"Merge Tagged PR 6598"

.ci/scripts/linux/docker.sh

@@ -51,6 +51,9 @@ mkdir -p AppDir/usr/optional/libgcc_s
 # Deploy yuzu's needed dependencies
 ./linuxdeploy-x86_64.AppImage --appdir AppDir --plugin qt
 
+# Workaround for libQt5MultimediaGstTools indirectly requiring libwayland-client and breaking Vulkan usage on end-user systems
+find AppDir -type f -regex '.*libwayland-client\.so.*' -delete -print
+
 # Workaround for building yuzu with GCC 10 but also trying to distribute it to Ubuntu 18.04 et al.
 # See https://github.com/darealshinji/AppImageKit-checkrt
 cp exec-x86_64.so AppDir/usr/optional/exec.so

.ci/templates/build-msvc.yml

@@ -6,7 +6,7 @@ parameters:
 steps:
 - script: choco install vulkan-sdk
   displayName: 'Install vulkan-sdk'
-- script: refreshenv && mkdir build && cd build && cmake -G "Visual Studio 17 2022" -A x64 -DYUZU_USE_BUNDLED_QT=1 -DYUZU_USE_BUNDLED_SDL2=1 -DYUZU_USE_QT_WEB_ENGINE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DYUZU_TESTS=OFF -DUSE_DISCORD_PRESENCE=ON -DENABLE_QT_TRANSLATION=ON -DDISPLAY_VERSION=${{ parameters['version'] }} -DCMAKE_BUILD_TYPE=Release -DYUZU_TESTS=OFF -DYUZU_USE_BUNDLED_VCPKG=ON .. && cd ..
+- script: refreshenv && mkdir build && cd build && cmake -G "Visual Studio 16 2019" -A x64 -DYUZU_USE_BUNDLED_QT=1 -DYUZU_USE_BUNDLED_SDL2=1 -DYUZU_USE_QT_WEB_ENGINE=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${COMPAT} -DYUZU_TESTS=OFF -DUSE_DISCORD_PRESENCE=ON -DENABLE_QT_TRANSLATION=ON -DDISPLAY_VERSION=${{ parameters['version'] }} -DCMAKE_BUILD_TYPE=Release -DYUZU_TESTS=OFF -DYUZU_USE_BUNDLED_VCPKG=ON .. && cd ..
   displayName: 'Configure CMake'
 - task: MSBuild@1
   displayName: 'Build'

.ci/yuzu-mainline-step2.yml

@@ -47,7 +47,7 @@ stages:
     timeoutInMinutes: 120
     displayName: 'msvc'
     pool:
-      vmImage: windows-2022
+      vmImage: windows-2019
     steps:
     - template: ./templates/sync-source.yml
       parameters:

.ci/yuzu-patreon-step2.yml

@@ -12,7 +12,7 @@ stages:
     timeoutInMinutes: 120
     displayName: 'windows-msvc'
     pool:
-      vmImage: windows-2022
+      vmImage: windows-2019
     steps:
     - template: ./templates/sync-source.yml
       parameters:

src/common/uint128.h

@@ -30,10 +30,6 @@ namespace Common {
 #else
     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;
@@ -41,7 +37,6 @@ namespace Common {
     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;

src/common/x64/native_clock.cpp

@@ -75,8 +75,8 @@ NativeClock::NativeClock(u64 emulated_cpu_frequency_, u64 emulated_clock_frequen
 }
 
 u64 NativeClock::GetRTSC() {
-    TimePoint current_time_point{};
     TimePoint new_time_point{};
+    TimePoint current_time_point{};
 
     current_time_point.pack = Common::AtomicLoad128(time_point.pack.data());
     do {

src/core/arm/dynarmic/arm_dynarmic_32.cpp

@@ -125,7 +125,9 @@ public:
     }
 
     void AddTicks(u64 ticks) override {
-        ASSERT_MSG(!parent.uses_wall_clock, "This should never happen - dynarmic ticking disabled");
+        if (parent.uses_wall_clock) {
+            return;
+        }
 
         // Divide the number of ticks by the amount of CPU cores. TODO(Subv): This yields only a
         // rough approximation of the amount of executed ticks in the system, it may be thrown off
@@ -142,7 +144,12 @@ public:
     }
 
     u64 GetTicksRemaining() override {
-        ASSERT_MSG(!parent.uses_wall_clock, "This should never happen - dynarmic ticking disabled");
+        if (parent.uses_wall_clock) {
+            if (!parent.interrupt_handlers[parent.core_index].IsInterrupted()) {
+                return minimum_run_cycles;
+            }
+            return 0U;
+        }
 
         return std::max<s64>(parent.system.CoreTiming().GetDowncount(), 0);
     }
@@ -172,7 +179,7 @@ public:
     Core::Memory::Memory& memory;
     std::size_t num_interpreted_instructions{};
     bool debugger_enabled{};
-    static constexpr u64 minimum_run_cycles = 1000U;
+    static constexpr u64 minimum_run_cycles = 10000U;
 };
 
 std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable* page_table) const {
@@ -200,7 +207,7 @@ std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable*
 
     // Timing
     config.wall_clock_cntpct = uses_wall_clock;
-    config.enable_cycle_counting = !uses_wall_clock;
+    config.enable_cycle_counting = true;
 
     // Code cache size
     config.code_cache_size = 512_MiB;

src/core/arm/dynarmic/arm_dynarmic_64.cpp

@@ -166,7 +166,9 @@ public:
     }
 
     void AddTicks(u64 ticks) override {
-        ASSERT_MSG(!parent.uses_wall_clock, "This should never happen - dynarmic ticking disabled");
+        if (parent.uses_wall_clock) {
+            return;
+        }
 
         // Divide the number of ticks by the amount of CPU cores. TODO(Subv): This yields only a
         // rough approximation of the amount of executed ticks in the system, it may be thrown off
@@ -181,7 +183,12 @@ public:
     }
 
     u64 GetTicksRemaining() override {
-        ASSERT_MSG(!parent.uses_wall_clock, "This should never happen - dynarmic ticking disabled");
+        if (parent.uses_wall_clock) {
+            if (!parent.interrupt_handlers[parent.core_index].IsInterrupted()) {
+                return minimum_run_cycles;
+            }
+            return 0U;
+        }
 
         return std::max<s64>(parent.system.CoreTiming().GetDowncount(), 0);
     }
@@ -216,7 +223,7 @@ public:
     u64 tpidrro_el0 = 0;
     u64 tpidr_el0 = 0;
     bool debugger_enabled{};
-    static constexpr u64 minimum_run_cycles = 1000U;
+    static constexpr u64 minimum_run_cycles = 10000U;
 };
 
 std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable* page_table,
@@ -260,7 +267,7 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable*
 
     // Timing
     config.wall_clock_cntpct = uses_wall_clock;
-    config.enable_cycle_counting = !uses_wall_clock;
+    config.enable_cycle_counting = true;
 
     // Code cache size
     config.code_cache_size = 512_MiB;

src/core/core_timing.cpp

@@ -6,9 +6,7 @@
 #include <string>
 #include <tuple>
 
-#include "common/logging/log.h"
 #include "common/microprofile.h"
-#include "common/thread.h"
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
 #include "core/hardware_properties.h"
@@ -44,10 +42,10 @@ CoreTiming::CoreTiming()
 
 CoreTiming::~CoreTiming() = default;
 
-void CoreTiming::ThreadEntry(CoreTiming& instance, size_t id) {
-    const std::string name = "yuzu:HostTiming_" + std::to_string(id);
-    MicroProfileOnThreadCreate(name.c_str());
-    Common::SetCurrentThreadName(name.c_str());
+void CoreTiming::ThreadEntry(CoreTiming& instance) {
+    constexpr char name[] = "yuzu:HostTiming";
+    MicroProfileOnThreadCreate(name);
+    Common::SetCurrentThreadName(name);
     Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical);
     instance.on_thread_init();
     instance.ThreadLoop();
@@ -63,127 +61,100 @@ void CoreTiming::Initialize(std::function<void()>&& on_thread_init_) {
         -> std::optional<std::chrono::nanoseconds> { return std::nullopt; };
     ev_lost = CreateEvent("_lost_event", empty_timed_callback);
     if (is_multicore) {
-        worker_threads.emplace_back(ThreadEntry, std::ref(*this), 0);
+        timer_thread = std::make_unique<std::thread>(ThreadEntry, std::ref(*this));
     }
 }
 
 void CoreTiming::Shutdown() {
-    is_paused = true;
+    paused = true;
     shutting_down = true;
-    std::atomic_thread_fence(std::memory_order_release);
-
-    event_cv.notify_all();
-    wait_pause_cv.notify_all();
-    for (auto& thread : worker_threads) {
-        thread.join();
+    pause_event.Set();
+    event.Set();
+    if (timer_thread) {
+        timer_thread->join();
     }
-    worker_threads.clear();
     pause_callbacks.clear();
     ClearPendingEvents();
+    timer_thread.reset();
     has_started = false;
 }
 
-void CoreTiming::Pause(bool is_paused_) {
-    std::unique_lock main_lock(event_mutex);
-    if (is_paused_ == paused_state.load(std::memory_order_relaxed)) {
-        return;
-    }
-    if (is_multicore) {
-        is_paused = is_paused_;
-        event_cv.notify_all();
-        if (!is_paused_) {
-            wait_pause_cv.notify_all();
-        }
-    }
-    paused_state.store(is_paused_, std::memory_order_relaxed);
+void CoreTiming::Pause(bool is_paused) {
+    paused = is_paused;
+    pause_event.Set();
 
-    if (!is_paused_) {
+    if (!is_paused) {
         pause_end_time = GetGlobalTimeNs().count();
     }
 
     for (auto& cb : pause_callbacks) {
-        cb(is_paused_);
+        cb(is_paused);
     }
 }
 
-void CoreTiming::SyncPause(bool is_paused_) {
-    std::unique_lock main_lock(event_mutex);
-    if (is_paused_ == paused_state.load(std::memory_order_relaxed)) {
+void CoreTiming::SyncPause(bool is_paused) {
+    if (is_paused == paused && paused_set == paused) {
         return;
     }
 
-    if (is_multicore) {
-        is_paused = is_paused_;
-        event_cv.notify_all();
-        if (!is_paused_) {
-            wait_pause_cv.notify_all();
-        }
-    }
-    paused_state.store(is_paused_, std::memory_order_relaxed);
-    if (is_multicore) {
-        if (is_paused_) {
-            wait_signal_cv.wait(main_lock, [this] { return pause_count == worker_threads.size(); });
-        } else {
-            wait_signal_cv.wait(main_lock, [this] { return pause_count == 0; });
+    Pause(is_paused);
+    if (timer_thread) {
+        if (!is_paused) {
+            pause_event.Set();
         }
+        event.Set();
+        while (paused_set != is_paused)
+            ;
     }
 
-    if (!is_paused_) {
+    if (!is_paused) {
         pause_end_time = GetGlobalTimeNs().count();
     }
 
     for (auto& cb : pause_callbacks) {
-        cb(is_paused_);
+        cb(is_paused);
     }
 }
 
 bool CoreTiming::IsRunning() const {
-    return !paused_state.load(std::memory_order_acquire);
+    return !paused_set;
 }
 
 bool CoreTiming::HasPendingEvents() const {
-    std::unique_lock main_lock(event_mutex);
-    return !event_queue.empty() || pending_events.load(std::memory_order_relaxed) != 0;
+    return !(wait_set && event_queue.empty());
 }
 
 void CoreTiming::ScheduleEvent(std::chrono::nanoseconds ns_into_future,
                                const std::shared_ptr<EventType>& event_type,
                                std::uintptr_t user_data, bool absolute_time) {
+    {
+        std::scoped_lock scope{basic_lock};
+        const auto next_time{absolute_time ? ns_into_future : GetGlobalTimeNs() + ns_into_future};
 
-    std::unique_lock main_lock(event_mutex);
-    const auto next_time{absolute_time ? ns_into_future : GetGlobalTimeNs() + ns_into_future};
-
-    event_queue.emplace_back(Event{next_time.count(), event_fifo_id++, user_data, event_type, 0});
-    pending_events.fetch_add(1, std::memory_order_relaxed);
-
-    std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
-
-    if (is_multicore) {
-        event_cv.notify_one();
+        event_queue.emplace_back(
+            Event{next_time.count(), event_fifo_id++, user_data, event_type, 0});
+        std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
     }
+
+    event.Set();
 }
 
 void CoreTiming::ScheduleLoopingEvent(std::chrono::nanoseconds start_time,
                                       std::chrono::nanoseconds resched_time,
                                       const std::shared_ptr<EventType>& event_type,
                                       std::uintptr_t user_data, bool absolute_time) {
-    std::unique_lock main_lock(event_mutex);
+    std::scoped_lock scope{basic_lock};
     const auto next_time{absolute_time ? start_time : GetGlobalTimeNs() + start_time};
 
     event_queue.emplace_back(
         Event{next_time.count(), event_fifo_id++, user_data, event_type, resched_time.count()});
-    pending_events.fetch_add(1, std::memory_order_relaxed);
 
     std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
-
-    if (is_multicore) {
-        event_cv.notify_one();
-    }
 }
 
 void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type,
                                  std::uintptr_t user_data) {
-    std::unique_lock main_lock(event_mutex);
+    std::scoped_lock scope{basic_lock};
     const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) {
         return e.type.lock().get() == event_type.get() && e.user_data == user_data;
     });
@@ -192,7 +163,6 @@ void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type,
     if (itr != event_queue.end()) {
         event_queue.erase(itr, event_queue.end());
         std::make_heap(event_queue.begin(), event_queue.end(), std::greater<>());
-        pending_events.fetch_sub(1, std::memory_order_relaxed);
     }
 }
 
@@ -232,12 +202,11 @@ u64 CoreTiming::GetClockTicks() const {
 }
 
 void CoreTiming::ClearPendingEvents() {
-    std::unique_lock main_lock(event_mutex);
     event_queue.clear();
 }
 
 void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) {
-    std::unique_lock main_lock(event_mutex);
+    std::scoped_lock lock{basic_lock};
 
     const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) {
         return e.type.lock().get() == event_type.get();
@@ -251,28 +220,27 @@ void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) {
 }
 
 void CoreTiming::RegisterPauseCallback(PauseCallback&& callback) {
-    std::unique_lock main_lock(event_mutex);
+    std::scoped_lock lock{basic_lock};
     pause_callbacks.emplace_back(std::move(callback));
 }
 
 std::optional<s64> CoreTiming::Advance() {
+    std::scoped_lock lock{advance_lock, basic_lock};
     global_timer = GetGlobalTimeNs().count();
 
-    std::unique_lock main_lock(event_mutex);
     while (!event_queue.empty() && event_queue.front().time <= global_timer) {
         Event evt = std::move(event_queue.front());
         std::pop_heap(event_queue.begin(), event_queue.end(), std::greater<>());
         event_queue.pop_back();
 
         if (const auto event_type{evt.type.lock()}) {
-            event_mutex.unlock();
+            basic_lock.unlock();
 
             const auto new_schedule_time{event_type->callback(
                 evt.user_data, evt.time,
                 std::chrono::nanoseconds{GetGlobalTimeNs().count() - evt.time})};
 
-            event_mutex.lock();
-            pending_events.fetch_sub(1, std::memory_order_relaxed);
+            basic_lock.lock();
 
             if (evt.reschedule_time != 0) {
                 // If this event was scheduled into a pause, its time now is going to be way behind.
@@ -285,9 +253,9 @@ std::optional<s64> CoreTiming::Advance() {
                 const auto next_schedule_time{new_schedule_time.has_value()
                                                   ? new_schedule_time.value().count()
                                                   : evt.reschedule_time};
+
                 event_queue.emplace_back(
                     Event{next_time, event_fifo_id++, evt.user_data, evt.type, next_schedule_time});
-                pending_events.fetch_add(1, std::memory_order_relaxed);
                 std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
             }
         }
@@ -304,34 +272,27 @@ std::optional<s64> CoreTiming::Advance() {
 }
 
 void CoreTiming::ThreadLoop() {
-    const auto predicate = [this] { return !event_queue.empty() || is_paused; };
     has_started = true;
     while (!shutting_down) {
-        while (!is_paused && !shutting_down) {
+        while (!paused) {
+            paused_set = false;
             const auto next_time = Advance();
             if (next_time) {
                 if (*next_time > 0) {
                     std::chrono::nanoseconds next_time_ns = std::chrono::nanoseconds(*next_time);
-                    std::unique_lock main_lock(event_mutex);
-                    event_cv.wait_for(main_lock, next_time_ns, predicate);
+                    event.WaitFor(next_time_ns);
                 }
             } else {
-                std::unique_lock main_lock(event_mutex);
-                event_cv.wait(main_lock, predicate);
+                wait_set = true;
+                event.Wait();
             }
+            wait_set = false;
         }
-        std::unique_lock main_lock(event_mutex);
-        pause_count++;
-        if (pause_count == worker_threads.size()) {
-            clock->Pause(true);
-            wait_signal_cv.notify_all();
-        }
-        wait_pause_cv.wait(main_lock, [this] { return !is_paused || shutting_down; });
-        pause_count--;
-        if (pause_count == 0) {
-            clock->Pause(false);
-            wait_signal_cv.notify_all();
-        }
+
+        paused_set = true;
+        clock->Pause(true);
+        pause_event.Wait();
+        clock->Pause(false);
     }
 }
 

src/core/core_timing.h

@@ -5,7 +5,6 @@
 
 #include <atomic>
 #include <chrono>
-#include <condition_variable>
 #include <functional>
 #include <memory>
 #include <mutex>
@@ -15,6 +14,7 @@
 #include <vector>
 
 #include "common/common_types.h"
+#include "common/thread.h"
 #include "common/wall_clock.h"
 
 namespace Core::Timing {
@@ -143,7 +143,7 @@ private:
     /// Clear all pending events. This should ONLY be done on exit.
     void ClearPendingEvents();
 
-    static void ThreadEntry(CoreTiming& instance, size_t id);
+    static void ThreadEntry(CoreTiming& instance);
     void ThreadLoop();
 
     std::unique_ptr<Common::WallClock> clock;
@@ -156,24 +156,21 @@ private:
     // accomodated by the standard adaptor class.
     std::vector<Event> event_queue;
     u64 event_fifo_id = 0;
-    std::atomic<size_t> pending_events{};
 
     std::shared_ptr<EventType> ev_lost;
+    Common::Event event{};
+    Common::Event pause_event{};
+    std::mutex basic_lock;
+    std::mutex advance_lock;
+    std::unique_ptr<std::thread> timer_thread;
+    std::atomic<bool> paused{};
+    std::atomic<bool> paused_set{};
+    std::atomic<bool> wait_set{};
+    std::atomic<bool> shutting_down{};
     std::atomic<bool> has_started{};
     std::function<void()> on_thread_init{};
 
-    std::vector<std::thread> worker_threads;
-
-    std::condition_variable event_cv;
-    std::condition_variable wait_pause_cv;
-    std::condition_variable wait_signal_cv;
-    mutable std::mutex event_mutex;
-
-    std::atomic<bool> paused_state{};
-    bool is_paused{};
-    bool shutting_down{};
     bool is_multicore{};
-    size_t pause_count{};
     s64 pause_end_time{};
 
     /// Cycle timing

src/tests/core/core_timing.cpp

@@ -8,7 +8,6 @@
 #include <chrono>
 #include <cstdlib>
 #include <memory>
-#include <mutex>
 #include <optional>
 #include <string>
 
@@ -23,15 +22,14 @@ std::array<s64, 5> delays{};
 
 std::bitset<CB_IDS.size()> callbacks_ran_flags;
 u64 expected_callback = 0;
-std::mutex control_mutex;
 
 template <unsigned int IDX>
 std::optional<std::chrono::nanoseconds> HostCallbackTemplate(std::uintptr_t user_data, s64 time,
                                                              std::chrono::nanoseconds ns_late) {
-    std::unique_lock<std::mutex> lk(control_mutex);
     static_assert(IDX < CB_IDS.size(), "IDX out of range");
     callbacks_ran_flags.set(IDX);
     REQUIRE(CB_IDS[IDX] == user_data);
+    REQUIRE(CB_IDS[IDX] == CB_IDS[calls_order[expected_callback]]);
     delays[IDX] = ns_late.count();
     ++expected_callback;
     return std::nullopt;

src/video_core/gpu_thread.cpp

@@ -31,8 +31,7 @@ static void RunThread(std::stop_token stop_token, Core::System& system,
     VideoCore::RasterizerInterface* const rasterizer = renderer.ReadRasterizer();
 
     while (!stop_token.stop_requested()) {
-        CommandDataContainer next;
-        state.queue.Pop(next, stop_token);
+        CommandDataContainer next = state.queue.PopWait(stop_token);
         if (stop_token.stop_requested()) {
             break;
         }

src/video_core/gpu_thread.h

@@ -10,7 +10,7 @@
 #include <thread>
 #include <variant>
 
-#include "common/bounded_threadsafe_queue.h"
+#include "common/threadsafe_queue.h"
 #include "video_core/framebuffer_config.h"
 
 namespace Tegra {
@@ -96,7 +96,7 @@ struct CommandDataContainer {
 
 /// Struct used to synchronize the GPU thread
 struct SynchState final {
-    using CommandQueue = Common::MPSCQueue<CommandDataContainer>;
+    using CommandQueue = Common::MPSCQueue<CommandDataContainer, true>;
     std::mutex write_lock;
     CommandQueue queue;
     u64 last_fence{};

src/video_core/macro/macro_hle.cpp

@@ -3,6 +3,8 @@
 
 #include <array>
 #include <vector>
+#include "common/scope_exit.h"
+#include "video_core/dirty_flags.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/macro/macro.h"
 #include "video_core/macro/macro_hle.h"
@@ -58,6 +60,7 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.regs.index_array.first = parameters[3];
     maxwell3d.regs.reg_array[0x446] = element_base; // vertex id base?
     maxwell3d.regs.index_array.count = parameters[1];
+    maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
     maxwell3d.regs.vb_element_base = element_base;
     maxwell3d.regs.vb_base_instance = base_instance;
     maxwell3d.mme_draw.instance_count = instance_count;
@@ -80,10 +83,67 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
 }
 
-constexpr std::array<std::pair<u64, HLEFunction>, 3> hle_funcs{{
+// Multidraw Indirect
+void HLE_3f5e74b9c9a50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
+    SCOPE_EXIT({
+        // Clean everything.
+        maxwell3d.regs.reg_array[0x446] = 0x0; // vertex id base?
+        maxwell3d.regs.index_array.count = 0;
+        maxwell3d.regs.vb_element_base = 0x0;
+        maxwell3d.regs.vb_base_instance = 0x0;
+        maxwell3d.mme_draw.instance_count = 0;
+        maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+        maxwell3d.CallMethodFromMME(0x8e4, 0x0);
+        maxwell3d.CallMethodFromMME(0x8e5, 0x0);
+        maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+        maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+    });
+    const u32 start_indirect = parameters[0];
+    const u32 end_indirect = parameters[1];
+    if (start_indirect >= end_indirect) {
+        // Nothing to do.
+        return;
+    }
+    const auto topology =
+        static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[2]);
+    maxwell3d.regs.draw.topology.Assign(topology);
+    const u32 padding = parameters[3];
+    const std::size_t max_draws = parameters[4];
+
+    const u32 indirect_words = 5 + padding;
+    const std::size_t first_draw = start_indirect;
+    const std::size_t effective_draws = end_indirect - start_indirect;
+    const std::size_t last_draw = start_indirect + std::min(effective_draws, max_draws);
+
+    for (std::size_t index = first_draw; index < last_draw; index++) {
+        const std::size_t base = index * indirect_words + 5;
+        const u32 num_vertices = parameters[base];
+        const u32 instance_count = parameters[base + 1];
+        const u32 first_index = parameters[base + 2];
+        const u32 base_vertex = parameters[base + 3];
+        const u32 base_instance = parameters[base + 4];
+        maxwell3d.regs.index_array.first = first_index;
+        maxwell3d.regs.reg_array[0x446] = base_vertex;
+        maxwell3d.regs.index_array.count = num_vertices;
+        maxwell3d.regs.vb_element_base = base_vertex;
+        maxwell3d.regs.vb_base_instance = base_instance;
+        maxwell3d.mme_draw.instance_count = instance_count;
+        maxwell3d.CallMethodFromMME(0x8e3, 0x640);
+        maxwell3d.CallMethodFromMME(0x8e4, base_vertex);
+        maxwell3d.CallMethodFromMME(0x8e5, base_instance);
+        maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+        if (maxwell3d.ShouldExecute()) {
+            maxwell3d.Rasterizer().Draw(true, true);
+        }
+        maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
+    }
+}
+
+constexpr std::array<std::pair<u64, HLEFunction>, 4> hle_funcs{{
     {0x771BB18C62444DA0, &HLE_771BB18C62444DA0},
     {0x0D61FC9FAAC9FCAD, &HLE_0D61FC9FAAC9FCAD},
     {0x0217920100488FF7, &HLE_0217920100488FF7},
+    {0x3f5e74b9c9a50164, &HLE_3f5e74b9c9a50164},
 }};
 
 class HLEMacroImpl final : public CachedMacro {
@@ -99,6 +159,7 @@ private:
     Engines::Maxwell3D& maxwell3d;
     HLEFunction func;
 };
+
 } // Anonymous namespace
 
 HLEMacro::HLEMacro(Engines::Maxwell3D& maxwell3d_) : maxwell3d{maxwell3d_} {}

src/yuzu/main.cpp

@@ -9,6 +9,10 @@
 #ifdef __APPLE__
 #include <unistd.h> // for chdir
 #endif
+#ifdef __linux__
+#include <csignal>
+#include <sys/socket.h>
+#endif
 
 // VFS includes must be before glad as they will conflict with Windows file api, which uses defines.
 #include "applets/qt_controller.h"
@@ -260,6 +264,10 @@ GMainWindow::GMainWindow(bool has_broken_vulkan)
       config{std::make_unique<Config>(*system)},
       vfs{std::make_shared<FileSys::RealVfsFilesystem>()},
       provider{std::make_unique<FileSys::ManualContentProvider>()} {
+#ifdef __linux__
+    SetupSigInterrupts();
+#endif
+
     Common::Log::Initialize();
     LoadTranslation();
 
@@ -379,6 +387,8 @@ GMainWindow::GMainWindow(bool has_broken_vulkan)
     SDL_EnableScreenSaver();
 #endif
 
+    SetupPrepareForSleep();
+
     Common::Log::Start();
 
     QStringList args = QApplication::arguments();
@@ -463,7 +473,13 @@ GMainWindow::~GMainWindow() {
     if (render_window->parent() == nullptr) {
         delete render_window;
     }
+
     system->GetRoomNetwork().Shutdown();
+
+#ifdef __linux__
+    ::close(sig_interrupt_fds[0]);
+    ::close(sig_interrupt_fds[1]);
+#endif
 }
 
 void GMainWindow::RegisterMetaTypes() {
@@ -1314,6 +1330,43 @@ void GMainWindow::OnDisplayTitleBars(bool show) {
     }
 }
 
+void GMainWindow::SetupPrepareForSleep() {
+#ifdef __linux__
+    auto bus = QDBusConnection::systemBus();
+    if (bus.isConnected()) {
+        const bool success = bus.connect(
+            QStringLiteral("org.freedesktop.login1"), QStringLiteral("/org/freedesktop/login1"),
+            QStringLiteral("org.freedesktop.login1.Manager"), QStringLiteral("PrepareForSleep"),
+            QStringLiteral("b"), this, SLOT(OnPrepareForSleep(bool)));
+
+        if (!success) {
+            LOG_WARNING(Frontend, "Couldn't register PrepareForSleep signal");
+        }
+    } else {
+        LOG_WARNING(Frontend, "QDBusConnection system bus is not connected");
+    }
+#endif // __linux__
+}
+
+void GMainWindow::OnPrepareForSleep(bool prepare_sleep) {
+    if (emu_thread == nullptr) {
+        return;
+    }
+
+    if (prepare_sleep) {
+        if (emu_thread->IsRunning()) {
+            auto_paused = true;
+            OnPauseGame();
+        }
+    } else {
+        if (!emu_thread->IsRunning() && auto_paused) {
+            auto_paused = false;
+            RequestGameResume();
+            OnStartGame();
+        }
+    }
+}
+
 #ifdef __linux__
 static std::optional<QDBusObjectPath> HoldWakeLockLinux(u32 window_id = 0) {
     if (!QDBusConnection::sessionBus().isConnected()) {
@@ -1353,6 +1406,52 @@ static void ReleaseWakeLockLinux(QDBusObjectPath lock) {
                             QString::fromLatin1("org.freedesktop.portal.Request"));
     unlocker.call(QString::fromLatin1("Close"));
 }
+
+std::array<int, 3> GMainWindow::sig_interrupt_fds{0, 0, 0};
+
+void GMainWindow::SetupSigInterrupts() {
+    if (sig_interrupt_fds[2] == 1) {
+        return;
+    }
+    socketpair(AF_UNIX, SOCK_STREAM, 0, sig_interrupt_fds.data());
+    sig_interrupt_fds[2] = 1;
+
+    struct sigaction sa;
+    sa.sa_handler = &GMainWindow::HandleSigInterrupt;
+    sigemptyset(&sa.sa_mask);
+    sa.sa_flags = SA_RESETHAND;
+    sigaction(SIGINT, &sa, nullptr);
+    sigaction(SIGTERM, &sa, nullptr);
+
+    sig_interrupt_notifier = new QSocketNotifier(sig_interrupt_fds[1], QSocketNotifier::Read, this);
+    connect(sig_interrupt_notifier, &QSocketNotifier::activated, this,
+            &GMainWindow::OnSigInterruptNotifierActivated);
+    connect(this, &GMainWindow::SigInterrupt, this, &GMainWindow::close);
+}
+
+void GMainWindow::HandleSigInterrupt(int sig) {
+    if (sig == SIGINT) {
+        exit(1);
+    }
+
+    // Calling into Qt directly from a signal handler is not safe,
+    // so wake up a QSocketNotifier with this hacky write call instead.
+    char a = 1;
+    int ret = write(sig_interrupt_fds[0], &a, sizeof(a));
+    (void)ret;
+}
+
+void GMainWindow::OnSigInterruptNotifierActivated() {
+    sig_interrupt_notifier->setEnabled(false);
+
+    char a;
+    int ret = read(sig_interrupt_fds[1], &a, sizeof(a));
+    (void)ret;
+
+    sig_interrupt_notifier->setEnabled(true);
+
+    emit SigInterrupt();
+}
 #endif // __linux__
 
 void GMainWindow::PreventOSSleep() {

src/yuzu/main.h

@@ -164,6 +164,8 @@ signals:
     void WebBrowserExtractOfflineRomFS();
     void WebBrowserClosed(Service::AM::Applets::WebExitReason exit_reason, std::string last_url);
 
+    void SigInterrupt();
+
 public slots:
     void OnLoadComplete();
     void OnExecuteProgram(std::size_t program_index);
@@ -203,7 +205,7 @@ private:
     void ConnectMenuEvents();
     void UpdateMenuState();
 
-    MultiplayerState* multiplayer_state = nullptr;
+    void SetupPrepareForSleep();
 
     void PreventOSSleep();
     void AllowOSSleep();
@@ -252,12 +254,19 @@ private:
     void RequestGameResume();
     void closeEvent(QCloseEvent* event) override;
 
+#ifdef __linux__
+    void SetupSigInterrupts();
+    static void HandleSigInterrupt(int);
+    void OnSigInterruptNotifierActivated();
+#endif
+
 private slots:
     void OnStartGame();
     void OnRestartGame();
     void OnPauseGame();
     void OnPauseContinueGame();
     void OnStopGame();
+    void OnPrepareForSleep(bool prepare_sleep);
     void OnMenuReportCompatibility();
     void OnOpenModsPage();
     void OnOpenQuickstartGuide();
@@ -348,6 +357,8 @@ private:
     std::unique_ptr<DiscordRPC::DiscordInterface> discord_rpc;
     std::shared_ptr<InputCommon::InputSubsystem> input_subsystem;
 
+    MultiplayerState* multiplayer_state = nullptr;
+
     GRenderWindow* render_window;
     GameList* game_list;
     LoadingScreen* loading_screen;
@@ -420,6 +431,9 @@ private:
     bool is_tas_recording_dialog_active{};
 
 #ifdef __linux__
+    QSocketNotifier* sig_interrupt_notifier;
+    static std::array<int, 3> sig_interrupt_fds;
+
     QDBusObjectPath wake_lock{};
 #endif