common/CMakeLists: Amend boost dependency

When #2247 was created, thread_queue_list.h was the only user of
boost-related code, however #2252 moved the page table struct into
common, which makes use of Boost.ICL, so we need to add the dependency
to the common library's link interface again.

CMakeLists.txt

@@ -163,12 +163,6 @@ else()
     set(CMAKE_EXE_LINKER_FLAGS_RELEASE "/DEBUG /MANIFEST:NO /INCREMENTAL:NO /OPT:REF,ICF" CACHE STRING "" FORCE)
 endif()
 
-# Fix GCC C++17 and Boost.ICL incompatibility (needed to build dynarmic)
-# See https://bugzilla.redhat.com/show_bug.cgi?id=1485641#c1
-if (CMAKE_COMPILER_IS_GNUCC)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-new-ttp-matching")
-endif()
-
 # Set file offset size to 64 bits.
 #
 # On modern Unixes, this is typically already the case. The lone exception is
@@ -185,9 +179,9 @@ set_property(DIRECTORY APPEND PROPERTY
 # System imported libraries
 # ======================
 
-find_package(Boost 1.63.0 QUIET)
+find_package(Boost 1.66.0 QUIET)
 if (NOT Boost_FOUND)
-    message(STATUS "Boost 1.63.0 or newer not found, falling back to externals")
+    message(STATUS "Boost 1.66.0 or newer not found, falling back to externals")
 
     set(BOOST_ROOT "${PROJECT_SOURCE_DIR}/externals/boost")
     set(Boost_NO_SYSTEM_PATHS OFF)

src/common/CMakeLists.txt

@@ -92,10 +92,14 @@ add_library(common STATIC
     logging/text_formatter.cpp
     logging/text_formatter.h
     math_util.h
+    memory_hook.cpp
+    memory_hook.h
     microprofile.cpp
     microprofile.h
     microprofileui.h
     misc.cpp
+    page_table.cpp
+    page_table.h
     param_package.cpp
     param_package.h
     quaternion.h
@@ -114,6 +118,8 @@ add_library(common STATIC
     threadsafe_queue.h
     timer.cpp
     timer.h
+    uint128.cpp
+    uint128.h
     vector_math.h
     web_result.h
 )

src/common/memory_hook.cpp

@@ -2,10 +2,10 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include "core/memory_hook.h"
+#include "common/memory_hook.h"
 
-namespace Memory {
+namespace Common {
 
 MemoryHook::~MemoryHook() = default;
 
-} // namespace Memory
+} // namespace Common

src/common/memory_hook.h

@@ -9,7 +9,7 @@
 
 #include "common/common_types.h"
 
-namespace Memory {
+namespace Common {
 
 /**
  * Memory hooks have two purposes:
@@ -44,4 +44,4 @@ public:
 };
 
 using MemoryHookPointer = std::shared_ptr<MemoryHook>;
-} // namespace Memory
+} // namespace Common

src/common/page_table.cpp

@@ -0,0 +1,29 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/page_table.h"
+
+namespace Common {
+
+PageTable::PageTable(std::size_t page_size_in_bits) : page_size_in_bits{page_size_in_bits} {}
+
+PageTable::~PageTable() = default;
+
+void PageTable::Resize(std::size_t address_space_width_in_bits) {
+    const std::size_t num_page_table_entries = 1ULL
+                                               << (address_space_width_in_bits - page_size_in_bits);
+
+    pointers.resize(num_page_table_entries);
+    attributes.resize(num_page_table_entries);
+
+    // The default is a 39-bit address space, which causes an initial 1GB allocation size. If the
+    // vector size is subsequently decreased (via resize), the vector might not automatically
+    // actually reallocate/resize its underlying allocation, which wastes up to ~800 MB for
+    // 36-bit titles. Call shrink_to_fit to reduce capacity to what's actually in use.
+
+    pointers.shrink_to_fit();
+    attributes.shrink_to_fit();
+}
+
+} // namespace Common

src/common/page_table.h

@@ -0,0 +1,80 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <vector>
+#include <boost/icl/interval_map.hpp>
+#include "common/common_types.h"
+#include "common/memory_hook.h"
+
+namespace Common {
+
+enum class PageType : u8 {
+    /// Page is unmapped and should cause an access error.
+    Unmapped,
+    /// Page is mapped to regular memory. This is the only type you can get pointers to.
+    Memory,
+    /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and
+    /// invalidation
+    RasterizerCachedMemory,
+    /// Page is mapped to a I/O region. Writing and reading to this page is handled by functions.
+    Special,
+};
+
+struct SpecialRegion {
+    enum class Type {
+        DebugHook,
+        IODevice,
+    } type;
+
+    MemoryHookPointer handler;
+
+    bool operator<(const SpecialRegion& other) const {
+        return std::tie(type, handler) < std::tie(other.type, other.handler);
+    }
+
+    bool operator==(const SpecialRegion& other) const {
+        return std::tie(type, handler) == std::tie(other.type, other.handler);
+    }
+};
+
+/**
+ * A (reasonably) fast way of allowing switchable and remappable process address spaces. It loosely
+ * mimics the way a real CPU page table works.
+ */
+struct PageTable {
+    explicit PageTable(std::size_t page_size_in_bits);
+    ~PageTable();
+
+    /**
+     * Resizes the page table to be able to accomodate enough pages within
+     * a given address space.
+     *
+     * @param address_space_width_in_bits The address size width in bits.
+     */
+    void Resize(std::size_t address_space_width_in_bits);
+
+    /**
+     * Vector of memory pointers backing each page. An entry can only be non-null if the
+     * corresponding entry in the `attributes` vector is of type `Memory`.
+     */
+    std::vector<u8*> pointers;
+
+    /**
+     * Contains MMIO handlers that back memory regions whose entries in the `attribute` vector is
+     * of type `Special`.
+     */
+    boost::icl::interval_map<VAddr, std::set<SpecialRegion>> special_regions;
+
+    /**
+     * Vector of fine grained page attributes. If it is set to any value other than `Memory`, then
+     * the corresponding entry in `pointers` MUST be set to null.
+     */
+    std::vector<PageType> attributes;
+
+    const std::size_t page_size_in_bits{};
+};
+
+} // namespace Common

src/common/thread_queue_list.h

@@ -6,7 +6,6 @@
 
 #include <array>
 #include <deque>
-#include <boost/range/algorithm_ext/erase.hpp>
 
 namespace Common {
 
@@ -111,8 +110,9 @@ struct ThreadQueueList {
     }
 
     void remove(Priority priority, const T& thread_id) {
-        Queue* cur = &queues[priority];
-        boost::remove_erase(cur->data, thread_id);
+        Queue* const cur = &queues[priority];
+        const auto iter = std::remove(cur->data.begin(), cur->data.end(), thread_id);
+        cur->data.erase(iter, cur->data.end());
     }
 
     void rotate(Priority priority) {

src/common/uint128.cpp

@@ -0,0 +1,41 @@
+#ifdef _MSC_VER
+#include <intrin.h>
+
+#pragma intrinsic(_umul128)
+#endif
+#include <cstring>
+#include "common/uint128.h"
+
+namespace Common {
+
+u128 Multiply64Into128(u64 a, u64 b) {
+    u128 result;
+#ifdef _MSC_VER
+    result[0] = _umul128(a, b, &result[1]);
+#else
+    unsigned __int128 tmp = a;
+    tmp *= b;
+    std::memcpy(&result, &tmp, sizeof(u128));
+#endif
+    return result;
+}
+
+std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor) {
+    u64 remainder = dividend[0] % divisor;
+    u64 accum = dividend[0] / divisor;
+    if (dividend[1] == 0)
+        return {accum, remainder};
+    // We ignore dividend[1] / divisor as that overflows
+    const u64 first_segment = (dividend[1] % divisor) << 32;
+    accum += (first_segment / divisor) << 32;
+    const u64 second_segment = (first_segment % divisor) << 32;
+    accum += (second_segment / divisor);
+    remainder += second_segment % divisor;
+    if (remainder >= divisor) {
+        accum++;
+        remainder -= divisor;
+    }
+    return {accum, remainder};
+}
+
+} // namespace Common

src/common/uint128.h

@@ -0,0 +1,14 @@
+
+#include <utility>
+#include "common/common_types.h"
+
+namespace Common {
+
+// This function multiplies 2 u64 values and produces a u128 value;
+u128 Multiply64Into128(u64 a, u64 b);
+
+// This function divides a u128 by a u32 value and produces two u64 values:
+// the result of division and the remainder
+std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor);
+
+} // namespace Common

src/core/CMakeLists.txt

@@ -419,8 +419,6 @@ add_library(core STATIC
     loader/deconstructed_rom_directory.h
     loader/elf.cpp
     loader/elf.h
-    loader/linker.cpp
-    loader/linker.h
     loader/loader.cpp
     loader/loader.h
     loader/nax.cpp
@@ -437,8 +435,6 @@ add_library(core STATIC
     loader/xci.h
     memory.cpp
     memory.h
-    memory_hook.cpp
-    memory_hook.h
     memory_setup.h
     perf_stats.cpp
     perf_stats.h

src/core/arm/dynarmic/arm_dynarmic.cpp

@@ -12,6 +12,7 @@
 #include "core/core.h"
 #include "core/core_cpu.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/svc.h"
@@ -119,7 +120,7 @@ public:
         return std::max(parent.core_timing.GetDowncount(), 0);
     }
     u64 GetCNTPCT() override {
-        return parent.core_timing.GetTicks();
+        return Timing::CpuCyclesToClockCycles(parent.core_timing.GetTicks());
     }
 
     ARM_Dynarmic& parent;
@@ -151,7 +152,7 @@ std::unique_ptr<Dynarmic::A64::Jit> ARM_Dynarmic::MakeJit() const {
     config.tpidr_el0 = &cb->tpidr_el0;
     config.dczid_el0 = 4;
     config.ctr_el0 = 0x8444c004;
-    config.cntfrq_el0 = 19200000; // Value from fusee.
+    config.cntfrq_el0 = Timing::CNTFREQ;
 
     // Unpredictable instructions
     config.define_unpredictable_behaviour = true;

src/core/arm/dynarmic/arm_dynarmic.h

@@ -12,7 +12,7 @@
 #include "core/arm/exclusive_monitor.h"
 #include "core/arm/unicorn/arm_unicorn.h"
 
-namespace Memory {
+namespace Common {
 struct PageTable;
 }
 
@@ -70,7 +70,7 @@ private:
     Timing::CoreTiming& core_timing;
     DynarmicExclusiveMonitor& exclusive_monitor;
 
-    Memory::PageTable* current_page_table = nullptr;
+    Common::PageTable* current_page_table = nullptr;
 };
 
 class DynarmicExclusiveMonitor final : public ExclusiveMonitor {

src/core/core_timing_util.cpp

@@ -7,6 +7,7 @@
 #include <cinttypes>
 #include <limits>
 #include "common/logging/log.h"
+#include "common/uint128.h"
 
 namespace Core::Timing {
 
@@ -60,4 +61,9 @@ s64 nsToCycles(u64 ns) {
     return (BASE_CLOCK_RATE * static_cast<s64>(ns)) / 1000000000;
 }
 
+u64 CpuCyclesToClockCycles(u64 ticks) {
+    const u128 temporal = Common::Multiply64Into128(ticks, CNTFREQ);
+    return Common::Divide128On32(temporal, static_cast<u32>(BASE_CLOCK_RATE)).first;
+}
+
 } // namespace Core::Timing

src/core/core_timing_util.h

@@ -11,6 +11,7 @@ namespace Core::Timing {
 // The below clock rate is based on Switch's clockspeed being widely known as 1.020GHz
 // The exact value used is of course unverified.
 constexpr u64 BASE_CLOCK_RATE = 1019215872; // Switch clock speed is 1020MHz un/docked
+constexpr u64 CNTFREQ = 19200000;           // Value from fusee.
 
 inline s64 msToCycles(int ms) {
     // since ms is int there is no way to overflow
@@ -61,4 +62,6 @@ inline u64 cyclesToMs(s64 cycles) {
     return cycles * 1000 / BASE_CLOCK_RATE;
 }
 
+u64 CpuCyclesToClockCycles(u64 ticks);
+
 } // namespace Core::Timing

src/core/hle/ipc_helpers.h

@@ -274,6 +274,20 @@ inline void ResponseBuilder::Push(u64 value) {
     Push(static_cast<u32>(value >> 32));
 }
 
+template <>
+inline void ResponseBuilder::Push(float value) {
+    u32 integral;
+    std::memcpy(&integral, &value, sizeof(u32));
+    Push(integral);
+}
+
+template <>
+inline void ResponseBuilder::Push(double value) {
+    u64 integral;
+    std::memcpy(&integral, &value, sizeof(u64));
+    Push(integral);
+}
+
 template <>
 inline void ResponseBuilder::Push(bool value) {
     Push(static_cast<u8>(value));
@@ -415,6 +429,22 @@ inline s64 RequestParser::Pop() {
     return static_cast<s64>(Pop<u64>());
 }
 
+template <>
+inline float RequestParser::Pop() {
+    const u32 value = Pop<u32>();
+    float real;
+    std::memcpy(&real, &value, sizeof(real));
+    return real;
+}
+
+template <>
+inline double RequestParser::Pop() {
+    const u64 value = Pop<u64>();
+    float real;
+    std::memcpy(&real, &value, sizeof(real));
+    return real;
+}
+
 template <>
 inline bool RequestParser::Pop() {
     return Pop<u8>() != 0;

src/core/hle/kernel/mutex.cpp

@@ -2,6 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <map>
 #include <utility>
 #include <vector>
 
@@ -9,11 +10,8 @@
 #include "core/core.h"
 #include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/handle_table.h"
-#include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/mutex.h"
 #include "core/hle/kernel/object.h"
-#include "core/hle/kernel/process.h"
-#include "core/hle/kernel/scheduler.h"
 #include "core/hle/kernel/thread.h"
 #include "core/hle/result.h"
 #include "core/memory.h"
@@ -59,47 +57,41 @@ static void TransferMutexOwnership(VAddr mutex_addr, SharedPtr<Thread> current_t
     }
 }
 
-Mutex::Mutex(Core::System& system) : system{system} {}
-Mutex::~Mutex() = default;
-
-ResultCode Mutex::TryAcquire(VAddr address, Handle holding_thread_handle,
+ResultCode Mutex::TryAcquire(HandleTable& handle_table, VAddr address, Handle holding_thread_handle,
                              Handle requesting_thread_handle) {
     // The mutex address must be 4-byte aligned
     if ((address % sizeof(u32)) != 0) {
         return ERR_INVALID_ADDRESS;
     }
 
-    const auto& handle_table = system.Kernel().CurrentProcess()->GetHandleTable();
-    Thread* const current_thread = system.CurrentScheduler().GetCurrentThread();
     SharedPtr<Thread> holding_thread = handle_table.Get<Thread>(holding_thread_handle);
     SharedPtr<Thread> requesting_thread = handle_table.Get<Thread>(requesting_thread_handle);
 
     // TODO(Subv): It is currently unknown if it is possible to lock a mutex in behalf of another
     // thread.
-    ASSERT(requesting_thread == current_thread);
+    ASSERT(requesting_thread == GetCurrentThread());
 
-    const u32 addr_value = Memory::Read32(address);
+    u32 addr_value = Memory::Read32(address);
 
     // If the mutex isn't being held, just return success.
     if (addr_value != (holding_thread_handle | Mutex::MutexHasWaitersFlag)) {
         return RESULT_SUCCESS;
     }
 
-    if (holding_thread == nullptr) {
+    if (holding_thread == nullptr)
         return ERR_INVALID_HANDLE;
-    }
 
     // Wait until the mutex is released
-    current_thread->SetMutexWaitAddress(address);
-    current_thread->SetWaitHandle(requesting_thread_handle);
+    GetCurrentThread()->SetMutexWaitAddress(address);
+    GetCurrentThread()->SetWaitHandle(requesting_thread_handle);
 
-    current_thread->SetStatus(ThreadStatus::WaitMutex);
-    current_thread->InvalidateWakeupCallback();
+    GetCurrentThread()->SetStatus(ThreadStatus::WaitMutex);
+    GetCurrentThread()->InvalidateWakeupCallback();
 
     // Update the lock holder thread's priority to prevent priority inversion.
-    holding_thread->AddMutexWaiter(current_thread);
+    holding_thread->AddMutexWaiter(GetCurrentThread());
 
-    system.PrepareReschedule();
+    Core::System::GetInstance().PrepareReschedule();
 
     return RESULT_SUCCESS;
 }
@@ -110,8 +102,7 @@ ResultCode Mutex::Release(VAddr address) {
         return ERR_INVALID_ADDRESS;
     }
 
-    auto* const current_thread = system.CurrentScheduler().GetCurrentThread();
-    auto [thread, num_waiters] = GetHighestPriorityMutexWaitingThread(current_thread, address);
+    auto [thread, num_waiters] = GetHighestPriorityMutexWaitingThread(GetCurrentThread(), address);
 
     // There are no more threads waiting for the mutex, release it completely.
     if (thread == nullptr) {
@@ -120,7 +111,7 @@ ResultCode Mutex::Release(VAddr address) {
     }
 
     // Transfer the ownership of the mutex from the previous owner to the new one.
-    TransferMutexOwnership(address, current_thread, thread);
+    TransferMutexOwnership(address, GetCurrentThread(), thread);
 
     u32 mutex_value = thread->GetWaitHandle();
 

src/core/hle/kernel/mutex.h

@@ -5,34 +5,32 @@
 #pragma once
 
 #include "common/common_types.h"
+#include "core/hle/kernel/object.h"
 
 union ResultCode;
 
-namespace Core {
-class System;
-}
-
 namespace Kernel {
 
+class HandleTable;
+class Thread;
+
 class Mutex final {
 public:
-    explicit Mutex(Core::System& system);
-    ~Mutex();
-
     /// Flag that indicates that a mutex still has threads waiting for it.
     static constexpr u32 MutexHasWaitersFlag = 0x40000000;
     /// Mask of the bits in a mutex address value that contain the mutex owner.
     static constexpr u32 MutexOwnerMask = 0xBFFFFFFF;
 
     /// Attempts to acquire a mutex at the specified address.
-    ResultCode TryAcquire(VAddr address, Handle holding_thread_handle,
-                          Handle requesting_thread_handle);
+    static ResultCode TryAcquire(HandleTable& handle_table, VAddr address,
+                                 Handle holding_thread_handle, Handle requesting_thread_handle);
 
     /// Releases the mutex at the specified address.
-    ResultCode Release(VAddr address);
+    static ResultCode Release(VAddr address);
 
 private:
-    Core::System& system;
+    Mutex() = default;
+    ~Mutex() = default;
 };
 
 } // namespace Kernel

src/core/hle/kernel/process.cpp

@@ -31,7 +31,7 @@ namespace {
  */
 void SetupMainThread(Process& owner_process, KernelCore& kernel, VAddr entry_point, u32 priority) {
     // Setup page table so we can write to memory
-    SetCurrentPageTable(&owner_process.VMManager().page_table);
+    Memory::SetCurrentPageTable(&owner_process.VMManager().page_table);
 
     // Initialize new "main" thread
     const VAddr stack_top = owner_process.VMManager().GetTLSIORegionEndAddress();
@@ -231,8 +231,7 @@ void Process::LoadModule(CodeSet module_, VAddr base_addr) {
 }
 
 Process::Process(Core::System& system)
-    : WaitObject{system.Kernel()}, address_arbiter{system}, mutex{system}, system{system} {}
-
+    : WaitObject{system.Kernel()}, address_arbiter{system}, system{system} {}
 Process::~Process() = default;
 
 void Process::Acquire(Thread* thread) {

src/core/hle/kernel/process.h

@@ -14,7 +14,6 @@
 #include "common/common_types.h"
 #include "core/hle/kernel/address_arbiter.h"
 #include "core/hle/kernel/handle_table.h"
-#include "core/hle/kernel/mutex.h"
 #include "core/hle/kernel/process_capability.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/hle/kernel/wait_object.h"
@@ -166,16 +165,6 @@ public:
         return address_arbiter;
     }
 
-    /// Gets a reference to the process' mutex lock.
-    Mutex& GetMutex() {
-        return mutex;
-    }
-
-    /// Gets a const reference to the process' mutex lock
-    const Mutex& GetMutex() const {
-        return mutex;
-    }
-
     /// Gets the current status of the process
     ProcessStatus GetStatus() const {
         return status;
@@ -338,11 +327,6 @@ private:
     /// Per-process address arbiter.
     AddressArbiter address_arbiter;
 
-    /// The per-process mutex lock instance used for handling various
-    /// forms of services, such as lock arbitration, and condition
-    /// variable related facilities.
-    Mutex mutex;
-
     /// Random values for svcGetInfo RandomEntropy
     std::array<u64, RANDOM_ENTROPY_SIZE> random_entropy;
 

src/core/hle/kernel/scheduler.cpp

@@ -96,7 +96,7 @@ void Scheduler::SwitchContext(Thread* new_thread) {
         auto* const thread_owner_process = current_thread->GetOwnerProcess();
         if (previous_process != thread_owner_process) {
             system.Kernel().MakeCurrentProcess(thread_owner_process);
-            SetCurrentPageTable(&thread_owner_process->VMManager().page_table);
+            Memory::SetCurrentPageTable(&thread_owner_process->VMManager().page_table);
         }
 
         cpu_core.LoadContext(new_thread->GetContext());
@@ -199,8 +199,7 @@ void Scheduler::YieldWithoutLoadBalancing(Thread* thread) {
     ASSERT(thread->GetPriority() < THREADPRIO_COUNT);
 
     // Yield this thread -- sleep for zero time and force reschedule to different thread
-    WaitCurrentThread_Sleep();
-    GetCurrentThread()->WakeAfterDelay(0);
+    GetCurrentThread()->Sleep(0);
 }
 
 void Scheduler::YieldWithLoadBalancing(Thread* thread) {
@@ -215,8 +214,7 @@ void Scheduler::YieldWithLoadBalancing(Thread* thread) {
     ASSERT(priority < THREADPRIO_COUNT);
 
     // Sleep for zero time to be able to force reschedule to different thread
-    WaitCurrentThread_Sleep();
-    GetCurrentThread()->WakeAfterDelay(0);
+    GetCurrentThread()->Sleep(0);
 
     Thread* suggested_thread = nullptr;
 

src/core/hle/kernel/svc.cpp

@@ -551,9 +551,9 @@ static ResultCode ArbitrateLock(Handle holding_thread_handle, VAddr mutex_addr,
         return ERR_INVALID_ADDRESS;
     }
 
-    auto* const current_process = Core::System::GetInstance().Kernel().CurrentProcess();
-    return current_process->GetMutex().TryAcquire(mutex_addr, holding_thread_handle,
-                                                  requesting_thread_handle);
+    auto& handle_table = Core::CurrentProcess()->GetHandleTable();
+    return Mutex::TryAcquire(handle_table, mutex_addr, holding_thread_handle,
+                             requesting_thread_handle);
 }
 
 /// Unlock a mutex
@@ -571,8 +571,7 @@ static ResultCode ArbitrateUnlock(VAddr mutex_addr) {
         return ERR_INVALID_ADDRESS;
     }
 
-    auto* const current_process = Core::System::GetInstance().Kernel().CurrentProcess();
-    return current_process->GetMutex().Release(mutex_addr);
+    return Mutex::Release(mutex_addr);
 }
 
 enum class BreakType : u32 {
@@ -1285,10 +1284,14 @@ static ResultCode StartThread(Handle thread_handle) {
 
 /// Called when a thread exits
 static void ExitThread() {
-    LOG_TRACE(Kernel_SVC, "called, pc=0x{:08X}", Core::CurrentArmInterface().GetPC());
+    auto& system = Core::System::GetInstance();
 
-    ExitCurrentThread();
-    Core::System::GetInstance().PrepareReschedule();
+    LOG_TRACE(Kernel_SVC, "called, pc=0x{:08X}", system.CurrentArmInterface().GetPC());
+
+    auto* const current_thread = system.CurrentScheduler().GetCurrentThread();
+    current_thread->Stop();
+    system.CurrentScheduler().RemoveThread(current_thread);
+    system.PrepareReschedule();
 }
 
 /// Sleep the current thread
@@ -1301,32 +1304,32 @@ static void SleepThread(s64 nanoseconds) {
         YieldAndWaitForLoadBalancing = -2,
     };
 
+    auto& system = Core::System::GetInstance();
+    auto& scheduler = system.CurrentScheduler();
+    auto* const current_thread = scheduler.GetCurrentThread();
+
     if (nanoseconds <= 0) {
-        auto& scheduler{Core::System::GetInstance().CurrentScheduler()};
         switch (static_cast<SleepType>(nanoseconds)) {
         case SleepType::YieldWithoutLoadBalancing:
-            scheduler.YieldWithoutLoadBalancing(GetCurrentThread());
+            scheduler.YieldWithoutLoadBalancing(current_thread);
             break;
         case SleepType::YieldWithLoadBalancing:
-            scheduler.YieldWithLoadBalancing(GetCurrentThread());
+            scheduler.YieldWithLoadBalancing(current_thread);
             break;
         case SleepType::YieldAndWaitForLoadBalancing:
-            scheduler.YieldAndWaitForLoadBalancing(GetCurrentThread());
+            scheduler.YieldAndWaitForLoadBalancing(current_thread);
             break;
         default:
             UNREACHABLE_MSG("Unimplemented sleep yield type '{:016X}'!", nanoseconds);
         }
     } else {
-        // Sleep current thread and check for next thread to schedule
-        WaitCurrentThread_Sleep();
-
-        // Create an event to wake the thread up after the specified nanosecond delay has passed
-        GetCurrentThread()->WakeAfterDelay(nanoseconds);
+        current_thread->Sleep(nanoseconds);
     }
 
     // Reschedule all CPU cores
-    for (std::size_t i = 0; i < Core::NUM_CPU_CORES; ++i)
-        Core::System::GetInstance().CpuCore(i).PrepareReschedule();
+    for (std::size_t i = 0; i < Core::NUM_CPU_CORES; ++i) {
+        system.CpuCore(i).PrepareReschedule();
+    }
 }
 
 /// Wait process wide key atomic
@@ -1337,15 +1340,11 @@ static ResultCode WaitProcessWideKeyAtomic(VAddr mutex_addr, VAddr condition_var
         "called mutex_addr={:X}, condition_variable_addr={:X}, thread_handle=0x{:08X}, timeout={}",
         mutex_addr, condition_variable_addr, thread_handle, nano_seconds);
 
-    auto* const current_process = Core::System::GetInstance().Kernel().CurrentProcess();
-    const auto& handle_table = current_process->GetHandleTable();
+    const auto& handle_table = Core::CurrentProcess()->GetHandleTable();
     SharedPtr<Thread> thread = handle_table.Get<Thread>(thread_handle);
     ASSERT(thread);
 
-    const auto release_result = current_process->GetMutex().Release(mutex_addr);
-    if (release_result.IsError()) {
-        return release_result;
-    }
+    CASCADE_CODE(Mutex::Release(mutex_addr));
 
     SharedPtr<Thread> current_thread = GetCurrentThread();
     current_thread->SetCondVarWaitAddress(condition_variable_addr);

src/core/hle/kernel/thread.cpp

@@ -7,8 +7,6 @@
 #include <optional>
 #include <vector>
 
-#include <boost/range/algorithm_ext/erase.hpp>
-
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "common/logging/log.h"
@@ -68,17 +66,6 @@ void Thread::Stop() {
     owner_process->FreeTLSSlot(tls_address);
 }
 
-void WaitCurrentThread_Sleep() {
-    Thread* thread = GetCurrentThread();
-    thread->SetStatus(ThreadStatus::WaitSleep);
-}
-
-void ExitCurrentThread() {
-    Thread* thread = GetCurrentThread();
-    thread->Stop();
-    Core::System::GetInstance().CurrentScheduler().RemoveThread(thread);
-}
-
 void Thread::WakeAfterDelay(s64 nanoseconds) {
     // Don't schedule a wakeup if the thread wants to wait forever
     if (nanoseconds == -1)
@@ -269,8 +256,8 @@ void Thread::AddMutexWaiter(SharedPtr<Thread> thread) {
     if (thread->lock_owner == this) {
         // If the thread is already waiting for this thread to release the mutex, ensure that the
         // waiters list is consistent and return without doing anything.
-        auto itr = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
-        ASSERT(itr != wait_mutex_threads.end());
+        const auto iter = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
+        ASSERT(iter != wait_mutex_threads.end());
         return;
     }
 
@@ -278,11 +265,16 @@ void Thread::AddMutexWaiter(SharedPtr<Thread> thread) {
     ASSERT(thread->lock_owner == nullptr);
 
     // Ensure that the thread is not already in the list of mutex waiters
-    auto itr = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
-    ASSERT(itr == wait_mutex_threads.end());
+    const auto iter = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
+    ASSERT(iter == wait_mutex_threads.end());
 
+    // Keep the list in an ordered fashion
+    const auto insertion_point = std::find_if(
+        wait_mutex_threads.begin(), wait_mutex_threads.end(),
+        [&thread](const auto& entry) { return entry->GetPriority() > thread->GetPriority(); });
+    wait_mutex_threads.insert(insertion_point, thread);
     thread->lock_owner = this;
-    wait_mutex_threads.emplace_back(std::move(thread));
+
     UpdatePriority();
 }
 
@@ -290,32 +282,44 @@ void Thread::RemoveMutexWaiter(SharedPtr<Thread> thread) {
     ASSERT(thread->lock_owner == this);
 
     // Ensure that the thread is in the list of mutex waiters
-    auto itr = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
-    ASSERT(itr != wait_mutex_threads.end());
+    const auto iter = std::find(wait_mutex_threads.begin(), wait_mutex_threads.end(), thread);
+    ASSERT(iter != wait_mutex_threads.end());
+
+    wait_mutex_threads.erase(iter);
 
-    boost::remove_erase(wait_mutex_threads, thread);
     thread->lock_owner = nullptr;
     UpdatePriority();
 }
 
 void Thread::UpdatePriority() {
-    // Find the highest priority among all the threads that are waiting for this thread's lock
+    // If any of the threads waiting on the mutex have a higher priority
+    // (taking into account priority inheritance), then this thread inherits
+    // that thread's priority.
     u32 new_priority = nominal_priority;
-    for (const auto& thread : wait_mutex_threads) {
-        if (thread->nominal_priority < new_priority)
-            new_priority = thread->nominal_priority;
+    if (!wait_mutex_threads.empty()) {
+        if (wait_mutex_threads.front()->current_priority < new_priority) {
+            new_priority = wait_mutex_threads.front()->current_priority;
+        }
     }
 
-    if (new_priority == current_priority)
+    if (new_priority == current_priority) {
         return;
+    }
 
     scheduler->SetThreadPriority(this, new_priority);
-
     current_priority = new_priority;
 
+    if (!lock_owner) {
+        return;
+    }
+
+    // Ensure that the thread is within the correct location in the waiting list.
+    auto old_owner = lock_owner;
+    lock_owner->RemoveMutexWaiter(this);
+    old_owner->AddMutexWaiter(this);
+
     // Recursively update the priority of the thread that depends on the priority of this one.
-    if (lock_owner)
-        lock_owner->UpdatePriority();
+    lock_owner->UpdatePriority();
 }
 
 void Thread::ChangeCore(u32 core, u64 mask) {
@@ -391,6 +395,14 @@ void Thread::SetActivity(ThreadActivity value) {
     }
 }
 
+void Thread::Sleep(s64 nanoseconds) {
+    // Sleep current thread and check for next thread to schedule
+    SetStatus(ThreadStatus::WaitSleep);
+
+    // Create an event to wake the thread up after the specified nanosecond delay has passed
+    WakeAfterDelay(nanoseconds);
+}
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 
 /**

src/core/hle/kernel/thread.h

@@ -383,6 +383,9 @@ public:
 
     void SetActivity(ThreadActivity value);
 
+    /// Sleeps this thread for the given amount of nanoseconds.
+    void Sleep(s64 nanoseconds);
+
 private:
     explicit Thread(KernelCore& kernel);
     ~Thread() override;
@@ -398,8 +401,14 @@ private:
     VAddr entry_point = 0;
     VAddr stack_top = 0;
 
-    u32 nominal_priority = 0; ///< Nominal thread priority, as set by the emulated application
-    u32 current_priority = 0; ///< Current thread priority, can be temporarily changed
+    /// Nominal thread priority, as set by the emulated application.
+    /// The nominal priority is the thread priority without priority
+    /// inheritance taken into account.
+    u32 nominal_priority = 0;
+
+    /// Current thread priority. This may change over the course of the
+    /// thread's lifetime in order to facilitate priority inheritance.
+    u32 current_priority = 0;
 
     u64 total_cpu_time_ticks = 0; ///< Total CPU running ticks.
     u64 last_running_ticks = 0;   ///< CPU tick when thread was last running
@@ -460,14 +469,4 @@ private:
  */
 Thread* GetCurrentThread();
 
-/**
- * Waits the current thread on a sleep
- */
-void WaitCurrentThread_Sleep();
-
-/**
- * Stops the current thread and removes it from the thread_list
- */
-void ExitCurrentThread();
-
 } // namespace Kernel

src/core/hle/kernel/vm_manager.cpp

@@ -7,13 +7,13 @@
 #include <utility>
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "common/memory_hook.h"
 #include "core/arm/arm_interface.h"
 #include "core/core.h"
 #include "core/file_sys/program_metadata.h"
 #include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
-#include "core/memory_hook.h"
 #include "core/memory_setup.h"
 
 namespace Kernel {
@@ -177,7 +177,7 @@ ResultVal<VAddr> VMManager::FindFreeRegion(u64 size) const {
 
 ResultVal<VMManager::VMAHandle> VMManager::MapMMIO(VAddr target, PAddr paddr, u64 size,
                                                    MemoryState state,
-                                                   Memory::MemoryHookPointer mmio_handler) {
+                                                   Common::MemoryHookPointer mmio_handler) {
     // This is the appropriately sized VMA that will turn into our allocation.
     CASCADE_RESULT(VMAIter vma_handle, CarveVMA(target, size));
     VirtualMemoryArea& final_vma = vma_handle->second;
@@ -624,7 +624,7 @@ void VMManager::ClearPageTable() {
     std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr);
     page_table.special_regions.clear();
     std::fill(page_table.attributes.begin(), page_table.attributes.end(),
-              Memory::PageType::Unmapped);
+              Common::PageType::Unmapped);
 }
 
 VMManager::CheckResults VMManager::CheckRangeState(VAddr address, u64 size, MemoryState state_mask,

src/core/hle/kernel/vm_manager.h

@@ -9,9 +9,10 @@
 #include <tuple>
 #include <vector>
 #include "common/common_types.h"
+#include "common/memory_hook.h"
+#include "common/page_table.h"
 #include "core/hle/result.h"
 #include "core/memory.h"
-#include "core/memory_hook.h"
 
 namespace FileSys {
 enum class ProgramAddressSpaceType : u8;
@@ -290,7 +291,7 @@ struct VirtualMemoryArea {
     // Settings for type = MMIO
     /// Physical address of the register area this VMA maps to.
     PAddr paddr = 0;
-    Memory::MemoryHookPointer mmio_handler = nullptr;
+    Common::MemoryHookPointer mmio_handler = nullptr;
 
     /// Tests if this area can be merged to the right with `next`.
     bool CanBeMergedWith(const VirtualMemoryArea& next) const;
@@ -368,7 +369,7 @@ public:
      * @param mmio_handler The handler that will implement read and write for this MMIO region.
      */
     ResultVal<VMAHandle> MapMMIO(VAddr target, PAddr paddr, u64 size, MemoryState state,
-                                 Memory::MemoryHookPointer mmio_handler);
+                                 Common::MemoryHookPointer mmio_handler);
 
     /// Unmaps a range of addresses, splitting VMAs as necessary.
     ResultCode UnmapRange(VAddr target, u64 size);
@@ -509,7 +510,7 @@ public:
 
     /// Each VMManager has its own page table, which is set as the main one when the owning process
     /// is scheduled.
-    Memory::PageTable page_table;
+    Common::PageTable page_table{Memory::PAGE_BITS};
 
 private:
     using VMAIter = VMAMap::iterator;

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

@@ -2,10 +2,10 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <algorithm>
 #include <array>
 #include <cinttypes>
 #include <cstring>
-#include <stack>
 #include "audio_core/audio_renderer.h"
 #include "core/core.h"
 #include "core/file_sys/savedata_factory.h"
@@ -93,38 +93,84 @@ void IWindowController::AcquireForegroundRights(Kernel::HLERequestContext& ctx)
 }
 
 IAudioController::IAudioController() : ServiceFramework("IAudioController") {
+    // clang-format off
     static const FunctionInfo functions[] = {
         {0, &IAudioController::SetExpectedMasterVolume, "SetExpectedMasterVolume"},
-        {1, &IAudioController::GetMainAppletExpectedMasterVolume,
-         "GetMainAppletExpectedMasterVolume"},
-        {2, &IAudioController::GetLibraryAppletExpectedMasterVolume,
-         "GetLibraryAppletExpectedMasterVolume"},
-        {3, nullptr, "ChangeMainAppletMasterVolume"},
-        {4, nullptr, "SetTransparentVolumeRate"},
+        {1, &IAudioController::GetMainAppletExpectedMasterVolume, "GetMainAppletExpectedMasterVolume"},
+        {2, &IAudioController::GetLibraryAppletExpectedMasterVolume, "GetLibraryAppletExpectedMasterVolume"},
+        {3, &IAudioController::ChangeMainAppletMasterVolume, "ChangeMainAppletMasterVolume"},
+        {4, &IAudioController::SetTransparentAudioRate, "SetTransparentVolumeRate"},
     };
+    // clang-format on
+
     RegisterHandlers(functions);
 }
 
 IAudioController::~IAudioController() = default;
 
 void IAudioController::SetExpectedMasterVolume(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_AM, "(STUBBED) called");
+    IPC::RequestParser rp{ctx};
+    const float main_applet_volume_tmp = rp.Pop<float>();
+    const float library_applet_volume_tmp = rp.Pop<float>();
+
+    LOG_DEBUG(Service_AM, "called. main_applet_volume={}, library_applet_volume={}",
+              main_applet_volume_tmp, library_applet_volume_tmp);
+
+    // Ensure the volume values remain within the 0-100% range
+    main_applet_volume = std::clamp(main_applet_volume_tmp, min_allowed_volume, max_allowed_volume);
+    library_applet_volume =
+        std::clamp(library_applet_volume_tmp, min_allowed_volume, max_allowed_volume);
+
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);
 }
 
 void IAudioController::GetMainAppletExpectedMasterVolume(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_AM, "(STUBBED) called");
+    LOG_DEBUG(Service_AM, "called. main_applet_volume={}", main_applet_volume);
     IPC::ResponseBuilder rb{ctx, 3};
     rb.Push(RESULT_SUCCESS);
-    rb.Push(volume);
+    rb.Push(main_applet_volume);
 }
 
 void IAudioController::GetLibraryAppletExpectedMasterVolume(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_AM, "(STUBBED) called");
+    LOG_DEBUG(Service_AM, "called. library_applet_volume={}", library_applet_volume);
     IPC::ResponseBuilder rb{ctx, 3};
     rb.Push(RESULT_SUCCESS);
-    rb.Push(volume);
+    rb.Push(library_applet_volume);
+}
+
+void IAudioController::ChangeMainAppletMasterVolume(Kernel::HLERequestContext& ctx) {
+    struct Parameters {
+        float volume;
+        s64 fade_time_ns;
+    };
+    static_assert(sizeof(Parameters) == 16);
+
+    IPC::RequestParser rp{ctx};
+    const auto parameters = rp.PopRaw<Parameters>();
+
+    LOG_DEBUG(Service_AM, "called. volume={}, fade_time_ns={}", parameters.volume,
+              parameters.fade_time_ns);
+
+    main_applet_volume = std::clamp(parameters.volume, min_allowed_volume, max_allowed_volume);
+    fade_time_ns = std::chrono::nanoseconds{parameters.fade_time_ns};
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
+}
+
+void IAudioController::SetTransparentAudioRate(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const float transparent_volume_rate_tmp = rp.Pop<float>();
+
+    LOG_DEBUG(Service_AM, "called. transparent_volume_rate={}", transparent_volume_rate_tmp);
+
+    // Clamp volume range to 0-100%.
+    transparent_volume_rate =
+        std::clamp(transparent_volume_rate_tmp, min_allowed_volume, max_allowed_volume);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
 }
 
 IDisplayController::IDisplayController() : ServiceFramework("IDisplayController") {

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

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <chrono>
 #include <memory>
 #include <queue>
 #include "core/hle/kernel/writable_event.h"
@@ -81,8 +82,21 @@ private:
     void SetExpectedMasterVolume(Kernel::HLERequestContext& ctx);
     void GetMainAppletExpectedMasterVolume(Kernel::HLERequestContext& ctx);
     void GetLibraryAppletExpectedMasterVolume(Kernel::HLERequestContext& ctx);
+    void ChangeMainAppletMasterVolume(Kernel::HLERequestContext& ctx);
+    void SetTransparentAudioRate(Kernel::HLERequestContext& ctx);
 
-    u32 volume{100};
+    static constexpr float min_allowed_volume = 0.0f;
+    static constexpr float max_allowed_volume = 1.0f;
+
+    float main_applet_volume{0.25f};
+    float library_applet_volume{max_allowed_volume};
+    float transparent_volume_rate{min_allowed_volume};
+
+    // Volume transition fade time in nanoseconds.
+    // e.g. If the main applet volume was 0% and was changed to 50%
+    //      with a fade of 50ns, then over the course of 50ns,
+    //      the volume will gradually fade up to 50%
+    std::chrono::nanoseconds fade_time_ns{0};
 };
 
 class IDisplayController final : public ServiceFramework<IDisplayController> {

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

@@ -8,6 +8,7 @@
 #include <vector>
 
 #include <opus.h>
+#include <opus_multistream.h>
 
 #include "common/assert.h"
 #include "common/logging/log.h"
@@ -18,12 +19,12 @@
 namespace Service::Audio {
 namespace {
 struct OpusDeleter {
-    void operator()(void* ptr) const {
-        operator delete(ptr);
+    void operator()(OpusMSDecoder* ptr) const {
+        opus_multistream_decoder_destroy(ptr);
     }
 };
 
-using OpusDecoderPtr = std::unique_ptr<OpusDecoder, OpusDeleter>;
+using OpusDecoderPtr = std::unique_ptr<OpusMSDecoder, OpusDeleter>;
 
 struct OpusPacketHeader {
     // Packet size in bytes.
@@ -33,7 +34,7 @@ struct OpusPacketHeader {
 };
 static_assert(sizeof(OpusPacketHeader) == 0x8, "OpusHeader is an invalid size");
 
-class OpusDecoderStateBase {
+class OpusDecoderState {
 public:
     /// Describes extra behavior that may be asked of the decoding context.
     enum class ExtraBehavior {
@@ -49,22 +50,13 @@ public:
         Enabled,
     };
 
-    virtual ~OpusDecoderStateBase() = default;
-
-    // Decodes interleaved Opus packets. Optionally allows reporting time taken to
-    // perform the decoding, as well as any relevant extra behavior.
-    virtual void DecodeInterleaved(Kernel::HLERequestContext& ctx, PerfTime perf_time,
-                                   ExtraBehavior extra_behavior) = 0;
-};
-
-// Represents the decoder state for a non-multistream decoder.
-class OpusDecoderState final : public OpusDecoderStateBase {
-public:
     explicit OpusDecoderState(OpusDecoderPtr decoder, u32 sample_rate, u32 channel_count)
         : decoder{std::move(decoder)}, sample_rate{sample_rate}, channel_count{channel_count} {}
 
+    // Decodes interleaved Opus packets. Optionally allows reporting time taken to
+    // perform the decoding, as well as any relevant extra behavior.
     void DecodeInterleaved(Kernel::HLERequestContext& ctx, PerfTime perf_time,
-                           ExtraBehavior extra_behavior) override {
+                           ExtraBehavior extra_behavior) {
         if (perf_time == PerfTime::Disabled) {
             DecodeInterleavedHelper(ctx, nullptr, extra_behavior);
         } else {
@@ -135,7 +127,7 @@ private:
 
         const int frame_size = (static_cast<int>(raw_output_sz / sizeof(s16) / channel_count));
         const auto out_sample_count =
-            opus_decode(decoder.get(), frame, hdr.size, output.data(), frame_size, 0);
+            opus_multistream_decode(decoder.get(), frame, hdr.size, output.data(), frame_size, 0);
         if (out_sample_count < 0) {
             LOG_ERROR(Audio,
                       "Incorrect sample count received from opus_decode, "
@@ -158,7 +150,7 @@ private:
     void ResetDecoderContext() {
         ASSERT(decoder != nullptr);
 
-        opus_decoder_ctl(decoder.get(), OPUS_RESET_STATE);
+        opus_multistream_decoder_ctl(decoder.get(), OPUS_RESET_STATE);
     }
 
     OpusDecoderPtr decoder;
@@ -168,7 +160,7 @@ private:
 
 class IHardwareOpusDecoderManager final : public ServiceFramework<IHardwareOpusDecoderManager> {
 public:
-    explicit IHardwareOpusDecoderManager(std::unique_ptr<OpusDecoderStateBase> decoder_state)
+    explicit IHardwareOpusDecoderManager(OpusDecoderState decoder_state)
         : ServiceFramework("IHardwareOpusDecoderManager"), decoder_state{std::move(decoder_state)} {
         // clang-format off
         static const FunctionInfo functions[] = {
@@ -190,35 +182,51 @@ private:
     void DecodeInterleavedOld(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Audio, "called");
 
-        decoder_state->DecodeInterleaved(ctx, OpusDecoderStateBase::PerfTime::Disabled,
-                                         OpusDecoderStateBase::ExtraBehavior::None);
+        decoder_state.DecodeInterleaved(ctx, OpusDecoderState::PerfTime::Disabled,
+                                        OpusDecoderState::ExtraBehavior::None);
     }
 
     void DecodeInterleavedWithPerfOld(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Audio, "called");
 
-        decoder_state->DecodeInterleaved(ctx, OpusDecoderStateBase::PerfTime::Enabled,
-                                         OpusDecoderStateBase::ExtraBehavior::None);
+        decoder_state.DecodeInterleaved(ctx, OpusDecoderState::PerfTime::Enabled,
+                                        OpusDecoderState::ExtraBehavior::None);
     }
 
     void DecodeInterleaved(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Audio, "called");
 
         IPC::RequestParser rp{ctx};
-        const auto extra_behavior = rp.Pop<bool>()
-                                        ? OpusDecoderStateBase::ExtraBehavior::ResetContext
-                                        : OpusDecoderStateBase::ExtraBehavior::None;
+        const auto extra_behavior = rp.Pop<bool>() ? OpusDecoderState::ExtraBehavior::ResetContext
+                                                   : OpusDecoderState::ExtraBehavior::None;
 
-        decoder_state->DecodeInterleaved(ctx, OpusDecoderStateBase::PerfTime::Enabled,
-                                         extra_behavior);
+        decoder_state.DecodeInterleaved(ctx, OpusDecoderState::PerfTime::Enabled, extra_behavior);
     }
 
-    std::unique_ptr<OpusDecoderStateBase> decoder_state;
+    OpusDecoderState decoder_state;
 };
 
 std::size_t WorkerBufferSize(u32 channel_count) {
     ASSERT_MSG(channel_count == 1 || channel_count == 2, "Invalid channel count");
-    return opus_decoder_get_size(static_cast<int>(channel_count));
+    constexpr int num_streams = 1;
+    const int num_stereo_streams = channel_count == 2 ? 1 : 0;
+    return opus_multistream_decoder_get_size(num_streams, num_stereo_streams);
+}
+
+// Creates the mapping table that maps the input channels to the particular
+// output channels. In the stereo case, we map the left and right input channels
+// to the left and right output channels respectively.
+//
+// However, in the monophonic case, we only map the one available channel
+// to the sole output channel. We specify 255 for the would-be right channel
+// as this is a special value defined by Opus to indicate to the decoder to
+// ignore that channel.
+std::array<u8, 2> CreateMappingTable(u32 channel_count) {
+    if (channel_count == 2) {
+        return {{0, 1}};
+    }
+
+    return {{0, 255}};
 }
 } // Anonymous namespace
 
@@ -259,9 +267,15 @@ void HwOpus::OpenOpusDecoder(Kernel::HLERequestContext& ctx) {
     const std::size_t worker_sz = WorkerBufferSize(channel_count);
     ASSERT_MSG(buffer_sz >= worker_sz, "Worker buffer too large");
 
-    OpusDecoderPtr decoder{static_cast<OpusDecoder*>(operator new(worker_sz))};
-    if (const int err = opus_decoder_init(decoder.get(), sample_rate, channel_count)) {
-        LOG_ERROR(Audio, "Failed to init opus decoder with error={}", err);
+    const int num_stereo_streams = channel_count == 2 ? 1 : 0;
+    const auto mapping_table = CreateMappingTable(channel_count);
+
+    int error = 0;
+    OpusDecoderPtr decoder{
+        opus_multistream_decoder_create(sample_rate, static_cast<int>(channel_count), 1,
+                                        num_stereo_streams, mapping_table.data(), &error)};
+    if (error != OPUS_OK || decoder == nullptr) {
+        LOG_ERROR(Audio, "Failed to create Opus decoder (error={}).", error);
         IPC::ResponseBuilder rb{ctx, 2};
         // TODO(ogniK): Use correct error code
         rb.Push(ResultCode(-1));
@@ -271,7 +285,7 @@ void HwOpus::OpenOpusDecoder(Kernel::HLERequestContext& ctx) {
     IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);
     rb.PushIpcInterface<IHardwareOpusDecoderManager>(
-        std::make_unique<OpusDecoderState>(std::move(decoder), sample_rate, channel_count));
+        OpusDecoderState{std::move(decoder), sample_rate, channel_count});
 }
 
 HwOpus::HwOpus() : ServiceFramework("hwopus") {

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

@@ -733,7 +733,10 @@ FSP_SRV::FSP_SRV() : ServiceFramework("fsp-srv") {
 FSP_SRV::~FSP_SRV() = default;
 
 void FSP_SRV::SetCurrentProcess(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_FS, "(STUBBED) called");
+    IPC::RequestParser rp{ctx};
+    current_process_id = rp.Pop<u64>();
+
+    LOG_DEBUG(Service_FS, "called. current_process_id=0x{:016X}", current_process_id);
 
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);

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

@@ -32,6 +32,7 @@ private:
     void OpenPatchDataStorageByCurrentProcess(Kernel::HLERequestContext& ctx);
 
     FileSys::VirtualFile romfs;
+    u64 current_process_id = 0;
 };
 
 } // namespace Service::FileSystem

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp

@@ -10,6 +10,7 @@
 #include "core/core.h"
 #include "core/hle/service/nvdrv/devices/nvhost_as_gpu.h"
 #include "core/hle/service/nvdrv/devices/nvmap.h"
+#include "core/memory.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_base.h"
@@ -178,7 +179,7 @@ u32 nvhost_as_gpu::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& ou
     auto& gpu = system_instance.GPU();
     auto cpu_addr = gpu.MemoryManager().GpuToCpuAddress(params.offset);
     ASSERT(cpu_addr);
-    gpu.FlushAndInvalidateRegion(*cpu_addr, itr->second.size);
+    gpu.FlushAndInvalidateRegion(ToCacheAddr(Memory::GetPointer(*cpu_addr)), itr->second.size);
 
     params.offset = gpu.MemoryManager().UnmapBuffer(params.offset, itr->second.size);
 

src/core/loader/linker.cpp

@@ -1,147 +0,0 @@
-// Copyright 2018 yuzu emulator team
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <vector>
-
-#include "common/common_funcs.h"
-#include "common/logging/log.h"
-#include "common/swap.h"
-#include "core/loader/linker.h"
-#include "core/memory.h"
-
-namespace Loader {
-
-enum class RelocationType : u32 { ABS64 = 257, GLOB_DAT = 1025, JUMP_SLOT = 1026, RELATIVE = 1027 };
-
-enum DynamicType : u32 {
-    DT_NULL = 0,
-    DT_PLTRELSZ = 2,
-    DT_STRTAB = 5,
-    DT_SYMTAB = 6,
-    DT_RELA = 7,
-    DT_RELASZ = 8,
-    DT_STRSZ = 10,
-    DT_JMPREL = 23,
-};
-
-struct Elf64_Rela {
-    u64_le offset;
-    RelocationType type;
-    u32_le symbol;
-    s64_le addend;
-};
-static_assert(sizeof(Elf64_Rela) == 0x18, "Elf64_Rela has incorrect size.");
-
-struct Elf64_Dyn {
-    u64_le tag;
-    u64_le value;
-};
-static_assert(sizeof(Elf64_Dyn) == 0x10, "Elf64_Dyn has incorrect size.");
-
-struct Elf64_Sym {
-    u32_le name;
-    INSERT_PADDING_BYTES(0x2);
-    u16_le shndx;
-    u64_le value;
-    u64_le size;
-};
-static_assert(sizeof(Elf64_Sym) == 0x18, "Elf64_Sym has incorrect size.");
-
-void Linker::WriteRelocations(std::vector<u8>& program_image, const std::vector<Symbol>& symbols,
-                              u64 relocation_offset, u64 size, VAddr load_base) {
-    for (u64 i = 0; i < size; i += sizeof(Elf64_Rela)) {
-        Elf64_Rela rela;
-        std::memcpy(&rela, &program_image[relocation_offset + i], sizeof(Elf64_Rela));
-
-        const Symbol& symbol = symbols[rela.symbol];
-        switch (rela.type) {
-        case RelocationType::RELATIVE: {
-            const u64 value = load_base + rela.addend;
-            if (!symbol.name.empty()) {
-                exports[symbol.name] = value;
-            }
-            std::memcpy(&program_image[rela.offset], &value, sizeof(u64));
-            break;
-        }
-        case RelocationType::JUMP_SLOT:
-        case RelocationType::GLOB_DAT:
-            if (!symbol.value) {
-                imports[symbol.name] = {rela.offset + load_base, 0};
-            } else {
-                exports[symbol.name] = symbol.value;
-                std::memcpy(&program_image[rela.offset], &symbol.value, sizeof(u64));
-            }
-            break;
-        case RelocationType::ABS64:
-            if (!symbol.value) {
-                imports[symbol.name] = {rela.offset + load_base, rela.addend};
-            } else {
-                const u64 value = symbol.value + rela.addend;
-                exports[symbol.name] = value;
-                std::memcpy(&program_image[rela.offset], &value, sizeof(u64));
-            }
-            break;
-        default:
-            LOG_CRITICAL(Loader, "Unknown relocation type: {}", static_cast<int>(rela.type));
-            break;
-        }
-    }
-}
-
-void Linker::Relocate(std::vector<u8>& program_image, u32 dynamic_section_offset, VAddr load_base) {
-    std::map<u64, u64> dynamic;
-    while (dynamic_section_offset < program_image.size()) {
-        Elf64_Dyn dyn;
-        std::memcpy(&dyn, &program_image[dynamic_section_offset], sizeof(Elf64_Dyn));
-        dynamic_section_offset += sizeof(Elf64_Dyn);
-
-        if (dyn.tag == DT_NULL) {
-            break;
-        }
-        dynamic[dyn.tag] = dyn.value;
-    }
-
-    u64 offset = dynamic[DT_SYMTAB];
-    std::vector<Symbol> symbols;
-    while (offset < program_image.size()) {
-        Elf64_Sym sym;
-        std::memcpy(&sym, &program_image[offset], sizeof(Elf64_Sym));
-        offset += sizeof(Elf64_Sym);
-
-        if (sym.name >= dynamic[DT_STRSZ]) {
-            break;
-        }
-
-        std::string name = reinterpret_cast<char*>(&program_image[dynamic[DT_STRTAB] + sym.name]);
-        if (sym.value) {
-            exports[name] = load_base + sym.value;
-            symbols.emplace_back(std::move(name), load_base + sym.value);
-        } else {
-            symbols.emplace_back(std::move(name), 0);
-        }
-    }
-
-    if (dynamic.find(DT_RELA) != dynamic.end()) {
-        WriteRelocations(program_image, symbols, dynamic[DT_RELA], dynamic[DT_RELASZ], load_base);
-    }
-
-    if (dynamic.find(DT_JMPREL) != dynamic.end()) {
-        WriteRelocations(program_image, symbols, dynamic[DT_JMPREL], dynamic[DT_PLTRELSZ],
-                         load_base);
-    }
-}
-
-void Linker::ResolveImports() {
-    // Resolve imports
-    for (const auto& import : imports) {
-        const auto& search = exports.find(import.first);
-        if (search != exports.end()) {
-            Memory::Write64(import.second.ea, search->second + import.second.addend);
-        } else {
-            LOG_ERROR(Loader, "Unresolved import: {}", import.first);
-        }
-    }
-}
-
-} // namespace Loader

src/core/loader/linker.h

@@ -1,36 +0,0 @@
-// Copyright 2018 yuzu emulator team
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <map>
-#include <string>
-#include "common/common_types.h"
-
-namespace Loader {
-
-class Linker {
-protected:
-    struct Symbol {
-        Symbol(std::string&& name, u64 value) : name(std::move(name)), value(value) {}
-        std::string name;
-        u64 value;
-    };
-
-    struct Import {
-        VAddr ea;
-        s64 addend;
-    };
-
-    void WriteRelocations(std::vector<u8>& program_image, const std::vector<Symbol>& symbols,
-                          u64 relocation_offset, u64 size, VAddr load_base);
-    void Relocate(std::vector<u8>& program_image, u32 dynamic_section_offset, VAddr load_base);
-
-    void ResolveImports();
-
-    std::map<std::string, Import> imports;
-    std::map<std::string, VAddr> exports;
-};
-
-} // namespace Loader

src/core/loader/nro.h

@@ -4,10 +4,10 @@
 
 #pragma once
 
+#include <memory>
 #include <string>
 #include <vector>
 #include "common/common_types.h"
-#include "core/loader/linker.h"
 #include "core/loader/loader.h"
 
 namespace FileSys {
@@ -21,7 +21,7 @@ class Process;
 namespace Loader {
 
 /// Loads an NRO file
-class AppLoader_NRO final : public AppLoader, Linker {
+class AppLoader_NRO final : public AppLoader {
 public:
     explicit AppLoader_NRO(FileSys::VirtualFile file);
     ~AppLoader_NRO() override;

src/core/loader/nso.h

@@ -6,8 +6,8 @@
 
 #include <optional>
 #include "common/common_types.h"
+#include "common/swap.h"
 #include "core/file_sys/patch_manager.h"
-#include "core/loader/linker.h"
 #include "core/loader/loader.h"
 
 namespace Kernel {
@@ -26,7 +26,7 @@ struct NSOArgumentHeader {
 static_assert(sizeof(NSOArgumentHeader) == 0x20, "NSOArgumentHeader has incorrect size.");
 
 /// Loads an NSO file
-class AppLoader_NSO final : public AppLoader, Linker {
+class AppLoader_NSO final : public AppLoader {
 public:
     explicit AppLoader_NSO(FileSys::VirtualFile file);
 

src/core/memory.cpp

@@ -10,6 +10,7 @@
 #include "common/assert.h"
 #include "common/common_types.h"
 #include "common/logging/log.h"
+#include "common/page_table.h"
 #include "common/swap.h"
 #include "core/arm/arm_interface.h"
 #include "core/core.h"
@@ -18,13 +19,14 @@
 #include "core/hle/lock.h"
 #include "core/memory.h"
 #include "core/memory_setup.h"
+#include "video_core/gpu.h"
 #include "video_core/renderer_base.h"
 
 namespace Memory {
 
-static PageTable* current_page_table = nullptr;
+static Common::PageTable* current_page_table = nullptr;
 
-void SetCurrentPageTable(PageTable* page_table) {
+void SetCurrentPageTable(Common::PageTable* page_table) {
     current_page_table = page_table;
 
     auto& system = Core::System::GetInstance();
@@ -36,39 +38,20 @@ void SetCurrentPageTable(PageTable* page_table) {
     }
 }
 
-PageTable* GetCurrentPageTable() {
+Common::PageTable* GetCurrentPageTable() {
     return current_page_table;
 }
 
-PageTable::PageTable() = default;
-
-PageTable::PageTable(std::size_t address_space_width_in_bits) {
-    Resize(address_space_width_in_bits);
-}
-
-PageTable::~PageTable() = default;
-
-void PageTable::Resize(std::size_t address_space_width_in_bits) {
-    const std::size_t num_page_table_entries = 1ULL << (address_space_width_in_bits - PAGE_BITS);
-
-    pointers.resize(num_page_table_entries);
-    attributes.resize(num_page_table_entries);
-
-    // The default is a 39-bit address space, which causes an initial 1GB allocation size. If the
-    // vector size is subsequently decreased (via resize), the vector might not automatically
-    // actually reallocate/resize its underlying allocation, which wastes up to ~800 MB for
-    // 36-bit titles. Call shrink_to_fit to reduce capacity to what's actually in use.
-
-    pointers.shrink_to_fit();
-    attributes.shrink_to_fit();
-}
-
-static void MapPages(PageTable& page_table, VAddr base, u64 size, u8* memory, PageType type) {
+static void MapPages(Common::PageTable& page_table, VAddr base, u64 size, u8* memory,
+                     Common::PageType type) {
     LOG_DEBUG(HW_Memory, "Mapping {} onto {:016X}-{:016X}", fmt::ptr(memory), base * PAGE_SIZE,
               (base + size) * PAGE_SIZE);
 
-    RasterizerFlushVirtualRegion(base << PAGE_BITS, size * PAGE_SIZE,
-                                 FlushMode::FlushAndInvalidate);
+    // During boot, current_page_table might not be set yet, in which case we need not flush
+    if (current_page_table) {
+        Core::System::GetInstance().GPU().FlushAndInvalidateRegion(base << PAGE_BITS,
+                                                                   size * PAGE_SIZE);
+    }
 
     VAddr end = base + size;
     ASSERT_MSG(end <= page_table.pointers.size(), "out of range mapping at {:016X}",
@@ -88,41 +71,47 @@ static void MapPages(PageTable& page_table, VAddr base, u64 size, u8* memory, Pa
     }
 }
 
-void MapMemoryRegion(PageTable& page_table, VAddr base, u64 size, u8* target) {
+void MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size, u8* target) {
     ASSERT_MSG((size & PAGE_MASK) == 0, "non-page aligned size: {:016X}", size);
     ASSERT_MSG((base & PAGE_MASK) == 0, "non-page aligned base: {:016X}", base);
-    MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, target, PageType::Memory);
+    MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, target, Common::PageType::Memory);
 }
 
-void MapIoRegion(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer mmio_handler) {
+void MapIoRegion(Common::PageTable& page_table, VAddr base, u64 size,
+                 Common::MemoryHookPointer mmio_handler) {
     ASSERT_MSG((size & PAGE_MASK) == 0, "non-page aligned size: {:016X}", size);
     ASSERT_MSG((base & PAGE_MASK) == 0, "non-page aligned base: {:016X}", base);
-    MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, nullptr, PageType::Special);
+    MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, nullptr, Common::PageType::Special);
 
     auto interval = boost::icl::discrete_interval<VAddr>::closed(base, base + size - 1);
-    SpecialRegion region{SpecialRegion::Type::IODevice, std::move(mmio_handler)};
-    page_table.special_regions.add(std::make_pair(interval, std::set<SpecialRegion>{region}));
+    Common::SpecialRegion region{Common::SpecialRegion::Type::IODevice, std::move(mmio_handler)};
+    page_table.special_regions.add(
+        std::make_pair(interval, std::set<Common::SpecialRegion>{region}));
 }
 
-void UnmapRegion(PageTable& page_table, VAddr base, u64 size) {
+void UnmapRegion(Common::PageTable& page_table, VAddr base, u64 size) {
     ASSERT_MSG((size & PAGE_MASK) == 0, "non-page aligned size: {:016X}", size);
     ASSERT_MSG((base & PAGE_MASK) == 0, "non-page aligned base: {:016X}", base);
-    MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, nullptr, PageType::Unmapped);
+    MapPages(page_table, base / PAGE_SIZE, size / PAGE_SIZE, nullptr, Common::PageType::Unmapped);
 
     auto interval = boost::icl::discrete_interval<VAddr>::closed(base, base + size - 1);
     page_table.special_regions.erase(interval);
 }
 
-void AddDebugHook(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer hook) {
+void AddDebugHook(Common::PageTable& page_table, VAddr base, u64 size,
+                  Common::MemoryHookPointer hook) {
     auto interval = boost::icl::discrete_interval<VAddr>::closed(base, base + size - 1);
-    SpecialRegion region{SpecialRegion::Type::DebugHook, std::move(hook)};
-    page_table.special_regions.add(std::make_pair(interval, std::set<SpecialRegion>{region}));
+    Common::SpecialRegion region{Common::SpecialRegion::Type::DebugHook, std::move(hook)};
+    page_table.special_regions.add(
+        std::make_pair(interval, std::set<Common::SpecialRegion>{region}));
 }
 
-void RemoveDebugHook(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer hook) {
+void RemoveDebugHook(Common::PageTable& page_table, VAddr base, u64 size,
+                     Common::MemoryHookPointer hook) {
     auto interval = boost::icl::discrete_interval<VAddr>::closed(base, base + size - 1);
-    SpecialRegion region{SpecialRegion::Type::DebugHook, std::move(hook)};
-    page_table.special_regions.subtract(std::make_pair(interval, std::set<SpecialRegion>{region}));
+    Common::SpecialRegion region{Common::SpecialRegion::Type::DebugHook, std::move(hook)};
+    page_table.special_regions.subtract(
+        std::make_pair(interval, std::set<Common::SpecialRegion>{region}));
 }
 
 /**
@@ -171,19 +160,19 @@ T Read(const VAddr vaddr) {
         return value;
     }
 
-    PageType type = current_page_table->attributes[vaddr >> PAGE_BITS];
+    Common::PageType type = current_page_table->attributes[vaddr >> PAGE_BITS];
     switch (type) {
-    case PageType::Unmapped:
+    case Common::PageType::Unmapped:
         LOG_ERROR(HW_Memory, "Unmapped Read{} @ 0x{:08X}", sizeof(T) * 8, vaddr);
         return 0;
-    case PageType::Memory:
+    case Common::PageType::Memory:
         ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr);
         break;
-    case PageType::RasterizerCachedMemory: {
-        RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::Flush);
-
+    case Common::PageType::RasterizerCachedMemory: {
+        auto host_ptr{GetPointerFromVMA(vaddr)};
+        Core::System::GetInstance().GPU().FlushRegion(ToCacheAddr(host_ptr), sizeof(T));
         T value;
-        std::memcpy(&value, GetPointerFromVMA(vaddr), sizeof(T));
+        std::memcpy(&value, host_ptr, sizeof(T));
         return value;
     }
     default:
@@ -201,18 +190,19 @@ void Write(const VAddr vaddr, const T data) {
         return;
     }
 
-    PageType type = current_page_table->attributes[vaddr >> PAGE_BITS];
+    Common::PageType type = current_page_table->attributes[vaddr >> PAGE_BITS];
     switch (type) {
-    case PageType::Unmapped:
+    case Common::PageType::Unmapped:
         LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8,
                   static_cast<u32>(data), vaddr);
         return;
-    case PageType::Memory:
+    case Common::PageType::Memory:
         ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr);
         break;
-    case PageType::RasterizerCachedMemory: {
-        RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::Invalidate);
-        std::memcpy(GetPointerFromVMA(vaddr), &data, sizeof(T));
+    case Common::PageType::RasterizerCachedMemory: {
+        auto host_ptr{GetPointerFromVMA(vaddr)};
+        Core::System::GetInstance().GPU().InvalidateRegion(ToCacheAddr(host_ptr), sizeof(T));
+        std::memcpy(host_ptr, &data, sizeof(T));
         break;
     }
     default:
@@ -227,10 +217,10 @@ bool IsValidVirtualAddress(const Kernel::Process& process, const VAddr vaddr) {
     if (page_pointer)
         return true;
 
-    if (page_table.attributes[vaddr >> PAGE_BITS] == PageType::RasterizerCachedMemory)
+    if (page_table.attributes[vaddr >> PAGE_BITS] == Common::PageType::RasterizerCachedMemory)
         return true;
 
-    if (page_table.attributes[vaddr >> PAGE_BITS] != PageType::Special)
+    if (page_table.attributes[vaddr >> PAGE_BITS] != Common::PageType::Special)
         return false;
 
     return false;
@@ -250,7 +240,8 @@ u8* GetPointer(const VAddr vaddr) {
         return page_pointer + (vaddr & PAGE_MASK);
     }
 
-    if (current_page_table->attributes[vaddr >> PAGE_BITS] == PageType::RasterizerCachedMemory) {
+    if (current_page_table->attributes[vaddr >> PAGE_BITS] ==
+        Common::PageType::RasterizerCachedMemory) {
         return GetPointerFromVMA(vaddr);
     }
 
@@ -284,20 +275,20 @@ void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) {
 
     u64 num_pages = ((vaddr + size - 1) >> PAGE_BITS) - (vaddr >> PAGE_BITS) + 1;
     for (unsigned i = 0; i < num_pages; ++i, vaddr += PAGE_SIZE) {
-        PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
+        Common::PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
 
         if (cached) {
             // Switch page type to cached if now cached
             switch (page_type) {
-            case PageType::Unmapped:
+            case Common::PageType::Unmapped:
                 // It is not necessary for a process to have this region mapped into its address
                 // space, for example, a system module need not have a VRAM mapping.
                 break;
-            case PageType::Memory:
-                page_type = PageType::RasterizerCachedMemory;
+            case Common::PageType::Memory:
+                page_type = Common::PageType::RasterizerCachedMemory;
                 current_page_table->pointers[vaddr >> PAGE_BITS] = nullptr;
                 break;
-            case PageType::RasterizerCachedMemory:
+            case Common::PageType::RasterizerCachedMemory:
                 // There can be more than one GPU region mapped per CPU region, so it's common that
                 // this area is already marked as cached.
                 break;
@@ -307,23 +298,23 @@ void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) {
         } else {
             // Switch page type to uncached if now uncached
             switch (page_type) {
-            case PageType::Unmapped:
+            case Common::PageType::Unmapped:
                 // It is not necessary for a process to have this region mapped into its address
                 // space, for example, a system module need not have a VRAM mapping.
                 break;
-            case PageType::Memory:
+            case Common::PageType::Memory:
                 // There can be more than one GPU region mapped per CPU region, so it's common that
                 // this area is already unmarked as cached.
                 break;
-            case PageType::RasterizerCachedMemory: {
+            case Common::PageType::RasterizerCachedMemory: {
                 u8* pointer = GetPointerFromVMA(vaddr & ~PAGE_MASK);
                 if (pointer == nullptr) {
                     // It's possible that this function has been called while updating the pagetable
                     // after unmapping a VMA. In that case the underlying VMA will no longer exist,
                     // and we should just leave the pagetable entry blank.
-                    page_type = PageType::Unmapped;
+                    page_type = Common::PageType::Unmapped;
                 } else {
-                    page_type = PageType::Memory;
+                    page_type = Common::PageType::Memory;
                     current_page_table->pointers[vaddr >> PAGE_BITS] = pointer;
                 }
                 break;
@@ -335,47 +326,6 @@ void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached) {
     }
 }
 
-void RasterizerFlushVirtualRegion(VAddr start, u64 size, FlushMode mode) {
-    auto& system_instance = Core::System::GetInstance();
-
-    // Since pages are unmapped on shutdown after video core is shutdown, the renderer may be
-    // null here
-    if (!system_instance.IsPoweredOn()) {
-        return;
-    }
-
-    const VAddr end = start + size;
-
-    const auto CheckRegion = [&](VAddr region_start, VAddr region_end) {
-        if (start >= region_end || end <= region_start) {
-            // No overlap with region
-            return;
-        }
-
-        const VAddr overlap_start = std::max(start, region_start);
-        const VAddr overlap_end = std::min(end, region_end);
-        const VAddr overlap_size = overlap_end - overlap_start;
-
-        auto& gpu = system_instance.GPU();
-        switch (mode) {
-        case FlushMode::Flush:
-            gpu.FlushRegion(overlap_start, overlap_size);
-            break;
-        case FlushMode::Invalidate:
-            gpu.InvalidateRegion(overlap_start, overlap_size);
-            break;
-        case FlushMode::FlushAndInvalidate:
-            gpu.FlushAndInvalidateRegion(overlap_start, overlap_size);
-            break;
-        }
-    };
-
-    const auto& vm_manager = Core::CurrentProcess()->VMManager();
-
-    CheckRegion(vm_manager.GetCodeRegionBaseAddress(), vm_manager.GetCodeRegionEndAddress());
-    CheckRegion(vm_manager.GetHeapRegionBaseAddress(), vm_manager.GetHeapRegionEndAddress());
-}
-
 u8 Read8(const VAddr addr) {
     return Read<u8>(addr);
 }
@@ -406,24 +356,24 @@ void ReadBlock(const Kernel::Process& process, const VAddr src_addr, void* dest_
         const VAddr current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
 
         switch (page_table.attributes[page_index]) {
-        case PageType::Unmapped: {
+        case Common::PageType::Unmapped: {
             LOG_ERROR(HW_Memory,
                       "Unmapped ReadBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
                       current_vaddr, src_addr, size);
             std::memset(dest_buffer, 0, copy_amount);
             break;
         }
-        case PageType::Memory: {
+        case Common::PageType::Memory: {
             DEBUG_ASSERT(page_table.pointers[page_index]);
 
             const u8* src_ptr = page_table.pointers[page_index] + page_offset;
             std::memcpy(dest_buffer, src_ptr, copy_amount);
             break;
         }
-        case PageType::RasterizerCachedMemory: {
-            RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::Flush);
-            std::memcpy(dest_buffer, GetPointerFromVMA(process, current_vaddr), copy_amount);
+        case Common::PageType::RasterizerCachedMemory: {
+            const auto& host_ptr{GetPointerFromVMA(process, current_vaddr)};
+            Core::System::GetInstance().GPU().FlushRegion(ToCacheAddr(host_ptr), copy_amount);
+            std::memcpy(dest_buffer, host_ptr, copy_amount);
             break;
         }
         default:
@@ -470,23 +420,23 @@ void WriteBlock(const Kernel::Process& process, const VAddr dest_addr, const voi
         const VAddr current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
 
         switch (page_table.attributes[page_index]) {
-        case PageType::Unmapped: {
+        case Common::PageType::Unmapped: {
             LOG_ERROR(HW_Memory,
                       "Unmapped WriteBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
                       current_vaddr, dest_addr, size);
             break;
         }
-        case PageType::Memory: {
+        case Common::PageType::Memory: {
             DEBUG_ASSERT(page_table.pointers[page_index]);
 
             u8* dest_ptr = page_table.pointers[page_index] + page_offset;
             std::memcpy(dest_ptr, src_buffer, copy_amount);
             break;
         }
-        case PageType::RasterizerCachedMemory: {
-            RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::Invalidate);
-            std::memcpy(GetPointerFromVMA(process, current_vaddr), src_buffer, copy_amount);
+        case Common::PageType::RasterizerCachedMemory: {
+            const auto& host_ptr{GetPointerFromVMA(process, current_vaddr)};
+            Core::System::GetInstance().GPU().InvalidateRegion(ToCacheAddr(host_ptr), copy_amount);
+            std::memcpy(host_ptr, src_buffer, copy_amount);
             break;
         }
         default:
@@ -516,23 +466,23 @@ void ZeroBlock(const Kernel::Process& process, const VAddr dest_addr, const std:
         const VAddr current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
 
         switch (page_table.attributes[page_index]) {
-        case PageType::Unmapped: {
+        case Common::PageType::Unmapped: {
             LOG_ERROR(HW_Memory,
                       "Unmapped ZeroBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
                       current_vaddr, dest_addr, size);
             break;
         }
-        case PageType::Memory: {
+        case Common::PageType::Memory: {
             DEBUG_ASSERT(page_table.pointers[page_index]);
 
             u8* dest_ptr = page_table.pointers[page_index] + page_offset;
             std::memset(dest_ptr, 0, copy_amount);
             break;
         }
-        case PageType::RasterizerCachedMemory: {
-            RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::Invalidate);
-            std::memset(GetPointerFromVMA(process, current_vaddr), 0, copy_amount);
+        case Common::PageType::RasterizerCachedMemory: {
+            const auto& host_ptr{GetPointerFromVMA(process, current_vaddr)};
+            Core::System::GetInstance().GPU().InvalidateRegion(ToCacheAddr(host_ptr), copy_amount);
+            std::memset(host_ptr, 0, copy_amount);
             break;
         }
         default:
@@ -558,23 +508,23 @@ void CopyBlock(const Kernel::Process& process, VAddr dest_addr, VAddr src_addr,
         const VAddr current_vaddr = static_cast<VAddr>((page_index << PAGE_BITS) + page_offset);
 
         switch (page_table.attributes[page_index]) {
-        case PageType::Unmapped: {
+        case Common::PageType::Unmapped: {
             LOG_ERROR(HW_Memory,
                       "Unmapped CopyBlock @ 0x{:016X} (start address = 0x{:016X}, size = {})",
                       current_vaddr, src_addr, size);
             ZeroBlock(process, dest_addr, copy_amount);
             break;
         }
-        case PageType::Memory: {
+        case Common::PageType::Memory: {
             DEBUG_ASSERT(page_table.pointers[page_index]);
             const u8* src_ptr = page_table.pointers[page_index] + page_offset;
             WriteBlock(process, dest_addr, src_ptr, copy_amount);
             break;
         }
-        case PageType::RasterizerCachedMemory: {
-            RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::Flush);
-            WriteBlock(process, dest_addr, GetPointerFromVMA(process, current_vaddr), copy_amount);
+        case Common::PageType::RasterizerCachedMemory: {
+            const auto& host_ptr{GetPointerFromVMA(process, current_vaddr)};
+            Core::System::GetInstance().GPU().FlushRegion(ToCacheAddr(host_ptr), copy_amount);
+            WriteBlock(process, dest_addr, host_ptr, copy_amount);
             break;
         }
         default:

src/core/memory.h

@@ -10,7 +10,10 @@
 #include <vector>
 #include <boost/icl/interval_map.hpp>
 #include "common/common_types.h"
-#include "core/memory_hook.h"
+
+namespace Common {
+struct PageTable;
+}
 
 namespace Kernel {
 class Process;
@@ -26,71 +29,6 @@ constexpr std::size_t PAGE_BITS = 12;
 constexpr u64 PAGE_SIZE = 1ULL << PAGE_BITS;
 constexpr u64 PAGE_MASK = PAGE_SIZE - 1;
 
-enum class PageType : u8 {
-    /// Page is unmapped and should cause an access error.
-    Unmapped,
-    /// Page is mapped to regular memory. This is the only type you can get pointers to.
-    Memory,
-    /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and
-    /// invalidation
-    RasterizerCachedMemory,
-    /// Page is mapped to a I/O region. Writing and reading to this page is handled by functions.
-    Special,
-};
-
-struct SpecialRegion {
-    enum class Type {
-        DebugHook,
-        IODevice,
-    } type;
-
-    MemoryHookPointer handler;
-
-    bool operator<(const SpecialRegion& other) const {
-        return std::tie(type, handler) < std::tie(other.type, other.handler);
-    }
-
-    bool operator==(const SpecialRegion& other) const {
-        return std::tie(type, handler) == std::tie(other.type, other.handler);
-    }
-};
-
-/**
- * A (reasonably) fast way of allowing switchable and remappable process address spaces. It loosely
- * mimics the way a real CPU page table works.
- */
-struct PageTable {
-    explicit PageTable();
-    explicit PageTable(std::size_t address_space_width_in_bits);
-    ~PageTable();
-
-    /**
-     * Resizes the page table to be able to accomodate enough pages within
-     * a given address space.
-     *
-     * @param address_space_width_in_bits The address size width in bits.
-     */
-    void Resize(std::size_t address_space_width_in_bits);
-
-    /**
-     * Vector of memory pointers backing each page. An entry can only be non-null if the
-     * corresponding entry in the `attributes` vector is of type `Memory`.
-     */
-    std::vector<u8*> pointers;
-
-    /**
-     * Contains MMIO handlers that back memory regions whose entries in the `attribute` vector is
-     * of type `Special`.
-     */
-    boost::icl::interval_map<VAddr, std::set<SpecialRegion>> special_regions;
-
-    /**
-     * Vector of fine grained page attributes. If it is set to any value other than `Memory`, then
-     * the corresponding entry in `pointers` MUST be set to null.
-     */
-    std::vector<PageType> attributes;
-};
-
 /// Virtual user-space memory regions
 enum : VAddr {
     /// Read-only page containing kernel and system configuration values.
@@ -116,8 +54,8 @@ enum : VAddr {
 };
 
 /// Currently active page table
-void SetCurrentPageTable(PageTable* page_table);
-PageTable* GetCurrentPageTable();
+void SetCurrentPageTable(Common::PageTable* page_table);
+Common::PageTable* GetCurrentPageTable();
 
 /// Determines if the given VAddr is valid for the specified process.
 bool IsValidVirtualAddress(const Kernel::Process& process, VAddr vaddr);
@@ -161,10 +99,4 @@ enum class FlushMode {
  */
 void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached);
 
-/**
- * Flushes and invalidates any externally cached rasterizer resources touching the given virtual
- * address region.
- */
-void RasterizerFlushVirtualRegion(VAddr start, u64 size, FlushMode mode);
-
 } // namespace Memory

src/core/memory_setup.h

@@ -5,7 +5,11 @@
 #pragma once
 
 #include "common/common_types.h"
-#include "core/memory_hook.h"
+#include "common/memory_hook.h"
+
+namespace Common {
+struct PageTable;
+}
 
 namespace Memory {
 
@@ -17,7 +21,7 @@ namespace Memory {
  * @param size The amount of bytes to map. Must be page-aligned.
  * @param target Buffer with the memory backing the mapping. Must be of length at least `size`.
  */
-void MapMemoryRegion(PageTable& page_table, VAddr base, u64 size, u8* target);
+void MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size, u8* target);
 
 /**
  * Maps a region of the emulated process address space as a IO region.
@@ -26,11 +30,14 @@ void MapMemoryRegion(PageTable& page_table, VAddr base, u64 size, u8* target);
  * @param size The amount of bytes to map. Must be page-aligned.
  * @param mmio_handler The handler that backs the mapping.
  */
-void MapIoRegion(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer mmio_handler);
+void MapIoRegion(Common::PageTable& page_table, VAddr base, u64 size,
+                 Common::MemoryHookPointer mmio_handler);
 
-void UnmapRegion(PageTable& page_table, VAddr base, u64 size);
+void UnmapRegion(Common::PageTable& page_table, VAddr base, u64 size);
 
-void AddDebugHook(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer hook);
-void RemoveDebugHook(PageTable& page_table, VAddr base, u64 size, MemoryHookPointer hook);
+void AddDebugHook(Common::PageTable& page_table, VAddr base, u64 size,
+                  Common::MemoryHookPointer hook);
+void RemoveDebugHook(Common::PageTable& page_table, VAddr base, u64 size,
+                     Common::MemoryHookPointer hook);
 
 } // namespace Memory

src/input_common/sdl/sdl_impl.cpp

@@ -475,12 +475,11 @@ SDLState::SDLState() {
 
     initialized = true;
     if (start_thread) {
-        poll_thread = std::thread([&] {
+        poll_thread = std::thread([this] {
             using namespace std::chrono_literals;
-            SDL_Event event;
             while (initialized) {
                 SDL_PumpEvents();
-                std::this_thread::sleep_for(std::chrono::duration(10ms));
+                std::this_thread::sleep_for(10ms);
             }
         });
     }

src/tests/core/arm/arm_test_common.cpp

@@ -4,6 +4,7 @@
 
 #include <algorithm>
 
+#include "common/page_table.h"
 #include "core/core.h"
 #include "core/hle/kernel/process.h"
 #include "core/memory.h"
@@ -22,7 +23,7 @@ TestEnvironment::TestEnvironment(bool mutable_memory_)
     std::fill(page_table->pointers.begin(), page_table->pointers.end(), nullptr);
     page_table->special_regions.clear();
     std::fill(page_table->attributes.begin(), page_table->attributes.end(),
-              Memory::PageType::Unmapped);
+              Common::PageType::Unmapped);
 
     Memory::MapIoRegion(*page_table, 0x00000000, 0x80000000, test_memory);
     Memory::MapIoRegion(*page_table, 0x80000000, 0x80000000, test_memory);

src/tests/core/arm/arm_test_common.h

@@ -9,10 +9,10 @@
 #include <vector>
 
 #include "common/common_types.h"
+#include "common/memory_hook.h"
 #include "core/hle/kernel/kernel.h"
-#include "core/memory_hook.h"
 
-namespace Memory {
+namespace Common {
 struct PageTable;
 }
 
@@ -58,7 +58,7 @@ public:
 
 private:
     friend struct TestMemory;
-    struct TestMemory final : Memory::MemoryHook {
+    struct TestMemory final : Common::MemoryHook {
         explicit TestMemory(TestEnvironment* env_) : env(env_) {}
         TestEnvironment* env;
 
@@ -86,7 +86,7 @@ private:
     bool mutable_memory;
     std::shared_ptr<TestMemory> test_memory;
     std::vector<WriteRecord> write_records;
-    Memory::PageTable* page_table = nullptr;
+    Common::PageTable* page_table = nullptr;
     Kernel::KernelCore kernel;
 };
 

src/video_core/CMakeLists.txt

@@ -123,6 +123,8 @@ if (ENABLE_VULKAN)
         renderer_vulkan/vk_memory_manager.h
         renderer_vulkan/vk_resource_manager.cpp
         renderer_vulkan/vk_resource_manager.h
+        renderer_vulkan/vk_sampler_cache.cpp
+        renderer_vulkan/vk_sampler_cache.h
         renderer_vulkan/vk_scheduler.cpp
         renderer_vulkan/vk_scheduler.h
         renderer_vulkan/vk_stream_buffer.cpp

src/video_core/dma_pusher.cpp

@@ -55,12 +55,9 @@ bool DmaPusher::Step() {
     }
 
     // Push buffer non-empty, read a word
-    const auto address = gpu.MemoryManager().GpuToCpuAddress(dma_get);
-    ASSERT_MSG(address, "Invalid GPU address");
-
     command_headers.resize(command_list_header.size);
-
-    Memory::ReadBlock(*address, command_headers.data(), command_list_header.size * sizeof(u32));
+    gpu.MemoryManager().ReadBlock(dma_get, command_headers.data(),
+                                  command_list_header.size * sizeof(u32));
 
     for (const CommandHeader& command_header : command_headers) {
 

src/video_core/engines/kepler_memory.cpp

@@ -9,6 +9,7 @@
 #include "video_core/engines/kepler_memory.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_base.h"
 
 namespace Tegra::Engines {
 
@@ -40,17 +41,13 @@ void KeplerMemory::ProcessData(u32 data) {
     ASSERT_MSG(regs.exec.linear, "Non-linear uploads are not supported");
     ASSERT(regs.dest.x == 0 && regs.dest.y == 0 && regs.dest.z == 0);
 
-    const GPUVAddr address = regs.dest.Address();
-    const auto dest_address =
-        memory_manager.GpuToCpuAddress(address + state.write_offset * sizeof(u32));
-    ASSERT_MSG(dest_address, "Invalid GPU address");
-
     // We have to invalidate the destination region to evict any outdated surfaces from the cache.
-    // We do this before actually writing the new data because the destination address might contain
-    // a dirty surface that will have to be written back to memory.
-    Core::System::GetInstance().GPU().InvalidateRegion(*dest_address, sizeof(u32));
+    // We do this before actually writing the new data because the destination address might
+    // contain a dirty surface that will have to be written back to memory.
+    const GPUVAddr address{regs.dest.Address() + state.write_offset * sizeof(u32)};
+    rasterizer.InvalidateRegion(ToCacheAddr(memory_manager.GetPointer(address)), sizeof(u32));
+    memory_manager.Write32(address, data);
 
-    Memory::Write32(*dest_address, data);
     system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();
 
     state.write_offset++;

src/video_core/engines/maxwell_3d.cpp

@@ -270,11 +270,9 @@ void Maxwell3D::ProcessMacroBind(u32 data) {
 }
 
 void Maxwell3D::ProcessQueryGet() {
-    GPUVAddr sequence_address = regs.query.QueryAddress();
+    const GPUVAddr sequence_address{regs.query.QueryAddress()};
     // Since the sequence address is given as a GPU VAddr, we have to convert it to an application
     // VAddr before writing.
-    const auto address = memory_manager.GpuToCpuAddress(sequence_address);
-    ASSERT_MSG(address, "Invalid GPU address");
 
     // TODO(Subv): Support the other query units.
     ASSERT_MSG(regs.query.query_get.unit == Regs::QueryUnit::Crop,
@@ -309,7 +307,7 @@ void Maxwell3D::ProcessQueryGet() {
             // Write the current query sequence to the sequence address.
             // TODO(Subv): Find out what happens if you use a long query type but mark it as a short
             // query.
-            Memory::Write32(*address, sequence);
+            memory_manager.Write32(sequence_address, sequence);
         } else {
             // Write the 128-bit result structure in long mode. Note: We emulate an infinitely fast
             // GPU, this command may actually take a while to complete in real hardware due to GPU
@@ -318,7 +316,7 @@ void Maxwell3D::ProcessQueryGet() {
             query_result.value = result;
             // TODO(Subv): Generate a real GPU timestamp and write it here instead of CoreTiming
             query_result.timestamp = system.CoreTiming().GetTicks();
-            Memory::WriteBlock(*address, &query_result, sizeof(query_result));
+            memory_manager.WriteBlock(sequence_address, &query_result, sizeof(query_result));
         }
         dirty_flags.OnMemoryWrite();
         break;
@@ -393,10 +391,12 @@ void Maxwell3D::ProcessCBData(u32 value) {
     // Don't allow writing past the end of the buffer.
     ASSERT(regs.const_buffer.cb_pos + sizeof(u32) <= regs.const_buffer.cb_size);
 
-    const auto address = memory_manager.GpuToCpuAddress(buffer_address + regs.const_buffer.cb_pos);
-    ASSERT_MSG(address, "Invalid GPU address");
+    const GPUVAddr address{buffer_address + regs.const_buffer.cb_pos};
+
+    u8* ptr{memory_manager.GetPointer(address)};
+    rasterizer.InvalidateRegion(ToCacheAddr(ptr), sizeof(u32));
+    memory_manager.Write32(address, value);
 
-    Memory::Write32(*address, value);
     dirty_flags.OnMemoryWrite();
 
     // Increment the current buffer position.
@@ -404,14 +404,10 @@ void Maxwell3D::ProcessCBData(u32 value) {
 }
 
 Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
-    const GPUVAddr tic_base_address = regs.tic.TICAddress();
-
-    const GPUVAddr tic_address_gpu = tic_base_address + tic_index * sizeof(Texture::TICEntry);
-    const auto tic_address_cpu = memory_manager.GpuToCpuAddress(tic_address_gpu);
-    ASSERT_MSG(tic_address_cpu, "Invalid GPU address");
+    const GPUVAddr tic_address_gpu{regs.tic.TICAddress() + tic_index * sizeof(Texture::TICEntry)};
 
     Texture::TICEntry tic_entry;
-    Memory::ReadBlock(*tic_address_cpu, &tic_entry, sizeof(Texture::TICEntry));
+    memory_manager.ReadBlock(tic_address_gpu, &tic_entry, sizeof(Texture::TICEntry));
 
     ASSERT_MSG(tic_entry.header_version == Texture::TICHeaderVersion::BlockLinear ||
                    tic_entry.header_version == Texture::TICHeaderVersion::Pitch,
@@ -429,14 +425,10 @@ Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
 }
 
 Texture::TSCEntry Maxwell3D::GetTSCEntry(u32 tsc_index) const {
-    const GPUVAddr tsc_base_address = regs.tsc.TSCAddress();
-
-    const GPUVAddr tsc_address_gpu = tsc_base_address + tsc_index * sizeof(Texture::TSCEntry);
-    const auto tsc_address_cpu = memory_manager.GpuToCpuAddress(tsc_address_gpu);
-    ASSERT_MSG(tsc_address_cpu, "Invalid GPU address");
+    const GPUVAddr tsc_address_gpu{regs.tsc.TSCAddress() + tsc_index * sizeof(Texture::TSCEntry)};
 
     Texture::TSCEntry tsc_entry;
-    Memory::ReadBlock(*tsc_address_cpu, &tsc_entry, sizeof(Texture::TSCEntry));
+    memory_manager.ReadBlock(tsc_address_gpu, &tsc_entry, sizeof(Texture::TSCEntry));
     return tsc_entry;
 }
 
@@ -455,10 +447,7 @@ std::vector<Texture::FullTextureInfo> Maxwell3D::GetStageTextures(Regs::ShaderSt
     for (GPUVAddr current_texture = tex_info_buffer.address + TextureInfoOffset;
          current_texture < tex_info_buffer_end; current_texture += sizeof(Texture::TextureHandle)) {
 
-        const auto address = memory_manager.GpuToCpuAddress(current_texture);
-        ASSERT_MSG(address, "Invalid GPU address");
-
-        const Texture::TextureHandle tex_handle{Memory::Read32(*address)};
+        const Texture::TextureHandle tex_handle{memory_manager.Read32(current_texture)};
 
         Texture::FullTextureInfo tex_info{};
         // TODO(Subv): Use the shader to determine which textures are actually accessed.
@@ -493,10 +482,7 @@ Texture::FullTextureInfo Maxwell3D::GetStageTexture(Regs::ShaderStage stage,
 
     ASSERT(tex_info_address < tex_info_buffer.address + tex_info_buffer.size);
 
-    const auto tex_address_cpu = memory_manager.GpuToCpuAddress(tex_info_address);
-    ASSERT_MSG(tex_address_cpu, "Invalid GPU address");
-
-    const Texture::TextureHandle tex_handle{Memory::Read32(*tex_address_cpu)};
+    const Texture::TextureHandle tex_handle{memory_manager.Read32(tex_info_address)};
 
     Texture::FullTextureInfo tex_info{};
     tex_info.index = static_cast<u32>(offset);

src/video_core/engines/maxwell_dma.cpp

@@ -9,6 +9,7 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/maxwell_dma.h"
 #include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_base.h"
 #include "video_core/textures/decoders.h"
 
 namespace Tegra::Engines {
@@ -42,11 +43,6 @@ void MaxwellDMA::HandleCopy() {
     const GPUVAddr source = regs.src_address.Address();
     const GPUVAddr dest = regs.dst_address.Address();
 
-    const auto source_cpu = memory_manager.GpuToCpuAddress(source);
-    const auto dest_cpu = memory_manager.GpuToCpuAddress(dest);
-    ASSERT_MSG(source_cpu, "Invalid source GPU address");
-    ASSERT_MSG(dest_cpu, "Invalid destination GPU address");
-
     // TODO(Subv): Perform more research and implement all features of this engine.
     ASSERT(regs.exec.enable_swizzle == 0);
     ASSERT(regs.exec.query_mode == Regs::QueryMode::None);
@@ -69,7 +65,7 @@ void MaxwellDMA::HandleCopy() {
         // buffer of length `x_count`, otherwise we copy a 2D image of dimensions (x_count,
         // y_count).
         if (!regs.exec.enable_2d) {
-            Memory::CopyBlock(*dest_cpu, *source_cpu, regs.x_count);
+            memory_manager.CopyBlock(dest, source, regs.x_count);
             return;
         }
 
@@ -78,9 +74,9 @@ void MaxwellDMA::HandleCopy() {
         // rectangle. There is no need to manually flush/invalidate the regions because
         // CopyBlock does that for us.
         for (u32 line = 0; line < regs.y_count; ++line) {
-            const VAddr source_line = *source_cpu + line * regs.src_pitch;
-            const VAddr dest_line = *dest_cpu + line * regs.dst_pitch;
-            Memory::CopyBlock(dest_line, source_line, regs.x_count);
+            const GPUVAddr source_line = source + line * regs.src_pitch;
+            const GPUVAddr dest_line = dest + line * regs.dst_pitch;
+            memory_manager.CopyBlock(dest_line, source_line, regs.x_count);
         }
         return;
     }
@@ -89,15 +85,18 @@ void MaxwellDMA::HandleCopy() {
 
     const std::size_t copy_size = regs.x_count * regs.y_count;
 
+    auto source_ptr{memory_manager.GetPointer(source)};
+    auto dst_ptr{memory_manager.GetPointer(dest)};
+
     const auto FlushAndInvalidate = [&](u32 src_size, u64 dst_size) {
         // TODO(Subv): For now, manually flush the regions until we implement GPU-accelerated
         // copying.
-        Core::System::GetInstance().GPU().FlushRegion(*source_cpu, src_size);
+        rasterizer.FlushRegion(ToCacheAddr(source_ptr), src_size);
 
         // We have to invalidate the destination region to evict any outdated surfaces from the
         // cache. We do this before actually writing the new data because the destination address
         // might contain a dirty surface that will have to be written back to memory.
-        Core::System::GetInstance().GPU().InvalidateRegion(*dest_cpu, dst_size);
+        rasterizer.InvalidateRegion(ToCacheAddr(dst_ptr), dst_size);
     };
 
     if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
@@ -110,8 +109,8 @@ void MaxwellDMA::HandleCopy() {
                            copy_size * src_bytes_per_pixel);
 
         Texture::UnswizzleSubrect(regs.x_count, regs.y_count, regs.dst_pitch,
-                                  regs.src_params.size_x, src_bytes_per_pixel, *source_cpu,
-                                  *dest_cpu, regs.src_params.BlockHeight(), regs.src_params.pos_x,
+                                  regs.src_params.size_x, src_bytes_per_pixel, source_ptr, dst_ptr,
+                                  regs.src_params.BlockHeight(), regs.src_params.pos_x,
                                   regs.src_params.pos_y);
     } else {
         ASSERT(regs.dst_params.size_z == 1);
@@ -124,7 +123,7 @@ void MaxwellDMA::HandleCopy() {
 
         // If the input is linear and the output is tiled, swizzle the input and copy it over.
         Texture::SwizzleSubrect(regs.x_count, regs.y_count, regs.src_pitch, regs.dst_params.size_x,
-                                src_bpp, *dest_cpu, *source_cpu, regs.dst_params.BlockHeight());
+                                src_bpp, dst_ptr, source_ptr, regs.dst_params.BlockHeight());
     }
 }
 

src/video_core/gpu.cpp

@@ -274,7 +274,6 @@ void GPU::ProcessSemaphoreTriggerMethod() {
     const auto op =
         static_cast<GpuSemaphoreOperation>(regs.semaphore_trigger & semaphoreOperationMask);
     if (op == GpuSemaphoreOperation::WriteLong) {
-        auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
         struct Block {
             u32 sequence;
             u32 zeros = 0;
@@ -286,11 +285,9 @@ void GPU::ProcessSemaphoreTriggerMethod() {
         // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of
         // CoreTiming
         block.timestamp = Core::System::GetInstance().CoreTiming().GetTicks();
-        Memory::WriteBlock(*address, &block, sizeof(block));
+        memory_manager->WriteBlock(regs.smaphore_address.SmaphoreAddress(), &block, sizeof(block));
     } else {
-        const auto address =
-            memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
-        const u32 word = Memory::Read32(*address);
+        const u32 word{memory_manager->Read32(regs.smaphore_address.SmaphoreAddress())};
         if ((op == GpuSemaphoreOperation::AcquireEqual && word == regs.semaphore_sequence) ||
             (op == GpuSemaphoreOperation::AcquireGequal &&
              static_cast<s32>(word - regs.semaphore_sequence) > 0) ||
@@ -317,13 +314,11 @@ void GPU::ProcessSemaphoreTriggerMethod() {
 }
 
 void GPU::ProcessSemaphoreRelease() {
-    const auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
-    Memory::Write32(*address, regs.semaphore_release);
+    memory_manager->Write32(regs.smaphore_address.SmaphoreAddress(), regs.semaphore_release);
 }
 
 void GPU::ProcessSemaphoreAcquire() {
-    const auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
-    const u32 word = Memory::Read32(*address);
+    const u32 word = memory_manager->Read32(regs.smaphore_address.SmaphoreAddress());
     const auto value = regs.semaphore_acquire;
     if (word != value) {
         regs.acquire_active = true;

src/video_core/gpu.h

@@ -11,6 +11,11 @@
 #include "video_core/dma_pusher.h"
 #include "video_core/memory_manager.h"
 
+using CacheAddr = std::uintptr_t;
+inline CacheAddr ToCacheAddr(const void* host_ptr) {
+    return reinterpret_cast<CacheAddr>(host_ptr);
+}
+
 namespace Core {
 class System;
 }
@@ -209,13 +214,13 @@ public:
         std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
-    virtual void FlushRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushRegion(CacheAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be invalidated
-    virtual void InvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void InvalidateRegion(CacheAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed and invalidated
-    virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
 
 private:
     void ProcessBindMethod(const MethodCall& method_call);

src/video_core/gpu_asynch.cpp

@@ -22,15 +22,15 @@ void GPUAsynch::SwapBuffers(
     gpu_thread.SwapBuffers(std::move(framebuffer));
 }
 
-void GPUAsynch::FlushRegion(VAddr addr, u64 size) {
+void GPUAsynch::FlushRegion(CacheAddr addr, u64 size) {
     gpu_thread.FlushRegion(addr, size);
 }
 
-void GPUAsynch::InvalidateRegion(VAddr addr, u64 size) {
+void GPUAsynch::InvalidateRegion(CacheAddr addr, u64 size) {
     gpu_thread.InvalidateRegion(addr, size);
 }
 
-void GPUAsynch::FlushAndInvalidateRegion(VAddr addr, u64 size) {
+void GPUAsynch::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     gpu_thread.FlushAndInvalidateRegion(addr, size);
 }
 

src/video_core/gpu_asynch.h

@@ -26,9 +26,9 @@ public:
     void PushGPUEntries(Tegra::CommandList&& entries) override;
     void SwapBuffers(
         std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) override;
-    void FlushRegion(VAddr addr, u64 size) override;
-    void InvalidateRegion(VAddr addr, u64 size) override;
-    void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
+    void FlushRegion(CacheAddr addr, u64 size) override;
+    void InvalidateRegion(CacheAddr addr, u64 size) override;
+    void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
 
 private:
     GPUThread::ThreadManager gpu_thread;

src/video_core/gpu_synch.cpp

@@ -22,15 +22,15 @@ void GPUSynch::SwapBuffers(
     renderer.SwapBuffers(std::move(framebuffer));
 }
 
-void GPUSynch::FlushRegion(VAddr addr, u64 size) {
+void GPUSynch::FlushRegion(CacheAddr addr, u64 size) {
     renderer.Rasterizer().FlushRegion(addr, size);
 }
 
-void GPUSynch::InvalidateRegion(VAddr addr, u64 size) {
+void GPUSynch::InvalidateRegion(CacheAddr addr, u64 size) {
     renderer.Rasterizer().InvalidateRegion(addr, size);
 }
 
-void GPUSynch::FlushAndInvalidateRegion(VAddr addr, u64 size) {
+void GPUSynch::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     renderer.Rasterizer().FlushAndInvalidateRegion(addr, size);
 }
 

src/video_core/gpu_synch.h

@@ -21,9 +21,9 @@ public:
     void PushGPUEntries(Tegra::CommandList&& entries) override;
     void SwapBuffers(
         std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) override;
-    void FlushRegion(VAddr addr, u64 size) override;
-    void InvalidateRegion(VAddr addr, u64 size) override;
-    void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
+    void FlushRegion(CacheAddr addr, u64 size) override;
+    void InvalidateRegion(CacheAddr addr, u64 size) override;
+    void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
 };
 
 } // namespace VideoCommon

src/video_core/gpu_thread.cpp

@@ -5,7 +5,6 @@
 #include "common/assert.h"
 #include "common/microprofile.h"
 #include "core/frontend/scope_acquire_window_context.h"
-#include "core/settings.h"
 #include "video_core/dma_pusher.h"
 #include "video_core/gpu.h"
 #include "video_core/gpu_thread.h"
@@ -13,38 +12,13 @@
 
 namespace VideoCommon::GPUThread {
 
-/// Executes a single GPU thread command
-static void ExecuteCommand(CommandData* command, VideoCore::RendererBase& renderer,
-                           Tegra::DmaPusher& dma_pusher) {
-    if (const auto submit_list = std::get_if<SubmitListCommand>(command)) {
-        dma_pusher.Push(std::move(submit_list->entries));
-        dma_pusher.DispatchCalls();
-    } else if (const auto data = std::get_if<SwapBuffersCommand>(command)) {
-        renderer.SwapBuffers(data->framebuffer);
-    } else if (const auto data = std::get_if<FlushRegionCommand>(command)) {
-        renderer.Rasterizer().FlushRegion(data->addr, data->size);
-    } else if (const auto data = std::get_if<InvalidateRegionCommand>(command)) {
-        renderer.Rasterizer().InvalidateRegion(data->addr, data->size);
-    } else if (const auto data = std::get_if<FlushAndInvalidateRegionCommand>(command)) {
-        renderer.Rasterizer().FlushAndInvalidateRegion(data->addr, data->size);
-    } else {
-        UNREACHABLE();
-    }
-}
-
 /// Runs the GPU thread
 static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_pusher,
                       SynchState& state) {
-
     MicroProfileOnThreadCreate("GpuThread");
 
-    auto WaitForWakeup = [&]() {
-        std::unique_lock<std::mutex> lock{state.signal_mutex};
-        state.signal_condition.wait(lock, [&] { return !state.is_idle || !state.is_running; });
-    };
-
     // Wait for first GPU command before acquiring the window context
-    WaitForWakeup();
+    state.WaitForCommands();
 
     // If emulation was stopped during disk shader loading, abort before trying to acquire context
     if (!state.is_running) {
@@ -53,100 +27,72 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
 
     Core::Frontend::ScopeAcquireWindowContext acquire_context{renderer.GetRenderWindow()};
 
+    CommandDataContainer next;
     while (state.is_running) {
-        if (!state.is_running) {
-            return;
+        state.WaitForCommands();
+        while (!state.queue.Empty()) {
+            state.queue.Pop(next);
+            if (const auto submit_list = std::get_if<SubmitListCommand>(&next.data)) {
+                dma_pusher.Push(std::move(submit_list->entries));
+                dma_pusher.DispatchCalls();
+            } else if (const auto data = std::get_if<SwapBuffersCommand>(&next.data)) {
+                state.DecrementFramesCounter();
+                renderer.SwapBuffers(std::move(data->framebuffer));
+            } else if (const auto data = std::get_if<FlushRegionCommand>(&next.data)) {
+                renderer.Rasterizer().FlushRegion(data->addr, data->size);
+            } else if (const auto data = std::get_if<InvalidateRegionCommand>(&next.data)) {
+                renderer.Rasterizer().InvalidateRegion(data->addr, data->size);
+            } else if (const auto data = std::get_if<EndProcessingCommand>(&next.data)) {
+                return;
+            } else {
+                UNREACHABLE();
+            }
         }
-
-        {
-            // Thread has been woken up, so make the previous write queue the next read queue
-            std::lock_guard<std::mutex> lock{state.signal_mutex};
-            std::swap(state.push_queue, state.pop_queue);
-        }
-
-        // Execute all of the GPU commands
-        while (!state.pop_queue->empty()) {
-            ExecuteCommand(&state.pop_queue->front(), renderer, dma_pusher);
-            state.pop_queue->pop();
-        }
-
-        state.UpdateIdleState();
-
-        // Signal that the GPU thread has finished processing commands
-        if (state.is_idle) {
-            state.idle_condition.notify_one();
-        }
-
-        // Wait for CPU thread to send more GPU commands
-        WaitForWakeup();
     }
 }
 
 ThreadManager::ThreadManager(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_pusher)
     : renderer{renderer}, dma_pusher{dma_pusher}, thread{RunThread, std::ref(renderer),
-                                                         std::ref(dma_pusher), std::ref(state)},
-      thread_id{thread.get_id()} {}
+                                                         std::ref(dma_pusher), std::ref(state)} {}
 
 ThreadManager::~ThreadManager() {
-    {
-        // Notify GPU thread that a shutdown is pending
-        std::lock_guard<std::mutex> lock{state.signal_mutex};
-        state.is_running = false;
-    }
-
-    state.signal_condition.notify_one();
+    // Notify GPU thread that a shutdown is pending
+    PushCommand(EndProcessingCommand());
     thread.join();
 }
 
 void ThreadManager::SubmitList(Tegra::CommandList&& entries) {
-    if (entries.empty()) {
-        return;
-    }
-
-    PushCommand(SubmitListCommand(std::move(entries)), false, false);
+    PushCommand(SubmitListCommand(std::move(entries)));
 }
 
 void ThreadManager::SwapBuffers(
     std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) {
-    PushCommand(SwapBuffersCommand(std::move(framebuffer)), true, false);
+    state.IncrementFramesCounter();
+    PushCommand(SwapBuffersCommand(std::move(framebuffer)));
+    state.WaitForFrames();
 }
 
-void ThreadManager::FlushRegion(VAddr addr, u64 size) {
-    // Block the CPU when using accurate emulation
-    PushCommand(FlushRegionCommand(addr, size), Settings::values.use_accurate_gpu_emulation, false);
+void ThreadManager::FlushRegion(CacheAddr addr, u64 size) {
+    PushCommand(FlushRegionCommand(addr, size));
 }
 
-void ThreadManager::InvalidateRegion(VAddr addr, u64 size) {
-    PushCommand(InvalidateRegionCommand(addr, size), true, true);
+void ThreadManager::InvalidateRegion(CacheAddr addr, u64 size) {
+    if (state.queue.Empty()) {
+        // It's quicker to invalidate a single region on the CPU if the queue is already empty
+        renderer.Rasterizer().InvalidateRegion(addr, size);
+    } else {
+        PushCommand(InvalidateRegionCommand(addr, size));
+    }
 }
 
-void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) {
+void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
+    // Skip flush on asynch mode, as FlushAndInvalidateRegion is not used for anything too important
     InvalidateRegion(addr, size);
 }
 
-void ThreadManager::PushCommand(CommandData&& command_data, bool wait_for_idle, bool allow_on_cpu) {
-    {
-        std::lock_guard<std::mutex> lock{state.signal_mutex};
-
-        if ((allow_on_cpu && state.is_idle) || IsGpuThread()) {
-            // Execute the command synchronously on the current thread
-            ExecuteCommand(&command_data, renderer, dma_pusher);
-            return;
-        }
-
-        // Push the command to the GPU thread
-        state.UpdateIdleState();
-        state.push_queue->emplace(command_data);
-    }
-
-    // Signal the GPU thread that commands are pending
-    state.signal_condition.notify_one();
-
-    if (wait_for_idle) {
-        // Wait for the GPU to be idle (all commands to be executed)
-        std::unique_lock<std::mutex> lock{state.idle_mutex};
-        state.idle_condition.wait(lock, [this] { return static_cast<bool>(state.is_idle); });
-    }
+void ThreadManager::PushCommand(CommandData&& command_data) {
+    state.queue.Push(CommandDataContainer(std::move(command_data)));
+    state.SignalCommands();
 }
 
 } // namespace VideoCommon::GPUThread

src/video_core/gpu_thread.h

@@ -13,6 +13,9 @@
 #include <thread>
 #include <variant>
 
+#include "common/threadsafe_queue.h"
+#include "video_core/gpu.h"
+
 namespace Tegra {
 struct FramebufferConfig;
 class DmaPusher;
@@ -24,6 +27,9 @@ class RendererBase;
 
 namespace VideoCommon::GPUThread {
 
+/// Command to signal to the GPU thread that processing has ended
+struct EndProcessingCommand final {};
+
 /// Command to signal to the GPU thread that a command list is ready for processing
 struct SubmitListCommand final {
     explicit SubmitListCommand(Tegra::CommandList&& entries) : entries{std::move(entries)} {}
@@ -36,59 +42,110 @@ struct SwapBuffersCommand final {
     explicit SwapBuffersCommand(std::optional<const Tegra::FramebufferConfig> framebuffer)
         : framebuffer{std::move(framebuffer)} {}
 
-    std::optional<const Tegra::FramebufferConfig> framebuffer;
+    std::optional<Tegra::FramebufferConfig> framebuffer;
 };
 
 /// Command to signal to the GPU thread to flush a region
 struct FlushRegionCommand final {
-    explicit constexpr FlushRegionCommand(VAddr addr, u64 size) : addr{addr}, size{size} {}
+    explicit constexpr FlushRegionCommand(CacheAddr addr, u64 size) : addr{addr}, size{size} {}
 
-    const VAddr addr;
-    const u64 size;
+    CacheAddr addr;
+    u64 size;
 };
 
 /// Command to signal to the GPU thread to invalidate a region
 struct InvalidateRegionCommand final {
-    explicit constexpr InvalidateRegionCommand(VAddr addr, u64 size) : addr{addr}, size{size} {}
+    explicit constexpr InvalidateRegionCommand(CacheAddr addr, u64 size) : addr{addr}, size{size} {}
 
-    const VAddr addr;
-    const u64 size;
+    CacheAddr addr;
+    u64 size;
 };
 
 /// Command to signal to the GPU thread to flush and invalidate a region
 struct FlushAndInvalidateRegionCommand final {
-    explicit constexpr FlushAndInvalidateRegionCommand(VAddr addr, u64 size)
+    explicit constexpr FlushAndInvalidateRegionCommand(CacheAddr addr, u64 size)
         : addr{addr}, size{size} {}
 
-    const VAddr addr;
-    const u64 size;
+    CacheAddr addr;
+    u64 size;
 };
 
-using CommandData = std::variant<SubmitListCommand, SwapBuffersCommand, FlushRegionCommand,
-                                 InvalidateRegionCommand, FlushAndInvalidateRegionCommand>;
+using CommandData =
+    std::variant<EndProcessingCommand, SubmitListCommand, SwapBuffersCommand, FlushRegionCommand,
+                 InvalidateRegionCommand, FlushAndInvalidateRegionCommand>;
+
+struct CommandDataContainer {
+    CommandDataContainer() = default;
+
+    CommandDataContainer(CommandData&& data) : data{std::move(data)} {}
+
+    CommandDataContainer& operator=(const CommandDataContainer& t) {
+        data = std::move(t.data);
+        return *this;
+    }
+
+    CommandData data;
+};
 
 /// Struct used to synchronize the GPU thread
 struct SynchState final {
-    std::atomic<bool> is_running{true};
-    std::atomic<bool> is_idle{true};
-    std::condition_variable signal_condition;
-    std::mutex signal_mutex;
-    std::condition_variable idle_condition;
-    std::mutex idle_mutex;
+    std::atomic_bool is_running{true};
+    std::atomic_int queued_frame_count{};
+    std::mutex frames_mutex;
+    std::mutex commands_mutex;
+    std::condition_variable commands_condition;
+    std::condition_variable frames_condition;
 
-    // We use two queues for sending commands to the GPU thread, one for writing (push_queue) to and
-    // one for reading from (pop_queue). These are swapped whenever the current pop_queue becomes
-    // empty. This allows for efficient thread-safe access, as it does not require any copies.
-
-    using CommandQueue = std::queue<CommandData>;
-    std::array<CommandQueue, 2> command_queues;
-    CommandQueue* push_queue{&command_queues[0]};
-    CommandQueue* pop_queue{&command_queues[1]};
-
-    void UpdateIdleState() {
-        std::lock_guard<std::mutex> lock{idle_mutex};
-        is_idle = command_queues[0].empty() && command_queues[1].empty();
+    void IncrementFramesCounter() {
+        std::lock_guard<std::mutex> lock{frames_mutex};
+        ++queued_frame_count;
     }
+
+    void DecrementFramesCounter() {
+        {
+            std::lock_guard<std::mutex> lock{frames_mutex};
+            --queued_frame_count;
+
+            if (queued_frame_count) {
+                return;
+            }
+        }
+        frames_condition.notify_one();
+    }
+
+    void WaitForFrames() {
+        {
+            std::lock_guard<std::mutex> lock{frames_mutex};
+            if (!queued_frame_count) {
+                return;
+            }
+        }
+
+        // Wait for the GPU to be idle (all commands to be executed)
+        {
+            std::unique_lock<std::mutex> lock{frames_mutex};
+            frames_condition.wait(lock, [this] { return !queued_frame_count; });
+        }
+    }
+
+    void SignalCommands() {
+        {
+            std::unique_lock<std::mutex> lock{commands_mutex};
+            if (queue.Empty()) {
+                return;
+            }
+        }
+
+        commands_condition.notify_one();
+    }
+
+    void WaitForCommands() {
+        std::unique_lock<std::mutex> lock{commands_mutex};
+        commands_condition.wait(lock, [this] { return !queue.Empty(); });
+    }
+
+    using CommandQueue = Common::SPSCQueue<CommandDataContainer>;
+    CommandQueue queue;
 };
 
 /// Class used to manage the GPU thread
@@ -105,22 +162,17 @@ public:
         std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer);
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
-    void FlushRegion(VAddr addr, u64 size);
+    void FlushRegion(CacheAddr addr, u64 size);
 
     /// Notify rasterizer that any caches of the specified region should be invalidated
-    void InvalidateRegion(VAddr addr, u64 size);
+    void InvalidateRegion(CacheAddr addr, u64 size);
 
     /// Notify rasterizer that any caches of the specified region should be flushed and invalidated
-    void FlushAndInvalidateRegion(VAddr addr, u64 size);
+    void FlushAndInvalidateRegion(CacheAddr addr, u64 size);
 
 private:
     /// Pushes a command to be executed by the GPU thread
-    void PushCommand(CommandData&& command_data, bool wait_for_idle, bool allow_on_cpu);
-
-    /// Returns true if this is called by the GPU thread
-    bool IsGpuThread() const {
-        return std::this_thread::get_id() == thread_id;
-    }
+    void PushCommand(CommandData&& command_data);
 
 private:
     SynchState state;

src/video_core/memory_manager.cpp

@@ -5,6 +5,7 @@
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "core/memory.h"
 #include "video_core/memory_manager.h"
 
 namespace Tegra {
@@ -162,15 +163,51 @@ std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) {
     return base_addr + (gpu_addr & PAGE_MASK);
 }
 
-std::vector<GPUVAddr> MemoryManager::CpuToGpuAddress(VAddr cpu_addr) const {
-    std::vector<GPUVAddr> results;
-    for (const auto& region : mapped_regions) {
-        if (cpu_addr >= region.cpu_addr && cpu_addr < (region.cpu_addr + region.size)) {
-            const u64 offset{cpu_addr - region.cpu_addr};
-            results.push_back(region.gpu_addr + offset);
-        }
-    }
-    return results;
+u8 MemoryManager::Read8(GPUVAddr addr) {
+    return Memory::Read8(*GpuToCpuAddress(addr));
+}
+
+u16 MemoryManager::Read16(GPUVAddr addr) {
+    return Memory::Read16(*GpuToCpuAddress(addr));
+}
+
+u32 MemoryManager::Read32(GPUVAddr addr) {
+    return Memory::Read32(*GpuToCpuAddress(addr));
+}
+
+u64 MemoryManager::Read64(GPUVAddr addr) {
+    return Memory::Read64(*GpuToCpuAddress(addr));
+}
+
+void MemoryManager::Write8(GPUVAddr addr, u8 data) {
+    Memory::Write8(*GpuToCpuAddress(addr), data);
+}
+
+void MemoryManager::Write16(GPUVAddr addr, u16 data) {
+    Memory::Write16(*GpuToCpuAddress(addr), data);
+}
+
+void MemoryManager::Write32(GPUVAddr addr, u32 data) {
+    Memory::Write32(*GpuToCpuAddress(addr), data);
+}
+
+void MemoryManager::Write64(GPUVAddr addr, u64 data) {
+    Memory::Write64(*GpuToCpuAddress(addr), data);
+}
+
+u8* MemoryManager::GetPointer(GPUVAddr addr) {
+    return Memory::GetPointer(*GpuToCpuAddress(addr));
+}
+
+void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size) {
+    std::memcpy(dest_buffer, GetPointer(src_addr), size);
+}
+void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size) {
+    std::memcpy(GetPointer(dest_addr), src_buffer, size);
+}
+
+void MemoryManager::CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, std::size_t size) {
+    std::memcpy(GetPointer(dest_addr), GetPointer(src_addr), size);
 }
 
 VAddr& MemoryManager::PageSlot(GPUVAddr gpu_addr) {

src/video_core/memory_manager.h

@@ -27,12 +27,27 @@ public:
     GPUVAddr UnmapBuffer(GPUVAddr gpu_addr, u64 size);
     GPUVAddr GetRegionEnd(GPUVAddr region_start) const;
     std::optional<VAddr> GpuToCpuAddress(GPUVAddr gpu_addr);
-    std::vector<GPUVAddr> CpuToGpuAddress(VAddr cpu_addr) const;
 
     static constexpr u64 PAGE_BITS = 16;
     static constexpr u64 PAGE_SIZE = 1 << PAGE_BITS;
     static constexpr u64 PAGE_MASK = PAGE_SIZE - 1;
 
+    u8 Read8(GPUVAddr addr);
+    u16 Read16(GPUVAddr addr);
+    u32 Read32(GPUVAddr addr);
+    u64 Read64(GPUVAddr addr);
+
+    void Write8(GPUVAddr addr, u8 data);
+    void Write16(GPUVAddr addr, u16 data);
+    void Write32(GPUVAddr addr, u32 data);
+    void Write64(GPUVAddr addr, u64 data);
+
+    u8* GetPointer(GPUVAddr vaddr);
+
+    void ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size);
+    void WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size);
+    void CopyBlock(VAddr dest_addr, VAddr src_addr, std::size_t size);
+
 private:
     enum class PageStatus : u64 {
         Unmapped = 0xFFFFFFFFFFFFFFFFULL,

src/video_core/morton.cpp

@@ -6,7 +6,6 @@
 #include <cstring>
 #include "common/assert.h"
 #include "common/common_types.h"
-#include "core/memory.h"
 #include "video_core/morton.h"
 #include "video_core/surface.h"
 #include "video_core/textures/decoders.h"
@@ -16,12 +15,12 @@ namespace VideoCore {
 using Surface::GetBytesPerPixel;
 using Surface::PixelFormat;
 
-using MortonCopyFn = void (*)(u32, u32, u32, u32, u32, u32, u8*, std::size_t, VAddr);
+using MortonCopyFn = void (*)(u32, u32, u32, u32, u32, u32, u8*, u8*);
 using ConversionArray = std::array<MortonCopyFn, Surface::MaxPixelFormat>;
 
 template <bool morton_to_linear, PixelFormat format>
 static void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth, u32 depth,
-                       u32 tile_width_spacing, u8* buffer, std::size_t buffer_size, VAddr addr) {
+                       u32 tile_width_spacing, u8* buffer, u8* addr) {
     constexpr u32 bytes_per_pixel = GetBytesPerPixel(format);
 
     // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual
@@ -34,150 +33,146 @@ static void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth
                                          stride, height, depth, block_height, block_depth,
                                          tile_width_spacing);
     } else {
-        Tegra::Texture::CopySwizzledData(
-            (stride + tile_size_x - 1) / tile_size_x, (height + tile_size_y - 1) / tile_size_y,
-            depth, bytes_per_pixel, bytes_per_pixel, Memory::GetPointer(addr), buffer, false,
-            block_height, block_depth, tile_width_spacing);
+        Tegra::Texture::CopySwizzledData((stride + tile_size_x - 1) / tile_size_x,
+                                         (height + tile_size_y - 1) / tile_size_y, depth,
+                                         bytes_per_pixel, bytes_per_pixel, addr, buffer, false,
+                                         block_height, block_depth, tile_width_spacing);
     }
 }
 
 static constexpr ConversionArray morton_to_linear_fns = {
-    // clang-format off
-        MortonCopy<true, PixelFormat::ABGR8U>,
-        MortonCopy<true, PixelFormat::ABGR8S>,
-        MortonCopy<true, PixelFormat::ABGR8UI>,
-        MortonCopy<true, PixelFormat::B5G6R5U>,
-        MortonCopy<true, PixelFormat::A2B10G10R10U>,
-        MortonCopy<true, PixelFormat::A1B5G5R5U>,
-        MortonCopy<true, PixelFormat::R8U>,
-        MortonCopy<true, PixelFormat::R8UI>,
-        MortonCopy<true, PixelFormat::RGBA16F>,
-        MortonCopy<true, PixelFormat::RGBA16U>,
-        MortonCopy<true, PixelFormat::RGBA16UI>,
-        MortonCopy<true, PixelFormat::R11FG11FB10F>,
-        MortonCopy<true, PixelFormat::RGBA32UI>,
-        MortonCopy<true, PixelFormat::DXT1>,
-        MortonCopy<true, PixelFormat::DXT23>,
-        MortonCopy<true, PixelFormat::DXT45>,
-        MortonCopy<true, PixelFormat::DXN1>,
-        MortonCopy<true, PixelFormat::DXN2UNORM>,
-        MortonCopy<true, PixelFormat::DXN2SNORM>,
-        MortonCopy<true, PixelFormat::BC7U>,
-        MortonCopy<true, PixelFormat::BC6H_UF16>,
-        MortonCopy<true, PixelFormat::BC6H_SF16>,
-        MortonCopy<true, PixelFormat::ASTC_2D_4X4>,
-        MortonCopy<true, PixelFormat::BGRA8>,
-        MortonCopy<true, PixelFormat::RGBA32F>,
-        MortonCopy<true, PixelFormat::RG32F>,
-        MortonCopy<true, PixelFormat::R32F>,
-        MortonCopy<true, PixelFormat::R16F>,
-        MortonCopy<true, PixelFormat::R16U>,
-        MortonCopy<true, PixelFormat::R16S>,
-        MortonCopy<true, PixelFormat::R16UI>,
-        MortonCopy<true, PixelFormat::R16I>,
-        MortonCopy<true, PixelFormat::RG16>,
-        MortonCopy<true, PixelFormat::RG16F>,
-        MortonCopy<true, PixelFormat::RG16UI>,
-        MortonCopy<true, PixelFormat::RG16I>,
-        MortonCopy<true, PixelFormat::RG16S>,
-        MortonCopy<true, PixelFormat::RGB32F>,
-        MortonCopy<true, PixelFormat::RGBA8_SRGB>,
-        MortonCopy<true, PixelFormat::RG8U>,
-        MortonCopy<true, PixelFormat::RG8S>,
-        MortonCopy<true, PixelFormat::RG32UI>,
-        MortonCopy<true, PixelFormat::R32UI>,
-        MortonCopy<true, PixelFormat::ASTC_2D_8X8>,
-        MortonCopy<true, PixelFormat::ASTC_2D_8X5>,
-        MortonCopy<true, PixelFormat::ASTC_2D_5X4>,
-        MortonCopy<true, PixelFormat::BGRA8_SRGB>,
-        MortonCopy<true, PixelFormat::DXT1_SRGB>,
-        MortonCopy<true, PixelFormat::DXT23_SRGB>,
-        MortonCopy<true, PixelFormat::DXT45_SRGB>,
-        MortonCopy<true, PixelFormat::BC7U_SRGB>,
-        MortonCopy<true, PixelFormat::ASTC_2D_4X4_SRGB>,
-        MortonCopy<true, PixelFormat::ASTC_2D_8X8_SRGB>,
-        MortonCopy<true, PixelFormat::ASTC_2D_8X5_SRGB>,
-        MortonCopy<true, PixelFormat::ASTC_2D_5X4_SRGB>,
-        MortonCopy<true, PixelFormat::ASTC_2D_5X5>,
-        MortonCopy<true, PixelFormat::ASTC_2D_5X5_SRGB>,
-        MortonCopy<true, PixelFormat::ASTC_2D_10X8>,
-        MortonCopy<true, PixelFormat::ASTC_2D_10X8_SRGB>,
-        MortonCopy<true, PixelFormat::Z32F>,
-        MortonCopy<true, PixelFormat::Z16>,
-        MortonCopy<true, PixelFormat::Z24S8>,
-        MortonCopy<true, PixelFormat::S8Z24>,
-        MortonCopy<true, PixelFormat::Z32FS8>,
-    // clang-format on
+    MortonCopy<true, PixelFormat::ABGR8U>,
+    MortonCopy<true, PixelFormat::ABGR8S>,
+    MortonCopy<true, PixelFormat::ABGR8UI>,
+    MortonCopy<true, PixelFormat::B5G6R5U>,
+    MortonCopy<true, PixelFormat::A2B10G10R10U>,
+    MortonCopy<true, PixelFormat::A1B5G5R5U>,
+    MortonCopy<true, PixelFormat::R8U>,
+    MortonCopy<true, PixelFormat::R8UI>,
+    MortonCopy<true, PixelFormat::RGBA16F>,
+    MortonCopy<true, PixelFormat::RGBA16U>,
+    MortonCopy<true, PixelFormat::RGBA16UI>,
+    MortonCopy<true, PixelFormat::R11FG11FB10F>,
+    MortonCopy<true, PixelFormat::RGBA32UI>,
+    MortonCopy<true, PixelFormat::DXT1>,
+    MortonCopy<true, PixelFormat::DXT23>,
+    MortonCopy<true, PixelFormat::DXT45>,
+    MortonCopy<true, PixelFormat::DXN1>,
+    MortonCopy<true, PixelFormat::DXN2UNORM>,
+    MortonCopy<true, PixelFormat::DXN2SNORM>,
+    MortonCopy<true, PixelFormat::BC7U>,
+    MortonCopy<true, PixelFormat::BC6H_UF16>,
+    MortonCopy<true, PixelFormat::BC6H_SF16>,
+    MortonCopy<true, PixelFormat::ASTC_2D_4X4>,
+    MortonCopy<true, PixelFormat::BGRA8>,
+    MortonCopy<true, PixelFormat::RGBA32F>,
+    MortonCopy<true, PixelFormat::RG32F>,
+    MortonCopy<true, PixelFormat::R32F>,
+    MortonCopy<true, PixelFormat::R16F>,
+    MortonCopy<true, PixelFormat::R16U>,
+    MortonCopy<true, PixelFormat::R16S>,
+    MortonCopy<true, PixelFormat::R16UI>,
+    MortonCopy<true, PixelFormat::R16I>,
+    MortonCopy<true, PixelFormat::RG16>,
+    MortonCopy<true, PixelFormat::RG16F>,
+    MortonCopy<true, PixelFormat::RG16UI>,
+    MortonCopy<true, PixelFormat::RG16I>,
+    MortonCopy<true, PixelFormat::RG16S>,
+    MortonCopy<true, PixelFormat::RGB32F>,
+    MortonCopy<true, PixelFormat::RGBA8_SRGB>,
+    MortonCopy<true, PixelFormat::RG8U>,
+    MortonCopy<true, PixelFormat::RG8S>,
+    MortonCopy<true, PixelFormat::RG32UI>,
+    MortonCopy<true, PixelFormat::R32UI>,
+    MortonCopy<true, PixelFormat::ASTC_2D_8X8>,
+    MortonCopy<true, PixelFormat::ASTC_2D_8X5>,
+    MortonCopy<true, PixelFormat::ASTC_2D_5X4>,
+    MortonCopy<true, PixelFormat::BGRA8_SRGB>,
+    MortonCopy<true, PixelFormat::DXT1_SRGB>,
+    MortonCopy<true, PixelFormat::DXT23_SRGB>,
+    MortonCopy<true, PixelFormat::DXT45_SRGB>,
+    MortonCopy<true, PixelFormat::BC7U_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_4X4_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_8X8_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_8X5_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_5X4_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_5X5>,
+    MortonCopy<true, PixelFormat::ASTC_2D_5X5_SRGB>,
+    MortonCopy<true, PixelFormat::ASTC_2D_10X8>,
+    MortonCopy<true, PixelFormat::ASTC_2D_10X8_SRGB>,
+    MortonCopy<true, PixelFormat::Z32F>,
+    MortonCopy<true, PixelFormat::Z16>,
+    MortonCopy<true, PixelFormat::Z24S8>,
+    MortonCopy<true, PixelFormat::S8Z24>,
+    MortonCopy<true, PixelFormat::Z32FS8>,
 };
 
 static constexpr ConversionArray linear_to_morton_fns = {
-    // clang-format off
-        MortonCopy<false, PixelFormat::ABGR8U>,
-        MortonCopy<false, PixelFormat::ABGR8S>,
-        MortonCopy<false, PixelFormat::ABGR8UI>,
-        MortonCopy<false, PixelFormat::B5G6R5U>,
-        MortonCopy<false, PixelFormat::A2B10G10R10U>,
-        MortonCopy<false, PixelFormat::A1B5G5R5U>,
-        MortonCopy<false, PixelFormat::R8U>,
-        MortonCopy<false, PixelFormat::R8UI>,
-        MortonCopy<false, PixelFormat::RGBA16F>,
-        MortonCopy<false, PixelFormat::RGBA16U>,
-        MortonCopy<false, PixelFormat::RGBA16UI>,
-        MortonCopy<false, PixelFormat::R11FG11FB10F>,
-        MortonCopy<false, PixelFormat::RGBA32UI>,
-        MortonCopy<false, PixelFormat::DXT1>,
-        MortonCopy<false, PixelFormat::DXT23>,
-        MortonCopy<false, PixelFormat::DXT45>,
-        MortonCopy<false, PixelFormat::DXN1>,
-        MortonCopy<false, PixelFormat::DXN2UNORM>,
-        MortonCopy<false, PixelFormat::DXN2SNORM>,
-        MortonCopy<false, PixelFormat::BC7U>,
-        MortonCopy<false, PixelFormat::BC6H_UF16>,
-        MortonCopy<false, PixelFormat::BC6H_SF16>,
-        // TODO(Subv): Swizzling ASTC formats are not supported
-        nullptr,
-        MortonCopy<false, PixelFormat::BGRA8>,
-        MortonCopy<false, PixelFormat::RGBA32F>,
-        MortonCopy<false, PixelFormat::RG32F>,
-        MortonCopy<false, PixelFormat::R32F>,
-        MortonCopy<false, PixelFormat::R16F>,
-        MortonCopy<false, PixelFormat::R16U>,
-        MortonCopy<false, PixelFormat::R16S>,
-        MortonCopy<false, PixelFormat::R16UI>,
-        MortonCopy<false, PixelFormat::R16I>,
-        MortonCopy<false, PixelFormat::RG16>,
-        MortonCopy<false, PixelFormat::RG16F>,
-        MortonCopy<false, PixelFormat::RG16UI>,
-        MortonCopy<false, PixelFormat::RG16I>,
-        MortonCopy<false, PixelFormat::RG16S>,
-        MortonCopy<false, PixelFormat::RGB32F>,
-        MortonCopy<false, PixelFormat::RGBA8_SRGB>,
-        MortonCopy<false, PixelFormat::RG8U>,
-        MortonCopy<false, PixelFormat::RG8S>,
-        MortonCopy<false, PixelFormat::RG32UI>,
-        MortonCopy<false, PixelFormat::R32UI>,
-        nullptr,
-        nullptr,
-        nullptr,
-        MortonCopy<false, PixelFormat::BGRA8_SRGB>,
-        MortonCopy<false, PixelFormat::DXT1_SRGB>,
-        MortonCopy<false, PixelFormat::DXT23_SRGB>,
-        MortonCopy<false, PixelFormat::DXT45_SRGB>,
-        MortonCopy<false, PixelFormat::BC7U_SRGB>,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        nullptr,
-        MortonCopy<false, PixelFormat::Z32F>,
-        MortonCopy<false, PixelFormat::Z16>,
-        MortonCopy<false, PixelFormat::Z24S8>,
-        MortonCopy<false, PixelFormat::S8Z24>,
-        MortonCopy<false, PixelFormat::Z32FS8>,
-    // clang-format on
+    MortonCopy<false, PixelFormat::ABGR8U>,
+    MortonCopy<false, PixelFormat::ABGR8S>,
+    MortonCopy<false, PixelFormat::ABGR8UI>,
+    MortonCopy<false, PixelFormat::B5G6R5U>,
+    MortonCopy<false, PixelFormat::A2B10G10R10U>,
+    MortonCopy<false, PixelFormat::A1B5G5R5U>,
+    MortonCopy<false, PixelFormat::R8U>,
+    MortonCopy<false, PixelFormat::R8UI>,
+    MortonCopy<false, PixelFormat::RGBA16F>,
+    MortonCopy<false, PixelFormat::RGBA16U>,
+    MortonCopy<false, PixelFormat::RGBA16UI>,
+    MortonCopy<false, PixelFormat::R11FG11FB10F>,
+    MortonCopy<false, PixelFormat::RGBA32UI>,
+    MortonCopy<false, PixelFormat::DXT1>,
+    MortonCopy<false, PixelFormat::DXT23>,
+    MortonCopy<false, PixelFormat::DXT45>,
+    MortonCopy<false, PixelFormat::DXN1>,
+    MortonCopy<false, PixelFormat::DXN2UNORM>,
+    MortonCopy<false, PixelFormat::DXN2SNORM>,
+    MortonCopy<false, PixelFormat::BC7U>,
+    MortonCopy<false, PixelFormat::BC6H_UF16>,
+    MortonCopy<false, PixelFormat::BC6H_SF16>,
+    // TODO(Subv): Swizzling ASTC formats are not supported
+    nullptr,
+    MortonCopy<false, PixelFormat::BGRA8>,
+    MortonCopy<false, PixelFormat::RGBA32F>,
+    MortonCopy<false, PixelFormat::RG32F>,
+    MortonCopy<false, PixelFormat::R32F>,
+    MortonCopy<false, PixelFormat::R16F>,
+    MortonCopy<false, PixelFormat::R16U>,
+    MortonCopy<false, PixelFormat::R16S>,
+    MortonCopy<false, PixelFormat::R16UI>,
+    MortonCopy<false, PixelFormat::R16I>,
+    MortonCopy<false, PixelFormat::RG16>,
+    MortonCopy<false, PixelFormat::RG16F>,
+    MortonCopy<false, PixelFormat::RG16UI>,
+    MortonCopy<false, PixelFormat::RG16I>,
+    MortonCopy<false, PixelFormat::RG16S>,
+    MortonCopy<false, PixelFormat::RGB32F>,
+    MortonCopy<false, PixelFormat::RGBA8_SRGB>,
+    MortonCopy<false, PixelFormat::RG8U>,
+    MortonCopy<false, PixelFormat::RG8S>,
+    MortonCopy<false, PixelFormat::RG32UI>,
+    MortonCopy<false, PixelFormat::R32UI>,
+    nullptr,
+    nullptr,
+    nullptr,
+    MortonCopy<false, PixelFormat::BGRA8_SRGB>,
+    MortonCopy<false, PixelFormat::DXT1_SRGB>,
+    MortonCopy<false, PixelFormat::DXT23_SRGB>,
+    MortonCopy<false, PixelFormat::DXT45_SRGB>,
+    MortonCopy<false, PixelFormat::BC7U_SRGB>,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    nullptr,
+    MortonCopy<false, PixelFormat::Z32F>,
+    MortonCopy<false, PixelFormat::Z16>,
+    MortonCopy<false, PixelFormat::Z24S8>,
+    MortonCopy<false, PixelFormat::S8Z24>,
+    MortonCopy<false, PixelFormat::Z32FS8>,
 };
 
 static MortonCopyFn GetSwizzleFunction(MortonSwizzleMode mode, Surface::PixelFormat format) {
@@ -191,45 +186,6 @@ static MortonCopyFn GetSwizzleFunction(MortonSwizzleMode mode, Surface::PixelFor
     return morton_to_linear_fns[static_cast<std::size_t>(format)];
 }
 
-/// 8x8 Z-Order coordinate from 2D coordinates
-static u32 MortonInterleave(u32 x, u32 y) {
-    static const u32 xlut[] = {0x00, 0x01, 0x04, 0x05, 0x10, 0x11, 0x14, 0x15};
-    static const u32 ylut[] = {0x00, 0x02, 0x08, 0x0a, 0x20, 0x22, 0x28, 0x2a};
-    return xlut[x % 8] + ylut[y % 8];
-}
-
-/// Calculates the offset of the position of the pixel in Morton order
-static u32 GetMortonOffset(u32 x, u32 y, u32 bytes_per_pixel) {
-    // Images are split into 8x8 tiles. Each tile is composed of four 4x4 subtiles each
-    // of which is composed of four 2x2 subtiles each of which is composed of four texels.
-    // Each structure is embedded into the next-bigger one in a diagonal pattern, e.g.
-    // texels are laid out in a 2x2 subtile like this:
-    // 2 3
-    // 0 1
-    //
-    // The full 8x8 tile has the texels arranged like this:
-    //
-    // 42 43 46 47 58 59 62 63
-    // 40 41 44 45 56 57 60 61
-    // 34 35 38 39 50 51 54 55
-    // 32 33 36 37 48 49 52 53
-    // 10 11 14 15 26 27 30 31
-    // 08 09 12 13 24 25 28 29
-    // 02 03 06 07 18 19 22 23
-    // 00 01 04 05 16 17 20 21
-    //
-    // This pattern is what's called Z-order curve, or Morton order.
-
-    const unsigned int block_height = 8;
-    const unsigned int coarse_x = x & ~7;
-
-    u32 i = MortonInterleave(x, y);
-
-    const unsigned int offset = coarse_x * block_height;
-
-    return (i + offset) * bytes_per_pixel;
-}
-
 static u32 MortonInterleave128(u32 x, u32 y) {
     // 128x128 Z-Order coordinate from 2D coordinates
     static constexpr u32 xlut[] = {
@@ -325,14 +281,14 @@ static u32 GetMortonOffset128(u32 x, u32 y, u32 bytes_per_pixel) {
 
 void MortonSwizzle(MortonSwizzleMode mode, Surface::PixelFormat format, u32 stride,
                    u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing,
-                   u8* buffer, std::size_t buffer_size, VAddr addr) {
-
+                   u8* buffer, u8* addr) {
     GetSwizzleFunction(mode, format)(stride, block_height, height, block_depth, depth,
-                                     tile_width_spacing, buffer, buffer_size, addr);
+                                     tile_width_spacing, buffer, addr);
 }
 
-void MortonCopyPixels128(u32 width, u32 height, u32 bytes_per_pixel, u32 linear_bytes_per_pixel,
-                         u8* morton_data, u8* linear_data, bool morton_to_linear) {
+void MortonCopyPixels128(MortonSwizzleMode mode, u32 width, u32 height, u32 bytes_per_pixel,
+                         u32 linear_bytes_per_pixel, u8* morton_data, u8* linear_data) {
+    const bool morton_to_linear = mode == MortonSwizzleMode::MortonToLinear;
     u8* data_ptrs[2];
     for (u32 y = 0; y < height; ++y) {
         for (u32 x = 0; x < width; ++x) {

src/video_core/morton.h

@@ -13,9 +13,9 @@ enum class MortonSwizzleMode { MortonToLinear, LinearToMorton };
 
 void MortonSwizzle(MortonSwizzleMode mode, VideoCore::Surface::PixelFormat format, u32 stride,
                    u32 block_height, u32 height, u32 block_depth, u32 depth, u32 tile_width_spacing,
-                   u8* buffer, std::size_t buffer_size, VAddr addr);
+                   u8* buffer, u8* addr);
 
-void MortonCopyPixels128(u32 width, u32 height, u32 bytes_per_pixel, u32 linear_bytes_per_pixel,
-                         u8* morton_data, u8* linear_data, bool morton_to_linear);
+void MortonCopyPixels128(MortonSwizzleMode mode, u32 width, u32 height, u32 bytes_per_pixel,
+                         u32 linear_bytes_per_pixel, u8* morton_data, u8* linear_data);
 
 } // namespace VideoCore

src/video_core/rasterizer_cache.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include <mutex>
 #include <set>
 #include <unordered_map>
 
@@ -12,14 +13,26 @@
 
 #include "common/common_types.h"
 #include "core/settings.h"
+#include "video_core/gpu.h"
 #include "video_core/rasterizer_interface.h"
 
 class RasterizerCacheObject {
 public:
+    explicit RasterizerCacheObject(const u8* host_ptr)
+        : host_ptr{host_ptr}, cache_addr{ToCacheAddr(host_ptr)} {}
+
     virtual ~RasterizerCacheObject();
 
+    CacheAddr GetCacheAddr() const {
+        return cache_addr;
+    }
+
+    const u8* GetHostPtr() const {
+        return host_ptr;
+    }
+
     /// Gets the address of the shader in guest memory, required for cache management
-    virtual VAddr GetAddr() const = 0;
+    virtual VAddr GetCpuAddr() const = 0;
 
     /// Gets the size of the shader in guest memory, required for cache management
     virtual std::size_t GetSizeInBytes() const = 0;
@@ -58,6 +71,8 @@ private:
     bool is_registered{};      ///< Whether the object is currently registered with the cache
     bool is_dirty{};           ///< Whether the object is dirty (out of sync with guest memory)
     u64 last_modified_ticks{}; ///< When the object was last modified, used for in-order flushing
+    CacheAddr cache_addr{};    ///< Cache address memory, unique from emulated virtual address space
+    const u8* host_ptr{};      ///< Pointer to the memory backing this cached region
 };
 
 template <class T>
@@ -68,7 +83,9 @@ public:
     explicit RasterizerCache(VideoCore::RasterizerInterface& rasterizer) : rasterizer{rasterizer} {}
 
     /// Write any cached resources overlapping the specified region back to memory
-    void FlushRegion(Tegra::GPUVAddr addr, size_t size) {
+    void FlushRegion(CacheAddr addr, std::size_t size) {
+        std::lock_guard<std::recursive_mutex> lock{mutex};
+
         const auto& objects{GetSortedObjectsFromRegion(addr, size)};
         for (auto& object : objects) {
             FlushObject(object);
@@ -76,7 +93,9 @@ public:
     }
 
     /// Mark the specified region as being invalidated
-    void InvalidateRegion(VAddr addr, u64 size) {
+    void InvalidateRegion(CacheAddr addr, u64 size) {
+        std::lock_guard<std::recursive_mutex> lock{mutex};
+
         const auto& objects{GetSortedObjectsFromRegion(addr, size)};
         for (auto& object : objects) {
             if (!object->IsRegistered()) {
@@ -89,48 +108,60 @@ public:
 
     /// Invalidates everything in the cache
     void InvalidateAll() {
+        std::lock_guard<std::recursive_mutex> lock{mutex};
+
         while (interval_cache.begin() != interval_cache.end()) {
             Unregister(*interval_cache.begin()->second.begin());
         }
     }
 
 protected:
-    /// Tries to get an object from the cache with the specified address
-    T TryGet(VAddr addr) const {
+    /// Tries to get an object from the cache with the specified cache address
+    T TryGet(CacheAddr addr) const {
         const auto iter = map_cache.find(addr);
         if (iter != map_cache.end())
             return iter->second;
         return nullptr;
     }
 
+    T TryGet(const void* addr) const {
+        const auto iter = map_cache.find(ToCacheAddr(addr));
+        if (iter != map_cache.end())
+            return iter->second;
+        return nullptr;
+    }
+
     /// Register an object into the cache
     void Register(const T& object) {
+        std::lock_guard<std::recursive_mutex> lock{mutex};
+
         object->SetIsRegistered(true);
         interval_cache.add({GetInterval(object), ObjectSet{object}});
-        map_cache.insert({object->GetAddr(), object});
-        rasterizer.UpdatePagesCachedCount(object->GetAddr(), object->GetSizeInBytes(), 1);
+        map_cache.insert({object->GetCacheAddr(), object});
+        rasterizer.UpdatePagesCachedCount(object->GetCpuAddr(), object->GetSizeInBytes(), 1);
     }
 
     /// Unregisters an object from the cache
     void Unregister(const T& object) {
-        object->SetIsRegistered(false);
-        rasterizer.UpdatePagesCachedCount(object->GetAddr(), object->GetSizeInBytes(), -1);
-        // Only flush if use_accurate_gpu_emulation is enabled, as it incurs a performance hit
-        if (Settings::values.use_accurate_gpu_emulation) {
-            FlushObject(object);
-        }
+        std::lock_guard<std::recursive_mutex> lock{mutex};
 
+        object->SetIsRegistered(false);
+        rasterizer.UpdatePagesCachedCount(object->GetCpuAddr(), object->GetSizeInBytes(), -1);
         interval_cache.subtract({GetInterval(object), ObjectSet{object}});
-        map_cache.erase(object->GetAddr());
+        map_cache.erase(object->GetCacheAddr());
     }
 
     /// Returns a ticks counter used for tracking when cached objects were last modified
     u64 GetModifiedTicks() {
+        std::lock_guard<std::recursive_mutex> lock{mutex};
+
         return ++modified_ticks;
     }
 
     /// Flushes the specified object, updating appropriate cache state as needed
     void FlushObject(const T& object) {
+        std::lock_guard<std::recursive_mutex> lock{mutex};
+
         if (!object->IsDirty()) {
             return;
         }
@@ -140,7 +171,7 @@ protected:
 
 private:
     /// Returns a list of cached objects from the specified memory region, ordered by access time
-    std::vector<T> GetSortedObjectsFromRegion(VAddr addr, u64 size) {
+    std::vector<T> GetSortedObjectsFromRegion(CacheAddr addr, u64 size) {
         if (size == 0) {
             return {};
         }
@@ -164,17 +195,18 @@ private:
     }
 
     using ObjectSet = std::set<T>;
-    using ObjectCache = std::unordered_map<VAddr, T>;
-    using IntervalCache = boost::icl::interval_map<VAddr, ObjectSet>;
+    using ObjectCache = std::unordered_map<CacheAddr, T>;
+    using IntervalCache = boost::icl::interval_map<CacheAddr, ObjectSet>;
     using ObjectInterval = typename IntervalCache::interval_type;
 
     static auto GetInterval(const T& object) {
-        return ObjectInterval::right_open(object->GetAddr(),
-                                          object->GetAddr() + object->GetSizeInBytes());
+        return ObjectInterval::right_open(object->GetCacheAddr(),
+                                          object->GetCacheAddr() + object->GetSizeInBytes());
     }
 
     ObjectCache map_cache;
     IntervalCache interval_cache; ///< Cache of objects
     u64 modified_ticks{};         ///< Counter of cache state ticks, used for in-order flushing
     VideoCore::RasterizerInterface& rasterizer;
+    std::recursive_mutex mutex;
 };

src/video_core/rasterizer_interface.h

@@ -35,14 +35,14 @@ public:
     virtual void FlushAll() = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
-    virtual void FlushRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushRegion(CacheAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be invalidated
-    virtual void InvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void InvalidateRegion(CacheAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     /// and invalidated
-    virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
 
     /// Attempt to use a faster method to perform a surface copy
     virtual bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
@@ -63,7 +63,7 @@ public:
     }
 
     /// Increase/decrease the number of object in pages touching the specified region
-    virtual void UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta) {}
+    virtual void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {}
 
     /// Initialize disk cached resources for the game being emulated
     virtual void LoadDiskResources(const std::atomic_bool& stop_loading = false,

src/video_core/renderer_opengl/gl_buffer_cache.cpp

@@ -13,24 +13,28 @@
 
 namespace OpenGL {
 
+CachedBufferEntry::CachedBufferEntry(VAddr cpu_addr, std::size_t size, GLintptr offset,
+                                     std::size_t alignment, u8* host_ptr)
+    : cpu_addr{cpu_addr}, size{size}, offset{offset}, alignment{alignment}, RasterizerCacheObject{
+                                                                                host_ptr} {}
+
 OGLBufferCache::OGLBufferCache(RasterizerOpenGL& rasterizer, std::size_t size)
     : RasterizerCache{rasterizer}, stream_buffer(size, true) {}
 
 GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size,
                                       std::size_t alignment, bool cache) {
     auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
-    const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
-    ASSERT_MSG(cpu_addr, "Invalid GPU address");
 
     // Cache management is a big overhead, so only cache entries with a given size.
     // TODO: Figure out which size is the best for given games.
     cache &= size >= 2048;
 
+    const auto& host_ptr{memory_manager.GetPointer(gpu_addr)};
     if (cache) {
-        auto entry = TryGet(*cpu_addr);
+        auto entry = TryGet(host_ptr);
         if (entry) {
-            if (entry->size >= size && entry->alignment == alignment) {
-                return entry->offset;
+            if (entry->GetSize() >= size && entry->GetAlignment() == alignment) {
+                return entry->GetOffset();
             }
             Unregister(entry);
         }
@@ -39,17 +43,17 @@ GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size
     AlignBuffer(alignment);
     const GLintptr uploaded_offset = buffer_offset;
 
-    Memory::ReadBlock(*cpu_addr, buffer_ptr, size);
+    if (!host_ptr) {
+        return uploaded_offset;
+    }
 
+    std::memcpy(buffer_ptr, host_ptr, size);
     buffer_ptr += size;
     buffer_offset += size;
 
     if (cache) {
-        auto entry = std::make_shared<CachedBufferEntry>();
-        entry->offset = uploaded_offset;
-        entry->size = size;
-        entry->alignment = alignment;
-        entry->addr = *cpu_addr;
+        auto entry = std::make_shared<CachedBufferEntry>(
+            *memory_manager.GpuToCpuAddress(gpu_addr), size, uploaded_offset, alignment, host_ptr);
         Register(entry);
     }
 

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -17,22 +17,39 @@ namespace OpenGL {
 
 class RasterizerOpenGL;
 
-struct CachedBufferEntry final : public RasterizerCacheObject {
-    VAddr GetAddr() const override {
-        return addr;
+class CachedBufferEntry final : public RasterizerCacheObject {
+public:
+    explicit CachedBufferEntry(VAddr cpu_addr, std::size_t size, GLintptr offset,
+                               std::size_t alignment, u8* host_ptr);
+
+    VAddr GetCpuAddr() const override {
+        return cpu_addr;
     }
 
     std::size_t GetSizeInBytes() const override {
         return size;
     }
 
+    std::size_t GetSize() const {
+        return size;
+    }
+
+    GLintptr GetOffset() const {
+        return offset;
+    }
+
+    std::size_t GetAlignment() const {
+        return alignment;
+    }
+
     // We do not have to flush this cache as things in it are never modified by us.
     void Flush() override {}
 
-    VAddr addr;
-    std::size_t size;
-    GLintptr offset;
-    std::size_t alignment;
+private:
+    VAddr cpu_addr{};
+    std::size_t size{};
+    GLintptr offset{};
+    std::size_t alignment{};
 };
 
 class OGLBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> {

src/video_core/renderer_opengl/gl_global_cache.cpp

@@ -7,7 +7,6 @@
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "core/core.h"
-#include "core/memory.h"
 #include "video_core/renderer_opengl/gl_global_cache.h"
 #include "video_core/renderer_opengl/gl_rasterizer.h"
 #include "video_core/renderer_opengl/gl_shader_decompiler.h"
@@ -15,12 +14,13 @@
 
 namespace OpenGL {
 
-CachedGlobalRegion::CachedGlobalRegion(VAddr addr, u32 size) : addr{addr}, size{size} {
+CachedGlobalRegion::CachedGlobalRegion(VAddr cpu_addr, u32 size, u8* host_ptr)
+    : cpu_addr{cpu_addr}, size{size}, RasterizerCacheObject{host_ptr} {
     buffer.Create();
     // Bind and unbind the buffer so it gets allocated by the driver
     glBindBuffer(GL_SHADER_STORAGE_BUFFER, buffer.handle);
     glBindBuffer(GL_SHADER_STORAGE_BUFFER, 0);
-    LabelGLObject(GL_BUFFER, buffer.handle, addr, "GlobalMemory");
+    LabelGLObject(GL_BUFFER, buffer.handle, cpu_addr, "GlobalMemory");
 }
 
 void CachedGlobalRegion::Reload(u32 size_) {
@@ -35,10 +35,10 @@ void CachedGlobalRegion::Reload(u32 size_) {
 
     // TODO(Rodrigo): Get rid of Memory::GetPointer with a staging buffer
     glBindBuffer(GL_SHADER_STORAGE_BUFFER, buffer.handle);
-    glBufferData(GL_SHADER_STORAGE_BUFFER, size, Memory::GetPointer(addr), GL_DYNAMIC_DRAW);
+    glBufferData(GL_SHADER_STORAGE_BUFFER, size, GetHostPtr(), GL_DYNAMIC_DRAW);
 }
 
-GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(VAddr addr, u32 size) const {
+GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const {
     const auto search{reserve.find(addr)};
     if (search == reserve.end()) {
         return {};
@@ -46,11 +46,14 @@ GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(VAddr addr, u32
     return search->second;
 }
 
-GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(VAddr addr, u32 size) {
-    GlobalRegion region{TryGetReservedGlobalRegion(addr, size)};
+GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(Tegra::GPUVAddr addr, u32 size,
+                                                              u8* host_ptr) {
+    GlobalRegion region{TryGetReservedGlobalRegion(ToCacheAddr(host_ptr), size)};
     if (!region) {
         // No reserved surface available, create a new one and reserve it
-        region = std::make_shared<CachedGlobalRegion>(addr, size);
+        auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
+        const auto cpu_addr = *memory_manager.GpuToCpuAddress(addr);
+        region = std::make_shared<CachedGlobalRegion>(cpu_addr, size, host_ptr);
         ReserveGlobalRegion(region);
     }
     region->Reload(size);
@@ -58,7 +61,7 @@ GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(VAddr addr, u32 si
 }
 
 void GlobalRegionCacheOpenGL::ReserveGlobalRegion(GlobalRegion region) {
-    reserve.insert_or_assign(region->GetAddr(), std::move(region));
+    reserve.insert_or_assign(region->GetCacheAddr(), std::move(region));
 }
 
 GlobalRegionCacheOpenGL::GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer)
@@ -69,22 +72,20 @@ GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion(
     Tegra::Engines::Maxwell3D::Regs::ShaderStage stage) {
 
     auto& gpu{Core::System::GetInstance().GPU()};
-    const auto cbufs = gpu.Maxwell3D().state.shader_stages[static_cast<u64>(stage)];
-    const auto cbuf_addr = gpu.MemoryManager().GpuToCpuAddress(
-        cbufs.const_buffers[global_region.GetCbufIndex()].address + global_region.GetCbufOffset());
-    ASSERT(cbuf_addr);
-
-    const auto actual_addr_gpu = Memory::Read64(*cbuf_addr);
-    const auto size = Memory::Read32(*cbuf_addr + 8);
-    const auto actual_addr = gpu.MemoryManager().GpuToCpuAddress(actual_addr_gpu);
-    ASSERT(actual_addr);
+    auto& memory_manager{gpu.MemoryManager()};
+    const auto cbufs{gpu.Maxwell3D().state.shader_stages[static_cast<u64>(stage)]};
+    const auto addr{cbufs.const_buffers[global_region.GetCbufIndex()].address +
+                    global_region.GetCbufOffset()};
+    const auto actual_addr{memory_manager.Read64(addr)};
+    const auto size{memory_manager.Read32(addr + 8)};
 
     // Look up global region in the cache based on address
-    GlobalRegion region = TryGet(*actual_addr);
+    const auto& host_ptr{memory_manager.GetPointer(actual_addr)};
+    GlobalRegion region{TryGet(host_ptr)};
 
     if (!region) {
         // No global region found - create a new one
-        region = GetUncachedGlobalRegion(*actual_addr, size);
+        region = GetUncachedGlobalRegion(actual_addr, size, host_ptr);
         Register(region);
     }
 

src/video_core/renderer_opengl/gl_global_cache.h

@@ -27,14 +27,12 @@ using GlobalRegion = std::shared_ptr<CachedGlobalRegion>;
 
 class CachedGlobalRegion final : public RasterizerCacheObject {
 public:
-    explicit CachedGlobalRegion(VAddr addr, u32 size);
+    explicit CachedGlobalRegion(VAddr cpu_addr, u32 size, u8* host_ptr);
 
-    /// Gets the address of the shader in guest memory, required for cache management
-    VAddr GetAddr() const override {
-        return addr;
+    VAddr GetCpuAddr() const override {
+        return cpu_addr;
     }
 
-    /// Gets the size of the shader in guest memory, required for cache management
     std::size_t GetSizeInBytes() const override {
         return size;
     }
@@ -53,9 +51,8 @@ public:
     }
 
 private:
-    VAddr addr{};
+    VAddr cpu_addr{};
     u32 size{};
-
     OGLBuffer buffer;
 };
 
@@ -68,11 +65,11 @@ public:
                                  Tegra::Engines::Maxwell3D::Regs::ShaderStage stage);
 
 private:
-    GlobalRegion TryGetReservedGlobalRegion(VAddr addr, u32 size) const;
-    GlobalRegion GetUncachedGlobalRegion(VAddr addr, u32 size);
+    GlobalRegion TryGetReservedGlobalRegion(CacheAddr addr, u32 size) const;
+    GlobalRegion GetUncachedGlobalRegion(Tegra::GPUVAddr addr, u32 size, u8* host_ptr);
     void ReserveGlobalRegion(GlobalRegion region);
 
-    std::unordered_map<VAddr, GlobalRegion> reserve;
+    std::unordered_map<CacheAddr, GlobalRegion> reserve;
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_primitive_assembler.cpp

@@ -46,10 +46,7 @@ GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size
     auto [dst_pointer, index_offset] = buffer_cache.ReserveMemory(map_size);
 
     auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
-    const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
-    ASSERT_MSG(cpu_addr, "Invalid GPU address");
-
-    const u8* source{Memory::GetPointer(*cpu_addr)};
+    const u8* source{memory_manager.GetPointer(gpu_addr)};
 
     for (u32 primitive = 0; primitive < count / 4; ++primitive) {
         for (std::size_t i = 0; i < TRIANGLES_PER_QUAD; ++i) {
@@ -64,4 +61,4 @@ GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size
     return index_offset;
 }
 
-} // namespace OpenGL
+} // namespace OpenGL

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -449,7 +449,7 @@ static constexpr auto RangeFromInterval(Map& map, const Interval& interval) {
     return boost::make_iterator_range(map.equal_range(interval));
 }
 
-void RasterizerOpenGL::UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta) {
+void RasterizerOpenGL::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {
     const u64 page_start{addr >> Memory::PAGE_BITS};
     const u64 page_end{(addr + size + Memory::PAGE_SIZE - 1) >> Memory::PAGE_BITS};
 
@@ -747,20 +747,26 @@ void RasterizerOpenGL::DrawArrays() {
 
 void RasterizerOpenGL::FlushAll() {}
 
-void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    if (!addr || !size) {
+        return;
+    }
     res_cache.FlushRegion(addr, size);
 }
 
-void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    if (!addr || !size) {
+        return;
+    }
     res_cache.InvalidateRegion(addr, size);
     shader_cache.InvalidateRegion(addr, size);
     global_cache.InvalidateRegion(addr, size);
     buffer_cache.InvalidateRegion(addr, size);
 }
 
-void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     FlushRegion(addr, size);
     InvalidateRegion(addr, size);
 }
@@ -782,7 +788,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
 
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
 
-    const auto& surface{res_cache.TryFindFramebufferSurface(framebuffer_addr)};
+    const auto& surface{res_cache.TryFindFramebufferSurface(Memory::GetPointer(framebuffer_addr))};
     if (!surface) {
         return {};
     }

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -57,9 +57,9 @@ public:
     void DrawArrays() override;
     void Clear() override;
     void FlushAll() override;
-    void FlushRegion(VAddr addr, u64 size) override;
-    void InvalidateRegion(VAddr addr, u64 size) override;
-    void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
+    void FlushRegion(CacheAddr addr, u64 size) override;
+    void InvalidateRegion(CacheAddr addr, u64 size) override;
+    void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
     bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                const Tegra::Engines::Fermi2D::Regs::Surface& dst,
                                const Common::Rectangle<u32>& src_rect,
@@ -67,7 +67,7 @@ public:
     bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
                            u32 pixel_stride) override;
     bool AccelerateDrawBatch(bool is_indexed) override;
-    void UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta) override;
+    void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) override;
     void LoadDiskResources(const std::atomic_bool& stop_loading,
                            const VideoCore::DiskResourceLoadCallback& callback) override;
 

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -57,10 +57,9 @@ static void ApplyTextureDefaults(GLuint texture, u32 max_mip_level) {
 
 void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) {
     auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
-    const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr_)};
 
-    addr = cpu_addr ? *cpu_addr : 0;
     gpu_addr = gpu_addr_;
+    host_ptr = memory_manager.GetPointer(gpu_addr_);
     size_in_bytes = SizeInBytesRaw();
 
     if (IsPixelFormatASTC(pixel_format)) {
@@ -446,7 +445,7 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params,
             MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level),
                           params.MipBlockHeight(mip_level), params.MipHeight(mip_level),
                           params.MipBlockDepth(mip_level), 1, params.tile_width_spacing,
-                          gl_buffer.data() + offset_gl, gl_size, params.addr + offset);
+                          gl_buffer.data() + offset_gl, params.host_ptr + offset);
             offset += layer_size;
             offset_gl += gl_size;
         }
@@ -455,7 +454,7 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params,
         MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level),
                       params.MipBlockHeight(mip_level), params.MipHeight(mip_level),
                       params.MipBlockDepth(mip_level), depth, params.tile_width_spacing,
-                      gl_buffer.data(), gl_buffer.size(), params.addr + offset);
+                      gl_buffer.data(), params.host_ptr + offset);
     }
 }
 
@@ -513,9 +512,9 @@ void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surfac
                               "reinterpretation but the texture is tiled.");
         }
         const std::size_t remaining_size = dst_params.size_in_bytes - src_params.size_in_bytes;
-
+        auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
         glBufferSubData(GL_PIXEL_PACK_BUFFER, src_params.size_in_bytes, remaining_size,
-                        Memory::GetPointer(dst_params.addr + src_params.size_in_bytes));
+                        memory_manager.GetPointer(dst_params.gpu_addr + src_params.size_in_bytes));
     }
 
     glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
@@ -563,8 +562,8 @@ void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surfac
 }
 
 CachedSurface::CachedSurface(const SurfaceParams& params)
-    : params(params), gl_target(SurfaceTargetToGL(params.target)),
-      cached_size_in_bytes(params.size_in_bytes) {
+    : params{params}, gl_target{SurfaceTargetToGL(params.target)},
+      cached_size_in_bytes{params.size_in_bytes}, RasterizerCacheObject{params.host_ptr} {
     texture.Create(gl_target);
 
     // TODO(Rodrigo): Using params.GetRect() returns a different size than using its Mip*(0)
@@ -603,7 +602,7 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
 
     ApplyTextureDefaults(texture.handle, params.max_mip_level);
 
-    OpenGL::LabelGLObject(GL_TEXTURE, texture.handle, params.addr, params.IdentityString());
+    OpenGL::LabelGLObject(GL_TEXTURE, texture.handle, params.gpu_addr, params.IdentityString());
 
     // Clamp size to mapped GPU memory region
     // TODO(bunnei): Super Mario Odyssey maps a 0x40000 byte region and then uses it for a 0x80000
@@ -616,6 +615,8 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
         LOG_ERROR(HW_GPU, "Surface size {} exceeds region size {}", params.size_in_bytes, max_size);
         cached_size_in_bytes = max_size;
     }
+
+    cpu_addr = *memory_manager.GpuToCpuAddress(params.gpu_addr);
 }
 
 MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 192, 64));
@@ -633,10 +634,9 @@ void CachedSurface::LoadGLBuffer() {
         const u32 bpp = params.GetFormatBpp() / 8;
         const u32 copy_size = params.width * bpp;
         if (params.pitch == copy_size) {
-            std::memcpy(gl_buffer[0].data(), Memory::GetPointer(params.addr),
-                        params.size_in_bytes_gl);
+            std::memcpy(gl_buffer[0].data(), params.host_ptr, params.size_in_bytes_gl);
         } else {
-            const u8* start = Memory::GetPointer(params.addr);
+            const u8* start{params.host_ptr};
             u8* write_to = gl_buffer[0].data();
             for (u32 h = params.height; h > 0; h--) {
                 std::memcpy(write_to, start, copy_size);
@@ -680,8 +680,6 @@ void CachedSurface::FlushGLBuffer() {
     glPixelStorei(GL_PACK_ROW_LENGTH, 0);
     Tegra::Texture::ConvertFromHostToGuest(gl_buffer[0].data(), params.pixel_format, params.width,
                                            params.height, params.depth, true, true);
-    const u8* const texture_src_data = Memory::GetPointer(params.addr);
-    ASSERT(texture_src_data);
     if (params.is_tiled) {
         ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}",
                    params.block_width, static_cast<u32>(params.target));
@@ -691,9 +689,9 @@ void CachedSurface::FlushGLBuffer() {
         const u32 bpp = params.GetFormatBpp() / 8;
         const u32 copy_size = params.width * bpp;
         if (params.pitch == copy_size) {
-            std::memcpy(Memory::GetPointer(params.addr), gl_buffer[0].data(), GetSizeInBytes());
+            std::memcpy(params.host_ptr, gl_buffer[0].data(), GetSizeInBytes());
         } else {
-            u8* start = Memory::GetPointer(params.addr);
+            u8* start{params.host_ptr};
             const u8* read_to = gl_buffer[0].data();
             for (u32 h = params.height; h > 0; h--) {
                 std::memcpy(start, read_to, copy_size);
@@ -927,12 +925,12 @@ void RasterizerCacheOpenGL::LoadSurface(const Surface& surface) {
 }
 
 Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool preserve_contents) {
-    if (params.addr == 0 || params.height * params.width == 0) {
+    if (params.gpu_addr == 0 || params.height * params.width == 0) {
         return {};
     }
 
     // Look up surface in the cache based on address
-    Surface surface{TryGet(params.addr)};
+    Surface surface{TryGet(params.host_ptr)};
     if (surface) {
         if (surface->GetSurfaceParams().IsCompatibleSurface(params)) {
             // Use the cached surface as-is unless it's not synced with memory
@@ -981,14 +979,16 @@ void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface,
                                                    const Surface& dst_surface) {
     const auto& init_params{src_surface->GetSurfaceParams()};
     const auto& dst_params{dst_surface->GetSurfaceParams()};
-    VAddr address = init_params.addr;
-    const std::size_t layer_size = dst_params.LayerMemorySize();
+    auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
+    Tegra::GPUVAddr address{init_params.gpu_addr};
+    const std::size_t layer_size{dst_params.LayerMemorySize()};
     for (u32 layer = 0; layer < dst_params.depth; layer++) {
         for (u32 mipmap = 0; mipmap < dst_params.max_mip_level; mipmap++) {
-            const VAddr sub_address = address + dst_params.GetMipmapLevelOffset(mipmap);
-            const Surface& copy = TryGet(sub_address);
-            if (!copy)
+            const Tegra::GPUVAddr sub_address{address + dst_params.GetMipmapLevelOffset(mipmap)};
+            const Surface& copy{TryGet(memory_manager.GetPointer(sub_address))};
+            if (!copy) {
                 continue;
+            }
             const auto& src_params{copy->GetSurfaceParams()};
             const u32 width{std::min(src_params.width, dst_params.MipWidth(mipmap))};
             const u32 height{std::min(src_params.height, dst_params.MipHeight(mipmap))};
@@ -1163,7 +1163,8 @@ void RasterizerCacheOpenGL::AccurateCopySurface(const Surface& src_surface,
     const auto& dst_params{dst_surface->GetSurfaceParams()};
 
     // Flush enough memory for both the source and destination surface
-    FlushRegion(src_params.addr, std::max(src_params.MemorySize(), dst_params.MemorySize()));
+    FlushRegion(ToCacheAddr(src_params.host_ptr),
+                std::max(src_params.MemorySize(), dst_params.MemorySize()));
 
     LoadSurface(dst_surface);
 }
@@ -1215,8 +1216,8 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface,
     return new_surface;
 }
 
-Surface RasterizerCacheOpenGL::TryFindFramebufferSurface(VAddr addr) const {
-    return TryGet(addr);
+Surface RasterizerCacheOpenGL::TryFindFramebufferSurface(const u8* host_ptr) const {
+    return TryGet(host_ptr);
 }
 
 void RasterizerCacheOpenGL::ReserveSurface(const Surface& surface) {
@@ -1243,9 +1244,10 @@ static std::optional<u32> TryFindBestMipMap(std::size_t memory, const SurfacePar
     return {};
 }
 
-static std::optional<u32> TryFindBestLayer(VAddr addr, const SurfaceParams params, u32 mipmap) {
-    const std::size_t size = params.LayerMemorySize();
-    VAddr start = params.addr + params.GetMipmapLevelOffset(mipmap);
+static std::optional<u32> TryFindBestLayer(Tegra::GPUVAddr addr, const SurfaceParams params,
+                                           u32 mipmap) {
+    const std::size_t size{params.LayerMemorySize()};
+    Tegra::GPUVAddr start{params.gpu_addr + params.GetMipmapLevelOffset(mipmap)};
     for (u32 i = 0; i < params.depth; i++) {
         if (start == addr) {
             return {i};
@@ -1267,7 +1269,7 @@ static bool LayerFitReinterpretSurface(RasterizerCacheOpenGL& cache, const Surfa
             src_params.height == dst_params.MipHeight(*level) &&
             src_params.block_height >= dst_params.MipBlockHeight(*level)) {
             const std::optional<u32> slot =
-                TryFindBestLayer(render_surface->GetAddr(), dst_params, *level);
+                TryFindBestLayer(render_surface->GetSurfaceParams().gpu_addr, dst_params, *level);
             if (slot.has_value()) {
                 glCopyImageSubData(render_surface->Texture().handle,
                                    SurfaceTargetToGL(src_params.target), 0, 0, 0, 0,
@@ -1283,8 +1285,8 @@ static bool LayerFitReinterpretSurface(RasterizerCacheOpenGL& cache, const Surfa
 }
 
 static bool IsReinterpretInvalid(const Surface render_surface, const Surface blitted_surface) {
-    const VAddr bound1 = blitted_surface->GetAddr() + blitted_surface->GetMemorySize();
-    const VAddr bound2 = render_surface->GetAddr() + render_surface->GetMemorySize();
+    const VAddr bound1 = blitted_surface->GetCpuAddr() + blitted_surface->GetMemorySize();
+    const VAddr bound2 = render_surface->GetCpuAddr() + render_surface->GetMemorySize();
     if (bound2 > bound1)
         return true;
     const auto& dst_params = blitted_surface->GetSurfaceParams();
@@ -1327,7 +1329,8 @@ void RasterizerCacheOpenGL::SignalPreDrawCall() {
 void RasterizerCacheOpenGL::SignalPostDrawCall() {
     for (u32 i = 0; i < Maxwell::NumRenderTargets; i++) {
         if (current_color_buffers[i] != nullptr) {
-            Surface intersect = CollideOnReinterpretedSurface(current_color_buffers[i]->GetAddr());
+            Surface intersect =
+                CollideOnReinterpretedSurface(current_color_buffers[i]->GetCacheAddr());
             if (intersect != nullptr) {
                 PartialReinterpretSurface(current_color_buffers[i], intersect);
                 texception = true;

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -296,7 +296,7 @@ struct SurfaceParams {
     bool is_array;
     bool srgb_conversion;
     // Parameters used for caching
-    VAddr addr;
+    u8* host_ptr;
     Tegra::GPUVAddr gpu_addr;
     std::size_t size_in_bytes;
     std::size_t size_in_bytes_gl;
@@ -345,10 +345,10 @@ class RasterizerOpenGL;
 
 class CachedSurface final : public RasterizerCacheObject {
 public:
-    CachedSurface(const SurfaceParams& params);
+    explicit CachedSurface(const SurfaceParams& params);
 
-    VAddr GetAddr() const override {
-        return params.addr;
+    VAddr GetCpuAddr() const override {
+        return cpu_addr;
     }
 
     std::size_t GetSizeInBytes() const override {
@@ -432,6 +432,7 @@ private:
     std::size_t memory_size;
     bool reinterpreted = false;
     bool must_reload = false;
+    VAddr cpu_addr{};
 };
 
 class RasterizerCacheOpenGL final : public RasterizerCache<Surface> {
@@ -449,7 +450,7 @@ public:
     Surface GetColorBufferSurface(std::size_t index, bool preserve_contents);
 
     /// Tries to find a framebuffer using on the provided CPU address
-    Surface TryFindFramebufferSurface(VAddr addr) const;
+    Surface TryFindFramebufferSurface(const u8* host_ptr) const;
 
     /// Copies the contents of one surface to another
     void FermiCopySurface(const Tegra::Engines::Fermi2D::Regs::Surface& src_config,
@@ -506,12 +507,12 @@ private:
     std::array<Surface, Maxwell::NumRenderTargets> current_color_buffers;
     Surface last_depth_buffer;
 
-    using SurfaceIntervalCache = boost::icl::interval_map<VAddr, Surface>;
+    using SurfaceIntervalCache = boost::icl::interval_map<CacheAddr, Surface>;
     using SurfaceInterval = typename SurfaceIntervalCache::interval_type;
 
     static auto GetReinterpretInterval(const Surface& object) {
-        return SurfaceInterval::right_open(object->GetAddr() + 1,
-                                           object->GetAddr() + object->GetMemorySize() - 1);
+        return SurfaceInterval::right_open(object->GetCacheAddr() + 1,
+                                           object->GetCacheAddr() + object->GetMemorySize() - 1);
     }
 
     // Reinterpreted surfaces are very fragil as the game may keep rendering into them.
@@ -523,7 +524,7 @@ private:
         reinterpret_surface->MarkReinterpreted();
     }
 
-    Surface CollideOnReinterpretedSurface(VAddr addr) const {
+    Surface CollideOnReinterpretedSurface(CacheAddr addr) const {
         const SurfaceInterval interval{addr};
         for (auto& pair :
              boost::make_iterator_range(reinterpreted_surfaces.equal_range(interval))) {

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -32,19 +32,16 @@ struct UnspecializedShader {
 namespace {
 
 /// Gets the address for the specified shader stage program
-VAddr GetShaderAddress(Maxwell::ShaderProgram program) {
-    const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
-    const auto& shader_config = gpu.regs.shader_config[static_cast<std::size_t>(program)];
-    const auto address = gpu.memory_manager.GpuToCpuAddress(gpu.regs.code_address.CodeAddress() +
-                                                            shader_config.offset);
-    ASSERT_MSG(address, "Invalid GPU address");
-    return *address;
+Tegra::GPUVAddr GetShaderAddress(Maxwell::ShaderProgram program) {
+    const auto& gpu{Core::System::GetInstance().GPU().Maxwell3D()};
+    const auto& shader_config{gpu.regs.shader_config[static_cast<std::size_t>(program)]};
+    return gpu.regs.code_address.CodeAddress() + shader_config.offset;
 }
 
 /// Gets the shader program code from memory for the specified address
-ProgramCode GetShaderCode(VAddr addr) {
+ProgramCode GetShaderCode(const u8* host_ptr) {
     ProgramCode program_code(VideoCommon::Shader::MAX_PROGRAM_LENGTH);
-    Memory::ReadBlock(addr, program_code.data(), program_code.size() * sizeof(u64));
+    std::memcpy(program_code.data(), host_ptr, program_code.size() * sizeof(u64));
     return program_code;
 }
 
@@ -214,12 +211,13 @@ std::set<GLenum> GetSupportedFormats() {
 
 } // namespace
 
-CachedShader::CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderProgram program_type,
-                           ShaderDiskCacheOpenGL& disk_cache,
+CachedShader::CachedShader(VAddr cpu_addr, u64 unique_identifier,
+                           Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache,
                            const PrecompiledPrograms& precompiled_programs,
-                           ProgramCode&& program_code, ProgramCode&& program_code_b)
-    : addr{addr}, unique_identifier{unique_identifier}, program_type{program_type},
-      disk_cache{disk_cache}, precompiled_programs{precompiled_programs} {
+                           ProgramCode&& program_code, ProgramCode&& program_code_b, u8* host_ptr)
+    : host_ptr{host_ptr}, cpu_addr{cpu_addr}, unique_identifier{unique_identifier},
+      program_type{program_type}, disk_cache{disk_cache},
+      precompiled_programs{precompiled_programs}, RasterizerCacheObject{host_ptr} {
 
     const std::size_t code_size = CalculateProgramSize(program_code);
     const std::size_t code_size_b =
@@ -243,12 +241,13 @@ CachedShader::CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderPro
     disk_cache.SaveRaw(raw);
 }
 
-CachedShader::CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderProgram program_type,
-                           ShaderDiskCacheOpenGL& disk_cache,
+CachedShader::CachedShader(VAddr cpu_addr, u64 unique_identifier,
+                           Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache,
                            const PrecompiledPrograms& precompiled_programs,
-                           GLShader::ProgramResult result)
-    : addr{addr}, unique_identifier{unique_identifier}, program_type{program_type},
-      disk_cache{disk_cache}, precompiled_programs{precompiled_programs} {
+                           GLShader::ProgramResult result, u8* host_ptr)
+    : cpu_addr{cpu_addr}, unique_identifier{unique_identifier}, program_type{program_type},
+      disk_cache{disk_cache}, precompiled_programs{precompiled_programs}, RasterizerCacheObject{
+                                                                              host_ptr} {
 
     code = std::move(result.first);
     entries = result.second;
@@ -271,7 +270,7 @@ std::tuple<GLuint, BaseBindings> CachedShader::GetProgramHandle(GLenum primitive
                 disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings));
             }
 
-            LabelGLObject(GL_PROGRAM, program->handle, addr);
+            LabelGLObject(GL_PROGRAM, program->handle, cpu_addr);
         }
 
         handle = program->handle;
@@ -323,7 +322,7 @@ GLuint CachedShader::LazyGeometryProgram(CachedProgram& target_program, BaseBind
         disk_cache.SaveUsage(GetUsage(primitive_mode, base_bindings));
     }
 
-    LabelGLObject(GL_PROGRAM, target_program->handle, addr, debug_name);
+    LabelGLObject(GL_PROGRAM, target_program->handle, cpu_addr, debug_name);
 
     return target_program->handle;
 };
@@ -486,29 +485,32 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
         return last_shaders[static_cast<u32>(program)];
     }
 
-    const VAddr program_addr{GetShaderAddress(program)};
+    auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
+    const Tegra::GPUVAddr program_addr{GetShaderAddress(program)};
 
     // Look up shader in the cache based on address
-    Shader shader{TryGet(program_addr)};
+    const auto& host_ptr{memory_manager.GetPointer(program_addr)};
+    Shader shader{TryGet(host_ptr)};
 
     if (!shader) {
         // No shader found - create a new one
-        ProgramCode program_code = GetShaderCode(program_addr);
+        ProgramCode program_code{GetShaderCode(host_ptr)};
         ProgramCode program_code_b;
         if (program == Maxwell::ShaderProgram::VertexA) {
-            program_code_b = GetShaderCode(GetShaderAddress(Maxwell::ShaderProgram::VertexB));
+            program_code_b = GetShaderCode(
+                memory_manager.GetPointer(GetShaderAddress(Maxwell::ShaderProgram::VertexB)));
         }
         const u64 unique_identifier = GetUniqueIdentifier(program, program_code, program_code_b);
-
+        const VAddr cpu_addr{*memory_manager.GpuToCpuAddress(program_addr)};
         const auto found = precompiled_shaders.find(unique_identifier);
         if (found != precompiled_shaders.end()) {
             shader =
-                std::make_shared<CachedShader>(program_addr, unique_identifier, program, disk_cache,
-                                               precompiled_programs, found->second);
+                std::make_shared<CachedShader>(cpu_addr, unique_identifier, program, disk_cache,
+                                               precompiled_programs, found->second, host_ptr);
         } else {
             shader = std::make_shared<CachedShader>(
-                program_addr, unique_identifier, program, disk_cache, precompiled_programs,
-                std::move(program_code), std::move(program_code_b));
+                cpu_addr, unique_identifier, program, disk_cache, precompiled_programs,
+                std::move(program_code), std::move(program_code_b), host_ptr);
         }
         Register(shader);
     }

src/video_core/renderer_opengl/gl_shader_cache.h

@@ -39,18 +39,18 @@ using PrecompiledShaders = std::unordered_map<u64, GLShader::ProgramResult>;
 
 class CachedShader final : public RasterizerCacheObject {
 public:
-    explicit CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderProgram program_type,
-                          ShaderDiskCacheOpenGL& disk_cache,
+    explicit CachedShader(VAddr cpu_addr, u64 unique_identifier,
+                          Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache,
                           const PrecompiledPrograms& precompiled_programs,
-                          ProgramCode&& program_code, ProgramCode&& program_code_b);
+                          ProgramCode&& program_code, ProgramCode&& program_code_b, u8* host_ptr);
 
-    explicit CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderProgram program_type,
-                          ShaderDiskCacheOpenGL& disk_cache,
+    explicit CachedShader(VAddr cpu_addr, u64 unique_identifier,
+                          Maxwell::ShaderProgram program_type, ShaderDiskCacheOpenGL& disk_cache,
                           const PrecompiledPrograms& precompiled_programs,
-                          GLShader::ProgramResult result);
+                          GLShader::ProgramResult result, u8* host_ptr);
 
-    VAddr GetAddr() const override {
-        return addr;
+    VAddr GetCpuAddr() const override {
+        return cpu_addr;
     }
 
     std::size_t GetSizeInBytes() const override {
@@ -91,7 +91,8 @@ private:
 
     ShaderDiskCacheUsage GetUsage(GLenum primitive_mode, BaseBindings base_bindings) const;
 
-    VAddr addr{};
+    u8* host_ptr{};
+    VAddr cpu_addr{};
     u64 unique_identifier{};
     Maxwell::ShaderProgram program_type{};
     ShaderDiskCacheOpenGL& disk_cache;

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -164,12 +164,13 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf
         // Reset the screen info's display texture to its own permanent texture
         screen_info.display_texture = screen_info.texture.resource.handle;
 
-        Memory::RasterizerFlushVirtualRegion(framebuffer_addr, size_in_bytes,
-                                             Memory::FlushMode::Flush);
+        rasterizer->FlushRegion(ToCacheAddr(Memory::GetPointer(framebuffer_addr)), size_in_bytes);
 
-        VideoCore::MortonCopyPixels128(framebuffer.width, framebuffer.height, bytes_per_pixel, 4,
-                                       Memory::GetPointer(framebuffer_addr),
-                                       gl_framebuffer_data.data(), true);
+        constexpr u32 linear_bpp = 4;
+        VideoCore::MortonCopyPixels128(VideoCore::MortonSwizzleMode::MortonToLinear,
+                                       framebuffer.width, framebuffer.height, bytes_per_pixel,
+                                       linear_bpp, Memory::GetPointer(framebuffer_addr),
+                                       gl_framebuffer_data.data());
 
         glPixelStorei(GL_UNPACK_ROW_LENGTH, static_cast<GLint>(framebuffer.stride));
 

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -17,6 +17,11 @@
 
 namespace Vulkan {
 
+CachedBufferEntry::CachedBufferEntry(VAddr cpu_addr, std::size_t size, u64 offset,
+                                     std::size_t alignment, u8* host_ptr)
+    : cpu_addr{cpu_addr}, size{size}, offset{offset}, alignment{alignment}, RasterizerCacheObject{
+                                                                                host_ptr} {}
+
 VKBufferCache::VKBufferCache(Tegra::MemoryManager& tegra_memory_manager,
                              VideoCore::RasterizerInterface& rasterizer, const VKDevice& device,
                              VKMemoryManager& memory_manager, VKScheduler& scheduler, u64 size)
@@ -37,16 +42,18 @@ VKBufferCache::~VKBufferCache() = default;
 u64 VKBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size, u64 alignment,
                                 bool cache) {
     const auto cpu_addr{tegra_memory_manager.GpuToCpuAddress(gpu_addr)};
-    ASSERT(cpu_addr);
+    ASSERT_MSG(cpu_addr, "Invalid GPU address");
 
     // Cache management is a big overhead, so only cache entries with a given size.
     // TODO: Figure out which size is the best for given games.
     cache &= size >= 2048;
 
+    const auto& host_ptr{Memory::GetPointer(*cpu_addr)};
     if (cache) {
-        if (auto entry = TryGet(*cpu_addr); entry) {
-            if (entry->size >= size && entry->alignment == alignment) {
-                return entry->offset;
+        auto entry = TryGet(host_ptr);
+        if (entry) {
+            if (entry->GetSize() >= size && entry->GetAlignment() == alignment) {
+                return entry->GetOffset();
             }
             Unregister(entry);
         }
@@ -55,17 +62,17 @@ u64 VKBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size, u64
     AlignBuffer(alignment);
     const u64 uploaded_offset = buffer_offset;
 
-    Memory::ReadBlock(*cpu_addr, buffer_ptr, size);
+    if (!host_ptr) {
+        return uploaded_offset;
+    }
 
+    std::memcpy(buffer_ptr, host_ptr, size);
     buffer_ptr += size;
     buffer_offset += size;
 
     if (cache) {
-        auto entry = std::make_shared<CachedBufferEntry>();
-        entry->offset = uploaded_offset;
-        entry->size = size;
-        entry->alignment = alignment;
-        entry->addr = *cpu_addr;
+        auto entry = std::make_shared<CachedBufferEntry>(*cpu_addr, size, uploaded_offset,
+                                                         alignment, host_ptr);
         Register(entry);
     }
 

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -24,22 +24,39 @@ class VKFence;
 class VKMemoryManager;
 class VKStreamBuffer;
 
-struct CachedBufferEntry final : public RasterizerCacheObject {
-    VAddr GetAddr() const override {
-        return addr;
+class CachedBufferEntry final : public RasterizerCacheObject {
+public:
+    explicit CachedBufferEntry(VAddr cpu_addr, std::size_t size, u64 offset, std::size_t alignment,
+                               u8* host_ptr);
+
+    VAddr GetCpuAddr() const override {
+        return cpu_addr;
     }
 
     std::size_t GetSizeInBytes() const override {
         return size;
     }
 
+    std::size_t GetSize() const {
+        return size;
+    }
+
+    u64 GetOffset() const {
+        return offset;
+    }
+
+    std::size_t GetAlignment() const {
+        return alignment;
+    }
+
     // We do not have to flush this cache as things in it are never modified by us.
     void Flush() override {}
 
-    VAddr addr;
-    std::size_t size;
-    u64 offset;
-    std::size_t alignment;
+private:
+    VAddr cpu_addr{};
+    std::size_t size{};
+    u64 offset{};
+    std::size_t alignment{};
 };
 
 class VKBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> {

src/video_core/renderer_vulkan/vk_sampler_cache.cpp

@@ -0,0 +1,81 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstring>
+#include <optional>
+#include <unordered_map>
+
+#include "common/assert.h"
+#include "common/cityhash.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/maxwell_to_vk.h"
+#include "video_core/renderer_vulkan/vk_sampler_cache.h"
+#include "video_core/textures/texture.h"
+
+namespace Vulkan {
+
+static std::optional<vk::BorderColor> TryConvertBorderColor(std::array<float, 4> color) {
+    // TODO(Rodrigo): Manage integer border colors
+    if (color == std::array<float, 4>{0, 0, 0, 0}) {
+        return vk::BorderColor::eFloatTransparentBlack;
+    } else if (color == std::array<float, 4>{0, 0, 0, 1}) {
+        return vk::BorderColor::eFloatOpaqueBlack;
+    } else if (color == std::array<float, 4>{1, 1, 1, 1}) {
+        return vk::BorderColor::eFloatOpaqueWhite;
+    } else {
+        return {};
+    }
+}
+
+std::size_t SamplerCacheKey::Hash() const {
+    static_assert(sizeof(raw) % sizeof(u64) == 0);
+    return static_cast<std::size_t>(
+        Common::CityHash64(reinterpret_cast<const char*>(raw.data()), sizeof(raw) / sizeof(u64)));
+}
+
+bool SamplerCacheKey::operator==(const SamplerCacheKey& rhs) const {
+    return raw == rhs.raw;
+}
+
+VKSamplerCache::VKSamplerCache(const VKDevice& device) : device{device} {}
+
+VKSamplerCache::~VKSamplerCache() = default;
+
+vk::Sampler VKSamplerCache::GetSampler(const Tegra::Texture::TSCEntry& tsc) {
+    const auto [entry, is_cache_miss] = cache.try_emplace(SamplerCacheKey{tsc});
+    auto& sampler = entry->second;
+    if (is_cache_miss) {
+        sampler = CreateSampler(tsc);
+    }
+    return *sampler;
+}
+
+UniqueSampler VKSamplerCache::CreateSampler(const Tegra::Texture::TSCEntry& tsc) {
+    const float max_anisotropy = tsc.GetMaxAnisotropy();
+    const bool has_anisotropy = max_anisotropy > 1.0f;
+
+    const auto border_color = tsc.GetBorderColor();
+    const auto vk_border_color = TryConvertBorderColor(border_color);
+    UNIMPLEMENTED_IF_MSG(!vk_border_color, "Unimplemented border color {} {} {} {}",
+                         border_color[0], border_color[1], border_color[2], border_color[3]);
+
+    constexpr bool unnormalized_coords = false;
+
+    const vk::SamplerCreateInfo sampler_ci(
+        {}, MaxwellToVK::Sampler::Filter(tsc.mag_filter),
+        MaxwellToVK::Sampler::Filter(tsc.min_filter),
+        MaxwellToVK::Sampler::MipmapMode(tsc.mipmap_filter),
+        MaxwellToVK::Sampler::WrapMode(tsc.wrap_u), MaxwellToVK::Sampler::WrapMode(tsc.wrap_v),
+        MaxwellToVK::Sampler::WrapMode(tsc.wrap_p), tsc.GetLodBias(), has_anisotropy,
+        max_anisotropy, tsc.depth_compare_enabled,
+        MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func), tsc.GetMinLod(),
+        tsc.GetMaxLod(), vk_border_color.value_or(vk::BorderColor::eFloatTransparentBlack),
+        unnormalized_coords);
+
+    const auto& dld = device.GetDispatchLoader();
+    const auto dev = device.GetLogical();
+    return dev.createSamplerUnique(sampler_ci, nullptr, dld);
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_sampler_cache.h

@@ -0,0 +1,56 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <unordered_map>
+
+#include "common/common_types.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/textures/texture.h"
+
+namespace Vulkan {
+
+class VKDevice;
+
+struct SamplerCacheKey final : public Tegra::Texture::TSCEntry {
+    std::size_t Hash() const;
+
+    bool operator==(const SamplerCacheKey& rhs) const;
+
+    bool operator!=(const SamplerCacheKey& rhs) const {
+        return !operator==(rhs);
+    }
+};
+
+} // namespace Vulkan
+
+namespace std {
+
+template <>
+struct hash<Vulkan::SamplerCacheKey> {
+    std::size_t operator()(const Vulkan::SamplerCacheKey& k) const noexcept {
+        return k.Hash();
+    }
+};
+
+} // namespace std
+
+namespace Vulkan {
+
+class VKSamplerCache {
+public:
+    explicit VKSamplerCache(const VKDevice& device);
+    ~VKSamplerCache();
+
+    vk::Sampler GetSampler(const Tegra::Texture::TSCEntry& tsc);
+
+private:
+    UniqueSampler CreateSampler(const Tegra::Texture::TSCEntry& tsc);
+
+    const VKDevice& device;
+    std::unordered_map<SamplerCacheKey, UniqueSampler> cache;
+};
+
+} // namespace Vulkan

src/video_core/textures/decoders.cpp

@@ -6,7 +6,6 @@
 #include <cstring>
 #include "common/alignment.h"
 #include "common/assert.h"
-#include "core/memory.h"
 #include "video_core/gpu.h"
 #include "video_core/textures/decoders.h"
 #include "video_core/textures/texture.h"
@@ -230,18 +229,18 @@ u32 BytesPerPixel(TextureFormat format) {
     }
 }
 
-void UnswizzleTexture(u8* const unswizzled_data, VAddr address, u32 tile_size_x, u32 tile_size_y,
+void UnswizzleTexture(u8* const unswizzled_data, u8* address, u32 tile_size_x, u32 tile_size_y,
                       u32 bytes_per_pixel, u32 width, u32 height, u32 depth, u32 block_height,
                       u32 block_depth, u32 width_spacing) {
     CopySwizzledData((width + tile_size_x - 1) / tile_size_x,
                      (height + tile_size_y - 1) / tile_size_y, depth, bytes_per_pixel,
-                     bytes_per_pixel, Memory::GetPointer(address), unswizzled_data, true,
-                     block_height, block_depth, width_spacing);
+                     bytes_per_pixel, address, unswizzled_data, true, block_height, block_depth,
+                     width_spacing);
 }
 
-std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size_x, u32 tile_size_y,
-                                 u32 bytes_per_pixel, u32 width, u32 height, u32 depth,
-                                 u32 block_height, u32 block_depth, u32 width_spacing) {
+std::vector<u8> UnswizzleTexture(u8* address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel,
+                                 u32 width, u32 height, u32 depth, u32 block_height,
+                                 u32 block_depth, u32 width_spacing) {
     std::vector<u8> unswizzled_data(width * height * depth * bytes_per_pixel);
     UnswizzleTexture(unswizzled_data.data(), address, tile_size_x, tile_size_y, bytes_per_pixel,
                      width, height, depth, block_height, block_depth, width_spacing);
@@ -249,8 +248,7 @@ std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size_x, u32 tile_size_y
 }
 
 void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width,
-                    u32 bytes_per_pixel, VAddr swizzled_data, VAddr unswizzled_data,
-                    u32 block_height) {
+                    u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height) {
     const u32 image_width_in_gobs{(swizzled_width * bytes_per_pixel + (gob_size_x - 1)) /
                                   gob_size_x};
     for (u32 line = 0; line < subrect_height; ++line) {
@@ -262,17 +260,17 @@ void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32
             const u32 gob_address =
                 gob_address_y + (x * bytes_per_pixel / gob_size_x) * gob_size * block_height;
             const u32 swizzled_offset = gob_address + table[(x * bytes_per_pixel) % gob_size_x];
-            const VAddr source_line = unswizzled_data + line * source_pitch + x * bytes_per_pixel;
-            const VAddr dest_addr = swizzled_data + swizzled_offset;
+            u8* source_line = unswizzled_data + line * source_pitch + x * bytes_per_pixel;
+            u8* dest_addr = swizzled_data + swizzled_offset;
 
-            Memory::CopyBlock(dest_addr, source_line, bytes_per_pixel);
+            std::memcpy(dest_addr, source_line, bytes_per_pixel);
         }
     }
 }
 
 void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 swizzled_width,
-                      u32 bytes_per_pixel, VAddr swizzled_data, VAddr unswizzled_data,
-                      u32 block_height, u32 offset_x, u32 offset_y) {
+                      u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height,
+                      u32 offset_x, u32 offset_y) {
     for (u32 line = 0; line < subrect_height; ++line) {
         const u32 y2 = line + offset_y;
         const u32 gob_address_y = (y2 / (gob_size_y * block_height)) * gob_size * block_height +
@@ -282,10 +280,10 @@ void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32
             const u32 x2 = (x + offset_x) * bytes_per_pixel;
             const u32 gob_address = gob_address_y + (x2 / gob_size_x) * gob_size * block_height;
             const u32 swizzled_offset = gob_address + table[x2 % gob_size_x];
-            const VAddr dest_line = unswizzled_data + line * dest_pitch + x * bytes_per_pixel;
-            const VAddr source_addr = swizzled_data + swizzled_offset;
+            u8* dest_line = unswizzled_data + line * dest_pitch + x * bytes_per_pixel;
+            u8* source_addr = swizzled_data + swizzled_offset;
 
-            Memory::CopyBlock(dest_line, source_addr, bytes_per_pixel);
+            std::memcpy(dest_line, source_addr, bytes_per_pixel);
         }
     }
 }

src/video_core/textures/decoders.h

@@ -17,14 +17,14 @@ inline std::size_t GetGOBSize() {
 }
 
 /// Unswizzles a swizzled texture without changing its format.
-void UnswizzleTexture(u8* unswizzled_data, VAddr address, u32 tile_size_x, u32 tile_size_y,
+void UnswizzleTexture(u8* unswizzled_data, u8* address, u32 tile_size_x, u32 tile_size_y,
                       u32 bytes_per_pixel, u32 width, u32 height, u32 depth,
                       u32 block_height = TICEntry::DefaultBlockHeight,
                       u32 block_depth = TICEntry::DefaultBlockHeight, u32 width_spacing = 0);
 
 /// Unswizzles a swizzled texture without changing its format.
-std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size_x, u32 tile_size_y,
-                                 u32 bytes_per_pixel, u32 width, u32 height, u32 depth,
+std::vector<u8> UnswizzleTexture(u8* address, u32 tile_size_x, u32 tile_size_y, u32 bytes_per_pixel,
+                                 u32 width, u32 height, u32 depth,
                                  u32 block_height = TICEntry::DefaultBlockHeight,
                                  u32 block_depth = TICEntry::DefaultBlockHeight,
                                  u32 width_spacing = 0);
@@ -44,12 +44,11 @@ std::size_t CalculateSize(bool tiled, u32 bytes_per_pixel, u32 width, u32 height
 
 /// Copies an untiled subrectangle into a tiled surface.
 void SwizzleSubrect(u32 subrect_width, u32 subrect_height, u32 source_pitch, u32 swizzled_width,
-                    u32 bytes_per_pixel, VAddr swizzled_data, VAddr unswizzled_data,
-                    u32 block_height);
+                    u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height);
 
 /// Copies a tiled subrectangle into a linear surface.
 void UnswizzleSubrect(u32 subrect_width, u32 subrect_height, u32 dest_pitch, u32 swizzled_width,
-                      u32 bytes_per_pixel, VAddr swizzled_data, VAddr unswizzled_data,
-                      u32 block_height, u32 offset_x, u32 offset_y);
+                      u32 bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data, u32 block_height,
+                      u32 offset_x, u32 offset_y);
 
 } // namespace Tegra::Texture

src/video_core/textures/texture.h

@@ -283,31 +283,36 @@ enum class TextureMipmapFilter : u32 {
 
 struct TSCEntry {
     union {
-        BitField<0, 3, WrapMode> wrap_u;
-        BitField<3, 3, WrapMode> wrap_v;
-        BitField<6, 3, WrapMode> wrap_p;
-        BitField<9, 1, u32> depth_compare_enabled;
-        BitField<10, 3, DepthCompareFunc> depth_compare_func;
-        BitField<13, 1, u32> srgb_conversion;
-        BitField<20, 3, u32> max_anisotropy;
+        struct {
+            union {
+                BitField<0, 3, WrapMode> wrap_u;
+                BitField<3, 3, WrapMode> wrap_v;
+                BitField<6, 3, WrapMode> wrap_p;
+                BitField<9, 1, u32> depth_compare_enabled;
+                BitField<10, 3, DepthCompareFunc> depth_compare_func;
+                BitField<13, 1, u32> srgb_conversion;
+                BitField<20, 3, u32> max_anisotropy;
+            };
+            union {
+                BitField<0, 2, TextureFilter> mag_filter;
+                BitField<4, 2, TextureFilter> min_filter;
+                BitField<6, 2, TextureMipmapFilter> mipmap_filter;
+                BitField<9, 1, u32> cubemap_interface_filtering;
+                BitField<12, 13, u32> mip_lod_bias;
+            };
+            union {
+                BitField<0, 12, u32> min_lod_clamp;
+                BitField<12, 12, u32> max_lod_clamp;
+                BitField<24, 8, u32> srgb_border_color_r;
+            };
+            union {
+                BitField<12, 8, u32> srgb_border_color_g;
+                BitField<20, 8, u32> srgb_border_color_b;
+            };
+            std::array<f32, 4> border_color;
+        };
+        std::array<u8, 0x20> raw;
     };
-    union {
-        BitField<0, 2, TextureFilter> mag_filter;
-        BitField<4, 2, TextureFilter> min_filter;
-        BitField<6, 2, TextureMipmapFilter> mipmap_filter;
-        BitField<9, 1, u32> cubemap_interface_filtering;
-        BitField<12, 13, u32> mip_lod_bias;
-    };
-    union {
-        BitField<0, 12, u32> min_lod_clamp;
-        BitField<12, 12, u32> max_lod_clamp;
-        BitField<24, 8, u32> srgb_border_color_r;
-    };
-    union {
-        BitField<12, 8, u32> srgb_border_color_g;
-        BitField<20, 8, u32> srgb_border_color_b;
-    };
-    std::array<f32, 4> border_color;
 
     float GetMaxAnisotropy() const {
         return static_cast<float>(1U << max_anisotropy);

src/yuzu/bootmanager.cpp

@@ -24,8 +24,6 @@ void EmuThread::run() {
 
     MicroProfileOnThreadCreate("EmuThread");
 
-    stop_run = false;
-
     emit LoadProgress(VideoCore::LoadCallbackStage::Prepare, 0, 0);
 
     Core::System::GetInstance().Renderer().Rasterizer().LoadDiskResources(
@@ -40,7 +38,7 @@ void EmuThread::run() {
         render_window->DoneCurrent();
     }
 
-    // holds whether the cpu was running during the last iteration,
+    // Holds whether the cpu was running during the last iteration,
     // so that the DebugModeLeft signal can be emitted before the
     // next execution step
     bool was_active = false;

src/yuzu/configuration/config.cpp

@@ -209,7 +209,7 @@ void Config::ReadPlayerValues() {
     for (std::size_t p = 0; p < Settings::values.players.size(); ++p) {
         auto& player = Settings::values.players[p];
 
-        player.connected = qt_config->value(QString("player_%1_connected").arg(p), false).toBool();
+        player.connected = ReadSetting(QString("player_%1_connected").arg(p), false).toBool();
 
         player.type = static_cast<Settings::ControllerType>(
             qt_config
@@ -269,7 +269,7 @@ void Config::ReadPlayerValues() {
 }
 
 void Config::ReadDebugValues() {
-    Settings::values.debug_pad_enabled = qt_config->value("debug_pad_enabled", false).toBool();
+    Settings::values.debug_pad_enabled = ReadSetting("debug_pad_enabled", false).toBool();
     for (int i = 0; i < Settings::NativeButton::NumButtons; ++i) {
         std::string default_param = InputCommon::GenerateKeyboardParam(default_buttons[i]);
         Settings::values.debug_pad_buttons[i] =
@@ -298,7 +298,7 @@ void Config::ReadDebugValues() {
 }
 
 void Config::ReadKeyboardValues() {
-    Settings::values.keyboard_enabled = qt_config->value("keyboard_enabled", false).toBool();
+    Settings::values.keyboard_enabled = ReadSetting("keyboard_enabled", false).toBool();
 
     std::transform(default_keyboard_keys.begin(), default_keyboard_keys.end(),
                    Settings::values.keyboard_keys.begin(), InputCommon::GenerateKeyboardParam);
@@ -311,7 +311,7 @@ void Config::ReadKeyboardValues() {
 }
 
 void Config::ReadMouseValues() {
-    Settings::values.mouse_enabled = qt_config->value("mouse_enabled", false).toBool();
+    Settings::values.mouse_enabled = ReadSetting("mouse_enabled", false).toBool();
 
     for (int i = 0; i < Settings::NativeMouseButton::NumMouseButtons; ++i) {
         std::string default_param = InputCommon::GenerateKeyboardParam(default_mouse_buttons[i]);
@@ -327,16 +327,14 @@ void Config::ReadMouseValues() {
 }
 
 void Config::ReadTouchscreenValues() {
-    Settings::values.touchscreen.enabled = qt_config->value("touchscreen_enabled", true).toBool();
+    Settings::values.touchscreen.enabled = ReadSetting("touchscreen_enabled", true).toBool();
     Settings::values.touchscreen.device =
-        qt_config->value("touchscreen_device", "engine:emu_window").toString().toStdString();
+        ReadSetting("touchscreen_device", "engine:emu_window").toString().toStdString();
 
-    Settings::values.touchscreen.finger = qt_config->value("touchscreen_finger", 0).toUInt();
-    Settings::values.touchscreen.rotation_angle = qt_config->value("touchscreen_angle", 0).toUInt();
-    Settings::values.touchscreen.diameter_x =
-        qt_config->value("touchscreen_diameter_x", 15).toUInt();
-    Settings::values.touchscreen.diameter_y =
-        qt_config->value("touchscreen_diameter_y", 15).toUInt();
+    Settings::values.touchscreen.finger = ReadSetting("touchscreen_finger", 0).toUInt();
+    Settings::values.touchscreen.rotation_angle = ReadSetting("touchscreen_angle", 0).toUInt();
+    Settings::values.touchscreen.diameter_x = ReadSetting("touchscreen_diameter_x", 15).toUInt();
+    Settings::values.touchscreen.diameter_y = ReadSetting("touchscreen_diameter_y", 15).toUInt();
     qt_config->endGroup();
 }
 
@@ -357,42 +355,41 @@ void Config::ReadValues() {
     ReadTouchscreenValues();
 
     Settings::values.motion_device =
-        qt_config->value("motion_device", "engine:motion_emu,update_period:100,sensitivity:0.01")
+        ReadSetting("motion_device", "engine:motion_emu,update_period:100,sensitivity:0.01")
             .toString()
             .toStdString();
 
     qt_config->beginGroup("Core");
-    Settings::values.use_cpu_jit = qt_config->value("use_cpu_jit", true).toBool();
-    Settings::values.use_multi_core = qt_config->value("use_multi_core", false).toBool();
+    Settings::values.use_cpu_jit = ReadSetting("use_cpu_jit", true).toBool();
+    Settings::values.use_multi_core = ReadSetting("use_multi_core", false).toBool();
     qt_config->endGroup();
 
     qt_config->beginGroup("Renderer");
-    Settings::values.resolution_factor = qt_config->value("resolution_factor", 1.0).toFloat();
-    Settings::values.use_frame_limit = qt_config->value("use_frame_limit", true).toBool();
-    Settings::values.frame_limit = qt_config->value("frame_limit", 100).toInt();
-    Settings::values.use_disk_shader_cache =
-        qt_config->value("use_disk_shader_cache", false).toBool();
+    Settings::values.resolution_factor = ReadSetting("resolution_factor", 1.0).toFloat();
+    Settings::values.use_frame_limit = ReadSetting("use_frame_limit", true).toBool();
+    Settings::values.frame_limit = ReadSetting("frame_limit", 100).toInt();
+    Settings::values.use_disk_shader_cache = ReadSetting("use_disk_shader_cache", true).toBool();
     Settings::values.use_accurate_gpu_emulation =
-        qt_config->value("use_accurate_gpu_emulation", false).toBool();
+        ReadSetting("use_accurate_gpu_emulation", false).toBool();
     Settings::values.use_asynchronous_gpu_emulation =
-        qt_config->value("use_asynchronous_gpu_emulation", false).toBool();
+        ReadSetting("use_asynchronous_gpu_emulation", false).toBool();
 
-    Settings::values.bg_red = qt_config->value("bg_red", 0.0).toFloat();
-    Settings::values.bg_green = qt_config->value("bg_green", 0.0).toFloat();
-    Settings::values.bg_blue = qt_config->value("bg_blue", 0.0).toFloat();
+    Settings::values.bg_red = ReadSetting("bg_red", 0.0).toFloat();
+    Settings::values.bg_green = ReadSetting("bg_green", 0.0).toFloat();
+    Settings::values.bg_blue = ReadSetting("bg_blue", 0.0).toFloat();
     qt_config->endGroup();
 
     qt_config->beginGroup("Audio");
-    Settings::values.sink_id = qt_config->value("output_engine", "auto").toString().toStdString();
+    Settings::values.sink_id = ReadSetting("output_engine", "auto").toString().toStdString();
     Settings::values.enable_audio_stretching =
-        qt_config->value("enable_audio_stretching", true).toBool();
+        ReadSetting("enable_audio_stretching", true).toBool();
     Settings::values.audio_device_id =
-        qt_config->value("output_device", "auto").toString().toStdString();
-    Settings::values.volume = qt_config->value("volume", 1).toFloat();
+        ReadSetting("output_device", "auto").toString().toStdString();
+    Settings::values.volume = ReadSetting("volume", 1).toFloat();
     qt_config->endGroup();
 
     qt_config->beginGroup("Data Storage");
-    Settings::values.use_virtual_sd = qt_config->value("use_virtual_sd", true).toBool();
+    Settings::values.use_virtual_sd = ReadSetting("use_virtual_sd", true).toBool();
     FileUtil::GetUserPath(
         FileUtil::UserPath::NANDDir,
         qt_config
@@ -410,30 +407,30 @@ void Config::ReadValues() {
     qt_config->endGroup();
 
     qt_config->beginGroup("Core");
-    Settings::values.use_cpu_jit = qt_config->value("use_cpu_jit", true).toBool();
-    Settings::values.use_multi_core = qt_config->value("use_multi_core", false).toBool();
+    Settings::values.use_cpu_jit = ReadSetting("use_cpu_jit", true).toBool();
+    Settings::values.use_multi_core = ReadSetting("use_multi_core", false).toBool();
     qt_config->endGroup();
 
     qt_config->beginGroup("System");
-    Settings::values.use_docked_mode = qt_config->value("use_docked_mode", false).toBool();
-    Settings::values.enable_nfc = qt_config->value("enable_nfc", true).toBool();
+    Settings::values.use_docked_mode = ReadSetting("use_docked_mode", false).toBool();
+    Settings::values.enable_nfc = ReadSetting("enable_nfc", true).toBool();
 
-    Settings::values.current_user = std::clamp<int>(qt_config->value("current_user", 0).toInt(), 0,
-                                                    Service::Account::MAX_USERS - 1);
+    Settings::values.current_user =
+        std::clamp<int>(ReadSetting("current_user", 0).toInt(), 0, Service::Account::MAX_USERS - 1);
 
-    Settings::values.language_index = qt_config->value("language_index", 1).toInt();
+    Settings::values.language_index = ReadSetting("language_index", 1).toInt();
 
-    const auto rng_seed_enabled = qt_config->value("rng_seed_enabled", false).toBool();
+    const auto rng_seed_enabled = ReadSetting("rng_seed_enabled", false).toBool();
     if (rng_seed_enabled) {
-        Settings::values.rng_seed = qt_config->value("rng_seed", 0).toULongLong();
+        Settings::values.rng_seed = ReadSetting("rng_seed", 0).toULongLong();
     } else {
         Settings::values.rng_seed = std::nullopt;
     }
 
-    const auto custom_rtc_enabled = qt_config->value("custom_rtc_enabled", false).toBool();
+    const auto custom_rtc_enabled = ReadSetting("custom_rtc_enabled", false).toBool();
     if (custom_rtc_enabled) {
         Settings::values.custom_rtc =
-            std::chrono::seconds(qt_config->value("custom_rtc", 0).toULongLong());
+            std::chrono::seconds(ReadSetting("custom_rtc", 0).toULongLong());
     } else {
         Settings::values.custom_rtc = std::nullopt;
     }
@@ -441,35 +438,35 @@ void Config::ReadValues() {
     qt_config->endGroup();
 
     qt_config->beginGroup("Miscellaneous");
-    Settings::values.log_filter = qt_config->value("log_filter", "*:Info").toString().toStdString();
-    Settings::values.use_dev_keys = qt_config->value("use_dev_keys", false).toBool();
+    Settings::values.log_filter = ReadSetting("log_filter", "*:Info").toString().toStdString();
+    Settings::values.use_dev_keys = ReadSetting("use_dev_keys", false).toBool();
     qt_config->endGroup();
 
     qt_config->beginGroup("Debugging");
-    Settings::values.use_gdbstub = qt_config->value("use_gdbstub", false).toBool();
-    Settings::values.gdbstub_port = qt_config->value("gdbstub_port", 24689).toInt();
-    Settings::values.program_args = qt_config->value("program_args", "").toString().toStdString();
-    Settings::values.dump_exefs = qt_config->value("dump_exefs", false).toBool();
-    Settings::values.dump_nso = qt_config->value("dump_nso", false).toBool();
+    Settings::values.use_gdbstub = ReadSetting("use_gdbstub", false).toBool();
+    Settings::values.gdbstub_port = ReadSetting("gdbstub_port", 24689).toInt();
+    Settings::values.program_args = ReadSetting("program_args", "").toString().toStdString();
+    Settings::values.dump_exefs = ReadSetting("dump_exefs", false).toBool();
+    Settings::values.dump_nso = ReadSetting("dump_nso", false).toBool();
     qt_config->endGroup();
 
     qt_config->beginGroup("WebService");
-    Settings::values.enable_telemetry = qt_config->value("enable_telemetry", true).toBool();
+    Settings::values.enable_telemetry = ReadSetting("enable_telemetry", true).toBool();
     Settings::values.web_api_url =
-        qt_config->value("web_api_url", "https://api.yuzu-emu.org").toString().toStdString();
-    Settings::values.yuzu_username = qt_config->value("yuzu_username").toString().toStdString();
-    Settings::values.yuzu_token = qt_config->value("yuzu_token").toString().toStdString();
+        ReadSetting("web_api_url", "https://api.yuzu-emu.org").toString().toStdString();
+    Settings::values.yuzu_username = ReadSetting("yuzu_username").toString().toStdString();
+    Settings::values.yuzu_token = ReadSetting("yuzu_token").toString().toStdString();
     qt_config->endGroup();
 
     const auto size = qt_config->beginReadArray("DisabledAddOns");
     for (int i = 0; i < size; ++i) {
         qt_config->setArrayIndex(i);
-        const auto title_id = qt_config->value("title_id", 0).toULongLong();
+        const auto title_id = ReadSetting("title_id", 0).toULongLong();
         std::vector<std::string> out;
         const auto d_size = qt_config->beginReadArray("disabled");
         for (int j = 0; j < d_size; ++j) {
             qt_config->setArrayIndex(j);
-            out.push_back(qt_config->value("d", "").toString().toStdString());
+            out.push_back(ReadSetting("d", "").toString().toStdString());
         }
         qt_config->endArray();
         Settings::values.disabled_addons.insert_or_assign(title_id, out);
@@ -477,41 +474,38 @@ void Config::ReadValues() {
     qt_config->endArray();
 
     qt_config->beginGroup("UI");
-    UISettings::values.theme = qt_config->value("theme", UISettings::themes[0].second).toString();
+    UISettings::values.theme = ReadSetting("theme", UISettings::themes[0].second).toString();
     UISettings::values.enable_discord_presence =
-        qt_config->value("enable_discord_presence", true).toBool();
+        ReadSetting("enable_discord_presence", true).toBool();
     UISettings::values.screenshot_resolution_factor =
-        static_cast<u16>(qt_config->value("screenshot_resolution_factor", 0).toUInt());
-    UISettings::values.select_user_on_boot =
-        qt_config->value("select_user_on_boot", false).toBool();
+        static_cast<u16>(ReadSetting("screenshot_resolution_factor", 0).toUInt());
+    UISettings::values.select_user_on_boot = ReadSetting("select_user_on_boot", false).toBool();
 
     qt_config->beginGroup("UIGameList");
-    UISettings::values.show_unknown = qt_config->value("show_unknown", true).toBool();
-    UISettings::values.show_add_ons = qt_config->value("show_add_ons", true).toBool();
-    UISettings::values.icon_size = qt_config->value("icon_size", 64).toUInt();
-    UISettings::values.row_1_text_id = qt_config->value("row_1_text_id", 3).toUInt();
-    UISettings::values.row_2_text_id = qt_config->value("row_2_text_id", 2).toUInt();
+    UISettings::values.show_unknown = ReadSetting("show_unknown", true).toBool();
+    UISettings::values.show_add_ons = ReadSetting("show_add_ons", true).toBool();
+    UISettings::values.icon_size = ReadSetting("icon_size", 64).toUInt();
+    UISettings::values.row_1_text_id = ReadSetting("row_1_text_id", 3).toUInt();
+    UISettings::values.row_2_text_id = ReadSetting("row_2_text_id", 2).toUInt();
     qt_config->endGroup();
 
     qt_config->beginGroup("UILayout");
-    UISettings::values.geometry = qt_config->value("geometry").toByteArray();
-    UISettings::values.state = qt_config->value("state").toByteArray();
-    UISettings::values.renderwindow_geometry =
-        qt_config->value("geometryRenderWindow").toByteArray();
-    UISettings::values.gamelist_header_state =
-        qt_config->value("gameListHeaderState").toByteArray();
+    UISettings::values.geometry = ReadSetting("geometry").toByteArray();
+    UISettings::values.state = ReadSetting("state").toByteArray();
+    UISettings::values.renderwindow_geometry = ReadSetting("geometryRenderWindow").toByteArray();
+    UISettings::values.gamelist_header_state = ReadSetting("gameListHeaderState").toByteArray();
     UISettings::values.microprofile_geometry =
-        qt_config->value("microProfileDialogGeometry").toByteArray();
+        ReadSetting("microProfileDialogGeometry").toByteArray();
     UISettings::values.microprofile_visible =
-        qt_config->value("microProfileDialogVisible", false).toBool();
+        ReadSetting("microProfileDialogVisible", false).toBool();
     qt_config->endGroup();
 
     qt_config->beginGroup("Paths");
-    UISettings::values.roms_path = qt_config->value("romsPath").toString();
-    UISettings::values.symbols_path = qt_config->value("symbolsPath").toString();
-    UISettings::values.gamedir = qt_config->value("gameListRootDir", ".").toString();
-    UISettings::values.gamedir_deepscan = qt_config->value("gameListDeepScan", false).toBool();
-    UISettings::values.recent_files = qt_config->value("recentFiles").toStringList();
+    UISettings::values.roms_path = ReadSetting("romsPath").toString();
+    UISettings::values.symbols_path = ReadSetting("symbolsPath").toString();
+    UISettings::values.gamedir = ReadSetting("gameListRootDir", ".").toString();
+    UISettings::values.gamedir_deepscan = ReadSetting("gameListDeepScan", false).toBool();
+    UISettings::values.recent_files = ReadSetting("recentFiles").toStringList();
     qt_config->endGroup();
 
     qt_config->beginGroup("Shortcuts");
@@ -524,8 +518,8 @@ void Config::ReadValues() {
             qt_config->beginGroup(hotkey);
             UISettings::values.shortcuts.emplace_back(UISettings::Shortcut(
                 group + "/" + hotkey,
-                UISettings::ContextualShortcut(qt_config->value("KeySeq").toString(),
-                                               qt_config->value("Context").toInt())));
+                UISettings::ContextualShortcut(ReadSetting("KeySeq").toString(),
+                                               ReadSetting("Context").toInt())));
             qt_config->endGroup();
         }
 
@@ -533,16 +527,16 @@ void Config::ReadValues() {
     }
     qt_config->endGroup();
 
-    UISettings::values.single_window_mode = qt_config->value("singleWindowMode", true).toBool();
-    UISettings::values.fullscreen = qt_config->value("fullscreen", false).toBool();
-    UISettings::values.display_titlebar = qt_config->value("displayTitleBars", true).toBool();
-    UISettings::values.show_filter_bar = qt_config->value("showFilterBar", true).toBool();
-    UISettings::values.show_status_bar = qt_config->value("showStatusBar", true).toBool();
-    UISettings::values.confirm_before_closing = qt_config->value("confirmClose", true).toBool();
-    UISettings::values.first_start = qt_config->value("firstStart", true).toBool();
-    UISettings::values.callout_flags = qt_config->value("calloutFlags", 0).toUInt();
-    UISettings::values.show_console = qt_config->value("showConsole", false).toBool();
-    UISettings::values.profile_index = qt_config->value("profileIndex", 0).toUInt();
+    UISettings::values.single_window_mode = ReadSetting("singleWindowMode", true).toBool();
+    UISettings::values.fullscreen = ReadSetting("fullscreen", false).toBool();
+    UISettings::values.display_titlebar = ReadSetting("displayTitleBars", true).toBool();
+    UISettings::values.show_filter_bar = ReadSetting("showFilterBar", true).toBool();
+    UISettings::values.show_status_bar = ReadSetting("showStatusBar", true).toBool();
+    UISettings::values.confirm_before_closing = ReadSetting("confirmClose", true).toBool();
+    UISettings::values.first_start = ReadSetting("firstStart", true).toBool();
+    UISettings::values.callout_flags = ReadSetting("calloutFlags", 0).toUInt();
+    UISettings::values.show_console = ReadSetting("showConsole", false).toBool();
+    UISettings::values.profile_index = ReadSetting("profileIndex", 0).toUInt();
 
     ApplyDefaultProfileIfInputInvalid();
 
@@ -553,62 +547,79 @@ void Config::SavePlayerValues() {
     for (std::size_t p = 0; p < Settings::values.players.size(); ++p) {
         const auto& player = Settings::values.players[p];
 
-        qt_config->setValue(QString("player_%1_connected").arg(p), player.connected);
-        qt_config->setValue(QString("player_%1_type").arg(p), static_cast<u8>(player.type));
+        WriteSetting(QString("player_%1_connected").arg(p), player.connected, false);
+        WriteSetting(QString("player_%1_type").arg(p), static_cast<u8>(player.type),
+                     static_cast<u8>(Settings::ControllerType::DualJoycon));
 
-        qt_config->setValue(QString("player_%1_body_color_left").arg(p), player.body_color_left);
-        qt_config->setValue(QString("player_%1_body_color_right").arg(p), player.body_color_right);
-        qt_config->setValue(QString("player_%1_button_color_left").arg(p),
-                            player.button_color_left);
-        qt_config->setValue(QString("player_%1_button_color_right").arg(p),
-                            player.button_color_right);
+        WriteSetting(QString("player_%1_body_color_left").arg(p), player.body_color_left,
+                     Settings::JOYCON_BODY_NEON_BLUE);
+        WriteSetting(QString("player_%1_body_color_right").arg(p), player.body_color_right,
+                     Settings::JOYCON_BODY_NEON_RED);
+        WriteSetting(QString("player_%1_button_color_left").arg(p), player.button_color_left,
+                     Settings::JOYCON_BUTTONS_NEON_BLUE);
+        WriteSetting(QString("player_%1_button_color_right").arg(p), player.button_color_right,
+                     Settings::JOYCON_BUTTONS_NEON_RED);
 
         for (int i = 0; i < Settings::NativeButton::NumButtons; ++i) {
-            qt_config->setValue(QString("player_%1_").arg(p) +
-                                    QString::fromStdString(Settings::NativeButton::mapping[i]),
-                                QString::fromStdString(player.buttons[i]));
+            std::string default_param = InputCommon::GenerateKeyboardParam(default_buttons[i]);
+            WriteSetting(QString("player_%1_").arg(p) +
+                             QString::fromStdString(Settings::NativeButton::mapping[i]),
+                         QString::fromStdString(player.buttons[i]),
+                         QString::fromStdString(default_param));
         }
         for (int i = 0; i < Settings::NativeAnalog::NumAnalogs; ++i) {
-            qt_config->setValue(QString("player_%1_").arg(p) +
-                                    QString::fromStdString(Settings::NativeAnalog::mapping[i]),
-                                QString::fromStdString(player.analogs[i]));
+            std::string default_param = InputCommon::GenerateAnalogParamFromKeys(
+                default_analogs[i][0], default_analogs[i][1], default_analogs[i][2],
+                default_analogs[i][3], default_analogs[i][4], 0.5f);
+            WriteSetting(QString("player_%1_").arg(p) +
+                             QString::fromStdString(Settings::NativeAnalog::mapping[i]),
+                         QString::fromStdString(player.analogs[i]),
+                         QString::fromStdString(default_param));
         }
     }
 }
 
 void Config::SaveDebugValues() {
-    qt_config->setValue("debug_pad_enabled", Settings::values.debug_pad_enabled);
+    WriteSetting("debug_pad_enabled", Settings::values.debug_pad_enabled, false);
     for (int i = 0; i < Settings::NativeButton::NumButtons; ++i) {
-        qt_config->setValue(QString("debug_pad_") +
-                                QString::fromStdString(Settings::NativeButton::mapping[i]),
-                            QString::fromStdString(Settings::values.debug_pad_buttons[i]));
+        std::string default_param = InputCommon::GenerateKeyboardParam(default_buttons[i]);
+        WriteSetting(QString("debug_pad_") +
+                         QString::fromStdString(Settings::NativeButton::mapping[i]),
+                     QString::fromStdString(Settings::values.debug_pad_buttons[i]),
+                     QString::fromStdString(default_param));
     }
     for (int i = 0; i < Settings::NativeAnalog::NumAnalogs; ++i) {
-        qt_config->setValue(QString("debug_pad_") +
-                                QString::fromStdString(Settings::NativeAnalog::mapping[i]),
-                            QString::fromStdString(Settings::values.debug_pad_analogs[i]));
+        std::string default_param = InputCommon::GenerateAnalogParamFromKeys(
+            default_analogs[i][0], default_analogs[i][1], default_analogs[i][2],
+            default_analogs[i][3], default_analogs[i][4], 0.5f);
+        WriteSetting(QString("debug_pad_") +
+                         QString::fromStdString(Settings::NativeAnalog::mapping[i]),
+                     QString::fromStdString(Settings::values.debug_pad_analogs[i]),
+                     QString::fromStdString(default_param));
     }
 }
 
 void Config::SaveMouseValues() {
-    qt_config->setValue("mouse_enabled", Settings::values.mouse_enabled);
+    WriteSetting("mouse_enabled", Settings::values.mouse_enabled, false);
 
     for (int i = 0; i < Settings::NativeMouseButton::NumMouseButtons; ++i) {
-        qt_config->setValue(QString("mouse_") +
-                                QString::fromStdString(Settings::NativeMouseButton::mapping[i]),
-                            QString::fromStdString(Settings::values.mouse_buttons[i]));
+        std::string default_param = InputCommon::GenerateKeyboardParam(default_mouse_buttons[i]);
+        WriteSetting(QString("mouse_") +
+                         QString::fromStdString(Settings::NativeMouseButton::mapping[i]),
+                     QString::fromStdString(Settings::values.mouse_buttons[i]),
+                     QString::fromStdString(default_param));
     }
 }
 
 void Config::SaveTouchscreenValues() {
-    qt_config->setValue("touchscreen_enabled", Settings::values.touchscreen.enabled);
-    qt_config->setValue("touchscreen_device",
-                        QString::fromStdString(Settings::values.touchscreen.device));
+    WriteSetting("touchscreen_enabled", Settings::values.touchscreen.enabled, true);
+    WriteSetting("touchscreen_device", QString::fromStdString(Settings::values.touchscreen.device),
+                 "engine:emu_window");
 
-    qt_config->setValue("touchscreen_finger", Settings::values.touchscreen.finger);
-    qt_config->setValue("touchscreen_angle", Settings::values.touchscreen.rotation_angle);
-    qt_config->setValue("touchscreen_diameter_x", Settings::values.touchscreen.diameter_x);
-    qt_config->setValue("touchscreen_diameter_y", Settings::values.touchscreen.diameter_y);
+    WriteSetting("touchscreen_finger", Settings::values.touchscreen.finger, 0);
+    WriteSetting("touchscreen_angle", Settings::values.touchscreen.rotation_angle, 0);
+    WriteSetting("touchscreen_diameter_x", Settings::values.touchscreen.diameter_x, 15);
+    WriteSetting("touchscreen_diameter_y", Settings::values.touchscreen.diameter_y, 15);
 }
 
 void Config::SaveValues() {
@@ -619,91 +630,96 @@ void Config::SaveValues() {
     SaveMouseValues();
     SaveTouchscreenValues();
 
-    qt_config->setValue("motion_device", QString::fromStdString(Settings::values.motion_device));
-    qt_config->setValue("keyboard_enabled", Settings::values.keyboard_enabled);
+    WriteSetting("motion_device", QString::fromStdString(Settings::values.motion_device),
+                 "engine:motion_emu,update_period:100,sensitivity:0.01");
+    WriteSetting("keyboard_enabled", Settings::values.keyboard_enabled, false);
 
     qt_config->endGroup();
 
     qt_config->beginGroup("Core");
-    qt_config->setValue("use_cpu_jit", Settings::values.use_cpu_jit);
-    qt_config->setValue("use_multi_core", Settings::values.use_multi_core);
+    WriteSetting("use_cpu_jit", Settings::values.use_cpu_jit, true);
+    WriteSetting("use_multi_core", Settings::values.use_multi_core, false);
     qt_config->endGroup();
 
     qt_config->beginGroup("Renderer");
-    qt_config->setValue("resolution_factor", (double)Settings::values.resolution_factor);
-    qt_config->setValue("use_frame_limit", Settings::values.use_frame_limit);
-    qt_config->setValue("frame_limit", Settings::values.frame_limit);
-    qt_config->setValue("use_disk_shader_cache", Settings::values.use_disk_shader_cache);
-    qt_config->setValue("use_accurate_gpu_emulation", Settings::values.use_accurate_gpu_emulation);
-    qt_config->setValue("use_asynchronous_gpu_emulation",
-                        Settings::values.use_asynchronous_gpu_emulation);
+    WriteSetting("resolution_factor", (double)Settings::values.resolution_factor, 1.0);
+    WriteSetting("use_frame_limit", Settings::values.use_frame_limit, true);
+    WriteSetting("frame_limit", Settings::values.frame_limit, 100);
+    WriteSetting("use_disk_shader_cache", Settings::values.use_disk_shader_cache, true);
+    WriteSetting("use_accurate_gpu_emulation", Settings::values.use_accurate_gpu_emulation, false);
+    WriteSetting("use_asynchronous_gpu_emulation", Settings::values.use_asynchronous_gpu_emulation,
+                 false);
 
     // Cast to double because Qt's written float values are not human-readable
-    qt_config->setValue("bg_red", (double)Settings::values.bg_red);
-    qt_config->setValue("bg_green", (double)Settings::values.bg_green);
-    qt_config->setValue("bg_blue", (double)Settings::values.bg_blue);
+    WriteSetting("bg_red", (double)Settings::values.bg_red, 0.0);
+    WriteSetting("bg_green", (double)Settings::values.bg_green, 0.0);
+    WriteSetting("bg_blue", (double)Settings::values.bg_blue, 0.0);
     qt_config->endGroup();
 
     qt_config->beginGroup("Audio");
-    qt_config->setValue("output_engine", QString::fromStdString(Settings::values.sink_id));
-    qt_config->setValue("enable_audio_stretching", Settings::values.enable_audio_stretching);
-    qt_config->setValue("output_device", QString::fromStdString(Settings::values.audio_device_id));
-    qt_config->setValue("volume", Settings::values.volume);
+    WriteSetting("output_engine", QString::fromStdString(Settings::values.sink_id), "auto");
+    WriteSetting("enable_audio_stretching", Settings::values.enable_audio_stretching, true);
+    WriteSetting("output_device", QString::fromStdString(Settings::values.audio_device_id), "auto");
+    WriteSetting("volume", Settings::values.volume, 1.0f);
     qt_config->endGroup();
 
     qt_config->beginGroup("Data Storage");
-    qt_config->setValue("use_virtual_sd", Settings::values.use_virtual_sd);
-    qt_config->setValue("nand_directory",
-                        QString::fromStdString(FileUtil::GetUserPath(FileUtil::UserPath::NANDDir)));
-    qt_config->setValue("sdmc_directory",
-                        QString::fromStdString(FileUtil::GetUserPath(FileUtil::UserPath::SDMCDir)));
+    WriteSetting("use_virtual_sd", Settings::values.use_virtual_sd, true);
+    WriteSetting("nand_directory",
+                 QString::fromStdString(FileUtil::GetUserPath(FileUtil::UserPath::NANDDir)),
+                 QString::fromStdString(FileUtil::GetUserPath(FileUtil::UserPath::NANDDir)));
+    WriteSetting("sdmc_directory",
+                 QString::fromStdString(FileUtil::GetUserPath(FileUtil::UserPath::SDMCDir)),
+                 QString::fromStdString(FileUtil::GetUserPath(FileUtil::UserPath::SDMCDir)));
     qt_config->endGroup();
 
     qt_config->beginGroup("System");
-    qt_config->setValue("use_docked_mode", Settings::values.use_docked_mode);
-    qt_config->setValue("enable_nfc", Settings::values.enable_nfc);
-    qt_config->setValue("current_user", Settings::values.current_user);
-    qt_config->setValue("language_index", Settings::values.language_index);
+    WriteSetting("use_docked_mode", Settings::values.use_docked_mode, false);
+    WriteSetting("enable_nfc", Settings::values.enable_nfc, true);
+    WriteSetting("current_user", Settings::values.current_user, 0);
+    WriteSetting("language_index", Settings::values.language_index, 1);
 
-    qt_config->setValue("rng_seed_enabled", Settings::values.rng_seed.has_value());
-    qt_config->setValue("rng_seed", Settings::values.rng_seed.value_or(0));
+    WriteSetting("rng_seed_enabled", Settings::values.rng_seed.has_value(), false);
+    WriteSetting("rng_seed", Settings::values.rng_seed.value_or(0), 0);
 
-    qt_config->setValue("custom_rtc_enabled", Settings::values.custom_rtc.has_value());
-    qt_config->setValue("custom_rtc",
-                        QVariant::fromValue<long long>(
-                            Settings::values.custom_rtc.value_or(std::chrono::seconds{}).count()));
+    WriteSetting("custom_rtc_enabled", Settings::values.custom_rtc.has_value(), false);
+    WriteSetting("custom_rtc",
+                 QVariant::fromValue<long long>(
+                     Settings::values.custom_rtc.value_or(std::chrono::seconds{}).count()),
+                 0);
 
     qt_config->endGroup();
 
     qt_config->beginGroup("Miscellaneous");
-    qt_config->setValue("log_filter", QString::fromStdString(Settings::values.log_filter));
-    qt_config->setValue("use_dev_keys", Settings::values.use_dev_keys);
+    WriteSetting("log_filter", QString::fromStdString(Settings::values.log_filter), "*:Info");
+    WriteSetting("use_dev_keys", Settings::values.use_dev_keys, false);
     qt_config->endGroup();
 
     qt_config->beginGroup("Debugging");
-    qt_config->setValue("use_gdbstub", Settings::values.use_gdbstub);
-    qt_config->setValue("gdbstub_port", Settings::values.gdbstub_port);
-    qt_config->setValue("program_args", QString::fromStdString(Settings::values.program_args));
-    qt_config->setValue("dump_exefs", Settings::values.dump_exefs);
-    qt_config->setValue("dump_nso", Settings::values.dump_nso);
+    WriteSetting("use_gdbstub", Settings::values.use_gdbstub, false);
+    WriteSetting("gdbstub_port", Settings::values.gdbstub_port, 24689);
+    WriteSetting("program_args", QString::fromStdString(Settings::values.program_args), "");
+    WriteSetting("dump_exefs", Settings::values.dump_exefs, false);
+    WriteSetting("dump_nso", Settings::values.dump_nso, false);
     qt_config->endGroup();
 
     qt_config->beginGroup("WebService");
-    qt_config->setValue("enable_telemetry", Settings::values.enable_telemetry);
-    qt_config->setValue("web_api_url", QString::fromStdString(Settings::values.web_api_url));
-    qt_config->setValue("yuzu_username", QString::fromStdString(Settings::values.yuzu_username));
-    qt_config->setValue("yuzu_token", QString::fromStdString(Settings::values.yuzu_token));
+    WriteSetting("enable_telemetry", Settings::values.enable_telemetry, true);
+    WriteSetting("web_api_url", QString::fromStdString(Settings::values.web_api_url),
+                 "https://api.yuzu-emu.org");
+    WriteSetting("yuzu_username", QString::fromStdString(Settings::values.yuzu_username));
+    WriteSetting("yuzu_token", QString::fromStdString(Settings::values.yuzu_token));
     qt_config->endGroup();
 
     qt_config->beginWriteArray("DisabledAddOns");
     int i = 0;
     for (const auto& elem : Settings::values.disabled_addons) {
         qt_config->setArrayIndex(i);
-        qt_config->setValue("title_id", QVariant::fromValue<u64>(elem.first));
+        WriteSetting("title_id", QVariant::fromValue<u64>(elem.first), 0);
         qt_config->beginWriteArray("disabled");
         for (std::size_t j = 0; j < elem.second.size(); ++j) {
             qt_config->setArrayIndex(static_cast<int>(j));
-            qt_config->setValue("d", QString::fromStdString(elem.second[j]));
+            WriteSetting("d", QString::fromStdString(elem.second[j]), "");
         }
         qt_config->endArray();
         ++i;
@@ -711,60 +727,86 @@ void Config::SaveValues() {
     qt_config->endArray();
 
     qt_config->beginGroup("UI");
-    qt_config->setValue("theme", UISettings::values.theme);
-    qt_config->setValue("enable_discord_presence", UISettings::values.enable_discord_presence);
-    qt_config->setValue("screenshot_resolution_factor",
-                        UISettings::values.screenshot_resolution_factor);
-    qt_config->setValue("select_user_on_boot", UISettings::values.select_user_on_boot);
+    WriteSetting("theme", UISettings::values.theme, UISettings::themes[0].second);
+    WriteSetting("enable_discord_presence", UISettings::values.enable_discord_presence, true);
+    WriteSetting("screenshot_resolution_factor", UISettings::values.screenshot_resolution_factor,
+                 0);
+    WriteSetting("select_user_on_boot", UISettings::values.select_user_on_boot, false);
 
     qt_config->beginGroup("UIGameList");
-    qt_config->setValue("show_unknown", UISettings::values.show_unknown);
-    qt_config->setValue("show_add_ons", UISettings::values.show_add_ons);
-    qt_config->setValue("icon_size", UISettings::values.icon_size);
-    qt_config->setValue("row_1_text_id", UISettings::values.row_1_text_id);
-    qt_config->setValue("row_2_text_id", UISettings::values.row_2_text_id);
+    WriteSetting("show_unknown", UISettings::values.show_unknown, true);
+    WriteSetting("show_add_ons", UISettings::values.show_add_ons, true);
+    WriteSetting("icon_size", UISettings::values.icon_size, 64);
+    WriteSetting("row_1_text_id", UISettings::values.row_1_text_id, 3);
+    WriteSetting("row_2_text_id", UISettings::values.row_2_text_id, 2);
     qt_config->endGroup();
 
     qt_config->beginGroup("UILayout");
-    qt_config->setValue("geometry", UISettings::values.geometry);
-    qt_config->setValue("state", UISettings::values.state);
-    qt_config->setValue("geometryRenderWindow", UISettings::values.renderwindow_geometry);
-    qt_config->setValue("gameListHeaderState", UISettings::values.gamelist_header_state);
-    qt_config->setValue("microProfileDialogGeometry", UISettings::values.microprofile_geometry);
-    qt_config->setValue("microProfileDialogVisible", UISettings::values.microprofile_visible);
+    WriteSetting("geometry", UISettings::values.geometry);
+    WriteSetting("state", UISettings::values.state);
+    WriteSetting("geometryRenderWindow", UISettings::values.renderwindow_geometry);
+    WriteSetting("gameListHeaderState", UISettings::values.gamelist_header_state);
+    WriteSetting("microProfileDialogGeometry", UISettings::values.microprofile_geometry);
+    WriteSetting("microProfileDialogVisible", UISettings::values.microprofile_visible, false);
     qt_config->endGroup();
 
     qt_config->beginGroup("Paths");
-    qt_config->setValue("romsPath", UISettings::values.roms_path);
-    qt_config->setValue("symbolsPath", UISettings::values.symbols_path);
-    qt_config->setValue("screenshotPath", UISettings::values.screenshot_path);
-    qt_config->setValue("gameListRootDir", UISettings::values.gamedir);
-    qt_config->setValue("gameListDeepScan", UISettings::values.gamedir_deepscan);
-    qt_config->setValue("recentFiles", UISettings::values.recent_files);
+    WriteSetting("romsPath", UISettings::values.roms_path);
+    WriteSetting("symbolsPath", UISettings::values.symbols_path);
+    WriteSetting("screenshotPath", UISettings::values.screenshot_path);
+    WriteSetting("gameListRootDir", UISettings::values.gamedir, ".");
+    WriteSetting("gameListDeepScan", UISettings::values.gamedir_deepscan, false);
+    WriteSetting("recentFiles", UISettings::values.recent_files);
     qt_config->endGroup();
 
     qt_config->beginGroup("Shortcuts");
     for (auto shortcut : UISettings::values.shortcuts) {
-        qt_config->setValue(shortcut.first + "/KeySeq", shortcut.second.first);
-        qt_config->setValue(shortcut.first + "/Context", shortcut.second.second);
+        WriteSetting(shortcut.first + "/KeySeq", shortcut.second.first);
+        WriteSetting(shortcut.first + "/Context", shortcut.second.second);
     }
     qt_config->endGroup();
 
-    qt_config->setValue("singleWindowMode", UISettings::values.single_window_mode);
-    qt_config->setValue("fullscreen", UISettings::values.fullscreen);
-    qt_config->setValue("displayTitleBars", UISettings::values.display_titlebar);
-    qt_config->setValue("showFilterBar", UISettings::values.show_filter_bar);
-    qt_config->setValue("showStatusBar", UISettings::values.show_status_bar);
-    qt_config->setValue("confirmClose", UISettings::values.confirm_before_closing);
-    qt_config->setValue("firstStart", UISettings::values.first_start);
-    qt_config->setValue("calloutFlags", UISettings::values.callout_flags);
-    qt_config->setValue("showConsole", UISettings::values.show_console);
-    qt_config->setValue("profileIndex", UISettings::values.profile_index);
+    WriteSetting("singleWindowMode", UISettings::values.single_window_mode, true);
+    WriteSetting("fullscreen", UISettings::values.fullscreen, false);
+    WriteSetting("displayTitleBars", UISettings::values.display_titlebar, true);
+    WriteSetting("showFilterBar", UISettings::values.show_filter_bar, true);
+    WriteSetting("showStatusBar", UISettings::values.show_status_bar, true);
+    WriteSetting("confirmClose", UISettings::values.confirm_before_closing, true);
+    WriteSetting("firstStart", UISettings::values.first_start, true);
+    WriteSetting("calloutFlags", UISettings::values.callout_flags, 0);
+    WriteSetting("showConsole", UISettings::values.show_console, false);
+    WriteSetting("profileIndex", UISettings::values.profile_index, 0);
     qt_config->endGroup();
 }
 
+QVariant Config::ReadSetting(const QString& name) const {
+    return qt_config->value(name);
+}
+
+QVariant Config::ReadSetting(const QString& name, const QVariant& default_value) const {
+    QVariant result;
+    if (qt_config->value(name + "/default", false).toBool()) {
+        result = default_value;
+    } else {
+        result = qt_config->value(name, default_value);
+    }
+    return result;
+}
+
+void Config::WriteSetting(const QString& name, const QVariant& value) {
+    qt_config->setValue(name, value);
+}
+
+void Config::WriteSetting(const QString& name, const QVariant& value,
+                          const QVariant& default_value) {
+    qt_config->setValue(name + "/default", value == default_value);
+    qt_config->setValue(name, value);
+}
+
 void Config::Reload() {
     ReadValues();
+    // To apply default value changes
+    SaveValues();
     Settings::Apply();
 }
 

src/yuzu/configuration/config.h

@@ -42,6 +42,11 @@ private:
     void SaveMouseValues();
     void SaveTouchscreenValues();
 
+    QVariant ReadSetting(const QString& name) const;
+    QVariant ReadSetting(const QString& name, const QVariant& default_value) const;
+    void WriteSetting(const QString& name, const QVariant& value);
+    void WriteSetting(const QString& name, const QVariant& value, const QVariant& default_value);
+
     std::unique_ptr<QSettings> qt_config;
     std::string qt_config_loc;
 };

src/yuzu/debugger/graphics/graphics_surface.cpp

@@ -383,13 +383,12 @@ void GraphicsSurfaceWidget::OnUpdate() {
     // TODO: Implement a good way to visualize alpha components!
 
     QImage decoded_image(surface_width, surface_height, QImage::Format_ARGB32);
-    std::optional<VAddr> address = gpu.MemoryManager().GpuToCpuAddress(surface_address);
 
     // TODO(bunnei): Will not work with BCn formats that swizzle 4x4 tiles.
     // Needs to be fixed if we plan to use this feature more, otherwise we may remove it.
     auto unswizzled_data = Tegra::Texture::UnswizzleTexture(
-        *address, 1, 1, Tegra::Texture::BytesPerPixel(surface_format), surface_width,
-        surface_height, 1U);
+        gpu.MemoryManager().GetPointer(surface_address), 1, 1,
+        Tegra::Texture::BytesPerPixel(surface_format), surface_width, surface_height, 1U);
 
     auto texture_data = Tegra::Texture::DecodeTexture(unswizzled_data, surface_format,
                                                       surface_width, surface_height);

src/yuzu_cmd/yuzu.cpp

@@ -114,9 +114,9 @@ int main(int argc, char** argv) {
     };
 
     while (optind < argc) {
-        char arg = getopt_long(argc, argv, "g:fhvp::", long_options, &option_index);
+        int arg = getopt_long(argc, argv, "g:fhvp::", long_options, &option_index);
         if (arg != -1) {
-            switch (arg) {
+            switch (static_cast<char>(arg)) {
             case 'g':
                 errno = 0;
                 gdb_port = strtoul(optarg, &endarg, 0);