12 files changed, 132 insertions, 133 deletions

diff --git a/src/common/assert.cpp b/src/common/assert.cpp
index b44570528..6026b7dc2 100644
--- a/src/common/assert.cpp
+++ b/src/common/assert.cpp
@@ -6,8 +6,13 @@
 
 #include "common/settings.h"
 
-void assert_handle_failure() {
+void assert_fail_impl() {
     if (Settings::values.use_debug_asserts) {
         Crash();
     }
 }
+
+[[noreturn]] void unreachable_impl() {
+    Crash();
+    throw std::runtime_error("Unreachable code");
+}
diff --git a/src/common/assert.h b/src/common/assert.h
index dbfd8abaf..8c927fcc0 100644
--- a/src/common/assert.h
+++ b/src/common/assert.h
@@ -9,44 +9,43 @@
 // Sometimes we want to try to continue even after hitting an assert.
 // However touching this file yields a global recompilation as this header is included almost
 // everywhere. So let's just move the handling of the failed assert to a single cpp file.
-void assert_handle_failure();
 
-// For asserts we'd like to keep all the junk executed when an assert happens away from the
-// important code in the function. One way of doing this is to put all the relevant code inside a
-// lambda and force the compiler to not inline it. Unfortunately, MSVC seems to have no syntax to
-// specify __declspec on lambda functions, so what we do instead is define a noinline wrapper
-// template that calls the lambda. This seems to generate an extra instruction at the call-site
-// compared to the ideal implementation (which wouldn't support ASSERT_MSG parameters), but is good
-// enough for our purposes.
-template <typename Fn>
-#if defined(_MSC_VER)
-[[msvc::noinline]]
-#elif defined(__GNUC__)
-[[gnu::cold, gnu::noinline]]
+void assert_fail_impl();
+[[noreturn]] void unreachable_impl();
+
+#ifdef _MSC_VER
+#define YUZU_NO_INLINE __declspec(noinline)
+#else
+#define YUZU_NO_INLINE __attribute__((noinline))
 #endif
-static void
-assert_noinline_call(const Fn& fn) {
-    fn();
-    assert_handle_failure();
-}
 
 #define ASSERT(_a_)                                                                                \
-    do                                                                                             \
-        if (!(_a_)) {                                                                              \
-            assert_noinline_call([] { LOG_CRITICAL(Debug, "Assertion Failed!"); });                \
+    ([&]() YUZU_NO_INLINE {                                                                        \
+        if (!(_a_)) [[unlikely]] {                                                                 \
+            LOG_CRITICAL(Debug, "Assertion Failed!");                                              \
+            assert_fail_impl();                                                                    \
         }                                                                                          \
-    while (0)
+    }())
 
 #define ASSERT_MSG(_a_, ...)                                                                       \
-    do                                                                                             \
-        if (!(_a_)) {                                                                              \
-            assert_noinline_call([&] { LOG_CRITICAL(Debug, "Assertion Failed!\n" __VA_ARGS__); }); \
+    ([&]() YUZU_NO_INLINE {                                                                        \
+        if (!(_a_)) [[unlikely]] {                                                                 \
+            LOG_CRITICAL(Debug, "Assertion Failed!\n" __VA_ARGS__);                                \
+            assert_fail_impl();                                                                    \
         }                                                                                          \
-    while (0)
+    }())
+
+#define UNREACHABLE()                                                                              \
+    do {                                                                                           \
+        LOG_CRITICAL(Debug, "Unreachable code!");                                                  \
+        unreachable_impl();                                                                        \
+    } while (0)
 
-#define UNREACHABLE() assert_noinline_call([] { LOG_CRITICAL(Debug, "Unreachable code!"); })
 #define UNREACHABLE_MSG(...)                                                                       \
-    assert_noinline_call([&] { LOG_CRITICAL(Debug, "Unreachable code!\n" __VA_ARGS__); })
+    do {                                                                                           \
+        LOG_CRITICAL(Debug, "Unreachable code!\n" __VA_ARGS__);                                    \
+        unreachable_impl();                                                                        \
+    } while (0)
 
 #ifdef _DEBUG
 #define DEBUG_ASSERT(_a_) ASSERT(_a_)
diff --git a/src/common/bounded_threadsafe_queue.h b/src/common/bounded_threadsafe_queue.h
index e83064c7f..7e465549b 100644
--- a/src/common/bounded_threadsafe_queue.h
+++ b/src/common/bounded_threadsafe_queue.h
@@ -1,10 +1,7 @@
 // SPDX-FileCopyrightText: Copyright (c) 2020 Erik Rigtorp <erik@rigtorp.se>
 // SPDX-License-Identifier: MIT
+
 #pragma once
-#ifdef _MSC_VER
-#pragma warning(push)
-#pragma warning(disable : 4324)
-#endif
 
 #include <atomic>
 #include <bit>
@@ -12,105 +9,63 @@
 #include <memory>
 #include <mutex>
 #include <new>
-#include <stdexcept>
 #include <stop_token>
 #include <type_traits>
 #include <utility>
 
 namespace Common {
-namespace mpsc {
+
 #if defined(__cpp_lib_hardware_interference_size)
 constexpr size_t hardware_interference_size = std::hardware_destructive_interference_size;
 #else
 constexpr size_t hardware_interference_size = 64;
 #endif
 
-template <typename T>
-using AlignedAllocator = std::allocator<T>;
-
-template <typename T>
-struct Slot {
-    ~Slot() noexcept {
-        if (turn.test()) {
-            destroy();
-        }
-    }
-
-    template <typename... Args>
-    void construct(Args&&... args) noexcept {
-        static_assert(std::is_nothrow_constructible_v<T, Args&&...>,
-                      "T must be nothrow constructible with Args&&...");
-        std::construct_at(reinterpret_cast<T*>(&storage), std::forward<Args>(args)...);
-    }
-
-    void destroy() noexcept {
-        static_assert(std::is_nothrow_destructible_v<T>, "T must be nothrow destructible");
-        std::destroy_at(reinterpret_cast<T*>(&storage));
-    }
-
-    T&& move() noexcept {
-        return reinterpret_cast<T&&>(storage);
-    }
-
-    // Align to avoid false sharing between adjacent slots
-    alignas(hardware_interference_size) std::atomic_flag turn{};
-    struct aligned_store {
-        struct type {
-            alignas(T) unsigned char data[sizeof(T)];
-        };
-    };
-    typename aligned_store::type storage;
-};
+#ifdef _MSC_VER
+#pragma warning(push)
+#pragma warning(disable : 4324)
+#endif
 
-template <typename T, typename Allocator = AlignedAllocator<Slot<T>>>
-class Queue {
+template <typename T, size_t capacity = 0x400>
+class MPSCQueue {
 public:
-    explicit Queue(const size_t capacity, const Allocator& allocator = Allocator())
-        : allocator_(allocator) {
-        if (capacity < 1) {
-            throw std::invalid_argument("capacity < 1");
-        }
-        // Ensure that the queue length is an integer power of 2
-        // This is so that idx(i) can be a simple i & mask_ insted of i % capacity
-        // https://github.com/rigtorp/MPMCQueue/pull/36
-        if (!std::has_single_bit(capacity)) {
-            throw std::invalid_argument("capacity must be an integer power of 2");
-        }
-
-        mask_ = capacity - 1;
-
+    explicit MPSCQueue() : allocator{std::allocator<Slot<T>>()} {
         // Allocate one extra slot to prevent false sharing on the last slot
-        slots_ = allocator_.allocate(mask_ + 2);
+        slots = allocator.allocate(capacity + 1);
         // Allocators are not required to honor alignment for over-aligned types
         // (see http://eel.is/c++draft/allocator.requirements#10) so we verify
         // alignment here
-        if (reinterpret_cast<uintptr_t>(slots_) % alignof(Slot<T>) != 0) {
-            allocator_.deallocate(slots_, mask_ + 2);
+        if (reinterpret_cast<uintptr_t>(slots) % alignof(Slot<T>) != 0) {
+            allocator.deallocate(slots, capacity + 1);
             throw std::bad_alloc();
         }
-        for (size_t i = 0; i < mask_ + 1; ++i) {
-            std::construct_at(&slots_[i]);
+        for (size_t i = 0; i < capacity; ++i) {
+            std::construct_at(&slots[i]);
         }
+        static_assert(std::has_single_bit(capacity), "capacity must be an integer power of 2");
         static_assert(alignof(Slot<T>) == hardware_interference_size,
                       "Slot must be aligned to cache line boundary to prevent false sharing");
         static_assert(sizeof(Slot<T>) % hardware_interference_size == 0,
                       "Slot size must be a multiple of cache line size to prevent "
                       "false sharing between adjacent slots");
-        static_assert(sizeof(Queue) % hardware_interference_size == 0,
+        static_assert(sizeof(MPSCQueue) % hardware_interference_size == 0,
                       "Queue size must be a multiple of cache line size to "
                       "prevent false sharing between adjacent queues");
     }
 
-    ~Queue() noexcept {
-        for (size_t i = 0; i < mask_ + 1; ++i) {
-            slots_[i].~Slot();
+    ~MPSCQueue() noexcept {
+        for (size_t i = 0; i < capacity; ++i) {
+            std::destroy_at(&slots[i]);
         }
-        allocator_.deallocate(slots_, mask_ + 2);
+        allocator.deallocate(slots, capacity + 1);
     }
 
-    // non-copyable and non-movable
-    Queue(const Queue&) = delete;
-    Queue& operator=(const Queue&) = delete;
+    // The queue must be both non-copyable and non-movable
+    MPSCQueue(const MPSCQueue&) = delete;
+    MPSCQueue& operator=(const MPSCQueue&) = delete;
+
+    MPSCQueue(MPSCQueue&&) = delete;
+    MPSCQueue& operator=(MPSCQueue&&) = delete;
 
     void Push(const T& v) noexcept {
         static_assert(std::is_nothrow_copy_constructible_v<T>,
@@ -125,8 +80,8 @@ public:
 
     void Pop(T& v, std::stop_token stop) noexcept {
         auto const tail = tail_.fetch_add(1);
-        auto& slot = slots_[idx(tail)];
-        if (false == slot.turn.test()) {
+        auto& slot = slots[idx(tail)];
+        if (!slot.turn.test()) {
             std::unique_lock lock{cv_mutex};
             cv.wait(lock, stop, [&slot] { return slot.turn.test(); });
         }
@@ -137,12 +92,46 @@ public:
     }
 
 private:
+    template <typename U = T>
+    struct Slot {
+        ~Slot() noexcept {
+            if (turn.test()) {
+                destroy();
+            }
+        }
+
+        template <typename... Args>
+        void construct(Args&&... args) noexcept {
+            static_assert(std::is_nothrow_constructible_v<U, Args&&...>,
+                          "T must be nothrow constructible with Args&&...");
+            std::construct_at(reinterpret_cast<U*>(&storage), std::forward<Args>(args)...);
+        }
+
+        void destroy() noexcept {
+            static_assert(std::is_nothrow_destructible_v<U>, "T must be nothrow destructible");
+            std::destroy_at(reinterpret_cast<U*>(&storage));
+        }
+
+        U&& move() noexcept {
+            return reinterpret_cast<U&&>(storage);
+        }
+
+        // Align to avoid false sharing between adjacent slots
+        alignas(hardware_interference_size) std::atomic_flag turn{};
+        struct aligned_store {
+            struct type {
+                alignas(U) unsigned char data[sizeof(U)];
+            };
+        };
+        typename aligned_store::type storage;
+    };
+
     template <typename... Args>
     void emplace(Args&&... args) noexcept {
         static_assert(std::is_nothrow_constructible_v<T, Args&&...>,
                       "T must be nothrow constructible with Args&&...");
         auto const head = head_.fetch_add(1);
-        auto& slot = slots_[idx(head)];
+        auto& slot = slots[idx(head)];
         slot.turn.wait(true);
         slot.construct(std::forward<Args>(args)...);
         slot.turn.test_and_set();
@@ -150,31 +139,29 @@ private:
     }
 
     constexpr size_t idx(size_t i) const noexcept {
-        return i & mask_;
+        return i & mask;
     }
 
-    std::conditional_t<true, std::condition_variable_any, std::condition_variable> cv;
-    std::mutex cv_mutex;
-    size_t mask_;
-    Slot<T>* slots_;
-    [[no_unique_address]] Allocator allocator_;
+    static constexpr size_t mask = capacity - 1;
 
     // Align to avoid false sharing between head_ and tail_
     alignas(hardware_interference_size) std::atomic<size_t> head_{0};
     alignas(hardware_interference_size) std::atomic<size_t> tail_{0};
 
+    std::mutex cv_mutex;
+    std::condition_variable_any cv;
+
+    Slot<T>* slots;
+    [[no_unique_address]] std::allocator<Slot<T>> allocator;
+
     static_assert(std::is_nothrow_copy_assignable_v<T> || std::is_nothrow_move_assignable_v<T>,
                   "T must be nothrow copy or move assignable");
 
     static_assert(std::is_nothrow_destructible_v<T>, "T must be nothrow destructible");
 };
-} // namespace mpsc
-
-template <typename T, typename Allocator = mpsc::AlignedAllocator<mpsc::Slot<T>>>
-using MPSCQueue = mpsc::Queue<T, Allocator>;
-
-} // namespace Common
 
 #ifdef _MSC_VER
 #pragma warning(pop)
 #endif
+
+} // namespace Common
diff --git a/src/common/detached_tasks.cpp b/src/common/detached_tasks.cpp
index c1362631e..ec31d0b88 100644
--- a/src/common/detached_tasks.cpp
+++ b/src/common/detached_tasks.cpp
@@ -33,9 +33,9 @@ void DetachedTasks::AddTask(std::function<void()> task) {
     ++instance->count;
     std::thread([task{std::move(task)}]() {
         task();
-        std::unique_lock lock{instance->mutex};
+        std::unique_lock thread_lock{instance->mutex};
         --instance->count;
-        std::notify_all_at_thread_exit(instance->cv, std::move(lock));
+        std::notify_all_at_thread_exit(instance->cv, std::move(thread_lock));
     }).detach();
 }
 
diff --git a/src/common/page_table.h b/src/common/page_table.h
index fcbd12a43..1ad3a9f8b 100644
--- a/src/common/page_table.h
+++ b/src/common/page_table.h
@@ -15,6 +15,9 @@ enum class PageType : u8 {
     Unmapped,
     /// Page is mapped to regular memory. This is the only type you can get pointers to.
     Memory,
+    /// Page is mapped to regular memory, but inaccessible from CPU fastmem and must use
+    /// the callbacks.
+    DebugMemory,
     /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and
     /// invalidation
     RasterizerCachedMemory,
diff --git a/src/common/param_package.cpp b/src/common/param_package.cpp
index bbf20f5eb..462502e34 100644
--- a/src/common/param_package.cpp
+++ b/src/common/param_package.cpp
@@ -76,7 +76,7 @@ std::string ParamPackage::Serialize() const {
 std::string ParamPackage::Get(const std::string& key, const std::string& default_value) const {
     auto pair = data.find(key);
     if (pair == data.end()) {
-        LOG_DEBUG(Common, "key '{}' not found", key);
+        LOG_TRACE(Common, "key '{}' not found", key);
         return default_value;
     }
 
@@ -86,7 +86,7 @@ std::string ParamPackage::Get(const std::string& key, const std::string& default
 int ParamPackage::Get(const std::string& key, int default_value) const {
     auto pair = data.find(key);
     if (pair == data.end()) {
-        LOG_DEBUG(Common, "key '{}' not found", key);
+        LOG_TRACE(Common, "key '{}' not found", key);
         return default_value;
     }
 
@@ -101,7 +101,7 @@ int ParamPackage::Get(const std::string& key, int default_value) const {
 float ParamPackage::Get(const std::string& key, float default_value) const {
     auto pair = data.find(key);
     if (pair == data.end()) {
-        LOG_DEBUG(Common, "key {} not found", key);
+        LOG_TRACE(Common, "key {} not found", key);
         return default_value;
     }
 
diff --git a/src/common/settings.cpp b/src/common/settings.cpp
index 6ffab63af..751549583 100644
--- a/src/common/settings.cpp
+++ b/src/common/settings.cpp
@@ -147,7 +147,7 @@ void UpdateRescalingInfo() {
         info.down_shift = 0;
         break;
     default:
-        UNREACHABLE();
+        ASSERT(false);
         info.up_scale = 1;
         info.down_shift = 0;
     }
diff --git a/src/common/thread.cpp b/src/common/thread.cpp
index f932a7290..919e33af9 100644
--- a/src/common/thread.cpp
+++ b/src/common/thread.cpp
@@ -47,6 +47,9 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) {
     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;
@@ -59,9 +62,10 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) {
 void SetCurrentThreadPriority(ThreadPriority new_priority) {
     pthread_t this_thread = pthread_self();
 
-    s32 max_prio = sched_get_priority_max(SCHED_OTHER);
-    s32 min_prio = sched_get_priority_min(SCHED_OTHER);
-    u32 level = static_cast<u32>(new_priority) + 1;
+    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) {
@@ -70,7 +74,7 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) {
         params.sched_priority = min_prio - ((min_prio - max_prio) * level) / 4;
     }
 
-    pthread_setschedparam(this_thread, SCHED_OTHER, &params);
+    pthread_setschedparam(this_thread, scheduling_type, &params);
 }
 
 #endif
diff --git a/src/common/thread.h b/src/common/thread.h
index a63122516..1552f58e0 100644
--- a/src/common/thread.h
+++ b/src/common/thread.h
@@ -92,6 +92,7 @@ enum class ThreadPriority : u32 {
     Normal = 1,
     High = 2,
     VeryHigh = 3,
+    Critical = 4,
 };
 
 void SetCurrentThreadPriority(ThreadPriority new_priority);
diff --git a/src/common/uint128.h b/src/common/uint128.h
index f890ffec2..199d0f55e 100644
--- a/src/common/uint128.h
+++ b/src/common/uint128.h
@@ -31,12 +31,17 @@ namespace Common {
     return _udiv128(r[1], r[0], d, &remainder);
 #endif
 #else
+#ifdef __SIZEOF_INT128__
+    const auto product = static_cast<unsigned __int128>(a) * static_cast<unsigned __int128>(b);
+    return static_cast<u64>(product / d);
+#else
     const u64 diva = a / d;
     const u64 moda = a % d;
     const u64 divb = b / d;
     const u64 modb = b % d;
     return diva * b + moda * divb + moda * modb / d;
 #endif
+#endif
 }
 
 // This function multiplies 2 u64 values and produces a u128 value;
diff --git a/src/common/x64/native_clock.cpp b/src/common/x64/native_clock.cpp
index 1b7194503..6aaa8cdf9 100644
--- a/src/common/x64/native_clock.cpp
+++ b/src/common/x64/native_clock.cpp
@@ -75,8 +75,8 @@ NativeClock::NativeClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequen
 }
 
 u64 NativeClock::GetRTSC() {
-    TimePoint new_time_point{};
     TimePoint current_time_point{};
+    TimePoint new_time_point{};
 
     current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
     do {
@@ -89,8 +89,7 @@ u64 NativeClock::GetRTSC() {
         new_time_point.inner.accumulated_ticks = current_time_point.inner.accumulated_ticks + diff;
     } while (!Common::AtomicCompareAndSwap(time_point.pack.data(), new_time_point.pack,
                                            current_time_point.pack, current_time_point.pack));
-    /// The clock cannot be more precise than the guest timer, remove the lower bits
-    return new_time_point.inner.accumulated_ticks & inaccuracy_mask;
+    return new_time_point.inner.accumulated_ticks;
 }
 
 void NativeClock::Pause(bool is_paused) {
diff --git a/src/common/x64/native_clock.h b/src/common/x64/native_clock.h
index 30d2ba2e9..38ae7a462 100644
--- a/src/common/x64/native_clock.h
+++ b/src/common/x64/native_clock.h
@@ -37,12 +37,8 @@ private:
         } inner;
     };
 
-    /// value used to reduce the native clocks accuracy as some apss rely on
-    /// undefined behavior where the level of accuracy in the clock shouldn't
-    /// be higher.
-    static constexpr u64 inaccuracy_mask = ~(UINT64_C(0x400) - 1);
-
     TimePoint time_point;
+
     // factors
     u64 clock_rtsc_factor{};
     u64 cpu_rtsc_factor{};