    bool supports_fast_mem = true;
-    void* signal_stack_memory = nullptr;
     ankerl::unordered_dense::map<u64, CodeBlockInfo> code_block_infos;
     std::shared_mutex code_block_infos_mutex;
     struct sigaction old_sa_segv;
     struct sigaction old_sa_bus;
-    std::size_t signal_stack_size;
+    std::unique_ptr<uint8_t[]> signal_stack_memory;
+    bool supports_fast_mem = true;
 public:
     SigHandler() noexcept {
-        signal_stack_size = std::max<size_t>(SIGSTKSZ, 2 * 1024 * 1024);
+        auto const stack_size = std::max<size_t>(SIGSTKSZ, 2 * 1024 * 1024);
-        signal_stack_memory = mmap(nullptr, signal_stack_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+        signal_stack_memory = std::make_unique<uint8_t[]>(stack_size);
 
         stack_t signal_stack{};
-        signal_stack.ss_sp = signal_stack_memory;
+        signal_stack.ss_sp = signal_stack_memory.get();
-        signal_stack.ss_size = signal_stack_size;
+        signal_stack.ss_size = stack_size;
         signal_stack.ss_flags = 0;
         if (sigaltstack(&signal_stack, nullptr) != 0) {
             fmt::print(stderr, "dynarmic: POSIX SigHandler: init failure at sigaltstack\n");
@@ -87,7 +86,7 @@ public:
             supports_fast_mem = false;
             return;
         }
-#ifdef __APPLE__
+#if defined(__APPLE__)
         if (sigaction(SIGBUS, &sa, &old_sa_bus) != 0) {
             fmt::print(stderr, "dynarmic: POSIX SigHandler: could not set SIGBUS handler\n");
             supports_fast_mem = false;
@@ -96,10 +95,6 @@ public:
 #endif
     }
 
-    ~SigHandler() noexcept {
-        munmap(signal_stack_memory, signal_stack_size);
-    }
-
     void AddCodeBlock(u64 offset, CodeBlockInfo cbi) noexcept {
         std::unique_lock guard(code_block_infos_mutex);
         code_block_infos.insert_or_assign(offset, cbi);
@@ -109,14 +104,17 @@ public:
         code_block_infos.erase(offset);
     }
 
-    bool SupportsFastmem() const noexcept { return supports_fast_mem; }
+    [[nodiscard]] inline bool SupportsFastmem() const noexcept {
+        return supports_fast_mem;
+    }
+
+    static void RegisterHandler();
+    static void SigAction(int sig, siginfo_t* info, void* raw_context);
 };
 
-std::mutex handler_lock;
 std::optional<SigHandler> sig_handler;
 
-void RegisterHandler() {
+void SigHandler::RegisterHandler() {
-    std::lock_guard<std::mutex> guard(handler_lock);
     if (!sig_handler) {
         sig_handler.emplace();
     }
@@ -125,51 +123,27 @@ void RegisterHandler() {
 void SigHandler::SigAction(int sig, siginfo_t* info, void* raw_context) {
     DEBUG_ASSERT(sig == SIGSEGV || sig == SIGBUS);
     CTX_DECLARE(raw_context);
-#if defined(ARCHITECTURE_x86_64)
     {
         std::shared_lock guard(sig_handler->code_block_infos_mutex);
         if (auto const iter = sig_handler->FindCodeBlockInfo(CTX_PC); iter != sig_handler->code_block_infos.end()) {
             FakeCall fc = iter->second.cb(CTX_PC);
+#if defined(ARCHITECTURE_x86_64)
             CTX_SP -= sizeof(u64);
             *std::bit_cast<u64*>(CTX_SP) = fc.ret_rip;
             CTX_PC = fc.call_rip;
-            return;
-        }
-    }
-    fmt::print(stderr, "Unhandled {} at rip {:#018x}\n", sig == SIGSEGV ? "SIGSEGV" : "SIGBUS", CTX_PC);
 #elif defined(ARCHITECTURE_arm64)
-    {
-        std::shared_lock guard(sig_handler->code_block_infos_mutex);
-        if (const auto iter = sig_handler->FindCodeBlockInfo(CTX_PC); iter != sig_handler->code_block_infos.end()) {
-            FakeCall fc = iter->second.cb(CTX_PC);
             CTX_PC = fc.call_pc;
-            return;
-        }
-    }
-    fmt::print(stderr, "Unhandled {} at pc {:#018x}\n", sig == SIGSEGV ? "SIGSEGV" : "SIGBUS", CTX_PC);
 #elif defined(ARCHITECTURE_riscv64)
-    {
+            CTX_PC = fc.call_sepc;
-        std::shared_lock guard(sig_handler->code_block_infos_mutex);
-        if (const auto iter = sig_handler->FindCodeBlockInfo(CTX_SEPC); iter != sig_handler->code_block_infos.end()) {
-            FakeCall fc = iter->second.cb(CTX_SEPC);
-            CTX_SEPC = fc.call_sepc;
-            return;
-        }
-    }
-    fmt::print(stderr, "Unhandled {} at pc {:#018x}\n", sig == SIGSEGV ? "SIGSEGV" : "SIGBUS", CTX_SEPC);
 #elif defined(ARCHITECTURE_loongarch64)
-    {
-        std::shared_lock guard(sig_handler->code_block_infos_mutex);
-        if (const auto iter = sig_handler->FindCodeBlockInfo(CTX_PC); iter != sig_handler->code_block_infos.end()) {
-            FakeCall fc = iter->second.cb(CTX_PC);
             CTX_PC = fc.call_pc;
+#else
+            ASSERT(false);
+#endif
             return;
         }
     }
-    fmt::print(stderr, "Unhandled {} at pc {:#018x}\n", sig == SIGSEGV ? "SIGSEGV" : "SIGBUS", CTX_PC);
+    LOG_ERROR(Core, "Unhandled {} at {:#018x}\n", sig == SIGSEGV ? "SIGSEGV" : "SIGBUS", CTX_PC);
-#else
-#    error "Invalid architecture"
-#endif
 
     struct sigaction* retry_sa = sig == SIGSEGV ? &sig_handler->old_sa_segv : &sig_handler->old_sa_bus;
     if (retry_sa->sa_flags & SA_SIGINFO) {
@@ -191,8 +165,9 @@ void SigHandler::SigAction(int sig, siginfo_t* info, void* raw_context) {
 struct ExceptionHandler::Impl final {
     Impl(u64 offset_, u64 size_)
         : offset(offset_)
-            , size(size_) {
+        , size(size_)
-        RegisterHandler();
+    {
+        SigHandler::RegisterHandler();
     }
 
     void SetCallback(std::function<FakeCall(u64)> cb) {

src/dynarmic/src/dynarmic/backend/loongarch64/emit_loongarch64.cpp

@@ -1,6 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
+#include <variant>
 #include "dynarmic/backend/loongarch64/emit_loongarch64.h"
 
 #include "dynarmic/backend/loongarch64/a32_jitstate.h"
@@ -95,7 +96,8 @@ EmittedBlockInfo EmitLoongArch64(lagoon_assembler_t& as, IR::Block block, const
 
     // TODO: Emit Terminal
     const auto term = block.GetTerminal();
-    const IR::Term::LinkBlock* link_block_term = boost::get<IR::Term::LinkBlock>(&term);
+    const IR::Term::LeafTerminal* leaft_term = std::get_if<IR::Term::LeafTerminal>(&term);
+    const IR::Term::LinkBlock* link_block_term = std::get_if<IR::Term::LinkBlock>(leaft_term);
     ASSERT(link_block_term);
     la_load_immediate64(&as, Xscratch0, link_block_term->next.Value());
     la_st_w(&as, Xscratch0, Xstate, static_cast<int32_t>(offsetof(A32JitState, regs) + sizeof(u32) * 15));

src/dynarmic/src/dynarmic/backend/x64/a32_emit_x64.cpp

@@ -247,7 +247,7 @@ void A32EmitX64::GenTerminalHandlers() {
     calculate_location_descriptor();
     code.mov(eax, dword[code.ABI_JIT_PTR + offsetof(A32JitState, rsb_ptr)]);
     code.sub(eax, 1);
-    code.and_(eax, u32(A32JitState::RSB_PTR_MASK));
+    code.and_(eax, u32(A32JitState::RSBPtrMask));
     code.mov(dword[code.ABI_JIT_PTR + offsetof(A32JitState, rsb_ptr)], eax);
     code.cmp(rbx, qword[code.ABI_JIT_PTR + offsetof(A32JitState, rsb_location_descriptors) + rax * sizeof(u64)]);
     if (conf.HasOptimization(OptimizationFlag::FastDispatch)) {

src/dynarmic/src/dynarmic/backend/x64/a32_interface.cpp

@@ -37,9 +37,9 @@ using namespace Backend::X64;
 
 static RunCodeCallbacks GenRunCodeCallbacks(A32::UserCallbacks* cb, CodePtr (*LookupBlock)(void* lookup_block_arg), void* arg, const A32::UserConfig& conf) {
     return RunCodeCallbacks{
-        ArgCallback(LookupBlock, reinterpret_cast<u64>(arg)),
+        std::make_unique<ArgCallback>(LookupBlock, reinterpret_cast<u64>(arg)),
-        ArgCallback(Devirtualize<&A32::UserCallbacks::AddTicks>(cb)),
+        std::make_unique<ArgCallback>(Devirtualize<&A32::UserCallbacks::AddTicks>(cb)),
-        ArgCallback(Devirtualize<&A32::UserCallbacks::GetTicksRemaining>(cb)),
+        std::make_unique<ArgCallback>(Devirtualize<&A32::UserCallbacks::GetTicksRemaining>(cb)),
         conf.enable_cycle_counting,
     };
 }
@@ -79,7 +79,7 @@ struct Jit::Impl {
         jit_interface->is_executing = true;
         const CodePtr current_codeptr = [this] {
             // RSB optimization
-            const u32 new_rsb_ptr = (jit_state.rsb_ptr - 1) & A32JitState::RSB_PTR_MASK;
+            const u32 new_rsb_ptr = (jit_state.rsb_ptr - 1) & A32JitState::RSBPtrMask;
             if (jit_state.GetUniqueHash() == jit_state.rsb_location_descriptors[new_rsb_ptr]) {
                 jit_state.rsb_ptr = new_rsb_ptr;
                 return reinterpret_cast<CodePtr>(jit_state.rsb_codeptrs[new_rsb_ptr]);

src/dynarmic/src/dynarmic/backend/x64/a32_jitstate.h

@@ -27,9 +27,6 @@ struct A32JitState {
 
     A32JitState() { ResetRSB(); }
 
-    static constexpr std::size_t RSB_SIZE = 8;  // MUST be a power of 2.
-    static constexpr std::size_t RSB_PTR_MASK = RSB_SIZE - 1;
-
     std::array<u32, 16> Reg{};  // Current register file.
     // TODO: Mode-specific register sets unimplemented.
 
@@ -39,9 +36,8 @@ struct A32JitState {
     u32 cpsr_q = 0;
     u32 cpsr_nzcv = 0;
     u32 cpsr_jaifm = 0;
-    u32 fpsr_exc = 0;
+    u32 Cpsr() const;
-    u32 fpsr_qc = 0;
+    void SetCpsr(u32 cpsr);
-    u32 fpsr_nzcv = 0;
 
     alignas(16) std::array<u32, 64> ExtReg{};  // Extension registers.
 
@@ -53,19 +49,21 @@ struct A32JitState {
     // Exclusive state
     u32 exclusive_state = 0;
 
+    static constexpr std::size_t RSBSize = 8;  // MUST be a power of 2.
+    static constexpr std::size_t RSBPtrMask = RSBSize - 1;
     u32 rsb_ptr = 0;
-    std::array<u64, RSB_SIZE> rsb_location_descriptors;
+    std::array<u64, RSBSize> rsb_location_descriptors;
-    std::array<u64, RSB_SIZE> rsb_codeptrs;
+    std::array<u64, RSBSize> rsb_codeptrs;
-
-    u32 Cpsr() const;
-    void SetCpsr(u32 cpsr);
-
     void ResetRSB();
+
+    u32 fpsr_exc = 0;
+    u32 fpsr_qc = 0;
+    u32 fpsr_nzcv = 0;
     u32 Fpscr() const;
     void SetFpscr(u32 FPSCR);
 
     u64 GetUniqueHash() const noexcept {
-        return (u64(upper_location_descriptor) << 32) | (u64(Reg[15]));
+        return (static_cast<u64>(upper_location_descriptor) << 32) | (static_cast<u64>(Reg[15]));
     }
 
     void TransferJitState(const A32JitState& src, bool reset_rsb) {

src/dynarmic/src/dynarmic/backend/x64/a64_emit_x64.cpp

@@ -208,7 +208,7 @@ void A64EmitX64::GenTerminalHandlers() {
     calculate_location_descriptor();
     code.mov(eax, dword[code.ABI_JIT_PTR + offsetof(A64JitState, rsb_ptr)]);
     code.sub(eax, 1);
-    code.and_(eax, u32(A64JitState::RSB_PTR_MASK));
+    code.and_(eax, u32(A64JitState::RSBPtrMask));
     code.mov(dword[code.ABI_JIT_PTR + offsetof(A64JitState, rsb_ptr)], eax);
     code.cmp(rbx, qword[code.ABI_JIT_PTR + offsetof(A64JitState, rsb_location_descriptors) + rax * sizeof(u64)]);
     if (conf.HasOptimization(OptimizationFlag::FastDispatch)) {

src/dynarmic/src/dynarmic/backend/x64/a64_interface.cpp

@@ -33,9 +33,9 @@ using namespace Backend::X64;
 
 static RunCodeCallbacks GenRunCodeCallbacks(A64::UserCallbacks* cb, CodePtr (*LookupBlock)(void* lookup_block_arg), void* arg, const A64::UserConfig& conf) {
     return RunCodeCallbacks{
-        ArgCallback(LookupBlock, reinterpret_cast<u64>(arg)),
+        std::make_unique<ArgCallback>(LookupBlock, reinterpret_cast<u64>(arg)),
-        ArgCallback(Devirtualize<&A64::UserCallbacks::AddTicks>(cb)),
+        std::make_unique<ArgCallback>(Devirtualize<&A64::UserCallbacks::AddTicks>(cb)),
-        ArgCallback(Devirtualize<&A64::UserCallbacks::GetTicksRemaining>(cb)),
+        std::make_unique<ArgCallback>(Devirtualize<&A64::UserCallbacks::GetTicksRemaining>(cb)),
         conf.enable_cycle_counting,
     };
 }
@@ -78,7 +78,7 @@ public:
         // TODO: Check code alignment
         const CodePtr current_code_ptr = [this] {
             // RSB optimization
-            const u32 new_rsb_ptr = (jit_state.rsb_ptr - 1) & A64JitState::RSB_PTR_MASK;
+            const u32 new_rsb_ptr = (jit_state.rsb_ptr - 1) & A64JitState::RSBPtrMask;
             if (jit_state.GetUniqueHash() == jit_state.rsb_location_descriptors[new_rsb_ptr]) {
                 jit_state.rsb_ptr = new_rsb_ptr;
                 return CodePtr(jit_state.rsb_codeptrs[new_rsb_ptr]);

src/dynarmic/src/dynarmic/backend/x64/a64_jitstate.h

@@ -29,19 +29,18 @@ struct A64JitState {
 
     A64JitState() { ResetRSB(); }
 
-    // Exclusive state stuff
-    static constexpr u64 RESERVATION_GRANULE_MASK = 0xFFFF'FFFF'FFFF'FFF0ull;
-    // Return stack buffer
-    static constexpr size_t RSB_SIZE = 8;  // MUST be a power of 2.
-    static constexpr size_t RSB_PTR_MASK = RSB_SIZE - 1;
-
     std::array<u64, 31> reg{};
     u64 sp = 0;
     u64 pc = 0;
+
     u32 cpsr_nzcv = 0;
-    u32 fpsr_exc = 0;
+
-    u32 fpsr_qc = 0;
+    u32 GetPstate() const {
-    u32 fpcr = 0;
+        return NZCV::FromX64(cpsr_nzcv);
+    }
+    void SetPstate(u32 new_pstate) {
+        cpsr_nzcv = NZCV::ToX64(new_pstate);
+    }
 
     alignas(16) std::array<u64, 64> vec{};  // Extension registers.
 
@@ -51,31 +50,29 @@ struct A64JitState {
     volatile u32 halt_reason = 0;
 
     // Exclusive state
+    static constexpr u64 RESERVATION_GRANULE_MASK = 0xFFFF'FFFF'FFFF'FFF0ull;
     u8 exclusive_state = 0;
+
+    static constexpr size_t RSBSize = 8;  // MUST be a power of 2.
+    static constexpr size_t RSBPtrMask = RSBSize - 1;
     u32 rsb_ptr = 0;
-    std::array<u64, RSB_SIZE> rsb_location_descriptors;
+    std::array<u64, RSBSize> rsb_location_descriptors;
-    std::array<u64, RSB_SIZE> rsb_codeptrs;
+    std::array<u64, RSBSize> rsb_codeptrs;
-
-    u32 GetPstate() const {
-        return NZCV::FromX64(cpsr_nzcv);
-    }
-
-    void SetPstate(u32 new_pstate) {
-        cpsr_nzcv = NZCV::ToX64(new_pstate);
-    }
-
     void ResetRSB() {
         rsb_location_descriptors.fill(0xFFFFFFFFFFFFFFFFull);
         rsb_codeptrs.fill(0);
     }
 
+    u32 fpsr_exc = 0;
+    u32 fpsr_qc = 0;
+    u32 fpcr = 0;
     u32 GetFpcr() const;
     u32 GetFpsr() const;
     void SetFpcr(u32 value);
     void SetFpsr(u32 value);
 
     u64 GetUniqueHash() const noexcept {
-        const u64 fpcr_u64 = u64(fpcr & A64::LocationDescriptor::fpcr_mask) << A64::LocationDescriptor::fpcr_shift;
+        const u64 fpcr_u64 = static_cast<u64>(fpcr & A64::LocationDescriptor::fpcr_mask) << A64::LocationDescriptor::fpcr_shift;
         const u64 pc_u64 = pc & A64::LocationDescriptor::pc_mask;
         return pc_u64 | fpcr_u64;
     }

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

@@ -61,6 +61,73 @@ namespace {
 constexpr size_t CONSTANT_POOL_SIZE = 2 * 1024 * 1024;
 constexpr size_t PRELUDE_COMMIT_SIZE = 16 * 1024 * 1024;
 
+class CustomXbyakAllocator : public Xbyak::Allocator {
+public:
+#ifdef _WIN32
+    uint8_t* alloc(size_t size) override {
+        void* p = VirtualAlloc(nullptr, size, MEM_RESERVE, PAGE_READWRITE);
+        if (p == nullptr) {
+            using Xbyak::Error;
+            XBYAK_THROW(Xbyak::ERR_CANT_ALLOC);
+        }
+        return static_cast<uint8_t*>(p);
+    }
+
+    void free(uint8_t* p) override {
+        VirtualFree(static_cast<void*>(p), 0, MEM_RELEASE);
+    }
+
+    bool useProtect() const override { return false; }
+#else
+    static constexpr size_t DYNARMIC_PAGE_SIZE = 4096;
+
+    // Can't subclass Xbyak::MmapAllocator because it is not a pure interface
+    // and doesn't expose its construtor
+    uint8_t* alloc(size_t size) override {
+        // Waste a page to store the size
+        size += DYNARMIC_PAGE_SIZE;
+
+        int mode = MAP_PRIVATE;
+#if defined(MAP_ANONYMOUS)
+        mode |= MAP_ANONYMOUS;
+#elif defined(MAP_ANON)
+        mode |= MAP_ANON;
+#else
+#   error "not supported"
+#endif
+#ifdef MAP_JIT
+        mode |= MAP_JIT;
+#endif
+        int prot = PROT_READ | PROT_WRITE;
+#ifdef PROT_MPROTECT
+        // https://man.netbsd.org/mprotect.2 specifies that an mprotect() that is LESS
+        // restrictive than the original mapping MUST fail
+        prot |= PROT_MPROTECT(PROT_READ) | PROT_MPROTECT(PROT_WRITE) | PROT_MPROTECT(PROT_EXEC);
+#endif
+        void* p = mmap(nullptr, size, prot, mode, -1, 0);
+        if (p == MAP_FAILED) {
+            using Xbyak::Error;
+            XBYAK_THROW(Xbyak::ERR_CANT_ALLOC);
+        }
+        std::memcpy(p, &size, sizeof(size_t));
+        return static_cast<uint8_t*>(p) + DYNARMIC_PAGE_SIZE;
+    }
+
+    void free(uint8_t* p) override {
+        size_t size;
+        std::memcpy(&size, p - DYNARMIC_PAGE_SIZE, sizeof(size_t));
+        munmap(p - DYNARMIC_PAGE_SIZE, size);
+    }
+
+#    ifdef DYNARMIC_ENABLE_NO_EXECUTE_SUPPORT
+    bool useProtect() const override { return false; }
+#    endif
+#endif
+};
+
+// This is threadsafe as Xbyak::Allocator does not contain any state; it is a pure interface.
+CustomXbyakAllocator s_allocator;
+
 #ifdef DYNARMIC_ENABLE_NO_EXECUTE_SUPPORT
 void ProtectMemory(const void* base, size_t size, bool is_executable) {
 #    ifdef _WIN32
@@ -78,9 +145,11 @@ void ProtectMemory(const void* base, size_t size, bool is_executable) {
 
 HostFeature GetHostFeatures() {
     HostFeature features = {};
+
 #ifdef DYNARMIC_ENABLE_CPU_FEATURE_DETECTION
     using Cpu = Xbyak::util::Cpu;
-    Xbyak::util::Cpu cpu_info{};
+    Xbyak::util::Cpu cpu_info;
+
     if (cpu_info.has(Cpu::tSSSE3))
         features |= HostFeature::SSSE3;
     if (cpu_info.has(Cpu::tSSE41))
@@ -127,6 +196,7 @@ HostFeature GetHostFeatures() {
         features |= HostFeature::GFNI;
     if (cpu_info.has(Cpu::tWAITPKG))
         features |= HostFeature::WAITPKG;
+
     if (cpu_info.has(Cpu::tBMI2)) {
         // BMI2 instructions such as pdep and pext have been very slow up until Zen 3.
         // Check for Zen 3 or newer by its family (0x19).
@@ -144,6 +214,7 @@ HostFeature GetHostFeatures() {
         }
     }
 #endif
+
     return features;
 }
 
@@ -162,27 +233,23 @@ bool IsUnderRosetta() {
 
 }  // anonymous namespace
 
-BlockOfCode::BlockOfCode(RunCodeCallbacks cb, JitStateInfo jsi, size_t total_code_size, std::function<void(BlockOfCode&)> rcp)
-    : Xbyak::CodeGenerator(total_code_size
 #ifdef DYNARMIC_ENABLE_NO_EXECUTE_SUPPORT
-        , Xbyak::DontSetProtectRWE
+static const auto default_cg_mode = Xbyak::DontSetProtectRWE;
 #else
-        , nullptr //Allow RWE
+static const auto default_cg_mode = nullptr; //Allow RWE
 #endif
-        , nullptr)
+
-    , constant_pool(*this, CONSTANT_POOL_SIZE)
+BlockOfCode::BlockOfCode(RunCodeCallbacks cb, JitStateInfo jsi, size_t total_code_size, std::function<void(BlockOfCode&)> rcp)
-    , jsi(jsi)
+        : Xbyak::CodeGenerator(total_code_size, default_cg_mode, &s_allocator)
         , cb(std::move(cb))
-{
+        , jsi(jsi)
+        , constant_pool(*this, CONSTANT_POOL_SIZE)
+        , host_features(GetHostFeatures()) {
     EnableWriting();
     EnsureMemoryCommitted(PRELUDE_COMMIT_SIZE);
     GenRunCode(rcp);
 }
 
-bool BlockOfCode::HasHostFeature(HostFeature feature) const noexcept {
-    return (GetHostFeatures() & feature) == feature;
-}
-
 void BlockOfCode::PreludeComplete() {
     prelude_complete = true;
     code_begin = getCurr();
@@ -274,7 +341,7 @@ void BlockOfCode::GenRunCode(std::function<void(BlockOfCode&)> rcp) {
     mov(rbx, ABI_PARAM2); // save temporarily in non-volatile register
 
     if (cb.enable_cycle_counting) {
-        cb.GetTicksRemaining.EmitCall(*this);
+        cb.GetTicksRemaining->EmitCall(*this);
         mov(qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, cycles_to_run)], ABI_RETURN);
         mov(qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, cycles_remaining)], ABI_RETURN);
     }
@@ -321,7 +388,7 @@ void BlockOfCode::GenRunCode(std::function<void(BlockOfCode&)> rcp) {
         cmp(qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, cycles_remaining)], 0);
         jng(return_to_caller);
     }
-    cb.LookupBlock.EmitCall(*this);
+    cb.LookupBlock->EmitCall(*this);
     jmp(ABI_RETURN);
 
     align();
@@ -334,7 +401,7 @@ void BlockOfCode::GenRunCode(std::function<void(BlockOfCode&)> rcp) {
         jng(return_to_caller_mxcsr_already_exited);
     }
     SwitchMxcsrOnEntry();
-    cb.LookupBlock.EmitCall(*this);
+    cb.LookupBlock->EmitCall(*this);
     jmp(ABI_RETURN);
 
     align();
@@ -348,7 +415,7 @@ void BlockOfCode::GenRunCode(std::function<void(BlockOfCode&)> rcp) {
     L(return_to_caller_mxcsr_already_exited);
 
     if (cb.enable_cycle_counting) {
-        cb.AddTicks.EmitCall(*this, [this](RegList param) {
+        cb.AddTicks->EmitCall(*this, [this](RegList param) {
             mov(param[0], qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, cycles_to_run)]);
             sub(param[0], qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, cycles_remaining)]);
         });
@@ -388,18 +455,18 @@ void BlockOfCode::UpdateTicks() {
         return;
     }
 
-    cb.AddTicks.EmitCall(*this, [this](RegList param) {
+    cb.AddTicks->EmitCall(*this, [this](RegList param) {
         mov(param[0], qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, cycles_to_run)]);
         sub(param[0], qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, cycles_remaining)]);
     });
 
-    cb.GetTicksRemaining.EmitCall(*this);
+    cb.GetTicksRemaining->EmitCall(*this);
     mov(qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, cycles_to_run)], ABI_RETURN);
     mov(qword[rsp + ABI_SHADOW_SPACE + offsetof(StackLayout, cycles_remaining)], ABI_RETURN);
 }
 
 void BlockOfCode::LookupBlock() {
-    cb.LookupBlock.EmitCall(*this);
+    cb.LookupBlock->EmitCall(*this);
 }
 
 void BlockOfCode::LoadRequiredFlagsForCondFromRax(IR::Cond cond) {
@@ -453,7 +520,7 @@ void BlockOfCode::LoadRequiredFlagsForCondFromRax(IR::Cond cond) {
 }
 
 Xbyak::Address BlockOfCode::Const(const Xbyak::AddressFrame& frame, u64 lower, u64 upper) {
-    return constant_pool.GetConstant(*this, frame, lower, upper);
+    return constant_pool.GetConstant(frame, lower, upper);
 }
 
 CodePtr BlockOfCode::GetCodeBegin() const {

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

@@ -31,9 +31,9 @@ namespace Dynarmic::Backend::X64 {
 using CodePtr = const void*;
 
 struct RunCodeCallbacks {
-    ArgCallback LookupBlock;
+    std::unique_ptr<Callback> LookupBlock;
-    ArgCallback AddTicks;
+    std::unique_ptr<Callback> AddTicks;
-    ArgCallback GetTicksRemaining;
+    std::unique_ptr<Callback> GetTicksRemaining;
     bool enable_cycle_counting;
 };
 
@@ -166,24 +166,27 @@ public:
 
     JitStateInfo GetJitStateInfo() const { return jsi; }
 
-    bool HasHostFeature(HostFeature feature) const noexcept;
+    bool HasHostFeature(HostFeature feature) const {
+        return (host_features & feature) == feature;
+    }
 
 private:
     using RunCodeFuncType = HaltReason (*)(void*, CodePtr);
     static constexpr size_t MXCSR_ALREADY_EXITED = 1 << 0;
     static constexpr size_t FORCE_RETURN = 1 << 1;
 
-    ConstantPool constant_pool;
-    JitStateInfo jsi;
-    std::array<const void*, 4> return_from_run_code;
-    RunCodeFuncType run_code = nullptr;
-    RunCodeFuncType step_code = nullptr;
     RunCodeCallbacks cb;
+    JitStateInfo jsi;
     CodePtr code_begin = nullptr;
 #ifdef _WIN32
     size_t committed_size = 0;
 #endif
+    ConstantPool constant_pool;
+    RunCodeFuncType run_code = nullptr;
+    RunCodeFuncType step_code = nullptr;
+    std::array<const void*, 4> return_from_run_code;
     bool prelude_complete = false;
+    const HostFeature host_features;
 
     void GenRunCode(std::function<void(BlockOfCode&)> rcp);
 };

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

@@ -16,7 +16,8 @@
 namespace Dynarmic::Backend::X64 {
 
 ConstantPool::ConstantPool(BlockOfCode& code, size_t size)
-    : insertion_point(0)
+    : code(code)
+    , insertion_point(0)
 {
     code.EnsureMemoryCommitted(align_size + size);
     code.int3();
@@ -24,17 +25,17 @@ ConstantPool::ConstantPool(BlockOfCode& code, size_t size)
     pool = std::span<ConstantT>(reinterpret_cast<ConstantT*>(code.AllocateFromCodeSpace(size)), size / align_size);
 }
 
-Xbyak::Address ConstantPool::GetConstant(BlockOfCode& code, const Xbyak::AddressFrame& frame, u64 lower, u64 upper) {
+Xbyak::Address ConstantPool::GetConstant(const Xbyak::AddressFrame& frame, u64 lower, u64 upper) {
     const auto constant = ConstantT(lower, upper);
-    auto it = constant_info.find(constant);
+    auto iter = constant_info.find(constant);
-    if (it == constant_info.end()) {
+    if (iter == constant_info.end()) {
         ASSERT(insertion_point < pool.size());
         ConstantT& target_constant = pool[insertion_point];
         target_constant = constant;
-        it = constant_info.insert({constant, &target_constant}).first;
+        iter = constant_info.insert({constant, &target_constant}).first;
         ++insertion_point;
     }
-    return frame[code.rip + it->second];
+    return frame[code.rip + iter->second];
 }
 
 }  // namespace Dynarmic::Backend::X64

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

@@ -29,7 +29,7 @@ class ConstantPool final {
 public:
     ConstantPool(BlockOfCode& code, size_t size);
 
-    Xbyak::Address GetConstant(BlockOfCode& code, const Xbyak::AddressFrame& frame, u64 lower, u64 upper = 0);
+    Xbyak::Address GetConstant(const Xbyak::AddressFrame& frame, u64 lower, u64 upper = 0);
 
 private:
     static constexpr size_t align_size = 16;  // bytes
@@ -45,6 +45,7 @@ private:
 
     ankerl::unordered_dense::map<ConstantT, void*, ConstantHash> constant_info;
     std::span<ConstantT> pool;
+    BlockOfCode& code;
     std::size_t insertion_point;
 };
 

src/dynarmic/src/dynarmic/backend/x64/host_feature.h

@@ -12,7 +12,7 @@
 
 namespace Dynarmic::Backend::X64 {
 
-enum class HostFeature : u32 {
+enum class HostFeature : u64 {
     SSSE3 = 1ULL << 0,
     SSE41 = 1ULL << 1,
     SSE42 = 1ULL << 2,

src/dynarmic/src/dynarmic/backend/x64/jitstate_info.h

@@ -1,6 +1,3 @@
-// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
-// SPDX-License-Identifier: GPL-3.0-or-later
-
 /* This file is part of the dynarmic project.
  * Copyright (c) 2016 MerryMage
  * SPDX-License-Identifier: 0BSD
@@ -18,7 +15,7 @@ struct JitStateInfo {
             : offsetof_guest_MXCSR(offsetof(JitStateType, guest_MXCSR))
             , offsetof_asimd_MXCSR(offsetof(JitStateType, asimd_MXCSR))
             , offsetof_rsb_ptr(offsetof(JitStateType, rsb_ptr))
-            , rsb_ptr_mask(JitStateType::RSB_PTR_MASK)
+            , rsb_ptr_mask(JitStateType::RSBPtrMask)
             , offsetof_rsb_location_descriptors(offsetof(JitStateType, rsb_location_descriptors))
             , offsetof_rsb_codeptrs(offsetof(JitStateType, rsb_codeptrs))
             , offsetof_cpsr_nzcv(offsetof(JitStateType, cpsr_nzcv))

src/dynarmic/src/dynarmic/common/context.h

@@ -136,14 +136,14 @@
 #    endif
 #elif defined(ARCHITECTURE_riscv64)
 #   if defined(__FreeBSD__)
-#       define CTX_SEPC (mctx.mc_gpregs.gp_sepc)
+#       define CTX_PC (mctx.mc_gpregs.gp_sepc)
 #       define CTX_SP (mctx.mc_gpregs.gp_sp)
 #   elif defined(__linux__)
-#       define CTX_SEPC (mctx.__gregs[REG_PC])
+#       define CTX_PC (mctx.__gregs[REG_PC])
 #       define CTX_SP (mctx.__gregs[REG_SP])
 #   elif defined(__OpenBSD__)
 // https://github.com/openbsd/src/blob/master/sys/arch/riscv64/include/signal.h
-#        define CTX_SEPC (ucontext->sc_sepc)
+#        define CTX_PC (ucontext->sc_sepc)
 #        define CTX_SP (ucontext->sc_sp)
 #    else
 #        error "unknown platform"

src/qt_common/discord/discord_impl.cpp

@@ -74,8 +74,8 @@ static constexpr char DEFAULT_DISCORD_IMAGE[] =
     "https://git.eden-emu.dev/eden-emu/eden/raw/branch/master/dist/qt_themes/default/icons/256x256/"
     "eden.png";
 
-void DiscordImpl::UpdateGameStatus(bool use_default) {
+void DiscordImpl::UpdateGameStatus(std::string_view game_url, bool has_boxart) {
-    const std::string url = use_default ? std::string{DEFAULT_DISCORD_IMAGE} : game_url;
+    const std::string url = std::string{has_boxart ? game_url : DEFAULT_DISCORD_IMAGE};
     s64 start_time = std::chrono::duration_cast<std::chrono::seconds>(
                          std::chrono::system_clock::now().time_since_epoch())
                          .count();
@@ -98,7 +98,7 @@ void DiscordImpl::Update() {
 
         // Used to format Icon URL for yuzu website game compatibility page
         std::string icon_name = GetGameString(game_title);
-        game_url = fmt::format(
+        auto const game_url = fmt::format(
             "https://raw.githubusercontent.com/eden-emulator/boxart/refs/heads/master/img/{}.png",
             icon_name);
 
@@ -117,7 +117,7 @@ void DiscordImpl::Update() {
         };
 
         auto res = client.send(request);
-        UpdateGameStatus(res && res->status == 200);
+        UpdateGameStatus(game_url, res && res->status == 200);
 
         return;
     }

src/qt_common/discord/discord_impl.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: 2018 Citra Emulator Project
@@ -26,11 +26,9 @@ public:
 
 private:
     std::string GetGameString(const std::string& title);
-    void UpdateGameStatus(bool use_default);
+    void UpdateGameStatus(std::string_view game_url, bool use_default);
 
-    std::string game_url{};
     std::string game_title{};
-
     Core::System& system;
 };
 

src/shader_recompiler/backend/glasm/emit_glasm_instructions.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -665,6 +665,8 @@ void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU3
                      const IR::Value& clamp, const IR::Value& segmentation_mask);
 void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 index,
                           const IR::Value& clamp, const IR::Value& segmentation_mask);
+void EmitQuadBroadcast(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 lane);
+void EmitQuadSwap(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 direction);
 void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, ScalarF32 op_a, ScalarF32 op_b,
                      ScalarU32 swizzle);
 void EmitDPdxFine(EmitContext& ctx, IR::Inst& inst, ScalarF32 op_a);

src/shader_recompiler/backend/glasm/emit_glasm_warp.cpp

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -97,6 +100,24 @@ void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, Sca
     Shuffle(ctx, inst, value, index, clamp, segmentation_mask, "XOR");
 }
 
+void EmitQuadBroadcast(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 lane) {
+    const Register ret{ctx.reg_alloc.Define(inst)};
+    ctx.Add("AND.U RC.x,{}.threadid,~3;"
+            "AND.U RC.y,{},3;"
+            "OR.U RC.x,RC.x,RC.y;"
+            "SHFIDX.U {},{},RC.x,0x1C03;"
+            "MOV.U {}.x,{}.y;",
+            ctx.stage_name, lane, ret, value, ret, ret);
+}
+
+void EmitQuadSwap(EmitContext& ctx, IR::Inst& inst, ScalarU32 value, ScalarU32 direction) {
+    const Register ret{ctx.reg_alloc.Define(inst)};
+    ctx.Add("ADD.U RC.x,{},1;"
+            "SHFXOR.U {},{},RC.x,0x1C03;"
+            "MOV.U {}.x,{}.y;",
+            direction, ret, value, ret, ret);
+}
+
 void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, ScalarF32 op_a, ScalarF32 op_b,
                      ScalarU32 swizzle) {
     const auto ret{ctx.reg_alloc.Define(inst)};

src/shader_recompiler/backend/glsl/emit_glsl_instructions.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -743,6 +743,10 @@ void EmitShuffleDown(EmitContext& ctx, IR::Inst& inst, std::string_view value,
 void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view value,
                           std::string_view index, std::string_view clamp,
                           std::string_view segmentation_mask);
+void EmitQuadBroadcast(EmitContext& ctx, IR::Inst& inst, std::string_view value,
+                       std::string_view lane);
+void EmitQuadSwap(EmitContext& ctx, IR::Inst& inst, std::string_view value,
+                  std::string_view direction);
 void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,
                      std::string_view swizzle);
 void EmitDPdxFine(EmitContext& ctx, IR::Inst& inst, std::string_view op_a);

src/shader_recompiler/backend/glsl/emit_glsl_warp.cpp

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -200,6 +203,18 @@ void EmitShuffleButterfly(EmitContext& ctx, IR::Inst& inst, std::string_view val
     ctx.AddU32("{}=shfl_in_bounds?shfl_result:{};", inst, value);
 }
 
+void EmitQuadBroadcast(EmitContext& ctx, IR::Inst& inst, std::string_view value,
+                       std::string_view lane) {
+    const auto src_thread_id{fmt::format("(({}&~3)|({}& 3))", THREAD_ID, lane)};
+    ctx.AddU32("{}=readInvocationARB({},{});", inst, value, src_thread_id);
+}
+
+void EmitQuadSwap(EmitContext& ctx, IR::Inst& inst, std::string_view value,
+                  std::string_view direction) {
+    const auto src_thread_id{fmt::format("({}^({}+1))", THREAD_ID, direction)};
+    ctx.AddU32("{}=readInvocationARB({},{});", inst, value, src_thread_id);
+}
+
 void EmitFSwizzleAdd(EmitContext& ctx, IR::Inst& inst, std::string_view op_a, std::string_view op_b,
                      std::string_view swizzle) {
     const auto mask{fmt::format("({}>>((gl_SubGroupInvocationARB&3)<<1))&3", swizzle)};

src/shader_recompiler/backend/spirv/emit_spirv.cpp

@@ -322,6 +322,11 @@ void DefineEntryPoint(const IR::Program& program, EmitContext& ctx, Id main) {
         if (ctx.runtime_info.force_early_z) {
             ctx.AddExecutionMode(main, spv::ExecutionMode::EarlyFragmentTests);
         }
+        if (ctx.profile.support_shader_quad_control && program.info.uses_quad_shuffles) {
+            ctx.AddExtension("SPV_KHR_quad_control");
+            ctx.AddCapability(spv::Capability::QuadControlKHR);
+            ctx.AddExecutionMode(main, spv::ExecutionMode::RequireFullQuadsKHR);
+        }
         break;
     default:
         throw NotImplementedException("Stage {}", program.stage);
@@ -335,7 +340,7 @@ void SetupDenormControl(const Profile& profile, const IR::Program& program, Emit
     if (info.uses_fp32_denorms_flush && info.uses_fp32_denorms_preserve) {
         LOG_DEBUG(Shader_SPIRV, "Fp32 denorm flush and preserve on the same shader");
     } else if (info.uses_fp32_denorms_flush) {
-        if (profile.support_fp32_denorm_flush) {
+        if (profile.support_fp32_denorm_flush && !profile.has_broken_fp32_denorm_flush) {
             ctx.AddCapability(spv::Capability::DenormFlushToZero);
             ctx.AddExecutionMode(main_func, spv::ExecutionMode::DenormFlushToZero, 32U);
         } else {
@@ -443,6 +448,12 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
             ctx.AddCapability(spv::Capability::GroupNonUniformVote);
         }
     }
+    if (info.uses_quad_shuffles) {
+        if (profile.support_quad_shuffles) {
+            ctx.AddCapability(spv::Capability::GroupNonUniformQuad);
+        }
+        ctx.AddCapability(spv::Capability::GroupNonUniformShuffle);
+    }
     if (info.uses_int64_bit_atomics && profile.support_int64_atomics) {
         ctx.AddCapability(spv::Capability::Int64Atomics);
     }

src/shader_recompiler/backend/spirv/emit_spirv_instructions.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -622,6 +622,8 @@ Id EmitShuffleDown(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clam
                    Id segmentation_mask);
 Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id clamp,
                         Id segmentation_mask);
+Id EmitQuadBroadcast(EmitContext& ctx, Id value, Id lane);
+Id EmitQuadSwap(EmitContext& ctx, Id value, Id direction);
 Id EmitFSwizzleAdd(EmitContext& ctx, Id op_a, Id op_b, Id swizzle);
 Id EmitDPdxFine(EmitContext& ctx, Id op_a);
 Id EmitDPdyFine(EmitContext& ctx, Id op_a);

src/shader_recompiler/backend/spirv/emit_spirv_warp.cpp

@@ -260,6 +260,21 @@ Id EmitShuffleButterfly(EmitContext& ctx, IR::Inst* inst, Id value, Id index, Id
     return SelectValue(ctx, in_range, value, src_thread_id);
 }
 
+Id EmitQuadBroadcast(EmitContext& ctx, Id value, Id lane) {
+    if (ctx.profile.support_quad_shuffles) {
+        return ctx.OpGroupNonUniformQuadBroadcast(ctx.U32[1], SubgroupScope(ctx), value, lane);
+    }
+    const Id base{ctx.OpBitwiseAnd(ctx.U32[1], GetThreadId(ctx), ctx.Const(~3u))};
+    const Id local_lane{ctx.OpBitwiseAnd(ctx.U32[1], lane, ctx.Const(3u))};
+    const Id src_thread_id{ctx.OpBitwiseOr(ctx.U32[1], base, local_lane)};
+    return ctx.OpGroupNonUniformShuffle(ctx.U32[1], SubgroupScope(ctx), value, src_thread_id);
+}
+
+Id EmitQuadSwap(EmitContext& ctx, Id value, Id direction) {
+    const Id xor_mask{ctx.OpIAdd(ctx.U32[1], direction, ctx.Const(1u))};
+    return ctx.OpGroupNonUniformShuffleXor(ctx.U32[1], SubgroupScope(ctx), value, xor_mask);
+}
+
 Id EmitFSwizzleAdd(EmitContext& ctx, Id op_a, Id op_b, Id swizzle) {
     const Id three{ctx.Const(3U)};
     Id mask{GetThreadId(ctx)};

src/shader_recompiler/frontend/ir/ir_emitter.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -2100,6 +2100,14 @@ U32 IREmitter::ShuffleButterfly(const IR::U32& value, const IR::U32& index, cons
     return Inst<U32>(Opcode::ShuffleButterfly, value, index, clamp, seg_mask);
 }
 
+U32 IREmitter::QuadBroadcast(const IR::U32& value, const IR::U32& lane) {
+    return Inst<U32>(Opcode::QuadBroadcast, value, lane);
+}
+
+U32 IREmitter::QuadSwap(const IR::U32& value, const IR::U32& direction) {
+    return Inst<U32>(Opcode::QuadSwap, value, direction);
+}
+
 F32 IREmitter::FSwizzleAdd(const F32& a, const F32& b, const U32& swizzle, FpControl control) {
     return Inst<F32>(Opcode::FSwizzleAdd, Flags{control}, a, b, swizzle);
 }

src/shader_recompiler/frontend/ir/ir_emitter.h

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -394,6 +394,8 @@ public:
                                   const IR::U32& seg_mask);
     [[nodiscard]] U32 ShuffleButterfly(const IR::U32& value, const IR::U32& index,
                                        const IR::U32& clamp, const IR::U32& seg_mask);
+    [[nodiscard]] U32 QuadBroadcast(const IR::U32& value, const IR::U32& lane);
+    [[nodiscard]] U32 QuadSwap(const IR::U32& value, const IR::U32& direction);
     [[nodiscard]] F32 FSwizzleAdd(const F32& a, const F32& b, const U32& swizzle,
                                   FpControl control = {});
 

src/shader_recompiler/frontend/ir/opcodes.cpp

@@ -10,7 +10,7 @@ namespace Shader::IR {
 
 namespace Detail {
 
-OpcodeMeta META_TABLE[532] = {
+OpcodeMeta META_TABLE[534] = {
 #define OPCODE(name_token, type_token, ...)                                                        \
     {                                                                                              \
         .name{#name_token},                                                                        \
@@ -21,7 +21,7 @@ OpcodeMeta META_TABLE[532] = {
 #undef OPCODE
 };
 
-u8 NUM_ARGS[532] = {
+u8 NUM_ARGS[534] = {
 #define OPCODE(name_token, type_token, ...) u8(CalculateNumArgsOf(Opcode::name_token)),
 #include "opcodes.inc"
 #undef OPCODE

src/shader_recompiler/frontend/ir/opcodes.h

@@ -57,12 +57,12 @@ static constexpr Type F64x2{Type::F64x2};
 static constexpr Type F64x3{Type::F64x3};
 static constexpr Type F64x4{Type::F64x4};
 
-extern OpcodeMeta META_TABLE[532];
+extern OpcodeMeta META_TABLE[534];
 constexpr size_t CalculateNumArgsOf(Opcode op) noexcept {
     const auto& arg_types = META_TABLE[size_t(op)].arg_types;
     return size_t(std::distance(arg_types.begin(), std::ranges::find(arg_types, Type::Void)));
 }
-extern u8 NUM_ARGS[532];
+extern u8 NUM_ARGS[534];
 } // namespace Detail
 
 /// Get return type of an opcode

src/shader_recompiler/frontend/ir/opcodes.inc

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -579,6 +582,8 @@ OPCODE(ShuffleIndex,                                        U32,            U32,
 OPCODE(ShuffleUp,                                           U32,            U32,            U32,            U32,            U32,                            )
 OPCODE(ShuffleDown,                                         U32,            U32,            U32,            U32,            U32,                            )
 OPCODE(ShuffleButterfly,                                    U32,            U32,            U32,            U32,            U32,                            )
+OPCODE(QuadBroadcast,                                       U32,            U32,            U32,                                                                    )
+OPCODE(QuadSwap,                                            U32,            U32,            U32,                                                                    )
 OPCODE(FSwizzleAdd,                                         F32,            F32,            F32,            U32,                                            )
 OPCODE(DPdxFine,                                            F32,            F32,                                                                            )
 OPCODE(DPdyFine,                                            F32,            F32,                                                                            )

src/shader_recompiler/frontend/maxwell/translate/impl/warp_shuffle.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
 // SPDX-License-Identifier: GPL-3.0-or-later
 
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
@@ -36,7 +36,10 @@ enum class ShuffleMode : u64 {
     }
 }
 
-void Shuffle(TranslatorVisitor& v, u64 insn, const IR::U32& index, const IR::U32& mask) {
+constexpr u32 QUAD_MASK = (28u << 8) | 3u;
+
+void Shuffle(TranslatorVisitor& v, u64 insn, const IR::U32& index, const IR::U32& mask,
+             bool index_is_imm, u32 index_imm, bool mask_is_imm, u32 mask_imm) {
     union {
         u64 insn;
         BitField<0, 8, IR::Reg> dest_reg;
@@ -45,6 +48,21 @@ void Shuffle(TranslatorVisitor& v, u64 insn, const IR::U32& index, const IR::U32
         BitField<48, 3, IR::Pred> pred;
     } const shfl{insn};
 
+    const bool is_quad_candidate{mask_is_imm && mask_imm == QUAD_MASK && index_is_imm &&
+                                  v.env.ShaderStage() == Stage::Fragment};
+    if (is_quad_candidate) {
+        if (shfl.mode == ShuffleMode::IDX && index_imm <= 3) {
+            v.X(shfl.dest_reg, v.ir.QuadBroadcast(v.X(shfl.src_reg), v.ir.Imm32(index_imm)));
+            v.ir.SetPred(shfl.pred, v.ir.Imm1(true));
+            return;
+        }
+        if (shfl.mode == ShuffleMode::BFLY && index_imm >= 1 && index_imm <= 3) {
+            v.X(shfl.dest_reg, v.ir.QuadSwap(v.X(shfl.src_reg), v.ir.Imm32(index_imm - 1)));
+            v.ir.SetPred(shfl.pred, v.ir.Imm1(true));
+            return;
+        }
+    }
+
     const IR::U32 result{ShuffleOperation(v.ir, v.X(shfl.src_reg), index, mask, shfl.mode)};
     v.ir.SetPred(shfl.pred, v.ir.GetInBoundsFromOp(result));
     v.X(shfl.dest_reg, result);
@@ -59,11 +77,14 @@ void TranslatorVisitor::SHFL(u64 insn) {
         BitField<29, 1, u64> src_b_flag;
         BitField<34, 13, u64> src_b_imm;
     } const flags{insn};
-    const IR::U32 src_a{flags.src_a_flag != 0 ? ir.Imm32(static_cast<u32>(flags.src_a_imm))
+    const bool index_is_imm{flags.src_a_flag != 0};
+    const bool mask_is_imm{flags.src_b_flag != 0};
+    const IR::U32 src_a{index_is_imm ? ir.Imm32(static_cast<u32>(flags.src_a_imm))
                                      : GetReg20(insn)};
-    const IR::U32 src_b{flags.src_b_flag != 0 ? ir.Imm32(static_cast<u32>(flags.src_b_imm))
+    const IR::U32 src_b{mask_is_imm ? ir.Imm32(static_cast<u32>(flags.src_b_imm))
                                     : GetReg39(insn)};
-    Shuffle(*this, insn, src_a, src_b);
+    Shuffle(*this, insn, src_a, src_b, index_is_imm, static_cast<u32>(flags.src_a_imm),
+            mask_is_imm, static_cast<u32>(flags.src_b_imm));
 }
 
 } // namespace Shader::Maxwell

src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp

@@ -498,6 +498,10 @@ void VisitUsages(Info& info, IR::Inst& inst) {
     case IR::Opcode::ShuffleButterfly:
         info.uses_subgroup_shuffles = true;
         break;
+    case IR::Opcode::QuadBroadcast:
+    case IR::Opcode::QuadSwap:
+        info.uses_quad_shuffles = true;
+        break;
     case IR::Opcode::GetCbufU8:
     case IR::Opcode::GetCbufS8:
     case IR::Opcode::GetCbufU16:

src/shader_recompiler/profile.h

@@ -37,6 +37,8 @@ struct Profile {
     bool support_explicit_workgroup_layout{};
     bool support_workgroup_layout_8bit_access{};
     bool support_workgroup_layout_16bit_access{};
+    bool support_shader_quad_control{};
+    bool support_quad_shuffles{};
     bool support_vote{};
     u32 supported_subgroup_stages{0x7F};
     bool support_viewport_index_layer_non_geometry{};
@@ -86,6 +88,8 @@ struct Profile {
     bool has_broken_signed_operations{};
     /// Float controls break when fp16 is enabled
     bool has_broken_fp16_float_controls{};
+    /// Declaring fp32 denorm flush to zero miscompiles on some drivers
+    bool has_broken_fp32_denorm_flush{};
     /// Dynamic vec4 indexing is broken on some OpenGL drivers
     bool has_gl_component_indexing_bug{};
     /// The precise type qualifier is broken in the fragment stage of some drivers

src/shader_recompiler/shader_info.h

@@ -252,6 +252,7 @@ struct Info {
     bool uses_is_helper_invocation{};
     bool uses_subgroup_invocation_id{};
     bool uses_subgroup_shuffles{};
+    bool uses_quad_shuffles{};
     std::array<bool, 30> uses_patches{};
 
     std::array<Interpolation, 32> interpolation{};

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -231,6 +231,7 @@ ShaderCache::ShaderCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
           .has_broken_unsigned_image_offsets = true,
           .has_broken_signed_operations = true,
           .has_broken_fp16_float_controls = false,
+          .has_broken_fp32_denorm_flush = false,
           .has_gl_component_indexing_bug = device.HasComponentIndexingBug(),
           .has_gl_precise_bug = device.HasPreciseBug(),
           .has_gl_cbuf_ftou_bug = device.HasCbufFtouBug(),

src/video_core/renderer_vulkan/pipeline_helper.h

@@ -370,20 +370,9 @@ inline void PushImageDescriptors(TextureCache& texture_cache,
                 const VkImageView null_image_view{texture_cache.GetImageView(VideoCommon::NULL_IMAGE_VIEW_ID).Handle(desc.type)};
                 if (null_image_view != VK_NULL_HANDLE) vk_image_view = null_image_view;
             }
-            const Sampler& sampler{texture_cache.GetSampler(sampler_id)};
+            Sampler& sampler{texture_cache.GetSampler(sampler_id)};
-            const bool use_fallback_sampler{sampler.HasAddedAnisotropy() &&
+            guest_descriptor_queue.AddSampledImage(vk_image_view,
-                                            !image_view.SupportsAnisotropy()};
+                                                   sampler.HandleFor(image_view, desc.is_depth));
-            VkSampler vk_sampler{use_fallback_sampler ? sampler.HandleWithDefaultAnisotropy()
-                                                      : sampler.Handle()};
-            if (sampler.HasLinearFiltering() &&
-                VideoCore::Surface::IsPixelFormatInteger(image_view.format)) {
-                vk_sampler = sampler.HandleWithNearestFilter();
-            }
-            if (desc.is_depth && sampler.HasDepthComparison() &&
-                !image_view.SupportsDepthComparison()) {
-                vk_sampler = sampler.HandleWithoutDepthComparison();
-            }
-            guest_descriptor_queue.AddSampledImage(vk_image_view, vk_sampler);
             const bool element_rescaled{texture_cache.IsRescaling(image_view)};
             is_rescaled |= element_rescaled;
         }

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -404,6 +404,8 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
             device.IsWorkgroupMemoryExplicitLayout8BitAccessSupported(),
         .support_workgroup_layout_16bit_access =
             device.IsWorkgroupMemoryExplicitLayout16BitAccessSupported(),
+        .support_shader_quad_control = device.IsKhrShaderQuadControlSupported(),
+        .support_quad_shuffles = device.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_QUAD_BIT),
         .support_vote = device.IsSubgroupFeatureSupported(VK_SUBGROUP_FEATURE_VOTE_BIT),
         .supported_subgroup_stages = supported_subgroup_stages,
         .support_viewport_index_layer_non_geometry =
@@ -444,6 +446,7 @@ PipelineCache::PipelineCache(Tegra::MaxwellDeviceMemoryManager& device_memory_,
         .has_broken_unsigned_image_offsets = false,
         .has_broken_signed_operations = false,
         .has_broken_fp16_float_controls = driver_id == VK_DRIVER_ID_NVIDIA_PROPRIETARY,
+        .has_broken_fp32_denorm_flush = driver_id == VK_DRIVER_ID_QUALCOMM_PROPRIETARY,
         .ignore_nan_fp_comparisons = false,
         .has_broken_spirv_subgroup_mask_vector_extract_dynamic = false,
         .has_broken_robust =

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -51,6 +51,7 @@ using VideoCore::Surface::BytesPerBlock;
 using VideoCore::Surface::HasAlpha;
 using VideoCore::Surface::IsPixelFormatASTC;
 using VideoCore::Surface::IsPixelFormatInteger;
+using VideoCore::Surface::IsPixelFormatSRGB;
 using VideoCore::Surface::SurfaceType;
 
 namespace {
@@ -59,6 +60,17 @@ constexpr bool ENABLE_MSAA_RESOLVE_CONSUME = true;
 constexpr bool ENABLE_MSAA_COLOR_DISCARD = true;
 constexpr bool ENABLE_MSAA_DEPTH_STENCIL_DISCARD = true;
 
+[[nodiscard]] constexpr bool NeedsExplicitBorderColorFormat(VkFormat format) {
+    switch (format) {
+    case VK_FORMAT_B4G4R4A4_UNORM_PACK16:
+    case VK_FORMAT_B5G6R5_UNORM_PACK16:
+    case VK_FORMAT_B5G5R5A1_UNORM_PACK16:
+        return true;
+    default:
+        return false;
+    }
+}
+
 constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
     if (color == std::array<float, 4>{0, 0, 0, 0}) {
         return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
@@ -1802,10 +1814,6 @@ bool TextureCacheRuntime::CanReportMemoryUsage() const {
     return device.CanReportMemoryUsage();
 }
 
-std::optional<size_t> TextureCacheRuntime::GetSamplerHeapBudget() const {
-    return device.GetSamplerHeapBudget();
-}
-
 void TextureCacheRuntime::FlushDeferredClear() {
     scheduler.FlushDeferredClear();
 }
@@ -2517,9 +2525,23 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
         supports_depth_comparison =
             (properties3.optimalTilingFeatures &
              VK_FORMAT_FEATURE_2_SAMPLED_IMAGE_DEPTH_COMPARISON_BIT) != 0;
+        supports_minmax_filter = (properties3.optimalTilingFeatures &
+                                  VK_FORMAT_FEATURE_2_SAMPLED_IMAGE_FILTER_MINMAX_BIT) != 0;
     } else {
         supports_depth_comparison = true;
+        supports_minmax_filter =
+            (device->GetPhysical().GetFormatProperties(format_info.format).optimalTilingFeatures &
+             VK_FORMAT_FEATURE_SAMPLED_IMAGE_FILTER_MINMAX_BIT) != 0;
     }
+    requires_border_color_format = NeedsExplicitBorderColorFormat(format_info.format);
+    swizzle_mapping = VkComponentMapping{
+        .r = ComponentSwizzle(swizzle[0]),
+        .g = ComponentSwizzle(swizzle[1]),
+        .b = ComponentSwizzle(swizzle[2]),
+        .a = ComponentSwizzle(swizzle[3]),
+    };
+    has_identity_swizzle = swizzle[0] == SwizzleSource::R && swizzle[1] == SwizzleSource::G &&
+                           swizzle[2] == SwizzleSource::B && swizzle[3] == SwizzleSource::A;
     const VkImageUsageFlags requested_view_usage = ImageUsageFlags(format_info, format);
     const VkImageUsageFlags image_usage = image.UsageFlags();
     const VkImageUsageFlags clamped_view_usage = requested_view_usage & image_usage;
@@ -2544,12 +2566,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
         .image = image.Handle(),
         .viewType = VkImageViewType{},
         .format = format_info.format,
-        .components{
+        .components = swizzle_mapping,
-            .r = ComponentSwizzle(swizzle[0]),
-            .g = ComponentSwizzle(swizzle[1]),
-            .b = ComponentSwizzle(swizzle[2]),
-            .a = ComponentSwizzle(swizzle[3]),
-        },
         .subresourceRange = MakeSubresourceRange(aspect_mask, info.range),
     };
     const auto create = [&](TextureType tex_type, std::optional<u32> num_layers) {
@@ -2718,92 +2735,223 @@ vk::ImageView ImageView::MakeView(VkFormat vk_format, VkImageAspectFlags aspect_
     });
 }
 
+CustomBorderColorBudget::~CustomBorderColorBudget() {
+    Release();
+}
+
+CustomBorderColorBudget::CustomBorderColorBudget(CustomBorderColorBudget&& rhs) noexcept
+    : device_ptr{std::exchange(rhs.device_ptr, nullptr)}, held{std::exchange(rhs.held, 0)} {}
+
+CustomBorderColorBudget& CustomBorderColorBudget::operator=(
+    CustomBorderColorBudget&& rhs) noexcept {
+    if (this != &rhs) {
+        Release();
+        device_ptr = std::exchange(rhs.device_ptr, nullptr);
+        held = std::exchange(rhs.held, 0);
+    }
+    return *this;
+}
+
+bool CustomBorderColorBudget::TryAcquire(const Device& device, size_t count) {
+    if (!device.TryReserveCustomBorderColorSamplers(count)) {
+        return false;
+    }
+    device_ptr = &device;
+    held += count;
+    return true;
+}
+
+void CustomBorderColorBudget::Release() noexcept {
+    if (device_ptr != nullptr) {
+        device_ptr->ReleaseCustomBorderColorSamplers(held);
+    }
+    device_ptr = nullptr;
+    held = 0;
+}
+
 Sampler::Sampler(TextureCacheRuntime& runtime, const Tegra::Texture::TSCEntry& tsc) {
     const auto& device = runtime.device;
-    const bool has_custom_border_extension = runtime.device.IsExtCustomBorderColorSupported();
+    device_ptr = &device;
-    const bool has_format_undefined =
+    border_color = tsc.BorderColor();
-        has_custom_border_extension && runtime.device.IsCustomBorderColorWithoutFormatSupported();
+    srgb_border_color = tsc.SrgbBorderColor();
-    const bool has_custom_border_colors =
-        has_format_undefined && runtime.device.IsCustomBorderColorsSupported();
-    const auto color = tsc.BorderColor();
 
+    const f32 max_anisotropy = std::clamp(tsc.MaxAnisotropy(), 1.0f, 16.0f);
+    default_anisotropy = static_cast<f32>(1U << tsc.max_anisotropy);
+
+    const VkFilter mag_filter{MaxwellToVK::Sampler::Filter(tsc.mag_filter)};
+    const VkFilter min_filter{MaxwellToVK::Sampler::Filter(tsc.min_filter)};
+    const VkSamplerMipmapMode mipmap_mode{MaxwellToVK::Sampler::MipmapMode(tsc.mipmap_filter)};
+
+    const VkSamplerAddressMode wrap_u{
+        MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_u, tsc.mag_filter)};
+    const VkSamplerAddressMode wrap_v{
+        MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_v, tsc.mag_filter)};
+    const VkSamplerAddressMode wrap_p{
+        MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_p, tsc.mag_filter)};
+    const bool samples_border = wrap_u == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER ||
+                                wrap_v == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER ||
+                                wrap_p == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
+
+    reduction_mode = MaxwellToVK::SamplerReduction(tsc.reduction_filter);
+
+    has_added_anisotropy = max_anisotropy > default_anisotropy;
+    has_linear_filtering = mag_filter == VK_FILTER_LINEAR || min_filter == VK_FILTER_LINEAR ||
+                           mipmap_mode == VK_SAMPLER_MIPMAP_MODE_LINEAR;
+    has_depth_comparison = tsc.depth_compare_enabled != 0;
+    has_minmax_reduction = reduction_mode != VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT;
+    has_srgb_border_color = tsc.srgb_conversion != 0 && srgb_border_color != border_color;
+    if (has_minmax_reduction && !device.IsExtSamplerFilterMinmaxSupported()) {
+        LOG_WARNING(Render_Vulkan, "VK_EXT_sampler_filter_minmax is required");
+        has_minmax_reduction = false;
+    }
+    has_custom_border_colors = samples_border && device.IsCustomBorderColorUsable();
+    needs_swizzle_mapping = has_custom_border_colors && device.NeedsBorderColorSwizzleMapping();
+
+    if (has_custom_border_colors && GPU::Logging::IsActive()) {
+        GPU::Logging::GPULogger::GetInstance().LogExtensionUsage(
+            "VK_EXT_custom_border_color", "Sampler::Sampler");
+    }
+    if (device.IsExtBorderColorSwizzleSupported() && GPU::Logging::IsActive()) {
+        GPU::Logging::GPULogger::GetInstance().LogExtensionUsage(
+            "VK_EXT_border_color_swizzle", "Sampler::Sampler");
+    }
+
+    f32 min_lod = 0.0f;
+    f32 max_lod = 0.25f;
+    if (tsc.mipmap_filter != TextureMipmapFilter::None) {
+        min_lod = tsc.MinLod();
+        max_lod = tsc.MaxLod();
+    }
+    base_ci = VkSamplerCreateInfo{
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .magFilter = mag_filter,
+        .minFilter = min_filter,
+        .mipmapMode = mipmap_mode,
+        .addressModeU = wrap_u,
+        .addressModeV = wrap_v,
+        .addressModeW = wrap_p,
+        .mipLodBias = tsc.LodBias(),
+        .anisotropyEnable = static_cast<VkBool32>(max_anisotropy > 1.0f),
+        .maxAnisotropy = max_anisotropy,
+        .compareEnable = static_cast<VkBool32>(tsc.depth_compare_enabled),
+        .compareOp = MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func),
+        .minLod = min_lod,
+        .maxLod = max_lod,
+        .borderColor = VK_BORDER_COLOR_FLOAT_CUSTOM_EXT,
+        .unnormalizedCoordinates = VK_FALSE,
+    };
+    variants.reserve(1);
+    Emplace(VariantKey{});
+}
+
+Sampler::VariantKey Sampler::MakeKey(const ImageView& image_view, bool is_depth) const noexcept {
+    VariantKey key{};
+    key.reduce_anisotropy = has_added_anisotropy && !image_view.SupportsAnisotropy();
+    key.force_nearest = has_linear_filtering && IsPixelFormatInteger(image_view.format);
+    key.drop_depth_comparison =
+        is_depth && has_depth_comparison && !image_view.SupportsDepthComparison();
+    key.drop_reduction = has_minmax_reduction && !image_view.SupportsMinmaxFilter();
+    key.drop_custom_border = has_custom_border_colors && image_view.RequiresBorderColorFormat();
+    key.srgb_border = has_srgb_border_color && IsPixelFormatSRGB(image_view.format);
+    if (needs_swizzle_mapping && !key.drop_custom_border && !image_view.HasIdentitySwizzle()) {
+        const VkComponentMapping& mapping = image_view.Swizzle();
+        key.swizzle = {mapping.r, mapping.g, mapping.b, mapping.a};
+    }
+    return key;
+}
+
+VkSampler Sampler::Find(const VariantKey& key) const noexcept {
+    const auto it = std::ranges::find(variants, key, &Variant::key);
+    if (it == variants.end()) {
+        return VK_NULL_HANDLE;
+    }
+    return *it->sampler;
+}
+
+VkSampler Sampler::Emplace(VariantKey key) {
+    bool custom_border = has_custom_border_colors && !key.drop_custom_border;
+    if (custom_border && !custom_border_color_budget.TryAcquire(*device_ptr, 1)) {
+        static bool warned_budget = false;
+        if (!warned_budget) {
+            warned_budget = true;
+        }
+        custom_border = false;
+        key.drop_custom_border = true;
+        key.swizzle = {};
+        if (const VkSampler existing = Find(key); existing != VK_NULL_HANDLE) {
+            return existing;
+        }
+    }
+
+    std::array<float, 4> color = border_color;
+    if (key.srgb_border) {
+        color = srgb_border_color;
+    }
     const VkSamplerCustomBorderColorCreateInfoEXT border_ci{
         .sType = VK_STRUCTURE_TYPE_SAMPLER_CUSTOM_BORDER_COLOR_CREATE_INFO_EXT,
         .pNext = nullptr,
         .customBorderColor = std::bit_cast<VkClearColorValue>(color),
         .format = VK_FORMAT_UNDEFINED,
     };
-    const void* pnext = nullptr;
+    const VkSamplerBorderColorComponentMappingCreateInfoEXT mapping_ci{
-    if (has_custom_border_colors) {
+        .sType = VK_STRUCTURE_TYPE_SAMPLER_BORDER_COLOR_COMPONENT_MAPPING_CREATE_INFO_EXT,
-        pnext = &border_ci;
+        .pNext = &border_ci,
-        if (GPU::Logging::IsActive()) {
+        .components{key.swizzle[0], key.swizzle[1], key.swizzle[2], key.swizzle[3]},
-            GPU::Logging::GPULogger::GetInstance().LogExtensionUsage(
+        .srgb = VK_FALSE,
-                "VK_EXT_custom_border_color", "Sampler::Sampler");
+    };
+    const void* chain = nullptr;
+    if (custom_border) {
+        chain = &border_ci;
+        if (key.HasSwizzle()) {
+            chain = &mapping_ci;
         }
     }
-    if (device.IsExtBorderColorSwizzleSupported() && GPU::Logging::IsActive()) {
-        GPU::Logging::GPULogger::GetInstance().LogExtensionUsage(
-            "VK_EXT_border_color_swizzle", "Sampler::Sampler");
-    }
     const VkSamplerReductionModeCreateInfoEXT reduction_ci{
         .sType = VK_STRUCTURE_TYPE_SAMPLER_REDUCTION_MODE_CREATE_INFO_EXT,
-        .pNext = pnext,
+        .pNext = chain,
-        .reductionMode = MaxwellToVK::SamplerReduction(tsc.reduction_filter),
+        .reductionMode = reduction_mode,
     };
-    if (runtime.device.IsExtSamplerFilterMinmaxSupported()) {
+    if (has_minmax_reduction && !key.drop_reduction) {
-        pnext = &reduction_ci;
+        chain = &reduction_ci;
-    } else if (reduction_ci.reductionMode != VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT) {
-        LOG_WARNING(Render_Vulkan, "VK_EXT_sampler_filter_minmax is required");
     }
-    // Some games have samplers with garbage. Sanitize them here.
-    const f32 max_anisotropy = std::clamp(tsc.MaxAnisotropy(), 1.0f, 16.0f);
 
-    const VkFilter mag_filter{MaxwellToVK::Sampler::Filter(tsc.mag_filter)};
+    VkSamplerCreateInfo create_info = base_ci;
-    const VkFilter min_filter{MaxwellToVK::Sampler::Filter(tsc.min_filter)};
+    create_info.pNext = chain;
-    const VkSamplerMipmapMode mipmap_mode{MaxwellToVK::Sampler::MipmapMode(tsc.mipmap_filter)};
+    if (key.force_nearest) {
-    const bool has_linear_filtering{mag_filter == VK_FILTER_LINEAR ||
+        create_info.magFilter = VK_FILTER_NEAREST;
-                                    min_filter == VK_FILTER_LINEAR ||
+        create_info.minFilter = VK_FILTER_NEAREST;
-                                    mipmap_mode == VK_SAMPLER_MIPMAP_MODE_LINEAR};
+        create_info.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST;
-
+        create_info.anisotropyEnable = VK_FALSE;
-    const auto create_sampler = [&](const f32 anisotropy, bool force_nearest,
+        create_info.maxAnisotropy = 1.0f;
-                                    bool disable_compare = false) {
+    } else if (key.reduce_anisotropy) {
-        return device.GetLogical().CreateSampler(VkSamplerCreateInfo{
+        create_info.anisotropyEnable = static_cast<VkBool32>(default_anisotropy > 1.0f);
-            .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
+        create_info.maxAnisotropy = default_anisotropy;
-            .pNext = pnext,
+    }
-            .flags = 0,
+    if (key.drop_depth_comparison) {
-            .magFilter = force_nearest ? VK_FILTER_NEAREST : mag_filter,
+        create_info.compareEnable = VK_FALSE;
-            .minFilter = force_nearest ? VK_FILTER_NEAREST : min_filter,
+    }
-            .mipmapMode = force_nearest ? VK_SAMPLER_MIPMAP_MODE_NEAREST : mipmap_mode,
+    if (!custom_border) {
-            .addressModeU = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_u, tsc.mag_filter),
+        create_info.borderColor = ConvertBorderColor(color);
-            .addressModeV = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_v, tsc.mag_filter),
+    }
-            .addressModeW = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_p, tsc.mag_filter),
+    variants.push_back(Variant{
-            .mipLodBias = tsc.LodBias(),
+        .key = key,
-            .anisotropyEnable =
+        .sampler = device_ptr->GetLogical().CreateSampler(create_info),
-                static_cast<VkBool32>(!force_nearest && anisotropy > 1.0f ? VK_TRUE : VK_FALSE),
-            .maxAnisotropy = force_nearest ? 1.0f : anisotropy,
-            .compareEnable = disable_compare ? VK_FALSE
-                                             : static_cast<VkBool32>(tsc.depth_compare_enabled),
-            .compareOp = MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func),
-            .minLod = tsc.mipmap_filter == TextureMipmapFilter::None ? 0.0f : tsc.MinLod(),
-            .maxLod = tsc.mipmap_filter == TextureMipmapFilter::None ? 0.25f : tsc.MaxLod(),
-            .borderColor = has_custom_border_colors ? VK_BORDER_COLOR_FLOAT_CUSTOM_EXT
-                                                    : ConvertBorderColor(color),
-            .unnormalizedCoordinates = VK_FALSE,
     });
-    };
+    return *variants.back().sampler;
+}
 
-    sampler = create_sampler(max_anisotropy, false);
+VkSampler Sampler::HandleFor(const ImageView& image_view, bool is_depth) {
-
+    VariantKey key = MakeKey(image_view, is_depth);
-    const f32 max_anisotropy_default = static_cast<f32>(1U << tsc.max_anisotropy);
+    if (variants.size() >= MAX_VARIANTS) {
-    if (max_anisotropy > max_anisotropy_default) {
+        key.srgb_border = false;
-        sampler_default_anisotropy = create_sampler(max_anisotropy_default, false);
+        key.swizzle = {};
     }
-    if (has_linear_filtering) {
+    if (const VkSampler existing = Find(key); existing != VK_NULL_HANDLE) {
-        sampler_nearest = create_sampler(1.0f, true);
+        return existing;
-    }
-    if (tsc.depth_compare_enabled) {
-        sampler_noncompare = create_sampler(max_anisotropy, false, true);
     }
+    return Emplace(key);
 }
 
 Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM_RT> color_buffers,

src/video_core/renderer_vulkan/vk_texture_cache.h

@@ -68,8 +68,6 @@ public:
 
     bool CanReportMemoryUsage() const;
 
-    std::optional<size_t> GetSamplerHeapBudget() const;
-
     bool CanDownloadMsaa(const VideoCommon::ImageInfo& info) const;
 
     [[nodiscard]] VkImage AcquireMsaaScratchImage(const VkImageCreateInfo& image_ci);
@@ -446,6 +444,22 @@ public:
         return supports_depth_comparison;
     }
 
+    [[nodiscard]] bool RequiresBorderColorFormat() const noexcept {
+        return requires_border_color_format;
+    }
+
+    [[nodiscard]] bool SupportsMinmaxFilter() const noexcept {
+        return supports_minmax_filter;
+    }
+
+    [[nodiscard]] const VkComponentMapping& Swizzle() const noexcept {
+        return swizzle_mapping;
+    }
+
+    [[nodiscard]] bool HasIdentitySwizzle() const noexcept {
+        return has_identity_swizzle;
+    }
+
     [[nodiscard]] GPUVAddr GpuAddr() const noexcept {
         return gpu_addr;
     }
@@ -478,48 +492,91 @@ private:
     VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT;
     u32 buffer_size = 0;
 
+    VkComponentMapping swizzle_mapping{};
+
     bool supports_depth_comparison = false;
+    bool requires_border_color_format = false;
+    bool supports_minmax_filter = false;
+    bool has_identity_swizzle = true;
 };
 
 class ImageAlloc : public VideoCommon::ImageAllocBase {};
 
+class CustomBorderColorBudget {
+public:
+    CustomBorderColorBudget() = default;
+    ~CustomBorderColorBudget();
+
+    CustomBorderColorBudget(const CustomBorderColorBudget&) = delete;
+    CustomBorderColorBudget& operator=(const CustomBorderColorBudget&) = delete;
+
+    CustomBorderColorBudget(CustomBorderColorBudget&& rhs) noexcept;
+    CustomBorderColorBudget& operator=(CustomBorderColorBudget&& rhs) noexcept;
+
+    bool TryAcquire(const Device& device, size_t count);
+
+private:
+    void Release() noexcept;
+
+    const Device* device_ptr = nullptr;
+    size_t held = 0;
+};
+
 class Sampler {
 public:
     explicit Sampler(TextureCacheRuntime&, const Tegra::Texture::TSCEntry&);
 
     [[nodiscard]] VkSampler Handle() const noexcept {
-        return *sampler;
+        return *variants.front().sampler;
     }
 
-    [[nodiscard]] VkSampler HandleWithDefaultAnisotropy() const noexcept {
+    [[nodiscard]] VkSampler HandleFor(const ImageView& image_view, bool is_depth);
-        return *sampler_default_anisotropy;
-    }
-
-    [[nodiscard]] bool HasAddedAnisotropy() const noexcept {
-        return static_cast<bool>(sampler_default_anisotropy);
-    }
-
-    [[nodiscard]] VkSampler HandleWithNearestFilter() const noexcept {
-        return *sampler_nearest;
-    }
-
-    [[nodiscard]] bool HasLinearFiltering() const noexcept {
-        return static_cast<bool>(sampler_nearest);
-    }
-
-    [[nodiscard]] VkSampler HandleWithoutDepthComparison() const noexcept {
-        return *sampler_noncompare;
-    }
-
-    [[nodiscard]] bool HasDepthComparison() const noexcept {
-        return static_cast<bool>(sampler_noncompare);
-    }
 
 private:
+    struct VariantKey {
+        bool reduce_anisotropy;
+        bool force_nearest;
+        bool drop_depth_comparison;
+        bool drop_reduction;
+        bool drop_custom_border;
+        bool srgb_border;
+        std::array<VkComponentSwizzle, 4> swizzle;
+
+        bool operator==(const VariantKey&) const noexcept = default;
+
+        [[nodiscard]] bool HasSwizzle() const noexcept {
+            return swizzle != std::array<VkComponentSwizzle, 4>{};
+        }
+    };
+
+    struct Variant {
+        VariantKey key;
         vk::Sampler sampler;
-    vk::Sampler sampler_default_anisotropy;
+    };
-    vk::Sampler sampler_nearest;
+
-    vk::Sampler sampler_noncompare;
+    static constexpr size_t MAX_VARIANTS = 32;
+
+    [[nodiscard]] VariantKey MakeKey(const ImageView& image_view, bool is_depth) const noexcept;
+    [[nodiscard]] VkSampler Find(const VariantKey& key) const noexcept;
+    VkSampler Emplace(VariantKey key);
+
+    CustomBorderColorBudget custom_border_color_budget;
+    std::vector<Variant> variants;
+
+    const Device* device_ptr{nullptr};
+    VkSamplerCreateInfo base_ci{};
+    VkSamplerReductionModeEXT reduction_mode{VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE_EXT};
+    std::array<float, 4> border_color{};
+    std::array<float, 4> srgb_border_color{};
+    f32 default_anisotropy{1.0f};
+
+    bool has_added_anisotropy{};
+    bool has_linear_filtering{};
+    bool has_depth_comparison{};
+    bool has_minmax_reduction{};
+    bool has_custom_border_colors{};
+    bool has_srgb_border_color{};
+    bool needs_swizzle_mapping{};
 };
 
 struct TextureCacheParams {

src/video_core/texture_cache/texture_cache.h

@@ -1896,67 +1896,10 @@ SamplerId TextureCache<P>::FindSampler(const TSCEntry& config, bool compute) {
     const auto [pair, is_new] = channel_state->samplers.try_emplace(config);
     if (is_new) {
         pair->second = slot_samplers.insert(runtime, config);
-        EnforceSamplerBudget();
     }
     return pair->second;
 }
 
-template <class P>
-std::optional<size_t> TextureCache<P>::QuerySamplerBudget() const {
-    if constexpr (requires { runtime.GetSamplerHeapBudget(); }) {
-        return runtime.GetSamplerHeapBudget();
-    } else {
-        return std::nullopt;
-    }
-}
-
-template <class P>
-void TextureCache<P>::EnforceSamplerBudget() {
-    if (auto const budget = QuerySamplerBudget(); budget) {
-        if (slot_samplers.size() < *budget) {
-            return;
-        }
-        if (!channel_state) {
-            return;
-        }
-        if (last_sampler_gc_frame == frame_tick) {
-            return;
-        }
-        last_sampler_gc_frame = frame_tick;
-        TrimInactiveSamplers(*budget);
-    }
-}
-
-template <class P>
-void TextureCache<P>::TrimInactiveSamplers(size_t budget) {
-    if (channel_state->samplers.size() > 0) {
-        constexpr size_t SAMPLER_GC_SLACK = 1024;
-        ankerl::unordered_dense::set<SamplerId> active_sampler_ids;
-        for (auto const& e : channel_state->sampler_ids)
-            active_sampler_ids.insert(e.second);
-        // Elements in the map must be necesarily valid
-        size_t removed = 0;
-        for (auto it = channel_state->samplers.begin(); it != channel_state->samplers.end();) {
-            const SamplerId sampler_id = it->second;
-            if (!sampler_id || sampler_id == CORRUPT_ID) {
-                it = channel_state->samplers.erase(it);
-            } else if (std::ranges::find(active_sampler_ids, sampler_id) != active_sampler_ids.end()) {
-                ++it;
-            } else {
-                slot_samplers.erase(sampler_id);
-                it = channel_state->samplers.erase(it);
-                ++removed;
-                if (slot_samplers.size() + SAMPLER_GC_SLACK <= budget) {
-                    break;
-                }
-            }
-        }
-        if (removed != 0) {
-            LOG_WARNING(HW_GPU, "Sampler cache exceeded {} entries on this driver; reclaimed {} inactive samplers", budget, removed);
-        }
-    }
-}
-
 template <class P>
 ImageViewId TextureCache<P>::FindColorBuffer(size_t index) {
     const auto& regs = maxwell3d->regs;

src/video_core/texture_cache/texture_cache_base.h

@@ -420,9 +420,6 @@ private:
 
     void QueueAsyncDecode(Image& image, ImageId image_id);
     void TickAsyncDecode();
-    void EnforceSamplerBudget();
-    void TrimInactiveSamplers(size_t budget);
-    std::optional<size_t> QuerySamplerBudget() const;
 
     void QueueAsyncUnswizzle(Image& image, ImageId image_id);
     void TickAsyncUnswizzle();
@@ -509,7 +506,6 @@ private:
 
     u64 modification_tick = 0;
     u64 frame_tick = 0;
-    u64 last_sampler_gc_frame = (std::numeric_limits<u64>::max)();
 
     Common::ThreadWorker texture_decode_worker{1, "TextureDecoder", {},
                                                Common::ThreadPlacement::Efficiency};

src/video_core/textures/texture.cpp

@@ -5,6 +5,7 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <array>
+#include <cmath>
 
 #include "common/cityhash.h"
 #include "common/settings.h"
@@ -17,53 +18,25 @@ namespace Tegra::Texture {
 
 namespace {
 
-[[maybe_unused]] constexpr std::array<float, 256> SRGB_CONVERSION_LUT = {
+float SrgbToLinear(u32 value) {
-    0.000000f, 0.000000f, 0.000000f, 0.000012f, 0.000021f, 0.000033f, 0.000046f, 0.000062f,
+    const float encoded = static_cast<float>(value) / 255.0f;
-    0.000081f, 0.000102f, 0.000125f, 0.000151f, 0.000181f, 0.000214f, 0.000251f, 0.000293f,
+    if (encoded <= 0.04045f) {
-    0.000338f, 0.000388f, 0.000443f, 0.000503f, 0.000568f, 0.000639f, 0.000715f, 0.000798f,
+        return encoded / 12.92f;
-    0.000887f, 0.000983f, 0.001085f, 0.001195f, 0.001312f, 0.001437f, 0.001569f, 0.001710f,
+    }
-    0.001860f, 0.002019f, 0.002186f, 0.002364f, 0.002551f, 0.002748f, 0.002955f, 0.003174f,
+    return std::pow((encoded + 0.055f) / 1.055f, 2.4f);
-    0.003403f, 0.003643f, 0.003896f, 0.004160f, 0.004436f, 0.004725f, 0.005028f, 0.005343f,
+}
-    0.005672f, 0.006015f, 0.006372f, 0.006744f, 0.007130f, 0.007533f, 0.007950f, 0.008384f,
-    0.008834f, 0.009301f, 0.009785f, 0.010286f, 0.010805f, 0.011342f, 0.011898f, 0.012472f,
-    0.013066f, 0.013680f, 0.014313f, 0.014967f, 0.015641f, 0.016337f, 0.017054f, 0.017793f,
-    0.018554f, 0.019337f, 0.020144f, 0.020974f, 0.021828f, 0.022706f, 0.023609f, 0.024536f,
-    0.025489f, 0.026468f, 0.027473f, 0.028504f, 0.029563f, 0.030649f, 0.031762f, 0.032904f,
-    0.034074f, 0.035274f, 0.036503f, 0.037762f, 0.039050f, 0.040370f, 0.041721f, 0.043103f,
-    0.044518f, 0.045964f, 0.047444f, 0.048956f, 0.050503f, 0.052083f, 0.053699f, 0.055349f,
-    0.057034f, 0.058755f, 0.060513f, 0.062307f, 0.064139f, 0.066008f, 0.067915f, 0.069861f,
-    0.071845f, 0.073869f, 0.075933f, 0.078037f, 0.080182f, 0.082369f, 0.084597f, 0.086867f,
-    0.089180f, 0.091535f, 0.093935f, 0.096378f, 0.098866f, 0.101398f, 0.103977f, 0.106601f,
-    0.109271f, 0.111988f, 0.114753f, 0.117565f, 0.120426f, 0.123335f, 0.126293f, 0.129301f,
-    0.132360f, 0.135469f, 0.138629f, 0.141841f, 0.145105f, 0.148421f, 0.151791f, 0.155214f,
-    0.158691f, 0.162224f, 0.165810f, 0.169453f, 0.173152f, 0.176907f, 0.180720f, 0.184589f,
-    0.188517f, 0.192504f, 0.196549f, 0.200655f, 0.204820f, 0.209046f, 0.213334f, 0.217682f,
-    0.222093f, 0.226567f, 0.231104f, 0.235704f, 0.240369f, 0.245099f, 0.249894f, 0.254754f,
-    0.259681f, 0.264674f, 0.269736f, 0.274864f, 0.280062f, 0.285328f, 0.290664f, 0.296070f,
-    0.301546f, 0.307094f, 0.312713f, 0.318404f, 0.324168f, 0.330006f, 0.335916f, 0.341902f,
-    0.347962f, 0.354097f, 0.360309f, 0.366597f, 0.372961f, 0.379403f, 0.385924f, 0.392524f,
-    0.399202f, 0.405960f, 0.412798f, 0.419718f, 0.426719f, 0.433802f, 0.440967f, 0.448216f,
-    0.455548f, 0.462965f, 0.470465f, 0.478052f, 0.485725f, 0.493484f, 0.501329f, 0.509263f,
-    0.517285f, 0.525396f, 0.533595f, 0.541885f, 0.550265f, 0.558736f, 0.567299f, 0.575954f,
-    0.584702f, 0.593542f, 0.602477f, 0.611507f, 0.620632f, 0.629852f, 0.639168f, 0.648581f,
-    0.658092f, 0.667700f, 0.677408f, 0.687214f, 0.697120f, 0.707127f, 0.717234f, 0.727443f,
-    0.737753f, 0.748167f, 0.758685f, 0.769305f, 0.780031f, 0.790861f, 0.801798f, 0.812839f,
-    0.823989f, 0.835246f, 0.846611f, 0.858085f, 0.869668f, 0.881360f, 0.893164f, 0.905078f,
-    0.917104f, 0.929242f, 0.941493f, 0.953859f, 0.966338f, 1.000000f, 1.000000f, 1.000000f,
-};
 
 } // Anonymous namespace
 
 std::array<float, 4> TSCEntry::BorderColor() const noexcept {
-    // TODO: Handle SRGB correctly. Using this breaks shadows in some games (Xenoblade).
-    // if (!srgb_conversion) {
-    //    return border_color;
-    //}
-    // return {SRGB_CONVERSION_LUT[srgb_border_color_r], SRGB_CONVERSION_LUT[srgb_border_color_g],
-    //        SRGB_CONVERSION_LUT[srgb_border_color_b], border_color[3]};
     return border_color;
 }
 
+std::array<float, 4> TSCEntry::SrgbBorderColor() const noexcept {
+    return {SrgbToLinear(srgb_border_color_r), SrgbToLinear(srgb_border_color_g),
+            SrgbToLinear(srgb_border_color_b), border_color[3]};
+}
+
 float TSCEntry::MaxAnisotropy() const noexcept {
     const bool is_suitable_mipmap_filter = mipmap_filter != TextureMipmapFilter::None;
     const bool has_regular_lods = min_lod_clamp == 0 && max_lod_clamp >= 256;

src/video_core/textures/texture.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2026 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -378,6 +381,8 @@ struct TSCEntry {
 
     std::array<float, 4> BorderColor() const noexcept;
 
+    std::array<float, 4> SrgbBorderColor() const noexcept;
+
     float MaxAnisotropy() const noexcept;
 
     float MinLod() const {

src/video_core/vulkan_common/vulkan_device.cpp

@@ -508,17 +508,9 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
         LOG_WARNING(Render_Vulkan, "Qualcomm drivers require scaled vertex format emulation.");
         has_broken_descriptor_aliasing = true;
         LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken descriptor aliasing.");
-        LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken custom border color.");
-        RemoveExtensionFeature(extensions.custom_border_color, features.custom_border_color,
-                               VK_EXT_CUSTOM_BORDER_COLOR_EXTENSION_NAME);
-        LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken border color swizzle.");
-        RemoveExtensionFeature(extensions.border_color_swizzle, features.border_color_swizzle,
-                               VK_EXT_BORDER_COLOR_SWIZZLE_EXTENSION_NAME);
         LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken color write enable.");
         RemoveExtensionFeature(extensions.color_write_enable, features.color_write_enable,
                                VK_EXT_COLOR_WRITE_ENABLE_EXTENSION_NAME);
-        LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken shader float controls.");
-        RemoveExtension(extensions.shader_float_controls, VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME);
         LOG_WARNING(Render_Vulkan, "Qualcomm drivers have broken shader atomic int64.");
         RemoveExtensionFeature(extensions.shader_atomic_int64, features.shader_atomic_int64,
                                VK_KHR_SHADER_ATOMIC_INT64_EXTENSION_NAME);
@@ -619,21 +611,6 @@ Device::Device(VkInstance instance_, vk::PhysicalDevice physical_, VkSurfaceKHR
         }
     }
 
-    if (is_qualcomm) {
-        const size_t sampler_limit = properties.properties.limits.maxSamplerAllocationCount;
-        if (sampler_limit > 0) {
-            constexpr size_t MIN_SAMPLER_BUDGET = 1024U;
-            const size_t reserved = sampler_limit / 4U;
-            const size_t derived_budget =
-                (std::max)(MIN_SAMPLER_BUDGET, sampler_limit - reserved);
-            sampler_heap_budget = derived_budget;
-            LOG_WARNING(Render_Vulkan,
-                        "Qualcomm driver reports max {} samplers; reserving {} (25%) and "
-                        "allowing Eden to use {} (75%) to avoid heap exhaustion",
-                        sampler_limit, reserved, sampler_heap_budget);
-        }
-    }
-
     if (extensions.sampler_filter_minmax && is_amd) {
         // Disable ext_sampler_filter_minmax on AMD GCN4 and lower as it is broken.
         if (!features.shader_float16_int8.shaderFloat16) {
@@ -1146,6 +1123,11 @@ bool Device::GetSuitability(bool requires_swapchain) {
             VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_5_PROPERTIES_KHR;
         SetNext(next, properties.maintenance5);
     }
+    if (extensions.custom_border_color) {
+        properties.custom_border_color.sType =
+            VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_PROPERTIES_EXT;
+        SetNext(next, properties.custom_border_color);
+    }
 
     // Perform the property fetch.
     physical.GetProperties2(properties2);
@@ -1242,9 +1224,7 @@ void Device::RemoveUnsuitableExtensions() {
     // VK_EXT_border_color_swizzle
     if (extensions.border_color_swizzle) {
         extensions.border_color_swizzle =
-            extensions.custom_border_color &&
+            extensions.custom_border_color && features.border_color_swizzle.borderColorSwizzle;
-            features.border_color_swizzle.borderColorSwizzle &&
-            features.border_color_swizzle.borderColorSwizzleFromImage;
     }
     RemoveExtensionFeatureIfUnsuitable(extensions.border_color_swizzle,
                                        features.border_color_swizzle,
@@ -1406,6 +1386,11 @@ void Device::RemoveUnsuitableExtensions() {
                                VK_KHR_PIPELINE_EXECUTABLE_PROPERTIES_EXTENSION_NAME);
     }
 
+    // VK_KHR_shader_quad_control
+    extensions.shader_quad_control = features.shader_quad_control.shaderQuadControl;
+    RemoveExtensionFeatureIfUnsuitable(extensions.shader_quad_control, features.shader_quad_control,
+                                       VK_KHR_SHADER_QUAD_CONTROL_EXTENSION_NAME);
+
     // VK_KHR_workgroup_memory_explicit_layout
     extensions.workgroup_memory_explicit_layout =
         features.workgroup_memory_explicit_layout.workgroupMemoryExplicitLayout &&
@@ -1490,11 +1475,26 @@ void Device::SetupFamilies(VkSurfaceKHR surface) {
     }
 }
 
-std::optional<size_t> Device::GetSamplerHeapBudget() const {
+bool Device::TryReserveCustomBorderColorSamplers(size_t count) const {
-    if (sampler_heap_budget == 0) {
+    const size_t limit = properties.custom_border_color.maxCustomBorderColorSamplers;
-        return std::nullopt;
+    if (limit == 0) {
+        return true;
     }
-    return sampler_heap_budget;
+    size_t used = custom_border_color_samplers_used.load(std::memory_order_relaxed);
+    while (used + count <= limit) {
+        if (custom_border_color_samplers_used.compare_exchange_weak(
+                used, used + count, std::memory_order_relaxed, std::memory_order_relaxed)) {
+            return true;
+        }
+    }
+    return false;
+}
+
+void Device::ReleaseCustomBorderColorSamplers(size_t count) const {
+    if (count == 0) {
+        return;
+    }
+    custom_border_color_samplers_used.fetch_sub(count, std::memory_order_relaxed);
 }
 
 u64 Device::GetDeviceMemoryUsage() const {

src/video_core/vulkan_common/vulkan_device.h

@@ -6,6 +6,7 @@
 
 #pragma once
 
+#include <atomic>
 #include <optional>
 #include <set>
 #include <span>
@@ -74,6 +75,7 @@ VK_DEFINE_HANDLE(VmaAllocator)
     FEATURE(KHR, Maintenance6, MAINTENANCE_6, maintenance6)                                        \
     FEATURE(KHR, PipelineExecutableProperties, PIPELINE_EXECUTABLE_PROPERTIES,                     \
             pipeline_executable_properties)                                                        \
+    FEATURE(KHR, ShaderQuadControl, SHADER_QUAD_CONTROL, shader_quad_control)                      \
     FEATURE(KHR, WorkgroupMemoryExplicitLayout, WORKGROUP_MEMORY_EXPLICIT_LAYOUT,                  \
             workgroup_memory_explicit_layout)
 
@@ -585,6 +587,11 @@ FN_MAX_LIMIT_LIST
                features.features.shaderInt16;
     }
 
+    /// Returns true if the device supports VK_KHR_shader_quad_control.
+    bool IsKhrShaderQuadControlSupported() const {
+        return extensions.shader_quad_control && features.shader_quad_control.shaderQuadControl;
+    }
+
     /// Returns true if the device supports VK_KHR_image_format_list.
     bool IsKhrImageFormatListSupported() const {
         return extensions.image_format_list || instance_version >= VK_API_VERSION_1_2;
@@ -699,20 +706,18 @@ FN_MAX_LIMIT_LIST
         return features.transform_feedback.geometryStreams;
     }
 
-    /// Returns true if the device supports VK_EXT_custom_border_color.
+    /// Returns true if custom border colors can be created without a format.
-    bool IsExtCustomBorderColorSupported() const {
+    bool IsCustomBorderColorUsable() const {
-        return extensions.custom_border_color;
+        return extensions.custom_border_color &&
+               features.custom_border_color.customBorderColors &&
+               features.custom_border_color.customBorderColorWithoutFormat;
     }
 
-    /// Returns true if customBorderColors feature is available.
+    /// Takes budget for samplers carrying a custom border color, false when exhausted.
-    bool IsCustomBorderColorsSupported() const {
+    bool TryReserveCustomBorderColorSamplers(size_t count) const;
-        return features.custom_border_color.customBorderColors;
-    }
 
-    /// Returns true if customBorderColorWithoutFormat feature is available.
+    /// Gives back budget taken by TryReserveCustomBorderColorSamplers.
-    bool IsCustomBorderColorWithoutFormatSupported() const {
+    void ReleaseCustomBorderColorSamplers(size_t count) const;
-        return features.custom_border_color.customBorderColorWithoutFormat;
-    }
 
     /// Returns true if the device supports VK_EXT_color_write_enable.
     bool IsExtColorWriteEnableSupported() const {
@@ -724,6 +729,12 @@ FN_MAX_LIMIT_LIST
         return extensions.border_color_swizzle;
     }
 
+    /// Returns true if samplers must be carried with border color swizzle mapping.
+    bool NeedsBorderColorSwizzleMapping() const {
+        return extensions.border_color_swizzle &&
+               !features.border_color_swizzle.borderColorSwizzleFromImage;
+    }
+
     /// Returns true if borderColorSwizzleFromImage is available.
     bool IsBorderColorSwizzleFromImageSupported() const {
         return features.border_color_swizzle.borderColorSwizzleFromImage;
@@ -919,8 +930,6 @@ FN_MAX_LIMIT_LIST
         return has_broken_parallel_compiling;
     }
 
-    std::optional<size_t> GetSamplerHeapBudget() const;
-
     /// Returns the vendor name reported from Vulkan.
     std::string_view GetVendorName() const {
         return properties.driver.driverName;
@@ -1173,6 +1182,7 @@ private:
         VkPhysicalDeviceTransformFeedbackPropertiesEXT transform_feedback{};
         VkPhysicalDeviceMaintenance5PropertiesKHR maintenance5{};
         VkPhysicalDeviceDepthStencilResolveProperties depth_stencil_resolve{};
+        VkPhysicalDeviceCustomBorderColorPropertiesEXT custom_border_color{};
 
         VkPhysicalDeviceProperties properties{};
     };
@@ -1211,7 +1221,7 @@ private:
     bool dynamic_state3_alpha_to_coverage{};
     bool dynamic_state3_alpha_to_one{};
     bool supports_conditional_barriers{};      ///< Allows barriers in conditional control flow.
-    size_t sampler_heap_budget{};              ///< Sampler budget for buggy drivers (0 = unlimited).
+    mutable std::atomic<size_t> custom_border_color_samplers_used{};
     u64 device_access_memory{};                ///< Total size of device local memory in bytes.
     u32 sets_per_pool{};                       ///< Sets per Description Pool
     NvidiaArchitecture nvidia_arch{NvidiaArchitecture::Arch_AmpereOrNewer};