arm_dynarmic: Remove IsExecuting check from PrepareReschedule

No longer required. HaltExecution is a no-op if it is not currently executing.

CMakeLists.txt

@@ -188,7 +188,7 @@ if (ENABLE_SDL2)
     if (YUZU_USE_BUNDLED_SDL2)
         # Detect toolchain and platform
         if ((MSVC_VERSION GREATER_EQUAL 1910 AND MSVC_VERSION LESS 1920) AND ARCHITECTURE_x86_64)
-            set(SDL2_VER "SDL2-2.0.5")
+            set(SDL2_VER "SDL2-2.0.8")
         else()
             message(FATAL_ERROR "No bundled SDL2 binaries for your toolchain. Disable YUZU_USE_BUNDLED_SDL2 and provide your own.")
         endif()

src/audio_core/audio_renderer.cpp

@@ -26,6 +26,18 @@ AudioRenderer::AudioRenderer(AudioRendererParameter params,
     QueueMixedBuffer(2);
 }
 
+u32 AudioRenderer::GetSampleRate() const {
+    return worker_params.sample_rate;
+}
+
+u32 AudioRenderer::GetSampleCount() const {
+    return worker_params.sample_count;
+}
+
+u32 AudioRenderer::GetMixBufferCount() const {
+    return worker_params.mix_buffer_count;
+}
+
 std::vector<u8> AudioRenderer::UpdateAudioRenderer(const std::vector<u8>& input_params) {
     // Copy UpdateDataHeader struct
     UpdateDataHeader config{};

src/audio_core/audio_renderer.h

@@ -26,7 +26,7 @@ enum class PlayState : u8 {
 struct AudioRendererParameter {
     u32_le sample_rate;
     u32_le sample_count;
-    u32_le unknown_8;
+    u32_le mix_buffer_count;
     u32_le unknown_c;
     u32_le voice_count;
     u32_le sink_count;
@@ -160,6 +160,9 @@ public:
     std::vector<u8> UpdateAudioRenderer(const std::vector<u8>& input_params);
     void QueueMixedBuffer(Buffer::Tag tag);
     void ReleaseAndQueueBuffers();
+    u32 GetSampleRate() const;
+    u32 GetSampleCount() const;
+    u32 GetMixBufferCount() const;
 
 private:
     class VoiceState {

src/audio_core/cubeb_sink.cpp

@@ -4,6 +4,7 @@
 
 #include <algorithm>
 #include <cstring>
+#include <mutex>
 
 #include "audio_core/cubeb_sink.h"
 #include "audio_core/stream.h"
@@ -66,6 +67,8 @@ public:
             return;
         }
 
+        std::lock_guard lock{queue_mutex};
+
         queue.reserve(queue.size() + samples.size() * GetNumChannels());
 
         if (is_6_channel) {
@@ -94,6 +97,7 @@ private:
     u32 num_channels{};
     bool is_6_channel{};
 
+    std::mutex queue_mutex;
     std::vector<s16> queue;
 
     static long DataCallback(cubeb_stream* stream, void* user_data, const void* input_buffer,
@@ -153,6 +157,8 @@ long SinkStreamImpl::DataCallback(cubeb_stream* stream, void* user_data, const v
         return {};
     }
 
+    std::lock_guard lock{impl->queue_mutex};
+
     const size_t frames_to_write{
         std::min(impl->queue.size() / impl->GetNumChannels(), static_cast<size_t>(num_frames))};
 

src/core/arm/dynarmic/arm_dynarmic.cpp

@@ -86,7 +86,16 @@ public:
     }
 
     void AddTicks(u64 ticks) override {
-        CoreTiming::AddTicks(ticks - num_interpreted_instructions);
+        // 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
+        // if not all cores are doing a similar amount of work. Instead of doing this, we should
+        // device a way so that timing is consistent across all cores without increasing the ticks 4
+        // times.
+        u64 amortized_ticks = (ticks - num_interpreted_instructions) / Core::NUM_CPU_CORES;
+        // Always execute at least one tick.
+        amortized_ticks = std::max<u64>(amortized_ticks, 1);
+
+        CoreTiming::AddTicks(amortized_ticks);
         num_interpreted_instructions = 0;
     }
     u64 GetTicksRemaining() override {
@@ -234,9 +243,7 @@ void ARM_Dynarmic::LoadContext(const ThreadContext& ctx) {
 }
 
 void ARM_Dynarmic::PrepareReschedule() {
-    if (jit->IsExecuting()) {
-        jit->HaltExecution();
-    }
+    jit->HaltExecution();
 }
 
 void ARM_Dynarmic::ClearInstructionCache() {

src/core/core_cpu.cpp

@@ -14,6 +14,7 @@
 #include "core/core_timing.h"
 #include "core/hle/kernel/scheduler.h"
 #include "core/hle/kernel/thread.h"
+#include "core/hle/lock.h"
 #include "core/settings.h"
 
 namespace Core {
@@ -90,6 +91,7 @@ void Cpu::RunLoop(bool tight_loop) {
         LOG_TRACE(Core, "Core-{} idling", core_index);
 
         if (IsMainCore()) {
+            // TODO(Subv): Only let CoreTiming idle if all 4 cores are idling.
             CoreTiming::Idle();
             CoreTiming::Advance();
         }
@@ -125,6 +127,8 @@ void Cpu::Reschedule() {
     }
 
     reschedule_pending = false;
+    // Lock the global kernel mutex when we manipulate the HLE state
+    std::lock_guard<std::recursive_mutex> lock(HLE::g_hle_lock);
     scheduler->Reschedule();
 }
 

src/core/core_cpu.h

@@ -79,7 +79,7 @@ private:
     std::shared_ptr<CpuBarrier> cpu_barrier;
     std::shared_ptr<Kernel::Scheduler> scheduler;
 
-    bool reschedule_pending{};
+    std::atomic<bool> reschedule_pending = false;
     size_t core_index;
 };
 

src/core/core_timing.cpp

@@ -135,11 +135,9 @@ void ClearPendingEvents() {
 void ScheduleEvent(s64 cycles_into_future, const EventType* event_type, u64 userdata) {
     ASSERT(event_type != nullptr);
     s64 timeout = GetTicks() + cycles_into_future;
-
     // If this event needs to be scheduled before the next advance(), force one early
     if (!is_global_timer_sane)
         ForceExceptionCheck(cycles_into_future);
-
     event_queue.emplace_back(Event{timeout, event_fifo_id++, userdata, event_type});
     std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>());
 }

src/core/file_sys/card_image.cpp

@@ -107,19 +107,19 @@ VirtualFile XCI::GetNCAFileByType(NCAContentType type) const {
     return nullptr;
 }
 
-std::vector<std::shared_ptr<VfsFile>> XCI::GetFiles() const {
+std::vector<VirtualFile> XCI::GetFiles() const {
     return {};
 }
 
-std::vector<std::shared_ptr<VfsDirectory>> XCI::GetSubdirectories() const {
-    return std::vector<std::shared_ptr<VfsDirectory>>();
+std::vector<VirtualDir> XCI::GetSubdirectories() const {
+    return {};
 }
 
 std::string XCI::GetName() const {
     return file->GetName();
 }
 
-std::shared_ptr<VfsDirectory> XCI::GetParentDirectory() const {
+VirtualDir XCI::GetParentDirectory() const {
     return file->GetContainingDirectory();
 }
 

src/core/file_sys/card_image.h

@@ -71,13 +71,13 @@ public:
     std::shared_ptr<NCA> GetNCAByType(NCAContentType type) const;
     VirtualFile GetNCAFileByType(NCAContentType type) const;
 
-    std::vector<std::shared_ptr<VfsFile>> GetFiles() const override;
+    std::vector<VirtualFile> GetFiles() const override;
 
-    std::vector<std::shared_ptr<VfsDirectory>> GetSubdirectories() const override;
+    std::vector<VirtualDir> GetSubdirectories() const override;
 
     std::string GetName() const override;
 
-    std::shared_ptr<VfsDirectory> GetParentDirectory() const override;
+    VirtualDir GetParentDirectory() const override;
 
 protected:
     bool ReplaceFileWithSubdirectory(VirtualFile file, VirtualDir dir) override;

src/core/file_sys/partition_filesystem.h

@@ -13,7 +13,7 @@
 #include "core/file_sys/vfs.h"
 
 namespace Loader {
-enum class ResultStatus;
+enum class ResultStatus : u16;
 }
 
 namespace FileSys {

src/core/file_sys/program_metadata.h

@@ -13,7 +13,7 @@
 #include "partition_filesystem.h"
 
 namespace Loader {
-enum class ResultStatus;
+enum class ResultStatus : u16;
 }
 
 namespace FileSys {

src/core/file_sys/vfs.cpp

@@ -74,15 +74,15 @@ VirtualFile VfsFilesystem::CopyFile(std::string_view old_path_, std::string_view
     return new_file;
 }
 
-VirtualFile VfsFilesystem::MoveFile(std::string_view old_path_, std::string_view new_path_) {
-    const auto old_path = FileUtil::SanitizePath(old_path_);
-    const auto new_path = FileUtil::SanitizePath(new_path_);
+VirtualFile VfsFilesystem::MoveFile(std::string_view old_path, std::string_view new_path) {
+    const auto sanitized_old_path = FileUtil::SanitizePath(old_path);
+    const auto sanitized_new_path = FileUtil::SanitizePath(new_path);
 
     // Again, non-default impls are highly encouraged to provide a more optimized version of this.
-    auto out = CopyFile(old_path_, new_path_);
+    auto out = CopyFile(sanitized_old_path, sanitized_new_path);
     if (out == nullptr)
         return nullptr;
-    if (DeleteFile(old_path))
+    if (DeleteFile(sanitized_old_path))
         return out;
     return nullptr;
 }
@@ -137,15 +137,15 @@ VirtualDir VfsFilesystem::CopyDirectory(std::string_view old_path_, std::string_
     return new_dir;
 }
 
-VirtualDir VfsFilesystem::MoveDirectory(std::string_view old_path_, std::string_view new_path_) {
-    const auto old_path = FileUtil::SanitizePath(old_path_);
-    const auto new_path = FileUtil::SanitizePath(new_path_);
+VirtualDir VfsFilesystem::MoveDirectory(std::string_view old_path, std::string_view new_path) {
+    const auto sanitized_old_path = FileUtil::SanitizePath(old_path);
+    const auto sanitized_new_path = FileUtil::SanitizePath(new_path);
 
     // Non-default impls are highly encouraged to provide a more optimized version of this.
-    auto out = CopyDirectory(old_path_, new_path_);
+    auto out = CopyDirectory(sanitized_old_path, sanitized_new_path);
     if (out == nullptr)
         return nullptr;
-    if (DeleteDirectory(old_path))
+    if (DeleteDirectory(sanitized_old_path))
         return out;
     return nullptr;
 }

src/core/file_sys/vfs.h

@@ -15,9 +15,9 @@
 
 namespace FileSys {
 
-struct VfsFilesystem;
-struct VfsFile;
-struct VfsDirectory;
+class VfsDirectory;
+class VfsFile;
+class VfsFilesystem;
 
 // Convenience typedefs to use Vfs* interfaces
 using VirtualFilesystem = std::shared_ptr<VfsFilesystem>;
@@ -34,8 +34,9 @@ enum class VfsEntryType {
 // A class representing an abstract filesystem. A default implementation given the root VirtualDir
 // is provided for convenience, but if the Vfs implementation has any additional state or
 // functionality, they will need to override.
-struct VfsFilesystem : NonCopyable {
-    VfsFilesystem(VirtualDir root);
+class VfsFilesystem : NonCopyable {
+public:
+    explicit VfsFilesystem(VirtualDir root);
     virtual ~VfsFilesystem();
 
     // Gets the friendly name for the filesystem.
@@ -81,7 +82,8 @@ protected:
 };
 
 // A class representing a file in an abstract filesystem.
-struct VfsFile : NonCopyable {
+class VfsFile : NonCopyable {
+public:
     virtual ~VfsFile();
 
     // Retrieves the file name.
@@ -179,7 +181,8 @@ struct VfsFile : NonCopyable {
 };
 
 // A class representing a directory in an abstract filesystem.
-struct VfsDirectory : NonCopyable {
+class VfsDirectory : NonCopyable {
+public:
     virtual ~VfsDirectory();
 
     // Retrives the file located at path as if the current directory was root. Returns nullptr if
@@ -295,7 +298,8 @@ protected:
 
 // A convenience partial-implementation of VfsDirectory that stubs out methods that should only work
 // if writable. This is to avoid redundant empty methods everywhere.
-struct ReadOnlyVfsDirectory : public VfsDirectory {
+class ReadOnlyVfsDirectory : public VfsDirectory {
+public:
     bool IsWritable() const override;
     bool IsReadable() const override;
     std::shared_ptr<VfsDirectory> CreateSubdirectory(std::string_view name) override;

src/core/file_sys/vfs_offset.h

@@ -15,7 +15,8 @@ namespace FileSys {
 // Similar to seeking to an offset.
 // If the file is writable, operations that would write past the end of the offset file will expand
 // the size of this wrapper.
-struct OffsetVfsFile : public VfsFile {
+class OffsetVfsFile : public VfsFile {
+public:
     OffsetVfsFile(std::shared_ptr<VfsFile> file, size_t size, size_t offset = 0,
                   std::string new_name = "", VirtualDir new_parent = nullptr);
 

src/core/file_sys/vfs_vector.h

@@ -10,7 +10,8 @@ namespace FileSys {
 
 // An implementation of VfsDirectory that maintains two vectors for subdirectories and files.
 // Vector data is supplied upon construction.
-struct VectorVfsDirectory : public VfsDirectory {
+class VectorVfsDirectory : public VfsDirectory {
+public:
     explicit VectorVfsDirectory(std::vector<VirtualFile> files = {},
                                 std::vector<VirtualDir> dirs = {}, VirtualDir parent = nullptr,
                                 std::string name = "");

src/core/hle/kernel/kernel.cpp

@@ -11,7 +11,7 @@
 
 namespace Kernel {
 
-unsigned int Object::next_object_id;
+std::atomic<u32> Object::next_object_id{0};
 
 /// Initialize the kernel
 void Init() {

src/core/hle/kernel/object.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <atomic>
 #include <string>
 #include <utility>
 
@@ -42,8 +43,8 @@ public:
     virtual ~Object();
 
     /// Returns a unique identifier for the object. For debugging purposes only.
-    unsigned int GetObjectId() const {
-        return object_id;
+    u32 GetObjectId() const {
+        return object_id.load(std::memory_order_relaxed);
     }
 
     virtual std::string GetTypeName() const {
@@ -61,23 +62,23 @@ public:
     bool IsWaitable() const;
 
 public:
-    static unsigned int next_object_id;
+    static std::atomic<u32> next_object_id;
 
 private:
     friend void intrusive_ptr_add_ref(Object*);
     friend void intrusive_ptr_release(Object*);
 
-    unsigned int ref_count = 0;
-    unsigned int object_id = next_object_id++;
+    std::atomic<u32> ref_count{0};
+    std::atomic<u32> object_id{next_object_id++};
 };
 
 // Special functions used by boost::instrusive_ptr to do automatic ref-counting
 inline void intrusive_ptr_add_ref(Object* object) {
-    ++object->ref_count;
+    object->ref_count.fetch_add(1, std::memory_order_relaxed);
 }
 
 inline void intrusive_ptr_release(Object* object) {
-    if (--object->ref_count == 0) {
+    if (object->ref_count.fetch_sub(1, std::memory_order_acq_rel) == 1) {
         delete object;
     }
 }

src/core/hle/kernel/svc.cpp

@@ -532,7 +532,6 @@ static ResultCode CreateThread(Handle* out_handle, VAddr entry_point, u64 arg, V
     CASCADE_RESULT(thread->guest_handle, g_handle_table.Create(thread));
     *out_handle = thread->guest_handle;
 
-    Core::System::GetInstance().PrepareReschedule();
     Core::System::GetInstance().CpuCore(thread->processor_id).PrepareReschedule();
 
     LOG_TRACE(Kernel_SVC,
@@ -706,8 +705,7 @@ static ResultCode SignalProcessWideKey(VAddr condition_variable_addr, s32 target
             Handle owner_handle = static_cast<Handle>(mutex_val & Mutex::MutexOwnerMask);
             auto owner = g_handle_table.Get<Thread>(owner_handle);
             ASSERT(owner);
-            ASSERT(thread->status != ThreadStatus::Running);
-            thread->status = ThreadStatus::WaitMutex;
+            ASSERT(thread->status == ThreadStatus::WaitMutex);
             thread->wakeup_callback = nullptr;
 
             owner->AddMutexWaiter(thread);

src/core/hle/kernel/thread.cpp

@@ -23,6 +23,7 @@
 #include "core/hle/kernel/object.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/thread.h"
+#include "core/hle/lock.h"
 #include "core/hle/result.h"
 #include "core/memory.h"
 
@@ -104,6 +105,10 @@ void ExitCurrentThread() {
  */
 static void ThreadWakeupCallback(u64 thread_handle, int cycles_late) {
     const auto proper_handle = static_cast<Handle>(thread_handle);
+
+    // Lock the global kernel mutex when we enter the kernel HLE.
+    std::lock_guard<std::recursive_mutex> lock(HLE::g_hle_lock);
+
     SharedPtr<Thread> thread = wakeup_callback_handle_table.Get<Thread>(proper_handle);
     if (thread == nullptr) {
         LOG_CRITICAL(Kernel, "Callback fired for invalid thread {:08X}", proper_handle);
@@ -155,8 +160,10 @@ void Thread::WakeAfterDelay(s64 nanoseconds) {
     if (nanoseconds == -1)
         return;
 
-    CoreTiming::ScheduleEvent(CoreTiming::nsToCycles(nanoseconds), ThreadWakeupEventType,
-                              callback_handle);
+    // This function might be called from any thread so we have to be cautious and use the
+    // thread-safe version of ScheduleEvent.
+    CoreTiming::ScheduleEventThreadsafe(CoreTiming::nsToCycles(nanoseconds), ThreadWakeupEventType,
+                                        callback_handle);
 }
 
 void Thread::CancelWakeupTimer() {
@@ -419,12 +426,33 @@ VAddr Thread::GetCommandBufferAddress() const {
 }
 
 void Thread::AddMutexWaiter(SharedPtr<Thread> thread) {
+    if (thread->lock_owner == this) {
+        // If the thread is already waiting for this thread to release the mutex, ensure that the
+        // waiters list is consistent and return without doing anything.
+        auto itr = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
+        ASSERT(itr != wait_mutex_threads.end());
+        return;
+    }
+
+    // A thread can't wait on two different mutexes at the same time.
+    ASSERT(thread->lock_owner == nullptr);
+
+    // Ensure that the thread is not already in the list of mutex waiters
+    auto itr = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
+    ASSERT(itr == wait_mutex_threads.end());
+
     thread->lock_owner = this;
     wait_mutex_threads.emplace_back(std::move(thread));
     UpdatePriority();
 }
 
 void Thread::RemoveMutexWaiter(SharedPtr<Thread> thread) {
+    ASSERT(thread->lock_owner == this);
+
+    // Ensure that the thread is in the list of mutex waiters
+    auto itr = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
+    ASSERT(itr != wait_mutex_threads.end());
+
     boost::remove_erase(wait_mutex_threads, thread);
     thread->lock_owner = nullptr;
     UpdatePriority();

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

@@ -20,9 +20,9 @@ public:
     explicit IAudioRenderer(AudioCore::AudioRendererParameter audren_params)
         : ServiceFramework("IAudioRenderer") {
         static const FunctionInfo functions[] = {
-            {0, nullptr, "GetAudioRendererSampleRate"},
-            {1, nullptr, "GetAudioRendererSampleCount"},
-            {2, nullptr, "GetAudioRendererMixBufferCount"},
+            {0, &IAudioRenderer::GetAudioRendererSampleRate, "GetAudioRendererSampleRate"},
+            {1, &IAudioRenderer::GetAudioRendererSampleCount, "GetAudioRendererSampleCount"},
+            {2, &IAudioRenderer::GetAudioRendererMixBufferCount, "GetAudioRendererMixBufferCount"},
             {3, nullptr, "GetAudioRendererState"},
             {4, &IAudioRenderer::RequestUpdateAudioRenderer, "RequestUpdateAudioRenderer"},
             {5, &IAudioRenderer::StartAudioRenderer, "StartAudioRenderer"},
@@ -45,6 +45,27 @@ private:
         system_event->Signal();
     }
 
+    void GetAudioRendererSampleRate(Kernel::HLERequestContext& ctx) {
+        IPC::ResponseBuilder rb{ctx, 3};
+        rb.Push(RESULT_SUCCESS);
+        rb.Push<u32>(renderer->GetSampleRate());
+        LOG_DEBUG(Service_Audio, "called");
+    }
+
+    void GetAudioRendererSampleCount(Kernel::HLERequestContext& ctx) {
+        IPC::ResponseBuilder rb{ctx, 3};
+        rb.Push(RESULT_SUCCESS);
+        rb.Push<u32>(renderer->GetSampleCount());
+        LOG_DEBUG(Service_Audio, "called");
+    }
+
+    void GetAudioRendererMixBufferCount(Kernel::HLERequestContext& ctx) {
+        IPC::ResponseBuilder rb{ctx, 3};
+        rb.Push(RESULT_SUCCESS);
+        rb.Push<u32>(renderer->GetMixBufferCount());
+        LOG_DEBUG(Service_Audio, "called");
+    }
+
     void RequestUpdateAudioRenderer(Kernel::HLERequestContext& ctx) {
         ctx.WriteBuffer(renderer->UpdateAudioRenderer(ctx.ReadBuffer()));
         IPC::ResponseBuilder rb{ctx, 2};
@@ -169,7 +190,8 @@ AudRenU::AudRenU() : ServiceFramework("audren:u") {
         {1, &AudRenU::GetAudioRendererWorkBufferSize, "GetAudioRendererWorkBufferSize"},
         {2, &AudRenU::GetAudioDevice, "GetAudioDevice"},
         {3, nullptr, "OpenAudioRendererAuto"},
-        {4, nullptr, "GetAudioDeviceServiceWithRevisionInfo"},
+        {4, &AudRenU::GetAudioDeviceServiceWithRevisionInfo,
+         "GetAudioDeviceServiceWithRevisionInfo"},
     };
     RegisterHandlers(functions);
 }
@@ -189,7 +211,7 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
     auto params = rp.PopRaw<AudioCore::AudioRendererParameter>();
 
-    u64 buffer_sz = Common::AlignUp(4 * params.unknown_8, 0x40);
+    u64 buffer_sz = Common::AlignUp(4 * params.mix_buffer_count, 0x40);
     buffer_sz += params.unknown_c * 1024;
     buffer_sz += 0x940 * (params.unknown_c + 1);
     buffer_sz += 0x3F0 * params.voice_count;
@@ -197,7 +219,7 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
     buffer_sz += Common::AlignUp(8 * params.voice_count, 0x10);
     buffer_sz +=
         Common::AlignUp((0x3C0 * (params.sink_count + params.unknown_c) + 4 * params.sample_count) *
-                            (params.unknown_8 + 6),
+                            (params.mix_buffer_count + 6),
                         0x40);
 
     if (IsFeatureSupported(AudioFeatures::Splitter, params.revision)) {
@@ -253,6 +275,16 @@ void AudRenU::GetAudioDevice(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_Audio, "called");
 }
 
+void AudRenU::GetAudioDeviceServiceWithRevisionInfo(Kernel::HLERequestContext& ctx) {
+    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
+
+    rb.Push(RESULT_SUCCESS);
+    rb.PushIpcInterface<Audio::IAudioDevice>();
+
+    LOG_WARNING(Service_Audio, "(STUBBED) called"); // TODO(ogniK): Figure out what is different
+                                                    // based on the current revision
+}
+
 bool AudRenU::IsFeatureSupported(AudioFeatures feature, u32_le revision) const {
     u32_be version_num = (revision - Common::MakeMagic('R', 'E', 'V', '0')); // Byte swap
     switch (feature) {

src/core/hle/service/audio/audren_u.h

@@ -22,6 +22,7 @@ private:
     void OpenAudioRenderer(Kernel::HLERequestContext& ctx);
     void GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx);
     void GetAudioDevice(Kernel::HLERequestContext& ctx);
+    void GetAudioDeviceServiceWithRevisionInfo(Kernel::HLERequestContext& ctx);
 
     enum class AudioFeatures : u32 {
         Splitter,

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

@@ -291,6 +291,7 @@ private:
 class Hid final : public ServiceFramework<Hid> {
 public:
     Hid() : ServiceFramework("hid") {
+        // clang-format off
         static const FunctionInfo functions[] = {
             {0, &Hid::CreateAppletResource, "CreateAppletResource"},
             {1, &Hid::ActivateDebugPad, "ActivateDebugPad"},
@@ -333,15 +334,13 @@ public:
             {102, &Hid::SetSupportedNpadIdType, "SetSupportedNpadIdType"},
             {103, &Hid::ActivateNpad, "ActivateNpad"},
             {104, nullptr, "DeactivateNpad"},
-            {106, &Hid::AcquireNpadStyleSetUpdateEventHandle,
-             "AcquireNpadStyleSetUpdateEventHandle"},
-            {107, nullptr, "DisconnectNpad"},
+            {106, &Hid::AcquireNpadStyleSetUpdateEventHandle, "AcquireNpadStyleSetUpdateEventHandle"},
+            {107, &Hid::DisconnectNpad, "DisconnectNpad"},
             {108, &Hid::GetPlayerLedPattern, "GetPlayerLedPattern"},
             {109, nullptr, "ActivateNpadWithRevision"},
             {120, &Hid::SetNpadJoyHoldType, "SetNpadJoyHoldType"},
             {121, &Hid::GetNpadJoyHoldType, "GetNpadJoyHoldType"},
-            {122, &Hid::SetNpadJoyAssignmentModeSingleByDefault,
-             "SetNpadJoyAssignmentModeSingleByDefault"},
+            {122, &Hid::SetNpadJoyAssignmentModeSingleByDefault, "SetNpadJoyAssignmentModeSingleByDefault"},
             {123, nullptr, "SetNpadJoyAssignmentModeSingleByDefault"},
             {124, &Hid::SetNpadJoyAssignmentModeDual, "SetNpadJoyAssignmentModeDual"},
             {125, &Hid::MergeSingleJoyAsDualJoy, "MergeSingleJoyAsDualJoy"},
@@ -398,6 +397,8 @@ public:
             {1000, nullptr, "SetNpadCommunicationMode"},
             {1001, nullptr, "GetNpadCommunicationMode"},
         };
+        // clang-format on
+
         RegisterHandlers(functions);
 
         event = Kernel::Event::Create(Kernel::ResetType::OneShot, "hid:EventHandle");
@@ -496,6 +497,12 @@ private:
         LOG_WARNING(Service_HID, "(STUBBED) called");
     }
 
+    void DisconnectNpad(Kernel::HLERequestContext& ctx) {
+        IPC::ResponseBuilder rb{ctx, 2};
+        rb.Push(RESULT_SUCCESS);
+        LOG_WARNING(Service_HID, "(STUBBED) called");
+    }
+
     void GetPlayerLedPattern(Kernel::HLERequestContext& ctx) {
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(RESULT_SUCCESS);

src/core/hle/service/service.h

@@ -23,7 +23,7 @@ class HLERequestContext;
 } // namespace Kernel
 
 namespace FileSys {
-struct VfsFilesystem;
+class VfsFilesystem;
 }
 
 namespace Service {

src/core/loader/loader.cpp

@@ -126,7 +126,7 @@ constexpr std::array<const char*, 36> RESULT_MESSAGES{
 };
 
 std::string GetMessageForResultStatus(ResultStatus status) {
-    return GetMessageForResultStatus(static_cast<size_t>(status));
+    return GetMessageForResultStatus(static_cast<u16>(status));
 }
 
 std::string GetMessageForResultStatus(u16 status) {

src/core/loader/loader.h

@@ -56,7 +56,7 @@ FileType GuessFromFilename(const std::string& name);
 std::string GetFileTypeString(FileType type);
 
 /// Return type for functions in Loader namespace
-enum class ResultStatus {
+enum class ResultStatus : u16 {
     Success,
     ErrorAlreadyLoaded,
     ErrorNotImplemented,

src/video_core/engines/shader_bytecode.h

@@ -200,6 +200,14 @@ enum class IMinMaxExchange : u64 {
     XHi = 3,
 };
 
+enum class XmadMode : u64 {
+    None = 0,
+    CLo = 1,
+    CHi = 2,
+    CSfu = 3,
+    CBcc = 4,
+};
+
 enum class FlowCondition : u64 {
     Always = 0xF,
     Fcsm_Tr = 0x1C, // TODO(bunnei): What is this used for?
@@ -456,6 +464,18 @@ union Instruction {
         }
     } bra;
 
+    union {
+        BitField<20, 16, u64> imm20_16;
+        BitField<36, 1, u64> product_shift_left;
+        BitField<37, 1, u64> merge_37;
+        BitField<48, 1, u64> sign_a;
+        BitField<49, 1, u64> sign_b;
+        BitField<50, 3, XmadMode> mode;
+        BitField<52, 1, u64> high_b;
+        BitField<53, 1, u64> high_a;
+        BitField<56, 1, u64> merge_56;
+    } xmad;
+
     union {
         BitField<20, 14, u64> offset;
         BitField<34, 5, u64> index;
@@ -593,6 +613,7 @@ public:
         IntegerSetPredicate,
         PredicateSetPredicate,
         Conversion,
+        Xmad,
         Unknown,
     };
 
@@ -782,10 +803,10 @@ private:
             INST("010010110101----", Id::ISET_C, Type::IntegerSet, "ISET_C"),
             INST("0011011-0101----", Id::ISET_IMM, Type::IntegerSet, "ISET_IMM"),
             INST("0101000010010---", Id::PSETP, Type::PredicateSetPredicate, "PSETP"),
-            INST("0011011-00------", Id::XMAD_IMM, Type::Arithmetic, "XMAD_IMM"),
-            INST("0100111---------", Id::XMAD_CR, Type::Arithmetic, "XMAD_CR"),
-            INST("010100010-------", Id::XMAD_RC, Type::Arithmetic, "XMAD_RC"),
-            INST("0101101100------", Id::XMAD_RR, Type::Arithmetic, "XMAD_RR"),
+            INST("0011011-00------", Id::XMAD_IMM, Type::Xmad, "XMAD_IMM"),
+            INST("0100111---------", Id::XMAD_CR, Type::Xmad, "XMAD_CR"),
+            INST("010100010-------", Id::XMAD_RC, Type::Xmad, "XMAD_RC"),
+            INST("0101101100------", Id::XMAD_RR, Type::Xmad, "XMAD_RR"),
         };
 #undef INST
         std::stable_sort(table.begin(), table.end(), [](const auto& a, const auto& b) {

src/video_core/gpu.cpp

@@ -46,6 +46,7 @@ u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
     case RenderTargetFormat::RGBA32_FLOAT:
     case RenderTargetFormat::RGBA32_UINT:
         return 16;
+    case RenderTargetFormat::RGBA16_UINT:
     case RenderTargetFormat::RGBA16_FLOAT:
     case RenderTargetFormat::RG32_FLOAT:
         return 8;
@@ -67,6 +68,7 @@ u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
     case RenderTargetFormat::R16_UINT:
     case RenderTargetFormat::R16_SINT:
     case RenderTargetFormat::R16_FLOAT:
+    case RenderTargetFormat::RG8_UNORM:
     case RenderTargetFormat::RG8_SNORM:
         return 2;
     case RenderTargetFormat::R8_UNORM:

src/video_core/gpu.h

@@ -20,6 +20,7 @@ enum class RenderTargetFormat : u32 {
     NONE = 0x0,
     RGBA32_FLOAT = 0xC0,
     RGBA32_UINT = 0xC2,
+    RGBA16_UINT = 0xC9,
     RGBA16_FLOAT = 0xCA,
     RG32_FLOAT = 0xCB,
     BGRA8_UNORM = 0xCF,
@@ -35,6 +36,7 @@ enum class RenderTargetFormat : u32 {
     R11G11B10_FLOAT = 0xE0,
     R32_FLOAT = 0xE5,
     B5G6R5_UNORM = 0xE8,
+    RG8_UNORM = 0xEA,
     RG8_SNORM = 0xEB,
     R16_UNORM = 0xEE,
     R16_SNORM = 0xEF,

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -36,30 +36,21 @@ MICROPROFILE_DEFINE(OpenGL_Drawing, "OpenGL", "Drawing", MP_RGB(128, 128, 192));
 MICROPROFILE_DEFINE(OpenGL_Blits, "OpenGL", "Blits", MP_RGB(100, 100, 255));
 MICROPROFILE_DEFINE(OpenGL_CacheManagement, "OpenGL", "Cache Mgmt", MP_RGB(100, 255, 100));
 
-RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window) : emu_window{window} {
+RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window)
+    : emu_window{window}, stream_buffer(GL_ARRAY_BUFFER, STREAM_BUFFER_SIZE) {
     // Create sampler objects
     for (size_t i = 0; i < texture_samplers.size(); ++i) {
         texture_samplers[i].Create();
         state.texture_units[i].sampler = texture_samplers[i].sampler.handle;
     }
 
-    // Create SSBOs
-    for (size_t stage = 0; stage < ssbos.size(); ++stage) {
-        for (size_t buffer = 0; buffer < ssbos[stage].size(); ++buffer) {
-            ssbos[stage][buffer].Create();
-            state.draw.const_buffers[stage][buffer].ssbo = ssbos[stage][buffer].handle;
-        }
-    }
-
     GLint ext_num;
     glGetIntegerv(GL_NUM_EXTENSIONS, &ext_num);
     for (GLint i = 0; i < ext_num; i++) {
         const std::string_view extension{
             reinterpret_cast<const char*>(glGetStringi(GL_EXTENSIONS, i))};
 
-        if (extension == "GL_ARB_buffer_storage") {
-            has_ARB_buffer_storage = true;
-        } else if (extension == "GL_ARB_direct_state_access") {
+        if (extension == "GL_ARB_direct_state_access") {
             has_ARB_direct_state_access = true;
         } else if (extension == "GL_ARB_separate_shader_objects") {
             has_ARB_separate_shader_objects = true;
@@ -86,47 +77,31 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window) : emu_wind
 
     hw_vao.Create();
 
-    stream_buffer = OGLStreamBuffer::MakeBuffer(has_ARB_buffer_storage, GL_ARRAY_BUFFER);
-    stream_buffer->Create(STREAM_BUFFER_SIZE, STREAM_BUFFER_SIZE / 2);
-    state.draw.vertex_buffer = stream_buffer->GetHandle();
+    state.draw.vertex_buffer = stream_buffer.GetHandle();
 
     shader_program_manager = std::make_unique<GLShader::ProgramManager>();
     state.draw.shader_program = 0;
     state.draw.vertex_array = hw_vao.handle;
     state.Apply();
 
-    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, stream_buffer->GetHandle());
-
-    for (unsigned index = 0; index < uniform_buffers.size(); ++index) {
-        auto& buffer = uniform_buffers[index];
-        buffer.Create();
-        glBindBuffer(GL_UNIFORM_BUFFER, buffer.handle);
-        glBufferData(GL_UNIFORM_BUFFER, sizeof(GLShader::MaxwellUniformData), nullptr,
-                     GL_STREAM_COPY);
-        glBindBufferBase(GL_UNIFORM_BUFFER, index, buffer.handle);
-    }
+    glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, stream_buffer.GetHandle());
 
     glEnable(GL_BLEND);
 
+    glGetIntegerv(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT, &uniform_buffer_alignment);
+
     LOG_CRITICAL(Render_OpenGL, "Sync fixed function OpenGL state here!");
 }
 
-RasterizerOpenGL::~RasterizerOpenGL() {
-    if (stream_buffer != nullptr) {
-        state.draw.vertex_buffer = stream_buffer->GetHandle();
-        state.Apply();
-        stream_buffer->Release();
-    }
-}
+RasterizerOpenGL::~RasterizerOpenGL() {}
 
 std::pair<u8*, GLintptr> RasterizerOpenGL::SetupVertexArrays(u8* array_ptr,
                                                              GLintptr buffer_offset) {
     MICROPROFILE_SCOPE(OpenGL_VAO);
     const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
-    const auto& memory_manager = Core::System::GetInstance().GPU().memory_manager;
 
     state.draw.vertex_array = hw_vao.handle;
-    state.draw.vertex_buffer = stream_buffer->GetHandle();
+    state.draw.vertex_buffer = stream_buffer.GetHandle();
     state.Apply();
 
     // Upload all guest vertex arrays sequentially to our buffer
@@ -141,16 +116,15 @@ std::pair<u8*, GLintptr> RasterizerOpenGL::SetupVertexArrays(u8* array_ptr,
         ASSERT(end > start);
         u64 size = end - start + 1;
 
-        // Copy vertex array data
-        Memory::ReadBlock(*memory_manager->GpuToCpuAddress(start), array_ptr, size);
+        GLintptr vertex_buffer_offset;
+        std::tie(array_ptr, buffer_offset, vertex_buffer_offset) =
+            UploadMemory(array_ptr, buffer_offset, start, size);
 
         // Bind the vertex array to the buffer at the current offset.
-        glBindVertexBuffer(index, stream_buffer->GetHandle(), buffer_offset, vertex_array.stride);
+        glBindVertexBuffer(index, stream_buffer.GetHandle(), vertex_buffer_offset,
+                           vertex_array.stride);
 
         ASSERT_MSG(vertex_array.divisor == 0, "Vertex buffer divisor unimplemented");
-
-        array_ptr += size;
-        buffer_offset += size;
     }
 
     // Use the vertex array as-is, assumes that the data is formatted correctly for OpenGL.
@@ -201,22 +175,12 @@ static GLShader::ProgramCode GetShaderProgramCode(Maxwell::ShaderProgram program
     return program_code;
 }
 
-void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
-    // Helper function for uploading uniform data
-    const auto copy_buffer = [&](GLuint handle, GLintptr offset, GLsizeiptr size) {
-        if (has_ARB_direct_state_access) {
-            glCopyNamedBufferSubData(stream_buffer->GetHandle(), handle, offset, 0, size);
-        } else {
-            glBindBuffer(GL_COPY_WRITE_BUFFER, handle);
-            glCopyBufferSubData(GL_ARRAY_BUFFER, GL_COPY_WRITE_BUFFER, offset, 0, size);
-        }
-    };
-
+std::pair<u8*, GLintptr> RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
     auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
 
     // Next available bindpoints to use when uploading the const buffers and textures to the GLSL
     // shaders. The constbuffer bindpoint starts after the shader stage configuration bind points.
-    u32 current_constbuffer_bindpoint = static_cast<u32>(uniform_buffers.size());
+    u32 current_constbuffer_bindpoint = Tegra::Engines::Maxwell3D::Regs::MaxShaderStage;
     u32 current_texture_bindpoint = 0;
 
     for (size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
@@ -228,22 +192,21 @@ void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
             continue;
         }
 
+        std::tie(buffer_ptr, buffer_offset) =
+            AlignBuffer(buffer_ptr, buffer_offset, static_cast<size_t>(uniform_buffer_alignment));
+
         const size_t stage{index == 0 ? 0 : index - 1}; // Stage indices are 0 - 5
 
         GLShader::MaxwellUniformData ubo{};
         ubo.SetFromRegs(gpu.state.shader_stages[stage]);
         std::memcpy(buffer_ptr, &ubo, sizeof(ubo));
 
-        // Flush the buffer so that the GPU can see the data we just wrote.
-        glFlushMappedBufferRange(GL_ARRAY_BUFFER, buffer_offset, sizeof(ubo));
+        // Bind the buffer
+        glBindBufferRange(GL_UNIFORM_BUFFER, stage, stream_buffer.GetHandle(), buffer_offset,
+                          sizeof(ubo));
 
-        // Upload uniform data as one UBO per stage
-        const GLintptr ubo_offset = buffer_offset;
-        copy_buffer(uniform_buffers[stage].handle, ubo_offset,
-                    sizeof(GLShader::MaxwellUniformData));
-
-        buffer_ptr += sizeof(GLShader::MaxwellUniformData);
-        buffer_offset += sizeof(GLShader::MaxwellUniformData);
+        buffer_ptr += sizeof(ubo);
+        buffer_offset += sizeof(ubo);
 
         GLShader::ShaderSetup setup{GetShaderProgramCode(program)};
         GLShader::ShaderEntries shader_resources;
@@ -282,9 +245,9 @@ void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
             static_cast<Maxwell::ShaderStage>(stage));
 
         // Configure the const buffers for this shader stage.
-        current_constbuffer_bindpoint =
-            SetupConstBuffers(static_cast<Maxwell::ShaderStage>(stage), gl_stage_program,
-                              current_constbuffer_bindpoint, shader_resources.const_buffer_entries);
+        std::tie(buffer_ptr, buffer_offset, current_constbuffer_bindpoint) = SetupConstBuffers(
+            buffer_ptr, buffer_offset, static_cast<Maxwell::ShaderStage>(stage), gl_stage_program,
+            current_constbuffer_bindpoint, shader_resources.const_buffer_entries);
 
         // Configure the textures for this shader stage.
         current_texture_bindpoint =
@@ -299,6 +262,8 @@ void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
     }
 
     shader_program_manager->UseTrivialGeometryShader();
+
+    return {buffer_ptr, buffer_offset};
 }
 
 size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
@@ -432,6 +397,31 @@ void RasterizerOpenGL::Clear() {
     }
 }
 
+std::pair<u8*, GLintptr> RasterizerOpenGL::AlignBuffer(u8* buffer_ptr, GLintptr buffer_offset,
+                                                       size_t alignment) {
+    // Align the offset, not the mapped pointer
+    GLintptr offset_aligned =
+        static_cast<GLintptr>(Common::AlignUp(static_cast<size_t>(buffer_offset), alignment));
+    return {buffer_ptr + (offset_aligned - buffer_offset), offset_aligned};
+}
+
+std::tuple<u8*, GLintptr, GLintptr> RasterizerOpenGL::UploadMemory(u8* buffer_ptr,
+                                                                   GLintptr buffer_offset,
+                                                                   Tegra::GPUVAddr gpu_addr,
+                                                                   size_t size, size_t alignment) {
+    std::tie(buffer_ptr, buffer_offset) = AlignBuffer(buffer_ptr, buffer_offset, alignment);
+    GLintptr uploaded_offset = buffer_offset;
+
+    const auto& memory_manager = Core::System::GetInstance().GPU().memory_manager;
+    const boost::optional<VAddr> cpu_addr{memory_manager->GpuToCpuAddress(gpu_addr)};
+    Memory::ReadBlock(*cpu_addr, buffer_ptr, size);
+
+    buffer_ptr += size;
+    buffer_offset += size;
+
+    return {buffer_ptr, buffer_offset, uploaded_offset};
+}
+
 void RasterizerOpenGL::DrawArrays() {
     if (accelerate_draw == AccelDraw::Disabled)
         return;
@@ -456,7 +446,7 @@ void RasterizerOpenGL::DrawArrays() {
     const u64 index_buffer_size{regs.index_array.count * regs.index_array.FormatSizeInBytes()};
     const unsigned vertex_num{is_indexed ? regs.index_array.count : regs.vertex_buffer.count};
 
-    state.draw.vertex_buffer = stream_buffer->GetHandle();
+    state.draw.vertex_buffer = stream_buffer.GetHandle();
     state.Apply();
 
     size_t buffer_size = CalculateVertexArraysSize();
@@ -466,41 +456,31 @@ void RasterizerOpenGL::DrawArrays() {
     }
 
     // Uniform space for the 5 shader stages
-    buffer_size = Common::AlignUp<size_t>(buffer_size, 4) +
-                  sizeof(GLShader::MaxwellUniformData) * Maxwell::MaxShaderStage;
+    buffer_size =
+        Common::AlignUp<size_t>(buffer_size, 4) +
+        (sizeof(GLShader::MaxwellUniformData) + uniform_buffer_alignment) * Maxwell::MaxShaderStage;
+
+    // Add space for at least 18 constant buffers
+    buffer_size += Maxwell::MaxConstBuffers * (MaxConstbufferSize + uniform_buffer_alignment);
 
     u8* buffer_ptr;
     GLintptr buffer_offset;
-    std::tie(buffer_ptr, buffer_offset) =
-        stream_buffer->Map(static_cast<GLsizeiptr>(buffer_size), 4);
+    std::tie(buffer_ptr, buffer_offset, std::ignore) =
+        stream_buffer.Map(static_cast<GLsizeiptr>(buffer_size), 4);
+    u8* buffer_ptr_base = buffer_ptr;
 
-    u8* offseted_buffer;
-    std::tie(offseted_buffer, buffer_offset) = SetupVertexArrays(buffer_ptr, buffer_offset);
-
-    offseted_buffer =
-        reinterpret_cast<u8*>(Common::AlignUp(reinterpret_cast<size_t>(offseted_buffer), 4));
-    buffer_offset = Common::AlignUp<size_t>(buffer_offset, 4);
+    std::tie(buffer_ptr, buffer_offset) = SetupVertexArrays(buffer_ptr, buffer_offset);
 
     // If indexed mode, copy the index buffer
     GLintptr index_buffer_offset = 0;
     if (is_indexed) {
-        const auto& memory_manager = Core::System::GetInstance().GPU().memory_manager;
-        const boost::optional<VAddr> index_data_addr{
-            memory_manager->GpuToCpuAddress(regs.index_array.StartAddress())};
-        Memory::ReadBlock(*index_data_addr, offseted_buffer, index_buffer_size);
-
-        index_buffer_offset = buffer_offset;
-        offseted_buffer += index_buffer_size;
-        buffer_offset += index_buffer_size;
+        std::tie(buffer_ptr, buffer_offset, index_buffer_offset) = UploadMemory(
+            buffer_ptr, buffer_offset, regs.index_array.StartAddress(), index_buffer_size);
     }
 
-    offseted_buffer =
-        reinterpret_cast<u8*>(Common::AlignUp(reinterpret_cast<size_t>(offseted_buffer), 4));
-    buffer_offset = Common::AlignUp<size_t>(buffer_offset, 4);
+    std::tie(buffer_ptr, buffer_offset) = SetupShaders(buffer_ptr, buffer_offset);
 
-    SetupShaders(offseted_buffer, buffer_offset);
-
-    stream_buffer->Unmap();
+    stream_buffer.Unmap(buffer_ptr - buffer_ptr_base);
 
     shader_program_manager->ApplyTo(state);
     state.Apply();
@@ -647,36 +627,23 @@ void RasterizerOpenGL::SamplerInfo::SyncWithConfig(const Tegra::Texture::TSCEntr
     }
 }
 
-u32 RasterizerOpenGL::SetupConstBuffers(Maxwell::ShaderStage stage, GLuint program,
-                                        u32 current_bindpoint,
-                                        const std::vector<GLShader::ConstBufferEntry>& entries) {
+std::tuple<u8*, GLintptr, u32> RasterizerOpenGL::SetupConstBuffers(
+    u8* buffer_ptr, GLintptr buffer_offset, Maxwell::ShaderStage stage, GLuint program,
+    u32 current_bindpoint, const std::vector<GLShader::ConstBufferEntry>& entries) {
     const auto& gpu = Core::System::GetInstance().GPU();
     const auto& maxwell3d = gpu.Maxwell3D();
 
-    // Reset all buffer draw state for this stage.
-    for (auto& buffer : state.draw.const_buffers[static_cast<size_t>(stage)]) {
-        buffer.bindpoint = 0;
-        buffer.enabled = false;
-    }
-
     // Upload only the enabled buffers from the 16 constbuffers of each shader stage
     const auto& shader_stage = maxwell3d.state.shader_stages[static_cast<size_t>(stage)];
 
     for (u32 bindpoint = 0; bindpoint < entries.size(); ++bindpoint) {
         const auto& used_buffer = entries[bindpoint];
         const auto& buffer = shader_stage.const_buffers[used_buffer.GetIndex()];
-        auto& buffer_draw_state =
-            state.draw.const_buffers[static_cast<size_t>(stage)][used_buffer.GetIndex()];
 
         if (!buffer.enabled) {
             continue;
         }
 
-        buffer_draw_state.enabled = true;
-        buffer_draw_state.bindpoint = current_bindpoint + bindpoint;
-
-        boost::optional<VAddr> addr = gpu.memory_manager->GpuToCpuAddress(buffer.address);
-
         size_t size = 0;
 
         if (used_buffer.IsIndirect()) {
@@ -698,25 +665,26 @@ u32 RasterizerOpenGL::SetupConstBuffers(Maxwell::ShaderStage stage, GLuint progr
         size = Common::AlignUp(size, sizeof(GLvec4));
         ASSERT_MSG(size <= MaxConstbufferSize, "Constbuffer too big");
 
-        std::vector<u8> data(size);
-        Memory::ReadBlock(*addr, data.data(), data.size());
+        GLintptr const_buffer_offset;
+        std::tie(buffer_ptr, buffer_offset, const_buffer_offset) =
+            UploadMemory(buffer_ptr, buffer_offset, buffer.address, size,
+                         static_cast<size_t>(uniform_buffer_alignment));
 
-        glBindBuffer(GL_UNIFORM_BUFFER, buffer_draw_state.ssbo);
-        glBufferData(GL_UNIFORM_BUFFER, data.size(), data.data(), GL_DYNAMIC_DRAW);
-        glBindBuffer(GL_UNIFORM_BUFFER, 0);
+        glBindBufferRange(GL_UNIFORM_BUFFER, current_bindpoint + bindpoint,
+                          stream_buffer.GetHandle(), const_buffer_offset, size);
 
         // Now configure the bindpoint of the buffer inside the shader
         const std::string buffer_name = used_buffer.GetName();
         const GLuint index =
             glGetProgramResourceIndex(program, GL_UNIFORM_BLOCK, buffer_name.c_str());
         if (index != GL_INVALID_INDEX) {
-            glUniformBlockBinding(program, index, buffer_draw_state.bindpoint);
+            glUniformBlockBinding(program, index, current_bindpoint + bindpoint);
         }
     }
 
     state.Apply();
 
-    return current_bindpoint + static_cast<u32>(entries.size());
+    return {buffer_ptr, buffer_offset, current_bindpoint + static_cast<u32>(entries.size())};
 }
 
 u32 RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, GLuint program, u32 current_unit,

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -7,6 +7,7 @@
 #include <array>
 #include <cstddef>
 #include <memory>
+#include <tuple>
 #include <utility>
 #include <vector>
 #include <glad/glad.h>
@@ -100,9 +101,10 @@ private:
      * @param entries Vector describing the buffers that are actually used in the guest shader.
      * @returns The next available bindpoint for use in the next shader stage.
      */
-    u32 SetupConstBuffers(Tegra::Engines::Maxwell3D::Regs::ShaderStage stage, GLuint program,
-                          u32 current_bindpoint,
-                          const std::vector<GLShader::ConstBufferEntry>& entries);
+    std::tuple<u8*, GLintptr, u32> SetupConstBuffers(
+        u8* buffer_ptr, GLintptr buffer_offset, Tegra::Engines::Maxwell3D::Regs::ShaderStage stage,
+        GLuint program, u32 current_bindpoint,
+        const std::vector<GLShader::ConstBufferEntry>& entries);
 
     /*
      * Configures the current textures to use for the draw command.
@@ -139,7 +141,6 @@ private:
     /// Syncs the blend state to match the guest state
     void SyncBlendState();
 
-    bool has_ARB_buffer_storage = false;
     bool has_ARB_direct_state_access = false;
     bool has_ARB_separate_shader_objects = false;
     bool has_ARB_vertex_attrib_binding = false;
@@ -155,22 +156,24 @@ private:
     OGLVertexArray hw_vao;
 
     std::array<SamplerInfo, GLShader::NumTextureSamplers> texture_samplers;
-    std::array<std::array<OGLBuffer, Tegra::Engines::Maxwell3D::Regs::MaxConstBuffers>,
-               Tegra::Engines::Maxwell3D::Regs::MaxShaderStage>
-        ssbos;
 
     static constexpr size_t STREAM_BUFFER_SIZE = 128 * 1024 * 1024;
-    std::unique_ptr<OGLStreamBuffer> stream_buffer;
+    OGLStreamBuffer stream_buffer;
     OGLBuffer uniform_buffer;
     OGLFramebuffer framebuffer;
+    GLint uniform_buffer_alignment;
 
     size_t CalculateVertexArraysSize() const;
 
     std::pair<u8*, GLintptr> SetupVertexArrays(u8* array_ptr, GLintptr buffer_offset);
 
-    std::array<OGLBuffer, Tegra::Engines::Maxwell3D::Regs::MaxShaderStage> uniform_buffers;
+    std::pair<u8*, GLintptr> SetupShaders(u8* buffer_ptr, GLintptr buffer_offset);
 
-    void SetupShaders(u8* buffer_ptr, GLintptr buffer_offset);
+    std::pair<u8*, GLintptr> AlignBuffer(u8* buffer_ptr, GLintptr buffer_offset, size_t alignment);
+
+    std::tuple<u8*, GLintptr, GLintptr> UploadMemory(u8* buffer_ptr, GLintptr buffer_offset,
+                                                     Tegra::GPUVAddr gpu_addr, size_t size,
+                                                     size_t alignment = 4);
 
     enum class AccelDraw { Disabled, Arrays, Indexed };
     AccelDraw accelerate_draw = AccelDraw::Disabled;

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -101,6 +101,7 @@ static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_form
     {GL_R8, GL_RED, GL_UNSIGNED_BYTE, ComponentType::UNorm, false},                    // R8
     {GL_R8UI, GL_RED_INTEGER, GL_UNSIGNED_BYTE, ComponentType::UInt, false},           // R8UI
     {GL_RGBA16F, GL_RGBA, GL_HALF_FLOAT, ComponentType::Float, false},                 // RGBA16F
+    {GL_RGBA16UI, GL_RGBA, GL_UNSIGNED_SHORT, ComponentType::UInt, false},             // RGBA16UI
     {GL_R11F_G11F_B10F, GL_RGB, GL_UNSIGNED_INT_10F_11F_11F_REV, ComponentType::Float,
      false},                                                                     // R11FG11FB10F
     {GL_RGBA32UI, GL_RGBA_INTEGER, GL_UNSIGNED_INT, ComponentType::UInt, false}, // RGBA32UI
@@ -134,6 +135,7 @@ static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_form
     {GL_RG16_SNORM, GL_RG, GL_SHORT, ComponentType::SNorm, false},             // RG16S
     {GL_RGB32F, GL_RGB, GL_FLOAT, ComponentType::Float, false},                // RGB32F
     {GL_SRGB8_ALPHA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, ComponentType::UNorm, false}, // SRGBA8
+    {GL_RG8, GL_RG, GL_UNSIGNED_BYTE, ComponentType::UNorm, false},                       // RG8U
     {GL_RG8, GL_RG, GL_BYTE, ComponentType::SNorm, false},                                // RG8S
 
     // DepthStencil formats
@@ -234,32 +236,57 @@ void MortonCopy(u32 stride, u32 block_height, u32 height, std::vector<u8>& gl_bu
 static constexpr std::array<void (*)(u32, u32, u32, std::vector<u8>&, Tegra::GPUVAddr),
                             SurfaceParams::MaxPixelFormat>
     morton_to_gl_fns = {
-        MortonCopy<true, PixelFormat::ABGR8U>,       MortonCopy<true, PixelFormat::ABGR8S>,
-        MortonCopy<true, PixelFormat::B5G6R5>,       MortonCopy<true, PixelFormat::A2B10G10R10>,
-        MortonCopy<true, PixelFormat::A1B5G5R5>,     MortonCopy<true, PixelFormat::R8>,
-        MortonCopy<true, PixelFormat::R8UI>,         MortonCopy<true, PixelFormat::RGBA16F>,
-        MortonCopy<true, PixelFormat::R11FG11FB10F>, MortonCopy<true, PixelFormat::RGBA32UI>,
-        MortonCopy<true, PixelFormat::DXT1>,         MortonCopy<true, PixelFormat::DXT23>,
-        MortonCopy<true, PixelFormat::DXT45>,        MortonCopy<true, PixelFormat::DXN1>,
-        MortonCopy<true, PixelFormat::DXN2UNORM>,    MortonCopy<true, PixelFormat::DXN2SNORM>,
-        MortonCopy<true, PixelFormat::BC7U>,         MortonCopy<true, PixelFormat::ASTC_2D_4X4>,
-        MortonCopy<true, PixelFormat::G8R8>,         MortonCopy<true, PixelFormat::BGRA8>,
-        MortonCopy<true, PixelFormat::RGBA32F>,      MortonCopy<true, PixelFormat::RG32F>,
-        MortonCopy<true, PixelFormat::R32F>,         MortonCopy<true, PixelFormat::R16F>,
-        MortonCopy<true, PixelFormat::R16UNORM>,     MortonCopy<true, PixelFormat::R16S>,
-        MortonCopy<true, PixelFormat::R16UI>,        MortonCopy<true, PixelFormat::R16I>,
-        MortonCopy<true, PixelFormat::RG16>,         MortonCopy<true, PixelFormat::RG16F>,
-        MortonCopy<true, PixelFormat::RG16UI>,       MortonCopy<true, PixelFormat::RG16I>,
-        MortonCopy<true, PixelFormat::RG16S>,        MortonCopy<true, PixelFormat::RGB32F>,
-        MortonCopy<true, PixelFormat::SRGBA8>,       MortonCopy<true, PixelFormat::RG8S>,
-        MortonCopy<true, PixelFormat::Z24S8>,        MortonCopy<true, PixelFormat::S8Z24>,
-        MortonCopy<true, PixelFormat::Z32F>,         MortonCopy<true, PixelFormat::Z16>,
+        // clang-format off
+        MortonCopy<true, PixelFormat::ABGR8U>,
+        MortonCopy<true, PixelFormat::ABGR8S>,
+        MortonCopy<true, PixelFormat::B5G6R5>,
+        MortonCopy<true, PixelFormat::A2B10G10R10>,
+        MortonCopy<true, PixelFormat::A1B5G5R5>,
+        MortonCopy<true, PixelFormat::R8>,
+        MortonCopy<true, PixelFormat::R8UI>,
+        MortonCopy<true, PixelFormat::RGBA16F>,
+        MortonCopy<true, PixelFormat::RGBA16UI>,
+        MortonCopy<true, PixelFormat::R11FG11FB10F>,
+        MortonCopy<true, PixelFormat::RGBA32UI>,
+        MortonCopy<true, PixelFormat::DXT1>,
+        MortonCopy<true, PixelFormat::DXT23>,
+        MortonCopy<true, PixelFormat::DXT45>,
+        MortonCopy<true, PixelFormat::DXN1>,
+        MortonCopy<true, PixelFormat::DXN2UNORM>,
+        MortonCopy<true, PixelFormat::DXN2SNORM>,
+        MortonCopy<true, PixelFormat::BC7U>,
+        MortonCopy<true, PixelFormat::ASTC_2D_4X4>,
+        MortonCopy<true, PixelFormat::G8R8>,
+        MortonCopy<true, PixelFormat::BGRA8>,
+        MortonCopy<true, PixelFormat::RGBA32F>,
+        MortonCopy<true, PixelFormat::RG32F>,
+        MortonCopy<true, PixelFormat::R32F>,
+        MortonCopy<true, PixelFormat::R16F>,
+        MortonCopy<true, PixelFormat::R16UNORM>,
+        MortonCopy<true, PixelFormat::R16S>,
+        MortonCopy<true, PixelFormat::R16UI>,
+        MortonCopy<true, PixelFormat::R16I>,
+        MortonCopy<true, PixelFormat::RG16>,
+        MortonCopy<true, PixelFormat::RG16F>,
+        MortonCopy<true, PixelFormat::RG16UI>,
+        MortonCopy<true, PixelFormat::RG16I>,
+        MortonCopy<true, PixelFormat::RG16S>,
+        MortonCopy<true, PixelFormat::RGB32F>,
+        MortonCopy<true, PixelFormat::SRGBA8>,
+        MortonCopy<true, PixelFormat::RG8U>,
+        MortonCopy<true, PixelFormat::RG8S>,
+        MortonCopy<true, PixelFormat::Z24S8>,
+        MortonCopy<true, PixelFormat::S8Z24>,
+        MortonCopy<true, PixelFormat::Z32F>,
+        MortonCopy<true, PixelFormat::Z16>,
         MortonCopy<true, PixelFormat::Z32FS8>,
+        // clang-format on
 };
 
 static constexpr std::array<void (*)(u32, u32, u32, std::vector<u8>&, Tegra::GPUVAddr),
                             SurfaceParams::MaxPixelFormat>
     gl_to_morton_fns = {
+        // clang-format off
         MortonCopy<false, PixelFormat::ABGR8U>,
         MortonCopy<false, PixelFormat::ABGR8S>,
         MortonCopy<false, PixelFormat::B5G6R5>,
@@ -268,6 +295,7 @@ static constexpr std::array<void (*)(u32, u32, u32, std::vector<u8>&, Tegra::GPU
         MortonCopy<false, PixelFormat::R8>,
         MortonCopy<false, PixelFormat::R8UI>,
         MortonCopy<false, PixelFormat::RGBA16F>,
+        MortonCopy<false, PixelFormat::RGBA16UI>,
         MortonCopy<false, PixelFormat::R11FG11FB10F>,
         MortonCopy<false, PixelFormat::RGBA32UI>,
         // TODO(Subv): Swizzling DXT1/DXT23/DXT45/DXN1/DXN2/BC7U/ASTC_2D_4X4 formats is not
@@ -297,12 +325,14 @@ static constexpr std::array<void (*)(u32, u32, u32, std::vector<u8>&, Tegra::GPU
         MortonCopy<false, PixelFormat::RG16S>,
         MortonCopy<false, PixelFormat::RGB32F>,
         MortonCopy<false, PixelFormat::SRGBA8>,
+        MortonCopy<false, PixelFormat::RG8U>,
         MortonCopy<false, PixelFormat::RG8S>,
         MortonCopy<false, PixelFormat::Z24S8>,
         MortonCopy<false, PixelFormat::S8Z24>,
         MortonCopy<false, PixelFormat::Z32F>,
         MortonCopy<false, PixelFormat::Z16>,
         MortonCopy<false, PixelFormat::Z32FS8>,
+        // clang-format on
 };
 
 // Allocate an uninitialized texture of appropriate size and format for the surface

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -31,43 +31,45 @@ struct SurfaceParams {
         R8 = 5,
         R8UI = 6,
         RGBA16F = 7,
-        R11FG11FB10F = 8,
-        RGBA32UI = 9,
-        DXT1 = 10,
-        DXT23 = 11,
-        DXT45 = 12,
-        DXN1 = 13, // This is also known as BC4
-        DXN2UNORM = 14,
-        DXN2SNORM = 15,
-        BC7U = 16,
-        ASTC_2D_4X4 = 17,
-        G8R8 = 18,
-        BGRA8 = 19,
-        RGBA32F = 20,
-        RG32F = 21,
-        R32F = 22,
-        R16F = 23,
-        R16UNORM = 24,
-        R16S = 25,
-        R16UI = 26,
-        R16I = 27,
-        RG16 = 28,
-        RG16F = 29,
-        RG16UI = 30,
-        RG16I = 31,
-        RG16S = 32,
-        RGB32F = 33,
-        SRGBA8 = 34,
-        RG8S = 35,
+        RGBA16UI = 8,
+        R11FG11FB10F = 9,
+        RGBA32UI = 10,
+        DXT1 = 11,
+        DXT23 = 12,
+        DXT45 = 13,
+        DXN1 = 14, // This is also known as BC4
+        DXN2UNORM = 15,
+        DXN2SNORM = 16,
+        BC7U = 17,
+        ASTC_2D_4X4 = 18,
+        G8R8 = 19,
+        BGRA8 = 20,
+        RGBA32F = 21,
+        RG32F = 22,
+        R32F = 23,
+        R16F = 24,
+        R16UNORM = 25,
+        R16S = 26,
+        R16UI = 27,
+        R16I = 28,
+        RG16 = 29,
+        RG16F = 30,
+        RG16UI = 31,
+        RG16I = 32,
+        RG16S = 33,
+        RGB32F = 34,
+        SRGBA8 = 35,
+        RG8U = 36,
+        RG8S = 37,
 
         MaxColorFormat,
 
         // DepthStencil formats
-        Z24S8 = 36,
-        S8Z24 = 37,
-        Z32F = 38,
-        Z16 = 39,
-        Z32FS8 = 40,
+        Z24S8 = 38,
+        S8Z24 = 39,
+        Z32F = 40,
+        Z16 = 41,
+        Z32FS8 = 42,
 
         MaxDepthStencilFormat,
 
@@ -113,6 +115,7 @@ struct SurfaceParams {
             1, // R8
             1, // R8UI
             1, // RGBA16F
+            1, // RGBA16UI
             1, // R11FG11FB10F
             1, // RGBA32UI
             4, // DXT1
@@ -140,6 +143,7 @@ struct SurfaceParams {
             1, // RG16S
             1, // RGB32F
             1, // SRGBA8
+            1, // RG8U
             1, // RG8S
             1, // Z24S8
             1, // S8Z24
@@ -165,6 +169,7 @@ struct SurfaceParams {
             8,   // R8
             8,   // R8UI
             64,  // RGBA16F
+            64,  // RGBA16UI
             32,  // R11FG11FB10F
             128, // RGBA32UI
             64,  // DXT1
@@ -192,6 +197,7 @@ struct SurfaceParams {
             32,  // RG16S
             96,  // RGB32F
             32,  // SRGBA8
+            16,  // RG8U
             16,  // RG8S
             32,  // Z24S8
             32,  // S8Z24
@@ -241,6 +247,8 @@ struct SurfaceParams {
             return PixelFormat::A2B10G10R10;
         case Tegra::RenderTargetFormat::RGBA16_FLOAT:
             return PixelFormat::RGBA16F;
+        case Tegra::RenderTargetFormat::RGBA16_UINT:
+            return PixelFormat::RGBA16UI;
         case Tegra::RenderTargetFormat::RGBA32_FLOAT:
             return PixelFormat::RGBA32F;
         case Tegra::RenderTargetFormat::RG32_FLOAT:
@@ -265,6 +273,8 @@ struct SurfaceParams {
             return PixelFormat::RG16;
         case Tegra::RenderTargetFormat::RG16_SNORM:
             return PixelFormat::RG16S;
+        case Tegra::RenderTargetFormat::RG8_UNORM:
+            return PixelFormat::RG8U;
         case Tegra::RenderTargetFormat::RG8_SNORM:
             return PixelFormat::RG8S;
         case Tegra::RenderTargetFormat::R16_FLOAT:
@@ -432,6 +442,7 @@ struct SurfaceParams {
         case Tegra::RenderTargetFormat::RG16_UNORM:
         case Tegra::RenderTargetFormat::R16_UNORM:
         case Tegra::RenderTargetFormat::B5G6R5_UNORM:
+        case Tegra::RenderTargetFormat::RG8_UNORM:
             return ComponentType::UNorm;
         case Tegra::RenderTargetFormat::RGBA8_SNORM:
         case Tegra::RenderTargetFormat::RG16_SNORM:
@@ -447,6 +458,7 @@ struct SurfaceParams {
         case Tegra::RenderTargetFormat::R32_FLOAT:
             return ComponentType::Float;
         case Tegra::RenderTargetFormat::RGBA32_UINT:
+        case Tegra::RenderTargetFormat::RGBA16_UINT:
         case Tegra::RenderTargetFormat::RG16_UINT:
         case Tegra::RenderTargetFormat::R8_UINT:
         case Tegra::RenderTargetFormat::R16_UINT:

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -376,6 +376,8 @@ public:
             return value;
         } else if (type == GLSLRegister::Type::Integer) {
             return "floatBitsToInt(" + value + ')';
+        } else if (type == GLSLRegister::Type::UnsignedInteger) {
+            return "floatBitsToUint(" + value + ')';
         } else {
             UNREACHABLE();
         }
@@ -1630,6 +1632,99 @@ private:
             }
             break;
         }
+        case OpCode::Type::Xmad: {
+            ASSERT_MSG(!instr.xmad.sign_a, "Unimplemented");
+            ASSERT_MSG(!instr.xmad.sign_b, "Unimplemented");
+
+            std::string op_a{regs.GetRegisterAsInteger(instr.gpr8, 0, instr.xmad.sign_a)};
+            std::string op_b;
+            std::string op_c;
+
+            // TODO(bunnei): Needs to be fixed once op_a or op_b is signed
+            ASSERT_MSG(instr.xmad.sign_a == instr.xmad.sign_b, "Unimplemented");
+            const bool is_signed{instr.xmad.sign_a == 1};
+
+            bool is_merge{};
+            switch (opcode->GetId()) {
+            case OpCode::Id::XMAD_CR: {
+                is_merge = instr.xmad.merge_56;
+                op_b += regs.GetUniform(instr.cbuf34.index, instr.cbuf34.offset,
+                                        instr.xmad.sign_b ? GLSLRegister::Type::Integer
+                                                          : GLSLRegister::Type::UnsignedInteger);
+                op_c += regs.GetRegisterAsInteger(instr.gpr39, 0, is_signed);
+                break;
+            }
+            case OpCode::Id::XMAD_RR: {
+                is_merge = instr.xmad.merge_37;
+                op_b += regs.GetRegisterAsInteger(instr.gpr20, 0, instr.xmad.sign_b);
+                op_c += regs.GetRegisterAsInteger(instr.gpr39, 0, is_signed);
+                break;
+            }
+            case OpCode::Id::XMAD_RC: {
+                op_b += regs.GetRegisterAsInteger(instr.gpr39, 0, instr.xmad.sign_b);
+                op_c += regs.GetUniform(instr.cbuf34.index, instr.cbuf34.offset,
+                                        is_signed ? GLSLRegister::Type::Integer
+                                                  : GLSLRegister::Type::UnsignedInteger);
+                break;
+            }
+            case OpCode::Id::XMAD_IMM: {
+                is_merge = instr.xmad.merge_37;
+                op_b += std::to_string(instr.xmad.imm20_16);
+                op_c += regs.GetRegisterAsInteger(instr.gpr39, 0, is_signed);
+                break;
+            }
+            default: {
+                LOG_CRITICAL(HW_GPU, "Unhandled XMAD instruction: {}", opcode->GetName());
+                UNREACHABLE();
+            }
+            }
+
+            // TODO(bunnei): Ensure this is right with signed operands
+            if (instr.xmad.high_a) {
+                op_a = "((" + op_a + ") >> 16)";
+            } else {
+                op_a = "((" + op_a + ") & 0xFFFF)";
+            }
+
+            std::string src2 = '(' + op_b + ')'; // Preserve original source 2
+            if (instr.xmad.high_b) {
+                op_b = '(' + src2 + " >> 16)";
+            } else {
+                op_b = '(' + src2 + " & 0xFFFF)";
+            }
+
+            std::string product = '(' + op_a + " * " + op_b + ')';
+            if (instr.xmad.product_shift_left) {
+                product = '(' + product + " << 16)";
+            }
+
+            switch (instr.xmad.mode) {
+            case Tegra::Shader::XmadMode::None:
+                break;
+            case Tegra::Shader::XmadMode::CLo:
+                op_c = "((" + op_c + ") & 0xFFFF)";
+                break;
+            case Tegra::Shader::XmadMode::CHi:
+                op_c = "((" + op_c + ") >> 16)";
+                break;
+            case Tegra::Shader::XmadMode::CBcc:
+                op_c = "((" + op_c + ") + (" + src2 + "<< 16))";
+                break;
+            default: {
+                LOG_CRITICAL(HW_GPU, "Unhandled XMAD mode: {}",
+                             static_cast<u32>(instr.xmad.mode.Value()));
+                UNREACHABLE();
+            }
+            }
+
+            std::string sum{'(' + product + " + " + op_c + ')'};
+            if (is_merge) {
+                sum = "((" + sum + " & 0xFFFF) | (" + src2 + "<< 16))";
+            }
+
+            regs.SetRegisterToInteger(instr.gpr0, is_signed, 0, sum, 1, 1);
+            break;
+        }
         default: {
             switch (opcode->GetId()) {
             case OpCode::Id::EXIT: {

src/video_core/renderer_opengl/gl_state.cpp

@@ -203,21 +203,6 @@ void OpenGLState::Apply() const {
         }
     }
 
-    // Constbuffers
-    for (std::size_t stage = 0; stage < draw.const_buffers.size(); ++stage) {
-        for (std::size_t buffer_id = 0; buffer_id < draw.const_buffers[stage].size(); ++buffer_id) {
-            const auto& current = cur_state.draw.const_buffers[stage][buffer_id];
-            const auto& new_state = draw.const_buffers[stage][buffer_id];
-
-            if (current.enabled != new_state.enabled || current.bindpoint != new_state.bindpoint ||
-                current.ssbo != new_state.ssbo) {
-                if (new_state.enabled) {
-                    glBindBufferBase(GL_UNIFORM_BUFFER, new_state.bindpoint, new_state.ssbo);
-                }
-            }
-        }
-    }
-
     // Framebuffer
     if (draw.read_framebuffer != cur_state.draw.read_framebuffer) {
         glBindFramebuffer(GL_READ_FRAMEBUFFER, draw.read_framebuffer);

src/video_core/renderer_opengl/gl_state.h

@@ -119,12 +119,6 @@ public:
         GLuint uniform_buffer;   // GL_UNIFORM_BUFFER_BINDING
         GLuint shader_program;   // GL_CURRENT_PROGRAM
         GLuint program_pipeline; // GL_PROGRAM_PIPELINE_BINDING
-        struct ConstBufferConfig {
-            bool enabled = false;
-            GLuint bindpoint;
-            GLuint ssbo;
-        };
-        std::array<std::array<ConstBufferConfig, Regs::MaxConstBuffers>, 5> const_buffers;
     } draw;
 
     struct {

src/video_core/renderer_opengl/gl_stream_buffer.cpp

@@ -9,174 +9,91 @@
 #include "video_core/renderer_opengl/gl_state.h"
 #include "video_core/renderer_opengl/gl_stream_buffer.h"
 
-class OrphanBuffer : public OGLStreamBuffer {
-public:
-    explicit OrphanBuffer(GLenum target) : OGLStreamBuffer(target) {}
-    ~OrphanBuffer() override;
+OGLStreamBuffer::OGLStreamBuffer(GLenum target, GLsizeiptr size, bool prefer_coherent)
+    : gl_target(target), buffer_size(size) {
+    gl_buffer.Create();
+    glBindBuffer(gl_target, gl_buffer.handle);
 
-private:
-    void Create(size_t size, size_t sync_subdivide) override;
-    void Release() override;
+    GLsizeiptr allocate_size = size;
+    if (target == GL_ARRAY_BUFFER) {
+        // On AMD GPU there is a strange crash in indexed drawing. The crash happens when the buffer
+        // read position is near the end and is an out-of-bound access to the vertex buffer. This is
+        // probably a bug in the driver and is related to the usage of vec3<byte> attributes in the
+        // vertex array. Doubling the allocation size for the vertex buffer seems to avoid the
+        // crash.
+        allocate_size *= 2;
+    }
 
-    std::pair<u8*, GLintptr> Map(size_t size, size_t alignment) override;
-    void Unmap() override;
+    if (GLAD_GL_ARB_buffer_storage) {
+        persistent = true;
+        coherent = prefer_coherent;
+        GLbitfield flags =
+            GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | (coherent ? GL_MAP_COHERENT_BIT : 0);
+        glBufferStorage(gl_target, allocate_size, nullptr, flags);
+        mapped_ptr = static_cast<u8*>(glMapBufferRange(
+            gl_target, 0, buffer_size, flags | (coherent ? 0 : GL_MAP_FLUSH_EXPLICIT_BIT)));
+    } else {
+        glBufferData(gl_target, allocate_size, nullptr, GL_STREAM_DRAW);
+    }
+}
 
-    std::vector<u8> data;
-};
-
-class StorageBuffer : public OGLStreamBuffer {
-public:
-    explicit StorageBuffer(GLenum target) : OGLStreamBuffer(target) {}
-    ~StorageBuffer() override;
-
-private:
-    void Create(size_t size, size_t sync_subdivide) override;
-    void Release() override;
-
-    std::pair<u8*, GLintptr> Map(size_t size, size_t alignment) override;
-    void Unmap() override;
-
-    struct Fence {
-        OGLSync sync;
-        size_t offset;
-    };
-    std::deque<Fence> head;
-    std::deque<Fence> tail;
-
-    u8* mapped_ptr;
-};
-
-OGLStreamBuffer::OGLStreamBuffer(GLenum target) {
-    gl_target = target;
+OGLStreamBuffer::~OGLStreamBuffer() {
+    if (persistent) {
+        glBindBuffer(gl_target, gl_buffer.handle);
+        glUnmapBuffer(gl_target);
+    }
+    gl_buffer.Release();
 }
 
 GLuint OGLStreamBuffer::GetHandle() const {
     return gl_buffer.handle;
 }
 
-std::unique_ptr<OGLStreamBuffer> OGLStreamBuffer::MakeBuffer(bool storage_buffer, GLenum target) {
-    if (storage_buffer) {
-        return std::make_unique<StorageBuffer>(target);
-    }
-    return std::make_unique<OrphanBuffer>(target);
+GLsizeiptr OGLStreamBuffer::GetSize() const {
+    return buffer_size;
 }
 
-OrphanBuffer::~OrphanBuffer() {
-    Release();
-}
-
-void OrphanBuffer::Create(size_t size, size_t /*sync_subdivide*/) {
-    buffer_pos = 0;
-    buffer_size = size;
-    data.resize(buffer_size);
-
-    if (gl_buffer.handle == 0) {
-        gl_buffer.Create();
-        glBindBuffer(gl_target, gl_buffer.handle);
-    }
-
-    glBufferData(gl_target, static_cast<GLsizeiptr>(buffer_size), nullptr, GL_STREAM_DRAW);
-}
-
-void OrphanBuffer::Release() {
-    gl_buffer.Release();
-}
-
-std::pair<u8*, GLintptr> OrphanBuffer::Map(size_t size, size_t alignment) {
-    buffer_pos = Common::AlignUp(buffer_pos, alignment);
-
-    if (buffer_pos + size > buffer_size) {
-        Create(std::max(buffer_size, size), 0);
-    }
-
-    mapped_size = size;
-    return std::make_pair(&data[buffer_pos], static_cast<GLintptr>(buffer_pos));
-}
-
-void OrphanBuffer::Unmap() {
-    glBufferSubData(gl_target, static_cast<GLintptr>(buffer_pos),
-                    static_cast<GLsizeiptr>(mapped_size), &data[buffer_pos]);
-    buffer_pos += mapped_size;
-}
-
-StorageBuffer::~StorageBuffer() {
-    Release();
-}
-
-void StorageBuffer::Create(size_t size, size_t sync_subdivide) {
-    if (gl_buffer.handle != 0)
-        return;
-
-    buffer_pos = 0;
-    buffer_size = size;
-    buffer_sync_subdivide = std::max<size_t>(sync_subdivide, 1);
-
-    gl_buffer.Create();
-    glBindBuffer(gl_target, gl_buffer.handle);
-
-    glBufferStorage(gl_target, static_cast<GLsizeiptr>(buffer_size), nullptr,
-                    GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT);
-    mapped_ptr = reinterpret_cast<u8*>(
-        glMapBufferRange(gl_target, 0, static_cast<GLsizeiptr>(buffer_size),
-                         GL_MAP_WRITE_BIT | GL_MAP_PERSISTENT_BIT | GL_MAP_FLUSH_EXPLICIT_BIT));
-}
-
-void StorageBuffer::Release() {
-    if (gl_buffer.handle == 0)
-        return;
-
-    glUnmapBuffer(gl_target);
-
-    gl_buffer.Release();
-    head.clear();
-    tail.clear();
-}
-
-std::pair<u8*, GLintptr> StorageBuffer::Map(size_t size, size_t alignment) {
+std::tuple<u8*, GLintptr, bool> OGLStreamBuffer::Map(GLsizeiptr size, GLintptr alignment) {
     ASSERT(size <= buffer_size);
-
-    OGLSync sync;
-
-    buffer_pos = Common::AlignUp(buffer_pos, alignment);
-    size_t effective_offset = Common::AlignDown(buffer_pos, buffer_sync_subdivide);
-
-    if (!head.empty() &&
-        (effective_offset > head.back().offset || buffer_pos + size > buffer_size)) {
-        ASSERT(head.back().sync.handle == 0);
-        head.back().sync.Create();
-    }
-
-    if (buffer_pos + size > buffer_size) {
-        if (!tail.empty()) {
-            std::swap(sync, tail.back().sync);
-            tail.clear();
-        }
-        std::swap(tail, head);
-        buffer_pos = 0;
-        effective_offset = 0;
-    }
-
-    while (!tail.empty() && buffer_pos + size > tail.front().offset) {
-        std::swap(sync, tail.front().sync);
-        tail.pop_front();
-    }
-
-    if (sync.handle != 0) {
-        glClientWaitSync(sync.handle, GL_SYNC_FLUSH_COMMANDS_BIT, GL_TIMEOUT_IGNORED);
-        sync.Release();
-    }
-
-    if (head.empty() || effective_offset > head.back().offset) {
-        head.emplace_back();
-        head.back().offset = effective_offset;
-    }
-
+    ASSERT(alignment <= buffer_size);
     mapped_size = size;
-    return std::make_pair(&mapped_ptr[buffer_pos], static_cast<GLintptr>(buffer_pos));
+
+    if (alignment > 0) {
+        buffer_pos = Common::AlignUp<size_t>(buffer_pos, alignment);
+    }
+
+    bool invalidate = false;
+    if (buffer_pos + size > buffer_size) {
+        buffer_pos = 0;
+        invalidate = true;
+
+        if (persistent) {
+            glUnmapBuffer(gl_target);
+        }
+    }
+
+    if (invalidate | !persistent) {
+        GLbitfield flags = GL_MAP_WRITE_BIT | (persistent ? GL_MAP_PERSISTENT_BIT : 0) |
+                           (coherent ? GL_MAP_COHERENT_BIT : GL_MAP_FLUSH_EXPLICIT_BIT) |
+                           (invalidate ? GL_MAP_INVALIDATE_BUFFER_BIT : GL_MAP_UNSYNCHRONIZED_BIT);
+        mapped_ptr = static_cast<u8*>(
+            glMapBufferRange(gl_target, buffer_pos, buffer_size - buffer_pos, flags));
+        mapped_offset = buffer_pos;
+    }
+
+    return std::make_tuple(mapped_ptr + buffer_pos - mapped_offset, buffer_pos, invalidate);
 }
 
-void StorageBuffer::Unmap() {
-    glFlushMappedBufferRange(gl_target, static_cast<GLintptr>(buffer_pos),
-                             static_cast<GLsizeiptr>(mapped_size));
-    buffer_pos += mapped_size;
+void OGLStreamBuffer::Unmap(GLsizeiptr size) {
+    ASSERT(size <= mapped_size);
+
+    if (!coherent && size > 0) {
+        glFlushMappedBufferRange(gl_target, buffer_pos - mapped_offset, size);
+    }
+
+    if (!persistent) {
+        glUnmapBuffer(gl_target);
+    }
+
+    buffer_pos += size;
 }

src/video_core/renderer_opengl/gl_stream_buffer.h

@@ -2,35 +2,41 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#pragma once
-
-#include <memory>
+#include <tuple>
 #include <glad/glad.h>
 #include "common/common_types.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 
 class OGLStreamBuffer : private NonCopyable {
 public:
-    explicit OGLStreamBuffer(GLenum target);
-    virtual ~OGLStreamBuffer() = default;
-
-public:
-    static std::unique_ptr<OGLStreamBuffer> MakeBuffer(bool storage_buffer, GLenum target);
-
-    virtual void Create(size_t size, size_t sync_subdivide) = 0;
-    virtual void Release() {}
+    explicit OGLStreamBuffer(GLenum target, GLsizeiptr size, bool prefer_coherent = false);
+    ~OGLStreamBuffer();
 
     GLuint GetHandle() const;
+    GLsizeiptr GetSize() const;
 
-    virtual std::pair<u8*, GLintptr> Map(size_t size, size_t alignment) = 0;
-    virtual void Unmap() = 0;
+    /*
+     * Allocates a linear chunk of memory in the GPU buffer with at least "size" bytes
+     * and the optional alignment requirement.
+     * If the buffer is full, the whole buffer is reallocated which invalidates old chunks.
+     * The return values are the pointer to the new chunk, the offset within the buffer,
+     * and the invalidation flag for previous chunks.
+     * The actual used size must be specified on unmapping the chunk.
+     */
+    std::tuple<u8*, GLintptr, bool> Map(GLsizeiptr size, GLintptr alignment = 0);
 
-protected:
+    void Unmap(GLsizeiptr size);
+
+private:
     OGLBuffer gl_buffer;
     GLenum gl_target;
 
-    size_t buffer_pos = 0;
-    size_t buffer_size = 0;
-    size_t buffer_sync_subdivide = 0;
-    size_t mapped_size = 0;
+    bool coherent = false;
+    bool persistent = false;
+
+    GLintptr buffer_pos = 0;
+    GLsizeiptr buffer_size = 0;
+    GLintptr mapped_offset = 0;
+    GLsizeiptr mapped_size = 0;
+    u8* mapped_ptr = nullptr;
 };

src/video_core/renderer_opengl/maxwell_to_gl.h

@@ -91,6 +91,8 @@ inline GLenum PrimitiveTopology(Maxwell::PrimitiveTopology topology) {
     switch (topology) {
     case Maxwell::PrimitiveTopology::Points:
         return GL_POINTS;
+    case Maxwell::PrimitiveTopology::LineStrip:
+        return GL_LINE_STRIP;
     case Maxwell::PrimitiveTopology::Triangles:
         return GL_TRIANGLES;
     case Maxwell::PrimitiveTopology::TriangleStrip: