audio_in/out_system: Pass Initialize members by value where applicable

applet_resource_user_id isn't actually modified and is just assigned to
a member variable, so this doesn't need to be a mutable reference.

Similarly, the device name itself isn't modified and is only moved. We
pass by value here, since we can still perform the move, but eliminate a
sneaky set of calls that can unintentionally destroy the original
string. Given how nested the calls are, it's good to get rid of this
potential vector for a use-after-move bug.

src/audio_core/in/audio_in_system.cpp

@@ -56,7 +56,7 @@ Result System::IsConfigValid(const std::string_view device_name,
     return ResultSuccess;
 }
 
-Result System::Initialize(std::string& device_name, const AudioInParameter& in_params,
+Result System::Initialize(std::string device_name, const AudioInParameter& in_params,
                           const u32 handle_, const u64 applet_resource_user_id_) {
     auto result{IsConfigValid(device_name, in_params)};
     if (result.IsError()) {

src/audio_core/in/audio_in_system.h

@@ -97,7 +97,7 @@ public:
      * @param applet_resource_user_id - Unused.
      * @return Result code.
      */
-    Result Initialize(std::string& device_name, const AudioInParameter& in_params, u32 handle,
+    Result Initialize(std::string device_name, const AudioInParameter& in_params, u32 handle,
                       u64 applet_resource_user_id);
 
     /**

src/audio_core/out/audio_out_system.cpp

@@ -49,8 +49,8 @@ Result System::IsConfigValid(std::string_view device_name,
     return Service::Audio::ERR_INVALID_CHANNEL_COUNT;
 }
 
-Result System::Initialize(std::string& device_name, const AudioOutParameter& in_params, u32 handle_,
-                          u64& applet_resource_user_id_) {
+Result System::Initialize(std::string device_name, const AudioOutParameter& in_params, u32 handle_,
+                          u64 applet_resource_user_id_) {
     auto result = IsConfigValid(device_name, in_params);
     if (result.IsError()) {
         return result;

src/audio_core/out/audio_out_system.h

@@ -88,8 +88,8 @@ public:
      * @param applet_resource_user_id - Unused.
      * @return Result code.
      */
-    Result Initialize(std::string& device_name, const AudioOutParameter& in_params, u32 handle,
-                      u64& applet_resource_user_id);
+    Result Initialize(std::string device_name, const AudioOutParameter& in_params, u32 handle,
+                      u64 applet_resource_user_id);
 
     /**
      * Start this system.

src/common/concepts.h

@@ -3,24 +3,14 @@
 
 #pragma once
 
+#include <iterator>
 #include <type_traits>
 
 namespace Common {
 
-// Check if type is like an STL container
+// Check if type satisfies the ContiguousContainer named requirement.
 template <typename T>
-concept IsSTLContainer = requires(T t) {
-    typename T::value_type;
-    typename T::iterator;
-    typename T::const_iterator;
-    // TODO(ogniK): Replace below is std::same_as<void> when MSVC supports it.
-    t.begin();
-    t.end();
-    t.cbegin();
-    t.cend();
-    t.data();
-    t.size();
-};
+concept IsContiguousContainer = std::contiguous_iterator<typename T::iterator>;
 
 // TODO: Replace with std::derived_from when the <concepts> header
 //       is available on all supported platforms.

src/common/fs/file.h

@@ -209,8 +209,8 @@ public:
 
     /**
      * Helper function which deduces the value type of a contiguous STL container used in ReadSpan.
-     * If T is not a contiguous STL container as defined by the concept IsSTLContainer, this calls
-     * ReadObject and T must be a trivially copyable object.
+     * If T is not a contiguous container as defined by the concept IsContiguousContainer, this
+     * calls ReadObject and T must be a trivially copyable object.
      *
      * See ReadSpan for more details if T is a contiguous container.
      * See ReadObject for more details if T is a trivially copyable object.
@@ -223,7 +223,7 @@ public:
      */
     template <typename T>
     [[nodiscard]] size_t Read(T& data) const {
-        if constexpr (IsSTLContainer<T>) {
+        if constexpr (IsContiguousContainer<T>) {
             using ContiguousType = typename T::value_type;
             static_assert(std::is_trivially_copyable_v<ContiguousType>,
                           "Data type must be trivially copyable.");
@@ -235,8 +235,8 @@ public:
 
     /**
      * Helper function which deduces the value type of a contiguous STL container used in WriteSpan.
-     * If T is not a contiguous STL container as defined by the concept IsSTLContainer, this calls
-     * WriteObject and T must be a trivially copyable object.
+     * If T is not a contiguous STL container as defined by the concept IsContiguousContainer, this
+     * calls WriteObject and T must be a trivially copyable object.
      *
      * See WriteSpan for more details if T is a contiguous container.
      * See WriteObject for more details if T is a trivially copyable object.
@@ -249,7 +249,7 @@ public:
      */
     template <typename T>
     [[nodiscard]] size_t Write(const T& data) const {
-        if constexpr (IsSTLContainer<T>) {
+        if constexpr (IsContiguousContainer<T>) {
             using ContiguousType = typename T::value_type;
             static_assert(std::is_trivially_copyable_v<ContiguousType>,
                           "Data type must be trivially copyable.");

src/common/input.h

@@ -100,7 +100,6 @@ enum class CameraError {
 enum class VibrationAmplificationType {
     Linear,
     Exponential,
-    Test,
 };
 
 // Analog properties for calibration
@@ -325,6 +324,10 @@ public:
         return VibrationError::NotSupported;
     }
 
+    virtual bool IsVibrationEnabled() {
+        return false;
+    }
+
     virtual PollingError SetPollingMode([[maybe_unused]] PollingMode polling_mode) {
         return PollingError::NotSupported;
     }

src/core/core.cpp

@@ -384,6 +384,7 @@ struct System::Impl {
         kernel.ShutdownCores();
         cpu_manager.Shutdown();
         debugger.reset();
+        services->KillNVNFlinger();
         kernel.CloseServices();
         services.reset();
         service_manager.reset();

src/core/hid/emulated_controller.cpp

@@ -970,14 +970,7 @@ bool EmulatedController::SetVibration(std::size_t device_index, VibrationValue v
            Common::Input::VibrationError::None;
 }
 
-bool EmulatedController::TestVibration(std::size_t device_index) {
-    if (device_index >= output_devices.size()) {
-        return false;
-    }
-    if (!output_devices[device_index]) {
-        return false;
-    }
-
+bool EmulatedController::IsVibrationEnabled(std::size_t device_index) {
     const auto player_index = NpadIdTypeToIndex(npad_id_type);
     const auto& player = Settings::values.players.GetValue()[player_index];
 
@@ -985,31 +978,15 @@ bool EmulatedController::TestVibration(std::size_t device_index) {
         return false;
     }
 
-    const Common::Input::VibrationStatus test_vibration = {
-        .low_amplitude = 0.001f,
-        .low_frequency = DEFAULT_VIBRATION_VALUE.low_frequency,
-        .high_amplitude = 0.001f,
-        .high_frequency = DEFAULT_VIBRATION_VALUE.high_frequency,
-        .type = Common::Input::VibrationAmplificationType::Test,
-    };
+    if (device_index >= output_devices.size()) {
+        return false;
+    }
 
-    const Common::Input::VibrationStatus zero_vibration = {
-        .low_amplitude = DEFAULT_VIBRATION_VALUE.low_amplitude,
-        .low_frequency = DEFAULT_VIBRATION_VALUE.low_frequency,
-        .high_amplitude = DEFAULT_VIBRATION_VALUE.high_amplitude,
-        .high_frequency = DEFAULT_VIBRATION_VALUE.high_frequency,
-        .type = Common::Input::VibrationAmplificationType::Test,
-    };
+    if (!output_devices[device_index]) {
+        return false;
+    }
 
-    // Send a slight vibration to test for rumble support
-    output_devices[device_index]->SetVibration(test_vibration);
-
-    // Wait for about 15ms to ensure the controller is ready for the stop command
-    std::this_thread::sleep_for(std::chrono::milliseconds(15));
-
-    // Stop any vibration and return the result
-    return output_devices[device_index]->SetVibration(zero_vibration) ==
-           Common::Input::VibrationError::None;
+    return output_devices[device_index]->IsVibrationEnabled();
 }
 
 bool EmulatedController::SetPollingMode(Common::Input::PollingMode polling_mode) {
@@ -1048,6 +1025,7 @@ bool EmulatedController::HasNfc() const {
     case NpadStyleIndex::JoyconRight:
     case NpadStyleIndex::JoyconDual:
     case NpadStyleIndex::ProController:
+    case NpadStyleIndex::Handheld:
         break;
     default:
         return false;
@@ -1234,12 +1212,6 @@ bool EmulatedController::IsConnected(bool get_temporary_value) const {
     return is_connected;
 }
 
-bool EmulatedController::IsVibrationEnabled() const {
-    const auto player_index = NpadIdTypeToIndex(npad_id_type);
-    const auto& player = Settings::values.players.GetValue()[player_index];
-    return player.vibration_enabled;
-}
-
 NpadIdType EmulatedController::GetNpadIdType() const {
     std::scoped_lock lock{mutex};
     return npad_id_type;

src/core/hid/emulated_controller.h

@@ -206,9 +206,6 @@ public:
      */
     bool IsConnected(bool get_temporary_value = false) const;
 
-    /// Returns true if vibration is enabled
-    bool IsVibrationEnabled() const;
-
     /// Removes all callbacks created from input devices
     void UnloadInput();
 
@@ -339,7 +336,7 @@ public:
      * Sends a small vibration to the output device
      * @return true if SetVibration was successfull
      */
-    bool TestVibration(std::size_t device_index);
+    bool IsVibrationEnabled(std::size_t device_index);
 
     /**
      * Sets the desired data to be polled from a controller

src/core/hle/kernel/hle_ipc.h

@@ -304,7 +304,7 @@ public:
      */
     template <typename T, typename = std::enable_if_t<!std::is_pointer_v<T>>>
     std::size_t WriteBuffer(const T& data, std::size_t buffer_index = 0) const {
-        if constexpr (Common::IsSTLContainer<T>) {
+        if constexpr (Common::IsContiguousContainer<T>) {
             using ContiguousType = typename T::value_type;
             static_assert(std::is_trivially_copyable_v<ContiguousType>,
                           "Container to WriteBuffer must contain trivially copyable objects");

src/core/hle/kernel/kernel.cpp

@@ -48,8 +48,8 @@ namespace Kernel {
 
 struct KernelCore::Impl {
     explicit Impl(Core::System& system_, KernelCore& kernel_)
-        : time_manager{system_},
-          service_threads_manager{1, "ServiceThreadsManager"}, system{system_} {}
+        : time_manager{system_}, service_threads_manager{1, "ServiceThreadsManager"},
+          service_thread_barrier{2}, system{system_} {}
 
     void SetMulticore(bool is_multi) {
         is_multicore = is_multi;
@@ -737,7 +737,12 @@ struct KernelCore::Impl {
     }
 
     void ClearServiceThreads() {
-        service_threads_manager.QueueWork([this]() { service_threads.clear(); });
+        service_threads_manager.QueueWork([this] {
+            service_threads.clear();
+            default_service_thread.reset();
+            service_thread_barrier.Sync();
+        });
+        service_thread_barrier.Sync();
     }
 
     std::mutex server_objects_lock;
@@ -802,6 +807,7 @@ struct KernelCore::Impl {
     std::unordered_set<std::shared_ptr<ServiceThread>> service_threads;
     std::weak_ptr<ServiceThread> default_service_thread;
     Common::ThreadWorker service_threads_manager;
+    Common::Barrier service_thread_barrier;
 
     std::array<KThread*, Core::Hardware::NUM_CPU_CORES> shutdown_threads;
     std::array<std::unique_ptr<Kernel::KScheduler>, Core::Hardware::NUM_CPU_CORES> schedulers{};

src/core/hle/service/audio/audin_u.cpp

@@ -17,7 +17,7 @@ using namespace AudioCore::AudioIn;
 class IAudioIn final : public ServiceFramework<IAudioIn> {
 public:
     explicit IAudioIn(Core::System& system_, Manager& manager, size_t session_id,
-                      std::string& device_name, const AudioInParameter& in_params, u32 handle,
+                      const std::string& device_name, const AudioInParameter& in_params, u32 handle,
                       u64 applet_resource_user_id)
         : ServiceFramework{system_, "IAudioIn"},
           service_context{system_, "IAudioIn"}, event{service_context.CreateEvent("AudioInEvent")},

src/core/hle/service/audio/audout_u.cpp

@@ -24,7 +24,7 @@ using namespace AudioCore::AudioOut;
 class IAudioOut final : public ServiceFramework<IAudioOut> {
 public:
     explicit IAudioOut(Core::System& system_, AudioCore::AudioOut::Manager& manager,
-                       size_t session_id, std::string& device_name,
+                       size_t session_id, const std::string& device_name,
                        const AudioOutParameter& in_params, u32 handle, u64 applet_resource_user_id)
         : ServiceFramework{system_, "IAudioOut", ServiceThreadType::CreateNew},
           service_context{system_, "IAudioOut"}, event{service_context.CreateEvent(

src/core/hle/service/hid/controllers/npad.cpp

@@ -868,7 +868,7 @@ bool Controller_NPad::VibrateControllerAtIndex(Core::HID::NpadIdType npad_id,
         return false;
     }
 
-    if (!controller.device->IsVibrationEnabled()) {
+    if (!controller.device->IsVibrationEnabled(device_index)) {
         if (controller.vibration[device_index].latest_vibration_value.low_amplitude != 0.0f ||
             controller.vibration[device_index].latest_vibration_value.high_amplitude != 0.0f) {
             // Send an empty vibration to stop any vibrations.
@@ -1001,7 +1001,7 @@ void Controller_NPad::InitializeVibrationDeviceAtIndex(Core::HID::NpadIdType npa
     }
 
     controller.vibration[device_index].device_mounted =
-        controller.device->TestVibration(device_index);
+        controller.device->IsVibrationEnabled(device_index);
 }
 
 void Controller_NPad::SetPermitVibrationSession(bool permit_vibration_session) {

src/core/hle/service/nfp/amiibo_crypto.cpp

@@ -9,6 +9,7 @@
 #include <mbedtls/hmac_drbg.h>
 
 #include "common/fs/file.h"
+#include "common/fs/fs.h"
 #include "common/fs/path_util.h"
 #include "common/logging/log.h"
 #include "core/hle/service/mii/mii_manager.h"
@@ -279,7 +280,7 @@ bool LoadKeys(InternalKey& locked_secret, InternalKey& unfixed_info) {
                                        Common::FS::FileType::BinaryFile};
 
     if (!keys_file.IsOpen()) {
-        LOG_ERROR(Service_NFP, "No keys detected");
+        LOG_ERROR(Service_NFP, "Failed to open key file");
         return false;
     }
 
@@ -295,6 +296,11 @@ bool LoadKeys(InternalKey& locked_secret, InternalKey& unfixed_info) {
     return true;
 }
 
+bool IsKeyAvailable() {
+    const auto yuzu_keys_dir = Common::FS::GetYuzuPath(Common::FS::YuzuPath::KeysDir);
+    return Common::FS::Exists(yuzu_keys_dir / "key_retail.bin");
+}
+
 bool DecodeAmiibo(const EncryptedNTAG215File& encrypted_tag_data, NTAG215File& tag_data) {
     InternalKey locked_secret{};
     InternalKey unfixed_info{};

src/core/hle/service/nfp/amiibo_crypto.h

@@ -91,6 +91,9 @@ void Cipher(const DerivedKeys& keys, const NTAG215File& in_data, NTAG215File& ou
 /// Loads both amiibo keys from key_retail.bin
 bool LoadKeys(InternalKey& locked_secret, InternalKey& unfixed_info);
 
+/// Returns true if key_retail.bin exist
+bool IsKeyAvailable();
+
 /// Decodes encripted amiibo data returns true if output is valid
 bool DecodeAmiibo(const EncryptedNTAG215File& encrypted_tag_data, NTAG215File& tag_data);
 

src/core/hle/service/nfp/nfp_device.cpp

@@ -17,6 +17,7 @@
 #include "core/hle/ipc_helpers.h"
 #include "core/hle/kernel/k_event.h"
 #include "core/hle/service/mii/mii_manager.h"
+#include "core/hle/service/mii/types.h"
 #include "core/hle/service/nfp/amiibo_crypto.h"
 #include "core/hle/service/nfp/nfp.h"
 #include "core/hle/service/nfp/nfp_device.h"
@@ -233,6 +234,14 @@ Result NfpDevice::Mount(MountTarget mount_target_) {
         return NotAnAmiibo;
     }
 
+    // Mark amiibos as read only when keys are missing
+    if (!AmiiboCrypto::IsKeyAvailable()) {
+        LOG_ERROR(Service_NFP, "No keys detected");
+        device_state = DeviceState::TagMounted;
+        mount_target = MountTarget::Rom;
+        return ResultSuccess;
+    }
+
     if (!AmiiboCrypto::DecodeAmiibo(encrypted_tag_data, tag_data)) {
         LOG_ERROR(Service_NFP, "Can't decode amiibo {}", device_state);
         return CorruptedData;

src/core/hle/service/nfp/nfp_device.h

@@ -8,7 +8,6 @@
 
 #include "common/common_funcs.h"
 #include "core/hle/service/kernel_helpers.h"
-#include "core/hle/service/mii/types.h"
 #include "core/hle/service/nfp/nfp_types.h"
 #include "core/hle/service/service.h"
 

src/core/hle/service/nfp/nfp_types.h

@@ -17,11 +17,6 @@ enum class ServiceType : u32 {
     System,
 };
 
-enum class State : u32 {
-    NonInitialized,
-    Initialized,
-};
-
 enum class DeviceState : u32 {
     Initialized,
     SearchingForTag,

src/core/hle/service/nfp/nfp_user.cpp

@@ -6,12 +6,9 @@
 
 #include "common/logging/log.h"
 #include "core/core.h"
-#include "core/hid/emulated_controller.h"
-#include "core/hid/hid_core.h"
 #include "core/hid/hid_types.h"
 #include "core/hle/ipc_helpers.h"
 #include "core/hle/kernel/k_event.h"
-#include "core/hle/service/mii/mii_manager.h"
 #include "core/hle/service/nfp/nfp_device.h"
 #include "core/hle/service/nfp/nfp_result.h"
 #include "core/hle/service/nfp/nfp_user.h"

src/core/hle/service/nfp/nfp_user.h

@@ -4,8 +4,7 @@
 #pragma once
 
 #include "core/hle/service/kernel_helpers.h"
-#include "core/hle/service/nfp/nfp.h"
-#include "core/hle/service/nfp/nfp_types.h"
+#include "core/hle/service/service.h"
 
 namespace Service::NFP {
 class NfpDevice;
@@ -15,6 +14,11 @@ public:
     explicit IUser(Core::System& system_);
 
 private:
+    enum class State : u32 {
+        NonInitialized,
+        Initialized,
+    };
+
     void Initialize(Kernel::HLERequestContext& ctx);
     void Finalize(Kernel::HLERequestContext& ctx);
     void ListDevices(Kernel::HLERequestContext& ctx);

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

@@ -102,15 +102,19 @@ NVFlinger::~NVFlinger() {
         system.CoreTiming().UnscheduleEvent(single_composition_event, {});
     }
 
+    ShutdownLayers();
+
+    if (nvdrv) {
+        nvdrv->Close(disp_fd);
+    }
+}
+
+void NVFlinger::ShutdownLayers() {
     for (auto& display : displays) {
         for (size_t layer = 0; layer < display.GetNumLayers(); ++layer) {
             display.GetLayer(layer).Core().NotifyShutdown();
         }
     }
-
-    if (nvdrv) {
-        nvdrv->Close(disp_fd);
-    }
 }
 
 void NVFlinger::SetNVDrvInstance(std::shared_ptr<Nvidia::Module> instance) {

src/core/hle/service/nvflinger/nvflinger.h

@@ -48,6 +48,8 @@ public:
     explicit NVFlinger(Core::System& system_, HosBinderDriverServer& hos_binder_driver_server_);
     ~NVFlinger();
 
+    void ShutdownLayers();
+
     /// Sets the NVDrv module instance to use to send buffers to the GPU.
     void SetNVDrvInstance(std::shared_ptr<Nvidia::Module> instance);
 

src/core/hle/service/service.cpp

@@ -303,4 +303,8 @@ Services::Services(std::shared_ptr<SM::ServiceManager>& sm, Core::System& system
 
 Services::~Services() = default;
 
+void Services::KillNVNFlinger() {
+    nv_flinger->ShutdownLayers();
+}
+
 } // namespace Service

src/core/hle/service/service.h

@@ -238,6 +238,8 @@ public:
     explicit Services(std::shared_ptr<SM::ServiceManager>& sm, Core::System& system);
     ~Services();
 
+    void KillNVNFlinger();
+
 private:
     std::unique_ptr<NVFlinger::HosBinderDriverServer> hos_binder_driver_server;
     std::unique_ptr<NVFlinger::NVFlinger> nv_flinger;

src/input_common/drivers/gc_adapter.cpp

@@ -324,7 +324,7 @@ bool GCAdapter::GetGCEndpoint(libusb_device* device) {
     return true;
 }
 
-Common::Input::VibrationError GCAdapter::SetRumble(
+Common::Input::VibrationError GCAdapter::SetVibration(
     const PadIdentifier& identifier, const Common::Input::VibrationStatus& vibration) {
     const auto mean_amplitude = (vibration.low_amplitude + vibration.high_amplitude) * 0.5f;
     const auto processed_amplitude =
@@ -338,6 +338,10 @@ Common::Input::VibrationError GCAdapter::SetRumble(
     return Common::Input::VibrationError::None;
 }
 
+bool GCAdapter::IsVibrationEnabled([[maybe_unused]] const PadIdentifier& identifier) {
+    return rumble_enabled;
+}
+
 void GCAdapter::UpdateVibrations() {
     // Use 8 states to keep the switching between on/off fast enough for
     // a human to feel different vibration strenght

src/input_common/drivers/gc_adapter.h

@@ -25,9 +25,11 @@ public:
     explicit GCAdapter(std::string input_engine_);
     ~GCAdapter() override;
 
-    Common::Input::VibrationError SetRumble(
+    Common::Input::VibrationError SetVibration(
         const PadIdentifier& identifier, const Common::Input::VibrationStatus& vibration) override;
 
+    bool IsVibrationEnabled(const PadIdentifier& identifier) override;
+
     /// Used for automapping features
     std::vector<Common::ParamPackage> GetInputDevices() const override;
     ButtonMapping GetButtonMappingForDevice(const Common::ParamPackage& params) override;

src/input_common/drivers/sdl_driver.cpp

@@ -114,6 +114,20 @@ public:
         }
         return false;
     }
+
+    void EnableVibration(bool is_enabled) {
+        has_vibration = is_enabled;
+        is_vibration_tested = true;
+    }
+
+    bool HasVibration() const {
+        return has_vibration;
+    }
+
+    bool IsVibrationTested() const {
+        return is_vibration_tested;
+    }
+
     /**
      * The Pad identifier of the joystick
      */
@@ -236,6 +250,8 @@ private:
     u64 last_motion_update{};
     bool has_gyro{false};
     bool has_accel{false};
+    bool has_vibration{false};
+    bool is_vibration_tested{false};
     BasicMotion motion;
 };
 
@@ -517,7 +533,7 @@ std::vector<Common::ParamPackage> SDLDriver::GetInputDevices() const {
     return devices;
 }
 
-Common::Input::VibrationError SDLDriver::SetRumble(
+Common::Input::VibrationError SDLDriver::SetVibration(
     const PadIdentifier& identifier, const Common::Input::VibrationStatus& vibration) {
     const auto joystick =
         GetSDLJoystickByGUID(identifier.guid.RawString(), static_cast<int>(identifier.port));
@@ -546,13 +562,6 @@ Common::Input::VibrationError SDLDriver::SetRumble(
         .type = Common::Input::VibrationAmplificationType::Exponential,
     };
 
-    if (vibration.type == Common::Input::VibrationAmplificationType::Test) {
-        if (!joystick->RumblePlay(new_vibration)) {
-            return Common::Input::VibrationError::Unknown;
-        }
-        return Common::Input::VibrationError::None;
-    }
-
     vibration_queue.Push(VibrationRequest{
         .identifier = identifier,
         .vibration = new_vibration,
@@ -561,6 +570,45 @@ Common::Input::VibrationError SDLDriver::SetRumble(
     return Common::Input::VibrationError::None;
 }
 
+bool SDLDriver::IsVibrationEnabled(const PadIdentifier& identifier) {
+    const auto joystick =
+        GetSDLJoystickByGUID(identifier.guid.RawString(), static_cast<int>(identifier.port));
+
+    constexpr Common::Input::VibrationStatus test_vibration{
+        .low_amplitude = 1,
+        .low_frequency = 160.0f,
+        .high_amplitude = 1,
+        .high_frequency = 320.0f,
+        .type = Common::Input::VibrationAmplificationType::Exponential,
+    };
+
+    constexpr Common::Input::VibrationStatus zero_vibration{
+        .low_amplitude = 0,
+        .low_frequency = 160.0f,
+        .high_amplitude = 0,
+        .high_frequency = 320.0f,
+        .type = Common::Input::VibrationAmplificationType::Exponential,
+    };
+
+    if (joystick->IsVibrationTested()) {
+        return joystick->HasVibration();
+    }
+
+    // First vibration might fail
+    joystick->RumblePlay(test_vibration);
+
+    // Wait for about 15ms to ensure the controller is ready for the stop command
+    std::this_thread::sleep_for(std::chrono::milliseconds(15));
+
+    if (!joystick->RumblePlay(zero_vibration)) {
+        joystick->EnableVibration(false);
+        return false;
+    }
+
+    joystick->EnableVibration(true);
+    return true;
+}
+
 void SDLDriver::SendVibrations() {
     while (!vibration_queue.Empty()) {
         VibrationRequest request;

src/input_common/drivers/sdl_driver.h

@@ -61,9 +61,11 @@ public:
 
     bool IsStickInverted(const Common::ParamPackage& params) override;
 
-    Common::Input::VibrationError SetRumble(
+    Common::Input::VibrationError SetVibration(
         const PadIdentifier& identifier, const Common::Input::VibrationStatus& vibration) override;
 
+    bool IsVibrationEnabled(const PadIdentifier& identifier) override;
+
 private:
     struct VibrationRequest {
         PadIdentifier identifier;

src/input_common/input_engine.h

@@ -108,12 +108,17 @@ public:
                          [[maybe_unused]] const Common::Input::LedStatus& led_status) {}
 
     // Sets rumble to a controller
-    virtual Common::Input::VibrationError SetRumble(
+    virtual Common::Input::VibrationError SetVibration(
         [[maybe_unused]] const PadIdentifier& identifier,
         [[maybe_unused]] const Common::Input::VibrationStatus& vibration) {
         return Common::Input::VibrationError::NotSupported;
     }
 
+    // Returns true if device supports vibrations
+    virtual bool IsVibrationEnabled([[maybe_unused]] const PadIdentifier& identifier) {
+        return false;
+    }
+
     // Sets polling mode to a controller
     virtual Common::Input::PollingError SetPollingMode(
         [[maybe_unused]] const PadIdentifier& identifier,

src/input_common/input_poller.cpp

@@ -763,7 +763,11 @@ public:
 
     Common::Input::VibrationError SetVibration(
         const Common::Input::VibrationStatus& vibration_status) override {
-        return input_engine->SetRumble(identifier, vibration_status);
+        return input_engine->SetVibration(identifier, vibration_status);
+    }
+
+    bool IsVibrationEnabled() override {
+        return input_engine->IsVibrationEnabled(identifier);
     }
 
     Common::Input::PollingError SetPollingMode(Common::Input::PollingMode polling_mode) override {

src/shader_recompiler/frontend/ir/microinstruction.cpp

@@ -325,6 +325,11 @@ void Inst::AddPhiOperand(Block* predecessor, const Value& value) {
     phi_args.emplace_back(predecessor, value);
 }
 
+void Inst::ErasePhiOperand(size_t index) {
+    const auto operand_it{phi_args.begin() + static_cast<ptrdiff_t>(index)};
+    phi_args.erase(operand_it);
+}
+
 void Inst::OrderPhiArgs() {
     if (op != Opcode::Phi) {
         throw LogicError("{} is not a Phi instruction", op);

src/shader_recompiler/frontend/ir/value.h

@@ -178,9 +178,13 @@ public:
 
     /// Get a pointer to the block of a phi argument.
     [[nodiscard]] Block* PhiBlock(size_t index) const;
+
     /// Add phi operand to a phi instruction.
     void AddPhiOperand(Block* predecessor, const Value& value);
 
+    // Erase the phi operand at the given index.
+    void ErasePhiOperand(size_t index);
+
     /// Orders the Phi arguments from farthest away to nearest.
     void OrderPhiArgs();
 

src/shader_recompiler/ir_opt/dead_code_elimination_pass.cpp

@@ -1,25 +1,105 @@
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <algorithm>
+
+#include <boost/container/small_vector.hpp>
+
 #include "shader_recompiler/frontend/ir/basic_block.h"
 #include "shader_recompiler/frontend/ir/value.h"
 #include "shader_recompiler/ir_opt/passes.h"
 
 namespace Shader::Optimization {
-
-void DeadCodeEliminationPass(IR::Program& program) {
+namespace {
+template <bool TEST_USES>
+void DeadInstElimination(IR::Block* const block) {
     // We iterate over the instructions in reverse order.
     // This is because removing an instruction reduces the number of uses for earlier instructions.
-    for (IR::Block* const block : program.post_order_blocks) {
-        auto it{block->end()};
-        while (it != block->begin()) {
-            --it;
-            if (!it->HasUses() && !it->MayHaveSideEffects()) {
-                it->Invalidate();
-                it = block->Instructions().erase(it);
+    auto it{block->end()};
+    while (it != block->begin()) {
+        --it;
+        if constexpr (TEST_USES) {
+            if (it->HasUses() || it->MayHaveSideEffects()) {
+                continue;
+            }
+        }
+        it->Invalidate();
+        it = block->Instructions().erase(it);
+    }
+}
+
+void DeletedPhiArgElimination(IR::Program& program, std::span<const IR::Block*> dead_blocks) {
+    for (IR::Block* const block : program.blocks) {
+        for (IR::Inst& phi : *block) {
+            if (!IR::IsPhi(phi)) {
+                continue;
+            }
+            for (size_t i = 0; i < phi.NumArgs(); ++i) {
+                if (std::ranges::find(dead_blocks, phi.PhiBlock(i)) == dead_blocks.end()) {
+                    continue;
+                }
+                // Phi operand at this index is an unreachable block
+                phi.ErasePhiOperand(i);
+                --i;
             }
         }
     }
 }
 
+void DeadBranchElimination(IR::Program& program) {
+    boost::container::small_vector<const IR::Block*, 3> dead_blocks;
+    const auto begin_it{program.syntax_list.begin()};
+    for (auto node_it = begin_it; node_it != program.syntax_list.end(); ++node_it) {
+        if (node_it->type != IR::AbstractSyntaxNode::Type::If) {
+            continue;
+        }
+        IR::Inst* const cond_ref{node_it->data.if_node.cond.Inst()};
+        const IR::U1 cond{cond_ref->Arg(0)};
+        if (!cond.IsImmediate()) {
+            continue;
+        }
+        if (cond.U1()) {
+            continue;
+        }
+        // False immediate condition. Remove condition ref, erase the entire branch.
+        cond_ref->Invalidate();
+        // Account for nested if-statements within the if(false) branch
+        u32 nested_ifs{1u};
+        while (node_it->type != IR::AbstractSyntaxNode::Type::EndIf || nested_ifs > 0) {
+            node_it = program.syntax_list.erase(node_it);
+            switch (node_it->type) {
+            case IR::AbstractSyntaxNode::Type::If:
+                ++nested_ifs;
+                break;
+            case IR::AbstractSyntaxNode::Type::EndIf:
+                --nested_ifs;
+                break;
+            case IR::AbstractSyntaxNode::Type::Block: {
+                IR::Block* const block{node_it->data.block};
+                DeadInstElimination<false>(block);
+                dead_blocks.push_back(block);
+                break;
+            }
+            default:
+                break;
+            }
+        }
+        // Erase EndIf node of the if(false) branch
+        node_it = program.syntax_list.erase(node_it);
+        // Account for loop increment
+        --node_it;
+    }
+    if (!dead_blocks.empty()) {
+        DeletedPhiArgElimination(program, std::span(dead_blocks.data(), dead_blocks.size()));
+    }
+}
+} // namespace
+
+void DeadCodeEliminationPass(IR::Program& program) {
+    DeadBranchElimination(program);
+    for (IR::Block* const block : program.post_order_blocks) {
+        DeadInstElimination<true>(block);
+    }
+}
+
 } // namespace Shader::Optimization

src/tests/video_core/buffer_base.cpp

@@ -44,7 +44,7 @@ public:
 
     [[nodiscard]] unsigned Count() const noexcept {
         unsigned count = 0;
-        for (const auto [index, value] : page_table) {
+        for (const auto& [index, value] : page_table) {
             count += value;
         }
         return count;

src/video_core/engines/maxwell_3d.cpp

@@ -117,10 +117,18 @@ void Maxwell3D::InitializeRegisterDefaults() {
 
     shadow_state = regs;
 
-    mme_inline[MAXWELL3D_REG_INDEX(draw.end)] = true;
-    mme_inline[MAXWELL3D_REG_INDEX(draw.begin)] = true;
-    mme_inline[MAXWELL3D_REG_INDEX(vertex_buffer.count)] = true;
-    mme_inline[MAXWELL3D_REG_INDEX(index_buffer.count)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(draw.end)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(draw.begin)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(vertex_buffer.first)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(vertex_buffer.count)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(index_buffer.first)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(index_buffer.count)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(index_buffer32_first)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(index_buffer16_first)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(index_buffer8_first)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(draw_inline_index)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(inline_index_2x16.even)] = true;
+    draw_command[MAXWELL3D_REG_INDEX(inline_index_4x8.index0)] = true;
 }
 
 void Maxwell3D::ProcessMacro(u32 method, const u32* base_start, u32 amount, bool is_last_call) {
@@ -208,25 +216,6 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
         return ProcessCBBind(3);
     case MAXWELL3D_REG_INDEX(bind_groups[4].raw_config):
         return ProcessCBBind(4);
-    case MAXWELL3D_REG_INDEX(draw.end):
-        return DrawArrays();
-    case MAXWELL3D_REG_INDEX(index_buffer32_first):
-        regs.index_buffer.count = regs.index_buffer32_first.count;
-        regs.index_buffer.first = regs.index_buffer32_first.first;
-        dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
-        return DrawArrays();
-    case MAXWELL3D_REG_INDEX(index_buffer16_first):
-        regs.index_buffer.count = regs.index_buffer16_first.count;
-        regs.index_buffer.first = regs.index_buffer16_first.first;
-        dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
-        return DrawArrays();
-    case MAXWELL3D_REG_INDEX(index_buffer8_first):
-        regs.index_buffer.count = regs.index_buffer8_first.count;
-        regs.index_buffer.first = regs.index_buffer8_first.first;
-        dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
-        // a macro calls this one over and over, should it increase instancing?
-        // Used by Hades and likely other Vulkan games.
-        return DrawArrays();
     case MAXWELL3D_REG_INDEX(topology_override):
         use_topology_override = true;
         return;
@@ -261,14 +250,13 @@ void Maxwell3D::CallMacroMethod(u32 method, const std::vector<u32>& parameters)
 
     // Execute the current macro.
     macro_engine->Execute(macro_positions[entry], parameters);
-    if (mme_draw.current_mode != MMEDrawMode::Undefined) {
-        FlushMMEInlineDraw();
-    }
+
+    ProcessDeferredDraw();
 }
 
 void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
-    // It is an error to write to a register other than the current macro's ARG register before it
-    // has finished execution.
+    // It is an error to write to a register other than the current macro's ARG register before
+    // it has finished execution.
     if (executing_macro != 0) {
         ASSERT(method == executing_macro + 1);
     }
@@ -283,9 +271,33 @@ void Maxwell3D::CallMethod(u32 method, u32 method_argument, bool is_last_call) {
     ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid Maxwell3D register, increase the size of the Regs structure");
 
-    const u32 argument = ProcessShadowRam(method, method_argument);
-    ProcessDirtyRegisters(method, argument);
-    ProcessMethodCall(method, argument, method_argument, is_last_call);
+    if (draw_command[method]) {
+        regs.reg_array[method] = method_argument;
+        deferred_draw_method.push_back(method);
+        auto u32_to_u8 = [&](const u32 argument) {
+            inline_index_draw_indexes.push_back(static_cast<u8>(argument & 0x000000ff));
+            inline_index_draw_indexes.push_back(static_cast<u8>((argument & 0x0000ff00) >> 8));
+            inline_index_draw_indexes.push_back(static_cast<u8>((argument & 0x00ff0000) >> 16));
+            inline_index_draw_indexes.push_back(static_cast<u8>((argument & 0xff000000) >> 24));
+        };
+        if (MAXWELL3D_REG_INDEX(draw_inline_index) == method) {
+            u32_to_u8(method_argument);
+        } else if (MAXWELL3D_REG_INDEX(inline_index_2x16.even) == method) {
+            u32_to_u8(regs.inline_index_2x16.even);
+            u32_to_u8(regs.inline_index_2x16.odd);
+        } else if (MAXWELL3D_REG_INDEX(inline_index_4x8.index0) == method) {
+            u32_to_u8(regs.inline_index_4x8.index0);
+            u32_to_u8(regs.inline_index_4x8.index1);
+            u32_to_u8(regs.inline_index_4x8.index2);
+            u32_to_u8(regs.inline_index_4x8.index3);
+        }
+    } else {
+        ProcessDeferredDraw();
+
+        const u32 argument = ProcessShadowRam(method, method_argument);
+        ProcessDirtyRegisters(method, argument);
+        ProcessMethodCall(method, argument, method_argument, is_last_call);
+    }
 }
 
 void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
@@ -326,55 +338,6 @@ void Maxwell3D::CallMultiMethod(u32 method, const u32* base_start, u32 amount,
     }
 }
 
-void Maxwell3D::StepInstance(const MMEDrawMode expected_mode, const u32 count) {
-    if (mme_draw.current_mode == MMEDrawMode::Undefined) {
-        if (mme_draw.gl_begin_consume) {
-            mme_draw.current_mode = expected_mode;
-            mme_draw.current_count = count;
-            mme_draw.instance_count = 1;
-            mme_draw.gl_begin_consume = false;
-            mme_draw.gl_end_count = 0;
-        }
-        return;
-    } else {
-        if (mme_draw.current_mode == expected_mode && count == mme_draw.current_count &&
-            mme_draw.instance_mode && mme_draw.gl_begin_consume) {
-            mme_draw.instance_count++;
-            mme_draw.gl_begin_consume = false;
-            return;
-        } else {
-            FlushMMEInlineDraw();
-        }
-    }
-    // Tail call in case it needs to retry.
-    StepInstance(expected_mode, count);
-}
-
-void Maxwell3D::CallMethodFromMME(u32 method, u32 method_argument) {
-    if (mme_inline[method]) {
-        regs.reg_array[method] = method_argument;
-        if (method == MAXWELL3D_REG_INDEX(vertex_buffer.count) ||
-            method == MAXWELL3D_REG_INDEX(index_buffer.count)) {
-            const MMEDrawMode expected_mode = method == MAXWELL3D_REG_INDEX(vertex_buffer.count)
-                                                  ? MMEDrawMode::Array
-                                                  : MMEDrawMode::Indexed;
-            StepInstance(expected_mode, method_argument);
-        } else if (method == MAXWELL3D_REG_INDEX(draw.begin)) {
-            mme_draw.instance_mode =
-                (regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent) ||
-                (regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Unchanged);
-            mme_draw.gl_begin_consume = true;
-        } else {
-            mme_draw.gl_end_count++;
-        }
-    } else {
-        if (mme_draw.current_mode != MMEDrawMode::Undefined) {
-            FlushMMEInlineDraw();
-        }
-        CallMethod(method, method_argument, true);
-    }
-}
-
 void Maxwell3D::ProcessTopologyOverride() {
     using PrimitiveTopology = Maxwell3D::Regs::PrimitiveTopology;
     using PrimitiveTopologyOverride = Maxwell3D::Regs::PrimitiveTopologyOverride;
@@ -404,41 +367,6 @@ void Maxwell3D::ProcessTopologyOverride() {
     }
 }
 
-void Maxwell3D::FlushMMEInlineDraw() {
-    LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(),
-              regs.vertex_buffer.count);
-    ASSERT_MSG(!(regs.index_buffer.count && regs.vertex_buffer.count), "Both indexed and direct?");
-    ASSERT(mme_draw.instance_count == mme_draw.gl_end_count);
-
-    // Both instance configuration registers can not be set at the same time.
-    ASSERT_MSG(regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::First ||
-                   regs.draw.instance_id != Maxwell3D::Regs::Draw::InstanceId::Unchanged,
-               "Illegal combination of instancing parameters");
-
-    ProcessTopologyOverride();
-
-    const bool is_indexed = mme_draw.current_mode == MMEDrawMode::Indexed;
-    if (ShouldExecute()) {
-        rasterizer->Draw(is_indexed, true);
-    }
-
-    // TODO(bunnei): Below, we reset vertex count so that we can use these registers to determine if
-    // the game is trying to draw indexed or direct mode. This needs to be verified on HW still -
-    // it's possible that it is incorrect and that there is some other register used to specify the
-    // drawing mode.
-    if (is_indexed) {
-        regs.index_buffer.count = 0;
-    } else {
-        regs.vertex_buffer.count = 0;
-    }
-    mme_draw.current_mode = MMEDrawMode::Undefined;
-    mme_draw.current_count = 0;
-    mme_draw.instance_count = 0;
-    mme_draw.instance_mode = false;
-    mme_draw.gl_begin_consume = false;
-    mme_draw.gl_end_count = 0;
-}
-
 void Maxwell3D::ProcessMacroUpload(u32 data) {
     macro_engine->AddCode(regs.load_mme.instruction_ptr++, data);
 }
@@ -573,42 +501,6 @@ void Maxwell3D::ProcessSyncPoint() {
     rasterizer->SignalSyncPoint(sync_point);
 }
 
-void Maxwell3D::DrawArrays() {
-    LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(),
-              regs.vertex_buffer.count);
-    ASSERT_MSG(!(regs.index_buffer.count && regs.vertex_buffer.count), "Both indexed and direct?");
-
-    // Both instance configuration registers can not be set at the same time.
-    ASSERT_MSG(regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::First ||
-                   regs.draw.instance_id != Maxwell3D::Regs::Draw::InstanceId::Unchanged,
-               "Illegal combination of instancing parameters");
-
-    ProcessTopologyOverride();
-
-    if (regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent) {
-        // Increment the current instance *before* drawing.
-        state.current_instance++;
-    } else if (regs.draw.instance_id != Maxwell3D::Regs::Draw::InstanceId::Unchanged) {
-        // Reset the current instance to 0.
-        state.current_instance = 0;
-    }
-
-    const bool is_indexed{regs.index_buffer.count && !regs.vertex_buffer.count};
-    if (ShouldExecute()) {
-        rasterizer->Draw(is_indexed, false);
-    }
-
-    // TODO(bunnei): Below, we reset vertex count so that we can use these registers to determine if
-    // the game is trying to draw indexed or direct mode. This needs to be verified on HW still -
-    // it's possible that it is incorrect and that there is some other register used to specify the
-    // drawing mode.
-    if (is_indexed) {
-        regs.index_buffer.count = 0;
-    } else {
-        regs.vertex_buffer.count = 0;
-    }
-}
-
 std::optional<u64> Maxwell3D::GetQueryResult() {
     switch (regs.report_semaphore.query.report) {
     case Regs::ReportSemaphore::Report::Payload:
@@ -691,4 +583,95 @@ void Maxwell3D::ProcessClearBuffers() {
     rasterizer->Clear();
 }
 
+void Maxwell3D::ProcessDeferredDraw() {
+    if (deferred_draw_method.empty()) {
+        return;
+    }
+
+    enum class DrawMode {
+        Undefined,
+        General,
+        Instance,
+    };
+    DrawMode draw_mode{DrawMode::Undefined};
+    u32 instance_count = 1;
+
+    auto first_method = deferred_draw_method[0];
+    if (MAXWELL3D_REG_INDEX(draw.begin) == first_method) {
+        // The minimum number of methods for drawing must be greater than or equal to
+        // 3[draw.begin->vertex(index)count->draw.end] to avoid errors in index mode drawing
+        if (deferred_draw_method.size() < 3) {
+            return;
+        }
+        draw_mode = (regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Subsequent) ||
+                            (regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::Unchanged)
+                        ? DrawMode::Instance
+                        : DrawMode::General;
+    } else if (MAXWELL3D_REG_INDEX(index_buffer32_first) == first_method ||
+               MAXWELL3D_REG_INDEX(index_buffer16_first) == first_method ||
+               MAXWELL3D_REG_INDEX(index_buffer8_first) == first_method) {
+        draw_mode = DrawMode::General;
+    }
+
+    // Drawing will only begin with draw.begin or index_buffer method, other methods directly
+    // clear
+    if (draw_mode == DrawMode::Undefined) {
+        deferred_draw_method.clear();
+        return;
+    }
+
+    if (draw_mode == DrawMode::Instance) {
+        ASSERT_MSG(deferred_draw_method.size() % 4 == 0, "Instance mode method size error");
+        instance_count = static_cast<u32>(deferred_draw_method.size()) / 4;
+    } else {
+        if (MAXWELL3D_REG_INDEX(index_buffer32_first) == first_method) {
+            regs.index_buffer.count = regs.index_buffer32_first.count;
+            regs.index_buffer.first = regs.index_buffer32_first.first;
+            dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+        } else if (MAXWELL3D_REG_INDEX(index_buffer32_first) == first_method) {
+            regs.index_buffer.count = regs.index_buffer16_first.count;
+            regs.index_buffer.first = regs.index_buffer16_first.first;
+            dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+        } else if (MAXWELL3D_REG_INDEX(index_buffer32_first) == first_method) {
+            regs.index_buffer.count = regs.index_buffer8_first.count;
+            regs.index_buffer.first = regs.index_buffer8_first.first;
+            dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
+        } else {
+            auto second_method = deferred_draw_method[1];
+            if (MAXWELL3D_REG_INDEX(draw_inline_index) == second_method ||
+                MAXWELL3D_REG_INDEX(inline_index_2x16.even) == second_method ||
+                MAXWELL3D_REG_INDEX(inline_index_4x8.index0) == second_method) {
+                regs.index_buffer.count = static_cast<u32>(inline_index_draw_indexes.size() / 4);
+                regs.index_buffer.format = Regs::IndexFormat::UnsignedInt;
+            }
+        }
+    }
+
+    LOG_TRACE(HW_GPU, "called, topology={}, count={}", regs.draw.topology.Value(),
+              regs.vertex_buffer.count);
+
+    ASSERT_MSG(!(regs.index_buffer.count && regs.vertex_buffer.count), "Both indexed and direct?");
+
+    // Both instance configuration registers can not be set at the same time.
+    ASSERT_MSG(regs.draw.instance_id == Maxwell3D::Regs::Draw::InstanceId::First ||
+                   regs.draw.instance_id != Maxwell3D::Regs::Draw::InstanceId::Unchanged,
+               "Illegal combination of instancing parameters");
+
+    ProcessTopologyOverride();
+
+    const bool is_indexed = regs.index_buffer.count && !regs.vertex_buffer.count;
+    if (ShouldExecute()) {
+        rasterizer->Draw(is_indexed, instance_count);
+    }
+
+    if (is_indexed) {
+        regs.index_buffer.count = 0;
+    } else {
+        regs.vertex_buffer.count = 0;
+    }
+
+    deferred_draw_method.clear();
+    inline_index_draw_indexes.clear();
+}
+
 } // namespace Tegra::Engines

src/video_core/engines/maxwell_3d.h

@@ -1739,14 +1739,11 @@ public:
             Footprint_1x1_Virtual = 2,
         };
 
-        struct InlineIndex4x8Align {
+        struct InlineIndex4x8 {
             union {
                 BitField<0, 30, u32> count;
                 BitField<30, 2, u32> start;
             };
-        };
-
-        struct InlineIndex4x8Index {
             union {
                 BitField<0, 8, u32> index0;
                 BitField<8, 8, u32> index1;
@@ -2836,8 +2833,7 @@ public:
                 u32 depth_write_enabled;                                               ///< 0x12E8
                 u32 alpha_test_enabled;                                                ///< 0x12EC
                 INSERT_PADDING_BYTES_NOINIT(0x10);
-                InlineIndex4x8Align inline_index_4x8_align;                            ///< 0x1300
-                InlineIndex4x8Index inline_index_4x8_index;                            ///< 0x1304
+                InlineIndex4x8 inline_index_4x8;                                       ///< 0x1300
                 D3DCullMode d3d_cull_mode;                                             ///< 0x1308
                 ComparisonOp depth_test_func;                                          ///< 0x130C
                 f32 alpha_test_ref;                                                    ///< 0x1310
@@ -3048,8 +3044,6 @@ public:
         };
 
         std::array<ShaderStageInfo, Regs::MaxShaderStage> shader_stages;
-
-        u32 current_instance = 0; ///< Current instance to be used to simulate instanced rendering.
     };
 
     State state{};
@@ -3064,11 +3058,6 @@ public:
     void CallMultiMethod(u32 method, const u32* base_start, u32 amount,
                          u32 methods_pending) override;
 
-    /// Write the value to the register identified by method.
-    void CallMethodFromMME(u32 method, u32 method_argument);
-
-    void FlushMMEInlineDraw();
-
     bool ShouldExecute() const {
         return execute_on;
     }
@@ -3081,21 +3070,6 @@ public:
         return *rasterizer;
     }
 
-    enum class MMEDrawMode : u32 {
-        Undefined,
-        Array,
-        Indexed,
-    };
-
-    struct MMEDrawState {
-        MMEDrawMode current_mode{MMEDrawMode::Undefined};
-        u32 current_count{};
-        u32 instance_count{};
-        bool instance_mode{};
-        bool gl_begin_consume{};
-        u32 gl_end_count{};
-    } mme_draw;
-
     struct DirtyState {
         using Flags = std::bitset<std::numeric_limits<u8>::max()>;
         using Table = std::array<u8, Regs::NUM_REGS>;
@@ -3105,6 +3079,8 @@ public:
         Tables tables{};
     } dirty;
 
+    std::vector<u8> inline_index_draw_indexes;
+
 private:
     void InitializeRegisterDefaults();
 
@@ -3164,14 +3140,10 @@ private:
     /// Handles a write to the CB_BIND register.
     void ProcessCBBind(size_t stage_index);
 
-    /// Handles a write to the VERTEX_END_GL register, triggering a draw.
-    void DrawArrays();
-
     /// Handles use of topology overrides (e.g., to avoid using a topology assigned from a macro)
     void ProcessTopologyOverride();
 
-    // Handles a instance drawcall from MME
-    void StepInstance(MMEDrawMode expected_mode, u32 count);
+    void ProcessDeferredDraw();
 
     /// Returns a query's value or an empty object if the value will be deferred through a cache.
     std::optional<u64> GetQueryResult();
@@ -3184,8 +3156,6 @@ private:
     /// Start offsets of each macro in macro_memory
     std::array<u32, 0x80> macro_positions{};
 
-    std::array<bool, Regs::NUM_REGS> mme_inline{};
-
     /// Macro method that is currently being executed / being fed parameters.
     u32 executing_macro = 0;
     /// Parameters that have been submitted to the macro call so far.
@@ -3198,6 +3168,9 @@ private:
 
     bool execute_on{true};
     bool use_topology_override{false};
+
+    std::array<bool, Regs::NUM_REGS> draw_command{};
+    std::vector<u32> deferred_draw_method;
 };
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \
@@ -3402,8 +3375,7 @@ ASSERT_REG_POSITION(alpha_to_coverage_dither, 0x12E0);
 ASSERT_REG_POSITION(blend_per_target_enabled, 0x12E4);
 ASSERT_REG_POSITION(depth_write_enabled, 0x12E8);
 ASSERT_REG_POSITION(alpha_test_enabled, 0x12EC);
-ASSERT_REG_POSITION(inline_index_4x8_align, 0x1300);
-ASSERT_REG_POSITION(inline_index_4x8_index, 0x1304);
+ASSERT_REG_POSITION(inline_index_4x8, 0x1300);
 ASSERT_REG_POSITION(d3d_cull_mode, 0x1308);
 ASSERT_REG_POSITION(depth_test_func, 0x130C);
 ASSERT_REG_POSITION(alpha_test_ref, 0x1310);

src/video_core/macro/macro_hle.cpp

@@ -22,35 +22,29 @@ void HLE_771BB18C62444DA0(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.regs.draw.topology.Assign(
         static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0] & 0x3ffffff));
     maxwell3d.regs.global_base_instance_index = parameters[5];
-    maxwell3d.mme_draw.instance_count = instance_count;
     maxwell3d.regs.global_base_vertex_index = parameters[3];
     maxwell3d.regs.index_buffer.count = parameters[1];
     maxwell3d.regs.index_buffer.first = parameters[4];
 
     if (maxwell3d.ShouldExecute()) {
-        maxwell3d.Rasterizer().Draw(true, true);
+        maxwell3d.Rasterizer().Draw(true, instance_count);
     }
     maxwell3d.regs.index_buffer.count = 0;
-    maxwell3d.mme_draw.instance_count = 0;
-    maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
 }
 
 void HLE_0D61FC9FAAC9FCAD(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
-    const u32 count = (maxwell3d.GetRegisterValue(0xD1B) & parameters[2]);
+    const u32 instance_count = (maxwell3d.GetRegisterValue(0xD1B) & parameters[2]);
 
     maxwell3d.regs.vertex_buffer.first = parameters[3];
     maxwell3d.regs.vertex_buffer.count = parameters[1];
     maxwell3d.regs.global_base_instance_index = parameters[4];
     maxwell3d.regs.draw.topology.Assign(
         static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]));
-    maxwell3d.mme_draw.instance_count = count;
 
     if (maxwell3d.ShouldExecute()) {
-        maxwell3d.Rasterizer().Draw(false, true);
+        maxwell3d.Rasterizer().Draw(false, instance_count);
     }
     maxwell3d.regs.vertex_buffer.count = 0;
-    maxwell3d.mme_draw.instance_count = 0;
-    maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
 }
 
 void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>& parameters) {
@@ -63,24 +57,21 @@ void HLE_0217920100488FF7(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
     maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
     maxwell3d.regs.global_base_vertex_index = element_base;
     maxwell3d.regs.global_base_instance_index = base_instance;
-    maxwell3d.mme_draw.instance_count = instance_count;
-    maxwell3d.CallMethodFromMME(0x8e3, 0x640);
-    maxwell3d.CallMethodFromMME(0x8e4, element_base);
-    maxwell3d.CallMethodFromMME(0x8e5, base_instance);
+    maxwell3d.CallMethod(0x8e3, 0x640, true);
+    maxwell3d.CallMethod(0x8e4, element_base, true);
+    maxwell3d.CallMethod(0x8e5, base_instance, true);
     maxwell3d.regs.draw.topology.Assign(
         static_cast<Tegra::Engines::Maxwell3D::Regs::PrimitiveTopology>(parameters[0]));
     if (maxwell3d.ShouldExecute()) {
-        maxwell3d.Rasterizer().Draw(true, true);
+        maxwell3d.Rasterizer().Draw(true, instance_count);
     }
     maxwell3d.regs.vertex_id_base = 0x0;
     maxwell3d.regs.index_buffer.count = 0;
     maxwell3d.regs.global_base_vertex_index = 0x0;
     maxwell3d.regs.global_base_instance_index = 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.CallMethod(0x8e3, 0x640, true);
+    maxwell3d.CallMethod(0x8e4, 0x0, true);
+    maxwell3d.CallMethod(0x8e5, 0x0, true);
 }
 
 // Multidraw Indirect
@@ -91,11 +82,9 @@ void HLE_3F5E74B9C9A50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
         maxwell3d.regs.index_buffer.count = 0;
         maxwell3d.regs.global_base_vertex_index = 0x0;
         maxwell3d.regs.global_base_instance_index = 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.CallMethod(0x8e3, 0x640, true);
+        maxwell3d.CallMethod(0x8e4, 0x0, true);
+        maxwell3d.CallMethod(0x8e5, 0x0, true);
         maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
     });
     const u32 start_indirect = parameters[0];
@@ -127,15 +116,13 @@ void HLE_3F5E74B9C9A50164(Engines::Maxwell3D& maxwell3d, const std::vector<u32>&
         maxwell3d.regs.index_buffer.count = num_vertices;
         maxwell3d.regs.global_base_vertex_index = base_vertex;
         maxwell3d.regs.global_base_instance_index = base_instance;
-        maxwell3d.mme_draw.instance_count = instance_count;
-        maxwell3d.CallMethodFromMME(0x8e3, 0x640);
-        maxwell3d.CallMethodFromMME(0x8e4, base_vertex);
-        maxwell3d.CallMethodFromMME(0x8e5, base_instance);
+        maxwell3d.CallMethod(0x8e3, 0x640, true);
+        maxwell3d.CallMethod(0x8e4, base_vertex, true);
+        maxwell3d.CallMethod(0x8e5, base_instance, true);
         maxwell3d.dirty.flags[VideoCommon::Dirty::IndexBuffer] = true;
         if (maxwell3d.ShouldExecute()) {
-            maxwell3d.Rasterizer().Draw(true, true);
+            maxwell3d.Rasterizer().Draw(true, instance_count);
         }
-        maxwell3d.mme_draw.current_mode = Engines::Maxwell3D::MMEDrawMode::Undefined;
     }
 }
 

src/video_core/macro/macro_interpreter.cpp

@@ -335,7 +335,7 @@ void MacroInterpreterImpl::SetMethodAddress(u32 address) {
 }
 
 void MacroInterpreterImpl::Send(u32 value) {
-    maxwell3d.CallMethodFromMME(method_address.address, value);
+    maxwell3d.CallMethod(method_address.address, value, true);
     // Increment the method address by the method increment.
     method_address.address.Assign(method_address.address.Value() +
                                   method_address.increment.Value());

src/video_core/macro/macro_jit_x64.cpp

@@ -346,7 +346,7 @@ void MacroJITx64Impl::Compile_Read(Macro::Opcode opcode) {
 }
 
 void Send(Engines::Maxwell3D* maxwell3d, Macro::MethodAddress method_address, u32 value) {
-    maxwell3d->CallMethodFromMME(method_address.address, value);
+    maxwell3d->CallMethod(method_address.address, value, true);
 }
 
 void MacroJITx64Impl::Compile_Send(Xbyak::Reg32 value) {

src/video_core/rasterizer_interface.h

@@ -40,7 +40,7 @@ public:
     virtual ~RasterizerInterface() = default;
 
     /// Dispatches a draw invocation
-    virtual void Draw(bool is_indexed, bool is_instanced) = 0;
+    virtual void Draw(bool is_indexed, u32 instance_count) = 0;
 
     /// Clear the current framebuffer
     virtual void Clear() = 0;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -205,7 +205,7 @@ void RasterizerOpenGL::Clear() {
     ++num_queued_commands;
 }
 
-void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
+void RasterizerOpenGL::Draw(bool is_indexed, u32 instance_count) {
     MICROPROFILE_SCOPE(OpenGL_Drawing);
 
     SCOPE_EXIT({ gpu.TickWork(); });
@@ -222,14 +222,15 @@ void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
     pipeline->SetEngine(maxwell3d, gpu_memory);
     pipeline->Configure(is_indexed);
 
+    BindInlineIndexBuffer();
+
     SyncState();
 
     const GLenum primitive_mode = MaxwellToGL::PrimitiveTopology(maxwell3d->regs.draw.topology);
     BeginTransformFeedback(pipeline, primitive_mode);
 
     const GLuint base_instance = static_cast<GLuint>(maxwell3d->regs.global_base_instance_index);
-    const GLsizei num_instances =
-        static_cast<GLsizei>(is_instanced ? maxwell3d->mme_draw.instance_count : 1);
+    const GLsizei num_instances = static_cast<GLsizei>(instance_count);
     if (is_indexed) {
         const GLint base_vertex = static_cast<GLint>(maxwell3d->regs.global_base_vertex_index);
         const GLsizei num_vertices = static_cast<GLsizei>(maxwell3d->regs.index_buffer.count);
@@ -1129,6 +1130,16 @@ void RasterizerOpenGL::ReleaseChannel(s32 channel_id) {
     query_cache.EraseChannel(channel_id);
 }
 
+void RasterizerOpenGL::BindInlineIndexBuffer() {
+    if (maxwell3d->inline_index_draw_indexes.empty()) {
+        return;
+    }
+    const auto data_count = static_cast<u32>(maxwell3d->inline_index_draw_indexes.size());
+    auto buffer = Buffer(buffer_cache_runtime, *this, 0, data_count);
+    buffer.ImmediateUpload(0, maxwell3d->inline_index_draw_indexes);
+    buffer_cache_runtime.BindIndexBuffer(buffer, 0, data_count);
+}
+
 AccelerateDMA::AccelerateDMA(BufferCache& buffer_cache_) : buffer_cache{buffer_cache_} {}
 
 bool AccelerateDMA::BufferCopy(GPUVAddr src_address, GPUVAddr dest_address, u64 amount) {

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -68,7 +68,7 @@ public:
                               StateTracker& state_tracker_);
     ~RasterizerOpenGL() override;
 
-    void Draw(bool is_indexed, bool is_instanced) override;
+    void Draw(bool is_indexed, u32 instance_count) override;
     void Clear() override;
     void DispatchCompute() override;
     void ResetCounter(VideoCore::QueryType type) override;
@@ -199,6 +199,8 @@ private:
     /// End a transform feedback
     void EndTransformFeedback();
 
+    void BindInlineIndexBuffer();
+
     Tegra::GPU& gpu;
 
     const Device& device;

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -127,11 +127,10 @@ VkRect2D GetScissorState(const Maxwell& regs, size_t index, u32 up_scale = 1, u3
     return scissor;
 }
 
-DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_instanced,
-                          bool is_indexed) {
+DrawParams MakeDrawParams(const Maxwell& regs, u32 num_instances, bool is_indexed) {
     DrawParams params{
         .base_instance = regs.global_base_instance_index,
-        .num_instances = is_instanced ? num_instances : 1,
+        .num_instances = num_instances,
         .base_vertex = is_indexed ? regs.global_base_vertex_index : regs.vertex_buffer.first,
         .num_vertices = is_indexed ? regs.index_buffer.count : regs.vertex_buffer.count,
         .first_index = is_indexed ? regs.index_buffer.first : 0,
@@ -177,7 +176,7 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
 
 RasterizerVulkan::~RasterizerVulkan() = default;
 
-void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
+void RasterizerVulkan::Draw(bool is_indexed, u32 instance_count) {
     MICROPROFILE_SCOPE(Vulkan_Drawing);
 
     SCOPE_EXIT({ gpu.TickWork(); });
@@ -194,13 +193,15 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
     pipeline->SetEngine(maxwell3d, gpu_memory);
     pipeline->Configure(is_indexed);
 
+    BindInlineIndexBuffer();
+
     BeginTransformFeedback();
 
     UpdateDynamicStates();
 
     const auto& regs{maxwell3d->regs};
-    const u32 num_instances{maxwell3d->mme_draw.instance_count};
-    const DrawParams draw_params{MakeDrawParams(regs, num_instances, is_instanced, is_indexed)};
+    const u32 num_instances{instance_count};
+    const DrawParams draw_params{MakeDrawParams(regs, num_instances, is_indexed)};
     scheduler.Record([draw_params](vk::CommandBuffer cmdbuf) {
         if (draw_params.is_indexed) {
             cmdbuf.DrawIndexed(draw_params.num_vertices, draw_params.num_instances,
@@ -1009,4 +1010,17 @@ void RasterizerVulkan::ReleaseChannel(s32 channel_id) {
     query_cache.EraseChannel(channel_id);
 }
 
+void RasterizerVulkan::BindInlineIndexBuffer() {
+    if (maxwell3d->inline_index_draw_indexes.empty()) {
+        return;
+    }
+    const auto data_count = static_cast<u32>(maxwell3d->inline_index_draw_indexes.size());
+    auto buffer = buffer_cache_runtime.UploadStagingBuffer(data_count);
+    std::memcpy(buffer.mapped_span.data(), maxwell3d->inline_index_draw_indexes.data(), data_count);
+    buffer_cache_runtime.BindIndexBuffer(
+        maxwell3d->regs.draw.topology, maxwell3d->regs.index_buffer.format,
+        maxwell3d->regs.index_buffer.first, maxwell3d->regs.index_buffer.count, buffer.buffer,
+        static_cast<u32>(buffer.offset), data_count);
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -64,7 +64,7 @@ public:
                               StateTracker& state_tracker_, Scheduler& scheduler_);
     ~RasterizerVulkan() override;
 
-    void Draw(bool is_indexed, bool is_instanced) override;
+    void Draw(bool is_indexed, u32 instance_count) override;
     void Clear() override;
     void DispatchCompute() override;
     void ResetCounter(VideoCore::QueryType type) override;
@@ -141,6 +141,8 @@ private:
 
     void UpdateVertexInput(Tegra::Engines::Maxwell3D::Regs& regs);
 
+    void BindInlineIndexBuffer();
+
     Tegra::GPU& gpu;
 
     ScreenInfo& screen_info;