travis/macos: Use macpack to bundle dependencies

This appears to properly handle the ffmpeg libraries that dylibbundler
failed to patch.

.travis/macos/deps.sh

@@ -1,5 +1,6 @@
 #!/bin/sh -ex
 
 brew update
-brew install dylibbundler p7zip qt5 sdl2 ccache
+brew install p7zip qt5 sdl2 ccache
 brew outdated cmake || brew upgrade cmake
+pip3 install macpack

.travis/macos/upload.sh

@@ -11,92 +11,19 @@ mkdir "$REV_NAME"
 cp build/bin/yuzu-cmd "$REV_NAME"
 cp -r build/bin/yuzu.app "$REV_NAME"
 
-# move qt libs into app bundle for deployment
-$(brew --prefix)/opt/qt5/bin/macdeployqt "${REV_NAME}/yuzu.app"
+# move libs into folder for deployment
+macpack "${REV_NAME}/yuzu.app/Contents/MacOS/yuzu" -d "../Frameworks"
+# move qt frameworks into app bundle for deployment
+$(brew --prefix)/opt/qt5/bin/macdeployqt "${REV_NAME}/yuzu.app" -executable="${REV_NAME}/yuzu.app/Contents/MacOS/yuzu"
 
-# move SDL2 libs into folder for deployment
-dylibbundler -b -x "${REV_NAME}/yuzu-cmd" -cd -d "${REV_NAME}/libs" -p "@executable_path/libs/"
-
-# Make the changes to make the yuzu app standalone (i.e. not dependent on the current brew installation).
-# To do this, the absolute references to each and every QT framework must be re-written to point to the local frameworks
-# (in the Contents/Frameworks folder).
-# The "install_name_tool" is used to do so.
-
-# Coreutils is a hack to coerce Homebrew to point to the absolute Cellar path (symlink dereferenced). i.e:
-# ls -l /usr/local/opt/qt5:: /usr/local/opt/qt5 -> ../Cellar/qt5/5.6.1-1
-# grealpath ../Cellar/qt5/5.6.1-1:: /usr/local/Cellar/qt5/5.6.1-1
-brew install coreutils || brew upgrade coreutils || true
-
-REV_NAME_ALT=$REV_NAME/
-# grealpath is located in coreutils, there is no "realpath" for OS X :(
-QT_BREWS_PATH=$(grealpath "$(brew --prefix qt5)")
-BREW_PATH=$(brew --prefix)
-QT_VERSION_NUM=5
-
-$BREW_PATH/opt/qt5/bin/macdeployqt "${REV_NAME_ALT}yuzu.app" \
-    -executable="${REV_NAME_ALT}yuzu.app/Contents/MacOS/yuzu"
-
-# These are the files that macdeployqt packed into Contents/Frameworks/ - we don't want those, so we replace them.
-declare -a macos_libs=("QtCore" "QtWidgets" "QtGui" "QtOpenGL" "QtPrintSupport")
-
-for macos_lib in "${macos_libs[@]}"
-do
-    SC_FRAMEWORK_PART=$macos_lib.framework/Versions/$QT_VERSION_NUM/$macos_lib
-    # Replace macdeployqt versions of the Frameworks with our own (from /usr/local/opt/qt5/lib/)
-    cp "$BREW_PATH/opt/qt5/lib/$SC_FRAMEWORK_PART" "${REV_NAME_ALT}yuzu.app/Contents/Frameworks/$SC_FRAMEWORK_PART"
-
-    # Replace references within the embedded Framework files with "internal" versions.
-    for macos_lib2 in "${macos_libs[@]}"
-    do
-        # Since brew references both the non-symlinked and symlink paths of QT5, it needs to be duplicated.
-        # /usr/local/Cellar/qt5/5.6.1-1/lib and /usr/local/opt/qt5/lib both resolve to the same files.
-        # So the two lines below are effectively duplicates when resolved as a path, but as strings, they aren't.
-        RM_FRAMEWORK_PART=$macos_lib2.framework/Versions/$QT_VERSION_NUM/$macos_lib2
-        install_name_tool -change \
-            $QT_BREWS_PATH/lib/$RM_FRAMEWORK_PART \
-            @executable_path/../Frameworks/$RM_FRAMEWORK_PART \
-            "${REV_NAME_ALT}yuzu.app/Contents/Frameworks/$SC_FRAMEWORK_PART"
-        install_name_tool -change \
-            "$BREW_PATH/opt/qt5/lib/$RM_FRAMEWORK_PART" \
-            @executable_path/../Frameworks/$RM_FRAMEWORK_PART \
-            "${REV_NAME_ALT}yuzu.app/Contents/Frameworks/$SC_FRAMEWORK_PART"
-    done
-done
-
-# Handles `This application failed to start because it could not find or load the Qt platform plugin "cocoa"`
-# Which manifests itself as:
-# "Exception Type: EXC_CRASH (SIGABRT) | Exception Codes: 0x0000000000000000, 0x0000000000000000 | Exception Note: EXC_CORPSE_NOTIFY"
-# There may be more dylibs needed to be fixed...
-declare -a macos_plugins=("Plugins/platforms/libqcocoa.dylib")
-
-for macos_lib in "${macos_plugins[@]}"
-do
-    install_name_tool -id @executable_path/../$macos_lib "${REV_NAME_ALT}yuzu.app/Contents/$macos_lib"
-    for macos_lib2 in "${macos_libs[@]}"
-    do
-        RM_FRAMEWORK_PART=$macos_lib2.framework/Versions/$QT_VERSION_NUM/$macos_lib2
-        install_name_tool -change \
-            $QT_BREWS_PATH/lib/$RM_FRAMEWORK_PART \
-            @executable_path/../Frameworks/$RM_FRAMEWORK_PART \
-            "${REV_NAME_ALT}yuzu.app/Contents/$macos_lib"
-        install_name_tool -change \
-            "$BREW_PATH/opt/qt5/lib/$RM_FRAMEWORK_PART" \
-            @executable_path/../Frameworks/$RM_FRAMEWORK_PART \
-            "${REV_NAME_ALT}yuzu.app/Contents/$macos_lib"
-    done
-done
-
-for macos_lib in "${macos_libs[@]}"
-do
-    # Debugging info for Travis-CI
-    otool -L "${REV_NAME_ALT}yuzu.app/Contents/Frameworks/$macos_lib.framework/Versions/$QT_VERSION_NUM/$macos_lib"
-done
+# move libs into folder for deployment
+macpack "${REV_NAME}/yuzu-cmd" -d "libs"
 
 # Make the yuzu.app application launch a debugging terminal.
 # Store away the actual binary
-mv ${REV_NAME_ALT}yuzu.app/Contents/MacOS/yuzu ${REV_NAME_ALT}yuzu.app/Contents/MacOS/yuzu-bin
+mv ${REV_NAME}/yuzu.app/Contents/MacOS/yuzu ${REV_NAME}/yuzu.app/Contents/MacOS/yuzu-bin
 
-cat > ${REV_NAME_ALT}yuzu.app/Contents/MacOS/yuzu <<EOL
+cat > ${REV_NAME}/yuzu.app/Contents/MacOS/yuzu <<EOL
 #!/usr/bin/env bash
 cd "\`dirname "\$0"\`"
 chmod +x yuzu-bin
@@ -105,6 +32,9 @@ EOL
 # Content that will serve as the launching script for yuzu (within the .app folder)
 
 # Make the launching script executable
-chmod +x ${REV_NAME_ALT}yuzu.app/Contents/MacOS/yuzu
+chmod +x ${REV_NAME}/yuzu.app/Contents/MacOS/yuzu
+
+# Verify loader instructions
+find "$REV_NAME" -exec otool -L {} \;
 
 . .travis/common/post-upload.sh

CMakeLists.txt

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

src/common/CMakeLists.txt

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

src/common/bit_field.h

@@ -34,6 +34,7 @@
 #include <limits>
 #include <type_traits>
 #include "common/common_funcs.h"
+#include "common/swap.h"
 
 /*
  * Abstract bitfield class
@@ -108,7 +109,7 @@
  * symptoms.
  */
 #pragma pack(1)
-template <std::size_t Position, std::size_t Bits, typename T>
+template <std::size_t Position, std::size_t Bits, typename T, typename EndianTag = LETag>
 struct BitField {
 private:
     // UnderlyingType is T for non-enum types and the underlying type of T if
@@ -121,6 +122,8 @@ private:
     // We store the value as the unsigned type to avoid undefined behaviour on value shifting
     using StorageType = std::make_unsigned_t<UnderlyingType>;
 
+    using StorageTypeWithEndian = typename AddEndian<StorageType, EndianTag>::type;
+
 public:
     /// Constants to allow limited introspection of fields if needed
     static constexpr std::size_t position = Position;
@@ -170,7 +173,7 @@ public:
     }
 
     constexpr FORCE_INLINE void Assign(const T& value) {
-        storage = (storage & ~mask) | FormatValue(value);
+        storage = (static_cast<StorageType>(storage) & ~mask) | FormatValue(value);
     }
 
     constexpr T Value() const {
@@ -182,7 +185,7 @@ public:
     }
 
 private:
-    StorageType storage;
+    StorageTypeWithEndian storage;
 
     static_assert(bits + position <= 8 * sizeof(T), "Bitfield out of range");
 
@@ -193,3 +196,6 @@ private:
     static_assert(std::is_trivially_copyable_v<T>, "T must be trivially copyable in a BitField");
 };
 #pragma pack()
+
+template <std::size_t Position, std::size_t Bits, typename T>
+using BitFieldBE = BitField<Position, Bits, T, BETag>;

src/common/common_types.h

@@ -40,10 +40,9 @@ using s64 = std::int64_t; ///< 64-bit signed int
 using f32 = float;  ///< 32-bit floating point
 using f64 = double; ///< 64-bit floating point
 
-// TODO: It would be nice to eventually replace these with strong types that prevent accidental
-// conversion between each other.
-using VAddr = u64; ///< Represents a pointer in the userspace virtual address space.
-using PAddr = u64; ///< Represents a pointer in the ARM11 physical address space.
+using VAddr = u64;    ///< Represents a pointer in the userspace virtual address space.
+using PAddr = u64;    ///< Represents a pointer in the ARM11 physical address space.
+using GPUVAddr = u64; ///< Represents a pointer in the GPU virtual address space.
 
 using u128 = std::array<std::uint64_t, 2>;
 static_assert(sizeof(u128) == 16, "u128 must be 128 bits wide");

src/common/memory_hook.cpp

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

src/common/memory_hook.h

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

src/common/page_table.cpp

@@ -0,0 +1,31 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/page_table.h"
+
+namespace Common {
+
+PageTable::PageTable(std::size_t page_size_in_bits) : page_size_in_bits{page_size_in_bits} {}
+
+PageTable::~PageTable() = default;
+
+void PageTable::Resize(std::size_t address_space_width_in_bits) {
+    const std::size_t num_page_table_entries = 1ULL
+                                               << (address_space_width_in_bits - page_size_in_bits);
+
+    pointers.resize(num_page_table_entries);
+    attributes.resize(num_page_table_entries);
+    backing_addr.resize(num_page_table_entries);
+
+    // The default is a 39-bit address space, which causes an initial 1GB allocation size. If the
+    // vector size is subsequently decreased (via resize), the vector might not automatically
+    // actually reallocate/resize its underlying allocation, which wastes up to ~800 MB for
+    // 36-bit titles. Call shrink_to_fit to reduce capacity to what's actually in use.
+
+    pointers.shrink_to_fit();
+    attributes.shrink_to_fit();
+    backing_addr.shrink_to_fit();
+}
+
+} // namespace Common

src/common/page_table.h

@@ -0,0 +1,84 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <vector>
+#include <boost/icl/interval_map.hpp>
+#include "common/common_types.h"
+#include "common/memory_hook.h"
+
+namespace Common {
+
+enum class PageType : u8 {
+    /// Page is unmapped and should cause an access error.
+    Unmapped,
+    /// Page is mapped to regular memory. This is the only type you can get pointers to.
+    Memory,
+    /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and
+    /// invalidation
+    RasterizerCachedMemory,
+    /// Page is mapped to a I/O region. Writing and reading to this page is handled by functions.
+    Special,
+    /// Page is allocated for use.
+    Allocated,
+};
+
+struct SpecialRegion {
+    enum class Type {
+        DebugHook,
+        IODevice,
+    } type;
+
+    MemoryHookPointer handler;
+
+    bool operator<(const SpecialRegion& other) const {
+        return std::tie(type, handler) < std::tie(other.type, other.handler);
+    }
+
+    bool operator==(const SpecialRegion& other) const {
+        return std::tie(type, handler) == std::tie(other.type, other.handler);
+    }
+};
+
+/**
+ * A (reasonably) fast way of allowing switchable and remappable process address spaces. It loosely
+ * mimics the way a real CPU page table works.
+ */
+struct PageTable {
+    explicit PageTable(std::size_t page_size_in_bits);
+    ~PageTable();
+
+    /**
+     * Resizes the page table to be able to accomodate enough pages within
+     * a given address space.
+     *
+     * @param address_space_width_in_bits The address size width in bits.
+     */
+    void Resize(std::size_t address_space_width_in_bits);
+
+    /**
+     * Vector of memory pointers backing each page. An entry can only be non-null if the
+     * corresponding entry in the `attributes` vector is of type `Memory`.
+     */
+    std::vector<u8*> pointers;
+
+    /**
+     * Contains MMIO handlers that back memory regions whose entries in the `attribute` vector is
+     * of type `Special`.
+     */
+    boost::icl::interval_map<u64, std::set<SpecialRegion>> special_regions;
+
+    /**
+     * Vector of fine grained page attributes. If it is set to any value other than `Memory`, then
+     * the corresponding entry in `pointers` MUST be set to null.
+     */
+    std::vector<PageType> attributes;
+
+    std::vector<u64> backing_addr;
+
+    const std::size_t page_size_in_bits{};
+};
+
+} // namespace Common

src/common/swap.h

@@ -17,6 +17,8 @@
 
 #pragma once
 
+#include <type_traits>
+
 #if defined(_MSC_VER)
 #include <cstdlib>
 #elif defined(__linux__)
@@ -170,7 +172,7 @@ struct swap_struct_t {
     using swapped_t = swap_struct_t;
 
 protected:
-    T value = T();
+    T value;
 
     static T swap(T v) {
         return F::swap(v);
@@ -605,52 +607,154 @@ struct swap_double_t {
     }
 };
 
+template <typename T>
+struct swap_enum_t {
+    static_assert(std::is_enum_v<T>);
+    using base = std::underlying_type_t<T>;
+
+public:
+    swap_enum_t() = default;
+    swap_enum_t(const T& v) : value(swap(v)) {}
+
+    swap_enum_t& operator=(const T& v) {
+        value = swap(v);
+        return *this;
+    }
+
+    operator T() const {
+        return swap(value);
+    }
+
+    explicit operator base() const {
+        return static_cast<base>(swap(value));
+    }
+
+protected:
+    T value{};
+    // clang-format off
+    using swap_t = std::conditional_t<
+        std::is_same_v<base, u16>, swap_16_t<u16>, std::conditional_t<
+        std::is_same_v<base, s16>, swap_16_t<s16>, std::conditional_t<
+        std::is_same_v<base, u32>, swap_32_t<u32>, std::conditional_t<
+        std::is_same_v<base, s32>, swap_32_t<s32>, std::conditional_t<
+        std::is_same_v<base, u64>, swap_64_t<u64>, std::conditional_t<
+        std::is_same_v<base, s64>, swap_64_t<s64>, void>>>>>>;
+    // clang-format on
+    static T swap(T x) {
+        return static_cast<T>(swap_t::swap(static_cast<base>(x)));
+    }
+};
+
+struct SwapTag {}; // Use the different endianness from the system
+struct KeepTag {}; // Use the same endianness as the system
+
+template <typename T, typename Tag>
+struct AddEndian;
+
+// KeepTag specializations
+
+template <typename T>
+struct AddEndian<T, KeepTag> {
+    using type = T;
+};
+
+// SwapTag specializations
+
+template <>
+struct AddEndian<u8, SwapTag> {
+    using type = u8;
+};
+
+template <>
+struct AddEndian<u16, SwapTag> {
+    using type = swap_struct_t<u16, swap_16_t<u16>>;
+};
+
+template <>
+struct AddEndian<u32, SwapTag> {
+    using type = swap_struct_t<u32, swap_32_t<u32>>;
+};
+
+template <>
+struct AddEndian<u64, SwapTag> {
+    using type = swap_struct_t<u64, swap_64_t<u64>>;
+};
+
+template <>
+struct AddEndian<s8, SwapTag> {
+    using type = s8;
+};
+
+template <>
+struct AddEndian<s16, SwapTag> {
+    using type = swap_struct_t<s16, swap_16_t<s16>>;
+};
+
+template <>
+struct AddEndian<s32, SwapTag> {
+    using type = swap_struct_t<s32, swap_32_t<s32>>;
+};
+
+template <>
+struct AddEndian<s64, SwapTag> {
+    using type = swap_struct_t<s64, swap_64_t<s64>>;
+};
+
+template <>
+struct AddEndian<float, SwapTag> {
+    using type = swap_struct_t<float, swap_float_t<float>>;
+};
+
+template <>
+struct AddEndian<double, SwapTag> {
+    using type = swap_struct_t<double, swap_double_t<double>>;
+};
+
+template <typename T>
+struct AddEndian<T, SwapTag> {
+    static_assert(std::is_enum_v<T>);
+    using type = swap_enum_t<T>;
+};
+
+// Alias LETag/BETag as KeepTag/SwapTag depending on the system
 #if COMMON_LITTLE_ENDIAN
-using u16_le = u16;
-using u32_le = u32;
-using u64_le = u64;
 
-using s16_le = s16;
-using s32_le = s32;
-using s64_le = s64;
+using LETag = KeepTag;
+using BETag = SwapTag;
 
-using float_le = float;
-using double_le = double;
-
-using u64_be = swap_struct_t<u64, swap_64_t<u64>>;
-using s64_be = swap_struct_t<s64, swap_64_t<s64>>;
-
-using u32_be = swap_struct_t<u32, swap_32_t<u32>>;
-using s32_be = swap_struct_t<s32, swap_32_t<s32>>;
-
-using u16_be = swap_struct_t<u16, swap_16_t<u16>>;
-using s16_be = swap_struct_t<s16, swap_16_t<s16>>;
-
-using float_be = swap_struct_t<float, swap_float_t<float>>;
-using double_be = swap_struct_t<double, swap_double_t<double>>;
 #else
 
-using u64_le = swap_struct_t<u64, swap_64_t<u64>>;
-using s64_le = swap_struct_t<s64, swap_64_t<s64>>;
-
-using u32_le = swap_struct_t<u32, swap_32_t<u32>>;
-using s32_le = swap_struct_t<s32, swap_32_t<s32>>;
-
-using u16_le = swap_struct_t<u16, swap_16_t<u16>>;
-using s16_le = swap_struct_t<s16, swap_16_t<s16>>;
-
-using float_le = swap_struct_t<float, swap_float_t<float>>;
-using double_le = swap_struct_t<double, swap_double_t<double>>;
-
-using u16_be = u16;
-using u32_be = u32;
-using u64_be = u64;
-
-using s16_be = s16;
-using s32_be = s32;
-using s64_be = s64;
-
-using float_be = float;
-using double_be = double;
+using BETag = KeepTag;
+using LETag = SwapTag;
 
 #endif
+
+// Aliases for LE types
+using u16_le = AddEndian<u16, LETag>::type;
+using u32_le = AddEndian<u32, LETag>::type;
+using u64_le = AddEndian<u64, LETag>::type;
+
+using s16_le = AddEndian<s16, LETag>::type;
+using s32_le = AddEndian<s32, LETag>::type;
+using s64_le = AddEndian<s64, LETag>::type;
+
+template <typename T>
+using enum_le = std::enable_if_t<std::is_enum_v<T>, typename AddEndian<T, LETag>::type>;
+
+using float_le = AddEndian<float, LETag>::type;
+using double_le = AddEndian<double, LETag>::type;
+
+// Aliases for BE types
+using u16_be = AddEndian<u16, BETag>::type;
+using u32_be = AddEndian<u32, BETag>::type;
+using u64_be = AddEndian<u64, BETag>::type;
+
+using s16_be = AddEndian<s16, BETag>::type;
+using s32_be = AddEndian<s32, BETag>::type;
+using s64_be = AddEndian<s64, BETag>::type;
+
+template <typename T>
+using enum_be = std::enable_if_t<std::is_enum_v<T>, typename AddEndian<T, BETag>::type>;
+
+using float_be = AddEndian<float, BETag>::type;
+using double_be = AddEndian<double, BETag>::type;

src/common/thread_queue_list.h

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

src/common/uint128.cpp

@@ -0,0 +1,45 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#ifdef _MSC_VER
+#include <intrin.h>
+
+#pragma intrinsic(_umul128)
+#endif
+#include <cstring>
+#include "common/uint128.h"
+
+namespace Common {
+
+u128 Multiply64Into128(u64 a, u64 b) {
+    u128 result;
+#ifdef _MSC_VER
+    result[0] = _umul128(a, b, &result[1]);
+#else
+    unsigned __int128 tmp = a;
+    tmp *= b;
+    std::memcpy(&result, &tmp, sizeof(u128));
+#endif
+    return result;
+}
+
+std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor) {
+    u64 remainder = dividend[0] % divisor;
+    u64 accum = dividend[0] / divisor;
+    if (dividend[1] == 0)
+        return {accum, remainder};
+    // We ignore dividend[1] / divisor as that overflows
+    const u64 first_segment = (dividend[1] % divisor) << 32;
+    accum += (first_segment / divisor) << 32;
+    const u64 second_segment = (first_segment % divisor) << 32;
+    accum += (second_segment / divisor);
+    remainder += second_segment % divisor;
+    if (remainder >= divisor) {
+        accum++;
+        remainder -= divisor;
+    }
+    return {accum, remainder};
+}
+
+} // namespace Common

src/common/uint128.h

@@ -0,0 +1,19 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <utility>
+#include "common/common_types.h"
+
+namespace Common {
+
+// This function multiplies 2 u64 values and produces a u128 value;
+u128 Multiply64Into128(u64 a, u64 b);
+
+// This function divides a u128 by a u32 value and produces two u64 values:
+// the result of division and the remainder
+std::pair<u64, u64> Divide128On32(u128 dividend, u32 divisor);
+
+} // namespace Common

src/core/CMakeLists.txt

@@ -31,6 +31,8 @@ add_library(core STATIC
     file_sys/bis_factory.h
     file_sys/card_image.cpp
     file_sys/card_image.h
+    file_sys/cheat_engine.cpp
+    file_sys/cheat_engine.h
     file_sys/content_archive.cpp
     file_sys/content_archive.h
     file_sys/control_metadata.cpp
@@ -107,6 +109,8 @@ add_library(core STATIC
     hle/kernel/client_port.h
     hle/kernel/client_session.cpp
     hle/kernel/client_session.h
+    hle/kernel/code_set.cpp
+    hle/kernel/code_set.h
     hle/kernel/errors.h
     hle/kernel/handle_table.cpp
     hle/kernel/handle_table.h
@@ -419,8 +423,6 @@ add_library(core STATIC
     loader/deconstructed_rom_directory.h
     loader/elf.cpp
     loader/elf.h
-    loader/linker.cpp
-    loader/linker.h
     loader/loader.cpp
     loader/loader.h
     loader/nax.cpp
@@ -437,8 +439,6 @@ add_library(core STATIC
     loader/xci.h
     memory.cpp
     memory.h
-    memory_hook.cpp
-    memory_hook.h
     memory_setup.h
     perf_stats.cpp
     perf_stats.h

src/core/arm/dynarmic/arm_dynarmic.cpp

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

src/core/arm/dynarmic/arm_dynarmic.h

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

src/core/core.cpp

@@ -32,6 +32,7 @@
 #include "core/perf_stats.h"
 #include "core/settings.h"
 #include "core/telemetry_session.h"
+#include "file_sys/cheat_engine.h"
 #include "frontend/applets/profile_select.h"
 #include "frontend/applets/software_keyboard.h"
 #include "frontend/applets/web_browser.h"
@@ -205,6 +206,7 @@ struct System::Impl {
         GDBStub::Shutdown();
         Service::Shutdown();
         service_manager.reset();
+        cheat_engine.reset();
         telemetry_session.reset();
         gpu_core.reset();
 
@@ -255,6 +257,8 @@ struct System::Impl {
     CpuCoreManager cpu_core_manager;
     bool is_powered_on = false;
 
+    std::unique_ptr<FileSys::CheatEngine> cheat_engine;
+
     /// Frontend applets
     std::unique_ptr<Core::Frontend::ProfileSelectApplet> profile_selector;
     std::unique_ptr<Core::Frontend::SoftwareKeyboardApplet> software_keyboard;
@@ -453,6 +457,13 @@ Tegra::DebugContext* System::GetGPUDebugContext() const {
     return impl->debug_context.get();
 }
 
+void System::RegisterCheatList(const std::vector<FileSys::CheatList>& list,
+                               const std::string& build_id, VAddr code_region_start,
+                               VAddr code_region_end) {
+    impl->cheat_engine = std::make_unique<FileSys::CheatEngine>(*this, list, build_id,
+                                                                code_region_start, code_region_end);
+}
+
 void System::SetFilesystem(std::shared_ptr<FileSys::VfsFilesystem> vfs) {
     impl->virtual_filesystem = std::move(vfs);
 }

src/core/core.h

@@ -20,6 +20,7 @@ class WebBrowserApplet;
 } // namespace Core::Frontend
 
 namespace FileSys {
+class CheatList;
 class VfsFilesystem;
 } // namespace FileSys
 
@@ -253,6 +254,9 @@ public:
 
     std::shared_ptr<FileSys::VfsFilesystem> GetFilesystem() const;
 
+    void RegisterCheatList(const std::vector<FileSys::CheatList>& list, const std::string& build_id,
+                           VAddr code_region_start, VAddr code_region_end);
+
     void SetProfileSelector(std::unique_ptr<Frontend::ProfileSelectApplet> applet);
 
     const Frontend::ProfileSelectApplet& GetProfileSelector() const;

src/core/core_timing_util.cpp

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

src/core/core_timing_util.h

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

src/core/file_sys/cheat_engine.cpp

@@ -0,0 +1,490 @@
+// Copyright 2018 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <locale>
+#include "common/hex_util.h"
+#include "common/microprofile.h"
+#include "common/swap.h"
+#include "core/core.h"
+#include "core/core_timing.h"
+#include "core/core_timing_util.h"
+#include "core/file_sys/cheat_engine.h"
+#include "core/hle/kernel/process.h"
+#include "core/hle/service/hid/controllers/npad.h"
+#include "core/hle/service/hid/hid.h"
+#include "core/hle/service/sm/sm.h"
+
+namespace FileSys {
+
+constexpr s64 CHEAT_ENGINE_TICKS = static_cast<s64>(Core::Timing::BASE_CLOCK_RATE / 60);
+constexpr u32 KEYPAD_BITMASK = 0x3FFFFFF;
+
+u64 Cheat::Address() const {
+    u64 out;
+    std::memcpy(&out, raw.data(), sizeof(u64));
+    return Common::swap64(out) & 0xFFFFFFFFFF;
+}
+
+u64 Cheat::ValueWidth(u64 offset) const {
+    return Value(offset, width);
+}
+
+u64 Cheat::Value(u64 offset, u64 width) const {
+    u64 out;
+    std::memcpy(&out, raw.data() + offset, sizeof(u64));
+    out = Common::swap64(out);
+    if (width == 8)
+        return out;
+    return out & ((1ull << (width * CHAR_BIT)) - 1);
+}
+
+u32 Cheat::KeypadValue() const {
+    u32 out;
+    std::memcpy(&out, raw.data(), sizeof(u32));
+    return Common::swap32(out) & 0x0FFFFFFF;
+}
+
+void CheatList::SetMemoryParameters(VAddr main_begin, VAddr heap_begin, VAddr main_end,
+                                    VAddr heap_end, MemoryWriter writer, MemoryReader reader) {
+    this->main_region_begin = main_begin;
+    this->main_region_end = main_end;
+    this->heap_region_begin = heap_begin;
+    this->heap_region_end = heap_end;
+    this->writer = writer;
+    this->reader = reader;
+}
+
+MICROPROFILE_DEFINE(Cheat_Engine, "Add-Ons", "Cheat Engine", MP_RGB(70, 200, 70));
+
+void CheatList::Execute() {
+    MICROPROFILE_SCOPE(Cheat_Engine);
+
+    std::fill(scratch.begin(), scratch.end(), 0);
+    in_standard = false;
+    for (std::size_t i = 0; i < master_list.size(); ++i) {
+        LOG_DEBUG(Common_Filesystem, "Executing block #{:08X} ({})", i, master_list[i].first);
+        current_block = i;
+        ExecuteBlock(master_list[i].second);
+    }
+
+    in_standard = true;
+    for (std::size_t i = 0; i < standard_list.size(); ++i) {
+        LOG_DEBUG(Common_Filesystem, "Executing block #{:08X} ({})", i, standard_list[i].first);
+        current_block = i;
+        ExecuteBlock(standard_list[i].second);
+    }
+}
+
+CheatList::CheatList(const Core::System& system_, ProgramSegment master, ProgramSegment standard)
+    : master_list{std::move(master)}, standard_list{std::move(standard)}, system{&system_} {}
+
+bool CheatList::EvaluateConditional(const Cheat& cheat) const {
+    using ComparisonFunction = bool (*)(u64, u64);
+    constexpr std::array<ComparisonFunction, 6> comparison_functions{
+        [](u64 a, u64 b) { return a > b; },  [](u64 a, u64 b) { return a >= b; },
+        [](u64 a, u64 b) { return a < b; },  [](u64 a, u64 b) { return a <= b; },
+        [](u64 a, u64 b) { return a == b; }, [](u64 a, u64 b) { return a != b; },
+    };
+
+    if (cheat.type == CodeType::ConditionalInput) {
+        const auto applet_resource =
+            system->ServiceManager().GetService<Service::HID::Hid>("hid")->GetAppletResource();
+        if (applet_resource == nullptr) {
+            LOG_WARNING(
+                Common_Filesystem,
+                "Attempted to evaluate input conditional, but applet resource is not initialized!");
+            return false;
+        }
+
+        const auto press_state =
+            applet_resource
+                ->GetController<Service::HID::Controller_NPad>(Service::HID::HidController::NPad)
+                .GetAndResetPressState();
+        return ((press_state & cheat.KeypadValue()) & KEYPAD_BITMASK) != 0;
+    }
+
+    ASSERT(cheat.type == CodeType::Conditional);
+
+    const auto offset =
+        cheat.memory_type == MemoryType::MainNSO ? main_region_begin : heap_region_begin;
+    ASSERT(static_cast<u8>(cheat.comparison_op.Value()) < 6);
+    auto* function = comparison_functions[static_cast<u8>(cheat.comparison_op.Value())];
+    const auto addr = cheat.Address() + offset;
+
+    return function(reader(cheat.width, SanitizeAddress(addr)), cheat.ValueWidth(8));
+}
+
+void CheatList::ProcessBlockPairs(const Block& block) {
+    block_pairs.clear();
+
+    u64 scope = 0;
+    std::map<u64, u64> pairs;
+
+    for (std::size_t i = 0; i < block.size(); ++i) {
+        const auto& cheat = block[i];
+
+        switch (cheat.type) {
+        case CodeType::Conditional:
+        case CodeType::ConditionalInput:
+            pairs.insert_or_assign(scope, i);
+            ++scope;
+            break;
+        case CodeType::EndConditional: {
+            --scope;
+            const auto idx = pairs.at(scope);
+            block_pairs.insert_or_assign(idx, i);
+            break;
+        }
+        case CodeType::Loop: {
+            if (cheat.end_of_loop) {
+                --scope;
+                const auto idx = pairs.at(scope);
+                block_pairs.insert_or_assign(idx, i);
+            } else {
+                pairs.insert_or_assign(scope, i);
+                ++scope;
+            }
+            break;
+        }
+        }
+    }
+}
+
+void CheatList::WriteImmediate(const Cheat& cheat) {
+    const auto offset =
+        cheat.memory_type == MemoryType::MainNSO ? main_region_begin : heap_region_begin;
+    const auto& register_3 = scratch.at(cheat.register_3);
+
+    const auto addr = cheat.Address() + offset + register_3;
+    LOG_DEBUG(Common_Filesystem, "writing value={:016X} to addr={:016X}", addr,
+              cheat.Value(8, cheat.width));
+    writer(cheat.width, SanitizeAddress(addr), cheat.ValueWidth(8));
+}
+
+void CheatList::BeginConditional(const Cheat& cheat) {
+    if (EvaluateConditional(cheat)) {
+        return;
+    }
+
+    const auto iter = block_pairs.find(current_index);
+    ASSERT(iter != block_pairs.end());
+    current_index = iter->second - 1;
+}
+
+void CheatList::EndConditional(const Cheat& cheat) {
+    LOG_DEBUG(Common_Filesystem, "Ending conditional block.");
+}
+
+void CheatList::Loop(const Cheat& cheat) {
+    if (cheat.end_of_loop.Value())
+        ASSERT(!cheat.end_of_loop.Value());
+
+    auto& register_3 = scratch.at(cheat.register_3);
+    const auto iter = block_pairs.find(current_index);
+    ASSERT(iter != block_pairs.end());
+    ASSERT(iter->first < iter->second);
+
+    const s32 initial_value = static_cast<s32>(cheat.Value(4, sizeof(s32)));
+    for (s32 i = initial_value; i >= 0; --i) {
+        register_3 = static_cast<u64>(i);
+        for (std::size_t c = iter->first + 1; c < iter->second; ++c) {
+            current_index = c;
+            ExecuteSingleCheat(
+                (in_standard ? standard_list : master_list)[current_block].second[c]);
+        }
+    }
+
+    current_index = iter->second;
+}
+
+void CheatList::LoadImmediate(const Cheat& cheat) {
+    auto& register_3 = scratch.at(cheat.register_3);
+
+    LOG_DEBUG(Common_Filesystem, "setting register={:01X} equal to value={:016X}", cheat.register_3,
+              cheat.Value(4, 8));
+    register_3 = cheat.Value(4, 8);
+}
+
+void CheatList::LoadIndexed(const Cheat& cheat) {
+    const auto offset =
+        cheat.memory_type == MemoryType::MainNSO ? main_region_begin : heap_region_begin;
+    auto& register_3 = scratch.at(cheat.register_3);
+
+    const auto addr = (cheat.load_from_register.Value() ? register_3 : offset) + cheat.Address();
+    LOG_DEBUG(Common_Filesystem, "writing indexed value to register={:01X}, addr={:016X}",
+              cheat.register_3, addr);
+    register_3 = reader(cheat.width, SanitizeAddress(addr));
+}
+
+void CheatList::StoreIndexed(const Cheat& cheat) {
+    const auto& register_3 = scratch.at(cheat.register_3);
+
+    const auto addr =
+        register_3 + (cheat.add_additional_register.Value() ? scratch.at(cheat.register_6) : 0);
+    LOG_DEBUG(Common_Filesystem, "writing value={:016X} to addr={:016X}",
+              cheat.Value(4, cheat.width), addr);
+    writer(cheat.width, SanitizeAddress(addr), cheat.ValueWidth(4));
+}
+
+void CheatList::RegisterArithmetic(const Cheat& cheat) {
+    using ArithmeticFunction = u64 (*)(u64, u64);
+    constexpr std::array<ArithmeticFunction, 5> arithmetic_functions{
+        [](u64 a, u64 b) { return a + b; },  [](u64 a, u64 b) { return a - b; },
+        [](u64 a, u64 b) { return a * b; },  [](u64 a, u64 b) { return a << b; },
+        [](u64 a, u64 b) { return a >> b; },
+    };
+
+    using ArithmeticOverflowCheck = bool (*)(u64, u64);
+    constexpr std::array<ArithmeticOverflowCheck, 5> arithmetic_overflow_checks{
+        [](u64 a, u64 b) { return a > (std::numeric_limits<u64>::max() - b); },       // a + b
+        [](u64 a, u64 b) { return a > (std::numeric_limits<u64>::max() + b); },       // a - b
+        [](u64 a, u64 b) { return a > (std::numeric_limits<u64>::max() / b); },       // a * b
+        [](u64 a, u64 b) { return b >= 64 || (a & ~((1ull << (64 - b)) - 1)) != 0; }, // a << b
+        [](u64 a, u64 b) { return b >= 64 || (a & ((1ull << b) - 1)) != 0; },         // a >> b
+    };
+
+    static_assert(sizeof(arithmetic_functions) == sizeof(arithmetic_overflow_checks),
+                  "Missing or have extra arithmetic overflow checks compared to functions!");
+
+    auto& register_3 = scratch.at(cheat.register_3);
+
+    ASSERT(static_cast<u8>(cheat.arithmetic_op.Value()) < 5);
+    auto* function = arithmetic_functions[static_cast<u8>(cheat.arithmetic_op.Value())];
+    auto* overflow_function =
+        arithmetic_overflow_checks[static_cast<u8>(cheat.arithmetic_op.Value())];
+    LOG_DEBUG(Common_Filesystem, "performing arithmetic with register={:01X}, value={:016X}",
+              cheat.register_3, cheat.ValueWidth(4));
+
+    if (overflow_function(register_3, cheat.ValueWidth(4))) {
+        LOG_WARNING(Common_Filesystem,
+                    "overflow will occur when performing arithmetic operation={:02X} with operands "
+                    "a={:016X}, b={:016X}!",
+                    static_cast<u8>(cheat.arithmetic_op.Value()), register_3, cheat.ValueWidth(4));
+    }
+
+    register_3 = function(register_3, cheat.ValueWidth(4));
+}
+
+void CheatList::BeginConditionalInput(const Cheat& cheat) {
+    if (EvaluateConditional(cheat))
+        return;
+
+    const auto iter = block_pairs.find(current_index);
+    ASSERT(iter != block_pairs.end());
+    current_index = iter->second - 1;
+}
+
+VAddr CheatList::SanitizeAddress(VAddr in) const {
+    if ((in < main_region_begin || in >= main_region_end) &&
+        (in < heap_region_begin || in >= heap_region_end)) {
+        LOG_ERROR(Common_Filesystem,
+                  "Cheat attempting to access memory at invalid address={:016X}, if this persists, "
+                  "the cheat may be incorrect. However, this may be normal early in execution if "
+                  "the game has not properly set up yet.",
+                  in);
+        return 0; ///< Invalid addresses will hard crash
+    }
+
+    return in;
+}
+
+void CheatList::ExecuteSingleCheat(const Cheat& cheat) {
+    using CheatOperationFunction = void (CheatList::*)(const Cheat&);
+    constexpr std::array<CheatOperationFunction, 9> cheat_operation_functions{
+        &CheatList::WriteImmediate,        &CheatList::BeginConditional,
+        &CheatList::EndConditional,        &CheatList::Loop,
+        &CheatList::LoadImmediate,         &CheatList::LoadIndexed,
+        &CheatList::StoreIndexed,          &CheatList::RegisterArithmetic,
+        &CheatList::BeginConditionalInput,
+    };
+
+    const auto index = static_cast<u8>(cheat.type.Value());
+    ASSERT(index < sizeof(cheat_operation_functions));
+    const auto op = cheat_operation_functions[index];
+    (this->*op)(cheat);
+}
+
+void CheatList::ExecuteBlock(const Block& block) {
+    encountered_loops.clear();
+
+    ProcessBlockPairs(block);
+    for (std::size_t i = 0; i < block.size(); ++i) {
+        current_index = i;
+        ExecuteSingleCheat(block[i]);
+        i = current_index;
+    }
+}
+
+CheatParser::~CheatParser() = default;
+
+CheatList CheatParser::MakeCheatList(const Core::System& system, CheatList::ProgramSegment master,
+                                     CheatList::ProgramSegment standard) const {
+    return {system, std::move(master), std::move(standard)};
+}
+
+TextCheatParser::~TextCheatParser() = default;
+
+CheatList TextCheatParser::Parse(const Core::System& system, const std::vector<u8>& data) const {
+    std::stringstream ss;
+    ss.write(reinterpret_cast<const char*>(data.data()), data.size());
+
+    std::vector<std::string> lines;
+    std::string stream_line;
+    while (std::getline(ss, stream_line)) {
+        // Remove a trailing \r
+        if (!stream_line.empty() && stream_line.back() == '\r')
+            stream_line.pop_back();
+        lines.push_back(std::move(stream_line));
+    }
+
+    CheatList::ProgramSegment master_list;
+    CheatList::ProgramSegment standard_list;
+
+    for (std::size_t i = 0; i < lines.size(); ++i) {
+        auto line = lines[i];
+
+        if (!line.empty() && (line[0] == '[' || line[0] == '{')) {
+            const auto master = line[0] == '{';
+            const auto begin = master ? line.find('{') : line.find('[');
+            const auto end = master ? line.rfind('}') : line.rfind(']');
+
+            ASSERT(begin != std::string::npos && end != std::string::npos);
+
+            const std::string patch_name{line.begin() + begin + 1, line.begin() + end};
+            CheatList::Block block{};
+
+            while (i < lines.size() - 1) {
+                line = lines[++i];
+                if (!line.empty() && (line[0] == '[' || line[0] == '{')) {
+                    --i;
+                    break;
+                }
+
+                if (line.size() < 8)
+                    continue;
+
+                Cheat out{};
+                out.raw = ParseSingleLineCheat(line);
+                block.push_back(out);
+            }
+
+            (master ? master_list : standard_list).emplace_back(patch_name, block);
+        }
+    }
+
+    return MakeCheatList(system, master_list, standard_list);
+}
+
+std::array<u8, 16> TextCheatParser::ParseSingleLineCheat(const std::string& line) const {
+    std::array<u8, 16> out{};
+
+    if (line.size() < 8)
+        return out;
+
+    const auto word1 = Common::HexStringToArray<sizeof(u32)>(std::string_view{line.data(), 8});
+    std::memcpy(out.data(), word1.data(), sizeof(u32));
+
+    if (line.size() < 17 || line[8] != ' ')
+        return out;
+
+    const auto word2 = Common::HexStringToArray<sizeof(u32)>(std::string_view{line.data() + 9, 8});
+    std::memcpy(out.data() + sizeof(u32), word2.data(), sizeof(u32));
+
+    if (line.size() < 26 || line[17] != ' ') {
+        // Perform shifting in case value is truncated early.
+        const auto type = static_cast<CodeType>((out[0] & 0xF0) >> 4);
+        if (type == CodeType::Loop || type == CodeType::LoadImmediate ||
+            type == CodeType::StoreIndexed || type == CodeType::RegisterArithmetic) {
+            std::memcpy(out.data() + 8, out.data() + 4, sizeof(u32));
+            std::memset(out.data() + 4, 0, sizeof(u32));
+        }
+
+        return out;
+    }
+
+    const auto word3 = Common::HexStringToArray<sizeof(u32)>(std::string_view{line.data() + 18, 8});
+    std::memcpy(out.data() + 2 * sizeof(u32), word3.data(), sizeof(u32));
+
+    if (line.size() < 35 || line[26] != ' ') {
+        // Perform shifting in case value is truncated early.
+        const auto type = static_cast<CodeType>((out[0] & 0xF0) >> 4);
+        if (type == CodeType::WriteImmediate || type == CodeType::Conditional) {
+            std::memcpy(out.data() + 12, out.data() + 8, sizeof(u32));
+            std::memset(out.data() + 8, 0, sizeof(u32));
+        }
+
+        return out;
+    }
+
+    const auto word4 = Common::HexStringToArray<sizeof(u32)>(std::string_view{line.data() + 27, 8});
+    std::memcpy(out.data() + 3 * sizeof(u32), word4.data(), sizeof(u32));
+
+    return out;
+}
+
+u64 MemoryReadImpl(u32 width, VAddr addr) {
+    switch (width) {
+    case 1:
+        return Memory::Read8(addr);
+    case 2:
+        return Memory::Read16(addr);
+    case 4:
+        return Memory::Read32(addr);
+    case 8:
+        return Memory::Read64(addr);
+    default:
+        UNREACHABLE();
+        return 0;
+    }
+}
+
+void MemoryWriteImpl(u32 width, VAddr addr, u64 value) {
+    switch (width) {
+    case 1:
+        Memory::Write8(addr, static_cast<u8>(value));
+        break;
+    case 2:
+        Memory::Write16(addr, static_cast<u16>(value));
+        break;
+    case 4:
+        Memory::Write32(addr, static_cast<u32>(value));
+        break;
+    case 8:
+        Memory::Write64(addr, value);
+        break;
+    default:
+        UNREACHABLE();
+    }
+}
+
+CheatEngine::CheatEngine(Core::System& system, std::vector<CheatList> cheats_,
+                         const std::string& build_id, VAddr code_region_start,
+                         VAddr code_region_end)
+    : cheats{std::move(cheats_)}, core_timing{system.CoreTiming()} {
+    event = core_timing.RegisterEvent(
+        "CheatEngine::FrameCallback::" + build_id,
+        [this](u64 userdata, s64 cycles_late) { FrameCallback(userdata, cycles_late); });
+    core_timing.ScheduleEvent(CHEAT_ENGINE_TICKS, event);
+
+    const auto& vm_manager = system.CurrentProcess()->VMManager();
+    for (auto& list : this->cheats) {
+        list.SetMemoryParameters(code_region_start, vm_manager.GetHeapRegionBaseAddress(),
+                                 code_region_end, vm_manager.GetHeapRegionEndAddress(),
+                                 &MemoryWriteImpl, &MemoryReadImpl);
+    }
+}
+
+CheatEngine::~CheatEngine() {
+    core_timing.UnscheduleEvent(event, 0);
+}
+
+void CheatEngine::FrameCallback(u64 userdata, s64 cycles_late) {
+    for (auto& list : cheats) {
+        list.Execute();
+    }
+
+    core_timing.ScheduleEvent(CHEAT_ENGINE_TICKS - cycles_late, event);
+}
+
+} // namespace FileSys

src/core/file_sys/cheat_engine.h

@@ -0,0 +1,234 @@
+// Copyright 2018 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <map>
+#include <set>
+#include <vector>
+#include "common/bit_field.h"
+#include "common/common_types.h"
+
+namespace Core {
+class System;
+}
+
+namespace Core::Timing {
+class CoreTiming;
+struct EventType;
+} // namespace Core::Timing
+
+namespace FileSys {
+
+enum class CodeType : u32 {
+    // 0TMR00AA AAAAAAAA YYYYYYYY YYYYYYYY
+    // Writes a T sized value Y to the address A added to the value of register R in memory domain M
+    WriteImmediate = 0,
+
+    // 1TMC00AA AAAAAAAA YYYYYYYY YYYYYYYY
+    // Compares the T sized value Y to the value at address A in memory domain M using the
+    // conditional function C. If success, continues execution. If failure, jumps to the matching
+    // EndConditional statement.
+    Conditional = 1,
+
+    // 20000000
+    // Terminates a Conditional or ConditionalInput block.
+    EndConditional = 2,
+
+    // 300R0000 VVVVVVVV
+    // Starts looping V times, storing the current count in register R.
+    // Loop block is terminated with a matching 310R0000.
+    Loop = 3,
+
+    // 400R0000 VVVVVVVV VVVVVVVV
+    // Sets the value of register R to the value V.
+    LoadImmediate = 4,
+
+    // 5TMRI0AA AAAAAAAA
+    // Sets the value of register R to the value of width T at address A in memory domain M, with
+    // the current value of R added to the address if I == 1.
+    LoadIndexed = 5,
+
+    // 6T0RIFG0 VVVVVVVV VVVVVVVV
+    // Writes the value V of width T to the memory address stored in register R. Adds the value of
+    // register G to the final calculation if F is nonzero. Increments the value of register R by T
+    // after operation if I is nonzero.
+    StoreIndexed = 6,
+
+    // 7T0RA000 VVVVVVVV
+    // Performs the arithmetic operation A on the value in register R and the value V of width T,
+    // storing the result in register R.
+    RegisterArithmetic = 7,
+
+    // 8KKKKKKK
+    // Checks to see if any of the buttons defined by the bitmask K are pressed. If any are,
+    // execution continues. If none are, execution skips to the next EndConditional command.
+    ConditionalInput = 8,
+};
+
+enum class MemoryType : u32 {
+    // Addressed relative to start of main NSO
+    MainNSO = 0,
+
+    // Addressed relative to start of heap
+    Heap = 1,
+};
+
+enum class ArithmeticOp : u32 {
+    Add = 0,
+    Sub = 1,
+    Mult = 2,
+    LShift = 3,
+    RShift = 4,
+};
+
+enum class ComparisonOp : u32 {
+    GreaterThan = 1,
+    GreaterThanEqual = 2,
+    LessThan = 3,
+    LessThanEqual = 4,
+    Equal = 5,
+    Inequal = 6,
+};
+
+union Cheat {
+    std::array<u8, 16> raw;
+
+    BitField<4, 4, CodeType> type;
+    BitField<0, 4, u32> width; // Can be 1, 2, 4, or 8. Measured in bytes.
+    BitField<0, 4, u32> end_of_loop;
+    BitField<12, 4, MemoryType> memory_type;
+    BitField<8, 4, u32> register_3;
+    BitField<8, 4, ComparisonOp> comparison_op;
+    BitField<20, 4, u32> load_from_register;
+    BitField<20, 4, u32> increment_register;
+    BitField<20, 4, ArithmeticOp> arithmetic_op;
+    BitField<16, 4, u32> add_additional_register;
+    BitField<28, 4, u32> register_6;
+
+    u64 Address() const;
+    u64 ValueWidth(u64 offset) const;
+    u64 Value(u64 offset, u64 width) const;
+    u32 KeypadValue() const;
+};
+
+class CheatParser;
+
+// Represents a full collection of cheats for a game. The Execute function should be called every
+// interval that all cheats should be executed. Clients should not directly instantiate this class
+// (hence private constructor), they should instead receive an instance from CheatParser, which
+// guarantees the list is always in an acceptable state.
+class CheatList {
+public:
+    friend class CheatParser;
+
+    using Block = std::vector<Cheat>;
+    using ProgramSegment = std::vector<std::pair<std::string, Block>>;
+
+    // (width in bytes, address, value)
+    using MemoryWriter = void (*)(u32, VAddr, u64);
+    // (width in bytes, address) -> value
+    using MemoryReader = u64 (*)(u32, VAddr);
+
+    void SetMemoryParameters(VAddr main_begin, VAddr heap_begin, VAddr main_end, VAddr heap_end,
+                             MemoryWriter writer, MemoryReader reader);
+
+    void Execute();
+
+private:
+    CheatList(const Core::System& system_, ProgramSegment master, ProgramSegment standard);
+
+    void ProcessBlockPairs(const Block& block);
+    void ExecuteSingleCheat(const Cheat& cheat);
+
+    void ExecuteBlock(const Block& block);
+
+    bool EvaluateConditional(const Cheat& cheat) const;
+
+    // Individual cheat operations
+    void WriteImmediate(const Cheat& cheat);
+    void BeginConditional(const Cheat& cheat);
+    void EndConditional(const Cheat& cheat);
+    void Loop(const Cheat& cheat);
+    void LoadImmediate(const Cheat& cheat);
+    void LoadIndexed(const Cheat& cheat);
+    void StoreIndexed(const Cheat& cheat);
+    void RegisterArithmetic(const Cheat& cheat);
+    void BeginConditionalInput(const Cheat& cheat);
+
+    VAddr SanitizeAddress(VAddr in) const;
+
+    // Master Codes are defined as codes that cannot be disabled and are run prior to all
+    // others.
+    ProgramSegment master_list;
+    // All other codes
+    ProgramSegment standard_list;
+
+    bool in_standard = false;
+
+    // 16 (0x0-0xF) scratch registers that can be used by cheats
+    std::array<u64, 16> scratch{};
+
+    MemoryWriter writer = nullptr;
+    MemoryReader reader = nullptr;
+
+    u64 main_region_begin{};
+    u64 heap_region_begin{};
+    u64 main_region_end{};
+    u64 heap_region_end{};
+
+    u64 current_block{};
+    // The current index of the cheat within the current Block
+    u64 current_index{};
+
+    // The 'stack' of the program. When a conditional or loop statement is encountered, its index is
+    // pushed onto this queue. When a end block is encountered, the condition is checked.
+    std::map<u64, u64> block_pairs;
+
+    std::set<u64> encountered_loops;
+
+    const Core::System* system;
+};
+
+// Intermediary class that parses a text file or other disk format for storing cheats into a
+// CheatList object, that can be used for execution.
+class CheatParser {
+public:
+    virtual ~CheatParser();
+
+    virtual CheatList Parse(const Core::System& system, const std::vector<u8>& data) const = 0;
+
+protected:
+    CheatList MakeCheatList(const Core::System& system_, CheatList::ProgramSegment master,
+                            CheatList::ProgramSegment standard) const;
+};
+
+// CheatParser implementation that parses text files
+class TextCheatParser final : public CheatParser {
+public:
+    ~TextCheatParser() override;
+
+    CheatList Parse(const Core::System& system, const std::vector<u8>& data) const override;
+
+private:
+    std::array<u8, 16> ParseSingleLineCheat(const std::string& line) const;
+};
+
+// Class that encapsulates a CheatList and manages its interaction with memory and CoreTiming
+class CheatEngine final {
+public:
+    CheatEngine(Core::System& system_, std::vector<CheatList> cheats_, const std::string& build_id,
+                VAddr code_region_start, VAddr code_region_end);
+    ~CheatEngine();
+
+private:
+    void FrameCallback(u64 userdata, s64 cycles_late);
+
+    std::vector<CheatList> cheats;
+
+    Core::Timing::EventType* event;
+    Core::Timing::CoreTiming& core_timing;
+};
+
+} // namespace FileSys

src/core/file_sys/content_archive.h

@@ -24,13 +24,26 @@ namespace FileSys {
 
 union NCASectionHeader;
 
+/// Describes the type of content within an NCA archive.
 enum class NCAContentType : u8 {
+    /// Executable-related data
     Program = 0,
+
+    /// Metadata.
     Meta = 1,
+
+    /// Access control data.
     Control = 2,
+
+    /// Information related to the game manual
+    /// e.g. Legal information, etc.
     Manual = 3,
+
+    /// System data.
     Data = 4,
-    Data_Unknown5 = 5, ///< Seems to be used on some system archives
+
+    /// Data that can be accessed by applications.
+    PublicData = 5,
 };
 
 enum class NCASectionCryptoType : u8 {

src/core/file_sys/patch_manager.cpp

@@ -7,6 +7,7 @@
 #include <cstddef>
 #include <cstring>
 
+#include "common/file_util.h"
 #include "common/hex_util.h"
 #include "common/logging/log.h"
 #include "core/file_sys/content_archive.h"
@@ -232,6 +233,57 @@ bool PatchManager::HasNSOPatch(const std::array<u8, 32>& build_id_) const {
     return !CollectPatches(patch_dirs, build_id).empty();
 }
 
+static std::optional<CheatList> ReadCheatFileFromFolder(const Core::System& system, u64 title_id,
+                                                        const std::array<u8, 0x20>& build_id_,
+                                                        const VirtualDir& base_path, bool upper) {
+    const auto build_id_raw = Common::HexArrayToString(build_id_, upper);
+    const auto build_id = build_id_raw.substr(0, sizeof(u64) * 2);
+    const auto file = base_path->GetFile(fmt::format("{}.txt", build_id));
+
+    if (file == nullptr) {
+        LOG_INFO(Common_Filesystem, "No cheats file found for title_id={:016X}, build_id={}",
+                 title_id, build_id);
+        return std::nullopt;
+    }
+
+    std::vector<u8> data(file->GetSize());
+    if (file->Read(data.data(), data.size()) != data.size()) {
+        LOG_INFO(Common_Filesystem, "Failed to read cheats file for title_id={:016X}, build_id={}",
+                 title_id, build_id);
+        return std::nullopt;
+    }
+
+    TextCheatParser parser;
+    return parser.Parse(system, data);
+}
+
+std::vector<CheatList> PatchManager::CreateCheatList(const Core::System& system,
+                                                     const std::array<u8, 32>& build_id_) const {
+    const auto load_dir = Service::FileSystem::GetModificationLoadRoot(title_id);
+    auto patch_dirs = load_dir->GetSubdirectories();
+    std::sort(patch_dirs.begin(), patch_dirs.end(),
+              [](const VirtualDir& l, const VirtualDir& r) { return l->GetName() < r->GetName(); });
+
+    std::vector<CheatList> out;
+    out.reserve(patch_dirs.size());
+    for (const auto& subdir : patch_dirs) {
+        auto cheats_dir = subdir->GetSubdirectory("cheats");
+        if (cheats_dir != nullptr) {
+            auto res = ReadCheatFileFromFolder(system, title_id, build_id_, cheats_dir, true);
+            if (res.has_value()) {
+                out.push_back(std::move(*res));
+                continue;
+            }
+
+            res = ReadCheatFileFromFolder(system, title_id, build_id_, cheats_dir, false);
+            if (res.has_value())
+                out.push_back(std::move(*res));
+        }
+    }
+
+    return out;
+}
+
 static void ApplyLayeredFS(VirtualFile& romfs, u64 title_id, ContentRecordType type) {
     const auto load_dir = Service::FileSystem::GetModificationLoadRoot(title_id);
     if ((type != ContentRecordType::Program && type != ContentRecordType::Data) ||
@@ -403,6 +455,8 @@ std::map<std::string, std::string, std::less<>> PatchManager::GetPatchVersionNam
             }
             if (IsDirValidAndNonEmpty(mod->GetSubdirectory("romfs")))
                 AppendCommaIfNotEmpty(types, "LayeredFS");
+            if (IsDirValidAndNonEmpty(mod->GetSubdirectory("cheats")))
+                AppendCommaIfNotEmpty(types, "Cheats");
 
             if (types.empty())
                 continue;

src/core/file_sys/patch_manager.h

@@ -8,9 +8,14 @@
 #include <memory>
 #include <string>
 #include "common/common_types.h"
+#include "core/file_sys/cheat_engine.h"
 #include "core/file_sys/nca_metadata.h"
 #include "core/file_sys/vfs.h"
 
+namespace Core {
+class System;
+}
+
 namespace FileSys {
 
 class NCA;
@@ -45,6 +50,10 @@ public:
     // Used to prevent expensive copies in NSO loader.
     bool HasNSOPatch(const std::array<u8, 0x20>& build_id) const;
 
+    // Creates a CheatList object with all
+    std::vector<CheatList> CreateCheatList(const Core::System& system,
+                                           const std::array<u8, 0x20>& build_id) const;
+
     // Currently tracked RomFS patches:
     // - Game Updates
     // - LayeredFS

src/core/file_sys/registered_cache.cpp

@@ -94,7 +94,7 @@ static ContentRecordType GetCRTypeFromNCAType(NCAContentType type) {
     case NCAContentType::Control:
         return ContentRecordType::Control;
     case NCAContentType::Data:
-    case NCAContentType::Data_Unknown5:
+    case NCAContentType::PublicData:
         return ContentRecordType::Data;
     case NCAContentType::Manual:
         // TODO(DarkLordZach): Peek at NCA contents to differentiate Manual and Legal.

src/core/hle/ipc.h

@@ -39,10 +39,10 @@ struct CommandHeader {
     union {
         u32_le raw_low;
         BitField<0, 16, CommandType> type;
-        BitField<16, 4, u32_le> num_buf_x_descriptors;
-        BitField<20, 4, u32_le> num_buf_a_descriptors;
-        BitField<24, 4, u32_le> num_buf_b_descriptors;
-        BitField<28, 4, u32_le> num_buf_w_descriptors;
+        BitField<16, 4, u32> num_buf_x_descriptors;
+        BitField<20, 4, u32> num_buf_a_descriptors;
+        BitField<24, 4, u32> num_buf_b_descriptors;
+        BitField<28, 4, u32> num_buf_w_descriptors;
     };
 
     enum class BufferDescriptorCFlag : u32 {
@@ -53,28 +53,28 @@ struct CommandHeader {
 
     union {
         u32_le raw_high;
-        BitField<0, 10, u32_le> data_size;
+        BitField<0, 10, u32> data_size;
         BitField<10, 4, BufferDescriptorCFlag> buf_c_descriptor_flags;
-        BitField<31, 1, u32_le> enable_handle_descriptor;
+        BitField<31, 1, u32> enable_handle_descriptor;
     };
 };
 static_assert(sizeof(CommandHeader) == 8, "CommandHeader size is incorrect");
 
 union HandleDescriptorHeader {
     u32_le raw_high;
-    BitField<0, 1, u32_le> send_current_pid;
-    BitField<1, 4, u32_le> num_handles_to_copy;
-    BitField<5, 4, u32_le> num_handles_to_move;
+    BitField<0, 1, u32> send_current_pid;
+    BitField<1, 4, u32> num_handles_to_copy;
+    BitField<5, 4, u32> num_handles_to_move;
 };
 static_assert(sizeof(HandleDescriptorHeader) == 4, "HandleDescriptorHeader size is incorrect");
 
 struct BufferDescriptorX {
     union {
-        BitField<0, 6, u32_le> counter_bits_0_5;
-        BitField<6, 3, u32_le> address_bits_36_38;
-        BitField<9, 3, u32_le> counter_bits_9_11;
-        BitField<12, 4, u32_le> address_bits_32_35;
-        BitField<16, 16, u32_le> size;
+        BitField<0, 6, u32> counter_bits_0_5;
+        BitField<6, 3, u32> address_bits_36_38;
+        BitField<9, 3, u32> counter_bits_9_11;
+        BitField<12, 4, u32> address_bits_32_35;
+        BitField<16, 16, u32> size;
     };
 
     u32_le address_bits_0_31;
@@ -103,10 +103,10 @@ struct BufferDescriptorABW {
     u32_le address_bits_0_31;
 
     union {
-        BitField<0, 2, u32_le> flags;
-        BitField<2, 3, u32_le> address_bits_36_38;
-        BitField<24, 4, u32_le> size_bits_32_35;
-        BitField<28, 4, u32_le> address_bits_32_35;
+        BitField<0, 2, u32> flags;
+        BitField<2, 3, u32> address_bits_36_38;
+        BitField<24, 4, u32> size_bits_32_35;
+        BitField<28, 4, u32> address_bits_32_35;
     };
 
     VAddr Address() const {
@@ -128,8 +128,8 @@ struct BufferDescriptorC {
     u32_le address_bits_0_31;
 
     union {
-        BitField<0, 16, u32_le> address_bits_32_47;
-        BitField<16, 16, u32_le> size;
+        BitField<0, 16, u32> address_bits_32_47;
+        BitField<16, 16, u32> size;
     };
 
     VAddr Address() const {
@@ -167,8 +167,8 @@ struct DomainMessageHeader {
         struct {
             union {
                 BitField<0, 8, CommandType> command;
-                BitField<8, 8, u32_le> input_object_count;
-                BitField<16, 16, u32_le> size;
+                BitField<8, 8, u32> input_object_count;
+                BitField<16, 16, u32> size;
             };
             u32_le object_id;
             INSERT_PADDING_WORDS(2);

src/core/hle/ipc_helpers.h

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

src/core/hle/kernel/code_set.cpp

@@ -0,0 +1,12 @@
+// Copyright 2019 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "core/hle/kernel/code_set.h"
+
+namespace Kernel {
+
+CodeSet::CodeSet() = default;
+CodeSet::~CodeSet() = default;
+
+} // namespace Kernel

src/core/hle/kernel/code_set.h

@@ -0,0 +1,90 @@
+// Copyright 2019 yuzu emulator team
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <cstddef>
+#include <memory>
+#include <vector>
+
+#include "common/common_types.h"
+
+namespace Kernel {
+
+/**
+ * Represents executable data that may be loaded into a kernel process.
+ *
+ * A code set consists of three basic segments:
+ *   - A code (AKA text) segment,
+ *   - A read-only data segment (rodata)
+ *   - A data segment
+ *
+ * The code segment is the portion of the object file that contains
+ * executable instructions.
+ *
+ * The read-only data segment in the portion of the object file that
+ * contains (as one would expect) read-only data, such as fixed constant
+ * values and data structures.
+ *
+ * The data segment is similar to the read-only data segment -- it contains
+ * variables and data structures that have predefined values, however,
+ * entities within this segment can be modified.
+ */
+struct CodeSet final {
+    /// A single segment within a code set.
+    struct Segment final {
+        /// The byte offset that this segment is located at.
+        std::size_t offset = 0;
+
+        /// The address to map this segment to.
+        VAddr addr = 0;
+
+        /// The size of this segment in bytes.
+        u32 size = 0;
+    };
+
+    explicit CodeSet();
+    ~CodeSet();
+
+    CodeSet(const CodeSet&) = delete;
+    CodeSet& operator=(const CodeSet&) = delete;
+
+    CodeSet(CodeSet&&) = default;
+    CodeSet& operator=(CodeSet&&) = default;
+
+    Segment& CodeSegment() {
+        return segments[0];
+    }
+
+    const Segment& CodeSegment() const {
+        return segments[0];
+    }
+
+    Segment& RODataSegment() {
+        return segments[1];
+    }
+
+    const Segment& RODataSegment() const {
+        return segments[1];
+    }
+
+    Segment& DataSegment() {
+        return segments[2];
+    }
+
+    const Segment& DataSegment() const {
+        return segments[2];
+    }
+
+    /// The overall data that backs this code set.
+    std::shared_ptr<std::vector<u8>> memory;
+
+    /// The segments that comprise this code set.
+    std::array<Segment, 3> segments;
+
+    /// The entry point address for this code set.
+    VAddr entrypoint = 0;
+};
+
+} // namespace Kernel

src/core/hle/kernel/process.cpp

@@ -9,6 +9,7 @@
 #include "common/logging/log.h"
 #include "core/core.h"
 #include "core/file_sys/program_metadata.h"
+#include "core/hle/kernel/code_set.h"
 #include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/kernel.h"
 #include "core/hle/kernel/process.h"
@@ -31,7 +32,7 @@ namespace {
  */
 void SetupMainThread(Process& owner_process, KernelCore& kernel, VAddr entry_point, u32 priority) {
     // Setup page table so we can write to memory
-    SetCurrentPageTable(&owner_process.VMManager().page_table);
+    Memory::SetCurrentPageTable(&owner_process.VMManager().page_table);
 
     // Initialize new "main" thread
     const VAddr stack_top = owner_process.VMManager().GetTLSIORegionEndAddress();
@@ -50,9 +51,6 @@ void SetupMainThread(Process& owner_process, KernelCore& kernel, VAddr entry_poi
 }
 } // Anonymous namespace
 
-CodeSet::CodeSet() = default;
-CodeSet::~CodeSet() = default;
-
 SharedPtr<Process> Process::Create(Core::System& system, std::string&& name) {
     auto& kernel = system.Kernel();
 
@@ -212,7 +210,7 @@ void Process::FreeTLSSlot(VAddr tls_address) {
 }
 
 void Process::LoadModule(CodeSet module_, VAddr base_addr) {
-    const auto MapSegment = [&](CodeSet::Segment& segment, VMAPermission permissions,
+    const auto MapSegment = [&](const CodeSet::Segment& segment, VMAPermission permissions,
                                 MemoryState memory_state) {
         const auto vma = vm_manager
                              .MapMemoryBlock(segment.addr + base_addr, module_.memory,
@@ -222,9 +220,9 @@ void Process::LoadModule(CodeSet module_, VAddr base_addr) {
     };
 
     // Map CodeSet segments
-    MapSegment(module_.CodeSegment(), VMAPermission::ReadExecute, MemoryState::CodeStatic);
-    MapSegment(module_.RODataSegment(), VMAPermission::Read, MemoryState::CodeMutable);
-    MapSegment(module_.DataSegment(), VMAPermission::ReadWrite, MemoryState::CodeMutable);
+    MapSegment(module_.CodeSegment(), VMAPermission::ReadExecute, MemoryState::Code);
+    MapSegment(module_.RODataSegment(), VMAPermission::Read, MemoryState::CodeData);
+    MapSegment(module_.DataSegment(), VMAPermission::ReadWrite, MemoryState::CodeData);
 
     // Clear instruction cache in CPU JIT
     system.InvalidateCpuInstructionCaches();

src/core/hle/kernel/process.h

@@ -7,7 +7,6 @@
 #include <array>
 #include <bitset>
 #include <cstddef>
-#include <memory>
 #include <string>
 #include <vector>
 #include <boost/container/static_vector.hpp>
@@ -34,6 +33,8 @@ class KernelCore;
 class ResourceLimit;
 class Thread;
 
+struct CodeSet;
+
 struct AddressMapping {
     // Address and size must be page-aligned
     VAddr address;
@@ -66,46 +67,6 @@ enum class ProcessStatus {
     DebugBreak,
 };
 
-struct CodeSet final {
-    struct Segment {
-        std::size_t offset = 0;
-        VAddr addr = 0;
-        u32 size = 0;
-    };
-
-    explicit CodeSet();
-    ~CodeSet();
-
-    Segment& CodeSegment() {
-        return segments[0];
-    }
-
-    const Segment& CodeSegment() const {
-        return segments[0];
-    }
-
-    Segment& RODataSegment() {
-        return segments[1];
-    }
-
-    const Segment& RODataSegment() const {
-        return segments[1];
-    }
-
-    Segment& DataSegment() {
-        return segments[2];
-    }
-
-    const Segment& DataSegment() const {
-        return segments[2];
-    }
-
-    std::shared_ptr<std::vector<u8>> memory;
-
-    std::array<Segment, 3> segments;
-    VAddr entrypoint = 0;
-};
-
 class Process final : public WaitObject {
 public:
     enum : u64 {

src/core/hle/kernel/scheduler.cpp

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

src/core/hle/kernel/svc.cpp

@@ -1285,10 +1285,14 @@ static ResultCode StartThread(Handle thread_handle) {
 
 /// Called when a thread exits
 static void ExitThread() {
-    LOG_TRACE(Kernel_SVC, "called, pc=0x{:08X}", Core::CurrentArmInterface().GetPC());
+    auto& system = Core::System::GetInstance();
 
-    ExitCurrentThread();
-    Core::System::GetInstance().PrepareReschedule();
+    LOG_TRACE(Kernel_SVC, "called, pc=0x{:08X}", system.CurrentArmInterface().GetPC());
+
+    auto* const current_thread = system.CurrentScheduler().GetCurrentThread();
+    current_thread->Stop();
+    system.CurrentScheduler().RemoveThread(current_thread);
+    system.PrepareReschedule();
 }
 
 /// Sleep the current thread
@@ -1301,32 +1305,32 @@ static void SleepThread(s64 nanoseconds) {
         YieldAndWaitForLoadBalancing = -2,
     };
 
+    auto& system = Core::System::GetInstance();
+    auto& scheduler = system.CurrentScheduler();
+    auto* const current_thread = scheduler.GetCurrentThread();
+
     if (nanoseconds <= 0) {
-        auto& scheduler{Core::System::GetInstance().CurrentScheduler()};
         switch (static_cast<SleepType>(nanoseconds)) {
         case SleepType::YieldWithoutLoadBalancing:
-            scheduler.YieldWithoutLoadBalancing(GetCurrentThread());
+            scheduler.YieldWithoutLoadBalancing(current_thread);
             break;
         case SleepType::YieldWithLoadBalancing:
-            scheduler.YieldWithLoadBalancing(GetCurrentThread());
+            scheduler.YieldWithLoadBalancing(current_thread);
             break;
         case SleepType::YieldAndWaitForLoadBalancing:
-            scheduler.YieldAndWaitForLoadBalancing(GetCurrentThread());
+            scheduler.YieldAndWaitForLoadBalancing(current_thread);
             break;
         default:
             UNREACHABLE_MSG("Unimplemented sleep yield type '{:016X}'!", nanoseconds);
         }
     } else {
-        // Sleep current thread and check for next thread to schedule
-        WaitCurrentThread_Sleep();
-
-        // Create an event to wake the thread up after the specified nanosecond delay has passed
-        GetCurrentThread()->WakeAfterDelay(nanoseconds);
+        current_thread->Sleep(nanoseconds);
     }
 
     // Reschedule all CPU cores
-    for (std::size_t i = 0; i < Core::NUM_CPU_CORES; ++i)
-        Core::System::GetInstance().CpuCore(i).PrepareReschedule();
+    for (std::size_t i = 0; i < Core::NUM_CPU_CORES; ++i) {
+        system.CpuCore(i).PrepareReschedule();
+    }
 }
 
 /// Wait process wide key atomic

src/core/hle/kernel/thread.cpp

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

src/core/hle/kernel/thread.h

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

src/core/hle/kernel/vm_manager.cpp

@@ -7,29 +7,29 @@
 #include <utility>
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "common/memory_hook.h"
 #include "core/arm/arm_interface.h"
 #include "core/core.h"
 #include "core/file_sys/program_metadata.h"
 #include "core/hle/kernel/errors.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/memory.h"
-#include "core/memory_hook.h"
 #include "core/memory_setup.h"
 
 namespace Kernel {
 namespace {
 const char* GetMemoryStateName(MemoryState state) {
     static constexpr const char* names[] = {
-        "Unmapped",         "Io",
-        "Normal",           "CodeStatic",
-        "CodeMutable",      "Heap",
-        "Shared",           "Unknown1",
-        "ModuleCodeStatic", "ModuleCodeMutable",
-        "IpcBuffer0",       "Stack",
-        "ThreadLocal",      "TransferMemoryIsolated",
-        "TransferMemory",   "ProcessMemory",
-        "Inaccessible",     "IpcBuffer1",
-        "IpcBuffer3",       "KernelStack",
+        "Unmapped",       "Io",
+        "Normal",         "Code",
+        "CodeData",       "Heap",
+        "Shared",         "Unknown1",
+        "ModuleCode",     "ModuleCodeData",
+        "IpcBuffer0",     "Stack",
+        "ThreadLocal",    "TransferMemoryIsolated",
+        "TransferMemory", "ProcessMemory",
+        "Inaccessible",   "IpcBuffer1",
+        "IpcBuffer3",     "KernelStack",
     };
 
     return names[ToSvcMemoryState(state)];
@@ -177,7 +177,7 @@ ResultVal<VAddr> VMManager::FindFreeRegion(u64 size) const {
 
 ResultVal<VMManager::VMAHandle> VMManager::MapMMIO(VAddr target, PAddr paddr, u64 size,
                                                    MemoryState state,
-                                                   Memory::MemoryHookPointer mmio_handler) {
+                                                   Common::MemoryHookPointer mmio_handler) {
     // This is the appropriately sized VMA that will turn into our allocation.
     CASCADE_RESULT(VMAIter vma_handle, CarveVMA(target, size));
     VirtualMemoryArea& final_vma = vma_handle->second;
@@ -624,7 +624,7 @@ void VMManager::ClearPageTable() {
     std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr);
     page_table.special_regions.clear();
     std::fill(page_table.attributes.begin(), page_table.attributes.end(),
-              Memory::PageType::Unmapped);
+              Common::PageType::Unmapped);
 }
 
 VMManager::CheckResults VMManager::CheckRangeState(VAddr address, u64 size, MemoryState state_mask,

src/core/hle/kernel/vm_manager.h

@@ -9,9 +9,10 @@
 #include <tuple>
 #include <vector>
 #include "common/common_types.h"
+#include "common/memory_hook.h"
+#include "common/page_table.h"
 #include "core/hle/result.h"
 #include "core/memory.h"
-#include "core/memory_hook.h"
 
 namespace FileSys {
 enum class ProgramAddressSpaceType : u8;
@@ -164,12 +165,12 @@ enum class MemoryState : u32 {
     Unmapped               = 0x00,
     Io                     = 0x01 | FlagMapped,
     Normal                 = 0x02 | FlagMapped | FlagQueryPhysicalAddressAllowed,
-    CodeStatic             = 0x03 | CodeFlags  | FlagMapProcess,
-    CodeMutable            = 0x04 | CodeFlags  | FlagMapProcess | FlagCodeMemory,
+    Code                   = 0x03 | CodeFlags  | FlagMapProcess,
+    CodeData               = 0x04 | DataFlags  | FlagMapProcess | FlagCodeMemory,
     Heap                   = 0x05 | DataFlags  | FlagCodeMemory,
     Shared                 = 0x06 | FlagMapped | FlagMemoryPoolAllocated,
-    ModuleCodeStatic       = 0x08 | CodeFlags  | FlagModule | FlagMapProcess,
-    ModuleCodeMutable      = 0x09 | DataFlags  | FlagModule | FlagMapProcess | FlagCodeMemory,
+    ModuleCode             = 0x08 | CodeFlags  | FlagModule | FlagMapProcess,
+    ModuleCodeData         = 0x09 | DataFlags  | FlagModule | FlagMapProcess | FlagCodeMemory,
 
     IpcBuffer0             = 0x0A | FlagMapped | FlagQueryPhysicalAddressAllowed | FlagMemoryPoolAllocated |
                                     IPCFlags | FlagSharedDevice | FlagSharedDeviceAligned,
@@ -290,7 +291,7 @@ struct VirtualMemoryArea {
     // Settings for type = MMIO
     /// Physical address of the register area this VMA maps to.
     PAddr paddr = 0;
-    Memory::MemoryHookPointer mmio_handler = nullptr;
+    Common::MemoryHookPointer mmio_handler = nullptr;
 
     /// Tests if this area can be merged to the right with `next`.
     bool CanBeMergedWith(const VirtualMemoryArea& next) const;
@@ -368,7 +369,7 @@ public:
      * @param mmio_handler The handler that will implement read and write for this MMIO region.
      */
     ResultVal<VMAHandle> MapMMIO(VAddr target, PAddr paddr, u64 size, MemoryState state,
-                                 Memory::MemoryHookPointer mmio_handler);
+                                 Common::MemoryHookPointer mmio_handler);
 
     /// Unmaps a range of addresses, splitting VMAs as necessary.
     ResultCode UnmapRange(VAddr target, u64 size);
@@ -509,7 +510,7 @@ public:
 
     /// Each VMManager has its own page table, which is set as the main one when the owning process
     /// is scheduled.
-    Memory::PageTable page_table;
+    Common::PageTable page_table{Memory::PAGE_BITS};
 
 private:
     using VMAIter = VMAMap::iterator;
@@ -616,6 +617,9 @@ private:
     VAddr new_map_region_base = 0;
     VAddr new_map_region_end = 0;
 
+    VAddr main_code_region_base = 0;
+    VAddr main_code_region_end = 0;
+
     VAddr tls_io_region_base = 0;
     VAddr tls_io_region_end = 0;
 

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

@@ -2,10 +2,10 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include <algorithm>
 #include <array>
 #include <cinttypes>
 #include <cstring>
-#include <stack>
 #include "audio_core/audio_renderer.h"
 #include "core/core.h"
 #include "core/file_sys/savedata_factory.h"
@@ -93,38 +93,84 @@ void IWindowController::AcquireForegroundRights(Kernel::HLERequestContext& ctx)
 }
 
 IAudioController::IAudioController() : ServiceFramework("IAudioController") {
+    // clang-format off
     static const FunctionInfo functions[] = {
         {0, &IAudioController::SetExpectedMasterVolume, "SetExpectedMasterVolume"},
-        {1, &IAudioController::GetMainAppletExpectedMasterVolume,
-         "GetMainAppletExpectedMasterVolume"},
-        {2, &IAudioController::GetLibraryAppletExpectedMasterVolume,
-         "GetLibraryAppletExpectedMasterVolume"},
-        {3, nullptr, "ChangeMainAppletMasterVolume"},
-        {4, nullptr, "SetTransparentVolumeRate"},
+        {1, &IAudioController::GetMainAppletExpectedMasterVolume, "GetMainAppletExpectedMasterVolume"},
+        {2, &IAudioController::GetLibraryAppletExpectedMasterVolume, "GetLibraryAppletExpectedMasterVolume"},
+        {3, &IAudioController::ChangeMainAppletMasterVolume, "ChangeMainAppletMasterVolume"},
+        {4, &IAudioController::SetTransparentAudioRate, "SetTransparentVolumeRate"},
     };
+    // clang-format on
+
     RegisterHandlers(functions);
 }
 
 IAudioController::~IAudioController() = default;
 
 void IAudioController::SetExpectedMasterVolume(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_AM, "(STUBBED) called");
+    IPC::RequestParser rp{ctx};
+    const float main_applet_volume_tmp = rp.Pop<float>();
+    const float library_applet_volume_tmp = rp.Pop<float>();
+
+    LOG_DEBUG(Service_AM, "called. main_applet_volume={}, library_applet_volume={}",
+              main_applet_volume_tmp, library_applet_volume_tmp);
+
+    // Ensure the volume values remain within the 0-100% range
+    main_applet_volume = std::clamp(main_applet_volume_tmp, min_allowed_volume, max_allowed_volume);
+    library_applet_volume =
+        std::clamp(library_applet_volume_tmp, min_allowed_volume, max_allowed_volume);
+
     IPC::ResponseBuilder rb{ctx, 2};
     rb.Push(RESULT_SUCCESS);
 }
 
 void IAudioController::GetMainAppletExpectedMasterVolume(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_AM, "(STUBBED) called");
+    LOG_DEBUG(Service_AM, "called. main_applet_volume={}", main_applet_volume);
     IPC::ResponseBuilder rb{ctx, 3};
     rb.Push(RESULT_SUCCESS);
-    rb.Push(volume);
+    rb.Push(main_applet_volume);
 }
 
 void IAudioController::GetLibraryAppletExpectedMasterVolume(Kernel::HLERequestContext& ctx) {
-    LOG_WARNING(Service_AM, "(STUBBED) called");
+    LOG_DEBUG(Service_AM, "called. library_applet_volume={}", library_applet_volume);
     IPC::ResponseBuilder rb{ctx, 3};
     rb.Push(RESULT_SUCCESS);
-    rb.Push(volume);
+    rb.Push(library_applet_volume);
+}
+
+void IAudioController::ChangeMainAppletMasterVolume(Kernel::HLERequestContext& ctx) {
+    struct Parameters {
+        float volume;
+        s64 fade_time_ns;
+    };
+    static_assert(sizeof(Parameters) == 16);
+
+    IPC::RequestParser rp{ctx};
+    const auto parameters = rp.PopRaw<Parameters>();
+
+    LOG_DEBUG(Service_AM, "called. volume={}, fade_time_ns={}", parameters.volume,
+              parameters.fade_time_ns);
+
+    main_applet_volume = std::clamp(parameters.volume, min_allowed_volume, max_allowed_volume);
+    fade_time_ns = std::chrono::nanoseconds{parameters.fade_time_ns};
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
+}
+
+void IAudioController::SetTransparentAudioRate(Kernel::HLERequestContext& ctx) {
+    IPC::RequestParser rp{ctx};
+    const float transparent_volume_rate_tmp = rp.Pop<float>();
+
+    LOG_DEBUG(Service_AM, "called. transparent_volume_rate={}", transparent_volume_rate_tmp);
+
+    // Clamp volume range to 0-100%.
+    transparent_volume_rate =
+        std::clamp(transparent_volume_rate_tmp, min_allowed_volume, max_allowed_volume);
+
+    IPC::ResponseBuilder rb{ctx, 2};
+    rb.Push(RESULT_SUCCESS);
 }
 
 IDisplayController::IDisplayController() : ServiceFramework("IDisplayController") {
@@ -169,7 +215,21 @@ IDisplayController::IDisplayController() : ServiceFramework("IDisplayController"
 
 IDisplayController::~IDisplayController() = default;
 
-IDebugFunctions::IDebugFunctions() : ServiceFramework("IDebugFunctions") {}
+IDebugFunctions::IDebugFunctions() : ServiceFramework{"IDebugFunctions"} {
+    // clang-format off
+    static const FunctionInfo functions[] = {
+        {0, nullptr, "NotifyMessageToHomeMenuForDebug"},
+        {1, nullptr, "OpenMainApplication"},
+        {10, nullptr, "EmulateButtonEvent"},
+        {20, nullptr, "InvalidateTransitionLayer"},
+        {30, nullptr, "RequestLaunchApplicationWithUserAndArgumentForDebug"},
+        {40, nullptr, "GetAppletResourceUsageInfo"},
+    };
+    // clang-format on
+
+    RegisterHandlers(functions);
+}
+
 IDebugFunctions::~IDebugFunctions() = default;
 
 ISelfController::ISelfController(std::shared_ptr<NVFlinger::NVFlinger> nvflinger)

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

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

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

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

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

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

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

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

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

@@ -41,20 +41,20 @@ private:
     struct PadState {
         union {
             u32_le raw{};
-            BitField<0, 1, u32_le> a;
-            BitField<1, 1, u32_le> b;
-            BitField<2, 1, u32_le> x;
-            BitField<3, 1, u32_le> y;
-            BitField<4, 1, u32_le> l;
-            BitField<5, 1, u32_le> r;
-            BitField<6, 1, u32_le> zl;
-            BitField<7, 1, u32_le> zr;
-            BitField<8, 1, u32_le> plus;
-            BitField<9, 1, u32_le> minus;
-            BitField<10, 1, u32_le> d_left;
-            BitField<11, 1, u32_le> d_up;
-            BitField<12, 1, u32_le> d_right;
-            BitField<13, 1, u32_le> d_down;
+            BitField<0, 1, u32> a;
+            BitField<1, 1, u32> b;
+            BitField<2, 1, u32> x;
+            BitField<3, 1, u32> y;
+            BitField<4, 1, u32> l;
+            BitField<5, 1, u32> r;
+            BitField<6, 1, u32> zl;
+            BitField<7, 1, u32> zr;
+            BitField<8, 1, u32> plus;
+            BitField<9, 1, u32> minus;
+            BitField<10, 1, u32> d_left;
+            BitField<11, 1, u32> d_up;
+            BitField<12, 1, u32> d_right;
+            BitField<13, 1, u32> d_down;
         };
     };
     static_assert(sizeof(PadState) == 0x4, "PadState is an invalid size");
@@ -62,7 +62,7 @@ private:
     struct Attributes {
         union {
             u32_le raw{};
-            BitField<0, 1, u32_le> connected;
+            BitField<0, 1, u32> connected;
         };
     };
     static_assert(sizeof(Attributes) == 0x4, "Attributes is an invalid size");

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

@@ -39,13 +39,13 @@ public:
         union {
             u32_le raw{};
 
-            BitField<0, 1, u32_le> pro_controller;
-            BitField<1, 1, u32_le> handheld;
-            BitField<2, 1, u32_le> joycon_dual;
-            BitField<3, 1, u32_le> joycon_left;
-            BitField<4, 1, u32_le> joycon_right;
+            BitField<0, 1, u32> pro_controller;
+            BitField<1, 1, u32> handheld;
+            BitField<2, 1, u32> joycon_dual;
+            BitField<3, 1, u32> joycon_left;
+            BitField<4, 1, u32> joycon_right;
 
-            BitField<6, 1, u32_le> pokeball; // TODO(ogniK): Confirm when possible
+            BitField<6, 1, u32> pokeball; // TODO(ogniK): Confirm when possible
         };
     };
     static_assert(sizeof(NPadType) == 4, "NPadType is an invalid size");
@@ -150,43 +150,43 @@ private:
         union {
             u64_le raw{};
             // Button states
-            BitField<0, 1, u64_le> a;
-            BitField<1, 1, u64_le> b;
-            BitField<2, 1, u64_le> x;
-            BitField<3, 1, u64_le> y;
-            BitField<4, 1, u64_le> l_stick;
-            BitField<5, 1, u64_le> r_stick;
-            BitField<6, 1, u64_le> l;
-            BitField<7, 1, u64_le> r;
-            BitField<8, 1, u64_le> zl;
-            BitField<9, 1, u64_le> zr;
-            BitField<10, 1, u64_le> plus;
-            BitField<11, 1, u64_le> minus;
+            BitField<0, 1, u64> a;
+            BitField<1, 1, u64> b;
+            BitField<2, 1, u64> x;
+            BitField<3, 1, u64> y;
+            BitField<4, 1, u64> l_stick;
+            BitField<5, 1, u64> r_stick;
+            BitField<6, 1, u64> l;
+            BitField<7, 1, u64> r;
+            BitField<8, 1, u64> zl;
+            BitField<9, 1, u64> zr;
+            BitField<10, 1, u64> plus;
+            BitField<11, 1, u64> minus;
 
             // D-Pad
-            BitField<12, 1, u64_le> d_left;
-            BitField<13, 1, u64_le> d_up;
-            BitField<14, 1, u64_le> d_right;
-            BitField<15, 1, u64_le> d_down;
+            BitField<12, 1, u64> d_left;
+            BitField<13, 1, u64> d_up;
+            BitField<14, 1, u64> d_right;
+            BitField<15, 1, u64> d_down;
 
             // Left JoyStick
-            BitField<16, 1, u64_le> l_stick_left;
-            BitField<17, 1, u64_le> l_stick_up;
-            BitField<18, 1, u64_le> l_stick_right;
-            BitField<19, 1, u64_le> l_stick_down;
+            BitField<16, 1, u64> l_stick_left;
+            BitField<17, 1, u64> l_stick_up;
+            BitField<18, 1, u64> l_stick_right;
+            BitField<19, 1, u64> l_stick_down;
 
             // Right JoyStick
-            BitField<20, 1, u64_le> r_stick_left;
-            BitField<21, 1, u64_le> r_stick_up;
-            BitField<22, 1, u64_le> r_stick_right;
-            BitField<23, 1, u64_le> r_stick_down;
+            BitField<20, 1, u64> r_stick_left;
+            BitField<21, 1, u64> r_stick_up;
+            BitField<22, 1, u64> r_stick_right;
+            BitField<23, 1, u64> r_stick_down;
 
             // Not always active?
-            BitField<24, 1, u64_le> left_sl;
-            BitField<25, 1, u64_le> left_sr;
+            BitField<24, 1, u64> left_sl;
+            BitField<25, 1, u64> left_sr;
 
-            BitField<26, 1, u64_le> right_sl;
-            BitField<27, 1, u64_le> right_sr;
+            BitField<26, 1, u64> right_sl;
+            BitField<27, 1, u64> right_sr;
         };
     };
     static_assert(sizeof(ControllerPadState) == 8, "ControllerPadState is an invalid size");
@@ -200,12 +200,12 @@ private:
     struct ConnectionState {
         union {
             u32_le raw{};
-            BitField<0, 1, u32_le> IsConnected;
-            BitField<1, 1, u32_le> IsWired;
-            BitField<2, 1, u32_le> IsLeftJoyConnected;
-            BitField<3, 1, u32_le> IsLeftJoyWired;
-            BitField<4, 1, u32_le> IsRightJoyConnected;
-            BitField<5, 1, u32_le> IsRightJoyWired;
+            BitField<0, 1, u32> IsConnected;
+            BitField<1, 1, u32> IsWired;
+            BitField<2, 1, u32> IsLeftJoyConnected;
+            BitField<3, 1, u32> IsLeftJoyWired;
+            BitField<4, 1, u32> IsRightJoyConnected;
+            BitField<5, 1, u32> IsRightJoyWired;
         };
     };
     static_assert(sizeof(ConnectionState) == 4, "ConnectionState is an invalid size");
@@ -240,23 +240,23 @@ private:
     struct NPadProperties {
         union {
             s64_le raw{};
-            BitField<11, 1, s64_le> is_vertical;
-            BitField<12, 1, s64_le> is_horizontal;
-            BitField<13, 1, s64_le> use_plus;
-            BitField<14, 1, s64_le> use_minus;
+            BitField<11, 1, s64> is_vertical;
+            BitField<12, 1, s64> is_horizontal;
+            BitField<13, 1, s64> use_plus;
+            BitField<14, 1, s64> use_minus;
         };
     };
 
     struct NPadDevice {
         union {
             u32_le raw{};
-            BitField<0, 1, s32_le> pro_controller;
-            BitField<1, 1, s32_le> handheld;
-            BitField<2, 1, s32_le> handheld_left;
-            BitField<3, 1, s32_le> handheld_right;
-            BitField<4, 1, s32_le> joycon_left;
-            BitField<5, 1, s32_le> joycon_right;
-            BitField<6, 1, s32_le> pokeball;
+            BitField<0, 1, s32> pro_controller;
+            BitField<1, 1, s32> handheld;
+            BitField<2, 1, s32> handheld_left;
+            BitField<3, 1, s32> handheld_right;
+            BitField<4, 1, s32> joycon_left;
+            BitField<5, 1, s32> joycon_right;
+            BitField<6, 1, s32> pokeball;
         };
     };
 

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

@@ -33,8 +33,8 @@ private:
     struct Attributes {
         union {
             u32 raw{};
-            BitField<0, 1, u32_le> start_touch;
-            BitField<1, 1, u32_le> end_touch;
+            BitField<0, 1, u32> start_touch;
+            BitField<1, 1, u32> end_touch;
         };
     };
     static_assert(sizeof(Attributes) == 0x4, "Attributes is an invalid size");

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

@@ -4,6 +4,9 @@
 
 #pragma once
 
+#include "core/hle/service/hid/controllers/controller_base.h"
+#include "core/hle/service/service.h"
+
 #include "controllers/controller_base.h"
 #include "core/hle/service/service.h"
 

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

@@ -319,15 +319,14 @@ public:
         }
 
         ASSERT(vm_manager
-                   .MirrorMemory(*map_address, nro_addr, nro_size,
-                                 Kernel::MemoryState::ModuleCodeStatic)
+                   .MirrorMemory(*map_address, nro_addr, nro_size, Kernel::MemoryState::ModuleCode)
                    .IsSuccess());
         ASSERT(vm_manager.UnmapRange(nro_addr, nro_size).IsSuccess());
 
         if (bss_size > 0) {
             ASSERT(vm_manager
                        .MirrorMemory(*map_address + nro_size, bss_addr, bss_size,
-                                     Kernel::MemoryState::ModuleCodeStatic)
+                                     Kernel::MemoryState::ModuleCode)
                        .IsSuccess());
             ASSERT(vm_manager.UnmapRange(bss_addr, bss_size).IsSuccess());
         }
@@ -388,8 +387,7 @@ public:
         const auto& nro_size = iter->second.size;
 
         ASSERT(vm_manager
-                   .MirrorMemory(heap_addr, mapped_addr, nro_size,
-                                 Kernel::MemoryState::ModuleCodeStatic)
+                   .MirrorMemory(heap_addr, mapped_addr, nro_size, Kernel::MemoryState::ModuleCode)
                    .IsSuccess());
         ASSERT(vm_manager.UnmapRange(mapped_addr, nro_size).IsSuccess());
 

src/core/hle/service/lm/lm.cpp

@@ -42,7 +42,7 @@ private:
         union {
             BitField<0, 16, Flags> flags;
             BitField<16, 8, Severity> severity;
-            BitField<24, 8, u32_le> verbosity;
+            BitField<24, 8, u32> verbosity;
         };
         u32_le payload_size;
 

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

@@ -19,11 +19,11 @@ public:
     virtual ~nvdevice() = default;
     union Ioctl {
         u32_le raw;
-        BitField<0, 8, u32_le> cmd;
-        BitField<8, 8, u32_le> group;
-        BitField<16, 14, u32_le> length;
-        BitField<30, 1, u32_le> is_in;
-        BitField<31, 1, u32_le> is_out;
+        BitField<0, 8, u32> cmd;
+        BitField<8, 8, u32> group;
+        BitField<16, 14, u32> length;
+        BitField<30, 1, u32> is_in;
+        BitField<31, 1, u32> is_out;
     };
 
     /**

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

@@ -10,6 +10,7 @@
 #include "core/core.h"
 #include "core/hle/service/nvdrv/devices/nvhost_as_gpu.h"
 #include "core/hle/service/nvdrv/devices/nvmap.h"
+#include "core/memory.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_base.h"
@@ -88,7 +89,7 @@ u32 nvhost_as_gpu::Remap(const std::vector<u8>& input, std::vector<u8>& output)
     for (const auto& entry : entries) {
         LOG_WARNING(Service_NVDRV, "remap entry, offset=0x{:X} handle=0x{:X} pages=0x{:X}",
                     entry.offset, entry.nvmap_handle, entry.pages);
-        Tegra::GPUVAddr offset = static_cast<Tegra::GPUVAddr>(entry.offset) << 0x10;
+        GPUVAddr offset = static_cast<GPUVAddr>(entry.offset) << 0x10;
         auto object = nvmap_dev->GetObject(entry.nvmap_handle);
         if (!object) {
             LOG_CRITICAL(Service_NVDRV, "nvmap {} is an invalid handle!", entry.nvmap_handle);
@@ -101,7 +102,7 @@ u32 nvhost_as_gpu::Remap(const std::vector<u8>& input, std::vector<u8>& output)
         u64 size = static_cast<u64>(entry.pages) << 0x10;
         ASSERT(size <= object->size);
 
-        Tegra::GPUVAddr returned = gpu.MemoryManager().MapBufferEx(object->addr, offset, size);
+        GPUVAddr returned = gpu.MemoryManager().MapBufferEx(object->addr, offset, size);
         ASSERT(returned == offset);
     }
     std::memcpy(output.data(), entries.data(), output.size());
@@ -172,16 +173,8 @@ u32 nvhost_as_gpu::UnmapBuffer(const std::vector<u8>& input, std::vector<u8>& ou
         return 0;
     }
 
-    auto& system_instance = Core::System::GetInstance();
-
-    // Remove this memory region from the rasterizer cache.
-    auto& gpu = system_instance.GPU();
-    auto cpu_addr = gpu.MemoryManager().GpuToCpuAddress(params.offset);
-    ASSERT(cpu_addr);
-    gpu.FlushAndInvalidateRegion(*cpu_addr, itr->second.size);
-
-    params.offset = gpu.MemoryManager().UnmapBuffer(params.offset, itr->second.size);
-
+    params.offset = Core::System::GetInstance().GPU().MemoryManager().UnmapBuffer(params.offset,
+                                                                                  itr->second.size);
     buffer_mappings.erase(itr->second.offset);
 
     std::memcpy(output.data(), &params, output.size());

src/core/loader/elf.cpp

@@ -9,6 +9,7 @@
 #include "common/common_types.h"
 #include "common/file_util.h"
 #include "common/logging/log.h"
+#include "core/hle/kernel/code_set.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/loader/elf.h"

src/core/loader/linker.cpp

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

src/core/loader/linker.h

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

src/core/loader/nro.cpp

@@ -14,6 +14,7 @@
 #include "core/file_sys/romfs_factory.h"
 #include "core/file_sys/vfs_offset.h"
 #include "core/gdbstub/gdbstub.h"
+#include "core/hle/kernel/code_set.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/hle/service/filesystem/filesystem.h"

src/core/loader/nro.h

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

src/core/loader/nso.cpp

@@ -7,10 +7,13 @@
 #include <lz4.h>
 #include "common/common_funcs.h"
 #include "common/file_util.h"
+#include "common/hex_util.h"
 #include "common/logging/log.h"
 #include "common/swap.h"
+#include "core/core.h"
 #include "core/file_sys/patch_manager.h"
 #include "core/gdbstub/gdbstub.h"
+#include "core/hle/kernel/code_set.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/vm_manager.h"
 #include "core/loader/nso.h"
@@ -164,6 +167,16 @@ std::optional<VAddr> AppLoader_NSO::LoadModule(Kernel::Process& process,
         std::memcpy(program_image.data(), pi_header.data() + 0x100, program_image.size());
     }
 
+    // Apply cheats if they exist and the program has a valid title ID
+    if (pm) {
+        auto& system = Core::System::GetInstance();
+        const auto cheats = pm->CreateCheatList(system, nso_header.build_id);
+        if (!cheats.empty()) {
+            system.RegisterCheatList(cheats, Common::HexArrayToString(nso_header.build_id),
+                                     load_base, load_base + program_image.size());
+        }
+    }
+
     // Load codeset for current process
     codeset.memory = std::make_shared<std::vector<u8>>(std::move(program_image));
     process.LoadModule(std::move(codeset), load_base);

src/core/loader/nso.h

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

src/core/memory.cpp

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

src/core/memory.h

@@ -6,11 +6,11 @@
 
 #include <cstddef>
 #include <string>
-#include <tuple>
-#include <vector>
-#include <boost/icl/interval_map.hpp>
 #include "common/common_types.h"
-#include "core/memory_hook.h"
+
+namespace Common {
+struct PageTable;
+}
 
 namespace Kernel {
 class Process;
@@ -26,71 +26,6 @@ constexpr std::size_t PAGE_BITS = 12;
 constexpr u64 PAGE_SIZE = 1ULL << PAGE_BITS;
 constexpr u64 PAGE_MASK = PAGE_SIZE - 1;
 
-enum class PageType : u8 {
-    /// Page is unmapped and should cause an access error.
-    Unmapped,
-    /// Page is mapped to regular memory. This is the only type you can get pointers to.
-    Memory,
-    /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and
-    /// invalidation
-    RasterizerCachedMemory,
-    /// Page is mapped to a I/O region. Writing and reading to this page is handled by functions.
-    Special,
-};
-
-struct SpecialRegion {
-    enum class Type {
-        DebugHook,
-        IODevice,
-    } type;
-
-    MemoryHookPointer handler;
-
-    bool operator<(const SpecialRegion& other) const {
-        return std::tie(type, handler) < std::tie(other.type, other.handler);
-    }
-
-    bool operator==(const SpecialRegion& other) const {
-        return std::tie(type, handler) == std::tie(other.type, other.handler);
-    }
-};
-
-/**
- * A (reasonably) fast way of allowing switchable and remappable process address spaces. It loosely
- * mimics the way a real CPU page table works.
- */
-struct PageTable {
-    explicit PageTable();
-    explicit PageTable(std::size_t address_space_width_in_bits);
-    ~PageTable();
-
-    /**
-     * Resizes the page table to be able to accomodate enough pages within
-     * a given address space.
-     *
-     * @param address_space_width_in_bits The address size width in bits.
-     */
-    void Resize(std::size_t address_space_width_in_bits);
-
-    /**
-     * Vector of memory pointers backing each page. An entry can only be non-null if the
-     * corresponding entry in the `attributes` vector is of type `Memory`.
-     */
-    std::vector<u8*> pointers;
-
-    /**
-     * Contains MMIO handlers that back memory regions whose entries in the `attribute` vector is
-     * of type `Special`.
-     */
-    boost::icl::interval_map<VAddr, std::set<SpecialRegion>> special_regions;
-
-    /**
-     * Vector of fine grained page attributes. If it is set to any value other than `Memory`, then
-     * the corresponding entry in `pointers` MUST be set to null.
-     */
-    std::vector<PageType> attributes;
-};
-
 /// Virtual user-space memory regions
 enum : VAddr {
     /// Read-only page containing kernel and system configuration values.
@@ -116,8 +51,8 @@ enum : VAddr {
 };
 
 /// Currently active page table
-void SetCurrentPageTable(PageTable* page_table);
-PageTable* GetCurrentPageTable();
+void SetCurrentPageTable(Common::PageTable* page_table);
+Common::PageTable* GetCurrentPageTable();
 
 /// Determines if the given VAddr is valid for the specified process.
 bool IsValidVirtualAddress(const Kernel::Process& process, VAddr vaddr);
@@ -161,10 +96,4 @@ enum class FlushMode {
  */
 void RasterizerMarkRegionCached(VAddr vaddr, u64 size, bool cached);
 
-/**
- * Flushes and invalidates any externally cached rasterizer resources touching the given virtual
- * address region.
- */
-void RasterizerFlushVirtualRegion(VAddr start, u64 size, FlushMode mode);
-
 } // namespace Memory

src/core/memory_setup.h

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

src/input_common/sdl/sdl.h

@@ -24,17 +24,19 @@ namespace InputCommon::SDL {
 
 class State {
 public:
-    /// Unresisters SDL device factories and shut them down.
+    using Pollers = std::vector<std::unique_ptr<Polling::DevicePoller>>;
+
+    /// Unregisters SDL device factories and shut them down.
     virtual ~State() = default;
 
-    virtual std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> GetPollers(
-        InputCommon::Polling::DeviceType type) = 0;
+    virtual Pollers GetPollers(Polling::DeviceType type) = 0;
 };
 
 class NullState : public State {
 public:
-    std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> GetPollers(
-        InputCommon::Polling::DeviceType type) override {}
+    Pollers GetPollers(Polling::DeviceType type) override {
+        return {};
+    }
 };
 
 std::unique_ptr<State> Init();

src/input_common/sdl/sdl_impl.cpp

@@ -475,12 +475,11 @@ SDLState::SDLState() {
 
     initialized = true;
     if (start_thread) {
-        poll_thread = std::thread([&] {
+        poll_thread = std::thread([this] {
             using namespace std::chrono_literals;
-            SDL_Event event;
             while (initialized) {
                 SDL_PumpEvents();
-                std::this_thread::sleep_for(std::chrono::duration(10ms));
+                std::this_thread::sleep_for(10ms);
             }
         });
     }
@@ -651,9 +650,9 @@ private:
 };
 } // namespace Polling
 
-std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> SDLState::GetPollers(
-    InputCommon::Polling::DeviceType type) {
-    std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> pollers;
+SDLState::Pollers SDLState::GetPollers(InputCommon::Polling::DeviceType type) {
+    Pollers pollers;
+
     switch (type) {
     case InputCommon::Polling::DeviceType::Analog:
         pollers.emplace_back(std::make_unique<Polling::SDLAnalogPoller>(*this));
@@ -661,8 +660,9 @@ std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> SDLState::GetPo
     case InputCommon::Polling::DeviceType::Button:
         pollers.emplace_back(std::make_unique<Polling::SDLButtonPoller>(*this));
         break;
-        return pollers;
     }
+
+    return pollers;
 }
 
 } // namespace SDL

src/input_common/sdl/sdl_impl.h

@@ -25,7 +25,7 @@ public:
     /// Initializes and registers SDL device factories
     SDLState();
 
-    /// Unresisters SDL device factories and shut them down.
+    /// Unregisters SDL device factories and shut them down.
     ~SDLState() override;
 
     /// Handle SDL_Events for joysticks from SDL_PollEvent
@@ -35,8 +35,7 @@ public:
     std::shared_ptr<SDLJoystick> GetSDLJoystickByGUID(const std::string& guid, int port);
 
     /// Get all DevicePoller that use the SDL backend for a specific device type
-    std::vector<std::unique_ptr<InputCommon::Polling::DevicePoller>> GetPollers(
-        InputCommon::Polling::DeviceType type) override;
+    Pollers GetPollers(Polling::DeviceType type) override;
 
     /// Used by the Pollers during config
     std::atomic<bool> polling = false;

src/tests/CMakeLists.txt

@@ -1,4 +1,5 @@
 add_executable(tests
+    common/bit_field.cpp
     common/param_package.cpp
     common/ring_buffer.cpp
     core/arm/arm_test_common.cpp

src/tests/common/bit_field.cpp

@@ -0,0 +1,90 @@
+// Copyright 2019 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <array>
+#include <cstring>
+#include <type_traits>
+#include <catch2/catch.hpp>
+#include "common/bit_field.h"
+
+TEST_CASE("BitField", "[common]") {
+    enum class TestEnum : u32 {
+        A = 0b10111101,
+        B = 0b10101110,
+        C = 0b00001111,
+    };
+
+    union LEBitField {
+        u32_le raw;
+        BitField<0, 6, u32> a;
+        BitField<6, 4, s32> b;
+        BitField<10, 8, TestEnum> c;
+        BitField<18, 14, u32> d;
+    } le_bitfield;
+
+    union BEBitField {
+        u32_be raw;
+        BitFieldBE<0, 6, u32> a;
+        BitFieldBE<6, 4, s32> b;
+        BitFieldBE<10, 8, TestEnum> c;
+        BitFieldBE<18, 14, u32> d;
+    } be_bitfield;
+
+    static_assert(sizeof(LEBitField) == sizeof(u32));
+    static_assert(sizeof(BEBitField) == sizeof(u32));
+    static_assert(std::is_trivially_copyable_v<LEBitField>);
+    static_assert(std::is_trivially_copyable_v<BEBitField>);
+
+    std::array<u8, 4> raw{{
+        0b01101100,
+        0b11110110,
+        0b10111010,
+        0b11101100,
+    }};
+
+    std::memcpy(&le_bitfield, &raw, sizeof(raw));
+    std::memcpy(&be_bitfield, &raw, sizeof(raw));
+
+    // bit fields: 11101100101110'10111101'1001'101100
+    REQUIRE(le_bitfield.raw == 0b11101100'10111010'11110110'01101100);
+    REQUIRE(le_bitfield.a == 0b101100);
+    REQUIRE(le_bitfield.b == -7); // 1001 as two's complement
+    REQUIRE(le_bitfield.c == TestEnum::A);
+    REQUIRE(le_bitfield.d == 0b11101100101110);
+
+    le_bitfield.a.Assign(0b000111);
+    le_bitfield.b.Assign(-1);
+    le_bitfield.c.Assign(TestEnum::C);
+    le_bitfield.d.Assign(0b01010101010101);
+    std::memcpy(&raw, &le_bitfield, sizeof(raw));
+    // bit fields: 01010101010101'00001111'1111'000111
+    REQUIRE(le_bitfield.raw == 0b01010101'01010100'00111111'11000111);
+    REQUIRE(raw == std::array<u8, 4>{{
+                       0b11000111,
+                       0b00111111,
+                       0b01010100,
+                       0b01010101,
+                   }});
+
+    // bit fields: 01101100111101'10101110'1011'101100
+    REQUIRE(be_bitfield.raw == 0b01101100'11110110'10111010'11101100);
+    REQUIRE(be_bitfield.a == 0b101100);
+    REQUIRE(be_bitfield.b == -5); // 1011 as two's complement
+    REQUIRE(be_bitfield.c == TestEnum::B);
+    REQUIRE(be_bitfield.d == 0b01101100111101);
+
+    be_bitfield.a.Assign(0b000111);
+    be_bitfield.b.Assign(-1);
+    be_bitfield.c.Assign(TestEnum::C);
+    be_bitfield.d.Assign(0b01010101010101);
+    std::memcpy(&raw, &be_bitfield, sizeof(raw));
+    // bit fields: 01010101010101'00001111'1111'000111
+    REQUIRE(be_bitfield.raw == 0b01010101'01010100'00111111'11000111);
+    REQUIRE(raw == std::array<u8, 4>{{
+                       0b01010101,
+                       0b01010100,
+                       0b00111111,
+                       0b11000111,
+                   }});
+}

src/tests/core/arm/arm_test_common.cpp

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

src/tests/core/arm/arm_test_common.h

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

src/video_core/CMakeLists.txt

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

src/video_core/dma_pusher.cpp

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

src/video_core/dma_pusher.h

@@ -9,7 +9,6 @@
 
 #include "common/bit_field.h"
 #include "common/common_types.h"
-#include "video_core/memory_manager.h"
 
 namespace Tegra {
 

src/video_core/engines/kepler_memory.cpp

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

src/video_core/engines/maxwell_3d.cpp

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

src/video_core/engines/maxwell_dma.cpp

@@ -9,6 +9,7 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/maxwell_dma.h"
 #include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_base.h"
 #include "video_core/textures/decoders.h"
 
 namespace Tegra::Engines {
@@ -42,11 +43,6 @@ void MaxwellDMA::HandleCopy() {
     const GPUVAddr source = regs.src_address.Address();
     const GPUVAddr dest = regs.dst_address.Address();
 
-    const auto source_cpu = memory_manager.GpuToCpuAddress(source);
-    const auto dest_cpu = memory_manager.GpuToCpuAddress(dest);
-    ASSERT_MSG(source_cpu, "Invalid source GPU address");
-    ASSERT_MSG(dest_cpu, "Invalid destination GPU address");
-
     // TODO(Subv): Perform more research and implement all features of this engine.
     ASSERT(regs.exec.enable_swizzle == 0);
     ASSERT(regs.exec.query_mode == Regs::QueryMode::None);
@@ -69,7 +65,7 @@ void MaxwellDMA::HandleCopy() {
         // buffer of length `x_count`, otherwise we copy a 2D image of dimensions (x_count,
         // y_count).
         if (!regs.exec.enable_2d) {
-            Memory::CopyBlock(*dest_cpu, *source_cpu, regs.x_count);
+            memory_manager.CopyBlock(dest, source, regs.x_count);
             return;
         }
 
@@ -78,9 +74,9 @@ void MaxwellDMA::HandleCopy() {
         // rectangle. There is no need to manually flush/invalidate the regions because
         // CopyBlock does that for us.
         for (u32 line = 0; line < regs.y_count; ++line) {
-            const VAddr source_line = *source_cpu + line * regs.src_pitch;
-            const VAddr dest_line = *dest_cpu + line * regs.dst_pitch;
-            Memory::CopyBlock(dest_line, source_line, regs.x_count);
+            const GPUVAddr source_line = source + line * regs.src_pitch;
+            const GPUVAddr dest_line = dest + line * regs.dst_pitch;
+            memory_manager.CopyBlock(dest_line, source_line, regs.x_count);
         }
         return;
     }
@@ -89,15 +85,28 @@ void MaxwellDMA::HandleCopy() {
 
     const std::size_t copy_size = regs.x_count * regs.y_count;
 
+    auto source_ptr{memory_manager.GetPointer(source)};
+    auto dst_ptr{memory_manager.GetPointer(dest)};
+
+    if (!source_ptr) {
+        LOG_ERROR(HW_GPU, "source_ptr is invalid");
+        return;
+    }
+
+    if (!dst_ptr) {
+        LOG_ERROR(HW_GPU, "dst_ptr is invalid");
+        return;
+    }
+
     const auto FlushAndInvalidate = [&](u32 src_size, u64 dst_size) {
         // TODO(Subv): For now, manually flush the regions until we implement GPU-accelerated
         // copying.
-        Core::System::GetInstance().GPU().FlushRegion(*source_cpu, src_size);
+        rasterizer.FlushRegion(ToCacheAddr(source_ptr), src_size);
 
         // We have to invalidate the destination region to evict any outdated surfaces from the
         // cache. We do this before actually writing the new data because the destination address
         // might contain a dirty surface that will have to be written back to memory.
-        Core::System::GetInstance().GPU().InvalidateRegion(*dest_cpu, dst_size);
+        rasterizer.InvalidateRegion(ToCacheAddr(dst_ptr), dst_size);
     };
 
     if (regs.exec.is_dst_linear && !regs.exec.is_src_linear) {
@@ -110,8 +119,8 @@ void MaxwellDMA::HandleCopy() {
                            copy_size * src_bytes_per_pixel);
 
         Texture::UnswizzleSubrect(regs.x_count, regs.y_count, regs.dst_pitch,
-                                  regs.src_params.size_x, src_bytes_per_pixel, *source_cpu,
-                                  *dest_cpu, regs.src_params.BlockHeight(), regs.src_params.pos_x,
+                                  regs.src_params.size_x, src_bytes_per_pixel, source_ptr, dst_ptr,
+                                  regs.src_params.BlockHeight(), regs.src_params.pos_x,
                                   regs.src_params.pos_y);
     } else {
         ASSERT(regs.dst_params.size_z == 1);
@@ -124,7 +133,7 @@ void MaxwellDMA::HandleCopy() {
 
         // If the input is linear and the output is tiled, swizzle the input and copy it over.
         Texture::SwizzleSubrect(regs.x_count, regs.y_count, regs.src_pitch, regs.dst_params.size_x,
-                                src_bpp, *dest_cpu, *source_cpu, regs.dst_params.BlockHeight());
+                                src_bpp, dst_ptr, source_ptr, regs.dst_params.BlockHeight());
     }
 }
 

src/video_core/gpu.cpp

@@ -12,6 +12,7 @@
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/engines/maxwell_dma.h"
 #include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
 #include "video_core/renderer_base.h"
 
 namespace Tegra {
@@ -274,7 +275,6 @@ void GPU::ProcessSemaphoreTriggerMethod() {
     const auto op =
         static_cast<GpuSemaphoreOperation>(regs.semaphore_trigger & semaphoreOperationMask);
     if (op == GpuSemaphoreOperation::WriteLong) {
-        auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
         struct Block {
             u32 sequence;
             u32 zeros = 0;
@@ -286,11 +286,9 @@ void GPU::ProcessSemaphoreTriggerMethod() {
         // TODO(Kmather73): Generate a real GPU timestamp and write it here instead of
         // CoreTiming
         block.timestamp = Core::System::GetInstance().CoreTiming().GetTicks();
-        Memory::WriteBlock(*address, &block, sizeof(block));
+        memory_manager->WriteBlock(regs.smaphore_address.SmaphoreAddress(), &block, sizeof(block));
     } else {
-        const auto address =
-            memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
-        const u32 word = Memory::Read32(*address);
+        const u32 word{memory_manager->Read<u32>(regs.smaphore_address.SmaphoreAddress())};
         if ((op == GpuSemaphoreOperation::AcquireEqual && word == regs.semaphore_sequence) ||
             (op == GpuSemaphoreOperation::AcquireGequal &&
              static_cast<s32>(word - regs.semaphore_sequence) > 0) ||
@@ -317,13 +315,11 @@ void GPU::ProcessSemaphoreTriggerMethod() {
 }
 
 void GPU::ProcessSemaphoreRelease() {
-    const auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
-    Memory::Write32(*address, regs.semaphore_release);
+    memory_manager->Write<u32>(regs.smaphore_address.SmaphoreAddress(), regs.semaphore_release);
 }
 
 void GPU::ProcessSemaphoreAcquire() {
-    const auto address = memory_manager->GpuToCpuAddress(regs.smaphore_address.SmaphoreAddress());
-    const u32 word = Memory::Read32(*address);
+    const u32 word = memory_manager->Read<u32>(regs.smaphore_address.SmaphoreAddress());
     const auto value = regs.semaphore_acquire;
     if (word != value) {
         regs.acquire_active = true;

src/video_core/gpu.h

@@ -9,7 +9,11 @@
 #include "common/common_types.h"
 #include "core/hle/service/nvflinger/buffer_queue.h"
 #include "video_core/dma_pusher.h"
-#include "video_core/memory_manager.h"
+
+using CacheAddr = std::uintptr_t;
+inline CacheAddr ToCacheAddr(const void* host_ptr) {
+    return reinterpret_cast<CacheAddr>(host_ptr);
+}
 
 namespace Core {
 class System;
@@ -119,6 +123,8 @@ enum class EngineID {
     MAXWELL_DMA_COPY_A = 0xB0B5,
 };
 
+class MemoryManager;
+
 class GPU {
 public:
     explicit GPU(Core::System& system, VideoCore::RendererBase& renderer);
@@ -209,13 +215,13 @@ public:
         std::optional<std::reference_wrapper<const Tegra::FramebufferConfig>> framebuffer) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
-    virtual void FlushRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushRegion(CacheAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be invalidated
-    virtual void InvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void InvalidateRegion(CacheAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed and invalidated
-    virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
 
 private:
     void ProcessBindMethod(const MethodCall& method_call);
@@ -239,9 +245,8 @@ protected:
 private:
     std::unique_ptr<Tegra::MemoryManager> memory_manager;
 
-    /// Mapping of command subchannels to their bound engine ids.
+    /// Mapping of command subchannels to their bound engine ids
     std::array<EngineID, 8> bound_engines = {};
-
     /// 3D engine
     std::unique_ptr<Engines::Maxwell3D> maxwell_3d;
     /// 2D engine

src/video_core/gpu_asynch.cpp

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

src/video_core/gpu_asynch.h

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

src/video_core/gpu_synch.cpp

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

src/video_core/gpu_synch.h

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

src/video_core/gpu_thread.cpp

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

src/video_core/gpu_thread.h

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

src/video_core/memory_manager.cpp

@@ -5,181 +5,446 @@
 #include "common/alignment.h"
 #include "common/assert.h"
 #include "common/logging/log.h"
+#include "core/core.h"
+#include "core/memory.h"
+#include "video_core/gpu.h"
 #include "video_core/memory_manager.h"
+#include "video_core/rasterizer_interface.h"
+#include "video_core/renderer_base.h"
 
 namespace Tegra {
 
 MemoryManager::MemoryManager() {
-    // Mark the first page as reserved, so that 0 is not a valid GPUVAddr. Otherwise, games might
-    // try to use 0 as a valid address, which is also used to mean nullptr. This fixes a bug with
-    // Undertale using 0 for a render target.
-    PageSlot(0) = static_cast<u64>(PageStatus::Reserved);
+    std::fill(page_table.pointers.begin(), page_table.pointers.end(), nullptr);
+    std::fill(page_table.attributes.begin(), page_table.attributes.end(),
+              Common::PageType::Unmapped);
+    page_table.Resize(address_space_width);
+
+    // Initialize the map with a single free region covering the entire managed space.
+    VirtualMemoryArea initial_vma;
+    initial_vma.size = address_space_end;
+    vma_map.emplace(initial_vma.base, initial_vma);
+
+    UpdatePageTableForVMA(initial_vma);
 }
 
 GPUVAddr MemoryManager::AllocateSpace(u64 size, u64 align) {
-    const std::optional<GPUVAddr> gpu_addr{FindFreeBlock(0, size, align, PageStatus::Unmapped)};
+    const u64 aligned_size{Common::AlignUp(size, page_size)};
+    const GPUVAddr gpu_addr{FindFreeRegion(address_space_base, aligned_size)};
 
-    ASSERT_MSG(gpu_addr, "unable to find available GPU memory");
+    AllocateMemory(gpu_addr, 0, aligned_size);
 
-    for (u64 offset{}; offset < size; offset += PAGE_SIZE) {
-        VAddr& slot{PageSlot(*gpu_addr + offset)};
-
-        ASSERT(slot == static_cast<u64>(PageStatus::Unmapped));
-
-        slot = static_cast<u64>(PageStatus::Allocated);
-    }
-
-    return *gpu_addr;
+    return gpu_addr;
 }
 
 GPUVAddr MemoryManager::AllocateSpace(GPUVAddr gpu_addr, u64 size, u64 align) {
-    for (u64 offset{}; offset < size; offset += PAGE_SIZE) {
-        VAddr& slot{PageSlot(gpu_addr + offset)};
+    const u64 aligned_size{Common::AlignUp(size, page_size)};
 
-        ASSERT(slot == static_cast<u64>(PageStatus::Unmapped));
-
-        slot = static_cast<u64>(PageStatus::Allocated);
-    }
+    AllocateMemory(gpu_addr, 0, aligned_size);
 
     return gpu_addr;
 }
 
 GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, u64 size) {
-    const std::optional<GPUVAddr> gpu_addr{FindFreeBlock(0, size, PAGE_SIZE, PageStatus::Unmapped)};
+    const u64 aligned_size{Common::AlignUp(size, page_size)};
+    const GPUVAddr gpu_addr{FindFreeRegion(address_space_base, aligned_size)};
 
-    ASSERT_MSG(gpu_addr, "unable to find available GPU memory");
+    MapBackingMemory(gpu_addr, Memory::GetPointer(cpu_addr), aligned_size, cpu_addr);
 
-    for (u64 offset{}; offset < size; offset += PAGE_SIZE) {
-        VAddr& slot{PageSlot(*gpu_addr + offset)};
-
-        ASSERT(slot == static_cast<u64>(PageStatus::Unmapped));
-
-        slot = cpu_addr + offset;
-    }
-
-    const MappedRegion region{cpu_addr, *gpu_addr, size};
-    mapped_regions.push_back(region);
-
-    return *gpu_addr;
+    return gpu_addr;
 }
 
 GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size) {
-    ASSERT((gpu_addr & PAGE_MASK) == 0);
+    ASSERT((gpu_addr & page_mask) == 0);
 
-    if (PageSlot(gpu_addr) != static_cast<u64>(PageStatus::Allocated)) {
-        // Page has been already mapped. In this case, we must find a new area of memory to use that
-        // is different than the specified one. Super Mario Odyssey hits this scenario when changing
-        // areas, but we do not want to overwrite the old pages.
-        // TODO(bunnei): We need to write a hardware test to confirm this behavior.
+    const u64 aligned_size{Common::AlignUp(size, page_size)};
 
-        LOG_ERROR(HW_GPU, "attempting to map addr 0x{:016X}, which is not available!", gpu_addr);
-
-        const std::optional<GPUVAddr> new_gpu_addr{
-            FindFreeBlock(gpu_addr, size, PAGE_SIZE, PageStatus::Allocated)};
-
-        ASSERT_MSG(new_gpu_addr, "unable to find available GPU memory");
-
-        gpu_addr = *new_gpu_addr;
-    }
-
-    for (u64 offset{}; offset < size; offset += PAGE_SIZE) {
-        VAddr& slot{PageSlot(gpu_addr + offset)};
-
-        ASSERT(slot == static_cast<u64>(PageStatus::Allocated));
-
-        slot = cpu_addr + offset;
-    }
-
-    const MappedRegion region{cpu_addr, gpu_addr, size};
-    mapped_regions.push_back(region);
+    MapBackingMemory(gpu_addr, Memory::GetPointer(cpu_addr), aligned_size, cpu_addr);
 
     return gpu_addr;
 }
 
 GPUVAddr MemoryManager::UnmapBuffer(GPUVAddr gpu_addr, u64 size) {
-    ASSERT((gpu_addr & PAGE_MASK) == 0);
+    ASSERT((gpu_addr & page_mask) == 0);
 
-    for (u64 offset{}; offset < size; offset += PAGE_SIZE) {
-        VAddr& slot{PageSlot(gpu_addr + offset)};
+    const u64 aligned_size{Common::AlignUp(size, page_size)};
+    const CacheAddr cache_addr{ToCacheAddr(GetPointer(gpu_addr))};
 
-        ASSERT(slot != static_cast<u64>(PageStatus::Allocated) &&
-               slot != static_cast<u64>(PageStatus::Unmapped));
+    Core::System::GetInstance().Renderer().Rasterizer().FlushAndInvalidateRegion(cache_addr,
+                                                                                 aligned_size);
+    UnmapRange(gpu_addr, aligned_size);
 
-        slot = static_cast<u64>(PageStatus::Unmapped);
-    }
-
-    // Delete the region mappings that are contained within the unmapped region
-    mapped_regions.erase(std::remove_if(mapped_regions.begin(), mapped_regions.end(),
-                                        [&](const MappedRegion& region) {
-                                            return region.gpu_addr <= gpu_addr &&
-                                                   region.gpu_addr + region.size < gpu_addr + size;
-                                        }),
-                         mapped_regions.end());
     return gpu_addr;
 }
 
-GPUVAddr MemoryManager::GetRegionEnd(GPUVAddr region_start) const {
-    for (const auto& region : mapped_regions) {
-        const GPUVAddr region_end{region.gpu_addr + region.size};
-        if (region_start >= region.gpu_addr && region_start < region_end) {
-            return region_end;
+GPUVAddr MemoryManager::FindFreeRegion(GPUVAddr region_start, u64 size) {
+    // Find the first Free VMA.
+    const VMAHandle vma_handle{std::find_if(vma_map.begin(), vma_map.end(), [&](const auto& vma) {
+        if (vma.second.type != VirtualMemoryArea::Type::Unmapped) {
+            return false;
         }
-    }
-    return {};
-}
 
-std::optional<GPUVAddr> MemoryManager::FindFreeBlock(GPUVAddr region_start, u64 size, u64 align,
-                                                     PageStatus status) {
-    GPUVAddr gpu_addr{region_start};
-    u64 free_space{};
-    align = (align + PAGE_MASK) & ~PAGE_MASK;
+        const VAddr vma_end{vma.second.base + vma.second.size};
+        return vma_end > region_start && vma_end >= region_start + size;
+    })};
 
-    while (gpu_addr + free_space < MAX_ADDRESS) {
-        if (PageSlot(gpu_addr + free_space) == static_cast<u64>(status)) {
-            free_space += PAGE_SIZE;
-            if (free_space >= size) {
-                return gpu_addr;
-            }
-        } else {
-            gpu_addr += free_space + PAGE_SIZE;
-            free_space = 0;
-            gpu_addr = Common::AlignUp(gpu_addr, align);
-        }
-    }
-
-    return {};
-}
-
-std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) {
-    const VAddr base_addr{PageSlot(gpu_addr)};
-
-    if (base_addr == static_cast<u64>(PageStatus::Allocated) ||
-        base_addr == static_cast<u64>(PageStatus::Unmapped) ||
-        base_addr == static_cast<u64>(PageStatus::Reserved)) {
+    if (vma_handle == vma_map.end()) {
         return {};
     }
 
-    return base_addr + (gpu_addr & PAGE_MASK);
+    return std::max(region_start, vma_handle->second.base);
 }
 
-std::vector<GPUVAddr> MemoryManager::CpuToGpuAddress(VAddr cpu_addr) const {
-    std::vector<GPUVAddr> results;
-    for (const auto& region : mapped_regions) {
-        if (cpu_addr >= region.cpu_addr && cpu_addr < (region.cpu_addr + region.size)) {
-            const u64 offset{cpu_addr - region.cpu_addr};
-            results.push_back(region.gpu_addr + offset);
+bool MemoryManager::IsAddressValid(GPUVAddr addr) const {
+    return (addr >> page_bits) < page_table.pointers.size();
+}
+
+std::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr addr) {
+    if (!IsAddressValid(addr)) {
+        return {};
+    }
+
+    VAddr cpu_addr{page_table.backing_addr[addr >> page_bits]};
+    if (cpu_addr) {
+        return cpu_addr + (addr & page_mask);
+    }
+
+    return {};
+}
+
+template <typename T>
+T MemoryManager::Read(GPUVAddr addr) {
+    if (!IsAddressValid(addr)) {
+        return {};
+    }
+
+    const u8* page_pointer{page_table.pointers[addr >> page_bits]};
+    if (page_pointer) {
+        // NOTE: Avoid adding any extra logic to this fast-path block
+        T value;
+        std::memcpy(&value, &page_pointer[addr & page_mask], sizeof(T));
+        return value;
+    }
+
+    switch (page_table.attributes[addr >> page_bits]) {
+    case Common::PageType::Unmapped:
+        LOG_ERROR(HW_GPU, "Unmapped Read{} @ 0x{:08X}", sizeof(T) * 8, addr);
+        return 0;
+    case Common::PageType::Memory:
+        ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", addr);
+        break;
+    default:
+        UNREACHABLE();
+    }
+    return {};
+}
+
+template <typename T>
+void MemoryManager::Write(GPUVAddr addr, T data) {
+    if (!IsAddressValid(addr)) {
+        return;
+    }
+
+    u8* page_pointer{page_table.pointers[addr >> page_bits]};
+    if (page_pointer) {
+        // NOTE: Avoid adding any extra logic to this fast-path block
+        std::memcpy(&page_pointer[addr & page_mask], &data, sizeof(T));
+        return;
+    }
+
+    switch (page_table.attributes[addr >> page_bits]) {
+    case Common::PageType::Unmapped:
+        LOG_ERROR(HW_GPU, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8,
+                  static_cast<u32>(data), addr);
+        return;
+    case Common::PageType::Memory:
+        ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", addr);
+        break;
+    default:
+        UNREACHABLE();
+    }
+}
+
+template u8 MemoryManager::Read<u8>(GPUVAddr addr);
+template u16 MemoryManager::Read<u16>(GPUVAddr addr);
+template u32 MemoryManager::Read<u32>(GPUVAddr addr);
+template u64 MemoryManager::Read<u64>(GPUVAddr addr);
+template void MemoryManager::Write<u8>(GPUVAddr addr, u8 data);
+template void MemoryManager::Write<u16>(GPUVAddr addr, u16 data);
+template void MemoryManager::Write<u32>(GPUVAddr addr, u32 data);
+template void MemoryManager::Write<u64>(GPUVAddr addr, u64 data);
+
+u8* MemoryManager::GetPointer(GPUVAddr addr) {
+    if (!IsAddressValid(addr)) {
+        return {};
+    }
+
+    u8* page_pointer{page_table.pointers[addr >> page_bits]};
+    if (page_pointer) {
+        return page_pointer + (addr & page_mask);
+    }
+
+    LOG_ERROR(HW_GPU, "Unknown GetPointer @ 0x{:016X}", addr);
+    return {};
+}
+
+void MemoryManager::ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size) {
+    std::memcpy(dest_buffer, GetPointer(src_addr), size);
+}
+void MemoryManager::WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size) {
+    std::memcpy(GetPointer(dest_addr), src_buffer, size);
+}
+
+void MemoryManager::CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, std::size_t size) {
+    std::memcpy(GetPointer(dest_addr), GetPointer(src_addr), size);
+}
+
+void MemoryManager::MapPages(GPUVAddr base, u64 size, u8* memory, Common::PageType type,
+                             VAddr backing_addr) {
+    LOG_DEBUG(HW_GPU, "Mapping {} onto {:016X}-{:016X}", fmt::ptr(memory), base * page_size,
+              (base + size) * page_size);
+
+    const VAddr end{base + size};
+    ASSERT_MSG(end <= page_table.pointers.size(), "out of range mapping at {:016X}",
+               base + page_table.pointers.size());
+
+    std::fill(page_table.attributes.begin() + base, page_table.attributes.begin() + end, type);
+
+    if (memory == nullptr) {
+        std::fill(page_table.pointers.begin() + base, page_table.pointers.begin() + end, memory);
+        std::fill(page_table.backing_addr.begin() + base, page_table.backing_addr.begin() + end,
+                  backing_addr);
+    } else {
+        while (base != end) {
+            page_table.pointers[base] = memory;
+            page_table.backing_addr[base] = backing_addr;
+
+            base += 1;
+            memory += page_size;
+            backing_addr += page_size;
         }
     }
-    return results;
 }
 
-VAddr& MemoryManager::PageSlot(GPUVAddr gpu_addr) {
-    auto& block{page_table[(gpu_addr >> (PAGE_BITS + PAGE_TABLE_BITS)) & PAGE_TABLE_MASK]};
-    if (!block) {
-        block = std::make_unique<PageBlock>();
-        block->fill(static_cast<VAddr>(PageStatus::Unmapped));
+void MemoryManager::MapMemoryRegion(GPUVAddr base, u64 size, u8* target, VAddr backing_addr) {
+    ASSERT_MSG((size & page_mask) == 0, "non-page aligned size: {:016X}", size);
+    ASSERT_MSG((base & page_mask) == 0, "non-page aligned base: {:016X}", base);
+    MapPages(base / page_size, size / page_size, target, Common::PageType::Memory, backing_addr);
+}
+
+void MemoryManager::UnmapRegion(GPUVAddr base, u64 size) {
+    ASSERT_MSG((size & page_mask) == 0, "non-page aligned size: {:016X}", size);
+    ASSERT_MSG((base & page_mask) == 0, "non-page aligned base: {:016X}", base);
+    MapPages(base / page_size, size / page_size, nullptr, Common::PageType::Unmapped);
+}
+
+bool VirtualMemoryArea::CanBeMergedWith(const VirtualMemoryArea& next) const {
+    ASSERT(base + size == next.base);
+    if (type != next.type) {
+        return {};
+    }
+    if (type == VirtualMemoryArea::Type::Allocated && (offset + size != next.offset)) {
+        return {};
+    }
+    if (type == VirtualMemoryArea::Type::Mapped && backing_memory + size != next.backing_memory) {
+        return {};
+    }
+    return true;
+}
+
+MemoryManager::VMAHandle MemoryManager::FindVMA(GPUVAddr target) const {
+    if (target >= address_space_end) {
+        return vma_map.end();
+    } else {
+        return std::prev(vma_map.upper_bound(target));
+    }
+}
+
+MemoryManager::VMAIter MemoryManager::Allocate(VMAIter vma_handle) {
+    VirtualMemoryArea& vma{vma_handle->second};
+
+    vma.type = VirtualMemoryArea::Type::Allocated;
+    vma.backing_addr = 0;
+    vma.backing_memory = {};
+    UpdatePageTableForVMA(vma);
+
+    return MergeAdjacent(vma_handle);
+}
+
+MemoryManager::VMAHandle MemoryManager::AllocateMemory(GPUVAddr target, std::size_t offset,
+                                                       u64 size) {
+
+    // This is the appropriately sized VMA that will turn into our allocation.
+    VMAIter vma_handle{CarveVMA(target, size)};
+    VirtualMemoryArea& vma{vma_handle->second};
+
+    ASSERT(vma.size == size);
+
+    vma.offset = offset;
+
+    return Allocate(vma_handle);
+}
+
+MemoryManager::VMAHandle MemoryManager::MapBackingMemory(GPUVAddr target, u8* memory, u64 size,
+                                                         VAddr backing_addr) {
+    // This is the appropriately sized VMA that will turn into our allocation.
+    VMAIter vma_handle{CarveVMA(target, size)};
+    VirtualMemoryArea& vma{vma_handle->second};
+
+    ASSERT(vma.size == size);
+
+    vma.type = VirtualMemoryArea::Type::Mapped;
+    vma.backing_memory = memory;
+    vma.backing_addr = backing_addr;
+    UpdatePageTableForVMA(vma);
+
+    return MergeAdjacent(vma_handle);
+}
+
+void MemoryManager::UnmapRange(GPUVAddr target, u64 size) {
+    VMAIter vma{CarveVMARange(target, size)};
+    const VAddr target_end{target + size};
+    const VMAIter end{vma_map.end()};
+
+    // The comparison against the end of the range must be done using addresses since VMAs can be
+    // merged during this process, causing invalidation of the iterators.
+    while (vma != end && vma->second.base < target_end) {
+        // Unmapped ranges return to allocated state and can be reused
+        // This behavior is used by Super Mario Odyssey, Sonic Forces, and likely other games
+        vma = std::next(Allocate(vma));
+    }
+
+    ASSERT(FindVMA(target)->second.size >= size);
+}
+
+MemoryManager::VMAIter MemoryManager::StripIterConstness(const VMAHandle& iter) {
+    // This uses a neat C++ trick to convert a const_iterator to a regular iterator, given
+    // non-const access to its container.
+    return vma_map.erase(iter, iter); // Erases an empty range of elements
+}
+
+MemoryManager::VMAIter MemoryManager::CarveVMA(GPUVAddr base, u64 size) {
+    ASSERT_MSG((size & page_mask) == 0, "non-page aligned size: 0x{:016X}", size);
+    ASSERT_MSG((base & page_mask) == 0, "non-page aligned base: 0x{:016X}", base);
+
+    VMAIter vma_handle{StripIterConstness(FindVMA(base))};
+    if (vma_handle == vma_map.end()) {
+        // Target address is outside the managed range
+        return {};
+    }
+
+    const VirtualMemoryArea& vma{vma_handle->second};
+    if (vma.type == VirtualMemoryArea::Type::Mapped) {
+        // Region is already allocated
+        return {};
+    }
+
+    const VAddr start_in_vma{base - vma.base};
+    const VAddr end_in_vma{start_in_vma + size};
+
+    ASSERT_MSG(end_in_vma <= vma.size, "region size 0x{:016X} is less than required size 0x{:016X}",
+               vma.size, end_in_vma);
+
+    if (end_in_vma < vma.size) {
+        // Split VMA at the end of the allocated region
+        SplitVMA(vma_handle, end_in_vma);
+    }
+    if (start_in_vma != 0) {
+        // Split VMA at the start of the allocated region
+        vma_handle = SplitVMA(vma_handle, start_in_vma);
+    }
+
+    return vma_handle;
+}
+
+MemoryManager::VMAIter MemoryManager::CarveVMARange(GPUVAddr target, u64 size) {
+    ASSERT_MSG((size & page_mask) == 0, "non-page aligned size: 0x{:016X}", size);
+    ASSERT_MSG((target & page_mask) == 0, "non-page aligned base: 0x{:016X}", target);
+
+    const VAddr target_end{target + size};
+    ASSERT(target_end >= target);
+    ASSERT(size > 0);
+
+    VMAIter begin_vma{StripIterConstness(FindVMA(target))};
+    const VMAIter i_end{vma_map.lower_bound(target_end)};
+    if (std::any_of(begin_vma, i_end, [](const auto& entry) {
+            return entry.second.type == VirtualMemoryArea::Type::Unmapped;
+        })) {
+        return {};
+    }
+
+    if (target != begin_vma->second.base) {
+        begin_vma = SplitVMA(begin_vma, target - begin_vma->second.base);
+    }
+
+    VMAIter end_vma{StripIterConstness(FindVMA(target_end))};
+    if (end_vma != vma_map.end() && target_end != end_vma->second.base) {
+        end_vma = SplitVMA(end_vma, target_end - end_vma->second.base);
+    }
+
+    return begin_vma;
+}
+
+MemoryManager::VMAIter MemoryManager::SplitVMA(VMAIter vma_handle, u64 offset_in_vma) {
+    VirtualMemoryArea& old_vma{vma_handle->second};
+    VirtualMemoryArea new_vma{old_vma}; // Make a copy of the VMA
+
+    // For now, don't allow no-op VMA splits (trying to split at a boundary) because it's probably
+    // a bug. This restriction might be removed later.
+    ASSERT(offset_in_vma < old_vma.size);
+    ASSERT(offset_in_vma > 0);
+
+    old_vma.size = offset_in_vma;
+    new_vma.base += offset_in_vma;
+    new_vma.size -= offset_in_vma;
+
+    switch (new_vma.type) {
+    case VirtualMemoryArea::Type::Unmapped:
+        break;
+    case VirtualMemoryArea::Type::Allocated:
+        new_vma.offset += offset_in_vma;
+        break;
+    case VirtualMemoryArea::Type::Mapped:
+        new_vma.backing_memory += offset_in_vma;
+        break;
+    }
+
+    ASSERT(old_vma.CanBeMergedWith(new_vma));
+
+    return vma_map.emplace_hint(std::next(vma_handle), new_vma.base, new_vma);
+}
+
+MemoryManager::VMAIter MemoryManager::MergeAdjacent(VMAIter iter) {
+    const VMAIter next_vma{std::next(iter)};
+    if (next_vma != vma_map.end() && iter->second.CanBeMergedWith(next_vma->second)) {
+        iter->second.size += next_vma->second.size;
+        vma_map.erase(next_vma);
+    }
+
+    if (iter != vma_map.begin()) {
+        VMAIter prev_vma{std::prev(iter)};
+        if (prev_vma->second.CanBeMergedWith(iter->second)) {
+            prev_vma->second.size += iter->second.size;
+            vma_map.erase(iter);
+            iter = prev_vma;
+        }
+    }
+
+    return iter;
+}
+
+void MemoryManager::UpdatePageTableForVMA(const VirtualMemoryArea& vma) {
+    switch (vma.type) {
+    case VirtualMemoryArea::Type::Unmapped:
+        UnmapRegion(vma.base, vma.size);
+        break;
+    case VirtualMemoryArea::Type::Allocated:
+        MapMemoryRegion(vma.base, vma.size, nullptr, vma.backing_addr);
+        break;
+    case VirtualMemoryArea::Type::Mapped:
+        MapMemoryRegion(vma.base, vma.size, vma.backing_memory, vma.backing_addr);
+        break;
     }
-    return (*block)[(gpu_addr >> PAGE_BITS) & PAGE_BLOCK_MASK];
 }
 
 } // namespace Tegra

src/video_core/memory_manager.h

@@ -1,67 +1,148 @@
-// Copyright 2018 yuzu emulator team
+// Copyright 2018 yuzu emulator team
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
 #pragma once
 
-#include <array>
-#include <memory>
+#include <map>
 #include <optional>
-#include <vector>
 
 #include "common/common_types.h"
+#include "common/page_table.h"
 
 namespace Tegra {
 
-/// Virtual addresses in the GPU's memory map are 64 bit.
-using GPUVAddr = u64;
+/**
+ * Represents a VMA in an address space. A VMA is a contiguous region of virtual addressing space
+ * with homogeneous attributes across its extents. In this particular implementation each VMA is
+ * also backed by a single host memory allocation.
+ */
+struct VirtualMemoryArea {
+    enum class Type : u8 {
+        Unmapped,
+        Allocated,
+        Mapped,
+    };
+
+    /// Virtual base address of the region.
+    GPUVAddr base{};
+    /// Size of the region.
+    u64 size{};
+    /// Memory area mapping type.
+    Type type{Type::Unmapped};
+    /// CPU memory mapped address corresponding to this memory area.
+    VAddr backing_addr{};
+    /// Offset into the backing_memory the mapping starts from.
+    std::size_t offset{};
+    /// Pointer backing this VMA.
+    u8* backing_memory{};
+
+    /// Tests if this area can be merged to the right with `next`.
+    bool CanBeMergedWith(const VirtualMemoryArea& next) const;
+};
 
 class MemoryManager final {
 public:
     MemoryManager();
 
     GPUVAddr AllocateSpace(u64 size, u64 align);
-    GPUVAddr AllocateSpace(GPUVAddr gpu_addr, u64 size, u64 align);
+    GPUVAddr AllocateSpace(GPUVAddr addr, u64 size, u64 align);
     GPUVAddr MapBufferEx(VAddr cpu_addr, u64 size);
-    GPUVAddr MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size);
-    GPUVAddr UnmapBuffer(GPUVAddr gpu_addr, u64 size);
-    GPUVAddr GetRegionEnd(GPUVAddr region_start) const;
-    std::optional<VAddr> GpuToCpuAddress(GPUVAddr gpu_addr);
-    std::vector<GPUVAddr> CpuToGpuAddress(VAddr cpu_addr) const;
+    GPUVAddr MapBufferEx(VAddr cpu_addr, GPUVAddr addr, u64 size);
+    GPUVAddr UnmapBuffer(GPUVAddr addr, u64 size);
+    std::optional<VAddr> GpuToCpuAddress(GPUVAddr addr);
 
-    static constexpr u64 PAGE_BITS = 16;
-    static constexpr u64 PAGE_SIZE = 1 << PAGE_BITS;
-    static constexpr u64 PAGE_MASK = PAGE_SIZE - 1;
+    template <typename T>
+    T Read(GPUVAddr addr);
+
+    template <typename T>
+    void Write(GPUVAddr addr, T data);
+
+    u8* GetPointer(GPUVAddr addr);
+
+    void ReadBlock(GPUVAddr src_addr, void* dest_buffer, std::size_t size);
+    void WriteBlock(GPUVAddr dest_addr, const void* src_buffer, std::size_t size);
+    void CopyBlock(GPUVAddr dest_addr, GPUVAddr src_addr, std::size_t size);
 
 private:
-    enum class PageStatus : u64 {
-        Unmapped = 0xFFFFFFFFFFFFFFFFULL,
-        Allocated = 0xFFFFFFFFFFFFFFFEULL,
-        Reserved = 0xFFFFFFFFFFFFFFFDULL,
-    };
+    using VMAMap = std::map<GPUVAddr, VirtualMemoryArea>;
+    using VMAHandle = VMAMap::const_iterator;
+    using VMAIter = VMAMap::iterator;
 
-    std::optional<GPUVAddr> FindFreeBlock(GPUVAddr region_start, u64 size, u64 align,
-                                          PageStatus status);
-    VAddr& PageSlot(GPUVAddr gpu_addr);
+    bool IsAddressValid(GPUVAddr addr) const;
+    void MapPages(GPUVAddr base, u64 size, u8* memory, Common::PageType type,
+                  VAddr backing_addr = 0);
+    void MapMemoryRegion(GPUVAddr base, u64 size, u8* target, VAddr backing_addr);
+    void UnmapRegion(GPUVAddr base, u64 size);
 
-    static constexpr u64 MAX_ADDRESS{0x10000000000ULL};
-    static constexpr u64 PAGE_TABLE_BITS{10};
-    static constexpr u64 PAGE_TABLE_SIZE{1 << PAGE_TABLE_BITS};
-    static constexpr u64 PAGE_TABLE_MASK{PAGE_TABLE_SIZE - 1};
-    static constexpr u64 PAGE_BLOCK_BITS{14};
-    static constexpr u64 PAGE_BLOCK_SIZE{1 << PAGE_BLOCK_BITS};
-    static constexpr u64 PAGE_BLOCK_MASK{PAGE_BLOCK_SIZE - 1};
+    /// Finds the VMA in which the given address is included in, or `vma_map.end()`.
+    VMAHandle FindVMA(GPUVAddr target) const;
 
-    using PageBlock = std::array<VAddr, PAGE_BLOCK_SIZE>;
-    std::array<std::unique_ptr<PageBlock>, PAGE_TABLE_SIZE> page_table{};
+    VMAHandle AllocateMemory(GPUVAddr target, std::size_t offset, u64 size);
 
-    struct MappedRegion {
-        VAddr cpu_addr;
-        GPUVAddr gpu_addr;
-        u64 size;
-    };
+    /**
+     * Maps an unmanaged host memory pointer at a given address.
+     *
+     * @param target The guest address to start the mapping at.
+     * @param memory The memory to be mapped.
+     * @param size Size of the mapping.
+     * @param state MemoryState tag to attach to the VMA.
+     */
+    VMAHandle MapBackingMemory(GPUVAddr target, u8* memory, u64 size, VAddr backing_addr);
 
-    std::vector<MappedRegion> mapped_regions;
+    /// Unmaps a range of addresses, splitting VMAs as necessary.
+    void UnmapRange(GPUVAddr target, u64 size);
+
+    /// Converts a VMAHandle to a mutable VMAIter.
+    VMAIter StripIterConstness(const VMAHandle& iter);
+
+    /// Marks as the specfied VMA as allocated.
+    VMAIter Allocate(VMAIter vma);
+
+    /**
+     * Carves a VMA of a specific size at the specified address by splitting Free VMAs while doing
+     * the appropriate error checking.
+     */
+    VMAIter CarveVMA(GPUVAddr base, u64 size);
+
+    /**
+     * Splits the edges of the given range of non-Free VMAs so that there is a VMA split at each
+     * end of the range.
+     */
+    VMAIter CarveVMARange(GPUVAddr base, u64 size);
+
+    /**
+     * Splits a VMA in two, at the specified offset.
+     * @returns the right side of the split, with the original iterator becoming the left side.
+     */
+    VMAIter SplitVMA(VMAIter vma, u64 offset_in_vma);
+
+    /**
+     * Checks for and merges the specified VMA with adjacent ones if possible.
+     * @returns the merged VMA or the original if no merging was possible.
+     */
+    VMAIter MergeAdjacent(VMAIter vma);
+
+    /// Updates the pages corresponding to this VMA so they match the VMA's attributes.
+    void UpdatePageTableForVMA(const VirtualMemoryArea& vma);
+
+    /// Finds a free (unmapped region) of the specified size starting at the specified address.
+    GPUVAddr FindFreeRegion(GPUVAddr region_start, u64 size);
+
+private:
+    static constexpr u64 page_bits{16};
+    static constexpr u64 page_size{1 << page_bits};
+    static constexpr u64 page_mask{page_size - 1};
+
+    /// Address space in bits, this is fairly arbitrary but sufficiently large.
+    static constexpr u32 address_space_width{39};
+    /// Start address for mapping, this is fairly arbitrary but must be non-zero.
+    static constexpr GPUVAddr address_space_base{0x100000};
+    /// End of address space, based on address space in bits.
+    static constexpr GPUVAddr address_space_end{1ULL << address_space_width};
+
+    Common::PageTable page_table{page_bits};
+    VMAMap vma_map;
 };
 
 } // namespace Tegra

src/video_core/morton.cpp

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

src/video_core/morton.h

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

src/video_core/rasterizer_cache.h

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

src/video_core/rasterizer_interface.h

@@ -9,7 +9,6 @@
 #include "common/common_types.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/gpu.h"
-#include "video_core/memory_manager.h"
 
 namespace VideoCore {
 
@@ -35,14 +34,14 @@ public:
     virtual void FlushAll() = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
-    virtual void FlushRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushRegion(CacheAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be invalidated
-    virtual void InvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void InvalidateRegion(CacheAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     /// and invalidated
-    virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
 
     /// Attempt to use a faster method to perform a surface copy
     virtual bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
@@ -63,7 +62,7 @@ public:
     }
 
     /// Increase/decrease the number of object in pages touching the specified region
-    virtual void UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta) {}
+    virtual void UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {}
 
     /// Initialize disk cached resources for the game being emulated
     virtual void LoadDiskResources(const std::atomic_bool& stop_loading = false,

src/video_core/renderer_opengl/gl_buffer_cache.cpp

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

src/video_core/renderer_opengl/gl_buffer_cache.h

@@ -17,22 +17,39 @@ namespace OpenGL {
 
 class RasterizerOpenGL;
 
-struct CachedBufferEntry final : public RasterizerCacheObject {
-    VAddr GetAddr() const override {
-        return addr;
+class CachedBufferEntry final : public RasterizerCacheObject {
+public:
+    explicit CachedBufferEntry(VAddr cpu_addr, std::size_t size, GLintptr offset,
+                               std::size_t alignment, u8* host_ptr);
+
+    VAddr GetCpuAddr() const override {
+        return cpu_addr;
     }
 
     std::size_t GetSizeInBytes() const override {
         return size;
     }
 
+    std::size_t GetSize() const {
+        return size;
+    }
+
+    GLintptr GetOffset() const {
+        return offset;
+    }
+
+    std::size_t GetAlignment() const {
+        return alignment;
+    }
+
     // We do not have to flush this cache as things in it are never modified by us.
     void Flush() override {}
 
-    VAddr addr;
-    std::size_t size;
-    GLintptr offset;
-    std::size_t alignment;
+private:
+    VAddr cpu_addr{};
+    std::size_t size{};
+    GLintptr offset{};
+    std::size_t alignment{};
 };
 
 class OGLBufferCache final : public RasterizerCache<std::shared_ptr<CachedBufferEntry>> {
@@ -41,7 +58,7 @@ public:
 
     /// Uploads data from a guest GPU address. Returns host's buffer offset where it's been
     /// allocated.
-    GLintptr UploadMemory(Tegra::GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
+    GLintptr UploadMemory(GPUVAddr gpu_addr, std::size_t size, std::size_t alignment = 4,
                           bool cache = true);
 
     /// Uploads from a host memory. Returns host's buffer offset where it's been allocated.

src/video_core/renderer_opengl/gl_global_cache.cpp

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

src/video_core/renderer_opengl/gl_global_cache.h

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

src/video_core/renderer_opengl/gl_primitive_assembler.cpp

@@ -40,16 +40,12 @@ GLintptr PrimitiveAssembler::MakeQuadArray(u32 first, u32 count) {
     return index_offset;
 }
 
-GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size_t index_size,
-                                             u32 count) {
+GLintptr PrimitiveAssembler::MakeQuadIndexed(GPUVAddr gpu_addr, std::size_t index_size, u32 count) {
     const std::size_t map_size{CalculateQuadSize(count)};
     auto [dst_pointer, index_offset] = buffer_cache.ReserveMemory(map_size);
 
     auto& memory_manager = Core::System::GetInstance().GPU().MemoryManager();
-    const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr)};
-    ASSERT_MSG(cpu_addr, "Invalid GPU address");
-
-    const u8* source{Memory::GetPointer(*cpu_addr)};
+    const u8* source{memory_manager.GetPointer(gpu_addr)};
 
     for (u32 primitive = 0; primitive < count / 4; ++primitive) {
         for (std::size_t i = 0; i < TRIANGLES_PER_QUAD; ++i) {
@@ -64,4 +60,4 @@ GLintptr PrimitiveAssembler::MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size
     return index_offset;
 }
 
-} // namespace OpenGL
+} // namespace OpenGL

src/video_core/renderer_opengl/gl_primitive_assembler.h

@@ -24,7 +24,7 @@ public:
 
     GLintptr MakeQuadArray(u32 first, u32 count);
 
-    GLintptr MakeQuadIndexed(Tegra::GPUVAddr gpu_addr, std::size_t index_size, u32 count);
+    GLintptr MakeQuadIndexed(GPUVAddr gpu_addr, std::size_t index_size, u32 count);
 
 private:
     OGLBufferCache& buffer_cache;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -225,8 +225,8 @@ void RasterizerOpenGL::SetupVertexBuffer(GLuint vao) {
         if (!vertex_array.IsEnabled())
             continue;
 
-        const Tegra::GPUVAddr start = vertex_array.StartAddress();
-        const Tegra::GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
+        const GPUVAddr start = vertex_array.StartAddress();
+        const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
 
         ASSERT(end > start);
         const u64 size = end - start + 1;
@@ -421,8 +421,8 @@ std::size_t RasterizerOpenGL::CalculateVertexArraysSize() const {
         if (!regs.vertex_array[index].IsEnabled())
             continue;
 
-        const Tegra::GPUVAddr start = regs.vertex_array[index].StartAddress();
-        const Tegra::GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
+        const GPUVAddr start = regs.vertex_array[index].StartAddress();
+        const GPUVAddr end = regs.vertex_array_limit[index].LimitAddress();
 
         ASSERT(end > start);
         size += end - start + 1;
@@ -449,7 +449,7 @@ static constexpr auto RangeFromInterval(Map& map, const Interval& interval) {
     return boost::make_iterator_range(map.equal_range(interval));
 }
 
-void RasterizerOpenGL::UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta) {
+void RasterizerOpenGL::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {
     const u64 page_start{addr >> Memory::PAGE_BITS};
     const u64 page_end{(addr + size + Memory::PAGE_SIZE - 1) >> Memory::PAGE_BITS};
 
@@ -747,20 +747,26 @@ void RasterizerOpenGL::DrawArrays() {
 
 void RasterizerOpenGL::FlushAll() {}
 
-void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::FlushRegion(CacheAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    if (!addr || !size) {
+        return;
+    }
     res_cache.FlushRegion(addr, size);
 }
 
-void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::InvalidateRegion(CacheAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
+    if (!addr || !size) {
+        return;
+    }
     res_cache.InvalidateRegion(addr, size);
     shader_cache.InvalidateRegion(addr, size);
     global_cache.InvalidateRegion(addr, size);
     buffer_cache.InvalidateRegion(addr, size);
 }
 
-void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     FlushRegion(addr, size);
     InvalidateRegion(addr, size);
 }
@@ -782,7 +788,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
 
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
 
-    const auto& surface{res_cache.TryFindFramebufferSurface(framebuffer_addr)};
+    const auto& surface{res_cache.TryFindFramebufferSurface(Memory::GetPointer(framebuffer_addr))};
     if (!surface) {
         return {};
     }

src/video_core/renderer_opengl/gl_rasterizer.h

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

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -55,12 +55,11 @@ static void ApplyTextureDefaults(GLuint texture, u32 max_mip_level) {
     }
 }
 
-void SurfaceParams::InitCacheParameters(Tegra::GPUVAddr gpu_addr_) {
+void SurfaceParams::InitCacheParameters(GPUVAddr gpu_addr_) {
     auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
-    const auto cpu_addr{memory_manager.GpuToCpuAddress(gpu_addr_)};
 
-    addr = cpu_addr ? *cpu_addr : 0;
     gpu_addr = gpu_addr_;
+    host_ptr = memory_manager.GetPointer(gpu_addr_);
     size_in_bytes = SizeInBytesRaw();
 
     if (IsPixelFormatASTC(pixel_format)) {
@@ -223,7 +222,7 @@ std::size_t SurfaceParams::InnerMemorySize(bool force_gl, bool layer_only,
 }
 
 /*static*/ SurfaceParams SurfaceParams::CreateForDepthBuffer(
-    u32 zeta_width, u32 zeta_height, Tegra::GPUVAddr zeta_address, Tegra::DepthFormat format,
+    u32 zeta_width, u32 zeta_height, GPUVAddr zeta_address, Tegra::DepthFormat format,
     u32 block_width, u32 block_height, u32 block_depth,
     Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type) {
     SurfaceParams params{};
@@ -446,7 +445,7 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params,
             MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level),
                           params.MipBlockHeight(mip_level), params.MipHeight(mip_level),
                           params.MipBlockDepth(mip_level), 1, params.tile_width_spacing,
-                          gl_buffer.data() + offset_gl, gl_size, params.addr + offset);
+                          gl_buffer.data() + offset_gl, params.host_ptr + offset);
             offset += layer_size;
             offset_gl += gl_size;
         }
@@ -455,7 +454,7 @@ void SwizzleFunc(const MortonSwizzleMode& mode, const SurfaceParams& params,
         MortonSwizzle(mode, params.pixel_format, params.MipWidth(mip_level),
                       params.MipBlockHeight(mip_level), params.MipHeight(mip_level),
                       params.MipBlockDepth(mip_level), depth, params.tile_width_spacing,
-                      gl_buffer.data(), gl_buffer.size(), params.addr + offset);
+                      gl_buffer.data(), params.host_ptr + offset);
     }
 }
 
@@ -513,9 +512,9 @@ void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surfac
                               "reinterpretation but the texture is tiled.");
         }
         const std::size_t remaining_size = dst_params.size_in_bytes - src_params.size_in_bytes;
-
+        auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
         glBufferSubData(GL_PIXEL_PACK_BUFFER, src_params.size_in_bytes, remaining_size,
-                        Memory::GetPointer(dst_params.addr + src_params.size_in_bytes));
+                        memory_manager.GetPointer(dst_params.gpu_addr + src_params.size_in_bytes));
     }
 
     glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
@@ -563,8 +562,14 @@ void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surfac
 }
 
 CachedSurface::CachedSurface(const SurfaceParams& params)
-    : params(params), gl_target(SurfaceTargetToGL(params.target)),
-      cached_size_in_bytes(params.size_in_bytes) {
+    : params{params}, gl_target{SurfaceTargetToGL(params.target)},
+      cached_size_in_bytes{params.size_in_bytes}, RasterizerCacheObject{params.host_ptr} {
+
+    const auto optional_cpu_addr{
+        Core::System::GetInstance().GPU().MemoryManager().GpuToCpuAddress(params.gpu_addr)};
+    ASSERT_MSG(optional_cpu_addr, "optional_cpu_addr is invalid");
+    cpu_addr = *optional_cpu_addr;
+
     texture.Create(gl_target);
 
     // TODO(Rodrigo): Using params.GetRect() returns a different size than using its Mip*(0)
@@ -603,19 +608,7 @@ CachedSurface::CachedSurface(const SurfaceParams& params)
 
     ApplyTextureDefaults(texture.handle, params.max_mip_level);
 
-    OpenGL::LabelGLObject(GL_TEXTURE, texture.handle, params.addr, params.IdentityString());
-
-    // Clamp size to mapped GPU memory region
-    // TODO(bunnei): Super Mario Odyssey maps a 0x40000 byte region and then uses it for a 0x80000
-    // R32F render buffer. We do not yet know if this is a game bug or something else, but this
-    // check is necessary to prevent flushing from overwriting unmapped memory.
-
-    auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
-    const u64 max_size{memory_manager.GetRegionEnd(params.gpu_addr) - params.gpu_addr};
-    if (cached_size_in_bytes > max_size) {
-        LOG_ERROR(HW_GPU, "Surface size {} exceeds region size {}", params.size_in_bytes, max_size);
-        cached_size_in_bytes = max_size;
-    }
+    OpenGL::LabelGLObject(GL_TEXTURE, texture.handle, params.gpu_addr, params.IdentityString());
 }
 
 MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 192, 64));
@@ -633,10 +626,9 @@ void CachedSurface::LoadGLBuffer() {
         const u32 bpp = params.GetFormatBpp() / 8;
         const u32 copy_size = params.width * bpp;
         if (params.pitch == copy_size) {
-            std::memcpy(gl_buffer[0].data(), Memory::GetPointer(params.addr),
-                        params.size_in_bytes_gl);
+            std::memcpy(gl_buffer[0].data(), params.host_ptr, params.size_in_bytes_gl);
         } else {
-            const u8* start = Memory::GetPointer(params.addr);
+            const u8* start{params.host_ptr};
             u8* write_to = gl_buffer[0].data();
             for (u32 h = params.height; h > 0; h--) {
                 std::memcpy(write_to, start, copy_size);
@@ -680,8 +672,6 @@ void CachedSurface::FlushGLBuffer() {
     glPixelStorei(GL_PACK_ROW_LENGTH, 0);
     Tegra::Texture::ConvertFromHostToGuest(gl_buffer[0].data(), params.pixel_format, params.width,
                                            params.height, params.depth, true, true);
-    const u8* const texture_src_data = Memory::GetPointer(params.addr);
-    ASSERT(texture_src_data);
     if (params.is_tiled) {
         ASSERT_MSG(params.block_width == 1, "Block width is defined as {} on texture type {}",
                    params.block_width, static_cast<u32>(params.target));
@@ -691,9 +681,9 @@ void CachedSurface::FlushGLBuffer() {
         const u32 bpp = params.GetFormatBpp() / 8;
         const u32 copy_size = params.width * bpp;
         if (params.pitch == copy_size) {
-            std::memcpy(Memory::GetPointer(params.addr), gl_buffer[0].data(), GetSizeInBytes());
+            std::memcpy(params.host_ptr, gl_buffer[0].data(), GetSizeInBytes());
         } else {
-            u8* start = Memory::GetPointer(params.addr);
+            u8* start{params.host_ptr};
             const u8* read_to = gl_buffer[0].data();
             for (u32 h = params.height; h > 0; h--) {
                 std::memcpy(start, read_to, copy_size);
@@ -927,12 +917,12 @@ void RasterizerCacheOpenGL::LoadSurface(const Surface& surface) {
 }
 
 Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool preserve_contents) {
-    if (params.addr == 0 || params.height * params.width == 0) {
+    if (!params.IsValid()) {
         return {};
     }
 
     // Look up surface in the cache based on address
-    Surface surface{TryGet(params.addr)};
+    Surface surface{TryGet(params.host_ptr)};
     if (surface) {
         if (surface->GetSurfaceParams().IsCompatibleSurface(params)) {
             // Use the cached surface as-is unless it's not synced with memory
@@ -943,7 +933,7 @@ Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool pres
             // 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)};
-            UnregisterSurface(surface);
+            Unregister(surface);
             Register(new_surface);
             if (new_surface->IsUploaded()) {
                 RegisterReinterpretSurface(new_surface);
@@ -951,7 +941,7 @@ Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, bool pres
             return new_surface;
         } else {
             // Delete the old surface before creating a new one to prevent collisions.
-            UnregisterSurface(surface);
+            Unregister(surface);
         }
     }
 
@@ -981,14 +971,16 @@ void RasterizerCacheOpenGL::FastLayeredCopySurface(const Surface& src_surface,
                                                    const Surface& dst_surface) {
     const auto& init_params{src_surface->GetSurfaceParams()};
     const auto& dst_params{dst_surface->GetSurfaceParams()};
-    VAddr address = init_params.addr;
-    const std::size_t layer_size = dst_params.LayerMemorySize();
+    auto& memory_manager{Core::System::GetInstance().GPU().MemoryManager()};
+    GPUVAddr address{init_params.gpu_addr};
+    const std::size_t layer_size{dst_params.LayerMemorySize()};
     for (u32 layer = 0; layer < dst_params.depth; layer++) {
         for (u32 mipmap = 0; mipmap < dst_params.max_mip_level; mipmap++) {
-            const VAddr sub_address = address + dst_params.GetMipmapLevelOffset(mipmap);
-            const Surface& copy = TryGet(sub_address);
-            if (!copy)
+            const GPUVAddr sub_address{address + dst_params.GetMipmapLevelOffset(mipmap)};
+            const Surface& copy{TryGet(memory_manager.GetPointer(sub_address))};
+            if (!copy) {
                 continue;
+            }
             const auto& src_params{copy->GetSurfaceParams()};
             const u32 width{std::min(src_params.width, dst_params.MipWidth(mipmap))};
             const u32 height{std::min(src_params.height, dst_params.MipHeight(mipmap))};
@@ -1163,7 +1155,8 @@ void RasterizerCacheOpenGL::AccurateCopySurface(const Surface& src_surface,
     const auto& dst_params{dst_surface->GetSurfaceParams()};
 
     // Flush enough memory for both the source and destination surface
-    FlushRegion(src_params.addr, std::max(src_params.MemorySize(), dst_params.MemorySize()));
+    FlushRegion(ToCacheAddr(src_params.host_ptr),
+                std::max(src_params.MemorySize(), dst_params.MemorySize()));
 
     LoadSurface(dst_surface);
 }
@@ -1215,8 +1208,8 @@ Surface RasterizerCacheOpenGL::RecreateSurface(const Surface& old_surface,
     return new_surface;
 }
 
-Surface RasterizerCacheOpenGL::TryFindFramebufferSurface(VAddr addr) const {
-    return TryGet(addr);
+Surface RasterizerCacheOpenGL::TryFindFramebufferSurface(const u8* host_ptr) const {
+    return TryGet(host_ptr);
 }
 
 void RasterizerCacheOpenGL::ReserveSurface(const Surface& surface) {
@@ -1243,9 +1236,9 @@ static std::optional<u32> TryFindBestMipMap(std::size_t memory, const SurfacePar
     return {};
 }
 
-static std::optional<u32> TryFindBestLayer(VAddr addr, const SurfaceParams params, u32 mipmap) {
-    const std::size_t size = params.LayerMemorySize();
-    VAddr start = params.addr + params.GetMipmapLevelOffset(mipmap);
+static std::optional<u32> TryFindBestLayer(GPUVAddr addr, const SurfaceParams params, u32 mipmap) {
+    const std::size_t size{params.LayerMemorySize()};
+    GPUVAddr start{params.gpu_addr + params.GetMipmapLevelOffset(mipmap)};
     for (u32 i = 0; i < params.depth; i++) {
         if (start == addr) {
             return {i};
@@ -1267,7 +1260,7 @@ static bool LayerFitReinterpretSurface(RasterizerCacheOpenGL& cache, const Surfa
             src_params.height == dst_params.MipHeight(*level) &&
             src_params.block_height >= dst_params.MipBlockHeight(*level)) {
             const std::optional<u32> slot =
-                TryFindBestLayer(render_surface->GetAddr(), dst_params, *level);
+                TryFindBestLayer(render_surface->GetSurfaceParams().gpu_addr, dst_params, *level);
             if (slot.has_value()) {
                 glCopyImageSubData(render_surface->Texture().handle,
                                    SurfaceTargetToGL(src_params.target), 0, 0, 0, 0,
@@ -1283,8 +1276,8 @@ static bool LayerFitReinterpretSurface(RasterizerCacheOpenGL& cache, const Surfa
 }
 
 static bool IsReinterpretInvalid(const Surface render_surface, const Surface blitted_surface) {
-    const VAddr bound1 = blitted_surface->GetAddr() + blitted_surface->GetMemorySize();
-    const VAddr bound2 = render_surface->GetAddr() + render_surface->GetMemorySize();
+    const VAddr bound1 = blitted_surface->GetCpuAddr() + blitted_surface->GetMemorySize();
+    const VAddr bound2 = render_surface->GetCpuAddr() + render_surface->GetMemorySize();
     if (bound2 > bound1)
         return true;
     const auto& dst_params = blitted_surface->GetSurfaceParams();
@@ -1302,12 +1295,12 @@ static bool IsReinterpretInvalidSecond(const Surface render_surface,
 bool RasterizerCacheOpenGL::PartialReinterpretSurface(Surface triggering_surface,
                                                       Surface intersect) {
     if (IsReinterpretInvalid(triggering_surface, intersect)) {
-        UnregisterSurface(intersect);
+        Unregister(intersect);
         return false;
     }
     if (!LayerFitReinterpretSurface(*this, triggering_surface, intersect)) {
         if (IsReinterpretInvalidSecond(triggering_surface, intersect)) {
-            UnregisterSurface(intersect);
+            Unregister(intersect);
             return false;
         }
         FlushObject(intersect);
@@ -1327,7 +1320,8 @@ void RasterizerCacheOpenGL::SignalPreDrawCall() {
 void RasterizerCacheOpenGL::SignalPostDrawCall() {
     for (u32 i = 0; i < Maxwell::NumRenderTargets; i++) {
         if (current_color_buffers[i] != nullptr) {
-            Surface intersect = CollideOnReinterpretedSurface(current_color_buffers[i]->GetAddr());
+            Surface intersect =
+                CollideOnReinterpretedSurface(current_color_buffers[i]->GetCacheAddr());
             if (intersect != nullptr) {
                 PartialReinterpretSurface(current_color_buffers[i], intersect);
                 texception = true;

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -109,6 +109,11 @@ struct SurfaceParams {
         return size;
     }
 
+    /// Returns true if the parameters constitute a valid rasterizer surface.
+    bool IsValid() const {
+        return gpu_addr && host_ptr && height && width;
+    }
+
     /// Returns the exact size of the memory occupied by a layer in a texture in VRAM, including
     /// mipmaps.
     std::size_t LayerMemorySize() const {
@@ -210,7 +215,7 @@ struct SurfaceParams {
 
     /// Creates SurfaceParams for a depth buffer configuration
     static SurfaceParams CreateForDepthBuffer(
-        u32 zeta_width, u32 zeta_height, Tegra::GPUVAddr zeta_address, Tegra::DepthFormat format,
+        u32 zeta_width, u32 zeta_height, GPUVAddr zeta_address, Tegra::DepthFormat format,
         u32 block_width, u32 block_height, u32 block_depth,
         Tegra::Engines::Maxwell3D::Regs::InvMemoryLayout type);
 
@@ -232,7 +237,7 @@ struct SurfaceParams {
     }
 
     /// Initializes parameters for caching, should be called after everything has been initialized
-    void InitCacheParameters(Tegra::GPUVAddr gpu_addr);
+    void InitCacheParameters(GPUVAddr gpu_addr);
 
     std::string TargetName() const {
         switch (target) {
@@ -296,8 +301,8 @@ struct SurfaceParams {
     bool is_array;
     bool srgb_conversion;
     // Parameters used for caching
-    VAddr addr;
-    Tegra::GPUVAddr gpu_addr;
+    u8* host_ptr;
+    GPUVAddr gpu_addr;
     std::size_t size_in_bytes;
     std::size_t size_in_bytes_gl;
 
@@ -345,10 +350,10 @@ class RasterizerOpenGL;
 
 class CachedSurface final : public RasterizerCacheObject {
 public:
-    CachedSurface(const SurfaceParams& params);
+    explicit CachedSurface(const SurfaceParams& params);
 
-    VAddr GetAddr() const override {
-        return params.addr;
+    VAddr GetCpuAddr() const override {
+        return cpu_addr;
     }
 
     std::size_t GetSizeInBytes() const override {
@@ -432,6 +437,7 @@ private:
     std::size_t memory_size;
     bool reinterpreted = false;
     bool must_reload = false;
+    VAddr cpu_addr{};
 };
 
 class RasterizerCacheOpenGL final : public RasterizerCache<Surface> {
@@ -449,7 +455,7 @@ public:
     Surface GetColorBufferSurface(std::size_t index, bool preserve_contents);
 
     /// Tries to find a framebuffer using on the provided CPU address
-    Surface TryFindFramebufferSurface(VAddr addr) const;
+    Surface TryFindFramebufferSurface(const u8* host_ptr) const;
 
     /// Copies the contents of one surface to another
     void FermiCopySurface(const Tegra::Engines::Fermi2D::Regs::Surface& src_config,
@@ -506,12 +512,12 @@ private:
     std::array<Surface, Maxwell::NumRenderTargets> current_color_buffers;
     Surface last_depth_buffer;
 
-    using SurfaceIntervalCache = boost::icl::interval_map<VAddr, Surface>;
+    using SurfaceIntervalCache = boost::icl::interval_map<CacheAddr, Surface>;
     using SurfaceInterval = typename SurfaceIntervalCache::interval_type;
 
     static auto GetReinterpretInterval(const Surface& object) {
-        return SurfaceInterval::right_open(object->GetAddr() + 1,
-                                           object->GetAddr() + object->GetMemorySize() - 1);
+        return SurfaceInterval::right_open(object->GetCacheAddr() + 1,
+                                           object->GetCacheAddr() + object->GetMemorySize() - 1);
     }
 
     // Reinterpreted surfaces are very fragil as the game may keep rendering into them.
@@ -523,7 +529,7 @@ private:
         reinterpret_surface->MarkReinterpreted();
     }
 
-    Surface CollideOnReinterpretedSurface(VAddr addr) const {
+    Surface CollideOnReinterpretedSurface(CacheAddr addr) const {
         const SurfaceInterval interval{addr};
         for (auto& pair :
              boost::make_iterator_range(reinterpreted_surfaces.equal_range(interval))) {
@@ -532,13 +538,17 @@ private:
         return nullptr;
     }
 
+    void Register(const Surface& object) {
+        RasterizerCache<Surface>::Register(object);
+    }
+
     /// Unregisters an object from the cache
-    void UnregisterSurface(const Surface& object) {
+    void Unregister(const Surface& object) {
         if (object->IsReinterpreted()) {
             auto interval = GetReinterpretInterval(object);
             reinterpreted_surfaces.erase(interval);
         }
-        Unregister(object);
+        RasterizerCache<Surface>::Unregister(object);
     }
 };