vk_buffer_cache: Implement a buffer cache

This buffer cache is just like OpenGL's buffer cache with some minor
style changes. It uses VKStreamBuffer.

src/audio_core/audio_renderer.h

@@ -46,16 +46,18 @@ struct AudioRendererParameter {
     u32_le sample_rate;
     u32_le sample_count;
     u32_le mix_buffer_count;
-    u32_le unknown_c;
+    u32_le submix_count;
     u32_le voice_count;
     u32_le sink_count;
     u32_le effect_count;
-    u32_le unknown_1c;
-    u8 unknown_20;
-    INSERT_PADDING_BYTES(3);
+    u32_le performance_frame_count;
+    u8 is_voice_drop_enabled;
+    u8 unknown_21;
+    u8 unknown_22;
+    u8 execution_mode;
     u32_le splitter_count;
-    u32_le unknown_2c;
-    INSERT_PADDING_WORDS(1);
+    u32_le num_splitter_send_channels;
+    u32_le unknown_30;
     u32_le revision;
 };
 static_assert(sizeof(AudioRendererParameter) == 52, "AudioRendererParameter is an invalid size");

src/audio_core/cubeb_sink.cpp

@@ -12,6 +12,10 @@
 #include "common/ring_buffer.h"
 #include "core/settings.h"
 
+#ifdef _MSC_VER
+#include <objbase.h>
+#endif
+
 namespace AudioCore {
 
 class CubebSinkStream final : public SinkStream {
@@ -108,6 +112,11 @@ private:
 };
 
 CubebSink::CubebSink(std::string_view target_device_name) {
+    // Cubeb requires COM to be initialized on the thread calling cubeb_init on Windows
+#ifdef _MSC_VER
+    com_init_result = CoInitializeEx(nullptr, COINIT_MULTITHREADED);
+#endif
+
     if (cubeb_init(&ctx, "yuzu", nullptr) != CUBEB_OK) {
         LOG_CRITICAL(Audio_Sink, "cubeb_init failed");
         return;
@@ -142,6 +151,12 @@ CubebSink::~CubebSink() {
     }
 
     cubeb_destroy(ctx);
+
+#ifdef _MSC_VER
+    if (SUCCEEDED(com_init_result)) {
+        CoUninitialize();
+    }
+#endif
 }
 
 SinkStream& CubebSink::AcquireSinkStream(u32 sample_rate, u32 num_channels,

src/audio_core/cubeb_sink.h

@@ -25,6 +25,10 @@ private:
     cubeb* ctx{};
     cubeb_devid output_device{};
     std::vector<SinkStreamPtr> sink_streams;
+
+#ifdef _MSC_VER
+    u32 com_init_result = 0;
+#endif
 };
 
 std::vector<std::string> ListCubebSinkDevices();

src/common/color.h

@@ -55,36 +55,36 @@ constexpr u8 Convert8To6(u8 value) {
 /**
  * Decode a color stored in RGBA8 format
  * @param bytes Pointer to encoded source color
- * @return Result color decoded as Math::Vec4<u8>
+ * @return Result color decoded as Common::Vec4<u8>
  */
-inline Math::Vec4<u8> DecodeRGBA8(const u8* bytes) {
+inline Common::Vec4<u8> DecodeRGBA8(const u8* bytes) {
     return {bytes[3], bytes[2], bytes[1], bytes[0]};
 }
 
 /**
  * Decode a color stored in RGB8 format
  * @param bytes Pointer to encoded source color
- * @return Result color decoded as Math::Vec4<u8>
+ * @return Result color decoded as Common::Vec4<u8>
  */
-inline Math::Vec4<u8> DecodeRGB8(const u8* bytes) {
+inline Common::Vec4<u8> DecodeRGB8(const u8* bytes) {
     return {bytes[2], bytes[1], bytes[0], 255};
 }
 
 /**
  * Decode a color stored in RG8 (aka HILO8) format
  * @param bytes Pointer to encoded source color
- * @return Result color decoded as Math::Vec4<u8>
+ * @return Result color decoded as Common::Vec4<u8>
  */
-inline Math::Vec4<u8> DecodeRG8(const u8* bytes) {
+inline Common::Vec4<u8> DecodeRG8(const u8* bytes) {
     return {bytes[1], bytes[0], 0, 255};
 }
 
 /**
  * Decode a color stored in RGB565 format
  * @param bytes Pointer to encoded source color
- * @return Result color decoded as Math::Vec4<u8>
+ * @return Result color decoded as Common::Vec4<u8>
  */
-inline Math::Vec4<u8> DecodeRGB565(const u8* bytes) {
+inline Common::Vec4<u8> DecodeRGB565(const u8* bytes) {
     u16_le pixel;
     std::memcpy(&pixel, bytes, sizeof(pixel));
     return {Convert5To8((pixel >> 11) & 0x1F), Convert6To8((pixel >> 5) & 0x3F),
@@ -94,9 +94,9 @@ inline Math::Vec4<u8> DecodeRGB565(const u8* bytes) {
 /**
  * Decode a color stored in RGB5A1 format
  * @param bytes Pointer to encoded source color
- * @return Result color decoded as Math::Vec4<u8>
+ * @return Result color decoded as Common::Vec4<u8>
  */
-inline Math::Vec4<u8> DecodeRGB5A1(const u8* bytes) {
+inline Common::Vec4<u8> DecodeRGB5A1(const u8* bytes) {
     u16_le pixel;
     std::memcpy(&pixel, bytes, sizeof(pixel));
     return {Convert5To8((pixel >> 11) & 0x1F), Convert5To8((pixel >> 6) & 0x1F),
@@ -106,9 +106,9 @@ inline Math::Vec4<u8> DecodeRGB5A1(const u8* bytes) {
 /**
  * Decode a color stored in RGBA4 format
  * @param bytes Pointer to encoded source color
- * @return Result color decoded as Math::Vec4<u8>
+ * @return Result color decoded as Common::Vec4<u8>
  */
-inline Math::Vec4<u8> DecodeRGBA4(const u8* bytes) {
+inline Common::Vec4<u8> DecodeRGBA4(const u8* bytes) {
     u16_le pixel;
     std::memcpy(&pixel, bytes, sizeof(pixel));
     return {Convert4To8((pixel >> 12) & 0xF), Convert4To8((pixel >> 8) & 0xF),
@@ -138,9 +138,9 @@ inline u32 DecodeD24(const u8* bytes) {
 /**
  * Decode a depth value and a stencil value stored in D24S8 format
  * @param bytes Pointer to encoded source values
- * @return Resulting values stored as a Math::Vec2
+ * @return Resulting values stored as a Common::Vec2
  */
-inline Math::Vec2<u32> DecodeD24S8(const u8* bytes) {
+inline Common::Vec2<u32> DecodeD24S8(const u8* bytes) {
     return {static_cast<u32>((bytes[2] << 16) | (bytes[1] << 8) | bytes[0]), bytes[3]};
 }
 
@@ -149,7 +149,7 @@ inline Math::Vec2<u32> DecodeD24S8(const u8* bytes) {
  * @param color Source color to encode
  * @param bytes Destination pointer to store encoded color
  */
-inline void EncodeRGBA8(const Math::Vec4<u8>& color, u8* bytes) {
+inline void EncodeRGBA8(const Common::Vec4<u8>& color, u8* bytes) {
     bytes[3] = color.r();
     bytes[2] = color.g();
     bytes[1] = color.b();
@@ -161,7 +161,7 @@ inline void EncodeRGBA8(const Math::Vec4<u8>& color, u8* bytes) {
  * @param color Source color to encode
  * @param bytes Destination pointer to store encoded color
  */
-inline void EncodeRGB8(const Math::Vec4<u8>& color, u8* bytes) {
+inline void EncodeRGB8(const Common::Vec4<u8>& color, u8* bytes) {
     bytes[2] = color.r();
     bytes[1] = color.g();
     bytes[0] = color.b();
@@ -172,7 +172,7 @@ inline void EncodeRGB8(const Math::Vec4<u8>& color, u8* bytes) {
  * @param color Source color to encode
  * @param bytes Destination pointer to store encoded color
  */
-inline void EncodeRG8(const Math::Vec4<u8>& color, u8* bytes) {
+inline void EncodeRG8(const Common::Vec4<u8>& color, u8* bytes) {
     bytes[1] = color.r();
     bytes[0] = color.g();
 }
@@ -181,7 +181,7 @@ inline void EncodeRG8(const Math::Vec4<u8>& color, u8* bytes) {
  * @param color Source color to encode
  * @param bytes Destination pointer to store encoded color
  */
-inline void EncodeRGB565(const Math::Vec4<u8>& color, u8* bytes) {
+inline void EncodeRGB565(const Common::Vec4<u8>& color, u8* bytes) {
     const u16_le data =
         (Convert8To5(color.r()) << 11) | (Convert8To6(color.g()) << 5) | Convert8To5(color.b());
 
@@ -193,7 +193,7 @@ inline void EncodeRGB565(const Math::Vec4<u8>& color, u8* bytes) {
  * @param color Source color to encode
  * @param bytes Destination pointer to store encoded color
  */
-inline void EncodeRGB5A1(const Math::Vec4<u8>& color, u8* bytes) {
+inline void EncodeRGB5A1(const Common::Vec4<u8>& color, u8* bytes) {
     const u16_le data = (Convert8To5(color.r()) << 11) | (Convert8To5(color.g()) << 6) |
                         (Convert8To5(color.b()) << 1) | Convert8To1(color.a());
 
@@ -205,7 +205,7 @@ inline void EncodeRGB5A1(const Math::Vec4<u8>& color, u8* bytes) {
  * @param color Source color to encode
  * @param bytes Destination pointer to store encoded color
  */
-inline void EncodeRGBA4(const Math::Vec4<u8>& color, u8* bytes) {
+inline void EncodeRGBA4(const Common::Vec4<u8>& color, u8* bytes) {
     const u16 data = (Convert8To4(color.r()) << 12) | (Convert8To4(color.g()) << 8) |
                      (Convert8To4(color.b()) << 4) | Convert8To4(color.a());
 

src/common/math_util.h

@@ -7,7 +7,7 @@
 #include <cstdlib>
 #include <type_traits>
 
-namespace MathUtil {
+namespace Common {
 
 constexpr float PI = 3.14159265f;
 
@@ -41,4 +41,4 @@ struct Rectangle {
     }
 };
 
-} // namespace MathUtil
+} // namespace Common

src/common/quaternion.h

@@ -6,12 +6,12 @@
 
 #include "common/vector_math.h"
 
-namespace Math {
+namespace Common {
 
 template <typename T>
 class Quaternion {
 public:
-    Math::Vec3<T> xyz;
+    Vec3<T> xyz;
     T w{};
 
     Quaternion<decltype(-T{})> Inverse() const {
@@ -38,12 +38,12 @@ public:
 };
 
 template <typename T>
-auto QuaternionRotate(const Quaternion<T>& q, const Math::Vec3<T>& v) {
+auto QuaternionRotate(const Quaternion<T>& q, const Vec3<T>& v) {
     return v + 2 * Cross(q.xyz, Cross(q.xyz, v) + v * q.w);
 }
 
-inline Quaternion<float> MakeQuaternion(const Math::Vec3<float>& axis, float angle) {
+inline Quaternion<float> MakeQuaternion(const Vec3<float>& axis, float angle) {
     return {axis * std::sin(angle / 2), std::cos(angle / 2)};
 }
 
-} // namespace Math
+} // namespace Common

src/common/vector_math.h

@@ -33,7 +33,7 @@
 #include <cmath>
 #include <type_traits>
 
-namespace Math {
+namespace Common {
 
 template <typename T>
 class Vec2;
@@ -690,4 +690,4 @@ constexpr Vec4<T> MakeVec(const T& x, const Vec3<T>& yzw) {
     return MakeVec(x, yzw[0], yzw[1], yzw[2]);
 }
 
-} // namespace Math
+} // namespace Common

src/core/frontend/emu_window.cpp

@@ -67,7 +67,7 @@ static bool IsWithinTouchscreen(const Layout::FramebufferLayout& layout, unsigne
             framebuffer_x >= layout.screen.left && framebuffer_x < layout.screen.right);
 }
 
-std::tuple<unsigned, unsigned> EmuWindow::ClipToTouchScreen(unsigned new_x, unsigned new_y) {
+std::tuple<unsigned, unsigned> EmuWindow::ClipToTouchScreen(unsigned new_x, unsigned new_y) const {
     new_x = std::max(new_x, framebuffer_layout.screen.left);
     new_x = std::min(new_x, framebuffer_layout.screen.right - 1);
 

src/core/frontend/emu_window.h

@@ -166,7 +166,7 @@ private:
     /**
      * Clip the provided coordinates to be inside the touchscreen area.
      */
-    std::tuple<unsigned, unsigned> ClipToTouchScreen(unsigned new_x, unsigned new_y);
+    std::tuple<unsigned, unsigned> ClipToTouchScreen(unsigned new_x, unsigned new_y) const;
 };
 
 } // namespace Core::Frontend

src/core/frontend/framebuffer_layout.cpp

@@ -12,12 +12,12 @@ namespace Layout {
 
 // Finds the largest size subrectangle contained in window area that is confined to the aspect ratio
 template <class T>
-static MathUtil::Rectangle<T> maxRectangle(MathUtil::Rectangle<T> window_area,
-                                           float screen_aspect_ratio) {
+static Common::Rectangle<T> MaxRectangle(Common::Rectangle<T> window_area,
+                                         float screen_aspect_ratio) {
     float scale = std::min(static_cast<float>(window_area.GetWidth()),
                            window_area.GetHeight() / screen_aspect_ratio);
-    return MathUtil::Rectangle<T>{0, 0, static_cast<T>(std::round(scale)),
-                                  static_cast<T>(std::round(scale * screen_aspect_ratio))};
+    return Common::Rectangle<T>{0, 0, static_cast<T>(std::round(scale)),
+                                static_cast<T>(std::round(scale * screen_aspect_ratio))};
 }
 
 FramebufferLayout DefaultFrameLayout(unsigned width, unsigned height) {
@@ -29,8 +29,8 @@ FramebufferLayout DefaultFrameLayout(unsigned width, unsigned height) {
 
     const float emulation_aspect_ratio{static_cast<float>(ScreenUndocked::Height) /
                                        ScreenUndocked::Width};
-    MathUtil::Rectangle<unsigned> screen_window_area{0, 0, width, height};
-    MathUtil::Rectangle<unsigned> screen = maxRectangle(screen_window_area, emulation_aspect_ratio);
+    Common::Rectangle<unsigned> screen_window_area{0, 0, width, height};
+    Common::Rectangle<unsigned> screen = MaxRectangle(screen_window_area, emulation_aspect_ratio);
 
     float window_aspect_ratio = static_cast<float>(height) / width;
 

src/core/frontend/framebuffer_layout.h

@@ -16,7 +16,7 @@ struct FramebufferLayout {
     unsigned width{ScreenUndocked::Width};
     unsigned height{ScreenUndocked::Height};
 
-    MathUtil::Rectangle<unsigned> screen;
+    Common::Rectangle<unsigned> screen;
 
     /**
      * Returns the ration of pixel size of the screen, compared to the native size of the undocked

src/core/frontend/input.h

@@ -124,7 +124,7 @@ using AnalogDevice = InputDevice<std::tuple<float, float>>;
  *   Orientation is determined by right-hand rule.
  *   Units: deg/sec
  */
-using MotionDevice = InputDevice<std::tuple<Math::Vec3<float>, Math::Vec3<float>>>;
+using MotionDevice = InputDevice<std::tuple<Common::Vec3<float>, Common::Vec3<float>>>;
 
 /**
  * A touch device is an input device that returns a tuple of two floats and a bool. The floats are

src/core/hle/kernel/errors.h

@@ -14,6 +14,7 @@ constexpr ResultCode ERR_MAX_CONNECTIONS_REACHED{ErrorModule::Kernel, 7};
 constexpr ResultCode ERR_INVALID_CAPABILITY_DESCRIPTOR{ErrorModule::Kernel, 14};
 constexpr ResultCode ERR_INVALID_SIZE{ErrorModule::Kernel, 101};
 constexpr ResultCode ERR_INVALID_ADDRESS{ErrorModule::Kernel, 102};
+constexpr ResultCode ERR_OUT_OF_MEMORY{ErrorModule::Kernel, 104};
 constexpr ResultCode ERR_HANDLE_TABLE_FULL{ErrorModule::Kernel, 105};
 constexpr ResultCode ERR_INVALID_ADDRESS_STATE{ErrorModule::Kernel, 106};
 constexpr ResultCode ERR_INVALID_MEMORY_PERMISSIONS{ErrorModule::Kernel, 108};

src/core/hle/kernel/handle_table.cpp

@@ -14,32 +14,47 @@
 namespace Kernel {
 namespace {
 constexpr u16 GetSlot(Handle handle) {
-    return handle >> 15;
+    return static_cast<u16>(handle >> 15);
 }
 
 constexpr u16 GetGeneration(Handle handle) {
-    return handle & 0x7FFF;
+    return static_cast<u16>(handle & 0x7FFF);
 }
 } // Anonymous namespace
 
 HandleTable::HandleTable() {
-    next_generation = 1;
     Clear();
 }
 
 HandleTable::~HandleTable() = default;
 
+ResultCode HandleTable::SetSize(s32 handle_table_size) {
+    if (static_cast<u32>(handle_table_size) > MAX_COUNT) {
+        return ERR_OUT_OF_MEMORY;
+    }
+
+    // Values less than or equal to zero indicate to use the maximum allowable
+    // size for the handle table in the actual kernel, so we ignore the given
+    // value in that case, since we assume this by default unless this function
+    // is called.
+    if (handle_table_size > 0) {
+        table_size = static_cast<u16>(handle_table_size);
+    }
+
+    return RESULT_SUCCESS;
+}
+
 ResultVal<Handle> HandleTable::Create(SharedPtr<Object> obj) {
     DEBUG_ASSERT(obj != nullptr);
 
-    u16 slot = next_free_slot;
-    if (slot >= generations.size()) {
+    const u16 slot = next_free_slot;
+    if (slot >= table_size) {
         LOG_ERROR(Kernel, "Unable to allocate Handle, too many slots in use.");
         return ERR_HANDLE_TABLE_FULL;
     }
     next_free_slot = generations[slot];
 
-    u16 generation = next_generation++;
+    const u16 generation = next_generation++;
 
     // Overflow count so it fits in the 15 bits dedicated to the generation in the handle.
     // Horizon OS uses zero to represent an invalid handle, so skip to 1.
@@ -64,10 +79,11 @@ ResultVal<Handle> HandleTable::Duplicate(Handle handle) {
 }
 
 ResultCode HandleTable::Close(Handle handle) {
-    if (!IsValid(handle))
+    if (!IsValid(handle)) {
         return ERR_INVALID_HANDLE;
+    }
 
-    u16 slot = GetSlot(handle);
+    const u16 slot = GetSlot(handle);
 
     objects[slot] = nullptr;
 
@@ -77,10 +93,10 @@ ResultCode HandleTable::Close(Handle handle) {
 }
 
 bool HandleTable::IsValid(Handle handle) const {
-    std::size_t slot = GetSlot(handle);
-    u16 generation = GetGeneration(handle);
+    const std::size_t slot = GetSlot(handle);
+    const u16 generation = GetGeneration(handle);
 
-    return slot < MAX_COUNT && objects[slot] != nullptr && generations[slot] == generation;
+    return slot < table_size && objects[slot] != nullptr && generations[slot] == generation;
 }
 
 SharedPtr<Object> HandleTable::GetGeneric(Handle handle) const {
@@ -97,7 +113,7 @@ SharedPtr<Object> HandleTable::GetGeneric(Handle handle) const {
 }
 
 void HandleTable::Clear() {
-    for (u16 i = 0; i < MAX_COUNT; ++i) {
+    for (u16 i = 0; i < table_size; ++i) {
         generations[i] = i + 1;
         objects[i] = nullptr;
     }

src/core/hle/kernel/handle_table.h

@@ -49,6 +49,20 @@ public:
     HandleTable();
     ~HandleTable();
 
+    /**
+     * Sets the number of handles that may be in use at one time
+     * for this handle table.
+     *
+     * @param handle_table_size The desired size to limit the handle table to.
+     *
+     * @returns an error code indicating if initialization was successful.
+     *          If initialization was not successful, then ERR_OUT_OF_MEMORY
+     *          will be returned.
+     *
+     * @pre handle_table_size must be within the range [0, 1024]
+     */
+    ResultCode SetSize(s32 handle_table_size);
+
     /**
      * Allocates a handle for the given object.
      * @return The created Handle or one of the following errors:
@@ -103,14 +117,21 @@ private:
      */
     std::array<u16, MAX_COUNT> generations;
 
+    /**
+     * The limited size of the handle table. This can be specified by process
+     * capabilities in order to restrict the overall number of handles that
+     * can be created in a process instance
+     */
+    u16 table_size = static_cast<u16>(MAX_COUNT);
+
     /**
      * Global counter of the number of created handles. Stored in `generations` when a handle is
      * created, and wraps around to 1 when it hits 0x8000.
      */
-    u16 next_generation;
+    u16 next_generation = 1;
 
     /// Head of the free slots linked list.
-    u16 next_free_slot;
+    u16 next_free_slot = 0;
 };
 
 } // namespace Kernel

src/core/hle/kernel/process.cpp

@@ -99,7 +99,13 @@ ResultCode Process::LoadFromMetadata(const FileSys::ProgramMetadata& metadata) {
     vm_manager.Reset(metadata.GetAddressSpaceType());
 
     const auto& caps = metadata.GetKernelCapabilities();
-    return capabilities.InitializeForUserProcess(caps.data(), caps.size(), vm_manager);
+    const auto capability_init_result =
+        capabilities.InitializeForUserProcess(caps.data(), caps.size(), vm_manager);
+    if (capability_init_result.IsError()) {
+        return capability_init_result;
+    }
+
+    return handle_table.SetSize(capabilities.GetHandleTableSize());
 }
 
 void Process::Run(VAddr entry_point, s32 main_thread_priority, u32 stack_size) {

src/core/hle/kernel/process_capability.cpp

@@ -96,7 +96,7 @@ void ProcessCapabilities::InitializeForMetadatalessProcess() {
     interrupt_capabilities.set();
 
     // Allow using the maximum possible amount of handles
-    handle_table_size = static_cast<u32>(HandleTable::MAX_COUNT);
+    handle_table_size = static_cast<s32>(HandleTable::MAX_COUNT);
 
     // Allow all debugging capabilities.
     is_debuggable = true;
@@ -337,7 +337,7 @@ ResultCode ProcessCapabilities::HandleHandleTableFlags(u32 flags) {
         return ERR_RESERVED_VALUE;
     }
 
-    handle_table_size = (flags >> 16) & 0x3FF;
+    handle_table_size = static_cast<s32>((flags >> 16) & 0x3FF);
     return RESULT_SUCCESS;
 }
 

src/core/hle/kernel/process_capability.h

@@ -156,7 +156,7 @@ public:
     }
 
     /// Gets the number of total allowable handles for the process' handle table.
-    u32 GetHandleTableSize() const {
+    s32 GetHandleTableSize() const {
         return handle_table_size;
     }
 
@@ -252,7 +252,7 @@ private:
     u64 core_mask = 0;
     u64 priority_mask = 0;
 
-    u32 handle_table_size = 0;
+    s32 handle_table_size = 0;
     u32 kernel_version = 0;
 
     ProgramType program_type = ProgramType::SysModule;

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

@@ -262,20 +262,20 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service_Audio, "called");
 
     u64 buffer_sz = Common::AlignUp(4 * params.mix_buffer_count, 0x40);
-    buffer_sz += params.unknown_c * 1024;
-    buffer_sz += 0x940 * (params.unknown_c + 1);
+    buffer_sz += params.submix_count * 1024;
+    buffer_sz += 0x940 * (params.submix_count + 1);
     buffer_sz += 0x3F0 * params.voice_count;
-    buffer_sz += Common::AlignUp(8 * (params.unknown_c + 1), 0x10);
+    buffer_sz += Common::AlignUp(8 * (params.submix_count + 1), 0x10);
     buffer_sz += Common::AlignUp(8 * params.voice_count, 0x10);
-    buffer_sz +=
-        Common::AlignUp((0x3C0 * (params.sink_count + params.unknown_c) + 4 * params.sample_count) *
-                            (params.mix_buffer_count + 6),
-                        0x40);
+    buffer_sz += Common::AlignUp(
+        (0x3C0 * (params.sink_count + params.submix_count) + 4 * params.sample_count) *
+            (params.mix_buffer_count + 6),
+        0x40);
 
     if (IsFeatureSupported(AudioFeatures::Splitter, params.revision)) {
-        u32 count = params.unknown_c + 1;
+        const u32 count = params.submix_count + 1;
         u64 node_count = Common::AlignUp(count, 0x40);
-        u64 node_state_buffer_sz =
+        const u64 node_state_buffer_sz =
             4 * (node_count * node_count) + 0xC * node_count + 2 * (node_count / 8);
         u64 edge_matrix_buffer_sz = 0;
         node_count = Common::AlignUp(count * count, 0x40);
@@ -289,19 +289,19 @@ void AudRenU::GetAudioRendererWorkBufferSize(Kernel::HLERequestContext& ctx) {
 
     buffer_sz += 0x20 * (params.effect_count + 4 * params.voice_count) + 0x50;
     if (IsFeatureSupported(AudioFeatures::Splitter, params.revision)) {
-        buffer_sz += 0xE0 * params.unknown_2c;
+        buffer_sz += 0xE0 * params.num_splitter_send_channels;
         buffer_sz += 0x20 * params.splitter_count;
-        buffer_sz += Common::AlignUp(4 * params.unknown_2c, 0x10);
+        buffer_sz += Common::AlignUp(4 * params.num_splitter_send_channels, 0x10);
     }
     buffer_sz = Common::AlignUp(buffer_sz, 0x40) + 0x170 * params.sink_count;
     u64 output_sz = buffer_sz + 0x280 * params.sink_count + 0x4B0 * params.effect_count +
                     ((params.voice_count * 256) | 0x40);
 
-    if (params.unknown_1c >= 1) {
+    if (params.performance_frame_count >= 1) {
         output_sz = Common::AlignUp(((16 * params.sink_count + 16 * params.effect_count +
                                       16 * params.voice_count + 16) +
                                      0x658) *
-                                            (params.unknown_1c + 1) +
+                                            (params.performance_frame_count + 1) +
                                         0xc0,
                                     0x40) +
                     output_sz;

src/core/hle/service/hid/hid.h

@@ -15,7 +15,7 @@ namespace Kernel {
 class SharedMemory;
 }
 
-namespace SM {
+namespace Service::SM {
 class ServiceManager;
 }
 

src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp

@@ -23,7 +23,7 @@ u32 nvdisp_disp0::ioctl(Ioctl command, const std::vector<u8>& input, std::vector
 
 void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u32 height,
                         u32 stride, NVFlinger::BufferQueue::BufferTransformFlags transform,
-                        const MathUtil::Rectangle<int>& crop_rect) {
+                        const Common::Rectangle<int>& crop_rect) {
     VAddr addr = nvmap_dev->GetObjectAddress(buffer_handle);
     LOG_TRACE(Service,
               "Drawing from address {:X} offset {:08X} Width {} Height {} Stride {} Format {}",

src/core/hle/service/nvdrv/devices/nvdisp_disp0.h

@@ -25,7 +25,7 @@ public:
     /// Performs a screen flip, drawing the buffer pointed to by the handle.
     void flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u32 height, u32 stride,
               NVFlinger::BufferQueue::BufferTransformFlags transform,
-              const MathUtil::Rectangle<int>& crop_rect);
+              const Common::Rectangle<int>& crop_rect);
 
 private:
     std::shared_ptr<nvmap> nvmap_dev;

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

@@ -63,7 +63,7 @@ const IGBPBuffer& BufferQueue::RequestBuffer(u32 slot) const {
 }
 
 void BufferQueue::QueueBuffer(u32 slot, BufferTransformFlags transform,
-                              const MathUtil::Rectangle<int>& crop_rect) {
+                              const Common::Rectangle<int>& crop_rect) {
     auto itr = std::find_if(queue.begin(), queue.end(),
                             [&](const Buffer& buffer) { return buffer.slot == slot; });
     ASSERT(itr != queue.end());

src/core/hle/service/nvflinger/buffer_queue.h

@@ -67,14 +67,14 @@ public:
         Status status = Status::Free;
         IGBPBuffer igbp_buffer;
         BufferTransformFlags transform;
-        MathUtil::Rectangle<int> crop_rect;
+        Common::Rectangle<int> crop_rect;
     };
 
     void SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer);
     std::optional<u32> DequeueBuffer(u32 width, u32 height);
     const IGBPBuffer& RequestBuffer(u32 slot) const;
     void QueueBuffer(u32 slot, BufferTransformFlags transform,
-                     const MathUtil::Rectangle<int>& crop_rect);
+                     const Common::Rectangle<int>& crop_rect);
     std::optional<std::reference_wrapper<const Buffer>> AcquireBuffer();
     void ReleaseBuffer(u32 slot);
     u32 Query(QueryType type);

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

@@ -420,7 +420,7 @@ public:
         u32_le fence_is_valid;
         std::array<Fence, 2> fences;
 
-        MathUtil::Rectangle<int> GetCropRect() const {
+        Common::Rectangle<int> GetCropRect() const {
             return {crop_left, crop_top, crop_right, crop_bottom};
         }
     };

src/core/memory.cpp

@@ -71,15 +71,20 @@ static void MapPages(PageTable& page_table, VAddr base, u64 size, u8* memory, Pa
                                  FlushMode::FlushAndInvalidate);
 
     VAddr end = base + size;
-    while (base != end) {
-        ASSERT_MSG(base < page_table.pointers.size(), "out of range mapping at {:016X}", base);
+    ASSERT_MSG(end <= page_table.pointers.size(), "out of range mapping at {:016X}",
+               base + page_table.pointers.size());
 
-        page_table.attributes[base] = type;
-        page_table.pointers[base] = memory;
+    std::fill(page_table.attributes.begin() + base, page_table.attributes.begin() + end, type);
 
-        base += 1;
-        if (memory != nullptr)
+    if (memory == nullptr) {
+        std::fill(page_table.pointers.begin() + base, page_table.pointers.begin() + end, memory);
+    } else {
+        while (base != end) {
+            page_table.pointers[base] = memory;
+
+            base += 1;
             memory += PAGE_SIZE;
+        }
     }
 }
 

src/input_common/motion_emu.cpp

@@ -32,12 +32,12 @@ public:
     }
 
     void BeginTilt(int x, int y) {
-        mouse_origin = Math::MakeVec(x, y);
+        mouse_origin = Common::MakeVec(x, y);
         is_tilting = true;
     }
 
     void Tilt(int x, int y) {
-        auto mouse_move = Math::MakeVec(x, y) - mouse_origin;
+        auto mouse_move = Common::MakeVec(x, y) - mouse_origin;
         if (is_tilting) {
             std::lock_guard<std::mutex> guard(tilt_mutex);
             if (mouse_move.x == 0 && mouse_move.y == 0) {
@@ -45,7 +45,7 @@ public:
             } else {
                 tilt_direction = mouse_move.Cast<float>();
                 tilt_angle =
-                    std::clamp(tilt_direction.Normalize() * sensitivity, 0.0f, MathUtil::PI * 0.5f);
+                    std::clamp(tilt_direction.Normalize() * sensitivity, 0.0f, Common::PI * 0.5f);
             }
         }
     }
@@ -56,7 +56,7 @@ public:
         is_tilting = false;
     }
 
-    std::tuple<Math::Vec3<float>, Math::Vec3<float>> GetStatus() {
+    std::tuple<Common::Vec3<float>, Common::Vec3<float>> GetStatus() {
         std::lock_guard<std::mutex> guard(status_mutex);
         return status;
     }
@@ -66,17 +66,17 @@ private:
     const std::chrono::steady_clock::duration update_duration;
     const float sensitivity;
 
-    Math::Vec2<int> mouse_origin;
+    Common::Vec2<int> mouse_origin;
 
     std::mutex tilt_mutex;
-    Math::Vec2<float> tilt_direction;
+    Common::Vec2<float> tilt_direction;
     float tilt_angle = 0;
 
     bool is_tilting = false;
 
     Common::Event shutdown_event;
 
-    std::tuple<Math::Vec3<float>, Math::Vec3<float>> status;
+    std::tuple<Common::Vec3<float>, Common::Vec3<float>> status;
     std::mutex status_mutex;
 
     // Note: always keep the thread declaration at the end so that other objects are initialized
@@ -85,8 +85,8 @@ private:
 
     void MotionEmuThread() {
         auto update_time = std::chrono::steady_clock::now();
-        Math::Quaternion<float> q = MakeQuaternion(Math::Vec3<float>(), 0);
-        Math::Quaternion<float> old_q;
+        Common::Quaternion<float> q = Common::MakeQuaternion(Common::Vec3<float>(), 0);
+        Common::Quaternion<float> old_q;
 
         while (!shutdown_event.WaitUntil(update_time)) {
             update_time += update_duration;
@@ -96,18 +96,18 @@ private:
                 std::lock_guard<std::mutex> guard(tilt_mutex);
 
                 // Find the quaternion describing current 3DS tilting
-                q = MakeQuaternion(Math::MakeVec(-tilt_direction.y, 0.0f, tilt_direction.x),
-                                   tilt_angle);
+                q = Common::MakeQuaternion(
+                    Common::MakeVec(-tilt_direction.y, 0.0f, tilt_direction.x), tilt_angle);
             }
 
             auto inv_q = q.Inverse();
 
             // Set the gravity vector in world space
-            auto gravity = Math::MakeVec(0.0f, -1.0f, 0.0f);
+            auto gravity = Common::MakeVec(0.0f, -1.0f, 0.0f);
 
             // Find the angular rate vector in world space
             auto angular_rate = ((q - old_q) * inv_q).xyz * 2;
-            angular_rate *= 1000 / update_millisecond / MathUtil::PI * 180;
+            angular_rate *= 1000 / update_millisecond / Common::PI * 180;
 
             // Transform the two vectors from world space to 3DS space
             gravity = QuaternionRotate(inv_q, gravity);
@@ -131,7 +131,7 @@ public:
         device = std::make_shared<MotionEmuDevice>(update_millisecond, sensitivity);
     }
 
-    std::tuple<Math::Vec3<float>, Math::Vec3<float>> GetStatus() const override {
+    std::tuple<Common::Vec3<float>, Common::Vec3<float>> GetStatus() const override {
         return device->GetStatus();
     }
 

src/video_core/CMakeLists.txt

@@ -104,6 +104,8 @@ add_library(video_core STATIC
 if (ENABLE_VULKAN)
     target_sources(video_core PRIVATE
         renderer_vulkan/declarations.h
+        renderer_vulkan/vk_buffer_cache.cpp
+        renderer_vulkan/vk_buffer_cache.h
         renderer_vulkan/vk_device.cpp
         renderer_vulkan/vk_device.h
         renderer_vulkan/vk_memory_manager.cpp
@@ -111,7 +113,9 @@ if (ENABLE_VULKAN)
         renderer_vulkan/vk_resource_manager.cpp
         renderer_vulkan/vk_resource_manager.h
         renderer_vulkan/vk_scheduler.cpp
-        renderer_vulkan/vk_scheduler.h)
+        renderer_vulkan/vk_scheduler.h
+        renderer_vulkan/vk_stream_buffer.cpp
+        renderer_vulkan/vk_stream_buffer.h)
 
     target_include_directories(video_core PRIVATE ../../externals/Vulkan-Headers/include)
     target_compile_definitions(video_core PRIVATE HAS_VULKAN)

src/video_core/engines/fermi_2d.cpp

@@ -44,10 +44,10 @@ void Fermi2D::HandleSurfaceCopy() {
     const u32 src_blit_y2{
         static_cast<u32>((regs.blit_src_y + (regs.blit_dst_height * regs.blit_dv_dy)) >> 32)};
 
-    const MathUtil::Rectangle<u32> src_rect{src_blit_x1, src_blit_y1, src_blit_x2, src_blit_y2};
-    const MathUtil::Rectangle<u32> dst_rect{regs.blit_dst_x, regs.blit_dst_y,
-                                            regs.blit_dst_x + regs.blit_dst_width,
-                                            regs.blit_dst_y + regs.blit_dst_height};
+    const Common::Rectangle<u32> src_rect{src_blit_x1, src_blit_y1, src_blit_x2, src_blit_y2};
+    const Common::Rectangle<u32> dst_rect{regs.blit_dst_x, regs.blit_dst_y,
+                                          regs.blit_dst_x + regs.blit_dst_width,
+                                          regs.blit_dst_y + regs.blit_dst_height};
 
     if (!rasterizer.AccelerateSurfaceCopy(regs.src, regs.dst, src_rect, dst_rect)) {
         UNIMPLEMENTED();

src/video_core/engines/maxwell_3d.cpp

@@ -107,21 +107,23 @@ void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
 void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
     auto debug_context = system.GetGPUDebugContext();
 
+    const u32 method = method_call.method;
+
     // It is an error to write to a register other than the current macro's ARG register before it
     // has finished execution.
     if (executing_macro != 0) {
-        ASSERT(method_call.method == executing_macro + 1);
+        ASSERT(method == executing_macro + 1);
     }
 
     // Methods after 0xE00 are special, they're actually triggers for some microcode that was
     // uploaded to the GPU during initialization.
-    if (method_call.method >= MacroRegistersStart) {
+    if (method >= MacroRegistersStart) {
         // We're trying to execute a macro
         if (executing_macro == 0) {
             // A macro call must begin by writing the macro method's register, not its argument.
-            ASSERT_MSG((method_call.method % 2) == 0,
+            ASSERT_MSG((method % 2) == 0,
                        "Can't start macro execution by writing to the ARGS register");
-            executing_macro = method_call.method;
+            executing_macro = method;
         }
 
         macro_params.push_back(method_call.argument);
@@ -133,66 +135,62 @@ void Maxwell3D::CallMethod(const GPU::MethodCall& method_call) {
         return;
     }
 
-    ASSERT_MSG(method_call.method < Regs::NUM_REGS,
+    ASSERT_MSG(method < Regs::NUM_REGS,
                "Invalid Maxwell3D register, increase the size of the Regs structure");
 
     if (debug_context) {
         debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandLoaded, nullptr);
     }
 
-    if (regs.reg_array[method_call.method] != method_call.argument) {
-        regs.reg_array[method_call.method] = method_call.argument;
+    if (regs.reg_array[method] != method_call.argument) {
+        regs.reg_array[method] = method_call.argument;
         // Color buffers
         constexpr u32 first_rt_reg = MAXWELL3D_REG_INDEX(rt);
         constexpr u32 registers_per_rt = sizeof(regs.rt[0]) / sizeof(u32);
-        if (method_call.method >= first_rt_reg &&
-            method_call.method < first_rt_reg + registers_per_rt * Regs::NumRenderTargets) {
-            const std::size_t rt_index = (method_call.method - first_rt_reg) / registers_per_rt;
-            dirty_flags.color_buffer |= 1u << static_cast<u32>(rt_index);
+        if (method >= first_rt_reg &&
+            method < first_rt_reg + registers_per_rt * Regs::NumRenderTargets) {
+            const std::size_t rt_index = (method - first_rt_reg) / registers_per_rt;
+            dirty_flags.color_buffer.set(rt_index);
         }
 
         // Zeta buffer
         constexpr u32 registers_in_zeta = sizeof(regs.zeta) / sizeof(u32);
-        if (method_call.method == MAXWELL3D_REG_INDEX(zeta_enable) ||
-            method_call.method == MAXWELL3D_REG_INDEX(zeta_width) ||
-            method_call.method == MAXWELL3D_REG_INDEX(zeta_height) ||
-            (method_call.method >= MAXWELL3D_REG_INDEX(zeta) &&
-             method_call.method < MAXWELL3D_REG_INDEX(zeta) + registers_in_zeta)) {
+        if (method == MAXWELL3D_REG_INDEX(zeta_enable) ||
+            method == MAXWELL3D_REG_INDEX(zeta_width) ||
+            method == MAXWELL3D_REG_INDEX(zeta_height) ||
+            (method >= MAXWELL3D_REG_INDEX(zeta) &&
+             method < MAXWELL3D_REG_INDEX(zeta) + registers_in_zeta)) {
             dirty_flags.zeta_buffer = true;
         }
 
         // Shader
         constexpr u32 shader_registers_count =
             sizeof(regs.shader_config[0]) * Regs::MaxShaderProgram / sizeof(u32);
-        if (method_call.method >= MAXWELL3D_REG_INDEX(shader_config[0]) &&
-            method_call.method < MAXWELL3D_REG_INDEX(shader_config[0]) + shader_registers_count) {
+        if (method >= MAXWELL3D_REG_INDEX(shader_config[0]) &&
+            method < MAXWELL3D_REG_INDEX(shader_config[0]) + shader_registers_count) {
             dirty_flags.shaders = true;
         }
 
         // Vertex format
-        if (method_call.method >= MAXWELL3D_REG_INDEX(vertex_attrib_format) &&
-            method_call.method <
-                MAXWELL3D_REG_INDEX(vertex_attrib_format) + regs.vertex_attrib_format.size()) {
+        if (method >= MAXWELL3D_REG_INDEX(vertex_attrib_format) &&
+            method < MAXWELL3D_REG_INDEX(vertex_attrib_format) + regs.vertex_attrib_format.size()) {
             dirty_flags.vertex_attrib_format = true;
         }
 
         // Vertex buffer
-        if (method_call.method >= MAXWELL3D_REG_INDEX(vertex_array) &&
-            method_call.method < MAXWELL3D_REG_INDEX(vertex_array) + 4 * 32) {
-            dirty_flags.vertex_array |=
-                1u << ((method_call.method - MAXWELL3D_REG_INDEX(vertex_array)) >> 2);
-        } else if (method_call.method >= MAXWELL3D_REG_INDEX(vertex_array_limit) &&
-                   method_call.method < MAXWELL3D_REG_INDEX(vertex_array_limit) + 2 * 32) {
-            dirty_flags.vertex_array |=
-                1u << ((method_call.method - MAXWELL3D_REG_INDEX(vertex_array_limit)) >> 1);
-        } else if (method_call.method >= MAXWELL3D_REG_INDEX(instanced_arrays) &&
-                   method_call.method < MAXWELL3D_REG_INDEX(instanced_arrays) + 32) {
-            dirty_flags.vertex_array |=
-                1u << (method_call.method - MAXWELL3D_REG_INDEX(instanced_arrays));
+        if (method >= MAXWELL3D_REG_INDEX(vertex_array) &&
+            method < MAXWELL3D_REG_INDEX(vertex_array) + 4 * 32) {
+            dirty_flags.vertex_array.set((method - MAXWELL3D_REG_INDEX(vertex_array)) >> 2);
+        } else if (method >= MAXWELL3D_REG_INDEX(vertex_array_limit) &&
+                   method < MAXWELL3D_REG_INDEX(vertex_array_limit) + 2 * 32) {
+            dirty_flags.vertex_array.set((method - MAXWELL3D_REG_INDEX(vertex_array_limit)) >> 1);
+        } else if (method >= MAXWELL3D_REG_INDEX(instanced_arrays) &&
+                   method < MAXWELL3D_REG_INDEX(instanced_arrays) + 32) {
+            dirty_flags.vertex_array.set(method - MAXWELL3D_REG_INDEX(instanced_arrays));
         }
     }
 
-    switch (method_call.method) {
+    switch (method) {
     case MAXWELL3D_REG_INDEX(macros.data): {
         ProcessMacroUpload(method_call.argument);
         break;

src/video_core/engines/maxwell_3d.h

@@ -5,8 +5,10 @@
 #pragma once
 
 #include <array>
+#include <bitset>
 #include <unordered_map>
 #include <vector>
+
 #include "common/assert.h"
 #include "common/bit_field.h"
 #include "common/common_funcs.h"
@@ -503,7 +505,7 @@ public:
             f32 translate_z;
             INSERT_PADDING_WORDS(2);
 
-            MathUtil::Rectangle<s32> GetRect() const {
+            Common::Rectangle<s32> GetRect() const {
                 return {
                     GetX(),               // left
                     GetY() + GetHeight(), // top
@@ -1094,19 +1096,18 @@ public:
     MemoryManager& memory_manager;
 
     struct DirtyFlags {
-        u8 color_buffer = 0xFF;
-        bool zeta_buffer = true;
-
-        bool shaders = true;
+        std::bitset<8> color_buffer{0xFF};
+        std::bitset<32> vertex_array{0xFFFFFFFF};
 
         bool vertex_attrib_format = true;
-        u32 vertex_array = 0xFFFFFFFF;
+        bool zeta_buffer = true;
+        bool shaders = true;
 
         void OnMemoryWrite() {
-            color_buffer = 0xFF;
             zeta_buffer = true;
             shaders = true;
-            vertex_array = 0xFFFFFFFF;
+            color_buffer.set();
+            vertex_array.set();
         }
     };
 

src/video_core/gpu.h

@@ -100,7 +100,7 @@ struct FramebufferConfig {
 
     using TransformFlags = Service::NVFlinger::BufferQueue::BufferTransformFlags;
     TransformFlags transform_flags;
-    MathUtil::Rectangle<int> crop_rect;
+    Common::Rectangle<int> crop_rect;
 };
 
 namespace Engines {

src/video_core/rasterizer_cache.h

@@ -129,6 +129,15 @@ protected:
         return ++modified_ticks;
     }
 
+    /// Flushes the specified object, updating appropriate cache state as needed
+    void FlushObject(const T& object) {
+        if (!object->IsDirty()) {
+            return;
+        }
+        object->Flush();
+        object->MarkAsModified(false, *this);
+    }
+
 private:
     /// Returns a list of cached objects from the specified memory region, ordered by access time
     std::vector<T> GetSortedObjectsFromRegion(VAddr addr, u64 size) {
@@ -154,15 +163,6 @@ private:
         return objects;
     }
 
-    /// Flushes the specified object, updating appropriate cache state as needed
-    void FlushObject(const T& object) {
-        if (!object->IsDirty()) {
-            return;
-        }
-        object->Flush();
-        object->MarkAsModified(false, *this);
-    }
-
     using ObjectSet = std::set<T>;
     using ObjectCache = std::unordered_map<VAddr, T>;
     using IntervalCache = boost::icl::interval_map<VAddr, ObjectSet>;

src/video_core/rasterizer_interface.h

@@ -47,8 +47,8 @@ public:
     /// Attempt to use a faster method to perform a surface copy
     virtual bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                        const Tegra::Engines::Fermi2D::Regs::Surface& dst,
-                                       const MathUtil::Rectangle<u32>& src_rect,
-                                       const MathUtil::Rectangle<u32>& dst_rect) {
+                                       const Common::Rectangle<u32>& src_rect,
+                                       const Common::Rectangle<u32>& dst_rect) {
         return false;
     }
 

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -102,8 +102,8 @@ struct FramebufferCacheKey {
 
 RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& window, Core::System& system,
                                    ScreenInfo& info)
-    : res_cache{*this}, shader_cache{*this, system}, emu_window{window}, screen_info{info},
-      buffer_cache(*this, STREAM_BUFFER_SIZE), global_cache{*this} {
+    : res_cache{*this}, shader_cache{*this, system}, global_cache{*this}, emu_window{window},
+      screen_info{info}, buffer_cache(*this, STREAM_BUFFER_SIZE) {
     // Create sampler objects
     for (std::size_t i = 0; i < texture_samplers.size(); ++i) {
         texture_samplers[i].Create();
@@ -200,7 +200,7 @@ GLuint RasterizerOpenGL::SetupVertexFormat() {
     }
 
     // Rebinding the VAO invalidates the vertex buffer bindings.
-    gpu.dirty_flags.vertex_array = 0xFFFFFFFF;
+    gpu.dirty_flags.vertex_array.set();
 
     state.draw.vertex_array = vao_entry.handle;
     return vao_entry.handle;
@@ -210,14 +210,14 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
     auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
     const auto& regs = gpu.regs;
 
-    if (!gpu.dirty_flags.vertex_array)
+    if (gpu.dirty_flags.vertex_array.none())
         return;
 
     MICROPROFILE_SCOPE(OpenGL_VB);
 
     // Upload all guest vertex arrays sequentially to our buffer
     for (u32 index = 0; index < Maxwell::NumVertexArrays; ++index) {
-        if (~gpu.dirty_flags.vertex_array & (1u << index))
+        if (!gpu.dirty_flags.vertex_array[index])
             continue;
 
         const auto& vertex_array = regs.vertex_array[index];
@@ -244,7 +244,7 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
         }
     }
 
-    gpu.dirty_flags.vertex_array = 0;
+    gpu.dirty_flags.vertex_array.reset();
 }
 
 DrawParameters RasterizerOpenGL::SetupDraw() {
@@ -488,13 +488,13 @@ std::pair<bool, bool> RasterizerOpenGL::ConfigureFramebuffers(
     OpenGLState& current_state, bool using_color_fb, bool using_depth_fb, bool preserve_contents,
     std::optional<std::size_t> single_color_target) {
     MICROPROFILE_SCOPE(OpenGL_Framebuffer);
-    const auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
+    auto& gpu = Core::System::GetInstance().GPU().Maxwell3D();
     const auto& regs = gpu.regs;
 
     const FramebufferConfigState fb_config_state{using_color_fb, using_depth_fb, preserve_contents,
                                                  single_color_target};
-    if (fb_config_state == current_framebuffer_config_state && gpu.dirty_flags.color_buffer == 0 &&
-        !gpu.dirty_flags.zeta_buffer) {
+    if (fb_config_state == current_framebuffer_config_state &&
+        gpu.dirty_flags.color_buffer.none() && !gpu.dirty_flags.zeta_buffer) {
         // Only skip if the previous ConfigureFramebuffers call was from the same kind (multiple or
         // single color targets). This is done because the guest registers may not change but the
         // host framebuffer may contain different attachments
@@ -721,10 +721,10 @@ void RasterizerOpenGL::DrawArrays() {
     // Add space for at least 18 constant buffers
     buffer_size += Maxwell::MaxConstBuffers * (MaxConstbufferSize + uniform_buffer_alignment);
 
-    bool invalidate = buffer_cache.Map(buffer_size);
+    const bool invalidate = buffer_cache.Map(buffer_size);
     if (invalidate) {
         // As all cached buffers are invalidated, we need to recheck their state.
-        gpu.dirty_flags.vertex_array = 0xFFFFFFFF;
+        gpu.dirty_flags.vertex_array.set();
     }
 
     const GLuint vao = SetupVertexFormat();
@@ -738,9 +738,13 @@ void RasterizerOpenGL::DrawArrays() {
     shader_program_manager->ApplyTo(state);
     state.Apply();
 
+    res_cache.SignalPreDrawCall();
+
     // Execute draw call
     params.DispatchDraw();
 
+    res_cache.SignalPostDrawCall();
+
     // Disable scissor test
     state.viewports[0].scissor.enabled = false;
 
@@ -779,8 +783,8 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
 
 bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                              const Tegra::Engines::Fermi2D::Regs::Surface& dst,
-                                             const MathUtil::Rectangle<u32>& src_rect,
-                                             const MathUtil::Rectangle<u32>& dst_rect) {
+                                             const Common::Rectangle<u32>& src_rect,
+                                             const Common::Rectangle<u32>& dst_rect) {
     MICROPROFILE_SCOPE(OpenGL_Blits);
     res_cache.FermiCopySurface(src, dst, src_rect, dst_rect);
     return true;
@@ -1034,7 +1038,7 @@ void RasterizerOpenGL::SyncViewport(OpenGLState& current_state) {
     for (std::size_t i = 0; i < viewport_count; i++) {
         auto& viewport = current_state.viewports[i];
         const auto& src = regs.viewports[i];
-        const MathUtil::Rectangle<s32> viewport_rect{regs.viewport_transform[i].GetRect()};
+        const Common::Rectangle<s32> viewport_rect{regs.viewport_transform[i].GetRect()};
         viewport.x = viewport_rect.left;
         viewport.y = viewport_rect.bottom;
         viewport.width = viewport_rect.GetWidth();

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -62,8 +62,8 @@ public:
     void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
     bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                const Tegra::Engines::Fermi2D::Regs::Surface& dst,
-                               const MathUtil::Rectangle<u32>& src_rect,
-                               const MathUtil::Rectangle<u32>& dst_rect) override;
+                               const Common::Rectangle<u32>& src_rect,
+                               const Common::Rectangle<u32>& dst_rect) override;
     bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
                            u32 pixel_stride) override;
     bool AccelerateDrawBatch(bool is_indexed) override;

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
+#include <optional>
 #include <glad/glad.h>
 
 #include "common/alignment.h"
@@ -399,7 +400,7 @@ static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType
     return format;
 }
 
-MathUtil::Rectangle<u32> SurfaceParams::GetRect(u32 mip_level) const {
+Common::Rectangle<u32> SurfaceParams::GetRect(u32 mip_level) const {
     u32 actual_height{std::max(1U, unaligned_height >> mip_level)};
     if (IsPixelFormatASTC(pixel_format)) {
         // ASTC formats must stop at the ATSC block size boundary
@@ -549,6 +550,8 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
     // alternatives. This signals a bug on those functions.
     const auto width = static_cast<GLsizei>(params.MipWidth(0));
     const auto height = static_cast<GLsizei>(params.MipHeight(0));
+    memory_size = params.MemorySize();
+    reinterpreted = false;
 
     const auto& format_tuple = GetFormatTuple(params.pixel_format, params.component_type);
     gl_internal_format = format_tuple.internal_format;
@@ -962,30 +965,31 @@ Surface RasterizerCacheOpenGL::GetColorBufferSurface(std::size_t index, bool pre
     auto& gpu{Core::System::GetInstance().GPU().Maxwell3D()};
     const auto& regs{gpu.regs};
 
-    if ((gpu.dirty_flags.color_buffer & (1u << static_cast<u32>(index))) == 0) {
+    if (!gpu.dirty_flags.color_buffer[index]) {
         return last_color_buffers[index];
     }
-    gpu.dirty_flags.color_buffer &= ~(1u << static_cast<u32>(index));
+    gpu.dirty_flags.color_buffer.reset(index);
 
     ASSERT(index < Tegra::Engines::Maxwell3D::Regs::NumRenderTargets);
 
     if (index >= regs.rt_control.count) {
-        return last_color_buffers[index] = {};
+        return current_color_buffers[index] = {};
     }
 
     if (regs.rt[index].Address() == 0 || regs.rt[index].format == Tegra::RenderTargetFormat::NONE) {
-        return last_color_buffers[index] = {};
+        return current_color_buffers[index] = {};
     }
 
     const SurfaceParams color_params{SurfaceParams::CreateForFramebuffer(index)};
 
-    return last_color_buffers[index] = GetSurface(color_params, preserve_contents);
+    return current_color_buffers[index] = GetSurface(color_params, preserve_contents);
 }
 
 void RasterizerCacheOpenGL::LoadSurface(const Surface& surface) {
     surface->LoadGLBuffer();
     surface->UploadGLTexture(read_framebuffer.handle, draw_framebuffer.handle);
     surface->MarkAsModified(false, *this);
+    surface->MarkForReload(false);
 }
 
 Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool preserve_contents) {
@@ -997,18 +1001,23 @@ Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool pres
     Surface surface{TryGet(params.addr)};
     if (surface) {
         if (surface->GetSurfaceParams().IsCompatibleSurface(params)) {
-            // Use the cached surface as-is
+            // Use the cached surface as-is unless it's not synced with memory
+            if (surface->MustReload())
+                LoadSurface(surface);
             return surface;
         } else if (preserve_contents) {
             // If surface parameters changed and we care about keeping the previous data, recreate
             // the surface from the old one
             Surface new_surface{RecreateSurface(surface, params)};
-            Unregister(surface);
+            UnregisterSurface(surface);
             Register(new_surface);
+            if (new_surface->IsUploaded()) {
+                RegisterReinterpretSurface(new_surface);
+            }
             return new_surface;
         } else {
             // Delete the old surface before creating a new one to prevent collisions.
-            Unregister(surface);
+            UnregisterSurface(surface);
         }
     }
 
@@ -1062,8 +1071,8 @@ void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface,
 }
 
 static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface,
-                        const MathUtil::Rectangle<u32>& src_rect,
-                        const MathUtil::Rectangle<u32>& dst_rect, GLuint read_fb_handle,
+                        const Common::Rectangle<u32>& src_rect,
+                        const Common::Rectangle<u32>& dst_rect, GLuint read_fb_handle,
                         GLuint draw_fb_handle, GLenum src_attachment = 0, GLenum dst_attachment = 0,
                         std::size_t cubemap_face = 0) {
 
@@ -1193,7 +1202,7 @@ static bool BlitSurface(const Surface& src_surface, const Surface& dst_surface,
 void RasterizerCacheOpenGL::FermiCopySurface(
     const Tegra::Engines::Fermi2D::Regs::Surface& src_config,
     const Tegra::Engines::Fermi2D::Regs::Surface& dst_config,
-    const MathUtil::Rectangle<u32>& src_rect, const MathUtil::Rectangle<u32>& dst_rect) {
+    const Common::Rectangle<u32>& src_rect, const Common::Rectangle<u32>& dst_rect) {
 
     const auto& src_params = SurfaceParams::CreateForFermiCopySurface(src_config);
     const auto& dst_params = SurfaceParams::CreateForFermiCopySurface(dst_config);
@@ -1290,4 +1299,107 @@ Surface RasterizerCacheOpenGL::TryGetReservedSurface(const SurfaceParams& params
     return {};
 }
 
+static std::optional<u32> TryFindBestMipMap(std::size_t memory, const SurfaceParams params,
+                                            u32 height) {
+    for (u32 i = 0; i < params.max_mip_level; i++) {
+        if (memory == params.GetMipmapSingleSize(i) && params.MipHeight(i) == height) {
+            return {i};
+        }
+    }
+    return {};
+}
+
+static std::optional<u32> TryFindBestLayer(VAddr addr, const SurfaceParams params, u32 mipmap) {
+    const std::size_t size = params.LayerMemorySize();
+    VAddr start = params.addr + params.GetMipmapLevelOffset(mipmap);
+    for (u32 i = 0; i < params.depth; i++) {
+        if (start == addr) {
+            return {i};
+        }
+        start += size;
+    }
+    return {};
+}
+
+static bool LayerFitReinterpretSurface(RasterizerCacheOpenGL& cache, const Surface render_surface,
+                                       const Surface blitted_surface) {
+    const auto& dst_params = blitted_surface->GetSurfaceParams();
+    const auto& src_params = render_surface->GetSurfaceParams();
+    const std::size_t src_memory_size = src_params.size_in_bytes;
+    const std::optional<u32> level =
+        TryFindBestMipMap(src_memory_size, dst_params, src_params.height);
+    if (level.has_value()) {
+        if (src_params.width == dst_params.MipWidthGobAligned(*level) &&
+            src_params.height == dst_params.MipHeight(*level) &&
+            src_params.block_height >= dst_params.MipBlockHeight(*level)) {
+            const std::optional<u32> slot =
+                TryFindBestLayer(render_surface->GetAddr(), dst_params, *level);
+            if (slot.has_value()) {
+                glCopyImageSubData(render_surface->Texture().handle,
+                                   SurfaceTargetToGL(src_params.target), 0, 0, 0, 0,
+                                   blitted_surface->Texture().handle,
+                                   SurfaceTargetToGL(dst_params.target), *level, 0, 0, *slot,
+                                   dst_params.MipWidth(*level), dst_params.MipHeight(*level), 1);
+                blitted_surface->MarkAsModified(true, cache);
+                return true;
+            }
+        }
+    }
+    return false;
+}
+
+static bool IsReinterpretInvalid(const Surface render_surface, const Surface blitted_surface) {
+    const VAddr bound1 = blitted_surface->GetAddr() + blitted_surface->GetMemorySize();
+    const VAddr bound2 = render_surface->GetAddr() + render_surface->GetMemorySize();
+    if (bound2 > bound1)
+        return true;
+    const auto& dst_params = blitted_surface->GetSurfaceParams();
+    const auto& src_params = render_surface->GetSurfaceParams();
+    return (dst_params.component_type != src_params.component_type);
+}
+
+static bool IsReinterpretInvalidSecond(const Surface render_surface,
+                                       const Surface blitted_surface) {
+    const auto& dst_params = blitted_surface->GetSurfaceParams();
+    const auto& src_params = render_surface->GetSurfaceParams();
+    return (dst_params.height > src_params.height && dst_params.width > src_params.width);
+}
+
+bool RasterizerCacheOpenGL::PartialReinterpretSurface(Surface triggering_surface,
+                                                      Surface intersect) {
+    if (IsReinterpretInvalid(triggering_surface, intersect)) {
+        UnregisterSurface(intersect);
+        return false;
+    }
+    if (!LayerFitReinterpretSurface(*this, triggering_surface, intersect)) {
+        if (IsReinterpretInvalidSecond(triggering_surface, intersect)) {
+            UnregisterSurface(intersect);
+            return false;
+        }
+        FlushObject(intersect);
+        FlushObject(triggering_surface);
+        intersect->MarkForReload(true);
+    }
+    return true;
+}
+
+void RasterizerCacheOpenGL::SignalPreDrawCall() {
+    if (texception && GLAD_GL_ARB_texture_barrier) {
+        glTextureBarrier();
+    }
+    texception = false;
+}
+
+void RasterizerCacheOpenGL::SignalPostDrawCall() {
+    for (u32 i = 0; i < Maxwell::NumRenderTargets; i++) {
+        if (current_color_buffers[i] != nullptr) {
+            Surface intersect = CollideOnReinterpretedSurface(current_color_buffers[i]->GetAddr());
+            if (intersect != nullptr) {
+                PartialReinterpretSurface(current_color_buffers[i], intersect);
+                texception = true;
+            }
+        }
+    }
+}
+
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -28,12 +28,13 @@ namespace OpenGL {
 
 class CachedSurface;
 using Surface = std::shared_ptr<CachedSurface>;
-using SurfaceSurfaceRect_Tuple = std::tuple<Surface, Surface, MathUtil::Rectangle<u32>>;
+using SurfaceSurfaceRect_Tuple = std::tuple<Surface, Surface, Common::Rectangle<u32>>;
 
 using SurfaceTarget = VideoCore::Surface::SurfaceTarget;
 using SurfaceType = VideoCore::Surface::SurfaceType;
 using PixelFormat = VideoCore::Surface::PixelFormat;
 using ComponentType = VideoCore::Surface::ComponentType;
+using Maxwell = Tegra::Engines::Maxwell3D::Regs;
 
 struct SurfaceParams {
     enum class SurfaceClass {
@@ -71,7 +72,7 @@ struct SurfaceParams {
     }
 
     /// Returns the rectangle corresponding to this surface
-    MathUtil::Rectangle<u32> GetRect(u32 mip_level = 0) const;
+    Common::Rectangle<u32> GetRect(u32 mip_level = 0) const;
 
     /// Returns the total size of this surface in bytes, adjusted for compression
     std::size_t SizeInBytesRaw(bool ignore_tiled = false) const {
@@ -140,10 +141,18 @@ struct SurfaceParams {
         return offset;
     }
 
+    std::size_t GetMipmapSingleSize(u32 mip_level) const {
+        return InnerMipmapMemorySize(mip_level, false, is_layered);
+    }
+
     u32 MipWidth(u32 mip_level) const {
         return std::max(1U, width >> mip_level);
     }
 
+    u32 MipWidthGobAligned(u32 mip_level) const {
+        return Common::AlignUp(std::max(1U, width >> mip_level), 64U * 8U / GetFormatBpp());
+    }
+
     u32 MipHeight(u32 mip_level) const {
         return std::max(1U, height >> mip_level);
     }
@@ -346,6 +355,10 @@ public:
         return cached_size_in_bytes;
     }
 
+    std::size_t GetMemorySize() const {
+        return memory_size;
+    }
+
     void Flush() override {
         FlushGLBuffer();
     }
@@ -395,6 +408,26 @@ public:
                        Tegra::Texture::SwizzleSource swizzle_z,
                        Tegra::Texture::SwizzleSource swizzle_w);
 
+    void MarkReinterpreted() {
+        reinterpreted = true;
+    }
+
+    bool IsReinterpreted() const {
+        return reinterpreted;
+    }
+
+    void MarkForReload(bool reload) {
+        must_reload = reload;
+    }
+
+    bool MustReload() const {
+        return must_reload;
+    }
+
+    bool IsUploaded() const {
+        return params.identity == SurfaceParams::SurfaceClass::Uploaded;
+    }
+
 private:
     void UploadGLMipmapTexture(u32 mip_map, GLuint read_fb_handle, GLuint draw_fb_handle);
 
@@ -408,6 +441,9 @@ private:
     GLenum gl_internal_format{};
     std::size_t cached_size_in_bytes{};
     std::array<GLenum, 4> swizzle{GL_RED, GL_GREEN, GL_BLUE, GL_ALPHA};
+    std::size_t memory_size;
+    bool reinterpreted = false;
+    bool must_reload = false;
 };
 
 class RasterizerCacheOpenGL final : public RasterizerCache<Surface> {
@@ -430,8 +466,11 @@ public:
     /// Copies the contents of one surface to another
     void FermiCopySurface(const Tegra::Engines::Fermi2D::Regs::Surface& src_config,
                           const Tegra::Engines::Fermi2D::Regs::Surface& dst_config,
-                          const MathUtil::Rectangle<u32>& src_rect,
-                          const MathUtil::Rectangle<u32>& dst_rect);
+                          const Common::Rectangle<u32>& src_rect,
+                          const Common::Rectangle<u32>& dst_rect);
+
+    void SignalPreDrawCall();
+    void SignalPostDrawCall();
 
 private:
     void LoadSurface(const Surface& surface);
@@ -449,6 +488,10 @@ private:
     /// Tries to get a reserved surface for the specified parameters
     Surface TryGetReservedSurface(const SurfaceParams& params);
 
+    // Partialy reinterpret a surface based on a triggering_surface that collides with it.
+    // returns true if the reinterpret was successful, false in case it was not.
+    bool PartialReinterpretSurface(Surface triggering_surface, Surface intersect);
+
     /// Performs a slow but accurate surface copy, flushing to RAM and reinterpreting the data
     void AccurateCopySurface(const Surface& src_surface, const Surface& dst_surface);
     void FastLayeredCopySurface(const Surface& src_surface, const Surface& dst_surface);
@@ -465,12 +508,50 @@ private:
     OGLFramebuffer read_framebuffer;
     OGLFramebuffer draw_framebuffer;
 
+    bool texception = false;
+
     /// Use a Pixel Buffer Object to download the previous texture and then upload it to the new one
     /// using the new format.
     OGLBuffer copy_pbo;
 
-    std::array<Surface, Tegra::Engines::Maxwell3D::Regs::NumRenderTargets> last_color_buffers;
+    std::array<Surface, Maxwell::NumRenderTargets> last_color_buffers;
+    std::array<Surface, Maxwell::NumRenderTargets> current_color_buffers;
     Surface last_depth_buffer;
+
+    using SurfaceIntervalCache = boost::icl::interval_map<VAddr, Surface>;
+    using SurfaceInterval = typename SurfaceIntervalCache::interval_type;
+
+    static auto GetReinterpretInterval(const Surface& object) {
+        return SurfaceInterval::right_open(object->GetAddr() + 1,
+                                           object->GetAddr() + object->GetMemorySize() - 1);
+    }
+
+    // Reinterpreted surfaces are very fragil as the game may keep rendering into them.
+    SurfaceIntervalCache reinterpreted_surfaces;
+
+    void RegisterReinterpretSurface(Surface reinterpret_surface) {
+        auto interval = GetReinterpretInterval(reinterpret_surface);
+        reinterpreted_surfaces.insert({interval, reinterpret_surface});
+        reinterpret_surface->MarkReinterpreted();
+    }
+
+    Surface CollideOnReinterpretedSurface(VAddr addr) const {
+        const SurfaceInterval interval{addr};
+        for (auto& pair :
+             boost::make_iterator_range(reinterpreted_surfaces.equal_range(interval))) {
+            return pair.second;
+        }
+        return nullptr;
+    }
+
+    /// Unregisters an object from the cache
+    void UnregisterSurface(const Surface& object) {
+        if (object->IsReinterpreted()) {
+            auto interval = GetReinterpretInterval(object);
+            reinterpreted_surfaces.erase(interval);
+        }
+        Unregister(object);
+    }
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_shader_disk_cache.cpp

@@ -2,8 +2,6 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#pragma once
-
 #include <cstring>
 #include <fmt/format.h>
 #include <lz4.h>

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -257,6 +257,7 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture,
                                                  const Tegra::FramebufferConfig& framebuffer) {
     texture.width = framebuffer.width;
     texture.height = framebuffer.height;
+    texture.pixel_format = framebuffer.pixel_format;
 
     GLint internal_format;
     switch (framebuffer.pixel_format) {

src/video_core/renderer_opengl/renderer_opengl.h

@@ -39,7 +39,7 @@ struct TextureInfo {
 /// Structure used for storing information about the display target for the Switch screen
 struct ScreenInfo {
     GLuint display_texture;
-    const MathUtil::Rectangle<float> display_texcoords{0.0f, 0.0f, 1.0f, 1.0f};
+    const Common::Rectangle<float> display_texcoords{0.0f, 0.0f, 1.0f, 1.0f};
     TextureInfo texture;
 };
 
@@ -102,7 +102,7 @@ private:
 
     /// Used for transforming the framebuffer orientation
     Tegra::FramebufferConfig::TransformFlags framebuffer_transform_flags;
-    MathUtil::Rectangle<int> framebuffer_crop_rect;
+    Common::Rectangle<int> framebuffer_crop_rect;
 };
 
 } // namespace OpenGL

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -0,0 +1,116 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstring>
+#include <memory>
+#include <optional>
+#include <tuple>
+
+#include "common/alignment.h"
+#include "core/core.h"
+#include "core/memory.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_buffer_cache.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_stream_buffer.h"
+
+namespace Vulkan {
+
+VKBufferCache::VKBufferCache(Tegra::MemoryManager& tegra_memory_manager,
+                             VideoCore::RasterizerInterface& rasterizer, const VKDevice& device,
+                             VKMemoryManager& memory_manager, VKScheduler& scheduler, u64 size)
+    : RasterizerCache{rasterizer}, tegra_memory_manager{tegra_memory_manager} {
+    const auto usage = vk::BufferUsageFlagBits::eVertexBuffer |
+                       vk::BufferUsageFlagBits::eIndexBuffer |
+                       vk::BufferUsageFlagBits::eUniformBuffer;
+    const auto access = vk::AccessFlagBits::eVertexAttributeRead | vk::AccessFlagBits::eIndexRead |
+                        vk::AccessFlagBits::eUniformRead;
+    stream_buffer =
+        std::make_unique<VKStreamBuffer>(device, memory_manager, scheduler, size, usage, access,
+                                         vk::PipelineStageFlagBits::eAllCommands);
+    buffer_handle = stream_buffer->GetBuffer();
+}
+
+VKBufferCache::~VKBufferCache() = default;
+
+u64 VKBufferCache::UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size, u64 alignment,
+                                bool cache) {
+    const auto cpu_addr{tegra_memory_manager.GpuToCpuAddress(gpu_addr)};
+    ASSERT(cpu_addr);
+
+    // Cache management is a big overhead, so only cache entries with a given size.
+    // TODO: Figure out which size is the best for given games.
+    cache &= size >= 2048;
+
+    if (cache) {
+        if (auto entry = TryGet(*cpu_addr); entry) {
+            if (entry->size >= size && entry->alignment == alignment) {
+                return entry->offset;
+            }
+            Unregister(entry);
+        }
+    }
+
+    AlignBuffer(alignment);
+    const u64 uploaded_offset = buffer_offset;
+
+    Memory::ReadBlock(*cpu_addr, buffer_ptr, size);
+
+    buffer_ptr += size;
+    buffer_offset += size;
+
+    if (cache) {
+        auto entry = std::make_shared<CachedBufferEntry>();
+        entry->offset = uploaded_offset;
+        entry->size = size;
+        entry->alignment = alignment;
+        entry->addr = *cpu_addr;
+        Register(entry);
+    }
+
+    return uploaded_offset;
+}
+
+u64 VKBufferCache::UploadHostMemory(const u8* raw_pointer, std::size_t size, u64 alignment) {
+    AlignBuffer(alignment);
+    std::memcpy(buffer_ptr, raw_pointer, size);
+    const u64 uploaded_offset = buffer_offset;
+
+    buffer_ptr += size;
+    buffer_offset += size;
+    return uploaded_offset;
+}
+
+std::tuple<u8*, u64> VKBufferCache::ReserveMemory(std::size_t size, u64 alignment) {
+    AlignBuffer(alignment);
+    u8* const uploaded_ptr = buffer_ptr;
+    const u64 uploaded_offset = buffer_offset;
+
+    buffer_ptr += size;
+    buffer_offset += size;
+    return {uploaded_ptr, uploaded_offset};
+}
+
+void VKBufferCache::Reserve(std::size_t max_size) {
+    bool invalidate;
+    std::tie(buffer_ptr, buffer_offset_base, invalidate) = stream_buffer->Reserve(max_size);
+    buffer_offset = buffer_offset_base;
+
+    if (invalidate) {
+        InvalidateAll();
+    }
+}
+
+VKExecutionContext VKBufferCache::Send(VKExecutionContext exctx) {
+    return stream_buffer->Send(exctx, buffer_offset - buffer_offset_base);
+}
+
+void VKBufferCache::AlignBuffer(std::size_t alignment) {
+    // Align the offset, not the mapped pointer
+    const u64 offset_aligned = Common::AlignUp(buffer_offset, alignment);
+    buffer_ptr += offset_aligned - buffer_offset;
+    buffer_offset = offset_aligned;
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -0,0 +1,87 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <tuple>
+
+#include "common/common_types.h"
+#include "video_core/gpu.h"
+#include "video_core/rasterizer_cache.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+
+namespace Tegra {
+class MemoryManager;
+}
+
+namespace Vulkan {
+
+class VKDevice;
+class VKFence;
+class VKMemoryManager;
+class VKStreamBuffer;
+
+struct CachedBufferEntry final : public RasterizerCacheObject {
+    VAddr GetAddr() const override {
+        return addr;
+    }
+
+    std::size_t GetSizeInBytes() const override {
+        return size;
+    }
+
+    // We do not have to flush this cache as things in it are never modified by us.
+    void Flush() override {}
+
+    VAddr addr;
+    std::size_t size;
+    u64 offset;
+    std::size_t alignment;
+};
+
+class VKBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> {
+public:
+    explicit VKBufferCache(Tegra::MemoryManager& tegra_memory_manager, VideoCore::RasterizerInterface& rasterizer,
+                           const VKDevice& device, VKMemoryManager& memory_manager,
+                           VKScheduler& scheduler, u64 size);
+    ~VKBufferCache();
+
+    /// Uploads data from a guest GPU address. Returns host's buffer offset where it's been
+    /// allocated.
+    u64 UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size, u64 alignment = 4,
+                     bool cache = true);
+
+    /// Uploads from a host memory. Returns host's buffer offset where it's been allocated.
+    u64 UploadHostMemory(const u8* raw_pointer, std::size_t size, u64 alignment = 4);
+
+    /// Reserves memory to be used by host's CPU. Returns mapped address and offset.
+    std::tuple<u8*, u64> ReserveMemory(std::size_t size, u64 alignment = 4);
+
+    /// Reserves a region of memory to be used in subsequent upload/reserve operations.
+    void Reserve(std::size_t max_size);
+
+    /// Ensures that the set data is sent to the device.
+    [[nodiscard]] VKExecutionContext Send(VKExecutionContext exctx);
+
+    /// Returns the buffer cache handle.
+    vk::Buffer GetBuffer() const {
+        return buffer_handle;
+    }
+
+private:
+    void AlignBuffer(std::size_t alignment);
+
+    Tegra::MemoryManager& tegra_memory_manager;
+
+    std::unique_ptr<VKStreamBuffer> stream_buffer;
+    vk::Buffer buffer_handle;
+
+    u8* buffer_ptr = nullptr;
+    u64 buffer_offset = 0;
+    u64 buffer_offset_base = 0;
+};
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_memory_manager.cpp

@@ -238,7 +238,7 @@ bool VKMemoryManager::AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32
 
 VKMemoryCommitImpl::VKMemoryCommitImpl(VKMemoryAllocation* allocation, vk::DeviceMemory memory,
                                        u8* data, u64 begin, u64 end)
-    : allocation{allocation}, memory{memory}, data{data}, interval(std::make_pair(begin, end)) {}
+    : interval(std::make_pair(begin, end)), memory{memory}, allocation{allocation}, data{data} {}
 
 VKMemoryCommitImpl::~VKMemoryCommitImpl() {
     allocation->Free(this);

src/video_core/renderer_vulkan/vk_resource_manager.cpp

@@ -125,11 +125,12 @@ void VKFence::Protect(VKResource* resource) {
     protected_resources.push_back(resource);
 }
 
-void VKFence::Unprotect(const VKResource* resource) {
+void VKFence::Unprotect(VKResource* resource) {
     const auto it = std::find(protected_resources.begin(), protected_resources.end(), resource);
-    if (it != protected_resources.end()) {
-        protected_resources.erase(it);
-    }
+    ASSERT(it != protected_resources.end());
+
+    resource->OnFenceRemoval(this);
+    protected_resources.erase(it);
 }
 
 VKFenceWatch::VKFenceWatch() = default;
@@ -141,12 +142,11 @@ VKFenceWatch::~VKFenceWatch() {
 }
 
 void VKFenceWatch::Wait() {
-    if (!fence) {
+    if (fence == nullptr) {
         return;
     }
     fence->Wait();
     fence->Unprotect(this);
-    fence = nullptr;
 }
 
 void VKFenceWatch::Watch(VKFence& new_fence) {

src/video_core/renderer_vulkan/vk_resource_manager.h

@@ -63,7 +63,7 @@ public:
     void Protect(VKResource* resource);
 
     /// Removes protection for a resource.
-    void Unprotect(const VKResource* resource);
+    void Unprotect(VKResource* resource);
 
     /// Retreives the fence.
     operator vk::Fence() const {

src/video_core/renderer_vulkan/vk_stream_buffer.cpp

@@ -0,0 +1,90 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <algorithm>
+#include <memory>
+#include <optional>
+#include <vector>
+
+#include "common/assert.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_device.h"
+#include "video_core/renderer_vulkan/vk_memory_manager.h"
+#include "video_core/renderer_vulkan/vk_resource_manager.h"
+#include "video_core/renderer_vulkan/vk_scheduler.h"
+#include "video_core/renderer_vulkan/vk_stream_buffer.h"
+
+namespace Vulkan {
+
+constexpr u64 WATCHES_INITIAL_RESERVE = 0x4000;
+constexpr u64 WATCHES_RESERVE_CHUNK = 0x1000;
+
+VKStreamBuffer::VKStreamBuffer(const VKDevice& device, VKMemoryManager& memory_manager,
+                               VKScheduler& scheduler, u64 size, vk::BufferUsageFlags usage,
+                               vk::AccessFlags access, vk::PipelineStageFlags pipeline_stage)
+    : device{device}, scheduler{scheduler}, buffer_size{size}, access{access}, pipeline_stage{
+                                                                                   pipeline_stage} {
+    CreateBuffers(memory_manager, usage);
+    ReserveWatches(WATCHES_INITIAL_RESERVE);
+}
+
+VKStreamBuffer::~VKStreamBuffer() = default;
+
+std::tuple<u8*, u64, bool> VKStreamBuffer::Reserve(u64 size) {
+    ASSERT(size <= buffer_size);
+    mapped_size = size;
+
+    if (offset + size > buffer_size) {
+        // The buffer would overflow, save the amount of used buffers, signal an invalidation and
+        // reset the state.
+        invalidation_mark = used_watches;
+        used_watches = 0;
+        offset = 0;
+    }
+
+    return {mapped_pointer + offset, offset, invalidation_mark.has_value()};
+}
+
+VKExecutionContext VKStreamBuffer::Send(VKExecutionContext exctx, u64 size) {
+    ASSERT_MSG(size <= mapped_size, "Reserved size is too small");
+
+    if (invalidation_mark) {
+        // TODO(Rodrigo): Find a better way to invalidate than waiting for all watches to finish.
+        exctx = scheduler.Flush();
+        std::for_each(watches.begin(), watches.begin() + *invalidation_mark,
+                      [&](auto& resource) { resource->Wait(); });
+        invalidation_mark = std::nullopt;
+    }
+
+    if (used_watches + 1 >= watches.size()) {
+        // Ensure that there are enough watches.
+        ReserveWatches(WATCHES_RESERVE_CHUNK);
+    }
+    // Add a watch for this allocation.
+    watches[used_watches++]->Watch(exctx.GetFence());
+
+    offset += size;
+
+    return exctx;
+}
+
+void VKStreamBuffer::CreateBuffers(VKMemoryManager& memory_manager, vk::BufferUsageFlags usage) {
+    const vk::BufferCreateInfo buffer_ci({}, buffer_size, usage, vk::SharingMode::eExclusive, 0,
+                                         nullptr);
+
+    const auto dev = device.GetLogical();
+    const auto& dld = device.GetDispatchLoader();
+    buffer = dev.createBufferUnique(buffer_ci, nullptr, dld);
+    commit = memory_manager.Commit(*buffer, true);
+    mapped_pointer = commit->GetData();
+}
+
+void VKStreamBuffer::ReserveWatches(std::size_t grow_size) {
+    const std::size_t previous_size = watches.size();
+    watches.resize(previous_size + grow_size);
+    std::generate(watches.begin() + previous_size, watches.end(),
+                  []() { return std::make_unique<VKFenceWatch>(); });
+}
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_stream_buffer.h

@@ -0,0 +1,72 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <optional>
+#include <tuple>
+#include <vector>
+
+#include "common/common_types.h"
+#include "video_core/renderer_vulkan/declarations.h"
+#include "video_core/renderer_vulkan/vk_memory_manager.h"
+
+namespace Vulkan {
+
+class VKDevice;
+class VKFence;
+class VKFenceWatch;
+class VKResourceManager;
+class VKScheduler;
+
+class VKStreamBuffer {
+public:
+    explicit VKStreamBuffer(const VKDevice& device, VKMemoryManager& memory_manager,
+                            VKScheduler& scheduler, u64 size, vk::BufferUsageFlags usage,
+                            vk::AccessFlags access, vk::PipelineStageFlags pipeline_stage);
+    ~VKStreamBuffer();
+
+    /**
+     * Reserves a region of memory from the stream buffer.
+     * @param size Size to reserve.
+     * @returns A tuple in the following order: Raw memory pointer (with offset added), buffer
+     * offset and a boolean that's true when buffer has been invalidated.
+     */
+    std::tuple<u8*, u64, bool> Reserve(u64 size);
+
+    /// Ensures that "size" bytes of memory are available to the GPU, potentially recording a copy.
+    [[nodiscard]] VKExecutionContext Send(VKExecutionContext exctx, u64 size);
+
+    vk::Buffer GetBuffer() const {
+        return *buffer;
+    }
+
+private:
+    /// Creates Vulkan buffer handles committing the required the required memory.
+    void CreateBuffers(VKMemoryManager& memory_manager, vk::BufferUsageFlags usage);
+
+    /// Increases the amount of watches available.
+    void ReserveWatches(std::size_t grow_size);
+
+    const VKDevice& device;                      ///< Vulkan device manager.
+    VKScheduler& scheduler;                      ///< Command scheduler.
+    const u64 buffer_size;                       ///< Total size of the stream buffer.
+    const vk::AccessFlags access;                ///< Access usage of this stream buffer.
+    const vk::PipelineStageFlags pipeline_stage; ///< Pipeline usage of this stream buffer.
+
+    UniqueBuffer buffer;   ///< Mapped buffer.
+    VKMemoryCommit commit; ///< Memory commit.
+    u8* mapped_pointer{};  ///< Pointer to the host visible commit
+
+    u64 offset{};      ///< Buffer iterator.
+    u64 mapped_size{}; ///< Size reserved for the current copy.
+
+    std::vector<std::unique_ptr<VKFenceWatch>> watches; ///< Total watches
+    std::size_t used_watches{}; ///< Count of watches, reset on invalidation.
+    std::optional<std::size_t>
+        invalidation_mark{}; ///< Number of watches used in the current invalidation.
+};
+
+} // namespace Vulkan

src/yuzu/debugger/graphics/graphics_surface.cpp

@@ -398,7 +398,7 @@ void GraphicsSurfaceWidget::OnUpdate() {
 
     for (unsigned int y = 0; y < surface_height; ++y) {
         for (unsigned int x = 0; x < surface_width; ++x) {
-            Math::Vec4<u8> color;
+            Common::Vec4<u8> color;
             color[0] = texture_data[x + y * surface_width + 0];
             color[1] = texture_data[x + y * surface_width + 1];
             color[2] = texture_data[x + y * surface_width + 2];