loader: nro: Fix process initialization using ProgramMetadata default.

- We need this for homebrew process initialization.

CMakeLists.txt

@@ -164,7 +164,7 @@ if (ENABLE_SDL2)
             set(SDL2_LIBRARIES "SDL2::SDL2")
         endif()
 
-        include_directories(${SDL2_INCLUDE_DIRS})
+        include_directories(SYSTEM ${SDL2_INCLUDE_DIRS})
         add_library(SDL2 INTERFACE)
         target_link_libraries(SDL2 INTERFACE "${SDL2_LIBRARIES}")
     endif()

externals/CMakeLists.txt

@@ -16,11 +16,11 @@ add_library(catch-single-include INTERFACE)
 target_include_directories(catch-single-include INTERFACE catch/single_include)
 
 # libfmt
-pkg_check_modules(FMT IMPORTED_TARGET GLOBAL fmt>=6.1.0)
+pkg_check_modules(FMT IMPORTED_TARGET GLOBAL fmt>=6.2.0)
 if (FMT_FOUND)
     add_library(fmt::fmt ALIAS PkgConfig::FMT)
 else()
-    message(STATUS "fmt 6.1.0 or newer not found, falling back to externals")
+    message(STATUS "fmt 6.2.0 or newer not found, falling back to externals")
     add_subdirectory(fmt)
     add_library(fmt::fmt ALIAS fmt)
 endif()

src/common/file_util.cpp

@@ -888,7 +888,14 @@ std::string SanitizePath(std::string_view path_, DirectorySeparator directory_se
     }
 
     std::replace(path.begin(), path.end(), type1, type2);
-    path.erase(std::unique(path.begin(), path.end(),
+
+    auto start = path.begin();
+#ifdef _WIN32
+    // allow network paths which start with a double backslash (e.g. \\server\share)
+    if (start != path.end())
+        ++start;
+#endif
+    path.erase(std::unique(start, path.end(),
                            [type2](char c1, char c2) { return c1 == type2 && c2 == type2; }),
                path.end());
     return std::string(RemoveTrailingSlash(path));
@@ -967,6 +974,34 @@ bool IOFile::Flush() {
     return IsOpen() && 0 == std::fflush(m_file);
 }
 
+std::size_t IOFile::ReadImpl(void* data, std::size_t length, std::size_t data_size) const {
+    if (!IsOpen()) {
+        return std::numeric_limits<std::size_t>::max();
+    }
+
+    if (length == 0) {
+        return 0;
+    }
+
+    DEBUG_ASSERT(data != nullptr);
+
+    return std::fread(data, data_size, length, m_file);
+}
+
+std::size_t IOFile::WriteImpl(const void* data, std::size_t length, std::size_t data_size) {
+    if (!IsOpen()) {
+        return std::numeric_limits<std::size_t>::max();
+    }
+
+    if (length == 0) {
+        return 0;
+    }
+
+    DEBUG_ASSERT(data != nullptr);
+
+    return std::fwrite(data, data_size, length, m_file);
+}
+
 bool IOFile::Resize(u64 size) {
     return IsOpen() && 0 ==
 #ifdef _WIN32

src/common/file_util.h

@@ -222,22 +222,15 @@ public:
         static_assert(std::is_trivially_copyable_v<T>,
                       "Given array does not consist of trivially copyable objects");
 
-        if (!IsOpen()) {
-            return std::numeric_limits<std::size_t>::max();
-        }
-
-        return std::fread(data, sizeof(T), length, m_file);
+        return ReadImpl(data, length, sizeof(T));
     }
 
     template <typename T>
     std::size_t WriteArray(const T* data, std::size_t length) {
         static_assert(std::is_trivially_copyable_v<T>,
                       "Given array does not consist of trivially copyable objects");
-        if (!IsOpen()) {
-            return std::numeric_limits<std::size_t>::max();
-        }
 
-        return std::fwrite(data, sizeof(T), length, m_file);
+        return WriteImpl(data, length, sizeof(T));
     }
 
     template <typename T>
@@ -278,6 +271,9 @@ public:
     }
 
 private:
+    std::size_t ReadImpl(void* data, std::size_t length, std::size_t data_size) const;
+    std::size_t WriteImpl(const void* data, std::size_t length, std::size_t data_size);
+
     std::FILE* m_file = nullptr;
 };
 

src/core/arm/dynarmic/arm_dynarmic_64.cpp

@@ -20,6 +20,7 @@
 #include "core/hle/kernel/scheduler.h"
 #include "core/hle/kernel/svc.h"
 #include "core/memory.h"
+#include "core/settings.h"
 
 namespace Core {
 
@@ -144,6 +145,8 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable&
     config.page_table_address_space_bits = address_space_bits;
     config.silently_mirror_page_table = false;
     config.absolute_offset_page_table = true;
+    config.detect_misaligned_access_via_page_table = 16 | 32 | 64 | 128;
+    config.only_detect_misalignment_via_page_table_on_page_boundary = true;
 
     // Multi-process state
     config.processor_id = core_index;
@@ -159,8 +162,11 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable&
     // Unpredictable instructions
     config.define_unpredictable_behaviour = true;
 
-    config.detect_misaligned_access_via_page_table = 16 | 32 | 64 | 128;
-    config.only_detect_misalignment_via_page_table_on_page_boundary = true;
+    // Optimizations
+    if (Settings::values.disable_cpu_opt) {
+        config.enable_optimizations = false;
+        config.enable_fast_dispatch = false;
+    }
 
     return std::make_shared<Dynarmic::A64::Jit>(config);
 }

src/core/file_sys/program_metadata.cpp

@@ -51,6 +51,17 @@ Loader::ResultStatus ProgramMetadata::Load(VirtualFile file) {
     return Loader::ResultStatus::Success;
 }
 
+/*static*/ ProgramMetadata ProgramMetadata::GetDefault() {
+    ProgramMetadata result;
+
+    result.LoadManual(
+        true /*is_64_bit*/, FileSys::ProgramAddressSpaceType::Is39Bit /*address_space*/,
+        0x2c /*main_thread_prio*/, 0 /*main_thread_core*/, 0x00100000 /*main_thread_stack_size*/,
+        {}, 0xFFFFFFFFFFFFFFFF /*filesystem_permissions*/, {} /*capabilities*/);
+
+    return result;
+}
+
 void ProgramMetadata::LoadManual(bool is_64_bit, ProgramAddressSpaceType address_space,
                                  s32 main_thread_prio, u32 main_thread_core,
                                  u32 main_thread_stack_size, u64 title_id,

src/core/file_sys/program_metadata.h

@@ -44,9 +44,13 @@ public:
     ProgramMetadata();
     ~ProgramMetadata();
 
+    /// Gets a default ProgramMetadata configuration, should only be used for homebrew formats where
+    /// we do not have an NPDM file
+    static ProgramMetadata GetDefault();
+
     Loader::ResultStatus Load(VirtualFile file);
 
-    // Load from parameters instead of NPDM file, used for KIP
+    /// Load from parameters instead of NPDM file, used for KIP
     void LoadManual(bool is_64_bit, ProgramAddressSpaceType address_space, s32 main_thread_prio,
                     u32 main_thread_core, u32 main_thread_stack_size, u64 title_id,
                     u64 filesystem_permissions, KernelCapabilityDescriptors capabilities);

src/core/hle/kernel/hle_ipc.cpp

@@ -282,19 +282,19 @@ ResultCode HLERequestContext::WriteToOutgoingCommandBuffer(Thread& thread) {
     return RESULT_SUCCESS;
 }
 
-std::vector<u8> HLERequestContext::ReadBuffer(int buffer_index) const {
+std::vector<u8> HLERequestContext::ReadBuffer(std::size_t buffer_index) const {
     std::vector<u8> buffer;
-    const bool is_buffer_a{BufferDescriptorA().size() > std::size_t(buffer_index) &&
+    const bool is_buffer_a{BufferDescriptorA().size() > buffer_index &&
                            BufferDescriptorA()[buffer_index].Size()};
     auto& memory = Core::System::GetInstance().Memory();
 
     if (is_buffer_a) {
-        ASSERT_MSG(BufferDescriptorA().size() > std::size_t(buffer_index),
+        ASSERT_MSG(BufferDescriptorA().size() > buffer_index,
                    "BufferDescriptorA invalid buffer_index {}", buffer_index);
         buffer.resize(BufferDescriptorA()[buffer_index].Size());
         memory.ReadBlock(BufferDescriptorA()[buffer_index].Address(), buffer.data(), buffer.size());
     } else {
-        ASSERT_MSG(BufferDescriptorX().size() > std::size_t(buffer_index),
+        ASSERT_MSG(BufferDescriptorX().size() > buffer_index,
                    "BufferDescriptorX invalid buffer_index {}", buffer_index);
         buffer.resize(BufferDescriptorX()[buffer_index].Size());
         memory.ReadBlock(BufferDescriptorX()[buffer_index].Address(), buffer.data(), buffer.size());
@@ -304,13 +304,13 @@ std::vector<u8> HLERequestContext::ReadBuffer(int buffer_index) const {
 }
 
 std::size_t HLERequestContext::WriteBuffer(const void* buffer, std::size_t size,
-                                           int buffer_index) const {
+                                           std::size_t buffer_index) const {
     if (size == 0) {
         LOG_WARNING(Core, "skip empty buffer write");
         return 0;
     }
 
-    const bool is_buffer_b{BufferDescriptorB().size() > std::size_t(buffer_index) &&
+    const bool is_buffer_b{BufferDescriptorB().size() > buffer_index &&
                            BufferDescriptorB()[buffer_index].Size()};
     const std::size_t buffer_size{GetWriteBufferSize(buffer_index)};
     if (size > buffer_size) {
@@ -321,13 +321,13 @@ std::size_t HLERequestContext::WriteBuffer(const void* buffer, std::size_t size,
 
     auto& memory = Core::System::GetInstance().Memory();
     if (is_buffer_b) {
-        ASSERT_MSG(BufferDescriptorB().size() > std::size_t(buffer_index),
+        ASSERT_MSG(BufferDescriptorB().size() > buffer_index,
                    "BufferDescriptorB invalid buffer_index {}", buffer_index);
         ASSERT_MSG(BufferDescriptorB()[buffer_index].Size() >= size,
                    "BufferDescriptorB buffer_index {} is not large enough", buffer_index);
         memory.WriteBlock(BufferDescriptorB()[buffer_index].Address(), buffer, size);
     } else {
-        ASSERT_MSG(BufferDescriptorC().size() > std::size_t(buffer_index),
+        ASSERT_MSG(BufferDescriptorC().size() > buffer_index,
                    "BufferDescriptorC invalid buffer_index {}", buffer_index);
         ASSERT_MSG(BufferDescriptorC()[buffer_index].Size() >= size,
                    "BufferDescriptorC buffer_index {} is not large enough", buffer_index);
@@ -337,17 +337,17 @@ std::size_t HLERequestContext::WriteBuffer(const void* buffer, std::size_t size,
     return size;
 }
 
-std::size_t HLERequestContext::GetReadBufferSize(int buffer_index) const {
-    const bool is_buffer_a{BufferDescriptorA().size() > std::size_t(buffer_index) &&
+std::size_t HLERequestContext::GetReadBufferSize(std::size_t buffer_index) const {
+    const bool is_buffer_a{BufferDescriptorA().size() > buffer_index &&
                            BufferDescriptorA()[buffer_index].Size()};
     if (is_buffer_a) {
-        ASSERT_MSG(BufferDescriptorA().size() > std::size_t(buffer_index),
+        ASSERT_MSG(BufferDescriptorA().size() > buffer_index,
                    "BufferDescriptorA invalid buffer_index {}", buffer_index);
         ASSERT_MSG(BufferDescriptorA()[buffer_index].Size() > 0,
                    "BufferDescriptorA buffer_index {} is empty", buffer_index);
         return BufferDescriptorA()[buffer_index].Size();
     } else {
-        ASSERT_MSG(BufferDescriptorX().size() > std::size_t(buffer_index),
+        ASSERT_MSG(BufferDescriptorX().size() > buffer_index,
                    "BufferDescriptorX invalid buffer_index {}", buffer_index);
         ASSERT_MSG(BufferDescriptorX()[buffer_index].Size() > 0,
                    "BufferDescriptorX buffer_index {} is empty", buffer_index);
@@ -355,15 +355,15 @@ std::size_t HLERequestContext::GetReadBufferSize(int buffer_index) const {
     }
 }
 
-std::size_t HLERequestContext::GetWriteBufferSize(int buffer_index) const {
-    const bool is_buffer_b{BufferDescriptorB().size() > std::size_t(buffer_index) &&
+std::size_t HLERequestContext::GetWriteBufferSize(std::size_t buffer_index) const {
+    const bool is_buffer_b{BufferDescriptorB().size() > buffer_index &&
                            BufferDescriptorB()[buffer_index].Size()};
     if (is_buffer_b) {
-        ASSERT_MSG(BufferDescriptorB().size() > std::size_t(buffer_index),
+        ASSERT_MSG(BufferDescriptorB().size() > buffer_index,
                    "BufferDescriptorB invalid buffer_index {}", buffer_index);
         return BufferDescriptorB()[buffer_index].Size();
     } else {
-        ASSERT_MSG(BufferDescriptorC().size() > std::size_t(buffer_index),
+        ASSERT_MSG(BufferDescriptorC().size() > buffer_index,
                    "BufferDescriptorC invalid buffer_index {}", buffer_index);
         return BufferDescriptorC()[buffer_index].Size();
     }

src/core/hle/kernel/hle_ipc.h

@@ -179,10 +179,11 @@ public:
     }
 
     /// Helper function to read a buffer using the appropriate buffer descriptor
-    std::vector<u8> ReadBuffer(int buffer_index = 0) const;
+    std::vector<u8> ReadBuffer(std::size_t buffer_index = 0) const;
 
     /// Helper function to write a buffer using the appropriate buffer descriptor
-    std::size_t WriteBuffer(const void* buffer, std::size_t size, int buffer_index = 0) const;
+    std::size_t WriteBuffer(const void* buffer, std::size_t size,
+                            std::size_t buffer_index = 0) const;
 
     /* Helper function to write a buffer using the appropriate buffer descriptor
      *
@@ -194,7 +195,8 @@ public:
      */
     template <typename ContiguousContainer,
               typename = std::enable_if_t<!std::is_pointer_v<ContiguousContainer>>>
-    std::size_t WriteBuffer(const ContiguousContainer& container, int buffer_index = 0) const {
+    std::size_t WriteBuffer(const ContiguousContainer& container,
+                            std::size_t buffer_index = 0) const {
         using ContiguousType = typename ContiguousContainer::value_type;
 
         static_assert(std::is_trivially_copyable_v<ContiguousType>,
@@ -205,10 +207,10 @@ public:
     }
 
     /// Helper function to get the size of the input buffer
-    std::size_t GetReadBufferSize(int buffer_index = 0) const;
+    std::size_t GetReadBufferSize(std::size_t buffer_index = 0) const;
 
     /// Helper function to get the size of the output buffer
-    std::size_t GetWriteBufferSize(int buffer_index = 0) const;
+    std::size_t GetWriteBufferSize(std::size_t buffer_index = 0) const;
 
     template <typename T>
     std::shared_ptr<T> GetCopyObject(std::size_t index) {

src/core/hle/kernel/memory/slab_heap.h

@@ -51,7 +51,7 @@ public:
     }
 
     void Free(void* obj) {
-        Node* node = reinterpret_cast<Node*>(obj);
+        Node* node = static_cast<Node*>(obj);
 
         Node* cur_head = head.load();
         do {
@@ -145,7 +145,7 @@ public:
     }
 
     T* Allocate() {
-        T* obj = reinterpret_cast<T*>(AllocateImpl());
+        T* obj = static_cast<T*>(AllocateImpl());
         if (obj != nullptr) {
             new (obj) T();
         }

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

@@ -903,7 +903,7 @@ private:
     void PopOutData(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Service_AM, "called");
 
-        const auto storage = applet->GetBroker().PopNormalDataToGame();
+        auto storage = applet->GetBroker().PopNormalDataToGame();
         if (storage == nullptr) {
             LOG_ERROR(Service_AM,
                       "storage is a nullptr. There is no data in the current normal channel");
@@ -934,7 +934,7 @@ private:
     void PopInteractiveOutData(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Service_AM, "called");
 
-        const auto storage = applet->GetBroker().PopInteractiveDataToGame();
+        auto storage = applet->GetBroker().PopInteractiveDataToGame();
         if (storage == nullptr) {
             LOG_ERROR(Service_AM,
                       "storage is a nullptr. There is no data in the current interactive channel");

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

@@ -92,7 +92,7 @@ private:
     }
 
     void RequestUpdateImpl(Kernel::HLERequestContext& ctx) {
-        LOG_WARNING(Service_Audio, "(STUBBED) called");
+        LOG_DEBUG(Service_Audio, "(STUBBED) called");
 
         ctx.WriteBuffer(renderer->UpdateAudioRenderer(ctx.ReadBuffer()));
         IPC::ResponseBuilder rb{ctx, 2};
@@ -252,8 +252,6 @@ private:
     }
 
     void GetAudioDeviceOutputVolume(Kernel::HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
-
         const auto device_name_buffer = ctx.ReadBuffer();
         const std::string name = Common::StringFromBuffer(device_name_buffer);
 

src/core/hle/service/es/es.cpp

@@ -76,7 +76,6 @@ private:
     }
 
     void ImportTicket(Kernel::HLERequestContext& ctx) {
-        IPC::RequestParser rp{ctx};
         const auto ticket = ctx.ReadBuffer();
         const auto cert = ctx.ReadBuffer(1);
 

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

@@ -107,6 +107,7 @@ void Controller_NPad::InitNewlyAddedControler(std::size_t controller_idx) {
     switch (controller_type) {
     case NPadControllerType::None:
         UNREACHABLE();
+        break;
     case NPadControllerType::Handheld:
         controller.joy_styles.handheld.Assign(1);
         controller.device_type.handheld.Assign(1);
@@ -363,6 +364,7 @@ void Controller_NPad::OnUpdate(const Core::Timing::CoreTiming& core_timing, u8*
         switch (controller_type) {
         case NPadControllerType::None:
             UNREACHABLE();
+            break;
         case NPadControllerType::Handheld:
             handheld_entry.connection_status.raw = 0;
             handheld_entry.connection_status.IsWired.Assign(1);
@@ -500,7 +502,7 @@ void Controller_NPad::SetNpadMode(u32 npad_id, NPadAssignments assignment_mode)
 
 void Controller_NPad::VibrateController(const std::vector<u32>& controller_ids,
                                         const std::vector<Vibration>& vibrations) {
-    LOG_WARNING(Service_HID, "(STUBBED) called");
+    LOG_DEBUG(Service_HID, "(STUBBED) called");
 
     if (!can_controllers_vibrate) {
         return;

src/core/hle/service/time/standard_network_system_clock_core.h

@@ -23,7 +23,7 @@ public:
         standard_network_clock_sufficient_accuracy = value;
     }
 
-    bool IsStandardNetworkSystemClockAccuracySufficient(Core::System& system) {
+    bool IsStandardNetworkSystemClockAccuracySufficient(Core::System& system) const {
         SystemClockContext context{};
         if (GetClockContext(system, context) != RESULT_SUCCESS) {
             return {};

src/core/hle/service/time/steady_clock_core.h

@@ -16,6 +16,7 @@ namespace Service::Time::Clock {
 class SteadyClockCore {
 public:
     SteadyClockCore() = default;
+    virtual ~SteadyClockCore() = default;
 
     const Common::UUID& GetClockSourceId() const {
         return clock_source_id;

src/core/hle/service/time/system_clock_context_update_callback.h

@@ -20,7 +20,7 @@ namespace Service::Time::Clock {
 class SystemClockContextUpdateCallback {
 public:
     SystemClockContextUpdateCallback();
-    ~SystemClockContextUpdateCallback();
+    virtual ~SystemClockContextUpdateCallback();
 
     bool NeedUpdate(const SystemClockContext& value) const;
 

src/core/hle/service/time/system_clock_core.cpp

@@ -9,7 +9,7 @@
 namespace Service::Time::Clock {
 
 SystemClockCore::SystemClockCore(SteadyClockCore& steady_clock_core)
-    : steady_clock_core{steady_clock_core}, is_initialized{} {
+    : steady_clock_core{steady_clock_core} {
     context.steady_time_point.clock_source_id = steady_clock_core.GetClockSourceId();
 }
 

src/core/hle/service/time/system_clock_core.h

@@ -22,7 +22,7 @@ class SystemClockContextUpdateCallback;
 class SystemClockCore {
 public:
     explicit SystemClockCore(SteadyClockCore& steady_clock_core);
-    ~SystemClockCore();
+    virtual ~SystemClockCore();
 
     SteadyClockCore& GetSteadyClockCore() const {
         return steady_clock_core;

src/core/loader/elf.cpp

@@ -398,6 +398,11 @@ AppLoader_ELF::LoadResult AppLoader_ELF::Load(Kernel::Process& process) {
     Kernel::CodeSet codeset = elf_reader.LoadInto(base_address);
     const VAddr entry_point = codeset.entrypoint;
 
+    // Setup the process code layout
+    if (process.LoadFromMetadata(FileSys::ProgramMetadata::GetDefault(), buffer.size()).IsError()) {
+        return {ResultStatus::ErrorNotInitialized, {}};
+    }
+
     process.LoadModule(std::move(codeset), entry_point);
 
     is_loaded = true;

src/core/loader/nro.cpp

@@ -131,7 +131,7 @@ static constexpr u32 PageAlignSize(u32 size) {
 }
 
 static bool LoadNroImpl(Kernel::Process& process, const std::vector<u8>& data,
-                        const std::string& name, VAddr load_base) {
+                        const std::string& name) {
     if (data.size() < sizeof(NroHeader)) {
         return {};
     }
@@ -187,19 +187,25 @@ static bool LoadNroImpl(Kernel::Process& process, const std::vector<u8>& data,
     codeset.DataSegment().size += bss_size;
     program_image.resize(static_cast<u32>(program_image.size()) + bss_size);
 
+    // Setup the process code layout
+    if (process.LoadFromMetadata(FileSys::ProgramMetadata::GetDefault(), program_image.size())
+            .IsError()) {
+        return false;
+    }
+
     // Load codeset for current process
     codeset.memory = std::move(program_image);
-    process.LoadModule(std::move(codeset), load_base);
+    process.LoadModule(std::move(codeset), process.PageTable().GetCodeRegionStart());
 
     // Register module with GDBStub
-    GDBStub::RegisterModule(name, load_base, load_base);
+    GDBStub::RegisterModule(name, process.PageTable().GetCodeRegionStart(),
+                            process.PageTable().GetCodeRegionEnd());
 
     return true;
 }
 
-bool AppLoader_NRO::LoadNro(Kernel::Process& process, const FileSys::VfsFile& file,
-                            VAddr load_base) {
-    return LoadNroImpl(process, file.ReadAllBytes(), file.GetName(), load_base);
+bool AppLoader_NRO::LoadNro(Kernel::Process& process, const FileSys::VfsFile& file) {
+    return LoadNroImpl(process, file.ReadAllBytes(), file.GetName());
 }
 
 AppLoader_NRO::LoadResult AppLoader_NRO::Load(Kernel::Process& process) {
@@ -207,10 +213,7 @@ AppLoader_NRO::LoadResult AppLoader_NRO::Load(Kernel::Process& process) {
         return {ResultStatus::ErrorAlreadyLoaded, {}};
     }
 
-    // Load NRO
-    const VAddr base_address = process.PageTable().GetCodeRegionStart();
-
-    if (!LoadNro(process, *file, base_address)) {
+    if (!LoadNro(process, *file)) {
         return {ResultStatus::ErrorLoadingNRO, {}};
     }
 

src/core/loader/nro.h

@@ -47,7 +47,7 @@ public:
     bool IsRomFSUpdatable() const override;
 
 private:
-    bool LoadNro(Kernel::Process& process, const FileSys::VfsFile& file, VAddr load_base);
+    bool LoadNro(Kernel::Process& process, const FileSys::VfsFile& file);
 
     std::vector<u8> icon_data;
     std::unique_ptr<FileSys::NACP> nacp;

src/core/loader/nso.cpp

@@ -37,7 +37,7 @@ static_assert(sizeof(MODHeader) == 0x1c, "MODHeader has incorrect size.");
 
 std::vector<u8> DecompressSegment(const std::vector<u8>& compressed_data,
                                   const NSOSegmentHeader& header) {
-    const std::vector<u8> uncompressed_data =
+    std::vector<u8> uncompressed_data =
         Common::Compression::DecompressDataLZ4(compressed_data, header.size);
 
     ASSERT_MSG(uncompressed_data.size() == header.size, "{} != {}", header.size,

src/core/settings.h

@@ -464,6 +464,7 @@ struct Values {
     bool dump_nso;
     bool reporting_services;
     bool quest_flag;
+    bool disable_cpu_opt;
 
     // BCAT
     std::string bcat_backend;

src/video_core/dma_pusher.cpp

@@ -12,7 +12,7 @@
 
 namespace Tegra {
 
-DmaPusher::DmaPusher(GPU& gpu) : gpu(gpu) {}
+DmaPusher::DmaPusher(Core::System& system, GPU& gpu) : gpu{gpu}, system{system} {}
 
 DmaPusher::~DmaPusher() = default;
 
@@ -26,7 +26,7 @@ void DmaPusher::DispatchCalls() {
 
     dma_pushbuffer_subindex = 0;
 
-    while (Core::System::GetInstance().IsPoweredOn()) {
+    while (system.IsPoweredOn()) {
         if (!Step()) {
             break;
         }

src/video_core/dma_pusher.h

@@ -10,6 +10,10 @@
 #include "common/bit_field.h"
 #include "common/common_types.h"
 
+namespace Core {
+class System;
+}
+
 namespace Tegra {
 
 enum class SubmissionMode : u32 {
@@ -56,7 +60,7 @@ using CommandList = std::vector<Tegra::CommandListHeader>;
  */
 class DmaPusher {
 public:
-    explicit DmaPusher(GPU& gpu);
+    explicit DmaPusher(Core::System& system, GPU& gpu);
     ~DmaPusher();
 
     void Push(CommandList&& entries) {
@@ -72,8 +76,6 @@ private:
 
     void CallMethod(u32 argument) const;
 
-    GPU& gpu;
-
     std::vector<CommandHeader> command_headers; ///< Buffer for list of commands fetched at once
 
     std::queue<CommandList> dma_pushbuffer; ///< Queue of command lists to be processed
@@ -92,6 +94,9 @@ private:
 
     GPUVAddr dma_mget{};  ///< main pushbuffer last read address
     bool ib_enable{true}; ///< IB mode enabled
+
+    GPU& gpu;
+    Core::System& system;
 };
 
 } // namespace Tegra

src/video_core/gpu.cpp

@@ -27,7 +27,7 @@ GPU::GPU(Core::System& system, std::unique_ptr<VideoCore::RendererBase>&& render
     : system{system}, renderer{std::move(renderer_)}, is_async{is_async} {
     auto& rasterizer{renderer->Rasterizer()};
     memory_manager = std::make_unique<Tegra::MemoryManager>(system, rasterizer);
-    dma_pusher = std::make_unique<Tegra::DmaPusher>(*this);
+    dma_pusher = std::make_unique<Tegra::DmaPusher>(system, *this);
     maxwell_3d = std::make_unique<Engines::Maxwell3D>(system, rasterizer, *memory_manager);
     fermi_2d = std::make_unique<Engines::Fermi2D>(rasterizer);
     kepler_compute = std::make_unique<Engines::KeplerCompute>(system, rasterizer, *memory_manager);

src/video_core/renderer_opengl/gl_shader_decompiler.cpp

@@ -1145,6 +1145,7 @@ private:
                 return {"gl_FragCoord"s + GetSwizzle(element), Type::Float};
             default:
                 UNREACHABLE();
+                return {"0", Type::Int};
             }
         case Attribute::Index::FrontColor:
             return {"gl_Color"s + GetSwizzle(element), Type::Float};

src/video_core/renderer_opengl/maxwell_to_gl.h

@@ -191,6 +191,7 @@ inline GLenum TextureFilterMode(Tegra::Texture::TextureFilter filter_mode,
         case Tegra::Texture::TextureMipmapFilter::Linear:
             return GL_LINEAR_MIPMAP_LINEAR;
         }
+        break;
     }
     case Tegra::Texture::TextureFilter::Nearest: {
         switch (mip_filter_mode) {
@@ -201,6 +202,7 @@ inline GLenum TextureFilterMode(Tegra::Texture::TextureFilter filter_mode,
         case Tegra::Texture::TextureMipmapFilter::Linear:
             return GL_NEAREST_MIPMAP_LINEAR;
         }
+        break;
     }
     }
     LOG_ERROR(Render_OpenGL, "Unimplemented texture filter mode={}", static_cast<u32>(filter_mode));

src/video_core/renderer_vulkan/renderer_vulkan.cpp

@@ -42,7 +42,7 @@
 #include <vulkan/vulkan_win32.h>
 #endif
 
-#ifdef __linux__
+#if !defined(_WIN32) && !defined(__APPLE__)
 #include <X11/Xlib.h>
 #include <vulkan/vulkan_wayland.h>
 #include <vulkan/vulkan_xlib.h>
@@ -119,7 +119,7 @@ vk::Instance CreateInstance(Common::DynamicLibrary& library, vk::InstanceDispatc
         extensions.push_back(VK_KHR_WIN32_SURFACE_EXTENSION_NAME);
         break;
 #endif
-#ifdef __linux__
+#if !defined(_WIN32) && !defined(__APPLE__)
     case Core::Frontend::WindowSystemType::X11:
         extensions.push_back(VK_KHR_XLIB_SURFACE_EXTENSION_NAME);
         break;
@@ -345,7 +345,7 @@ bool RendererVulkan::CreateSurface() {
         }
     }
 #endif
-#ifdef __linux__
+#if !defined(_WIN32) && !defined(__APPLE__)
     if (window_info.type == Core::Frontend::WindowSystemType::X11) {
         const VkXlibSurfaceCreateInfoKHR xlib_ci{
             VK_STRUCTURE_TYPE_XLIB_SURFACE_CREATE_INFO_KHR, nullptr, 0,

src/video_core/renderer_vulkan/shaders/quad_indexed.comp

@@ -0,0 +1,50 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+/*
+ * Build instructions:
+ * $ glslangValidator -V quad_indexed.comp -o output.spv
+ * $ spirv-opt -O --strip-debug output.spv -o optimized.spv
+ * $ xxd -i optimized.spv
+ *
+ * Then copy that bytecode to the C++ file
+ */
+
+#version 460 core
+
+layout (local_size_x = 1024) in;
+
+layout (std430, set = 0, binding = 0) readonly buffer InputBuffer {
+    uint input_indexes[];
+};
+
+layout (std430, set = 0, binding = 1) writeonly buffer OutputBuffer {
+    uint output_indexes[];
+};
+
+layout (push_constant) uniform PushConstants {
+    uint base_vertex;
+    int index_shift; // 0: uint8, 1: uint16, 2: uint32
+};
+
+void main() {
+    int primitive = int(gl_GlobalInvocationID.x);
+    if (primitive * 6 >= output_indexes.length()) {
+        return;
+    }
+
+    int index_size = 8 << index_shift;
+    int flipped_shift = 2 - index_shift;
+    int mask = (1 << flipped_shift) - 1;
+
+    const int quad_swizzle[6] = int[](0, 1, 2, 0, 2, 3);
+    for (uint vertex = 0; vertex < 6; ++vertex) {
+        int offset = primitive * 4 + quad_swizzle[vertex];
+        int int_offset = offset >> flipped_shift;
+        int bit_offset = (offset & mask) * index_size;
+        uint packed_input = input_indexes[int_offset];
+        uint index = bitfieldExtract(packed_input, bit_offset, index_size);
+        output_indexes[primitive * 6 + vertex] = index + base_vertex;
+    }
+}

src/video_core/renderer_vulkan/vk_compute_pass.cpp

@@ -135,11 +135,11 @@ VkDescriptorUpdateTemplateEntryKHR BuildQuadArrayPassDescriptorUpdateTemplateEnt
     return entry;
 }
 
-VkPushConstantRange BuildQuadArrayPassPushConstantRange() {
+VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
     VkPushConstantRange range;
     range.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;
     range.offset = 0;
-    range.size = sizeof(u32);
+    range.size = static_cast<u32>(size);
     return range;
 }
 
@@ -220,7 +220,130 @@ constexpr u8 uint8_pass[] = {
     0xf9, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00,
     0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
 
-std::array<VkDescriptorSetLayoutBinding, 2> BuildUint8PassDescriptorSetBindings() {
+// Quad indexed SPIR-V module. Generated from the "shaders/" directory.
+constexpr u8 QUAD_INDEXED_SPV[] = {
+    0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x7c, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x11, 0x00, 0x02, 0x00, 0x01, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x06, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x47, 0x4c, 0x53, 0x4c, 0x2e, 0x73, 0x74, 0x64, 0x2e, 0x34, 0x35, 0x30,
+    0x00, 0x00, 0x00, 0x00, 0x0e, 0x00, 0x03, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x0f, 0x00, 0x06, 0x00, 0x05, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x6d, 0x61, 0x69, 0x6e,
+    0x00, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x10, 0x00, 0x06, 0x00, 0x04, 0x00, 0x00, 0x00,
+    0x11, 0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x04, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x04, 0x00, 0x15, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x04, 0x00, 0x16, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x05, 0x00, 0x16, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x16, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x04, 0x00, 0x18, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x05, 0x00, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x22, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+    0x23, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x22, 0x00, 0x00, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x56, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x04, 0x00, 0x00, 0x00, 0x48, 0x00, 0x04, 0x00, 0x57, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x18, 0x00, 0x00, 0x00, 0x48, 0x00, 0x05, 0x00, 0x57, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x23, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x03, 0x00, 0x57, 0x00, 0x00, 0x00,
+    0x03, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x59, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x59, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x47, 0x00, 0x04, 0x00, 0x72, 0x00, 0x00, 0x00, 0x0b, 0x00, 0x00, 0x00,
+    0x19, 0x00, 0x00, 0x00, 0x13, 0x00, 0x02, 0x00, 0x02, 0x00, 0x00, 0x00, 0x21, 0x00, 0x03, 0x00,
+    0x03, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x20, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x07, 0x00, 0x00, 0x00,
+    0x07, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x15, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
+    0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x17, 0x00, 0x04, 0x00, 0x0a, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x0b, 0x00, 0x00, 0x00,
+    0x0c, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
+    0x0d, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x0e, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x13, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00, 0x15, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x16, 0x00, 0x00, 0x00, 0x15, 0x00, 0x00, 0x00,
+    0x20, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x16, 0x00, 0x00, 0x00,
+    0x3b, 0x00, 0x04, 0x00, 0x17, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x14, 0x00, 0x02, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x21, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x04, 0x00, 0x22, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x22, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x23, 0x00, 0x00, 0x00,
+    0x24, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x25, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x26, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x2b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00,
+    0x3b, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x3f, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x04, 0x00, 0x41, 0x00, 0x00, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x42, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x43, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x09, 0x00, 0x41, 0x00, 0x00, 0x00,
+    0x44, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00,
+    0x42, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x00, 0x00, 0x43, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00,
+    0x46, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0x41, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x03, 0x00,
+    0x56, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x1e, 0x00, 0x03, 0x00, 0x57, 0x00, 0x00, 0x00,
+    0x56, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x58, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x57, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x58, 0x00, 0x00, 0x00, 0x59, 0x00, 0x00, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x5b, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x20, 0x00, 0x04, 0x00, 0x69, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x70, 0x00, 0x00, 0x00,
+    0x00, 0x04, 0x00, 0x00, 0x2b, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00,
+    0x01, 0x00, 0x00, 0x00, 0x2c, 0x00, 0x06, 0x00, 0x0a, 0x00, 0x00, 0x00, 0x72, 0x00, 0x00, 0x00,
+    0x70, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00, 0x71, 0x00, 0x00, 0x00, 0x36, 0x00, 0x05, 0x00,
+    0x02, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+    0xf8, 0x00, 0x02, 0x00, 0x05, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x04, 0x00, 0x46, 0x00, 0x00, 0x00,
+    0x47, 0x00, 0x00, 0x00, 0x07, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00,
+    0xf8, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00, 0xf6, 0x00, 0x04, 0x00, 0x73, 0x00, 0x00, 0x00,
+    0x76, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00, 0x75, 0x00, 0x00, 0x00,
+    0xf8, 0x00, 0x02, 0x00, 0x75, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x0e, 0x00, 0x00, 0x00,
+    0x0f, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x0f, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00,
+    0x44, 0x00, 0x05, 0x00, 0x09, 0x00, 0x00, 0x00, 0x19, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00,
+    0x19, 0x00, 0x00, 0x00, 0xaf, 0x00, 0x05, 0x00, 0x1b, 0x00, 0x00, 0x00, 0x1c, 0x00, 0x00, 0x00,
+    0x14, 0x00, 0x00, 0x00, 0x1a, 0x00, 0x00, 0x00, 0xf7, 0x00, 0x03, 0x00, 0x1e, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0xfa, 0x00, 0x04, 0x00, 0x1c, 0x00, 0x00, 0x00, 0x1d, 0x00, 0x00, 0x00,
+    0x1e, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf9, 0x00, 0x02, 0x00,
+    0x73, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1e, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00,
+    0x26, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
+    0x3d, 0x00, 0x04, 0x00, 0x06, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00,
+    0xc4, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x21, 0x00, 0x00, 0x00,
+    0x28, 0x00, 0x00, 0x00, 0x82, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00,
+    0x2b, 0x00, 0x00, 0x00, 0x28, 0x00, 0x00, 0x00, 0xc4, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x31, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00, 0x2e, 0x00, 0x00, 0x00, 0x82, 0x00, 0x05, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00, 0x31, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
+    0xf9, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00,
+    0xf5, 0x00, 0x07, 0x00, 0x09, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x0d, 0x00, 0x00, 0x00,
+    0x1e, 0x00, 0x00, 0x00, 0x6f, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0xb0, 0x00, 0x05, 0x00,
+    0x1b, 0x00, 0x00, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00,
+    0xf6, 0x00, 0x04, 0x00, 0x37, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0xfa, 0x00, 0x04, 0x00, 0x3c, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x00, 0x37, 0x00, 0x00, 0x00,
+    0xf8, 0x00, 0x02, 0x00, 0x36, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x40, 0x00, 0x00, 0x00, 0x11, 0x00, 0x00, 0x00, 0x3f, 0x00, 0x00, 0x00, 0x3e, 0x00, 0x03, 0x00,
+    0x47, 0x00, 0x00, 0x00, 0x44, 0x00, 0x00, 0x00, 0x41, 0x00, 0x05, 0x00, 0x07, 0x00, 0x00, 0x00,
+    0x48, 0x00, 0x00, 0x00, 0x47, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
+    0x06, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x49, 0x00, 0x00, 0x00,
+    0xc3, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x4e, 0x00, 0x00, 0x00, 0x4a, 0x00, 0x00, 0x00,
+    0x2e, 0x00, 0x00, 0x00, 0xc7, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00, 0x52, 0x00, 0x00, 0x00,
+    0x4a, 0x00, 0x00, 0x00, 0x32, 0x00, 0x00, 0x00, 0x84, 0x00, 0x05, 0x00, 0x06, 0x00, 0x00, 0x00,
+    0x54, 0x00, 0x00, 0x00, 0x52, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00,
+    0x5b, 0x00, 0x00, 0x00, 0x5c, 0x00, 0x00, 0x00, 0x59, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00,
+    0x4e, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x5d, 0x00, 0x00, 0x00,
+    0x5c, 0x00, 0x00, 0x00, 0xcb, 0x00, 0x06, 0x00, 0x09, 0x00, 0x00, 0x00, 0x62, 0x00, 0x00, 0x00,
+    0x5d, 0x00, 0x00, 0x00, 0x54, 0x00, 0x00, 0x00, 0x29, 0x00, 0x00, 0x00, 0x7c, 0x00, 0x04, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x14, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00, 0x65, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00,
+    0x41, 0x00, 0x05, 0x00, 0x69, 0x00, 0x00, 0x00, 0x6a, 0x00, 0x00, 0x00, 0x24, 0x00, 0x00, 0x00,
+    0x42, 0x00, 0x00, 0x00, 0x3d, 0x00, 0x04, 0x00, 0x09, 0x00, 0x00, 0x00, 0x6b, 0x00, 0x00, 0x00,
+    0x6a, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00, 0x09, 0x00, 0x00, 0x00, 0x6c, 0x00, 0x00, 0x00,
+    0x62, 0x00, 0x00, 0x00, 0x6b, 0x00, 0x00, 0x00, 0x41, 0x00, 0x06, 0x00, 0x5b, 0x00, 0x00, 0x00,
+    0x6d, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x67, 0x00, 0x00, 0x00,
+    0x3e, 0x00, 0x03, 0x00, 0x6d, 0x00, 0x00, 0x00, 0x6c, 0x00, 0x00, 0x00, 0x80, 0x00, 0x05, 0x00,
+    0x09, 0x00, 0x00, 0x00, 0x6f, 0x00, 0x00, 0x00, 0x7b, 0x00, 0x00, 0x00, 0x25, 0x00, 0x00, 0x00,
+    0xf9, 0x00, 0x02, 0x00, 0x35, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x37, 0x00, 0x00, 0x00,
+    0xf9, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x76, 0x00, 0x00, 0x00,
+    0xf9, 0x00, 0x02, 0x00, 0x74, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x73, 0x00, 0x00, 0x00,
+    0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
+
+std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBindings() {
     std::array<VkDescriptorSetLayoutBinding, 2> bindings;
     bindings[0].binding = 0;
     bindings[0].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
@@ -235,7 +358,7 @@ std::array<VkDescriptorSetLayoutBinding, 2> BuildUint8PassDescriptorSetBindings(
     return bindings;
 }
 
-VkDescriptorUpdateTemplateEntryKHR BuildUint8PassDescriptorUpdateTemplateEntry() {
+VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
     VkDescriptorUpdateTemplateEntryKHR entry;
     entry.dstBinding = 0;
     entry.dstArrayElement = 0;
@@ -337,14 +460,14 @@ QuadArrayPass::QuadArrayPass(const VKDevice& device, VKScheduler& scheduler,
                              VKUpdateDescriptorQueue& update_descriptor_queue)
     : VKComputePass(device, descriptor_pool, BuildQuadArrayPassDescriptorSetLayoutBinding(),
                     BuildQuadArrayPassDescriptorUpdateTemplateEntry(),
-                    BuildQuadArrayPassPushConstantRange(), std::size(quad_array), quad_array),
+                    BuildComputePushConstantRange(sizeof(u32)), std::size(quad_array), quad_array),
       scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
       update_descriptor_queue{update_descriptor_queue} {}
 
 QuadArrayPass::~QuadArrayPass() = default;
 
 std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
-    const u32 num_triangle_vertices = num_vertices * 6 / 4;
+    const u32 num_triangle_vertices = (num_vertices / 4) * 6;
     const std::size_t staging_size = num_triangle_vertices * sizeof(u32);
     auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
 
@@ -383,8 +506,8 @@ std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32
 Uint8Pass::Uint8Pass(const VKDevice& device, VKScheduler& scheduler,
                      VKDescriptorPool& descriptor_pool, VKStagingBufferPool& staging_buffer_pool,
                      VKUpdateDescriptorQueue& update_descriptor_queue)
-    : VKComputePass(device, descriptor_pool, BuildUint8PassDescriptorSetBindings(),
-                    BuildUint8PassDescriptorUpdateTemplateEntry(), {}, std::size(uint8_pass),
+    : VKComputePass(device, descriptor_pool, BuildInputOutputDescriptorSetBindings(),
+                    BuildInputOutputDescriptorUpdateTemplate(), {}, std::size(uint8_pass),
                     uint8_pass),
       scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
       update_descriptor_queue{update_descriptor_queue} {}
@@ -425,4 +548,70 @@ std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buff
     return {*buffer.handle, 0};
 }
 
+QuadIndexedPass::QuadIndexedPass(const VKDevice& device, VKScheduler& scheduler,
+                                 VKDescriptorPool& descriptor_pool,
+                                 VKStagingBufferPool& staging_buffer_pool,
+                                 VKUpdateDescriptorQueue& update_descriptor_queue)
+    : VKComputePass(device, descriptor_pool, BuildInputOutputDescriptorSetBindings(),
+                    BuildInputOutputDescriptorUpdateTemplate(),
+                    BuildComputePushConstantRange(sizeof(u32) * 2), std::size(QUAD_INDEXED_SPV),
+                    QUAD_INDEXED_SPV),
+      scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
+      update_descriptor_queue{update_descriptor_queue} {}
+
+QuadIndexedPass::~QuadIndexedPass() = default;
+
+std::pair<VkBuffer, u64> QuadIndexedPass::Assemble(
+    Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format, u32 num_vertices, u32 base_vertex,
+    VkBuffer src_buffer, u64 src_offset) {
+    const u32 index_shift = [index_format] {
+        switch (index_format) {
+        case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedByte:
+            return 0;
+        case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedShort:
+            return 1;
+        case Tegra::Engines::Maxwell3D::Regs::IndexFormat::UnsignedInt:
+            return 2;
+        }
+        UNREACHABLE();
+        return 2;
+    }();
+    const u32 input_size = num_vertices << index_shift;
+    const u32 num_tri_vertices = (num_vertices / 4) * 6;
+
+    const std::size_t staging_size = num_tri_vertices * sizeof(u32);
+    auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
+
+    update_descriptor_queue.Acquire();
+    update_descriptor_queue.AddBuffer(src_buffer, src_offset, input_size);
+    update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
+    const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence());
+
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set,
+                      num_tri_vertices, base_vertex, index_shift](vk::CommandBuffer cmdbuf) {
+        static constexpr u32 dispatch_size = 1024;
+        const std::array push_constants = {base_vertex, index_shift};
+        cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
+        cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
+        cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(push_constants),
+                             &push_constants);
+        cmdbuf.Dispatch(Common::AlignUp(num_tri_vertices, dispatch_size) / dispatch_size, 1, 1);
+
+        VkBufferMemoryBarrier barrier;
+        barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
+        barrier.pNext = nullptr;
+        barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
+        barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
+        barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+        barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
+        barrier.buffer = buffer;
+        barrier.offset = 0;
+        barrier.size = static_cast<VkDeviceSize>(num_tri_vertices * sizeof(u32));
+        cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+                               VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
+    });
+    return {*buffer.handle, 0};
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_compute_pass.h

@@ -8,6 +8,7 @@
 #include <utility>
 #include <vector>
 #include "common/common_types.h"
+#include "video_core/engines/maxwell_3d.h"
 #include "video_core/renderer_vulkan/vk_descriptor_pool.h"
 #include "video_core/renderer_vulkan/wrapper.h"
 
@@ -73,4 +74,22 @@ private:
     VKUpdateDescriptorQueue& update_descriptor_queue;
 };
 
+class QuadIndexedPass final : public VKComputePass {
+public:
+    explicit QuadIndexedPass(const VKDevice& device, VKScheduler& scheduler,
+                             VKDescriptorPool& descriptor_pool,
+                             VKStagingBufferPool& staging_buffer_pool,
+                             VKUpdateDescriptorQueue& update_descriptor_queue);
+    ~QuadIndexedPass();
+
+    std::pair<VkBuffer, u64> Assemble(Tegra::Engines::Maxwell3D::Regs::IndexFormat index_format,
+                                      u32 num_vertices, u32 base_vertex, VkBuffer src_buffer,
+                                      u64 src_offset);
+
+private:
+    VKScheduler& scheduler;
+    VKStagingBufferPool& staging_buffer_pool;
+    VKUpdateDescriptorQueue& update_descriptor_queue;
+};
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -293,6 +293,7 @@ RasterizerVulkan::RasterizerVulkan(Core::System& system, Core::Frontend::EmuWind
       update_descriptor_queue(device, scheduler), renderpass_cache(device),
       quad_array_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
       uint8_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
+      quad_indexed_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue),
       texture_cache(system, *this, device, resource_manager, memory_manager, scheduler,
                     staging_pool),
       pipeline_cache(system, *this, device, scheduler, descriptor_pool, update_descriptor_queue,
@@ -844,18 +845,26 @@ void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawPar
                                         bool is_indexed) {
     const auto& regs = system.GPU().Maxwell3D().regs;
     switch (regs.draw.topology) {
-    case Maxwell::PrimitiveTopology::Quads:
-        if (params.is_indexed) {
-            UNIMPLEMENTED();
-        } else {
+    case Maxwell::PrimitiveTopology::Quads: {
+        if (!params.is_indexed) {
             const auto [buffer, offset] =
                 quad_array_pass.Assemble(params.num_vertices, params.base_vertex);
             buffer_bindings.SetIndexBinding(buffer, offset, VK_INDEX_TYPE_UINT32);
             params.base_vertex = 0;
             params.num_vertices = params.num_vertices * 6 / 4;
             params.is_indexed = true;
+            break;
         }
+        const GPUVAddr gpu_addr = regs.index_array.IndexStart();
+        auto [buffer, offset] = buffer_cache.UploadMemory(gpu_addr, CalculateIndexBufferSize());
+        std::tie(buffer, offset) = quad_indexed_pass.Assemble(
+            regs.index_array.format, params.num_vertices, params.base_vertex, buffer, offset);
+
+        buffer_bindings.SetIndexBinding(buffer, offset, VK_INDEX_TYPE_UINT32);
+        params.num_vertices = (params.num_vertices / 4) * 6;
+        params.base_vertex = 0;
         break;
+    }
     default: {
         if (!is_indexed) {
             break;

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -254,6 +254,7 @@ private:
     VKUpdateDescriptorQueue update_descriptor_queue;
     VKRenderPassCache renderpass_cache;
     QuadArrayPass quad_array_pass;
+    QuadIndexedPass quad_indexed_pass;
     Uint8Pass uint8_pass;
 
     VKTextureCache texture_cache;

src/video_core/renderer_vulkan/vk_stream_buffer.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
+#include <limits>
 #include <optional>
 #include <tuple>
 #include <vector>
@@ -22,22 +23,38 @@ namespace {
 constexpr u64 WATCHES_INITIAL_RESERVE = 0x4000;
 constexpr u64 WATCHES_RESERVE_CHUNK = 0x1000;
 
-constexpr u64 STREAM_BUFFER_SIZE = 256 * 1024 * 1024;
+constexpr u64 PREFERRED_STREAM_BUFFER_SIZE = 256 * 1024 * 1024;
 
-std::optional<u32> FindMemoryType(const VKDevice& device, u32 filter,
-                                  VkMemoryPropertyFlags wanted) {
-    const auto properties = device.GetPhysical().GetMemoryProperties();
-    for (u32 i = 0; i < properties.memoryTypeCount; i++) {
-        if (!(filter & (1 << i))) {
-            continue;
-        }
-        if ((properties.memoryTypes[i].propertyFlags & wanted) == wanted) {
+/// Find a memory type with the passed requirements
+std::optional<u32> FindMemoryType(const VkPhysicalDeviceMemoryProperties& properties,
+                                  VkMemoryPropertyFlags wanted,
+                                  u32 filter = std::numeric_limits<u32>::max()) {
+    for (u32 i = 0; i < properties.memoryTypeCount; ++i) {
+        const auto flags = properties.memoryTypes[i].propertyFlags;
+        if ((flags & wanted) == wanted && (filter & (1U << i)) != 0) {
             return i;
         }
     }
     return std::nullopt;
 }
 
+/// Get the preferred host visible memory type.
+u32 GetMemoryType(const VkPhysicalDeviceMemoryProperties& properties,
+                  u32 filter = std::numeric_limits<u32>::max()) {
+    // Prefer device local host visible allocations. Both AMD and Nvidia now provide one.
+    // Otherwise search for a host visible allocation.
+    static constexpr auto HOST_MEMORY =
+        VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT;
+    static constexpr auto DYNAMIC_MEMORY = HOST_MEMORY | VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
+
+    std::optional preferred_type = FindMemoryType(properties, DYNAMIC_MEMORY);
+    if (!preferred_type) {
+        preferred_type = FindMemoryType(properties, HOST_MEMORY);
+        ASSERT_MSG(preferred_type, "No host visible and coherent memory type found");
+    }
+    return preferred_type.value_or(0);
+}
+
 } // Anonymous namespace
 
 VKStreamBuffer::VKStreamBuffer(const VKDevice& device, VKScheduler& scheduler,
@@ -51,7 +68,7 @@ VKStreamBuffer::VKStreamBuffer(const VKDevice& device, VKScheduler& scheduler,
 VKStreamBuffer::~VKStreamBuffer() = default;
 
 std::tuple<u8*, u64, bool> VKStreamBuffer::Map(u64 size, u64 alignment) {
-    ASSERT(size <= STREAM_BUFFER_SIZE);
+    ASSERT(size <= stream_buffer_size);
     mapped_size = size;
 
     if (alignment > 0) {
@@ -61,7 +78,7 @@ std::tuple<u8*, u64, bool> VKStreamBuffer::Map(u64 size, u64 alignment) {
     WaitPendingOperations(offset);
 
     bool invalidated = false;
-    if (offset + size > STREAM_BUFFER_SIZE) {
+    if (offset + size > stream_buffer_size) {
         // The buffer would overflow, save the amount of used watches and reset the state.
         invalidation_mark = current_watch_cursor;
         current_watch_cursor = 0;
@@ -98,40 +115,37 @@ void VKStreamBuffer::Unmap(u64 size) {
 }
 
 void VKStreamBuffer::CreateBuffers(VkBufferUsageFlags usage) {
+    const auto memory_properties = device.GetPhysical().GetMemoryProperties();
+    const u32 preferred_type = GetMemoryType(memory_properties);
+    const u32 preferred_heap = memory_properties.memoryTypes[preferred_type].heapIndex;
+
+    // Substract from the preferred heap size some bytes to avoid getting out of memory.
+    const VkDeviceSize heap_size = memory_properties.memoryHeaps[preferred_heap].size;
+    const VkDeviceSize allocable_size = heap_size - 4 * 1024 * 1024;
+
     VkBufferCreateInfo buffer_ci;
     buffer_ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
     buffer_ci.pNext = nullptr;
     buffer_ci.flags = 0;
-    buffer_ci.size = STREAM_BUFFER_SIZE;
+    buffer_ci.size = std::min(PREFERRED_STREAM_BUFFER_SIZE, allocable_size);
     buffer_ci.usage = usage;
     buffer_ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
     buffer_ci.queueFamilyIndexCount = 0;
     buffer_ci.pQueueFamilyIndices = nullptr;
 
-    const auto& dev = device.GetLogical();
-    buffer = dev.CreateBuffer(buffer_ci);
+    buffer = device.GetLogical().CreateBuffer(buffer_ci);
+
+    const auto requirements = device.GetLogical().GetBufferMemoryRequirements(*buffer);
+    const u32 required_flags = requirements.memoryTypeBits;
+    stream_buffer_size = static_cast<u64>(requirements.size);
 
-    const auto& dld = device.GetDispatchLoader();
-    const auto requirements = dev.GetBufferMemoryRequirements(*buffer);
-    // Prefer device local host visible allocations (this should hit AMD's pinned memory).
-    auto type =
-        FindMemoryType(device, requirements.memoryTypeBits,
-                       VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
-                           VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
-    if (!type) {
-        // Otherwise search for a host visible allocation.
-        type = FindMemoryType(device, requirements.memoryTypeBits,
-                              VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
-                                  VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
-        ASSERT_MSG(type, "No host visible and coherent memory type found");
-    }
     VkMemoryAllocateInfo memory_ai;
     memory_ai.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
     memory_ai.pNext = nullptr;
     memory_ai.allocationSize = requirements.size;
-    memory_ai.memoryTypeIndex = *type;
+    memory_ai.memoryTypeIndex = GetMemoryType(memory_properties, required_flags);
 
-    memory = dev.AllocateMemory(memory_ai);
+    memory = device.GetLogical().AllocateMemory(memory_ai);
     buffer.BindMemory(*memory, 0);
 }
 

src/video_core/renderer_vulkan/vk_stream_buffer.h

@@ -56,8 +56,9 @@ private:
     const VKDevice& device; ///< Vulkan device manager.
     VKScheduler& scheduler; ///< Command scheduler.
 
-    vk::Buffer buffer;       ///< Mapped buffer.
-    vk::DeviceMemory memory; ///< Memory allocation.
+    vk::Buffer buffer;        ///< Mapped buffer.
+    vk::DeviceMemory memory;  ///< Memory allocation.
+    u64 stream_buffer_size{}; ///< Stream buffer size.
 
     u64 offset{};      ///< Buffer iterator.
     u64 mapped_size{}; ///< Size reserved for the current copy.

src/video_core/shader/track.cpp

@@ -76,12 +76,13 @@ std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, cons
                                                               s64 cursor) {
     if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
         // Constant buffer found, test if it's an immediate
-        const auto offset = cbuf->GetOffset();
+        const auto& offset = cbuf->GetOffset();
         if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
             auto track =
                 MakeTrackSampler<BindlessSamplerNode>(cbuf->GetIndex(), immediate->GetValue());
             return {tracked, track};
-        } else if (const auto operation = std::get_if<OperationNode>(&*offset)) {
+        }
+        if (const auto operation = std::get_if<OperationNode>(&*offset)) {
             const u32 bound_buffer = registry.GetBoundBuffer();
             if (bound_buffer != cbuf->GetIndex()) {
                 return {};
@@ -94,12 +95,12 @@ std::tuple<Node, TrackSampler> ShaderIR::TrackBindlessSampler(Node tracked, cons
             const auto offset_inm = std::get_if<ImmediateNode>(&*base_offset);
             const auto& gpu_driver = registry.AccessGuestDriverProfile();
             const u32 bindless_cv = NewCustomVariable();
-            const Node op =
+            Node op =
                 Operation(OperationCode::UDiv, gpr, Immediate(gpu_driver.GetTextureHandlerSize()));
 
             const Node cv_node = GetCustomVariable(bindless_cv);
             Node amend_op = Operation(OperationCode::Assign, cv_node, std::move(op));
-            const std::size_t amend_index = DeclareAmend(amend_op);
+            const std::size_t amend_index = DeclareAmend(std::move(amend_op));
             AmendNodeCv(amend_index, code[cursor]);
             // TODO Implement Bindless Index custom variable
             auto track = MakeTrackSampler<ArraySamplerNode>(cbuf->GetIndex(),
@@ -142,7 +143,7 @@ std::tuple<Node, u32, u32> ShaderIR::TrackCbuf(Node tracked, const NodeBlock& co
                                                s64 cursor) const {
     if (const auto cbuf = std::get_if<CbufNode>(&*tracked)) {
         // Constant buffer found, test if it's an immediate
-        const auto offset = cbuf->GetOffset();
+        const auto& offset = cbuf->GetOffset();
         if (const auto immediate = std::get_if<ImmediateNode>(&*offset)) {
             return {tracked, cbuf->GetIndex(), immediate->GetValue()};
         }

src/video_core/texture_cache/format_lookup_table.cpp

@@ -196,9 +196,9 @@ std::size_t FormatLookupTable::CalculateIndex(TextureFormat format, bool is_srgb
                                               ComponentType alpha_component) noexcept {
     const auto format_index = static_cast<std::size_t>(format);
     const auto red_index = static_cast<std::size_t>(red_component);
-    const auto green_index = static_cast<std::size_t>(red_component);
-    const auto blue_index = static_cast<std::size_t>(red_component);
-    const auto alpha_index = static_cast<std::size_t>(red_component);
+    const auto green_index = static_cast<std::size_t>(green_component);
+    const auto blue_index = static_cast<std::size_t>(blue_component);
+    const auto alpha_index = static_cast<std::size_t>(alpha_component);
     const std::size_t srgb_index = is_srgb ? 1 : 0;
 
     return format_index * PerFormat +

src/yuzu/configuration/config.cpp

@@ -532,6 +532,8 @@ void Config::ReadDebuggingValues() {
     Settings::values.reporting_services =
         ReadSetting(QStringLiteral("reporting_services"), false).toBool();
     Settings::values.quest_flag = ReadSetting(QStringLiteral("quest_flag"), false).toBool();
+    Settings::values.disable_cpu_opt =
+        ReadSetting(QStringLiteral("disable_cpu_opt"), false).toBool();
 
     qt_config->endGroup();
 }
@@ -1001,6 +1003,7 @@ void Config::SaveDebuggingValues() {
     WriteSetting(QStringLiteral("dump_exefs"), Settings::values.dump_exefs, false);
     WriteSetting(QStringLiteral("dump_nso"), Settings::values.dump_nso, false);
     WriteSetting(QStringLiteral("quest_flag"), Settings::values.quest_flag, false);
+    WriteSetting(QStringLiteral("disable_cpu_opt"), Settings::values.disable_cpu_opt, false);
 
     qt_config->endGroup();
 }

src/yuzu/configuration/configure_debug.cpp

@@ -36,6 +36,7 @@ void ConfigureDebug::SetConfiguration() {
     ui->homebrew_args_edit->setText(QString::fromStdString(Settings::values.program_args));
     ui->reporting_services->setChecked(Settings::values.reporting_services);
     ui->quest_flag->setChecked(Settings::values.quest_flag);
+    ui->disable_cpu_opt->setChecked(Settings::values.disable_cpu_opt);
     ui->enable_graphics_debugging->setEnabled(!Core::System::GetInstance().IsPoweredOn());
     ui->enable_graphics_debugging->setChecked(Settings::values.renderer_debug);
 }
@@ -48,6 +49,7 @@ void ConfigureDebug::ApplyConfiguration() {
     Settings::values.program_args = ui->homebrew_args_edit->text().toStdString();
     Settings::values.reporting_services = ui->reporting_services->isChecked();
     Settings::values.quest_flag = ui->quest_flag->isChecked();
+    Settings::values.disable_cpu_opt = ui->disable_cpu_opt->isChecked();
     Settings::values.renderer_debug = ui->enable_graphics_debugging->isChecked();
     Debugger::ToggleConsole();
     Log::Filter filter;

src/yuzu/configuration/configure_debug.ui

@@ -215,6 +215,13 @@
         </property>
        </widget>
       </item>
+      <item>
+       <widget class="QCheckBox" name="disable_cpu_opt">
+        <property name="text">
+         <string>Disable CPU JIT optimizations</string>
+        </property>
+       </widget>
+      </item>
      </layout>
     </widget>
    </item>

src/yuzu/main.cpp

@@ -802,10 +802,6 @@ void GMainWindow::ConnectMenuEvents() {
     connect(ui.action_Load_Folder, &QAction::triggered, this, &GMainWindow::OnMenuLoadFolder);
     connect(ui.action_Install_File_NAND, &QAction::triggered, this,
             &GMainWindow::OnMenuInstallToNAND);
-    connect(ui.action_Select_NAND_Directory, &QAction::triggered, this,
-            [this] { OnMenuSelectEmulatedDirectory(EmulatedDirectoryTarget::NAND); });
-    connect(ui.action_Select_SDMC_Directory, &QAction::triggered, this,
-            [this] { OnMenuSelectEmulatedDirectory(EmulatedDirectoryTarget::SDMC); });
     connect(ui.action_Exit, &QAction::triggered, this, &QMainWindow::close);
     connect(ui.action_Load_Amiibo, &QAction::triggered, this, &GMainWindow::OnLoadAmiibo);
 
@@ -940,16 +936,18 @@ bool GMainWindow::LoadROM(const QString& filename) {
         default:
             if (static_cast<u32>(result) >
                 static_cast<u32>(Core::System::ResultStatus::ErrorLoader)) {
-                LOG_CRITICAL(Frontend, "Failed to load ROM!");
                 const u16 loader_id = static_cast<u16>(Core::System::ResultStatus::ErrorLoader);
                 const u16 error_id = static_cast<u16>(result) - loader_id;
+                const std::string error_code = fmt::format("({:04X}-{:04X})", loader_id, error_id);
+                LOG_CRITICAL(Frontend, "Failed to load ROM! {}", error_code);
                 QMessageBox::critical(
-                    this, tr("Error while loading ROM!"),
+                    this,
+                    tr("Error while loading ROM! ").append(QString::fromStdString(error_code)),
                     QString::fromStdString(fmt::format(
-                        "While attempting to load the ROM requested, an error occured. Please "
-                        "refer to the yuzu wiki for more information or the yuzu discord for "
-                        "additional help.\n\nError Code: {:04X}-{:04X}\nError Description: {}",
-                        loader_id, error_id, static_cast<Loader::ResultStatus>(error_id))));
+                        "{}<br>Please follow <a href='https://yuzu-emu.org/help/quickstart/'>the "
+                        "yuzu quickstart guide</a> to redump your files.<br>You can refer "
+                        "to the yuzu wiki</a> or the yuzu Discord</a> for help.",
+                        static_cast<Loader::ResultStatus>(error_id))));
             } else {
                 QMessageBox::critical(
                     this, tr("Error while loading ROM!"),
@@ -1663,28 +1661,6 @@ void GMainWindow::OnMenuInstallToNAND() {
     }
 }
 
-void GMainWindow::OnMenuSelectEmulatedDirectory(EmulatedDirectoryTarget target) {
-    const auto res = QMessageBox::information(
-        this, tr("Changing Emulated Directory"),
-        tr("You are about to change the emulated %1 directory of the system. Please note "
-           "that this does not also move the contents of the previous directory to the "
-           "new one and you will have to do that yourself.")
-            .arg(target == EmulatedDirectoryTarget::SDMC ? tr("SD card") : tr("NAND")),
-        QMessageBox::StandardButtons{QMessageBox::Ok, QMessageBox::Cancel});
-
-    if (res == QMessageBox::Cancel)
-        return;
-
-    QString dir_path = QFileDialog::getExistingDirectory(this, tr("Select Directory"));
-    if (!dir_path.isEmpty()) {
-        FileUtil::GetUserPath(target == EmulatedDirectoryTarget::SDMC ? FileUtil::UserPath::SDMCDir
-                                                                      : FileUtil::UserPath::NANDDir,
-                              dir_path.toStdString());
-        Core::System::GetInstance().GetFileSystemController().CreateFactories(*vfs);
-        game_list->PopulateAsync(UISettings::values.game_dirs);
-    }
-}
-
 void GMainWindow::OnMenuRecentFile() {
     QAction* action = qobject_cast<QAction*>(sender());
     assert(action);
@@ -2095,27 +2071,25 @@ void GMainWindow::OnReinitializeKeys(ReinitializeKeyBehavior behavior) {
 
         QString errors;
         if (!pdm.HasFuses()) {
-            errors += tr("- Missing fuses - Cannot derive SBK\n");
+            errors += tr("Missing fuses");
         }
         if (!pdm.HasBoot0()) {
-            errors += tr("- Missing BOOT0 - Cannot derive master keys\n");
+            errors += tr(" - Missing BOOT0");
         }
         if (!pdm.HasPackage2()) {
-            errors += tr("- Missing BCPKG2-1-Normal-Main - Cannot derive general keys\n");
+            errors += tr(" - Missing BCPKG2-1-Normal-Main");
         }
         if (!pdm.HasProdInfo()) {
-            errors += tr("- Missing PRODINFO - Cannot derive title keys\n");
+            errors += tr(" - Missing PRODINFO");
         }
         if (!errors.isEmpty()) {
             QMessageBox::warning(
-                this, tr("Warning Missing Derivation Components"),
-                tr("The following are missing from your configuration that may hinder key "
-                   "derivation. It will be attempted but may not complete.<br><br>") +
-                    errors +
-                    tr("<br><br>You can get all of these and dump all of your games easily by "
-                       "following <a href='https://yuzu-emu.org/help/quickstart/'>the "
-                       "quickstart guide</a>. Alternatively, you can use another method of dumping "
-                       "to obtain all of your keys."));
+                this, tr("Derivation Components Missing"),
+                tr("Components are missing that may hinder key derivation from completing. "
+                   "<br>Please follow <a href='https://yuzu-emu.org/help/quickstart/'>the yuzu "
+                   "quickstart guide</a> to get all your keys and "
+                   "games.<br><br><small>(%1)</small>")
+                    .arg(errors));
         }
 
         QProgressDialog prog;

src/yuzu/main.h

@@ -196,8 +196,6 @@ private slots:
     void OnMenuLoadFile();
     void OnMenuLoadFolder();
     void OnMenuInstallToNAND();
-    /// Called whenever a user select the "File->Select -- Directory" where -- is NAND or SD Card
-    void OnMenuSelectEmulatedDirectory(EmulatedDirectoryTarget target);
     void OnMenuRecentFile();
     void OnConfigure();
     void OnLoadAmiibo();

src/yuzu/main.ui

@@ -64,8 +64,6 @@
     <addaction name="separator"/>
     <addaction name="menu_recent_files"/>
     <addaction name="separator"/>
-    <addaction name="action_Select_NAND_Directory"/>
-    <addaction name="action_Select_SDMC_Directory"/>
     <addaction name="separator"/>
     <addaction name="action_Load_Amiibo"/>
     <addaction name="separator"/>
@@ -217,22 +215,6 @@
     <string>Show Status Bar</string>
    </property>
   </action>
-  <action name="action_Select_NAND_Directory">
-   <property name="text">
-    <string>Select NAND Directory...</string>
-   </property>
-   <property name="toolTip">
-    <string>Selects a folder to use as the root of the emulated NAND</string>
-   </property>
-  </action>
-  <action name="action_Select_SDMC_Directory">
-   <property name="text">
-    <string>Select SD Card Directory...</string>
-   </property>
-   <property name="toolTip">
-    <string>Selects a folder to use as the root of the emulated SD card</string>
-   </property>
-  </action>
   <action name="action_Fullscreen">
    <property name="checkable">
     <bool>true</bool>

src/yuzu_cmd/config.cpp

@@ -425,6 +425,8 @@ void Config::ReadValues() {
     Settings::values.reporting_services =
         sdl2_config->GetBoolean("Debugging", "reporting_services", false);
     Settings::values.quest_flag = sdl2_config->GetBoolean("Debugging", "quest_flag", false);
+    Settings::values.disable_cpu_opt =
+        sdl2_config->GetBoolean("Debugging", "disable_cpu_opt", false);
 
     const auto title_list = sdl2_config->Get("AddOns", "title_ids", "");
     std::stringstream ss(title_list);

src/yuzu_cmd/default_ini.h

@@ -280,6 +280,9 @@ dump_nso=false
 # Determines whether or not yuzu will report to the game that the emulated console is in Kiosk Mode
 # false: Retail/Normal Mode (default), true: Kiosk Mode
 quest_flag =
+# Determines whether or not JIT CPU optimizations are enabled
+# false: Optimizations Enabled, true: Optimizations Disabled
+disable_cpu_opt =
 
 [WebService]
 # Whether or not to enable telemetry