Custom textures rewrite (#6452)

* common: Add thread pool from yuzu

* Is really useful for asynchronous operations like shader compilation and custom textures, will be used in following PRs

* core: Improve ImageInterface

* Provide a default implementation so frontends don't have to duplicate code registering the lodepng version

* Add a dds version too which we will use in the next commit

* rasterizer_cache: Rewrite custom textures

* There's just too much to talk about here, look at the PR description for more details

* rasterizer_cache: Implement basic pack configuration file

* custom_tex_manager: Flip dumped textures

* custom_tex_manager: Optimize custom texture hashing

* If no convertions are needed then we can hash the decoded data directly removing the needed for duplicate decode

* custom_tex_manager: Implement asynchronous texture loading

* The file loading and decoding is offloaded into worker threads, while the upload itself still occurs in the main thread to avoid having to manage shared contexts

* Address review comments

* custom_tex_manager: Introduce custom material support

* video_core: Move custom textures to separate directory

* Also split the files to make the code cleaner

* gl_texture_runtime: Generate mipmaps for material

* custom_tex_manager: Prevent memory overflow when preloading

* externals: Add dds-ktx as submodule

* string_util: Return vector from SplitString

* No code benefits from passing it as an argument

* custom_textures: Use json config file

* gl_rasterizer: Only bind material for unit 0

* Address review comments

.gitmodules

@@ -61,3 +61,6 @@
 [submodule "cryptopp"]
 	path = externals/cryptopp
 	url = https://github.com/weidai11/cryptopp.git
+[submodule "dds-ktx"]
+    path = externals/dds-ktx
+    url = https://github.com/septag/dds-ktx

externals/CMakeLists.txt

@@ -64,6 +64,10 @@ if(ANDROID)
     endforeach()
 endif()
 
+# dds-ktx
+add_library(dds-ktx INTERFACE)
+target_include_directories(dds-ktx INTERFACE ./dds-ktx)
+
 # fmt and Xbyak need to be added before dynarmic
 # libfmt
 option(FMT_INSTALL "" ON)
@@ -141,6 +145,10 @@ if (USE_DISCORD_PRESENCE)
     target_include_directories(discord-rpc INTERFACE ./discord-rpc/include)
 endif()
 
+# JSON
+add_library(json-headers INTERFACE)
+target_include_directories(json-headers INTERFACE ./json)
+
 if (ENABLE_WEB_SERVICE)
     find_package(OpenSSL 1.1)
     if (OPENSSL_FOUND)
@@ -156,9 +164,6 @@ if (ENABLE_WEB_SERVICE)
             DIRECTORY libressl
             DEFINITION OPENSSL_LIBS)
     endif()
-    # JSON
-    add_library(json-headers INTERFACE)
-    target_include_directories(json-headers INTERFACE ./json)
 
     if(ANDROID)
         add_subdirectory(android-ifaddrs)

externals/dds-ktx

@@ -0,0 +1 @@
+Subproject commit 42dd8aa6ded90b1ec06091522774feff51e83fc5

externals/zstd

@@ -1 +1 @@
-Subproject commit 63779c798237346c2b245c546c40b72a5a5913fe
+Subproject commit e47e674cd09583ff0503f0f6defd6d23d8b718d3

src/android/app/src/main/java/org/citra/citra_emu/features/settings/ui/SettingsFragmentPresenter.java

@@ -368,6 +368,7 @@ public final class SettingsFragmentPresenter {
         SettingSection utilitySection = mSettings.getSection(Settings.SECTION_UTILITY);
         Setting dumpTextures = utilitySection.getSetting(SettingsFile.KEY_DUMP_TEXTURES);
         Setting customTextures = utilitySection.getSetting(SettingsFile.KEY_CUSTOM_TEXTURES);
+        Setting asyncCustomLoading = utilitySection.getSetting(SettingsFile.KEY_ASYNC_CUSTOM_LOADING);
         //Setting preloadTextures = utilitySection.getSetting(SettingsFile.KEY_PRELOAD_TEXTURES);
 
         sl.add(new HeaderSetting(null, null, R.string.renderer, 0));
@@ -389,6 +390,7 @@ public final class SettingsFragmentPresenter {
         sl.add(new HeaderSetting(null, null, R.string.utility, 0));
         sl.add(new CheckBoxSetting(SettingsFile.KEY_DUMP_TEXTURES, Settings.SECTION_UTILITY, R.string.dump_textures, R.string.dump_textures_description, false, dumpTextures));
         sl.add(new CheckBoxSetting(SettingsFile.KEY_CUSTOM_TEXTURES, Settings.SECTION_UTILITY, R.string.custom_textures, R.string.custom_textures_description, false, customTextures));
+        sl.add(new CheckBoxSetting(SettingsFile.KEY_ASYNC_CUSTOM_LOADING, Settings.SECTION_UTILITY, R.string.async_custom_loading, R.string.async_custom_loading_description, true, asyncCustomLoading));
         //Disabled until custom texture implementation gets rewrite, current one overloads RAM and crashes Citra.
         //sl.add(new CheckBoxSetting(SettingsFile.KEY_PRELOAD_TEXTURES, Settings.SECTION_UTILITY, R.string.preload_textures, R.string.preload_textures_description, false, preloadTextures));
     }

src/android/app/src/main/java/org/citra/citra_emu/features/settings/utils/SettingsFile.java

@@ -73,6 +73,7 @@ public final class SettingsFile {
     public static final String KEY_DUMP_TEXTURES = "dump_textures";
     public static final String KEY_CUSTOM_TEXTURES = "custom_textures";
     public static final String KEY_PRELOAD_TEXTURES = "preload_textures";
+    public static final String KEY_ASYNC_CUSTOM_LOADING = "async_custom_loading";
 
     public static final String KEY_AUDIO_OUTPUT_ENGINE = "output_engine";
     public static final String KEY_ENABLE_AUDIO_STRETCHING = "enable_audio_stretching";

src/android/app/src/main/jni/CMakeLists.txt

@@ -25,8 +25,6 @@ add_library(citra-android SHARED
     game_settings.h
     id_cache.cpp
     id_cache.h
-    lodepng_image_interface.cpp
-    lodepng_image_interface.h
     mic.cpp
     mic.h
     native.cpp
@@ -36,6 +34,6 @@ add_library(citra-android SHARED
 )
 
 target_link_libraries(citra-android PRIVATE audio_core common core input_common network)
-target_link_libraries(citra-android PRIVATE android camera2ndk EGL glad inih jnigraphics lodepng log mediandk yuv)
+target_link_libraries(citra-android PRIVATE android camera2ndk EGL glad inih jnigraphics log mediandk yuv)
 
 set(CPACK_PACKAGE_EXECUTABLES ${CPACK_PACKAGE_EXECUTABLES} citra-android)

src/android/app/src/main/jni/cheats/cheat.cpp

@@ -62,8 +62,7 @@ JNIEXPORT void JNICALL Java_org_citra_citra_1emu_features_cheats_model_Cheat_set
 JNIEXPORT jint JNICALL Java_org_citra_citra_1emu_features_cheats_model_Cheat_isValidGatewayCode(
     JNIEnv* env, jclass, jstring j_code) {
     const std::string code = GetJString(env, j_code);
-    std::vector<std::string> code_lines;
-    Common::SplitString(code, '\n', code_lines);
+    const auto code_lines = Common::SplitString(code, '\n');
 
     for (int i = 0; i < code_lines.size(); ++i) {
         Cheats::GatewayCheat::CheatLine cheat_line(code_lines[i]);

src/android/app/src/main/jni/config.cpp

@@ -195,6 +195,7 @@ void Config::ReadValues() {
     ReadSetting("Utility", Settings::values.dump_textures);
     ReadSetting("Utility", Settings::values.custom_textures);
     ReadSetting("Utility", Settings::values.preload_textures);
+    ReadSetting("Utility", Settings::values.async_custom_loading);
 
     // Audio
     ReadSetting("Audio", Settings::values.audio_emulation);

src/android/app/src/main/jni/default_ini.h

@@ -213,6 +213,10 @@ custom_textures =
 # 0 (default): Off, 1: On
 preload_textures =
 
+# Loads custom textures asynchronously with background threads.
+# 0: Off, 1 (default): On
+async_custom_loading =
+
 [Audio]
 # Whether or not to enable DSP LLE
 # 0 (default): No, 1: Yes

src/android/app/src/main/jni/lodepng_image_interface.cpp

@@ -1,44 +0,0 @@
-// Copyright 2019 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <lodepng.h>
-#include "common/file_util.h"
-#include "common/logging/log.h"
-#include "jni/lodepng_image_interface.h"
-
-bool LodePNGImageInterface::DecodePNG(std::vector<u8>& dst, u32& width, u32& height,
-                                      const std::string& path) {
-    FileUtil::IOFile file(path, "rb");
-    size_t read_size = file.GetSize();
-    std::vector<u8> in(read_size);
-    if (file.ReadBytes(&in[0], read_size) != read_size) {
-        LOG_CRITICAL(Frontend, "Failed to decode {}", path);
-    }
-    u32 lodepng_ret = lodepng::decode(dst, width, height, in);
-    if (lodepng_ret) {
-        LOG_CRITICAL(Frontend, "Failed to decode {} because {}", path,
-                     lodepng_error_text(lodepng_ret));
-        return false;
-    }
-    return true;
-}
-
-bool LodePNGImageInterface::EncodePNG(const std::string& path, const std::vector<u8>& src,
-                                      u32 width, u32 height) {
-    std::vector<u8> out;
-    u32 lodepng_ret = lodepng::encode(out, src, width, height);
-    if (lodepng_ret) {
-        LOG_CRITICAL(Frontend, "Failed to encode {} because {}", path,
-                     lodepng_error_text(lodepng_ret));
-        return false;
-    }
-
-    FileUtil::IOFile file(path, "wb");
-    if (file.WriteBytes(&out[0], out.size()) != out.size()) {
-        LOG_CRITICAL(Frontend, "Failed to save encode to path={}", path);
-        return false;
-    }
-
-    return true;
-}

src/android/app/src/main/jni/lodepng_image_interface.h

@@ -1,14 +0,0 @@
-// Copyright 2019 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include "core/frontend/image_interface.h"
-
-class LodePNGImageInterface final : public Frontend::ImageInterface {
-public:
-    bool DecodePNG(std::vector<u8>& dst, u32& width, u32& height, const std::string& path) override;
-    bool EncodePNG(const std::string& path, const std::vector<u8>& src, u32 width,
-                   u32 height) override;
-};

src/android/app/src/main/jni/native.cpp

@@ -39,7 +39,6 @@
 #include "jni/game_settings.h"
 #include "jni/id_cache.h"
 #include "jni/input_manager.h"
-#include "jni/lodepng_image_interface.h"
 #include "jni/mic.h"
 #include "jni/native.h"
 #include "jni/ndk_motion.h"
@@ -184,9 +183,6 @@ static Core::System::ResultStatus RunCitra(const std::string& filepath) {
     system.RegisterMiiSelector(std::make_shared<MiiSelector::AndroidMiiSelector>());
     system.RegisterSoftwareKeyboard(std::make_shared<SoftwareKeyboard::AndroidKeyboard>());
 
-    // Register generic image interface
-    Core::System::GetInstance().RegisterImageInterface(std::make_shared<LodePNGImageInterface>());
-
     // Register real Mic factory
     Frontend::Mic::RegisterRealMicFactory(std::make_unique<Mic::AndroidFactory>());
 

src/android/app/src/main/res/values/strings.xml

@@ -116,6 +116,8 @@
     <string name="custom_textures_description">Uses custom textures found in load/textures/[GAME ID]</string>
     <string name="preload_textures">Preload custom textures</string>
     <string name="preload_textures_description">Loads all custom textures into memory. This feature can use a lot of memory.</string>
+    <string name="async_custom_loading">Async custom texture loading</string>
+    <string name="async_custom_loading_description">Loads custom textures in the background with worker threads to reduce loading stutter.</string>
     <!-- Premium strings -->
     <string name="premium_text">Premium</string>
     <string name="premium_settings_upsell">Upgrade to Premium and support Citra!</string>

src/citra/CMakeLists.txt

@@ -12,8 +12,6 @@ add_executable(citra
     emu_window/emu_window_sdl2_gl.h
     emu_window/emu_window_sdl2_sw.cpp
     emu_window/emu_window_sdl2_sw.h
-    lodepng_image_interface.cpp
-    lodepng_image_interface.h
     precompiled_headers.h
     resource.h
 )
@@ -21,7 +19,7 @@ add_executable(citra
 create_target_directory_groups(citra)
 
 target_link_libraries(citra PRIVATE common core input_common network)
-target_link_libraries(citra PRIVATE inih glad lodepng)
+target_link_libraries(citra PRIVATE inih glad)
 if (MSVC)
     target_link_libraries(citra PRIVATE getopt)
 endif()

src/citra/citra.cpp

@@ -15,7 +15,6 @@
 #include "citra/emu_window/emu_window_sdl2.h"
 #include "citra/emu_window/emu_window_sdl2_gl.h"
 #include "citra/emu_window/emu_window_sdl2_sw.h"
-#include "citra/lodepng_image_interface.h"
 #include "common/common_paths.h"
 #include "common/detached_tasks.h"
 #include "common/file_util.h"
@@ -359,9 +358,6 @@ int main(int argc, char** argv) {
     // Register frontend applets
     Frontend::RegisterDefaultApplets();
 
-    // Register generic image interface
-    Core::System::GetInstance().RegisterImageInterface(std::make_shared<LodePNGImageInterface>());
-
     EmuWindow_SDL2::InitializeSDL2();
 
     const auto create_emu_window = [](bool fullscreen,

src/citra/config.cpp

@@ -175,6 +175,7 @@ void Config::ReadValues() {
     ReadSetting("Utility", Settings::values.dump_textures);
     ReadSetting("Utility", Settings::values.custom_textures);
     ReadSetting("Utility", Settings::values.preload_textures);
+    ReadSetting("Utility", Settings::values.async_custom_loading);
 
     // Audio
     ReadSetting("Audio", Settings::values.audio_emulation);

src/citra/default_ini.h

@@ -232,6 +232,10 @@ custom_textures =
 # 0 (default): Off, 1: On
 preload_textures =
 
+# Loads custom textures asynchronously with background threads.
+# 0: Off, 1 (default): On
+async_custom_loading =
+
 [Audio]
 # Whether or not to enable DSP LLE
 # 0 (default): No, 1: Yes

src/citra/lodepng_image_interface.cpp

@@ -1,29 +0,0 @@
-// Copyright 2019 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <lodepng.h>
-#include "citra/lodepng_image_interface.h"
-#include "common/logging/log.h"
-
-bool LodePNGImageInterface::DecodePNG(std::vector<u8>& dst, u32& width, u32& height,
-                                      const std::string& path) {
-    u32 lodepng_ret = lodepng::decode(dst, width, height, path);
-    if (lodepng_ret) {
-        LOG_CRITICAL(Frontend, "Failed to decode {} because {}", path,
-                     lodepng_error_text(lodepng_ret));
-        return false;
-    }
-    return true;
-}
-
-bool LodePNGImageInterface::EncodePNG(const std::string& path, const std::vector<u8>& src,
-                                      u32 width, u32 height) {
-    u32 lodepng_ret = lodepng::encode(path, src, width, height);
-    if (lodepng_ret) {
-        LOG_CRITICAL(Frontend, "Failed to encode {} because {}", path,
-                     lodepng_error_text(lodepng_ret));
-        return false;
-    }
-    return true;
-}

src/citra/lodepng_image_interface.h

@@ -1,14 +0,0 @@
-// Copyright 2019 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include "core/frontend/image_interface.h"
-
-class LodePNGImageInterface final : public Frontend::ImageInterface {
-public:
-    bool DecodePNG(std::vector<u8>& dst, u32& width, u32& height, const std::string& path) override;
-    bool EncodePNG(const std::string& path, const std::vector<u8>& src, u32 width,
-                   u32 height) override;
-};

src/citra_qt/configuration/config.cpp

@@ -56,7 +56,7 @@ const std::array<std::array<int, 5>, Settings::NativeAnalog::NumAnalogs> Config:
 // This must be in alphabetical order according to action name as it must have the same order as
 // UISetting::values.shortcuts, which is alphabetically ordered.
 // clang-format off
-const std::array<UISettings::Shortcut, 27> Config::default_hotkeys {{
+const std::array<UISettings::Shortcut, 28> Config::default_hotkeys {{
      {QStringLiteral("Advance Frame"),            QStringLiteral("Main Window"), {QStringLiteral(""),     Qt::ApplicationShortcut}},
      {QStringLiteral("Capture Screenshot"),       QStringLiteral("Main Window"), {QStringLiteral("Ctrl+P"), Qt::WidgetWithChildrenShortcut}},
      {QStringLiteral("Continue/Pause Emulation"), QStringLiteral("Main Window"), {QStringLiteral("F4"),     Qt::WindowShortcut}},
@@ -84,6 +84,7 @@ const std::array<UISettings::Shortcut, 27> Config::default_hotkeys {{
      {QStringLiteral("Toggle Screen Layout"),     QStringLiteral("Main Window"), {QStringLiteral("F10"),    Qt::WindowShortcut}},
      {QStringLiteral("Toggle Status Bar"),        QStringLiteral("Main Window"), {QStringLiteral("Ctrl+S"), Qt::WindowShortcut}},
      {QStringLiteral("Toggle Texture Dumping"),   QStringLiteral("Main Window"), {QStringLiteral(""),       Qt::ApplicationShortcut}},
+     {QStringLiteral("Toggle Custom Textures"),   QStringLiteral("Main Window"), {QStringLiteral("F7"),     Qt::ApplicationShortcut}},
     }};
 // clang-format on
 
@@ -439,6 +440,7 @@ void Config::ReadUtilityValues() {
     ReadGlobalSetting(Settings::values.dump_textures);
     ReadGlobalSetting(Settings::values.custom_textures);
     ReadGlobalSetting(Settings::values.preload_textures);
+    ReadGlobalSetting(Settings::values.async_custom_loading);
 
     qt_config->endGroup();
 }
@@ -949,6 +951,7 @@ void Config::SaveUtilityValues() {
     WriteGlobalSetting(Settings::values.dump_textures);
     WriteGlobalSetting(Settings::values.custom_textures);
     WriteGlobalSetting(Settings::values.preload_textures);
+    WriteGlobalSetting(Settings::values.async_custom_loading);
 
     qt_config->endGroup();
 }

src/citra_qt/configuration/config.h

@@ -26,7 +26,7 @@ public:
 
     static const std::array<int, Settings::NativeButton::NumButtons> default_buttons;
     static const std::array<std::array<int, 5>, Settings::NativeAnalog::NumAnalogs> default_analogs;
-    static const std::array<UISettings::Shortcut, 27> default_hotkeys;
+    static const std::array<UISettings::Shortcut, 28> default_hotkeys;
 
 private:
     void Initialize(const std::string& config_name);

src/citra_qt/configuration/configure_enhancements.cpp

@@ -41,8 +41,10 @@ ConfigureEnhancements::ConfigureEnhancements(QWidget* parent)
     });
 
     ui->toggle_preload_textures->setEnabled(ui->toggle_custom_textures->isChecked());
+    ui->toggle_async_custom_loading->setEnabled(ui->toggle_custom_textures->isChecked());
     connect(ui->toggle_custom_textures, &QCheckBox::toggled, this, [this] {
         ui->toggle_preload_textures->setEnabled(ui->toggle_custom_textures->isChecked());
+        ui->toggle_async_custom_loading->setEnabled(ui->toggle_custom_textures->isChecked());
         if (!ui->toggle_preload_textures->isEnabled())
             ui->toggle_preload_textures->setChecked(false);
     });
@@ -83,6 +85,7 @@ void ConfigureEnhancements::SetConfiguration() {
     ui->toggle_dump_textures->setChecked(Settings::values.dump_textures.GetValue());
     ui->toggle_custom_textures->setChecked(Settings::values.custom_textures.GetValue());
     ui->toggle_preload_textures->setChecked(Settings::values.preload_textures.GetValue());
+    ui->toggle_async_custom_loading->setChecked(Settings::values.async_custom_loading.GetValue());
     bg_color =
         QColor::fromRgbF(Settings::values.bg_red.GetValue(), Settings::values.bg_green.GetValue(),
                          Settings::values.bg_blue.GetValue());
@@ -159,6 +162,8 @@ void ConfigureEnhancements::ApplyConfiguration() {
                                              ui->toggle_custom_textures, custom_textures);
     ConfigurationShared::ApplyPerGameSetting(&Settings::values.preload_textures,
                                              ui->toggle_preload_textures, preload_textures);
+    ConfigurationShared::ApplyPerGameSetting(&Settings::values.async_custom_loading,
+                                             ui->toggle_async_custom_loading, async_custom_loading);
 
     Settings::values.bg_red = static_cast<float>(bg_color.redF());
     Settings::values.bg_green = static_cast<float>(bg_color.greenF());
@@ -176,6 +181,8 @@ void ConfigureEnhancements::SetupPerGameUI() {
         ui->toggle_dump_textures->setEnabled(Settings::values.dump_textures.UsingGlobal());
         ui->toggle_custom_textures->setEnabled(Settings::values.custom_textures.UsingGlobal());
         ui->toggle_preload_textures->setEnabled(Settings::values.preload_textures.UsingGlobal());
+        ui->toggle_async_custom_loading->setEnabled(
+            Settings::values.async_custom_loading.UsingGlobal());
         return;
     }
 
@@ -195,6 +202,9 @@ void ConfigureEnhancements::SetupPerGameUI() {
                                             Settings::values.custom_textures, custom_textures);
     ConfigurationShared::SetColoredTristate(ui->toggle_preload_textures,
                                             Settings::values.preload_textures, preload_textures);
+    ConfigurationShared::SetColoredTristate(ui->toggle_async_custom_loading,
+                                            Settings::values.async_custom_loading,
+                                            async_custom_loading);
 
     ConfigurationShared::SetColoredComboBox(
         ui->resolution_factor_combobox, ui->widget_resolution,

src/citra_qt/configuration/configure_enhancements.h

@@ -44,5 +44,6 @@ private:
     ConfigurationShared::CheckState dump_textures;
     ConfigurationShared::CheckState custom_textures;
     ConfigurationShared::CheckState preload_textures;
+    ConfigurationShared::CheckState async_custom_loading;
     QColor bg_color;
 };

src/citra_qt/configuration/configure_enhancements.ui

@@ -494,6 +494,16 @@
         </property>
        </widget>
       </item>
+      <item>
+       <widget class="QCheckBox" name="toggle_async_custom_loading">
+        <property name="toolTip">
+         <string>&lt;html&gt;&lt;head/&gt;&lt;body&gt;&lt;p&gt;Load custom textures asynchronously with background threads to reduce loading stutter&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</string>
+        </property>
+        <property name="text">
+         <string>Async Custom Texture Loading</string>
+        </property>
+       </widget>
+      </item>
      </layout>
     </widget>
    </item>

src/citra_qt/dumping/option_set_dialog.cpp

@@ -206,13 +206,11 @@ void OptionSetDialog::SetCheckBoxDefaults(const std::string& initial_value) {
         }
     } else {
         // This is a combination of constants, splitted with + or |
-        std::vector<std::string> tmp;
-        Common::SplitString(initial_value, '+', tmp);
+        const auto tmp = Common::SplitString(initial_value, '+');
 
         std::vector<std::string> out;
-        std::vector<std::string> tmp2;
         for (const auto& str : tmp) {
-            Common::SplitString(str, '|', tmp2);
+            const auto tmp2 = Common::SplitString(str, '|');
             out.insert(out.end(), tmp2.begin(), tmp2.end());
         }
         for (int i = 0; i < ui->checkBoxLayout->count(); ++i) {

src/citra_qt/hotkeys.cpp

@@ -2,13 +2,13 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <QKeySequence>
 #include <QShortcut>
 #include <QtGlobal>
 #include "citra_qt/hotkeys.h"
 #include "citra_qt/uisettings.h"
 
 HotkeyRegistry::HotkeyRegistry() = default;
+
 HotkeyRegistry::~HotkeyRegistry() = default;
 
 void HotkeyRegistry::SaveHotkeys() {

src/citra_qt/hotkeys.h

@@ -5,11 +5,13 @@
 #pragma once
 
 #include <map>
+#include <QKeySequence>
+#include <QString>
 
 class QDialog;
-class QKeySequence;
 class QSettings;
 class QShortcut;
+class QWidget;
 
 class HotkeyRegistry final {
 public:

src/citra_qt/main.cpp

@@ -578,6 +578,8 @@ void GMainWindow::InitializeHotkeys() {
     });
     connect_shortcut(QStringLiteral("Toggle Texture Dumping"),
                      [&] { Settings::values.dump_textures = !Settings::values.dump_textures; });
+    connect_shortcut(QStringLiteral("Toggle Custom Textures"),
+                     [&] { Settings::values.custom_textures = !Settings::values.custom_textures; });
     // We use "static" here in order to avoid capturing by lambda due to a MSVC bug, which makes
     // the variable hold a garbage value after this function exits
     static constexpr u16 SPEED_LIMIT_STEP = 5;

src/citra_qt/qt_image_interface.cpp

@@ -8,11 +8,10 @@
 #include "common/logging/log.h"
 
 bool QtImageInterface::DecodePNG(std::vector<u8>& dst, u32& width, u32& height,
-                                 const std::string& path) {
-    QImage image(QString::fromStdString(path));
-
+                                 std::span<const u8> src) {
+    QImage image(QImage::fromData(src.data(), static_cast<int>(src.size())));
     if (image.isNull()) {
-        LOG_ERROR(Frontend, "Failed to open {} for decoding", path);
+        LOG_ERROR(Frontend, "Failed to decode png because image is null");
         return false;
     }
     width = image.width();
@@ -21,13 +20,15 @@ bool QtImageInterface::DecodePNG(std::vector<u8>& dst, u32& width, u32& height,
     image = image.convertToFormat(QImage::Format_RGBA8888);
 
     // Write RGBA8 to vector
-    dst = std::vector<u8>(image.constBits(), image.constBits() + (width * height * 4));
+    const size_t image_size = width * height * 4;
+    dst.resize(image_size);
+    std::memcpy(dst.data(), image.constBits(), image_size);
 
     return true;
 }
 
-bool QtImageInterface::EncodePNG(const std::string& path, const std::vector<u8>& src, u32 width,
-                                 u32 height) {
+bool QtImageInterface::EncodePNG(const std::string& path, u32 width, u32 height,
+                                 std::span<const u8> src) {
     QImage image(src.data(), width, height, QImage::Format_RGBA8888);
 
     if (!image.save(QString::fromStdString(path), "PNG")) {

src/citra_qt/qt_image_interface.h

@@ -8,7 +8,7 @@
 
 class QtImageInterface final : public Frontend::ImageInterface {
 public:
-    bool DecodePNG(std::vector<u8>& dst, u32& width, u32& height, const std::string& path) override;
-    bool EncodePNG(const std::string& path, const std::vector<u8>& src, u32 width,
-                   u32 height) override;
+    bool DecodePNG(std::vector<u8>& dst, u32& width, u32& height, std::span<const u8> src) override;
+    bool EncodePNG(const std::string& path, u32 width, u32 height,
+                   std::span<const u8> src) override;
 };

src/common/CMakeLists.txt

@@ -65,6 +65,8 @@ add_library(common STATIC
     common_precompiled_headers.h
     common_types.h
     construct.h
+    error.cpp
+    error.h
     file_util.cpp
     file_util.h
     hash.h
@@ -89,6 +91,7 @@ add_library(common STATIC
     misc.cpp
     param_package.cpp
     param_package.h
+    polyfill_thread.h
     precompiled_headers.h
     quaternion.h
     ring_buffer.h
@@ -113,9 +116,11 @@ add_library(common STATIC
     thread.cpp
     thread.h
     thread_queue_list.h
+    thread_worker.h
     threadsafe_queue.h
     timer.cpp
     timer.h
+    unique_function.h
     vector_math.h
     web_result.h
     x64/cpu_detect.cpp

src/common/error.cpp

@@ -0,0 +1,57 @@
+// Copyright 2013 Dolphin Emulator Project / 2014 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <cstddef>
+#ifdef _WIN32
+#include <windows.h>
+#else
+#include <cerrno>
+#include <cstring>
+#endif
+
+#include "common/error.h"
+
+namespace Common {
+
+std::string NativeErrorToString(int e) {
+#ifdef _WIN32
+    LPSTR err_str;
+
+    DWORD res = FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_ALLOCATE_BUFFER |
+                                   FORMAT_MESSAGE_IGNORE_INSERTS,
+                               nullptr, e, MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),
+                               reinterpret_cast<LPSTR>(&err_str), 1, nullptr);
+    if (!res) {
+        return "(FormatMessageA failed to format error)";
+    }
+    std::string ret(err_str);
+    LocalFree(err_str);
+    return ret;
+#else
+    char err_str[255];
+#if defined(__GLIBC__) && (_GNU_SOURCE || (_POSIX_C_SOURCE < 200112L && _XOPEN_SOURCE < 600)) ||   \
+    defined(ANDROID)
+    // Thread safe (GNU-specific)
+    const char* str = strerror_r(e, err_str, sizeof(err_str));
+    return std::string(str);
+#else
+    // Thread safe (XSI-compliant)
+    int second_err = strerror_r(e, err_str, sizeof(err_str));
+    if (second_err != 0) {
+        return "(strerror_r failed to format error)";
+    }
+    return std::string(err_str);
+#endif // GLIBC etc.
+#endif // _WIN32
+}
+
+std::string GetLastErrorMsg() {
+#ifdef _WIN32
+    return NativeErrorToString(GetLastError());
+#else
+    return NativeErrorToString(errno);
+#endif
+}
+
+} // namespace Common

src/common/error.h

@@ -0,0 +1,21 @@
+// Copyright 2013 Dolphin Emulator Project / 2014 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <string>
+
+namespace Common {
+
+// Generic function to get last error message.
+// Call directly after the command or use the error num.
+// This function might change the error code.
+// Defined in error.cpp.
+[[nodiscard]] std::string GetLastErrorMsg();
+
+// Like GetLastErrorMsg(), but passing an explicit error code.
+// Defined in error.cpp.
+[[nodiscard]] std::string NativeErrorToString(int e);
+
+} // namespace Common

src/common/file_util.cpp

@@ -960,7 +960,7 @@ std::string_view GetFilename(std::string_view path) {
     const auto name_index = path.find_last_of("\\/");
 
     if (name_index == std::string_view::npos) {
-        return {};
+        return path;
     }
 
     return path.substr(name_index + 1);

src/common/param_package.cpp

@@ -28,12 +28,9 @@ ParamPackage::ParamPackage(const std::string& serialized) {
         return;
     }
 
-    std::vector<std::string> pairs;
-    Common::SplitString(serialized, PARAM_SEPARATOR, pairs);
-
+    const auto pairs = Common::SplitString(serialized, PARAM_SEPARATOR);
     for (const std::string& pair : pairs) {
-        std::vector<std::string> key_value;
-        Common::SplitString(pair, KEY_VALUE_SEPARATOR, key_value);
+        auto key_value = Common::SplitString(pair, KEY_VALUE_SEPARATOR);
         if (key_value.size() != 2) {
             LOG_ERROR(Common, "invalid key pair {}", pair);
             continue;

src/common/polyfill_thread.h

@@ -0,0 +1,338 @@
+// Copyright 2022 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+//
+// TODO: remove this file when jthread is supported by all compilation targets
+//
+
+#pragma once
+
+#include <version>
+
+#ifdef __cpp_lib_jthread
+
+#include <stop_token>
+#include <thread>
+
+namespace Common {
+
+template <typename Condvar, typename Lock, typename Pred>
+void CondvarWait(Condvar& cv, Lock& lock, std::stop_token token, Pred&& pred) {
+    cv.wait(lock, token, std::move(pred));
+}
+
+} // namespace Common
+
+#else
+
+#include <atomic>
+#include <functional>
+#include <map>
+#include <memory>
+#include <mutex>
+#include <optional>
+#include <thread>
+#include <type_traits>
+#include <utility>
+
+namespace std {
+namespace polyfill {
+
+using stop_state_callback = size_t;
+
+class stop_state {
+public:
+    stop_state() = default;
+    ~stop_state() = default;
+
+    bool request_stop() {
+        unique_lock lk{m_lock};
+
+        if (m_stop_requested) {
+            // Already set, nothing to do.
+            return false;
+        }
+
+        // Mark stop requested.
+        m_stop_requested = true;
+
+        while (!m_callbacks.empty()) {
+            // Get an iterator to the first element.
+            const auto it = m_callbacks.begin();
+
+            // Move the callback function out of the map.
+            function<void()> f;
+            swap(it->second, f);
+
+            // Erase the now-empty map element.
+            m_callbacks.erase(it);
+
+            // Run the callback.
+            if (f) {
+                f();
+            }
+        }
+
+        return true;
+    }
+
+    bool stop_requested() const {
+        unique_lock lk{m_lock};
+        return m_stop_requested;
+    }
+
+    stop_state_callback insert_callback(function<void()> f) {
+        unique_lock lk{m_lock};
+
+        if (m_stop_requested) {
+            // Stop already requested. Don't insert anything,
+            // just run the callback synchronously.
+            if (f) {
+                f();
+            }
+            return 0;
+        }
+
+        // Insert the callback.
+        stop_state_callback ret = ++m_next_callback;
+        m_callbacks.emplace(ret, move(f));
+        return ret;
+    }
+
+    void remove_callback(stop_state_callback cb) {
+        unique_lock lk{m_lock};
+        m_callbacks.erase(cb);
+    }
+
+private:
+    mutable recursive_mutex m_lock;
+    map<stop_state_callback, function<void()>> m_callbacks;
+    stop_state_callback m_next_callback{0};
+    bool m_stop_requested{false};
+};
+
+} // namespace polyfill
+
+#ifndef __cpp_lib_concepts
+template <class T, class... Args>
+concept constructible_from = is_nothrow_destructible_v<T> && is_constructible_v<T, Args...>;
+#endif
+
+class stop_token;
+class stop_source;
+struct nostopstate_t {
+    explicit nostopstate_t() = default;
+};
+inline constexpr nostopstate_t nostopstate{};
+
+template <class Callback>
+class stop_callback;
+
+class stop_token {
+public:
+    stop_token() noexcept = default;
+
+    stop_token(const stop_token&) noexcept = default;
+    stop_token(stop_token&&) noexcept = default;
+    stop_token& operator=(const stop_token&) noexcept = default;
+    stop_token& operator=(stop_token&&) noexcept = default;
+    ~stop_token() = default;
+
+    void swap(stop_token& other) noexcept {
+        m_stop_state.swap(other.m_stop_state);
+    }
+
+    [[nodiscard]] bool stop_requested() const noexcept {
+        return m_stop_state && m_stop_state->stop_requested();
+    }
+    [[nodiscard]] bool stop_possible() const noexcept {
+        return m_stop_state != nullptr;
+    }
+
+private:
+    friend class stop_source;
+    template <typename Callback>
+    friend class stop_callback;
+    stop_token(shared_ptr<polyfill::stop_state> stop_state) : m_stop_state(move(stop_state)) {}
+
+private:
+    shared_ptr<polyfill::stop_state> m_stop_state;
+};
+
+class stop_source {
+public:
+    stop_source() : m_stop_state(make_shared<polyfill::stop_state>()) {}
+    explicit stop_source(nostopstate_t) noexcept {}
+
+    stop_source(const stop_source&) noexcept = default;
+    stop_source(stop_source&&) noexcept = default;
+    stop_source& operator=(const stop_source&) noexcept = default;
+    stop_source& operator=(stop_source&&) noexcept = default;
+    ~stop_source() = default;
+    void swap(stop_source& other) noexcept {
+        m_stop_state.swap(other.m_stop_state);
+    }
+
+    [[nodiscard]] stop_token get_token() const noexcept {
+        return stop_token(m_stop_state);
+    }
+    [[nodiscard]] bool stop_possible() const noexcept {
+        return m_stop_state != nullptr;
+    }
+    [[nodiscard]] bool stop_requested() const noexcept {
+        return m_stop_state && m_stop_state->stop_requested();
+    }
+    bool request_stop() noexcept {
+        return m_stop_state && m_stop_state->request_stop();
+    }
+
+private:
+    friend class jthread;
+    explicit stop_source(shared_ptr<polyfill::stop_state> stop_state)
+        : m_stop_state(move(stop_state)) {}
+
+private:
+    shared_ptr<polyfill::stop_state> m_stop_state;
+};
+
+template <typename Callback>
+class stop_callback {
+    static_assert(is_nothrow_destructible_v<Callback>);
+    static_assert(is_invocable_v<Callback>);
+
+public:
+    using callback_type = Callback;
+
+    template <typename C>
+    requires constructible_from<Callback, C>
+    explicit stop_callback(const stop_token& st,
+                           C&& cb) noexcept(is_nothrow_constructible_v<Callback, C>)
+        : m_stop_state(st.m_stop_state) {
+        if (m_stop_state) {
+            m_callback = m_stop_state->insert_callback(move(cb));
+        }
+    }
+    template <typename C>
+    requires constructible_from<Callback, C>
+    explicit stop_callback(stop_token&& st,
+                           C&& cb) noexcept(is_nothrow_constructible_v<Callback, C>)
+        : m_stop_state(move(st.m_stop_state)) {
+        if (m_stop_state) {
+            m_callback = m_stop_state->insert_callback(move(cb));
+        }
+    }
+    ~stop_callback() {
+        if (m_stop_state && m_callback) {
+            m_stop_state->remove_callback(m_callback);
+        }
+    }
+
+    stop_callback(const stop_callback&) = delete;
+    stop_callback(stop_callback&&) = delete;
+    stop_callback& operator=(const stop_callback&) = delete;
+    stop_callback& operator=(stop_callback&&) = delete;
+
+private:
+    shared_ptr<polyfill::stop_state> m_stop_state;
+    polyfill::stop_state_callback m_callback;
+};
+
+template <typename Callback>
+stop_callback(stop_token, Callback) -> stop_callback<Callback>;
+
+class jthread {
+public:
+    using id = thread::id;
+    using native_handle_type = thread::native_handle_type;
+
+    jthread() noexcept = default;
+
+    template <typename F, typename... Args,
+              typename = enable_if_t<!is_same_v<remove_cvref_t<F>, jthread>>>
+    explicit jthread(F&& f, Args&&... args)
+        : m_stop_state(make_shared<polyfill::stop_state>()),
+          m_thread(make_thread(move(f), move(args)...)) {}
+
+    ~jthread() {
+        if (joinable()) {
+            request_stop();
+            join();
+        }
+    }
+
+    jthread(const jthread&) = delete;
+    jthread(jthread&&) noexcept = default;
+    jthread& operator=(const jthread&) = delete;
+
+    jthread& operator=(jthread&& other) noexcept {
+        m_thread.swap(other.m_thread);
+        m_stop_state.swap(other.m_stop_state);
+        return *this;
+    }
+
+    void swap(jthread& other) noexcept {
+        m_thread.swap(other.m_thread);
+        m_stop_state.swap(other.m_stop_state);
+    }
+    [[nodiscard]] bool joinable() const noexcept {
+        return m_thread.joinable();
+    }
+    void join() {
+        m_thread.join();
+    }
+    void detach() {
+        m_thread.detach();
+        m_stop_state.reset();
+    }
+
+    [[nodiscard]] id get_id() const noexcept {
+        return m_thread.get_id();
+    }
+    [[nodiscard]] native_handle_type native_handle() {
+        return m_thread.native_handle();
+    }
+    [[nodiscard]] stop_source get_stop_source() noexcept {
+        return stop_source(m_stop_state);
+    }
+    [[nodiscard]] stop_token get_stop_token() const noexcept {
+        return stop_source(m_stop_state).get_token();
+    }
+    bool request_stop() noexcept {
+        return get_stop_source().request_stop();
+    }
+    [[nodiscard]] static unsigned int hardware_concurrency() noexcept {
+        return thread::hardware_concurrency();
+    }
+
+private:
+    template <typename F, typename... Args>
+    thread make_thread(F&& f, Args&&... args) {
+        if constexpr (is_invocable_v<decay_t<F>, stop_token, decay_t<Args>...>) {
+            return thread(move(f), get_stop_token(), move(args)...);
+        } else {
+            return thread(move(f), move(args)...);
+        }
+    }
+
+    shared_ptr<polyfill::stop_state> m_stop_state;
+    thread m_thread;
+};
+
+} // namespace std
+
+namespace Common {
+
+template <typename Condvar, typename Lock, typename Pred>
+void CondvarWait(Condvar& cv, Lock& lock, std::stop_token token, Pred pred) {
+    if (token.stop_requested()) {
+        return;
+    }
+
+    std::stop_callback callback(token, [&] { cv.notify_all(); });
+    cv.wait(lock, [&] { return pred() || token.stop_requested(); });
+}
+
+} // namespace Common
+
+#endif // __cpp_lib_jthread

src/common/settings.h

@@ -477,6 +477,7 @@ struct Values {
     SwitchableSetting<bool> dump_textures{false, "dump_textures"};
     SwitchableSetting<bool> custom_textures{false, "custom_textures"};
     SwitchableSetting<bool> preload_textures{false, "preload_textures"};
+    SwitchableSetting<bool> async_custom_loading{true, "async_custom_loading"};
 
     // Audio
     bool audio_muted;

src/common/string_util.cpp

@@ -102,15 +102,16 @@ void BuildCompleteFilename(std::string& _CompleteFilename, const std::string& _P
     _CompleteFilename += _Filename;
 }
 
-void SplitString(const std::string& str, const char delim, std::vector<std::string>& output) {
+std::vector<std::string> SplitString(const std::string& str, const char delim) {
     std::istringstream iss(str);
-    output.resize(1);
+    std::vector<std::string> output(1);
 
     while (std::getline(iss, *output.rbegin(), delim)) {
         output.emplace_back();
     }
 
     output.pop_back();
+    return output;
 }
 
 std::string TabsToSpaces(int tab_size, std::string in) {

src/common/string_util.h

@@ -29,7 +29,7 @@ namespace Common {
 
 [[nodiscard]] bool EndsWith(const std::string& value, const std::string& ending);
 
-void SplitString(const std::string& str, char delim, std::vector<std::string>& output);
+[[nodiscard]] std::vector<std::string> SplitString(const std::string& str, const char delim);
 
 // "C:/Windows/winhelp.exe" to "C:/Windows/", "winhelp", ".exe"
 bool SplitPath(const std::string& full_path, std::string* _pPath, std::string* _pFilename,

src/common/texture.cpp

@@ -3,12 +3,12 @@
 // Refer to the license.txt file included.
 
 #include <algorithm>
-#include <vector>
 #include "common/assert.h"
-#include "common/common_types.h"
+#include "common/texture.h"
 
 namespace Common {
-void FlipRGBA8Texture(std::vector<u8>& tex, u32 width, u32 height) {
+
+void FlipRGBA8Texture(std::span<u8> tex, u32 width, u32 height) {
     ASSERT(tex.size() == width * height * 4);
     const u32 line_size = width * 4;
     for (u32 line = 0; line < height / 2; line++) {
@@ -19,4 +19,5 @@ void FlipRGBA8Texture(std::vector<u8>& tex, u32 width, u32 height) {
                          tex.begin() + offset_2);
     }
 }
+
 } // namespace Common

src/common/texture.h

@@ -4,9 +4,11 @@
 
 #pragma once
 
-#include <vector>
+#include <span>
 #include "common/common_types.h"
 
 namespace Common {
-void FlipRGBA8Texture(std::vector<u8>& tex, u32 width, u32 height);
-}
+
+void FlipRGBA8Texture(std::span<u8> tex, u32 width, u32 height);
+
+} // namespace Common

src/common/thread.cpp

@@ -2,7 +2,9 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include "common/common_funcs.h"
+#include <string>
+
+#include "common/error.h"
 #include "common/logging/log.h"
 #include "common/thread.h"
 #ifdef __APPLE__
@@ -20,7 +22,6 @@
 #ifndef _WIN32
 #include <unistd.h>
 #endif
-#include <string>
 
 #ifdef __FreeBSD__
 #define cpu_set_t cpuset_t
@@ -28,6 +29,56 @@
 
 namespace Common {
 
+#ifdef _WIN32
+
+void SetCurrentThreadPriority(ThreadPriority new_priority) {
+    auto handle = GetCurrentThread();
+    int windows_priority = 0;
+    switch (new_priority) {
+    case ThreadPriority::Low:
+        windows_priority = THREAD_PRIORITY_BELOW_NORMAL;
+        break;
+    case ThreadPriority::Normal:
+        windows_priority = THREAD_PRIORITY_NORMAL;
+        break;
+    case ThreadPriority::High:
+        windows_priority = THREAD_PRIORITY_ABOVE_NORMAL;
+        break;
+    case ThreadPriority::VeryHigh:
+        windows_priority = THREAD_PRIORITY_HIGHEST;
+        break;
+    case ThreadPriority::Critical:
+        windows_priority = THREAD_PRIORITY_TIME_CRITICAL;
+        break;
+    default:
+        windows_priority = THREAD_PRIORITY_NORMAL;
+        break;
+    }
+    SetThreadPriority(handle, windows_priority);
+}
+
+#else
+
+void SetCurrentThreadPriority(ThreadPriority new_priority) {
+    pthread_t this_thread = pthread_self();
+
+    const auto scheduling_type = SCHED_OTHER;
+    s32 max_prio = sched_get_priority_max(scheduling_type);
+    s32 min_prio = sched_get_priority_min(scheduling_type);
+    u32 level = std::max(static_cast<u32>(new_priority) + 1, 4U);
+
+    struct sched_param params;
+    if (max_prio > min_prio) {
+        params.sched_priority = min_prio + ((max_prio - min_prio) * level) / 4;
+    } else {
+        params.sched_priority = min_prio - ((min_prio - max_prio) * level) / 4;
+    }
+
+    pthread_setschedparam(this_thread, scheduling_type, &params);
+}
+
+#endif
+
 #ifdef _MSC_VER
 
 // Sets the debugger-visible name of the current thread.
@@ -47,7 +98,7 @@ void SetCurrentThreadName(const char* name) {
 
     info.dwType = 0x1000;
     info.szName = name;
-    info.dwThreadID = static_cast<DWORD>(-1);
+    info.dwThreadID = std::numeric_limits<DWORD>::max();
     info.dwFlags = 0;
 
     __try {
@@ -81,6 +132,12 @@ void SetCurrentThreadName(const char* name) {
 }
 #endif
 
+#if defined(_WIN32)
+void SetCurrentThreadName(const char*) {
+    // Do Nothing on MingW
+}
+#endif
+
 #endif
 
 } // namespace Common

src/common/thread.h

@@ -10,13 +10,15 @@
 #include <cstddef>
 #include <mutex>
 #include <thread>
+#include "common/common_types.h"
+#include "common/polyfill_thread.h"
 
 namespace Common {
 
 class Event {
 public:
     void Set() {
-        std::lock_guard lk{mutex};
+        std::scoped_lock lk{mutex};
         if (!is_set) {
             is_set = true;
             condvar.notify_one();
@@ -54,6 +56,10 @@ public:
         is_set = false;
     }
 
+    [[nodiscard]] bool IsSet() {
+        return is_set;
+    }
+
 private:
     std::condition_variable condvar;
     std::mutex mutex;
@@ -65,7 +71,7 @@ public:
     explicit Barrier(std::size_t count_) : count(count_) {}
 
     /// Blocks until all "count" threads have called Sync()
-    void Sync() {
+    bool Sync(std::stop_token token = {}) {
         std::unique_lock lk{mutex};
         const std::size_t current_generation = generation;
 
@@ -73,25 +79,37 @@ public:
             generation++;
             waiting = 0;
             condvar.notify_all();
+            return true;
         } else {
-            condvar.wait(lk,
+            CondvarWait(condvar, lk, token,
                         [this, current_generation] { return current_generation != generation; });
+            return !token.stop_requested();
         }
     }
 
-    std::size_t Generation() const {
-        std::unique_lock lk(mutex);
+    std::size_t Generation() {
+        std::unique_lock lk{mutex};
         return generation;
     }
 
 private:
-    std::condition_variable condvar;
-    mutable std::mutex mutex;
+    std::condition_variable_any condvar;
+    std::mutex mutex;
     std::size_t count;
     std::size_t waiting = 0;
     std::size_t generation = 0; // Incremented once each time the barrier is used
 };
 
+enum class ThreadPriority : u32 {
+    Low = 0,
+    Normal = 1,
+    High = 2,
+    VeryHigh = 3,
+    Critical = 4,
+};
+
+void SetCurrentThreadPriority(ThreadPriority new_priority);
+
 void SetCurrentThreadName(const char* name);
 
 } // namespace Common

src/common/thread_worker.h

@@ -0,0 +1,123 @@
+// Copyright 2020 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <atomic>
+#include <condition_variable>
+#include <functional>
+#include <mutex>
+#include <string>
+#include <thread>
+#include <type_traits>
+#include <vector>
+#include <queue>
+
+#include "common/polyfill_thread.h"
+#include "common/thread.h"
+#include "common/unique_function.h"
+
+namespace Common {
+
+template <class StateType = void>
+class StatefulThreadWorker {
+    static constexpr bool with_state = !std::is_same_v<StateType, void>;
+
+    struct DummyCallable {
+        int operator()(size_t) const noexcept {
+            return 0;
+        }
+    };
+
+    using Task =
+        std::conditional_t<with_state, UniqueFunction<void, StateType*>, UniqueFunction<void>>;
+    using StateMaker =
+        std::conditional_t<with_state, std::function<StateType(size_t)>, DummyCallable>;
+
+public:
+    explicit StatefulThreadWorker(size_t num_workers, std::string_view name, StateMaker func = {})
+        : workers_queued{num_workers}, thread_name{name} {
+        const auto lambda = [this, func](std::stop_token stop_token, size_t index) {
+            Common::SetCurrentThreadName(thread_name.data());
+            {
+                [[maybe_unused]] std::conditional_t<with_state, StateType, int> state{func(index)};
+                while (!stop_token.stop_requested()) {
+                    Task task;
+                    {
+                        std::unique_lock lock{queue_mutex};
+                        if (requests.empty()) {
+                            wait_condition.notify_all();
+                        }
+                        Common::CondvarWait(condition, lock, stop_token,
+                                            [this] { return !requests.empty(); });
+                        if (stop_token.stop_requested()) {
+                            break;
+                        }
+                        task = std::move(requests.front());
+                        requests.pop();
+                    }
+                    if constexpr (with_state) {
+                        task(&state);
+                    } else {
+                        task();
+                    }
+                    ++work_done;
+                }
+            }
+            ++workers_stopped;
+            wait_condition.notify_all();
+        };
+        threads.reserve(num_workers);
+        for (size_t i = 0; i < num_workers; ++i) {
+            threads.emplace_back(lambda, i);
+        }
+    }
+
+    StatefulThreadWorker& operator=(const StatefulThreadWorker&) = delete;
+    StatefulThreadWorker(const StatefulThreadWorker&) = delete;
+
+    StatefulThreadWorker& operator=(StatefulThreadWorker&&) = delete;
+    StatefulThreadWorker(StatefulThreadWorker&&) = delete;
+
+    void QueueWork(Task work) {
+        {
+            std::unique_lock lock{queue_mutex};
+            requests.emplace(std::move(work));
+            ++work_scheduled;
+        }
+        condition.notify_one();
+    }
+
+    void WaitForRequests(std::stop_token stop_token = {}) {
+        std::stop_callback callback(stop_token, [this] {
+            for (auto& thread : threads) {
+                thread.request_stop();
+            }
+        });
+        std::unique_lock lock{queue_mutex};
+        wait_condition.wait(lock, [this] {
+            return workers_stopped >= workers_queued || work_done >= work_scheduled;
+        });
+    }
+
+    const std::size_t NumWorkers() const noexcept {
+        return threads.size();
+    }
+
+private:
+    std::queue<Task> requests;
+    std::mutex queue_mutex;
+    std::condition_variable_any condition;
+    std::condition_variable wait_condition;
+    std::atomic<size_t> work_scheduled{};
+    std::atomic<size_t> work_done{};
+    std::atomic<size_t> workers_stopped{};
+    std::atomic<size_t> workers_queued{};
+    std::string_view thread_name;
+    std::vector<std::jthread> threads;
+};
+
+using ThreadWorker = StatefulThreadWorker<>;
+
+} // namespace Common

src/common/unique_function.h

@@ -0,0 +1,62 @@
+// Copyright 2021 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <memory>
+#include <utility>
+
+namespace Common {
+
+/// General purpose function wrapper similar to std::function.
+/// Unlike std::function, the captured values don't have to be copyable.
+/// This class can be moved but not copied.
+template <typename ResultType, typename... Args>
+class UniqueFunction {
+    class CallableBase {
+    public:
+        virtual ~CallableBase() = default;
+        virtual ResultType operator()(Args&&...) = 0;
+    };
+
+    template <typename Functor>
+    class Callable final : public CallableBase {
+    public:
+        Callable(Functor&& functor_) : functor{std::move(functor_)} {}
+        ~Callable() override = default;
+
+        ResultType operator()(Args&&... args) override {
+            return functor(std::forward<Args>(args)...);
+        }
+
+    private:
+        Functor functor;
+    };
+
+public:
+    UniqueFunction() = default;
+
+    template <typename Functor>
+    UniqueFunction(Functor&& functor)
+        : callable{std::make_unique<Callable<Functor>>(std::move(functor))} {}
+
+    UniqueFunction& operator=(UniqueFunction&& rhs) noexcept = default;
+    UniqueFunction(UniqueFunction&& rhs) noexcept = default;
+
+    UniqueFunction& operator=(const UniqueFunction&) = delete;
+    UniqueFunction(const UniqueFunction&) = delete;
+
+    ResultType operator()(Args&&... args) const {
+        return (*callable)(std::forward<Args>(args)...);
+    }
+
+    explicit operator bool() const noexcept {
+        return static_cast<bool>(callable);
+    }
+
+private:
+    std::unique_ptr<CallableBase> callable;
+};
+
+} // namespace Common

src/core/CMakeLists.txt

@@ -36,8 +36,6 @@ add_library(core STATIC
     core.h
     core_timing.cpp
     core_timing.h
-    custom_tex_cache.cpp
-    custom_tex_cache.h
     dumping/backend.cpp
     dumping/backend.h
     file_sys/archive_backend.cpp
@@ -109,6 +107,7 @@ add_library(core STATIC
     frontend/emu_window.h
     frontend/framebuffer_layout.cpp
     frontend/framebuffer_layout.h
+    frontend/image_interface.cpp
     frontend/image_interface.h
     frontend/input.h
     frontend/mic.cpp
@@ -481,7 +480,8 @@ endif()
 create_target_directory_groups(core)
 
 target_link_libraries(core PUBLIC common PRIVATE audio_core network video_core)
-target_link_libraries(core PUBLIC Boost::boost PRIVATE cryptopp fmt::fmt open_source_archives Boost::serialization Boost::iostreams)
+target_link_libraries(core PRIVATE Boost::boost Boost::serialization Boost::iostreams)
+target_link_libraries(core PUBLIC dds-ktx PRIVATE cryptopp fmt::fmt lodepng open_source_archives)
 set_target_properties(core PROPERTIES INTERPROCEDURAL_OPTIMIZATION ${ENABLE_LTO})
 
 if (ENABLE_WEB_SERVICE)

src/core/cheats/gateway_cheat.cpp

@@ -216,13 +216,12 @@ GatewayCheat::GatewayCheat(std::string name_, std::vector<CheatLine> cheat_lines
 GatewayCheat::GatewayCheat(std::string name_, std::string code, std::string comments_)
     : name(std::move(name_)), comments(std::move(comments_)) {
 
-    std::vector<std::string> code_lines;
-    Common::SplitString(code, '\n', code_lines);
+    const auto code_lines = Common::SplitString(code, '\n');
 
     std::vector<CheatLine> temp_cheat_lines;
-    for (std::size_t i = 0; i < code_lines.size(); ++i) {
-        if (!code_lines[i].empty())
-            temp_cheat_lines.emplace_back(code_lines[i]);
+    for (const std::string& line : code_lines) {
+        if (!line.empty())
+            temp_cheat_lines.emplace_back(line);
     }
     cheat_lines = std::move(temp_cheat_lines);
 }
@@ -464,10 +463,10 @@ std::string GatewayCheat::ToString() const {
         result += EnabledText;
         result += '\n';
     }
-    std::vector<std::string> comment_lines;
-    Common::SplitString(comments, '\n', comment_lines);
-    for (const auto& comment_line : comment_lines)
+    const auto comment_lines = Common::SplitString(comments, '\n');
+    for (const auto& comment_line : comment_lines) {
         result += "*" + comment_line + '\n';
+    }
     result += GetCode() + '\n';
     return result;
 }

src/core/core.cpp

@@ -27,7 +27,7 @@
 #include "core/dumping/ffmpeg_backend.h"
 #endif
 #include "common/settings.h"
-#include "core/custom_tex_cache.h"
+#include "core/frontend/image_interface.h"
 #include "core/gdbstub/gdbstub.h"
 #include "core/global.h"
 #include "core/hle/kernel/client_port.h"
@@ -48,6 +48,7 @@
 #include "core/movie.h"
 #include "core/rpc/rpc_server.h"
 #include "network/network.h"
+#include "video_core/custom_textures/custom_tex_manager.h"
 #include "video_core/renderer_base.h"
 #include "video_core/video_core.h"
 
@@ -318,16 +319,15 @@ System::ResultStatus System::Load(Frontend::EmuWindow& emu_window, const std::st
                   static_cast<u32>(load_result));
     }
     perf_stats = std::make_unique<PerfStats>(title_id);
-    custom_tex_cache = std::make_unique<Core::CustomTexCache>();
 
     if (Settings::values.custom_textures) {
-        const u64 program_id = Kernel().GetCurrentProcess()->codeset->program_id;
-        FileUtil::CreateFullPath(fmt::format(
-            "{}textures/{:016X}/", FileUtil::GetUserPath(FileUtil::UserPath::LoadDir), program_id));
-        custom_tex_cache->FindCustomTextures(program_id);
+        custom_tex_manager->FindCustomTextures();
     }
     if (Settings::values.preload_textures) {
-        custom_tex_cache->PreloadTextures(*GetImageInterface());
+        custom_tex_manager->PreloadTextures();
+    }
+    if (Settings::values.dump_textures) {
+        custom_tex_manager->WriteConfig();
     }
 
     status = ResultStatus::Success;
@@ -432,6 +432,12 @@ System::ResultStatus System::Init(Frontend::EmuWindow& emu_window,
     video_dumper = std::make_unique<VideoDumper::NullBackend>();
 #endif
 
+    if (!registered_image_interface) {
+        registered_image_interface = std::make_shared<Frontend::ImageInterface>();
+    }
+
+    custom_tex_manager = std::make_unique<VideoCore::CustomTexManager>(*this);
+
     VideoCore::Init(emu_window, secondary_window, *this);
 
     LOG_DEBUG(Core, "Initialized OK");
@@ -505,12 +511,12 @@ const VideoDumper::Backend& System::VideoDumper() const {
     return *video_dumper;
 }
 
-Core::CustomTexCache& System::CustomTexCache() {
-    return *custom_tex_cache;
+VideoCore::CustomTexManager& System::CustomTexManager() {
+    return *custom_tex_manager;
 }
 
-const Core::CustomTexCache& System::CustomTexCache() const {
-    return *custom_tex_cache;
+const VideoCore::CustomTexManager& System::CustomTexManager() const {
+    return *custom_tex_manager;
 }
 
 void System::RegisterMiiSelector(std::shared_ptr<Frontend::MiiSelector> mii_selector) {

src/core/core.h

@@ -10,10 +10,8 @@
 #include <string>
 #include <boost/serialization/version.hpp>
 #include "common/common_types.h"
-#include "core/custom_tex_cache.h"
 #include "core/frontend/applets/mii_selector.h"
 #include "core/frontend/applets/swkbd.h"
-#include "core/frontend/image_interface.h"
 #include "core/loader/loader.h"
 #include "core/memory.h"
 #include "core/perf_stats.h"
@@ -23,7 +21,8 @@ class ARM_Interface;
 
 namespace Frontend {
 class EmuWindow;
-}
+class ImageInterface;
+} // namespace Frontend
 
 namespace Memory {
 class MemorySystem;
@@ -59,8 +58,9 @@ class Backend;
 }
 
 namespace VideoCore {
+class CustomTexManager;
 class RendererBase;
-}
+} // namespace VideoCore
 
 namespace Core {
 
@@ -253,10 +253,10 @@ public:
     [[nodiscard]] const Cheats::CheatEngine& CheatEngine() const;
 
     /// Gets a reference to the custom texture cache system
-    [[nodiscard]] Core::CustomTexCache& CustomTexCache();
+    [[nodiscard]] VideoCore::CustomTexManager& CustomTexManager();
 
     /// Gets a const reference to the custom texture cache system
-    [[nodiscard]] const Core::CustomTexCache& CustomTexCache() const;
+    [[nodiscard]] const VideoCore::CustomTexManager& CustomTexManager() const;
 
     /// Gets a reference to the video dumper backend
     [[nodiscard]] VideoDumper::Backend& VideoDumper();
@@ -362,7 +362,7 @@ private:
     std::unique_ptr<VideoDumper::Backend> video_dumper;
 
     /// Custom texture cache system
-    std::unique_ptr<Core::CustomTexCache> custom_tex_cache;
+    std::unique_ptr<VideoCore::CustomTexManager> custom_tex_manager;
 
     /// Image interface
     std::shared_ptr<Frontend::ImageInterface> registered_image_interface;

src/core/custom_tex_cache.cpp

@@ -1,109 +0,0 @@
-// Copyright 2019 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#include <fmt/format.h>
-#include "common/file_util.h"
-#include "common/texture.h"
-#include "core.h"
-#include "core/custom_tex_cache.h"
-
-namespace Core {
-CustomTexCache::CustomTexCache() = default;
-
-CustomTexCache::~CustomTexCache() = default;
-
-bool CustomTexCache::IsTextureDumped(u64 hash) const {
-    return dumped_textures.count(hash);
-}
-
-void CustomTexCache::SetTextureDumped(const u64 hash) {
-    dumped_textures.insert(hash);
-}
-
-bool CustomTexCache::IsTextureCached(u64 hash) const {
-    return custom_textures.count(hash);
-}
-
-const CustomTexInfo& CustomTexCache::LookupTexture(u64 hash) const {
-    return custom_textures.at(hash);
-}
-
-void CustomTexCache::CacheTexture(u64 hash, const std::vector<u8>& tex, u32 width, u32 height) {
-    custom_textures[hash] = {width, height, tex};
-}
-
-void CustomTexCache::AddTexturePath(u64 hash, const std::string& path) {
-    if (custom_texture_paths.count(hash))
-        LOG_ERROR(Core, "Textures {} and {} conflict!", custom_texture_paths[hash].path, path);
-    else
-        custom_texture_paths[hash] = {path, hash};
-}
-
-void CustomTexCache::FindCustomTextures(u64 program_id) {
-    // Custom textures are currently stored as
-    // [TitleID]/tex1_[width]x[height]_[64-bit hash]_[format].png
-
-    const std::string load_path = fmt::format(
-        "{}textures/{:016X}/", FileUtil::GetUserPath(FileUtil::UserPath::LoadDir), program_id);
-
-    if (FileUtil::Exists(load_path)) {
-        FileUtil::FSTEntry texture_dir;
-        std::vector<FileUtil::FSTEntry> textures;
-        // 64 nested folders should be plenty for most cases
-        FileUtil::ScanDirectoryTree(load_path, texture_dir, 64);
-        FileUtil::GetAllFilesFromNestedEntries(texture_dir, textures);
-
-        for (const auto& file : textures) {
-            if (file.isDirectory)
-                continue;
-            if (file.virtualName.substr(0, 5) != "tex1_")
-                continue;
-
-            u32 width;
-            u32 height;
-            u64 hash;
-            u32 format; // unused
-            // TODO: more modern way of doing this
-            if (std::sscanf(file.virtualName.c_str(), "tex1_%ux%u_%llX_%u.png", &width, &height,
-                            &hash, &format) == 4) {
-                AddTexturePath(hash, file.physicalName);
-            }
-        }
-    }
-}
-
-void CustomTexCache::PreloadTextures(Frontend::ImageInterface& image_interface) {
-    for (const auto& path : custom_texture_paths) {
-        const auto& path_info = path.second;
-        Core::CustomTexInfo tex_info;
-        if (image_interface.DecodePNG(tex_info.tex, tex_info.width, tex_info.height,
-                                      path_info.path)) {
-            // Make sure the texture size is a power of 2
-            std::bitset<32> width_bits(tex_info.width);
-            std::bitset<32> height_bits(tex_info.height);
-            if (width_bits.count() == 1 && height_bits.count() == 1) {
-                LOG_DEBUG(Render_OpenGL, "Loaded custom texture from {}", path_info.path);
-                Common::FlipRGBA8Texture(tex_info.tex, tex_info.width, tex_info.height);
-                CacheTexture(path_info.hash, tex_info.tex, tex_info.width, tex_info.height);
-            } else {
-                LOG_ERROR(Render_OpenGL, "Texture {} size is not a power of 2", path_info.path);
-            }
-        } else {
-            LOG_ERROR(Render_OpenGL, "Failed to load custom texture {}", path_info.path);
-        }
-    }
-}
-
-bool CustomTexCache::CustomTextureExists(u64 hash) const {
-    return custom_texture_paths.count(hash);
-}
-
-const CustomTexPathInfo& CustomTexCache::LookupTexturePathInfo(u64 hash) const {
-    return custom_texture_paths.at(hash);
-}
-
-bool CustomTexCache::IsTexturePathMapEmpty() const {
-    return custom_texture_paths.size() == 0;
-}
-} // namespace Core

src/core/custom_tex_cache.h

@@ -1,55 +0,0 @@
-// Copyright 2019 Citra Emulator Project
-// Licensed under GPLv2 or any later version
-// Refer to the license.txt file included.
-
-#pragma once
-
-#include <string>
-#include <unordered_map>
-#include <unordered_set>
-#include <vector>
-#include "common/common_types.h"
-
-namespace Frontend {
-class ImageInterface;
-} // namespace Frontend
-
-namespace Core {
-struct CustomTexInfo {
-    u32 width;
-    u32 height;
-    std::vector<u8> tex;
-};
-
-// This is to avoid parsing the filename multiple times
-struct CustomTexPathInfo {
-    std::string path;
-    u64 hash;
-};
-
-// TODO: think of a better name for this class...
-class CustomTexCache {
-public:
-    explicit CustomTexCache();
-    ~CustomTexCache();
-
-    bool IsTextureDumped(u64 hash) const;
-    void SetTextureDumped(u64 hash);
-
-    bool IsTextureCached(u64 hash) const;
-    const CustomTexInfo& LookupTexture(u64 hash) const;
-    void CacheTexture(u64 hash, const std::vector<u8>& tex, u32 width, u32 height);
-
-    void AddTexturePath(u64 hash, const std::string& path);
-    void FindCustomTextures(u64 program_id);
-    void PreloadTextures(Frontend::ImageInterface& image_interface);
-    bool CustomTextureExists(u64 hash) const;
-    const CustomTexPathInfo& LookupTexturePathInfo(u64 hash) const;
-    bool IsTexturePathMapEmpty() const;
-
-private:
-    std::unordered_set<u64> dumped_textures;
-    std::unordered_map<u64, CustomTexInfo> custom_textures;
-    std::unordered_map<u64, CustomTexPathInfo> custom_texture_paths;
-};
-} // namespace Core

src/core/dumping/ffmpeg_backend.cpp

@@ -813,8 +813,7 @@ std::vector<FormatInfo> ListFormats() {
     void* data = nullptr; // For libavformat to save the iteration state
     while ((current = av_muxer_iterate(&data))) {
 #endif
-        std::vector<std::string> extensions;
-        Common::SplitString(ToStdString(current->extensions), ',', extensions);
+        const auto extensions = Common::SplitString(ToStdString(current->extensions), ',');
 
         std::set<AVCodecID> supported_video_codecs;
         std::set<AVCodecID> supported_audio_codecs;

src/core/file_sys/path_parser.cpp

@@ -32,7 +32,7 @@ PathParser::PathParser(const Path& path) {
         return;
     }
 
-    Common::SplitString(path_string, '/', path_sequence);
+    path_sequence = Common::SplitString(path_string, '/');
 
     auto begin = path_sequence.begin();
     auto end = path_sequence.end();

src/core/frontend/image_interface.cpp

@@ -0,0 +1,65 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#define DDSKTX_IMPLEMENT
+#include <dds-ktx.h>
+#include <lodepng.h>
+#include "common/file_util.h"
+#include "common/logging/log.h"
+#include "core/frontend/image_interface.h"
+
+namespace Frontend {
+
+bool ImageInterface::DecodePNG(std::vector<u8>& dst, u32& width, u32& height,
+                               std::span<const u8> src) {
+    const u32 lodepng_ret = lodepng::decode(dst, width, height, src.data(), src.size());
+    if (lodepng_ret) {
+        LOG_ERROR(Frontend, "Failed to decode because {}", lodepng_error_text(lodepng_ret));
+        return false;
+    }
+    return true;
+}
+
+bool ImageInterface::DecodeDDS(std::vector<u8>& dst, u32& width, u32& height, ddsktx_format& format,
+                               std::span<const u8> src) {
+    ddsktx_texture_info tc{};
+    const int size = static_cast<int>(src.size());
+
+    if (!ddsktx_parse(&tc, src.data(), size, nullptr)) {
+        LOG_ERROR(Frontend, "Failed to decode");
+        return false;
+    }
+    width = tc.width;
+    height = tc.height;
+    format = tc.format;
+
+    ddsktx_sub_data sub_data{};
+    ddsktx_get_sub(&tc, &sub_data, src.data(), size, 0, 0, 0);
+
+    dst.resize(sub_data.size_bytes);
+    std::memcpy(dst.data(), sub_data.buff, sub_data.size_bytes);
+
+    return true;
+}
+
+bool ImageInterface::EncodePNG(const std::string& path, u32 width, u32 height,
+                               std::span<const u8> src) {
+    std::vector<u8> out;
+    const u32 lodepng_ret = lodepng::encode(out, src.data(), width, height);
+    if (lodepng_ret) {
+        LOG_ERROR(Frontend, "Failed to encode {} because {}", path,
+                  lodepng_error_text(lodepng_ret));
+        return false;
+    }
+
+    FileUtil::IOFile file{path, "wb"};
+    if (file.WriteBytes(out.data(), out.size()) != out.size()) {
+        LOG_ERROR(Frontend, "Failed to save encode to path {}", path);
+        return false;
+    }
+
+    return true;
+}
+
+} // namespace Frontend

src/core/frontend/image_interface.h

@@ -4,21 +4,26 @@
 
 #pragma once
 
+#include <span>
 #include <string>
 #include <vector>
+#include <dds-ktx.h>
 #include "common/common_types.h"
 
 namespace Frontend {
 
+/**
+ * Utility class that provides image decoding/encoding to the custom texture manager.
+ * Can be optionally overriden by frontends to provide a custom implementation.
+ */
 class ImageInterface {
 public:
     virtual ~ImageInterface() = default;
 
-    // Error logging should be handled by the frontend
-    virtual bool DecodePNG(std::vector<u8>& dst, u32& width, u32& height,
-                           const std::string& path) = 0;
-    virtual bool EncodePNG(const std::string& path, const std::vector<u8>& src, u32 width,
-                           u32 height) = 0;
+    virtual bool DecodePNG(std::vector<u8>& dst, u32& width, u32& height, std::span<const u8> src);
+    virtual bool DecodeDDS(std::vector<u8>& dst, u32& width, u32& height, ddsktx_format& format,
+                           std::span<const u8> src);
+    virtual bool EncodePNG(const std::string& path, u32 width, u32 height, std::span<const u8> src);
 };
 
 } // namespace Frontend

src/core/gdbstub/hio.cpp

@@ -136,8 +136,7 @@ void HandleHioReply(const u8* const command_buffer, const u32 command_length) {
     }
 
     const std::string command_str{command_pos, command_buffer + command_length};
-    std::vector<std::string> command_parts;
-    Common::SplitString(command_str, ',', command_parts);
+    const auto command_parts = Common::SplitString(command_str, ',');
 
     if (command_parts.empty() || command_parts.size() > 3) {
         LOG_WARNING(Debug_GDBStub, "Unexpected reply packet size: {}", command_parts);

src/core/hw/aes/key.cpp

@@ -458,8 +458,7 @@ void LoadPresetKeys() {
             continue;
         }
 
-        std::vector<std::string> parts;
-        Common::SplitString(line, '=', parts);
+        const auto parts = Common::SplitString(line, '=');
         if (parts.size() != 2) {
             LOG_ERROR(HW_AES, "Failed to parse {}", line);
             continue;

src/video_core/CMakeLists.txt

@@ -3,6 +3,12 @@ add_subdirectory(host_shaders)
 add_library(video_core STATIC
     command_processor.cpp
     command_processor.h
+    custom_textures/custom_format.cpp
+    custom_textures/custom_format.h
+    custom_textures/custom_tex_manager.cpp
+    custom_textures/custom_tex_manager.h
+    custom_textures/material.cpp
+    custom_textures/material.h
     debug_utils/debug_utils.cpp
     debug_utils/debug_utils.h
     geometry_pipeline.cpp
@@ -120,7 +126,7 @@ target_include_directories(video_core PRIVATE ${HOST_SHADERS_INCLUDE})
 create_target_directory_groups(video_core)
 
 target_link_libraries(video_core PUBLIC common core)
-target_link_libraries(video_core PRIVATE glad nihstro-headers Boost::serialization)
+target_link_libraries(video_core PRIVATE glad json-headers dds-ktx nihstro-headers Boost::serialization)
 set_target_properties(video_core PROPERTIES INTERPROCEDURAL_OPTIMIZATION ${ENABLE_LTO})
 
 if ("x86_64" IN_LIST ARCHITECTURE)

src/video_core/custom_textures/custom_format.cpp

@@ -0,0 +1,36 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "video_core/custom_textures/custom_format.h"
+
+namespace VideoCore {
+
+std::string_view CustomPixelFormatAsString(CustomPixelFormat format) {
+    switch (format) {
+    case CustomPixelFormat::RGBA8:
+        return "RGBA8";
+    case CustomPixelFormat::BC1:
+        return "BC1";
+    case CustomPixelFormat::BC3:
+        return "BC3";
+    case CustomPixelFormat::BC5:
+        return "BC5";
+    case CustomPixelFormat::BC7:
+        return "BC7";
+    case CustomPixelFormat::ASTC4:
+        return "ASTC4";
+    case CustomPixelFormat::ASTC6:
+        return "ASTC6";
+    case CustomPixelFormat::ASTC8:
+        return "ASTC8";
+    default:
+        return "NotReal";
+    }
+}
+
+bool IsCustomFormatCompressed(CustomPixelFormat format) {
+    return format != CustomPixelFormat::RGBA8 && format != CustomPixelFormat::Invalid;
+}
+
+} // namespace VideoCore

src/video_core/custom_textures/custom_format.h

@@ -0,0 +1,36 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <limits>
+#include <string_view>
+#include "common/common_types.h"
+
+namespace VideoCore {
+
+enum class CustomPixelFormat : u32 {
+    RGBA8 = 0,
+    BC1 = 1,
+    BC3 = 2,
+    BC5 = 3,
+    BC7 = 4,
+    ASTC4 = 5,
+    ASTC6 = 6,
+    ASTC8 = 7,
+    Invalid = std::numeric_limits<u32>::max(),
+};
+
+enum class CustomFileFormat : u32 {
+    None = 0,
+    PNG = 1,
+    DDS = 2,
+    KTX = 3,
+};
+
+std::string_view CustomPixelFormatAsString(CustomPixelFormat format);
+
+bool IsCustomFormatCompressed(CustomPixelFormat format);
+
+} // namespace VideoCore

src/video_core/custom_textures/custom_tex_manager.cpp

@@ -0,0 +1,352 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include <json.hpp>
+#include "common/file_util.h"
+#include "common/memory_detect.h"
+#include "common/microprofile.h"
+#include "common/settings.h"
+#include "common/string_util.h"
+#include "common/texture.h"
+#include "core/core.h"
+#include "core/frontend/image_interface.h"
+#include "video_core/custom_textures/custom_tex_manager.h"
+#include "video_core/rasterizer_cache/surface_params.h"
+
+namespace VideoCore {
+
+namespace {
+
+MICROPROFILE_DEFINE(CustomTexManager_TickFrame, "CustomTexManager", "TickFrame",
+                    MP_RGB(54, 16, 32));
+
+constexpr std::size_t MAX_UPLOADS_PER_TICK = 16;
+
+bool IsPow2(u32 value) {
+    return value != 0 && (value & (value - 1)) == 0;
+}
+
+CustomFileFormat MakeFileFormat(std::string_view ext) {
+    if (ext == "png") {
+        return CustomFileFormat::PNG;
+    } else if (ext == "dds") {
+        return CustomFileFormat::DDS;
+    } else if (ext == "ktx") {
+        return CustomFileFormat::KTX;
+    }
+    return CustomFileFormat::None;
+}
+
+MapType MakeMapType(std::string_view ext) {
+    if (ext == "norm") {
+        return MapType::Normal;
+    }
+    LOG_ERROR(Render, "Unknown material extension {}", ext);
+    return MapType::Color;
+}
+
+} // Anonymous namespace
+
+CustomTexManager::CustomTexManager(Core::System& system_)
+    : system{system_}, image_interface{*system.GetImageInterface()},
+      async_custom_loading{Settings::values.async_custom_loading.GetValue()} {}
+
+CustomTexManager::~CustomTexManager() = default;
+
+void CustomTexManager::TickFrame() {
+    MICROPROFILE_SCOPE(CustomTexManager_TickFrame);
+    if (!textures_loaded) {
+        return;
+    }
+    std::size_t num_uploads = 0;
+    for (auto it = async_uploads.begin(); it != async_uploads.end();) {
+        if (num_uploads >= MAX_UPLOADS_PER_TICK) {
+            return;
+        }
+        switch (it->material->state) {
+        case DecodeState::Decoded:
+            it->func();
+            num_uploads++;
+            [[fallthrough]];
+        case DecodeState::Failed:
+            it = async_uploads.erase(it);
+            continue;
+        default:
+            it++;
+            break;
+        }
+    }
+}
+
+void CustomTexManager::FindCustomTextures() {
+    if (textures_loaded) {
+        return;
+    }
+    if (!workers) {
+        CreateWorkers();
+    }
+
+    const u64 program_id = system.Kernel().GetCurrentProcess()->codeset->program_id;
+    const std::string load_path =
+        fmt::format("{}textures/{:016X}/", GetUserPath(FileUtil::UserPath::LoadDir), program_id);
+
+    if (!FileUtil::Exists(load_path)) {
+        FileUtil::CreateFullPath(load_path);
+    }
+    ReadConfig(load_path);
+
+    FileUtil::FSTEntry texture_dir;
+    std::vector<FileUtil::FSTEntry> textures;
+    FileUtil::ScanDirectoryTree(load_path, texture_dir, 64);
+    FileUtil::GetAllFilesFromNestedEntries(texture_dir, textures);
+
+    custom_textures.reserve(textures.size());
+    for (const FileUtil::FSTEntry& file : textures) {
+        if (file.isDirectory) {
+            continue;
+        }
+        custom_textures.push_back(std::make_unique<CustomTexture>(image_interface));
+        CustomTexture* const texture{custom_textures.back().get()};
+        if (!ParseFilename(file, texture)) {
+            continue;
+        }
+        auto& material = material_map[texture->hash];
+        if (!material) {
+            material = std::make_unique<Material>();
+        }
+        material->AddMapTexture(texture);
+    }
+    textures_loaded = true;
+}
+
+bool CustomTexManager::ParseFilename(const FileUtil::FSTEntry& file, CustomTexture* texture) {
+    auto parts = Common::SplitString(file.virtualName, '.');
+    if (parts.size() > 3) {
+        LOG_ERROR(Render, "Invalid filename {}, ignoring", file.virtualName);
+        return false;
+    }
+    // The last string should always be the file extension.
+    const CustomFileFormat file_format = MakeFileFormat(parts.back());
+    if (file_format == CustomFileFormat::None) {
+        return false;
+    }
+    if (file_format == CustomFileFormat::DDS && refuse_dds) {
+        LOG_ERROR(Render, "Legacy pack is attempting to use DDS textures, skipping!");
+        return false;
+    }
+    texture->file_format = file_format;
+    parts.pop_back();
+
+    // This means the texture is a material type other than color.
+    texture->type = MapType::Color;
+    if (parts.size() > 1) {
+        texture->type = MakeMapType(parts.back());
+        parts.pop_back();
+    }
+
+    // First check if the path is mapped directly to a hash
+    // before trying to parse the texture filename.
+    const auto it = path_to_hash_map.find(file.virtualName);
+    if (it != path_to_hash_map.end()) {
+        texture->hash = it->second;
+    } else {
+        u32 width;
+        u32 height;
+        u32 format;
+        unsigned long long hash{};
+        if (std::sscanf(parts.back().c_str(), "tex1_%ux%u_%llX_%u", &width, &height, &hash,
+                        &format) != 4) {
+            return false;
+        }
+        texture->hash = hash;
+    }
+
+    texture->path = file.physicalName;
+    return true;
+}
+
+void CustomTexManager::WriteConfig() {
+    const u64 program_id = system.Kernel().GetCurrentProcess()->codeset->program_id;
+    const std::string dump_path =
+        fmt::format("{}textures/{:016X}/", GetUserPath(FileUtil::UserPath::DumpDir), program_id);
+    const std::string pack_config = dump_path + "pack.json";
+    if (FileUtil::Exists(pack_config)) {
+        return;
+    }
+
+    nlohmann::ordered_json json;
+    json["author"] = "citra";
+    json["version"] = "1.0.0";
+    json["description"] = "A graphics pack";
+
+    auto& options = json["options"];
+    options["skip_mipmap"] = skip_mipmap;
+    options["flip_png_files"] = flip_png_files;
+    options["use_new_hash"] = use_new_hash;
+
+    FileUtil::IOFile file{pack_config, "w"};
+    const std::string output = json.dump(4);
+    file.WriteString(output);
+}
+
+void CustomTexManager::PreloadTextures() {
+    u64 size_sum = 0;
+    const u64 sys_mem = Common::GetMemInfo().total_physical_memory;
+    const u64 recommended_min_mem = 2 * size_t(1024 * 1024 * 1024);
+
+    // keep 2GB memory for system stability if system RAM is 4GB+ - use half of memory in other
+    // cases
+    const u64 max_mem =
+        (sys_mem / 2 < recommended_min_mem) ? (sys_mem / 2) : (sys_mem - recommended_min_mem);
+
+    workers->QueueWork([&]() {
+        for (auto& [hash, material] : material_map) {
+            if (size_sum > max_mem) {
+                LOG_WARNING(Render, "Aborting texture preload due to insufficient memory");
+                return;
+            }
+            material->LoadFromDisk(flip_png_files);
+            size_sum += material->size;
+        }
+    });
+    workers->WaitForRequests();
+    async_custom_loading = false;
+}
+
+void CustomTexManager::DumpTexture(const SurfaceParams& params, u32 level, std::span<u8> data,
+                                   u64 data_hash) {
+    const u64 program_id = system.Kernel().GetCurrentProcess()->codeset->program_id;
+    const u32 data_size = static_cast<u32>(data.size());
+    const u32 width = params.width;
+    const u32 height = params.height;
+    const PixelFormat format = params.pixel_format;
+
+    std::string dump_path = fmt::format(
+        "{}textures/{:016X}/", FileUtil::GetUserPath(FileUtil::UserPath::DumpDir), program_id);
+    if (!FileUtil::CreateFullPath(dump_path)) {
+        LOG_ERROR(Render, "Unable to create {}", dump_path);
+        return;
+    }
+
+    dump_path +=
+        fmt::format("tex1_{}x{}_{:016X}_{}_mip{}.png", width, height, data_hash, format, level);
+    if (dumped_textures.contains(data_hash) || FileUtil::Exists(dump_path)) {
+        return;
+    }
+
+    // Make sure the texture size is a power of 2.
+    // If not, the surface is probably a framebuffer
+    if (!IsPow2(width) || !IsPow2(height)) {
+        LOG_WARNING(Render, "Not dumping {:016X} because size isn't a power of 2 ({}x{})",
+                    data_hash, width, height);
+        return;
+    }
+
+    const u32 decoded_size = width * height * 4;
+    std::vector<u8> pixels(data_size + decoded_size);
+    std::memcpy(pixels.data(), data.data(), data_size);
+
+    auto dump = [this, width, height, params, data_size, decoded_size, pixels = std::move(pixels),
+                 dump_path = std::move(dump_path)]() mutable {
+        const std::span encoded = std::span{pixels}.first(data_size);
+        const std::span decoded = std::span{pixels}.last(decoded_size);
+        DecodeTexture(params, params.addr, params.end, encoded, decoded,
+                      params.type == SurfaceType::Color);
+        Common::FlipRGBA8Texture(decoded, width, height);
+        image_interface.EncodePNG(dump_path, width, height, decoded);
+    };
+    if (!workers) {
+        CreateWorkers();
+    }
+    workers->QueueWork(std::move(dump));
+    dumped_textures.insert(data_hash);
+}
+
+Material* CustomTexManager::GetMaterial(u64 data_hash) {
+    const auto it = material_map.find(data_hash);
+    if (it == material_map.end()) {
+        LOG_WARNING(Render, "Unable to find replacement for surface with hash {:016X}", data_hash);
+        return nullptr;
+    }
+    return it->second.get();
+}
+
+bool CustomTexManager::Decode(Material* material, std::function<bool()>&& upload) {
+    if (!async_custom_loading) {
+        material->LoadFromDisk(flip_png_files);
+        return upload();
+    }
+    if (material->IsUnloaded()) {
+        material->state = DecodeState::Pending;
+        workers->QueueWork([material, this] { material->LoadFromDisk(flip_png_files); });
+    }
+    async_uploads.push_back({
+        .material = material,
+        .func = std::move(upload),
+    });
+    return false;
+}
+
+void CustomTexManager::ReadConfig(const std::string& load_path) {
+    const std::string config_path = load_path + "pack.json";
+    FileUtil::IOFile file{config_path, "r"};
+    if (!file.IsOpen()) {
+        LOG_INFO(Render, "Unable to find pack config file, using legacy defaults");
+        refuse_dds = true;
+        return;
+    }
+    std::string config(file.GetSize(), '\0');
+    const std::size_t read_size = file.ReadBytes(config.data(), config.size());
+    if (!read_size) {
+        return;
+    }
+
+    nlohmann::json json = nlohmann::json::parse(config);
+
+    const auto& options = json["options"];
+    skip_mipmap = options["skip_mipmap"].get<bool>();
+    flip_png_files = options["flip_png_files"].get<bool>();
+    use_new_hash = options["use_new_hash"].get<bool>();
+    refuse_dds = skip_mipmap || !use_new_hash;
+
+    const auto& textures = json["textures"];
+    for (const auto& material : textures.items()) {
+        size_t idx{};
+        const u64 hash = std::stoull(material.key(), &idx, 16);
+        if (!idx) {
+            LOG_ERROR(Render, "Key {} is invalid, skipping", material.key());
+            continue;
+        }
+        const auto parse = [&](const std::string& file) {
+            const std::string filename{FileUtil::GetFilename(file)};
+            auto [it, new_hash] = path_to_hash_map.try_emplace(filename);
+            if (!new_hash) {
+                LOG_ERROR(Render,
+                          "File {} with key {} already exists and is mapped to {:#016X}, skipping",
+                          file, material.key(), path_to_hash_map[filename]);
+                return;
+            }
+            it->second = hash;
+        };
+        const auto value = material.value();
+        if (value.is_string()) {
+            const auto file = value.get<std::string>();
+            parse(file);
+        } else if (value.is_array()) {
+            const auto files = value.get<std::vector<std::string>>();
+            for (const std::string& file : files) {
+                parse(file);
+            }
+        } else {
+            LOG_ERROR(Render, "Material with key {} is invalid", material.key());
+        }
+    }
+}
+
+void CustomTexManager::CreateWorkers() {
+    const std::size_t num_workers = std::max(std::thread::hardware_concurrency(), 2U) - 1;
+    workers = std::make_unique<Common::ThreadWorker>(num_workers, "Custom textures");
+}
+
+} // namespace VideoCore

src/video_core/custom_textures/custom_tex_manager.h

@@ -0,0 +1,94 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <list>
+#include <span>
+#include <unordered_map>
+#include <unordered_set>
+#include "common/thread_worker.h"
+#include "video_core/custom_textures/material.h"
+
+namespace Core {
+class System;
+}
+
+namespace FileUtil {
+struct FSTEntry;
+}
+
+namespace VideoCore {
+
+class SurfaceParams;
+
+struct AsyncUpload {
+    const Material* material;
+    std::function<bool()> func;
+};
+
+class CustomTexManager {
+public:
+    explicit CustomTexManager(Core::System& system);
+    ~CustomTexManager();
+
+    /// Processes queued texture uploads
+    void TickFrame();
+
+    /// Searches the load directory assigned to program_id for any custom textures and loads them
+    void FindCustomTextures();
+
+    /// Saves the pack configuration file template to the dump directory if it doesn't exist.
+    void WriteConfig();
+
+    /// Preloads all registered custom textures
+    void PreloadTextures();
+
+    /// Saves the provided pixel data described by params to disk as png
+    void DumpTexture(const SurfaceParams& params, u32 level, std::span<u8> data, u64 data_hash);
+
+    /// Returns the material assigned to the provided data hash
+    Material* GetMaterial(u64 data_hash);
+
+    /// Decodes the textures in material to a consumable format and uploads it.
+    bool Decode(Material* material, std::function<bool()>&& upload);
+
+    /// True when mipmap uploads should be skipped (legacy packs only)
+    bool SkipMipmaps() const noexcept {
+        return skip_mipmap;
+    }
+
+    /// Returns true if the pack uses the new hashing method.
+    bool UseNewHash() const noexcept {
+        return use_new_hash;
+    }
+
+private:
+    /// Parses the custom texture filename (hash, material type, etc).
+    bool ParseFilename(const FileUtil::FSTEntry& file, CustomTexture* texture);
+
+    /// Reads the pack configuration file
+    void ReadConfig(const std::string& load_path);
+
+    /// Creates the thread workers.
+    void CreateWorkers();
+
+private:
+    Core::System& system;
+    Frontend::ImageInterface& image_interface;
+    std::unordered_set<u64> dumped_textures;
+    std::unordered_map<u64, std::unique_ptr<Material>> material_map;
+    std::unordered_map<std::string, u64> path_to_hash_map;
+    std::vector<std::unique_ptr<CustomTexture>> custom_textures;
+    std::list<AsyncUpload> async_uploads;
+    std::unique_ptr<Common::ThreadWorker> workers;
+    bool textures_loaded{false};
+    bool async_custom_loading{true};
+    bool skip_mipmap{true};
+    bool flip_png_files{true};
+    bool use_new_hash{false};
+    bool refuse_dds{false};
+};
+
+} // namespace VideoCore

src/video_core/custom_textures/material.cpp

@@ -0,0 +1,151 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "common/file_util.h"
+#include "common/logging/log.h"
+#include "common/texture.h"
+#include "core/frontend/image_interface.h"
+#include "video_core/custom_textures/material.h"
+
+namespace VideoCore {
+
+namespace {
+
+CustomPixelFormat ToCustomPixelFormat(ddsktx_format format) {
+    switch (format) {
+    case DDSKTX_FORMAT_RGBA8:
+        return CustomPixelFormat::RGBA8;
+    case DDSKTX_FORMAT_BC1:
+        return CustomPixelFormat::BC1;
+    case DDSKTX_FORMAT_BC3:
+        return CustomPixelFormat::BC3;
+    case DDSKTX_FORMAT_BC5:
+        return CustomPixelFormat::BC5;
+    case DDSKTX_FORMAT_BC7:
+        return CustomPixelFormat::BC7;
+    case DDSKTX_FORMAT_ASTC4x4:
+        return CustomPixelFormat::ASTC4;
+    case DDSKTX_FORMAT_ASTC6x6:
+        return CustomPixelFormat::ASTC6;
+    case DDSKTX_FORMAT_ASTC8x6:
+        return CustomPixelFormat::ASTC8;
+    default:
+        LOG_ERROR(Common, "Unknown dds/ktx pixel format {}", format);
+        return CustomPixelFormat::RGBA8;
+    }
+}
+
+std::string_view MapTypeName(MapType type) {
+    switch (type) {
+    case MapType::Color:
+        return "Color";
+    case MapType::Normal:
+        return "Normal";
+    default:
+        return "Invalid";
+    }
+}
+
+} // Anonymous namespace
+
+CustomTexture::CustomTexture(Frontend::ImageInterface& image_interface_)
+    : image_interface{image_interface_} {}
+
+CustomTexture::~CustomTexture() = default;
+
+void CustomTexture::LoadFromDisk(bool flip_png) {
+    FileUtil::IOFile file{path, "rb"};
+    std::vector<u8> input(file.GetSize());
+    if (file.ReadBytes(input.data(), input.size()) != input.size()) {
+        LOG_CRITICAL(Render, "Failed to open custom texture: {}", path);
+        return;
+    }
+    switch (file_format) {
+    case CustomFileFormat::PNG:
+        LoadPNG(input, flip_png);
+        break;
+    case CustomFileFormat::DDS:
+    case CustomFileFormat::KTX:
+        LoadDDS(input);
+        break;
+    default:
+        LOG_ERROR(Render, "Unknown file format {}", file_format);
+        return;
+    }
+}
+
+void CustomTexture::LoadPNG(std::span<const u8> input, bool flip_png) {
+    if (!image_interface.DecodePNG(data, width, height, input)) {
+        LOG_ERROR(Render, "Failed to decode png: {}", path);
+        return;
+    }
+    if (flip_png) {
+        Common::FlipRGBA8Texture(data, width, height);
+    }
+    format = CustomPixelFormat::RGBA8;
+}
+
+void CustomTexture::LoadDDS(std::span<const u8> input) {
+    ddsktx_format dds_format{};
+    image_interface.DecodeDDS(data, width, height, dds_format, input);
+    format = ToCustomPixelFormat(dds_format);
+}
+
+void Material::LoadFromDisk(bool flip_png) noexcept {
+    if (IsDecoded()) {
+        return;
+    }
+    for (CustomTexture* const texture : textures) {
+        if (!texture || texture->IsLoaded()) {
+            continue;
+        }
+        texture->LoadFromDisk(flip_png);
+        size += texture->data.size();
+        LOG_DEBUG(Render, "Loading {} map {} with hash {:#016X}", MapTypeName(texture->type),
+                  texture->path, texture->hash);
+    }
+    if (!textures[0]) {
+        LOG_ERROR(Render, "Unable to create material without color texture!");
+        state = DecodeState::Failed;
+        return;
+    }
+    width = textures[0]->width;
+    height = textures[0]->height;
+    format = textures[0]->format;
+    for (const CustomTexture* texture : textures) {
+        if (!texture) {
+            continue;
+        }
+        if (texture->width != width || texture->height != height) {
+            LOG_ERROR(Render,
+                      "{} map {} of material with hash {:#016X} has dimentions {}x{} "
+                      "which do not match the color texture dimentions {}x{}",
+                      MapTypeName(texture->type), texture->path, texture->hash, texture->width,
+                      texture->height, width, height);
+            state = DecodeState::Failed;
+            return;
+        }
+        if (texture->format != format) {
+            LOG_ERROR(
+                Render, "{} map {} is stored with {} format which does not match color format {}",
+                MapTypeName(texture->type), texture->path,
+                CustomPixelFormatAsString(texture->format), CustomPixelFormatAsString(format));
+            state = DecodeState::Failed;
+            return;
+        }
+    }
+    state = DecodeState::Decoded;
+}
+
+void Material::AddMapTexture(CustomTexture* texture) noexcept {
+    const std::size_t index = static_cast<std::size_t>(texture->type);
+    if (textures[index]) {
+        LOG_ERROR(Render, "Textures {} and {} are assigned to the same material, ignoring!",
+                  textures[index]->path, texture->path);
+        return;
+    }
+    textures[index] = texture;
+}
+
+} // namespace VideoCore

src/video_core/custom_textures/material.h

@@ -0,0 +1,99 @@
+// Copyright 2023 Citra Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include <array>
+#include <atomic>
+#include <span>
+#include <string>
+#include <vector>
+#include "video_core/custom_textures/custom_format.h"
+
+namespace Frontend {
+class ImageInterface;
+}
+
+namespace VideoCore {
+
+enum class MapType : u32 {
+    Color = 0,
+    Normal = 1,
+    MapCount = 2,
+};
+constexpr std::size_t MAX_MAPS = static_cast<std::size_t>(MapType::MapCount);
+
+enum class DecodeState : u32 {
+    None = 0,
+    Pending = 1,
+    Decoded = 2,
+    Failed = 3,
+};
+
+class CustomTexture {
+public:
+    explicit CustomTexture(Frontend::ImageInterface& image_interface);
+    ~CustomTexture();
+
+    void LoadFromDisk(bool flip_png);
+
+    [[nodiscard]] bool IsParsed() const noexcept {
+        return file_format != CustomFileFormat::None && hash != 0;
+    }
+
+    [[nodiscard]] bool IsLoaded() const noexcept {
+        return !data.empty();
+    }
+
+private:
+    void LoadPNG(std::span<const u8> input, bool flip_png);
+
+    void LoadDDS(std::span<const u8> input);
+
+public:
+    Frontend::ImageInterface& image_interface;
+    std::string path;
+    u32 width;
+    u32 height;
+    u64 hash;
+    CustomPixelFormat format;
+    CustomFileFormat file_format;
+    std::vector<u8> data;
+    MapType type;
+};
+
+struct Material {
+    u32 width;
+    u32 height;
+    u64 size;
+    CustomPixelFormat format;
+    std::array<CustomTexture*, MAX_MAPS> textures;
+    std::atomic<DecodeState> state{};
+
+    void LoadFromDisk(bool flip_png) noexcept;
+
+    void AddMapTexture(CustomTexture* texture) noexcept;
+
+    [[nodiscard]] CustomTexture* Map(MapType type) const noexcept {
+        return textures.at(static_cast<std::size_t>(type));
+    }
+
+    [[nodiscard]] bool IsPending() const noexcept {
+        return state == DecodeState::Pending;
+    }
+
+    [[nodiscard]] bool IsFailed() const noexcept {
+        return state == DecodeState::Failed;
+    }
+
+    [[nodiscard]] bool IsDecoded() const noexcept {
+        return state == DecodeState::Decoded;
+    }
+
+    [[nodiscard]] bool IsUnloaded() const noexcept {
+        return state == DecodeState::None;
+    }
+};
+
+} // namespace VideoCore

src/video_core/rasterizer_cache/pixel_format.h

@@ -44,7 +44,7 @@ enum class SurfaceType : u32 {
     Invalid = 5,
 };
 
-enum class TextureType : u32 {
+enum class TextureType : u16 {
     Texture2D = 0,
     CubeMap = 1,
 };
@@ -99,10 +99,10 @@ constexpr SurfaceType GetFormatType(PixelFormat format) {
     return FORMAT_MAP[index].type;
 }
 
-std::string_view PixelFormatAsString(PixelFormat format);
-
 bool CheckFormatsBlittable(PixelFormat source_format, PixelFormat dest_format);
 
+std::string_view PixelFormatAsString(PixelFormat format);
+
 PixelFormat PixelFormatFromTextureFormat(Pica::TexturingRegs::TextureFormat format);
 
 PixelFormat PixelFormatFromColorFormat(Pica::FramebufferRegs::ColorFormat format);

src/video_core/rasterizer_cache/rasterizer_cache.cpp

@@ -8,6 +8,7 @@
 #include "common/logging/log.h"
 #include "common/microprofile.h"
 #include "core/memory.h"
+#include "video_core/custom_textures/custom_tex_manager.h"
 #include "video_core/rasterizer_cache/rasterizer_cache.h"
 #include "video_core/regs.h"
 #include "video_core/renderer_base.h"
@@ -19,6 +20,15 @@ namespace {
 
 MICROPROFILE_DEFINE(RasterizerCache_CopySurface, "RasterizerCache", "CopySurface",
                     MP_RGB(128, 192, 64));
+MICROPROFILE_DEFINE(RasterizerCache_UploadSurface, "RasterizerCache", "UploadSurface",
+                    MP_RGB(128, 192, 64));
+MICROPROFILE_DEFINE(RasterizerCache_ComputeHash, "RasterizerCache", "ComputeHash",
+                    MP_RGB(32, 64, 192));
+MICROPROFILE_DEFINE(RasterizerCache_DownloadSurface, "RasterizerCache", "DownloadSurface",
+                    MP_RGB(128, 192, 64));
+MICROPROFILE_DEFINE(RasterizerCache_Invalidation, "RasterizerCache", "Invalidation",
+                    MP_RGB(128, 64, 192));
+MICROPROFILE_DEFINE(RasterizerCache_Flush, "RasterizerCache", "Flush", MP_RGB(128, 64, 192));
 
 constexpr auto RangeFromInterval(const auto& map, const auto& interval) {
     return boost::make_iterator_range(map.equal_range(interval));
@@ -119,11 +129,15 @@ auto FindMatch(const auto& surface_cache, const SurfaceParams& params, ScaleMatc
 
 } // Anonymous namespace
 
-RasterizerCache::RasterizerCache(Memory::MemorySystem& memory_, OpenGL::TextureRuntime& runtime_,
-                                 Pica::Regs& regs_, RendererBase& renderer_)
-    : memory{memory_}, runtime{runtime_}, regs{regs_}, renderer{renderer_},
-      resolution_scale_factor{renderer.GetResolutionScaleFactor()},
-      use_filter{Settings::values.texture_filter.GetValue() != Settings::TextureFilter::None} {}
+RasterizerCache::RasterizerCache(Memory::MemorySystem& memory_,
+                                 CustomTexManager& custom_tex_manager_,
+                                 OpenGL::TextureRuntime& runtime_, Pica::Regs& regs_,
+                                 RendererBase& renderer_)
+    : memory{memory_}, custom_tex_manager{custom_tex_manager_}, runtime{runtime_}, regs{regs_},
+      renderer{renderer_}, resolution_scale_factor{renderer.GetResolutionScaleFactor()},
+      use_filter{Settings::values.texture_filter.GetValue() != Settings::TextureFilter::None},
+      dump_textures{Settings::values.dump_textures.GetValue()},
+      use_custom_textures{Settings::values.custom_textures.GetValue()} {}
 
 RasterizerCache::~RasterizerCache() {
 #ifndef ANDROID
@@ -132,6 +146,32 @@ RasterizerCache::~RasterizerCache() {
 #endif
 }
 
+void RasterizerCache::TickFrame() {
+    custom_tex_manager.TickFrame();
+
+    const u32 scale_factor = renderer.GetResolutionScaleFactor();
+    const bool resolution_scale_changed = resolution_scale_factor != scale_factor;
+    const bool use_custom_texture_changed =
+        Settings::values.custom_textures.GetValue() != use_custom_textures;
+    const bool texture_filter_changed =
+        renderer.Settings().texture_filter_update_requested.exchange(false);
+
+    if (resolution_scale_changed || texture_filter_changed || use_custom_texture_changed) {
+        resolution_scale_factor = scale_factor;
+        use_filter = Settings::values.texture_filter.GetValue() != Settings::TextureFilter::None;
+        use_custom_textures = Settings::values.custom_textures.GetValue();
+        if (use_custom_textures) {
+            custom_tex_manager.FindCustomTextures();
+        }
+        FlushAll();
+        while (!surface_cache.empty()) {
+            UnregisterSurface(*surface_cache.begin()->second.begin());
+        }
+        texture_cube_cache.clear();
+        runtime.Reset();
+    }
+}
+
 bool RasterizerCache::AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config) {
     // Texture copy size is aligned to 16 byte units
     const u32 copy_size = Common::AlignDown(config.texture_copy.size, 16);
@@ -572,21 +612,6 @@ OpenGL::Framebuffer RasterizerCache::GetFramebufferSurfaces(bool using_color_fb,
                                                             bool using_depth_fb) {
     const auto& config = regs.framebuffer.framebuffer;
 
-    // Update resolution_scale_factor and reset cache if changed
-    const u32 scale_factor = renderer.GetResolutionScaleFactor();
-    const bool resolution_scale_changed = resolution_scale_factor != scale_factor;
-    const bool texture_filter_changed =
-        renderer.Settings().texture_filter_update_requested.exchange(false);
-
-    if (resolution_scale_changed || texture_filter_changed) {
-        resolution_scale_factor = scale_factor;
-        use_filter = Settings::values.texture_filter.GetValue() != Settings::TextureFilter::None;
-        FlushAll();
-        while (!surface_cache.empty())
-            UnregisterSurface(*surface_cache.begin()->second.begin());
-        texture_cube_cache.clear();
-    }
-
     const s32 framebuffer_width = config.GetWidth();
     const s32 framebuffer_height = config.GetHeight();
     const auto viewport_rect = regs.rasterizer.GetViewportRect();
@@ -818,11 +843,13 @@ void RasterizerCache::ValidateSurface(const SurfaceRef& surface, PAddr addr, u32
     // Filtered mipmaps often look really bad. We can achieve better quality by
     // generating them from the base level.
     if (surface->res_scale != 1 && level != 0) {
-        runtime.GenerateMipmaps(*surface, surface->levels - 1);
+        runtime.GenerateMipmaps(*surface);
     }
 }
 
 void RasterizerCache::UploadSurface(const SurfaceRef& surface, SurfaceInterval interval) {
+    MICROPROFILE_SCOPE(RasterizerCache_UploadSurface);
+
     const SurfaceParams load_info = surface->FromInterval(interval);
     ASSERT(load_info.addr >= surface->addr && load_info.end <= surface->end);
 
@@ -838,6 +865,18 @@ void RasterizerCache::UploadSurface(const SurfaceRef& surface, SurfaceInterval i
     DecodeTexture(load_info, load_info.addr, load_info.end, upload_data, staging.mapped,
                   runtime.NeedsConversion(surface->pixel_format));
 
+    if (use_custom_textures) {
+        const u64 hash = ComputeCustomHash(load_info, staging.mapped, upload_data);
+        if (UploadCustomSurface(surface, load_info, hash)) {
+            return;
+        }
+    }
+    if (dump_textures && !surface->is_custom) {
+        const u64 hash = Common::ComputeHash64(upload_data.data(), upload_data.size());
+        const u32 level = surface->LevelOf(load_info.addr);
+        custom_tex_manager.DumpTexture(load_info, level, upload_data, hash);
+    }
+
     const BufferTextureCopy upload = {
         .buffer_offset = 0,
         .buffer_size = staging.size,
@@ -847,7 +886,49 @@ void RasterizerCache::UploadSurface(const SurfaceRef& surface, SurfaceInterval i
     surface->Upload(upload, staging);
 }
 
+bool RasterizerCache::UploadCustomSurface(const SurfaceRef& surface, const SurfaceParams& load_info,
+                                          u64 hash) {
+    const u32 level = surface->LevelOf(load_info.addr);
+    const bool is_base_level = level == 0;
+    Material* material = custom_tex_manager.GetMaterial(hash);
+
+    if (!material) {
+        return surface->IsCustom();
+    }
+    if (!is_base_level && custom_tex_manager.SkipMipmaps()) {
+        return true;
+    }
+
+    surface->is_custom = true;
+
+    const auto upload = [this, level, surface, material]() -> bool {
+        if (!surface->IsCustom() && !surface->Swap(material)) {
+            LOG_ERROR(HW_GPU, "Custom compressed format {} unsupported by host GPU",
+                      material->format);
+            return false;
+        }
+        surface->UploadCustom(material, level);
+        if (custom_tex_manager.SkipMipmaps()) {
+            runtime.GenerateMipmaps(*surface);
+        }
+        return true;
+    };
+    return custom_tex_manager.Decode(material, std::move(upload));
+}
+
+u64 RasterizerCache::ComputeCustomHash(const SurfaceParams& load_info, std::span<u8> decoded,
+                                       std::span<u8> upload_data) {
+    MICROPROFILE_SCOPE(RasterizerCache_ComputeHash);
+
+    if (custom_tex_manager.UseNewHash()) {
+        return Common::ComputeHash64(upload_data.data(), upload_data.size());
+    }
+    return Common::ComputeHash64(decoded.data(), decoded.size());
+}
+
 void RasterizerCache::DownloadSurface(const SurfaceRef& surface, SurfaceInterval interval) {
+    MICROPROFILE_SCOPE(RasterizerCache_DownloadSurface);
+
     const SurfaceParams flush_info = surface->FromInterval(interval);
     const u32 flush_start = boost::icl::first(interval);
     const u32 flush_end = boost::icl::last_next(interval);

src/video_core/rasterizer_cache/rasterizer_cache.h

@@ -4,7 +4,9 @@
 
 #pragma once
 
+#include <functional>
 #include <unordered_map>
+#include <vector>
 #include <boost/icl/interval_map.hpp>
 #include <boost/icl/interval_set.hpp>
 #include "video_core/rasterizer_cache/surface_base.h"
@@ -28,6 +30,8 @@ enum class ScaleMatch {
     Ignore   // accept every scaled res
 };
 
+class CustomTexManager;
+struct CustomTexture;
 class RendererBase;
 
 class RasterizerCache : NonCopyable {
@@ -62,10 +66,13 @@ public:
     };
 
 public:
-    RasterizerCache(Memory::MemorySystem& memory, OpenGL::TextureRuntime& runtime, Pica::Regs& regs,
-                    RendererBase& renderer);
+    RasterizerCache(Memory::MemorySystem& memory, CustomTexManager& custom_tex_manager,
+                    OpenGL::TextureRuntime& runtime, Pica::Regs& regs, RendererBase& renderer);
     ~RasterizerCache();
 
+    /// Notify the cache that a new frame has been queued
+    void TickFrame();
+
     /// Perform hardware accelerated texture copy according to the provided configuration
     bool AccelerateTextureCopy(const GPU::Regs::DisplayTransferConfig& config);
 
@@ -126,6 +133,13 @@ private:
     /// Copies pixel data in interval from the guest VRAM to the host GPU surface
     void UploadSurface(const SurfaceRef& surface, SurfaceInterval interval);
 
+    /// Uploads a custom texture identified with hash to the target surface
+    bool UploadCustomSurface(const SurfaceRef& surface, const SurfaceParams& load_info, u64 hash);
+
+    /// Returns the hash used to lookup the custom surface.
+    u64 ComputeCustomHash(const SurfaceParams& load_info, std::span<u8> decoded,
+                          std::span<u8> upload_data);
+
     /// Copies pixel data in interval from the host GPU surface to the guest VRAM
     void DownloadSurface(const SurfaceRef& surface, SurfaceInterval interval);
 
@@ -158,6 +172,7 @@ private:
 
 private:
     Memory::MemorySystem& memory;
+    CustomTexManager& custom_tex_manager;
     OpenGL::TextureRuntime& runtime;
     Pica::Regs& regs;
     RendererBase& renderer;
@@ -168,7 +183,9 @@ private:
     u32 resolution_scale_factor;
     RenderTargets render_targets;
     std::unordered_map<TextureCubeConfig, TextureCube> texture_cube_cache;
-    bool use_filter{};
+    bool use_filter;
+    bool dump_textures;
+    bool use_custom_textures;
 };
 
 } // namespace VideoCore

src/video_core/rasterizer_cache/surface_base.cpp

@@ -3,6 +3,7 @@
 // Refer to the license.txt file included.
 
 #include "common/alignment.h"
+#include "video_core/custom_textures/material.h"
 #include "video_core/rasterizer_cache/surface_base.h"
 #include "video_core/texture/texture_decode.h"
 
@@ -101,6 +102,10 @@ SurfaceInterval SurfaceBase::GetCopyableInterval(const SurfaceParams& params) co
     return result;
 }
 
+bool SurfaceBase::HasNormalMap() const noexcept {
+    return material && material->Map(MapType::Normal) != nullptr;
+}
+
 ClearValue SurfaceBase::MakeClearValue(PAddr copy_addr, PixelFormat dst_format) {
     const SurfaceType dst_type = GetFormatType(dst_format);
     const std::array fill_buffer = MakeFillBuffer(copy_addr);

src/video_core/rasterizer_cache/surface_base.h

@@ -11,6 +11,8 @@ namespace VideoCore {
 
 using SurfaceRegions = boost::icl::interval_set<PAddr, std::less, SurfaceInterval>;
 
+struct Material;
+
 class SurfaceBase : public SurfaceParams {
 public:
     SurfaceBase(const SurfaceParams& params);
@@ -28,10 +30,17 @@ public:
     /// Returns the clear value used to validate another surface from this fill surface
     ClearValue MakeClearValue(PAddr copy_addr, PixelFormat dst_format);
 
+    /// Returns true if the surface contains a custom material with a normal map.
+    bool HasNormalMap() const noexcept;
+
     u64 ModificationTick() const noexcept {
         return modification_tick;
     }
 
+    bool IsCustom() const noexcept {
+        return is_custom && custom_format != CustomPixelFormat::Invalid;
+    }
+
     bool IsRegionValid(SurfaceInterval interval) const {
         return (invalid_regions.find(interval) == invalid_regions.end());
     }
@@ -57,6 +66,8 @@ private:
 
 public:
     bool registered = false;
+    bool is_custom = false;
+    const Material* material = nullptr;
     SurfaceRegions invalid_regions;
     u32 fill_size = 0;
     std::array<u8, 4> fill_data;

src/video_core/rasterizer_cache/surface_params.cpp

@@ -215,12 +215,12 @@ u32 SurfaceParams::LevelOf(PAddr level_addr) const {
     return level;
 }
 
-std::string SurfaceParams::DebugName(bool scaled) const noexcept {
+std::string SurfaceParams::DebugName(bool scaled, bool custom) const noexcept {
     const u32 scaled_width = scaled ? GetScaledWidth() : width;
     const u32 scaled_height = scaled ? GetScaledHeight() : height;
-    return fmt::format("Surface: {}x{} {} {} levels from {:#x} to {:#x} ({})", scaled_width,
+    return fmt::format("Surface: {}x{} {} {} levels from {:#x} to {:#x} ({}{})", scaled_width,
                        scaled_height, PixelFormatAsString(pixel_format), levels, addr, end,
-                       scaled ? "scaled" : "unscaled");
+                       custom ? "custom," : "", scaled ? "scaled" : "unscaled");
 }
 
 } // namespace VideoCore

src/video_core/rasterizer_cache/surface_params.h

@@ -4,6 +4,7 @@
 
 #pragma once
 
+#include "video_core/custom_textures/custom_format.h"
 #include "video_core/rasterizer_cache/utils.h"
 
 namespace VideoCore {
@@ -46,7 +47,7 @@ public:
     u32 LevelOf(PAddr addr) const;
 
     /// Returns a string identifier of the params object
-    std::string DebugName(bool scaled) const noexcept;
+    std::string DebugName(bool scaled, bool custom = false) const noexcept;
 
     [[nodiscard]] SurfaceInterval GetInterval() const noexcept {
         return SurfaceInterval{addr, end};
@@ -101,6 +102,7 @@ public:
     bool is_tiled = false;
     TextureType texture_type = TextureType::Texture2D;
     PixelFormat pixel_format = PixelFormat::Invalid;
+    CustomPixelFormat custom_format = CustomPixelFormat::Invalid;
     SurfaceType type = SurfaceType::Invalid;
 
     std::array<u32, MAX_PICA_LEVELS> mipmap_offsets{};

src/video_core/renderer_opengl/gl_blit_helper.cpp

@@ -13,7 +13,6 @@
 #include "video_core/host_shaders/texture_filtering/nearest_neighbor_frag.h"
 #include "video_core/host_shaders/texture_filtering/refine_frag.h"
 #include "video_core/host_shaders/texture_filtering/scale_force_frag.h"
-#include "video_core/host_shaders/texture_filtering/tex_coord_vert.h"
 #include "video_core/host_shaders/texture_filtering/x_gradient_frag.h"
 #include "video_core/host_shaders/texture_filtering/xbrz_freescale_frag.h"
 #include "video_core/host_shaders/texture_filtering/y_gradient_frag.h"
@@ -81,7 +80,7 @@ bool BlitHelper::Filter(Surface& surface, const VideoCore::TextureBlit& blit) {
     }
 
     const OpenGLState prev_state = OpenGLState::GetCurState();
-    state.texture_units[0].texture_2d = surface.Handle(false);
+    state.texture_units[0].texture_2d = surface.Handle(0);
 
     const auto filter{Settings::values.texture_filter.GetValue()};
     switch (filter) {
@@ -135,7 +134,7 @@ void BlitHelper::FilterAnime4K(Surface& surface, const VideoCore::TextureBlit& b
     auto LUMAD = setup_temp_tex(GL_R16F, GL_RED, temp_rect.GetWidth(), temp_rect.GetHeight());
 
     // Copy to SRC
-    glCopyImageSubData(surface.Handle(false), GL_TEXTURE_2D, 0, blit.src_rect.left,
+    glCopyImageSubData(surface.Handle(0), GL_TEXTURE_2D, 0, blit.src_rect.left,
                        blit.src_rect.bottom, 0, SRC.tex.handle, GL_TEXTURE_2D, 0, 0, 0, 0,
                        src_width, src_height, 1);
 
@@ -161,47 +160,42 @@ void BlitHelper::FilterAnime4K(Surface& surface, const VideoCore::TextureBlit& b
 }
 
 void BlitHelper::FilterBicubic(Surface& surface, const VideoCore::TextureBlit& blit) {
-    SetParams(bicubic_program, surface.width, surface.height, blit.src_rect);
+    SetParams(bicubic_program, surface.Extent(), blit.src_rect);
     Draw(bicubic_program, surface.Handle(), filter_fbo.handle, blit.dst_level, blit.dst_rect);
 }
 
 void BlitHelper::FilterNearest(Surface& surface, const VideoCore::TextureBlit& blit) {
-    state.texture_units[2].texture_2d = surface.Handle(false);
-    SetParams(nearest_program, surface.width, surface.height, blit.src_rect);
+    state.texture_units[2].texture_2d = surface.Handle(0);
+    SetParams(nearest_program, surface.Extent(), blit.src_rect);
     Draw(nearest_program, surface.Handle(), filter_fbo.handle, blit.dst_level, blit.dst_rect);
 }
 
 void BlitHelper::FilterScaleForce(Surface& surface, const VideoCore::TextureBlit& blit) {
-    SetParams(scale_force_program, surface.width, surface.height, blit.src_rect);
+    SetParams(scale_force_program, surface.Extent(), blit.src_rect);
     Draw(scale_force_program, surface.Handle(), filter_fbo.handle, blit.dst_level, blit.dst_rect);
 }
 
 void BlitHelper::FilterXbrz(Surface& surface, const VideoCore::TextureBlit& blit) {
     glProgramUniform1f(xbrz_program.handle, 2, static_cast<GLfloat>(surface.res_scale));
-    SetParams(xbrz_program, surface.width, surface.height, blit.src_rect);
+    SetParams(xbrz_program, surface.Extent(), blit.src_rect);
     Draw(xbrz_program, surface.Handle(), filter_fbo.handle, blit.dst_level, blit.dst_rect);
 }
 
-void BlitHelper::SetParams(OGLProgram& program, u32 src_width, u32 src_height,
+void BlitHelper::SetParams(OGLProgram& program, const VideoCore::Extent& src_extent,
                            Common::Rectangle<u32> src_rect) {
     glProgramUniform2f(
         program.handle, 0,
-        static_cast<float>(src_rect.right - src_rect.left) / static_cast<float>(src_width),
-        static_cast<float>(src_rect.top - src_rect.bottom) / static_cast<float>(src_height));
+        static_cast<float>(src_rect.right - src_rect.left) / static_cast<float>(src_extent.width),
+        static_cast<float>(src_rect.top - src_rect.bottom) / static_cast<float>(src_extent.height));
     glProgramUniform2f(program.handle, 1,
-                       static_cast<float>(src_rect.left) / static_cast<float>(src_width),
-                       static_cast<float>(src_rect.bottom) / static_cast<float>(src_height));
+                       static_cast<float>(src_rect.left) / static_cast<float>(src_extent.width),
+                       static_cast<float>(src_rect.bottom) / static_cast<float>(src_extent.height));
 }
 
 void BlitHelper::Draw(OGLProgram& program, GLuint dst_tex, GLuint dst_fbo, u32 dst_level,
                       Common::Rectangle<u32> dst_rect) {
     state.draw.draw_framebuffer = dst_fbo;
     state.draw.shader_program = program.handle;
-    state.scissor.enabled = true;
-    state.scissor.x = dst_rect.left;
-    state.scissor.y = dst_rect.bottom;
-    state.scissor.width = dst_rect.GetWidth();
-    state.scissor.height = dst_rect.GetHeight();
     state.viewport.x = dst_rect.left;
     state.viewport.y = dst_rect.bottom;
     state.viewport.width = dst_rect.GetWidth();
@@ -212,7 +206,6 @@ void BlitHelper::Draw(OGLProgram& program, GLuint dst_tex, GLuint dst_fbo, u32 d
                            dst_level);
     glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
 
-    glClear(GL_COLOR_BUFFER_BIT);
     glDrawArrays(GL_TRIANGLE_STRIP, 0, 4);
 }
 

src/video_core/renderer_opengl/gl_blit_helper.h

@@ -9,8 +9,9 @@
 #include "video_core/renderer_opengl/gl_state.h"
 
 namespace VideoCore {
+struct Extent;
 struct TextureBlit;
-}
+} // namespace VideoCore
 
 namespace OpenGL {
 
@@ -35,7 +36,7 @@ private:
 
     void FilterXbrz(Surface& surface, const VideoCore::TextureBlit& blit);
 
-    void SetParams(OGLProgram& program, u32 src_width, u32 src_height,
+    void SetParams(OGLProgram& program, const VideoCore::Extent& src_extent,
                    Common::Rectangle<u32> src_rect);
 
     void Draw(OGLProgram& program, GLuint dst_tex, GLuint dst_fbo, u32 dst_level,

src/video_core/renderer_opengl/gl_driver.cpp

@@ -6,6 +6,7 @@
 #include "common/assert.h"
 #include "common/settings.h"
 #include "core/telemetry_session.h"
+#include "video_core/custom_textures/custom_format.h"
 #include "video_core/renderer_opengl/gl_driver.h"
 
 namespace OpenGL {
@@ -103,6 +104,25 @@ bool Driver::HasDebugTool() {
     return false;
 }
 
+bool Driver::IsCustomFormatSupported(VideoCore::CustomPixelFormat format) const {
+    switch (format) {
+    case VideoCore::CustomPixelFormat::RGBA8:
+        return true;
+    case VideoCore::CustomPixelFormat::BC1:
+    case VideoCore::CustomPixelFormat::BC3:
+    case VideoCore::CustomPixelFormat::BC5:
+        return ext_texture_compression_s3tc;
+    case VideoCore::CustomPixelFormat::BC7:
+        return arb_texture_compression_bptc;
+    case VideoCore::CustomPixelFormat::ASTC4:
+    case VideoCore::CustomPixelFormat::ASTC6:
+    case VideoCore::CustomPixelFormat::ASTC8:
+        return is_gles;
+    default:
+        return false;
+    }
+}
+
 void Driver::ReportDriverInfo() {
     // Report the context version and the vendor string
     gl_version = std::string_view{reinterpret_cast<const char*>(glGetString(GL_VERSION))};
@@ -145,7 +165,9 @@ void Driver::CheckExtensionSupport() {
     arb_buffer_storage = GLAD_GL_ARB_buffer_storage;
     arb_clear_texture = GLAD_GL_ARB_clear_texture;
     arb_get_texture_sub_image = GLAD_GL_ARB_get_texture_sub_image;
+    arb_texture_compression_bptc = GLAD_GL_ARB_texture_compression_bptc;
     ext_clip_cull_distance = GLAD_GL_EXT_clip_cull_distance;
+    ext_texture_compression_s3tc = GLAD_GL_EXT_texture_compression_s3tc;
     is_suitable = GLAD_GL_VERSION_4_3 || GLAD_GL_ES_VERSION_3_1;
 }
 

src/video_core/renderer_opengl/gl_driver.h

@@ -11,6 +11,10 @@ namespace Core {
 class TelemetrySession;
 }
 
+namespace VideoCore {
+enum class CustomPixelFormat : u32;
+}
+
 namespace OpenGL {
 
 enum class Vendor {
@@ -51,6 +55,9 @@ public:
     /// Returns true if any debug tool is attached
     bool HasDebugTool();
 
+    /// Returns true if the driver supports the provided custom format
+    bool IsCustomFormatSupported(VideoCore::CustomPixelFormat format) const;
+
     /// Returns the vendor of the currently selected physical device
     Vendor GetVendor() const {
         return vendor;
@@ -114,6 +121,8 @@ private:
     bool arb_clear_texture{};
     bool arb_get_texture_sub_image{};
     bool ext_clip_cull_distance{};
+    bool ext_texture_compression_s3tc{};
+    bool arb_texture_compression_bptc{};
 
     std::string_view gl_version{};
     std::string_view gpu_vendor{};

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -73,11 +73,13 @@ GLenum MakeAttributeType(Pica::PipelineRegs::VertexAttributeFormat format) {
 
 } // Anonymous namespace
 
-RasterizerOpenGL::RasterizerOpenGL(Memory::MemorySystem& memory, VideoCore::RendererBase& renderer,
-                                   Driver& driver_)
+RasterizerOpenGL::RasterizerOpenGL(Memory::MemorySystem& memory,
+                                   VideoCore::CustomTexManager& custom_tex_manager,
+                                   VideoCore::RendererBase& renderer, Driver& driver_)
     : VideoCore::RasterizerAccelerated{memory}, driver{driver_}, runtime{driver, renderer},
-      res_cache{memory, runtime, regs, renderer}, texture_buffer_size{TextureBufferSize()},
-      vertex_buffer{driver, GL_ARRAY_BUFFER, VERTEX_BUFFER_SIZE},
+      res_cache{memory, custom_tex_manager, runtime, regs, renderer},
+      texture_buffer_size{TextureBufferSize()}, vertex_buffer{driver, GL_ARRAY_BUFFER,
+                                                              VERTEX_BUFFER_SIZE},
       uniform_buffer{driver, GL_UNIFORM_BUFFER, UNIFORM_BUFFER_SIZE},
       index_buffer{driver, GL_ELEMENT_ARRAY_BUFFER, INDEX_BUFFER_SIZE},
       texture_buffer{driver, GL_TEXTURE_BUFFER, texture_buffer_size}, texture_lf_buffer{
@@ -183,6 +185,10 @@ RasterizerOpenGL::RasterizerOpenGL(Memory::MemorySystem& memory, VideoCore::Rend
 
 RasterizerOpenGL::~RasterizerOpenGL() = default;
 
+void RasterizerOpenGL::TickFrame() {
+    res_cache.TickFrame();
+}
+
 void RasterizerOpenGL::LoadDiskResources(const std::atomic_bool& stop_loading,
                                          const VideoCore::DiskResourceLoadCallback& callback) {
     shader_program_manager->LoadDiskCache(stop_loading, callback);
@@ -420,7 +426,6 @@ bool RasterizerOpenGL::Draw(bool accelerate, bool is_indexed) {
     state.scissor.y = draw_rect.bottom;
     state.scissor.width = draw_rect.GetWidth();
     state.scissor.height = draw_rect.GetHeight();
-    state.Apply();
 
     const u32 res_scale = framebuffer.ResolutionScale();
     if (uniform_block_data.data.framebuffer_scale != res_scale) {
@@ -444,10 +449,11 @@ bool RasterizerOpenGL::Draw(bool accelerate, bool is_indexed) {
 
     // Sync and bind the texture surfaces
     SyncTextureUnits(framebuffer);
+    state.Apply();
 
     // Sync and bind the shader
     if (shader_dirty) {
-        SetShader();
+        shader_program_manager->UseFragmentShader(regs, use_custom_normal);
         shader_dirty = false;
     }
 
@@ -546,6 +552,7 @@ void RasterizerOpenGL::SyncTextureUnits(const Framebuffer& framebuffer) {
             // on the male character's face, which in the OpenGL default appear black.
             state.texture_units[texture_index].texture_2d = default_texture;
         } else if (!IsFeedbackLoop(texture_index, framebuffer, *surface)) {
+            BindMaterial(texture_index, *surface);
             state.texture_units[texture_index].texture_2d = surface->Handle();
         }
     }
@@ -589,6 +596,33 @@ void RasterizerOpenGL::BindTextureCube(const Pica::TexturingRegs::FullTextureCon
     state.texture_units[0].texture_2d = 0;
 }
 
+void RasterizerOpenGL::BindMaterial(u32 texture_index, Surface& surface) {
+    if (!surface.IsCustom() || texture_index != 0) {
+        return;
+    }
+
+    const auto bind_texture = [&](const TextureUnits::TextureUnit& unit, GLuint texture,
+                                  GLuint sampler) {
+        glActiveTexture(unit.Enum());
+        glBindTexture(GL_TEXTURE_2D, texture);
+        glBindSampler(unit.id, sampler);
+    };
+
+    const GLuint sampler = texture_samplers[texture_index].sampler.handle;
+    if (surface.HasNormalMap()) {
+        if (regs.lighting.disable) {
+            LOG_WARNING(Render_OpenGL, "Custom normal map used but scene has no light enabled");
+        }
+        bind_texture(TextureUnits::TextureNormalMap, surface.Handle(2), sampler);
+        use_custom_normal = true;
+    } else {
+        if (use_custom_normal) {
+            bind_texture(TextureUnits::TextureNormalMap, 0, 0);
+        }
+        use_custom_normal = false;
+    }
+}
+
 bool RasterizerOpenGL::IsFeedbackLoop(u32 texture_index, const Framebuffer& framebuffer,
                                       Surface& surface) {
     const GLuint color_attachment = framebuffer.Attachment(SurfaceType::Color);
@@ -622,7 +656,6 @@ void RasterizerOpenGL::UnbindSpecial() {
     state.image_shadow_texture_pz = 0;
     state.image_shadow_texture_nz = 0;
     state.image_shadow_buffer = 0;
-    state.Apply();
 }
 
 void RasterizerOpenGL::NotifyFixedFunctionPicaRegisterChanged(u32 id) {
@@ -823,10 +856,6 @@ void RasterizerOpenGL::SamplerInfo::SyncWithConfig(
     }
 }
 
-void RasterizerOpenGL::SetShader() {
-    shader_program_manager->UseFragmentShader(Pica::g_state.regs);
-}
-
 void RasterizerOpenGL::SyncClipEnabled() {
     state.clip_distance[1] = Pica::g_state.regs.rasterizer.clip_enable != 0;
 }

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -18,6 +18,10 @@ namespace VideoCore {
 class RendererBase;
 }
 
+namespace VideoCore {
+class CustomTexManager;
+}
+
 namespace OpenGL {
 
 class Driver;
@@ -25,10 +29,12 @@ class ShaderProgramManager;
 
 class RasterizerOpenGL : public VideoCore::RasterizerAccelerated {
 public:
-    explicit RasterizerOpenGL(Memory::MemorySystem& memory, VideoCore::RendererBase& renderer,
-                              Driver& driver);
+    explicit RasterizerOpenGL(Memory::MemorySystem& memory,
+                              VideoCore::CustomTexManager& custom_tex_manager,
+                              VideoCore::RendererBase& renderer, Driver& driver);
     ~RasterizerOpenGL() override;
 
+    void TickFrame();
     void LoadDiskResources(const std::atomic_bool& stop_loading,
                            const VideoCore::DiskResourceLoadCallback& callback) override;
 
@@ -74,9 +80,6 @@ private:
     /// Syncs the clip enabled status to match the PICA register
     void SyncClipEnabled();
 
-    /// Sets the OpenGL shader in accordance with the current PICA register state
-    void SetShader();
-
     /// Syncs the cull mode to match the PICA register
     void SyncCullMode();
 
@@ -126,6 +129,9 @@ private:
     /// Unbinds all special texture unit 0 texture configurations
     void UnbindSpecial();
 
+    /// Binds the custom material referenced by surface if it exists.
+    void BindMaterial(u32 texture_index, Surface& surface);
+
     /// Upload the uniform blocks to the uniform buffer object
     void UploadUniforms(bool accelerate_draw);
 
@@ -174,6 +180,7 @@ private:
     OGLTexture texture_buffer_lut_lf;
     OGLTexture texture_buffer_lut_rg;
     OGLTexture texture_buffer_lut_rgba;
+    bool use_custom_normal{};
 };
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_shader_gen.cpp

@@ -59,7 +59,7 @@ out gl_PerVertex {
     return out;
 }
 
-PicaFSConfig PicaFSConfig::BuildFromRegs(const Pica::Regs& regs) {
+PicaFSConfig PicaFSConfig::BuildFromRegs(const Pica::Regs& regs, bool use_normal) {
     PicaFSConfig res{};
 
     auto& state = res.state;
@@ -204,6 +204,8 @@ PicaFSConfig PicaFSConfig::BuildFromRegs(const Pica::Regs& regs) {
 
     state.shadow_texture_orthographic = regs.texturing.shadow.orthographic != 0;
 
+    state.use_custom_normal_map = use_normal;
+
     return res;
 }
 
@@ -297,6 +299,8 @@ static std::string SampleTexture(const PicaFSConfig& config, unsigned texture_un
             LOG_DEBUG(Render_OpenGL, "Using Texture3 without enabling it");
             return "vec4(0.0)";
         }
+    case 4:
+        return "texture(tex_normal, texcoord0)";
     default:
         UNREACHABLE();
         return "";
@@ -642,7 +646,12 @@ static void WriteLighting(std::string& out, const PicaFSConfig& config) {
     const auto Perturbation = [&] {
         return fmt::format("2.0 * ({}).rgb - 1.0", SampleTexture(config, lighting.bump_selector));
     };
-    if (lighting.bump_mode == LightingRegs::LightingBumpMode::NormalMap) {
+    if (config.state.use_custom_normal_map) {
+        const std::string normal_texel =
+            fmt::format("2.0 * ({}).rgb - 1.0", SampleTexture(config, 4));
+        out += fmt::format("vec3 surface_normal = {};\n", normal_texel);
+        out += "vec3 surface_tangent = vec3(1.0, 0.0, 0.0);\n";
+    } else if (lighting.bump_mode == LightingRegs::LightingBumpMode::NormalMap) {
         // Bump mapping is enabled using a normal map
         out += fmt::format("vec3 surface_normal = {};\n", Perturbation());
 
@@ -1207,6 +1216,7 @@ out vec4 color;
 uniform sampler2D tex0;
 uniform sampler2D tex1;
 uniform sampler2D tex2;
+uniform sampler2D tex_normal; //< Used for custom normal maps
 uniform samplerCube tex_cube;
 uniform samplerBuffer texture_buffer_lut_lf;
 uniform samplerBuffer texture_buffer_lut_rg;

src/video_core/renderer_opengl/gl_shader_gen.h

@@ -117,6 +117,7 @@ struct PicaFSConfigState {
 
     bool shadow_rendering;
     bool shadow_texture_orthographic;
+    bool use_custom_normal_map;
 };
 
 /**
@@ -130,7 +131,7 @@ struct PicaFSConfigState {
 struct PicaFSConfig : Common::HashableStruct<PicaFSConfigState> {
 
     /// Construct a PicaFSConfig with the given Pica register configuration.
-    static PicaFSConfig BuildFromRegs(const Pica::Regs& regs);
+    static PicaFSConfig BuildFromRegs(const Pica::Regs& regs, bool use_normal = false);
 
     bool TevStageUpdatesCombinerBufferColor(unsigned stage_index) const {
         return (stage_index < 4) && (state.combiner_buffer_input & (1 << stage_index));

src/video_core/renderer_opengl/gl_shader_manager.cpp

@@ -133,6 +133,7 @@ static void SetShaderSamplerBindings(GLuint shader) {
     SetShaderSamplerBinding(shader, "tex1", TextureUnits::PicaTexture(1));
     SetShaderSamplerBinding(shader, "tex2", TextureUnits::PicaTexture(2));
     SetShaderSamplerBinding(shader, "tex_cube", TextureUnits::TextureCube);
+    SetShaderSamplerBinding(shader, "tex_normal", TextureUnits::TextureNormalMap);
 
     // Set the texture samplers to correspond to different lookup table texture units
     SetShaderSamplerBinding(shader, "texture_buffer_lut_lf", TextureUnits::TextureBufferLUT_LF);
@@ -415,8 +416,8 @@ void ShaderProgramManager::UseTrivialGeometryShader() {
     impl->current.gs_hash = 0;
 }
 
-void ShaderProgramManager::UseFragmentShader(const Pica::Regs& regs) {
-    PicaFSConfig config = PicaFSConfig::BuildFromRegs(regs);
+void ShaderProgramManager::UseFragmentShader(const Pica::Regs& regs, bool use_normal) {
+    PicaFSConfig config = PicaFSConfig::BuildFromRegs(regs, use_normal);
     auto [handle, result] = impl->fragment_shaders.Get(config);
     impl->current.fs = handle;
     impl->current.fs_hash = config.Hash();

src/video_core/renderer_opengl/gl_shader_manager.h

@@ -41,7 +41,7 @@ public:
 
     void UseTrivialGeometryShader();
 
-    void UseFragmentShader(const Pica::Regs& config);
+    void UseFragmentShader(const Pica::Regs& config, bool use_normal);
 
     void ApplyTo(OpenGLState& state);
 

src/video_core/renderer_opengl/gl_state.h

@@ -26,6 +26,7 @@ constexpr TextureUnit TextureCube{6};
 constexpr TextureUnit TextureBufferLUT_LF{3};
 constexpr TextureUnit TextureBufferLUT_RG{4};
 constexpr TextureUnit TextureBufferLUT_RGBA{5};
+constexpr TextureUnit TextureNormalMap{7};
 
 } // namespace TextureUnits
 

src/video_core/renderer_opengl/gl_texture_runtime.cpp

@@ -3,7 +3,7 @@
 // Refer to the license.txt file included.
 
 #include "common/scope_exit.h"
-#include "common/settings.h"
+#include "video_core/custom_textures/material.h"
 #include "video_core/regs.h"
 #include "video_core/renderer_base.h"
 #include "video_core/renderer_opengl/gl_driver.h"
@@ -14,8 +14,10 @@ namespace OpenGL {
 
 namespace {
 
+using VideoCore::MapType;
 using VideoCore::PixelFormat;
 using VideoCore::SurfaceType;
+using VideoCore::TextureType;
 
 constexpr FormatTuple DEFAULT_TUPLE = {GL_RGBA8, GL_RGBA, GL_UNSIGNED_BYTE};
 
@@ -42,6 +44,17 @@ static constexpr std::array<FormatTuple, 5> COLOR_TUPLES_OES = {{
     {GL_RGBA4, GL_RGBA, GL_UNSIGNED_SHORT_4_4_4_4},   // RGBA4
 }};
 
+static constexpr std::array<FormatTuple, 8> CUSTOM_TUPLES = {{
+    DEFAULT_TUPLE,
+    {GL_COMPRESSED_RGBA_S3TC_DXT1_EXT, GL_COMPRESSED_RGBA_S3TC_DXT1_EXT, GL_UNSIGNED_BYTE},
+    {GL_COMPRESSED_RGBA_S3TC_DXT5_EXT, GL_COMPRESSED_RGBA_S3TC_DXT5_EXT, GL_UNSIGNED_BYTE},
+    {GL_COMPRESSED_RG_RGTC2, GL_COMPRESSED_RG_RGTC2, GL_UNSIGNED_BYTE},
+    {GL_COMPRESSED_RGBA_BPTC_UNORM_ARB, GL_COMPRESSED_RGBA_BPTC_UNORM_ARB, GL_UNSIGNED_BYTE},
+    {GL_COMPRESSED_RGBA_ASTC_4x4, GL_COMPRESSED_RGBA_ASTC_4x4, GL_UNSIGNED_BYTE},
+    {GL_COMPRESSED_RGBA_ASTC_6x6, GL_COMPRESSED_RGBA_ASTC_6x6, GL_UNSIGNED_BYTE},
+    {GL_COMPRESSED_RGBA_ASTC_8x6, GL_COMPRESSED_RGBA_ASTC_8x6, GL_UNSIGNED_BYTE},
+}};
+
 struct FramebufferInfo {
     GLuint color;
     GLuint depth;
@@ -109,17 +122,23 @@ TextureRuntime::TextureRuntime(const Driver& driver_, VideoCore::RendererBase& r
         read_fbos[i].Create();
     }
 
-    auto Register = [this](PixelFormat dest, std::unique_ptr<FormatReinterpreterBase>&& obj) {
+    auto add_reinterpreter = [this](PixelFormat dest,
+                                    std::unique_ptr<FormatReinterpreterBase>&& obj) {
         const u32 dst_index = static_cast<u32>(dest);
         return reinterpreters[dst_index].push_back(std::move(obj));
     };
 
-    Register(PixelFormat::RGBA8, std::make_unique<ShaderD24S8toRGBA8>());
-    Register(PixelFormat::RGB5A1, std::make_unique<RGBA4toRGB5A1>());
+    add_reinterpreter(PixelFormat::RGBA8, std::make_unique<ShaderD24S8toRGBA8>());
+    add_reinterpreter(PixelFormat::RGB5A1, std::make_unique<RGBA4toRGB5A1>());
 }
 
 TextureRuntime::~TextureRuntime() = default;
 
+void TextureRuntime::Reset() {
+    alloc_cache.clear();
+    framebuffer_cache.clear();
+}
+
 bool TextureRuntime::NeedsConversion(VideoCore::PixelFormat pixel_format) const {
     const bool should_convert = pixel_format == PixelFormat::RGBA8 || // Needs byteswap
                                 pixel_format == PixelFormat::RGB8;    // Is converted to RGBA8
@@ -153,51 +172,64 @@ const FormatTuple& TextureRuntime::GetFormatTuple(PixelFormat pixel_format) cons
     return DEFAULT_TUPLE;
 }
 
-void TextureRuntime::Recycle(const HostTextureTag tag, Allocation&& alloc) {
-    recycler.emplace(tag, std::move(alloc));
+const FormatTuple& TextureRuntime::GetFormatTuple(VideoCore::CustomPixelFormat pixel_format) {
+    const std::size_t format_index = static_cast<std::size_t>(pixel_format);
+    return CUSTOM_TUPLES[format_index];
 }
 
-Allocation TextureRuntime::Allocate(const VideoCore::SurfaceParams& params) {
-    const u32 width = params.width;
-    const u32 height = params.height;
-    const u32 levels = params.levels;
+void TextureRuntime::Recycle(const HostTextureTag tag, Allocation&& alloc) {
+    alloc_cache.emplace(tag, std::move(alloc));
+}
+
+Allocation TextureRuntime::Allocate(const VideoCore::SurfaceParams& params,
+                                    const VideoCore::Material* material) {
     const GLenum target = params.texture_type == VideoCore::TextureType::CubeMap
                               ? GL_TEXTURE_CUBE_MAP
                               : GL_TEXTURE_2D;
-    const auto& tuple = GetFormatTuple(params.pixel_format);
-
+    const bool is_custom = material != nullptr;
+    const bool has_normal = material && material->Map(MapType::Normal);
+    const auto& tuple =
+        is_custom ? GetFormatTuple(params.custom_format) : GetFormatTuple(params.pixel_format);
     const HostTextureTag key = {
+        .width = params.width,
+        .height = params.height,
+        .levels = params.levels,
+        .res_scale = params.res_scale,
         .tuple = tuple,
         .type = params.texture_type,
-        .width = width,
-        .height = height,
-        .levels = levels,
-        .res_scale = params.res_scale,
+        .is_custom = is_custom,
+        .has_normal = has_normal,
     };
 
-    if (auto it = recycler.find(key); it != recycler.end()) {
-        Allocation alloc = std::move(it->second);
-        ASSERT(alloc.res_scale == params.res_scale);
-        recycler.erase(it);
+    if (auto it = alloc_cache.find(key); it != alloc_cache.end()) {
+        auto alloc{std::move(it->second)};
+        alloc_cache.erase(it);
         return alloc;
     }
 
     const GLuint old_tex = OpenGLState::GetCurState().texture_units[0].texture_2d;
     glActiveTexture(GL_TEXTURE0);
 
-    std::array<OGLTexture, 2> textures{};
-    std::array<GLuint, 2> handles{};
+    std::array<OGLTexture, 3> textures{};
+    std::array<GLuint, 3> handles{};
 
-    textures[0] = MakeHandle(target, width, height, levels, tuple, params.DebugName(false));
+    textures[0] = MakeHandle(target, params.width, params.height, params.levels, tuple,
+                             params.DebugName(false));
     handles.fill(textures[0].handle);
 
     if (params.res_scale != 1) {
-        const u32 scaled_width = params.GetScaledWidth();
-        const u32 scaled_height = params.GetScaledHeight();
-        textures[1] =
-            MakeHandle(target, scaled_width, scaled_height, levels, tuple, params.DebugName(true));
+        const u32 scaled_width = is_custom ? params.width : params.GetScaledWidth();
+        const u32 scaled_height = is_custom ? params.height : params.GetScaledHeight();
+        const auto& scaled_tuple = is_custom ? GetFormatTuple(PixelFormat::RGBA8) : tuple;
+        textures[1] = MakeHandle(target, scaled_width, scaled_height, params.levels, scaled_tuple,
+                                 params.DebugName(true, is_custom));
         handles[1] = textures[1].handle;
     }
+    if (has_normal) {
+        textures[2] = MakeHandle(target, params.width, params.height, params.levels, tuple,
+                                 params.DebugName(true, is_custom));
+        handles[2] = textures[2].handle;
+    }
 
     glBindTexture(GL_TEXTURE_2D, old_tex);
 
@@ -311,15 +343,20 @@ bool TextureRuntime::BlitTextures(Surface& source, Surface& dest,
     return true;
 }
 
-void TextureRuntime::GenerateMipmaps(Surface& surface, u32 max_level) {
+void TextureRuntime::GenerateMipmaps(Surface& surface) {
     OpenGLState state = OpenGLState::GetCurState();
-    state.texture_units[0].texture_2d = surface.Handle();
+
+    const auto generate = [&](u32 index) {
+        state.texture_units[0].texture_2d = surface.Handle(index);
         state.Apply();
-
-    glActiveTexture(GL_TEXTURE0);
-    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, max_level);
-
+        glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAX_LEVEL, surface.levels - 1);
         glGenerateMipmap(GL_TEXTURE_2D);
+    };
+
+    generate(1);
+    if (surface.HasNormalMap()) {
+        generate(2);
+    }
 }
 
 const ReinterpreterList& TextureRuntime::GetPossibleReinterpretations(
@@ -340,21 +377,11 @@ Surface::~Surface() {
     if (pixel_format == PixelFormat::Invalid || !alloc) {
         return;
     }
-
-    const HostTextureTag tag = {
-        .tuple = alloc.tuple,
-        .type = texture_type,
-        .width = alloc.width,
-        .height = alloc.height,
-        .levels = alloc.levels,
-        .res_scale = alloc.res_scale,
-    };
-    runtime->Recycle(tag, std::move(alloc));
+    runtime->Recycle(MakeTag(), std::move(alloc));
 }
 
 void Surface::Upload(const VideoCore::BufferTextureCopy& upload,
                      const VideoCore::StagingData& staging) {
-    // Ensure no bad interactions with GL_UNPACK_ALIGNMENT
     ASSERT(stride * GetFormatBytesPerPixel(pixel_format) % 4 == 0);
 
     const GLuint old_tex = OpenGLState::GetCurState().texture_units[0].texture_2d;
@@ -362,12 +389,10 @@ void Surface::Upload(const VideoCore::BufferTextureCopy& upload,
     const u32 unscaled_height = upload.texture_rect.GetHeight();
     glPixelStorei(GL_UNPACK_ROW_LENGTH, unscaled_width);
 
-    // Bind the unscaled texture
     glActiveTexture(GL_TEXTURE0);
-    glBindTexture(GL_TEXTURE_2D, Handle(false));
+    glBindTexture(GL_TEXTURE_2D, Handle(0));
 
-    // Upload the requested rectangle of pixels
-    const auto& tuple = runtime->GetFormatTuple(pixel_format);
+    const auto& tuple = alloc.tuple;
     glTexSubImage2D(GL_TEXTURE_2D, upload.texture_level, upload.texture_rect.left,
                     upload.texture_rect.bottom, unscaled_width, unscaled_height, tuple.format,
                     tuple.type, staging.mapped.data());
@@ -381,17 +406,61 @@ void Surface::Upload(const VideoCore::BufferTextureCopy& upload,
         .src_rect = upload.texture_rect,
         .dst_rect = upload.texture_rect * res_scale,
     };
-
-    // If texture filtering is enabled attempt to upscale with that, otherwise fallback
-    // to normal blit.
     if (res_scale != 1 && !runtime->blit_helper.Filter(*this, blit)) {
         BlitScale(blit, true);
     }
 }
 
+void Surface::UploadCustom(const VideoCore::Material* material, u32 level) {
+    const GLuint old_tex = OpenGLState::GetCurState().texture_units[0].texture_2d;
+    const auto& tuple = alloc.tuple;
+    const u32 width = material->width;
+    const u32 height = material->height;
+    const auto color = material->textures[0];
+    const Common::Rectangle filter_rect{0U, height, width, 0U};
+
+    glActiveTexture(GL_TEXTURE0);
+    glPixelStorei(GL_UNPACK_ROW_LENGTH, width);
+
+    glBindTexture(GL_TEXTURE_2D, Handle(0));
+    if (VideoCore::IsCustomFormatCompressed(custom_format)) {
+        const GLsizei image_size = static_cast<GLsizei>(color->data.size());
+        glCompressedTexSubImage2D(GL_TEXTURE_2D, level, 0, 0, width, height, tuple.format,
+                                  image_size, color->data.data());
+    } else {
+        glTexSubImage2D(GL_TEXTURE_2D, level, 0, 0, width, height, tuple.format, tuple.type,
+                        color->data.data());
+    }
+
+    const VideoCore::TextureBlit blit = {
+        .src_rect = filter_rect,
+        .dst_rect = filter_rect,
+    };
+    if (res_scale != 1 && !runtime->blit_helper.Filter(*this, blit)) {
+        BlitScale(blit, true);
+    }
+    for (u32 i = 1; i < VideoCore::MAX_MAPS; i++) {
+        const auto texture = material->textures[i];
+        if (!texture) {
+            continue;
+        }
+        glBindTexture(GL_TEXTURE_2D, Handle(i + 1));
+        if (VideoCore::IsCustomFormatCompressed(custom_format)) {
+            const GLsizei image_size = static_cast<GLsizei>(texture->data.size());
+            glCompressedTexSubImage2D(GL_TEXTURE_2D, level, 0, 0, width, height, tuple.format,
+                                      image_size, texture->data.data());
+        } else {
+            glTexSubImage2D(GL_TEXTURE_2D, level, 0, 0, width, height, tuple.format, tuple.type,
+                            texture->data.data());
+        }
+    }
+
+    glPixelStorei(GL_UNPACK_ROW_LENGTH, 0);
+    glBindTexture(GL_TEXTURE_2D, old_tex);
+}
+
 void Surface::Download(const VideoCore::BufferTextureCopy& download,
                        const VideoCore::StagingData& staging) {
-    // Ensure no bad interactions with GL_PACK_ALIGNMENT
     ASSERT(stride * GetFormatBytesPerPixel(pixel_format) % 4 == 0);
 
     const u32 unscaled_width = download.texture_rect.GetWidth();
@@ -414,11 +483,9 @@ void Surface::Download(const VideoCore::BufferTextureCopy& download,
         return;
     }
 
-    const u32 fbo_index = FboIndex(type);
-
     OpenGLState state = OpenGLState::GetCurState();
     state.scissor.enabled = false;
-    state.draw.read_framebuffer = runtime->read_fbos[fbo_index].handle;
+    state.draw.read_framebuffer = runtime->read_fbos[FboIndex(type)].handle;
     state.Apply();
 
     Attach(GL_READ_FRAMEBUFFER, download.texture_level, 0, false);
@@ -428,14 +495,11 @@ void Surface::Download(const VideoCore::BufferTextureCopy& download,
     glReadPixels(download.texture_rect.left, download.texture_rect.bottom, unscaled_width,
                  unscaled_height, tuple.format, tuple.type, staging.mapped.data());
 
-    // Restore previous state
     glPixelStorei(GL_PACK_ROW_LENGTH, 0);
 }
 
 bool Surface::DownloadWithoutFbo(const VideoCore::BufferTextureCopy& download,
                                  const VideoCore::StagingData& staging) {
-    // If a partial download is requested and ARB_get_texture_sub_image (core in 4.5)
-    // is not available we cannot proceed further.
     const bool is_full_download = download.texture_rect == GetRect();
     const bool has_sub_image = driver->HasArbGetTextureSubImage();
     if (driver->IsOpenGLES() || (!is_full_download && !has_sub_image)) {
@@ -452,7 +516,7 @@ bool Surface::DownloadWithoutFbo(const VideoCore::BufferTextureCopy& download,
     // Prefer glGetTextureSubImage in most cases since it's the fastest and most convenient option
     if (has_sub_image) {
         const GLsizei buf_size = static_cast<GLsizei>(staging.mapped.size());
-        glGetTextureSubImage(Handle(false), download.texture_level, download.texture_rect.left,
+        glGetTextureSubImage(Handle(0), download.texture_level, download.texture_rect.left,
                              download.texture_rect.bottom, 0, download.texture_rect.GetWidth(),
                              download.texture_rect.GetHeight(), 1, tuple.format, tuple.type,
                              buf_size, staging.mapped.data());
@@ -461,7 +525,7 @@ bool Surface::DownloadWithoutFbo(const VideoCore::BufferTextureCopy& download,
 
     // This should only trigger for full texture downloads in oldish intel drivers
     // that only support up to 4.3
-    glBindTexture(GL_TEXTURE_2D, Handle(false));
+    glBindTexture(GL_TEXTURE_2D, Handle(0));
     glGetTexImage(GL_TEXTURE_2D, download.texture_level, tuple.format, tuple.type,
                   staging.mapped.data());
     glBindTexture(GL_TEXTURE_2D, old_tex);
@@ -470,20 +534,20 @@ bool Surface::DownloadWithoutFbo(const VideoCore::BufferTextureCopy& download,
 }
 
 void Surface::Attach(GLenum target, u32 level, u32 layer, bool scaled) {
-    const GLuint handle = Handle(scaled);
-    const GLenum textarget = texture_type == VideoCore::TextureType::CubeMap
+    const GLuint handle = Handle(static_cast<u32>(scaled));
+    const GLenum textarget = texture_type == TextureType::CubeMap
                                  ? GL_TEXTURE_CUBE_MAP_POSITIVE_X + layer
                                  : GL_TEXTURE_2D;
 
     switch (type) {
-    case VideoCore::SurfaceType::Color:
-    case VideoCore::SurfaceType::Texture:
+    case SurfaceType::Color:
+    case SurfaceType::Texture:
         glFramebufferTexture2D(target, GL_COLOR_ATTACHMENT0, textarget, handle, level);
         break;
-    case VideoCore::SurfaceType::Depth:
+    case SurfaceType::Depth:
         glFramebufferTexture2D(target, GL_DEPTH_ATTACHMENT, textarget, handle, level);
         break;
-    case VideoCore::SurfaceType::DepthStencil:
+    case SurfaceType::DepthStencil:
         glFramebufferTexture2D(target, GL_DEPTH_STENCIL_ATTACHMENT, textarget, handle, level);
         break;
     default:
@@ -491,6 +555,30 @@ void Surface::Attach(GLenum target, u32 level, u32 layer, bool scaled) {
     }
 }
 
+bool Surface::Swap(const VideoCore::Material* mat) {
+    const VideoCore::CustomPixelFormat format{mat->format};
+    if (!driver->IsCustomFormatSupported(format)) {
+        return false;
+    }
+    runtime->Recycle(MakeTag(), std::move(alloc));
+
+    SurfaceParams params = *this;
+    params.width = mat->width;
+    params.height = mat->height;
+    params.custom_format = mat->format;
+    alloc = runtime->Allocate(params, mat);
+
+    LOG_DEBUG(Render_OpenGL, "Swapped {}x{} {} surface at address {:#x} to {}x{} {}",
+              GetScaledWidth(), GetScaledHeight(), VideoCore::PixelFormatAsString(pixel_format),
+              addr, width, height, VideoCore::CustomPixelFormatAsString(format));
+
+    is_custom = true;
+    custom_format = format;
+    material = mat;
+
+    return true;
+}
+
 u32 Surface::GetInternalBytesPerPixel() const {
     // RGB8 is converted to RGBA8 on OpenGL ES since it doesn't support BGR8
     if (driver->IsOpenGLES() && pixel_format == VideoCore::PixelFormat::RGB8) {
@@ -518,6 +606,19 @@ void Surface::BlitScale(const VideoCore::TextureBlit& blit, bool up_scale) {
                       blit.dst_rect.right, blit.dst_rect.top, buffer_mask, filter);
 }
 
+HostTextureTag Surface::MakeTag() const noexcept {
+    return HostTextureTag{
+        .width = alloc.width,
+        .height = alloc.height,
+        .levels = alloc.levels,
+        .res_scale = alloc.res_scale,
+        .tuple = alloc.tuple,
+        .type = texture_type,
+        .is_custom = is_custom,
+        .has_normal = HasNormalMap(),
+    };
+}
+
 Framebuffer::Framebuffer(TextureRuntime& runtime, Surface* const color, u32 color_level,
                          Surface* const depth_stencil, u32 depth_level, const Pica::Regs& regs,
                          Common::Rectangle<u32> surfaces_rect)
@@ -527,7 +628,7 @@ Framebuffer::Framebuffer(TextureRuntime& runtime, Surface* const color, u32 colo
     const bool shadow_rendering = regs.framebuffer.IsShadowRendering();
     const bool has_stencil = regs.framebuffer.HasStencil();
     if (shadow_rendering && !color) {
-        return; // Framebuffer won't get used
+        return;
     }
 
     if (color) {
@@ -574,19 +675,15 @@ Framebuffer::Framebuffer(TextureRuntime& runtime, Surface* const color, u32 colo
                                color ? color->Handle() : 0, color_level);
         if (depth_stencil) {
             if (has_stencil) {
-                // Attach both depth and stencil
                 glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT,
                                        GL_TEXTURE_2D, depth_stencil->Handle(), depth_level);
             } else {
-                // Attach depth
                 glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D,
                                        depth_stencil->Handle(), depth_level);
-                // Clear stencil attachment
                 glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
                                        0);
             }
         } else {
-            // Clear both depth and stencil attachment
             glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
                                    0, 0);
         }

src/video_core/renderer_opengl/gl_texture_runtime.h

@@ -10,8 +10,9 @@
 #include "video_core/renderer_opengl/gl_format_reinterpreter.h"
 
 namespace VideoCore {
+struct Material;
 class RendererBase;
-}
+} // namespace VideoCore
 
 namespace OpenGL {
 
@@ -27,17 +28,19 @@ struct FormatTuple {
 };
 
 struct HostTextureTag {
-    FormatTuple tuple{};
-    VideoCore::TextureType type{};
-    u32 width = 0;
-    u32 height = 0;
-    u32 levels = 1;
-    u32 res_scale = 1;
+    u32 width;
+    u32 height;
+    u32 levels;
+    u32 res_scale;
+    FormatTuple tuple;
+    VideoCore::TextureType type;
+    bool is_custom;
+    bool has_normal;
 
     bool operator==(const HostTextureTag& other) const noexcept {
-        return std::tie(tuple, type, width, height, levels, res_scale) ==
+        return std::tie(tuple, type, width, height, levels, res_scale, is_custom, has_normal) ==
                std::tie(other.tuple, other.type, other.width, other.height, other.levels,
-                        other.res_scale);
+                        other.res_scale, other.is_custom, other.has_normal);
     }
 
     struct Hash {
@@ -46,10 +49,12 @@ struct HostTextureTag {
         }
     };
 };
+static_assert(std::has_unique_object_representations_v<HostTextureTag>,
+              "HostTextureTag is not suitable for hashing!");
 
 struct Allocation {
-    std::array<OGLTexture, 2> textures;
-    std::array<GLuint, 2> handles;
+    std::array<OGLTexture, 3> textures;
+    std::array<GLuint, 3> handles;
     FormatTuple tuple;
     u32 width;
     u32 height;
@@ -59,15 +64,10 @@ struct Allocation {
     operator bool() const noexcept {
         return textures[0].handle;
     }
-
-    bool Matches(u32 width_, u32 height_, u32 levels_, const FormatTuple& tuple_) const {
-        return std::tie(width, height, levels, tuple) == std::tie(width_, height_, levels_, tuple_);
-    }
 };
 
 class Surface;
 class Driver;
-struct CachedTextureCube;
 
 /**
  * Provides texture manipulation functions to the rasterizer cache
@@ -82,6 +82,9 @@ public:
     explicit TextureRuntime(const Driver& driver, VideoCore::RendererBase& renderer);
     ~TextureRuntime();
 
+    /// Clears all cached runtime resources
+    void Reset();
+
     /// Returns true if the provided pixel format cannot be used natively by the runtime.
     bool NeedsConversion(VideoCore::PixelFormat pixel_format) const;
 
@@ -90,12 +93,14 @@ public:
 
     /// Returns the OpenGL format tuple associated with the provided pixel format
     const FormatTuple& GetFormatTuple(VideoCore::PixelFormat pixel_format) const;
+    const FormatTuple& GetFormatTuple(VideoCore::CustomPixelFormat pixel_format);
 
     /// Takes back ownership of the allocation for recycling
     void Recycle(const HostTextureTag tag, Allocation&& alloc);
 
-    /// Allocates an OpenGL texture with the specified dimentions and format
-    Allocation Allocate(const VideoCore::SurfaceParams& params);
+    /// Allocates a texture with the specified dimentions and format
+    Allocation Allocate(const VideoCore::SurfaceParams& params,
+                        const VideoCore::Material* material = nullptr);
 
     /// Fills the rectangle of the texture with the clear value provided
     bool ClearTexture(Surface& surface, const VideoCore::TextureClear& clear);
@@ -107,7 +112,7 @@ public:
     bool BlitTextures(Surface& source, Surface& dest, const VideoCore::TextureBlit& blit);
 
     /// Generates mipmaps for all the available levels of the texture
-    void GenerateMipmaps(Surface& surface, u32 max_level);
+    void GenerateMipmaps(Surface& surface);
 
     /// Returns all source formats that support reinterpretation to the dest format
     const ReinterpreterList& GetPossibleReinterpretations(VideoCore::PixelFormat dest_format) const;
@@ -123,7 +128,7 @@ private:
     BlitHelper blit_helper;
     std::vector<u8> staging_buffer;
     std::array<ReinterpreterList, VideoCore::PIXEL_FORMAT_COUNT> reinterpreters;
-    std::unordered_multimap<HostTextureTag, Allocation, HostTextureTag::Hash> recycler;
+    std::unordered_multimap<HostTextureTag, Allocation, HostTextureTag::Hash> alloc_cache;
     std::unordered_map<u64, OGLFramebuffer, Common::IdentityHash<u64>> framebuffer_cache;
     std::array<OGLFramebuffer, 3> draw_fbos;
     std::array<OGLFramebuffer, 3> read_fbos;
@@ -140,9 +145,9 @@ public:
     Surface(Surface&& o) noexcept = default;
     Surface& operator=(Surface&& o) noexcept = default;
 
-    /// Returns the surface image handle
-    GLuint Handle(bool scaled = true) const noexcept {
-        return alloc.handles[static_cast<u32>(scaled)];
+    /// Returns the surface image handle at the provided index.
+    GLuint Handle(u32 index = 1) const noexcept {
+        return alloc.handles[index];
     }
 
     /// Returns the tuple of the surface allocation.
@@ -150,9 +155,20 @@ public:
         return alloc.tuple;
     }
 
+    /// Returns the extent of the underlying surface allocation
+    VideoCore::Extent Extent() const noexcept {
+        return {
+            .width = alloc.width,
+            .height = alloc.height,
+        };
+    }
+
     /// Uploads pixel data in staging to a rectangle region of the surface texture
     void Upload(const VideoCore::BufferTextureCopy& upload, const VideoCore::StagingData& staging);
 
+    /// Uploads the custom material to the surface allocation.
+    void UploadCustom(const VideoCore::Material* material, u32 level);
+
     /// Downloads pixel data to staging from a rectangle region of the surface texture
     void Download(const VideoCore::BufferTextureCopy& download,
                   const VideoCore::StagingData& staging);
@@ -160,6 +176,9 @@ public:
     /// Attaches a handle of surface to the specified framebuffer target
     void Attach(GLenum target, u32 level, u32 layer, bool scaled = true);
 
+    /// Swaps the internal allocation to match the provided material
+    bool Swap(const VideoCore::Material* material);
+
     /// Returns the bpp of the internal surface format
     u32 GetInternalBytesPerPixel() const;
 
@@ -171,6 +190,9 @@ private:
     bool DownloadWithoutFbo(const VideoCore::BufferTextureCopy& download,
                             const VideoCore::StagingData& staging);
 
+    /// Returns the texture tag of the current allocation
+    HostTextureTag MakeTag() const noexcept;
+
 private:
     const Driver* driver;
     TextureRuntime* runtime;

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -313,7 +313,8 @@ RendererOpenGL::RendererOpenGL(Core::System& system, Frontend::EmuWindow& window
     }
     frame_dumper.mailbox = std::make_unique<OGLVideoDumpingMailbox>();
     InitOpenGLObjects();
-    rasterizer = std::make_unique<RasterizerOpenGL>(system.Memory(), *this, driver);
+    rasterizer = std::make_unique<RasterizerOpenGL>(system.Memory(), system.CustomTexManager(),
+                                                    *this, driver);
 }
 
 RendererOpenGL::~RendererOpenGL() = default;
@@ -347,6 +348,8 @@ void RendererOpenGL::SwapBuffers() {
 
     EndFrame();
     prev_state.Apply();
+
+    rasterizer->TickFrame();
 }
 
 void RendererOpenGL::RenderScreenshot() {
@@ -1169,9 +1172,6 @@ void RendererOpenGL::TryPresent(int timeout_ms, bool is_secondary) {
     glBindFramebuffer(GL_READ_FRAMEBUFFER, 0);
 }
 
-/// Updates the framerate
-void RendererOpenGL::UpdateFramerate() {}
-
 void RendererOpenGL::PrepareVideoDumping() {
     auto* mailbox = static_cast<OGLVideoDumpingMailbox*>(frame_dumper.mailbox.get());
     {

src/video_core/renderer_opengl/renderer_opengl.h

@@ -94,7 +94,6 @@ private:
                                        float h);
     void DrawSingleScreenStereo(const ScreenInfo& screen_info_l, const ScreenInfo& screen_info_r,
                                 float x, float y, float w, float h);
-    void UpdateFramerate();
 
     // Loads framebuffer from emulated memory into the display information structure
     void LoadFBToScreenInfo(const GPU::Regs::FramebufferConfig& framebuffer,