Custom textures rewrite (#6452)

* common: Add thread pool from yuzu

* Is really useful for asynchronous operations like shader compilation and custom textures, will be used in following PRs

* core: Improve ImageInterface

* Provide a default implementation so frontends don't have to duplicate code registering the lodepng version

* Add a dds version too which we will use in the next commit

* rasterizer_cache: Rewrite custom textures

* There's just too much to talk about here, look at the PR description for more details

* rasterizer_cache: Implement basic pack configuration file

* custom_tex_manager: Flip dumped textures

* custom_tex_manager: Optimize custom texture hashing

* If no convertions are needed then we can hash the decoded data directly removing the needed for duplicate decode

* custom_tex_manager: Implement asynchronous texture loading

* The file loading and decoding is offloaded into worker threads, while the upload itself still occurs in the main thread to avoid having to manage shared contexts

* Address review comments

* custom_tex_manager: Introduce custom material support

* video_core: Move custom textures to separate directory

* Also split the files to make the code cleaner

* gl_texture_runtime: Generate mipmaps for material

* custom_tex_manager: Prevent memory overflow when preloading

* externals: Add dds-ktx as submodule

* string_util: Return vector from SplitString

* No code benefits from passing it as an argument

* custom_textures: Use json config file

* gl_rasterizer: Only bind material for unit 0

* Address review comments

src/common/CMakeLists.txt

@@ -65,6 +65,8 @@ add_library(common STATIC
     common_precompiled_headers.h
     common_types.h
     construct.h
+    error.cpp
+    error.h
     file_util.cpp
     file_util.h
     hash.h
@@ -89,6 +91,7 @@ add_library(common STATIC
     misc.cpp
     param_package.cpp
     param_package.h
+    polyfill_thread.h
     precompiled_headers.h
     quaternion.h
     ring_buffer.h
@@ -113,9 +116,11 @@ add_library(common STATIC
     thread.cpp
     thread.h
     thread_queue_list.h
+    thread_worker.h
     threadsafe_queue.h
     timer.cpp
     timer.h
+    unique_function.h
     vector_math.h
     web_result.h
     x64/cpu_detect.cpp

src/common/error.cpp

@@ -0,0 +1,57 @@
+// Copyright 2013 Dolphin Emulator Project / 2014 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstddef>
+#ifdef _WIN32
+#include <windows.h>
+#else
+#include <cerrno>
+#include <cstring>
+#endif
+
+#include "common/error.h"
+
+namespace Common {
+
+std::string NativeErrorToString(int e) {
+#ifdef _WIN32
+    LPSTR err_str;
+
+    DWORD res = FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_ALLOCATE_BUFFER |
+                                   FORMAT_MESSAGE_IGNORE_INSERTS,
+                               nullptr, e, MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),
+                               reinterpret_cast<LPSTR>(&err_str), 1, nullptr);
+    if (!res) {
+        return "(FormatMessageA failed to format error)";
+    }
+    std::string ret(err_str);
+    LocalFree(err_str);
+    return ret;
+#else
+    char err_str[255];
+#if defined(__GLIBC__) && (_GNU_SOURCE || (_POSIX_C_SOURCE < 200112L && _XOPEN_SOURCE < 600)) ||   \
+    defined(ANDROID)
+    // Thread safe (GNU-specific)
+    const char* str = strerror_r(e, err_str, sizeof(err_str));
+    return std::string(str);
+#else
+    // Thread safe (XSI-compliant)
+    int second_err = strerror_r(e, err_str, sizeof(err_str));
+    if (second_err != 0) {
+        return "(strerror_r failed to format error)";
+    }
+    return std::string(err_str);
+#endif // GLIBC etc.
+#endif // _WIN32
+}
+
+std::string GetLastErrorMsg() {
+#ifdef _WIN32
+    return NativeErrorToString(GetLastError());
+#else
+    return NativeErrorToString(errno);
+#endif
+}
+
+} // namespace Common

src/common/error.h

@@ -0,0 +1,21 @@
+// Copyright 2013 Dolphin Emulator Project / 2014 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <string>
+
+namespace Common {
+
+// Generic function to get last error message.
+// Call directly after the command or use the error num.
+// This function might change the error code.
+// Defined in error.cpp.
+[[nodiscard]] std::string GetLastErrorMsg();
+
+// Like GetLastErrorMsg(), but passing an explicit error code.
+// Defined in error.cpp.
+[[nodiscard]] std::string NativeErrorToString(int e);
+
+} // namespace Common

src/common/file_util.cpp

@@ -960,7 +960,7 @@ std::string_view GetFilename(std::string_view path) {
     const auto name_index = path.find_last_of("\\/");
 
     if (name_index == std::string_view::npos) {
-        return {};
+        return path;
     }
 
     return path.substr(name_index + 1);

src/common/param_package.cpp

@@ -28,12 +28,9 @@ ParamPackage::ParamPackage(const std::string& serialized) {
         return;
     }
 
-    std::vector<std::string> pairs;
-    Common::SplitString(serialized, PARAM_SEPARATOR, pairs);
-
+    const auto pairs = Common::SplitString(serialized, PARAM_SEPARATOR);
     for (const std::string& pair : pairs) {
-        std::vector<std::string> key_value;
-        Common::SplitString(pair, KEY_VALUE_SEPARATOR, key_value);
+        auto key_value = Common::SplitString(pair, KEY_VALUE_SEPARATOR);
         if (key_value.size() != 2) {
             LOG_ERROR(Common, "invalid key pair {}", pair);
             continue;

src/common/polyfill_thread.h

@@ -0,0 +1,338 @@
+// Copyright 2022 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+//
+// TODO: remove this file when jthread is supported by all compilation targets
+//
+
+#pragma once
+
+#include <version>
+
+#ifdef __cpp_lib_jthread
+
+#include <stop_token>
+#include <thread>
+
+namespace Common {
+
+template <typename Condvar, typename Lock, typename Pred>
+void CondvarWait(Condvar& cv, Lock& lock, std::stop_token token, Pred&& pred) {
+    cv.wait(lock, token, std::move(pred));
+}
+
+} // namespace Common
+
+#else
+
+#include <atomic>
+#include <functional>
+#include <map>
+#include <memory>
+#include <mutex>
+#include <optional>
+#include <thread>
+#include <type_traits>
+#include <utility>
+
+namespace std {
+namespace polyfill {
+
+using stop_state_callback = size_t;
+
+class stop_state {
+public:
+    stop_state() = default;
+    ~stop_state() = default;
+
+    bool request_stop() {
+        unique_lock lk{m_lock};
+
+        if (m_stop_requested) {
+            // Already set, nothing to do.
+            return false;
+        }
+
+        // Mark stop requested.
+        m_stop_requested = true;
+
+        while (!m_callbacks.empty()) {
+            // Get an iterator to the first element.
+            const auto it = m_callbacks.begin();
+
+            // Move the callback function out of the map.
+            function<void()> f;
+            swap(it->second, f);
+
+            // Erase the now-empty map element.
+            m_callbacks.erase(it);
+
+            // Run the callback.
+            if (f) {
+                f();
+            }
+        }
+
+        return true;
+    }
+
+    bool stop_requested() const {
+        unique_lock lk{m_lock};
+        return m_stop_requested;
+    }
+
+    stop_state_callback insert_callback(function<void()> f) {
+        unique_lock lk{m_lock};
+
+        if (m_stop_requested) {
+            // Stop already requested. Don't insert anything,
+            // just run the callback synchronously.
+            if (f) {
+                f();
+            }
+            return 0;
+        }
+
+        // Insert the callback.
+        stop_state_callback ret = ++m_next_callback;
+        m_callbacks.emplace(ret, move(f));
+        return ret;
+    }
+
+    void remove_callback(stop_state_callback cb) {
+        unique_lock lk{m_lock};
+        m_callbacks.erase(cb);
+    }
+
+private:
+    mutable recursive_mutex m_lock;
+    map<stop_state_callback, function<void()>> m_callbacks;
+    stop_state_callback m_next_callback{0};
+    bool m_stop_requested{false};
+};
+
+} // namespace polyfill
+
+#ifndef __cpp_lib_concepts
+template <class T, class... Args>
+concept constructible_from = is_nothrow_destructible_v<T> && is_constructible_v<T, Args...>;
+#endif
+
+class stop_token;
+class stop_source;
+struct nostopstate_t {
+    explicit nostopstate_t() = default;
+};
+inline constexpr nostopstate_t nostopstate{};
+
+template <class Callback>
+class stop_callback;
+
+class stop_token {
+public:
+    stop_token() noexcept = default;
+
+    stop_token(const stop_token&) noexcept = default;
+    stop_token(stop_token&&) noexcept = default;
+    stop_token& operator=(const stop_token&) noexcept = default;
+    stop_token& operator=(stop_token&&) noexcept = default;
+    ~stop_token() = default;
+
+    void swap(stop_token& other) noexcept {
+        m_stop_state.swap(other.m_stop_state);
+    }
+
+    [[nodiscard]] bool stop_requested() const noexcept {
+        return m_stop_state && m_stop_state->stop_requested();
+    }
+    [[nodiscard]] bool stop_possible() const noexcept {
+        return m_stop_state != nullptr;
+    }
+
+private:
+    friend class stop_source;
+    template <typename Callback>
+    friend class stop_callback;
+    stop_token(shared_ptr<polyfill::stop_state> stop_state) : m_stop_state(move(stop_state)) {}
+
+private:
+    shared_ptr<polyfill::stop_state> m_stop_state;
+};
+
+class stop_source {
+public:
+    stop_source() : m_stop_state(make_shared<polyfill::stop_state>()) {}
+    explicit stop_source(nostopstate_t) noexcept {}
+
+    stop_source(const stop_source&) noexcept = default;
+    stop_source(stop_source&&) noexcept = default;
+    stop_source& operator=(const stop_source&) noexcept = default;
+    stop_source& operator=(stop_source&&) noexcept = default;
+    ~stop_source() = default;
+    void swap(stop_source& other) noexcept {
+        m_stop_state.swap(other.m_stop_state);
+    }
+
+    [[nodiscard]] stop_token get_token() const noexcept {
+        return stop_token(m_stop_state);
+    }
+    [[nodiscard]] bool stop_possible() const noexcept {
+        return m_stop_state != nullptr;
+    }
+    [[nodiscard]] bool stop_requested() const noexcept {
+        return m_stop_state && m_stop_state->stop_requested();
+    }
+    bool request_stop() noexcept {
+        return m_stop_state && m_stop_state->request_stop();
+    }
+
+private:
+    friend class jthread;
+    explicit stop_source(shared_ptr<polyfill::stop_state> stop_state)
+        : m_stop_state(move(stop_state)) {}
+
+private:
+    shared_ptr<polyfill::stop_state> m_stop_state;
+};
+
+template <typename Callback>
+class stop_callback {
+    static_assert(is_nothrow_destructible_v<Callback>);
+    static_assert(is_invocable_v<Callback>);
+
+public:
+    using callback_type = Callback;
+
+    template <typename C>
+    requires constructible_from<Callback, C>
+    explicit stop_callback(const stop_token& st,
+                           C&& cb) noexcept(is_nothrow_constructible_v<Callback, C>)
+        : m_stop_state(st.m_stop_state) {
+        if (m_stop_state) {
+            m_callback = m_stop_state->insert_callback(move(cb));
+        }
+    }
+    template <typename C>
+    requires constructible_from<Callback, C>
+    explicit stop_callback(stop_token&& st,
+                           C&& cb) noexcept(is_nothrow_constructible_v<Callback, C>)
+        : m_stop_state(move(st.m_stop_state)) {
+        if (m_stop_state) {
+            m_callback = m_stop_state->insert_callback(move(cb));
+        }
+    }
+    ~stop_callback() {
+        if (m_stop_state && m_callback) {
+            m_stop_state->remove_callback(m_callback);
+        }
+    }
+
+    stop_callback(const stop_callback&) = delete;
+    stop_callback(stop_callback&&) = delete;
+    stop_callback& operator=(const stop_callback&) = delete;
+    stop_callback& operator=(stop_callback&&) = delete;
+
+private:
+    shared_ptr<polyfill::stop_state> m_stop_state;
+    polyfill::stop_state_callback m_callback;
+};
+
+template <typename Callback>
+stop_callback(stop_token, Callback) -> stop_callback<Callback>;
+
+class jthread {
+public:
+    using id = thread::id;
+    using native_handle_type = thread::native_handle_type;
+
+    jthread() noexcept = default;
+
+    template <typename F, typename... Args,
+              typename = enable_if_t<!is_same_v<remove_cvref_t<F>, jthread>>>
+    explicit jthread(F&& f, Args&&... args)
+        : m_stop_state(make_shared<polyfill::stop_state>()),
+          m_thread(make_thread(move(f), move(args)...)) {}
+
+    ~jthread() {
+        if (joinable()) {
+            request_stop();
+            join();
+        }
+    }
+
+    jthread(const jthread&) = delete;
+    jthread(jthread&&) noexcept = default;
+    jthread& operator=(const jthread&) = delete;
+
+    jthread& operator=(jthread&& other) noexcept {
+        m_thread.swap(other.m_thread);
+        m_stop_state.swap(other.m_stop_state);
+        return *this;
+    }
+
+    void swap(jthread& other) noexcept {
+        m_thread.swap(other.m_thread);
+        m_stop_state.swap(other.m_stop_state);
+    }
+    [[nodiscard]] bool joinable() const noexcept {
+        return m_thread.joinable();
+    }
+    void join() {
+        m_thread.join();
+    }
+    void detach() {
+        m_thread.detach();
+        m_stop_state.reset();
+    }
+
+    [[nodiscard]] id get_id() const noexcept {
+        return m_thread.get_id();
+    }
+    [[nodiscard]] native_handle_type native_handle() {
+        return m_thread.native_handle();
+    }
+    [[nodiscard]] stop_source get_stop_source() noexcept {
+        return stop_source(m_stop_state);
+    }
+    [[nodiscard]] stop_token get_stop_token() const noexcept {
+        return stop_source(m_stop_state).get_token();
+    }
+    bool request_stop() noexcept {
+        return get_stop_source().request_stop();
+    }
+    [[nodiscard]] static unsigned int hardware_concurrency() noexcept {
+        return thread::hardware_concurrency();
+    }
+
+private:
+    template <typename F, typename... Args>
+    thread make_thread(F&& f, Args&&... args) {
+        if constexpr (is_invocable_v<decay_t<F>, stop_token, decay_t<Args>...>) {
+            return thread(move(f), get_stop_token(), move(args)...);
+        } else {
+            return thread(move(f), move(args)...);
+        }
+    }
+
+    shared_ptr<polyfill::stop_state> m_stop_state;
+    thread m_thread;
+};
+
+} // namespace std
+
+namespace Common {
+
+template <typename Condvar, typename Lock, typename Pred>
+void CondvarWait(Condvar& cv, Lock& lock, std::stop_token token, Pred pred) {
+    if (token.stop_requested()) {
+        return;
+    }
+
+    std::stop_callback callback(token, [&] { cv.notify_all(); });
+    cv.wait(lock, [&] { return pred() || token.stop_requested(); });
+}
+
+} // namespace Common
+
+#endif // __cpp_lib_jthread

src/common/settings.h

@@ -477,6 +477,7 @@ struct Values {
     SwitchableSetting<bool> dump_textures{false, "dump_textures"};
     SwitchableSetting<bool> custom_textures{false, "custom_textures"};
     SwitchableSetting<bool> preload_textures{false, "preload_textures"};
+    SwitchableSetting<bool> async_custom_loading{true, "async_custom_loading"};
 
     // Audio
     bool audio_muted;

src/common/string_util.cpp

@@ -102,15 +102,16 @@ void BuildCompleteFilename(std::string& _CompleteFilename, const std::string& _P
     _CompleteFilename += _Filename;
 }
 
-void SplitString(const std::string& str, const char delim, std::vector<std::string>& output) {
+std::vector<std::string> SplitString(const std::string& str, const char delim) {
     std::istringstream iss(str);
-    output.resize(1);
+    std::vector<std::string> output(1);
 
     while (std::getline(iss, *output.rbegin(), delim)) {
         output.emplace_back();
     }
 
     output.pop_back();
+    return output;
 }
 
 std::string TabsToSpaces(int tab_size, std::string in) {

src/common/string_util.h

@@ -29,7 +29,7 @@ namespace Common {
 
 [[nodiscard]] bool EndsWith(const std::string& value, const std::string& ending);
 
-void SplitString(const std::string& str, char delim, std::vector<std::string>& output);
+[[nodiscard]] std::vector<std::string> SplitString(const std::string& str, const char delim);
 
 // "C:/Windows/winhelp.exe" to "C:/Windows/", "winhelp", ".exe"
 bool SplitPath(const std::string& full_path, std::string* _pPath, std::string* _pFilename,

src/common/texture.cpp

@@ -3,12 +3,12 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
-#include <vector>
 #include "common/assert.h"
-#include "common/common_types.h"
+#include "common/texture.h"
 
 namespace Common {
-void FlipRGBA8Texture(std::vector<u8>& tex, u32 width, u32 height) {
+
+void FlipRGBA8Texture(std::span<u8> tex, u32 width, u32 height) {
     ASSERT(tex.size() == width * height * 4);
     const u32 line_size = width * 4;
     for (u32 line = 0; line < height / 2; line++) {
@@ -19,4 +19,5 @@ void FlipRGBA8Texture(std::vector<u8>& tex, u32 width, u32 height) {
                          tex.begin() + offset_2);
     }
 }
+
 } // namespace Common

src/common/texture.h

@@ -4,9 +4,11 @@
 
 #pragma once
 
-#include <vector>
+#include <span>
 #include "common/common_types.h"
 
 namespace Common {
-void FlipRGBA8Texture(std::vector<u8>& tex, u32 width, u32 height);
-}
+
+void FlipRGBA8Texture(std::span<u8> tex, u32 width, u32 height);
+
+} // namespace Common

src/common/thread.cpp

@@ -2,7 +2,9 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include "common/common_funcs.h"
+#include <string>
+
+#include "common/error.h"
 #include "common/logging/log.h"
 #include "common/thread.h"
 #ifdef __APPLE__
@@ -20,7 +22,6 @@
 #ifndef _WIN32
 #include <unistd.h>
 #endif
-#include <string>
 
 #ifdef __FreeBSD__
 #define cpu_set_t cpuset_t
@@ -28,6 +29,56 @@
 
 namespace Common {
 
+#ifdef _WIN32
+
+void SetCurrentThreadPriority(ThreadPriority new_priority) {
+    auto handle = GetCurrentThread();
+    int windows_priority = 0;
+    switch (new_priority) {
+    case ThreadPriority::Low:
+        windows_priority = THREAD_PRIORITY_BELOW_NORMAL;
+        break;
+    case ThreadPriority::Normal:
+        windows_priority = THREAD_PRIORITY_NORMAL;
+        break;
+    case ThreadPriority::High:
+        windows_priority = THREAD_PRIORITY_ABOVE_NORMAL;
+        break;
+    case ThreadPriority::VeryHigh:
+        windows_priority = THREAD_PRIORITY_HIGHEST;
+        break;
+    case ThreadPriority::Critical:
+        windows_priority = THREAD_PRIORITY_TIME_CRITICAL;
+        break;
+    default:
+        windows_priority = THREAD_PRIORITY_NORMAL;
+        break;
+    }
+    SetThreadPriority(handle, windows_priority);
+}
+
+#else
+
+void SetCurrentThreadPriority(ThreadPriority new_priority) {
+    pthread_t this_thread = pthread_self();
+
+    const auto scheduling_type = SCHED_OTHER;
+    s32 max_prio = sched_get_priority_max(scheduling_type);
+    s32 min_prio = sched_get_priority_min(scheduling_type);
+    u32 level = std::max(static_cast<u32>(new_priority) + 1, 4U);
+
+    struct sched_param params;
+    if (max_prio > min_prio) {
+        params.sched_priority = min_prio + ((max_prio - min_prio) * level) / 4;
+    } else {
+        params.sched_priority = min_prio - ((min_prio - max_prio) * level) / 4;
+    }
+
+    pthread_setschedparam(this_thread, scheduling_type, &params);
+}
+
+#endif
+
 #ifdef _MSC_VER
 
 // Sets the debugger-visible name of the current thread.
@@ -47,7 +98,7 @@ void SetCurrentThreadName(const char* name) {
 
     info.dwType = 0x1000;
     info.szName = name;
-    info.dwThreadID = static_cast<DWORD>(-1);
+    info.dwThreadID = std::numeric_limits<DWORD>::max();
     info.dwFlags = 0;
 
     __try {
@@ -81,6 +132,12 @@ void SetCurrentThreadName(const char* name) {
 }
 #endif
 
+#if defined(_WIN32)
+void SetCurrentThreadName(const char*) {
+    // Do Nothing on MingW
+}
+#endif
+
 #endif
 
 } // namespace Common

src/common/thread.h

@@ -10,13 +10,15 @@
 #include <cstddef>
 #include <mutex>
 #include <thread>
+#include "common/common_types.h"
+#include "common/polyfill_thread.h"
 
 namespace Common {
 
 class Event {
 public:
     void Set() {
-        std::lock_guard lk{mutex};
+        std::scoped_lock lk{mutex};
         if (!is_set) {
             is_set = true;
             condvar.notify_one();
@@ -54,6 +56,10 @@ public:
         is_set = false;
     }
 
+    [[nodiscard]] bool IsSet() {
+        return is_set;
+    }
+
 private:
     std::condition_variable condvar;
     std::mutex mutex;
@@ -65,7 +71,7 @@ public:
     explicit Barrier(std::size_t count_) : count(count_) {}
 
     /// Blocks until all "count" threads have called Sync()
-    void Sync() {
+    bool Sync(std::stop_token token = {}) {
         std::unique_lock lk{mutex};
         const std::size_t current_generation = generation;
 
@@ -73,25 +79,37 @@ public:
             generation++;
             waiting = 0;
             condvar.notify_all();
+            return true;
         } else {
-            condvar.wait(lk,
-                         [this, current_generation] { return current_generation != generation; });
+            CondvarWait(condvar, lk, token,
+                        [this, current_generation] { return current_generation != generation; });
+            return !token.stop_requested();
         }
     }
 
-    std::size_t Generation() const {
-        std::unique_lock lk(mutex);
+    std::size_t Generation() {
+        std::unique_lock lk{mutex};
         return generation;
     }
 
 private:
-    std::condition_variable condvar;
-    mutable std::mutex mutex;
+    std::condition_variable_any condvar;
+    std::mutex mutex;
     std::size_t count;
     std::size_t waiting = 0;
     std::size_t generation = 0; // Incremented once each time the barrier is used
 };
 
+enum class ThreadPriority : u32 {
+    Low = 0,
+    Normal = 1,
+    High = 2,
+    VeryHigh = 3,
+    Critical = 4,
+};
+
+void SetCurrentThreadPriority(ThreadPriority new_priority);
+
 void SetCurrentThreadName(const char* name);
 
 } // namespace Common

src/common/thread_worker.h

@@ -0,0 +1,123 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <atomic>
+#include <condition_variable>
+#include <functional>
+#include <mutex>
+#include <string>
+#include <thread>
+#include <type_traits>
+#include <vector>
+#include <queue>
+
+#include "common/polyfill_thread.h"
+#include "common/thread.h"
+#include "common/unique_function.h"
+
+namespace Common {
+
+template <class StateType = void>
+class StatefulThreadWorker {
+    static constexpr bool with_state = !std::is_same_v<StateType, void>;
+
+    struct DummyCallable {
+        int operator()(size_t) const noexcept {
+            return 0;
+        }
+    };
+
+    using Task =
+        std::conditional_t<with_state, UniqueFunction<void, StateType*>, UniqueFunction<void>>;
+    using StateMaker =
+        std::conditional_t<with_state, std::function<StateType(size_t)>, DummyCallable>;
+
+public:
+    explicit StatefulThreadWorker(size_t num_workers, std::string_view name, StateMaker func = {})
+        : workers_queued{num_workers}, thread_name{name} {
+        const auto lambda = [this, func](std::stop_token stop_token, size_t index) {
+            Common::SetCurrentThreadName(thread_name.data());
+            {
+                [[maybe_unused]] std::conditional_t<with_state, StateType, int> state{func(index)};
+                while (!stop_token.stop_requested()) {
+                    Task task;
+                    {
+                        std::unique_lock lock{queue_mutex};
+                        if (requests.empty()) {
+                            wait_condition.notify_all();
+                        }
+                        Common::CondvarWait(condition, lock, stop_token,
+                                            [this] { return !requests.empty(); });
+                        if (stop_token.stop_requested()) {
+                            break;
+                        }
+                        task = std::move(requests.front());
+                        requests.pop();
+                    }
+                    if constexpr (with_state) {
+                        task(&state);
+                    } else {
+                        task();
+                    }
+                    ++work_done;
+                }
+            }
+            ++workers_stopped;
+            wait_condition.notify_all();
+        };
+        threads.reserve(num_workers);
+        for (size_t i = 0; i < num_workers; ++i) {
+            threads.emplace_back(lambda, i);
+        }
+    }
+
+    StatefulThreadWorker& operator=(const StatefulThreadWorker&) = delete;
+    StatefulThreadWorker(const StatefulThreadWorker&) = delete;
+
+    StatefulThreadWorker& operator=(StatefulThreadWorker&&) = delete;
+    StatefulThreadWorker(StatefulThreadWorker&&) = delete;
+
+    void QueueWork(Task work) {
+        {
+            std::unique_lock lock{queue_mutex};
+            requests.emplace(std::move(work));
+            ++work_scheduled;
+        }
+        condition.notify_one();
+    }
+
+    void WaitForRequests(std::stop_token stop_token = {}) {
+        std::stop_callback callback(stop_token, [this] {
+            for (auto& thread : threads) {
+                thread.request_stop();
+            }
+        });
+        std::unique_lock lock{queue_mutex};
+        wait_condition.wait(lock, [this] {
+            return workers_stopped >= workers_queued || work_done >= work_scheduled;
+        });
+    }
+
+    const std::size_t NumWorkers() const noexcept {
+        return threads.size();
+    }
+
+private:
+    std::queue<Task> requests;
+    std::mutex queue_mutex;
+    std::condition_variable_any condition;
+    std::condition_variable wait_condition;
+    std::atomic<size_t> work_scheduled{};
+    std::atomic<size_t> work_done{};
+    std::atomic<size_t> workers_stopped{};
+    std::atomic<size_t> workers_queued{};
+    std::string_view thread_name;
+    std::vector<std::jthread> threads;
+};
+
+using ThreadWorker = StatefulThreadWorker<>;
+
+} // namespace Common

src/common/unique_function.h

@@ -0,0 +1,62 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <utility>
+
+namespace Common {
+
+/// General purpose function wrapper similar to std::function.
+/// Unlike std::function, the captured values don't have to be copyable.
+/// This class can be moved but not copied.
+template <typename ResultType, typename... Args>
+class UniqueFunction {
+    class CallableBase {
+    public:
+        virtual ~CallableBase() = default;
+        virtual ResultType operator()(Args&&...) = 0;
+    };
+
+    template <typename Functor>
+    class Callable final : public CallableBase {
+    public:
+        Callable(Functor&& functor_) : functor{std::move(functor_)} {}
+        ~Callable() override = default;
+
+        ResultType operator()(Args&&... args) override {
+            return functor(std::forward<Args>(args)...);
+        }
+
+    private:
+        Functor functor;
+    };
+
+public:
+    UniqueFunction() = default;
+
+    template <typename Functor>
+    UniqueFunction(Functor&& functor)
+        : callable{std::make_unique<Callable<Functor>>(std::move(functor))} {}
+
+    UniqueFunction& operator=(UniqueFunction&& rhs) noexcept = default;
+    UniqueFunction(UniqueFunction&& rhs) noexcept = default;
+
+    UniqueFunction& operator=(const UniqueFunction&) = delete;
+    UniqueFunction(const UniqueFunction&) = delete;
+
+    ResultType operator()(Args&&... args) const {
+        return (*callable)(std::forward<Args>(args)...);
+    }
+
+    explicit operator bool() const noexcept {
+        return static_cast<bool>(callable);
+    }
+
+private:
+    std::unique_ptr<CallableBase> callable;
+};
+
+} // namespace Common