memory: Simplify rasterizer cache operations.

gpu: Use host address for caching instead of guest address.

.travis.yml

@@ -24,7 +24,7 @@ matrix:
     - os: osx
       env: NAME="macos build"
       sudo: false
-      osx_image: xcode10
+      osx_image: xcode10.1
       install: "./.travis/macos/deps.sh"
       script: "./.travis/macos/build.sh"
       after_success: "./.travis/macos/upload.sh"

.travis/macos/build.sh

@@ -2,7 +2,7 @@
 
 set -o pipefail
 
-export MACOSX_DEPLOYMENT_TARGET=10.13
+export MACOSX_DEPLOYMENT_TARGET=10.14
 export Qt5_DIR=$(brew --prefix)/opt/qt5
 export UNICORNDIR=$(pwd)/externals/unicorn
 export PATH="/usr/local/opt/ccache/libexec:$PATH"

CMakeModules/GenerateSCMRev.cmake

@@ -73,6 +73,7 @@ set(HASH_FILES
     "${VIDEO_CORE}/shader/decode/integer_set.cpp"
     "${VIDEO_CORE}/shader/decode/integer_set_predicate.cpp"
     "${VIDEO_CORE}/shader/decode/memory.cpp"
+    "${VIDEO_CORE}/shader/decode/texture.cpp"
     "${VIDEO_CORE}/shader/decode/other.cpp"
     "${VIDEO_CORE}/shader/decode/predicate_set_predicate.cpp"
     "${VIDEO_CORE}/shader/decode/predicate_set_register.cpp"

src/audio_core/cubeb_sink.cpp

@@ -12,7 +12,7 @@
 #include "common/ring_buffer.h"
 #include "core/settings.h"
 
-#ifdef _MSC_VER
+#ifdef _WIN32
 #include <objbase.h>
 #endif
 
@@ -113,7 +113,7 @@ private:
 
 CubebSink::CubebSink(std::string_view target_device_name) {
     // Cubeb requires COM to be initialized on the thread calling cubeb_init on Windows
-#ifdef _MSC_VER
+#ifdef _WIN32
     com_init_result = CoInitializeEx(nullptr, COINIT_MULTITHREADED);
 #endif
 
@@ -152,7 +152,7 @@ CubebSink::~CubebSink() {
 
     cubeb_destroy(ctx);
 
-#ifdef _MSC_VER
+#ifdef _WIN32
     if (SUCCEEDED(com_init_result)) {
         CoUninitialize();
     }

src/audio_core/cubeb_sink.h

@@ -26,7 +26,7 @@ private:
     cubeb_devid output_device{};
     std::vector<SinkStreamPtr> sink_streams;
 
-#ifdef _MSC_VER
+#ifdef _WIN32
     u32 com_init_result = 0;
 #endif
 };

src/common/CMakeLists.txt

@@ -47,6 +47,7 @@ add_custom_command(OUTPUT scm_rev.cpp
       "${VIDEO_CORE}/shader/decode/integer_set.cpp"
       "${VIDEO_CORE}/shader/decode/integer_set_predicate.cpp"
       "${VIDEO_CORE}/shader/decode/memory.cpp"
+      "${VIDEO_CORE}/shader/decode/texture.cpp"
       "${VIDEO_CORE}/shader/decode/other.cpp"
       "${VIDEO_CORE}/shader/decode/predicate_set_predicate.cpp"
       "${VIDEO_CORE}/shader/decode/predicate_set_register.cpp"

src/common/bit_field.h

@@ -111,12 +111,6 @@
 template <std::size_t Position, std::size_t Bits, typename T>
 struct BitField {
 private:
-    // We hide the copy assigment operator here, because the default copy
-    // assignment would copy the full storage value, rather than just the bits
-    // relevant to this particular bit field.
-    // We don't delete it because we want BitField to be trivially copyable.
-    constexpr BitField& operator=(const BitField&) = default;
-
     // UnderlyingType is T for non-enum types and the underlying type of T if
     // T is an enumeration. Note that T is wrapped within an enable_if in the
     // former case to workaround compile errors which arise when using
@@ -163,9 +157,13 @@ public:
     BitField(T val) = delete;
     BitField& operator=(T val) = delete;
 
-    // Force default constructor to be created
-    // so that we can use this within unions
-    constexpr BitField() = default;
+    constexpr BitField() noexcept = default;
+
+    constexpr BitField(const BitField&) noexcept = default;
+    constexpr BitField& operator=(const BitField&) noexcept = default;
+
+    constexpr BitField(BitField&&) noexcept = default;
+    constexpr BitField& operator=(BitField&&) noexcept = default;
 
     constexpr FORCE_INLINE operator T() const {
         return Value();

src/core/core.cpp

@@ -116,7 +116,7 @@ struct System::Impl {
         if (web_browser == nullptr)
             web_browser = std::make_unique<Core::Frontend::DefaultWebBrowserApplet>();
 
-        auto main_process = Kernel::Process::Create(kernel, "main");
+        auto main_process = Kernel::Process::Create(system, "main");
         kernel.MakeCurrentProcess(main_process.get());
 
         telemetry_session = std::make_unique<Core::TelemetrySession>();
@@ -190,13 +190,13 @@ struct System::Impl {
 
     void Shutdown() {
         // Log last frame performance stats
-        auto perf_results = GetAndResetPerfStats();
-        Telemetry().AddField(Telemetry::FieldType::Performance, "Shutdown_EmulationSpeed",
-                             perf_results.emulation_speed * 100.0);
-        Telemetry().AddField(Telemetry::FieldType::Performance, "Shutdown_Framerate",
-                             perf_results.game_fps);
-        Telemetry().AddField(Telemetry::FieldType::Performance, "Shutdown_Frametime",
-                             perf_results.frametime * 1000.0);
+        const auto perf_results = GetAndResetPerfStats();
+        telemetry_session->AddField(Telemetry::FieldType::Performance, "Shutdown_EmulationSpeed",
+                                    perf_results.emulation_speed * 100.0);
+        telemetry_session->AddField(Telemetry::FieldType::Performance, "Shutdown_Framerate",
+                                    perf_results.game_fps);
+        telemetry_session->AddField(Telemetry::FieldType::Performance, "Shutdown_Frametime",
+                                    perf_results.frametime * 1000.0);
 
         is_powered_on = false;
 

src/core/core.h

@@ -293,10 +293,6 @@ inline ARM_Interface& CurrentArmInterface() {
     return System::GetInstance().CurrentArmInterface();
 }
 
-inline TelemetrySession& Telemetry() {
-    return System::GetInstance().TelemetrySession();
-}
-
 inline Kernel::Process* CurrentProcess() {
     return System::GetInstance().CurrentProcess();
 }

src/core/core_cpu.cpp

@@ -11,6 +11,7 @@
 #endif
 #include "core/arm/exclusive_monitor.h"
 #include "core/arm/unicorn/arm_unicorn.h"
+#include "core/core.h"
 #include "core/core_cpu.h"
 #include "core/core_timing.h"
 #include "core/hle/kernel/scheduler.h"
@@ -49,9 +50,9 @@ bool CpuBarrier::Rendezvous() {
     return false;
 }
 
-Cpu::Cpu(Timing::CoreTiming& core_timing, ExclusiveMonitor& exclusive_monitor,
-         CpuBarrier& cpu_barrier, std::size_t core_index)
-    : cpu_barrier{cpu_barrier}, core_timing{core_timing}, core_index{core_index} {
+Cpu::Cpu(System& system, ExclusiveMonitor& exclusive_monitor, CpuBarrier& cpu_barrier,
+         std::size_t core_index)
+    : cpu_barrier{cpu_barrier}, core_timing{system.CoreTiming()}, core_index{core_index} {
     if (Settings::values.use_cpu_jit) {
 #ifdef ARCHITECTURE_x86_64
         arm_interface = std::make_unique<ARM_Dynarmic>(core_timing, exclusive_monitor, core_index);
@@ -63,7 +64,7 @@ Cpu::Cpu(Timing::CoreTiming& core_timing, ExclusiveMonitor& exclusive_monitor,
         arm_interface = std::make_unique<ARM_Unicorn>(core_timing);
     }
 
-    scheduler = std::make_unique<Kernel::Scheduler>(*arm_interface);
+    scheduler = std::make_unique<Kernel::Scheduler>(system, *arm_interface);
 }
 
 Cpu::~Cpu() = default;

src/core/core_cpu.h

@@ -15,6 +15,10 @@ namespace Kernel {
 class Scheduler;
 }
 
+namespace Core {
+class System;
+}
+
 namespace Core::Timing {
 class CoreTiming;
 }
@@ -45,8 +49,8 @@ private:
 
 class Cpu {
 public:
-    Cpu(Timing::CoreTiming& core_timing, ExclusiveMonitor& exclusive_monitor,
-        CpuBarrier& cpu_barrier, std::size_t core_index);
+    Cpu(System& system, ExclusiveMonitor& exclusive_monitor, CpuBarrier& cpu_barrier,
+        std::size_t core_index);
     ~Cpu();
 
     void RunLoop(bool tight_loop = true);

src/core/cpu_core_manager.cpp

@@ -27,8 +27,7 @@ void CpuCoreManager::Initialize(System& system) {
     exclusive_monitor = Cpu::MakeExclusiveMonitor(cores.size());
 
     for (std::size_t index = 0; index < cores.size(); ++index) {
-        cores[index] =
-            std::make_unique<Cpu>(system.CoreTiming(), *exclusive_monitor, *barrier, index);
+        cores[index] = std::make_unique<Cpu>(system, *exclusive_monitor, *barrier, index);
     }
 
     // Create threads for CPU cores 1-3, and build thread_to_cpu map

src/core/hle/ipc.h

@@ -4,10 +4,10 @@
 
 #pragma once
 
+#include "common/bit_field.h"
+#include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "common/swap.h"
-#include "core/hle/kernel/errors.h"
-#include "core/memory.h"
 
 namespace IPC {
 

src/core/hle/ipc_helpers.h

@@ -19,9 +19,12 @@
 #include "core/hle/kernel/hle_ipc.h"
 #include "core/hle/kernel/object.h"
 #include "core/hle/kernel/server_session.h"
+#include "core/hle/result.h"
 
 namespace IPC {
 
+constexpr ResultCode ERR_REMOTE_PROCESS_DEAD{ErrorModule::HIPC, 301};
+
 class RequestHelperBase {
 protected:
     Kernel::HLERequestContext* context = nullptr;
@@ -350,7 +353,7 @@ public:
     template <class T>
     std::shared_ptr<T> PopIpcInterface() {
         ASSERT(context->Session()->IsDomain());
-        ASSERT(context->GetDomainMessageHeader()->input_object_count > 0);
+        ASSERT(context->GetDomainMessageHeader().input_object_count > 0);
         return context->GetDomainRequestHandler<T>(Pop<u32>() - 1);
     }
 };
@@ -362,6 +365,11 @@ inline u32 RequestParser::Pop() {
     return cmdbuf[index++];
 }
 
+template <>
+inline s32 RequestParser::Pop() {
+    return static_cast<s32>(Pop<u32>());
+}
+
 template <typename T>
 void RequestParser::PopRaw(T& value) {
     std::memcpy(&value, cmdbuf + index, sizeof(T));
@@ -392,6 +400,16 @@ inline u64 RequestParser::Pop() {
     return msw << 32 | lsw;
 }
 
+template <>
+inline s8 RequestParser::Pop() {
+    return static_cast<s8>(Pop<u8>());
+}
+
+template <>
+inline s16 RequestParser::Pop() {
+    return static_cast<s16>(Pop<u16>());
+}
+
 template <>
 inline s64 RequestParser::Pop() {
     return static_cast<s64>(Pop<u64>());

src/core/hle/kernel/address_arbiter.cpp

@@ -42,7 +42,21 @@ void WakeThreads(const std::vector<SharedPtr<Thread>>& waiting_threads, s32 num_
 AddressArbiter::AddressArbiter(Core::System& system) : system{system} {}
 AddressArbiter::~AddressArbiter() = default;
 
-ResultCode AddressArbiter::SignalToAddress(VAddr address, s32 num_to_wake) {
+ResultCode AddressArbiter::SignalToAddress(VAddr address, SignalType type, s32 value,
+                                           s32 num_to_wake) {
+    switch (type) {
+    case SignalType::Signal:
+        return SignalToAddressOnly(address, num_to_wake);
+    case SignalType::IncrementAndSignalIfEqual:
+        return IncrementAndSignalToAddressIfEqual(address, value, num_to_wake);
+    case SignalType::ModifyByWaitingCountAndSignalIfEqual:
+        return ModifyByWaitingCountAndSignalToAddressIfEqual(address, value, num_to_wake);
+    default:
+        return ERR_INVALID_ENUM_VALUE;
+    }
+}
+
+ResultCode AddressArbiter::SignalToAddressOnly(VAddr address, s32 num_to_wake) {
     const std::vector<SharedPtr<Thread>> waiting_threads = GetThreadsWaitingOnAddress(address);
     WakeThreads(waiting_threads, num_to_wake);
     return RESULT_SUCCESS;
@@ -60,7 +74,7 @@ ResultCode AddressArbiter::IncrementAndSignalToAddressIfEqual(VAddr address, s32
     }
 
     Memory::Write32(address, static_cast<u32>(value + 1));
-    return SignalToAddress(address, num_to_wake);
+    return SignalToAddressOnly(address, num_to_wake);
 }
 
 ResultCode AddressArbiter::ModifyByWaitingCountAndSignalToAddressIfEqual(VAddr address, s32 value,
@@ -92,6 +106,20 @@ ResultCode AddressArbiter::ModifyByWaitingCountAndSignalToAddressIfEqual(VAddr a
     return RESULT_SUCCESS;
 }
 
+ResultCode AddressArbiter::WaitForAddress(VAddr address, ArbitrationType type, s32 value,
+                                          s64 timeout_ns) {
+    switch (type) {
+    case ArbitrationType::WaitIfLessThan:
+        return WaitForAddressIfLessThan(address, value, timeout_ns, false);
+    case ArbitrationType::DecrementAndWaitIfLessThan:
+        return WaitForAddressIfLessThan(address, value, timeout_ns, true);
+    case ArbitrationType::WaitIfEqual:
+        return WaitForAddressIfEqual(address, value, timeout_ns);
+    default:
+        return ERR_INVALID_ENUM_VALUE;
+    }
+}
+
 ResultCode AddressArbiter::WaitForAddressIfLessThan(VAddr address, s32 value, s64 timeout,
                                                     bool should_decrement) {
     // Ensure that we can read the address.
@@ -113,7 +141,7 @@ ResultCode AddressArbiter::WaitForAddressIfLessThan(VAddr address, s32 value, s6
         return RESULT_TIMEOUT;
     }
 
-    return WaitForAddress(address, timeout);
+    return WaitForAddressImpl(address, timeout);
 }
 
 ResultCode AddressArbiter::WaitForAddressIfEqual(VAddr address, s32 value, s64 timeout) {
@@ -130,10 +158,10 @@ ResultCode AddressArbiter::WaitForAddressIfEqual(VAddr address, s32 value, s64 t
         return RESULT_TIMEOUT;
     }
 
-    return WaitForAddress(address, timeout);
+    return WaitForAddressImpl(address, timeout);
 }
 
-ResultCode AddressArbiter::WaitForAddress(VAddr address, s64 timeout) {
+ResultCode AddressArbiter::WaitForAddressImpl(VAddr address, s64 timeout) {
     SharedPtr<Thread> current_thread = system.CurrentScheduler().GetCurrentThread();
     current_thread->SetArbiterWaitAddress(address);
     current_thread->SetStatus(ThreadStatus::WaitArb);

src/core/hle/kernel/address_arbiter.h

@@ -4,8 +4,10 @@
 
 #pragma once
 
+#include <vector>
+
 #include "common/common_types.h"
-#include "core/hle/kernel/address_arbiter.h"
+#include "core/hle/kernel/object.h"
 
 union ResultCode;
 
@@ -40,8 +42,15 @@ public:
     AddressArbiter(AddressArbiter&&) = default;
     AddressArbiter& operator=(AddressArbiter&&) = delete;
 
+    /// Signals an address being waited on with a particular signaling type.
+    ResultCode SignalToAddress(VAddr address, SignalType type, s32 value, s32 num_to_wake);
+
+    /// Waits on an address with a particular arbitration type.
+    ResultCode WaitForAddress(VAddr address, ArbitrationType type, s32 value, s64 timeout_ns);
+
+private:
     /// Signals an address being waited on.
-    ResultCode SignalToAddress(VAddr address, s32 num_to_wake);
+    ResultCode SignalToAddressOnly(VAddr address, s32 num_to_wake);
 
     /// Signals an address being waited on and increments its value if equal to the value argument.
     ResultCode IncrementAndSignalToAddressIfEqual(VAddr address, s32 value, s32 num_to_wake);
@@ -59,9 +68,8 @@ public:
     /// Waits on an address if the value passed is equal to the argument value.
     ResultCode WaitForAddressIfEqual(VAddr address, s32 value, s64 timeout);
 
-private:
     // Waits on the given address with a timeout in nanoseconds
-    ResultCode WaitForAddress(VAddr address, s64 timeout);
+    ResultCode WaitForAddressImpl(VAddr address, s64 timeout);
 
     // Gets the threads waiting on an address.
     std::vector<SharedPtr<Thread>> GetThreadsWaitingOnAddress(VAddr address) const;

src/core/hle/kernel/client_port.cpp

@@ -33,10 +33,11 @@ ResultVal<SharedPtr<ClientSession>> ClientPort::Connect() {
     // Create a new session pair, let the created sessions inherit the parent port's HLE handler.
     auto sessions = ServerSession::CreateSessionPair(kernel, server_port->GetName(), this);
 
-    if (server_port->hle_handler)
-        server_port->hle_handler->ClientConnected(std::get<SharedPtr<ServerSession>>(sessions));
-    else
-        server_port->pending_sessions.push_back(std::get<SharedPtr<ServerSession>>(sessions));
+    if (server_port->HasHLEHandler()) {
+        server_port->GetHLEHandler()->ClientConnected(std::get<SharedPtr<ServerSession>>(sessions));
+    } else {
+        server_port->AppendPendingSession(std::get<SharedPtr<ServerSession>>(sessions));
+    }
 
     // Wake the threads waiting on the ServerPort
     server_port->WakeupAllWaitingThreads();

src/core/hle/kernel/client_session.cpp

@@ -17,21 +17,11 @@ ClientSession::~ClientSession() {
     // This destructor will be called automatically when the last ClientSession handle is closed by
     // the emulated application.
 
-    // Local references to ServerSession and SessionRequestHandler are necessary to guarantee they
+    // A local reference to the ServerSession is necessary to guarantee it
     // will be kept alive until after ClientDisconnected() returns.
     SharedPtr<ServerSession> server = parent->server;
     if (server) {
-        std::shared_ptr<SessionRequestHandler> hle_handler = server->hle_handler;
-        if (hle_handler)
-            hle_handler->ClientDisconnected(server);
-
-        // TODO(Subv): Force a wake up of all the ServerSession's waiting threads and set
-        // their WaitSynchronization result to 0xC920181A.
-
-        // Clean up the list of client threads with pending requests, they are unneeded now that the
-        // client endpoint is closed.
-        server->pending_requesting_threads.clear();
-        server->currently_handling = nullptr;
+        server->ClientDisconnected();
     }
 
     parent->client = nullptr;

src/core/hle/kernel/client_session.h

@@ -36,14 +36,15 @@ public:
 
     ResultCode SendSyncRequest(SharedPtr<Thread> thread);
 
-    std::string name; ///< Name of client port (optional)
+private:
+    explicit ClientSession(KernelCore& kernel);
+    ~ClientSession() override;
 
     /// The parent session, which links to the server endpoint.
     std::shared_ptr<Session> parent;
 
-private:
-    explicit ClientSession(KernelCore& kernel);
-    ~ClientSession() override;
+    /// Name of the client session (optional)
+    std::string name;
 };
 
 } // namespace Kernel

src/core/hle/kernel/hle_ipc.cpp

@@ -86,7 +86,7 @@ HLERequestContext::~HLERequestContext() = default;
 void HLERequestContext::ParseCommandBuffer(const HandleTable& handle_table, u32_le* src_cmdbuf,
                                            bool incoming) {
     IPC::RequestParser rp(src_cmdbuf);
-    command_header = std::make_shared<IPC::CommandHeader>(rp.PopRaw<IPC::CommandHeader>());
+    command_header = rp.PopRaw<IPC::CommandHeader>();
 
     if (command_header->type == IPC::CommandType::Close) {
         // Close does not populate the rest of the IPC header
@@ -95,8 +95,7 @@ void HLERequestContext::ParseCommandBuffer(const HandleTable& handle_table, u32_
 
     // If handle descriptor is present, add size of it
     if (command_header->enable_handle_descriptor) {
-        handle_descriptor_header =
-            std::make_shared<IPC::HandleDescriptorHeader>(rp.PopRaw<IPC::HandleDescriptorHeader>());
+        handle_descriptor_header = rp.PopRaw<IPC::HandleDescriptorHeader>();
         if (handle_descriptor_header->send_current_pid) {
             rp.Skip(2, false);
         }
@@ -140,16 +139,15 @@ void HLERequestContext::ParseCommandBuffer(const HandleTable& handle_table, u32_
         // If this is an incoming message, only CommandType "Request" has a domain header
         // All outgoing domain messages have the domain header, if only incoming has it
         if (incoming || domain_message_header) {
-            domain_message_header =
-                std::make_shared<IPC::DomainMessageHeader>(rp.PopRaw<IPC::DomainMessageHeader>());
+            domain_message_header = rp.PopRaw<IPC::DomainMessageHeader>();
         } else {
-            if (Session()->IsDomain())
+            if (Session()->IsDomain()) {
                 LOG_WARNING(IPC, "Domain request has no DomainMessageHeader!");
+            }
         }
     }
 
-    data_payload_header =
-        std::make_shared<IPC::DataPayloadHeader>(rp.PopRaw<IPC::DataPayloadHeader>());
+    data_payload_header = rp.PopRaw<IPC::DataPayloadHeader>();
 
     data_payload_offset = rp.GetCurrentOffset();
 
@@ -264,11 +262,11 @@ ResultCode HLERequestContext::WriteToOutgoingCommandBuffer(Thread& thread) {
         // Write the domain objects to the command buffer, these go after the raw untranslated data.
         // TODO(Subv): This completely ignores C buffers.
         std::size_t domain_offset = size - domain_message_header->num_objects;
-        auto& request_handlers = server_session->domain_request_handlers;
 
-        for (auto& object : domain_objects) {
-            request_handlers.emplace_back(object);
-            dst_cmdbuf[domain_offset++] = static_cast<u32_le>(request_handlers.size());
+        for (const auto& object : domain_objects) {
+            server_session->AppendDomainRequestHandler(object);
+            dst_cmdbuf[domain_offset++] =
+                static_cast<u32_le>(server_session->NumDomainRequestHandlers());
         }
     }
 

src/core/hle/kernel/hle_ipc.h

@@ -6,6 +6,7 @@
 
 #include <array>
 #include <memory>
+#include <optional>
 #include <string>
 #include <type_traits>
 #include <vector>
@@ -15,6 +16,8 @@
 #include "core/hle/ipc.h"
 #include "core/hle/kernel/object.h"
 
+union ResultCode;
+
 namespace Service {
 class ServiceFrameworkBase;
 }
@@ -166,12 +169,12 @@ public:
         return buffer_c_desciptors;
     }
 
-    const IPC::DomainMessageHeader* GetDomainMessageHeader() const {
-        return domain_message_header.get();
+    const IPC::DomainMessageHeader& GetDomainMessageHeader() const {
+        return domain_message_header.value();
     }
 
     bool HasDomainMessageHeader() const {
-        return domain_message_header != nullptr;
+        return domain_message_header.has_value();
     }
 
     /// Helper function to read a buffer using the appropriate buffer descriptor
@@ -208,14 +211,12 @@ public:
 
     template <typename T>
     SharedPtr<T> GetCopyObject(std::size_t index) {
-        ASSERT(index < copy_objects.size());
-        return DynamicObjectCast<T>(copy_objects[index]);
+        return DynamicObjectCast<T>(copy_objects.at(index));
     }
 
     template <typename T>
     SharedPtr<T> GetMoveObject(std::size_t index) {
-        ASSERT(index < move_objects.size());
-        return DynamicObjectCast<T>(move_objects[index]);
+        return DynamicObjectCast<T>(move_objects.at(index));
     }
 
     void AddMoveObject(SharedPtr<Object> object) {
@@ -232,7 +233,7 @@ public:
 
     template <typename T>
     std::shared_ptr<T> GetDomainRequestHandler(std::size_t index) const {
-        return std::static_pointer_cast<T>(domain_request_handlers[index]);
+        return std::static_pointer_cast<T>(domain_request_handlers.at(index));
     }
 
     void SetDomainRequestHandlers(
@@ -272,10 +273,10 @@ private:
     boost::container::small_vector<SharedPtr<Object>, 8> copy_objects;
     boost::container::small_vector<std::shared_ptr<SessionRequestHandler>, 8> domain_objects;
 
-    std::shared_ptr<IPC::CommandHeader> command_header;
-    std::shared_ptr<IPC::HandleDescriptorHeader> handle_descriptor_header;
-    std::shared_ptr<IPC::DataPayloadHeader> data_payload_header;
-    std::shared_ptr<IPC::DomainMessageHeader> domain_message_header;
+    std::optional<IPC::CommandHeader> command_header;
+    std::optional<IPC::HandleDescriptorHeader> handle_descriptor_header;
+    std::optional<IPC::DataPayloadHeader> data_payload_header;
+    std::optional<IPC::DomainMessageHeader> domain_message_header;
     std::vector<IPC::BufferDescriptorX> buffer_x_desciptors;
     std::vector<IPC::BufferDescriptorABW> buffer_a_desciptors;
     std::vector<IPC::BufferDescriptorABW> buffer_b_desciptors;

src/core/hle/kernel/kernel.cpp

@@ -87,7 +87,7 @@ static void ThreadWakeupCallback(u64 thread_handle, [[maybe_unused]] int cycles_
 }
 
 struct KernelCore::Impl {
-    explicit Impl(Core::System& system) : address_arbiter{system}, system{system} {}
+    explicit Impl(Core::System& system) : system{system} {}
 
     void Initialize(KernelCore& kernel) {
         Shutdown();
@@ -138,8 +138,6 @@ struct KernelCore::Impl {
     std::vector<SharedPtr<Process>> process_list;
     Process* current_process = nullptr;
 
-    Kernel::AddressArbiter address_arbiter;
-
     SharedPtr<ResourceLimit> system_resource_limit;
 
     Core::Timing::EventType* thread_wakeup_event_type = nullptr;
@@ -192,14 +190,6 @@ const Process* KernelCore::CurrentProcess() const {
     return impl->current_process;
 }
 
-AddressArbiter& KernelCore::AddressArbiter() {
-    return impl->address_arbiter;
-}
-
-const AddressArbiter& KernelCore::AddressArbiter() const {
-    return impl->address_arbiter;
-}
-
 void KernelCore::AddNamedPort(std::string name, SharedPtr<ClientPort> port) {
     impl->named_ports.emplace(std::move(name), std::move(port));
 }

src/core/hle/kernel/kernel.h

@@ -75,12 +75,6 @@ public:
     /// Retrieves a const pointer to the current process.
     const Process* CurrentProcess() const;
 
-    /// Provides a reference to the kernel's address arbiter.
-    Kernel::AddressArbiter& AddressArbiter();
-
-    /// Provides a const reference to the kernel's address arbiter.
-    const Kernel::AddressArbiter& AddressArbiter() const;
-
     /// Adds a port to the named port table
     void AddNamedPort(std::string name, SharedPtr<ClientPort> port);
 

src/core/hle/kernel/process.cpp

@@ -53,9 +53,10 @@ void SetupMainThread(Process& owner_process, KernelCore& kernel, VAddr entry_poi
 CodeSet::CodeSet() = default;
 CodeSet::~CodeSet() = default;
 
-SharedPtr<Process> Process::Create(KernelCore& kernel, std::string&& name) {
-    SharedPtr<Process> process(new Process(kernel));
+SharedPtr<Process> Process::Create(Core::System& system, std::string&& name) {
+    auto& kernel = system.Kernel();
 
+    SharedPtr<Process> process(new Process(system));
     process->name = std::move(name);
     process->resource_limit = kernel.GetSystemResourceLimit();
     process->status = ProcessStatus::Created;
@@ -132,7 +133,7 @@ void Process::PrepareForTermination() {
             if (thread->GetOwnerProcess() != this)
                 continue;
 
-            if (thread == GetCurrentThread())
+            if (thread == system.CurrentScheduler().GetCurrentThread())
                 continue;
 
             // TODO(Subv): When are the other running/ready threads terminated?
@@ -144,7 +145,6 @@ void Process::PrepareForTermination() {
         }
     };
 
-    const auto& system = Core::System::GetInstance();
     stop_threads(system.Scheduler(0).GetThreadList());
     stop_threads(system.Scheduler(1).GetThreadList());
     stop_threads(system.Scheduler(2).GetThreadList());
@@ -227,14 +227,12 @@ void Process::LoadModule(CodeSet module_, VAddr base_addr) {
     MapSegment(module_.DataSegment(), VMAPermission::ReadWrite, MemoryState::CodeMutable);
 
     // Clear instruction cache in CPU JIT
-    Core::System::GetInstance().ArmInterface(0).ClearInstructionCache();
-    Core::System::GetInstance().ArmInterface(1).ClearInstructionCache();
-    Core::System::GetInstance().ArmInterface(2).ClearInstructionCache();
-    Core::System::GetInstance().ArmInterface(3).ClearInstructionCache();
+    system.InvalidateCpuInstructionCaches();
 }
 
-Kernel::Process::Process(KernelCore& kernel) : WaitObject{kernel} {}
-Kernel::Process::~Process() {}
+Process::Process(Core::System& system)
+    : WaitObject{system.Kernel()}, address_arbiter{system}, system{system} {}
+Process::~Process() = default;
 
 void Process::Acquire(Thread* thread) {
     ASSERT_MSG(!ShouldWait(thread), "Object unavailable!");

src/core/hle/kernel/process.h

@@ -12,12 +12,17 @@
 #include <vector>
 #include <boost/container/static_vector.hpp>
 #include "common/common_types.h"
+#include "core/hle/kernel/address_arbiter.h"
 #include "core/hle/kernel/handle_table.h"
 #include "core/hle/kernel/process_capability.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/hle/kernel/wait_object.h"
 #include "core/hle/result.h"
 
+namespace Core {
+class System;
+}
+
 namespace FileSys {
 class ProgramMetadata;
 }
@@ -116,7 +121,7 @@ public:
 
     static constexpr std::size_t RANDOM_ENTROPY_SIZE = 4;
 
-    static SharedPtr<Process> Create(KernelCore& kernel, std::string&& name);
+    static SharedPtr<Process> Create(Core::System& system, std::string&& name);
 
     std::string GetTypeName() const override {
         return "Process";
@@ -150,6 +155,16 @@ public:
         return handle_table;
     }
 
+    /// Gets a reference to the process' address arbiter.
+    AddressArbiter& GetAddressArbiter() {
+        return address_arbiter;
+    }
+
+    /// Gets a const reference to the process' address arbiter.
+    const AddressArbiter& GetAddressArbiter() const {
+        return address_arbiter;
+    }
+
     /// Gets the current status of the process
     ProcessStatus GetStatus() const {
         return status;
@@ -251,7 +266,7 @@ public:
     void FreeTLSSlot(VAddr tls_address);
 
 private:
-    explicit Process(KernelCore& kernel);
+    explicit Process(Core::System& system);
     ~Process() override;
 
     /// Checks if the specified thread should wait until this process is available.
@@ -309,9 +324,16 @@ private:
     /// Per-process handle table for storing created object handles in.
     HandleTable handle_table;
 
+    /// Per-process address arbiter.
+    AddressArbiter address_arbiter;
+
     /// Random values for svcGetInfo RandomEntropy
     std::array<u64, RANDOM_ENTROPY_SIZE> random_entropy;
 
+    /// System context
+    Core::System& system;
+
+    /// Name of this process
     std::string name;
 };
 

src/core/hle/kernel/scheduler.cpp

@@ -19,7 +19,8 @@ namespace Kernel {
 
 std::mutex Scheduler::scheduler_mutex;
 
-Scheduler::Scheduler(Core::ARM_Interface& cpu_core) : cpu_core(cpu_core) {}
+Scheduler::Scheduler(Core::System& system, Core::ARM_Interface& cpu_core)
+    : cpu_core{cpu_core}, system{system} {}
 
 Scheduler::~Scheduler() {
     for (auto& thread : thread_list) {
@@ -61,7 +62,7 @@ Thread* Scheduler::PopNextReadyThread() {
 
 void Scheduler::SwitchContext(Thread* new_thread) {
     Thread* const previous_thread = GetCurrentThread();
-    Process* const previous_process = Core::CurrentProcess();
+    Process* const previous_process = system.Kernel().CurrentProcess();
 
     UpdateLastContextSwitchTime(previous_thread, previous_process);
 
@@ -94,8 +95,8 @@ void Scheduler::SwitchContext(Thread* new_thread) {
 
         auto* const thread_owner_process = current_thread->GetOwnerProcess();
         if (previous_process != thread_owner_process) {
-            Core::System::GetInstance().Kernel().MakeCurrentProcess(thread_owner_process);
-            SetCurrentPageTable(&Core::CurrentProcess()->VMManager().page_table);
+            system.Kernel().MakeCurrentProcess(thread_owner_process);
+            SetCurrentPageTable(&thread_owner_process->VMManager().page_table);
         }
 
         cpu_core.LoadContext(new_thread->GetContext());
@@ -111,7 +112,7 @@ void Scheduler::SwitchContext(Thread* new_thread) {
 
 void Scheduler::UpdateLastContextSwitchTime(Thread* thread, Process* process) {
     const u64 prev_switch_ticks = last_context_switch_time;
-    const u64 most_recent_switch_ticks = Core::System::GetInstance().CoreTiming().GetTicks();
+    const u64 most_recent_switch_ticks = system.CoreTiming().GetTicks();
     const u64 update_ticks = most_recent_switch_ticks - prev_switch_ticks;
 
     if (thread != nullptr) {
@@ -223,8 +224,7 @@ void Scheduler::YieldWithLoadBalancing(Thread* thread) {
     // Take the first non-nullptr one
     for (unsigned cur_core = 0; cur_core < Core::NUM_CPU_CORES; ++cur_core) {
         const auto res =
-            Core::System::GetInstance().CpuCore(cur_core).Scheduler().GetNextSuggestedThread(
-                core, priority);
+            system.CpuCore(cur_core).Scheduler().GetNextSuggestedThread(core, priority);
 
         // If scheduler provides a suggested thread
         if (res != nullptr) {

src/core/hle/kernel/scheduler.h

@@ -13,7 +13,8 @@
 
 namespace Core {
 class ARM_Interface;
-}
+class System;
+} // namespace Core
 
 namespace Kernel {
 
@@ -21,7 +22,7 @@ class Process;
 
 class Scheduler final {
 public:
-    explicit Scheduler(Core::ARM_Interface& cpu_core);
+    explicit Scheduler(Core::System& system, Core::ARM_Interface& cpu_core);
     ~Scheduler();
 
     /// Returns whether there are any threads that are ready to run.
@@ -162,6 +163,7 @@ private:
     Core::ARM_Interface& cpu_core;
     u64 last_context_switch_time = 0;
 
+    Core::System& system;
     static std::mutex scheduler_mutex;
 };
 

src/core/hle/kernel/server_port.cpp

@@ -26,6 +26,10 @@ ResultVal<SharedPtr<ServerSession>> ServerPort::Accept() {
     return MakeResult(std::move(session));
 }
 
+void ServerPort::AppendPendingSession(SharedPtr<ServerSession> pending_session) {
+    pending_sessions.push_back(std::move(pending_session));
+}
+
 bool ServerPort::ShouldWait(Thread* thread) const {
     // If there are no pending sessions, we wait until a new one is added.
     return pending_sessions.empty();

src/core/hle/kernel/server_port.h

@@ -22,6 +22,8 @@ class SessionRequestHandler;
 
 class ServerPort final : public WaitObject {
 public:
+    using HLEHandler = std::shared_ptr<SessionRequestHandler>;
+
     /**
      * Creates a pair of ServerPort and an associated ClientPort.
      *
@@ -51,22 +53,27 @@ public:
      */
     ResultVal<SharedPtr<ServerSession>> Accept();
 
+    /// Whether or not this server port has an HLE handler available.
+    bool HasHLEHandler() const {
+        return hle_handler != nullptr;
+    }
+
+    /// Gets the HLE handler for this port.
+    HLEHandler GetHLEHandler() const {
+        return hle_handler;
+    }
+
     /**
      * Sets the HLE handler template for the port. ServerSessions crated by connecting to this port
      * will inherit a reference to this handler.
      */
-    void SetHleHandler(std::shared_ptr<SessionRequestHandler> hle_handler_) {
+    void SetHleHandler(HLEHandler hle_handler_) {
         hle_handler = std::move(hle_handler_);
     }
 
-    std::string name; ///< Name of port (optional)
-
-    /// ServerSessions waiting to be accepted by the port
-    std::vector<SharedPtr<ServerSession>> pending_sessions;
-
-    /// This session's HLE request handler template (optional)
-    /// ServerSessions created from this port inherit a reference to this handler.
-    std::shared_ptr<SessionRequestHandler> hle_handler;
+    /// Appends a ServerSession to the collection of ServerSessions
+    /// waiting to be accepted by this port.
+    void AppendPendingSession(SharedPtr<ServerSession> pending_session);
 
     bool ShouldWait(Thread* thread) const override;
     void Acquire(Thread* thread) override;
@@ -74,6 +81,16 @@ public:
 private:
     explicit ServerPort(KernelCore& kernel);
     ~ServerPort() override;
+
+    /// ServerSessions waiting to be accepted by the port
+    std::vector<SharedPtr<ServerSession>> pending_sessions;
+
+    /// This session's HLE request handler template (optional)
+    /// ServerSessions created from this port inherit a reference to this handler.
+    HLEHandler hle_handler;
+
+    /// Name of the port (optional)
+    std::string name;
 };
 
 } // namespace Kernel

src/core/hle/kernel/server_session.cpp

@@ -63,42 +63,71 @@ void ServerSession::Acquire(Thread* thread) {
     pending_requesting_threads.pop_back();
 }
 
-ResultCode ServerSession::HandleDomainSyncRequest(Kernel::HLERequestContext& context) {
-    auto* const domain_message_header = context.GetDomainMessageHeader();
-    if (domain_message_header) {
-        // Set domain handlers in HLE context, used for domain objects (IPC interfaces) as inputs
-        context.SetDomainRequestHandlers(domain_request_handlers);
-
-        // If there is a DomainMessageHeader, then this is CommandType "Request"
-        const u32 object_id{context.GetDomainMessageHeader()->object_id};
-        switch (domain_message_header->command) {
-        case IPC::DomainMessageHeader::CommandType::SendMessage:
-            if (object_id > domain_request_handlers.size()) {
-                LOG_CRITICAL(IPC,
-                             "object_id {} is too big! This probably means a recent service call "
-                             "to {} needed to return a new interface!",
-                             object_id, name);
-                UNREACHABLE();
-                return RESULT_SUCCESS; // Ignore error if asserts are off
-            }
-            return domain_request_handlers[object_id - 1]->HandleSyncRequest(context);
-
-        case IPC::DomainMessageHeader::CommandType::CloseVirtualHandle: {
-            LOG_DEBUG(IPC, "CloseVirtualHandle, object_id=0x{:08X}", object_id);
-
-            domain_request_handlers[object_id - 1] = nullptr;
-
-            IPC::ResponseBuilder rb{context, 2};
-            rb.Push(RESULT_SUCCESS);
-            return RESULT_SUCCESS;
-        }
-        }
-
-        LOG_CRITICAL(IPC, "Unknown domain command={}",
-                     static_cast<int>(domain_message_header->command.Value()));
-        ASSERT(false);
+void ServerSession::ClientDisconnected() {
+    // We keep a shared pointer to the hle handler to keep it alive throughout
+    // the call to ClientDisconnected, as ClientDisconnected invalidates the
+    // hle_handler member itself during the course of the function executing.
+    std::shared_ptr<SessionRequestHandler> handler = hle_handler;
+    if (handler) {
+        // Note that after this returns, this server session's hle_handler is
+        // invalidated (set to null).
+        handler->ClientDisconnected(this);
     }
 
+    // TODO(Subv): Force a wake up of all the ServerSession's waiting threads and set
+    // their WaitSynchronization result to 0xC920181A.
+
+    // Clean up the list of client threads with pending requests, they are unneeded now that the
+    // client endpoint is closed.
+    pending_requesting_threads.clear();
+    currently_handling = nullptr;
+}
+
+void ServerSession::AppendDomainRequestHandler(std::shared_ptr<SessionRequestHandler> handler) {
+    domain_request_handlers.push_back(std::move(handler));
+}
+
+std::size_t ServerSession::NumDomainRequestHandlers() const {
+    return domain_request_handlers.size();
+}
+
+ResultCode ServerSession::HandleDomainSyncRequest(Kernel::HLERequestContext& context) {
+    if (!context.HasDomainMessageHeader()) {
+        return RESULT_SUCCESS;
+    }
+
+    // Set domain handlers in HLE context, used for domain objects (IPC interfaces) as inputs
+    context.SetDomainRequestHandlers(domain_request_handlers);
+
+    // If there is a DomainMessageHeader, then this is CommandType "Request"
+    const auto& domain_message_header = context.GetDomainMessageHeader();
+    const u32 object_id{domain_message_header.object_id};
+    switch (domain_message_header.command) {
+    case IPC::DomainMessageHeader::CommandType::SendMessage:
+        if (object_id > domain_request_handlers.size()) {
+            LOG_CRITICAL(IPC,
+                         "object_id {} is too big! This probably means a recent service call "
+                         "to {} needed to return a new interface!",
+                         object_id, name);
+            UNREACHABLE();
+            return RESULT_SUCCESS; // Ignore error if asserts are off
+        }
+        return domain_request_handlers[object_id - 1]->HandleSyncRequest(context);
+
+    case IPC::DomainMessageHeader::CommandType::CloseVirtualHandle: {
+        LOG_DEBUG(IPC, "CloseVirtualHandle, object_id=0x{:08X}", object_id);
+
+        domain_request_handlers[object_id - 1] = nullptr;
+
+        IPC::ResponseBuilder rb{context, 2};
+        rb.Push(RESULT_SUCCESS);
+        return RESULT_SUCCESS;
+    }
+    }
+
+    LOG_CRITICAL(IPC, "Unknown domain command={}",
+                 static_cast<int>(domain_message_header.command.Value()));
+    ASSERT(false);
     return RESULT_SUCCESS;
 }
 

src/core/hle/kernel/server_session.h

@@ -46,6 +46,14 @@ public:
         return HANDLE_TYPE;
     }
 
+    Session* GetParent() {
+        return parent.get();
+    }
+
+    const Session* GetParent() const {
+        return parent.get();
+    }
+
     using SessionPair = std::tuple<SharedPtr<ServerSession>, SharedPtr<ClientSession>>;
 
     /**
@@ -78,23 +86,16 @@ public:
 
     void Acquire(Thread* thread) override;
 
-    std::string name;                ///< The name of this session (optional)
-    std::shared_ptr<Session> parent; ///< The parent session, which links to the client endpoint.
-    std::shared_ptr<SessionRequestHandler>
-        hle_handler; ///< This session's HLE request handler (applicable when not a domain)
+    /// Called when a client disconnection occurs.
+    void ClientDisconnected();
 
-    /// This is the list of domain request handlers (after conversion to a domain)
-    std::vector<std::shared_ptr<SessionRequestHandler>> domain_request_handlers;
+    /// Adds a new domain request handler to the collection of request handlers within
+    /// this ServerSession instance.
+    void AppendDomainRequestHandler(std::shared_ptr<SessionRequestHandler> handler);
 
-    /// List of threads that are pending a response after a sync request. This list is processed in
-    /// a LIFO manner, thus, the last request will be dispatched first.
-    /// TODO(Subv): Verify if this is indeed processed in LIFO using a hardware test.
-    std::vector<SharedPtr<Thread>> pending_requesting_threads;
-
-    /// Thread whose request is currently being handled. A request is considered "handled" when a
-    /// response is sent via svcReplyAndReceive.
-    /// TODO(Subv): Find a better name for this.
-    SharedPtr<Thread> currently_handling;
+    /// Retrieves the total number of domain request handlers that have been
+    /// appended to this ServerSession instance.
+    std::size_t NumDomainRequestHandlers() const;
 
     /// Returns true if the session has been converted to a domain, otherwise False
     bool IsDomain() const {
@@ -129,8 +130,30 @@ private:
     /// object handle.
     ResultCode HandleDomainSyncRequest(Kernel::HLERequestContext& context);
 
+    /// The parent session, which links to the client endpoint.
+    std::shared_ptr<Session> parent;
+
+    /// This session's HLE request handler (applicable when not a domain)
+    std::shared_ptr<SessionRequestHandler> hle_handler;
+
+    /// This is the list of domain request handlers (after conversion to a domain)
+    std::vector<std::shared_ptr<SessionRequestHandler>> domain_request_handlers;
+
+    /// List of threads that are pending a response after a sync request. This list is processed in
+    /// a LIFO manner, thus, the last request will be dispatched first.
+    /// TODO(Subv): Verify if this is indeed processed in LIFO using a hardware test.
+    std::vector<SharedPtr<Thread>> pending_requesting_threads;
+
+    /// Thread whose request is currently being handled. A request is considered "handled" when a
+    /// response is sent via svcReplyAndReceive.
+    /// TODO(Subv): Find a better name for this.
+    SharedPtr<Thread> currently_handling;
+
     /// When set to True, converts the session to a domain at the end of the command
     bool convert_to_domain{};
+
+    /// The name of this session (optional)
+    std::string name;
 };
 
 } // namespace Kernel

src/core/hle/kernel/shared_memory.cpp

@@ -6,7 +6,6 @@
 
 #include "common/assert.h"
 #include "common/logging/log.h"
-#include "core/core.h"
 #include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/shared_memory.h"
@@ -34,8 +33,8 @@ SharedPtr<SharedMemory> SharedMemory::Create(KernelCore& kernel, Process* owner_
         shared_memory->backing_block_offset = 0;
 
         // Refresh the address mappings for the current process.
-        if (Core::CurrentProcess() != nullptr) {
-            Core::CurrentProcess()->VMManager().RefreshMemoryBlockMappings(
+        if (kernel.CurrentProcess() != nullptr) {
+            kernel.CurrentProcess()->VMManager().RefreshMemoryBlockMappings(
                 shared_memory->backing_block.get());
         }
     } else {

src/core/hle/kernel/svc.cpp

@@ -20,6 +20,7 @@
 #include "core/hle/kernel/address_arbiter.h"
 #include "core/hle/kernel/client_port.h"
 #include "core/hle/kernel/client_session.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"
@@ -1478,21 +1479,10 @@ static ResultCode WaitForAddress(VAddr address, u32 type, s32 value, s64 timeout
         return ERR_INVALID_ADDRESS;
     }
 
-    auto& address_arbiter = Core::System::GetInstance().Kernel().AddressArbiter();
-    switch (static_cast<AddressArbiter::ArbitrationType>(type)) {
-    case AddressArbiter::ArbitrationType::WaitIfLessThan:
-        return address_arbiter.WaitForAddressIfLessThan(address, value, timeout, false);
-    case AddressArbiter::ArbitrationType::DecrementAndWaitIfLessThan:
-        return address_arbiter.WaitForAddressIfLessThan(address, value, timeout, true);
-    case AddressArbiter::ArbitrationType::WaitIfEqual:
-        return address_arbiter.WaitForAddressIfEqual(address, value, timeout);
-    default:
-        LOG_ERROR(Kernel_SVC,
-                  "Invalid arbitration type, expected WaitIfLessThan, DecrementAndWaitIfLessThan "
-                  "or WaitIfEqual but got {}",
-                  type);
-        return ERR_INVALID_ENUM_VALUE;
-    }
+    const auto arbitration_type = static_cast<AddressArbiter::ArbitrationType>(type);
+    auto& address_arbiter =
+        Core::System::GetInstance().Kernel().CurrentProcess()->GetAddressArbiter();
+    return address_arbiter.WaitForAddress(address, arbitration_type, value, timeout);
 }
 
 // Signals to an address (via Address Arbiter)
@@ -1510,22 +1500,10 @@ static ResultCode SignalToAddress(VAddr address, u32 type, s32 value, s32 num_to
         return ERR_INVALID_ADDRESS;
     }
 
-    auto& address_arbiter = Core::System::GetInstance().Kernel().AddressArbiter();
-    switch (static_cast<AddressArbiter::SignalType>(type)) {
-    case AddressArbiter::SignalType::Signal:
-        return address_arbiter.SignalToAddress(address, num_to_wake);
-    case AddressArbiter::SignalType::IncrementAndSignalIfEqual:
-        return address_arbiter.IncrementAndSignalToAddressIfEqual(address, value, num_to_wake);
-    case AddressArbiter::SignalType::ModifyByWaitingCountAndSignalIfEqual:
-        return address_arbiter.ModifyByWaitingCountAndSignalToAddressIfEqual(address, value,
-                                                                             num_to_wake);
-    default:
-        LOG_ERROR(Kernel_SVC,
-                  "Invalid signal type, expected Signal, IncrementAndSignalIfEqual "
-                  "or ModifyByWaitingCountAndSignalIfEqual but got {}",
-                  type);
-        return ERR_INVALID_ENUM_VALUE;
-    }
+    const auto signal_type = static_cast<AddressArbiter::SignalType>(type);
+    auto& address_arbiter =
+        Core::System::GetInstance().Kernel().CurrentProcess()->GetAddressArbiter();
+    return address_arbiter.SignalToAddress(address, signal_type, value, num_to_wake);
 }
 
 /// This returns the total CPU ticks elapsed since the CPU was powered-on

src/core/hle/result.h

@@ -8,19 +8,10 @@
 #include <utility>
 #include "common/assert.h"
 #include "common/bit_field.h"
-#include "common/common_funcs.h"
 #include "common/common_types.h"
 
 // All the constants in this file come from http://switchbrew.org/index.php?title=Error_codes
 
-/**
- * Detailed description of the error. Code 0 always means success.
- */
-enum class ErrorDescription : u32 {
-    Success = 0,
-    RemoteProcessDead = 301,
-};
-
 /**
  * Identifies the module which caused the error. Error codes can be propagated through a call
  * chain, meaning that this doesn't always correspond to the module where the API call made is
@@ -121,7 +112,7 @@ enum class ErrorModule : u32 {
     ShopN = 811,
 };
 
-/// Encapsulates a CTR-OS error code, allowing it to be separated into its constituent fields.
+/// Encapsulates a Horizon OS error code, allowing it to be separated into its constituent fields.
 union ResultCode {
     u32 raw;
 
@@ -134,17 +125,9 @@ union ResultCode {
 
     constexpr explicit ResultCode(u32 raw) : raw(raw) {}
 
-    constexpr ResultCode(ErrorModule module, ErrorDescription description)
-        : ResultCode(module, static_cast<u32>(description)) {}
-
     constexpr ResultCode(ErrorModule module_, u32 description_)
         : raw(module.FormatValue(module_) | description.FormatValue(description_)) {}
 
-    constexpr ResultCode& operator=(const ResultCode& o) {
-        raw = o.raw;
-        return *this;
-    }
-
     constexpr bool IsSuccess() const {
         return raw == 0;
     }

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

@@ -7,6 +7,7 @@
 #include "common/string_util.h"
 #include "core/core.h"
 #include "core/frontend/applets/software_keyboard.h"
+#include "core/hle/result.h"
 #include "core/hle/service/am/am.h"
 #include "core/hle/service/am/applets/software_keyboard.h"
 

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

@@ -9,10 +9,13 @@
 #include <vector>
 
 #include "common/common_funcs.h"
+#include "common/common_types.h"
 #include "common/swap.h"
 #include "core/hle/service/am/am.h"
 #include "core/hle/service/am/applets/applets.h"
 
+union ResultCode;
+
 namespace Service::AM::Applets {
 
 enum class KeysetDisable : u32 {

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

@@ -107,7 +107,9 @@ private:
     void StopAudioOut(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Service_Audio, "called");
 
-        audio_core.StopStream(stream);
+        if (stream->IsPlaying()) {
+            audio_core.StopStream(stream);
+        }
 
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(RESULT_SUCCESS);

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

@@ -9,43 +9,32 @@
 
 #include <opus.h>
 
-#include "common/common_funcs.h"
+#include "common/assert.h"
 #include "common/logging/log.h"
 #include "core/hle/ipc_helpers.h"
 #include "core/hle/kernel/hle_ipc.h"
 #include "core/hle/service/audio/hwopus.h"
 
 namespace Service::Audio {
-
+namespace {
 struct OpusDeleter {
     void operator()(void* ptr) const {
         operator delete(ptr);
     }
 };
 
-class IHardwareOpusDecoderManager final : public ServiceFramework<IHardwareOpusDecoderManager> {
+using OpusDecoderPtr = std::unique_ptr<OpusDecoder, OpusDeleter>;
+
+struct OpusPacketHeader {
+    // Packet size in bytes.
+    u32_be size;
+    // Indicates the final range of the codec's entropy coder.
+    u32_be final_range;
+};
+static_assert(sizeof(OpusPacketHeader) == 0x8, "OpusHeader is an invalid size");
+
+class OpusDecoderStateBase {
 public:
-    IHardwareOpusDecoderManager(std::unique_ptr<OpusDecoder, OpusDeleter> decoder, u32 sample_rate,
-                                u32 channel_count)
-        : ServiceFramework("IHardwareOpusDecoderManager"), decoder(std::move(decoder)),
-          sample_rate(sample_rate), channel_count(channel_count) {
-        // clang-format off
-        static const FunctionInfo functions[] = {
-            {0, &IHardwareOpusDecoderManager::DecodeInterleavedOld, "DecodeInterleavedOld"},
-            {1, nullptr, "SetContext"},
-            {2, nullptr, "DecodeInterleavedForMultiStreamOld"},
-            {3, nullptr, "SetContextForMultiStream"},
-            {4, &IHardwareOpusDecoderManager::DecodeInterleavedWithPerfOld, "DecodeInterleavedWithPerfOld"},
-            {5, nullptr, "DecodeInterleavedForMultiStreamWithPerfOld"},
-            {6, &IHardwareOpusDecoderManager::DecodeInterleaved, "DecodeInterleaved"},
-            {7, nullptr, "DecodeInterleavedForMultiStream"},
-        };
-        // clang-format on
-
-        RegisterHandlers(functions);
-    }
-
-private:
     /// Describes extra behavior that may be asked of the decoding context.
     enum class ExtraBehavior {
         /// No extra behavior.
@@ -55,30 +44,36 @@ private:
         ResetContext,
     };
 
-    void DecodeInterleavedOld(Kernel::HLERequestContext& ctx) {
-        LOG_DEBUG(Audio, "called");
+    enum class PerfTime {
+        Disabled,
+        Enabled,
+    };
 
-        DecodeInterleavedHelper(ctx, nullptr, ExtraBehavior::None);
-    }
-
-    void DecodeInterleavedWithPerfOld(Kernel::HLERequestContext& ctx) {
-        LOG_DEBUG(Audio, "called");
-
-        u64 performance = 0;
-        DecodeInterleavedHelper(ctx, &performance, ExtraBehavior::None);
-    }
-
-    void DecodeInterleaved(Kernel::HLERequestContext& ctx) {
-        LOG_DEBUG(Audio, "called");
-
-        IPC::RequestParser rp{ctx};
-        const auto extra_behavior =
-            rp.Pop<bool>() ? ExtraBehavior::ResetContext : ExtraBehavior::None;
-
-        u64 performance = 0;
-        DecodeInterleavedHelper(ctx, &performance, extra_behavior);
+    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} {}
+
+    void DecodeInterleaved(Kernel::HLERequestContext& ctx, PerfTime perf_time,
+                           ExtraBehavior extra_behavior) override {
+        if (perf_time == PerfTime::Disabled) {
+            DecodeInterleavedHelper(ctx, nullptr, extra_behavior);
+        } else {
+            u64 performance = 0;
+            DecodeInterleavedHelper(ctx, &performance, extra_behavior);
+        }
     }
 
+private:
     void DecodeInterleavedHelper(Kernel::HLERequestContext& ctx, u64* performance,
                                  ExtraBehavior extra_behavior) {
         u32 consumed = 0;
@@ -89,8 +84,7 @@ private:
             ResetDecoderContext();
         }
 
-        if (!Decoder_DecodeInterleaved(consumed, sample_count, ctx.ReadBuffer(), samples,
-                                       performance)) {
+        if (!DecodeOpusData(consumed, sample_count, ctx.ReadBuffer(), samples, performance)) {
             LOG_ERROR(Audio, "Failed to decode opus data");
             IPC::ResponseBuilder rb{ctx, 2};
             // TODO(ogniK): Use correct error code
@@ -109,27 +103,27 @@ private:
         ctx.WriteBuffer(samples.data(), samples.size() * sizeof(s16));
     }
 
-    bool Decoder_DecodeInterleaved(u32& consumed, u32& sample_count, const std::vector<u8>& input,
-                                   std::vector<opus_int16>& output, u64* out_performance_time) {
+    bool DecodeOpusData(u32& consumed, u32& sample_count, const std::vector<u8>& input,
+                        std::vector<opus_int16>& output, u64* out_performance_time) const {
         const auto start_time = std::chrono::high_resolution_clock::now();
         const std::size_t raw_output_sz = output.size() * sizeof(opus_int16);
-        if (sizeof(OpusHeader) > input.size()) {
+        if (sizeof(OpusPacketHeader) > input.size()) {
             LOG_ERROR(Audio, "Input is smaller than the header size, header_sz={}, input_sz={}",
-                      sizeof(OpusHeader), input.size());
+                      sizeof(OpusPacketHeader), input.size());
             return false;
         }
 
-        OpusHeader hdr{};
-        std::memcpy(&hdr, input.data(), sizeof(OpusHeader));
-        if (sizeof(OpusHeader) + static_cast<u32>(hdr.sz) > input.size()) {
+        OpusPacketHeader hdr{};
+        std::memcpy(&hdr, input.data(), sizeof(OpusPacketHeader));
+        if (sizeof(OpusPacketHeader) + static_cast<u32>(hdr.size) > input.size()) {
             LOG_ERROR(Audio, "Input does not fit in the opus header size. data_sz={}, input_sz={}",
-                      sizeof(OpusHeader) + static_cast<u32>(hdr.sz), input.size());
+                      sizeof(OpusPacketHeader) + static_cast<u32>(hdr.size), input.size());
             return false;
         }
 
-        const auto frame = input.data() + sizeof(OpusHeader);
+        const auto frame = input.data() + sizeof(OpusPacketHeader);
         const auto decoded_sample_count = opus_packet_get_nb_samples(
-            frame, static_cast<opus_int32>(input.size() - sizeof(OpusHeader)),
+            frame, static_cast<opus_int32>(input.size() - sizeof(OpusPacketHeader)),
             static_cast<opus_int32>(sample_rate));
         if (decoded_sample_count * channel_count * sizeof(u16) > raw_output_sz) {
             LOG_ERROR(
@@ -141,18 +135,18 @@ 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.sz, output.data(), frame_size, 0);
+            opus_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, "
                       "output_sample_count={}, frame_size={}, data_sz_from_hdr={}",
-                      out_sample_count, frame_size, static_cast<u32>(hdr.sz));
+                      out_sample_count, frame_size, static_cast<u32>(hdr.size));
             return false;
         }
 
         const auto end_time = std::chrono::high_resolution_clock::now() - start_time;
         sample_count = out_sample_count;
-        consumed = static_cast<u32>(sizeof(OpusHeader) + hdr.sz);
+        consumed = static_cast<u32>(sizeof(OpusPacketHeader) + hdr.size);
         if (out_performance_time != nullptr) {
             *out_performance_time =
                 std::chrono::duration_cast<std::chrono::milliseconds>(end_time).count();
@@ -167,21 +161,66 @@ private:
         opus_decoder_ctl(decoder.get(), OPUS_RESET_STATE);
     }
 
-    struct OpusHeader {
-        u32_be sz; // Needs to be BE for some odd reason
-        INSERT_PADDING_WORDS(1);
-    };
-    static_assert(sizeof(OpusHeader) == 0x8, "OpusHeader is an invalid size");
-
-    std::unique_ptr<OpusDecoder, OpusDeleter> decoder;
+    OpusDecoderPtr decoder;
     u32 sample_rate;
     u32 channel_count;
 };
 
-static std::size_t WorkerBufferSize(u32 channel_count) {
+class IHardwareOpusDecoderManager final : public ServiceFramework<IHardwareOpusDecoderManager> {
+public:
+    explicit IHardwareOpusDecoderManager(std::unique_ptr<OpusDecoderStateBase> decoder_state)
+        : ServiceFramework("IHardwareOpusDecoderManager"), decoder_state{std::move(decoder_state)} {
+        // clang-format off
+        static const FunctionInfo functions[] = {
+            {0, &IHardwareOpusDecoderManager::DecodeInterleavedOld, "DecodeInterleavedOld"},
+            {1, nullptr, "SetContext"},
+            {2, nullptr, "DecodeInterleavedForMultiStreamOld"},
+            {3, nullptr, "SetContextForMultiStream"},
+            {4, &IHardwareOpusDecoderManager::DecodeInterleavedWithPerfOld, "DecodeInterleavedWithPerfOld"},
+            {5, nullptr, "DecodeInterleavedForMultiStreamWithPerfOld"},
+            {6, &IHardwareOpusDecoderManager::DecodeInterleaved, "DecodeInterleaved"},
+            {7, nullptr, "DecodeInterleavedForMultiStream"},
+        };
+        // clang-format on
+
+        RegisterHandlers(functions);
+    }
+
+private:
+    void DecodeInterleavedOld(Kernel::HLERequestContext& ctx) {
+        LOG_DEBUG(Audio, "called");
+
+        decoder_state->DecodeInterleaved(ctx, OpusDecoderStateBase::PerfTime::Disabled,
+                                         OpusDecoderStateBase::ExtraBehavior::None);
+    }
+
+    void DecodeInterleavedWithPerfOld(Kernel::HLERequestContext& ctx) {
+        LOG_DEBUG(Audio, "called");
+
+        decoder_state->DecodeInterleaved(ctx, OpusDecoderStateBase::PerfTime::Enabled,
+                                         OpusDecoderStateBase::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;
+
+        decoder_state->DecodeInterleaved(ctx, OpusDecoderStateBase::PerfTime::Enabled,
+                                         extra_behavior);
+    }
+
+    std::unique_ptr<OpusDecoderStateBase> 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));
 }
+} // Anonymous namespace
 
 void HwOpus::GetWorkBufferSize(Kernel::HLERequestContext& ctx) {
     IPC::RequestParser rp{ctx};
@@ -220,8 +259,7 @@ 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");
 
-    std::unique_ptr<OpusDecoder, OpusDeleter> decoder{
-        static_cast<OpusDecoder*>(operator new(worker_sz))};
+    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);
         IPC::ResponseBuilder rb{ctx, 2};
@@ -232,8 +270,8 @@ void HwOpus::OpenOpusDecoder(Kernel::HLERequestContext& ctx) {
 
     IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<IHardwareOpusDecoderManager>(std::move(decoder), sample_rate,
-                                                     channel_count);
+    rb.PushIpcInterface<IHardwareOpusDecoderManager>(
+        std::make_unique<OpusDecoderState>(std::move(decoder), sample_rate, channel_count));
 }
 
 HwOpus::HwOpus() : ServiceFramework("hwopus") {

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/hle/service/service.cpp

@@ -11,7 +11,6 @@
 #include "core/hle/ipc.h"
 #include "core/hle/ipc_helpers.h"
 #include "core/hle/kernel/client_port.h"
-#include "core/hle/kernel/handle_table.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/server_port.h"
@@ -76,7 +75,8 @@ namespace Service {
  * Creates a function string for logging, complete with the name (or header code, depending
  * on what's passed in) the port name, and all the cmd_buff arguments.
  */
-[[maybe_unused]] static std::string MakeFunctionString(const char* name, const char* port_name,
+[[maybe_unused]] static std::string MakeFunctionString(std::string_view name,
+                                                       std::string_view port_name,
                                                        const u32* cmd_buff) {
     // Number of params == bits 0-5 + bits 6-11
     int num_params = (cmd_buff[0] & 0x3F) + ((cmd_buff[0] >> 6) & 0x3F);
@@ -158,9 +158,7 @@ void ServiceFrameworkBase::InvokeRequest(Kernel::HLERequestContext& ctx) {
         return ReportUnimplementedFunction(ctx, info);
     }
 
-    LOG_TRACE(
-        Service, "{}",
-        MakeFunctionString(info->name, GetServiceName().c_str(), ctx.CommandBuffer()).c_str());
+    LOG_TRACE(Service, "{}", MakeFunctionString(info->name, GetServiceName(), ctx.CommandBuffer()));
     handler_invoker(this, info->handler_callback, ctx);
 }
 
@@ -169,7 +167,7 @@ ResultCode ServiceFrameworkBase::HandleSyncRequest(Kernel::HLERequestContext& co
     case IPC::CommandType::Close: {
         IPC::ResponseBuilder rb{context, 2};
         rb.Push(RESULT_SUCCESS);
-        return ResultCode(ErrorModule::HIPC, ErrorDescription::RemoteProcessDead);
+        return IPC::ERR_REMOTE_PROCESS_DEAD;
     }
     case IPC::CommandType::ControlWithContext:
     case IPC::CommandType::Control: {

src/core/hle/service/sm/controller.cpp

@@ -30,7 +30,7 @@ void Controller::DuplicateSession(Kernel::HLERequestContext& ctx) {
 
     IPC::ResponseBuilder rb{ctx, 2, 0, 1, IPC::ResponseBuilder::Flags::AlwaysMoveHandles};
     rb.Push(RESULT_SUCCESS);
-    Kernel::SharedPtr<Kernel::ClientSession> session{ctx.Session()->parent->client};
+    Kernel::SharedPtr<Kernel::ClientSession> session{ctx.Session()->GetParent()->client};
     rb.PushMoveObjects(session);
 
     LOG_DEBUG(Service, "session={}", session->GetObjectId());

src/core/hle/service/sm/sm.h

@@ -67,7 +67,7 @@ public:
         if (port == nullptr) {
             return nullptr;
         }
-        return std::static_pointer_cast<T>(port->hle_handler);
+        return std::static_pointer_cast<T>(port->GetHLEHandler());
     }
 
     void InvokeControlRequest(Kernel::HLERequestContext& context);

src/core/hle/service/vi/vi.cpp

@@ -24,6 +24,7 @@
 #include "core/hle/service/nvdrv/nvdrv.h"
 #include "core/hle/service/nvflinger/buffer_queue.h"
 #include "core/hle/service/nvflinger/nvflinger.h"
+#include "core/hle/service/service.h"
 #include "core/hle/service/vi/vi.h"
 #include "core/hle/service/vi/vi_m.h"
 #include "core/hle/service/vi/vi_s.h"
@@ -33,6 +34,7 @@
 namespace Service::VI {
 
 constexpr ResultCode ERR_OPERATION_FAILED{ErrorModule::VI, 1};
+constexpr ResultCode ERR_PERMISSION_DENIED{ErrorModule::VI, 5};
 constexpr ResultCode ERR_UNSUPPORTED{ErrorModule::VI, 6};
 constexpr ResultCode ERR_NOT_FOUND{ErrorModule::VI, 7};
 
@@ -1203,26 +1205,40 @@ IApplicationDisplayService::IApplicationDisplayService(
     RegisterHandlers(functions);
 }
 
-Module::Interface::Interface(std::shared_ptr<Module> module, const char* name,
-                             std::shared_ptr<NVFlinger::NVFlinger> nv_flinger)
-    : ServiceFramework(name), module(std::move(module)), nv_flinger(std::move(nv_flinger)) {}
+static bool IsValidServiceAccess(Permission permission, Policy policy) {
+    if (permission == Permission::User) {
+        return policy == Policy::User;
+    }
 
-Module::Interface::~Interface() = default;
+    if (permission == Permission::System || permission == Permission::Manager) {
+        return policy == Policy::User || policy == Policy::Compositor;
+    }
 
-void Module::Interface::GetDisplayService(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_VI, "(STUBBED) called");
+    return false;
+}
+
+void detail::GetDisplayServiceImpl(Kernel::HLERequestContext& ctx,
+                                   std::shared_ptr<NVFlinger::NVFlinger> nv_flinger,
+                                   Permission permission) {
+    IPC::RequestParser rp{ctx};
+    const auto policy = rp.PopEnum<Policy>();
+
+    if (!IsValidServiceAccess(permission, policy)) {
+        IPC::ResponseBuilder rb{ctx, 2};
+        rb.Push(ERR_PERMISSION_DENIED);
+        return;
+    }
 
     IPC::ResponseBuilder rb{ctx, 2, 0, 1};
     rb.Push(RESULT_SUCCESS);
-    rb.PushIpcInterface<IApplicationDisplayService>(nv_flinger);
+    rb.PushIpcInterface<IApplicationDisplayService>(std::move(nv_flinger));
 }
 
 void InstallInterfaces(SM::ServiceManager& service_manager,
                        std::shared_ptr<NVFlinger::NVFlinger> nv_flinger) {
-    auto module = std::make_shared<Module>();
-    std::make_shared<VI_M>(module, nv_flinger)->InstallAsService(service_manager);
-    std::make_shared<VI_S>(module, nv_flinger)->InstallAsService(service_manager);
-    std::make_shared<VI_U>(module, nv_flinger)->InstallAsService(service_manager);
+    std::make_shared<VI_M>(nv_flinger)->InstallAsService(service_manager);
+    std::make_shared<VI_S>(nv_flinger)->InstallAsService(service_manager);
+    std::make_shared<VI_U>(nv_flinger)->InstallAsService(service_manager);
 }
 
 } // namespace Service::VI

src/core/hle/service/vi/vi.h

@@ -4,12 +4,21 @@
 
 #pragma once
 
-#include "core/hle/service/service.h"
+#include <memory>
+#include "common/common_types.h"
+
+namespace Kernel {
+class HLERequestContext;
+}
 
 namespace Service::NVFlinger {
 class NVFlinger;
 }
 
+namespace Service::SM {
+class ServiceManager;
+}
+
 namespace Service::VI {
 
 enum class DisplayResolution : u32 {
@@ -19,22 +28,25 @@ enum class DisplayResolution : u32 {
     UndockedHeight = 720,
 };
 
-class Module final {
-public:
-    class Interface : public ServiceFramework<Interface> {
-    public:
-        explicit Interface(std::shared_ptr<Module> module, const char* name,
-                           std::shared_ptr<NVFlinger::NVFlinger> nv_flinger);
-        ~Interface() override;
-
-        void GetDisplayService(Kernel::HLERequestContext& ctx);
-
-    protected:
-        std::shared_ptr<Module> module;
-        std::shared_ptr<NVFlinger::NVFlinger> nv_flinger;
-    };
+/// Permission level for a particular VI service instance
+enum class Permission {
+    User,
+    System,
+    Manager,
 };
 
+/// A policy type that may be requested via GetDisplayService and
+/// GetDisplayServiceWithProxyNameExchange
+enum class Policy {
+    User,
+    Compositor,
+};
+
+namespace detail {
+void GetDisplayServiceImpl(Kernel::HLERequestContext& ctx,
+                           std::shared_ptr<NVFlinger::NVFlinger> nv_flinger, Permission permission);
+} // namespace detail
+
 /// Registers all VI services with the specified service manager.
 void InstallInterfaces(SM::ServiceManager& service_manager,
                        std::shared_ptr<NVFlinger::NVFlinger> nv_flinger);

src/core/hle/service/vi/vi_m.cpp

@@ -2,12 +2,14 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "common/logging/log.h"
+#include "core/hle/service/vi/vi.h"
 #include "core/hle/service/vi/vi_m.h"
 
 namespace Service::VI {
 
-VI_M::VI_M(std::shared_ptr<Module> module, std::shared_ptr<NVFlinger::NVFlinger> nv_flinger)
-    : Module::Interface(std::move(module), "vi:m", std::move(nv_flinger)) {
+VI_M::VI_M(std::shared_ptr<NVFlinger::NVFlinger> nv_flinger)
+    : ServiceFramework{"vi:m"}, nv_flinger{std::move(nv_flinger)} {
     static const FunctionInfo functions[] = {
         {2, &VI_M::GetDisplayService, "GetDisplayService"},
         {3, nullptr, "GetDisplayServiceWithProxyNameExchange"},
@@ -17,4 +19,10 @@ VI_M::VI_M(std::shared_ptr<Module> module, std::shared_ptr<NVFlinger::NVFlinger>
 
 VI_M::~VI_M() = default;
 
+void VI_M::GetDisplayService(Kernel::HLERequestContext& ctx) {
+    LOG_DEBUG(Service_VI, "called");
+
+    detail::GetDisplayServiceImpl(ctx, nv_flinger, Permission::Manager);
+}
+
 } // namespace Service::VI

src/core/hle/service/vi/vi_m.h

@@ -4,14 +4,27 @@
 
 #pragma once
 
-#include "core/hle/service/vi/vi.h"
+#include "core/hle/service/service.h"
+
+namespace Kernel {
+class HLERequestContext;
+}
+
+namespace Service::NVFlinger {
+class NVFlinger;
+}
 
 namespace Service::VI {
 
-class VI_M final : public Module::Interface {
+class VI_M final : public ServiceFramework<VI_M> {
 public:
-    explicit VI_M(std::shared_ptr<Module> module, std::shared_ptr<NVFlinger::NVFlinger> nv_flinger);
+    explicit VI_M(std::shared_ptr<NVFlinger::NVFlinger> nv_flinger);
     ~VI_M() override;
+
+private:
+    void GetDisplayService(Kernel::HLERequestContext& ctx);
+
+    std::shared_ptr<NVFlinger::NVFlinger> nv_flinger;
 };
 
 } // namespace Service::VI

src/core/hle/service/vi/vi_s.cpp

@@ -2,12 +2,14 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "common/logging/log.h"
+#include "core/hle/service/vi/vi.h"
 #include "core/hle/service/vi/vi_s.h"
 
 namespace Service::VI {
 
-VI_S::VI_S(std::shared_ptr<Module> module, std::shared_ptr<NVFlinger::NVFlinger> nv_flinger)
-    : Module::Interface(std::move(module), "vi:s", std::move(nv_flinger)) {
+VI_S::VI_S(std::shared_ptr<NVFlinger::NVFlinger> nv_flinger)
+    : ServiceFramework{"vi:s"}, nv_flinger{std::move(nv_flinger)} {
     static const FunctionInfo functions[] = {
         {1, &VI_S::GetDisplayService, "GetDisplayService"},
         {3, nullptr, "GetDisplayServiceWithProxyNameExchange"},
@@ -17,4 +19,10 @@ VI_S::VI_S(std::shared_ptr<Module> module, std::shared_ptr<NVFlinger::NVFlinger>
 
 VI_S::~VI_S() = default;
 
+void VI_S::GetDisplayService(Kernel::HLERequestContext& ctx) {
+    LOG_DEBUG(Service_VI, "called");
+
+    detail::GetDisplayServiceImpl(ctx, nv_flinger, Permission::System);
+}
+
 } // namespace Service::VI

src/core/hle/service/vi/vi_s.h

@@ -4,14 +4,27 @@
 
 #pragma once
 
-#include "core/hle/service/vi/vi.h"
+#include "core/hle/service/service.h"
+
+namespace Kernel {
+class HLERequestContext;
+}
+
+namespace Service::NVFlinger {
+class NVFlinger;
+}
 
 namespace Service::VI {
 
-class VI_S final : public Module::Interface {
+class VI_S final : public ServiceFramework<VI_S> {
 public:
-    explicit VI_S(std::shared_ptr<Module> module, std::shared_ptr<NVFlinger::NVFlinger> nv_flinger);
+    explicit VI_S(std::shared_ptr<NVFlinger::NVFlinger> nv_flinger);
     ~VI_S() override;
+
+private:
+    void GetDisplayService(Kernel::HLERequestContext& ctx);
+
+    std::shared_ptr<NVFlinger::NVFlinger> nv_flinger;
 };
 
 } // namespace Service::VI

src/core/hle/service/vi/vi_u.cpp

@@ -2,12 +2,14 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "common/logging/log.h"
+#include "core/hle/service/vi/vi.h"
 #include "core/hle/service/vi/vi_u.h"
 
 namespace Service::VI {
 
-VI_U::VI_U(std::shared_ptr<Module> module, std::shared_ptr<NVFlinger::NVFlinger> nv_flinger)
-    : Module::Interface(std::move(module), "vi:u", std::move(nv_flinger)) {
+VI_U::VI_U(std::shared_ptr<NVFlinger::NVFlinger> nv_flinger)
+    : ServiceFramework{"vi:u"}, nv_flinger{std::move(nv_flinger)} {
     static const FunctionInfo functions[] = {
         {0, &VI_U::GetDisplayService, "GetDisplayService"},
     };
@@ -16,4 +18,10 @@ VI_U::VI_U(std::shared_ptr<Module> module, std::shared_ptr<NVFlinger::NVFlinger>
 
 VI_U::~VI_U() = default;
 
+void VI_U::GetDisplayService(Kernel::HLERequestContext& ctx) {
+    LOG_DEBUG(Service_VI, "called");
+
+    detail::GetDisplayServiceImpl(ctx, nv_flinger, Permission::User);
+}
+
 } // namespace Service::VI

src/core/hle/service/vi/vi_u.h

@@ -4,14 +4,27 @@
 
 #pragma once
 
-#include "core/hle/service/vi/vi.h"
+#include "core/hle/service/service.h"
+
+namespace Kernel {
+class HLERequestContext;
+}
+
+namespace Service::NVFlinger {
+class NVFlinger;
+}
 
 namespace Service::VI {
 
-class VI_U final : public Module::Interface {
+class VI_U final : public ServiceFramework<VI_U> {
 public:
-    explicit VI_U(std::shared_ptr<Module> module, std::shared_ptr<NVFlinger::NVFlinger> nv_flinger);
+    explicit VI_U(std::shared_ptr<NVFlinger::NVFlinger> nv_flinger);
     ~VI_U() override;
+
+private:
+    void GetDisplayService(Kernel::HLERequestContext& ctx);
+
+    std::shared_ptr<NVFlinger::NVFlinger> nv_flinger;
 };
 
 } // namespace Service::VI

src/core/memory.cpp

@@ -18,6 +18,7 @@
 #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 {
@@ -67,8 +68,11 @@ static void MapPages(PageTable& page_table, VAddr base, u64 size, u8* memory, Pa
     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}",
@@ -180,10 +184,10 @@ T Read(const VAddr vaddr) {
         ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr);
         break;
     case PageType::RasterizerCachedMemory: {
-        RasterizerFlushVirtualRegion(vaddr, sizeof(T), FlushMode::Flush);
-
+        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:
@@ -211,8 +215,9 @@ void Write(const VAddr vaddr, const T data) {
         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));
+        auto host_ptr{GetPointerFromVMA(vaddr)};
+        Core::System::GetInstance().GPU().InvalidateRegion(ToCacheAddr(host_ptr), sizeof(T));
+        std::memcpy(host_ptr, &data, sizeof(T));
         break;
     }
     default:
@@ -335,47 +340,6 @@ 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);
 }
@@ -421,9 +385,9 @@ void ReadBlock(const Kernel::Process& process, const VAddr src_addr, void* dest_
             break;
         }
         case PageType::RasterizerCachedMemory: {
-            RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::Flush);
-            std::memcpy(dest_buffer, GetPointerFromVMA(process, current_vaddr), copy_amount);
+            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:
@@ -484,9 +448,9 @@ void WriteBlock(const Kernel::Process& process, const VAddr dest_addr, const voi
             break;
         }
         case PageType::RasterizerCachedMemory: {
-            RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::Invalidate);
-            std::memcpy(GetPointerFromVMA(process, current_vaddr), src_buffer, copy_amount);
+            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:
@@ -530,9 +494,9 @@ void ZeroBlock(const Kernel::Process& process, const VAddr dest_addr, const std:
             break;
         }
         case PageType::RasterizerCachedMemory: {
-            RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::Invalidate);
-            std::memset(GetPointerFromVMA(process, current_vaddr), 0, copy_amount);
+            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:
@@ -572,9 +536,9 @@ void CopyBlock(const Kernel::Process& process, VAddr dest_addr, VAddr src_addr,
             break;
         }
         case PageType::RasterizerCachedMemory: {
-            RasterizerFlushVirtualRegion(current_vaddr, static_cast<u32>(copy_amount),
-                                         FlushMode::Flush);
-            WriteBlock(process, dest_addr, GetPointerFromVMA(process, current_vaddr), copy_amount);
+            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

@@ -161,10 +161,4 @@ 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/settings.cpp

@@ -91,7 +91,10 @@ void LogSettings() {
     LogSetting("Renderer_UseResolutionFactor", Settings::values.resolution_factor);
     LogSetting("Renderer_UseFrameLimit", Settings::values.use_frame_limit);
     LogSetting("Renderer_FrameLimit", Settings::values.frame_limit);
+    LogSetting("Renderer_UseDiskShaderCache", Settings::values.use_disk_shader_cache);
     LogSetting("Renderer_UseAccurateGpuEmulation", Settings::values.use_accurate_gpu_emulation);
+    LogSetting("Renderer_UseAsynchronousGpuEmulation",
+               Settings::values.use_asynchronous_gpu_emulation);
     LogSetting("Audio_OutputEngine", Settings::values.sink_id);
     LogSetting("Audio_EnableAudioStretching", Settings::values.enable_audio_stretching);
     LogSetting("Audio_OutputDevice", Settings::values.audio_device_id);

src/input_common/CMakeLists.txt

@@ -7,15 +7,18 @@ add_library(input_common STATIC
     main.h
     motion_emu.cpp
     motion_emu.h
-
-    $<$<BOOL:${SDL2_FOUND}>:sdl/sdl.cpp sdl/sdl.h>
+    sdl/sdl.cpp
+    sdl/sdl.h
 )
 
-create_target_directory_groups(input_common)
-
-target_link_libraries(input_common PUBLIC core PRIVATE common)
-
 if(SDL2_FOUND)
+    target_sources(input_common PRIVATE
+        sdl/sdl_impl.cpp
+        sdl/sdl_impl.h
+    )
     target_link_libraries(input_common PRIVATE SDL2)
     target_compile_definitions(input_common PRIVATE HAVE_SDL2)
 endif()
+
+create_target_directory_groups(input_common)
+target_link_libraries(input_common PUBLIC core PRIVATE common)

src/input_common/main.cpp

@@ -17,10 +17,7 @@ namespace InputCommon {
 
 static std::shared_ptr<Keyboard> keyboard;
 static std::shared_ptr<MotionEmu> motion_emu;
-
-#ifdef HAVE_SDL2
-static std::thread poll_thread;
-#endif
+static std::unique_ptr<SDL::State> sdl;
 
 void Init() {
     keyboard = std::make_shared<Keyboard>();
@@ -30,15 +27,7 @@ void Init() {
     motion_emu = std::make_shared<MotionEmu>();
     Input::RegisterFactory<Input::MotionDevice>("motion_emu", motion_emu);
 
-#ifdef HAVE_SDL2
-    SDL::Init();
-#endif
-}
-
-void StartJoystickEventHandler() {
-#ifdef HAVE_SDL2
-    poll_thread = std::thread(SDL::PollLoop);
-#endif
+    sdl = SDL::Init();
 }
 
 void Shutdown() {
@@ -47,11 +36,7 @@ void Shutdown() {
     Input::UnregisterFactory<Input::AnalogDevice>("analog_from_button");
     Input::UnregisterFactory<Input::MotionDevice>("motion_emu");
     motion_emu.reset();
-
-#ifdef HAVE_SDL2
-    SDL::Shutdown();
-    poll_thread.join();
-#endif
+    sdl.reset();
 }
 
 Keyboard* GetKeyboard() {
@@ -88,7 +73,7 @@ namespace Polling {
 
 std::vector<std::unique_ptr<DevicePoller>> GetPollers(DeviceType type) {
 #ifdef HAVE_SDL2
-    return SDL::Polling::GetPollers(type);
+    return sdl->GetPollers(type);
 #else
     return {};
 #endif

src/input_common/main.h

@@ -20,8 +20,6 @@ void Init();
 /// Deregisters all built-in input device factories and shuts them down.
 void Shutdown();
 
-void StartJoystickEventHandler();
-
 class Keyboard;
 
 /// Gets the keyboard button device factory.

src/input_common/sdl/sdl.cpp

@@ -1,631 +1,19 @@
-// Copyright 2017 Citra Emulator Project
+// Copyright 2018 Citra Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <algorithm>
-#include <atomic>
-#include <cmath>
-#include <functional>
-#include <iterator>
-#include <mutex>
-#include <string>
-#include <thread>
-#include <tuple>
-#include <unordered_map>
-#include <utility>
-#include <vector>
-#include <SDL.h>
-#include "common/assert.h"
-#include "common/logging/log.h"
-#include "common/math_util.h"
-#include "common/param_package.h"
-#include "common/threadsafe_queue.h"
-#include "input_common/main.h"
 #include "input_common/sdl/sdl.h"
+#ifdef HAVE_SDL2
+#include "input_common/sdl/sdl_impl.h"
+#endif
 
-namespace InputCommon {
+namespace InputCommon::SDL {
 
-namespace SDL {
-
-class SDLJoystick;
-class SDLButtonFactory;
-class SDLAnalogFactory;
-
-/// Map of GUID of a list of corresponding virtual Joysticks
-static std::unordered_map<std::string, std::vector<std::shared_ptr<SDLJoystick>>> joystick_map;
-static std::mutex joystick_map_mutex;
-
-static std::shared_ptr<SDLButtonFactory> button_factory;
-static std::shared_ptr<SDLAnalogFactory> analog_factory;
-
-/// Used by the Pollers during config
-static std::atomic<bool> polling;
-static Common::SPSCQueue<SDL_Event> event_queue;
-
-static std::atomic<bool> initialized = false;
-
-static std::string GetGUID(SDL_Joystick* joystick) {
-    SDL_JoystickGUID guid = SDL_JoystickGetGUID(joystick);
-    char guid_str[33];
-    SDL_JoystickGetGUIDString(guid, guid_str, sizeof(guid_str));
-    return guid_str;
+std::unique_ptr<State> Init() {
+#ifdef HAVE_SDL2
+    return std::make_unique<SDLState>();
+#else
+    return std::make_unique<NullState>();
+#endif
 }
-
-class SDLJoystick {
-public:
-    SDLJoystick(std::string guid_, int port_, SDL_Joystick* joystick,
-                decltype(&SDL_JoystickClose) deleter = &SDL_JoystickClose)
-        : guid{std::move(guid_)}, port{port_}, sdl_joystick{joystick, deleter} {}
-
-    void SetButton(int button, bool value) {
-        std::lock_guard<std::mutex> lock(mutex);
-        state.buttons[button] = value;
-    }
-
-    bool GetButton(int button) const {
-        std::lock_guard<std::mutex> lock(mutex);
-        return state.buttons.at(button);
-    }
-
-    void SetAxis(int axis, Sint16 value) {
-        std::lock_guard<std::mutex> lock(mutex);
-        state.axes[axis] = value;
-    }
-
-    float GetAxis(int axis) const {
-        std::lock_guard<std::mutex> lock(mutex);
-        return state.axes.at(axis) / 32767.0f;
-    }
-
-    std::tuple<float, float> GetAnalog(int axis_x, int axis_y) const {
-        float x = GetAxis(axis_x);
-        float y = GetAxis(axis_y);
-        y = -y; // 3DS uses an y-axis inverse from SDL
-
-        // Make sure the coordinates are in the unit circle,
-        // otherwise normalize it.
-        float r = x * x + y * y;
-        if (r > 1.0f) {
-            r = std::sqrt(r);
-            x /= r;
-            y /= r;
-        }
-
-        return std::make_tuple(x, y);
-    }
-
-    void SetHat(int hat, Uint8 direction) {
-        std::lock_guard<std::mutex> lock(mutex);
-        state.hats[hat] = direction;
-    }
-
-    bool GetHatDirection(int hat, Uint8 direction) const {
-        std::lock_guard<std::mutex> lock(mutex);
-        return (state.hats.at(hat) & direction) != 0;
-    }
-    /**
-     * The guid of the joystick
-     */
-    const std::string& GetGUID() const {
-        return guid;
-    }
-
-    /**
-     * The number of joystick from the same type that were connected before this joystick
-     */
-    int GetPort() const {
-        return port;
-    }
-
-    SDL_Joystick* GetSDLJoystick() const {
-        return sdl_joystick.get();
-    }
-
-    void SetSDLJoystick(SDL_Joystick* joystick,
-                        decltype(&SDL_JoystickClose) deleter = &SDL_JoystickClose) {
-        sdl_joystick =
-            std::unique_ptr<SDL_Joystick, decltype(&SDL_JoystickClose)>(joystick, deleter);
-    }
-
-private:
-    struct State {
-        std::unordered_map<int, bool> buttons;
-        std::unordered_map<int, Sint16> axes;
-        std::unordered_map<int, Uint8> hats;
-    } state;
-    std::string guid;
-    int port;
-    std::unique_ptr<SDL_Joystick, decltype(&SDL_JoystickClose)> sdl_joystick;
-    mutable std::mutex mutex;
-};
-
-/**
- * Get the nth joystick with the corresponding GUID
- */
-static std::shared_ptr<SDLJoystick> GetSDLJoystickByGUID(const std::string& guid, int port) {
-    std::lock_guard<std::mutex> lock(joystick_map_mutex);
-    const auto it = joystick_map.find(guid);
-    if (it != joystick_map.end()) {
-        while (it->second.size() <= port) {
-            auto joystick = std::make_shared<SDLJoystick>(guid, it->second.size(), nullptr,
-                                                          [](SDL_Joystick*) {});
-            it->second.emplace_back(std::move(joystick));
-        }
-        return it->second[port];
-    }
-    auto joystick = std::make_shared<SDLJoystick>(guid, 0, nullptr, [](SDL_Joystick*) {});
-    return joystick_map[guid].emplace_back(std::move(joystick));
-}
-
-/**
- * Check how many identical joysticks (by guid) were connected before the one with sdl_id and so tie
- * it to a SDLJoystick with the same guid and that port
- */
-static std::shared_ptr<SDLJoystick> GetSDLJoystickBySDLID(SDL_JoystickID sdl_id) {
-    std::lock_guard<std::mutex> lock(joystick_map_mutex);
-    auto sdl_joystick = SDL_JoystickFromInstanceID(sdl_id);
-    const std::string guid = GetGUID(sdl_joystick);
-    auto map_it = joystick_map.find(guid);
-    if (map_it != joystick_map.end()) {
-        auto vec_it = std::find_if(map_it->second.begin(), map_it->second.end(),
-                                   [&sdl_joystick](const std::shared_ptr<SDLJoystick>& joystick) {
-                                       return sdl_joystick == joystick->GetSDLJoystick();
-                                   });
-        if (vec_it != map_it->second.end()) {
-            // This is the common case: There is already an existing SDL_Joystick maped to a
-            // SDLJoystick. return the SDLJoystick
-            return *vec_it;
-        }
-        // Search for a SDLJoystick without a mapped SDL_Joystick...
-        auto nullptr_it = std::find_if(map_it->second.begin(), map_it->second.end(),
-                                       [](const std::shared_ptr<SDLJoystick>& joystick) {
-                                           return !joystick->GetSDLJoystick();
-                                       });
-        if (nullptr_it != map_it->second.end()) {
-            // ... and map it
-            (*nullptr_it)->SetSDLJoystick(sdl_joystick);
-            return *nullptr_it;
-        }
-        // There is no SDLJoystick without a mapped SDL_Joystick
-        // Create a new SDLJoystick
-        auto joystick = std::make_shared<SDLJoystick>(guid, map_it->second.size(), sdl_joystick);
-        return map_it->second.emplace_back(std::move(joystick));
-    }
-    auto joystick = std::make_shared<SDLJoystick>(guid, 0, sdl_joystick);
-    return joystick_map[guid].emplace_back(std::move(joystick));
-}
-
-void InitJoystick(int joystick_index) {
-    std::lock_guard<std::mutex> lock(joystick_map_mutex);
-    SDL_Joystick* sdl_joystick = SDL_JoystickOpen(joystick_index);
-    if (!sdl_joystick) {
-        LOG_ERROR(Input, "failed to open joystick {}", joystick_index);
-        return;
-    }
-    std::string guid = GetGUID(sdl_joystick);
-    if (joystick_map.find(guid) == joystick_map.end()) {
-        auto joystick = std::make_shared<SDLJoystick>(guid, 0, sdl_joystick);
-        joystick_map[guid].emplace_back(std::move(joystick));
-        return;
-    }
-    auto& joystick_guid_list = joystick_map[guid];
-    const auto it = std::find_if(
-        joystick_guid_list.begin(), joystick_guid_list.end(),
-        [](const std::shared_ptr<SDLJoystick>& joystick) { return !joystick->GetSDLJoystick(); });
-    if (it != joystick_guid_list.end()) {
-        (*it)->SetSDLJoystick(sdl_joystick);
-        return;
-    }
-    auto joystick = std::make_shared<SDLJoystick>(guid, joystick_guid_list.size(), sdl_joystick);
-    joystick_guid_list.emplace_back(std::move(joystick));
-}
-
-void CloseJoystick(SDL_Joystick* sdl_joystick) {
-    std::lock_guard<std::mutex> lock(joystick_map_mutex);
-    std::string guid = GetGUID(sdl_joystick);
-    // This call to guid is save since the joystick is guranteed to be in that map
-    auto& joystick_guid_list = joystick_map[guid];
-    const auto joystick_it =
-        std::find_if(joystick_guid_list.begin(), joystick_guid_list.end(),
-                     [&sdl_joystick](const std::shared_ptr<SDLJoystick>& joystick) {
-                         return joystick->GetSDLJoystick() == sdl_joystick;
-                     });
-    (*joystick_it)->SetSDLJoystick(nullptr, [](SDL_Joystick*) {});
-}
-
-void HandleGameControllerEvent(const SDL_Event& event) {
-    switch (event.type) {
-    case SDL_JOYBUTTONUP: {
-        auto joystick = GetSDLJoystickBySDLID(event.jbutton.which);
-        if (joystick) {
-            joystick->SetButton(event.jbutton.button, false);
-        }
-        break;
-    }
-    case SDL_JOYBUTTONDOWN: {
-        auto joystick = GetSDLJoystickBySDLID(event.jbutton.which);
-        if (joystick) {
-            joystick->SetButton(event.jbutton.button, true);
-        }
-        break;
-    }
-    case SDL_JOYHATMOTION: {
-        auto joystick = GetSDLJoystickBySDLID(event.jhat.which);
-        if (joystick) {
-            joystick->SetHat(event.jhat.hat, event.jhat.value);
-        }
-        break;
-    }
-    case SDL_JOYAXISMOTION: {
-        auto joystick = GetSDLJoystickBySDLID(event.jaxis.which);
-        if (joystick) {
-            joystick->SetAxis(event.jaxis.axis, event.jaxis.value);
-        }
-        break;
-    }
-    case SDL_JOYDEVICEREMOVED:
-        LOG_DEBUG(Input, "Controller removed with Instance_ID {}", event.jdevice.which);
-        CloseJoystick(SDL_JoystickFromInstanceID(event.jdevice.which));
-        break;
-    case SDL_JOYDEVICEADDED:
-        LOG_DEBUG(Input, "Controller connected with device index {}", event.jdevice.which);
-        InitJoystick(event.jdevice.which);
-        break;
-    }
-}
-
-void CloseSDLJoysticks() {
-    std::lock_guard<std::mutex> lock(joystick_map_mutex);
-    joystick_map.clear();
-}
-
-void PollLoop() {
-    if (SDL_Init(SDL_INIT_JOYSTICK) < 0) {
-        LOG_CRITICAL(Input, "SDL_Init(SDL_INIT_JOYSTICK) failed with: {}", SDL_GetError());
-        return;
-    }
-
-    SDL_Event event;
-    while (initialized) {
-        // Wait for 10 ms or until an event happens
-        if (SDL_WaitEventTimeout(&event, 10)) {
-            // Don't handle the event if we are configuring
-            if (polling) {
-                event_queue.Push(event);
-            } else {
-                HandleGameControllerEvent(event);
-            }
-        }
-    }
-    CloseSDLJoysticks();
-    SDL_QuitSubSystem(SDL_INIT_JOYSTICK);
-}
-
-class SDLButton final : public Input::ButtonDevice {
-public:
-    explicit SDLButton(std::shared_ptr<SDLJoystick> joystick_, int button_)
-        : joystick(std::move(joystick_)), button(button_) {}
-
-    bool GetStatus() const override {
-        return joystick->GetButton(button);
-    }
-
-private:
-    std::shared_ptr<SDLJoystick> joystick;
-    int button;
-};
-
-class SDLDirectionButton final : public Input::ButtonDevice {
-public:
-    explicit SDLDirectionButton(std::shared_ptr<SDLJoystick> joystick_, int hat_, Uint8 direction_)
-        : joystick(std::move(joystick_)), hat(hat_), direction(direction_) {}
-
-    bool GetStatus() const override {
-        return joystick->GetHatDirection(hat, direction);
-    }
-
-private:
-    std::shared_ptr<SDLJoystick> joystick;
-    int hat;
-    Uint8 direction;
-};
-
-class SDLAxisButton final : public Input::ButtonDevice {
-public:
-    explicit SDLAxisButton(std::shared_ptr<SDLJoystick> joystick_, int axis_, float threshold_,
-                           bool trigger_if_greater_)
-        : joystick(std::move(joystick_)), axis(axis_), threshold(threshold_),
-          trigger_if_greater(trigger_if_greater_) {}
-
-    bool GetStatus() const override {
-        float axis_value = joystick->GetAxis(axis);
-        if (trigger_if_greater)
-            return axis_value > threshold;
-        return axis_value < threshold;
-    }
-
-private:
-    std::shared_ptr<SDLJoystick> joystick;
-    int axis;
-    float threshold;
-    bool trigger_if_greater;
-};
-
-class SDLAnalog final : public Input::AnalogDevice {
-public:
-    SDLAnalog(std::shared_ptr<SDLJoystick> joystick_, int axis_x_, int axis_y_)
-        : joystick(std::move(joystick_)), axis_x(axis_x_), axis_y(axis_y_) {}
-
-    std::tuple<float, float> GetStatus() const override {
-        return joystick->GetAnalog(axis_x, axis_y);
-    }
-
-private:
-    std::shared_ptr<SDLJoystick> joystick;
-    int axis_x;
-    int axis_y;
-};
-
-/// A button device factory that creates button devices from SDL joystick
-class SDLButtonFactory final : public Input::Factory<Input::ButtonDevice> {
-public:
-    /**
-     * Creates a button device from a joystick button
-     * @param params contains parameters for creating the device:
-     *     - "guid": the guid of the joystick to bind
-     *     - "port": the nth joystick of the same type to bind
-     *     - "button"(optional): the index of the button to bind
-     *     - "hat"(optional): the index of the hat to bind as direction buttons
-     *     - "axis"(optional): the index of the axis to bind
-     *     - "direction"(only used for hat): the direction name of the hat to bind. Can be "up",
-     *         "down", "left" or "right"
-     *     - "threshold"(only used for axis): a float value in (-1.0, 1.0) which the button is
-     *         triggered if the axis value crosses
-     *     - "direction"(only used for axis): "+" means the button is triggered when the axis
-     * value is greater than the threshold; "-" means the button is triggered when the axis
-     * value is smaller than the threshold
-     */
-    std::unique_ptr<Input::ButtonDevice> Create(const Common::ParamPackage& params) override {
-        const std::string guid = params.Get("guid", "0");
-        const int port = params.Get("port", 0);
-
-        auto joystick = GetSDLJoystickByGUID(guid, port);
-
-        if (params.Has("hat")) {
-            const int hat = params.Get("hat", 0);
-            const std::string direction_name = params.Get("direction", "");
-            Uint8 direction;
-            if (direction_name == "up") {
-                direction = SDL_HAT_UP;
-            } else if (direction_name == "down") {
-                direction = SDL_HAT_DOWN;
-            } else if (direction_name == "left") {
-                direction = SDL_HAT_LEFT;
-            } else if (direction_name == "right") {
-                direction = SDL_HAT_RIGHT;
-            } else {
-                direction = 0;
-            }
-            // This is necessary so accessing GetHat with hat won't crash
-            joystick->SetHat(hat, SDL_HAT_CENTERED);
-            return std::make_unique<SDLDirectionButton>(joystick, hat, direction);
-        }
-
-        if (params.Has("axis")) {
-            const int axis = params.Get("axis", 0);
-            const float threshold = params.Get("threshold", 0.5f);
-            const std::string direction_name = params.Get("direction", "");
-            bool trigger_if_greater;
-            if (direction_name == "+") {
-                trigger_if_greater = true;
-            } else if (direction_name == "-") {
-                trigger_if_greater = false;
-            } else {
-                trigger_if_greater = true;
-                LOG_ERROR(Input, "Unknown direction '{}'", direction_name);
-            }
-            // This is necessary so accessing GetAxis with axis won't crash
-            joystick->SetAxis(axis, 0);
-            return std::make_unique<SDLAxisButton>(joystick, axis, threshold, trigger_if_greater);
-        }
-
-        const int button = params.Get("button", 0);
-        // This is necessary so accessing GetButton with button won't crash
-        joystick->SetButton(button, false);
-        return std::make_unique<SDLButton>(joystick, button);
-    }
-};
-
-/// An analog device factory that creates analog devices from SDL joystick
-class SDLAnalogFactory final : public Input::Factory<Input::AnalogDevice> {
-public:
-    /**
-     * Creates analog device from joystick axes
-     * @param params contains parameters for creating the device:
-     *     - "guid": the guid of the joystick to bind
-     *     - "port": the nth joystick of the same type
-     *     - "axis_x": the index of the axis to be bind as x-axis
-     *     - "axis_y": the index of the axis to be bind as y-axis
-     */
-    std::unique_ptr<Input::AnalogDevice> Create(const Common::ParamPackage& params) override {
-        const std::string guid = params.Get("guid", "0");
-        const int port = params.Get("port", 0);
-        const int axis_x = params.Get("axis_x", 0);
-        const int axis_y = params.Get("axis_y", 1);
-
-        auto joystick = GetSDLJoystickByGUID(guid, port);
-
-        // This is necessary so accessing GetAxis with axis_x and axis_y won't crash
-        joystick->SetAxis(axis_x, 0);
-        joystick->SetAxis(axis_y, 0);
-        return std::make_unique<SDLAnalog>(joystick, axis_x, axis_y);
-    }
-};
-
-void Init() {
-    using namespace Input;
-    RegisterFactory<ButtonDevice>("sdl", std::make_shared<SDLButtonFactory>());
-    RegisterFactory<AnalogDevice>("sdl", std::make_shared<SDLAnalogFactory>());
-    polling = false;
-    initialized = true;
-}
-
-void Shutdown() {
-    if (initialized) {
-        using namespace Input;
-        UnregisterFactory<ButtonDevice>("sdl");
-        UnregisterFactory<AnalogDevice>("sdl");
-        initialized = false;
-    }
-}
-
-Common::ParamPackage SDLEventToButtonParamPackage(const SDL_Event& event) {
-    Common::ParamPackage params({{"engine", "sdl"}});
-    switch (event.type) {
-    case SDL_JOYAXISMOTION: {
-        auto joystick = GetSDLJoystickBySDLID(event.jaxis.which);
-        params.Set("port", joystick->GetPort());
-        params.Set("guid", joystick->GetGUID());
-        params.Set("axis", event.jaxis.axis);
-        if (event.jaxis.value > 0) {
-            params.Set("direction", "+");
-            params.Set("threshold", "0.5");
-        } else {
-            params.Set("direction", "-");
-            params.Set("threshold", "-0.5");
-        }
-        break;
-    }
-    case SDL_JOYBUTTONUP: {
-        auto joystick = GetSDLJoystickBySDLID(event.jbutton.which);
-        params.Set("port", joystick->GetPort());
-        params.Set("guid", joystick->GetGUID());
-        params.Set("button", event.jbutton.button);
-        break;
-    }
-    case SDL_JOYHATMOTION: {
-        auto joystick = GetSDLJoystickBySDLID(event.jhat.which);
-        params.Set("port", joystick->GetPort());
-        params.Set("guid", joystick->GetGUID());
-        params.Set("hat", event.jhat.hat);
-        switch (event.jhat.value) {
-        case SDL_HAT_UP:
-            params.Set("direction", "up");
-            break;
-        case SDL_HAT_DOWN:
-            params.Set("direction", "down");
-            break;
-        case SDL_HAT_LEFT:
-            params.Set("direction", "left");
-            break;
-        case SDL_HAT_RIGHT:
-            params.Set("direction", "right");
-            break;
-        default:
-            return {};
-        }
-        break;
-    }
-    }
-    return params;
-}
-
-namespace Polling {
-
-class SDLPoller : public InputCommon::Polling::DevicePoller {
-public:
-    void Start() override {
-        event_queue.Clear();
-        polling = true;
-    }
-
-    void Stop() override {
-        polling = false;
-    }
-};
-
-class SDLButtonPoller final : public SDLPoller {
-public:
-    Common::ParamPackage GetNextInput() override {
-        SDL_Event event;
-        while (event_queue.Pop(event)) {
-            switch (event.type) {
-            case SDL_JOYAXISMOTION:
-                if (std::abs(event.jaxis.value / 32767.0) < 0.5) {
-                    break;
-                }
-            case SDL_JOYBUTTONUP:
-            case SDL_JOYHATMOTION:
-                return SDLEventToButtonParamPackage(event);
-            }
-        }
-        return {};
-    }
-};
-
-class SDLAnalogPoller final : public SDLPoller {
-public:
-    void Start() override {
-        SDLPoller::Start();
-
-        // Reset stored axes
-        analog_xaxis = -1;
-        analog_yaxis = -1;
-        analog_axes_joystick = -1;
-    }
-
-    Common::ParamPackage GetNextInput() override {
-        SDL_Event event;
-        while (event_queue.Pop(event)) {
-            if (event.type != SDL_JOYAXISMOTION || std::abs(event.jaxis.value / 32767.0) < 0.5) {
-                continue;
-            }
-            // An analog device needs two axes, so we need to store the axis for later and wait for
-            // a second SDL event. The axes also must be from the same joystick.
-            int axis = event.jaxis.axis;
-            if (analog_xaxis == -1) {
-                analog_xaxis = axis;
-                analog_axes_joystick = event.jaxis.which;
-            } else if (analog_yaxis == -1 && analog_xaxis != axis &&
-                       analog_axes_joystick == event.jaxis.which) {
-                analog_yaxis = axis;
-            }
-        }
-        Common::ParamPackage params;
-        if (analog_xaxis != -1 && analog_yaxis != -1) {
-            auto joystick = GetSDLJoystickBySDLID(event.jaxis.which);
-            params.Set("engine", "sdl");
-            params.Set("port", joystick->GetPort());
-            params.Set("guid", joystick->GetGUID());
-            params.Set("axis_x", analog_xaxis);
-            params.Set("axis_y", analog_yaxis);
-            analog_xaxis = -1;
-            analog_yaxis = -1;
-            analog_axes_joystick = -1;
-            return params;
-        }
-        return params;
-    }
-
-private:
-    int analog_xaxis = -1;
-    int analog_yaxis = -1;
-    SDL_JoystickID analog_axes_joystick = -1;
-};
-
-std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> GetPollers(
-    InputCommon::Polling::DeviceType type) {
-    std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> pollers;
-    switch (type) {
-    case InputCommon::Polling::DeviceType::Analog:
-        pollers.push_back(std::make_unique<SDLAnalogPoller>());
-        break;
-    case InputCommon::Polling::DeviceType::Button:
-        pollers.push_back(std::make_unique<SDLButtonPoller>());
-        break;
-    }
-    return pollers;
-}
-} // namespace Polling
-} // namespace SDL
-} // namespace InputCommon
+} // namespace InputCommon::SDL

src/input_common/sdl/sdl.h

@@ -1,4 +1,4 @@
-// Copyright 2017 Citra Emulator Project
+// Copyright 2018 Citra Emulator Project
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
@@ -7,45 +7,36 @@
 #include <memory>
 #include <vector>
 #include "core/frontend/input.h"
+#include "input_common/main.h"
 
 union SDL_Event;
+
 namespace Common {
 class ParamPackage;
-}
-namespace InputCommon {
-namespace Polling {
+} // namespace Common
+
+namespace InputCommon::Polling {
 class DevicePoller;
 enum class DeviceType;
-} // namespace Polling
-} // namespace InputCommon
+} // namespace InputCommon::Polling
 
-namespace InputCommon {
-namespace SDL {
+namespace InputCommon::SDL {
 
-/// Initializes and registers SDL device factories
-void Init();
+class State {
+public:
+    /// Unresisters SDL device factories and shut them down.
+    virtual ~State() = default;
 
-/// Unresisters SDL device factories and shut them down.
-void Shutdown();
+    virtual std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> GetPollers(
+        InputCommon::Polling::DeviceType type) = 0;
+};
 
-/// Needs to be called before SDL_QuitSubSystem.
-void CloseSDLJoysticks();
+class NullState : public State {
+public:
+    std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> GetPollers(
+        InputCommon::Polling::DeviceType type) override {}
+};
 
-/// Handle SDL_Events for joysticks from SDL_PollEvent
-void HandleGameControllerEvent(const SDL_Event& event);
+std::unique_ptr<State> Init();
 
-/// A Loop that calls HandleGameControllerEvent until Shutdown is called
-void PollLoop();
-
-/// Creates a ParamPackage from an SDL_Event that can directly be used to create a ButtonDevice
-Common::ParamPackage SDLEventToButtonParamPackage(const SDL_Event& event);
-
-namespace Polling {
-
-/// Get all DevicePoller that use the SDL backend for a specific device type
-std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> GetPollers(
-    InputCommon::Polling::DeviceType type);
-
-} // namespace Polling
-} // namespace SDL
-} // namespace InputCommon
+} // namespace InputCommon::SDL

src/input_common/sdl/sdl_impl.cpp

@@ -0,0 +1,669 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <atomic>
+#include <cmath>
+#include <functional>
+#include <iterator>
+#include <mutex>
+#include <string>
+#include <thread>
+#include <tuple>
+#include <unordered_map>
+#include <utility>
+#include <vector>
+#include <SDL.h>
+#include "common/assert.h"
+#include "common/logging/log.h"
+#include "common/math_util.h"
+#include "common/param_package.h"
+#include "common/threadsafe_queue.h"
+#include "core/frontend/input.h"
+#include "input_common/sdl/sdl_impl.h"
+
+namespace InputCommon {
+
+namespace SDL {
+
+static std::string GetGUID(SDL_Joystick* joystick) {
+    SDL_JoystickGUID guid = SDL_JoystickGetGUID(joystick);
+    char guid_str[33];
+    SDL_JoystickGetGUIDString(guid, guid_str, sizeof(guid_str));
+    return guid_str;
+}
+
+/// Creates a ParamPackage from an SDL_Event that can directly be used to create a ButtonDevice
+static Common::ParamPackage SDLEventToButtonParamPackage(SDLState& state, const SDL_Event& event);
+
+static int SDLEventWatcher(void* userdata, SDL_Event* event) {
+    SDLState* sdl_state = reinterpret_cast<SDLState*>(userdata);
+    // Don't handle the event if we are configuring
+    if (sdl_state->polling) {
+        sdl_state->event_queue.Push(*event);
+    } else {
+        sdl_state->HandleGameControllerEvent(*event);
+    }
+    return 0;
+}
+
+class SDLJoystick {
+public:
+    SDLJoystick(std::string guid_, int port_, SDL_Joystick* joystick,
+                decltype(&SDL_JoystickClose) deleter = &SDL_JoystickClose)
+        : guid{std::move(guid_)}, port{port_}, sdl_joystick{joystick, deleter} {}
+
+    void SetButton(int button, bool value) {
+        std::lock_guard<std::mutex> lock(mutex);
+        state.buttons[button] = value;
+    }
+
+    bool GetButton(int button) const {
+        std::lock_guard<std::mutex> lock(mutex);
+        return state.buttons.at(button);
+    }
+
+    void SetAxis(int axis, Sint16 value) {
+        std::lock_guard<std::mutex> lock(mutex);
+        state.axes[axis] = value;
+    }
+
+    float GetAxis(int axis) const {
+        std::lock_guard<std::mutex> lock(mutex);
+        return state.axes.at(axis) / 32767.0f;
+    }
+
+    std::tuple<float, float> GetAnalog(int axis_x, int axis_y) const {
+        float x = GetAxis(axis_x);
+        float y = GetAxis(axis_y);
+        y = -y; // 3DS uses an y-axis inverse from SDL
+
+        // Make sure the coordinates are in the unit circle,
+        // otherwise normalize it.
+        float r = x * x + y * y;
+        if (r > 1.0f) {
+            r = std::sqrt(r);
+            x /= r;
+            y /= r;
+        }
+
+        return std::make_tuple(x, y);
+    }
+
+    void SetHat(int hat, Uint8 direction) {
+        std::lock_guard<std::mutex> lock(mutex);
+        state.hats[hat] = direction;
+    }
+
+    bool GetHatDirection(int hat, Uint8 direction) const {
+        std::lock_guard<std::mutex> lock(mutex);
+        return (state.hats.at(hat) & direction) != 0;
+    }
+    /**
+     * The guid of the joystick
+     */
+    const std::string& GetGUID() const {
+        return guid;
+    }
+
+    /**
+     * The number of joystick from the same type that were connected before this joystick
+     */
+    int GetPort() const {
+        return port;
+    }
+
+    SDL_Joystick* GetSDLJoystick() const {
+        return sdl_joystick.get();
+    }
+
+    void SetSDLJoystick(SDL_Joystick* joystick,
+                        decltype(&SDL_JoystickClose) deleter = &SDL_JoystickClose) {
+        sdl_joystick =
+            std::unique_ptr<SDL_Joystick, decltype(&SDL_JoystickClose)>(joystick, deleter);
+    }
+
+private:
+    struct State {
+        std::unordered_map<int, bool> buttons;
+        std::unordered_map<int, Sint16> axes;
+        std::unordered_map<int, Uint8> hats;
+    } state;
+    std::string guid;
+    int port;
+    std::unique_ptr<SDL_Joystick, decltype(&SDL_JoystickClose)> sdl_joystick;
+    mutable std::mutex mutex;
+};
+
+/**
+ * Get the nth joystick with the corresponding GUID
+ */
+std::shared_ptr<SDLJoystick> SDLState::GetSDLJoystickByGUID(const std::string& guid, int port) {
+    std::lock_guard<std::mutex> lock(joystick_map_mutex);
+    const auto it = joystick_map.find(guid);
+    if (it != joystick_map.end()) {
+        while (it->second.size() <= port) {
+            auto joystick = std::make_shared<SDLJoystick>(guid, it->second.size(), nullptr,
+                                                          [](SDL_Joystick*) {});
+            it->second.emplace_back(std::move(joystick));
+        }
+        return it->second[port];
+    }
+    auto joystick = std::make_shared<SDLJoystick>(guid, 0, nullptr, [](SDL_Joystick*) {});
+    return joystick_map[guid].emplace_back(std::move(joystick));
+}
+
+/**
+ * Check how many identical joysticks (by guid) were connected before the one with sdl_id and so tie
+ * it to a SDLJoystick with the same guid and that port
+ */
+std::shared_ptr<SDLJoystick> SDLState::GetSDLJoystickBySDLID(SDL_JoystickID sdl_id) {
+    auto sdl_joystick = SDL_JoystickFromInstanceID(sdl_id);
+    const std::string guid = GetGUID(sdl_joystick);
+    std::lock_guard<std::mutex> lock(joystick_map_mutex);
+    auto map_it = joystick_map.find(guid);
+    if (map_it != joystick_map.end()) {
+        auto vec_it = std::find_if(map_it->second.begin(), map_it->second.end(),
+                                   [&sdl_joystick](const std::shared_ptr<SDLJoystick>& joystick) {
+                                       return sdl_joystick == joystick->GetSDLJoystick();
+                                   });
+        if (vec_it != map_it->second.end()) {
+            // This is the common case: There is already an existing SDL_Joystick maped to a
+            // SDLJoystick. return the SDLJoystick
+            return *vec_it;
+        }
+        // Search for a SDLJoystick without a mapped SDL_Joystick...
+        auto nullptr_it = std::find_if(map_it->second.begin(), map_it->second.end(),
+                                       [](const std::shared_ptr<SDLJoystick>& joystick) {
+                                           return !joystick->GetSDLJoystick();
+                                       });
+        if (nullptr_it != map_it->second.end()) {
+            // ... and map it
+            (*nullptr_it)->SetSDLJoystick(sdl_joystick);
+            return *nullptr_it;
+        }
+        // There is no SDLJoystick without a mapped SDL_Joystick
+        // Create a new SDLJoystick
+        auto joystick = std::make_shared<SDLJoystick>(guid, map_it->second.size(), sdl_joystick);
+        return map_it->second.emplace_back(std::move(joystick));
+    }
+    auto joystick = std::make_shared<SDLJoystick>(guid, 0, sdl_joystick);
+    return joystick_map[guid].emplace_back(std::move(joystick));
+}
+
+void SDLState::InitJoystick(int joystick_index) {
+    SDL_Joystick* sdl_joystick = SDL_JoystickOpen(joystick_index);
+    if (!sdl_joystick) {
+        LOG_ERROR(Input, "failed to open joystick {}", joystick_index);
+        return;
+    }
+    std::string guid = GetGUID(sdl_joystick);
+    std::lock_guard<std::mutex> lock(joystick_map_mutex);
+    if (joystick_map.find(guid) == joystick_map.end()) {
+        auto joystick = std::make_shared<SDLJoystick>(guid, 0, sdl_joystick);
+        joystick_map[guid].emplace_back(std::move(joystick));
+        return;
+    }
+    auto& joystick_guid_list = joystick_map[guid];
+    const auto it = std::find_if(
+        joystick_guid_list.begin(), joystick_guid_list.end(),
+        [](const std::shared_ptr<SDLJoystick>& joystick) { return !joystick->GetSDLJoystick(); });
+    if (it != joystick_guid_list.end()) {
+        (*it)->SetSDLJoystick(sdl_joystick);
+        return;
+    }
+    auto joystick = std::make_shared<SDLJoystick>(guid, joystick_guid_list.size(), sdl_joystick);
+    joystick_guid_list.emplace_back(std::move(joystick));
+}
+
+void SDLState::CloseJoystick(SDL_Joystick* sdl_joystick) {
+    std::string guid = GetGUID(sdl_joystick);
+    std::shared_ptr<SDLJoystick> joystick;
+    {
+        std::lock_guard<std::mutex> lock(joystick_map_mutex);
+        // This call to guid is safe since the joystick is guaranteed to be in the map
+        auto& joystick_guid_list = joystick_map[guid];
+        const auto joystick_it =
+            std::find_if(joystick_guid_list.begin(), joystick_guid_list.end(),
+                         [&sdl_joystick](const std::shared_ptr<SDLJoystick>& joystick) {
+                             return joystick->GetSDLJoystick() == sdl_joystick;
+                         });
+        joystick = *joystick_it;
+    }
+    // Destruct SDL_Joystick outside the lock guard because SDL can internally call event calback
+    // which locks the mutex again
+    joystick->SetSDLJoystick(nullptr, [](SDL_Joystick*) {});
+}
+
+void SDLState::HandleGameControllerEvent(const SDL_Event& event) {
+    switch (event.type) {
+    case SDL_JOYBUTTONUP: {
+        if (auto joystick = GetSDLJoystickBySDLID(event.jbutton.which)) {
+            joystick->SetButton(event.jbutton.button, false);
+        }
+        break;
+    }
+    case SDL_JOYBUTTONDOWN: {
+        if (auto joystick = GetSDLJoystickBySDLID(event.jbutton.which)) {
+            joystick->SetButton(event.jbutton.button, true);
+        }
+        break;
+    }
+    case SDL_JOYHATMOTION: {
+        if (auto joystick = GetSDLJoystickBySDLID(event.jhat.which)) {
+            joystick->SetHat(event.jhat.hat, event.jhat.value);
+        }
+        break;
+    }
+    case SDL_JOYAXISMOTION: {
+        if (auto joystick = GetSDLJoystickBySDLID(event.jaxis.which)) {
+            joystick->SetAxis(event.jaxis.axis, event.jaxis.value);
+        }
+        break;
+    }
+    case SDL_JOYDEVICEREMOVED:
+        LOG_DEBUG(Input, "Controller removed with Instance_ID {}", event.jdevice.which);
+        CloseJoystick(SDL_JoystickFromInstanceID(event.jdevice.which));
+        break;
+    case SDL_JOYDEVICEADDED:
+        LOG_DEBUG(Input, "Controller connected with device index {}", event.jdevice.which);
+        InitJoystick(event.jdevice.which);
+        break;
+    }
+}
+
+void SDLState::CloseJoysticks() {
+    std::lock_guard<std::mutex> lock(joystick_map_mutex);
+    joystick_map.clear();
+}
+
+class SDLButton final : public Input::ButtonDevice {
+public:
+    explicit SDLButton(std::shared_ptr<SDLJoystick> joystick_, int button_)
+        : joystick(std::move(joystick_)), button(button_) {}
+
+    bool GetStatus() const override {
+        return joystick->GetButton(button);
+    }
+
+private:
+    std::shared_ptr<SDLJoystick> joystick;
+    int button;
+};
+
+class SDLDirectionButton final : public Input::ButtonDevice {
+public:
+    explicit SDLDirectionButton(std::shared_ptr<SDLJoystick> joystick_, int hat_, Uint8 direction_)
+        : joystick(std::move(joystick_)), hat(hat_), direction(direction_) {}
+
+    bool GetStatus() const override {
+        return joystick->GetHatDirection(hat, direction);
+    }
+
+private:
+    std::shared_ptr<SDLJoystick> joystick;
+    int hat;
+    Uint8 direction;
+};
+
+class SDLAxisButton final : public Input::ButtonDevice {
+public:
+    explicit SDLAxisButton(std::shared_ptr<SDLJoystick> joystick_, int axis_, float threshold_,
+                           bool trigger_if_greater_)
+        : joystick(std::move(joystick_)), axis(axis_), threshold(threshold_),
+          trigger_if_greater(trigger_if_greater_) {}
+
+    bool GetStatus() const override {
+        float axis_value = joystick->GetAxis(axis);
+        if (trigger_if_greater)
+            return axis_value > threshold;
+        return axis_value < threshold;
+    }
+
+private:
+    std::shared_ptr<SDLJoystick> joystick;
+    int axis;
+    float threshold;
+    bool trigger_if_greater;
+};
+
+class SDLAnalog final : public Input::AnalogDevice {
+public:
+    SDLAnalog(std::shared_ptr<SDLJoystick> joystick_, int axis_x_, int axis_y_, float deadzone_)
+        : joystick(std::move(joystick_)), axis_x(axis_x_), axis_y(axis_y_), deadzone(deadzone_) {}
+
+    std::tuple<float, float> GetStatus() const override {
+        const auto [x, y] = joystick->GetAnalog(axis_x, axis_y);
+        const float r = std::sqrt((x * x) + (y * y));
+        if (r > deadzone) {
+            return std::make_tuple(x / r * (r - deadzone) / (1 - deadzone),
+                                   y / r * (r - deadzone) / (1 - deadzone));
+        }
+        return std::make_tuple<float, float>(0.0f, 0.0f);
+    }
+
+private:
+    std::shared_ptr<SDLJoystick> joystick;
+    const int axis_x;
+    const int axis_y;
+    const float deadzone;
+};
+
+/// A button device factory that creates button devices from SDL joystick
+class SDLButtonFactory final : public Input::Factory<Input::ButtonDevice> {
+public:
+    explicit SDLButtonFactory(SDLState& state_) : state(state_) {}
+
+    /**
+     * Creates a button device from a joystick button
+     * @param params contains parameters for creating the device:
+     *     - "guid": the guid of the joystick to bind
+     *     - "port": the nth joystick of the same type to bind
+     *     - "button"(optional): the index of the button to bind
+     *     - "hat"(optional): the index of the hat to bind as direction buttons
+     *     - "axis"(optional): the index of the axis to bind
+     *     - "direction"(only used for hat): the direction name of the hat to bind. Can be "up",
+     *         "down", "left" or "right"
+     *     - "threshold"(only used for axis): a float value in (-1.0, 1.0) which the button is
+     *         triggered if the axis value crosses
+     *     - "direction"(only used for axis): "+" means the button is triggered when the axis
+     * value is greater than the threshold; "-" means the button is triggered when the axis
+     * value is smaller than the threshold
+     */
+    std::unique_ptr<Input::ButtonDevice> Create(const Common::ParamPackage& params) override {
+        const std::string guid = params.Get("guid", "0");
+        const int port = params.Get("port", 0);
+
+        auto joystick = state.GetSDLJoystickByGUID(guid, port);
+
+        if (params.Has("hat")) {
+            const int hat = params.Get("hat", 0);
+            const std::string direction_name = params.Get("direction", "");
+            Uint8 direction;
+            if (direction_name == "up") {
+                direction = SDL_HAT_UP;
+            } else if (direction_name == "down") {
+                direction = SDL_HAT_DOWN;
+            } else if (direction_name == "left") {
+                direction = SDL_HAT_LEFT;
+            } else if (direction_name == "right") {
+                direction = SDL_HAT_RIGHT;
+            } else {
+                direction = 0;
+            }
+            // This is necessary so accessing GetHat with hat won't crash
+            joystick->SetHat(hat, SDL_HAT_CENTERED);
+            return std::make_unique<SDLDirectionButton>(joystick, hat, direction);
+        }
+
+        if (params.Has("axis")) {
+            const int axis = params.Get("axis", 0);
+            const float threshold = params.Get("threshold", 0.5f);
+            const std::string direction_name = params.Get("direction", "");
+            bool trigger_if_greater;
+            if (direction_name == "+") {
+                trigger_if_greater = true;
+            } else if (direction_name == "-") {
+                trigger_if_greater = false;
+            } else {
+                trigger_if_greater = true;
+                LOG_ERROR(Input, "Unknown direction {}", direction_name);
+            }
+            // This is necessary so accessing GetAxis with axis won't crash
+            joystick->SetAxis(axis, 0);
+            return std::make_unique<SDLAxisButton>(joystick, axis, threshold, trigger_if_greater);
+        }
+
+        const int button = params.Get("button", 0);
+        // This is necessary so accessing GetButton with button won't crash
+        joystick->SetButton(button, false);
+        return std::make_unique<SDLButton>(joystick, button);
+    }
+
+private:
+    SDLState& state;
+};
+
+/// An analog device factory that creates analog devices from SDL joystick
+class SDLAnalogFactory final : public Input::Factory<Input::AnalogDevice> {
+public:
+    explicit SDLAnalogFactory(SDLState& state_) : state(state_) {}
+    /**
+     * Creates analog device from joystick axes
+     * @param params contains parameters for creating the device:
+     *     - "guid": the guid of the joystick to bind
+     *     - "port": the nth joystick of the same type
+     *     - "axis_x": the index of the axis to be bind as x-axis
+     *     - "axis_y": the index of the axis to be bind as y-axis
+     */
+    std::unique_ptr<Input::AnalogDevice> Create(const Common::ParamPackage& params) override {
+        const std::string guid = params.Get("guid", "0");
+        const int port = params.Get("port", 0);
+        const int axis_x = params.Get("axis_x", 0);
+        const int axis_y = params.Get("axis_y", 1);
+        float deadzone = std::clamp(params.Get("deadzone", 0.0f), 0.0f, .99f);
+
+        auto joystick = state.GetSDLJoystickByGUID(guid, port);
+
+        // This is necessary so accessing GetAxis with axis_x and axis_y won't crash
+        joystick->SetAxis(axis_x, 0);
+        joystick->SetAxis(axis_y, 0);
+        return std::make_unique<SDLAnalog>(joystick, axis_x, axis_y, deadzone);
+    }
+
+private:
+    SDLState& state;
+};
+
+SDLState::SDLState() {
+    using namespace Input;
+    RegisterFactory<ButtonDevice>("sdl", std::make_shared<SDLButtonFactory>(*this));
+    RegisterFactory<AnalogDevice>("sdl", std::make_shared<SDLAnalogFactory>(*this));
+
+    // If the frontend is going to manage the event loop, then we dont start one here
+    start_thread = !SDL_WasInit(SDL_INIT_JOYSTICK);
+    if (start_thread && SDL_Init(SDL_INIT_JOYSTICK) < 0) {
+        LOG_CRITICAL(Input, "SDL_Init(SDL_INIT_JOYSTICK) failed with: {}", SDL_GetError());
+        return;
+    }
+    if (SDL_SetHint(SDL_HINT_JOYSTICK_ALLOW_BACKGROUND_EVENTS, "1") == SDL_FALSE) {
+        LOG_ERROR(Input, "Failed to set Hint for background events", SDL_GetError());
+    }
+
+    SDL_AddEventWatch(&SDLEventWatcher, this);
+
+    initialized = true;
+    if (start_thread) {
+        poll_thread = std::thread([&] {
+            using namespace std::chrono_literals;
+            SDL_Event event;
+            while (initialized) {
+                SDL_PumpEvents();
+                std::this_thread::sleep_for(std::chrono::duration(10ms));
+            }
+        });
+    }
+    // Because the events for joystick connection happens before we have our event watcher added, we
+    // can just open all the joysticks right here
+    for (int i = 0; i < SDL_NumJoysticks(); ++i) {
+        InitJoystick(i);
+    }
+}
+
+SDLState::~SDLState() {
+    using namespace Input;
+    UnregisterFactory<ButtonDevice>("sdl");
+    UnregisterFactory<AnalogDevice>("sdl");
+
+    CloseJoysticks();
+    SDL_DelEventWatch(&SDLEventWatcher, this);
+
+    initialized = false;
+    if (start_thread) {
+        poll_thread.join();
+        SDL_QuitSubSystem(SDL_INIT_JOYSTICK);
+    }
+}
+
+Common::ParamPackage SDLEventToButtonParamPackage(SDLState& state, const SDL_Event& event) {
+    Common::ParamPackage params({{"engine", "sdl"}});
+
+    switch (event.type) {
+    case SDL_JOYAXISMOTION: {
+        auto joystick = state.GetSDLJoystickBySDLID(event.jaxis.which);
+        params.Set("port", joystick->GetPort());
+        params.Set("guid", joystick->GetGUID());
+        params.Set("axis", event.jaxis.axis);
+        if (event.jaxis.value > 0) {
+            params.Set("direction", "+");
+            params.Set("threshold", "0.5");
+        } else {
+            params.Set("direction", "-");
+            params.Set("threshold", "-0.5");
+        }
+        break;
+    }
+    case SDL_JOYBUTTONUP: {
+        auto joystick = state.GetSDLJoystickBySDLID(event.jbutton.which);
+        params.Set("port", joystick->GetPort());
+        params.Set("guid", joystick->GetGUID());
+        params.Set("button", event.jbutton.button);
+        break;
+    }
+    case SDL_JOYHATMOTION: {
+        auto joystick = state.GetSDLJoystickBySDLID(event.jhat.which);
+        params.Set("port", joystick->GetPort());
+        params.Set("guid", joystick->GetGUID());
+        params.Set("hat", event.jhat.hat);
+        switch (event.jhat.value) {
+        case SDL_HAT_UP:
+            params.Set("direction", "up");
+            break;
+        case SDL_HAT_DOWN:
+            params.Set("direction", "down");
+            break;
+        case SDL_HAT_LEFT:
+            params.Set("direction", "left");
+            break;
+        case SDL_HAT_RIGHT:
+            params.Set("direction", "right");
+            break;
+        default:
+            return {};
+        }
+        break;
+    }
+    }
+    return params;
+}
+
+namespace Polling {
+
+class SDLPoller : public InputCommon::Polling::DevicePoller {
+public:
+    explicit SDLPoller(SDLState& state_) : state(state_) {}
+
+    void Start() override {
+        state.event_queue.Clear();
+        state.polling = true;
+    }
+
+    void Stop() override {
+        state.polling = false;
+    }
+
+protected:
+    SDLState& state;
+};
+
+class SDLButtonPoller final : public SDLPoller {
+public:
+    explicit SDLButtonPoller(SDLState& state_) : SDLPoller(state_) {}
+
+    Common::ParamPackage GetNextInput() override {
+        SDL_Event event;
+        while (state.event_queue.Pop(event)) {
+            switch (event.type) {
+            case SDL_JOYAXISMOTION:
+                if (std::abs(event.jaxis.value / 32767.0) < 0.5) {
+                    break;
+                }
+            case SDL_JOYBUTTONUP:
+            case SDL_JOYHATMOTION:
+                return SDLEventToButtonParamPackage(state, event);
+            }
+        }
+        return {};
+    }
+};
+
+class SDLAnalogPoller final : public SDLPoller {
+public:
+    explicit SDLAnalogPoller(SDLState& state_) : SDLPoller(state_) {}
+
+    void Start() override {
+        SDLPoller::Start();
+
+        // Reset stored axes
+        analog_xaxis = -1;
+        analog_yaxis = -1;
+        analog_axes_joystick = -1;
+    }
+
+    Common::ParamPackage GetNextInput() override {
+        SDL_Event event;
+        while (state.event_queue.Pop(event)) {
+            if (event.type != SDL_JOYAXISMOTION || std::abs(event.jaxis.value / 32767.0) < 0.5) {
+                continue;
+            }
+            // An analog device needs two axes, so we need to store the axis for later and wait for
+            // a second SDL event. The axes also must be from the same joystick.
+            int axis = event.jaxis.axis;
+            if (analog_xaxis == -1) {
+                analog_xaxis = axis;
+                analog_axes_joystick = event.jaxis.which;
+            } else if (analog_yaxis == -1 && analog_xaxis != axis &&
+                       analog_axes_joystick == event.jaxis.which) {
+                analog_yaxis = axis;
+            }
+        }
+        Common::ParamPackage params;
+        if (analog_xaxis != -1 && analog_yaxis != -1) {
+            auto joystick = state.GetSDLJoystickBySDLID(event.jaxis.which);
+            params.Set("engine", "sdl");
+            params.Set("port", joystick->GetPort());
+            params.Set("guid", joystick->GetGUID());
+            params.Set("axis_x", analog_xaxis);
+            params.Set("axis_y", analog_yaxis);
+            analog_xaxis = -1;
+            analog_yaxis = -1;
+            analog_axes_joystick = -1;
+            return params;
+        }
+        return params;
+    }
+
+private:
+    int analog_xaxis = -1;
+    int analog_yaxis = -1;
+    SDL_JoystickID analog_axes_joystick = -1;
+};
+} // namespace Polling
+
+std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> SDLState::GetPollers(
+    InputCommon::Polling::DeviceType type) {
+    std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> pollers;
+    switch (type) {
+    case InputCommon::Polling::DeviceType::Analog:
+        pollers.emplace_back(std::make_unique<Polling::SDLAnalogPoller>(*this));
+        break;
+    case InputCommon::Polling::DeviceType::Button:
+        pollers.emplace_back(std::make_unique<Polling::SDLButtonPoller>(*this));
+        break;
+        return pollers;
+    }
+}
+
+} // namespace SDL
+} // namespace InputCommon

src/input_common/sdl/sdl_impl.h

@@ -0,0 +1,64 @@
+// Copyright 2018 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <atomic>
+#include <memory>
+#include <thread>
+#include "common/threadsafe_queue.h"
+#include "input_common/sdl/sdl.h"
+
+union SDL_Event;
+using SDL_Joystick = struct _SDL_Joystick;
+using SDL_JoystickID = s32;
+
+namespace InputCommon::SDL {
+
+class SDLJoystick;
+class SDLButtonFactory;
+class SDLAnalogFactory;
+
+class SDLState : public State {
+public:
+    /// Initializes and registers SDL device factories
+    SDLState();
+
+    /// Unresisters SDL device factories and shut them down.
+    ~SDLState() override;
+
+    /// Handle SDL_Events for joysticks from SDL_PollEvent
+    void HandleGameControllerEvent(const SDL_Event& event);
+
+    std::shared_ptr<SDLJoystick> GetSDLJoystickBySDLID(SDL_JoystickID sdl_id);
+    std::shared_ptr<SDLJoystick> GetSDLJoystickByGUID(const std::string& guid, int port);
+
+    /// Get all DevicePoller that use the SDL backend for a specific device type
+    std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> GetPollers(
+        InputCommon::Polling::DeviceType type) override;
+
+    /// Used by the Pollers during config
+    std::atomic<bool> polling = false;
+    Common::SPSCQueue<SDL_Event> event_queue;
+
+private:
+    void InitJoystick(int joystick_index);
+    void CloseJoystick(SDL_Joystick* sdl_joystick);
+
+    /// Needs to be called before SDL_QuitSubSystem.
+    void CloseJoysticks();
+
+    /// Map of GUID of a list of corresponding virtual Joysticks
+    std::unordered_map<std::string, std::vector<std::shared_ptr<SDLJoystick>>> joystick_map;
+    std::mutex joystick_map_mutex;
+
+    std::shared_ptr<SDLButtonFactory> button_factory;
+    std::shared_ptr<SDLAnalogFactory> analog_factory;
+
+    bool start_thread = false;
+    std::atomic<bool> initialized = false;
+
+    std::thread poll_thread;
+};
+} // namespace InputCommon::SDL

src/tests/core/arm/arm_test_common.cpp

@@ -15,7 +15,7 @@ namespace ArmTests {
 TestEnvironment::TestEnvironment(bool mutable_memory_)
     : mutable_memory(mutable_memory_),
       test_memory(std::make_shared<TestMemory>(this)), kernel{Core::System::GetInstance()} {
-    auto process = Kernel::Process::Create(kernel, "");
+    auto process = Kernel::Process::Create(Core::System::GetInstance(), "");
     kernel.MakeCurrentProcess(process.get());
     page_table = &process->VMManager().page_table;
 

src/video_core/CMakeLists.txt

@@ -80,6 +80,7 @@ add_library(video_core STATIC
     shader/decode/hfma2.cpp
     shader/decode/conversion.cpp
     shader/decode/memory.cpp
+    shader/decode/texture.cpp
     shader/decode/float_set_predicate.cpp
     shader/decode/integer_set_predicate.cpp
     shader/decode/half_set_predicate.cpp
@@ -100,6 +101,8 @@ add_library(video_core STATIC
     surface.h
     textures/astc.cpp
     textures/astc.h
+    textures/convert.cpp
+    textures/convert.h
     textures/decoders.cpp
     textures/decoders.h
     textures/texture.h
@@ -110,6 +113,8 @@ add_library(video_core STATIC
 if (ENABLE_VULKAN)
     target_sources(video_core PRIVATE
         renderer_vulkan/declarations.h
+        renderer_vulkan/maxwell_to_vk.cpp
+        renderer_vulkan/maxwell_to_vk.h
         renderer_vulkan/vk_buffer_cache.cpp
         renderer_vulkan/vk_buffer_cache.h
         renderer_vulkan/vk_device.cpp
@@ -118,6 +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

@@ -39,7 +39,7 @@ bool DmaPusher::Step() {
     }
 
     const CommandList& command_list{dma_pushbuffer.front()};
-    const CommandListHeader& command_list_header{command_list[dma_pushbuffer_subindex++]};
+    const CommandListHeader command_list_header{command_list[dma_pushbuffer_subindex++]};
     GPUVAddr dma_get = command_list_header.addr;
     GPUVAddr dma_put = dma_get + command_list_header.size * sizeof(u32);
     bool non_main = command_list_header.is_non_main;

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 {
 
@@ -48,7 +49,8 @@ void KeplerMemory::ProcessData(u32 data) {
     // 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));
+    system.Renderer().Rasterizer().InvalidateRegion(ToCacheAddr(Memory::GetPointer(*dest_address)),
+                                                    sizeof(u32));
 
     Memory::Write32(*dest_address, data);
     system.GPU().Maxwell3D().dirty_flags.OnMemoryWrite();

src/video_core/engines/maxwell_3d.cpp

@@ -396,7 +396,10 @@ void Maxwell3D::ProcessCBData(u32 value) {
     const auto address = memory_manager.GpuToCpuAddress(buffer_address + regs.const_buffer.cb_pos);
     ASSERT_MSG(address, "Invalid GPU address");
 
-    Memory::Write32(*address, value);
+    u8* ptr{Memory::GetPointer(*address)};
+    rasterizer.InvalidateRegion(ToCacheAddr(ptr), sizeof(u32));
+    std::memcpy(ptr, &value, sizeof(u32));
+
     dirty_flags.OnMemoryWrite();
 
     // Increment the current buffer position.

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 {
@@ -92,12 +93,14 @@ void MaxwellDMA::HandleCopy() {
     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);
+        Core::System::GetInstance().Renderer().Rasterizer().FlushRegion(
+            ToCacheAddr(Memory::GetPointer(*source_cpu)), 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);
+        Core::System::GetInstance().Renderer().Rasterizer().InvalidateRegion(
+            ToCacheAddr(Memory::GetPointer(*dest_cpu)), dst_size);
     };
 
     if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {

src/video_core/engines/shader_bytecode.h

@@ -324,11 +324,11 @@ enum class TextureQueryType : u64 {
 
 enum class TextureProcessMode : u64 {
     None = 0,
-    LZ = 1,  // Unknown, appears to be the same as none.
+    LZ = 1,  // Load LOD of zero.
     LB = 2,  // Load Bias.
-    LL = 3,  // Load LOD (LevelOfDetail)
-    LBA = 6, // Load Bias. The A is unknown, does not appear to differ with LB
-    LLA = 7  // Load LOD. The A is unknown, does not appear to differ with LL
+    LL = 3,  // Load LOD.
+    LBA = 6, // Load Bias. The A is unknown, does not appear to differ with LB.
+    LLA = 7  // Load LOD. The A is unknown, does not appear to differ with LL.
 };
 
 enum class TextureMiscMode : u64 {
@@ -1445,6 +1445,7 @@ public:
         Flow,
         Synch,
         Memory,
+        Texture,
         FloatSet,
         FloatSetPredicate,
         IntegerSet,
@@ -1575,14 +1576,14 @@ private:
             INST("1110111101010---", Id::ST_L, Type::Memory, "ST_L"),
             INST("1110111011010---", Id::LDG, Type::Memory, "LDG"),
             INST("1110111011011---", Id::STG, Type::Memory, "STG"),
-            INST("110000----111---", Id::TEX, Type::Memory, "TEX"),
-            INST("1101111101001---", Id::TXQ, Type::Memory, "TXQ"),
-            INST("1101-00---------", Id::TEXS, Type::Memory, "TEXS"),
-            INST("1101101---------", Id::TLDS, Type::Memory, "TLDS"),
-            INST("110010----111---", Id::TLD4, Type::Memory, "TLD4"),
-            INST("1101111100------", Id::TLD4S, Type::Memory, "TLD4S"),
-            INST("110111110110----", Id::TMML_B, Type::Memory, "TMML_B"),
-            INST("1101111101011---", Id::TMML, Type::Memory, "TMML"),
+            INST("110000----111---", Id::TEX, Type::Texture, "TEX"),
+            INST("1101111101001---", Id::TXQ, Type::Texture, "TXQ"),
+            INST("1101-00---------", Id::TEXS, Type::Texture, "TEXS"),
+            INST("1101101---------", Id::TLDS, Type::Texture, "TLDS"),
+            INST("110010----111---", Id::TLD4, Type::Texture, "TLD4"),
+            INST("1101111100------", Id::TLD4S, Type::Texture, "TLD4S"),
+            INST("110111110110----", Id::TMML_B, Type::Texture, "TMML_B"),
+            INST("1101111101011---", Id::TMML, Type::Texture, "TMML"),
             INST("111000110000----", Id::EXIT, Type::Trivial, "EXIT"),
             INST("11100000--------", Id::IPA, Type::Trivial, "IPA"),
             INST("1111101111100---", Id::OUT_R, Type::Trivial, "OUT_R"),

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;
 }
@@ -123,7 +128,7 @@ class GPU {
 public:
     explicit GPU(Core::System& system, VideoCore::RendererBase& renderer);
 
-    ~GPU();
+    virtual ~GPU();
 
     struct MethodCall {
         u32 method{};
@@ -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

@@ -21,14 +21,14 @@ class ThreadManager;
 class GPUAsynch : public Tegra::GPU {
 public:
     explicit GPUAsynch(Core::System& system, VideoCore::RendererBase& renderer);
-    ~GPUAsynch();
+    ~GPUAsynch() override;
 
     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

@@ -16,14 +16,14 @@ namespace VideoCommon {
 class GPUSynch : public Tegra::GPU {
 public:
     explicit GPUSynch(Core::System& system, VideoCore::RendererBase& renderer);
-    ~GPUSynch();
+    ~GPUSynch() override;
 
     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
@@ -105,32 +162,24 @@ 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);
-
-    /// Waits the caller until the GPU thread is idle, used for synchronization
-    void WaitForIdle();
+    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;
-    std::thread thread;
-    std::thread::id thread_id;
     VideoCore::RendererBase& renderer;
     Tegra::DmaPusher& dma_pusher;
+    std::thread thread;
+    std::thread::id thread_id;
 };
 
 } // namespace VideoCommon::GPUThread

src/video_core/morton.cpp

@@ -16,12 +16,12 @@ 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*, VAddr);
 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, VAddr addr) {
     constexpr u32 bytes_per_pixel = GetBytesPerPixel(format);
 
     // With the BCn formats (DXT and DXN), each 4x4 tile is swizzled instead of just individual
@@ -42,142 +42,138 @@ static void MortonCopy(u32 stride, u32 block_height, u32 height, u32 block_depth
 }
 
 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 +187,6 @@ 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 +282,14 @@ 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, VAddr 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, VAddr 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

@@ -13,6 +13,11 @@
 
 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) {}
 
@@ -26,11 +31,12 @@ GLintptr OGLBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t 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) {
-        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 +45,17 @@ 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>(*cpu_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

@@ -15,12 +15,13 @@
 
 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_) {
@@ -35,7 +36,7 @@ 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 {
@@ -46,19 +47,19 @@ GlobalRegion GlobalRegionCacheOpenGL::TryGetReservedGlobalRegion(VAddr addr, u32
     return search->second;
 }
 
-GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(VAddr addr, u32 size) {
+GlobalRegion GlobalRegionCacheOpenGL::GetUncachedGlobalRegion(VAddr addr, u32 size, u8* host_ptr) {
     GlobalRegion region{TryGetReservedGlobalRegion(addr, size)};
     if (!region) {
         // No reserved surface available, create a new one and reserve it
-        region = std::make_shared<CachedGlobalRegion>(addr, size);
+        region = std::make_shared<CachedGlobalRegion>(addr, size, host_ptr);
         ReserveGlobalRegion(region);
     }
     region->Reload(size);
     return region;
 }
 
-void GlobalRegionCacheOpenGL::ReserveGlobalRegion(const GlobalRegion& region) {
-    reserve[region->GetAddr()] = region;
+void GlobalRegionCacheOpenGL::ReserveGlobalRegion(GlobalRegion region) {
+    reserve.insert_or_assign(region->GetCpuAddr(), std::move(region));
 }
 
 GlobalRegionCacheOpenGL::GlobalRegionCacheOpenGL(RasterizerOpenGL& rasterizer)
@@ -80,11 +81,12 @@ GlobalRegion GlobalRegionCacheOpenGL::GetGlobalRegion(
     ASSERT(actual_addr);
 
     // Look up global region in the cache based on address
-    GlobalRegion region = TryGet(*actual_addr);
+    const auto& host_ptr{Memory::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

@@ -27,15 +27,13 @@ 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 {
-        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 {
+    std::size_t GetSizeInBytes() const override {
         return size;
     }
 
@@ -53,9 +51,8 @@ public:
     }
 
 private:
-    VAddr addr{};
+    VAddr cpu_addr{};
     u32 size{};
-
     OGLBuffer buffer;
 };
 
@@ -69,8 +66,8 @@ public:
 
 private:
     GlobalRegion TryGetReservedGlobalRegion(VAddr addr, u32 size) const;
-    GlobalRegion GetUncachedGlobalRegion(VAddr addr, u32 size);
-    void ReserveGlobalRegion(const GlobalRegion& region);
+    GlobalRegion GetUncachedGlobalRegion(VAddr addr, u32 size, u8* host_ptr);
+    void ReserveGlobalRegion(GlobalRegion region);
 
     std::unordered_map<VAddr, GlobalRegion> reserve;
 };

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -102,8 +102,9 @@ struct FramebufferCacheKey {
 
 RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, Core::System& system,
                                    ScreenInfo& info)
-    : res_cache{*this}, shader_cache{*this, system}, global_cache{*this}, emu_window{window},
-      screen_info{info}, buffer_cache(*this, STREAM_BUFFER_SIZE) {
+    : res_cache{*this}, shader_cache{*this, system}, global_cache{*this},
+      emu_window{window}, system{system}, screen_info{info},
+      buffer_cache(*this, STREAM_BUFFER_SIZE) {
     // Create sampler objects
     for (std::size_t i = 0; i < texture_samplers.size(); ++i) {
         texture_samplers[i].Create();
@@ -118,7 +119,7 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, Core::Syst
 
     glGetIntegerv(GL_UNIFORM_BUFFER_OFFSET_ALIGNMENT, &uniform_buffer_alignment);
 
-    LOG_CRITICAL(Render_OpenGL, "Sync fixed function OpenGL state here!");
+    LOG_DEBUG(Render_OpenGL, "Sync fixed function OpenGL state here");
     CheckExtensions();
 }
 
@@ -138,7 +139,7 @@ void RasterizerOpenGL::CheckExtensions() {
 }
 
 GLuint RasterizerOpenGL::SetupVertexFormat() {
-    auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
+    auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
 
     if (!gpu.dirty_flags.vertex_attrib_format) {
@@ -177,7 +178,7 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
                 continue;
 
             const auto& buffer = regs.vertex_array[attrib.buffer];
-            LOG_TRACE(HW_GPU,
+            LOG_TRACE(Render_OpenGL,
                       "vertex attrib {}, count={}, size={}, type={}, offset={}, normalize={}",
                       index, attrib.ComponentCount(), attrib.SizeString(), attrib.TypeString(),
                       attrib.offset.Value(), attrib.IsNormalized());
@@ -207,7 +208,7 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
 }
 
 void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
-    auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
+    auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
 
     if (gpu.dirty_flags.vertex_array.none())
@@ -248,7 +249,7 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
 }
 
 DrawParameters RasterizerOpenGL::SetupDraw() {
-    const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
+    const auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
     const bool is_indexed = accelerate_draw == AccelDraw::Indexed;
 
@@ -297,7 +298,7 @@ DrawParameters RasterizerOpenGL::SetupDraw() {
 
 void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
     MICROPROFILE_SCOPE(OpenGL_Shader);
-    auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
+    auto& gpu = system.GPU().Maxwell3D();
 
     BaseBindings base_bindings;
     std::array<bool, Maxwell::NumClipDistances> clip_distances{};
@@ -343,9 +344,8 @@ void RasterizerOpenGL::SetupShaders(GLenum primitive_mode) {
             shader_program_manager->UseProgrammableFragmentShader(program_handle);
             break;
         default:
-            LOG_CRITICAL(HW_GPU, "Unimplemented shader index={}, enable={}, offset=0x{:08X}", index,
-                         shader_config.enable.Value(), shader_config.offset);
-            UNREACHABLE();
+            UNIMPLEMENTED_MSG("Unimplemented shader index={}, enable={}, offset=0x{:08X}", index,
+                              shader_config.enable.Value(), shader_config.offset);
         }
 
         const auto stage_enum = static_cast<Maxwell::ShaderStage>(stage);
@@ -414,7 +414,7 @@ void RasterizerOpenGL::SetupCachedFramebuffer(const FramebufferCacheKey& fbkey,
 }
 
 std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
 
     std::size_t size = 0;
     for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
@@ -432,7 +432,7 @@ std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
 }
 
 std::size_t RasterizerOpenGL::CalculateIndexBufferSize() const {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
 
     return static_cast<std::size_t>(regs.index_array.count) *
            static_cast<std::size_t>(regs.index_array.FormatSizeInBytes());
@@ -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};
 
@@ -488,7 +488,7 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
     OpenGLState& current_state, bool using_color_fb, bool using_depth_fb, bool preserve_contents,
     std::optional<std::size_t> single_color_target) {
     MICROPROFILE_SCOPE(OpenGL_Framebuffer);
-    auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
+    auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
 
     const FramebufferConfigState fb_config_state{using_color_fb, using_depth_fb, preserve_contents,
@@ -582,7 +582,7 @@ void RasterizerOpenGL::Clear() {
     const auto prev_state{state};
     SCOPE_EXIT({ prev_state.Apply(); });
 
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     bool use_color{};
     bool use_depth{};
     bool use_stencil{};
@@ -673,7 +673,7 @@ void RasterizerOpenGL::DrawArrays() {
         return;
 
     MICROPROFILE_SCOPE(OpenGL_Drawing);
-    auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
+    auto& gpu = system.GPU().Maxwell3D();
     const auto& regs = gpu.regs;
 
     ConfigureFramebuffers(state);
@@ -747,12 +747,12 @@ void RasterizerOpenGL::DrawArrays() {
 
 void RasterizerOpenGL::FlushAll() {}
 
-void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
     res_cache.FlushRegion(addr, size);
 }
 
-void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
     res_cache.InvalidateRegion(addr, size);
     shader_cache.InvalidateRegion(addr, size);
@@ -760,7 +760,7 @@ void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 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 +782,7 @@ 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 {};
     }
@@ -793,7 +793,10 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
         VideoCore::Surface::PixelFormatFromGPUPixelFormat(config.pixel_format)};
     ASSERT_MSG(params.width == config.width, "Framebuffer width is different");
     ASSERT_MSG(params.height == config.height, "Framebuffer height is different");
-    ASSERT_MSG(params.pixel_format == pixel_format, "Framebuffer pixel_format is different");
+
+    if (params.pixel_format != pixel_format) {
+        LOG_WARNING(Render_OpenGL, "Framebuffer pixel_format is different");
+    }
 
     screen_info.display_texture = surface->Texture().handle;
 
@@ -802,104 +805,87 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
 
 void RasterizerOpenGL::SamplerInfo::Create() {
     sampler.Create();
-    mag_filter = min_filter = Tegra::Texture::TextureFilter::Linear;
-    wrap_u = wrap_v = wrap_p = Tegra::Texture::WrapMode::Wrap;
-    uses_depth_compare = false;
+    mag_filter = Tegra::Texture::TextureFilter::Linear;
+    min_filter = Tegra::Texture::TextureFilter::Linear;
+    wrap_u = Tegra::Texture::WrapMode::Wrap;
+    wrap_v = Tegra::Texture::WrapMode::Wrap;
+    wrap_p = Tegra::Texture::WrapMode::Wrap;
+    use_depth_compare = false;
     depth_compare_func = Tegra::Texture::DepthCompareFunc::Never;
 
-    // default is GL_LINEAR_MIPMAP_LINEAR
+    // OpenGL's default is GL_LINEAR_MIPMAP_LINEAR
     glSamplerParameteri(sampler.handle, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
-    // Other attributes have correct defaults
     glSamplerParameteri(sampler.handle, GL_TEXTURE_COMPARE_FUNC, GL_NEVER);
+
+    // Other attributes have correct defaults
 }
 
 void RasterizerOpenGL::SamplerInfo::SyncWithConfig(const Tegra::Texture::TSCEntry& config) {
-    const GLuint s = sampler.handle;
+    const GLuint sampler_id = sampler.handle;
     if (mag_filter != config.mag_filter) {
         mag_filter = config.mag_filter;
         glSamplerParameteri(
-            s, GL_TEXTURE_MAG_FILTER,
+            sampler_id, GL_TEXTURE_MAG_FILTER,
             MaxwellToGL::TextureFilterMode(mag_filter, Tegra::Texture::TextureMipmapFilter::None));
     }
-    if (min_filter != config.min_filter || mip_filter != config.mip_filter) {
+    if (min_filter != config.min_filter || mipmap_filter != config.mipmap_filter) {
         min_filter = config.min_filter;
-        mip_filter = config.mip_filter;
-        glSamplerParameteri(s, GL_TEXTURE_MIN_FILTER,
-                            MaxwellToGL::TextureFilterMode(min_filter, mip_filter));
+        mipmap_filter = config.mipmap_filter;
+        glSamplerParameteri(sampler_id, GL_TEXTURE_MIN_FILTER,
+                            MaxwellToGL::TextureFilterMode(min_filter, mipmap_filter));
     }
 
     if (wrap_u != config.wrap_u) {
         wrap_u = config.wrap_u;
-        glSamplerParameteri(s, GL_TEXTURE_WRAP_S, MaxwellToGL::WrapMode(wrap_u));
+        glSamplerParameteri(sampler_id, GL_TEXTURE_WRAP_S, MaxwellToGL::WrapMode(wrap_u));
     }
     if (wrap_v != config.wrap_v) {
         wrap_v = config.wrap_v;
-        glSamplerParameteri(s, GL_TEXTURE_WRAP_T, MaxwellToGL::WrapMode(wrap_v));
+        glSamplerParameteri(sampler_id, GL_TEXTURE_WRAP_T, MaxwellToGL::WrapMode(wrap_v));
     }
     if (wrap_p != config.wrap_p) {
         wrap_p = config.wrap_p;
-        glSamplerParameteri(s, GL_TEXTURE_WRAP_R, MaxwellToGL::WrapMode(wrap_p));
+        glSamplerParameteri(sampler_id, GL_TEXTURE_WRAP_R, MaxwellToGL::WrapMode(wrap_p));
     }
 
-    if (uses_depth_compare != (config.depth_compare_enabled == 1)) {
-        uses_depth_compare = (config.depth_compare_enabled == 1);
-        if (uses_depth_compare) {
-            glSamplerParameteri(s, GL_TEXTURE_COMPARE_MODE, GL_COMPARE_REF_TO_TEXTURE);
-        } else {
-            glSamplerParameteri(s, GL_TEXTURE_COMPARE_MODE, GL_NONE);
-        }
+    if (const bool enabled = config.depth_compare_enabled == 1; use_depth_compare != enabled) {
+        use_depth_compare = enabled;
+        glSamplerParameteri(sampler_id, GL_TEXTURE_COMPARE_MODE,
+                            use_depth_compare ? GL_COMPARE_REF_TO_TEXTURE : GL_NONE);
     }
 
     if (depth_compare_func != config.depth_compare_func) {
         depth_compare_func = config.depth_compare_func;
-        glSamplerParameteri(s, GL_TEXTURE_COMPARE_FUNC,
+        glSamplerParameteri(sampler_id, GL_TEXTURE_COMPARE_FUNC,
                             MaxwellToGL::DepthCompareFunc(depth_compare_func));
     }
 
-    GLvec4 new_border_color;
-    if (config.srgb_conversion) {
-        new_border_color[0] = config.srgb_border_color_r / 255.0f;
-        new_border_color[1] = config.srgb_border_color_g / 255.0f;
-        new_border_color[2] = config.srgb_border_color_g / 255.0f;
-    } else {
-        new_border_color[0] = config.border_color_r;
-        new_border_color[1] = config.border_color_g;
-        new_border_color[2] = config.border_color_b;
-    }
-    new_border_color[3] = config.border_color_a;
-
-    if (border_color != new_border_color) {
+    if (const auto new_border_color = config.GetBorderColor(); border_color != new_border_color) {
         border_color = new_border_color;
-        glSamplerParameterfv(s, GL_TEXTURE_BORDER_COLOR, border_color.data());
+        glSamplerParameterfv(sampler_id, GL_TEXTURE_BORDER_COLOR, border_color.data());
     }
 
-    const float anisotropic_max = static_cast<float>(1 << config.max_anisotropy.Value());
-    if (anisotropic_max != max_anisotropic) {
-        max_anisotropic = anisotropic_max;
+    if (const float anisotropic = config.GetMaxAnisotropy(); max_anisotropic != anisotropic) {
+        max_anisotropic = anisotropic;
         if (GLAD_GL_ARB_texture_filter_anisotropic) {
-            glSamplerParameterf(s, GL_TEXTURE_MAX_ANISOTROPY, max_anisotropic);
+            glSamplerParameterf(sampler_id, GL_TEXTURE_MAX_ANISOTROPY, max_anisotropic);
         } else if (GLAD_GL_EXT_texture_filter_anisotropic) {
-            glSamplerParameterf(s, GL_TEXTURE_MAX_ANISOTROPY_EXT, max_anisotropic);
+            glSamplerParameterf(sampler_id, GL_TEXTURE_MAX_ANISOTROPY_EXT, max_anisotropic);
         }
     }
-    const float lod_min = static_cast<float>(config.min_lod_clamp.Value()) / 256.0f;
-    if (lod_min != min_lod) {
-        min_lod = lod_min;
-        glSamplerParameterf(s, GL_TEXTURE_MIN_LOD, min_lod);
+
+    if (const float min = config.GetMinLod(); min_lod != min) {
+        min_lod = min;
+        glSamplerParameterf(sampler_id, GL_TEXTURE_MIN_LOD, min_lod);
+    }
+    if (const float max = config.GetMaxLod(); max_lod != max) {
+        max_lod = max;
+        glSamplerParameterf(sampler_id, GL_TEXTURE_MAX_LOD, max_lod);
     }
 
-    const float lod_max = static_cast<float>(config.max_lod_clamp.Value()) / 256.0f;
-    if (lod_max != max_lod) {
-        max_lod = lod_max;
-        glSamplerParameterf(s, GL_TEXTURE_MAX_LOD, max_lod);
-    }
-    const u32 bias = config.mip_lod_bias.Value();
-    // Sign extend the 13-bit value.
-    constexpr u32 mask = 1U << (13 - 1);
-    const float bias_lod = static_cast<s32>((bias ^ mask) - mask) / 256.f;
-    if (lod_bias != bias_lod) {
-        lod_bias = bias_lod;
-        glSamplerParameterf(s, GL_TEXTURE_LOD_BIAS, lod_bias);
+    if (const float bias = config.GetLodBias(); lod_bias != bias) {
+        lod_bias = bias;
+        glSamplerParameterf(sampler_id, GL_TEXTURE_LOD_BIAS, lod_bias);
     }
 }
 
@@ -907,7 +893,7 @@ void RasterizerOpenGL::SetupConstBuffers(Tegra::Engines::Maxwell3D::Regs::Shader
                                          const Shader& shader, GLuint program_handle,
                                          BaseBindings base_bindings) {
     MICROPROFILE_SCOPE(OpenGL_UBO);
-    const auto& gpu = Core::System::GetInstance().GPU();
+    const auto& gpu = system.GPU();
     const auto& maxwell3d = gpu.Maxwell3D();
     const auto& shader_stage = maxwell3d.state.shader_stages[static_cast<std::size_t>(stage)];
     const auto& entries = shader->GetShaderEntries().const_buffers;
@@ -939,8 +925,8 @@ void RasterizerOpenGL::SetupConstBuffers(Tegra::Engines::Maxwell3D::Regs::Shader
             size = buffer.size;
 
             if (size > MaxConstbufferSize) {
-                LOG_CRITICAL(HW_GPU, "indirect constbuffer size {} exceeds maximum {}", size,
-                             MaxConstbufferSize);
+                LOG_WARNING(Render_OpenGL, "Indirect constbuffer size {} exceeds maximum {}", size,
+                            MaxConstbufferSize);
                 size = MaxConstbufferSize;
             }
         } else {
@@ -986,7 +972,7 @@ void RasterizerOpenGL::SetupGlobalRegions(Tegra::Engines::Maxwell3D::Regs::Shade
 void RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& shader,
                                      GLuint program_handle, BaseBindings base_bindings) {
     MICROPROFILE_SCOPE(OpenGL_Texture);
-    const auto& gpu = Core::System::GetInstance().GPU();
+    const auto& gpu = system.GPU();
     const auto& maxwell3d = gpu.Maxwell3D();
     const auto& entries = shader->GetShaderEntries().samplers;
 
@@ -1000,10 +986,9 @@ void RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& s
 
         texture_samplers[current_bindpoint].SyncWithConfig(texture.tsc);
 
-        Surface surface = res_cache.GetTextureSurface(texture, entry);
-        if (surface != nullptr) {
+        if (Surface surface = res_cache.GetTextureSurface(texture, entry); surface) {
             state.texture_units[current_bindpoint].texture =
-                entry.IsArray() ? surface->TextureLayer().handle : surface->Texture().handle;
+                surface->Texture(entry.IsArray()).handle;
             surface->UpdateSwizzle(texture.tic.x_source, texture.tic.y_source, texture.tic.z_source,
                                    texture.tic.w_source);
         } else {
@@ -1014,7 +999,7 @@ void RasterizerOpenGL::SetupTextures(Maxwell::ShaderStage stage, const Shader& s
 }
 
 void RasterizerOpenGL::SyncViewport(OpenGLState& current_state) {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     const bool geometry_shaders_enabled =
         regs.IsShaderConfigEnabled(static_cast<size_t>(Maxwell::ShaderProgram::Geometry));
     const std::size_t viewport_count =
@@ -1037,7 +1022,7 @@ void RasterizerOpenGL::SyncViewport(OpenGLState& current_state) {
 void RasterizerOpenGL::SyncClipEnabled(
     const std::array<bool, Maxwell::Regs::NumClipDistances>& clip_mask) {
 
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     const std::array<bool, Maxwell::Regs::NumClipDistances> reg_state{
         regs.clip_distance_enabled.c0 != 0, regs.clip_distance_enabled.c1 != 0,
         regs.clip_distance_enabled.c2 != 0, regs.clip_distance_enabled.c3 != 0,
@@ -1054,7 +1039,7 @@ void RasterizerOpenGL::SyncClipCoef() {
 }
 
 void RasterizerOpenGL::SyncCullMode() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
 
     state.cull.enabled = regs.cull.enabled != 0;
 
@@ -1078,14 +1063,14 @@ void RasterizerOpenGL::SyncCullMode() {
 }
 
 void RasterizerOpenGL::SyncPrimitiveRestart() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
 
     state.primitive_restart.enabled = regs.primitive_restart.enabled;
     state.primitive_restart.index = regs.primitive_restart.index;
 }
 
 void RasterizerOpenGL::SyncDepthTestState() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
 
     state.depth.test_enabled = regs.depth_test_enable != 0;
     state.depth.write_mask = regs.depth_write_enabled ? GL_TRUE : GL_FALSE;
@@ -1097,7 +1082,7 @@ void RasterizerOpenGL::SyncDepthTestState() {
 }
 
 void RasterizerOpenGL::SyncStencilTestState() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     state.stencil.test_enabled = regs.stencil_enable != 0;
 
     if (!regs.stencil_enable) {
@@ -1131,7 +1116,7 @@ void RasterizerOpenGL::SyncStencilTestState() {
 }
 
 void RasterizerOpenGL::SyncColorMask() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     const std::size_t count =
         regs.independent_blend_enable ? Tegra::Engines::Maxwell3D::Regs::NumRenderTargets : 1;
     for (std::size_t i = 0; i < count; i++) {
@@ -1145,18 +1130,18 @@ void RasterizerOpenGL::SyncColorMask() {
 }
 
 void RasterizerOpenGL::SyncMultiSampleState() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     state.multisample_control.alpha_to_coverage = regs.multisample_control.alpha_to_coverage != 0;
     state.multisample_control.alpha_to_one = regs.multisample_control.alpha_to_one != 0;
 }
 
 void RasterizerOpenGL::SyncFragmentColorClampState() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     state.fragment_color_clamp.enabled = regs.frag_color_clamp != 0;
 }
 
 void RasterizerOpenGL::SyncBlendState() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
 
     state.blend_color.red = regs.blend_color.r;
     state.blend_color.green = regs.blend_color.g;
@@ -1198,7 +1183,7 @@ void RasterizerOpenGL::SyncBlendState() {
 }
 
 void RasterizerOpenGL::SyncLogicOpState() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
 
     state.logic_op.enabled = regs.logic_op.enable != 0;
 
@@ -1212,7 +1197,7 @@ void RasterizerOpenGL::SyncLogicOpState() {
 }
 
 void RasterizerOpenGL::SyncScissorTest(OpenGLState& current_state) {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     const bool geometry_shaders_enabled =
         regs.IsShaderConfigEnabled(static_cast<size_t>(Maxwell::ShaderProgram::Geometry));
     const std::size_t viewport_count =
@@ -1234,21 +1219,17 @@ void RasterizerOpenGL::SyncScissorTest(OpenGLState& current_state) {
 }
 
 void RasterizerOpenGL::SyncTransformFeedback() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
-
-    if (regs.tfb_enabled != 0) {
-        LOG_CRITICAL(Render_OpenGL, "Transform feedbacks are not implemented");
-        UNREACHABLE();
-    }
+    const auto& regs = system.GPU().Maxwell3D().regs;
+    UNIMPLEMENTED_IF_MSG(regs.tfb_enabled != 0, "Transform feedbacks are not implemented");
 }
 
 void RasterizerOpenGL::SyncPointState() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     state.point.size = regs.point_size;
 }
 
 void RasterizerOpenGL::SyncPolygonOffset() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
+    const auto& regs = system.GPU().Maxwell3D().regs;
     state.polygon_offset.fill_enable = regs.polygon_offset_fill_enable != 0;
     state.polygon_offset.line_enable = regs.polygon_offset_line_enable != 0;
     state.polygon_offset.point_enable = regs.polygon_offset_point_enable != 0;
@@ -1258,13 +1239,9 @@ void RasterizerOpenGL::SyncPolygonOffset() {
 }
 
 void RasterizerOpenGL::CheckAlphaTests() {
-    const auto& regs = Core::System::GetInstance().GPU().Maxwell3D().regs;
-
-    if (regs.alpha_test_enabled != 0 && regs.rt_control.count > 1) {
-        LOG_CRITICAL(Render_OpenGL, "Alpha Testing is enabled with Multiple Render Targets, "
-                                    "this behavior is undefined.");
-        UNREACHABLE();
-    }
+    const auto& regs = system.GPU().Maxwell3D().regs;
+    UNIMPLEMENTED_IF_MSG(regs.alpha_test_enabled != 0 && regs.rt_control.count > 1,
+                         "Alpha Testing is enabled with more than one rendertarget");
 }
 
 } // namespace OpenGL

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;
 
@@ -94,11 +94,12 @@ private:
     private:
         Tegra::Texture::TextureFilter mag_filter = Tegra::Texture::TextureFilter::Nearest;
         Tegra::Texture::TextureFilter min_filter = Tegra::Texture::TextureFilter::Nearest;
-        Tegra::Texture::TextureMipmapFilter mip_filter = Tegra::Texture::TextureMipmapFilter::None;
+        Tegra::Texture::TextureMipmapFilter mipmap_filter =
+            Tegra::Texture::TextureMipmapFilter::None;
         Tegra::Texture::WrapMode wrap_u = Tegra::Texture::WrapMode::ClampToEdge;
         Tegra::Texture::WrapMode wrap_v = Tegra::Texture::WrapMode::ClampToEdge;
         Tegra::Texture::WrapMode wrap_p = Tegra::Texture::WrapMode::ClampToEdge;
-        bool uses_depth_compare = false;
+        bool use_depth_compare = false;
         Tegra::Texture::DepthCompareFunc depth_compare_func =
             Tegra::Texture::DepthCompareFunc::Always;
         GLvec4 border_color = {};
@@ -214,6 +215,7 @@ private:
     GlobalRegionCacheOpenGL global_cache;
 
     Core::Frontend::EmuWindow& emu_window;
+    Core::System& system;
 
     ScreenInfo& screen_info;
 

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -21,7 +21,7 @@
 #include "video_core/renderer_opengl/gl_rasterizer_cache.h"
 #include "video_core/renderer_opengl/utils.h"
 #include "video_core/surface.h"
-#include "video_core/textures/astc.h"
+#include "video_core/textures/convert.h"
 #include "video_core/textures/decoders.h"
 
 namespace OpenGL {
@@ -61,6 +61,7 @@ void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) {
 
     addr = cpu_addr ? *cpu_addr : 0;
     gpu_addr = gpu_addr_;
+    host_ptr = Memory::GetPointer(addr);
     size_in_bytes = SizeInBytesRaw();
 
     if (IsPixelFormatASTC(pixel_format)) {
@@ -400,6 +401,27 @@ static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType
     return format;
 }
 
+/// Returns the discrepant array target
+constexpr GLenum GetArrayDiscrepantTarget(SurfaceTarget target) {
+    switch (target) {
+    case SurfaceTarget::Texture1D:
+        return GL_TEXTURE_1D_ARRAY;
+    case SurfaceTarget::Texture2D:
+        return GL_TEXTURE_2D_ARRAY;
+    case SurfaceTarget::Texture3D:
+        return GL_NONE;
+    case SurfaceTarget::Texture1DArray:
+        return GL_TEXTURE_1D;
+    case SurfaceTarget::Texture2DArray:
+        return GL_TEXTURE_2D;
+    case SurfaceTarget::TextureCubemap:
+        return GL_TEXTURE_CUBE_MAP_ARRAY;
+    case SurfaceTarget::TextureCubeArray:
+        return GL_TEXTURE_CUBE_MAP;
+    }
+    return GL_NONE;
+}
+
 Common::Rectangle<u32> SurfaceParams::GetRect(u32 mip_level) const {
     u32 actual_height{std::max(1U, unaligned_height >> mip_level)};
     if (IsPixelFormatASTC(pixel_format)) {
@@ -425,7 +447,7 @@ 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.addr + offset);
             offset += layer_size;
             offset_gl += gl_size;
         }
@@ -434,7 +456,7 @@ 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.addr + offset);
     }
 }
 
@@ -542,8 +564,8 @@ 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)
@@ -597,103 +619,6 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
     }
 }
 
-static void ConvertS8Z24ToZ24S8(std::vector<u8>& data, u32 width, u32 height, bool reverse) {
-    union S8Z24 {
-        BitField<0, 24, u32> z24;
-        BitField<24, 8, u32> s8;
-    };
-    static_assert(sizeof(S8Z24) == 4, "S8Z24 is incorrect size");
-
-    union Z24S8 {
-        BitField<0, 8, u32> s8;
-        BitField<8, 24, u32> z24;
-    };
-    static_assert(sizeof(Z24S8) == 4, "Z24S8 is incorrect size");
-
-    S8Z24 s8z24_pixel{};
-    Z24S8 z24s8_pixel{};
-    constexpr auto bpp{GetBytesPerPixel(PixelFormat::S8Z24)};
-    for (std::size_t y = 0; y < height; ++y) {
-        for (std::size_t x = 0; x < width; ++x) {
-            const std::size_t offset{bpp * (y * width + x)};
-            if (reverse) {
-                std::memcpy(&z24s8_pixel, &data[offset], sizeof(Z24S8));
-                s8z24_pixel.s8.Assign(z24s8_pixel.s8);
-                s8z24_pixel.z24.Assign(z24s8_pixel.z24);
-                std::memcpy(&data[offset], &s8z24_pixel, sizeof(S8Z24));
-            } else {
-                std::memcpy(&s8z24_pixel, &data[offset], sizeof(S8Z24));
-                z24s8_pixel.s8.Assign(s8z24_pixel.s8);
-                z24s8_pixel.z24.Assign(s8z24_pixel.z24);
-                std::memcpy(&data[offset], &z24s8_pixel, sizeof(Z24S8));
-            }
-        }
-    }
-}
-
-/**
- * Helper function to perform software conversion (as needed) when loading a buffer from Switch
- * memory. This is for Maxwell pixel formats that cannot be represented as-is in OpenGL or with
- * typical desktop GPUs.
- */
-static void ConvertFormatAsNeeded_LoadGLBuffer(std::vector<u8>& data, PixelFormat pixel_format,
-                                               u32 width, u32 height, u32 depth) {
-    switch (pixel_format) {
-    case PixelFormat::ASTC_2D_4X4:
-    case PixelFormat::ASTC_2D_8X8:
-    case PixelFormat::ASTC_2D_8X5:
-    case PixelFormat::ASTC_2D_5X4:
-    case PixelFormat::ASTC_2D_5X5:
-    case PixelFormat::ASTC_2D_4X4_SRGB:
-    case PixelFormat::ASTC_2D_8X8_SRGB:
-    case PixelFormat::ASTC_2D_8X5_SRGB:
-    case PixelFormat::ASTC_2D_5X4_SRGB:
-    case PixelFormat::ASTC_2D_5X5_SRGB:
-    case PixelFormat::ASTC_2D_10X8:
-    case PixelFormat::ASTC_2D_10X8_SRGB: {
-        // Convert ASTC pixel formats to RGBA8, as most desktop GPUs do not support ASTC.
-        u32 block_width{};
-        u32 block_height{};
-        std::tie(block_width, block_height) = GetASTCBlockSize(pixel_format);
-        data =
-            Tegra::Texture::ASTC::Decompress(data, width, height, depth, block_width, block_height);
-        break;
-    }
-    case PixelFormat::S8Z24:
-        // Convert the S8Z24 depth format to Z24S8, as OpenGL does not support S8Z24.
-        ConvertS8Z24ToZ24S8(data, width, height, false);
-        break;
-    }
-}
-
-/**
- * Helper function to perform software conversion (as needed) when flushing a buffer from OpenGL to
- * Switch memory. This is for Maxwell pixel formats that cannot be represented as-is in OpenGL or
- * with typical desktop GPUs.
- */
-static void ConvertFormatAsNeeded_FlushGLBuffer(std::vector<u8>& data, PixelFormat pixel_format,
-                                                u32 width, u32 height) {
-    switch (pixel_format) {
-    case PixelFormat::ASTC_2D_4X4:
-    case PixelFormat::ASTC_2D_8X8:
-    case PixelFormat::ASTC_2D_4X4_SRGB:
-    case PixelFormat::ASTC_2D_8X8_SRGB:
-    case PixelFormat::ASTC_2D_5X5:
-    case PixelFormat::ASTC_2D_5X5_SRGB:
-    case PixelFormat::ASTC_2D_10X8:
-    case PixelFormat::ASTC_2D_10X8_SRGB: {
-        LOG_CRITICAL(HW_GPU, "Conversion of format {} after texture flushing is not implemented",
-                     static_cast<u32>(pixel_format));
-        UNREACHABLE();
-        break;
-    }
-    case PixelFormat::S8Z24:
-        // Convert the Z24S8 depth format to S8Z24, as OpenGL does not support S8Z24.
-        ConvertS8Z24ToZ24S8(data, width, height, true);
-        break;
-    }
-}
-
 MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 192, 64));
 void CachedSurface::LoadGLBuffer() {
     MICROPROFILE_SCOPE(OpenGL_SurfaceLoad);
@@ -709,10 +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);
@@ -722,8 +646,16 @@ void CachedSurface::LoadGLBuffer() {
         }
     }
     for (u32 i = 0; i < params.max_mip_level; i++) {
-        ConvertFormatAsNeeded_LoadGLBuffer(gl_buffer[i], params.pixel_format, params.MipWidth(i),
-                                           params.MipHeight(i), params.MipDepth(i));
+        const u32 width = params.MipWidth(i);
+        const u32 height = params.MipHeight(i);
+        const u32 depth = params.MipDepth(i);
+        if (VideoCore::Surface::IsPixelFormatASTC(params.pixel_format)) {
+            // Reserve size for RGBA8 conversion
+            constexpr std::size_t rgba_bpp = 4;
+            gl_buffer[i].resize(std::max(gl_buffer[i].size(), width * height * depth * rgba_bpp));
+        }
+        Tegra::Texture::ConvertFromGuestToHost(gl_buffer[i].data(), params.pixel_format, width,
+                                               height, depth, true, true);
     }
 }
 
@@ -746,10 +678,8 @@ void CachedSurface::FlushGLBuffer() {
     glGetTextureImage(texture.handle, 0, tuple.format, tuple.type,
                       static_cast<GLsizei>(gl_buffer[0].size()), gl_buffer[0].data());
     glPixelStorei(GL_PACK_ROW_LENGTH, 0);
-    ConvertFormatAsNeeded_FlushGLBuffer(gl_buffer[0], params.pixel_format, params.width,
-                                        params.height);
-    const u8* const texture_src_data = Memory::GetPointer(params.addr);
-    ASSERT(texture_src_data);
+    Tegra::Texture::ConvertFromHostToGuest(gl_buffer[0].data(), params.pixel_format, params.width,
+                                           params.height, params.depth, true, true);
     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));
@@ -759,9 +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);
@@ -884,20 +814,22 @@ void CachedSurface::UploadGLMipmapTexture(u32 mip_map, GLuint read_fb_handle,
     glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
 }
 
-void CachedSurface::EnsureTextureView() {
-    if (texture_view.handle != 0)
+void CachedSurface::EnsureTextureDiscrepantView() {
+    if (discrepant_view.handle != 0)
         return;
 
-    const GLenum target{TargetLayer()};
+    const GLenum target{GetArrayDiscrepantTarget(params.target)};
+    ASSERT(target != GL_NONE);
+
     const GLuint num_layers{target == GL_TEXTURE_CUBE_MAP_ARRAY ? 6u : 1u};
     constexpr GLuint min_layer = 0;
     constexpr GLuint min_level = 0;
 
-    glGenTextures(1, &texture_view.handle);
-    glTextureView(texture_view.handle, target, texture.handle, gl_internal_format, min_level,
+    glGenTextures(1, &discrepant_view.handle);
+    glTextureView(discrepant_view.handle, target, texture.handle, gl_internal_format, min_level,
                   params.max_mip_level, min_layer, num_layers);
-    ApplyTextureDefaults(texture_view.handle, params.max_mip_level);
-    glTextureParameteriv(texture_view.handle, GL_TEXTURE_SWIZZLE_RGBA,
+    ApplyTextureDefaults(discrepant_view.handle, params.max_mip_level);
+    glTextureParameteriv(discrepant_view.handle, GL_TEXTURE_SWIZZLE_RGBA,
                          reinterpret_cast<const GLint*>(swizzle.data()));
 }
 
@@ -923,8 +855,8 @@ void CachedSurface::UpdateSwizzle(Tegra::Texture::SwizzleSource swizzle_x,
     swizzle = {new_x, new_y, new_z, new_w};
     const auto swizzle_data = reinterpret_cast<const GLint*>(swizzle.data());
     glTextureParameteriv(texture.handle, GL_TEXTURE_SWIZZLE_RGBA, swizzle_data);
-    if (texture_view.handle != 0) {
-        glTextureParameteriv(texture_view.handle, GL_TEXTURE_SWIZZLE_RGBA, swizzle_data);
+    if (discrepant_view.handle != 0) {
+        glTextureParameteriv(discrepant_view.handle, GL_TEXTURE_SWIZZLE_RGBA, swizzle_data);
     }
 }
 
@@ -998,7 +930,7 @@ Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool pres
     }
 
     // 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
@@ -1052,7 +984,7 @@ void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface,
     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);
+            const Surface& copy = TryGet(Memory::GetPointer(sub_address));
             if (!copy)
                 continue;
             const auto& src_params{copy->GetSurfaceParams()};
@@ -1229,7 +1161,8 @@ 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);
 }
@@ -1281,8 +1214,8 @@ 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) {
@@ -1333,7 +1266,7 @@ 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->GetCpuAddr(), dst_params, *level);
             if (slot.has_value()) {
                 glCopyImageSubData(render_surface->Texture().handle,
                                    SurfaceTargetToGL(src_params.target), 0, 0, 0, 0,
@@ -1349,8 +1282,8 @@ 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();
@@ -1393,7 +1326,8 @@ 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

@@ -297,6 +297,7 @@ struct SurfaceParams {
     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;
@@ -345,9 +346,9 @@ class RasterizerOpenGL;
 
 class CachedSurface final : public RasterizerCacheObject {
 public:
-    CachedSurface(const SurfaceParams& params);
+    explicit CachedSurface(const SurfaceParams& params);
 
-    VAddr GetAddr() const override {
+    VAddr GetCpuAddr() const override {
         return params.addr;
     }
 
@@ -367,31 +368,19 @@ public:
         return texture;
     }
 
-    const OGLTexture& TextureLayer() {
-        if (params.is_array) {
-            return Texture();
+    const OGLTexture& Texture(bool as_array) {
+        if (params.is_array == as_array) {
+            return texture;
+        } else {
+            EnsureTextureDiscrepantView();
+            return discrepant_view;
         }
-        EnsureTextureView();
-        return texture_view;
     }
 
     GLenum Target() const {
         return gl_target;
     }
 
-    GLenum TargetLayer() const {
-        using VideoCore::Surface::SurfaceTarget;
-        switch (params.target) {
-        case SurfaceTarget::Texture1D:
-            return GL_TEXTURE_1D_ARRAY;
-        case SurfaceTarget::Texture2D:
-            return GL_TEXTURE_2D_ARRAY;
-        case SurfaceTarget::TextureCubemap:
-            return GL_TEXTURE_CUBE_MAP_ARRAY;
-        }
-        return Target();
-    }
-
     const SurfaceParams& GetSurfaceParams() const {
         return params;
     }
@@ -431,10 +420,10 @@ public:
 private:
     void UploadGLMipmapTexture(u32 mip_map, GLuint read_fb_handle, GLuint draw_fb_handle);
 
-    void EnsureTextureView();
+    void EnsureTextureDiscrepantView();
 
     OGLTexture texture;
-    OGLTexture texture_view;
+    OGLTexture discrepant_view;
     std::vector<std::vector<u8>> gl_buffer;
     SurfaceParams params{};
     GLenum gl_target{};
@@ -461,7 +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,
@@ -518,12 +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.
@@ -535,7 +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

@@ -42,9 +42,9 @@ VAddr GetShaderAddress(Maxwell::ShaderProgram program) {
 }
 
 /// 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 +214,13 @@ std::set<GLenum> GetSupportedFormats() {
 
 } // namespace
 
-CachedShader::CachedShader(VAddr addr, u64 unique_identifier, Maxwell::ShaderProgram program_type,
-                           ShaderDiskCacheOpenGL& disk_cache,
+CachedShader::CachedShader(VAddr guest_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}, guest_addr{guest_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 +244,13 @@ 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 guest_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)
+    : guest_addr{guest_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 +273,7 @@ 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, guest_addr);
         }
 
         handle = program->handle;
@@ -323,7 +325,7 @@ 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, guest_addr, debug_name);
 
     return target_program->handle;
 };
@@ -489,14 +491,17 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
     const VAddr program_addr{GetShaderAddress(program)};
 
     // Look up shader in the cache based on address
-    Shader shader{TryGet(program_addr)};
+    const auto& host_ptr{Memory::GetPointer(program_addr)};
+    Shader shader{TryGet(host_ptr)};
 
     if (!shader) {
         // No shader found - create a new one
-        ProgramCode program_code = GetShaderCode(program_addr);
+        const auto& host_ptr{Memory::GetPointer(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::GetPointer(GetShaderAddress(Maxwell::ShaderProgram::VertexB)));
         }
         const u64 unique_identifier = GetUniqueIdentifier(program, program_code, program_code_b);
 
@@ -504,11 +509,11 @@ Shader ShaderCacheOpenGL::GetStageProgram(Maxwell::ShaderProgram program) {
         if (found != precompiled_shaders.end()) {
             shader =
                 std::make_shared<CachedShader>(program_addr, unique_identifier, program, disk_cache,
-                                               precompiled_programs, found->second);
+                                               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));
+                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 guest_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 guest_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 guest_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 guest_addr{};
     u64 unique_identifier{};
     Maxwell::ShaderProgram program_type{};
     ShaderDiskCacheOpenGL& disk_cache;

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -5,7 +5,9 @@
 #include <array>
 #include <string>
 #include <string_view>
+#include <utility>
 #include <variant>
+#include <vector>
 
 #include <fmt/format.h>
 
@@ -717,7 +719,7 @@ private:
     }
 
     std::string GenerateTexture(Operation operation, const std::string& func,
-                                bool is_extra_int = false) {
+                                const std::vector<std::pair<Type, Node>>& extras) {
         constexpr std::array<const char*, 4> coord_constructors = {"float", "vec2", "vec3", "vec4"};
 
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
@@ -738,36 +740,47 @@ private:
             expr += Visit(operation[i]);
 
             const std::size_t next = i + 1;
-            if (next < count || has_array || has_shadow)
+            if (next < count)
                 expr += ", ";
         }
         if (has_array) {
-            expr += "float(ftoi(" + Visit(meta->array) + "))";
+            expr += ", float(ftoi(" + Visit(meta->array) + "))";
         }
         if (has_shadow) {
-            if (has_array)
-                expr += ", ";
-            expr += Visit(meta->depth_compare);
+            expr += ", " + Visit(meta->depth_compare);
         }
         expr += ')';
 
-        for (const Node extra : meta->extras) {
+        for (const auto& extra_pair : extras) {
+            const auto [type, operand] = extra_pair;
+            if (operand == nullptr) {
+                continue;
+            }
             expr += ", ";
-            if (is_extra_int) {
-                if (const auto immediate = std::get_if<ImmediateNode>(extra)) {
+
+            switch (type) {
+            case Type::Int:
+                if (const auto immediate = std::get_if<ImmediateNode>(operand)) {
                     // Inline the string as an immediate integer in GLSL (some extra arguments are
                     // required to be constant)
                     expr += std::to_string(static_cast<s32>(immediate->GetValue()));
                 } else {
-                    expr += "ftoi(" + Visit(extra) + ')';
+                    expr += "ftoi(" + Visit(operand) + ')';
                 }
-            } else {
-                expr += Visit(extra);
+                break;
+            case Type::Float:
+                expr += Visit(operand);
+                break;
+            default: {
+                const auto type_int = static_cast<u32>(type);
+                UNIMPLEMENTED_MSG("Unimplemented extra type={}", type_int);
+                expr += '0';
+                break;
+            }
             }
         }
 
-        expr += ')';
-        return expr;
+        return expr + ')';
     }
 
     std::string Assign(Operation operation) {
@@ -1146,7 +1159,7 @@ private:
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
         ASSERT(meta);
 
-        std::string expr = GenerateTexture(operation, "texture");
+        std::string expr = GenerateTexture(operation, "texture", {{Type::Float, meta->bias}});
         if (meta->sampler.IsShadow()) {
             expr = "vec4(" + expr + ')';
         }
@@ -1157,7 +1170,7 @@ private:
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
         ASSERT(meta);
 
-        std::string expr = GenerateTexture(operation, "textureLod");
+        std::string expr = GenerateTexture(operation, "textureLod", {{Type::Float, meta->lod}});
         if (meta->sampler.IsShadow()) {
             expr = "vec4(" + expr + ')';
         }
@@ -1168,7 +1181,8 @@ private:
         const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
         ASSERT(meta);
 
-        return GenerateTexture(operation, "textureGather", !meta->sampler.IsShadow()) +
+        const auto type = meta->sampler.IsShadow() ? Type::Float : Type::Int;
+        return GenerateTexture(operation, "textureGather", {{type, meta->component}}) +
                GetSwizzle(meta->element);
     }
 
@@ -1197,8 +1211,8 @@ private:
         ASSERT(meta);
 
         if (meta->element < 2) {
-            return "itof(int((" + GenerateTexture(operation, "textureQueryLod") + " * vec2(256))" +
-                   GetSwizzle(meta->element) + "))";
+            return "itof(int((" + GenerateTexture(operation, "textureQueryLod", {}) +
+                   " * vec2(256))" + GetSwizzle(meta->element) + "))";
         }
         return "0";
     }
@@ -1224,9 +1238,9 @@ private:
             else if (next < count)
                 expr += ", ";
         }
-        for (std::size_t i = 0; i < meta->extras.size(); ++i) {
+        if (meta->lod) {
             expr += ", ";
-            expr += CastOperand(Visit(meta->extras.at(i)), Type::Int);
+            expr += CastOperand(Visit(meta->lod), Type::Int);
         }
         expr += ')';
 

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));
 
@@ -244,6 +245,21 @@ void RendererOpenGL::InitOpenGLObjects() {
     LoadColorToActiveGLTexture(0, 0, 0, 0, screen_info.texture);
 }
 
+void RendererOpenGL::AddTelemetryFields() {
+    const char* const gl_version{reinterpret_cast<char const*>(glGetString(GL_VERSION))};
+    const char* const gpu_vendor{reinterpret_cast<char const*>(glGetString(GL_VENDOR))};
+    const char* const gpu_model{reinterpret_cast<char const*>(glGetString(GL_RENDERER))};
+
+    LOG_INFO(Render_OpenGL, "GL_VERSION: {}", gl_version);
+    LOG_INFO(Render_OpenGL, "GL_VENDOR: {}", gpu_vendor);
+    LOG_INFO(Render_OpenGL, "GL_RENDERER: {}", gpu_model);
+
+    auto& telemetry_session = system.TelemetrySession();
+    telemetry_session.AddField(Telemetry::FieldType::UserSystem, "GPU_Vendor", gpu_vendor);
+    telemetry_session.AddField(Telemetry::FieldType::UserSystem, "GPU_Model", gpu_model);
+    telemetry_session.AddField(Telemetry::FieldType::UserSystem, "GPU_OpenGL_Version", gl_version);
+}
+
 void RendererOpenGL::CreateRasterizer() {
     if (rasterizer) {
         return;
@@ -466,17 +482,7 @@ bool RendererOpenGL::Init() {
         glDebugMessageCallback(DebugHandler, nullptr);
     }
 
-    const char* gl_version{reinterpret_cast<char const*>(glGetString(GL_VERSION))};
-    const char* gpu_vendor{reinterpret_cast<char const*>(glGetString(GL_VENDOR))};
-    const char* gpu_model{reinterpret_cast<char const*>(glGetString(GL_RENDERER))};
-
-    LOG_INFO(Render_OpenGL, "GL_VERSION: {}", gl_version);
-    LOG_INFO(Render_OpenGL, "GL_VENDOR: {}", gpu_vendor);
-    LOG_INFO(Render_OpenGL, "GL_RENDERER: {}", gpu_model);
-
-    Core::Telemetry().AddField(Telemetry::FieldType::UserSystem, "GPU_Vendor", gpu_vendor);
-    Core::Telemetry().AddField(Telemetry::FieldType::UserSystem, "GPU_Model", gpu_model);
-    Core::Telemetry().AddField(Telemetry::FieldType::UserSystem, "GPU_OpenGL_Version", gl_version);
+    AddTelemetryFields();
 
     if (!GLAD_GL_VERSION_4_3) {
         return false;

src/video_core/renderer_opengl/renderer_opengl.h

@@ -60,6 +60,7 @@ public:
 
 private:
     void InitOpenGLObjects();
+    void AddTelemetryFields();
     void CreateRasterizer();
 
     void ConfigureFramebufferTexture(TextureInfo& texture,

src/video_core/renderer_vulkan/maxwell_to_vk.cpp

@@ -0,0 +1,483 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "common/logging/log.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/maxwell_to_vk.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/surface.h"
+
+namespace Vulkan::MaxwellToVK {
+
+namespace Sampler {
+
+vk::Filter Filter(Tegra::Texture::TextureFilter filter) {
+    switch (filter) {
+    case Tegra::Texture::TextureFilter::Linear:
+        return vk::Filter::eLinear;
+    case Tegra::Texture::TextureFilter::Nearest:
+        return vk::Filter::eNearest;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented sampler filter={}", static_cast<u32>(filter));
+    return {};
+}
+
+vk::SamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filter) {
+    switch (mipmap_filter) {
+    case Tegra::Texture::TextureMipmapFilter::None:
+        // TODO(Rodrigo): None seems to be mapped to OpenGL's mag and min filters without mipmapping
+        // (e.g. GL_NEAREST and GL_LINEAR). Vulkan doesn't have such a thing, find out if we have to
+        // use an image view with a single mipmap level to emulate this.
+        return vk::SamplerMipmapMode::eLinear;
+    case Tegra::Texture::TextureMipmapFilter::Linear:
+        return vk::SamplerMipmapMode::eLinear;
+    case Tegra::Texture::TextureMipmapFilter::Nearest:
+        return vk::SamplerMipmapMode::eNearest;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented sampler mipmap mode={}", static_cast<u32>(mipmap_filter));
+    return {};
+}
+
+vk::SamplerAddressMode WrapMode(Tegra::Texture::WrapMode wrap_mode) {
+    switch (wrap_mode) {
+    case Tegra::Texture::WrapMode::Wrap:
+        return vk::SamplerAddressMode::eRepeat;
+    case Tegra::Texture::WrapMode::Mirror:
+        return vk::SamplerAddressMode::eMirroredRepeat;
+    case Tegra::Texture::WrapMode::ClampToEdge:
+        return vk::SamplerAddressMode::eClampToEdge;
+    case Tegra::Texture::WrapMode::Border:
+        return vk::SamplerAddressMode::eClampToBorder;
+    case Tegra::Texture::WrapMode::ClampOGL:
+        // TODO(Rodrigo): GL_CLAMP was removed as of OpenGL 3.1, to implement GL_CLAMP, we can use
+        // eClampToBorder to get the border color of the texture, and then sample the edge to
+        // manually mix them. However the shader part of this is not yet implemented.
+        return vk::SamplerAddressMode::eClampToBorder;
+    case Tegra::Texture::WrapMode::MirrorOnceClampToEdge:
+        return vk::SamplerAddressMode::eMirrorClampToEdge;
+    case Tegra::Texture::WrapMode::MirrorOnceBorder:
+        UNIMPLEMENTED();
+        return vk::SamplerAddressMode::eMirrorClampToEdge;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented wrap mode={}", static_cast<u32>(wrap_mode));
+    return {};
+}
+
+vk::CompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_func) {
+    switch (depth_compare_func) {
+    case Tegra::Texture::DepthCompareFunc::Never:
+        return vk::CompareOp::eNever;
+    case Tegra::Texture::DepthCompareFunc::Less:
+        return vk::CompareOp::eLess;
+    case Tegra::Texture::DepthCompareFunc::LessEqual:
+        return vk::CompareOp::eLessOrEqual;
+    case Tegra::Texture::DepthCompareFunc::Equal:
+        return vk::CompareOp::eEqual;
+    case Tegra::Texture::DepthCompareFunc::NotEqual:
+        return vk::CompareOp::eNotEqual;
+    case Tegra::Texture::DepthCompareFunc::Greater:
+        return vk::CompareOp::eGreater;
+    case Tegra::Texture::DepthCompareFunc::GreaterEqual:
+        return vk::CompareOp::eGreaterOrEqual;
+    case Tegra::Texture::DepthCompareFunc::Always:
+        return vk::CompareOp::eAlways;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented sampler depth compare function={}",
+                      static_cast<u32>(depth_compare_func));
+    return {};
+}
+
+} // namespace Sampler
+
+struct FormatTuple {
+    vk::Format format;            ///< Vulkan format
+    ComponentType component_type; ///< Abstracted component type
+    bool attachable;              ///< True when this format can be used as an attachment
+};
+
+static constexpr std::array<FormatTuple, VideoCore::Surface::MaxPixelFormat> tex_format_tuples = {{
+    {vk::Format::eA8B8G8R8UnormPack32, ComponentType::UNorm, true},    // ABGR8U
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // ABGR8S
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // ABGR8UI
+    {vk::Format::eB5G6R5UnormPack16, ComponentType::UNorm, false},     // B5G6R5U
+    {vk::Format::eA2B10G10R10UnormPack32, ComponentType::UNorm, true}, // A2B10G10R10U
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // A1B5G5R5U
+    {vk::Format::eR8Unorm, ComponentType::UNorm, true},                // R8U
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // R8UI
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RGBA16F
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RGBA16U
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RGBA16UI
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // R11FG11FB10F
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RGBA32UI
+    {vk::Format::eBc1RgbaUnormBlock, ComponentType::UNorm, false},     // DXT1
+    {vk::Format::eBc2UnormBlock, ComponentType::UNorm, false},         // DXT23
+    {vk::Format::eBc3UnormBlock, ComponentType::UNorm, false},         // DXT45
+    {vk::Format::eBc4UnormBlock, ComponentType::UNorm, false},         // DXN1
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // DXN2UNORM
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // DXN2SNORM
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // BC7U
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // BC6H_UF16
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // BC6H_SF16
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // ASTC_2D_4X4
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // BGRA8
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RGBA32F
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RG32F
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // R32F
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // R16F
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // R16U
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // R16S
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // R16UI
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // R16I
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RG16
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RG16F
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RG16UI
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RG16I
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RG16S
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RGB32F
+    {vk::Format::eA8B8G8R8SrgbPack32, ComponentType::UNorm, true},     // RGBA8_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RG8U
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RG8S
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // RG32UI
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // R32UI
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // ASTC_2D_8X8
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // ASTC_2D_8X5
+    {vk::Format::eUndefined, ComponentType::Invalid, false},           // ASTC_2D_5X4
+
+    // Compressed sRGB formats
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // BGRA8_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // DXT1_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // DXT23_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // DXT45_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // BC7U_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // ASTC_2D_4X4_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // ASTC_2D_8X8_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // ASTC_2D_8X5_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // ASTC_2D_5X4_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // ASTC_2D_5X5
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // ASTC_2D_5X5_SRGB
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // ASTC_2D_10X8
+    {vk::Format::eUndefined, ComponentType::Invalid, false}, // ASTC_2D_10X8_SRGB
+
+    // Depth formats
+    {vk::Format::eD32Sfloat, ComponentType::Float, true}, // Z32F
+    {vk::Format::eD16Unorm, ComponentType::UNorm, true},  // Z16
+
+    // DepthStencil formats
+    {vk::Format::eD24UnormS8Uint, ComponentType::UNorm, true}, // Z24S8
+    {vk::Format::eD24UnormS8Uint, ComponentType::UNorm, true}, // S8Z24 (emulated)
+    {vk::Format::eUndefined, ComponentType::Invalid, false},   // Z32FS8
+}};
+
+static constexpr bool IsZetaFormat(PixelFormat pixel_format) {
+    return pixel_format >= PixelFormat::MaxColorFormat &&
+           pixel_format < PixelFormat::MaxDepthStencilFormat;
+}
+
+std::pair<vk::Format, bool> SurfaceFormat(const VKDevice& device, FormatType format_type,
+                                          PixelFormat pixel_format, ComponentType component_type) {
+    ASSERT(static_cast<std::size_t>(pixel_format) < tex_format_tuples.size());
+
+    const auto tuple = tex_format_tuples[static_cast<u32>(pixel_format)];
+    UNIMPLEMENTED_IF_MSG(tuple.format == vk::Format::eUndefined,
+                         "Unimplemented texture format with pixel format={} and component type={}",
+                         static_cast<u32>(pixel_format), static_cast<u32>(component_type));
+    ASSERT_MSG(component_type == tuple.component_type, "Component type mismatch");
+
+    auto usage = vk::FormatFeatureFlagBits::eSampledImage |
+                 vk::FormatFeatureFlagBits::eTransferDst | vk::FormatFeatureFlagBits::eTransferSrc;
+    if (tuple.attachable) {
+        usage |= IsZetaFormat(pixel_format) ? vk::FormatFeatureFlagBits::eDepthStencilAttachment
+                                            : vk::FormatFeatureFlagBits::eColorAttachment;
+    }
+    return {device.GetSupportedFormat(tuple.format, usage, format_type), tuple.attachable};
+}
+
+vk::ShaderStageFlagBits ShaderStage(Maxwell::ShaderStage stage) {
+    switch (stage) {
+    case Maxwell::ShaderStage::Vertex:
+        return vk::ShaderStageFlagBits::eVertex;
+    case Maxwell::ShaderStage::TesselationControl:
+        return vk::ShaderStageFlagBits::eTessellationControl;
+    case Maxwell::ShaderStage::TesselationEval:
+        return vk::ShaderStageFlagBits::eTessellationEvaluation;
+    case Maxwell::ShaderStage::Geometry:
+        return vk::ShaderStageFlagBits::eGeometry;
+    case Maxwell::ShaderStage::Fragment:
+        return vk::ShaderStageFlagBits::eFragment;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented shader stage={}", static_cast<u32>(stage));
+    return {};
+}
+
+vk::PrimitiveTopology PrimitiveTopology(Maxwell::PrimitiveTopology topology) {
+    switch (topology) {
+    case Maxwell::PrimitiveTopology::Points:
+        return vk::PrimitiveTopology::ePointList;
+    case Maxwell::PrimitiveTopology::Lines:
+        return vk::PrimitiveTopology::eLineList;
+    case Maxwell::PrimitiveTopology::LineStrip:
+        return vk::PrimitiveTopology::eLineStrip;
+    case Maxwell::PrimitiveTopology::Triangles:
+        return vk::PrimitiveTopology::eTriangleList;
+    case Maxwell::PrimitiveTopology::TriangleStrip:
+        return vk::PrimitiveTopology::eTriangleStrip;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented topology={}", static_cast<u32>(topology));
+    return {};
+}
+
+vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size) {
+    switch (type) {
+    case Maxwell::VertexAttribute::Type::SignedNorm:
+        break;
+    case Maxwell::VertexAttribute::Type::UnsignedNorm:
+        switch (size) {
+        case Maxwell::VertexAttribute::Size::Size_8_8_8_8:
+            return vk::Format::eR8G8B8A8Unorm;
+        default:
+            break;
+        }
+        break;
+    case Maxwell::VertexAttribute::Type::SignedInt:
+        break;
+    case Maxwell::VertexAttribute::Type::UnsignedInt:
+        switch (size) {
+        case Maxwell::VertexAttribute::Size::Size_32:
+            return vk::Format::eR32Uint;
+        default:
+            break;
+        }
+    case Maxwell::VertexAttribute::Type::UnsignedScaled:
+    case Maxwell::VertexAttribute::Type::SignedScaled:
+        break;
+    case Maxwell::VertexAttribute::Type::Float:
+        switch (size) {
+        case Maxwell::VertexAttribute::Size::Size_32_32_32_32:
+            return vk::Format::eR32G32B32A32Sfloat;
+        case Maxwell::VertexAttribute::Size::Size_32_32_32:
+            return vk::Format::eR32G32B32Sfloat;
+        case Maxwell::VertexAttribute::Size::Size_32_32:
+            return vk::Format::eR32G32Sfloat;
+        case Maxwell::VertexAttribute::Size::Size_32:
+            return vk::Format::eR32Sfloat;
+        default:
+            break;
+        }
+        break;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented vertex format of type={} and size={}", static_cast<u32>(type),
+                      static_cast<u32>(size));
+    return {};
+}
+
+vk::CompareOp ComparisonOp(Maxwell::ComparisonOp comparison) {
+    switch (comparison) {
+    case Maxwell::ComparisonOp::Never:
+    case Maxwell::ComparisonOp::NeverOld:
+        return vk::CompareOp::eNever;
+    case Maxwell::ComparisonOp::Less:
+    case Maxwell::ComparisonOp::LessOld:
+        return vk::CompareOp::eLess;
+    case Maxwell::ComparisonOp::Equal:
+    case Maxwell::ComparisonOp::EqualOld:
+        return vk::CompareOp::eEqual;
+    case Maxwell::ComparisonOp::LessEqual:
+    case Maxwell::ComparisonOp::LessEqualOld:
+        return vk::CompareOp::eLessOrEqual;
+    case Maxwell::ComparisonOp::Greater:
+    case Maxwell::ComparisonOp::GreaterOld:
+        return vk::CompareOp::eGreater;
+    case Maxwell::ComparisonOp::NotEqual:
+    case Maxwell::ComparisonOp::NotEqualOld:
+        return vk::CompareOp::eNotEqual;
+    case Maxwell::ComparisonOp::GreaterEqual:
+    case Maxwell::ComparisonOp::GreaterEqualOld:
+        return vk::CompareOp::eGreaterOrEqual;
+    case Maxwell::ComparisonOp::Always:
+    case Maxwell::ComparisonOp::AlwaysOld:
+        return vk::CompareOp::eAlways;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented comparison op={}", static_cast<u32>(comparison));
+    return {};
+}
+
+vk::IndexType IndexFormat(Maxwell::IndexFormat index_format) {
+    switch (index_format) {
+    case Maxwell::IndexFormat::UnsignedByte:
+        UNIMPLEMENTED_MSG("Vulkan does not support native u8 index format");
+        return vk::IndexType::eUint16;
+    case Maxwell::IndexFormat::UnsignedShort:
+        return vk::IndexType::eUint16;
+    case Maxwell::IndexFormat::UnsignedInt:
+        return vk::IndexType::eUint32;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented index_format={}", static_cast<u32>(index_format));
+    return {};
+}
+
+vk::StencilOp StencilOp(Maxwell::StencilOp stencil_op) {
+    switch (stencil_op) {
+    case Maxwell::StencilOp::Keep:
+    case Maxwell::StencilOp::KeepOGL:
+        return vk::StencilOp::eKeep;
+    case Maxwell::StencilOp::Zero:
+    case Maxwell::StencilOp::ZeroOGL:
+        return vk::StencilOp::eZero;
+    case Maxwell::StencilOp::Replace:
+    case Maxwell::StencilOp::ReplaceOGL:
+        return vk::StencilOp::eReplace;
+    case Maxwell::StencilOp::Incr:
+    case Maxwell::StencilOp::IncrOGL:
+        return vk::StencilOp::eIncrementAndClamp;
+    case Maxwell::StencilOp::Decr:
+    case Maxwell::StencilOp::DecrOGL:
+        return vk::StencilOp::eDecrementAndClamp;
+    case Maxwell::StencilOp::Invert:
+    case Maxwell::StencilOp::InvertOGL:
+        return vk::StencilOp::eInvert;
+    case Maxwell::StencilOp::IncrWrap:
+    case Maxwell::StencilOp::IncrWrapOGL:
+        return vk::StencilOp::eIncrementAndWrap;
+    case Maxwell::StencilOp::DecrWrap:
+    case Maxwell::StencilOp::DecrWrapOGL:
+        return vk::StencilOp::eDecrementAndWrap;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented stencil op={}", static_cast<u32>(stencil_op));
+    return {};
+}
+
+vk::BlendOp BlendEquation(Maxwell::Blend::Equation equation) {
+    switch (equation) {
+    case Maxwell::Blend::Equation::Add:
+    case Maxwell::Blend::Equation::AddGL:
+        return vk::BlendOp::eAdd;
+    case Maxwell::Blend::Equation::Subtract:
+    case Maxwell::Blend::Equation::SubtractGL:
+        return vk::BlendOp::eSubtract;
+    case Maxwell::Blend::Equation::ReverseSubtract:
+    case Maxwell::Blend::Equation::ReverseSubtractGL:
+        return vk::BlendOp::eReverseSubtract;
+    case Maxwell::Blend::Equation::Min:
+    case Maxwell::Blend::Equation::MinGL:
+        return vk::BlendOp::eMin;
+    case Maxwell::Blend::Equation::Max:
+    case Maxwell::Blend::Equation::MaxGL:
+        return vk::BlendOp::eMax;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented blend equation={}", static_cast<u32>(equation));
+    return {};
+}
+
+vk::BlendFactor BlendFactor(Maxwell::Blend::Factor factor) {
+    switch (factor) {
+    case Maxwell::Blend::Factor::Zero:
+    case Maxwell::Blend::Factor::ZeroGL:
+        return vk::BlendFactor::eZero;
+    case Maxwell::Blend::Factor::One:
+    case Maxwell::Blend::Factor::OneGL:
+        return vk::BlendFactor::eOne;
+    case Maxwell::Blend::Factor::SourceColor:
+    case Maxwell::Blend::Factor::SourceColorGL:
+        return vk::BlendFactor::eSrcColor;
+    case Maxwell::Blend::Factor::OneMinusSourceColor:
+    case Maxwell::Blend::Factor::OneMinusSourceColorGL:
+        return vk::BlendFactor::eOneMinusSrcColor;
+    case Maxwell::Blend::Factor::SourceAlpha:
+    case Maxwell::Blend::Factor::SourceAlphaGL:
+        return vk::BlendFactor::eSrcAlpha;
+    case Maxwell::Blend::Factor::OneMinusSourceAlpha:
+    case Maxwell::Blend::Factor::OneMinusSourceAlphaGL:
+        return vk::BlendFactor::eOneMinusSrcAlpha;
+    case Maxwell::Blend::Factor::DestAlpha:
+    case Maxwell::Blend::Factor::DestAlphaGL:
+        return vk::BlendFactor::eDstAlpha;
+    case Maxwell::Blend::Factor::OneMinusDestAlpha:
+    case Maxwell::Blend::Factor::OneMinusDestAlphaGL:
+        return vk::BlendFactor::eOneMinusDstAlpha;
+    case Maxwell::Blend::Factor::DestColor:
+    case Maxwell::Blend::Factor::DestColorGL:
+        return vk::BlendFactor::eDstColor;
+    case Maxwell::Blend::Factor::OneMinusDestColor:
+    case Maxwell::Blend::Factor::OneMinusDestColorGL:
+        return vk::BlendFactor::eOneMinusDstColor;
+    case Maxwell::Blend::Factor::SourceAlphaSaturate:
+    case Maxwell::Blend::Factor::SourceAlphaSaturateGL:
+        return vk::BlendFactor::eSrcAlphaSaturate;
+    case Maxwell::Blend::Factor::Source1Color:
+    case Maxwell::Blend::Factor::Source1ColorGL:
+        return vk::BlendFactor::eSrc1Color;
+    case Maxwell::Blend::Factor::OneMinusSource1Color:
+    case Maxwell::Blend::Factor::OneMinusSource1ColorGL:
+        return vk::BlendFactor::eOneMinusSrc1Color;
+    case Maxwell::Blend::Factor::Source1Alpha:
+    case Maxwell::Blend::Factor::Source1AlphaGL:
+        return vk::BlendFactor::eSrc1Alpha;
+    case Maxwell::Blend::Factor::OneMinusSource1Alpha:
+    case Maxwell::Blend::Factor::OneMinusSource1AlphaGL:
+        return vk::BlendFactor::eOneMinusSrc1Alpha;
+    case Maxwell::Blend::Factor::ConstantColor:
+    case Maxwell::Blend::Factor::ConstantColorGL:
+        return vk::BlendFactor::eConstantColor;
+    case Maxwell::Blend::Factor::OneMinusConstantColor:
+    case Maxwell::Blend::Factor::OneMinusConstantColorGL:
+        return vk::BlendFactor::eOneMinusConstantColor;
+    case Maxwell::Blend::Factor::ConstantAlpha:
+    case Maxwell::Blend::Factor::ConstantAlphaGL:
+        return vk::BlendFactor::eConstantAlpha;
+    case Maxwell::Blend::Factor::OneMinusConstantAlpha:
+    case Maxwell::Blend::Factor::OneMinusConstantAlphaGL:
+        return vk::BlendFactor::eOneMinusConstantAlpha;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented blend factor={}", static_cast<u32>(factor));
+    return {};
+}
+
+vk::FrontFace FrontFace(Maxwell::Cull::FrontFace front_face) {
+    switch (front_face) {
+    case Maxwell::Cull::FrontFace::ClockWise:
+        return vk::FrontFace::eClockwise;
+    case Maxwell::Cull::FrontFace::CounterClockWise:
+        return vk::FrontFace::eCounterClockwise;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented front face={}", static_cast<u32>(front_face));
+    return {};
+}
+
+vk::CullModeFlags CullFace(Maxwell::Cull::CullFace cull_face) {
+    switch (cull_face) {
+    case Maxwell::Cull::CullFace::Front:
+        return vk::CullModeFlagBits::eFront;
+    case Maxwell::Cull::CullFace::Back:
+        return vk::CullModeFlagBits::eBack;
+    case Maxwell::Cull::CullFace::FrontAndBack:
+        return vk::CullModeFlagBits::eFrontAndBack;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented cull face={}", static_cast<u32>(cull_face));
+    return {};
+}
+
+vk::ComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle) {
+    switch (swizzle) {
+    case Tegra::Texture::SwizzleSource::Zero:
+        return vk::ComponentSwizzle::eZero;
+    case Tegra::Texture::SwizzleSource::R:
+        return vk::ComponentSwizzle::eR;
+    case Tegra::Texture::SwizzleSource::G:
+        return vk::ComponentSwizzle::eG;
+    case Tegra::Texture::SwizzleSource::B:
+        return vk::ComponentSwizzle::eB;
+    case Tegra::Texture::SwizzleSource::A:
+        return vk::ComponentSwizzle::eA;
+    case Tegra::Texture::SwizzleSource::OneInt:
+    case Tegra::Texture::SwizzleSource::OneFloat:
+        return vk::ComponentSwizzle::eOne;
+    }
+    UNIMPLEMENTED_MSG("Unimplemented swizzle source={}", static_cast<u32>(swizzle));
+    return {};
+}
+
+} // namespace Vulkan::MaxwellToVK

src/video_core/renderer_vulkan/maxwell_to_vk.h

@@ -0,0 +1,58 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <utility>
+#include "common/common_types.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/surface.h"
+#include "video_core/textures/texture.h"
+
+namespace Vulkan::MaxwellToVK {
+
+using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+using PixelFormat = VideoCore::Surface::PixelFormat;
+using ComponentType = VideoCore::Surface::ComponentType;
+
+namespace Sampler {
+
+vk::Filter Filter(Tegra::Texture::TextureFilter filter);
+
+vk::SamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filter);
+
+vk::SamplerAddressMode WrapMode(Tegra::Texture::WrapMode wrap_mode);
+
+vk::CompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_func);
+
+} // namespace Sampler
+
+std::pair<vk::Format, bool> SurfaceFormat(const VKDevice& device, FormatType format_type,
+                                          PixelFormat pixel_format, ComponentType component_type);
+
+vk::ShaderStageFlagBits ShaderStage(Maxwell::ShaderStage stage);
+
+vk::PrimitiveTopology PrimitiveTopology(Maxwell::PrimitiveTopology topology);
+
+vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size);
+
+vk::CompareOp ComparisonOp(Maxwell::ComparisonOp comparison);
+
+vk::IndexType IndexFormat(Maxwell::IndexFormat index_format);
+
+vk::StencilOp StencilOp(Maxwell::StencilOp stencil_op);
+
+vk::BlendOp BlendEquation(Maxwell::Blend::Equation equation);
+
+vk::BlendFactor BlendFactor(Maxwell::Blend::Factor factor);
+
+vk::FrontFace FrontFace(Maxwell::Cull::FrontFace front_face);
+
+vk::CullModeFlags CullFace(Maxwell::Cull::CullFace cull_face);
+
+vk::ComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle);
+
+} // namespace Vulkan::MaxwellToVK

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_device.cpp

@@ -122,8 +122,7 @@ bool VKDevice::IsFormatSupported(vk::Format wanted_format, vk::FormatFeatureFlag
                                  FormatType format_type) const {
     const auto it = format_properties.find(wanted_format);
     if (it == format_properties.end()) {
-        LOG_CRITICAL(Render_Vulkan, "Unimplemented format query={}",
-                     static_cast<u32>(wanted_format));
+        LOG_CRITICAL(Render_Vulkan, "Unimplemented format query={}", vk::to_string(wanted_format));
         UNREACHABLE();
         return true;
     }
@@ -219,11 +218,19 @@ std::map<vk::Format, vk::FormatProperties> VKDevice::GetFormatProperties(
         format_properties.emplace(format, physical.getFormatProperties(format, dldi));
     };
     AddFormatQuery(vk::Format::eA8B8G8R8UnormPack32);
-    AddFormatQuery(vk::Format::eR5G6B5UnormPack16);
+    AddFormatQuery(vk::Format::eB5G6R5UnormPack16);
+    AddFormatQuery(vk::Format::eA2B10G10R10UnormPack32);
+    AddFormatQuery(vk::Format::eR8G8B8A8Srgb);
+    AddFormatQuery(vk::Format::eR8Unorm);
     AddFormatQuery(vk::Format::eD32Sfloat);
+    AddFormatQuery(vk::Format::eD16Unorm);
     AddFormatQuery(vk::Format::eD16UnormS8Uint);
     AddFormatQuery(vk::Format::eD24UnormS8Uint);
     AddFormatQuery(vk::Format::eD32SfloatS8Uint);
+    AddFormatQuery(vk::Format::eBc1RgbaUnormBlock);
+    AddFormatQuery(vk::Format::eBc2UnormBlock);
+    AddFormatQuery(vk::Format::eBc3UnormBlock);
+    AddFormatQuery(vk::Format::eBc4UnormBlock);
 
     return format_properties;
 }

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/shader/decode.cpp

@@ -165,6 +165,7 @@ u32 ShaderIR::DecodeInstr(NodeBlock& bb, u32 pc) {
         {OpCode::Type::Hfma2, &ShaderIR::DecodeHfma2},
         {OpCode::Type::Conversion, &ShaderIR::DecodeConversion},
         {OpCode::Type::Memory, &ShaderIR::DecodeMemory},
+        {OpCode::Type::Texture, &ShaderIR::DecodeTexture},
         {OpCode::Type::FloatSetPredicate, &ShaderIR::DecodeFloatSetPredicate},
         {OpCode::Type::IntegerSetPredicate, &ShaderIR::DecodeIntegerSetPredicate},
         {OpCode::Type::HalfSetPredicate, &ShaderIR::DecodeHalfSetPredicate},

src/video_core/shader/decode/memory.cpp

@@ -17,24 +17,6 @@ using Tegra::Shader::Attribute;
 using Tegra::Shader::Instruction;
 using Tegra::Shader::OpCode;
 using Tegra::Shader::Register;
-using Tegra::Shader::TextureMiscMode;
-using Tegra::Shader::TextureProcessMode;
-using Tegra::Shader::TextureType;
-
-static std::size_t GetCoordCount(TextureType texture_type) {
-    switch (texture_type) {
-    case TextureType::Texture1D:
-        return 1;
-    case TextureType::Texture2D:
-        return 2;
-    case TextureType::Texture3D:
-    case TextureType::TextureCube:
-        return 3;
-    default:
-        UNIMPLEMENTED_MSG("Unhandled texture type: {}", static_cast<u32>(texture_type));
-        return 0;
-    }
-}
 
 u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
     const Instruction instr = {program_code[pc]};
@@ -247,194 +229,6 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
         }
         break;
     }
-    case OpCode::Id::TEX: {
-        UNIMPLEMENTED_IF_MSG(instr.tex.UsesMiscMode(TextureMiscMode::AOFFI),
-                             "AOFFI is not implemented");
-
-        if (instr.tex.UsesMiscMode(TextureMiscMode::NODEP)) {
-            LOG_WARNING(HW_GPU, "TEX.NODEP implementation is incomplete");
-        }
-
-        const TextureType texture_type{instr.tex.texture_type};
-        const bool is_array = instr.tex.array != 0;
-        const bool depth_compare = instr.tex.UsesMiscMode(TextureMiscMode::DC);
-        const auto process_mode = instr.tex.GetTextureProcessMode();
-        WriteTexInstructionFloat(
-            bb, instr, GetTexCode(instr, texture_type, process_mode, depth_compare, is_array));
-        break;
-    }
-    case OpCode::Id::TEXS: {
-        const TextureType texture_type{instr.texs.GetTextureType()};
-        const bool is_array{instr.texs.IsArrayTexture()};
-        const bool depth_compare = instr.texs.UsesMiscMode(TextureMiscMode::DC);
-        const auto process_mode = instr.texs.GetTextureProcessMode();
-
-        if (instr.texs.UsesMiscMode(TextureMiscMode::NODEP)) {
-            LOG_WARNING(HW_GPU, "TEXS.NODEP implementation is incomplete");
-        }
-
-        const Node4 components =
-            GetTexsCode(instr, texture_type, process_mode, depth_compare, is_array);
-
-        if (instr.texs.fp32_flag) {
-            WriteTexsInstructionFloat(bb, instr, components);
-        } else {
-            WriteTexsInstructionHalfFloat(bb, instr, components);
-        }
-        break;
-    }
-    case OpCode::Id::TLD4: {
-        ASSERT(instr.tld4.array == 0);
-        UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(TextureMiscMode::AOFFI),
-                             "AOFFI is not implemented");
-        UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(TextureMiscMode::NDV),
-                             "NDV is not implemented");
-        UNIMPLEMENTED_IF_MSG(instr.tld4.UsesMiscMode(TextureMiscMode::PTP),
-                             "PTP is not implemented");
-
-        if (instr.tld4.UsesMiscMode(TextureMiscMode::NODEP)) {
-            LOG_WARNING(HW_GPU, "TLD4.NODEP implementation is incomplete");
-        }
-
-        const auto texture_type = instr.tld4.texture_type.Value();
-        const bool depth_compare = instr.tld4.UsesMiscMode(TextureMiscMode::DC);
-        const bool is_array = instr.tld4.array != 0;
-        WriteTexInstructionFloat(bb, instr,
-                                 GetTld4Code(instr, texture_type, depth_compare, is_array));
-        break;
-    }
-    case OpCode::Id::TLD4S: {
-        UNIMPLEMENTED_IF_MSG(instr.tld4s.UsesMiscMode(TextureMiscMode::AOFFI),
-                             "AOFFI is not implemented");
-        if (instr.tld4s.UsesMiscMode(TextureMiscMode::NODEP)) {
-            LOG_WARNING(HW_GPU, "TLD4S.NODEP implementation is incomplete");
-        }
-
-        const bool depth_compare = instr.tld4s.UsesMiscMode(TextureMiscMode::DC);
-        const Node op_a = GetRegister(instr.gpr8);
-        const Node op_b = GetRegister(instr.gpr20);
-
-        // TODO(Subv): Figure out how the sampler type is encoded in the TLD4S instruction.
-        std::vector<Node> coords;
-        if (depth_compare) {
-            // Note: TLD4S coordinate encoding works just like TEXS's
-            const Node op_y = GetRegister(instr.gpr8.Value() + 1);
-            coords.push_back(op_a);
-            coords.push_back(op_y);
-            coords.push_back(op_b);
-        } else {
-            coords.push_back(op_a);
-            coords.push_back(op_b);
-        }
-        std::vector<Node> extras;
-        extras.push_back(Immediate(static_cast<u32>(instr.tld4s.component)));
-
-        const auto& sampler =
-            GetSampler(instr.sampler, TextureType::Texture2D, false, depth_compare);
-
-        Node4 values;
-        for (u32 element = 0; element < values.size(); ++element) {
-            auto coords_copy = coords;
-            MetaTexture meta{sampler, {}, {}, extras, element};
-            values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
-        }
-
-        WriteTexsInstructionFloat(bb, instr, values);
-        break;
-    }
-    case OpCode::Id::TXQ: {
-        if (instr.txq.UsesMiscMode(TextureMiscMode::NODEP)) {
-            LOG_WARNING(HW_GPU, "TXQ.NODEP implementation is incomplete");
-        }
-
-        // TODO: The new commits on the texture refactor, change the way samplers work.
-        // Sadly, not all texture instructions specify the type of texture their sampler
-        // uses. This must be fixed at a later instance.
-        const auto& sampler =
-            GetSampler(instr.sampler, Tegra::Shader::TextureType::Texture2D, false, false);
-
-        u32 indexer = 0;
-        switch (instr.txq.query_type) {
-        case Tegra::Shader::TextureQueryType::Dimension: {
-            for (u32 element = 0; element < 4; ++element) {
-                if (!instr.txq.IsComponentEnabled(element)) {
-                    continue;
-                }
-                MetaTexture meta{sampler, {}, {}, {}, element};
-                const Node value =
-                    Operation(OperationCode::TextureQueryDimensions, meta, GetRegister(instr.gpr8));
-                SetTemporal(bb, indexer++, value);
-            }
-            for (u32 i = 0; i < indexer; ++i) {
-                SetRegister(bb, instr.gpr0.Value() + i, GetTemporal(i));
-            }
-            break;
-        }
-        default:
-            UNIMPLEMENTED_MSG("Unhandled texture query type: {}",
-                              static_cast<u32>(instr.txq.query_type.Value()));
-        }
-        break;
-    }
-    case OpCode::Id::TMML: {
-        UNIMPLEMENTED_IF_MSG(instr.tmml.UsesMiscMode(Tegra::Shader::TextureMiscMode::NDV),
-                             "NDV is not implemented");
-
-        if (instr.tmml.UsesMiscMode(TextureMiscMode::NODEP)) {
-            LOG_WARNING(HW_GPU, "TMML.NODEP implementation is incomplete");
-        }
-
-        auto texture_type = instr.tmml.texture_type.Value();
-        const bool is_array = instr.tmml.array != 0;
-        const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, false);
-
-        std::vector<Node> coords;
-
-        // TODO: Add coordinates for different samplers once other texture types are implemented.
-        switch (texture_type) {
-        case TextureType::Texture1D:
-            coords.push_back(GetRegister(instr.gpr8));
-            break;
-        case TextureType::Texture2D:
-            coords.push_back(GetRegister(instr.gpr8.Value() + 0));
-            coords.push_back(GetRegister(instr.gpr8.Value() + 1));
-            break;
-        default:
-            UNIMPLEMENTED_MSG("Unhandled texture type {}", static_cast<u32>(texture_type));
-
-            // Fallback to interpreting as a 2D texture for now
-            coords.push_back(GetRegister(instr.gpr8.Value() + 0));
-            coords.push_back(GetRegister(instr.gpr8.Value() + 1));
-            texture_type = TextureType::Texture2D;
-        }
-
-        for (u32 element = 0; element < 2; ++element) {
-            auto params = coords;
-            MetaTexture meta{sampler, {}, {}, {}, element};
-            const Node value = Operation(OperationCode::TextureQueryLod, meta, std::move(params));
-            SetTemporal(bb, element, value);
-        }
-        for (u32 element = 0; element < 2; ++element) {
-            SetRegister(bb, instr.gpr0.Value() + element, GetTemporal(element));
-        }
-
-        break;
-    }
-    case OpCode::Id::TLDS: {
-        const Tegra::Shader::TextureType texture_type{instr.tlds.GetTextureType()};
-        const bool is_array{instr.tlds.IsArrayTexture()};
-
-        UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(TextureMiscMode::AOFFI),
-                             "AOFFI is not implemented");
-        UNIMPLEMENTED_IF_MSG(instr.tlds.UsesMiscMode(TextureMiscMode::MZ), "MZ is not implemented");
-
-        if (instr.tlds.UsesMiscMode(TextureMiscMode::NODEP)) {
-            LOG_WARNING(HW_GPU, "TLDS.NODEP implementation is incomplete");
-        }
-
-        WriteTexsInstructionFloat(bb, instr, GetTldsCode(instr, texture_type, is_array));
-        break;
-    }
     default:
         UNIMPLEMENTED_MSG("Unhandled memory instruction: {}", opcode->get().GetName());
     }
@@ -442,291 +236,4 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
     return pc;
 }
 
-const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler, TextureType type,
-                                    bool is_array, bool is_shadow) {
-    const auto offset = static_cast<std::size_t>(sampler.index.Value());
-
-    // If this sampler has already been used, return the existing mapping.
-    const auto itr =
-        std::find_if(used_samplers.begin(), used_samplers.end(),
-                     [&](const Sampler& entry) { return entry.GetOffset() == offset; });
-    if (itr != used_samplers.end()) {
-        ASSERT(itr->GetType() == type && itr->IsArray() == is_array &&
-               itr->IsShadow() == is_shadow);
-        return *itr;
-    }
-
-    // Otherwise create a new mapping for this sampler
-    const std::size_t next_index = used_samplers.size();
-    const Sampler entry{offset, next_index, type, is_array, is_shadow};
-    return *used_samplers.emplace(entry).first;
-}
-
-void ShaderIR::WriteTexInstructionFloat(NodeBlock& bb, Instruction instr, const Node4& components) {
-    u32 dest_elem = 0;
-    for (u32 elem = 0; elem < 4; ++elem) {
-        if (!instr.tex.IsComponentEnabled(elem)) {
-            // Skip disabled components
-            continue;
-        }
-        SetTemporal(bb, dest_elem++, components[elem]);
-    }
-    // After writing values in temporals, move them to the real registers
-    for (u32 i = 0; i < dest_elem; ++i) {
-        SetRegister(bb, instr.gpr0.Value() + i, GetTemporal(i));
-    }
-}
-
-void ShaderIR::WriteTexsInstructionFloat(NodeBlock& bb, Instruction instr,
-                                         const Node4& components) {
-    // TEXS has two destination registers and a swizzle. The first two elements in the swizzle
-    // go into gpr0+0 and gpr0+1, and the rest goes into gpr28+0 and gpr28+1
-
-    u32 dest_elem = 0;
-    for (u32 component = 0; component < 4; ++component) {
-        if (!instr.texs.IsComponentEnabled(component))
-            continue;
-        SetTemporal(bb, dest_elem++, components[component]);
-    }
-
-    for (u32 i = 0; i < dest_elem; ++i) {
-        if (i < 2) {
-            // Write the first two swizzle components to gpr0 and gpr0+1
-            SetRegister(bb, instr.gpr0.Value() + i % 2, GetTemporal(i));
-        } else {
-            ASSERT(instr.texs.HasTwoDestinations());
-            // Write the rest of the swizzle components to gpr28 and gpr28+1
-            SetRegister(bb, instr.gpr28.Value() + i % 2, GetTemporal(i));
-        }
-    }
-}
-
-void ShaderIR::WriteTexsInstructionHalfFloat(NodeBlock& bb, Instruction instr,
-                                             const Node4& components) {
-    // TEXS.F16 destionation registers are packed in two registers in pairs (just like any half
-    // float instruction).
-
-    Node4 values;
-    u32 dest_elem = 0;
-    for (u32 component = 0; component < 4; ++component) {
-        if (!instr.texs.IsComponentEnabled(component))
-            continue;
-        values[dest_elem++] = components[component];
-    }
-    if (dest_elem == 0)
-        return;
-
-    std::generate(values.begin() + dest_elem, values.end(), [&]() { return Immediate(0); });
-
-    const Node first_value = Operation(OperationCode::HPack2, values[0], values[1]);
-    if (dest_elem <= 2) {
-        SetRegister(bb, instr.gpr0, first_value);
-        return;
-    }
-
-    SetTemporal(bb, 0, first_value);
-    SetTemporal(bb, 1, Operation(OperationCode::HPack2, values[2], values[3]));
-
-    SetRegister(bb, instr.gpr0, GetTemporal(0));
-    SetRegister(bb, instr.gpr28, GetTemporal(1));
-}
-
-Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
-                               TextureProcessMode process_mode, std::vector<Node> coords,
-                               Node array, Node depth_compare, u32 bias_offset) {
-    const bool is_array = array;
-    const bool is_shadow = depth_compare;
-
-    UNIMPLEMENTED_IF_MSG((texture_type == TextureType::Texture3D && (is_array || is_shadow)) ||
-                             (texture_type == TextureType::TextureCube && is_array && is_shadow),
-                         "This method is not supported.");
-
-    const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, is_shadow);
-
-    const bool lod_needed = process_mode == TextureProcessMode::LZ ||
-                            process_mode == TextureProcessMode::LL ||
-                            process_mode == TextureProcessMode::LLA;
-
-    // LOD selection (either via bias or explicit textureLod) not supported in GL for
-    // sampler2DArrayShadow and samplerCubeArrayShadow.
-    const bool gl_lod_supported =
-        !((texture_type == Tegra::Shader::TextureType::Texture2D && is_array && is_shadow) ||
-          (texture_type == Tegra::Shader::TextureType::TextureCube && is_array && is_shadow));
-
-    const OperationCode read_method =
-        lod_needed && gl_lod_supported ? OperationCode::TextureLod : OperationCode::Texture;
-
-    UNIMPLEMENTED_IF(process_mode != TextureProcessMode::None && !gl_lod_supported);
-
-    std::vector<Node> extras;
-    if (process_mode != TextureProcessMode::None && gl_lod_supported) {
-        if (process_mode == TextureProcessMode::LZ) {
-            extras.push_back(Immediate(0.0f));
-        } else {
-            // If present, lod or bias are always stored in the register indexed by the gpr20
-            // field with an offset depending on the usage of the other registers
-            extras.push_back(GetRegister(instr.gpr20.Value() + bias_offset));
-        }
-    }
-
-    Node4 values;
-    for (u32 element = 0; element < values.size(); ++element) {
-        auto copy_coords = coords;
-        MetaTexture meta{sampler, array, depth_compare, extras, element};
-        values[element] = Operation(read_method, meta, std::move(copy_coords));
-    }
-
-    return values;
-}
-
-Node4 ShaderIR::GetTexCode(Instruction instr, TextureType texture_type,
-                           TextureProcessMode process_mode, bool depth_compare, bool is_array) {
-    const bool lod_bias_enabled =
-        (process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ);
-
-    const auto [coord_count, total_coord_count] = ValidateAndGetCoordinateElement(
-        texture_type, depth_compare, is_array, lod_bias_enabled, 4, 5);
-    // If enabled arrays index is always stored in the gpr8 field
-    const u64 array_register = instr.gpr8.Value();
-    // First coordinate index is the gpr8 or gpr8 + 1 when arrays are used
-    const u64 coord_register = array_register + (is_array ? 1 : 0);
-
-    std::vector<Node> coords;
-    for (std::size_t i = 0; i < coord_count; ++i) {
-        coords.push_back(GetRegister(coord_register + i));
-    }
-    // 1D.DC in OpenGL the 2nd component is ignored.
-    if (depth_compare && !is_array && texture_type == TextureType::Texture1D) {
-        coords.push_back(Immediate(0.0f));
-    }
-
-    const Node array = is_array ? GetRegister(array_register) : nullptr;
-
-    Node dc{};
-    if (depth_compare) {
-        // Depth is always stored in the register signaled by gpr20 or in the next register if lod
-        // or bias are used
-        const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0);
-        dc = GetRegister(depth_register);
-    }
-
-    return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, 0);
-}
-
-Node4 ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type,
-                            TextureProcessMode process_mode, bool depth_compare, bool is_array) {
-    const bool lod_bias_enabled =
-        (process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ);
-
-    const auto [coord_count, total_coord_count] = ValidateAndGetCoordinateElement(
-        texture_type, depth_compare, is_array, lod_bias_enabled, 4, 4);
-    // If enabled arrays index is always stored in the gpr8 field
-    const u64 array_register = instr.gpr8.Value();
-    // First coordinate index is stored in gpr8 field or (gpr8 + 1) when arrays are used
-    const u64 coord_register = array_register + (is_array ? 1 : 0);
-    const u64 last_coord_register =
-        (is_array || !(lod_bias_enabled || depth_compare) || (coord_count > 2))
-            ? static_cast<u64>(instr.gpr20.Value())
-            : coord_register + 1;
-    const u32 bias_offset = coord_count > 2 ? 1 : 0;
-
-    std::vector<Node> coords;
-    for (std::size_t i = 0; i < coord_count; ++i) {
-        const bool last = (i == (coord_count - 1)) && (coord_count > 1);
-        coords.push_back(GetRegister(last ? last_coord_register : coord_register + i));
-    }
-
-    const Node array = is_array ? GetRegister(array_register) : nullptr;
-
-    Node dc{};
-    if (depth_compare) {
-        // Depth is always stored in the register signaled by gpr20 or in the next register if lod
-        // or bias are used
-        const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0);
-        dc = GetRegister(depth_register);
-    }
-
-    return GetTextureCode(instr, texture_type, process_mode, coords, array, dc, bias_offset);
-}
-
-Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool depth_compare,
-                            bool is_array) {
-    const std::size_t coord_count = GetCoordCount(texture_type);
-    const std::size_t total_coord_count = coord_count + (is_array ? 1 : 0);
-    const std::size_t total_reg_count = total_coord_count + (depth_compare ? 1 : 0);
-
-    // If enabled arrays index is always stored in the gpr8 field
-    const u64 array_register = instr.gpr8.Value();
-    // First coordinate index is the gpr8 or gpr8 + 1 when arrays are used
-    const u64 coord_register = array_register + (is_array ? 1 : 0);
-
-    std::vector<Node> coords;
-    for (size_t i = 0; i < coord_count; ++i)
-        coords.push_back(GetRegister(coord_register + i));
-
-    const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, depth_compare);
-
-    Node4 values;
-    for (u32 element = 0; element < values.size(); ++element) {
-        auto coords_copy = coords;
-        MetaTexture meta{sampler, GetRegister(array_register), {}, {}, element};
-        values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
-    }
-
-    return values;
-}
-
-Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is_array) {
-    const std::size_t type_coord_count = GetCoordCount(texture_type);
-    const bool lod_enabled = instr.tlds.GetTextureProcessMode() == TextureProcessMode::LL;
-
-    // If enabled arrays index is always stored in the gpr8 field
-    const u64 array_register = instr.gpr8.Value();
-    // if is array gpr20 is used
-    const u64 coord_register = is_array ? instr.gpr20.Value() : instr.gpr8.Value();
-
-    const u64 last_coord_register =
-        ((type_coord_count > 2) || (type_coord_count == 2 && !lod_enabled)) && !is_array
-            ? static_cast<u64>(instr.gpr20.Value())
-            : coord_register + 1;
-
-    std::vector<Node> coords;
-    for (std::size_t i = 0; i < type_coord_count; ++i) {
-        const bool last = (i == (type_coord_count - 1)) && (type_coord_count > 1);
-        coords.push_back(GetRegister(last ? last_coord_register : coord_register + i));
-    }
-
-    const Node array = is_array ? GetRegister(array_register) : nullptr;
-    // When lod is used always is in gpr20
-    const Node lod = lod_enabled ? GetRegister(instr.gpr20) : Immediate(0);
-
-    const auto& sampler = GetSampler(instr.sampler, texture_type, is_array, false);
-
-    Node4 values;
-    for (u32 element = 0; element < values.size(); ++element) {
-        auto coords_copy = coords;
-        MetaTexture meta{sampler, array, {}, {lod}, element};
-        values[element] = Operation(OperationCode::TexelFetch, meta, std::move(coords_copy));
-    }
-    return values;
-}
-
-std::tuple<std::size_t, std::size_t> ShaderIR::ValidateAndGetCoordinateElement(
-    TextureType texture_type, bool depth_compare, bool is_array, bool lod_bias_enabled,
-    std::size_t max_coords, std::size_t max_inputs) {
-    const std::size_t coord_count = GetCoordCount(texture_type);
-
-    std::size_t total_coord_count = coord_count + (is_array ? 1 : 0) + (depth_compare ? 1 : 0);
-    const std::size_t total_reg_count = total_coord_count + (lod_bias_enabled ? 1 : 0);
-    if (total_coord_count > max_coords || total_reg_count > max_inputs) {
-        UNIMPLEMENTED_MSG("Unsupported Texture operation");
-        total_coord_count = std::min(total_coord_count, max_coords);
-    }
-    // 1D.DC OpenGL is using a vec3 but 2nd component is ignored later.
-    total_coord_count +=
-        (depth_compare && !is_array && texture_type == TextureType::Texture1D) ? 1 : 0;
-
-    return {coord_count, total_coord_count};
-}
-
 } // namespace VideoCommon::Shader