Do some corrections in conversion shader instructions.

Corrects encodings for I2F, F2F, I2I and F2I
Implements Immediate variants of all four conversion types.
Add assertions to unimplemented stuffs.

.gitmodules

@@ -43,3 +43,6 @@
 [submodule "externals/zstd"]
     path = externals/zstd
     url = https://github.com/facebook/zstd
+[submodule "sirit"]
+    path = externals/sirit
+    url = https://github.com/ReinUsesLisp/sirit

externals/CMakeLists.txt

@@ -72,6 +72,11 @@ if (USE_DISCORD_PRESENCE)
     target_include_directories(discord-rpc INTERFACE ./discord-rpc/include)
 endif()
 
+# Sirit
+if (ENABLE_VULKAN)
+    add_subdirectory(sirit)
+endif()
+
 if (ENABLE_WEB_SERVICE)
     # LibreSSL
     set(LIBRESSL_SKIP_INSTALL ON CACHE BOOL "")

src/common/scope_exit.h

@@ -20,7 +20,7 @@ struct ScopeExitHelper {
 
 template <typename Func>
 ScopeExitHelper<Func> ScopeExit(Func&& func) {
-    return ScopeExitHelper<Func>(std::move(func));
+    return ScopeExitHelper<Func>(std::forward<Func>(func));
 }
 } // namespace detail
 

src/common/swap.h

@@ -21,11 +21,6 @@
 
 #if defined(_MSC_VER)
 #include <cstdlib>
-#elif defined(__linux__)
-#include <byteswap.h>
-#elif defined(__Bitrig__) || defined(__DragonFly__) || defined(__FreeBSD__) ||                     \
-    defined(__NetBSD__) || defined(__OpenBSD__)
-#include <sys/endian.h>
 #endif
 #include <cstring>
 #include "common/common_types.h"
@@ -62,86 +57,49 @@
 namespace Common {
 
 #ifdef _MSC_VER
-inline u16 swap16(u16 _data) {
-    return _byteswap_ushort(_data);
+[[nodiscard]] inline u16 swap16(u16 data) noexcept {
+    return _byteswap_ushort(data);
 }
-inline u32 swap32(u32 _data) {
-    return _byteswap_ulong(_data);
+[[nodiscard]] inline u32 swap32(u32 data) noexcept {
+    return _byteswap_ulong(data);
 }
-inline u64 swap64(u64 _data) {
-    return _byteswap_uint64(_data);
+[[nodiscard]] inline u64 swap64(u64 data) noexcept {
+    return _byteswap_uint64(data);
 }
-#elif defined(ARCHITECTURE_ARM) && (__ARM_ARCH >= 6)
-inline u16 swap16(u16 _data) {
-    u32 data = _data;
-    __asm__("rev16 %0, %1\n" : "=l"(data) : "l"(data));
-    return (u16)data;
-}
-inline u32 swap32(u32 _data) {
-    __asm__("rev %0, %1\n" : "=l"(_data) : "l"(_data));
-    return _data;
-}
-inline u64 swap64(u64 _data) {
-    return ((u64)swap32(_data) << 32) | swap32(_data >> 32);
-}
-#elif __linux__
-inline u16 swap16(u16 _data) {
-    return bswap_16(_data);
-}
-inline u32 swap32(u32 _data) {
-    return bswap_32(_data);
-}
-inline u64 swap64(u64 _data) {
-    return bswap_64(_data);
-}
-#elif __APPLE__
-inline __attribute__((always_inline)) u16 swap16(u16 _data) {
-    return (_data >> 8) | (_data << 8);
-}
-inline __attribute__((always_inline)) u32 swap32(u32 _data) {
-    return __builtin_bswap32(_data);
-}
-inline __attribute__((always_inline)) u64 swap64(u64 _data) {
-    return __builtin_bswap64(_data);
-}
-#elif defined(__Bitrig__) || defined(__OpenBSD__)
+#elif defined(__clang__) || defined(__GNUC__)
+#if defined(__Bitrig__) || defined(__OpenBSD__)
 // redefine swap16, swap32, swap64 as inline functions
 #undef swap16
 #undef swap32
 #undef swap64
-inline u16 swap16(u16 _data) {
-    return __swap16(_data);
+#endif
+[[nodiscard]] inline u16 swap16(u16 data) noexcept {
+    return __builtin_bswap16(data);
 }
-inline u32 swap32(u32 _data) {
-    return __swap32(_data);
+[[nodiscard]] inline u32 swap32(u32 data) noexcept {
+    return __builtin_bswap32(data);
 }
-inline u64 swap64(u64 _data) {
-    return __swap64(_data);
-}
-#elif defined(__DragonFly__) || defined(__FreeBSD__) || defined(__NetBSD__)
-inline u16 swap16(u16 _data) {
-    return bswap16(_data);
-}
-inline u32 swap32(u32 _data) {
-    return bswap32(_data);
-}
-inline u64 swap64(u64 _data) {
-    return bswap64(_data);
+[[nodiscard]] inline u64 swap64(u64 data) noexcept {
+    return __builtin_bswap64(data);
 }
 #else
-// Slow generic implementation.
-inline u16 swap16(u16 data) {
+// Generic implementation.
+[[nodiscard]] inline u16 swap16(u16 data) noexcept {
     return (data >> 8) | (data << 8);
 }
-inline u32 swap32(u32 data) {
-    return (swap16(data) << 16) | swap16(data >> 16);
+[[nodiscard]] inline u32 swap32(u32 data) noexcept {
+    return ((data & 0xFF000000U) >> 24) | ((data & 0x00FF0000U) >> 8) |
+           ((data & 0x0000FF00U) << 8) | ((data & 0x000000FFU) << 24);
 }
-inline u64 swap64(u64 data) {
-    return ((u64)swap32(data) << 32) | swap32(data >> 32);
+[[nodiscard]] inline u64 swap64(u64 data) noexcept {
+    return ((data & 0xFF00000000000000ULL) >> 56) | ((data & 0x00FF000000000000ULL) >> 40) |
+           ((data & 0x0000FF0000000000ULL) >> 24) | ((data & 0x000000FF00000000ULL) >> 8) |
+           ((data & 0x00000000FF000000ULL) << 8) | ((data & 0x0000000000FF0000ULL) << 24) |
+           ((data & 0x000000000000FF00ULL) << 40) | ((data & 0x00000000000000FFULL) << 56);
 }
 #endif
 
-inline float swapf(float f) {
+[[nodiscard]] inline float swapf(float f) noexcept {
     static_assert(sizeof(u32) == sizeof(float), "float must be the same size as uint32_t.");
 
     u32 value;
@@ -153,7 +111,7 @@ inline float swapf(float f) {
     return f;
 }
 
-inline double swapd(double f) {
+[[nodiscard]] inline double swapd(double f) noexcept {
     static_assert(sizeof(u64) == sizeof(double), "double must be the same size as uint64_t.");
 
     u64 value;

src/core/arm/dynarmic/arm_dynarmic.cpp

@@ -14,6 +14,7 @@
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
 #include "core/gdbstub/gdbstub.h"
+#include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/svc.h"
 #include "core/hle/kernel/vm_manager.h"
@@ -99,7 +100,7 @@ public:
     }
 
     void CallSVC(u32 swi) override {
-        Kernel::CallSVC(swi);
+        Kernel::CallSVC(parent.system, swi);
     }
 
     void AddTicks(u64 ticks) override {
@@ -112,14 +113,14 @@ public:
         // Always execute at least one tick.
         amortized_ticks = std::max<u64>(amortized_ticks, 1);
 
-        parent.core_timing.AddTicks(amortized_ticks);
+        parent.system.CoreTiming().AddTicks(amortized_ticks);
         num_interpreted_instructions = 0;
     }
     u64 GetTicksRemaining() override {
-        return std::max(parent.core_timing.GetDowncount(), 0);
+        return std::max(parent.system.CoreTiming().GetDowncount(), 0);
     }
     u64 GetCNTPCT() override {
-        return Timing::CpuCyclesToClockCycles(parent.core_timing.GetTicks());
+        return Timing::CpuCyclesToClockCycles(parent.system.CoreTiming().GetTicks());
     }
 
     ARM_Dynarmic& parent;
@@ -129,7 +130,7 @@ public:
 };
 
 std::unique_ptr<Dynarmic::A64::Jit> ARM_Dynarmic::MakeJit() const {
-    auto* current_process = Core::CurrentProcess();
+    auto* current_process = system.Kernel().CurrentProcess();
     auto** const page_table = current_process->VMManager().page_table.pointers.data();
 
     Dynarmic::A64::UserConfig config;
@@ -171,10 +172,10 @@ void ARM_Dynarmic::Step() {
     cb->InterpreterFallback(jit->GetPC(), 1);
 }
 
-ARM_Dynarmic::ARM_Dynarmic(Timing::CoreTiming& core_timing, ExclusiveMonitor& exclusive_monitor,
+ARM_Dynarmic::ARM_Dynarmic(System& system, ExclusiveMonitor& exclusive_monitor,
                            std::size_t core_index)
-    : cb(std::make_unique<ARM_Dynarmic_Callbacks>(*this)), inner_unicorn{core_timing},
-      core_index{core_index}, core_timing{core_timing},
+    : cb(std::make_unique<ARM_Dynarmic_Callbacks>(*this)), inner_unicorn{system},
+      core_index{core_index}, system{system},
       exclusive_monitor{dynamic_cast<DynarmicExclusiveMonitor&>(exclusive_monitor)} {
     ThreadContext ctx{};
     inner_unicorn.SaveContext(ctx);

src/core/arm/dynarmic/arm_dynarmic.h

@@ -12,19 +12,15 @@
 #include "core/arm/exclusive_monitor.h"
 #include "core/arm/unicorn/arm_unicorn.h"
 
-namespace Core::Timing {
-class CoreTiming;
-}
-
 namespace Core {
 
 class ARM_Dynarmic_Callbacks;
 class DynarmicExclusiveMonitor;
+class System;
 
 class ARM_Dynarmic final : public ARM_Interface {
 public:
-    ARM_Dynarmic(Timing::CoreTiming& core_timing, ExclusiveMonitor& exclusive_monitor,
-                 std::size_t core_index);
+    ARM_Dynarmic(System& system, ExclusiveMonitor& exclusive_monitor, std::size_t core_index);
     ~ARM_Dynarmic() override;
 
     void MapBackingMemory(VAddr address, std::size_t size, u8* memory,
@@ -63,7 +59,7 @@ private:
     ARM_Unicorn inner_unicorn;
 
     std::size_t core_index;
-    Timing::CoreTiming& core_timing;
+    System& system;
     DynarmicExclusiveMonitor& exclusive_monitor;
 };
 

src/core/arm/unicorn/arm_unicorn.cpp

@@ -10,7 +10,6 @@
 #include "core/core.h"
 #include "core/core_timing.h"
 #include "core/hle/kernel/svc.h"
-#include "core/memory.h"
 
 namespace Core {
 
@@ -49,20 +48,6 @@ static void CodeHook(uc_engine* uc, uint64_t address, uint32_t size, void* user_
     }
 }
 
-static void InterruptHook(uc_engine* uc, u32 intNo, void* user_data) {
-    u32 esr{};
-    CHECKED(uc_reg_read(uc, UC_ARM64_REG_ESR, &esr));
-
-    auto ec = esr >> 26;
-    auto iss = esr & 0xFFFFFF;
-
-    switch (ec) {
-    case 0x15: // SVC
-        Kernel::CallSVC(iss);
-        break;
-    }
-}
-
 static bool UnmappedMemoryHook(uc_engine* uc, uc_mem_type type, u64 addr, int size, u64 value,
                                void* user_data) {
     ARM_Interface::ThreadContext ctx{};
@@ -72,7 +57,7 @@ static bool UnmappedMemoryHook(uc_engine* uc, uc_mem_type type, u64 addr, int si
     return {};
 }
 
-ARM_Unicorn::ARM_Unicorn(Timing::CoreTiming& core_timing) : core_timing{core_timing} {
+ARM_Unicorn::ARM_Unicorn(System& system) : system{system} {
     CHECKED(uc_open(UC_ARCH_ARM64, UC_MODE_ARM, &uc));
 
     auto fpv = 3 << 20;
@@ -177,7 +162,7 @@ void ARM_Unicorn::Run() {
     if (GDBStub::IsServerEnabled()) {
         ExecuteInstructions(std::max(4000000, 0));
     } else {
-        ExecuteInstructions(std::max(core_timing.GetDowncount(), 0));
+        ExecuteInstructions(std::max(system.CoreTiming().GetDowncount(), 0));
     }
 }
 
@@ -190,7 +175,7 @@ MICROPROFILE_DEFINE(ARM_Jit_Unicorn, "ARM JIT", "Unicorn", MP_RGB(255, 64, 64));
 void ARM_Unicorn::ExecuteInstructions(int num_instructions) {
     MICROPROFILE_SCOPE(ARM_Jit_Unicorn);
     CHECKED(uc_emu_start(uc, GetPC(), 1ULL << 63, 0, num_instructions));
-    core_timing.AddTicks(num_instructions);
+    system.CoreTiming().AddTicks(num_instructions);
     if (GDBStub::IsServerEnabled()) {
         if (last_bkpt_hit && last_bkpt.type == GDBStub::BreakpointType::Execute) {
             uc_reg_write(uc, UC_ARM64_REG_PC, &last_bkpt.address);
@@ -273,4 +258,20 @@ void ARM_Unicorn::RecordBreak(GDBStub::BreakpointAddress bkpt) {
     last_bkpt_hit = true;
 }
 
+void ARM_Unicorn::InterruptHook(uc_engine* uc, u32 int_no, void* user_data) {
+    u32 esr{};
+    CHECKED(uc_reg_read(uc, UC_ARM64_REG_ESR, &esr));
+
+    const auto ec = esr >> 26;
+    const auto iss = esr & 0xFFFFFF;
+
+    auto* const arm_instance = static_cast<ARM_Unicorn*>(user_data);
+
+    switch (ec) {
+    case 0x15: // SVC
+        Kernel::CallSVC(arm_instance->system, iss);
+        break;
+    }
+}
+
 } // namespace Core

src/core/arm/unicorn/arm_unicorn.h

@@ -9,15 +9,13 @@
 #include "core/arm/arm_interface.h"
 #include "core/gdbstub/gdbstub.h"
 
-namespace Core::Timing {
-class CoreTiming;
-}
-
 namespace Core {
 
+class System;
+
 class ARM_Unicorn final : public ARM_Interface {
 public:
-    explicit ARM_Unicorn(Timing::CoreTiming& core_timing);
+    explicit ARM_Unicorn(System& system);
     ~ARM_Unicorn() override;
 
     void MapBackingMemory(VAddr address, std::size_t size, u8* memory,
@@ -47,8 +45,10 @@ public:
     void RecordBreak(GDBStub::BreakpointAddress bkpt);
 
 private:
+    static void InterruptHook(uc_engine* uc, u32 int_no, void* user_data);
+
     uc_engine* uc{};
-    Timing::CoreTiming& core_timing;
+    System& system;
     GDBStub::BreakpointAddress last_bkpt{};
     bool last_bkpt_hit = false;
 };

src/core/core_cpu.cpp

@@ -55,13 +55,13 @@ Cpu::Cpu(System& system, ExclusiveMonitor& exclusive_monitor, CpuBarrier& cpu_ba
     : 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);
+        arm_interface = std::make_unique<ARM_Dynarmic>(system, exclusive_monitor, core_index);
 #else
-        arm_interface = std::make_unique<ARM_Unicorn>();
+        arm_interface = std::make_unique<ARM_Unicorn>(system);
         LOG_WARNING(Core, "CPU JIT requested, but Dynarmic not available");
 #endif
     } else {
-        arm_interface = std::make_unique<ARM_Unicorn>(core_timing);
+        arm_interface = std::make_unique<ARM_Unicorn>(system);
     }
 
     scheduler = std::make_unique<Kernel::Scheduler>(system, *arm_interface);

src/core/frontend/emu_window.h

@@ -12,6 +12,23 @@
 
 namespace Core::Frontend {
 
+/**
+ * Represents a graphics context that can be used for background computation or drawing. If the
+ * graphics backend doesn't require the context, then the implementation of these methods can be
+ * stubs
+ */
+class GraphicsContext {
+public:
+    /// Makes the graphics context current for the caller thread
+    virtual void MakeCurrent() = 0;
+
+    /// Releases (dunno if this is the "right" word) the context from the caller thread
+    virtual void DoneCurrent() = 0;
+
+    /// Swap buffers to display the next frame
+    virtual void SwapBuffers() = 0;
+};
+
 /**
  * Abstraction class used to provide an interface between emulation code and the frontend
  * (e.g. SDL, QGLWidget, GLFW, etc...).
@@ -30,7 +47,7 @@ namespace Core::Frontend {
  * - DO NOT TREAT THIS CLASS AS A GUI TOOLKIT ABSTRACTION LAYER. That's not what it is. Please
  *   re-read the upper points again and think about it if you don't see this.
  */
-class EmuWindow {
+class EmuWindow : public GraphicsContext {
 public:
     /// Data structure to store emuwindow configuration
     struct WindowConfig {
@@ -40,17 +57,21 @@ public:
         std::pair<unsigned, unsigned> min_client_area_size;
     };
 
-    /// Swap buffers to display the next frame
-    virtual void SwapBuffers() = 0;
-
     /// Polls window events
     virtual void PollEvents() = 0;
 
-    /// Makes the graphics context current for the caller thread
-    virtual void MakeCurrent() = 0;
-
-    /// Releases (dunno if this is the "right" word) the GLFW context from the caller thread
-    virtual void DoneCurrent() = 0;
+    /**
+     * Returns a GraphicsContext that the frontend provides that is shared with the emu window. This
+     * context can be used from other threads for background graphics computation. If the frontend
+     * is using a graphics backend that doesn't need anything specific to run on a different thread,
+     * then it can use a stubbed implemenation for GraphicsContext.
+     *
+     * If the return value is null, then the core should assume that the frontend cannot provide a
+     * Shared Context
+     */
+    virtual std::unique_ptr<GraphicsContext> CreateSharedContext() const {
+        return nullptr;
+    }
 
     /**
      * Signal that a touch pressed event has occurred (e.g. mouse click pressed)

src/core/hle/kernel/client_port.h

@@ -25,7 +25,7 @@ public:
         return name;
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::ClientPort;
+    static constexpr HandleType HANDLE_TYPE = HandleType::ClientPort;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

src/core/hle/kernel/client_session.h

@@ -29,7 +29,7 @@ public:
         return name;
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::ClientSession;
+    static constexpr HandleType HANDLE_TYPE = HandleType::ClientSession;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

src/core/hle/kernel/process.h

@@ -85,7 +85,7 @@ public:
         return name;
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::Process;
+    static constexpr HandleType HANDLE_TYPE = HandleType::Process;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

src/core/hle/kernel/readable_event.h

@@ -31,7 +31,7 @@ public:
         return reset_type;
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::ReadableEvent;
+    static constexpr HandleType HANDLE_TYPE = HandleType::ReadableEvent;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

src/core/hle/kernel/resource_limit.h

@@ -41,7 +41,7 @@ public:
         return GetTypeName();
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::ResourceLimit;
+    static constexpr HandleType HANDLE_TYPE = HandleType::ResourceLimit;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

src/core/hle/kernel/server_port.h

@@ -43,7 +43,7 @@ public:
         return name;
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::ServerPort;
+    static constexpr HandleType HANDLE_TYPE = HandleType::ServerPort;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

src/core/hle/kernel/server_session.h

@@ -46,7 +46,7 @@ public:
         return name;
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::ServerSession;
+    static constexpr HandleType HANDLE_TYPE = HandleType::ServerSession;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

src/core/hle/kernel/shared_memory.h

@@ -76,7 +76,7 @@ public:
         return name;
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::SharedMemory;
+    static constexpr HandleType HANDLE_TYPE = HandleType::SharedMemory;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

src/core/hle/kernel/svc.h

@@ -6,8 +6,12 @@
 
 #include "common/common_types.h"
 
+namespace Core {
+class System;
+}
+
 namespace Kernel {
 
-void CallSVC(u32 immediate);
+void CallSVC(Core::System& system, u32 immediate);
 
 } // namespace Kernel

src/core/hle/kernel/svc_wrap.h

@@ -11,278 +11,312 @@
 
 namespace Kernel {
 
-static inline u64 Param(int n) {
-    return Core::CurrentArmInterface().GetReg(n);
+static inline u64 Param(const Core::System& system, int n) {
+    return system.CurrentArmInterface().GetReg(n);
 }
 
 /**
  * HLE a function return from the current ARM userland process
- * @param res Result to return
+ * @param system System context
+ * @param result Result to return
  */
-static inline void FuncReturn(u64 res) {
-    Core::CurrentArmInterface().SetReg(0, res);
+static inline void FuncReturn(Core::System& system, u64 result) {
+    system.CurrentArmInterface().SetReg(0, result);
 }
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Function wrappers that return type ResultCode
 
-template <ResultCode func(u64)>
-void SvcWrap() {
-    FuncReturn(func(Param(0)).raw);
+template <ResultCode func(Core::System&, u64)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0)).raw);
 }
 
-template <ResultCode func(u32)>
-void SvcWrap() {
-    FuncReturn(func(static_cast<u32>(Param(0))).raw);
+template <ResultCode func(Core::System&, u32)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, static_cast<u32>(Param(system, 0))).raw);
 }
 
-template <ResultCode func(u32, u32)>
-void SvcWrap() {
-    FuncReturn(func(static_cast<u32>(Param(0)), static_cast<u32>(Param(1))).raw);
+template <ResultCode func(Core::System&, u32, u32)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(
+        system,
+        func(system, static_cast<u32>(Param(system, 0)), static_cast<u32>(Param(system, 1))).raw);
 }
 
-template <ResultCode func(u32*)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32*)>
+void SvcWrap(Core::System& system) {
     u32 param = 0;
-    const u32 retval = func(&param).raw;
-    Core::CurrentArmInterface().SetReg(1, param);
-    FuncReturn(retval);
+    const u32 retval = func(system, &param).raw;
+    system.CurrentArmInterface().SetReg(1, param);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u32*, u32)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32*, u32)>
+void SvcWrap(Core::System& system) {
     u32 param_1 = 0;
-    u32 retval = func(&param_1, static_cast<u32>(Param(1))).raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval = func(system, &param_1, static_cast<u32>(Param(system, 1))).raw;
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u32*, u32*)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32*, u32*)>
+void SvcWrap(Core::System& system) {
     u32 param_1 = 0;
     u32 param_2 = 0;
-    const u32 retval = func(&param_1, &param_2).raw;
+    const u32 retval = func(system, &param_1, &param_2).raw;
 
-    auto& arm_interface = Core::CurrentArmInterface();
+    auto& arm_interface = system.CurrentArmInterface();
     arm_interface.SetReg(1, param_1);
     arm_interface.SetReg(2, param_2);
 
-    FuncReturn(retval);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u32*, u64)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32*, u64)>
+void SvcWrap(Core::System& system) {
     u32 param_1 = 0;
-    const u32 retval = func(&param_1, Param(1)).raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval = func(system, &param_1, Param(system, 1)).raw;
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u32*, u64, u32)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32*, u64, u32)>
+void SvcWrap(Core::System& system) {
     u32 param_1 = 0;
-    const u32 retval = func(&param_1, Param(1), static_cast<u32>(Param(2))).raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval =
+        func(system, &param_1, Param(system, 1), static_cast<u32>(Param(system, 2))).raw;
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u64*, u32)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u64*, u32)>
+void SvcWrap(Core::System& system) {
     u64 param_1 = 0;
-    const u32 retval = func(&param_1, static_cast<u32>(Param(1))).raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval = func(system, &param_1, static_cast<u32>(Param(system, 1))).raw;
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u64, s32)>
-void SvcWrap() {
-    FuncReturn(func(Param(0), static_cast<s32>(Param(1))).raw);
+template <ResultCode func(Core::System&, u64, s32)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), static_cast<s32>(Param(system, 1))).raw);
 }
 
-template <ResultCode func(u64, u32)>
-void SvcWrap() {
-    FuncReturn(func(Param(0), static_cast<u32>(Param(1))).raw);
+template <ResultCode func(Core::System&, u64, u32)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), static_cast<u32>(Param(system, 1))).raw);
 }
 
-template <ResultCode func(u64*, u64)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u64*, u64)>
+void SvcWrap(Core::System& system) {
     u64 param_1 = 0;
-    u32 retval = func(&param_1, Param(1)).raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval = func(system, &param_1, Param(system, 1)).raw;
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u64*, u32, u32)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u64*, u32, u32)>
+void SvcWrap(Core::System& system) {
     u64 param_1 = 0;
-    u32 retval = func(&param_1, static_cast<u32>(Param(1)), static_cast<u32>(Param(2))).raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval = func(system, &param_1, static_cast<u32>(Param(system, 1)),
+                            static_cast<u32>(Param(system, 2)))
+                           .raw;
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u32, u64)>
-void SvcWrap() {
-    FuncReturn(func(static_cast<u32>(Param(0)), Param(1)).raw);
+template <ResultCode func(Core::System&, u32, u64)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, static_cast<u32>(Param(system, 0)), Param(system, 1)).raw);
 }
 
-template <ResultCode func(u32, u32, u64)>
-void SvcWrap() {
-    FuncReturn(func(static_cast<u32>(Param(0)), static_cast<u32>(Param(1)), Param(2)).raw);
+template <ResultCode func(Core::System&, u32, u32, u64)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, static_cast<u32>(Param(system, 0)),
+                            static_cast<u32>(Param(system, 1)), Param(system, 2))
+                           .raw);
 }
 
-template <ResultCode func(u32, u32*, u64*)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32, u32*, u64*)>
+void SvcWrap(Core::System& system) {
     u32 param_1 = 0;
     u64 param_2 = 0;
-    ResultCode retval = func(static_cast<u32>(Param(2)), &param_1, &param_2);
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    Core::CurrentArmInterface().SetReg(2, param_2);
-    FuncReturn(retval.raw);
+    const ResultCode retval = func(system, static_cast<u32>(Param(system, 2)), &param_1, &param_2);
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    system.CurrentArmInterface().SetReg(2, param_2);
+    FuncReturn(system, retval.raw);
 }
 
-template <ResultCode func(u64, u64, u32, u32)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u64, u64, u32, u32)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), Param(system, 1),
+                            static_cast<u32>(Param(system, 2)), static_cast<u32>(Param(system, 3)))
+                           .raw);
+}
+
+template <ResultCode func(Core::System&, u64, u64, u32, u64)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), Param(system, 1),
+                            static_cast<u32>(Param(system, 2)), Param(system, 3))
+                           .raw);
+}
+
+template <ResultCode func(Core::System&, u32, u64, u32)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, static_cast<u32>(Param(system, 0)), Param(system, 1),
+                            static_cast<u32>(Param(system, 2)))
+                           .raw);
+}
+
+template <ResultCode func(Core::System&, u64, u64, u64)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), Param(system, 1), Param(system, 2)).raw);
+}
+
+template <ResultCode func(Core::System&, u64, u64, u32)>
+void SvcWrap(Core::System& system) {
     FuncReturn(
-        func(Param(0), Param(1), static_cast<u32>(Param(2)), static_cast<u32>(Param(3))).raw);
+        system,
+        func(system, Param(system, 0), Param(system, 1), static_cast<u32>(Param(system, 2))).raw);
 }
 
-template <ResultCode func(u64, u64, u32, u64)>
-void SvcWrap() {
-    FuncReturn(func(Param(0), Param(1), static_cast<u32>(Param(2)), Param(3)).raw);
+template <ResultCode func(Core::System&, u32, u64, u64, u32)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, static_cast<u32>(Param(system, 0)), Param(system, 1),
+                            Param(system, 2), static_cast<u32>(Param(system, 3)))
+                           .raw);
 }
 
-template <ResultCode func(u32, u64, u32)>
-void SvcWrap() {
-    FuncReturn(func(static_cast<u32>(Param(0)), Param(1), static_cast<u32>(Param(2))).raw);
-}
-
-template <ResultCode func(u64, u64, u64)>
-void SvcWrap() {
-    FuncReturn(func(Param(0), Param(1), Param(2)).raw);
-}
-
-template <ResultCode func(u64, u64, u32)>
-void SvcWrap() {
-    FuncReturn(func(Param(0), Param(1), static_cast<u32>(Param(2))).raw);
-}
-
-template <ResultCode func(u32, u64, u64, u32)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32, u64, u64)>
+void SvcWrap(Core::System& system) {
     FuncReturn(
-        func(static_cast<u32>(Param(0)), Param(1), Param(2), static_cast<u32>(Param(3))).raw);
+        system,
+        func(system, static_cast<u32>(Param(system, 0)), Param(system, 1), Param(system, 2)).raw);
 }
 
-template <ResultCode func(u32, u64, u64)>
-void SvcWrap() {
-    FuncReturn(func(static_cast<u32>(Param(0)), Param(1), Param(2)).raw);
-}
-
-template <ResultCode func(u32*, u64, u64, s64)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32*, u64, u64, s64)>
+void SvcWrap(Core::System& system) {
     u32 param_1 = 0;
-    ResultCode retval =
-        func(&param_1, Param(1), static_cast<u32>(Param(2)), static_cast<s64>(Param(3)));
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval.raw);
+    const u32 retval = func(system, &param_1, Param(system, 1), static_cast<u32>(Param(system, 2)),
+                            static_cast<s64>(Param(system, 3)))
+                           .raw;
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u64, u64, u32, s64)>
-void SvcWrap() {
-    FuncReturn(
-        func(Param(0), Param(1), static_cast<u32>(Param(2)), static_cast<s64>(Param(3))).raw);
+template <ResultCode func(Core::System&, u64, u64, u32, s64)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), Param(system, 1),
+                            static_cast<u32>(Param(system, 2)), static_cast<s64>(Param(system, 3)))
+                           .raw);
 }
 
-template <ResultCode func(u64*, u64, u64, u64)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u64*, u64, u64, u64)>
+void SvcWrap(Core::System& system) {
     u64 param_1 = 0;
-    u32 retval = func(&param_1, Param(1), Param(2), Param(3)).raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval =
+        func(system, &param_1, Param(system, 1), Param(system, 2), Param(system, 3)).raw;
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u32*, u64, u64, u64, u32, s32)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32*, u64, u64, u64, u32, s32)>
+void SvcWrap(Core::System& system) {
     u32 param_1 = 0;
-    u32 retval = func(&param_1, Param(1), Param(2), Param(3), static_cast<u32>(Param(4)),
-                      static_cast<s32>(Param(5)))
-                     .raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval = func(system, &param_1, Param(system, 1), Param(system, 2), Param(system, 3),
+                            static_cast<u32>(Param(system, 4)), static_cast<s32>(Param(system, 5)))
+                           .raw;
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u32*, u64, u64, u32)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, u32*, u64, u64, u32)>
+void SvcWrap(Core::System& system) {
     u32 param_1 = 0;
-    u32 retval = func(&param_1, Param(1), Param(2), static_cast<u32>(Param(3))).raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval = func(system, &param_1, Param(system, 1), Param(system, 2),
+                            static_cast<u32>(Param(system, 3)))
+                           .raw;
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(Handle*, u64, u32, u32)>
-void SvcWrap() {
+template <ResultCode func(Core::System&, Handle*, u64, u32, u32)>
+void SvcWrap(Core::System& system) {
     u32 param_1 = 0;
-    u32 retval =
-        func(&param_1, Param(1), static_cast<u32>(Param(2)), static_cast<u32>(Param(3))).raw;
-    Core::CurrentArmInterface().SetReg(1, param_1);
-    FuncReturn(retval);
+    const u32 retval = func(system, &param_1, Param(system, 1), static_cast<u32>(Param(system, 2)),
+                            static_cast<u32>(Param(system, 3)))
+                           .raw;
+
+    system.CurrentArmInterface().SetReg(1, param_1);
+    FuncReturn(system, retval);
 }
 
-template <ResultCode func(u64, u32, s32, s64)>
-void SvcWrap() {
-    FuncReturn(func(Param(0), static_cast<u32>(Param(1)), static_cast<s32>(Param(2)),
-                    static_cast<s64>(Param(3)))
-                   .raw);
+template <ResultCode func(Core::System&, u64, u32, s32, s64)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), static_cast<u32>(Param(system, 1)),
+                            static_cast<s32>(Param(system, 2)), static_cast<s64>(Param(system, 3)))
+                           .raw);
 }
 
-template <ResultCode func(u64, u32, s32, s32)>
-void SvcWrap() {
-    FuncReturn(func(Param(0), static_cast<u32>(Param(1)), static_cast<s32>(Param(2)),
-                    static_cast<s32>(Param(3)))
-                   .raw);
+template <ResultCode func(Core::System&, u64, u32, s32, s32)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system, Param(system, 0), static_cast<u32>(Param(system, 1)),
+                            static_cast<s32>(Param(system, 2)), static_cast<s32>(Param(system, 3)))
+                           .raw);
 }
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Function wrappers that return type u32
 
-template <u32 func()>
-void SvcWrap() {
-    FuncReturn(func());
+template <u32 func(Core::System&)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system));
 }
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Function wrappers that return type u64
 
-template <u64 func()>
-void SvcWrap() {
-    FuncReturn(func());
+template <u64 func(Core::System&)>
+void SvcWrap(Core::System& system) {
+    FuncReturn(system, func(system));
 }
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 /// Function wrappers that return type void
 
-template <void func()>
-void SvcWrap() {
-    func();
+template <void func(Core::System&)>
+void SvcWrap(Core::System& system) {
+    func(system);
 }
 
-template <void func(s64)>
-void SvcWrap() {
-    func(static_cast<s64>(Param(0)));
+template <void func(Core::System&, s64)>
+void SvcWrap(Core::System& system) {
+    func(system, static_cast<s64>(Param(system, 0)));
 }
 
-template <void func(u64, u64 len)>
-void SvcWrap() {
-    func(Param(0), Param(1));
+template <void func(Core::System&, u64, u64)>
+void SvcWrap(Core::System& system) {
+    func(system, Param(system, 0), Param(system, 1));
 }
 
-template <void func(u64, u64, u64)>
-void SvcWrap() {
-    func(Param(0), Param(1), Param(2));
+template <void func(Core::System&, u64, u64, u64)>
+void SvcWrap(Core::System& system) {
+    func(system, Param(system, 0), Param(system, 1), Param(system, 2));
 }
 
-template <void func(u32, u64, u64)>
-void SvcWrap() {
-    func(static_cast<u32>(Param(0)), Param(1), Param(2));
+template <void func(Core::System&, u32, u64, u64)>
+void SvcWrap(Core::System& system) {
+    func(system, static_cast<u32>(Param(system, 0)), Param(system, 1), Param(system, 2));
 }
 
 } // namespace Kernel

src/core/hle/kernel/thread.h

@@ -106,7 +106,7 @@ public:
         return "Thread";
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::Thread;
+    static constexpr HandleType HANDLE_TYPE = HandleType::Thread;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

src/core/hle/kernel/writable_event.h

@@ -37,7 +37,7 @@ public:
         return name;
     }
 
-    static const HandleType HANDLE_TYPE = HandleType::WritableEvent;
+    static constexpr HandleType HANDLE_TYPE = HandleType::WritableEvent;
     HandleType GetHandleType() const override {
         return HANDLE_TYPE;
     }

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

@@ -115,11 +115,12 @@ private:
 
     void Read(Kernel::HLERequestContext& ctx) {
         IPC::RequestParser rp{ctx};
-        const u64 unk = rp.Pop<u64>();
+        const u64 option = rp.Pop<u64>();
         const s64 offset = rp.Pop<s64>();
         const s64 length = rp.Pop<s64>();
 
-        LOG_DEBUG(Service_FS, "called, offset=0x{:X}, length={}", offset, length);
+        LOG_DEBUG(Service_FS, "called, option={}, offset=0x{:X}, length={}", option, offset,
+                  length);
 
         // Error checking
         if (length < 0) {
@@ -148,11 +149,12 @@ private:
 
     void Write(Kernel::HLERequestContext& ctx) {
         IPC::RequestParser rp{ctx};
-        const u64 unk = rp.Pop<u64>();
+        const u64 option = rp.Pop<u64>();
         const s64 offset = rp.Pop<s64>();
         const s64 length = rp.Pop<s64>();
 
-        LOG_DEBUG(Service_FS, "called, offset=0x{:X}, length={}", offset, length);
+        LOG_DEBUG(Service_FS, "called, option={}, offset=0x{:X}, length={}", option, offset,
+                  length);
 
         // Error checking
         if (length < 0) {
@@ -250,10 +252,7 @@ private:
     u64 next_entry_index = 0;
 
     void Read(Kernel::HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        const u64 unk = rp.Pop<u64>();
-
-        LOG_DEBUG(Service_FS, "called, unk=0x{:X}", unk);
+        LOG_DEBUG(Service_FS, "called.");
 
         // Calculate how many entries we can fit in the output buffer
         const u64 count_entries = ctx.GetWriteBufferSize() / sizeof(FileSys::Entry);

src/core/hle/service/ldr/ldr.cpp

@@ -93,12 +93,18 @@ public:
     }
 
     void LoadNrr(Kernel::HLERequestContext& ctx) {
+        struct Parameters {
+            u64_le process_id;
+            u64_le nrr_address;
+            u64_le nrr_size;
+        };
+
         IPC::RequestParser rp{ctx};
-        rp.Skip(2, false);
-        const VAddr nrr_addr{rp.Pop<VAddr>()};
-        const u64 nrr_size{rp.Pop<u64>()};
-        LOG_DEBUG(Service_LDR, "called with nrr_addr={:016X}, nrr_size={:016X}", nrr_addr,
-                  nrr_size);
+        const auto [process_id, nrr_address, nrr_size] = rp.PopRaw<Parameters>();
+
+        LOG_DEBUG(Service_LDR,
+                  "called with process_id={:016X}, nrr_address={:016X}, nrr_size={:016X}",
+                  process_id, nrr_address, nrr_size);
 
         if (!initialized) {
             LOG_ERROR(Service_LDR, "LDR:RO not initialized before use!");
@@ -116,24 +122,26 @@ public:
         }
 
         // NRR Address does not fall on 0x1000 byte boundary
-        if (!Common::Is4KBAligned(nrr_addr)) {
-            LOG_ERROR(Service_LDR, "NRR Address has invalid alignment (actual {:016X})!", nrr_addr);
+        if (!Common::Is4KBAligned(nrr_address)) {
+            LOG_ERROR(Service_LDR, "NRR Address has invalid alignment (actual {:016X})!",
+                      nrr_address);
             IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERROR_INVALID_ALIGNMENT);
             return;
         }
 
         // NRR Size is zero or causes overflow
-        if (nrr_addr + nrr_size <= nrr_addr || nrr_size == 0 || !Common::Is4KBAligned(nrr_size)) {
+        if (nrr_address + nrr_size <= nrr_address || nrr_size == 0 ||
+            !Common::Is4KBAligned(nrr_size)) {
             LOG_ERROR(Service_LDR, "NRR Size is invalid! (nrr_address={:016X}, nrr_size={:016X})",
-                      nrr_addr, nrr_size);
+                      nrr_address, nrr_size);
             IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERROR_INVALID_SIZE);
             return;
         }
         // Read NRR data from memory
         std::vector<u8> nrr_data(nrr_size);
-        Memory::ReadBlock(nrr_addr, nrr_data.data(), nrr_size);
+        Memory::ReadBlock(nrr_address, nrr_data.data(), nrr_size);
         NRRHeader header;
         std::memcpy(&header, nrr_data.data(), sizeof(NRRHeader));
 
@@ -174,7 +182,7 @@ public:
             hashes.emplace_back(hash);
         }
 
-        nrr.insert_or_assign(nrr_addr, std::move(hashes));
+        nrr.insert_or_assign(nrr_address, std::move(hashes));
 
         IPC::ResponseBuilder rb{ctx, 2};
         rb.Push(RESULT_SUCCESS);
@@ -188,23 +196,30 @@ public:
             return;
         }
 
-        IPC::RequestParser rp{ctx};
-        rp.Skip(2, false);
-        const auto nrr_addr{rp.Pop<VAddr>()};
-        LOG_DEBUG(Service_LDR, "called with nrr_addr={:016X}", nrr_addr);
+        struct Parameters {
+            u64_le process_id;
+            u64_le nrr_address;
+        };
 
-        if (!Common::Is4KBAligned(nrr_addr)) {
-            LOG_ERROR(Service_LDR, "NRR Address has invalid alignment (actual {:016X})!", nrr_addr);
+        IPC::RequestParser rp{ctx};
+        const auto [process_id, nrr_address] = rp.PopRaw<Parameters>();
+
+        LOG_DEBUG(Service_LDR, "called with process_id={:016X}, nrr_addr={:016X}", process_id,
+                  nrr_address);
+
+        if (!Common::Is4KBAligned(nrr_address)) {
+            LOG_ERROR(Service_LDR, "NRR Address has invalid alignment (actual {:016X})!",
+                      nrr_address);
             IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERROR_INVALID_ALIGNMENT);
             return;
         }
 
-        const auto iter = nrr.find(nrr_addr);
+        const auto iter = nrr.find(nrr_address);
         if (iter == nrr.end()) {
             LOG_ERROR(Service_LDR,
                       "Attempting to unload NRR which has not been loaded! (addr={:016X})",
-                      nrr_addr);
+                      nrr_address);
             IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERROR_INVALID_NRR_ADDRESS);
             return;
@@ -216,16 +231,22 @@ public:
     }
 
     void LoadNro(Kernel::HLERequestContext& ctx) {
+        struct Parameters {
+            u64_le process_id;
+            u64_le image_address;
+            u64_le image_size;
+            u64_le bss_address;
+            u64_le bss_size;
+        };
+
         IPC::RequestParser rp{ctx};
-        rp.Skip(2, false);
-        const VAddr nro_addr{rp.Pop<VAddr>()};
-        const u64 nro_size{rp.Pop<u64>()};
-        const VAddr bss_addr{rp.Pop<VAddr>()};
-        const u64 bss_size{rp.Pop<u64>()};
-        LOG_DEBUG(
-            Service_LDR,
-            "called with nro_addr={:016X}, nro_size={:016X}, bss_addr={:016X}, bss_size={:016X}",
-            nro_addr, nro_size, bss_addr, bss_size);
+        const auto [process_id, nro_address, nro_size, bss_address, bss_size] =
+            rp.PopRaw<Parameters>();
+
+        LOG_DEBUG(Service_LDR,
+                  "called with pid={:016X}, nro_addr={:016X}, nro_size={:016X}, bss_addr={:016X}, "
+                  "bss_size={:016X}",
+                  process_id, nro_address, nro_size, bss_address, bss_size);
 
         if (!initialized) {
             LOG_ERROR(Service_LDR, "LDR:RO not initialized before use!");
@@ -243,8 +264,9 @@ public:
         }
 
         // NRO Address does not fall on 0x1000 byte boundary
-        if (!Common::Is4KBAligned(nro_addr)) {
-            LOG_ERROR(Service_LDR, "NRO Address has invalid alignment (actual {:016X})!", nro_addr);
+        if (!Common::Is4KBAligned(nro_address)) {
+            LOG_ERROR(Service_LDR, "NRO Address has invalid alignment (actual {:016X})!",
+                      nro_address);
             IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERROR_INVALID_ALIGNMENT);
             return;
@@ -252,15 +274,15 @@ public:
 
         // NRO Size or BSS Size is zero or causes overflow
         const auto nro_size_valid =
-            nro_size != 0 && nro_addr + nro_size > nro_addr && Common::Is4KBAligned(nro_size);
-        const auto bss_size_valid =
-            nro_size + bss_size >= nro_size && (bss_size == 0 || bss_addr + bss_size > bss_addr);
+            nro_size != 0 && nro_address + nro_size > nro_address && Common::Is4KBAligned(nro_size);
+        const auto bss_size_valid = nro_size + bss_size >= nro_size &&
+                                    (bss_size == 0 || bss_address + bss_size > bss_address);
 
         if (!nro_size_valid || !bss_size_valid) {
             LOG_ERROR(Service_LDR,
                       "NRO Size or BSS Size is invalid! (nro_address={:016X}, nro_size={:016X}, "
                       "bss_address={:016X}, bss_size={:016X})",
-                      nro_addr, nro_size, bss_addr, bss_size);
+                      nro_address, nro_size, bss_address, bss_size);
             IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERROR_INVALID_SIZE);
             return;
@@ -268,7 +290,7 @@ public:
 
         // Read NRO data from memory
         std::vector<u8> nro_data(nro_size);
-        Memory::ReadBlock(nro_addr, nro_data.data(), nro_size);
+        Memory::ReadBlock(nro_address, nro_data.data(), nro_size);
 
         SHA256Hash hash{};
         mbedtls_sha256(nro_data.data(), nro_data.size(), hash.data(), 0);
@@ -318,17 +340,18 @@ public:
             return;
         }
 
-        ASSERT(vm_manager
-                   .MirrorMemory(*map_address, nro_addr, nro_size, Kernel::MemoryState::ModuleCode)
-                   .IsSuccess());
-        ASSERT(vm_manager.UnmapRange(nro_addr, nro_size).IsSuccess());
+        ASSERT(
+            vm_manager
+                .MirrorMemory(*map_address, nro_address, nro_size, Kernel::MemoryState::ModuleCode)
+                .IsSuccess());
+        ASSERT(vm_manager.UnmapRange(nro_address, nro_size).IsSuccess());
 
         if (bss_size > 0) {
             ASSERT(vm_manager
-                       .MirrorMemory(*map_address + nro_size, bss_addr, bss_size,
+                       .MirrorMemory(*map_address + nro_size, bss_address, bss_size,
                                      Kernel::MemoryState::ModuleCode)
                        .IsSuccess());
-            ASSERT(vm_manager.UnmapRange(bss_addr, bss_size).IsSuccess());
+            ASSERT(vm_manager.UnmapRange(bss_address, bss_size).IsSuccess());
         }
 
         vm_manager.ReprotectRange(*map_address, header.text_size,
@@ -348,13 +371,6 @@ public:
     }
 
     void UnloadNro(Kernel::HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-        rp.Skip(2, false);
-        const VAddr mapped_addr{rp.PopRaw<VAddr>()};
-        const VAddr heap_addr{rp.PopRaw<VAddr>()};
-        LOG_DEBUG(Service_LDR, "called with mapped_addr={:016X}, heap_addr={:016X}", mapped_addr,
-                  heap_addr);
-
         if (!initialized) {
             LOG_ERROR(Service_LDR, "LDR:RO not initialized before use!");
             IPC::ResponseBuilder rb{ctx, 2};
@@ -362,22 +378,30 @@ public:
             return;
         }
 
-        if (!Common::Is4KBAligned(mapped_addr) || !Common::Is4KBAligned(heap_addr)) {
-            LOG_ERROR(Service_LDR,
-                      "NRO/BSS Address has invalid alignment (actual nro_addr={:016X}, "
-                      "bss_addr={:016X})!",
-                      mapped_addr, heap_addr);
+        struct Parameters {
+            u64_le process_id;
+            u64_le nro_address;
+        };
+
+        IPC::RequestParser rp{ctx};
+        const auto [process_id, nro_address] = rp.PopRaw<Parameters>();
+        LOG_DEBUG(Service_LDR, "called with process_id={:016X}, nro_address=0x{:016X}", process_id,
+                  nro_address);
+
+        if (!Common::Is4KBAligned(nro_address)) {
+            LOG_ERROR(Service_LDR, "NRO address has invalid alignment (nro_address=0x{:016X})",
+                      nro_address);
             IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERROR_INVALID_ALIGNMENT);
             return;
         }
 
-        const auto iter = nro.find(mapped_addr);
+        const auto iter = nro.find(nro_address);
         if (iter == nro.end()) {
             LOG_ERROR(Service_LDR,
-                      "The NRO attempting to unmap was not mapped or has an invalid address "
-                      "(actual {:016X})!",
-                      mapped_addr);
+                      "The NRO attempting to be unmapped was not mapped or has an invalid address "
+                      "(nro_address=0x{:016X})!",
+                      nro_address);
             IPC::ResponseBuilder rb{ctx, 2};
             rb.Push(ERROR_INVALID_NRO_ADDRESS);
             return;
@@ -386,10 +410,7 @@ public:
         auto& vm_manager = Core::CurrentProcess()->VMManager();
         const auto& nro_size = iter->second.size;
 
-        ASSERT(vm_manager
-                   .MirrorMemory(heap_addr, mapped_addr, nro_size, Kernel::MemoryState::ModuleCode)
-                   .IsSuccess());
-        ASSERT(vm_manager.UnmapRange(mapped_addr, nro_size).IsSuccess());
+        ASSERT(vm_manager.UnmapRange(nro_address, nro_size).IsSuccess());
 
         Core::System::GetInstance().InvalidateCpuInstructionCaches();
 
@@ -459,11 +480,10 @@ private:
     std::map<VAddr, NROInfo> nro;
     std::map<VAddr, std::vector<SHA256Hash>> nrr;
 
-    bool IsValidNROHash(const SHA256Hash& hash) {
-        return std::any_of(
-            nrr.begin(), nrr.end(), [&hash](const std::pair<VAddr, std::vector<SHA256Hash>>& p) {
-                return std::find(p.second.begin(), p.second.end(), hash) != p.second.end();
-            });
+    bool IsValidNROHash(const SHA256Hash& hash) const {
+        return std::any_of(nrr.begin(), nrr.end(), [&hash](const auto& p) {
+            return std::find(p.second.begin(), p.second.end(), hash) != p.second.end();
+        });
     }
 
     static bool IsValidNRO(const NROHeader& header, u64 nro_size, u64 bss_size) {

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

@@ -21,12 +21,13 @@
 #include "core/hle/service/vi/display/vi_display.h"
 #include "core/hle/service/vi/layer/vi_layer.h"
 #include "core/perf_stats.h"
+#include "core/settings.h"
 #include "video_core/renderer_base.h"
 
 namespace Service::NVFlinger {
 
-constexpr std::size_t SCREEN_REFRESH_RATE = 60;
-constexpr s64 frame_ticks = static_cast<s64>(Core::Timing::BASE_CLOCK_RATE / SCREEN_REFRESH_RATE);
+constexpr s64 frame_ticks = static_cast<s64>(Core::Timing::BASE_CLOCK_RATE / 60);
+constexpr s64 frame_ticks_30fps = static_cast<s64>(Core::Timing::BASE_CLOCK_RATE / 30);
 
 NVFlinger::NVFlinger(Core::Timing::CoreTiming& core_timing) : core_timing{core_timing} {
     displays.emplace_back(0, "Default");
@@ -36,13 +37,15 @@ NVFlinger::NVFlinger(Core::Timing::CoreTiming& core_timing) : core_timing{core_t
     displays.emplace_back(4, "Null");
 
     // Schedule the screen composition events
-    composition_event =
-        core_timing.RegisterEvent("ScreenComposition", [this](u64 userdata, s64 cycles_late) {
+    const auto ticks = Settings::values.force_30fps_mode ? frame_ticks_30fps : frame_ticks;
+
+    composition_event = core_timing.RegisterEvent(
+        "ScreenComposition", [this, ticks](u64 userdata, s64 cycles_late) {
             Compose();
-            this->core_timing.ScheduleEvent(frame_ticks - cycles_late, composition_event);
+            this->core_timing.ScheduleEvent(ticks - cycles_late, composition_event);
         });
 
-    core_timing.ScheduleEvent(frame_ticks, composition_event);
+    core_timing.ScheduleEvent(ticks, composition_event);
 }
 
 NVFlinger::~NVFlinger() {
@@ -62,6 +65,7 @@ std::optional<u64> NVFlinger::OpenDisplay(std::string_view name) {
     const auto itr =
         std::find_if(displays.begin(), displays.end(),
                      [&](const VI::Display& display) { return display.GetName() == name; });
+
     if (itr == displays.end()) {
         return {};
     }

src/core/settings.h

@@ -393,6 +393,7 @@ struct Values {
     bool use_disk_shader_cache;
     bool use_accurate_gpu_emulation;
     bool use_asynchronous_gpu_emulation;
+    bool force_30fps_mode;
 
     float bg_red;
     float bg_green;

src/video_core/CMakeLists.txt

@@ -106,6 +106,8 @@ add_library(video_core STATIC
     textures/decoders.cpp
     textures/decoders.h
     textures/texture.h
+    texture_cache.cpp
+    texture_cache.h
     video_core.cpp
     video_core.h
 )
@@ -127,12 +129,14 @@ if (ENABLE_VULKAN)
         renderer_vulkan/vk_sampler_cache.h
         renderer_vulkan/vk_scheduler.cpp
         renderer_vulkan/vk_scheduler.h
+        renderer_vulkan/vk_shader_decompiler.cpp
+        renderer_vulkan/vk_shader_decompiler.h
         renderer_vulkan/vk_stream_buffer.cpp
         renderer_vulkan/vk_stream_buffer.h
         renderer_vulkan/vk_swapchain.cpp
         renderer_vulkan/vk_swapchain.h)
 
-    target_include_directories(video_core PRIVATE ../../externals/Vulkan-Headers/include)
+    target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include)
     target_compile_definitions(video_core PRIVATE HAS_VULKAN)
 endif()
 
@@ -140,3 +144,6 @@ create_target_directory_groups(video_core)
 
 target_link_libraries(video_core PUBLIC common core)
 target_link_libraries(video_core PRIVATE glad)
+if (ENABLE_VULKAN)
+    target_link_libraries(video_core PRIVATE sirit)
+endif()

src/video_core/engines/shader_bytecode.h

@@ -917,21 +917,34 @@ union Instruction {
     } iset;
 
     union {
-        BitField<8, 2, Register::Size> dest_size;
-        BitField<10, 2, Register::Size> src_size;
-        BitField<12, 1, u64> is_output_signed;
-        BitField<13, 1, u64> is_input_signed;
-        BitField<41, 2, u64> selector;
+        BitField<41, 2, u64> selector; // i2i and i2f only
         BitField<45, 1, u64> negate_a;
         BitField<49, 1, u64> abs_a;
+        BitField<10, 2, Register::Size> src_size;
+        BitField<13, 1, u64> is_input_signed;
+        BitField<8, 2, Register::Size> dst_size;
+        BitField<12, 1, u64> is_output_signed;
+
+        union {
+            BitField<39, 2, u64> tab5cb8_2;
+        } i2f;
 
         union {
             BitField<39, 2, F2iRoundingOp> rounding;
         } f2i;
 
         union {
-            BitField<39, 4, F2fRoundingOp> rounding;
+            BitField<8, 2, Register::Size> src_size;
+            BitField<10, 2, Register::Size> dst_size;
+            BitField<39, 4, u64> rounding;
+            // H0, H1 extract for F16 missing
+            BitField<41, 1, u64> selector; // Guessed as some games set it, TODO: reverse this value
+            F2fRoundingOp GetRoundingMode() const {
+                constexpr u64 rounding_mask = 0x0B;
+                return static_cast<F2fRoundingOp>(rounding.Value() & rounding_mask);
+            }
         } f2f;
+
     } conversion;
 
     union {
@@ -1678,7 +1691,7 @@ private:
             INST("0011100-00101---", Id::SHR_IMM, Type::Shift, "SHR_IMM"),
             INST("0100110011100---", Id::I2I_C, Type::Conversion, "I2I_C"),
             INST("0101110011100---", Id::I2I_R, Type::Conversion, "I2I_R"),
-            INST("01110001-1000---", Id::I2I_IMM, Type::Conversion, "I2I_IMM"),
+            INST("0011101-11100---", Id::I2I_IMM, Type::Conversion, "I2I_IMM"),
             INST("0100110010111---", Id::I2F_C, Type::Conversion, "I2F_C"),
             INST("0101110010111---", Id::I2F_R, Type::Conversion, "I2F_R"),
             INST("0011100-10111---", Id::I2F_IMM, Type::Conversion, "I2F_IMM"),

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -112,11 +112,26 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only,
     params.pixel_format = PixelFormatFromTextureFormat(config.tic.format, config.tic.r_type.Value(),
                                                        params.srgb_conversion);
 
-    if (params.pixel_format == PixelFormat::R16U && config.tsc.depth_compare_enabled) {
+    if (config.tsc.depth_compare_enabled) {
         // Some titles create a 'R16U' (normalized 16-bit) texture with depth_compare enabled,
         // then attempt to sample from it via a shadow sampler. Convert format to Z16 (which also
         // causes GetFormatType to properly return 'Depth' below).
-        params.pixel_format = PixelFormat::Z16;
+        if (GetFormatType(params.pixel_format) == SurfaceType::ColorTexture) {
+            switch (params.pixel_format) {
+            case PixelFormat::R16S:
+            case PixelFormat::R16U:
+            case PixelFormat::R16F:
+                params.pixel_format = PixelFormat::Z16;
+                break;
+            case PixelFormat::R32F:
+                params.pixel_format = PixelFormat::Z32F;
+                break;
+            default:
+                LOG_WARNING(HW_GPU, "Color texture format being used with depth compare: {}",
+                            static_cast<u32>(params.pixel_format));
+                break;
+            }
+        }
     }
 
     params.component_type = ComponentTypeFromTexture(config.tic.r_type.Value());
@@ -266,6 +281,10 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only,
     params.component_type = ComponentTypeFromRenderTarget(config.format);
     params.type = GetFormatType(params.pixel_format);
     params.width = config.width;
+    if (!params.is_tiled) {
+        const u32 bpp = params.GetFormatBpp() / 8;
+        params.pitch = config.width * bpp;
+    }
     params.height = config.height;
     params.unaligned_height = config.height;
     params.target = SurfaceTarget::Texture2D;
@@ -1175,10 +1194,16 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface,
         return new_surface;
     }
 
+    const bool old_compressed =
+        GetFormatTuple(old_params.pixel_format, old_params.component_type).compressed;
+    const bool new_compressed =
+        GetFormatTuple(new_params.pixel_format, new_params.component_type).compressed;
+    const bool compatible_formats =
+        GetFormatBpp(old_params.pixel_format) == GetFormatBpp(new_params.pixel_format) &&
+        !(old_compressed || new_compressed);
     // For compatible surfaces, we can just do fast glCopyImageSubData based copy
-    if (old_params.target == new_params.target && old_params.type == new_params.type &&
-        old_params.depth == new_params.depth && old_params.depth == 1 &&
-        GetFormatBpp(old_params.pixel_format) == GetFormatBpp(new_params.pixel_format)) {
+    if (old_params.target == new_params.target && old_params.depth == new_params.depth &&
+        old_params.depth == 1 && compatible_formats) {
         FastCopySurface(old_surface, new_surface);
         return new_surface;
     }
@@ -1193,7 +1218,7 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface,
     case SurfaceTarget::TextureCubemap:
     case SurfaceTarget::Texture2DArray:
     case SurfaceTarget::TextureCubeArray:
-        if (old_params.pixel_format == new_params.pixel_format)
+        if (compatible_formats)
             FastLayeredCopySurface(old_surface, new_surface);
         else {
             AccurateCopySurface(old_surface, new_surface);

src/video_core/renderer_vulkan/vk_shader_decompiler.h

@@ -0,0 +1,80 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <memory>
+#include <set>
+#include <utility>
+#include <vector>
+
+#include <sirit/sirit.h>
+
+#include "common/common_types.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/shader/shader_ir.h"
+
+namespace VideoCommon::Shader {
+class ShaderIR;
+}
+
+namespace Vulkan::VKShader {
+
+using Maxwell = Tegra::Engines::Maxwell3D::Regs;
+
+using SamplerEntry = VideoCommon::Shader::Sampler;
+
+constexpr u32 DESCRIPTOR_SET = 0;
+
+class ConstBufferEntry : public VideoCommon::Shader::ConstBuffer {
+public:
+    explicit constexpr ConstBufferEntry(const VideoCommon::Shader::ConstBuffer& entry, u32 index)
+        : VideoCommon::Shader::ConstBuffer{entry}, index{index} {}
+
+    constexpr u32 GetIndex() const {
+        return index;
+    }
+
+private:
+    u32 index{};
+};
+
+class GlobalBufferEntry {
+public:
+    explicit GlobalBufferEntry(u32 cbuf_index, u32 cbuf_offset)
+        : cbuf_index{cbuf_index}, cbuf_offset{cbuf_offset} {}
+
+    u32 GetCbufIndex() const {
+        return cbuf_index;
+    }
+
+    u32 GetCbufOffset() const {
+        return cbuf_offset;
+    }
+
+private:
+    u32 cbuf_index{};
+    u32 cbuf_offset{};
+};
+
+struct ShaderEntries {
+    u32 const_buffers_base_binding{};
+    u32 global_buffers_base_binding{};
+    u32 samplers_base_binding{};
+    std::vector<ConstBufferEntry> const_buffers;
+    std::vector<GlobalBufferEntry> global_buffers;
+    std::vector<SamplerEntry> samplers;
+    std::set<u32> attributes;
+    std::array<bool, Maxwell::NumClipDistances> clip_distances{};
+    std::size_t shader_length{};
+    Sirit::Id entry_function{};
+    std::vector<Sirit::Id> interfaces;
+};
+
+using DecompilerResult = std::pair<std::unique_ptr<Sirit::Module>, ShaderEntries>;
+
+DecompilerResult Decompile(const VideoCommon::Shader::ShaderIR& ir, Maxwell::ShaderStage stage);
+
+} // namespace Vulkan::VKShader

src/video_core/shader/decode/conversion.cpp

@@ -18,13 +18,29 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
     const auto opcode = OpCode::Decode(instr);
 
     switch (opcode->get().GetId()) {
-    case OpCode::Id::I2I_R: {
+    case OpCode::Id::I2I_R:
+    case OpCode::Id::I2I_C:
+    case OpCode::Id::I2I_IMM: {
         UNIMPLEMENTED_IF(instr.conversion.selector);
+        UNIMPLEMENTED_IF(instr.conversion.dst_size != Register::Size::Word);
+        UNIMPLEMENTED_IF(instr.alu.saturate_d);
 
         const bool input_signed = instr.conversion.is_input_signed;
         const bool output_signed = instr.conversion.is_output_signed;
 
-        Node value = GetRegister(instr.gpr20);
+        Node value = [&]() {
+            switch (opcode->get().GetId()) {
+            case OpCode::Id::I2I_R:
+                return GetRegister(instr.gpr20);
+            case OpCode::Id::I2I_C:
+                return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
+            case OpCode::Id::I2I_IMM:
+                return Immediate(instr.alu.GetSignedImm20_20());
+            default:
+                UNREACHABLE();
+                return Immediate(0);
+            }
+        }();
         value = ConvertIntegerSize(value, instr.conversion.src_size, input_signed);
 
         value = GetOperandAbsNegInteger(value, instr.conversion.abs_a, instr.conversion.negate_a,
@@ -38,17 +54,24 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
         break;
     }
     case OpCode::Id::I2F_R:
-    case OpCode::Id::I2F_C: {
-        UNIMPLEMENTED_IF(instr.conversion.dest_size != Register::Size::Word);
+    case OpCode::Id::I2F_C:
+    case OpCode::Id::I2F_IMM: {
+        UNIMPLEMENTED_IF(instr.conversion.dst_size != Register::Size::Word);
         UNIMPLEMENTED_IF(instr.conversion.selector);
         UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                              "Condition codes generation in I2F is not implemented");
 
         Node value = [&]() {
-            if (instr.is_b_gpr) {
+            switch (opcode->get().GetId()) {
+            case OpCode::Id::I2F_R:
                 return GetRegister(instr.gpr20);
-            } else {
+            case OpCode::Id::I2F_C:
                 return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
+            case OpCode::Id::I2F_IMM:
+                return Immediate(instr.alu.GetSignedImm20_20());
+            default:
+                UNREACHABLE();
+                return Immediate(0);
             }
         }();
         const bool input_signed = instr.conversion.is_input_signed;
@@ -62,24 +85,31 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
         break;
     }
     case OpCode::Id::F2F_R:
-    case OpCode::Id::F2F_C: {
-        UNIMPLEMENTED_IF(instr.conversion.dest_size != Register::Size::Word);
-        UNIMPLEMENTED_IF(instr.conversion.src_size != Register::Size::Word);
+    case OpCode::Id::F2F_C:
+    case OpCode::Id::F2F_IMM: {
+        UNIMPLEMENTED_IF(instr.conversion.f2f.dst_size != Register::Size::Word);
+        UNIMPLEMENTED_IF(instr.conversion.f2f.src_size != Register::Size::Word);
         UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                              "Condition codes generation in F2F is not implemented");
 
         Node value = [&]() {
-            if (instr.is_b_gpr) {
+            switch (opcode->get().GetId()) {
+            case OpCode::Id::F2F_R:
                 return GetRegister(instr.gpr20);
-            } else {
+            case OpCode::Id::F2F_C:
                 return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
+            case OpCode::Id::F2F_IMM:
+                return GetImmediate19(instr);
+            default:
+                UNREACHABLE();
+                return Immediate(0);
             }
         }();
 
         value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);
 
         value = [&]() {
-            switch (instr.conversion.f2f.rounding) {
+            switch (instr.conversion.f2f.GetRoundingMode()) {
             case Tegra::Shader::F2fRoundingOp::None:
                 return value;
             case Tegra::Shader::F2fRoundingOp::Round:
@@ -102,15 +132,22 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
         break;
     }
     case OpCode::Id::F2I_R:
-    case OpCode::Id::F2I_C: {
+    case OpCode::Id::F2I_C:
+    case OpCode::Id::F2I_IMM: {
         UNIMPLEMENTED_IF(instr.conversion.src_size != Register::Size::Word);
         UNIMPLEMENTED_IF_MSG(instr.generates_cc,
                              "Condition codes generation in F2I is not implemented");
         Node value = [&]() {
-            if (instr.is_b_gpr) {
+            switch (opcode->get().GetId()) {
+            case OpCode::Id::F2I_R:
                 return GetRegister(instr.gpr20);
-            } else {
+            case OpCode::Id::F2I_C:
                 return GetConstBuffer(instr.cbuf34.index, instr.cbuf34.GetOffset());
+            case OpCode::Id::F2I_IMM:
+                return GetImmediate19(instr);
+            default:
+                UNREACHABLE();
+                return Immediate(0);
             }
         }();
 
@@ -134,7 +171,7 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
         }();
         const bool is_signed = instr.conversion.is_output_signed;
         value = SignedOperation(OperationCode::ICastFloat, is_signed, PRECISE, value);
-        value = ConvertIntegerSize(value, instr.conversion.dest_size, is_signed);
+        value = ConvertIntegerSize(value, instr.conversion.dst_size, is_signed);
 
         SetRegister(bb, instr.gpr0, value);
         break;

src/video_core/surface.cpp

@@ -294,6 +294,8 @@ PixelFormat PixelFormatFromTextureFormat(Tegra::Texture::TextureFormat format,
         return PixelFormat::Z16;
     case Tegra::Texture::TextureFormat::Z24S8:
         return PixelFormat::Z24S8;
+    case Tegra::Texture::TextureFormat::ZF32_X24S8:
+        return PixelFormat::Z32FS8;
     case Tegra::Texture::TextureFormat::DXT1:
         return is_srgb ? PixelFormat::DXT1_SRGB : PixelFormat::DXT1;
     case Tegra::Texture::TextureFormat::DXT23:

src/video_core/texture_cache.cpp

@@ -0,0 +1,386 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/alignment.h"
+#include "common/assert.h"
+#include "common/cityhash.h"
+#include "common/common_types.h"
+#include "core/core.h"
+#include "video_core/surface.h"
+#include "video_core/texture_cache.h"
+#include "video_core/textures/decoders.h"
+#include "video_core/textures/texture.h"
+
+namespace VideoCommon {
+
+using VideoCore::Surface::SurfaceTarget;
+
+using VideoCore::Surface::ComponentTypeFromDepthFormat;
+using VideoCore::Surface::ComponentTypeFromRenderTarget;
+using VideoCore::Surface::ComponentTypeFromTexture;
+using VideoCore::Surface::PixelFormatFromDepthFormat;
+using VideoCore::Surface::PixelFormatFromRenderTargetFormat;
+using VideoCore::Surface::PixelFormatFromTextureFormat;
+using VideoCore::Surface::SurfaceTargetFromTextureType;
+
+constexpr u32 GetMipmapSize(bool uncompressed, u32 mip_size, u32 tile) {
+    return uncompressed ? mip_size : std::max(1U, (mip_size + tile - 1) / tile);
+}
+
+SurfaceParams SurfaceParams::CreateForTexture(Core::System& system,
+                                              const Tegra::Texture::FullTextureInfo& config) {
+    SurfaceParams params;
+    params.is_tiled = config.tic.IsTiled();
+    params.block_width = params.is_tiled ? config.tic.BlockWidth() : 0,
+    params.block_height = params.is_tiled ? config.tic.BlockHeight() : 0,
+    params.block_depth = params.is_tiled ? config.tic.BlockDepth() : 0,
+    params.tile_width_spacing = params.is_tiled ? (1 << config.tic.tile_width_spacing.Value()) : 1;
+    params.pixel_format =
+        PixelFormatFromTextureFormat(config.tic.format, config.tic.r_type.Value(), false);
+    params.component_type = ComponentTypeFromTexture(config.tic.r_type.Value());
+    params.type = GetFormatType(params.pixel_format);
+    params.target = SurfaceTargetFromTextureType(config.tic.texture_type);
+    params.width = Common::AlignUp(config.tic.Width(), GetCompressionFactor(params.pixel_format));
+    params.height = Common::AlignUp(config.tic.Height(), GetCompressionFactor(params.pixel_format));
+    params.depth = config.tic.Depth();
+    if (params.target == SurfaceTarget::TextureCubemap ||
+        params.target == SurfaceTarget::TextureCubeArray) {
+        params.depth *= 6;
+    }
+    params.pitch = params.is_tiled ? 0 : config.tic.Pitch();
+    params.unaligned_height = config.tic.Height();
+    params.num_levels = config.tic.max_mip_level + 1;
+
+    params.CalculateCachedValues();
+    return params;
+}
+
+SurfaceParams SurfaceParams::CreateForDepthBuffer(
+    Core::System& system, u32 zeta_width, u32 zeta_height, Tegra::DepthFormat format,
+    u32 block_width, u32 block_height, u32 block_depth,
+    Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type) {
+    SurfaceParams params;
+    params.is_tiled = type == Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout::BlockLinear;
+    params.block_width = 1 << std::min(block_width, 5U);
+    params.block_height = 1 << std::min(block_height, 5U);
+    params.block_depth = 1 << std::min(block_depth, 5U);
+    params.tile_width_spacing = 1;
+    params.pixel_format = PixelFormatFromDepthFormat(format);
+    params.component_type = ComponentTypeFromDepthFormat(format);
+    params.type = GetFormatType(params.pixel_format);
+    params.width = zeta_width;
+    params.height = zeta_height;
+    params.unaligned_height = zeta_height;
+    params.target = SurfaceTarget::Texture2D;
+    params.depth = 1;
+    params.num_levels = 1;
+
+    params.CalculateCachedValues();
+    return params;
+}
+
+SurfaceParams SurfaceParams::CreateForFramebuffer(Core::System& system, std::size_t index) {
+    const auto& config{system.GPU().Maxwell3D().regs.rt[index]};
+    SurfaceParams params;
+    params.is_tiled =
+        config.memory_layout.type == Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout::BlockLinear;
+    params.block_width = 1 << config.memory_layout.block_width;
+    params.block_height = 1 << config.memory_layout.block_height;
+    params.block_depth = 1 << config.memory_layout.block_depth;
+    params.tile_width_spacing = 1;
+    params.pixel_format = PixelFormatFromRenderTargetFormat(config.format);
+    params.component_type = ComponentTypeFromRenderTarget(config.format);
+    params.type = GetFormatType(params.pixel_format);
+    if (params.is_tiled) {
+        params.width = config.width;
+    } else {
+        const u32 bpp = GetFormatBpp(params.pixel_format) / CHAR_BIT;
+        params.pitch = config.width;
+        params.width = params.pitch / bpp;
+    }
+    params.height = config.height;
+    params.depth = 1;
+    params.unaligned_height = config.height;
+    params.target = SurfaceTarget::Texture2D;
+    params.num_levels = 1;
+
+    params.CalculateCachedValues();
+    return params;
+}
+
+SurfaceParams SurfaceParams::CreateForFermiCopySurface(
+    const Tegra::Engines::Fermi2D::Regs::Surface& config) {
+    SurfaceParams params{};
+    params.is_tiled = !config.linear;
+    params.block_width = params.is_tiled ? std::min(config.BlockWidth(), 32U) : 0,
+    params.block_height = params.is_tiled ? std::min(config.BlockHeight(), 32U) : 0,
+    params.block_depth = params.is_tiled ? std::min(config.BlockDepth(), 32U) : 0,
+    params.tile_width_spacing = 1;
+    params.pixel_format = PixelFormatFromRenderTargetFormat(config.format);
+    params.component_type = ComponentTypeFromRenderTarget(config.format);
+    params.type = GetFormatType(params.pixel_format);
+    params.width = config.width;
+    params.height = config.height;
+    params.unaligned_height = config.height;
+    // TODO(Rodrigo): Try to guess the surface target from depth and layer parameters
+    params.target = SurfaceTarget::Texture2D;
+    params.depth = 1;
+    params.num_levels = 1;
+
+    params.CalculateCachedValues();
+    return params;
+}
+
+u32 SurfaceParams::GetMipWidth(u32 level) const {
+    return std::max(1U, width >> level);
+}
+
+u32 SurfaceParams::GetMipHeight(u32 level) const {
+    return std::max(1U, height >> level);
+}
+
+u32 SurfaceParams::GetMipDepth(u32 level) const {
+    return IsLayered() ? depth : std::max(1U, depth >> level);
+}
+
+bool SurfaceParams::IsLayered() const {
+    switch (target) {
+    case SurfaceTarget::Texture1DArray:
+    case SurfaceTarget::Texture2DArray:
+    case SurfaceTarget::TextureCubeArray:
+    case SurfaceTarget::TextureCubemap:
+        return true;
+    default:
+        return false;
+    }
+}
+
+u32 SurfaceParams::GetMipBlockHeight(u32 level) const {
+    // Auto block resizing algorithm from:
+    // https://cgit.freedesktop.org/mesa/mesa/tree/src/gallium/drivers/nouveau/nv50/nv50_miptree.c
+    if (level == 0) {
+        return block_height;
+    }
+    const u32 height{GetMipHeight(level)};
+    const u32 default_block_height{GetDefaultBlockHeight(pixel_format)};
+    const u32 blocks_in_y{(height + default_block_height - 1) / default_block_height};
+    u32 block_height = 16;
+    while (block_height > 1 && blocks_in_y <= block_height * 4) {
+        block_height >>= 1;
+    }
+    return block_height;
+}
+
+u32 SurfaceParams::GetMipBlockDepth(u32 level) const {
+    if (level == 0)
+        return block_depth;
+    if (target != SurfaceTarget::Texture3D)
+        return 1;
+
+    const u32 depth{GetMipDepth(level)};
+    u32 block_depth = 32;
+    while (block_depth > 1 && depth * 2 <= block_depth) {
+        block_depth >>= 1;
+    }
+    if (block_depth == 32 && GetMipBlockHeight(level) >= 4) {
+        return 16;
+    }
+    return block_depth;
+}
+
+std::size_t SurfaceParams::GetGuestMipmapLevelOffset(u32 level) const {
+    std::size_t offset = 0;
+    for (u32 i = 0; i < level; i++) {
+        offset += GetInnerMipmapMemorySize(i, false, IsLayered(), false);
+    }
+    return offset;
+}
+
+std::size_t SurfaceParams::GetHostMipmapLevelOffset(u32 level) const {
+    std::size_t offset = 0;
+    for (u32 i = 0; i < level; i++) {
+        offset += GetInnerMipmapMemorySize(i, true, false, false);
+    }
+    return offset;
+}
+
+std::size_t SurfaceParams::GetGuestLayerSize() const {
+    return GetInnerMemorySize(false, true, false);
+}
+
+std::size_t SurfaceParams::GetHostLayerSize(u32 level) const {
+    return GetInnerMipmapMemorySize(level, true, IsLayered(), false);
+}
+
+bool SurfaceParams::IsFamiliar(const SurfaceParams& view_params) const {
+    if (std::tie(is_tiled, tile_width_spacing, pixel_format, component_type, type) !=
+        std::tie(view_params.is_tiled, view_params.tile_width_spacing, view_params.pixel_format,
+                 view_params.component_type, view_params.type)) {
+        return false;
+    }
+
+    const SurfaceTarget view_target{view_params.target};
+    if (view_target == target) {
+        return true;
+    }
+
+    switch (target) {
+    case SurfaceTarget::Texture1D:
+    case SurfaceTarget::Texture2D:
+    case SurfaceTarget::Texture3D:
+        return false;
+    case SurfaceTarget::Texture1DArray:
+        return view_target == SurfaceTarget::Texture1D;
+    case SurfaceTarget::Texture2DArray:
+        return view_target == SurfaceTarget::Texture2D;
+    case SurfaceTarget::TextureCubemap:
+        return view_target == SurfaceTarget::Texture2D ||
+               view_target == SurfaceTarget::Texture2DArray;
+    case SurfaceTarget::TextureCubeArray:
+        return view_target == SurfaceTarget::Texture2D ||
+               view_target == SurfaceTarget::Texture2DArray ||
+               view_target == SurfaceTarget::TextureCubemap;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented texture family={}", static_cast<u32>(target));
+        return false;
+    }
+}
+
+bool SurfaceParams::IsPixelFormatZeta() const {
+    return pixel_format >= VideoCore::Surface::PixelFormat::MaxColorFormat &&
+           pixel_format < VideoCore::Surface::PixelFormat::MaxDepthStencilFormat;
+}
+
+void SurfaceParams::CalculateCachedValues() {
+    guest_size_in_bytes = GetInnerMemorySize(false, false, false);
+
+    // ASTC is uncompressed in software, in emulated as RGBA8
+    if (IsPixelFormatASTC(pixel_format)) {
+        host_size_in_bytes = width * height * depth * 4;
+    } else {
+        host_size_in_bytes = GetInnerMemorySize(true, false, false);
+    }
+
+    switch (target) {
+    case SurfaceTarget::Texture1D:
+    case SurfaceTarget::Texture2D:
+    case SurfaceTarget::Texture3D:
+        num_layers = 1;
+        break;
+    case SurfaceTarget::Texture1DArray:
+    case SurfaceTarget::Texture2DArray:
+    case SurfaceTarget::TextureCubemap:
+    case SurfaceTarget::TextureCubeArray:
+        num_layers = depth;
+        break;
+    default:
+        UNREACHABLE();
+    }
+}
+
+std::size_t SurfaceParams::GetInnerMipmapMemorySize(u32 level, bool as_host_size, bool layer_only,
+                                                    bool uncompressed) const {
+    const bool tiled{as_host_size ? false : is_tiled};
+    const u32 tile_x{GetDefaultBlockWidth(pixel_format)};
+    const u32 tile_y{GetDefaultBlockHeight(pixel_format)};
+    const u32 width{GetMipmapSize(uncompressed, GetMipWidth(level), tile_x)};
+    const u32 height{GetMipmapSize(uncompressed, GetMipHeight(level), tile_y)};
+    const u32 depth{layer_only ? 1U : GetMipDepth(level)};
+    return Tegra::Texture::CalculateSize(tiled, GetBytesPerPixel(pixel_format), width, height,
+                                         depth, GetMipBlockHeight(level), GetMipBlockDepth(level));
+}
+
+std::size_t SurfaceParams::GetInnerMemorySize(bool as_host_size, bool layer_only,
+                                              bool uncompressed) const {
+    std::size_t size = 0;
+    for (u32 level = 0; level < num_levels; ++level) {
+        size += GetInnerMipmapMemorySize(level, as_host_size, layer_only, uncompressed);
+    }
+    if (!as_host_size && is_tiled) {
+        size = Common::AlignUp(size, Tegra::Texture::GetGOBSize() * block_height * block_depth);
+    }
+    return size;
+}
+
+std::map<u64, std::pair<u32, u32>> SurfaceParams::CreateViewOffsetMap() const {
+    std::map<u64, std::pair<u32, u32>> view_offset_map;
+    switch (target) {
+    case SurfaceTarget::Texture1D:
+    case SurfaceTarget::Texture2D:
+    case SurfaceTarget::Texture3D: {
+        constexpr u32 layer = 0;
+        for (u32 level = 0; level < num_levels; ++level) {
+            const std::size_t offset{GetGuestMipmapLevelOffset(level)};
+            view_offset_map.insert({offset, {layer, level}});
+        }
+        break;
+    }
+    case SurfaceTarget::Texture1DArray:
+    case SurfaceTarget::Texture2DArray:
+    case SurfaceTarget::TextureCubemap:
+    case SurfaceTarget::TextureCubeArray: {
+        const std::size_t layer_size{GetGuestLayerSize()};
+        for (u32 level = 0; level < num_levels; ++level) {
+            const std::size_t level_offset{GetGuestMipmapLevelOffset(level)};
+            for (u32 layer = 0; layer < num_layers; ++layer) {
+                const auto layer_offset{static_cast<std::size_t>(layer_size * layer)};
+                const std::size_t offset{level_offset + layer_offset};
+                view_offset_map.insert({offset, {layer, level}});
+            }
+        }
+        break;
+    }
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented surface target {}", static_cast<u32>(target));
+    }
+    return view_offset_map;
+}
+
+bool SurfaceParams::IsViewValid(const SurfaceParams& view_params, u32 layer, u32 level) const {
+    return IsDimensionValid(view_params, level) && IsDepthValid(view_params, level) &&
+           IsInBounds(view_params, layer, level);
+}
+
+bool SurfaceParams::IsDimensionValid(const SurfaceParams& view_params, u32 level) const {
+    return view_params.width == GetMipWidth(level) && view_params.height == GetMipHeight(level);
+}
+
+bool SurfaceParams::IsDepthValid(const SurfaceParams& view_params, u32 level) const {
+    if (view_params.target != SurfaceTarget::Texture3D) {
+        return true;
+    }
+    return view_params.depth == GetMipDepth(level);
+}
+
+bool SurfaceParams::IsInBounds(const SurfaceParams& view_params, u32 layer, u32 level) const {
+    return layer + view_params.num_layers <= num_layers &&
+           level + view_params.num_levels <= num_levels;
+}
+
+std::size_t HasheableSurfaceParams::Hash() const {
+    return static_cast<std::size_t>(
+        Common::CityHash64(reinterpret_cast<const char*>(this), sizeof(*this)));
+}
+
+bool HasheableSurfaceParams::operator==(const HasheableSurfaceParams& rhs) const {
+    return std::tie(is_tiled, block_width, block_height, block_depth, tile_width_spacing, width,
+                    height, depth, pitch, unaligned_height, num_levels, pixel_format,
+                    component_type, type, target) ==
+           std::tie(rhs.is_tiled, rhs.block_width, rhs.block_height, rhs.block_depth,
+                    rhs.tile_width_spacing, rhs.width, rhs.height, rhs.depth, rhs.pitch,
+                    rhs.unaligned_height, rhs.num_levels, rhs.pixel_format, rhs.component_type,
+                    rhs.type, rhs.target);
+}
+
+std::size_t ViewKey::Hash() const {
+    return static_cast<std::size_t>(
+        Common::CityHash64(reinterpret_cast<const char*>(this), sizeof(*this)));
+}
+
+bool ViewKey::operator==(const ViewKey& rhs) const {
+    return std::tie(base_layer, num_layers, base_level, num_levels) ==
+           std::tie(rhs.base_layer, rhs.num_layers, rhs.base_level, rhs.num_levels);
+}
+
+} // namespace VideoCommon

src/video_core/texture_cache.h

@@ -0,0 +1,586 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <list>
+#include <memory>
+#include <set>
+#include <tuple>
+#include <type_traits>
+#include <unordered_map>
+
+#include <boost/icl/interval_map.hpp>
+#include <boost/range/iterator_range.hpp>
+
+#include "common/assert.h"
+#include "common/common_types.h"
+#include "core/memory.h"
+#include "video_core/engines/fermi_2d.h"
+#include "video_core/engines/maxwell_3d.h"
+#include "video_core/gpu.h"
+#include "video_core/rasterizer_interface.h"
+#include "video_core/surface.h"
+
+namespace Core {
+class System;
+}
+
+namespace Tegra::Texture {
+struct FullTextureInfo;
+}
+
+namespace VideoCore {
+class RasterizerInterface;
+}
+
+namespace VideoCommon {
+
+class HasheableSurfaceParams {
+public:
+    std::size_t Hash() const;
+
+    bool operator==(const HasheableSurfaceParams& rhs) const;
+
+protected:
+    // Avoid creation outside of a managed environment.
+    HasheableSurfaceParams() = default;
+
+    bool is_tiled;
+    u32 block_width;
+    u32 block_height;
+    u32 block_depth;
+    u32 tile_width_spacing;
+    u32 width;
+    u32 height;
+    u32 depth;
+    u32 pitch;
+    u32 unaligned_height;
+    u32 num_levels;
+    VideoCore::Surface::PixelFormat pixel_format;
+    VideoCore::Surface::ComponentType component_type;
+    VideoCore::Surface::SurfaceType type;
+    VideoCore::Surface::SurfaceTarget target;
+};
+
+class SurfaceParams final : public HasheableSurfaceParams {
+public:
+    /// Creates SurfaceCachedParams from a texture configuration.
+    static SurfaceParams CreateForTexture(Core::System& system,
+                                          const Tegra::Texture::FullTextureInfo& config);
+
+    /// Creates SurfaceCachedParams for a depth buffer configuration.
+    static SurfaceParams CreateForDepthBuffer(
+        Core::System& system, u32 zeta_width, u32 zeta_height, Tegra::DepthFormat format,
+        u32 block_width, u32 block_height, u32 block_depth,
+        Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type);
+
+    /// Creates SurfaceCachedParams from a framebuffer configuration.
+    static SurfaceParams CreateForFramebuffer(Core::System& system, std::size_t index);
+
+    /// Creates SurfaceCachedParams from a Fermi2D surface configuration.
+    static SurfaceParams CreateForFermiCopySurface(
+        const Tegra::Engines::Fermi2D::Regs::Surface& config);
+
+    bool IsTiled() const {
+        return is_tiled;
+    }
+
+    u32 GetBlockWidth() const {
+        return block_width;
+    }
+
+    u32 GetTileWidthSpacing() const {
+        return tile_width_spacing;
+    }
+
+    u32 GetWidth() const {
+        return width;
+    }
+
+    u32 GetHeight() const {
+        return height;
+    }
+
+    u32 GetDepth() const {
+        return depth;
+    }
+
+    u32 GetPitch() const {
+        return pitch;
+    }
+
+    u32 GetNumLevels() const {
+        return num_levels;
+    }
+
+    VideoCore::Surface::PixelFormat GetPixelFormat() const {
+        return pixel_format;
+    }
+
+    VideoCore::Surface::ComponentType GetComponentType() const {
+        return component_type;
+    }
+
+    VideoCore::Surface::SurfaceTarget GetTarget() const {
+        return target;
+    }
+
+    VideoCore::Surface::SurfaceType GetType() const {
+        return type;
+    }
+
+    std::size_t GetGuestSizeInBytes() const {
+        return guest_size_in_bytes;
+    }
+
+    std::size_t GetHostSizeInBytes() const {
+        return host_size_in_bytes;
+    }
+
+    u32 GetNumLayers() const {
+        return num_layers;
+    }
+
+    /// Returns the width of a given mipmap level.
+    u32 GetMipWidth(u32 level) const;
+
+    /// Returns the height of a given mipmap level.
+    u32 GetMipHeight(u32 level) const;
+
+    /// Returns the depth of a given mipmap level.
+    u32 GetMipDepth(u32 level) const;
+
+    /// Returns true if these parameters are from a layered surface.
+    bool IsLayered() const;
+
+    /// Returns the block height of a given mipmap level.
+    u32 GetMipBlockHeight(u32 level) const;
+
+    /// Returns the block depth of a given mipmap level.
+    u32 GetMipBlockDepth(u32 level) const;
+
+    /// Returns the offset in bytes in guest memory of a given mipmap level.
+    std::size_t GetGuestMipmapLevelOffset(u32 level) const;
+
+    /// Returns the offset in bytes in host memory (linear) of a given mipmap level.
+    std::size_t GetHostMipmapLevelOffset(u32 level) const;
+
+    /// Returns the size of a layer in bytes in guest memory.
+    std::size_t GetGuestLayerSize() const;
+
+    /// Returns the size of a layer in bytes in host memory for a given mipmap level.
+    std::size_t GetHostLayerSize(u32 level) const;
+
+    /// Returns true if another surface can be familiar with this. This is a loosely defined term
+    /// that reflects the possibility of these two surface parameters potentially being part of a
+    /// bigger superset.
+    bool IsFamiliar(const SurfaceParams& view_params) const;
+
+    /// Returns true if the pixel format is a depth and/or stencil format.
+    bool IsPixelFormatZeta() const;
+
+    /// Creates a map that redirects an address difference to a layer and mipmap level.
+    std::map<u64, std::pair<u32, u32>> CreateViewOffsetMap() const;
+
+    /// Returns true if the passed surface view parameters is equal or a valid subset of this.
+    bool IsViewValid(const SurfaceParams& view_params, u32 layer, u32 level) const;
+
+private:
+    /// Calculates values that can be deduced from HasheableSurfaceParams.
+    void CalculateCachedValues();
+
+    /// Returns the size of a given mipmap level.
+    std::size_t GetInnerMipmapMemorySize(u32 level, bool as_host_size, bool layer_only,
+                                         bool uncompressed) const;
+
+    /// Returns the size of all mipmap levels and aligns as needed.
+    std::size_t GetInnerMemorySize(bool as_host_size, bool layer_only, bool uncompressed) const;
+
+    /// Returns true if the passed view width and height match the size of this params in a given
+    /// mipmap level.
+    bool IsDimensionValid(const SurfaceParams& view_params, u32 level) const;
+
+    /// Returns true if the passed view depth match the size of this params in a given mipmap level.
+    bool IsDepthValid(const SurfaceParams& view_params, u32 level) const;
+
+    /// Returns true if the passed view layers and mipmap levels are in bounds.
+    bool IsInBounds(const SurfaceParams& view_params, u32 layer, u32 level) const;
+
+    std::size_t guest_size_in_bytes;
+    std::size_t host_size_in_bytes;
+    u32 num_layers;
+};
+
+struct ViewKey {
+    std::size_t Hash() const;
+
+    bool operator==(const ViewKey& rhs) const;
+
+    u32 base_layer{};
+    u32 num_layers{};
+    u32 base_level{};
+    u32 num_levels{};
+};
+
+} // namespace VideoCommon
+
+namespace std {
+
+template <>
+struct hash<VideoCommon::SurfaceParams> {
+    std::size_t operator()(const VideoCommon::SurfaceParams& k) const noexcept {
+        return k.Hash();
+    }
+};
+
+template <>
+struct hash<VideoCommon::ViewKey> {
+    std::size_t operator()(const VideoCommon::ViewKey& k) const noexcept {
+        return k.Hash();
+    }
+};
+
+} // namespace std
+
+namespace VideoCommon {
+
+template <typename TView, typename TExecutionContext>
+class SurfaceBase {
+    static_assert(std::is_trivially_copyable_v<TExecutionContext>);
+
+public:
+    virtual void LoadBuffer() = 0;
+
+    virtual TExecutionContext FlushBuffer(TExecutionContext exctx) = 0;
+
+    virtual TExecutionContext UploadTexture(TExecutionContext exctx) = 0;
+
+    TView* TryGetView(VAddr view_addr, const SurfaceParams& view_params) {
+        if (view_addr < cpu_addr || !params.IsFamiliar(view_params)) {
+            // It can't be a view if it's in a prior address.
+            return {};
+        }
+
+        const auto relative_offset{static_cast<u64>(view_addr - cpu_addr)};
+        const auto it{view_offset_map.find(relative_offset)};
+        if (it == view_offset_map.end()) {
+            // Couldn't find an aligned view.
+            return {};
+        }
+        const auto [layer, level] = it->second;
+
+        if (!params.IsViewValid(view_params, layer, level)) {
+            return {};
+        }
+
+        return GetView(layer, view_params.GetNumLayers(), level, view_params.GetNumLevels());
+    }
+
+    VAddr GetCpuAddr() const {
+        ASSERT(is_registered);
+        return cpu_addr;
+    }
+
+    u8* GetHostPtr() const {
+        ASSERT(is_registered);
+        return host_ptr;
+    }
+
+    CacheAddr GetCacheAddr() const {
+        ASSERT(is_registered);
+        return cache_addr;
+    }
+
+    std::size_t GetSizeInBytes() const {
+        return params.GetGuestSizeInBytes();
+    }
+
+    void MarkAsModified(bool is_modified_) {
+        is_modified = is_modified_;
+    }
+
+    const SurfaceParams& GetSurfaceParams() const {
+        return params;
+    }
+
+    TView* GetView(VAddr view_addr, const SurfaceParams& view_params) {
+        TView* view{TryGetView(view_addr, view_params)};
+        ASSERT(view != nullptr);
+        return view;
+    }
+
+    void Register(VAddr cpu_addr_, u8* host_ptr_) {
+        ASSERT(!is_registered);
+        is_registered = true;
+        cpu_addr = cpu_addr_;
+        host_ptr = host_ptr_;
+        cache_addr = ToCacheAddr(host_ptr_);
+    }
+
+    void Register(VAddr cpu_addr_) {
+        Register(cpu_addr_, Memory::GetPointer(cpu_addr_));
+    }
+
+    void Unregister() {
+        ASSERT(is_registered);
+        is_registered = false;
+    }
+
+    bool IsRegistered() const {
+        return is_registered;
+    }
+
+protected:
+    explicit SurfaceBase(const SurfaceParams& params)
+        : params{params}, view_offset_map{params.CreateViewOffsetMap()} {}
+
+    ~SurfaceBase() = default;
+
+    virtual std::unique_ptr<TView> CreateView(const ViewKey& view_key) = 0;
+
+    bool IsModified() const {
+        return is_modified;
+    }
+
+    const SurfaceParams params;
+
+private:
+    TView* GetView(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels) {
+        const ViewKey key{base_layer, num_layers, base_level, num_levels};
+        const auto [entry, is_cache_miss] = views.try_emplace(key);
+        auto& view{entry->second};
+        if (is_cache_miss) {
+            view = CreateView(key);
+        }
+        return view.get();
+    }
+
+    const std::map<u64, std::pair<u32, u32>> view_offset_map;
+
+    VAddr cpu_addr{};
+    u8* host_ptr{};
+    CacheAddr cache_addr{};
+    bool is_modified{};
+    bool is_registered{};
+    std::unordered_map<ViewKey, std::unique_ptr<TView>> views;
+};
+
+template <typename TSurface, typename TView, typename TExecutionContext>
+class TextureCache {
+    static_assert(std::is_trivially_copyable_v<TExecutionContext>);
+    using ResultType = std::tuple<TView*, TExecutionContext>;
+    using IntervalMap = boost::icl::interval_map<CacheAddr, std::set<TSurface*>>;
+    using IntervalType = typename IntervalMap::interval_type;
+
+public:
+    void InvalidateRegion(CacheAddr addr, std::size_t size) {
+        for (TSurface* surface : GetSurfacesInRegion(addr, size)) {
+            if (!surface->IsRegistered()) {
+                // Skip duplicates
+                continue;
+            }
+            Unregister(surface);
+        }
+    }
+
+    ResultType GetTextureSurface(TExecutionContext exctx,
+                                 const Tegra::Texture::FullTextureInfo& config) {
+        auto& memory_manager{system.GPU().MemoryManager()};
+        const auto cpu_addr{memory_manager.GpuToCpuAddress(config.tic.Address())};
+        if (!cpu_addr) {
+            return {{}, exctx};
+        }
+        const auto params{SurfaceParams::CreateForTexture(system, config)};
+        return GetSurfaceView(exctx, *cpu_addr, params, true);
+    }
+
+    ResultType GetDepthBufferSurface(TExecutionContext exctx, bool preserve_contents) {
+        const auto& regs{system.GPU().Maxwell3D().regs};
+        if (!regs.zeta.Address() || !regs.zeta_enable) {
+            return {{}, exctx};
+        }
+
+        auto& memory_manager{system.GPU().MemoryManager()};
+        const auto cpu_addr{memory_manager.GpuToCpuAddress(regs.zeta.Address())};
+        if (!cpu_addr) {
+            return {{}, exctx};
+        }
+
+        const auto depth_params{SurfaceParams::CreateForDepthBuffer(
+            system, regs.zeta_width, regs.zeta_height, regs.zeta.format,
+            regs.zeta.memory_layout.block_width, regs.zeta.memory_layout.block_height,
+            regs.zeta.memory_layout.block_depth, regs.zeta.memory_layout.type)};
+        return GetSurfaceView(exctx, *cpu_addr, depth_params, preserve_contents);
+    }
+
+    ResultType GetColorBufferSurface(TExecutionContext exctx, std::size_t index,
+                                     bool preserve_contents) {
+        ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets);
+
+        const auto& regs{system.GPU().Maxwell3D().regs};
+        if (index >= regs.rt_control.count || regs.rt[index].Address() == 0 ||
+            regs.rt[index].format == Tegra::RenderTargetFormat::NONE) {
+            return {{}, exctx};
+        }
+
+        auto& memory_manager{system.GPU().MemoryManager()};
+        const auto& config{system.GPU().Maxwell3D().regs.rt[index]};
+        const auto cpu_addr{memory_manager.GpuToCpuAddress(
+            config.Address() + config.base_layer * config.layer_stride * sizeof(u32))};
+        if (!cpu_addr) {
+            return {{}, exctx};
+        }
+
+        return GetSurfaceView(exctx, *cpu_addr, SurfaceParams::CreateForFramebuffer(system, index),
+                              preserve_contents);
+    }
+
+    ResultType GetFermiSurface(TExecutionContext exctx,
+                               const Tegra::Engines::Fermi2D::Regs::Surface& config) {
+        const auto cpu_addr{system.GPU().MemoryManager().GpuToCpuAddress(config.Address())};
+        ASSERT(cpu_addr);
+        return GetSurfaceView(exctx, *cpu_addr, SurfaceParams::CreateForFermiCopySurface(config),
+                              true);
+    }
+
+    TSurface* TryFindFramebufferSurface(const u8* host_ptr) const {
+        const auto it{registered_surfaces.find(ToCacheAddr(host_ptr))};
+        return it != registered_surfaces.end() ? *it->second.begin() : nullptr;
+    }
+
+protected:
+    TextureCache(Core::System& system, VideoCore::RasterizerInterface& rasterizer)
+        : system{system}, rasterizer{rasterizer} {}
+
+    ~TextureCache() = default;
+
+    virtual ResultType TryFastGetSurfaceView(TExecutionContext exctx, VAddr cpu_addr, u8* host_ptr,
+                                             const SurfaceParams& params, bool preserve_contents,
+                                             const std::vector<TSurface*>& overlaps) = 0;
+
+    virtual std::unique_ptr<TSurface> CreateSurface(const SurfaceParams& params) = 0;
+
+    void Register(TSurface* surface, VAddr cpu_addr, u8* host_ptr) {
+        surface->Register(cpu_addr, host_ptr);
+        registered_surfaces.add({GetSurfaceInterval(surface), {surface}});
+        rasterizer.UpdatePagesCachedCount(surface->GetCpuAddr(), surface->GetSizeInBytes(), 1);
+    }
+
+    void Unregister(TSurface* surface) {
+        registered_surfaces.subtract({GetSurfaceInterval(surface), {surface}});
+        rasterizer.UpdatePagesCachedCount(surface->GetCpuAddr(), surface->GetSizeInBytes(), -1);
+        surface->Unregister();
+    }
+
+    TSurface* GetUncachedSurface(const SurfaceParams& params) {
+        if (TSurface* surface = TryGetReservedSurface(params); surface)
+            return surface;
+        // No reserved surface available, create a new one and reserve it
+        auto new_surface{CreateSurface(params)};
+        TSurface* surface{new_surface.get()};
+        ReserveSurface(params, std::move(new_surface));
+        return surface;
+    }
+
+    Core::System& system;
+
+private:
+    ResultType GetSurfaceView(TExecutionContext exctx, VAddr cpu_addr, const SurfaceParams& params,
+                              bool preserve_contents) {
+        const auto host_ptr{Memory::GetPointer(cpu_addr)};
+        const auto cache_addr{ToCacheAddr(host_ptr)};
+        const auto overlaps{GetSurfacesInRegion(cache_addr, params.GetGuestSizeInBytes())};
+        if (overlaps.empty()) {
+            return LoadSurfaceView(exctx, cpu_addr, host_ptr, params, preserve_contents);
+        }
+
+        if (overlaps.size() == 1) {
+            if (TView* view = overlaps[0]->TryGetView(cpu_addr, params); view)
+                return {view, exctx};
+        }
+
+        TView* fast_view;
+        std::tie(fast_view, exctx) =
+            TryFastGetSurfaceView(exctx, cpu_addr, host_ptr, params, preserve_contents, overlaps);
+
+        for (TSurface* surface : overlaps) {
+            if (!fast_view) {
+                // Flush even when we don't care about the contents, to preserve memory not written
+                // by the new surface.
+                exctx = surface->FlushBuffer(exctx);
+            }
+            Unregister(surface);
+        }
+
+        if (fast_view) {
+            return {fast_view, exctx};
+        }
+
+        return LoadSurfaceView(exctx, cpu_addr, host_ptr, params, preserve_contents);
+    }
+
+    ResultType LoadSurfaceView(TExecutionContext exctx, VAddr cpu_addr, u8* host_ptr,
+                               const SurfaceParams& params, bool preserve_contents) {
+        TSurface* new_surface{GetUncachedSurface(params)};
+        Register(new_surface, cpu_addr, host_ptr);
+        if (preserve_contents) {
+            exctx = LoadSurface(exctx, new_surface);
+        }
+        return {new_surface->GetView(cpu_addr, params), exctx};
+    }
+
+    TExecutionContext LoadSurface(TExecutionContext exctx, TSurface* surface) {
+        surface->LoadBuffer();
+        exctx = surface->UploadTexture(exctx);
+        surface->MarkAsModified(false);
+        return exctx;
+    }
+
+    std::vector<TSurface*> GetSurfacesInRegion(CacheAddr cache_addr, std::size_t size) const {
+        if (size == 0) {
+            return {};
+        }
+        const IntervalType interval{cache_addr, cache_addr + size};
+
+        std::vector<TSurface*> surfaces;
+        for (auto& pair : boost::make_iterator_range(registered_surfaces.equal_range(interval))) {
+            surfaces.push_back(*pair.second.begin());
+        }
+        return surfaces;
+    }
+
+    void ReserveSurface(const SurfaceParams& params, std::unique_ptr<TSurface> surface) {
+        surface_reserve[params].push_back(std::move(surface));
+    }
+
+    TSurface* TryGetReservedSurface(const SurfaceParams& params) {
+        auto search{surface_reserve.find(params)};
+        if (search == surface_reserve.end()) {
+            return {};
+        }
+        for (auto& surface : search->second) {
+            if (!surface->IsRegistered()) {
+                return surface.get();
+            }
+        }
+        return {};
+    }
+
+    IntervalType GetSurfaceInterval(TSurface* surface) const {
+        return IntervalType::right_open(surface->GetCacheAddr(),
+                                        surface->GetCacheAddr() + surface->GetSizeInBytes());
+    }
+
+    VideoCore::RasterizerInterface& rasterizer;
+
+    IntervalMap registered_surfaces;
+
+    /// The surface reserve is a "backup" cache, this is where we put unique surfaces that have
+    /// previously been used. This is to prevent surfaces from being constantly created and
+    /// destroyed when used with different surface parameters.
+    std::unordered_map<SurfaceParams, std::list<std::unique_ptr<TSurface>>> surface_reserve;
+};
+
+} // namespace VideoCommon

src/yuzu/bootmanager.cpp

@@ -1,6 +1,13 @@
+// Copyright 2014 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
 #include <QApplication>
 #include <QHBoxLayout>
 #include <QKeyEvent>
+#include <QOffscreenSurface>
+#include <QOpenGLWindow>
+#include <QPainter>
 #include <QScreen>
 #include <QWindow>
 #include <fmt/format.h>
@@ -82,13 +89,36 @@ void EmuThread::run() {
     render_window->moveContext();
 }
 
+class GGLContext : public Core::Frontend::GraphicsContext {
+public:
+    explicit GGLContext(QOpenGLContext* shared_context) : surface() {
+        context = std::make_unique<QOpenGLContext>(shared_context);
+        surface.setFormat(shared_context->format());
+        surface.create();
+    }
+
+    void MakeCurrent() override {
+        context->makeCurrent(&surface);
+    }
+
+    void DoneCurrent() override {
+        context->doneCurrent();
+    }
+
+    void SwapBuffers() override {}
+
+private:
+    std::unique_ptr<QOpenGLContext> context;
+    QOffscreenSurface surface;
+};
+
 // This class overrides paintEvent and resizeEvent to prevent the GUI thread from stealing GL
 // context.
 // The corresponding functionality is handled in EmuThread instead
-class GGLWidgetInternal : public QGLWidget {
+class GGLWidgetInternal : public QOpenGLWindow {
 public:
-    GGLWidgetInternal(QGLFormat fmt, GRenderWindow* parent)
-        : QGLWidget(fmt, parent), parent(parent) {}
+    GGLWidgetInternal(GRenderWindow* parent, QOpenGLContext* shared_context)
+        : QOpenGLWindow(shared_context), parent(parent) {}
 
     void paintEvent(QPaintEvent* ev) override {
         if (do_painting) {
@@ -101,9 +131,51 @@ public:
         parent->OnFramebufferSizeChanged();
     }
 
+    void keyPressEvent(QKeyEvent* event) override {
+        InputCommon::GetKeyboard()->PressKey(event->key());
+    }
+
+    void keyReleaseEvent(QKeyEvent* event) override {
+        InputCommon::GetKeyboard()->ReleaseKey(event->key());
+    }
+
+    void mousePressEvent(QMouseEvent* event) override {
+        if (event->source() == Qt::MouseEventSynthesizedBySystem)
+            return; // touch input is handled in TouchBeginEvent
+
+        const auto pos{event->pos()};
+        if (event->button() == Qt::LeftButton) {
+            const auto [x, y] = parent->ScaleTouch(pos);
+            parent->TouchPressed(x, y);
+        } else if (event->button() == Qt::RightButton) {
+            InputCommon::GetMotionEmu()->BeginTilt(pos.x(), pos.y());
+        }
+    }
+
+    void mouseMoveEvent(QMouseEvent* event) override {
+        if (event->source() == Qt::MouseEventSynthesizedBySystem)
+            return; // touch input is handled in TouchUpdateEvent
+
+        const auto pos{event->pos()};
+        const auto [x, y] = parent->ScaleTouch(pos);
+        parent->TouchMoved(x, y);
+        InputCommon::GetMotionEmu()->Tilt(pos.x(), pos.y());
+    }
+
+    void mouseReleaseEvent(QMouseEvent* event) override {
+        if (event->source() == Qt::MouseEventSynthesizedBySystem)
+            return; // touch input is handled in TouchEndEvent
+
+        if (event->button() == Qt::LeftButton)
+            parent->TouchReleased();
+        else if (event->button() == Qt::RightButton)
+            InputCommon::GetMotionEmu()->EndTilt();
+    }
+
     void DisablePainting() {
         do_painting = false;
     }
+
     void EnablePainting() {
         do_painting = true;
     }
@@ -114,7 +186,7 @@ private:
 };
 
 GRenderWindow::GRenderWindow(QWidget* parent, EmuThread* emu_thread)
-    : QWidget(parent), child(nullptr), emu_thread(emu_thread) {
+    : QWidget(parent), child(nullptr), context(nullptr), emu_thread(emu_thread) {
 
     setWindowTitle(QStringLiteral("yuzu %1 | %2-%3")
                        .arg(Common::g_build_name, Common::g_scm_branch, Common::g_scm_desc));
@@ -137,19 +209,19 @@ void GRenderWindow::moveContext() {
     auto thread = (QThread::currentThread() == qApp->thread() && emu_thread != nullptr)
                       ? emu_thread
                       : qApp->thread();
-    child->context()->moveToThread(thread);
+    context->moveToThread(thread);
 }
 
 void GRenderWindow::SwapBuffers() {
-    // In our multi-threaded QGLWidget use case we shouldn't need to call `makeCurrent`,
+    // In our multi-threaded QWidget use case we shouldn't need to call `makeCurrent`,
     // since we never call `doneCurrent` in this thread.
     // However:
     // - The Qt debug runtime prints a bogus warning on the console if `makeCurrent` wasn't called
     // since the last time `swapBuffers` was executed;
     // - On macOS, if `makeCurrent` isn't called explicitely, resizing the buffer breaks.
-    child->makeCurrent();
+    context->makeCurrent(child);
 
-    child->swapBuffers();
+    context->swapBuffers(child);
     if (!first_frame) {
         emit FirstFrameDisplayed();
         first_frame = true;
@@ -157,11 +229,11 @@ void GRenderWindow::SwapBuffers() {
 }
 
 void GRenderWindow::MakeCurrent() {
-    child->makeCurrent();
+    context->makeCurrent(child);
 }
 
 void GRenderWindow::DoneCurrent() {
-    child->doneCurrent();
+    context->doneCurrent();
 }
 
 void GRenderWindow::PollEvents() {}
@@ -174,14 +246,26 @@ void GRenderWindow::PollEvents() {}
 void GRenderWindow::OnFramebufferSizeChanged() {
     // Screen changes potentially incur a change in screen DPI, hence we should update the
     // framebuffer size
-    qreal pixelRatio = windowPixelRatio();
+    qreal pixelRatio = GetWindowPixelRatio();
     unsigned width = child->QPaintDevice::width() * pixelRatio;
     unsigned height = child->QPaintDevice::height() * pixelRatio;
     UpdateCurrentFramebufferLayout(width, height);
 }
 
+void GRenderWindow::ForwardKeyPressEvent(QKeyEvent* event) {
+    if (child) {
+        child->keyPressEvent(event);
+    }
+}
+
+void GRenderWindow::ForwardKeyReleaseEvent(QKeyEvent* event) {
+    if (child) {
+        child->keyReleaseEvent(event);
+    }
+}
+
 void GRenderWindow::BackupGeometry() {
-    geometry = ((QGLWidget*)this)->saveGeometry();
+    geometry = ((QWidget*)this)->saveGeometry();
 }
 
 void GRenderWindow::RestoreGeometry() {
@@ -199,18 +283,18 @@ QByteArray GRenderWindow::saveGeometry() {
     // If we are a top-level widget, store the current geometry
     // otherwise, store the last backup
     if (parent() == nullptr)
-        return ((QGLWidget*)this)->saveGeometry();
+        return ((QWidget*)this)->saveGeometry();
     else
         return geometry;
 }
 
-qreal GRenderWindow::windowPixelRatio() const {
+qreal GRenderWindow::GetWindowPixelRatio() const {
     // windowHandle() might not be accessible until the window is displayed to screen.
     return windowHandle() ? windowHandle()->screen()->devicePixelRatio() : 1.0f;
 }
 
 std::pair<unsigned, unsigned> GRenderWindow::ScaleTouch(const QPointF pos) const {
-    const qreal pixel_ratio = windowPixelRatio();
+    const qreal pixel_ratio = GetWindowPixelRatio();
     return {static_cast<unsigned>(std::max(std::round(pos.x() * pixel_ratio), qreal{0.0})),
             static_cast<unsigned>(std::max(std::round(pos.y() * pixel_ratio), qreal{0.0}))};
 }
@@ -220,47 +304,6 @@ void GRenderWindow::closeEvent(QCloseEvent* event) {
     QWidget::closeEvent(event);
 }
 
-void GRenderWindow::keyPressEvent(QKeyEvent* event) {
-    InputCommon::GetKeyboard()->PressKey(event->key());
-}
-
-void GRenderWindow::keyReleaseEvent(QKeyEvent* event) {
-    InputCommon::GetKeyboard()->ReleaseKey(event->key());
-}
-
-void GRenderWindow::mousePressEvent(QMouseEvent* event) {
-    if (event->source() == Qt::MouseEventSynthesizedBySystem)
-        return; // touch input is handled in TouchBeginEvent
-
-    auto pos = event->pos();
-    if (event->button() == Qt::LeftButton) {
-        const auto [x, y] = ScaleTouch(pos);
-        this->TouchPressed(x, y);
-    } else if (event->button() == Qt::RightButton) {
-        InputCommon::GetMotionEmu()->BeginTilt(pos.x(), pos.y());
-    }
-}
-
-void GRenderWindow::mouseMoveEvent(QMouseEvent* event) {
-    if (event->source() == Qt::MouseEventSynthesizedBySystem)
-        return; // touch input is handled in TouchUpdateEvent
-
-    auto pos = event->pos();
-    const auto [x, y] = ScaleTouch(pos);
-    this->TouchMoved(x, y);
-    InputCommon::GetMotionEmu()->Tilt(pos.x(), pos.y());
-}
-
-void GRenderWindow::mouseReleaseEvent(QMouseEvent* event) {
-    if (event->source() == Qt::MouseEventSynthesizedBySystem)
-        return; // touch input is handled in TouchEndEvent
-
-    if (event->button() == Qt::LeftButton)
-        this->TouchReleased();
-    else if (event->button() == Qt::RightButton)
-        InputCommon::GetMotionEmu()->EndTilt();
-}
-
 void GRenderWindow::TouchBeginEvent(const QTouchEvent* event) {
     // TouchBegin always has exactly one touch point, so take the .first()
     const auto [x, y] = ScaleTouch(event->touchPoints().first().pos());
@@ -313,35 +356,60 @@ void GRenderWindow::OnClientAreaResized(unsigned width, unsigned height) {
     NotifyClientAreaSizeChanged(std::make_pair(width, height));
 }
 
-void GRenderWindow::InitRenderTarget() {
-    if (child) {
-        delete child;
-    }
+std::unique_ptr<Core::Frontend::GraphicsContext> GRenderWindow::CreateSharedContext() const {
+    return std::make_unique<GGLContext>(shared_context.get());
+}
 
-    if (layout()) {
-        delete layout();
-    }
+void GRenderWindow::InitRenderTarget() {
+    shared_context.reset();
+    context.reset();
+
+    delete child;
+    child = nullptr;
+
+    delete container;
+    container = nullptr;
+
+    delete layout();
 
     first_frame = false;
 
     // TODO: One of these flags might be interesting: WA_OpaquePaintEvent, WA_NoBackground,
     // WA_DontShowOnScreen, WA_DeleteOnClose
-    QGLFormat fmt;
+    QSurfaceFormat fmt;
     fmt.setVersion(4, 3);
-    fmt.setProfile(QGLFormat::CoreProfile);
+    fmt.setProfile(QSurfaceFormat::CoreProfile);
+    // TODO: expose a setting for buffer value (ie default/single/double/triple)
+    fmt.setSwapBehavior(QSurfaceFormat::DefaultSwapBehavior);
+    shared_context = std::make_unique<QOpenGLContext>();
+    shared_context->setFormat(fmt);
+    shared_context->create();
+    context = std::make_unique<QOpenGLContext>();
+    context->setShareContext(shared_context.get());
+    context->setFormat(fmt);
+    context->create();
     fmt.setSwapInterval(false);
 
-    // Requests a forward-compatible context, which is required to get a 3.2+ context on OS X
-    fmt.setOption(QGL::NoDeprecatedFunctions);
+    child = new GGLWidgetInternal(this, shared_context.get());
+    container = QWidget::createWindowContainer(child, this);
 
-    child = new GGLWidgetInternal(fmt, this);
     QBoxLayout* layout = new QHBoxLayout(this);
-
-    resize(Layout::ScreenUndocked::Width, Layout::ScreenUndocked::Height);
-    layout->addWidget(child);
+    layout->addWidget(container);
     layout->setMargin(0);
     setLayout(layout);
 
+    // Reset minimum size to avoid unwanted resizes when this function is called for a second time.
+    setMinimumSize(1, 1);
+
+    // Show causes the window to actually be created and the OpenGL context as well, but we don't
+    // want the widget to be shown yet, so immediately hide it.
+    show();
+    hide();
+
+    resize(Layout::ScreenUndocked::Width, Layout::ScreenUndocked::Height);
+    child->resize(Layout::ScreenUndocked::Width, Layout::ScreenUndocked::Height);
+    container->resize(Layout::ScreenUndocked::Width, Layout::ScreenUndocked::Height);
+
     OnMinimalClientAreaChangeRequest(GetActiveConfig().min_client_area_size);
 
     OnFramebufferSizeChanged();

src/yuzu/bootmanager.h

@@ -7,9 +7,9 @@
 #include <atomic>
 #include <condition_variable>
 #include <mutex>
-#include <QGLWidget>
 #include <QImage>
 #include <QThread>
+#include <QWidget>
 #include "common/thread.h"
 #include "core/core.h"
 #include "core/frontend/emu_window.h"
@@ -21,6 +21,8 @@ class QTouchEvent;
 class GGLWidgetInternal;
 class GMainWindow;
 class GRenderWindow;
+class QSurface;
+class QOpenGLContext;
 
 namespace VideoCore {
 enum class LoadCallbackStage;
@@ -121,25 +123,21 @@ public:
     void MakeCurrent() override;
     void DoneCurrent() override;
     void PollEvents() override;
+    std::unique_ptr<Core::Frontend::GraphicsContext> CreateSharedContext() const override;
+
+    void ForwardKeyPressEvent(QKeyEvent* event);
+    void ForwardKeyReleaseEvent(QKeyEvent* event);
 
     void BackupGeometry();
     void RestoreGeometry();
     void restoreGeometry(const QByteArray& geometry); // overridden
     QByteArray saveGeometry();                        // overridden
 
-    qreal windowPixelRatio() const;
+    qreal GetWindowPixelRatio() const;
+    std::pair<unsigned, unsigned> ScaleTouch(const QPointF pos) const;
 
     void closeEvent(QCloseEvent* event) override;
-
-    void keyPressEvent(QKeyEvent* event) override;
-    void keyReleaseEvent(QKeyEvent* event) override;
-
-    void mousePressEvent(QMouseEvent* event) override;
-    void mouseMoveEvent(QMouseEvent* event) override;
-    void mouseReleaseEvent(QMouseEvent* event) override;
-
     bool event(QEvent* event) override;
-
     void focusOutEvent(QFocusEvent* event) override;
 
     void OnClientAreaResized(unsigned width, unsigned height);
@@ -161,7 +159,6 @@ signals:
     void FirstFrameDisplayed();
 
 private:
-    std::pair<unsigned, unsigned> ScaleTouch(const QPointF pos) const;
     void TouchBeginEvent(const QTouchEvent* event);
     void TouchUpdateEvent(const QTouchEvent* event);
     void TouchEndEvent();
@@ -169,11 +166,17 @@ private:
     void OnMinimalClientAreaChangeRequest(
         const std::pair<unsigned, unsigned>& minimal_size) override;
 
-    GGLWidgetInternal* child;
+    QWidget* container = nullptr;
+    GGLWidgetInternal* child = nullptr;
 
     QByteArray geometry;
 
     EmuThread* emu_thread;
+    // Context that backs the GGLWidgetInternal (and will be used by core to render)
+    std::unique_ptr<QOpenGLContext> context;
+    // Context that will be shared between all newly created contexts. This should never be made
+    // current
+    std::unique_ptr<QOpenGLContext> shared_context;
 
     /// Temporary storage of the screenshot taken
     QImage screenshot_image;

src/yuzu/configuration/config.cpp

@@ -394,6 +394,7 @@ void Config::ReadValues() {
         ReadSetting("use_accurate_gpu_emulation", false).toBool();
     Settings::values.use_asynchronous_gpu_emulation =
         ReadSetting("use_asynchronous_gpu_emulation", false).toBool();
+    Settings::values.force_30fps_mode = ReadSetting("force_30fps_mode", false).toBool();
 
     Settings::values.bg_red = ReadSetting("bg_red", 0.0).toFloat();
     Settings::values.bg_green = ReadSetting("bg_green", 0.0).toFloat();
@@ -523,8 +524,8 @@ void Config::ReadValues() {
     qt_config->beginGroup("Paths");
     UISettings::values.roms_path = ReadSetting("romsPath").toString();
     UISettings::values.symbols_path = ReadSetting("symbolsPath").toString();
-    UISettings::values.gamedir = ReadSetting("gameListRootDir", ".").toString();
-    UISettings::values.gamedir_deepscan = ReadSetting("gameListDeepScan", false).toBool();
+    UISettings::values.game_directory_path = ReadSetting("gameListRootDir", ".").toString();
+    UISettings::values.game_directory_deepscan = ReadSetting("gameListDeepScan", false).toBool();
     UISettings::values.recent_files = ReadSetting("recentFiles").toStringList();
     qt_config->endGroup();
 
@@ -664,6 +665,7 @@ void Config::SaveValues() {
     WriteSetting("use_accurate_gpu_emulation", Settings::values.use_accurate_gpu_emulation, false);
     WriteSetting("use_asynchronous_gpu_emulation", Settings::values.use_asynchronous_gpu_emulation,
                  false);
+    WriteSetting("force_30fps_mode", Settings::values.force_30fps_mode, false);
 
     // Cast to double because Qt's written float values are not human-readable
     WriteSetting("bg_red", (double)Settings::values.bg_red, 0.0);
@@ -768,8 +770,8 @@ void Config::SaveValues() {
     WriteSetting("romsPath", UISettings::values.roms_path);
     WriteSetting("symbolsPath", UISettings::values.symbols_path);
     WriteSetting("screenshotPath", UISettings::values.screenshot_path);
-    WriteSetting("gameListRootDir", UISettings::values.gamedir, ".");
-    WriteSetting("gameListDeepScan", UISettings::values.gamedir_deepscan, false);
+    WriteSetting("gameListRootDir", UISettings::values.game_directory_path, ".");
+    WriteSetting("gameListDeepScan", UISettings::values.game_directory_deepscan, false);
     WriteSetting("recentFiles", UISettings::values.recent_files);
     qt_config->endGroup();
 

src/yuzu/configuration/configure_general.cpp

@@ -28,7 +28,7 @@ ConfigureGeneral::ConfigureGeneral(QWidget* parent)
 ConfigureGeneral::~ConfigureGeneral() = default;
 
 void ConfigureGeneral::setConfiguration() {
-    ui->toggle_deepscan->setChecked(UISettings::values.gamedir_deepscan);
+    ui->toggle_deepscan->setChecked(UISettings::values.game_directory_deepscan);
     ui->toggle_check_exit->setChecked(UISettings::values.confirm_before_closing);
     ui->toggle_user_on_boot->setChecked(UISettings::values.select_user_on_boot);
     ui->theme_combobox->setCurrentIndex(ui->theme_combobox->findData(UISettings::values.theme));
@@ -36,7 +36,7 @@ void ConfigureGeneral::setConfiguration() {
 }
 
 void ConfigureGeneral::applyConfiguration() {
-    UISettings::values.gamedir_deepscan = ui->toggle_deepscan->isChecked();
+    UISettings::values.game_directory_deepscan = ui->toggle_deepscan->isChecked();
     UISettings::values.confirm_before_closing = ui->toggle_check_exit->isChecked();
     UISettings::values.select_user_on_boot = ui->toggle_user_on_boot->isChecked();
     UISettings::values.theme =

src/yuzu/configuration/configure_graphics.cpp

@@ -77,6 +77,8 @@ void ConfigureGraphics::setConfiguration() {
     ui->use_accurate_gpu_emulation->setChecked(Settings::values.use_accurate_gpu_emulation);
     ui->use_asynchronous_gpu_emulation->setEnabled(!Core::System::GetInstance().IsPoweredOn());
     ui->use_asynchronous_gpu_emulation->setChecked(Settings::values.use_asynchronous_gpu_emulation);
+    ui->force_30fps_mode->setEnabled(!Core::System::GetInstance().IsPoweredOn());
+    ui->force_30fps_mode->setChecked(Settings::values.force_30fps_mode);
     UpdateBackgroundColorButton(QColor::fromRgbF(Settings::values.bg_red, Settings::values.bg_green,
                                                  Settings::values.bg_blue));
 }
@@ -90,6 +92,7 @@ void ConfigureGraphics::applyConfiguration() {
     Settings::values.use_accurate_gpu_emulation = ui->use_accurate_gpu_emulation->isChecked();
     Settings::values.use_asynchronous_gpu_emulation =
         ui->use_asynchronous_gpu_emulation->isChecked();
+    Settings::values.force_30fps_mode = ui->force_30fps_mode->isChecked();
     Settings::values.bg_red = static_cast<float>(bg_color.redF());
     Settings::values.bg_green = static_cast<float>(bg_color.greenF());
     Settings::values.bg_blue = static_cast<float>(bg_color.blueF());

src/yuzu/configuration/configure_graphics.ui

@@ -70,6 +70,13 @@
           </property>
          </widget>
         </item>
+        <item>
+         <widget class="QCheckBox" name="force_30fps_mode">
+          <property name="text">
+           <string>Force 30 FPS mode</string>
+          </property>
+         </widget>
+        </item>
         <item>
          <layout class="QHBoxLayout" name="horizontalLayout">
           <item>

src/yuzu/game_list.cpp

@@ -467,9 +467,10 @@ void GameList::LoadInterfaceLayout() {
 const QStringList GameList::supported_file_extensions = {"nso", "nro", "nca", "xci", "nsp"};
 
 void GameList::RefreshGameDirectory() {
-    if (!UISettings::values.gamedir.isEmpty() && current_worker != nullptr) {
+    if (!UISettings::values.game_directory_path.isEmpty() && current_worker != nullptr) {
         LOG_INFO(Frontend, "Change detected in the games directory. Reloading game list.");
         search_field->clear();
-        PopulateAsync(UISettings::values.gamedir, UISettings::values.gamedir_deepscan);
+        PopulateAsync(UISettings::values.game_directory_path,
+                      UISettings::values.game_directory_deepscan);
     }
 }

src/yuzu/main.cpp

@@ -213,7 +213,8 @@ GMainWindow::GMainWindow()
     OnReinitializeKeys(ReinitializeKeyBehavior::NoWarning);
 
     game_list->LoadCompatibilityList();
-    game_list->PopulateAsync(UISettings::values.gamedir, UISettings::values.gamedir_deepscan);
+    game_list->PopulateAsync(UISettings::values.game_directory_path,
+                             UISettings::values.game_directory_deepscan);
 
     // Show one-time "callout" messages to the user
     ShowTelemetryCallout();
@@ -1278,8 +1279,8 @@ void GMainWindow::OnGameListOpenPerGameProperties(const std::string& file) {
 
         const auto reload = UISettings::values.is_game_list_reload_pending.exchange(false);
         if (reload) {
-            game_list->PopulateAsync(UISettings::values.gamedir,
-                                     UISettings::values.gamedir_deepscan);
+            game_list->PopulateAsync(UISettings::values.game_directory_path,
+                                     UISettings::values.game_directory_deepscan);
         }
 
         config->Save();
@@ -1367,7 +1368,8 @@ void GMainWindow::OnMenuInstallToNAND() {
     const auto success = [this]() {
         QMessageBox::information(this, tr("Successfully Installed"),
                                  tr("The file was successfully installed."));
-        game_list->PopulateAsync(UISettings::values.gamedir, UISettings::values.gamedir_deepscan);
+        game_list->PopulateAsync(UISettings::values.game_directory_path,
+                                 UISettings::values.game_directory_deepscan);
     };
 
     const auto failed = [this]() {
@@ -1494,8 +1496,8 @@ void GMainWindow::OnMenuInstallToNAND() {
 void GMainWindow::OnMenuSelectGameListRoot() {
     QString dir_path = QFileDialog::getExistingDirectory(this, tr("Select Directory"));
     if (!dir_path.isEmpty()) {
-        UISettings::values.gamedir = dir_path;
-        game_list->PopulateAsync(dir_path, UISettings::values.gamedir_deepscan);
+        UISettings::values.game_directory_path = dir_path;
+        game_list->PopulateAsync(dir_path, UISettings::values.game_directory_deepscan);
     }
 }
 
@@ -1517,7 +1519,8 @@ void GMainWindow::OnMenuSelectEmulatedDirectory(EmulatedDirectoryTarget target)
                                                                       : FileUtil::UserPath::NANDDir,
                               dir_path.toStdString());
         Service::FileSystem::CreateFactories(*vfs);
-        game_list->PopulateAsync(UISettings::values.gamedir, UISettings::values.gamedir_deepscan);
+        game_list->PopulateAsync(UISettings::values.game_directory_path,
+                                 UISettings::values.game_directory_deepscan);
     }
 }
 
@@ -1669,8 +1672,8 @@ void GMainWindow::OnConfigure() {
 
         const auto reload = UISettings::values.is_game_list_reload_pending.exchange(false);
         if (reload) {
-            game_list->PopulateAsync(UISettings::values.gamedir,
-                                     UISettings::values.gamedir_deepscan);
+            game_list->PopulateAsync(UISettings::values.game_directory_path,
+                                     UISettings::values.game_directory_deepscan);
         }
 
         config->Save();
@@ -1920,7 +1923,8 @@ void GMainWindow::OnReinitializeKeys(ReinitializeKeyBehavior behavior) {
     Service::FileSystem::CreateFactories(*vfs);
 
     if (behavior == ReinitializeKeyBehavior::Warning) {
-        game_list->PopulateAsync(UISettings::values.gamedir, UISettings::values.gamedir_deepscan);
+        game_list->PopulateAsync(UISettings::values.game_directory_path,
+                                 UISettings::values.game_directory_deepscan);
     }
 }
 
@@ -2027,6 +2031,18 @@ void GMainWindow::dragMoveEvent(QDragMoveEvent* event) {
     event->acceptProposedAction();
 }
 
+void GMainWindow::keyPressEvent(QKeyEvent* event) {
+    if (render_window) {
+        render_window->ForwardKeyPressEvent(event);
+    }
+}
+
+void GMainWindow::keyReleaseEvent(QKeyEvent* event) {
+    if (render_window) {
+        render_window->ForwardKeyReleaseEvent(event);
+    }
+}
+
 bool GMainWindow::ConfirmChangeGame() {
     if (emu_thread == nullptr)
         return true;
@@ -2094,7 +2110,8 @@ int main(int argc, char* argv[]) {
     QCoreApplication::setOrganizationName("yuzu team");
     QCoreApplication::setApplicationName("yuzu");
 
-    QApplication::setAttribute(Qt::AA_DontCheckOpenGLContextThreadAffinity);
+    // Enables the core to make the qt created contexts current on std::threads
+    QCoreApplication::setAttribute(Qt::AA_DontCheckOpenGLContextThreadAffinity);
     QApplication app(argc, argv);
 
     // Qt changes the locale and causes issues in float conversion using std::to_string() when

src/yuzu/main.h

@@ -252,4 +252,8 @@ protected:
     void dropEvent(QDropEvent* event) override;
     void dragEnterEvent(QDragEnterEvent* event) override;
     void dragMoveEvent(QDragMoveEvent* event) override;
+
+    // Overrides used to forward signals to the render window when the focus moves out.
+    void keyPressEvent(QKeyEvent* event) override;
+    void keyReleaseEvent(QKeyEvent* event) override;
 };

src/yuzu/ui_settings.h

@@ -55,8 +55,8 @@ struct Values {
     QString roms_path;
     QString symbols_path;
     QString screenshot_path;
-    QString gamedir;
-    bool gamedir_deepscan;
+    QString game_directory_path;
+    bool game_directory_deepscan;
     QStringList recent_files;
 
     QString theme;

src/yuzu_cmd/emu_window/emu_window_sdl2.cpp

@@ -19,6 +19,37 @@
 #include "input_common/sdl/sdl.h"
 #include "yuzu_cmd/emu_window/emu_window_sdl2.h"
 
+class SDLGLContext : public Core::Frontend::GraphicsContext {
+public:
+    explicit SDLGLContext() {
+        // create a hidden window to make the shared context against
+        window = SDL_CreateWindow("", SDL_WINDOWPOS_UNDEFINED, // x position
+                                  SDL_WINDOWPOS_UNDEFINED,     // y position
+                                  Layout::ScreenUndocked::Width, Layout::ScreenUndocked::Height,
+                                  SDL_WINDOW_OPENGL | SDL_WINDOW_HIDDEN);
+        context = SDL_GL_CreateContext(window);
+    }
+
+    ~SDLGLContext() {
+        SDL_GL_DeleteContext(context);
+        SDL_DestroyWindow(window);
+    }
+
+    void MakeCurrent() override {
+        SDL_GL_MakeCurrent(window, context);
+    }
+
+    void DoneCurrent() override {
+        SDL_GL_MakeCurrent(window, nullptr);
+    }
+
+    void SwapBuffers() override {}
+
+private:
+    SDL_Window* window;
+    SDL_GLContext context;
+};
+
 void EmuWindow_SDL2::OnMouseMotion(s32 x, s32 y) {
     TouchMoved((unsigned)std::max(x, 0), (unsigned)std::max(y, 0));
     InputCommon::GetMotionEmu()->Tilt(x, y);
@@ -153,6 +184,7 @@ EmuWindow_SDL2::EmuWindow_SDL2(bool fullscreen) {
     SDL_GL_SetAttribute(SDL_GL_GREEN_SIZE, 8);
     SDL_GL_SetAttribute(SDL_GL_BLUE_SIZE, 8);
     SDL_GL_SetAttribute(SDL_GL_ALPHA_SIZE, 0);
+    SDL_GL_SetAttribute(SDL_GL_SHARE_WITH_CURRENT_CONTEXT, 1);
 
     std::string window_title = fmt::format("yuzu {} | {}-{}", Common::g_build_fullname,
                                            Common::g_scm_branch, Common::g_scm_desc);
@@ -171,7 +203,6 @@ EmuWindow_SDL2::EmuWindow_SDL2(bool fullscreen) {
     if (fullscreen) {
         Fullscreen();
     }
-
     gl_context = SDL_GL_CreateContext(render_window);
 
     if (gl_context == nullptr) {
@@ -278,3 +309,7 @@ void EmuWindow_SDL2::OnMinimalClientAreaChangeRequest(
 
     SDL_SetWindowMinimumSize(render_window, minimal_size.first, minimal_size.second);
 }
+
+std::unique_ptr<Core::Frontend::GraphicsContext> EmuWindow_SDL2::CreateSharedContext() const {
+    return std::make_unique<SDLGLContext>();
+}

src/yuzu_cmd/emu_window/emu_window_sdl2.h

@@ -27,6 +27,8 @@ public:
     /// Releases the GL context from the caller thread
     void DoneCurrent() override;
 
+    std::unique_ptr<Core::Frontend::GraphicsContext> CreateSharedContext() const override;
+
     /// Whether the window is still open, and a close request hasn't yet been sent
     bool IsOpen() const;