bcat/backend: Make formatting of passphrase consistent in NullBackend::SetPassphrase()

Aligns the '=' to be consistent with the rest of the logs within this
source file.

.ci/scripts/linux/docker.sh

@@ -5,7 +5,7 @@ cd /yuzu
 ccache -s
 
 mkdir build || true && cd build
-cmake .. -G Ninja -DDISPLAY_VERSION=$1 -DYUZU_USE_BUNDLED_UNICORN=ON -DYUZU_USE_QT_WEB_ENGINE=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=/usr/lib/ccache/gcc -DCMAKE_CXX_COMPILER=/usr/lib/ccache/g++ -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${ENABLE_COMPATIBILITY_REPORTING:-"OFF"} -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DUSE_DISCORD_PRESENCE=ON
+cmake .. -G Ninja -DDISPLAY_VERSION=$1 -DYUZU_USE_BUNDLED_UNICORN=ON -DYUZU_USE_QT_WEB_ENGINE=ON -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=/usr/lib/ccache/gcc -DCMAKE_CXX_COMPILER=/usr/lib/ccache/g++ -DYUZU_ENABLE_COMPATIBILITY_REPORTING=${ENABLE_COMPATIBILITY_REPORTING:-"OFF"} -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DUSE_DISCORD_PRESENCE=ON -DENABLE_VULKAN=No
 
 ninja
 

.ci/scripts/windows/docker.sh

@@ -13,7 +13,7 @@ echo '' >> /bin/cmd
 chmod +x /bin/cmd
 
 mkdir build || true && cd build
-cmake .. -G Ninja -DDISPLAY_VERSION=$1 -DCMAKE_TOOLCHAIN_FILE="$(pwd)/../CMakeModules/MinGWCross.cmake" -DUSE_CCACHE=ON -DYUZU_USE_BUNDLED_UNICORN=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DCMAKE_BUILD_TYPE=Release
+cmake .. -G Ninja -DDISPLAY_VERSION=$1 -DCMAKE_TOOLCHAIN_FILE="$(pwd)/../CMakeModules/MinGWCross.cmake" -DUSE_CCACHE=ON -DYUZU_USE_BUNDLED_UNICORN=ON -DENABLE_COMPATIBILITY_LIST_DOWNLOAD=ON -DCMAKE_BUILD_TYPE=Release -DENABLE_VULKAN=No
 ninja
 
 # Clean up the dirty hacks

dist/qt_themes/colorful/style.qrc

@@ -10,6 +10,6 @@
         <file alias="256x256/plus_folder.png">icons/256x256/plus_folder.png</file>
     </qresource>
     <qresource prefix="colorful">
-        <file>style.qss</file>
+        <file alias="style.qss">../default/style.qss</file>
     </qresource>
 </RCC>

dist/qt_themes/colorful/style.qss

@@ -1,4 +0,0 @@
-/*
-    This file is intentionally left blank.
-    We do not want to apply any stylesheet for colorful, only icons.
-*/

dist/qt_themes/default/default.qrc

@@ -1,25 +1,18 @@
 <RCC>
     <qresource prefix="icons/default">
         <file alias="index.theme">icons/index.theme</file>
-      
         <file alias="16x16/checked.png">icons/16x16/checked.png</file>
-
         <file alias="16x16/failed.png">icons/16x16/failed.png</file>
-
         <file alias="16x16/lock.png">icons/16x16/lock.png</file>
-
         <file alias="48x48/bad_folder.png">icons/48x48/bad_folder.png</file>
-      
         <file alias="48x48/chip.png">icons/48x48/chip.png</file>
-
         <file alias="48x48/folder.png">icons/48x48/folder.png</file>
-
         <file alias="48x48/plus.png">icons/48x48/plus.png</file>
-      
         <file alias="48x48/sd_card.png">icons/48x48/sd_card.png</file>
-      
         <file alias="256x256/yuzu.png">icons/256x256/yuzu.png</file>
-
         <file alias="256x256/plus_folder.png">icons/256x256/plus_folder.png</file>
     </qresource>
+    <qresource prefix="default">
+        <file>style.qss</file>
+    </qresource>
 </RCC>

dist/qt_themes/default/style.qss

@@ -0,0 +1,35 @@
+QPushButton#TogglableStatusBarButton {
+    color: #959595;
+    border: 1px solid transparent;
+    background-color: transparent;
+    padding: 0px 3px 0px 3px;
+    text-align: center;
+}
+
+QPushButton#TogglableStatusBarButton:checked {
+    color: #000000;
+}
+
+QPushButton#TogglableStatusBarButton:hover {
+    border: 1px solid #76797C;
+}
+
+QPushButton#RendererStatusBarButton {
+    color: #656565;
+    border: 1px solid transparent;
+    background-color: transparent;
+    padding: 0px 3px 0px 3px;
+    text-align: center;
+}
+
+QPushButton#RendererStatusBarButton:hover {
+    border: 1px solid #76797C;
+}
+
+QPushButton#RendererStatusBarButton:checked {
+    color: #e85c00;
+}
+
+QPushButton#RendererStatusBarButton:!checked{
+    color: #0066ff;
+}

dist/qt_themes/qdarkstyle/style.qss

@@ -1236,3 +1236,41 @@ QToolButton:disabled,
 QPlainTextEdit:disabled {
     background-color: #2b2e31;
 }
+
+QPushButton#TogglableStatusBarButton {
+    min-width: 0px;
+    color: #656565;
+    border: 1px solid transparent;
+    background-color: transparent;
+    padding: 0px 3px 0px 3px;
+    text-align: center;
+}
+
+QPushButton#TogglableStatusBarButton:checked {
+    color: #ffffff;
+}
+
+QPushButton#TogglableStatusBarButton:hover {
+    border: 1px solid #76797C;
+}
+
+QPushButton#RendererStatusBarButton {
+    min-width: 0px;
+    color: #656565;
+    border: 1px solid transparent;
+    background-color: transparent;
+    padding: 0px 3px 0px 3px;
+    text-align: center;
+}
+
+QPushButton#RendererStatusBarButton:hover {
+    border: 1px solid #76797C;
+}
+
+QPushButton#RendererStatusBarButton:checked {
+    color: #e85c00;
+}
+
+QPushButton#RendererStatusBarButton:!checked{
+   color: #00ccdd;
+}

src/core/CMakeLists.txt

@@ -15,14 +15,14 @@ add_library(core STATIC
     constants.h
     core.cpp
     core.h
-    core_cpu.cpp
-    core_cpu.h
+    core_manager.cpp
+    core_manager.h
     core_timing.cpp
     core_timing.h
     core_timing_util.cpp
     core_timing_util.h
-    cpu_core_manager.cpp
-    cpu_core_manager.h
+    cpu_manager.cpp
+    cpu_manager.h
     crypto/aes_util.cpp
     crypto/aes_util.h
     crypto/encryption_layer.cpp
@@ -158,6 +158,8 @@ add_library(core STATIC
     hle/kernel/mutex.h
     hle/kernel/object.cpp
     hle/kernel/object.h
+    hle/kernel/physical_core.cpp
+    hle/kernel/physical_core.h
     hle/kernel/process.cpp
     hle/kernel/process.h
     hle/kernel/process_capability.cpp

src/core/arm/dynarmic/arm_dynarmic.cpp

@@ -10,11 +10,12 @@
 #include "common/microprofile.h"
 #include "core/arm/dynarmic/arm_dynarmic.h"
 #include "core/core.h"
-#include "core/core_cpu.h"
+#include "core/core_manager.h"
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
 #include "core/gdbstub/gdbstub.h"
 #include "core/hle/kernel/process.h"
+#include "core/hle/kernel/scheduler.h"
 #include "core/hle/kernel/svc.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
@@ -87,7 +88,7 @@ public:
             if (GDBStub::IsServerEnabled()) {
                 parent.jit->HaltExecution();
                 parent.SetPC(pc);
-                Kernel::Thread* thread = Kernel::GetCurrentThread();
+                Kernel::Thread* const thread = parent.system.CurrentScheduler().GetCurrentThread();
                 parent.SaveContext(thread->GetContext());
                 GDBStub::Break();
                 GDBStub::SendTrap(thread, 5);

src/core/arm/exclusive_monitor.cpp

@@ -2,10 +2,24 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#ifdef ARCHITECTURE_x86_64
+#include "core/arm/dynarmic/arm_dynarmic.h"
+#endif
 #include "core/arm/exclusive_monitor.h"
+#include "core/memory.h"
 
 namespace Core {
 
 ExclusiveMonitor::~ExclusiveMonitor() = default;
 
+std::unique_ptr<Core::ExclusiveMonitor> MakeExclusiveMonitor(Memory::Memory& memory,
+                                                             std::size_t num_cores) {
+#ifdef ARCHITECTURE_x86_64
+    return std::make_unique<Core::DynarmicExclusiveMonitor>(memory, num_cores);
+#else
+    // TODO(merry): Passthrough exclusive monitor
+    return nullptr;
+#endif
+}
+
 } // namespace Core

src/core/arm/exclusive_monitor.h

@@ -4,8 +4,14 @@
 
 #pragma once
 
+#include <memory>
+
 #include "common/common_types.h"
 
+namespace Memory {
+class Memory;
+}
+
 namespace Core {
 
 class ExclusiveMonitor {
@@ -22,4 +28,7 @@ public:
     virtual bool ExclusiveWrite128(std::size_t core_index, VAddr vaddr, u128 value) = 0;
 };
 
+std::unique_ptr<Core::ExclusiveMonitor> MakeExclusiveMonitor(Memory::Memory& memory,
+                                                             std::size_t num_cores);
+
 } // namespace Core

src/core/arm/unicorn/arm_unicorn.cpp

@@ -9,6 +9,7 @@
 #include "core/arm/unicorn/arm_unicorn.h"
 #include "core/core.h"
 #include "core/core_timing.h"
+#include "core/hle/kernel/scheduler.h"
 #include "core/hle/kernel/svc.h"
 
 namespace Core {
@@ -177,7 +178,7 @@ void ARM_Unicorn::ExecuteInstructions(std::size_t num_instructions) {
             uc_reg_write(uc, UC_ARM64_REG_PC, &last_bkpt.address);
         }
 
-        Kernel::Thread* thread = Kernel::GetCurrentThread();
+        Kernel::Thread* const thread = system.CurrentScheduler().GetCurrentThread();
         SaveContext(thread->GetContext());
         if (last_bkpt_hit || GDBStub::IsMemoryBreak() || GDBStub::GetCpuStepFlag()) {
             last_bkpt_hit = false;

src/core/core.cpp

@@ -11,9 +11,9 @@
 #include "common/string_util.h"
 #include "core/arm/exclusive_monitor.h"
 #include "core/core.h"
-#include "core/core_cpu.h"
+#include "core/core_manager.h"
 #include "core/core_timing.h"
-#include "core/cpu_core_manager.h"
+#include "core/cpu_manager.h"
 #include "core/file_sys/bis_factory.h"
 #include "core/file_sys/card_image.h"
 #include "core/file_sys/mode.h"
@@ -28,6 +28,7 @@
 #include "core/hardware_interrupt_manager.h"
 #include "core/hle/kernel/client_port.h"
 #include "core/hle/kernel/kernel.h"
+#include "core/hle/kernel/physical_core.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/scheduler.h"
 #include "core/hle/kernel/thread.h"
@@ -113,16 +114,25 @@ FileSys::VirtualFile GetGameFileFromPath(const FileSys::VirtualFilesystem& vfs,
 struct System::Impl {
     explicit Impl(System& system)
         : kernel{system}, fs_controller{system}, memory{system},
-          cpu_core_manager{system}, reporter{system}, applet_manager{system} {}
+          cpu_manager{system}, reporter{system}, applet_manager{system} {}
 
-    Cpu& CurrentCpuCore() {
-        return cpu_core_manager.GetCurrentCore();
+    CoreManager& CurrentCoreManager() {
+        return cpu_manager.GetCurrentCoreManager();
+    }
+
+    Kernel::PhysicalCore& CurrentPhysicalCore() {
+        const auto index = cpu_manager.GetActiveCoreIndex();
+        return kernel.PhysicalCore(index);
+    }
+
+    Kernel::PhysicalCore& GetPhysicalCore(std::size_t index) {
+        return kernel.PhysicalCore(index);
     }
 
     ResultStatus RunLoop(bool tight_loop) {
         status = ResultStatus::Success;
 
-        cpu_core_manager.RunLoop(tight_loop);
+        cpu_manager.RunLoop(tight_loop);
 
         return status;
     }
@@ -131,8 +141,8 @@ struct System::Impl {
         LOG_DEBUG(HW_Memory, "initialized OK");
 
         core_timing.Initialize();
-        cpu_core_manager.Initialize();
         kernel.Initialize();
+        cpu_manager.Initialize();
 
         const auto current_time = std::chrono::duration_cast<std::chrono::seconds>(
             std::chrono::system_clock::now().time_since_epoch());
@@ -205,7 +215,6 @@ struct System::Impl {
         // Main process has been loaded and been made current.
         // Begin GPU and CPU execution.
         gpu_core->Start();
-        cpu_core_manager.StartThreads();
 
         // Initialize cheat engine
         if (cheat_engine) {
@@ -259,7 +268,9 @@ struct System::Impl {
         is_powered_on = false;
         exit_lock = false;
 
-        gpu_core->WaitIdle();
+        if (gpu_core) {
+            gpu_core->WaitIdle();
+        }
 
         // Shutdown emulation session
         renderer.reset();
@@ -272,7 +283,7 @@ struct System::Impl {
         gpu_core.reset();
 
         // Close all CPU/threading state
-        cpu_core_manager.Shutdown();
+        cpu_manager.Shutdown();
 
         // Shutdown kernel and core timing
         kernel.Shutdown();
@@ -342,7 +353,7 @@ struct System::Impl {
     std::unique_ptr<Tegra::GPU> gpu_core;
     std::unique_ptr<Hardware::InterruptManager> interrupt_manager;
     Memory::Memory memory;
-    CpuCoreManager cpu_core_manager;
+    CpuManager cpu_manager;
     bool is_powered_on = false;
     bool exit_lock = false;
 
@@ -377,12 +388,12 @@ struct System::Impl {
 System::System() : impl{std::make_unique<Impl>(*this)} {}
 System::~System() = default;
 
-Cpu& System::CurrentCpuCore() {
-    return impl->CurrentCpuCore();
+CoreManager& System::CurrentCoreManager() {
+    return impl->CurrentCoreManager();
 }
 
-const Cpu& System::CurrentCpuCore() const {
-    return impl->CurrentCpuCore();
+const CoreManager& System::CurrentCoreManager() const {
+    return impl->CurrentCoreManager();
 }
 
 System::ResultStatus System::RunLoop(bool tight_loop) {
@@ -394,7 +405,7 @@ System::ResultStatus System::SingleStep() {
 }
 
 void System::InvalidateCpuInstructionCaches() {
-    impl->cpu_core_manager.InvalidateAllInstructionCaches();
+    impl->kernel.InvalidateAllInstructionCaches();
 }
 
 System::ResultStatus System::Load(Frontend::EmuWindow& emu_window, const std::string& filepath) {
@@ -406,13 +417,11 @@ bool System::IsPoweredOn() const {
 }
 
 void System::PrepareReschedule() {
-    CurrentCpuCore().PrepareReschedule();
+    impl->CurrentPhysicalCore().Stop();
 }
 
 void System::PrepareReschedule(const u32 core_index) {
-    if (core_index < GlobalScheduler().CpuCoresCount()) {
-        CpuCore(core_index).PrepareReschedule();
-    }
+    impl->kernel.PrepareReschedule(core_index);
 }
 
 PerfStatsResults System::GetAndResetPerfStats() {
@@ -428,31 +437,31 @@ const TelemetrySession& System::TelemetrySession() const {
 }
 
 ARM_Interface& System::CurrentArmInterface() {
-    return CurrentCpuCore().ArmInterface();
+    return impl->CurrentPhysicalCore().ArmInterface();
 }
 
 const ARM_Interface& System::CurrentArmInterface() const {
-    return CurrentCpuCore().ArmInterface();
+    return impl->CurrentPhysicalCore().ArmInterface();
 }
 
 std::size_t System::CurrentCoreIndex() const {
-    return CurrentCpuCore().CoreIndex();
+    return impl->cpu_manager.GetActiveCoreIndex();
 }
 
 Kernel::Scheduler& System::CurrentScheduler() {
-    return CurrentCpuCore().Scheduler();
+    return impl->CurrentPhysicalCore().Scheduler();
 }
 
 const Kernel::Scheduler& System::CurrentScheduler() const {
-    return CurrentCpuCore().Scheduler();
+    return impl->CurrentPhysicalCore().Scheduler();
 }
 
 Kernel::Scheduler& System::Scheduler(std::size_t core_index) {
-    return CpuCore(core_index).Scheduler();
+    return impl->GetPhysicalCore(core_index).Scheduler();
 }
 
 const Kernel::Scheduler& System::Scheduler(std::size_t core_index) const {
-    return CpuCore(core_index).Scheduler();
+    return impl->GetPhysicalCore(core_index).Scheduler();
 }
 
 /// Gets the global scheduler
@@ -474,28 +483,28 @@ const Kernel::Process* System::CurrentProcess() const {
 }
 
 ARM_Interface& System::ArmInterface(std::size_t core_index) {
-    return CpuCore(core_index).ArmInterface();
+    return impl->GetPhysicalCore(core_index).ArmInterface();
 }
 
 const ARM_Interface& System::ArmInterface(std::size_t core_index) const {
-    return CpuCore(core_index).ArmInterface();
+    return impl->GetPhysicalCore(core_index).ArmInterface();
 }
 
-Cpu& System::CpuCore(std::size_t core_index) {
-    return impl->cpu_core_manager.GetCore(core_index);
+CoreManager& System::GetCoreManager(std::size_t core_index) {
+    return impl->cpu_manager.GetCoreManager(core_index);
 }
 
-const Cpu& System::CpuCore(std::size_t core_index) const {
+const CoreManager& System::GetCoreManager(std::size_t core_index) const {
     ASSERT(core_index < NUM_CPU_CORES);
-    return impl->cpu_core_manager.GetCore(core_index);
+    return impl->cpu_manager.GetCoreManager(core_index);
 }
 
 ExclusiveMonitor& System::Monitor() {
-    return impl->cpu_core_manager.GetExclusiveMonitor();
+    return impl->kernel.GetExclusiveMonitor();
 }
 
 const ExclusiveMonitor& System::Monitor() const {
-    return impl->cpu_core_manager.GetExclusiveMonitor();
+    return impl->kernel.GetExclusiveMonitor();
 }
 
 Memory::Memory& System::Memory() {

src/core/core.h

@@ -93,7 +93,7 @@ class Memory;
 namespace Core {
 
 class ARM_Interface;
-class Cpu;
+class CoreManager;
 class ExclusiveMonitor;
 class FrameLimiter;
 class PerfStats;
@@ -218,10 +218,10 @@ public:
     const ARM_Interface& ArmInterface(std::size_t core_index) const;
 
     /// Gets a CPU interface to the CPU core with the specified index
-    Cpu& CpuCore(std::size_t core_index);
+    CoreManager& GetCoreManager(std::size_t core_index);
 
     /// Gets a CPU interface to the CPU core with the specified index
-    const Cpu& CpuCore(std::size_t core_index) const;
+    const CoreManager& GetCoreManager(std::size_t core_index) const;
 
     /// Gets a reference to the exclusive monitor
     ExclusiveMonitor& Monitor();
@@ -364,10 +364,10 @@ private:
     System();
 
     /// Returns the currently running CPU core
-    Cpu& CurrentCpuCore();
+    CoreManager& CurrentCoreManager();
 
     /// Returns the currently running CPU core
-    const Cpu& CurrentCpuCore() const;
+    const CoreManager& CurrentCoreManager() const;
 
     /**
      * Initialize the emulated system.

src/core/core_cpu.cpp

@@ -1,127 +0,0 @@
-// Copyright 2018 yuzu emulator team
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <condition_variable>
-#include <mutex>
-
-#include "common/logging/log.h"
-#ifdef ARCHITECTURE_x86_64
-#include "core/arm/dynarmic/arm_dynarmic.h"
-#endif
-#include "core/arm/exclusive_monitor.h"
-#include "core/arm/unicorn/arm_unicorn.h"
-#include "core/core.h"
-#include "core/core_cpu.h"
-#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 {
-
-void CpuBarrier::NotifyEnd() {
-    std::unique_lock lock{mutex};
-    end = true;
-    condition.notify_all();
-}
-
-bool CpuBarrier::Rendezvous() {
-    if (!Settings::values.use_multi_core) {
-        // Meaningless when running in single-core mode
-        return true;
-    }
-
-    if (!end) {
-        std::unique_lock lock{mutex};
-
-        --cores_waiting;
-        if (!cores_waiting) {
-            cores_waiting = NUM_CPU_CORES;
-            condition.notify_all();
-            return true;
-        }
-
-        condition.wait(lock);
-        return true;
-    }
-
-    return false;
-}
-
-Cpu::Cpu(System& system, ExclusiveMonitor& exclusive_monitor, CpuBarrier& cpu_barrier,
-         std::size_t core_index)
-    : cpu_barrier{cpu_barrier}, global_scheduler{system.GlobalScheduler()},
-      core_timing{system.CoreTiming()}, core_index{core_index} {
-#ifdef ARCHITECTURE_x86_64
-    arm_interface = std::make_unique<ARM_Dynarmic>(system, exclusive_monitor, core_index);
-#else
-    arm_interface = std::make_unique<ARM_Unicorn>(system);
-    LOG_WARNING(Core, "CPU JIT requested, but Dynarmic not available");
-#endif
-
-    scheduler = std::make_unique<Kernel::Scheduler>(system, *arm_interface, core_index);
-}
-
-Cpu::~Cpu() = default;
-
-std::unique_ptr<ExclusiveMonitor> Cpu::MakeExclusiveMonitor(
-    [[maybe_unused]] Memory::Memory& memory, [[maybe_unused]] std::size_t num_cores) {
-#ifdef ARCHITECTURE_x86_64
-    return std::make_unique<DynarmicExclusiveMonitor>(memory, num_cores);
-#else
-    // TODO(merry): Passthrough exclusive monitor
-    return nullptr;
-#endif
-}
-
-void Cpu::RunLoop(bool tight_loop) {
-    // Wait for all other CPU cores to complete the previous slice, such that they run in lock-step
-    if (!cpu_barrier.Rendezvous()) {
-        // If rendezvous failed, session has been killed
-        return;
-    }
-
-    Reschedule();
-
-    // If we don't have a currently active thread then don't execute instructions,
-    // instead advance to the next event and try to yield to the next thread
-    if (Kernel::GetCurrentThread() == nullptr) {
-        LOG_TRACE(Core, "Core-{} idling", core_index);
-        core_timing.Idle();
-    } else {
-        if (tight_loop) {
-            arm_interface->Run();
-        } else {
-            arm_interface->Step();
-        }
-        // We are stopping a run, exclusive state must be cleared
-        arm_interface->ClearExclusiveState();
-    }
-    core_timing.Advance();
-
-    Reschedule();
-}
-
-void Cpu::SingleStep() {
-    return RunLoop(false);
-}
-
-void Cpu::PrepareReschedule() {
-    arm_interface->PrepareReschedule();
-}
-
-void Cpu::Reschedule() {
-    // Lock the global kernel mutex when we manipulate the HLE state
-    std::lock_guard lock(HLE::g_hle_lock);
-
-    global_scheduler.SelectThread(core_index);
-    scheduler->TryDoContextSwitch();
-}
-
-void Cpu::Shutdown() {
-    scheduler->Shutdown();
-}
-
-} // namespace Core

src/core/core_cpu.h

@@ -1,120 +0,0 @@
-// Copyright 2018 yuzu emulator team
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <atomic>
-#include <condition_variable>
-#include <cstddef>
-#include <memory>
-#include <mutex>
-#include "common/common_types.h"
-
-namespace Kernel {
-class GlobalScheduler;
-class Scheduler;
-} // namespace Kernel
-
-namespace Core {
-class System;
-}
-
-namespace Core::Timing {
-class CoreTiming;
-}
-
-namespace Memory {
-class Memory;
-}
-
-namespace Core {
-
-class ARM_Interface;
-class ExclusiveMonitor;
-
-constexpr unsigned NUM_CPU_CORES{4};
-
-class CpuBarrier {
-public:
-    bool IsAlive() const {
-        return !end;
-    }
-
-    void NotifyEnd();
-
-    bool Rendezvous();
-
-private:
-    unsigned cores_waiting{NUM_CPU_CORES};
-    std::mutex mutex;
-    std::condition_variable condition;
-    std::atomic<bool> end{};
-};
-
-class Cpu {
-public:
-    Cpu(System& system, ExclusiveMonitor& exclusive_monitor, CpuBarrier& cpu_barrier,
-        std::size_t core_index);
-    ~Cpu();
-
-    void RunLoop(bool tight_loop = true);
-
-    void SingleStep();
-
-    void PrepareReschedule();
-
-    ARM_Interface& ArmInterface() {
-        return *arm_interface;
-    }
-
-    const ARM_Interface& ArmInterface() const {
-        return *arm_interface;
-    }
-
-    Kernel::Scheduler& Scheduler() {
-        return *scheduler;
-    }
-
-    const Kernel::Scheduler& Scheduler() const {
-        return *scheduler;
-    }
-
-    bool IsMainCore() const {
-        return core_index == 0;
-    }
-
-    std::size_t CoreIndex() const {
-        return core_index;
-    }
-
-    void Shutdown();
-
-    /**
-     * Creates an exclusive monitor to handle exclusive reads/writes.
-     *
-     * @param memory The current memory subsystem that the monitor may wish
-     *               to keep track of.
-     *
-     * @param num_cores The number of cores to assume about the CPU.
-     *
-     * @returns The constructed exclusive monitor instance, or nullptr if the current
-     *          CPU backend is unable to use an exclusive monitor.
-     */
-    static std::unique_ptr<ExclusiveMonitor> MakeExclusiveMonitor(Memory::Memory& memory,
-                                                                  std::size_t num_cores);
-
-private:
-    void Reschedule();
-
-    std::unique_ptr<ARM_Interface> arm_interface;
-    CpuBarrier& cpu_barrier;
-    Kernel::GlobalScheduler& global_scheduler;
-    std::unique_ptr<Kernel::Scheduler> scheduler;
-    Timing::CoreTiming& core_timing;
-
-    std::atomic<bool> reschedule_pending = false;
-    std::size_t core_index;
-};
-
-} // namespace Core

src/core/core_manager.cpp

@@ -0,0 +1,70 @@
+// Copyright 2018 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <condition_variable>
+#include <mutex>
+
+#include "common/logging/log.h"
+#ifdef ARCHITECTURE_x86_64
+#include "core/arm/dynarmic/arm_dynarmic.h"
+#endif
+#include "core/arm/exclusive_monitor.h"
+#include "core/arm/unicorn/arm_unicorn.h"
+#include "core/core.h"
+#include "core/core_manager.h"
+#include "core/core_timing.h"
+#include "core/hle/kernel/kernel.h"
+#include "core/hle/kernel/physical_core.h"
+#include "core/hle/kernel/scheduler.h"
+#include "core/hle/kernel/thread.h"
+#include "core/hle/lock.h"
+#include "core/settings.h"
+
+namespace Core {
+
+CoreManager::CoreManager(System& system, std::size_t core_index)
+    : global_scheduler{system.GlobalScheduler()}, physical_core{system.Kernel().PhysicalCore(
+                                                      core_index)},
+      core_timing{system.CoreTiming()}, core_index{core_index} {}
+
+CoreManager::~CoreManager() = default;
+
+void CoreManager::RunLoop(bool tight_loop) {
+    Reschedule();
+
+    // If we don't have a currently active thread then don't execute instructions,
+    // instead advance to the next event and try to yield to the next thread
+    if (Kernel::GetCurrentThread() == nullptr) {
+        LOG_TRACE(Core, "Core-{} idling", core_index);
+        core_timing.Idle();
+    } else {
+        if (tight_loop) {
+            physical_core.Run();
+        } else {
+            physical_core.Step();
+        }
+    }
+    core_timing.Advance();
+
+    Reschedule();
+}
+
+void CoreManager::SingleStep() {
+    return RunLoop(false);
+}
+
+void CoreManager::PrepareReschedule() {
+    physical_core.Stop();
+}
+
+void CoreManager::Reschedule() {
+    // Lock the global kernel mutex when we manipulate the HLE state
+    std::lock_guard lock(HLE::g_hle_lock);
+
+    global_scheduler.SelectThread(core_index);
+
+    physical_core.Scheduler().TryDoContextSwitch();
+}
+
+} // namespace Core

src/core/core_manager.h

@@ -0,0 +1,63 @@
+// Copyright 2018 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <atomic>
+#include <cstddef>
+#include <memory>
+#include "common/common_types.h"
+
+namespace Kernel {
+class GlobalScheduler;
+class PhysicalCore;
+} // namespace Kernel
+
+namespace Core {
+class System;
+}
+
+namespace Core::Timing {
+class CoreTiming;
+}
+
+namespace Memory {
+class Memory;
+}
+
+namespace Core {
+
+constexpr unsigned NUM_CPU_CORES{4};
+
+class CoreManager {
+public:
+    CoreManager(System& system, std::size_t core_index);
+    ~CoreManager();
+
+    void RunLoop(bool tight_loop = true);
+
+    void SingleStep();
+
+    void PrepareReschedule();
+
+    bool IsMainCore() const {
+        return core_index == 0;
+    }
+
+    std::size_t CoreIndex() const {
+        return core_index;
+    }
+
+private:
+    void Reschedule();
+
+    Kernel::GlobalScheduler& global_scheduler;
+    Kernel::PhysicalCore& physical_core;
+    Timing::CoreTiming& core_timing;
+
+    std::atomic<bool> reschedule_pending = false;
+    std::size_t core_index;
+};
+
+} // namespace Core

src/core/cpu_core_manager.cpp

@@ -1,152 +0,0 @@
-// Copyright 2018 yuzu emulator team
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include "common/assert.h"
-#include "core/arm/exclusive_monitor.h"
-#include "core/core.h"
-#include "core/core_cpu.h"
-#include "core/core_timing.h"
-#include "core/cpu_core_manager.h"
-#include "core/gdbstub/gdbstub.h"
-#include "core/settings.h"
-
-namespace Core {
-namespace {
-void RunCpuCore(const System& system, Cpu& cpu_state) {
-    while (system.IsPoweredOn()) {
-        cpu_state.RunLoop(true);
-    }
-}
-} // Anonymous namespace
-
-CpuCoreManager::CpuCoreManager(System& system) : system{system} {}
-CpuCoreManager::~CpuCoreManager() = default;
-
-void CpuCoreManager::Initialize() {
-    barrier = std::make_unique<CpuBarrier>();
-    exclusive_monitor = Cpu::MakeExclusiveMonitor(system.Memory(), cores.size());
-
-    for (std::size_t index = 0; index < cores.size(); ++index) {
-        cores[index] = std::make_unique<Cpu>(system, *exclusive_monitor, *barrier, index);
-    }
-}
-
-void CpuCoreManager::StartThreads() {
-    // Create threads for CPU cores 1-3, and build thread_to_cpu map
-    // CPU core 0 is run on the main thread
-    thread_to_cpu[std::this_thread::get_id()] = cores[0].get();
-    if (!Settings::values.use_multi_core) {
-        return;
-    }
-
-    for (std::size_t index = 0; index < core_threads.size(); ++index) {
-        core_threads[index] = std::make_unique<std::thread>(RunCpuCore, std::cref(system),
-                                                            std::ref(*cores[index + 1]));
-        thread_to_cpu[core_threads[index]->get_id()] = cores[index + 1].get();
-    }
-}
-
-void CpuCoreManager::Shutdown() {
-    barrier->NotifyEnd();
-    if (Settings::values.use_multi_core) {
-        for (auto& thread : core_threads) {
-            thread->join();
-            thread.reset();
-        }
-    }
-
-    thread_to_cpu.clear();
-    for (auto& cpu_core : cores) {
-        cpu_core->Shutdown();
-        cpu_core.reset();
-    }
-
-    exclusive_monitor.reset();
-    barrier.reset();
-}
-
-Cpu& CpuCoreManager::GetCore(std::size_t index) {
-    return *cores.at(index);
-}
-
-const Cpu& CpuCoreManager::GetCore(std::size_t index) const {
-    return *cores.at(index);
-}
-
-ExclusiveMonitor& CpuCoreManager::GetExclusiveMonitor() {
-    return *exclusive_monitor;
-}
-
-const ExclusiveMonitor& CpuCoreManager::GetExclusiveMonitor() const {
-    return *exclusive_monitor;
-}
-
-Cpu& CpuCoreManager::GetCurrentCore() {
-    if (Settings::values.use_multi_core) {
-        const auto& search = thread_to_cpu.find(std::this_thread::get_id());
-        ASSERT(search != thread_to_cpu.end());
-        ASSERT(search->second);
-        return *search->second;
-    }
-
-    // Otherwise, use single-threaded mode active_core variable
-    return *cores[active_core];
-}
-
-const Cpu& CpuCoreManager::GetCurrentCore() const {
-    if (Settings::values.use_multi_core) {
-        const auto& search = thread_to_cpu.find(std::this_thread::get_id());
-        ASSERT(search != thread_to_cpu.end());
-        ASSERT(search->second);
-        return *search->second;
-    }
-
-    // Otherwise, use single-threaded mode active_core variable
-    return *cores[active_core];
-}
-
-void CpuCoreManager::RunLoop(bool tight_loop) {
-    // Update thread_to_cpu in case Core 0 is run from a different host thread
-    thread_to_cpu[std::this_thread::get_id()] = cores[0].get();
-
-    if (GDBStub::IsServerEnabled()) {
-        GDBStub::HandlePacket();
-
-        // If the loop is halted and we want to step, use a tiny (1) number of instructions to
-        // execute. Otherwise, get out of the loop function.
-        if (GDBStub::GetCpuHaltFlag()) {
-            if (GDBStub::GetCpuStepFlag()) {
-                tight_loop = false;
-            } else {
-                return;
-            }
-        }
-    }
-
-    auto& core_timing = system.CoreTiming();
-    core_timing.ResetRun();
-    bool keep_running{};
-    do {
-        keep_running = false;
-        for (active_core = 0; active_core < NUM_CPU_CORES; ++active_core) {
-            core_timing.SwitchContext(active_core);
-            if (core_timing.CanCurrentContextRun()) {
-                cores[active_core]->RunLoop(tight_loop);
-            }
-            keep_running |= core_timing.CanCurrentContextRun();
-        }
-    } while (keep_running);
-
-    if (GDBStub::IsServerEnabled()) {
-        GDBStub::SetCpuStepFlag(false);
-    }
-}
-
-void CpuCoreManager::InvalidateAllInstructionCaches() {
-    for (auto& cpu : cores) {
-        cpu->ArmInterface().ClearInstructionCache();
-    }
-}
-
-} // namespace Core

src/core/cpu_core_manager.h

@@ -1,62 +0,0 @@
-// Copyright 2018 yuzu emulator team
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <array>
-#include <map>
-#include <memory>
-#include <thread>
-
-namespace Core {
-
-class Cpu;
-class CpuBarrier;
-class ExclusiveMonitor;
-class System;
-
-class CpuCoreManager {
-public:
-    explicit CpuCoreManager(System& system);
-    CpuCoreManager(const CpuCoreManager&) = delete;
-    CpuCoreManager(CpuCoreManager&&) = delete;
-
-    ~CpuCoreManager();
-
-    CpuCoreManager& operator=(const CpuCoreManager&) = delete;
-    CpuCoreManager& operator=(CpuCoreManager&&) = delete;
-
-    void Initialize();
-    void StartThreads();
-    void Shutdown();
-
-    Cpu& GetCore(std::size_t index);
-    const Cpu& GetCore(std::size_t index) const;
-
-    Cpu& GetCurrentCore();
-    const Cpu& GetCurrentCore() const;
-
-    ExclusiveMonitor& GetExclusiveMonitor();
-    const ExclusiveMonitor& GetExclusiveMonitor() const;
-
-    void RunLoop(bool tight_loop);
-
-    void InvalidateAllInstructionCaches();
-
-private:
-    static constexpr std::size_t NUM_CPU_CORES = 4;
-
-    std::unique_ptr<ExclusiveMonitor> exclusive_monitor;
-    std::unique_ptr<CpuBarrier> barrier;
-    std::array<std::unique_ptr<Cpu>, NUM_CPU_CORES> cores;
-    std::array<std::unique_ptr<std::thread>, NUM_CPU_CORES - 1> core_threads;
-    std::size_t active_core{}; ///< Active core, only used in single thread mode
-
-    /// Map of guest threads to CPU cores
-    std::map<std::thread::id, Cpu*> thread_to_cpu;
-
-    System& system;
-};
-
-} // namespace Core

src/core/cpu_manager.cpp

@@ -0,0 +1,81 @@
+// Copyright 2018 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "core/arm/exclusive_monitor.h"
+#include "core/core.h"
+#include "core/core_manager.h"
+#include "core/core_timing.h"
+#include "core/cpu_manager.h"
+#include "core/gdbstub/gdbstub.h"
+
+namespace Core {
+
+CpuManager::CpuManager(System& system) : system{system} {}
+CpuManager::~CpuManager() = default;
+
+void CpuManager::Initialize() {
+    for (std::size_t index = 0; index < core_managers.size(); ++index) {
+        core_managers[index] = std::make_unique<CoreManager>(system, index);
+    }
+}
+
+void CpuManager::Shutdown() {
+    for (auto& cpu_core : core_managers) {
+        cpu_core.reset();
+    }
+}
+
+CoreManager& CpuManager::GetCoreManager(std::size_t index) {
+    return *core_managers.at(index);
+}
+
+const CoreManager& CpuManager::GetCoreManager(std::size_t index) const {
+    return *core_managers.at(index);
+}
+
+CoreManager& CpuManager::GetCurrentCoreManager() {
+    // Otherwise, use single-threaded mode active_core variable
+    return *core_managers[active_core];
+}
+
+const CoreManager& CpuManager::GetCurrentCoreManager() const {
+    // Otherwise, use single-threaded mode active_core variable
+    return *core_managers[active_core];
+}
+
+void CpuManager::RunLoop(bool tight_loop) {
+    if (GDBStub::IsServerEnabled()) {
+        GDBStub::HandlePacket();
+
+        // If the loop is halted and we want to step, use a tiny (1) number of instructions to
+        // execute. Otherwise, get out of the loop function.
+        if (GDBStub::GetCpuHaltFlag()) {
+            if (GDBStub::GetCpuStepFlag()) {
+                tight_loop = false;
+            } else {
+                return;
+            }
+        }
+    }
+
+    auto& core_timing = system.CoreTiming();
+    core_timing.ResetRun();
+    bool keep_running{};
+    do {
+        keep_running = false;
+        for (active_core = 0; active_core < NUM_CPU_CORES; ++active_core) {
+            core_timing.SwitchContext(active_core);
+            if (core_timing.CanCurrentContextRun()) {
+                core_managers[active_core]->RunLoop(tight_loop);
+            }
+            keep_running |= core_timing.CanCurrentContextRun();
+        }
+    } while (keep_running);
+
+    if (GDBStub::IsServerEnabled()) {
+        GDBStub::SetCpuStepFlag(false);
+    }
+}
+
+} // namespace Core

src/core/cpu_manager.h

@@ -0,0 +1,50 @@
+// Copyright 2018 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <memory>
+
+namespace Core {
+
+class CoreManager;
+class System;
+
+class CpuManager {
+public:
+    explicit CpuManager(System& system);
+    CpuManager(const CpuManager&) = delete;
+    CpuManager(CpuManager&&) = delete;
+
+    ~CpuManager();
+
+    CpuManager& operator=(const CpuManager&) = delete;
+    CpuManager& operator=(CpuManager&&) = delete;
+
+    void Initialize();
+    void Shutdown();
+
+    CoreManager& GetCoreManager(std::size_t index);
+    const CoreManager& GetCoreManager(std::size_t index) const;
+
+    CoreManager& GetCurrentCoreManager();
+    const CoreManager& GetCurrentCoreManager() const;
+
+    std::size_t GetActiveCoreIndex() const {
+        return active_core;
+    }
+
+    void RunLoop(bool tight_loop);
+
+private:
+    static constexpr std::size_t NUM_CPU_CORES = 4;
+
+    std::array<std::unique_ptr<CoreManager>, NUM_CPU_CORES> core_managers;
+    std::size_t active_core{}; ///< Active core, only used in single thread mode
+
+    System& system;
+};
+
+} // namespace Core

src/core/frontend/emu_window.h

@@ -75,6 +75,13 @@ public:
         return nullptr;
     }
 
+    /// Returns if window is shown (not minimized)
+    virtual bool IsShown() const = 0;
+
+    /// Retrieves Vulkan specific handlers from the window
+    virtual void RetrieveVulkanHandlers(void* get_instance_proc_addr, void* instance,
+                                        void* surface) const = 0;
+
     /**
      * Signal that a touch pressed event has occurred (e.g. mouse click pressed)
      * @param framebuffer_x Framebuffer x-coordinate that was pressed

src/core/frontend/input.h

@@ -15,6 +15,13 @@
 
 namespace Input {
 
+enum class AnalogDirection : u8 {
+    RIGHT,
+    LEFT,
+    UP,
+    DOWN,
+};
+
 /// An abstract class template for an input device (a button, an analog input, etc.).
 template <typename StatusType>
 class InputDevice {
@@ -23,6 +30,9 @@ public:
     virtual StatusType GetStatus() const {
         return {};
     }
+    virtual bool GetAnalogDirectionStatus(AnalogDirection direction) const {
+        return {};
+    }
 };
 
 /// An abstract class template for a factory that can create input devices.

src/core/gdbstub/gdbstub.cpp

@@ -35,7 +35,7 @@
 #include "common/swap.h"
 #include "core/arm/arm_interface.h"
 #include "core/core.h"
-#include "core/core_cpu.h"
+#include "core/core_manager.h"
 #include "core/gdbstub/gdbstub.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/scheduler.h"

src/core/hle/kernel/address_arbiter.cpp

@@ -8,7 +8,6 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "core/core.h"
-#include "core/core_cpu.h"
 #include "core/hle/kernel/address_arbiter.h"
 #include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/scheduler.h"

src/core/hle/kernel/hle_ipc.cpp

@@ -284,13 +284,18 @@ ResultCode HLERequestContext::WriteToOutgoingCommandBuffer(Thread& thread) {
 
 std::vector<u8> HLERequestContext::ReadBuffer(int buffer_index) const {
     std::vector<u8> buffer;
-    const bool is_buffer_a{BufferDescriptorA().size() && BufferDescriptorA()[buffer_index].Size()};
+    const bool is_buffer_a{BufferDescriptorA().size() > buffer_index &&
+                           BufferDescriptorA()[buffer_index].Size()};
     auto& memory = Core::System::GetInstance().Memory();
 
     if (is_buffer_a) {
+        ASSERT_MSG(BufferDescriptorA().size() > buffer_index,
+                   "BufferDescriptorA invalid buffer_index {}", buffer_index);
         buffer.resize(BufferDescriptorA()[buffer_index].Size());
         memory.ReadBlock(BufferDescriptorA()[buffer_index].Address(), buffer.data(), buffer.size());
     } else {
+        ASSERT_MSG(BufferDescriptorX().size() > buffer_index,
+                   "BufferDescriptorX invalid buffer_index {}", buffer_index);
         buffer.resize(BufferDescriptorX()[buffer_index].Size());
         memory.ReadBlock(BufferDescriptorX()[buffer_index].Address(), buffer.data(), buffer.size());
     }
@@ -305,7 +310,8 @@ std::size_t HLERequestContext::WriteBuffer(const void* buffer, std::size_t size,
         return 0;
     }
 
-    const bool is_buffer_b{BufferDescriptorB().size() && BufferDescriptorB()[buffer_index].Size()};
+    const bool is_buffer_b{BufferDescriptorB().size() > buffer_index &&
+                           BufferDescriptorB()[buffer_index].Size()};
     const std::size_t buffer_size{GetWriteBufferSize(buffer_index)};
     if (size > buffer_size) {
         LOG_CRITICAL(Core, "size ({:016X}) is greater than buffer_size ({:016X})", size,
@@ -315,8 +321,16 @@ std::size_t HLERequestContext::WriteBuffer(const void* buffer, std::size_t size,
 
     auto& memory = Core::System::GetInstance().Memory();
     if (is_buffer_b) {
+        ASSERT_MSG(BufferDescriptorB().size() > buffer_index,
+                   "BufferDescriptorB invalid buffer_index {}", buffer_index);
+        ASSERT_MSG(BufferDescriptorB()[buffer_index].Size() >= size,
+                   "BufferDescriptorB buffer_index {} is not large enough", buffer_index);
         memory.WriteBlock(BufferDescriptorB()[buffer_index].Address(), buffer, size);
     } else {
+        ASSERT_MSG(BufferDescriptorC().size() > buffer_index,
+                   "BufferDescriptorC invalid buffer_index {}", buffer_index);
+        ASSERT_MSG(BufferDescriptorC()[buffer_index].Size() >= size,
+                   "BufferDescriptorC buffer_index {} is not large enough", buffer_index);
         memory.WriteBlock(BufferDescriptorC()[buffer_index].Address(), buffer, size);
     }
 
@@ -324,15 +338,35 @@ std::size_t HLERequestContext::WriteBuffer(const void* buffer, std::size_t size,
 }
 
 std::size_t HLERequestContext::GetReadBufferSize(int buffer_index) const {
-    const bool is_buffer_a{BufferDescriptorA().size() && BufferDescriptorA()[buffer_index].Size()};
-    return is_buffer_a ? BufferDescriptorA()[buffer_index].Size()
-                       : BufferDescriptorX()[buffer_index].Size();
+    const bool is_buffer_a{BufferDescriptorA().size() > buffer_index &&
+                           BufferDescriptorA()[buffer_index].Size()};
+    if (is_buffer_a) {
+        ASSERT_MSG(BufferDescriptorA().size() > buffer_index,
+                   "BufferDescriptorA invalid buffer_index {}", buffer_index);
+        ASSERT_MSG(BufferDescriptorA()[buffer_index].Size() > 0,
+                   "BufferDescriptorA buffer_index {} is empty", buffer_index);
+        return BufferDescriptorA()[buffer_index].Size();
+    } else {
+        ASSERT_MSG(BufferDescriptorX().size() > buffer_index,
+                   "BufferDescriptorX invalid buffer_index {}", buffer_index);
+        ASSERT_MSG(BufferDescriptorX()[buffer_index].Size() > 0,
+                   "BufferDescriptorX buffer_index {} is empty", buffer_index);
+        return BufferDescriptorX()[buffer_index].Size();
+    }
 }
 
 std::size_t HLERequestContext::GetWriteBufferSize(int buffer_index) const {
-    const bool is_buffer_b{BufferDescriptorB().size() && BufferDescriptorB()[buffer_index].Size()};
-    return is_buffer_b ? BufferDescriptorB()[buffer_index].Size()
-                       : BufferDescriptorC()[buffer_index].Size();
+    const bool is_buffer_b{BufferDescriptorB().size() > buffer_index &&
+                           BufferDescriptorB()[buffer_index].Size()};
+    if (is_buffer_b) {
+        ASSERT_MSG(BufferDescriptorB().size() > buffer_index,
+                   "BufferDescriptorB invalid buffer_index {}", buffer_index);
+        return BufferDescriptorB()[buffer_index].Size();
+    } else {
+        ASSERT_MSG(BufferDescriptorC().size() > buffer_index,
+                   "BufferDescriptorC invalid buffer_index {}", buffer_index);
+        return BufferDescriptorC()[buffer_index].Size();
+    }
 }
 
 std::string HLERequestContext::Description() const {

src/core/hle/kernel/kernel.cpp

@@ -3,13 +3,15 @@
 // Refer to the license.txt file included.
 
 #include <atomic>
+#include <functional>
 #include <memory>
 #include <mutex>
 #include <utility>
 
 #include "common/assert.h"
 #include "common/logging/log.h"
-
+#include "core/arm/arm_interface.h"
+#include "core/arm/exclusive_monitor.h"
 #include "core/core.h"
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
@@ -17,6 +19,7 @@
 #include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/handle_table.h"
 #include "core/hle/kernel/kernel.h"
+#include "core/hle/kernel/physical_core.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/resource_limit.h"
 #include "core/hle/kernel/scheduler.h"
@@ -98,6 +101,7 @@ struct KernelCore::Impl {
     void Initialize(KernelCore& kernel) {
         Shutdown();
 
+        InitializePhysicalCores();
         InitializeSystemResourceLimit(kernel);
         InitializeThreads();
         InitializePreemption();
@@ -121,6 +125,21 @@ struct KernelCore::Impl {
         global_scheduler.Shutdown();
 
         named_ports.clear();
+
+        for (auto& core : cores) {
+            core.Shutdown();
+        }
+        cores.clear();
+
+        exclusive_monitor.reset();
+    }
+
+    void InitializePhysicalCores() {
+        exclusive_monitor =
+            Core::MakeExclusiveMonitor(system.Memory(), global_scheduler.CpuCoresCount());
+        for (std::size_t i = 0; i < global_scheduler.CpuCoresCount(); i++) {
+            cores.emplace_back(system, i, *exclusive_monitor);
+        }
     }
 
     // Creates the default system resource limit
@@ -186,6 +205,9 @@ struct KernelCore::Impl {
     /// the ConnectToPort SVC.
     NamedPortTable named_ports;
 
+    std::unique_ptr<Core::ExclusiveMonitor> exclusive_monitor;
+    std::vector<Kernel::PhysicalCore> cores;
+
     // System context
     Core::System& system;
 };
@@ -240,6 +262,34 @@ const Kernel::GlobalScheduler& KernelCore::GlobalScheduler() const {
     return impl->global_scheduler;
 }
 
+Kernel::PhysicalCore& KernelCore::PhysicalCore(std::size_t id) {
+    return impl->cores[id];
+}
+
+const Kernel::PhysicalCore& KernelCore::PhysicalCore(std::size_t id) const {
+    return impl->cores[id];
+}
+
+Core::ExclusiveMonitor& KernelCore::GetExclusiveMonitor() {
+    return *impl->exclusive_monitor;
+}
+
+const Core::ExclusiveMonitor& KernelCore::GetExclusiveMonitor() const {
+    return *impl->exclusive_monitor;
+}
+
+void KernelCore::InvalidateAllInstructionCaches() {
+    for (std::size_t i = 0; i < impl->global_scheduler.CpuCoresCount(); i++) {
+        PhysicalCore(i).ArmInterface().ClearInstructionCache();
+    }
+}
+
+void KernelCore::PrepareReschedule(std::size_t id) {
+    if (id < impl->global_scheduler.CpuCoresCount()) {
+        impl->cores[id].Stop();
+    }
+}
+
 void KernelCore::AddNamedPort(std::string name, std::shared_ptr<ClientPort> port) {
     impl->named_ports.emplace(std::move(name), std::move(port));
 }

src/core/hle/kernel/kernel.h

@@ -11,8 +11,9 @@
 #include "core/hle/kernel/object.h"
 
 namespace Core {
+class ExclusiveMonitor;
 class System;
-}
+} // namespace Core
 
 namespace Core::Timing {
 class CoreTiming;
@@ -25,6 +26,7 @@ class AddressArbiter;
 class ClientPort;
 class GlobalScheduler;
 class HandleTable;
+class PhysicalCore;
 class Process;
 class ResourceLimit;
 class Thread;
@@ -84,6 +86,21 @@ public:
     /// Gets the sole instance of the global scheduler
     const Kernel::GlobalScheduler& GlobalScheduler() const;
 
+    /// Gets the an instance of the respective physical CPU core.
+    Kernel::PhysicalCore& PhysicalCore(std::size_t id);
+
+    /// Gets the an instance of the respective physical CPU core.
+    const Kernel::PhysicalCore& PhysicalCore(std::size_t id) const;
+
+    /// Stops execution of 'id' core, in order to reschedule a new thread.
+    void PrepareReschedule(std::size_t id);
+
+    Core::ExclusiveMonitor& GetExclusiveMonitor();
+
+    const Core::ExclusiveMonitor& GetExclusiveMonitor() const;
+
+    void InvalidateAllInstructionCaches();
+
     /// Adds a port to the named port table
     void AddNamedPort(std::string name, std::shared_ptr<ClientPort> port);
 

src/core/hle/kernel/physical_core.cpp

@@ -0,0 +1,51 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/logging/log.h"
+#include "core/arm/arm_interface.h"
+#ifdef ARCHITECTURE_x86_64
+#include "core/arm/dynarmic/arm_dynarmic.h"
+#endif
+#include "core/arm/exclusive_monitor.h"
+#include "core/arm/unicorn/arm_unicorn.h"
+#include "core/core.h"
+#include "core/hle/kernel/physical_core.h"
+#include "core/hle/kernel/scheduler.h"
+#include "core/hle/kernel/thread.h"
+
+namespace Kernel {
+
+PhysicalCore::PhysicalCore(Core::System& system, std::size_t id,
+                           Core::ExclusiveMonitor& exclusive_monitor)
+    : core_index{id} {
+#ifdef ARCHITECTURE_x86_64
+    arm_interface = std::make_unique<Core::ARM_Dynarmic>(system, exclusive_monitor, core_index);
+#else
+    arm_interface = std::make_shared<Core::ARM_Unicorn>(system);
+    LOG_WARNING(Core, "CPU JIT requested, but Dynarmic not available");
+#endif
+
+    scheduler = std::make_unique<Kernel::Scheduler>(system, *arm_interface, core_index);
+}
+
+PhysicalCore::~PhysicalCore() = default;
+
+void PhysicalCore::Run() {
+    arm_interface->Run();
+    arm_interface->ClearExclusiveState();
+}
+
+void PhysicalCore::Step() {
+    arm_interface->Step();
+}
+
+void PhysicalCore::Stop() {
+    arm_interface->PrepareReschedule();
+}
+
+void PhysicalCore::Shutdown() {
+    scheduler->Shutdown();
+}
+
+} // namespace Kernel

src/core/hle/kernel/physical_core.h

@@ -0,0 +1,77 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <cstddef>
+#include <memory>
+
+namespace Kernel {
+class Scheduler;
+} // namespace Kernel
+
+namespace Core {
+class ARM_Interface;
+class ExclusiveMonitor;
+class System;
+} // namespace Core
+
+namespace Kernel {
+
+class PhysicalCore {
+public:
+    PhysicalCore(Core::System& system, std::size_t id, Core::ExclusiveMonitor& exclusive_monitor);
+    ~PhysicalCore();
+
+    PhysicalCore(const PhysicalCore&) = delete;
+    PhysicalCore& operator=(const PhysicalCore&) = delete;
+
+    PhysicalCore(PhysicalCore&&) = default;
+    PhysicalCore& operator=(PhysicalCore&&) = default;
+
+    /// Execute current jit state
+    void Run();
+    /// Execute a single instruction in current jit.
+    void Step();
+    /// Stop JIT execution/exit
+    void Stop();
+
+    // Shutdown this physical core.
+    void Shutdown();
+
+    Core::ARM_Interface& ArmInterface() {
+        return *arm_interface;
+    }
+
+    const Core::ARM_Interface& ArmInterface() const {
+        return *arm_interface;
+    }
+
+    bool IsMainCore() const {
+        return core_index == 0;
+    }
+
+    bool IsSystemCore() const {
+        return core_index == 3;
+    }
+
+    std::size_t CoreIndex() const {
+        return core_index;
+    }
+
+    Kernel::Scheduler& Scheduler() {
+        return *scheduler;
+    }
+
+    const Kernel::Scheduler& Scheduler() const {
+        return *scheduler;
+    }
+
+private:
+    std::size_t core_index;
+    std::unique_ptr<Core::ARM_Interface> arm_interface;
+    std::unique_ptr<Kernel::Scheduler> scheduler;
+};
+
+} // namespace Kernel

src/core/hle/kernel/scheduler.cpp

@@ -14,7 +14,6 @@
 #include "common/logging/log.h"
 #include "core/arm/arm_interface.h"
 #include "core/core.h"
-#include "core/core_cpu.h"
 #include "core/core_timing.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/process.h"

src/core/hle/kernel/svc.cpp

@@ -15,7 +15,7 @@
 #include "common/string_util.h"
 #include "core/arm/exclusive_monitor.h"
 #include "core/core.h"
-#include "core/core_cpu.h"
+#include "core/core_manager.h"
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
 #include "core/hle/kernel/address_arbiter.h"
@@ -1863,10 +1863,14 @@ static ResultCode CreateTransferMemory(Core::System& system, Handle* handle, VAd
     }
 
     auto& kernel = system.Kernel();
-    auto transfer_mem_handle = TransferMemory::Create(kernel, addr, size, perms);
+    auto transfer_mem_handle = TransferMemory::Create(kernel, system.Memory(), addr, size, perms);
+
+    if (const auto reserve_result{transfer_mem_handle->Reserve()}; reserve_result.IsError()) {
+        return reserve_result;
+    }
 
     auto& handle_table = kernel.CurrentProcess()->GetHandleTable();
-    const auto result = handle_table.Create(std::move(transfer_mem_handle));
+    const auto result{handle_table.Create(std::move(transfer_mem_handle))};
     if (result.Failed()) {
         return result.Code();
     }

src/core/hle/kernel/thread.cpp

@@ -13,7 +13,6 @@
 #include "common/thread_queue_list.h"
 #include "core/arm/arm_interface.h"
 #include "core/core.h"
-#include "core/core_cpu.h"
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
 #include "core/hle/kernel/errors.h"
@@ -356,7 +355,7 @@ void Thread::SetActivity(ThreadActivity value) {
         // Set status if not waiting
         if (status == ThreadStatus::Ready || status == ThreadStatus::Running) {
             SetStatus(ThreadStatus::Paused);
-            Core::System::GetInstance().CpuCore(processor_id).PrepareReschedule();
+            kernel.PrepareReschedule(processor_id);
         }
     } else if (status == ThreadStatus::Paused) {
         // Ready to reschedule

src/core/hle/kernel/transfer_memory.cpp

@@ -8,15 +8,23 @@
 #include "core/hle/kernel/shared_memory.h"
 #include "core/hle/kernel/transfer_memory.h"
 #include "core/hle/result.h"
+#include "core/memory.h"
 
 namespace Kernel {
 
-TransferMemory::TransferMemory(KernelCore& kernel) : Object{kernel} {}
-TransferMemory::~TransferMemory() = default;
+TransferMemory::TransferMemory(KernelCore& kernel, Memory::Memory& memory)
+    : Object{kernel}, memory{memory} {}
 
-std::shared_ptr<TransferMemory> TransferMemory::Create(KernelCore& kernel, VAddr base_address,
-                                                       u64 size, MemoryPermission permissions) {
-    std::shared_ptr<TransferMemory> transfer_memory{std::make_shared<TransferMemory>(kernel)};
+TransferMemory::~TransferMemory() {
+    // Release memory region when transfer memory is destroyed
+    Reset();
+}
+
+std::shared_ptr<TransferMemory> TransferMemory::Create(KernelCore& kernel, Memory::Memory& memory,
+                                                       VAddr base_address, u64 size,
+                                                       MemoryPermission permissions) {
+    std::shared_ptr<TransferMemory> transfer_memory{
+        std::make_shared<TransferMemory>(kernel, memory)};
 
     transfer_memory->base_address = base_address;
     transfer_memory->memory_size = size;
@@ -27,7 +35,7 @@ std::shared_ptr<TransferMemory> TransferMemory::Create(KernelCore& kernel, VAddr
 }
 
 const u8* TransferMemory::GetPointer() const {
-    return backing_block.get()->data();
+    return memory.GetPointer(base_address);
 }
 
 u64 TransferMemory::GetSize() const {
@@ -62,6 +70,52 @@ ResultCode TransferMemory::MapMemory(VAddr address, u64 size, MemoryPermission p
     return RESULT_SUCCESS;
 }
 
+ResultCode TransferMemory::Reserve() {
+    auto& vm_manager{owner_process->VMManager()};
+    const auto check_range_result{vm_manager.CheckRangeState(
+        base_address, memory_size, MemoryState::FlagTransfer | MemoryState::FlagMemoryPoolAllocated,
+        MemoryState::FlagTransfer | MemoryState::FlagMemoryPoolAllocated, VMAPermission::All,
+        VMAPermission::ReadWrite, MemoryAttribute::Mask, MemoryAttribute::None,
+        MemoryAttribute::IpcAndDeviceMapped)};
+
+    if (check_range_result.Failed()) {
+        return check_range_result.Code();
+    }
+
+    auto [state_, permissions_, attribute] = *check_range_result;
+
+    if (const auto result{vm_manager.ReprotectRange(
+            base_address, memory_size, SharedMemory::ConvertPermissions(owner_permissions))};
+        result.IsError()) {
+        return result;
+    }
+
+    return vm_manager.SetMemoryAttribute(base_address, memory_size, MemoryAttribute::Mask,
+                                         attribute | MemoryAttribute::Locked);
+}
+
+ResultCode TransferMemory::Reset() {
+    auto& vm_manager{owner_process->VMManager()};
+    if (const auto result{vm_manager.CheckRangeState(
+            base_address, memory_size,
+            MemoryState::FlagTransfer | MemoryState::FlagMemoryPoolAllocated,
+            MemoryState::FlagTransfer | MemoryState::FlagMemoryPoolAllocated, VMAPermission::None,
+            VMAPermission::None, MemoryAttribute::Mask, MemoryAttribute::Locked,
+            MemoryAttribute::IpcAndDeviceMapped)};
+        result.Failed()) {
+        return result.Code();
+    }
+
+    if (const auto result{
+            vm_manager.ReprotectRange(base_address, memory_size, VMAPermission::ReadWrite)};
+        result.IsError()) {
+        return result;
+    }
+
+    return vm_manager.SetMemoryAttribute(base_address, memory_size, MemoryAttribute::Mask,
+                                         MemoryAttribute::None);
+}
+
 ResultCode TransferMemory::UnmapMemory(VAddr address, u64 size) {
     if (memory_size != size) {
         return ERR_INVALID_SIZE;

src/core/hle/kernel/transfer_memory.h

@@ -11,6 +11,10 @@
 
 union ResultCode;
 
+namespace Memory {
+class Memory;
+}
+
 namespace Kernel {
 
 class KernelCore;
@@ -26,12 +30,13 @@ enum class MemoryPermission : u32;
 ///
 class TransferMemory final : public Object {
 public:
-    explicit TransferMemory(KernelCore& kernel);
+    explicit TransferMemory(KernelCore& kernel, Memory::Memory& memory);
     ~TransferMemory() override;
 
     static constexpr HandleType HANDLE_TYPE = HandleType::TransferMemory;
 
-    static std::shared_ptr<TransferMemory> Create(KernelCore& kernel, VAddr base_address, u64 size,
+    static std::shared_ptr<TransferMemory> Create(KernelCore& kernel, Memory::Memory& memory,
+                                                  VAddr base_address, u64 size,
                                                   MemoryPermission permissions);
 
     TransferMemory(const TransferMemory&) = delete;
@@ -80,6 +85,14 @@ public:
     ///
     ResultCode UnmapMemory(VAddr address, u64 size);
 
+    /// Reserves the region to be used for the transfer memory, called after the transfer memory is
+    /// created.
+    ResultCode Reserve();
+
+    /// Resets the region previously used for the transfer memory, called after the transfer memory
+    /// is closed.
+    ResultCode Reset();
+
 private:
     /// Memory block backing this instance.
     std::shared_ptr<PhysicalMemory> backing_block;
@@ -98,6 +111,8 @@ private:
 
     /// Whether or not this transfer memory instance has mapped memory.
     bool is_mapped = false;
+
+    Memory::Memory& memory;
 };
 
 } // namespace Kernel

src/core/hle/kernel/vm_manager.cpp

@@ -544,7 +544,8 @@ MemoryInfo VMManager::QueryMemory(VAddr address) const {
 
 ResultCode VMManager::SetMemoryAttribute(VAddr address, u64 size, MemoryAttribute mask,
                                          MemoryAttribute attribute) {
-    constexpr auto ignore_mask = MemoryAttribute::Uncached | MemoryAttribute::DeviceMapped;
+    constexpr auto ignore_mask =
+        MemoryAttribute::Uncached | MemoryAttribute::DeviceMapped | MemoryAttribute::Locked;
     constexpr auto attribute_mask = ~ignore_mask;
 
     const auto result = CheckRangeState(

src/core/hle/kernel/vm_manager.h

@@ -98,6 +98,8 @@ enum class MemoryAttribute : u32 {
     DeviceMapped = 4,
     /// Uncached memory
     Uncached = 8,
+
+    IpcAndDeviceMapped = LockedForIPC | DeviceMapped,
 };
 
 constexpr MemoryAttribute operator|(MemoryAttribute lhs, MemoryAttribute rhs) {
@@ -654,6 +656,35 @@ public:
     /// is scheduled.
     Common::PageTable page_table{Memory::PAGE_BITS};
 
+    using CheckResults = ResultVal<std::tuple<MemoryState, VMAPermission, MemoryAttribute>>;
+
+    /// Checks if an address range adheres to the specified states provided.
+    ///
+    /// @param address         The starting address of the address range.
+    /// @param size            The size of the address range.
+    /// @param state_mask      The memory state mask.
+    /// @param state           The state to compare the individual VMA states against,
+    ///                        which is done in the form of: (vma.state & state_mask) != state.
+    /// @param permission_mask The memory permissions mask.
+    /// @param permissions     The permission to compare the individual VMA permissions against,
+    ///                        which is done in the form of:
+    ///                        (vma.permission & permission_mask) != permission.
+    /// @param attribute_mask  The memory attribute mask.
+    /// @param attribute       The memory attributes to compare the individual VMA attributes
+    ///                        against, which is done in the form of:
+    ///                        (vma.attributes & attribute_mask) != attribute.
+    /// @param ignore_mask     The memory attributes to ignore during the check.
+    ///
+    /// @returns If successful, returns a tuple containing the memory attributes
+    ///          (with ignored bits specified by ignore_mask unset), memory permissions, and
+    ///          memory state across the memory range.
+    /// @returns If not successful, returns ERR_INVALID_ADDRESS_STATE.
+    ///
+    CheckResults CheckRangeState(VAddr address, u64 size, MemoryState state_mask, MemoryState state,
+                                 VMAPermission permission_mask, VMAPermission permissions,
+                                 MemoryAttribute attribute_mask, MemoryAttribute attribute,
+                                 MemoryAttribute ignore_mask) const;
+
 private:
     using VMAIter = VMAMap::iterator;
 
@@ -707,35 +738,6 @@ private:
     /// Clears out the page table
     void ClearPageTable();
 
-    using CheckResults = ResultVal<std::tuple<MemoryState, VMAPermission, MemoryAttribute>>;
-
-    /// Checks if an address range adheres to the specified states provided.
-    ///
-    /// @param address         The starting address of the address range.
-    /// @param size            The size of the address range.
-    /// @param state_mask      The memory state mask.
-    /// @param state           The state to compare the individual VMA states against,
-    ///                        which is done in the form of: (vma.state & state_mask) != state.
-    /// @param permission_mask The memory permissions mask.
-    /// @param permissions     The permission to compare the individual VMA permissions against,
-    ///                        which is done in the form of:
-    ///                        (vma.permission & permission_mask) != permission.
-    /// @param attribute_mask  The memory attribute mask.
-    /// @param attribute       The memory attributes to compare the individual VMA attributes
-    ///                        against, which is done in the form of:
-    ///                        (vma.attributes & attribute_mask) != attribute.
-    /// @param ignore_mask     The memory attributes to ignore during the check.
-    ///
-    /// @returns If successful, returns a tuple containing the memory attributes
-    ///          (with ignored bits specified by ignore_mask unset), memory permissions, and
-    ///          memory state across the memory range.
-    /// @returns If not successful, returns ERR_INVALID_ADDRESS_STATE.
-    ///
-    CheckResults CheckRangeState(VAddr address, u64 size, MemoryState state_mask, MemoryState state,
-                                 VMAPermission permission_mask, VMAPermission permissions,
-                                 MemoryAttribute attribute_mask, MemoryAttribute attribute,
-                                 MemoryAttribute ignore_mask) const;
-
     /// Gets the amount of memory currently mapped (state != Unmapped) in a range.
     ResultVal<std::size_t> SizeOfAllocatedVMAsInRange(VAddr address, std::size_t size) const;
 

src/core/hle/kernel/wait_object.cpp

@@ -7,7 +7,6 @@
 #include "common/common_types.h"
 #include "common/logging/log.h"
 #include "core/core.h"
-#include "core/core_cpu.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/object.h"
 #include "core/hle/kernel/process.h"
@@ -51,17 +50,8 @@ std::shared_ptr<Thread> WaitObject::GetHighestPriorityReadyThread() const {
         if (ShouldWait(thread.get()))
             continue;
 
-        // A thread is ready to run if it's either in ThreadStatus::WaitSynch
-        // and the rest of the objects it is waiting on are ready.
-        bool ready_to_run = true;
-        if (thread_status == ThreadStatus::WaitSynch) {
-            ready_to_run = thread->AllWaitObjectsReady();
-        }
-
-        if (ready_to_run) {
-            candidate = thread.get();
-            candidate_priority = thread->GetPriority();
-        }
+        candidate = thread.get();
+        candidate_priority = thread->GetPriority();
     }
 
     return SharedFrom(candidate);
@@ -96,7 +86,7 @@ void WaitObject::WakeupWaitingThread(std::shared_ptr<Thread> thread) {
     }
     if (resume) {
         thread->ResumeFromWait();
-        Core::System::GetInstance().PrepareReschedule(thread->GetProcessorID());
+        kernel.PrepareReschedule(thread->GetProcessorID());
     }
 }
 

src/core/hle/service/am/am.cpp

@@ -709,8 +709,34 @@ void ICommonStateGetter::SetCpuBoostMode(Kernel::HLERequestContext& ctx) {
     apm_sys->SetCpuBoostMode(ctx);
 }
 
-IStorage::IStorage(std::vector<u8> buffer)
-    : ServiceFramework("IStorage"), buffer(std::move(buffer)) {
+IStorageImpl::~IStorageImpl() = default;
+
+class StorageDataImpl final : public IStorageImpl {
+public:
+    explicit StorageDataImpl(std::vector<u8>&& buffer) : buffer{std::move(buffer)} {}
+
+    std::vector<u8>& GetData() override {
+        return buffer;
+    }
+
+    const std::vector<u8>& GetData() const override {
+        return buffer;
+    }
+
+    std::size_t GetSize() const override {
+        return buffer.size();
+    }
+
+private:
+    std::vector<u8> buffer;
+};
+
+IStorage::IStorage(std::vector<u8>&& buffer)
+    : ServiceFramework("IStorage"), impl{std::make_shared<StorageDataImpl>(std::move(buffer))} {
+    Register();
+}
+
+void IStorage::Register() {
     // clang-format off
         static const FunctionInfo functions[] = {
             {0, &IStorage::Open, "Open"},
@@ -723,8 +749,13 @@ IStorage::IStorage(std::vector<u8> buffer)
 
 IStorage::~IStorage() = default;
 
-const std::vector<u8>& IStorage::GetData() const {
-    return buffer;
+void IStorage::Open(Kernel::HLERequestContext& ctx) {
+    LOG_DEBUG(Service_AM, "called");
+
+    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
+
+    rb.Push(RESULT_SUCCESS);
+    rb.PushIpcInterface<IStorageAccessor>(*this);
 }
 
 void ICommonStateGetter::GetOperationMode(Kernel::HLERequestContext& ctx) {
@@ -816,7 +847,7 @@ private:
         LOG_DEBUG(Service_AM, "called");
 
         IPC::RequestParser rp{ctx};
-        applet->GetBroker().PushNormalDataFromGame(*rp.PopIpcInterface<IStorage>());
+        applet->GetBroker().PushNormalDataFromGame(rp.PopIpcInterface<IStorage>());
 
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(RESULT_SUCCESS);
@@ -825,26 +856,25 @@ private:
     void PopOutData(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Service_AM, "called");
 
-        IPC::ResponseBuilder rb{ctx, 2, 0, 1};
-
         const auto storage = applet->GetBroker().PopNormalDataToGame();
         if (storage == nullptr) {
             LOG_ERROR(Service_AM,
                       "storage is a nullptr. There is no data in the current normal channel");
-
+            IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERR_NO_DATA_IN_CHANNEL);
             return;
         }
 
+        IPC::ResponseBuilder rb{ctx, 2, 0, 1};
         rb.Push(RESULT_SUCCESS);
-        rb.PushIpcInterface<IStorage>(std::move(*storage));
+        rb.PushIpcInterface<IStorage>(std::move(storage));
     }
 
     void PushInteractiveInData(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Service_AM, "called");
 
         IPC::RequestParser rp{ctx};
-        applet->GetBroker().PushInteractiveDataFromGame(*rp.PopIpcInterface<IStorage>());
+        applet->GetBroker().PushInteractiveDataFromGame(rp.PopIpcInterface<IStorage>());
 
         ASSERT(applet->IsInitialized());
         applet->ExecuteInteractive();
@@ -857,19 +887,18 @@ private:
     void PopInteractiveOutData(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Service_AM, "called");
 
-        IPC::ResponseBuilder rb{ctx, 2, 0, 1};
-
         const auto storage = applet->GetBroker().PopInteractiveDataToGame();
         if (storage == nullptr) {
             LOG_ERROR(Service_AM,
                       "storage is a nullptr. There is no data in the current interactive channel");
-
+            IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERR_NO_DATA_IN_CHANNEL);
             return;
         }
 
+        IPC::ResponseBuilder rb{ctx, 2, 0, 1};
         rb.Push(RESULT_SUCCESS);
-        rb.PushIpcInterface<IStorage>(std::move(*storage));
+        rb.PushIpcInterface<IStorage>(std::move(storage));
     }
 
     void GetPopOutDataEvent(Kernel::HLERequestContext& ctx) {
@@ -891,15 +920,6 @@ private:
     std::shared_ptr<Applets::Applet> applet;
 };
 
-void IStorage::Open(Kernel::HLERequestContext& ctx) {
-    LOG_DEBUG(Service_AM, "called");
-
-    IPC::ResponseBuilder rb{ctx, 2, 0, 1};
-
-    rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<IStorageAccessor>(*this);
-}
-
 IStorageAccessor::IStorageAccessor(IStorage& storage)
     : ServiceFramework("IStorageAccessor"), backing(storage) {
     // clang-format off
@@ -921,7 +941,7 @@ void IStorageAccessor::GetSize(Kernel::HLERequestContext& ctx) {
     IPC::ResponseBuilder rb{ctx, 4};
 
     rb.Push(RESULT_SUCCESS);
-    rb.Push(static_cast<u64>(backing.buffer.size()));
+    rb.Push(static_cast<u64>(backing.GetSize()));
 }
 
 void IStorageAccessor::Write(Kernel::HLERequestContext& ctx) {
@@ -932,17 +952,17 @@ void IStorageAccessor::Write(Kernel::HLERequestContext& ctx) {
 
     LOG_DEBUG(Service_AM, "called, offset={}, size={}", offset, data.size());
 
-    if (data.size() > backing.buffer.size() - offset) {
+    if (data.size() > backing.GetSize() - offset) {
         LOG_ERROR(Service_AM,
                   "offset is out of bounds, backing_buffer_sz={}, data_size={}, offset={}",
-                  backing.buffer.size(), data.size(), offset);
+                  backing.GetSize(), data.size(), offset);
 
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(ERR_SIZE_OUT_OF_BOUNDS);
         return;
     }
 
-    std::memcpy(backing.buffer.data() + offset, data.data(), data.size());
+    std::memcpy(backing.GetData().data() + offset, data.data(), data.size());
 
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);
@@ -956,16 +976,16 @@ void IStorageAccessor::Read(Kernel::HLERequestContext& ctx) {
 
     LOG_DEBUG(Service_AM, "called, offset={}, size={}", offset, size);
 
-    if (size > backing.buffer.size() - offset) {
+    if (size > backing.GetSize() - offset) {
         LOG_ERROR(Service_AM, "offset is out of bounds, backing_buffer_sz={}, size={}, offset={}",
-                  backing.buffer.size(), size, offset);
+                  backing.GetSize(), size, offset);
 
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(ERR_SIZE_OUT_OF_BOUNDS);
         return;
     }
 
-    ctx.WriteBuffer(backing.buffer.data() + offset, size);
+    ctx.WriteBuffer(backing.GetData().data() + offset, size);
 
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);
@@ -1031,7 +1051,7 @@ void ILibraryAppletCreator::CreateTransferMemoryStorage(Kernel::HLERequestContex
     rp.SetCurrentOffset(3);
     const auto handle{rp.Pop<Kernel::Handle>()};
 
-    const auto transfer_mem =
+    auto transfer_mem =
         system.CurrentProcess()->GetHandleTable().Get<Kernel::TransferMemory>(handle);
 
     if (transfer_mem == nullptr) {
@@ -1047,7 +1067,7 @@ void ILibraryAppletCreator::CreateTransferMemoryStorage(Kernel::HLERequestContex
 
     IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface(std::make_shared<IStorage>(std::move(memory)));
+    rb.PushIpcInterface<IStorage>(std::move(memory));
 }
 
 IApplicationFunctions::IApplicationFunctions(Core::System& system_)
@@ -1189,13 +1209,11 @@ void IApplicationFunctions::PopLaunchParameter(Kernel::HLERequestContext& ctx) {
         u64 build_id{};
         std::memcpy(&build_id, build_id_full.data(), sizeof(u64));
 
-        const auto data =
-            backend->GetLaunchParameter({system.CurrentProcess()->GetTitleID(), build_id});
-
+        auto data = backend->GetLaunchParameter({system.CurrentProcess()->GetTitleID(), build_id});
         if (data.has_value()) {
             IPC::ResponseBuilder rb{ctx, 2, 0, 1};
             rb.Push(RESULT_SUCCESS);
-            rb.PushIpcInterface<AM::IStorage>(*data);
+            rb.PushIpcInterface<IStorage>(std::move(*data));
             launch_popped_application_specific = true;
             return;
         }
@@ -1218,7 +1236,7 @@ void IApplicationFunctions::PopLaunchParameter(Kernel::HLERequestContext& ctx) {
         std::vector<u8> buffer(sizeof(LaunchParameterAccountPreselectedUser));
         std::memcpy(buffer.data(), &params, buffer.size());
 
-        rb.PushIpcInterface<AM::IStorage>(buffer);
+        rb.PushIpcInterface<IStorage>(std::move(buffer));
         launch_popped_account_preselect = true;
         return;
     }

src/core/hle/service/am/am.h

@@ -12,7 +12,8 @@
 
 namespace Kernel {
 class KernelCore;
-}
+class TransferMemory;
+} // namespace Kernel
 
 namespace Service::NVFlinger {
 class NVFlinger;
@@ -188,19 +189,36 @@ private:
     std::shared_ptr<AppletMessageQueue> msg_queue;
 };
 
+class IStorageImpl {
+public:
+    virtual ~IStorageImpl();
+    virtual std::vector<u8>& GetData() = 0;
+    virtual const std::vector<u8>& GetData() const = 0;
+    virtual std::size_t GetSize() const = 0;
+};
+
 class IStorage final : public ServiceFramework<IStorage> {
 public:
-    explicit IStorage(std::vector<u8> buffer);
+    explicit IStorage(std::vector<u8>&& buffer);
     ~IStorage() override;
 
-    const std::vector<u8>& GetData() const;
+    std::vector<u8>& GetData() {
+        return impl->GetData();
+    }
+
+    const std::vector<u8>& GetData() const {
+        return impl->GetData();
+    }
+
+    std::size_t GetSize() const {
+        return impl->GetSize();
+    }
 
 private:
+    void Register();
     void Open(Kernel::HLERequestContext& ctx);
 
-    std::vector<u8> buffer;
-
-    friend class IStorageAccessor;
+    std::shared_ptr<IStorageImpl> impl;
 };
 
 class IStorageAccessor final : public ServiceFramework<IStorageAccessor> {

src/core/hle/service/am/applets/applets.cpp

@@ -50,16 +50,17 @@ AppletDataBroker::RawChannelData AppletDataBroker::PeekDataToAppletForDebug() co
     return {std::move(out_normal), std::move(out_interactive)};
 }
 
-std::unique_ptr<IStorage> AppletDataBroker::PopNormalDataToGame() {
+std::shared_ptr<IStorage> AppletDataBroker::PopNormalDataToGame() {
     if (out_channel.empty())
         return nullptr;
 
     auto out = std::move(out_channel.front());
     out_channel.pop_front();
+    pop_out_data_event.writable->Clear();
     return out;
 }
 
-std::unique_ptr<IStorage> AppletDataBroker::PopNormalDataToApplet() {
+std::shared_ptr<IStorage> AppletDataBroker::PopNormalDataToApplet() {
     if (in_channel.empty())
         return nullptr;
 
@@ -68,16 +69,17 @@ std::unique_ptr<IStorage> AppletDataBroker::PopNormalDataToApplet() {
     return out;
 }
 
-std::unique_ptr<IStorage> AppletDataBroker::PopInteractiveDataToGame() {
+std::shared_ptr<IStorage> AppletDataBroker::PopInteractiveDataToGame() {
     if (out_interactive_channel.empty())
         return nullptr;
 
     auto out = std::move(out_interactive_channel.front());
     out_interactive_channel.pop_front();
+    pop_interactive_out_data_event.writable->Clear();
     return out;
 }
 
-std::unique_ptr<IStorage> AppletDataBroker::PopInteractiveDataToApplet() {
+std::shared_ptr<IStorage> AppletDataBroker::PopInteractiveDataToApplet() {
     if (in_interactive_channel.empty())
         return nullptr;
 
@@ -86,21 +88,21 @@ std::unique_ptr<IStorage> AppletDataBroker::PopInteractiveDataToApplet() {
     return out;
 }
 
-void AppletDataBroker::PushNormalDataFromGame(IStorage storage) {
-    in_channel.push_back(std::make_unique<IStorage>(storage));
+void AppletDataBroker::PushNormalDataFromGame(std::shared_ptr<IStorage>&& storage) {
+    in_channel.emplace_back(std::move(storage));
 }
 
-void AppletDataBroker::PushNormalDataFromApplet(IStorage storage) {
-    out_channel.push_back(std::make_unique<IStorage>(storage));
+void AppletDataBroker::PushNormalDataFromApplet(std::shared_ptr<IStorage>&& storage) {
+    out_channel.emplace_back(std::move(storage));
     pop_out_data_event.writable->Signal();
 }
 
-void AppletDataBroker::PushInteractiveDataFromGame(IStorage storage) {
-    in_interactive_channel.push_back(std::make_unique<IStorage>(storage));
+void AppletDataBroker::PushInteractiveDataFromGame(std::shared_ptr<IStorage>&& storage) {
+    in_interactive_channel.emplace_back(std::move(storage));
 }
 
-void AppletDataBroker::PushInteractiveDataFromApplet(IStorage storage) {
-    out_interactive_channel.push_back(std::make_unique<IStorage>(storage));
+void AppletDataBroker::PushInteractiveDataFromApplet(std::shared_ptr<IStorage>&& storage) {
+    out_interactive_channel.emplace_back(std::move(storage));
     pop_interactive_out_data_event.writable->Signal();
 }
 

src/core/hle/service/am/applets/applets.h

@@ -72,17 +72,17 @@ public:
     // Retrieves but does not pop the data sent to applet.
     RawChannelData PeekDataToAppletForDebug() const;
 
-    std::unique_ptr<IStorage> PopNormalDataToGame();
-    std::unique_ptr<IStorage> PopNormalDataToApplet();
+    std::shared_ptr<IStorage> PopNormalDataToGame();
+    std::shared_ptr<IStorage> PopNormalDataToApplet();
 
-    std::unique_ptr<IStorage> PopInteractiveDataToGame();
-    std::unique_ptr<IStorage> PopInteractiveDataToApplet();
+    std::shared_ptr<IStorage> PopInteractiveDataToGame();
+    std::shared_ptr<IStorage> PopInteractiveDataToApplet();
 
-    void PushNormalDataFromGame(IStorage storage);
-    void PushNormalDataFromApplet(IStorage storage);
+    void PushNormalDataFromGame(std::shared_ptr<IStorage>&& storage);
+    void PushNormalDataFromApplet(std::shared_ptr<IStorage>&& storage);
 
-    void PushInteractiveDataFromGame(IStorage storage);
-    void PushInteractiveDataFromApplet(IStorage storage);
+    void PushInteractiveDataFromGame(std::shared_ptr<IStorage>&& storage);
+    void PushInteractiveDataFromApplet(std::shared_ptr<IStorage>&& storage);
 
     void SignalStateChanged() const;
 
@@ -94,16 +94,16 @@ private:
     // Queues are named from applet's perspective
 
     // PopNormalDataToApplet and PushNormalDataFromGame
-    std::deque<std::unique_ptr<IStorage>> in_channel;
+    std::deque<std::shared_ptr<IStorage>> in_channel;
 
     // PopNormalDataToGame and PushNormalDataFromApplet
-    std::deque<std::unique_ptr<IStorage>> out_channel;
+    std::deque<std::shared_ptr<IStorage>> out_channel;
 
     // PopInteractiveDataToApplet and PushInteractiveDataFromGame
-    std::deque<std::unique_ptr<IStorage>> in_interactive_channel;
+    std::deque<std::shared_ptr<IStorage>> in_interactive_channel;
 
     // PopInteractiveDataToGame and PushInteractiveDataFromApplet
-    std::deque<std::unique_ptr<IStorage>> out_interactive_channel;
+    std::deque<std::shared_ptr<IStorage>> out_interactive_channel;
 
     Kernel::EventPair state_changed_event;
 

src/core/hle/service/am/applets/error.cpp

@@ -186,7 +186,7 @@ void Error::Execute() {
 
 void Error::DisplayCompleted() {
     complete = true;
-    broker.PushNormalDataFromApplet(IStorage{{}});
+    broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::vector<u8>{}));
     broker.SignalStateChanged();
 }
 

src/core/hle/service/am/applets/general_backend.cpp

@@ -20,7 +20,7 @@ namespace Service::AM::Applets {
 constexpr ResultCode ERROR_INVALID_PIN{ErrorModule::PCTL, 221};
 
 static void LogCurrentStorage(AppletDataBroker& broker, std::string_view prefix) {
-    std::unique_ptr<IStorage> storage = broker.PopNormalDataToApplet();
+    std::shared_ptr<IStorage> storage = broker.PopNormalDataToApplet();
     for (; storage != nullptr; storage = broker.PopNormalDataToApplet()) {
         const auto data = storage->GetData();
         LOG_INFO(Service_AM,
@@ -148,7 +148,7 @@ void Auth::AuthFinished(bool successful) {
     std::vector<u8> out(sizeof(Return));
     std::memcpy(out.data(), &return_, sizeof(Return));
 
-    broker.PushNormalDataFromApplet(IStorage{out});
+    broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::move(out)));
     broker.SignalStateChanged();
 }
 
@@ -198,7 +198,7 @@ void PhotoViewer::Execute() {
 }
 
 void PhotoViewer::ViewFinished() {
-    broker.PushNormalDataFromApplet(IStorage{{}});
+    broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::vector<u8>{}));
     broker.SignalStateChanged();
 }
 
@@ -234,8 +234,8 @@ void StubApplet::ExecuteInteractive() {
     LOG_WARNING(Service_AM, "called (STUBBED)");
     LogCurrentStorage(broker, "ExecuteInteractive");
 
-    broker.PushNormalDataFromApplet(IStorage{std::vector<u8>(0x1000)});
-    broker.PushInteractiveDataFromApplet(IStorage{std::vector<u8>(0x1000)});
+    broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::vector<u8>(0x1000)));
+    broker.PushInteractiveDataFromApplet(std::make_shared<IStorage>(std::vector<u8>(0x1000)));
     broker.SignalStateChanged();
 }
 
@@ -243,8 +243,8 @@ void StubApplet::Execute() {
     LOG_WARNING(Service_AM, "called (STUBBED)");
     LogCurrentStorage(broker, "Execute");
 
-    broker.PushNormalDataFromApplet(IStorage{std::vector<u8>(0x1000)});
-    broker.PushInteractiveDataFromApplet(IStorage{std::vector<u8>(0x1000)});
+    broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::vector<u8>(0x1000)));
+    broker.PushInteractiveDataFromApplet(std::make_shared<IStorage>(std::vector<u8>(0x1000)));
     broker.SignalStateChanged();
 }
 

src/core/hle/service/am/applets/profile_select.cpp

@@ -50,7 +50,7 @@ void ProfileSelect::ExecuteInteractive() {
 
 void ProfileSelect::Execute() {
     if (complete) {
-        broker.PushNormalDataFromApplet(IStorage{final_data});
+        broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::move(final_data)));
         return;
     }
 
@@ -71,7 +71,7 @@ void ProfileSelect::SelectionComplete(std::optional<Common::UUID> uuid) {
 
     final_data = std::vector<u8>(sizeof(UserSelectionOutput));
     std::memcpy(final_data.data(), &output, final_data.size());
-    broker.PushNormalDataFromApplet(IStorage{final_data});
+    broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::move(final_data)));
     broker.SignalStateChanged();
 }
 

src/core/hle/service/am/applets/software_keyboard.cpp

@@ -102,7 +102,8 @@ void SoftwareKeyboard::ExecuteInteractive() {
 
 void SoftwareKeyboard::Execute() {
     if (complete) {
-        broker.PushNormalDataFromApplet(IStorage{final_data});
+        broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::move(final_data)));
+        broker.SignalStateChanged();
         return;
     }
 
@@ -119,7 +120,7 @@ void SoftwareKeyboard::WriteText(std::optional<std::u16string> text) {
         std::vector<u8> output_sub(SWKBD_OUTPUT_BUFFER_SIZE);
 
         if (config.utf_8) {
-            const u64 size = text->size() + 8;
+            const u64 size = text->size() + sizeof(u64);
             const auto new_text = Common::UTF16ToUTF8(*text);
 
             std::memcpy(output_sub.data(), &size, sizeof(u64));
@@ -130,7 +131,7 @@ void SoftwareKeyboard::WriteText(std::optional<std::u16string> text) {
             std::memcpy(output_main.data() + 4, new_text.data(),
                         std::min(new_text.size(), SWKBD_OUTPUT_BUFFER_SIZE - 4));
         } else {
-            const u64 size = text->size() * 2 + 8;
+            const u64 size = text->size() * 2 + sizeof(u64);
             std::memcpy(output_sub.data(), &size, sizeof(u64));
             std::memcpy(output_sub.data() + 8, text->data(),
                         std::min(text->size() * 2, SWKBD_OUTPUT_BUFFER_SIZE - 8));
@@ -144,15 +145,15 @@ void SoftwareKeyboard::WriteText(std::optional<std::u16string> text) {
         final_data = output_main;
 
         if (complete) {
-            broker.PushNormalDataFromApplet(IStorage{output_main});
+            broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::move(output_main)));
             broker.SignalStateChanged();
         } else {
-            broker.PushInteractiveDataFromApplet(IStorage{output_sub});
+            broker.PushInteractiveDataFromApplet(std::make_shared<IStorage>(std::move(output_sub)));
         }
     } else {
         output_main[0] = 1;
         complete = true;
-        broker.PushNormalDataFromApplet(IStorage{output_main});
+        broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::move(output_main)));
         broker.SignalStateChanged();
     }
 }

src/core/hle/service/am/applets/web_browser.cpp

@@ -284,7 +284,7 @@ void WebBrowser::Finalize() {
     std::vector<u8> data(sizeof(WebCommonReturnValue));
     std::memcpy(data.data(), &out, sizeof(WebCommonReturnValue));
 
-    broker.PushNormalDataFromApplet(IStorage{data});
+    broker.PushNormalDataFromApplet(std::make_shared<IStorage>(std::move(data)));
     broker.SignalStateChanged();
 
     if (!temporary_dir.empty() && FileUtil::IsDirectory(temporary_dir)) {

src/core/hle/service/bcat/backend/backend.cpp

@@ -117,13 +117,13 @@ bool NullBackend::SynchronizeDirectory(TitleIDVersion title, std::string name,
 }
 
 bool NullBackend::Clear(u64 title_id) {
-    LOG_DEBUG(Service_BCAT, "called, title_id={:016X}");
+    LOG_DEBUG(Service_BCAT, "called, title_id={:016X}", title_id);
 
     return true;
 }
 
 void NullBackend::SetPassphrase(u64 title_id, const Passphrase& passphrase) {
-    LOG_DEBUG(Service_BCAT, "called, title_id={:016X}, passphrase = {}", title_id,
+    LOG_DEBUG(Service_BCAT, "called, title_id={:016X}, passphrase={}", title_id,
               Common::HexToString(passphrase));
 }
 

src/core/hle/service/filesystem/fsp_srv.cpp

@@ -420,7 +420,7 @@ public:
             return;
         }
 
-        IFile file(result.Unwrap());
+        auto file = std::make_shared<IFile>(result.Unwrap());
 
         IPC::ResponseBuilder rb{ctx, 2, 0, 1};
         rb.Push(RESULT_SUCCESS);
@@ -445,7 +445,7 @@ public:
             return;
         }
 
-        IDirectory directory(result.Unwrap());
+        auto directory = std::make_shared<IDirectory>(result.Unwrap());
 
         IPC::ResponseBuilder rb{ctx, 2, 0, 1};
         rb.Push(RESULT_SUCCESS);
@@ -794,8 +794,8 @@ void FSP_SRV::OpenFileSystemWithPatch(Kernel::HLERequestContext& ctx) {
 void FSP_SRV::OpenSdCardFileSystem(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_FS, "called");
 
-    IFileSystem filesystem(fsc.OpenSDMC().Unwrap(),
-                           SizeGetter::FromStorageId(fsc, FileSys::StorageId::SdCard));
+    auto filesystem = std::make_shared<IFileSystem>(
+        fsc.OpenSDMC().Unwrap(), SizeGetter::FromStorageId(fsc, FileSys::StorageId::SdCard));
 
     IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);
@@ -846,7 +846,8 @@ void FSP_SRV::OpenSaveDataFileSystem(Kernel::HLERequestContext& ctx) {
         id = FileSys::StorageId::NandSystem;
     }
 
-    IFileSystem filesystem(std::move(dir.Unwrap()), SizeGetter::FromStorageId(fsc, id));
+    auto filesystem =
+        std::make_shared<IFileSystem>(std::move(dir.Unwrap()), SizeGetter::FromStorageId(fsc, id));
 
     IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);
@@ -898,7 +899,7 @@ void FSP_SRV::OpenDataStorageByCurrentProcess(Kernel::HLERequestContext& ctx) {
         return;
     }
 
-    IStorage storage(std::move(romfs.Unwrap()));
+    auto storage = std::make_shared<IStorage>(std::move(romfs.Unwrap()));
 
     IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);
@@ -937,7 +938,8 @@ void FSP_SRV::OpenDataStorageByDataId(Kernel::HLERequestContext& ctx) {
 
     FileSys::PatchManager pm{title_id};
 
-    IStorage storage(pm.PatchRomFS(std::move(data.Unwrap()), 0, FileSys::ContentRecordType::Data));
+    auto storage = std::make_shared<IStorage>(
+        pm.PatchRomFS(std::move(data.Unwrap()), 0, FileSys::ContentRecordType::Data));
 
     IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);

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

@@ -250,6 +250,10 @@ void Controller_NPad::RequestPadStateUpdate(u32 npad_id) {
     auto& rstick_entry = npad_pad_states[controller_idx].r_stick;
     const auto& button_state = buttons[controller_idx];
     const auto& analog_state = sticks[controller_idx];
+    const auto [stick_l_x_f, stick_l_y_f] =
+        analog_state[static_cast<std::size_t>(JoystickId::Joystick_Left)]->GetStatus();
+    const auto [stick_r_x_f, stick_r_y_f] =
+        analog_state[static_cast<std::size_t>(JoystickId::Joystick_Right)]->GetStatus();
 
     using namespace Settings::NativeButton;
     pad_state.a.Assign(button_state[A - BUTTON_HID_BEGIN]->GetStatus());
@@ -270,23 +274,32 @@ void Controller_NPad::RequestPadStateUpdate(u32 npad_id) {
     pad_state.d_right.Assign(button_state[DRight - BUTTON_HID_BEGIN]->GetStatus());
     pad_state.d_down.Assign(button_state[DDown - BUTTON_HID_BEGIN]->GetStatus());
 
-    pad_state.l_stick_left.Assign(button_state[LStick_Left - BUTTON_HID_BEGIN]->GetStatus());
-    pad_state.l_stick_up.Assign(button_state[LStick_Up - BUTTON_HID_BEGIN]->GetStatus());
-    pad_state.l_stick_right.Assign(button_state[LStick_Right - BUTTON_HID_BEGIN]->GetStatus());
-    pad_state.l_stick_down.Assign(button_state[LStick_Down - BUTTON_HID_BEGIN]->GetStatus());
+    pad_state.l_stick_right.Assign(
+        analog_state[static_cast<std::size_t>(JoystickId::Joystick_Left)]->GetAnalogDirectionStatus(
+            Input::AnalogDirection::RIGHT));
+    pad_state.l_stick_left.Assign(
+        analog_state[static_cast<std::size_t>(JoystickId::Joystick_Left)]->GetAnalogDirectionStatus(
+            Input::AnalogDirection::LEFT));
+    pad_state.l_stick_up.Assign(
+        analog_state[static_cast<std::size_t>(JoystickId::Joystick_Left)]->GetAnalogDirectionStatus(
+            Input::AnalogDirection::UP));
+    pad_state.l_stick_down.Assign(
+        analog_state[static_cast<std::size_t>(JoystickId::Joystick_Left)]->GetAnalogDirectionStatus(
+            Input::AnalogDirection::DOWN));
 
-    pad_state.r_stick_left.Assign(button_state[RStick_Left - BUTTON_HID_BEGIN]->GetStatus());
-    pad_state.r_stick_up.Assign(button_state[RStick_Up - BUTTON_HID_BEGIN]->GetStatus());
-    pad_state.r_stick_right.Assign(button_state[RStick_Right - BUTTON_HID_BEGIN]->GetStatus());
-    pad_state.r_stick_down.Assign(button_state[RStick_Down - BUTTON_HID_BEGIN]->GetStatus());
+    pad_state.r_stick_up.Assign(analog_state[static_cast<std::size_t>(JoystickId::Joystick_Right)]
+                                    ->GetAnalogDirectionStatus(Input::AnalogDirection::RIGHT));
+    pad_state.r_stick_left.Assign(analog_state[static_cast<std::size_t>(JoystickId::Joystick_Right)]
+                                      ->GetAnalogDirectionStatus(Input::AnalogDirection::LEFT));
+    pad_state.r_stick_right.Assign(
+        analog_state[static_cast<std::size_t>(JoystickId::Joystick_Right)]
+            ->GetAnalogDirectionStatus(Input::AnalogDirection::UP));
+    pad_state.r_stick_down.Assign(analog_state[static_cast<std::size_t>(JoystickId::Joystick_Right)]
+                                      ->GetAnalogDirectionStatus(Input::AnalogDirection::DOWN));
 
     pad_state.left_sl.Assign(button_state[SL - BUTTON_HID_BEGIN]->GetStatus());
     pad_state.left_sr.Assign(button_state[SR - BUTTON_HID_BEGIN]->GetStatus());
 
-    const auto [stick_l_x_f, stick_l_y_f] =
-        analog_state[static_cast<std::size_t>(JoystickId::Joystick_Left)]->GetStatus();
-    const auto [stick_r_x_f, stick_r_y_f] =
-        analog_state[static_cast<std::size_t>(JoystickId::Joystick_Right)]->GetStatus();
     lstick_entry.x = static_cast<s32>(stick_l_x_f * HID_JOYSTICK_MAX);
     lstick_entry.y = static_cast<s32>(stick_l_y_f * HID_JOYSTICK_MAX);
     rstick_entry.x = static_cast<s32>(stick_r_x_f * HID_JOYSTICK_MAX);

src/core/hle/service/prepo/prepo.cpp

@@ -50,16 +50,16 @@ private:
         IPC::RequestParser rp{ctx};
         const auto process_id = rp.PopRaw<u64>();
 
-        const auto data1 = ctx.ReadBuffer(0);
-        const auto data2 = ctx.ReadBuffer(1);
+        std::vector<std::vector<u8>> data{ctx.ReadBuffer(0)};
+        if (Type == Core::Reporter::PlayReportType::New) {
+            data.emplace_back(ctx.ReadBuffer(1));
+        }
 
-        LOG_DEBUG(Service_PREPO,
-                  "called, type={:02X}, process_id={:016X}, data1_size={:016X}, data2_size={:016X}",
-                  static_cast<u8>(Type), process_id, data1.size(), data2.size());
+        LOG_DEBUG(Service_PREPO, "called, type={:02X}, process_id={:016X}, data1_size={:016X}",
+                  static_cast<u8>(Type), process_id, data[0].size());
 
         const auto& reporter{system.GetReporter()};
-        reporter.SavePlayReport(Type, system.CurrentProcess()->GetTitleID(), {data1, data2},
-                                process_id);
+        reporter.SavePlayReport(Type, system.CurrentProcess()->GetTitleID(), data, process_id);
 
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(RESULT_SUCCESS);
@@ -70,19 +70,19 @@ private:
         IPC::RequestParser rp{ctx};
         const auto user_id = rp.PopRaw<u128>();
         const auto process_id = rp.PopRaw<u64>();
-
-        const auto data1 = ctx.ReadBuffer(0);
-        const auto data2 = ctx.ReadBuffer(1);
+        std::vector<std::vector<u8>> data{ctx.ReadBuffer(0)};
+        if (Type == Core::Reporter::PlayReportType::New) {
+            data.emplace_back(ctx.ReadBuffer(1));
+        }
 
         LOG_DEBUG(
             Service_PREPO,
-            "called, type={:02X}, user_id={:016X}{:016X}, process_id={:016X}, data1_size={:016X}, "
-            "data2_size={:016X}",
-            static_cast<u8>(Type), user_id[1], user_id[0], process_id, data1.size(), data2.size());
+            "called, type={:02X}, user_id={:016X}{:016X}, process_id={:016X}, data1_size={:016X}",
+            static_cast<u8>(Type), user_id[1], user_id[0], process_id, data[0].size());
 
         const auto& reporter{system.GetReporter()};
-        reporter.SavePlayReport(Type, system.CurrentProcess()->GetTitleID(), {data1, data2},
-                                process_id, user_id);
+        reporter.SavePlayReport(Type, system.CurrentProcess()->GetTitleID(), data, process_id,
+                                user_id);
 
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(RESULT_SUCCESS);

src/core/hle/service/sockets/bsd.cpp

@@ -42,6 +42,26 @@ void BSD::Socket(Kernel::HLERequestContext& ctx) {
     rb.Push<u32>(0); // bsd errno
 }
 
+void BSD::Select(Kernel::HLERequestContext& ctx) {
+    LOG_WARNING(Service, "(STUBBED) called");
+
+    IPC::ResponseBuilder rb{ctx, 4};
+
+    rb.Push(RESULT_SUCCESS);
+    rb.Push<u32>(0); // ret
+    rb.Push<u32>(0); // bsd errno
+}
+
+void BSD::Bind(Kernel::HLERequestContext& ctx) {
+    LOG_WARNING(Service, "(STUBBED) called");
+
+    IPC::ResponseBuilder rb{ctx, 4};
+
+    rb.Push(RESULT_SUCCESS);
+    rb.Push<u32>(0); // ret
+    rb.Push<u32>(0); // bsd errno
+}
+
 void BSD::Connect(Kernel::HLERequestContext& ctx) {
     LOG_WARNING(Service, "(STUBBED) called");
 
@@ -52,6 +72,26 @@ void BSD::Connect(Kernel::HLERequestContext& ctx) {
     rb.Push<u32>(0); // bsd errno
 }
 
+void BSD::Listen(Kernel::HLERequestContext& ctx) {
+    LOG_WARNING(Service, "(STUBBED) called");
+
+    IPC::ResponseBuilder rb{ctx, 4};
+
+    rb.Push(RESULT_SUCCESS);
+    rb.Push<u32>(0); // ret
+    rb.Push<u32>(0); // bsd errno
+}
+
+void BSD::SetSockOpt(Kernel::HLERequestContext& ctx) {
+    LOG_WARNING(Service, "(STUBBED) called");
+
+    IPC::ResponseBuilder rb{ctx, 4};
+
+    rb.Push(RESULT_SUCCESS);
+    rb.Push<u32>(0); // ret
+    rb.Push<u32>(0); // bsd errno
+}
+
 void BSD::SendTo(Kernel::HLERequestContext& ctx) {
     LOG_WARNING(Service, "(STUBBED) called");
 
@@ -80,7 +120,7 @@ BSD::BSD(const char* name) : ServiceFramework(name) {
         {2, &BSD::Socket, "Socket"},
         {3, nullptr, "SocketExempt"},
         {4, nullptr, "Open"},
-        {5, nullptr, "Select"},
+        {5, &BSD::Select, "Select"},
         {6, nullptr, "Poll"},
         {7, nullptr, "Sysctl"},
         {8, nullptr, "Recv"},
@@ -88,15 +128,15 @@ BSD::BSD(const char* name) : ServiceFramework(name) {
         {10, nullptr, "Send"},
         {11, &BSD::SendTo, "SendTo"},
         {12, nullptr, "Accept"},
-        {13, nullptr, "Bind"},
+        {13, &BSD::Bind, "Bind"},
         {14, &BSD::Connect, "Connect"},
         {15, nullptr, "GetPeerName"},
         {16, nullptr, "GetSockName"},
         {17, nullptr, "GetSockOpt"},
-        {18, nullptr, "Listen"},
+        {18, &BSD::Listen, "Listen"},
         {19, nullptr, "Ioctl"},
         {20, nullptr, "Fcntl"},
-        {21, nullptr, "SetSockOpt"},
+        {21, &BSD::SetSockOpt, "SetSockOpt"},
         {22, nullptr, "Shutdown"},
         {23, nullptr, "ShutdownAllSockets"},
         {24, nullptr, "Write"},

src/core/hle/service/sockets/bsd.h

@@ -18,7 +18,11 @@ private:
     void RegisterClient(Kernel::HLERequestContext& ctx);
     void StartMonitoring(Kernel::HLERequestContext& ctx);
     void Socket(Kernel::HLERequestContext& ctx);
+    void Select(Kernel::HLERequestContext& ctx);
+    void Bind(Kernel::HLERequestContext& ctx);
     void Connect(Kernel::HLERequestContext& ctx);
+    void Listen(Kernel::HLERequestContext& ctx);
+    void SetSockOpt(Kernel::HLERequestContext& ctx);
     void SendTo(Kernel::HLERequestContext& ctx);
     void Close(Kernel::HLERequestContext& ctx);
 

src/core/settings.h

@@ -371,6 +371,11 @@ enum class SDMCSize : u64 {
     S1TB = 0x10000000000ULL,
 };
 
+enum class RendererBackend {
+    OpenGL = 0,
+    Vulkan = 1,
+};
+
 struct Values {
     // System
     bool use_docked_mode;
@@ -419,6 +424,10 @@ struct Values {
     SDMCSize sdmc_size;
 
     // Renderer
+    RendererBackend renderer_backend;
+    bool renderer_debug;
+    int vulkan_device;
+
     float resolution_factor;
     bool use_frame_limit;
     u16 frame_limit;

src/core/telemetry_session.cpp

@@ -46,6 +46,16 @@ static u64 GenerateTelemetryId() {
     return telemetry_id;
 }
 
+static const char* TranslateRenderer(Settings::RendererBackend backend) {
+    switch (backend) {
+    case Settings::RendererBackend::OpenGL:
+        return "OpenGL";
+    case Settings::RendererBackend::Vulkan:
+        return "Vulkan";
+    }
+    return "Unknown";
+}
+
 u64 GetTelemetryId() {
     u64 telemetry_id{};
     const std::string filename{FileUtil::GetUserPath(FileUtil::UserPath::ConfigDir) +
@@ -169,7 +179,7 @@ void TelemetrySession::AddInitialInfo(Loader::AppLoader& app_loader) {
     AddField(field_type, "Audio_SinkId", Settings::values.sink_id);
     AddField(field_type, "Audio_EnableAudioStretching", Settings::values.enable_audio_stretching);
     AddField(field_type, "Core_UseMultiCore", Settings::values.use_multi_core);
-    AddField(field_type, "Renderer_Backend", "OpenGL");
+    AddField(field_type, "Renderer_Backend", TranslateRenderer(Settings::values.renderer_backend));
     AddField(field_type, "Renderer_ResolutionFactor", Settings::values.resolution_factor);
     AddField(field_type, "Renderer_UseFrameLimit", Settings::values.use_frame_limit);
     AddField(field_type, "Renderer_FrameLimit", Settings::values.frame_limit);

src/input_common/main.cpp

@@ -41,6 +41,7 @@ void Shutdown() {
     Input::UnregisterFactory<Input::MotionDevice>("motion_emu");
     motion_emu.reset();
     sdl.reset();
+    udp.reset();
 }
 
 Keyboard* GetKeyboard() {

src/input_common/sdl/sdl_impl.cpp

@@ -342,6 +342,22 @@ public:
         return std::make_tuple<float, float>(0.0f, 0.0f);
     }
 
+    bool GetAnalogDirectionStatus(Input::AnalogDirection direction) const override {
+        const auto [x, y] = GetStatus();
+        const float directional_deadzone = 0.4f;
+        switch (direction) {
+        case Input::AnalogDirection::RIGHT:
+            return x > directional_deadzone;
+        case Input::AnalogDirection::LEFT:
+            return x < -directional_deadzone;
+        case Input::AnalogDirection::UP:
+            return y > directional_deadzone;
+        case Input::AnalogDirection::DOWN:
+            return y < -directional_deadzone;
+        }
+        return false;
+    }
+
 private:
     std::shared_ptr<SDLJoystick> joystick;
     const int axis_x;

src/input_common/udp/client.cpp

@@ -14,7 +14,6 @@
 #include "input_common/udp/client.h"
 #include "input_common/udp/protocol.h"
 
-using boost::asio::ip::address_v4;
 using boost::asio::ip::udp;
 
 namespace InputCommon::CemuhookUDP {
@@ -31,10 +30,10 @@ public:
 
     explicit Socket(const std::string& host, u16 port, u8 pad_index, u32 client_id,
                     SocketCallback callback)
-        : client_id(client_id), timer(io_service),
-          send_endpoint(udp::endpoint(address_v4::from_string(host), port)),
-          socket(io_service, udp::endpoint(udp::v4(), 0)), pad_index(pad_index),
-          callback(std::move(callback)) {}
+        : callback(std::move(callback)), timer(io_service),
+          socket(io_service, udp::endpoint(udp::v4(), 0)), client_id(client_id),
+          pad_index(pad_index),
+          send_endpoint(udp::endpoint(boost::asio::ip::make_address_v4(host), port)) {}
 
     void Stop() {
         io_service.stop();
@@ -126,7 +125,7 @@ static void SocketLoop(Socket* socket) {
 
 Client::Client(std::shared_ptr<DeviceStatus> status, const std::string& host, u16 port,
                u8 pad_index, u32 client_id)
-    : status(status) {
+    : status(std::move(status)) {
     StartCommunication(host, port, pad_index, client_id);
 }
 
@@ -207,7 +206,7 @@ void TestCommunication(const std::string& host, u16 port, u8 pad_index, u32 clie
         Common::Event success_event;
         SocketCallback callback{[](Response::Version version) {}, [](Response::PortInfo info) {},
                                 [&](Response::PadData data) { success_event.Set(); }};
-        Socket socket{host, port, pad_index, client_id, callback};
+        Socket socket{host, port, pad_index, client_id, std::move(callback)};
         std::thread worker_thread{SocketLoop, &socket};
         bool result = success_event.WaitFor(std::chrono::seconds(8));
         socket.Stop();
@@ -267,7 +266,7 @@ CalibrationConfigurationJob::CalibrationConfigurationJob(
                                         complete_event.Set();
                                     }
                                 }};
-        Socket socket{host, port, pad_index, client_id, callback};
+        Socket socket{host, port, pad_index, client_id, std::move(callback)};
         std::thread worker_thread{SocketLoop, &socket};
         complete_event.Wait();
         socket.Stop();

src/input_common/udp/client.h

@@ -11,7 +11,6 @@
 #include <string>
 #include <thread>
 #include <tuple>
-#include <vector>
 #include "common/common_types.h"
 #include "common/thread.h"
 #include "common/vector_math.h"

src/input_common/udp/protocol.h

@@ -7,7 +7,6 @@
 #include <array>
 #include <optional>
 #include <type_traits>
-#include <vector>
 #include <boost/crc.hpp>
 #include "common/bit_field.h"
 #include "common/swap.h"

src/input_common/udp/udp.cpp

@@ -2,7 +2,9 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include "common/logging/log.h"
+#include <mutex>
+#include <tuple>
+
 #include "common/param_package.h"
 #include "core/frontend/input.h"
 #include "core/settings.h"
@@ -14,7 +16,7 @@ namespace InputCommon::CemuhookUDP {
 class UDPTouchDevice final : public Input::TouchDevice {
 public:
     explicit UDPTouchDevice(std::shared_ptr<DeviceStatus> status_) : status(std::move(status_)) {}
-    std::tuple<float, float, bool> GetStatus() const {
+    std::tuple<float, float, bool> GetStatus() const override {
         std::lock_guard guard(status->update_mutex);
         return status->touch_status;
     }
@@ -26,7 +28,7 @@ private:
 class UDPMotionDevice final : public Input::MotionDevice {
 public:
     explicit UDPMotionDevice(std::shared_ptr<DeviceStatus> status_) : status(std::move(status_)) {}
-    std::tuple<Common::Vec3<float>, Common::Vec3<float>> GetStatus() const {
+    std::tuple<Common::Vec3<float>, Common::Vec3<float>> GetStatus() const override {
         std::lock_guard guard(status->update_mutex);
         return status->motion_status;
     }

src/input_common/udp/udp.h

@@ -2,15 +2,13 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#pragma once
+
 #include <memory>
-#include <unordered_map>
-#include "input_common/main.h"
-#include "input_common/udp/client.h"
 
 namespace InputCommon::CemuhookUDP {
 
-class UDPTouchDevice;
-class UDPMotionDevice;
+class Client;
 
 class State {
 public:

src/video_core/CMakeLists.txt

@@ -29,6 +29,8 @@ add_library(video_core STATIC
     gpu_synch.h
     gpu_thread.cpp
     gpu_thread.h
+    guest_driver.cpp
+    guest_driver.h
     macro_interpreter.cpp
     macro_interpreter.h
     memory_manager.cpp
@@ -154,6 +156,7 @@ if (ENABLE_VULKAN)
         renderer_vulkan/maxwell_to_vk.cpp
         renderer_vulkan/maxwell_to_vk.h
         renderer_vulkan/renderer_vulkan.h
+        renderer_vulkan/renderer_vulkan.cpp
         renderer_vulkan/vk_blit_screen.cpp
         renderer_vulkan/vk_blit_screen.h
         renderer_vulkan/vk_buffer_cache.cpp

src/video_core/buffer_cache/buffer_cache.h

@@ -101,7 +101,10 @@ public:
     void TickFrame() {
         ++epoch;
         while (!pending_destruction.empty()) {
-            if (pending_destruction.front()->GetEpoch() + 1 > epoch) {
+            // Delay at least 4 frames before destruction.
+            // This is due to triple buffering happening on some drivers.
+            static constexpr u64 epochs_to_destroy = 5;
+            if (pending_destruction.front()->GetEpoch() + epochs_to_destroy > epoch) {
                 break;
             }
             pending_destruction.pop_front();

src/video_core/engines/const_buffer_engine_interface.h

@@ -9,6 +9,7 @@
 #include "common/common_types.h"
 #include "video_core/engines/shader_bytecode.h"
 #include "video_core/engines/shader_type.h"
+#include "video_core/guest_driver.h"
 #include "video_core/textures/texture.h"
 
 namespace Tegra::Engines {
@@ -106,6 +107,9 @@ public:
     virtual SamplerDescriptor AccessBindlessSampler(ShaderType stage, u64 const_buffer,
                                                     u64 offset) const = 0;
     virtual u32 GetBoundBuffer() const = 0;
+
+    virtual VideoCore::GuestDriverProfile& AccessGuestDriverProfile() = 0;
+    virtual const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const = 0;
 };
 
 } // namespace Tegra::Engines

src/video_core/engines/kepler_compute.cpp

@@ -94,6 +94,14 @@ SamplerDescriptor KeplerCompute::AccessBindlessSampler(ShaderType stage, u64 con
     return result;
 }
 
+VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() {
+    return rasterizer.AccessGuestDriverProfile();
+}
+
+const VideoCore::GuestDriverProfile& KeplerCompute::AccessGuestDriverProfile() const {
+    return rasterizer.AccessGuestDriverProfile();
+}
+
 void KeplerCompute::ProcessLaunch() {
     const GPUVAddr launch_desc_loc = regs.launch_desc_loc.Address();
     memory_manager.ReadBlockUnsafe(launch_desc_loc, &launch_description,

src/video_core/engines/kepler_compute.h

@@ -218,6 +218,10 @@ public:
         return regs.tex_cb_index;
     }
 
+    VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override;
+
+    const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override;
+
 private:
     Core::System& system;
     VideoCore::RasterizerInterface& rasterizer;

src/video_core/engines/maxwell_3d.cpp

@@ -784,4 +784,12 @@ SamplerDescriptor Maxwell3D::AccessBindlessSampler(ShaderType stage, u64 const_b
     return result;
 }
 
+VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() {
+    return rasterizer.AccessGuestDriverProfile();
+}
+
+const VideoCore::GuestDriverProfile& Maxwell3D::AccessGuestDriverProfile() const {
+    return rasterizer.AccessGuestDriverProfile();
+}
+
 } // namespace Tegra::Engines

src/video_core/engines/maxwell_3d.h

@@ -704,8 +704,8 @@ public:
                 INSERT_UNION_PADDING_WORDS(0x15);
 
                 s32 stencil_back_func_ref;
-                u32 stencil_back_func_mask;
                 u32 stencil_back_mask;
+                u32 stencil_back_func_mask;
 
                 INSERT_UNION_PADDING_WORDS(0xC);
 
@@ -862,7 +862,11 @@ public:
 
                 float point_size;
 
-                INSERT_UNION_PADDING_WORDS(0x7);
+                INSERT_UNION_PADDING_WORDS(0x1);
+
+                u32 point_sprite_enable;
+
+                INSERT_UNION_PADDING_WORDS(0x5);
 
                 u32 zeta_enable;
 
@@ -1306,6 +1310,10 @@ public:
         return regs.tex_cb_index;
     }
 
+    VideoCore::GuestDriverProfile& AccessGuestDriverProfile() override;
+
+    const VideoCore::GuestDriverProfile& AccessGuestDriverProfile() const override;
+
     /// Memory for macro code - it's undetermined how big this is, however 1MB is much larger than
     /// we've seen used.
     using MacroMemory = std::array<u32, 0x40000>;
@@ -1454,8 +1462,8 @@ ASSERT_REG_POSITION(polygon_offset_fill_enable, 0x372);
 ASSERT_REG_POSITION(patch_vertices, 0x373);
 ASSERT_REG_POSITION(scissor_test, 0x380);
 ASSERT_REG_POSITION(stencil_back_func_ref, 0x3D5);
-ASSERT_REG_POSITION(stencil_back_func_mask, 0x3D6);
-ASSERT_REG_POSITION(stencil_back_mask, 0x3D7);
+ASSERT_REG_POSITION(stencil_back_mask, 0x3D6);
+ASSERT_REG_POSITION(stencil_back_func_mask, 0x3D7);
 ASSERT_REG_POSITION(color_mask_common, 0x3E4);
 ASSERT_REG_POSITION(rt_separate_frag_data, 0x3EB);
 ASSERT_REG_POSITION(depth_bounds, 0x3E7);
@@ -1490,6 +1498,7 @@ ASSERT_REG_POSITION(vb_element_base, 0x50D);
 ASSERT_REG_POSITION(vb_base_instance, 0x50E);
 ASSERT_REG_POSITION(clip_distance_enabled, 0x544);
 ASSERT_REG_POSITION(point_size, 0x546);
+ASSERT_REG_POSITION(point_sprite_enable, 0x548);
 ASSERT_REG_POSITION(zeta_enable, 0x54E);
 ASSERT_REG_POSITION(multisample_control, 0x54F);
 ASSERT_REG_POSITION(condition, 0x554);

src/video_core/engines/shader_bytecode.h

@@ -227,6 +227,28 @@ enum class AtomicOp : u64 {
     Exch = 8,
 };
 
+enum class GlobalAtomicOp : u64 {
+    Add = 0,
+    Min = 1,
+    Max = 2,
+    Inc = 3,
+    Dec = 4,
+    And = 5,
+    Or = 6,
+    Xor = 7,
+    Exch = 8,
+    SafeAdd = 10,
+};
+
+enum class GlobalAtomicType : u64 {
+    U32 = 0,
+    S32 = 1,
+    U64 = 2,
+    F32_FTZ_RN = 3,
+    F16x2_FTZ_RN = 4,
+    S64 = 5,
+};
+
 enum class UniformType : u64 {
     UnsignedByte = 0,
     SignedByte = 1,
@@ -602,6 +624,19 @@ enum class ShuffleOperation : u64 {
     Bfly = 3, // shuffleXorNV
 };
 
+enum class ShfType : u64 {
+    Bits32 = 0,
+    U64 = 2,
+    S64 = 3,
+};
+
+enum class ShfXmode : u64 {
+    None = 0,
+    HI = 1,
+    X = 2,
+    XHI = 3,
+};
+
 union Instruction {
     constexpr Instruction& operator=(const Instruction& instr) {
         value = instr.value;
@@ -753,6 +788,13 @@ union Instruction {
         BitField<39, 1, u64> wrap;
     } shr;
 
+    union {
+        BitField<37, 2, ShfType> type;
+        BitField<48, 2, ShfXmode> xmode;
+        BitField<50, 1, u64> wrap;
+        BitField<20, 6, u64> immediate;
+    } shf;
+
     union {
         BitField<39, 5, u64> shift_amount;
         BitField<48, 1, u64> negate_b;
@@ -957,6 +999,12 @@ union Instruction {
         BitField<46, 2, u64> cache_mode;
     } stg;
 
+    union {
+        BitField<52, 4, GlobalAtomicOp> operation;
+        BitField<49, 3, GlobalAtomicType> type;
+        BitField<28, 20, s64> offset;
+    } atom;
+
     union {
         BitField<52, 4, AtomicOp> operation;
         BitField<28, 2, AtomicType> type;
@@ -1095,6 +1143,11 @@ union Instruction {
         BitField<55, 1, u64> ftz;
     } fset;
 
+    union {
+        BitField<47, 1, u64> ftz;
+        BitField<48, 4, PredCondition> cond;
+    } fcmp;
+
     union {
         BitField<49, 1, u64> bf;
         BitField<35, 3, PredCondition> cond;
@@ -1675,6 +1728,7 @@ public:
         BFE_C,
         BFE_R,
         BFE_IMM,
+        BFI_RC,
         BFI_IMM_R,
         BRA,
         BRX,
@@ -1690,6 +1744,7 @@ public:
         ST_S,
         ST,    // Store in generic memory
         STG,   // Store in global memory
+        ATOM,  // Atomic operation on global memory
         ATOMS, // Atomic operation on shared memory
         AL2P,  // Transforms attribute memory into physical memory
         TEX,
@@ -1771,6 +1826,7 @@ public:
         ICMP_R,
         ICMP_CR,
         ICMP_IMM,
+        FCMP_R,
         MUFU,  // Multi-Function Operator
         RRO_C, // Range Reduction Operator
         RRO_R,
@@ -1994,6 +2050,7 @@ private:
             INST("1110111101010---", Id::ST_L, Type::Memory, "ST_L"),
             INST("101-------------", Id::ST, Type::Memory, "ST"),
             INST("1110111011011---", Id::STG, Type::Memory, "STG"),
+            INST("11101101--------", Id::ATOM, Type::Memory, "ATOM"),
             INST("11101100--------", Id::ATOMS, Type::Memory, "ATOMS"),
             INST("1110111110100---", Id::AL2P, Type::Memory, "AL2P"),
             INST("110000----111---", Id::TEX, Type::Texture, "TEX"),
@@ -2074,6 +2131,7 @@ private:
             INST("0101110100100---", Id::HSETP2_R, Type::HalfSetPredicate, "HSETP2_R"),
             INST("0111111-0-------", Id::HSETP2_IMM, Type::HalfSetPredicate, "HSETP2_IMM"),
             INST("0101110100011---", Id::HSET2_R, Type::HalfSet, "HSET2_R"),
+            INST("010110111010----", Id::FCMP_R, Type::Arithmetic, "FCMP_R"),
             INST("0101000010000---", Id::MUFU, Type::Arithmetic, "MUFU"),
             INST("0100110010010---", Id::RRO_C, Type::Arithmetic, "RRO_C"),
             INST("0101110010010---", Id::RRO_R, Type::Arithmetic, "RRO_R"),
@@ -2098,6 +2156,7 @@ private:
             INST("0100110000000---", Id::BFE_C, Type::Bfe, "BFE_C"),
             INST("0101110000000---", Id::BFE_R, Type::Bfe, "BFE_R"),
             INST("0011100-00000---", Id::BFE_IMM, Type::Bfe, "BFE_IMM"),
+            INST("0101001111110---", Id::BFI_RC, Type::Bfi, "BFI_RC"),
             INST("0011011-11110---", Id::BFI_IMM_R, Type::Bfi, "BFI_IMM_R"),
             INST("0100110001000---", Id::LOP_C, Type::ArithmeticInteger, "LOP_C"),
             INST("0101110001000---", Id::LOP_R, Type::ArithmeticInteger, "LOP_R"),

src/video_core/gpu.cpp

@@ -23,7 +23,7 @@ MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
 GPU::GPU(Core::System& system, VideoCore::RendererBase& renderer, bool is_async)
     : system{system}, renderer{renderer}, is_async{is_async} {
     auto& rasterizer{renderer.Rasterizer()};
-    memory_manager = std::make_unique<Tegra::MemoryManager>(system, rasterizer);
+    memory_manager = std::make_unique<Tegra::MemoryManager>(system);
     dma_pusher = std::make_unique<Tegra::DmaPusher>(*this);
     maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager);
     fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer);

src/video_core/gpu_thread.h

@@ -86,7 +86,7 @@ struct CommandDataContainer {
 struct SynchState final {
     std::atomic_bool is_running{true};
 
-    using CommandQueue = Common::SPSCQueue<CommandDataContainer>;
+    using CommandQueue = Common::MPSCQueue<CommandDataContainer>;
     CommandQueue queue;
     u64 last_fence{};
     std::atomic<u64> signaled_fence{};

src/video_core/guest_driver.cpp

@@ -0,0 +1,36 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <limits>
+
+#include "video_core/guest_driver.h"
+
+namespace VideoCore {
+
+void GuestDriverProfile::DeduceTextureHandlerSize(std::vector<u32>&& bound_offsets) {
+    if (texture_handler_size_deduced) {
+        return;
+    }
+    const std::size_t size = bound_offsets.size();
+    if (size < 2) {
+        return;
+    }
+    std::sort(bound_offsets.begin(), bound_offsets.end(), std::less{});
+    u32 min_val = std::numeric_limits<u32>::max();
+    for (std::size_t i = 1; i < size; ++i) {
+        if (bound_offsets[i] == bound_offsets[i - 1]) {
+            continue;
+        }
+        const u32 new_min = bound_offsets[i] - bound_offsets[i - 1];
+        min_val = std::min(min_val, new_min);
+    }
+    if (min_val > 2) {
+        return;
+    }
+    texture_handler_size_deduced = true;
+    texture_handler_size = min_texture_handler_size * min_val;
+}
+
+} // namespace VideoCore

src/video_core/guest_driver.h

@@ -0,0 +1,41 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <vector>
+
+#include "common/common_types.h"
+
+namespace VideoCore {
+
+/**
+ * The GuestDriverProfile class is used to learn about the GPU drivers behavior and collect
+ * information necessary for impossible to avoid HLE methods like shader tracks as they are
+ * Entscheidungsproblems.
+ */
+class GuestDriverProfile {
+public:
+    void DeduceTextureHandlerSize(std::vector<u32>&& bound_offsets);
+
+    u32 GetTextureHandlerSize() const {
+        return texture_handler_size;
+    }
+
+    bool TextureHandlerSizeKnown() const {
+        return texture_handler_size_deduced;
+    }
+
+private:
+    // Minimum size of texture handler any driver can use.
+    static constexpr u32 min_texture_handler_size = 4;
+    // This goes with Vulkan and OpenGL standards but Nvidia GPUs can easily
+    // use 4 bytes instead. Thus, certain drivers may squish the size.
+    static constexpr u32 default_texture_handler_size = 8;
+
+    u32 texture_handler_size = default_texture_handler_size;
+    bool texture_handler_size_deduced = false;
+};
+
+} // namespace VideoCore

src/video_core/memory_manager.cpp

@@ -9,13 +9,12 @@
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
+#include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
-#include "video_core/rasterizer_interface.h"
 
 namespace Tegra {
 
-MemoryManager::MemoryManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer)
-    : rasterizer{rasterizer}, system{system} {
+MemoryManager::MemoryManager(Core::System& system) : system{system} {
     std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr);
     std::fill(page_table.attributes.begin(), page_table.attributes.end(),
               Common::PageType::Unmapped);
@@ -84,7 +83,8 @@ GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) {
     const auto cpu_addr = GpuToCpuAddress(gpu_addr);
     ASSERT(cpu_addr);
 
-    rasterizer.FlushAndInvalidateRegion(cache_addr, aligned_size);
+    system.GPU().FlushAndInvalidateRegion(cache_addr, aligned_size);
+
     UnmapRange(gpu_addr, aligned_size);
     ASSERT(system.CurrentProcess()
                ->VMManager()
@@ -242,7 +242,7 @@ void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, const std::s
         switch (page_table.attributes[page_index]) {
         case Common::PageType::Memory: {
             const u8* src_ptr{page_table.pointers[page_index] + page_offset};
-            rasterizer.FlushRegion(ToCacheAddr(src_ptr), copy_amount);
+            system.GPU().FlushRegion(ToCacheAddr(src_ptr), copy_amount);
             std::memcpy(dest_buffer, src_ptr, copy_amount);
             break;
         }
@@ -292,7 +292,7 @@ void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, const
         switch (page_table.attributes[page_index]) {
         case Common::PageType::Memory: {
             u8* dest_ptr{page_table.pointers[page_index] + page_offset};
-            rasterizer.InvalidateRegion(ToCacheAddr(dest_ptr), copy_amount);
+            system.GPU().InvalidateRegion(ToCacheAddr(dest_ptr), copy_amount);
             std::memcpy(dest_ptr, src_buffer, copy_amount);
             break;
         }
@@ -340,7 +340,7 @@ void MemoryManager::CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, const std::
         switch (page_table.attributes[page_index]) {
         case Common::PageType::Memory: {
             const u8* src_ptr{page_table.pointers[page_index] + page_offset};
-            rasterizer.FlushRegion(ToCacheAddr(src_ptr), copy_amount);
+            system.GPU().FlushRegion(ToCacheAddr(src_ptr), copy_amount);
             WriteBlock(dest_addr, src_ptr, copy_amount);
             break;
         }

src/video_core/memory_manager.h

@@ -10,10 +10,6 @@
 #include "common/common_types.h"
 #include "common/page_table.h"
 
-namespace VideoCore {
-class RasterizerInterface;
-}
-
 namespace Core {
 class System;
 }
@@ -51,7 +47,7 @@ struct VirtualMemoryArea {
 
 class MemoryManager final {
 public:
-    explicit MemoryManager(Core::System& system, VideoCore::RasterizerInterface& rasterizer);
+    explicit MemoryManager(Core::System& system);
     ~MemoryManager();
 
     GPUVAddr AllocateSpace(u64 size, u64 align);
@@ -176,7 +172,6 @@ private:
 
     Common::PageTable page_table{page_bits};
     VMAMap vma_map;
-    VideoCore::RasterizerInterface& rasterizer;
 
     Core::System& system;
 };

src/video_core/rasterizer_interface.h

@@ -9,6 +9,7 @@
 #include "common/common_types.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/gpu.h"
+#include "video_core/guest_driver.h"
 
 namespace Tegra {
 class MemoryManager;
@@ -78,5 +79,18 @@ public:
     /// Initialize disk cached resources for the game being emulated
     virtual void LoadDiskResources(const std::atomic_bool& stop_loading = false,
                                    const DiskResourceLoadCallback& callback = {}) {}
+
+    /// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver.
+    GuestDriverProfile& AccessGuestDriverProfile() {
+        return guest_driver_profile;
+    }
+
+    /// Grant access to the Guest Driver Profile for recording/obtaining info on the guest driver.
+    const GuestDriverProfile& AccessGuestDriverProfile() const {
+        return guest_driver_profile;
+    }
+
+private:
+    GuestDriverProfile guest_driver_profile{};
 };
 } // namespace VideoCore

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -55,16 +55,20 @@ namespace {
 
 template <typename Engine, typename Entry>
 Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry& entry,
-                                               Tegra::Engines::ShaderType shader_type) {
+                                               Tegra::Engines::ShaderType shader_type,
+                                               std::size_t index = 0) {
     if (entry.IsBindless()) {
         const Tegra::Texture::TextureHandle tex_handle =
             engine.AccessConstBuffer32(shader_type, entry.GetBuffer(), entry.GetOffset());
         return engine.GetTextureInfo(tex_handle);
     }
+    const auto& gpu_profile = engine.AccessGuestDriverProfile();
+    const u32 offset =
+        entry.GetOffset() + static_cast<u32>(index * gpu_profile.GetTextureHandlerSize());
     if constexpr (std::is_same_v<Engine, Tegra::Engines::Maxwell3D>) {
-        return engine.GetStageTexture(shader_type, entry.GetOffset());
+        return engine.GetStageTexture(shader_type, offset);
     } else {
-        return engine.GetTexture(entry.GetOffset());
+        return engine.GetTexture(offset);
     }
 }
 
@@ -244,9 +248,6 @@ void RasterizerOpenGL::SetupVertexInstances(GLuint vao) {
 }
 
 GLintptr RasterizerOpenGL::SetupIndexBuffer() {
-    if (accelerate_draw != AccelDraw::Indexed) {
-        return 0;
-    }
     MICROPROFILE_SCOPE(OpenGL_Index);
     const auto& regs = system.GPU().Maxwell3D().regs;
     const std::size_t size = CalculateIndexBufferSize();
@@ -542,7 +543,8 @@ void RasterizerOpenGL::Clear() {
     }
 }
 
-void RasterizerOpenGL::DrawPrelude() {
+void RasterizerOpenGL::Draw(bool is_indexed, bool is_instanced) {
+    MICROPROFILE_SCOPE(OpenGL_Drawing);
     auto& gpu = system.GPU().Maxwell3D();
 
     SyncRasterizeEnable(state);
@@ -563,9 +565,6 @@ void RasterizerOpenGL::DrawPrelude() {
 
     buffer_cache.Acquire();
 
-    // Draw the vertex batch
-    const bool is_indexed = accelerate_draw == AccelDraw::Indexed;
-
     std::size_t buffer_size = CalculateVertexArraysSize();
 
     // Add space for index buffer
@@ -592,7 +591,11 @@ void RasterizerOpenGL::DrawPrelude() {
     // Upload vertex and index data.
     SetupVertexBuffer(vao);
     SetupVertexInstances(vao);
-    index_buffer_offset = SetupIndexBuffer();
+
+    GLintptr index_buffer_offset;
+    if (is_indexed) {
+        index_buffer_offset = SetupIndexBuffer();
+    }
 
     // Prepare packed bindings.
     bind_ubo_pushbuffer.Setup();
@@ -626,6 +629,7 @@ void RasterizerOpenGL::DrawPrelude() {
         // As all cached buffers are invalidated, we need to recheck their state.
         gpu.dirty.ResetVertexArrays();
     }
+    gpu.dirty.memory_general = false;
 
     shader_program_manager->ApplyTo(state);
     state.Apply();
@@ -633,106 +637,33 @@ void RasterizerOpenGL::DrawPrelude() {
     if (texture_cache.TextureBarrier()) {
         glTextureBarrier();
     }
+
+    const GLuint base_instance = static_cast<GLuint>(gpu.regs.vb_base_instance);
+    const GLsizei num_instances =
+        static_cast<GLsizei>(is_instanced ? gpu.mme_draw.instance_count : 1);
+    if (is_indexed) {
+        const GLenum index_format = MaxwellToGL::IndexFormat(gpu.regs.index_array.format);
+        const GLint base_vertex = static_cast<GLint>(gpu.regs.vb_element_base);
+        const GLsizei num_vertices = static_cast<GLsizei>(gpu.regs.index_array.count);
+        glDrawElementsInstancedBaseVertexBaseInstance(
+            primitive_mode, num_vertices, index_format,
+            reinterpret_cast<const void*>(index_buffer_offset), num_instances, base_vertex,
+            base_instance);
+    } else {
+        const GLint base_vertex = static_cast<GLint>(gpu.regs.vertex_buffer.first);
+        const GLsizei num_vertices = static_cast<GLsizei>(gpu.regs.vertex_buffer.count);
+        glDrawArraysInstancedBaseInstance(primitive_mode, base_vertex, num_vertices, num_instances,
+                                          base_instance);
+    }
 }
 
-struct DrawParams {
-    bool is_indexed{};
-    bool is_instanced{};
-    GLenum primitive_mode{};
-    GLint count{};
-    GLint base_vertex{};
-
-    // Indexed settings
-    GLenum index_format{};
-    GLintptr index_buffer_offset{};
-
-    // Instanced setting
-    GLint num_instances{};
-    GLint base_instance{};
-
-    void DispatchDraw() {
-        if (is_indexed) {
-            const auto index_buffer_ptr = reinterpret_cast<const void*>(index_buffer_offset);
-            if (is_instanced) {
-                glDrawElementsInstancedBaseVertexBaseInstance(primitive_mode, count, index_format,
-                                                              index_buffer_ptr, num_instances,
-                                                              base_vertex, base_instance);
-            } else {
-                glDrawElementsBaseVertex(primitive_mode, count, index_format, index_buffer_ptr,
-                                         base_vertex);
-            }
-        } else {
-            if (is_instanced) {
-                glDrawArraysInstancedBaseInstance(primitive_mode, base_vertex, count, num_instances,
-                                                  base_instance);
-            } else {
-                glDrawArrays(primitive_mode, base_vertex, count);
-            }
-        }
-    }
-};
-
 bool RasterizerOpenGL::DrawBatch(bool is_indexed) {
-    accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays;
-
-    MICROPROFILE_SCOPE(OpenGL_Drawing);
-
-    DrawPrelude();
-
-    auto& maxwell3d = system.GPU().Maxwell3D();
-    const auto& regs = maxwell3d.regs;
-    const auto current_instance = maxwell3d.state.current_instance;
-    DrawParams draw_call{};
-    draw_call.is_indexed = is_indexed;
-    draw_call.num_instances = static_cast<GLint>(1);
-    draw_call.base_instance = static_cast<GLint>(current_instance);
-    draw_call.is_instanced = current_instance > 0;
-    draw_call.primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology);
-    if (draw_call.is_indexed) {
-        draw_call.count = static_cast<GLint>(regs.index_array.count);
-        draw_call.base_vertex = static_cast<GLint>(regs.vb_element_base);
-        draw_call.index_format = MaxwellToGL::IndexFormat(regs.index_array.format);
-        draw_call.index_buffer_offset = index_buffer_offset;
-    } else {
-        draw_call.count = static_cast<GLint>(regs.vertex_buffer.count);
-        draw_call.base_vertex = static_cast<GLint>(regs.vertex_buffer.first);
-    }
-    draw_call.DispatchDraw();
-
-    maxwell3d.dirty.memory_general = false;
-    accelerate_draw = AccelDraw::Disabled;
+    Draw(is_indexed, false);
     return true;
 }
 
 bool RasterizerOpenGL::DrawMultiBatch(bool is_indexed) {
-    accelerate_draw = is_indexed ? AccelDraw::Indexed : AccelDraw::Arrays;
-
-    MICROPROFILE_SCOPE(OpenGL_Drawing);
-
-    DrawPrelude();
-
-    auto& maxwell3d = system.GPU().Maxwell3D();
-    const auto& regs = maxwell3d.regs;
-    const auto& draw_setup = maxwell3d.mme_draw;
-    DrawParams draw_call{};
-    draw_call.is_indexed = is_indexed;
-    draw_call.num_instances = static_cast<GLint>(draw_setup.instance_count);
-    draw_call.base_instance = static_cast<GLint>(regs.vb_base_instance);
-    draw_call.is_instanced = draw_setup.instance_count > 1;
-    draw_call.primitive_mode = MaxwellToGL::PrimitiveTopology(regs.draw.topology);
-    if (draw_call.is_indexed) {
-        draw_call.count = static_cast<GLint>(regs.index_array.count);
-        draw_call.base_vertex = static_cast<GLint>(regs.vb_element_base);
-        draw_call.index_format = MaxwellToGL::IndexFormat(regs.index_array.format);
-        draw_call.index_buffer_offset = index_buffer_offset;
-    } else {
-        draw_call.count = static_cast<GLint>(regs.vertex_buffer.count);
-        draw_call.base_vertex = static_cast<GLint>(regs.vertex_buffer.first);
-    }
-    draw_call.DispatchDraw();
-
-    maxwell3d.dirty.memory_general = false;
-    accelerate_draw = AccelDraw::Disabled;
+    Draw(is_indexed, true);
     return true;
 }
 
@@ -942,8 +873,15 @@ void RasterizerOpenGL::SetupDrawTextures(std::size_t stage_index, const Shader&
     u32 binding = device.GetBaseBindings(stage_index).sampler;
     for (const auto& entry : shader->GetShaderEntries().samplers) {
         const auto shader_type = static_cast<Tegra::Engines::ShaderType>(stage_index);
-        const auto texture = GetTextureInfo(maxwell3d, entry, shader_type);
-        SetupTexture(binding++, texture, entry);
+        if (!entry.IsIndexed()) {
+            const auto texture = GetTextureInfo(maxwell3d, entry, shader_type);
+            SetupTexture(binding++, texture, entry);
+        } else {
+            for (std::size_t i = 0; i < entry.Size(); ++i) {
+                const auto texture = GetTextureInfo(maxwell3d, entry, shader_type, i);
+                SetupTexture(binding++, texture, entry);
+            }
+        }
     }
 }
 
@@ -952,8 +890,17 @@ void RasterizerOpenGL::SetupComputeTextures(const Shader& kernel) {
     const auto& compute = system.GPU().KeplerCompute();
     u32 binding = 0;
     for (const auto& entry : kernel->GetShaderEntries().samplers) {
-        const auto texture = GetTextureInfo(compute, entry, Tegra::Engines::ShaderType::Compute);
-        SetupTexture(binding++, texture, entry);
+        if (!entry.IsIndexed()) {
+            const auto texture =
+                GetTextureInfo(compute, entry, Tegra::Engines::ShaderType::Compute);
+            SetupTexture(binding++, texture, entry);
+        } else {
+            for (std::size_t i = 0; i < entry.Size(); ++i) {
+                const auto texture =
+                    GetTextureInfo(compute, entry, Tegra::Engines::ShaderType::Compute, i);
+                SetupTexture(binding++, texture, entry);
+            }
+        }
     }
 }
 
@@ -1273,6 +1220,7 @@ void RasterizerOpenGL::SyncPointState() {
     // Limit the point size to 1 since nouveau sometimes sets a point size of 0 (and that's invalid
     // in OpenGL).
     state.point.program_control = regs.vp_point_size.enable != 0;
+    state.point.sprite = regs.point_sprite_enable != 0;
     state.point.size = std::max(1.0f, regs.point_size);
 }
 

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -103,7 +103,7 @@ private:
                            std::size_t size);
 
     /// Syncs all the state, shaders, render targets and textures setting before a draw call.
-    void DrawPrelude();
+    void Draw(bool is_indexed, bool is_instanced);
 
     /// Configures the current textures to use for the draw command.
     void SetupDrawTextures(std::size_t stage_index, const Shader& shader);
@@ -220,12 +220,7 @@ private:
 
     GLintptr SetupIndexBuffer();
 
-    GLintptr index_buffer_offset;
-
     void SetupShaders(GLenum primitive_mode);
-
-    enum class AccelDraw { Disabled, Arrays, Indexed };
-    AccelDraw accelerate_draw = AccelDraw::Disabled;
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -214,6 +214,7 @@ std::unique_ptr<ConstBufferLocker> MakeLocker(Core::System& system, ShaderType s
 }
 
 void FillLocker(ConstBufferLocker& locker, const ShaderDiskCacheUsage& usage) {
+    locker.SetBoundBuffer(usage.bound_buffer);
     for (const auto& key : usage.keys) {
         const auto [buffer, offset] = key.first;
         locker.InsertKey(buffer, offset, key.second);
@@ -418,7 +419,8 @@ bool CachedShader::EnsureValidLockerVariant() {
 
 ShaderDiskCacheUsage CachedShader::GetUsage(const ProgramVariant& variant,
                                             const ConstBufferLocker& locker) const {
-    return ShaderDiskCacheUsage{unique_identifier, variant, locker.GetKeys(),
+    return ShaderDiskCacheUsage{unique_identifier,         variant,
+                                locker.GetBoundBuffer(),   locker.GetKeys(),
                                 locker.GetBoundSamplers(), locker.GetBindlessSamplers()};
 }
 

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -391,6 +391,7 @@ public:
         DeclareVertex();
         DeclareGeometry();
         DeclareRegisters();
+        DeclareCustomVariables();
         DeclarePredicates();
         DeclareLocalMemory();
         DeclareInternalFlags();
@@ -503,6 +504,16 @@ private:
         }
     }
 
+    void DeclareCustomVariables() {
+        const u32 num_custom_variables = ir.GetNumCustomVariables();
+        for (u32 i = 0; i < num_custom_variables; ++i) {
+            code.AddLine("float {} = 0.0f;", GetCustomVariable(i));
+        }
+        if (num_custom_variables > 0) {
+            code.AddNewLine();
+        }
+    }
+
     void DeclarePredicates() {
         const auto& predicates = ir.GetPredicates();
         for (const auto pred : predicates) {
@@ -655,7 +666,8 @@ private:
         u32 binding = device.GetBaseBindings(stage).sampler;
         for (const auto& sampler : ir.GetSamplers()) {
             const std::string name = GetSampler(sampler);
-            const std::string description = fmt::format("layout (binding = {}) uniform", binding++);
+            const std::string description = fmt::format("layout (binding = {}) uniform", binding);
+            binding += sampler.IsIndexed() ? sampler.Size() : 1;
 
             std::string sampler_type = [&]() {
                 if (sampler.IsBuffer()) {
@@ -682,7 +694,11 @@ private:
                 sampler_type += "Shadow";
             }
 
-            code.AddLine("{} {} {};", description, sampler_type, name);
+            if (!sampler.IsIndexed()) {
+                code.AddLine("{} {} {};", description, sampler_type, name);
+            } else {
+                code.AddLine("{} {} {}[{}];", description, sampler_type, name, sampler.Size());
+            }
         }
         if (!ir.GetSamplers().empty()) {
             code.AddNewLine();
@@ -775,6 +791,11 @@ private:
             return {GetRegister(index), Type::Float};
         }
 
+        if (const auto cv = std::get_if<CustomVarNode>(&*node)) {
+            const u32 index = cv->GetIndex();
+            return {GetCustomVariable(index), Type::Float};
+        }
+
         if (const auto immediate = std::get_if<ImmediateNode>(&*node)) {
             const u32 value = immediate->GetValue();
             if (value < 10) {
@@ -1019,7 +1040,6 @@ private:
                 }
                 return {{"gl_ViewportIndex", Type::Int}};
             case 3:
-                UNIMPLEMENTED_MSG("Requires some state changes for gl_PointSize to work in shader");
                 return {{"gl_PointSize", Type::Float}};
             }
             return {};
@@ -1099,7 +1119,11 @@ private:
         } else if (!meta->ptp.empty()) {
             expr += "Offsets";
         }
-        expr += '(' + GetSampler(meta->sampler) + ", ";
+        if (!meta->sampler.IsIndexed()) {
+            expr += '(' + GetSampler(meta->sampler) + ", ";
+        } else {
+            expr += '(' + GetSampler(meta->sampler) + '[' + Visit(meta->index).AsUint() + "], ";
+        }
         expr += coord_constructors.at(count + (has_array ? 1 : 0) +
                                       (has_shadow && !separate_dc ? 1 : 0) - 1);
         expr += '(';
@@ -1311,6 +1335,8 @@ private:
             const std::string final_offset = fmt::format("({} - {}) >> 2", real, base);
             target = {fmt::format("{}[{}]", GetGlobalMemory(gmem->GetDescriptor()), final_offset),
                       Type::Uint};
+        } else if (const auto cv = std::get_if<CustomVarNode>(&*dest)) {
+            target = {GetCustomVariable(cv->GetIndex()), Type::Float};
         } else {
             UNREACHABLE_MSG("Assign called without a proper target");
         }
@@ -1858,10 +1884,7 @@ private:
 
     template <const std::string_view& opname, Type type>
     Expression Atomic(Operation operation) {
-        ASSERT(stage == ShaderType::Compute);
-        auto& smem = std::get<SmemNode>(*operation[0]);
-
-        return {fmt::format("atomic{}(smem[{} >> 2], {})", opname, Visit(smem.GetAddress()).AsInt(),
+        return {fmt::format("atomic{}({}, {})", opname, Visit(operation[0]).GetCode(),
                             Visit(operation[1]).As(type)),
                 type};
     }
@@ -2241,6 +2264,10 @@ private:
         return GetDeclarationWithSuffix(index, "gpr");
     }
 
+    std::string GetCustomVariable(u32 index) const {
+        return GetDeclarationWithSuffix(index, "custom_var");
+    }
+
     std::string GetPredicate(Tegra::Shader::Pred pred) const {
         return GetDeclarationWithSuffix(static_cast<u32>(pred), "pred");
     }

src/video_core/renderer_opengl/gl_shader_disk_cache.cpp

@@ -53,7 +53,7 @@ struct BindlessSamplerKey {
     Tegra::Engines::SamplerDescriptor sampler{};
 };
 
-constexpr u32 NativeVersion = 11;
+constexpr u32 NativeVersion = 12;
 
 // Making sure sizes doesn't change by accident
 static_assert(sizeof(ProgramVariant) == 20);
@@ -186,7 +186,8 @@ ShaderDiskCacheOpenGL::LoadTransferable() {
             u32 num_bound_samplers{};
             u32 num_bindless_samplers{};
             if (file.ReadArray(&usage.unique_identifier, 1) != 1 ||
-                file.ReadArray(&usage.variant, 1) != 1 || file.ReadArray(&num_keys, 1) != 1 ||
+                file.ReadArray(&usage.variant, 1) != 1 ||
+                file.ReadArray(&usage.bound_buffer, 1) != 1 || file.ReadArray(&num_keys, 1) != 1 ||
                 file.ReadArray(&num_bound_samplers, 1) != 1 ||
                 file.ReadArray(&num_bindless_samplers, 1) != 1) {
                 LOG_ERROR(Render_OpenGL, error_loading);
@@ -281,7 +282,9 @@ ShaderDiskCacheOpenGL::LoadPrecompiledFile(FileUtil::IOFile& file) {
         u32 num_bindless_samplers{};
         ShaderDiskCacheUsage usage;
         if (!LoadObjectFromPrecompiled(usage.unique_identifier) ||
-            !LoadObjectFromPrecompiled(usage.variant) || !LoadObjectFromPrecompiled(num_keys) ||
+            !LoadObjectFromPrecompiled(usage.variant) ||
+            !LoadObjectFromPrecompiled(usage.bound_buffer) ||
+            !LoadObjectFromPrecompiled(num_keys) ||
             !LoadObjectFromPrecompiled(num_bound_samplers) ||
             !LoadObjectFromPrecompiled(num_bindless_samplers)) {
             return {};
@@ -393,6 +396,7 @@ void ShaderDiskCacheOpenGL::SaveUsage(const ShaderDiskCacheUsage& usage) {
 
     if (file.WriteObject(TransferableEntryKind::Usage) != 1 ||
         file.WriteObject(usage.unique_identifier) != 1 || file.WriteObject(usage.variant) != 1 ||
+        file.WriteObject(usage.bound_buffer) != 1 ||
         file.WriteObject(static_cast<u32>(usage.keys.size())) != 1 ||
         file.WriteObject(static_cast<u32>(usage.bound_samplers.size())) != 1 ||
         file.WriteObject(static_cast<u32>(usage.bindless_samplers.size())) != 1) {
@@ -447,7 +451,7 @@ void ShaderDiskCacheOpenGL::SaveDump(const ShaderDiskCacheUsage& usage, GLuint p
     };
 
     if (!SaveObjectToPrecompiled(usage.unique_identifier) ||
-        !SaveObjectToPrecompiled(usage.variant) ||
+        !SaveObjectToPrecompiled(usage.variant) || !SaveObjectToPrecompiled(usage.bound_buffer) ||
         !SaveObjectToPrecompiled(static_cast<u32>(usage.keys.size())) ||
         !SaveObjectToPrecompiled(static_cast<u32>(usage.bound_samplers.size())) ||
         !SaveObjectToPrecompiled(static_cast<u32>(usage.bindless_samplers.size()))) {

src/video_core/renderer_opengl/gl_shader_disk_cache.h

@@ -79,6 +79,7 @@ static_assert(std::is_trivially_copyable_v<ProgramVariant>);
 struct ShaderDiskCacheUsage {
     u64 unique_identifier{};
     ProgramVariant variant;
+    u32 bound_buffer{};
     VideoCommon::Shader::KeyMap keys;
     VideoCommon::Shader::BoundSamplerMap bound_samplers;
     VideoCommon::Shader::BindlessSamplerMap bindless_samplers;

src/video_core/renderer_opengl/gl_state.cpp

@@ -128,6 +128,7 @@ void OpenGLState::ApplyClipDistances() {
 
 void OpenGLState::ApplyPointSize() {
     Enable(GL_PROGRAM_POINT_SIZE, cur_state.point.program_control, point.program_control);
+    Enable(GL_POINT_SPRITE, cur_state.point.sprite, point.sprite);
     if (UpdateValue(cur_state.point.size, point.size)) {
         glPointSize(point.size);
     }

src/video_core/renderer_opengl/gl_state.h

@@ -132,6 +132,7 @@ public:
 
     struct {
         bool program_control = false; // GL_PROGRAM_POINT_SIZE
+        bool sprite = false;          // GL_POINT_SPRITE
         GLfloat size = 1.0f;          // GL_POINT_SIZE
     } point;
 

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -176,6 +176,19 @@ GLint GetSwizzleSource(SwizzleSource source) {
     return GL_NONE;
 }
 
+GLenum GetComponent(PixelFormat format, bool is_first) {
+    switch (format) {
+    case PixelFormat::Z24S8:
+    case PixelFormat::Z32FS8:
+        return is_first ? GL_DEPTH_COMPONENT : GL_STENCIL_INDEX;
+    case PixelFormat::S8Z24:
+        return is_first ? GL_STENCIL_INDEX : GL_DEPTH_COMPONENT;
+    default:
+        UNREACHABLE();
+        return GL_DEPTH_COMPONENT;
+    }
+}
+
 void ApplyTextureDefaults(const SurfaceParams& params, GLuint texture) {
     if (params.IsBuffer()) {
         return;
@@ -184,7 +197,7 @@ void ApplyTextureDefaults(const SurfaceParams& params, GLuint texture) {
     glTextureParameteri(texture, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     glTextureParameteri(texture, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
     glTextureParameteri(texture, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
-    glTextureParameteri(texture, GL_TEXTURE_MAX_LEVEL, params.num_levels - 1);
+    glTextureParameteri(texture, GL_TEXTURE_MAX_LEVEL, static_cast<GLint>(params.num_levels - 1));
     if (params.num_levels == 1) {
         glTextureParameterf(texture, GL_TEXTURE_LOD_BIAS, 1000.0f);
     }
@@ -416,11 +429,21 @@ void CachedSurfaceView::ApplySwizzle(SwizzleSource x_source, SwizzleSource y_sou
     if (new_swizzle == swizzle)
         return;
     swizzle = new_swizzle;
-    const std::array<GLint, 4> gl_swizzle = {GetSwizzleSource(x_source), GetSwizzleSource(y_source),
-                                             GetSwizzleSource(z_source),
-                                             GetSwizzleSource(w_source)};
+    const std::array gl_swizzle = {GetSwizzleSource(x_source), GetSwizzleSource(y_source),
+                                   GetSwizzleSource(z_source), GetSwizzleSource(w_source)};
     const GLuint handle = GetTexture();
-    glTextureParameteriv(handle, GL_TEXTURE_SWIZZLE_RGBA, gl_swizzle.data());
+    const PixelFormat format = surface.GetSurfaceParams().pixel_format;
+    switch (format) {
+    case PixelFormat::Z24S8:
+    case PixelFormat::Z32FS8:
+    case PixelFormat::S8Z24:
+        glTextureParameteri(handle, GL_DEPTH_STENCIL_TEXTURE_MODE,
+                            GetComponent(format, x_source == SwizzleSource::R));
+        break;
+    default:
+        glTextureParameteriv(handle, GL_TEXTURE_SWIZZLE_RGBA, gl_swizzle.data());
+        break;
+    }
 }
 
 OGLTextureView CachedSurfaceView::CreateTextureView() const {
@@ -529,8 +552,11 @@ void TextureCacheOpenGL::ImageBlit(View& src_view, View& dst_view,
     const Common::Rectangle<u32>& dst_rect = copy_config.dst_rect;
     const bool is_linear = copy_config.filter == Tegra::Engines::Fermi2D::Filter::Linear;
 
-    glBlitFramebuffer(src_rect.left, src_rect.top, src_rect.right, src_rect.bottom, dst_rect.left,
-                      dst_rect.top, dst_rect.right, dst_rect.bottom, buffers,
+    glBlitFramebuffer(static_cast<GLint>(src_rect.left), static_cast<GLint>(src_rect.top),
+                      static_cast<GLint>(src_rect.right), static_cast<GLint>(src_rect.bottom),
+                      static_cast<GLint>(dst_rect.left), static_cast<GLint>(dst_rect.top),
+                      static_cast<GLint>(dst_rect.right), static_cast<GLint>(dst_rect.bottom),
+                      buffers,
                       is_linear && (buffers == GL_COLOR_BUFFER_BIT) ? GL_LINEAR : GL_NEAREST);
 }
 

src/video_core/renderer_opengl/maxwell_to_gl.h

@@ -47,8 +47,7 @@ inline GLenum VertexType(Maxwell::VertexAttribute attrib) {
         case Maxwell::VertexAttribute::Size::Size_10_10_10_2:
             return GL_UNSIGNED_INT_2_10_10_10_REV;
         default:
-            LOG_CRITICAL(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
-            UNREACHABLE();
+            LOG_ERROR(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
             return {};
         }
     case Maxwell::VertexAttribute::Type::SignedInt:
@@ -72,8 +71,7 @@ inline GLenum VertexType(Maxwell::VertexAttribute attrib) {
         case Maxwell::VertexAttribute::Size::Size_10_10_10_2:
             return GL_INT_2_10_10_10_REV;
         default:
-            LOG_CRITICAL(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
-            UNREACHABLE();
+            LOG_ERROR(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
             return {};
         }
     case Maxwell::VertexAttribute::Type::Float:
@@ -89,13 +87,19 @@ inline GLenum VertexType(Maxwell::VertexAttribute attrib) {
         case Maxwell::VertexAttribute::Size::Size_32_32_32_32:
             return GL_FLOAT;
         default:
-            LOG_CRITICAL(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
-            UNREACHABLE();
+            LOG_ERROR(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
+            return {};
+        }
+    case Maxwell::VertexAttribute::Type::UnsignedScaled:
+        switch (attrib.size) {
+        case Maxwell::VertexAttribute::Size::Size_8_8:
+            return GL_UNSIGNED_BYTE;
+        default:
+            LOG_ERROR(Render_OpenGL, "Unimplemented vertex size={}", attrib.SizeString());
             return {};
         }
     default:
-        LOG_CRITICAL(Render_OpenGL, "Unimplemented vertex type={}", attrib.TypeString());
-        UNREACHABLE();
+        LOG_ERROR(Render_OpenGL, "Unimplemented vertex type={}", attrib.TypeString());
         return {};
     }
 }

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -0,0 +1,265 @@
+// Copyright 2018 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <memory>
+#include <optional>
+#include <vector>
+
+#include <fmt/format.h>
+
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "common/telemetry.h"
+#include "core/core.h"
+#include "core/core_timing.h"
+#include "core/frontend/emu_window.h"
+#include "core/memory.h"
+#include "core/perf_stats.h"
+#include "core/settings.h"
+#include "core/telemetry_session.h"
+#include "video_core/gpu.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/renderer_vulkan.h"
+#include "video_core/renderer_vulkan/vk_blit_screen.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_memory_manager.h"
+#include "video_core/renderer_vulkan/vk_rasterizer.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_swapchain.h"
+
+namespace Vulkan {
+
+namespace {
+
+VkBool32 DebugCallback(VkDebugUtilsMessageSeverityFlagBitsEXT severity_,
+                       VkDebugUtilsMessageTypeFlagsEXT type,
+                       const VkDebugUtilsMessengerCallbackDataEXT* data,
+                       [[maybe_unused]] void* user_data) {
+    const vk::DebugUtilsMessageSeverityFlagBitsEXT severity{severity_};
+    const char* message{data->pMessage};
+
+    if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eError) {
+        LOG_CRITICAL(Render_Vulkan, "{}", message);
+    } else if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eWarning) {
+        LOG_WARNING(Render_Vulkan, "{}", message);
+    } else if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eInfo) {
+        LOG_INFO(Render_Vulkan, "{}", message);
+    } else if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eVerbose) {
+        LOG_DEBUG(Render_Vulkan, "{}", message);
+    }
+    return VK_FALSE;
+}
+
+std::string GetReadableVersion(u32 version) {
+    return fmt::format("{}.{}.{}", VK_VERSION_MAJOR(version), VK_VERSION_MINOR(version),
+                       VK_VERSION_PATCH(version));
+}
+
+std::string GetDriverVersion(const VKDevice& device) {
+    // Extracted from
+    // https://github.com/SaschaWillems/vulkan.gpuinfo.org/blob/5dddea46ea1120b0df14eef8f15ff8e318e35462/functions.php#L308-L314
+    const u32 version = device.GetDriverVersion();
+
+    if (device.GetDriverID() == vk::DriverIdKHR::eNvidiaProprietary) {
+        const u32 major = (version >> 22) & 0x3ff;
+        const u32 minor = (version >> 14) & 0x0ff;
+        const u32 secondary = (version >> 6) & 0x0ff;
+        const u32 tertiary = version & 0x003f;
+        return fmt::format("{}.{}.{}.{}", major, minor, secondary, tertiary);
+    }
+    if (device.GetDriverID() == vk::DriverIdKHR::eIntelProprietaryWindows) {
+        const u32 major = version >> 14;
+        const u32 minor = version & 0x3fff;
+        return fmt::format("{}.{}", major, minor);
+    }
+
+    return GetReadableVersion(version);
+}
+
+std::string BuildCommaSeparatedExtensions(std::vector<std::string> available_extensions) {
+    std::sort(std::begin(available_extensions), std::end(available_extensions));
+
+    static constexpr std::size_t AverageExtensionSize = 64;
+    std::string separated_extensions;
+    separated_extensions.reserve(available_extensions.size() * AverageExtensionSize);
+
+    const auto end = std::end(available_extensions);
+    for (auto extension = std::begin(available_extensions); extension != end; ++extension) {
+        if (const bool is_last = extension + 1 == end; is_last) {
+            separated_extensions += *extension;
+        } else {
+            separated_extensions += fmt::format("{},", *extension);
+        }
+    }
+    return separated_extensions;
+}
+
+} // Anonymous namespace
+
+RendererVulkan::RendererVulkan(Core::Frontend::EmuWindow& window, Core::System& system)
+    : RendererBase(window), system{system} {}
+
+RendererVulkan::~RendererVulkan() {
+    ShutDown();
+}
+
+void RendererVulkan::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
+    const auto& layout = render_window.GetFramebufferLayout();
+    if (framebuffer && layout.width > 0 && layout.height > 0 && render_window.IsShown()) {
+        const VAddr framebuffer_addr = framebuffer->address + framebuffer->offset;
+        const bool use_accelerated =
+            rasterizer->AccelerateDisplay(*framebuffer, framebuffer_addr, framebuffer->stride);
+        const bool is_srgb = use_accelerated && screen_info.is_srgb;
+        if (swapchain->HasFramebufferChanged(layout) || swapchain->GetSrgbState() != is_srgb) {
+            swapchain->Create(layout.width, layout.height, is_srgb);
+            blit_screen->Recreate();
+        }
+
+        scheduler->WaitWorker();
+
+        swapchain->AcquireNextImage();
+        const auto [fence, render_semaphore] = blit_screen->Draw(*framebuffer, use_accelerated);
+
+        scheduler->Flush(false, render_semaphore);
+
+        if (swapchain->Present(render_semaphore, fence)) {
+            blit_screen->Recreate();
+        }
+
+        render_window.SwapBuffers();
+        rasterizer->TickFrame();
+    }
+
+    render_window.PollEvents();
+}
+
+bool RendererVulkan::Init() {
+    PFN_vkGetInstanceProcAddr vkGetInstanceProcAddr{};
+    render_window.RetrieveVulkanHandlers(&vkGetInstanceProcAddr, &instance, &surface);
+    const vk::DispatchLoaderDynamic dldi(instance, vkGetInstanceProcAddr);
+
+    std::optional<vk::DebugUtilsMessengerEXT> callback;
+    if (Settings::values.renderer_debug && dldi.vkCreateDebugUtilsMessengerEXT) {
+        callback = CreateDebugCallback(dldi);
+        if (!callback) {
+            return false;
+        }
+    }
+
+    if (!PickDevices(dldi)) {
+        if (callback) {
+            instance.destroy(*callback, nullptr, dldi);
+        }
+        return false;
+    }
+    debug_callback = UniqueDebugUtilsMessengerEXT(
+        *callback, vk::ObjectDestroy<vk::Instance, vk::DispatchLoaderDynamic>(
+                       instance, nullptr, device->GetDispatchLoader()));
+
+    Report();
+
+    memory_manager = std::make_unique<VKMemoryManager>(*device);
+
+    resource_manager = std::make_unique<VKResourceManager>(*device);
+
+    const auto& framebuffer = render_window.GetFramebufferLayout();
+    swapchain = std::make_unique<VKSwapchain>(surface, *device);
+    swapchain->Create(framebuffer.width, framebuffer.height, false);
+
+    scheduler = std::make_unique<VKScheduler>(*device, *resource_manager);
+
+    rasterizer = std::make_unique<RasterizerVulkan>(system, render_window, screen_info, *device,
+                                                    *resource_manager, *memory_manager, *scheduler);
+
+    blit_screen = std::make_unique<VKBlitScreen>(system, render_window, *rasterizer, *device,
+                                                 *resource_manager, *memory_manager, *swapchain,
+                                                 *scheduler, screen_info);
+
+    return true;
+}
+
+void RendererVulkan::ShutDown() {
+    if (!device) {
+        return;
+    }
+    const auto dev = device->GetLogical();
+    const auto& dld = device->GetDispatchLoader();
+    if (dev && dld.vkDeviceWaitIdle) {
+        dev.waitIdle(dld);
+    }
+
+    rasterizer.reset();
+    blit_screen.reset();
+    scheduler.reset();
+    swapchain.reset();
+    memory_manager.reset();
+    resource_manager.reset();
+    device.reset();
+}
+
+std::optional<vk::DebugUtilsMessengerEXT> RendererVulkan::CreateDebugCallback(
+    const vk::DispatchLoaderDynamic& dldi) {
+    const vk::DebugUtilsMessengerCreateInfoEXT callback_ci(
+        {},
+        vk::DebugUtilsMessageSeverityFlagBitsEXT::eError |
+            vk::DebugUtilsMessageSeverityFlagBitsEXT::eWarning |
+            vk::DebugUtilsMessageSeverityFlagBitsEXT::eInfo |
+            vk::DebugUtilsMessageSeverityFlagBitsEXT::eVerbose,
+        vk::DebugUtilsMessageTypeFlagBitsEXT::eGeneral |
+            vk::DebugUtilsMessageTypeFlagBitsEXT::eValidation |
+            vk::DebugUtilsMessageTypeFlagBitsEXT::ePerformance,
+        &DebugCallback, nullptr);
+    vk::DebugUtilsMessengerEXT callback;
+    if (instance.createDebugUtilsMessengerEXT(&callback_ci, nullptr, &callback, dldi) !=
+        vk::Result::eSuccess) {
+        LOG_ERROR(Render_Vulkan, "Failed to create debug callback");
+        return {};
+    }
+    return callback;
+}
+
+bool RendererVulkan::PickDevices(const vk::DispatchLoaderDynamic& dldi) {
+    const auto devices = instance.enumeratePhysicalDevices(dldi);
+
+    // TODO(Rodrigo): Choose device from config file
+    const s32 device_index = Settings::values.vulkan_device;
+    if (device_index < 0 || device_index >= static_cast<s32>(devices.size())) {
+        LOG_ERROR(Render_Vulkan, "Invalid device index {}!", device_index);
+        return false;
+    }
+    const vk::PhysicalDevice physical_device = devices[device_index];
+
+    if (!VKDevice::IsSuitable(dldi, physical_device, surface)) {
+        return false;
+    }
+
+    device = std::make_unique<VKDevice>(dldi, physical_device, surface);
+    return device->Create(dldi, instance);
+}
+
+void RendererVulkan::Report() const {
+    const std::string vendor_name{device->GetVendorName()};
+    const std::string model_name{device->GetModelName()};
+    const std::string driver_version = GetDriverVersion(*device);
+    const std::string driver_name = fmt::format("{} {}", vendor_name, driver_version);
+
+    const std::string api_version = GetReadableVersion(device->GetApiVersion());
+
+    const std::string extensions = BuildCommaSeparatedExtensions(device->GetAvailableExtensions());
+
+    LOG_INFO(Render_Vulkan, "Driver: {}", driver_name);
+    LOG_INFO(Render_Vulkan, "Device: {}", model_name);
+    LOG_INFO(Render_Vulkan, "Vulkan: {}", api_version);
+
+    auto& telemetry_session = system.TelemetrySession();
+    constexpr auto field = Telemetry::FieldType::UserSystem;
+    telemetry_session.AddField(field, "GPU_Vendor", vendor_name);
+    telemetry_session.AddField(field, "GPU_Model", model_name);
+    telemetry_session.AddField(field, "GPU_Vulkan_Driver", driver_name);
+    telemetry_session.AddField(field, "GPU_Vulkan_Version", api_version);
+    telemetry_session.AddField(field, "GPU_Vulkan_Extensions", extensions);
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_device.cpp

@@ -400,8 +400,10 @@ std::vector<const char*> VKDevice::LoadExtensions(const vk::DispatchLoaderDynami
              VK_EXT_SHADER_VIEWPORT_INDEX_LAYER_EXTENSION_NAME, true);
         Test(extension, ext_subgroup_size_control, VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME,
              false);
-        Test(extension, nv_device_diagnostic_checkpoints,
-             VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME, true);
+        if (Settings::values.renderer_debug) {
+            Test(extension, nv_device_diagnostic_checkpoints,
+                 VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME, true);
+        }
     }
 
     if (khr_shader_float16_int8) {

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -571,7 +571,7 @@ RasterizerVulkan::Texceptions RasterizerVulkan::UpdateAttachments() {
             color_attachments[rt] = texture_cache.GetColorBufferSurface(rt, true);
         }
         if (color_attachments[rt] && WalkAttachmentOverlaps(*color_attachments[rt])) {
-            texceptions.set(rt);
+            texceptions[rt] = true;
         }
     }
 
@@ -579,7 +579,7 @@ RasterizerVulkan::Texceptions RasterizerVulkan::UpdateAttachments() {
         zeta_attachment = texture_cache.GetDepthBufferSurface(true);
     }
     if (zeta_attachment && WalkAttachmentOverlaps(*zeta_attachment)) {
-        texceptions.set(ZETA_TEXCEPTION_INDEX);
+        texceptions[ZETA_TEXCEPTION_INDEX] = true;
     }
 
     texture_cache.GuardRenderTargets(false);
@@ -1122,11 +1122,12 @@ RenderPassParams RasterizerVulkan::GetRenderPassParams(Texceptions texceptions)
 
     for (std::size_t rt = 0; rt < static_cast<std::size_t>(regs.rt_control.count); ++rt) {
         const auto& rendertarget = regs.rt[rt];
-        if (rendertarget.Address() == 0 || rendertarget.format == Tegra::RenderTargetFormat::NONE)
+        if (rendertarget.Address() == 0 || rendertarget.format == Tegra::RenderTargetFormat::NONE) {
             continue;
+        }
         renderpass_params.color_attachments.push_back(RenderPassParams::ColorAttachment{
             static_cast<u32>(rt), PixelFormatFromRenderTargetFormat(rendertarget.format),
-            texceptions.test(rt)});
+            texceptions[rt]});
     }
 
     renderpass_params.has_zeta = regs.zeta_enable;

src/video_core/renderer_vulkan/vk_shader_decompiler.cpp

@@ -353,6 +353,7 @@ private:
         DeclareFragment();
         DeclareCompute();
         DeclareRegisters();
+        DeclareCustomVariables();
         DeclarePredicates();
         DeclareLocalMemory();
         DeclareSharedMemory();
@@ -586,6 +587,15 @@ private:
         }
     }
 
+    void DeclareCustomVariables() {
+        const u32 num_custom_variables = ir.GetNumCustomVariables();
+        for (u32 i = 0; i < num_custom_variables; ++i) {
+            const Id id = OpVariable(t_prv_float, spv::StorageClass::Private, v_float_zero);
+            Name(id, fmt::format("custom_var_{}", i));
+            custom_variables.emplace(i, AddGlobalVariable(id));
+        }
+    }
+
     void DeclarePredicates() {
         for (const auto pred : ir.GetPredicates()) {
             const Id id = OpVariable(t_prv_bool, spv::StorageClass::Private, v_false);
@@ -982,6 +992,11 @@ private:
             return {OpLoad(t_float, registers.at(index)), Type::Float};
         }
 
+        if (const auto cv = std::get_if<CustomVarNode>(&*node)) {
+            const u32 index = cv->GetIndex();
+            return {OpLoad(t_float, custom_variables.at(index)), Type::Float};
+        }
+
         if (const auto immediate = std::get_if<ImmediateNode>(&*node)) {
             return {Constant(t_uint, immediate->GetValue()), Type::Uint};
         }
@@ -1123,15 +1138,7 @@ private:
         }
 
         if (const auto gmem = std::get_if<GmemNode>(&*node)) {
-            const Id gmem_buffer = global_buffers.at(gmem->GetDescriptor());
-            const Id real = AsUint(Visit(gmem->GetRealAddress()));
-            const Id base = AsUint(Visit(gmem->GetBaseAddress()));
-
-            Id offset = OpISub(t_uint, real, base);
-            offset = OpUDiv(t_uint, offset, Constant(t_uint, 4U));
-            return {OpLoad(t_float,
-                           OpAccessChain(t_gmem_float, gmem_buffer, Constant(t_uint, 0U), offset)),
-                    Type::Float};
+            return {OpLoad(t_uint, GetGlobalMemoryPointer(*gmem)), Type::Uint};
         }
 
         if (const auto lmem = std::get_if<LmemNode>(&*node)) {
@@ -1142,10 +1149,7 @@ private:
         }
 
         if (const auto smem = std::get_if<SmemNode>(&*node)) {
-            Id address = AsUint(Visit(smem->GetAddress()));
-            address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
-            const Id pointer = OpAccessChain(t_smem_uint, shared_memory, address);
-            return {OpLoad(t_uint, pointer), Type::Uint};
+            return {OpLoad(t_uint, GetSharedMemoryPointer(*smem)), Type::Uint};
         }
 
         if (const auto internal_flag = std::get_if<InternalFlagNode>(&*node)) {
@@ -1339,20 +1343,13 @@ private:
             target = {OpAccessChain(t_prv_float, local_memory, address), Type::Float};
 
         } else if (const auto smem = std::get_if<SmemNode>(&*dest)) {
-            ASSERT(stage == ShaderType::Compute);
-            Id address = AsUint(Visit(smem->GetAddress()));
-            address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
-            target = {OpAccessChain(t_smem_uint, shared_memory, address), Type::Uint};
+            target = {GetSharedMemoryPointer(*smem), Type::Uint};
 
         } else if (const auto gmem = std::get_if<GmemNode>(&*dest)) {
-            const Id real = AsUint(Visit(gmem->GetRealAddress()));
-            const Id base = AsUint(Visit(gmem->GetBaseAddress()));
-            const Id diff = OpISub(t_uint, real, base);
-            const Id offset = OpShiftRightLogical(t_uint, diff, Constant(t_uint, 2));
+            target = {GetGlobalMemoryPointer(*gmem), Type::Uint};
 
-            const Id gmem_buffer = global_buffers.at(gmem->GetDescriptor());
-            target = {OpAccessChain(t_gmem_float, gmem_buffer, Constant(t_uint, 0), offset),
-                      Type::Float};
+        } else if (const auto cv = std::get_if<CustomVarNode>(&*dest)) {
+            target = {custom_variables.at(cv->GetIndex()), Type::Float};
 
         } else {
             UNIMPLEMENTED();
@@ -1804,11 +1801,16 @@ private:
         return {};
     }
 
-    Expression UAtomicAdd(Operation operation) {
-        const auto& smem = std::get<SmemNode>(*operation[0]);
-        Id address = AsUint(Visit(smem.GetAddress()));
-        address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
-        const Id pointer = OpAccessChain(t_smem_uint, shared_memory, address);
+    Expression AtomicAdd(Operation operation) {
+        Id pointer;
+        if (const auto smem = std::get_if<SmemNode>(&*operation[0])) {
+            pointer = GetSharedMemoryPointer(*smem);
+        } else if (const auto gmem = std::get_if<GmemNode>(&*operation[0])) {
+            pointer = GetGlobalMemoryPointer(*gmem);
+        } else {
+            UNREACHABLE();
+            return {Constant(t_uint, 0), Type::Uint};
+        }
 
         const Id scope = Constant(t_uint, static_cast<u32>(spv::Scope::Device));
         const Id semantics = Constant(t_uint, 0U);
@@ -2243,6 +2245,22 @@ private:
         return {};
     }
 
+    Id GetGlobalMemoryPointer(const GmemNode& gmem) {
+        const Id real = AsUint(Visit(gmem.GetRealAddress()));
+        const Id base = AsUint(Visit(gmem.GetBaseAddress()));
+        const Id diff = OpISub(t_uint, real, base);
+        const Id offset = OpShiftRightLogical(t_uint, diff, Constant(t_uint, 2));
+        const Id buffer = global_buffers.at(gmem.GetDescriptor());
+        return OpAccessChain(t_gmem_uint, buffer, Constant(t_uint, 0), offset);
+    }
+
+    Id GetSharedMemoryPointer(const SmemNode& smem) {
+        ASSERT(stage == ShaderType::Compute);
+        Id address = AsUint(Visit(smem.GetAddress()));
+        address = OpShiftRightLogical(t_uint, address, Constant(t_uint, 2U));
+        return OpAccessChain(t_smem_uint, shared_memory, address);
+    }
+
     static constexpr std::array operation_decompilers = {
         &SPIRVDecompiler::Assign,
 
@@ -2389,7 +2407,7 @@ private:
         &SPIRVDecompiler::AtomicImageXor,
         &SPIRVDecompiler::AtomicImageExchange,
 
-        &SPIRVDecompiler::UAtomicAdd,
+        &SPIRVDecompiler::AtomicAdd,
 
         &SPIRVDecompiler::Branch,
         &SPIRVDecompiler::BranchIndirect,
@@ -2485,9 +2503,9 @@ private:
 
     Id t_smem_uint{};
 
-    const Id t_gmem_float = TypePointer(spv::StorageClass::StorageBuffer, t_float);
+    const Id t_gmem_uint = TypePointer(spv::StorageClass::StorageBuffer, t_uint);
     const Id t_gmem_array =
-        Name(Decorate(TypeRuntimeArray(t_float), spv::Decoration::ArrayStride, 4U), "GmemArray");
+        Name(Decorate(TypeRuntimeArray(t_uint), spv::Decoration::ArrayStride, 4U), "GmemArray");
     const Id t_gmem_struct = MemberDecorate(
         Decorate(TypeStruct(t_gmem_array), spv::Decoration::Block), 0, spv::Decoration::Offset, 0);
     const Id t_gmem_ssbo = TypePointer(spv::StorageClass::StorageBuffer, t_gmem_struct);
@@ -2508,6 +2526,7 @@ private:
     Id out_vertex{};
     Id in_vertex{};
     std::map<u32, Id> registers;
+    std::map<u32, Id> custom_variables;
     std::map<Tegra::Shader::Pred, Id> predicates;
     std::map<u32, Id> flow_variables;
     Id local_memory{};

src/video_core/shader/ast.h

@@ -65,8 +65,8 @@ public:
     void DetachSegment(ASTNode start, ASTNode end);
     void Remove(ASTNode node);
 
-    ASTNode first{};
-    ASTNode last{};
+    ASTNode first;
+    ASTNode last;
 };
 
 class ASTProgram {
@@ -299,9 +299,9 @@ private:
     friend class ASTZipper;
 
     ASTData data;
-    ASTNode parent{};
-    ASTNode next{};
-    ASTNode previous{};
+    ASTNode parent;
+    ASTNode next;
+    ASTNode previous;
     ASTZipper* manager{};
 };
 

src/video_core/shader/const_buffer_locker.cpp

@@ -66,6 +66,18 @@ std::optional<Tegra::Engines::SamplerDescriptor> ConstBufferLocker::ObtainBindle
     return value;
 }
 
+std::optional<u32> ConstBufferLocker::ObtainBoundBuffer() {
+    if (bound_buffer_saved) {
+        return bound_buffer;
+    }
+    if (!engine) {
+        return std::nullopt;
+    }
+    bound_buffer_saved = true;
+    bound_buffer = engine->GetBoundBuffer();
+    return bound_buffer;
+}
+
 void ConstBufferLocker::InsertKey(u32 buffer, u32 offset, u32 value) {
     keys.insert_or_assign({buffer, offset}, value);
 }
@@ -78,6 +90,11 @@ void ConstBufferLocker::InsertBindlessSampler(u32 buffer, u32 offset, SamplerDes
     bindless_samplers.insert_or_assign({buffer, offset}, sampler);
 }
 
+void ConstBufferLocker::SetBoundBuffer(u32 buffer) {
+    bound_buffer_saved = true;
+    bound_buffer = buffer;
+}
+
 bool ConstBufferLocker::IsConsistent() const {
     if (!engine) {
         return false;

src/video_core/shader/const_buffer_locker.h

@@ -10,6 +10,7 @@
 #include "common/hash.h"
 #include "video_core/engines/const_buffer_engine_interface.h"
 #include "video_core/engines/shader_type.h"
+#include "video_core/guest_driver.h"
 
 namespace VideoCommon::Shader {
 
@@ -40,6 +41,8 @@ public:
 
     std::optional<Tegra::Engines::SamplerDescriptor> ObtainBindlessSampler(u32 buffer, u32 offset);
 
+    std::optional<u32> ObtainBoundBuffer();
+
     /// Inserts a key.
     void InsertKey(u32 buffer, u32 offset, u32 value);
 
@@ -49,6 +52,9 @@ public:
     /// Inserts a bindless sampler key.
     void InsertBindlessSampler(u32 buffer, u32 offset, Tegra::Engines::SamplerDescriptor sampler);
 
+    /// Set the bound buffer for this locker.
+    void SetBoundBuffer(u32 buffer);
+
     /// Checks keys and samplers against engine's current const buffers. Returns true if they are
     /// the same value, false otherwise;
     bool IsConsistent() const;
@@ -71,12 +77,27 @@ public:
         return bindless_samplers;
     }
 
+    /// Gets bound buffer used on this shader
+    u32 GetBoundBuffer() const {
+        return bound_buffer;
+    }
+
+    /// Obtains access to the guest driver's profile.
+    VideoCore::GuestDriverProfile* AccessGuestDriverProfile() const {
+        if (engine) {
+            return &engine->AccessGuestDriverProfile();
+        }
+        return nullptr;
+    }
+
 private:
     const Tegra::Engines::ShaderType stage;
     Tegra::Engines::ConstBufferEngineInterface* engine = nullptr;
     KeyMap keys;
     BoundSamplerMap bound_samplers;
     BindlessSamplerMap bindless_samplers;
+    bool bound_buffer_saved{};
+    u32 bound_buffer{};
 };
 
 } // namespace VideoCommon::Shader

src/video_core/shader/decode.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include <cstring>
+#include <limits>
 #include <set>
 
 #include <fmt/format.h>
@@ -33,6 +34,52 @@ constexpr bool IsSchedInstruction(u32 offset, u32 main_offset) {
     return (absolute_offset % SchedPeriod) == 0;
 }
 
+void DeduceTextureHandlerSize(VideoCore::GuestDriverProfile* gpu_driver,
+                              const std::list<Sampler>& used_samplers) {
+    if (gpu_driver == nullptr) {
+        LOG_CRITICAL(HW_GPU, "GPU driver profile has not been created yet");
+        return;
+    }
+    if (gpu_driver->TextureHandlerSizeKnown() || used_samplers.size() <= 1) {
+        return;
+    }
+    u32 count{};
+    std::vector<u32> bound_offsets;
+    for (const auto& sampler : used_samplers) {
+        if (sampler.IsBindless()) {
+            continue;
+        }
+        ++count;
+        bound_offsets.emplace_back(sampler.GetOffset());
+    }
+    if (count > 1) {
+        gpu_driver->DeduceTextureHandlerSize(std::move(bound_offsets));
+    }
+}
+
+std::optional<u32> TryDeduceSamplerSize(const Sampler& sampler_to_deduce,
+                                        VideoCore::GuestDriverProfile* gpu_driver,
+                                        const std::list<Sampler>& used_samplers) {
+    if (gpu_driver == nullptr) {
+        LOG_CRITICAL(HW_GPU, "GPU Driver profile has not been created yet");
+        return std::nullopt;
+    }
+    const u32 base_offset = sampler_to_deduce.GetOffset();
+    u32 max_offset{std::numeric_limits<u32>::max()};
+    for (const auto& sampler : used_samplers) {
+        if (sampler.IsBindless()) {
+            continue;
+        }
+        if (sampler.GetOffset() > base_offset) {
+            max_offset = std::min(sampler.GetOffset(), max_offset);
+        }
+    }
+    if (max_offset == std::numeric_limits<u32>::max()) {
+        return std::nullopt;
+    }
+    return ((max_offset - base_offset) * 4) / gpu_driver->GetTextureHandlerSize();
+}
+
 } // Anonymous namespace
 
 class ASTDecoder {
@@ -315,4 +362,25 @@ u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
     return pc + 1;
 }
 
+void ShaderIR::PostDecode() {
+    // Deduce texture handler size if needed
+    auto gpu_driver = locker.AccessGuestDriverProfile();
+    DeduceTextureHandlerSize(gpu_driver, used_samplers);
+    // Deduce Indexed Samplers
+    if (!uses_indexed_samplers) {
+        return;
+    }
+    for (auto& sampler : used_samplers) {
+        if (!sampler.IsIndexed()) {
+            continue;
+        }
+        if (const auto size = TryDeduceSamplerSize(sampler, gpu_driver, used_samplers)) {
+            sampler.SetSize(*size);
+        } else {
+            LOG_CRITICAL(HW_GPU, "Failed to deduce size of indexed sampler");
+            sampler.SetSize(1);
+        }
+    }
+}
+
 } // namespace VideoCommon::Shader

src/video_core/shader/decode/arithmetic.cpp

@@ -21,7 +21,7 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) {
 
     Node op_a = GetRegister(instr.gpr8);
 
-    Node op_b = [&]() -> Node {
+    Node op_b = [&] {
         if (instr.is_b_imm) {
             return GetImmediate19(instr);
         } else if (instr.is_b_gpr) {
@@ -141,6 +141,15 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) {
         SetRegister(bb, instr.gpr0, value);
         break;
     }
+    case OpCode::Id::FCMP_R: {
+        UNIMPLEMENTED_IF(instr.fcmp.ftz == 0);
+        Node op_c = GetRegister(instr.gpr39);
+        Node comp = GetPredicateComparisonFloat(instr.fcmp.cond, std::move(op_c), Immediate(0.0f));
+        SetRegister(
+            bb, instr.gpr0,
+            Operation(OperationCode::Select, std::move(comp), std::move(op_a), std::move(op_b)));
+        break;
+    }
     case OpCode::Id::RRO_C:
     case OpCode::Id::RRO_R:
     case OpCode::Id::RRO_IMM: {

src/video_core/shader/decode/arithmetic_integer.cpp

@@ -297,7 +297,7 @@ void ShaderIR::WriteLop3Instruction(NodeBlock& bb, Register dest, Node op_a, Nod
     const Node one = Immediate(1);
     const Node two = Immediate(2);
 
-    Node value{};
+    Node value;
     for (u32 i = 0; i < lop_iterations; ++i) {
         const Node shift_amount = Immediate(i);
 

src/video_core/shader/decode/bfi.cpp

@@ -17,10 +17,13 @@ u32 ShaderIR::DecodeBfi(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
     const auto opcode = OpCode::Decode(instr);
 
-    const auto [base, packed_shift] = [&]() -> std::tuple<Node, Node> {
+    const auto [packed_shift, base] = [&]() -> std::pair<Node, Node> {
         switch (opcode->get().GetId()) {
+        case OpCode::Id::BFI_RC:
+            return {GetRegister(instr.gpr39),
+                    GetConstBuffer(instr.cbuf34.index, instr.cbuf34.offset)};
         case OpCode::Id::BFI_IMM_R:
-            return {GetRegister(instr.gpr39), Immediate(instr.alu.GetSignedImm20_20())};
+            return {Immediate(instr.alu.GetSignedImm20_20()), GetRegister(instr.gpr39)};
         default:
             UNREACHABLE();
             return {Immediate(0), Immediate(0)};

src/video_core/shader/decode/memory.cpp

@@ -19,9 +19,12 @@ namespace VideoCommon::Shader {
 using Tegra::Shader::AtomicOp;
 using Tegra::Shader::AtomicType;
 using Tegra::Shader::Attribute;
+using Tegra::Shader::GlobalAtomicOp;
+using Tegra::Shader::GlobalAtomicType;
 using Tegra::Shader::Instruction;
 using Tegra::Shader::OpCode;
 using Tegra::Shader::Register;
+using Tegra::Shader::StoreType;
 
 namespace {
 
@@ -61,6 +64,27 @@ u32 GetMemorySize(Tegra::Shader::UniformType uniform_type) {
     }
 }
 
+Node ExtractUnaligned(Node value, Node address, u32 mask, u32 size) {
+    Node offset = Operation(OperationCode::UBitwiseAnd, address, Immediate(mask));
+    offset = Operation(OperationCode::ULogicalShiftLeft, std::move(offset), Immediate(3));
+    return Operation(OperationCode::UBitfieldExtract, std::move(value), std::move(offset),
+                     Immediate(size));
+}
+
+Node InsertUnaligned(Node dest, Node value, Node address, u32 mask, u32 size) {
+    Node offset = Operation(OperationCode::UBitwiseAnd, std::move(address), Immediate(mask));
+    offset = Operation(OperationCode::ULogicalShiftLeft, std::move(offset), Immediate(3));
+    return Operation(OperationCode::UBitfieldInsert, std::move(dest), std::move(value),
+                     std::move(offset), Immediate(size));
+}
+
+Node Sign16Extend(Node value) {
+    Node sign = Operation(OperationCode::UBitwiseAnd, value, Immediate(1U << 15));
+    Node is_sign = Operation(OperationCode::LogicalUEqual, std::move(sign), Immediate(1U << 15));
+    Node extend = Operation(OperationCode::Select, is_sign, Immediate(0xFFFF0000), Immediate(0));
+    return Operation(OperationCode::UBitwiseOr, std::move(value), std::move(extend));
+}
+
 } // Anonymous namespace
 
 u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
@@ -136,26 +160,31 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         LOG_DEBUG(HW_GPU, "LD_L cache management mode: {}", static_cast<u64>(instr.ld_l.unknown));
         [[fallthrough]];
     case OpCode::Id::LD_S: {
-        const auto GetMemory = [&](s32 offset) {
+        const auto GetAddress = [&](s32 offset) {
             ASSERT(offset % 4 == 0);
             const Node immediate_offset = Immediate(static_cast<s32>(instr.smem_imm) + offset);
-            const Node address = Operation(OperationCode::IAdd, NO_PRECISE, GetRegister(instr.gpr8),
-                                           immediate_offset);
-            return opcode->get().GetId() == OpCode::Id::LD_S ? GetSharedMemory(address)
-                                                             : GetLocalMemory(address);
+            return Operation(OperationCode::IAdd, GetRegister(instr.gpr8), immediate_offset);
+        };
+        const auto GetMemory = [&](s32 offset) {
+            return opcode->get().GetId() == OpCode::Id::LD_S ? GetSharedMemory(GetAddress(offset))
+                                                             : GetLocalMemory(GetAddress(offset));
         };
 
         switch (instr.ldst_sl.type.Value()) {
-        case Tegra::Shader::StoreType::Bits32:
-        case Tegra::Shader::StoreType::Bits64:
-        case Tegra::Shader::StoreType::Bits128: {
-            const u32 count = [&]() {
+        case StoreType::Signed16:
+            SetRegister(bb, instr.gpr0,
+                        Sign16Extend(ExtractUnaligned(GetMemory(0), GetAddress(0), 0b10, 16)));
+            break;
+        case StoreType::Bits32:
+        case StoreType::Bits64:
+        case StoreType::Bits128: {
+            const u32 count = [&] {
                 switch (instr.ldst_sl.type.Value()) {
-                case Tegra::Shader::StoreType::Bits32:
+                case StoreType::Bits32:
                     return 1;
-                case Tegra::Shader::StoreType::Bits64:
+                case StoreType::Bits64:
                     return 2;
-                case Tegra::Shader::StoreType::Bits128:
+                case StoreType::Bits128:
                     return 4;
                 default:
                     UNREACHABLE();
@@ -212,12 +241,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             // To handle unaligned loads get the bytes used to dereference global memory and extract
             // those bytes from the loaded u32.
             if (IsUnaligned(type)) {
-                Node mask = Immediate(GetUnalignedMask(type));
-                Node offset = Operation(OperationCode::UBitwiseAnd, real_address, std::move(mask));
-                offset = Operation(OperationCode::ULogicalShiftLeft, offset, Immediate(3));
-
-                gmem = Operation(OperationCode::UBitfieldExtract, std::move(gmem),
-                                 std::move(offset), Immediate(size));
+                gmem = ExtractUnaligned(gmem, real_address, GetUnalignedMask(type), size);
             }
 
             SetTemporary(bb, i, gmem);
@@ -269,21 +293,28 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             return Operation(OperationCode::IAdd, NO_PRECISE, GetRegister(instr.gpr8), immediate);
         };
 
-        const auto set_memory = opcode->get().GetId() == OpCode::Id::ST_L
-                                    ? &ShaderIR::SetLocalMemory
-                                    : &ShaderIR::SetSharedMemory;
+        const bool is_local = opcode->get().GetId() == OpCode::Id::ST_L;
+        const auto set_memory = is_local ? &ShaderIR::SetLocalMemory : &ShaderIR::SetSharedMemory;
+        const auto get_memory = is_local ? &ShaderIR::GetLocalMemory : &ShaderIR::GetSharedMemory;
 
         switch (instr.ldst_sl.type.Value()) {
-        case Tegra::Shader::StoreType::Bits128:
+        case StoreType::Bits128:
             (this->*set_memory)(bb, GetAddress(12), GetRegister(instr.gpr0.Value() + 3));
             (this->*set_memory)(bb, GetAddress(8), GetRegister(instr.gpr0.Value() + 2));
             [[fallthrough]];
-        case Tegra::Shader::StoreType::Bits64:
+        case StoreType::Bits64:
             (this->*set_memory)(bb, GetAddress(4), GetRegister(instr.gpr0.Value() + 1));
             [[fallthrough]];
-        case Tegra::Shader::StoreType::Bits32:
+        case StoreType::Bits32:
             (this->*set_memory)(bb, GetAddress(0), GetRegister(instr.gpr0));
             break;
+        case StoreType::Signed16: {
+            Node address = GetAddress(0);
+            Node memory = (this->*get_memory)(address);
+            (this->*set_memory)(
+                bb, address, InsertUnaligned(memory, GetRegister(instr.gpr0), address, 0b10, 16));
+            break;
+        }
         default:
             UNIMPLEMENTED_MSG("{} unhandled type: {}", opcode->get().GetName(),
                               static_cast<u32>(instr.ldst_sl.type.Value()));
@@ -323,18 +354,32 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
             Node value = GetRegister(instr.gpr0.Value() + i);
 
             if (IsUnaligned(type)) {
-                Node mask = Immediate(GetUnalignedMask(type));
-                Node offset = Operation(OperationCode::UBitwiseAnd, real_address, std::move(mask));
-                offset = Operation(OperationCode::ULogicalShiftLeft, offset, Immediate(3));
-
-                value = Operation(OperationCode::UBitfieldInsert, gmem, std::move(value), offset,
-                                  Immediate(size));
+                const u32 mask = GetUnalignedMask(type);
+                value = InsertUnaligned(gmem, std::move(value), real_address, mask, size);
             }
 
             bb.push_back(Operation(OperationCode::Assign, gmem, value));
         }
         break;
     }
+    case OpCode::Id::ATOM: {
+        UNIMPLEMENTED_IF_MSG(instr.atom.operation != GlobalAtomicOp::Add, "operation={}",
+                             static_cast<int>(instr.atom.operation.Value()));
+        UNIMPLEMENTED_IF_MSG(instr.atom.type != GlobalAtomicType::S32, "type={}",
+                             static_cast<int>(instr.atom.type.Value()));
+
+        const auto [real_address, base_address, descriptor] =
+            TrackGlobalMemory(bb, instr, true, true);
+        if (!real_address || !base_address) {
+            // Tracking failed, skip atomic.
+            break;
+        }
+
+        Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
+        Node value = Operation(OperationCode::AtomicAdd, std::move(gmem), GetRegister(instr.gpr20));
+        SetRegister(bb, instr.gpr0, std::move(value));
+        break;
+    }
     case OpCode::Id::ATOMS: {
         UNIMPLEMENTED_IF_MSG(instr.atoms.operation != AtomicOp::Add, "operation={}",
                              static_cast<int>(instr.atoms.operation.Value()));
@@ -348,7 +393,7 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         Node memory = GetSharedMemory(std::move(address));
         Node data = GetRegister(instr.gpr20);
 
-        Node value = Operation(OperationCode::UAtomicAdd, std::move(memory), std::move(data));
+        Node value = Operation(OperationCode::AtomicAdd, std::move(memory), std::move(data));
         SetRegister(bb, instr.gpr0, std::move(value));
         break;
     }