Merge remote-tracking branch 'upstream/main' into gc2

.gitmodules

@@ -117,3 +117,6 @@
 	path = externals/sdl3_mixer
 	url = https://github.com/libsdl-org/SDL_mixer
 	shallow = true
+[submodule "externals/miniz"]
+	path = externals/miniz
+	url = https://github.com/richgel999/miniz

CMakeLists.txt

@@ -228,8 +228,10 @@ find_package(half 1.12.0 MODULE)
 find_package(magic_enum 0.9.7 CONFIG)
 find_package(PNG 1.6 MODULE)
 find_package(RenderDoc 1.6.0 MODULE)
-find_package(SDL3 3.1.2 CONFIG)
 find_package(SDL3_mixer 2.8.1 CONFIG)
+if (SDL3_mixer_FOUND)
+    find_package(SDL3 3.1.2 CONFIG)
+endif()
 find_package(stb MODULE)
 find_package(toml11 4.2.0 CONFIG)
 find_package(tsl-robin-map 1.3.0 CONFIG)
@@ -510,7 +512,7 @@ set(PAD_LIB src/core/libraries/pad/pad.cpp
             src/core/libraries/pad/pad_errors.h
 )
 
-set(SYSTEM_GESTURE_LIB 
+set(SYSTEM_GESTURE_LIB
             src/core/libraries/system_gesture/system_gesture.cpp
             src/core/libraries/system_gesture/system_gesture.h
 )
@@ -554,6 +556,8 @@ set(FIBER_LIB src/core/libraries/fiber/fiber_context.s
               src/core/libraries/fiber/fiber_error.h
 )
 
+set_source_files_properties(src/core/libraries/fiber/fiber_context.s PROPERTIES COMPILE_OPTIONS -Wno-unused-command-line-argument)
+
 set(VDEC_LIB src/core/libraries/videodec/videodec2_impl.cpp
              src/core/libraries/videodec/videodec2_impl.h
              src/core/libraries/videodec/videodec2.cpp
@@ -569,6 +573,8 @@ set(VDEC_LIB src/core/libraries/videodec/videodec2_impl.cpp
 set(NP_LIBS src/core/libraries/np/np_error.h
             src/core/libraries/np/np_common.cpp
             src/core/libraries/np/np_common.h
+            src/core/libraries/np/np_commerce.cpp
+            src/core/libraries/np/np_commerce.h
             src/core/libraries/np/np_manager.cpp
             src/core/libraries/np/np_manager.h
             src/core/libraries/np/np_score.cpp
@@ -689,7 +695,6 @@ set(COMMON src/common/logging/backend.cpp
            src/common/lru_cache.h
            src/common/error.cpp
            src/common/error.h
-           src/common/scope_exit.h
            src/common/fixed_value.h
            src/common/func_traits.h
            src/common/native_clock.cpp
@@ -703,6 +708,8 @@ set(COMMON src/common/logging/backend.cpp
            src/common/rdtsc.h
            src/common/recursive_lock.cpp
            src/common/recursive_lock.h
+           src/common/scope_exit.h
+           src/common/serdes.h
            src/common/sha1.h
            src/common/shared_first_mutex.h
            src/common/signal_context.h
@@ -982,6 +989,8 @@ set(VIDEO_CORE src/video_core/amdgpu/cb_db_extent.h
                src/video_core/renderer_vulkan/vk_pipeline_cache.h
                src/video_core/renderer_vulkan/vk_pipeline_common.cpp
                src/video_core/renderer_vulkan/vk_pipeline_common.h
+               src/video_core/renderer_vulkan/vk_pipeline_serialization.cpp
+               src/video_core/renderer_vulkan/vk_pipeline_serialization.h
                src/video_core/renderer_vulkan/vk_platform.cpp
                src/video_core/renderer_vulkan/vk_platform.h
                src/video_core/renderer_vulkan/vk_presenter.cpp
@@ -1019,6 +1028,8 @@ set(VIDEO_CORE src/video_core/amdgpu/cb_db_extent.h
                src/video_core/texture_cache/tile_manager.cpp
                src/video_core/texture_cache/tile_manager.h
                src/video_core/texture_cache/types.h
+               src/video_core/cache_storage.cpp
+               src/video_core/cache_storage.h
                src/video_core/page_manager.cpp
                src/video_core/page_manager.h
                src/video_core/multi_level_page_table.h
@@ -1073,7 +1084,8 @@ add_executable(shadps4
 create_target_directory_groups(shadps4)
 
 target_link_libraries(shadps4 PRIVATE magic_enum::magic_enum fmt::fmt toml11::toml11 tsl::robin_map xbyak::xbyak Tracy::TracyClient RenderDoc::API FFmpeg::ffmpeg Dear_ImGui gcn half::half ZLIB::ZLIB PNG::PNG)
-target_link_libraries(shadps4 PRIVATE Boost::headers GPUOpen::VulkanMemoryAllocator LibAtrac9 sirit Vulkan::Headers xxHash::xxhash Zydis::Zydis glslang::glslang SDL3::SDL3 SDL3_mixer::SDL3_mixer pugixml::pugixml stb::headers libusb::usb lfreist-hwinfo::hwinfo nlohmann_json::nlohmann_json)
+target_link_libraries(shadps4 PRIVATE Boost::headers GPUOpen::VulkanMemoryAllocator LibAtrac9 sirit Vulkan::Headers xxHash::xxhash Zydis::Zydis glslang::glslang SDL3::SDL3 SDL3_mixer::SDL3_mixer pugixml::pugixml)
+target_link_libraries(shadps4 PRIVATE stb::headers libusb::usb lfreist-hwinfo::hwinfo nlohmann_json::nlohmann_json miniz)
 
 target_compile_definitions(shadps4 PRIVATE IMGUI_USER_CONFIG="imgui/imgui_config.h")
 target_compile_definitions(Dear_ImGui PRIVATE IMGUI_USER_CONFIG="${PROJECT_SOURCE_DIR}/src/imgui/imgui_config.h")
@@ -1122,22 +1134,22 @@ if (APPLE)
 endif()
 
 if (WIN32)
-    target_link_libraries(shadps4 PRIVATE mincore wepoll)
+    target_link_libraries(shadps4 PRIVATE mincore wepoll wbemuuid)
 
     if (MSVC)
         # MSVC likes putting opinions on what people can use, disable:
-        add_definitions(-D_CRT_SECURE_NO_WARNINGS -D_CRT_NONSTDC_NO_DEPRECATE -D_SCL_SECURE_NO_WARNINGS)
+        add_compile_definitions(_CRT_SECURE_NO_WARNINGS _CRT_NONSTDC_NO_DEPRECATE _SCL_SECURE_NO_WARNINGS)
     endif()
 
-    add_definitions(-DNOMINMAX -DWIN32_LEAN_AND_MEAN)
+    add_compile_definitions(NOMINMAX WIN32_LEAN_AND_MEAN)
 
     if (MSVC)
         # Needed for conflicts with time.h of windows.h
-        add_definitions(-D_TIMESPEC_DEFINED)
+        add_compile_definitions(_TIMESPEC_DEFINED)
     endif()
 
     # Target Windows 10 RS5
-    add_definitions(-DNTDDI_VERSION=0x0A000006 -D_WIN32_WINNT=0x0A00 -DWINVER=0x0A00)
+    add_compile_definitions(NTDDI_VERSION=0x0A000006 _WIN32_WINNT=0x0A00 WINVER=0x0A00)
 
     if (MSVC)
         target_link_libraries(shadps4 PRIVATE clang_rt.builtins-x86_64.lib)
@@ -1169,7 +1181,7 @@ if (WIN32)
     target_sources(shadps4 PRIVATE src/shadps4.rc)
 endif()
 
-add_definitions(-DBOOST_ASIO_STANDALONE)
+add_compile_definitions(BOOST_ASIO_STANDALONE)
 
 target_include_directories(shadps4 PRIVATE ${CMAKE_CURRENT_SOURCE_DIR})
 

dist/net.shadps4.shadPS4.metainfo.xml

@@ -18,19 +18,19 @@
   <screenshots>
     <screenshot type="default">
       <image type="source" translate="no" >https://cdn.jsdelivr.net/gh/shadps4-emu/shadps4@main/documents/Screenshots/1.png</image>
-      <caption>Bloodborne</caption>
+      <caption>Bloodborne by From Software</caption>
     </screenshot>
     <screenshot>
       <image type="source" translate="no">https://cdn.jsdelivr.net/gh/shadps4-emu/shadps4@main/documents/Screenshots/2.png</image>
-      <caption>Hatsune Miku: Project DIVA Future Tone</caption>
+      <caption>Hatsune Miku Project DIVA Future Tone by SEGA</caption>
     </screenshot>
     <screenshot>
       <image type="source" translate="no">https://cdn.jsdelivr.net/gh/shadps4-emu/shadps4@main/documents/Screenshots/3.png</image>
-      <caption>Yakuza 0</caption>
+      <caption>Yakuza 0 by SEGA</caption>
     </screenshot>
     <screenshot>
       <image type="source" translate="no">https://cdn.jsdelivr.net/gh/shadps4-emu/shadps4@main/documents/Screenshots/4.png</image>
-      <caption>Persona 4 Golden</caption>
+      <caption>DRIVECLUB™ by Evolution Studios</caption>
     </screenshot>
   </screenshots>
   <categories>

externals/CMakeLists.txt

@@ -152,7 +152,7 @@ endif()
 # sirit
 add_subdirectory(sirit)
 if (WIN32)
-    target_compile_options(sirit PUBLIC "-Wno-error=unused-command-line-argument")
+    target_compile_options(sirit PRIVATE "-Wno-error=unused-command-line-argument")
 endif()
 
 # half
@@ -261,3 +261,6 @@ endif()
 #nlohmann json 
 set(JSON_BuildTests OFF CACHE INTERNAL "")
 add_subdirectory(json)
+
+# miniz
+add_subdirectory(miniz)

externals/miniz

@@ -0,0 +1 @@
+Subproject commit 174573d60290f447c13a2b1b3405de2b96e27d6c

src/common/config.cpp

@@ -177,7 +177,7 @@ static ConfigEntry<bool> isFullscreen(false);
 static ConfigEntry<string> fullscreenMode("Windowed");
 static ConfigEntry<string> presentMode("Mailbox");
 static ConfigEntry<bool> isHDRAllowed(false);
-static ConfigEntry<bool> fsrEnabled(true);
+static ConfigEntry<bool> fsrEnabled(false);
 static ConfigEntry<bool> rcasEnabled(true);
 static ConfigEntry<int> rcasAttenuation(250);
 
@@ -191,6 +191,8 @@ static ConfigEntry<bool> vkCrashDiagnostic(false);
 static ConfigEntry<bool> vkHostMarkers(false);
 static ConfigEntry<bool> vkGuestMarkers(false);
 static ConfigEntry<bool> rdocEnable(false);
+static ConfigEntry<bool> pipelineCacheEnable(false);
+static ConfigEntry<bool> pipelineCacheArchive(false);
 
 // Debug
 static ConfigEntry<bool> isDebugDump(false);
@@ -452,6 +454,14 @@ bool isRdocEnabled() {
     return rdocEnable.get();
 }
 
+bool isPipelineCacheEnabled() {
+    return pipelineCacheEnable.get();
+}
+
+bool isPipelineCacheArchived() {
+    return pipelineCacheArchive.get();
+}
+
 bool fpsColor() {
     return isFpsColor.get();
 }
@@ -603,6 +613,14 @@ void setRdocEnabled(bool enable, bool is_game_specific) {
     rdocEnable.set(enable, is_game_specific);
 }
 
+void setPipelineCacheEnabled(bool enable, bool is_game_specific) {
+    pipelineCacheEnable.set(enable, is_game_specific);
+}
+
+void setPipelineCacheArchived(bool enable, bool is_game_specific) {
+    pipelineCacheArchive.set(enable, is_game_specific);
+}
+
 void setVblankFreq(u32 value, bool is_game_specific) {
     vblankFrequency.set(value, is_game_specific);
 }
@@ -939,6 +957,8 @@ void load(const std::filesystem::path& path, bool is_game_specific) {
         vkHostMarkers.setFromToml(vk, "hostMarkers", is_game_specific);
         vkGuestMarkers.setFromToml(vk, "guestMarkers", is_game_specific);
         rdocEnable.setFromToml(vk, "rdocEnable", is_game_specific);
+        pipelineCacheEnable.setFromToml(vk, "pipelineCacheEnable", is_game_specific);
+        pipelineCacheArchive.setFromToml(vk, "pipelineCacheArchive", is_game_specific);
     }
 
     string current_version = {};
@@ -1107,6 +1127,8 @@ void save(const std::filesystem::path& path, bool is_game_specific) {
     vkHostMarkers.setTomlValue(data, "Vulkan", "hostMarkers", is_game_specific);
     vkGuestMarkers.setTomlValue(data, "Vulkan", "guestMarkers", is_game_specific);
     rdocEnable.setTomlValue(data, "Vulkan", "rdocEnable", is_game_specific);
+    pipelineCacheEnable.setTomlValue(data, "Vulkan", "pipelineCacheEnable", is_game_specific);
+    pipelineCacheArchive.setTomlValue(data, "Vulkan", "pipelineCacheArchive", is_game_specific);
 
     isDebugDump.setTomlValue(data, "Debug", "DebugDump", is_game_specific);
     isShaderDebug.setTomlValue(data, "Debug", "CollectShader", is_game_specific);
@@ -1237,6 +1259,8 @@ void setDefaultValues(bool is_game_specific) {
     vkHostMarkers.set(false, is_game_specific);
     vkGuestMarkers.set(false, is_game_specific);
     rdocEnable.set(false, is_game_specific);
+    pipelineCacheEnable.set(false, is_game_specific);
+    pipelineCacheArchive.set(false, is_game_specific);
 
     // GS - Debug
     isDebugDump.set(false, is_game_specific);
@@ -1286,6 +1310,7 @@ hotkey_pause = f9
 hotkey_reload_inputs = f8
 hotkey_toggle_mouse_to_joystick = f7
 hotkey_toggle_mouse_to_gyro = f6
+hotkey_toggle_mouse_to_touchpad = delete
 hotkey_quit = lctrl, lshift, end
 )";
 }

src/common/config.h

@@ -94,7 +94,11 @@ void setVkGuestMarkersEnabled(bool enable, bool is_game_specific = false);
 bool getEnableDiscordRPC();
 void setEnableDiscordRPC(bool enable);
 bool isRdocEnabled();
+bool isPipelineCacheEnabled();
+bool isPipelineCacheArchived();
 void setRdocEnabled(bool enable, bool is_game_specific = false);
+void setPipelineCacheEnabled(bool enable, bool is_game_specific = false);
+void setPipelineCacheArchived(bool enable, bool is_game_specific = false);
 std::string getLogType();
 void setLogType(const std::string& type, bool is_game_specific = false);
 std::string getLogFilter();

src/common/logging/filter.cpp

@@ -104,6 +104,7 @@ bool ParseFilterRule(Filter& instance, Iterator begin, Iterator end) {
     SUB(Lib, Move)                                                                                 \
     SUB(Lib, NpAuth)                                                                               \
     SUB(Lib, NpCommon)                                                                             \
+    SUB(Lib, NpCommerce)                                                                           \
     SUB(Lib, NpManager)                                                                            \
     SUB(Lib, NpScore)                                                                              \
     SUB(Lib, NpTrophy)                                                                             \

src/common/logging/types.h

@@ -70,6 +70,7 @@ enum class Class : u8 {
     Lib_Http2,               ///< The LibSceHttp2 implementation.
     Lib_SysModule,           ///< The LibSceSysModule implementation
     Lib_NpCommon,            ///< The LibSceNpCommon implementation
+    Lib_NpCommerce,          ///< The LibSceNpCommerce implementation
     Lib_NpAuth,              ///< The LibSceNpAuth implementation
     Lib_NpManager,           ///< The LibSceNpManager implementation
     Lib_NpScore,             ///< The LibSceNpScore implementation

src/common/path_util.cpp

@@ -127,6 +127,7 @@ static auto UserPaths = [] {
     create_path(PathType::MetaDataDir, user_dir / METADATA_DIR);
     create_path(PathType::CustomTrophy, user_dir / CUSTOM_TROPHY);
     create_path(PathType::CustomConfigs, user_dir / CUSTOM_CONFIGS);
+    create_path(PathType::CacheDir, user_dir / CACHE_DIR);
 
     std::ofstream notice_file(user_dir / CUSTOM_TROPHY / "Notice.txt");
     if (notice_file.is_open()) {

src/common/path_util.h

@@ -24,6 +24,7 @@ enum class PathType {
     MetaDataDir,    // Where game metadata (e.g. trophies and menu backgrounds) is stored.
     CustomTrophy,   // Where custom files for trophies are stored.
     CustomConfigs,  // Where custom files for different games are stored.
+    CacheDir,       // Where pipeline and shader cache is stored.
 };
 
 constexpr auto PORTABLE_DIR = "user";
@@ -42,6 +43,7 @@ constexpr auto PATCHES_DIR = "patches";
 constexpr auto METADATA_DIR = "game_data";
 constexpr auto CUSTOM_TROPHY = "custom_trophy";
 constexpr auto CUSTOM_CONFIGS = "custom_configs";
+constexpr auto CACHE_DIR = "cache";
 
 // Filenames
 constexpr auto LOG_FILE = "shad_log.txt";

src/common/serdes.h

@@ -0,0 +1,140 @@
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "common/assert.h"
+#include "common/types.h"
+
+#include <cstddef>
+
+namespace Serialization {
+
+template <typename T>
+concept Container = requires(T t) {
+    typename T::iterator;
+    { t.begin() } -> std::same_as<typename T::iterator>;
+    { t.end() } -> std::same_as<typename T::iterator>;
+    { t.size() } -> std::convertible_to<std::size_t>;
+};
+
+struct Archive {
+    void Alloc(size_t size) {
+        container.resize(size);
+    }
+
+    void Grow(size_t size) {
+        container.resize(container.size() + size);
+    }
+
+    void Merge(const Archive& ar) {
+        container.insert(container.end(), ar.container.cbegin(), ar.container.cend());
+        offset = container.size();
+    }
+
+    [[nodiscard]] size_t SizeBytes() const {
+        return container.size();
+    }
+
+    u8* CurrPtr() {
+        return container.data() + offset;
+    }
+
+    void Advance(size_t size) {
+        ASSERT(offset + size <= container.size());
+        offset += size;
+    }
+
+    std::vector<u8>&& TakeOff() {
+        offset = 0;
+        return std::move(container);
+    }
+
+    [[nodiscard]] bool IsEoS() const {
+        return offset >= container.size();
+    }
+
+    Archive() = default;
+    explicit Archive(std::vector<u8>&& v) : container{v} {}
+
+private:
+    u32 offset{};
+    std::vector<u8> container{};
+
+    friend struct Writer;
+    friend struct Reader;
+};
+
+struct Writer {
+    template <typename T>
+    void Write(const T* ptr, size_t size) {
+        if (ar.offset + size >= ar.container.size()) {
+            ar.Grow(size);
+        }
+        std::memcpy(ar.CurrPtr(), reinterpret_cast<const void*>(ptr), size);
+        ar.Advance(size);
+    }
+
+    template <typename T>
+        requires(!Container<T>)
+    void Write(const T& value) {
+        const auto size = sizeof(value);
+        Write(&value, size);
+    }
+
+    void Write(const auto& v) {
+        Write(v.size());
+        for (const auto& elem : v) {
+            Write(elem);
+        }
+    }
+
+    void Write(const std::string& s) {
+        Write(s.size());
+        Write(s.c_str(), s.size());
+    }
+
+    Writer() = delete;
+    explicit Writer(Archive& ar_) : ar{ar_} {}
+
+    Archive& ar;
+};
+
+struct Reader {
+    template <typename T>
+    void Read(T* ptr, size_t size) {
+        ASSERT(ar.offset + size <= ar.container.size());
+        std::memcpy(reinterpret_cast<void*>(ptr), ar.CurrPtr(), size);
+        ar.Advance(size);
+    }
+
+    template <typename T>
+        requires(!Container<T>)
+    void Read(T& value) {
+        const auto size = sizeof(value);
+        Read(&value, size);
+    }
+
+    void Read(auto& v) {
+        size_t num_elements{};
+        Read(num_elements);
+        for (int i = 0; i < num_elements; ++i) {
+            v.emplace_back();
+            Read(v.back());
+        }
+    }
+
+    void Read(std::string& s) {
+        size_t length{};
+        Read(length);
+        s.resize(length);
+        Read(s.data(), length);
+    }
+
+    Reader() = delete;
+    explicit Reader(Archive& ar_) : ar{ar_} {}
+
+    Archive& ar;
+};
+
+} // namespace Serialization

src/core/file_format/psf.cpp

@@ -36,6 +36,7 @@ bool PSF::Open(const std::filesystem::path& filepath) {
     }
 
     const u64 psfSize = file.GetSize();
+    ASSERT_MSG(psfSize != 0, "SFO file at {} is empty!", filepath.string());
     std::vector<u8> psf(psfSize);
     file.Seek(0);
     file.Read(psf);

src/core/file_sys/directories/base_directory.h

@@ -36,6 +36,18 @@ public:
         return ORBIS_KERNEL_ERROR_EBADF;
     }
 
+    virtual s64 write(const void* buf, u64 nbytes) {
+        return ORBIS_KERNEL_ERROR_EBADF;
+    }
+
+    virtual s64 writev(const Libraries::Kernel::OrbisKernelIovec* iov, s32 iovcnt) {
+        return ORBIS_KERNEL_ERROR_EBADF;
+    }
+
+    virtual s64 pwritev(const Libraries::Kernel::OrbisKernelIovec* iov, s32 iovcnt, s64 offset) {
+        return ORBIS_KERNEL_ERROR_EBADF;
+    }
+
     virtual s64 lseek(s64 offset, s32 whence) {
         return ORBIS_KERNEL_ERROR_EBADF;
     }

src/core/libraries/fiber/fiber.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include "fiber.h"

src/core/libraries/kernel/file_system.cpp

@@ -311,6 +311,9 @@ s64 PS4_SYSV_ABI write(s32 fd, const void* buf, u64 nbytes) {
     } else if (file->type == Core::FileSys::FileType::Socket) {
         // Socket functions handle errnos internally.
         return file->socket->SendPacket(buf, nbytes, 0, nullptr, 0);
+    } else if (file->type == Core::FileSys::FileType::Directory) {
+        *__Error() = POSIX_EBADF;
+        return -1;
     }
 
     return file->f.WriteRaw<u8>(buf, nbytes);
@@ -405,7 +408,11 @@ s64 PS4_SYSV_ABI writev(s32 fd, const OrbisKernelIovec* iov, s32 iovcnt) {
             return -1;
         }
         return result;
+    } else if (file->type == Core::FileSys::FileType::Directory) {
+        *__Error() = POSIX_EBADF;
+        return -1;
     }
+
     s64 total_written = 0;
     for (s32 i = 0; i < iovcnt; i++) {
         total_written += file->f.WriteRaw<u8>(iov[i].iov_base, iov[i].iov_len);
@@ -1047,7 +1054,11 @@ s64 PS4_SYSV_ABI posix_pwritev(s32 fd, const OrbisKernelIovec* iov, s32 iovcnt,
             return -1;
         }
         return result;
+    } else if (file->type == Core::FileSys::FileType::Directory) {
+        *__Error() = POSIX_EBADF;
+        return -1;
     }
+
     const s64 pos = file->f.Tell();
     SCOPE_EXIT {
         file->f.Seek(pos);

src/core/libraries/kernel/process.cpp

@@ -42,6 +42,16 @@ s32 PS4_SYSV_ABI sceKernelGetCompiledSdkVersion(s32* ver) {
 }
 
 s32 PS4_SYSV_ABI sceKernelGetCpumode() {
+    LOG_DEBUG(Lib_Kernel, "called");
+    auto& attrs = Common::ElfInfo::Instance().GetPSFAttributes();
+    u32 is_cpu6 = attrs.six_cpu_mode.Value();
+    u32 is_cpu7 = attrs.seven_cpu_mode.Value();
+    if (is_cpu6 == 1 && is_cpu7 == 1) {
+        return 2;
+    }
+    if (is_cpu7 == 1) {
+        return 5;
+    }
     return 0;
 }
 

src/core/libraries/kernel/threads/pthread.cpp

@@ -663,6 +663,10 @@ void RegisterThread(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("Z4QosVuAsA0", "libkernel", 1, "libkernel", posix_pthread_once);
     LIB_FUNCTION("EotR8a3ASf4", "libkernel", 1, "libkernel", posix_pthread_self);
     LIB_FUNCTION("OxhIB8LB-PQ", "libkernel", 1, "libkernel", posix_pthread_create);
+    LIB_FUNCTION("lZzFeSxPl08", "libkernel", 1, "libkernel", posix_pthread_setcancelstate);
+    LIB_FUNCTION("CBNtXOoef-E", "libkernel", 1, "libkernel", posix_sched_get_priority_max);
+    LIB_FUNCTION("m0iS6jNsXds", "libkernel", 1, "libkernel", posix_sched_get_priority_min);
+    LIB_FUNCTION("Xs9hdiD7sAA", "libkernel", 1, "libkernel", posix_pthread_setschedparam);
     LIB_FUNCTION("+U1R4WtXvoc", "libkernel", 1, "libkernel", posix_pthread_detach);
     LIB_FUNCTION("7Xl257M4VNI", "libkernel", 1, "libkernel", posix_pthread_equal);
     LIB_FUNCTION("h9CcP3J0oVM", "libkernel", 1, "libkernel", posix_pthread_join);

src/core/libraries/kernel/time.cpp

@@ -1,6 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <ctime>
 #include <thread>
 
 #include "common/assert.h"
@@ -485,25 +486,41 @@ Common::NativeClock* GetClock() {
 s32 PS4_SYSV_ABI sceKernelConvertUtcToLocaltime(time_t time, time_t* local_time,
                                                 struct OrbisTimesec* st, u64* dst_sec) {
     LOG_TRACE(Kernel, "Called");
+#ifdef _WIN32
+    TIME_ZONE_INFORMATION tz{};
+    DWORD res = GetTimeZoneInformation(&tz);
+    *local_time = time - tz.Bias;
+
+    if (st != nullptr) {
+        st->t = time;
+        st->west_sec = -tz.Bias * 60;
+        st->dst_sec = res == TIME_ZONE_ID_DAYLIGHT ? -_dstbias : 0;
+    }
+
+    if (dst_sec != nullptr) {
+        *dst_sec = res == TIME_ZONE_ID_DAYLIGHT ? -_dstbias : 0;
+    }
+#else
 #ifdef __APPLE__
     // std::chrono::current_zone() not available yet.
     const auto* time_zone = date::current_zone();
 #else
     const auto* time_zone = std::chrono::current_zone();
-#endif
+#endif // __APPLE__
     auto info = time_zone->get_info(std::chrono::system_clock::now());
 
     *local_time = info.offset.count() + info.save.count() * 60 + time;
 
     if (st != nullptr) {
         st->t = time;
-        st->west_sec = info.offset.count() * 60;
+        st->west_sec = info.offset.count();
         st->dst_sec = info.save.count() * 60;
     }
 
     if (dst_sec != nullptr) {
         *dst_sec = info.save.count() * 60;
     }
+#endif // _WIN32
 
     return ORBIS_OK;
 }
@@ -565,4 +582,4 @@ void RegisterTime(Core::Loader::SymbolsResolver* sym) {
     LIB_FUNCTION("-o5uEDpN+oY", "libkernel", 1, "libkernel", sceKernelConvertUtcToLocaltime);
 }
 
-} // namespace Libraries::Kernel
+} // namespace Libraries::Kernel

src/core/libraries/libs.cpp

@@ -32,6 +32,7 @@
 #include "core/libraries/network/ssl.h"
 #include "core/libraries/network/ssl2.h"
 #include "core/libraries/np/np_auth.h"
+#include "core/libraries/np/np_commerce.h"
 #include "core/libraries/np/np_common.h"
 #include "core/libraries/np/np_manager.h"
 #include "core/libraries/np/np_party.h"
@@ -93,6 +94,7 @@ void InitHLELibs(Core::Loader::SymbolsResolver* sym) {
     Libraries::SysModule::RegisterLib(sym);
     Libraries::Posix::RegisterLib(sym);
     Libraries::AudioIn::RegisterLib(sym);
+    Libraries::Np::NpCommerce::RegisterLib(sym);
     Libraries::Np::NpCommon::RegisterLib(sym);
     Libraries::Np::NpManager::RegisterLib(sym);
     Libraries::Np::NpScore::RegisterLib(sym);

src/core/libraries/network/http.cpp

@@ -712,8 +712,61 @@ int PS4_SYSV_ABI sceHttpUriCopy() {
     return ORBIS_OK;
 }
 
-int PS4_SYSV_ABI sceHttpUriEscape() {
+int PS4_SYSV_ABI sceHttpUriEscape(char* out, u64* require, u64 prepare, const char* in) {
-    LOG_ERROR(Lib_Http, "(STUBBED) called");
+    LOG_TRACE(Lib_Http, "called");
+
+    if (!in) {
+        LOG_ERROR(Lib_Http, "Invalid input string");
+        return ORBIS_HTTP_ERROR_INVALID_VALUE;
+    }
+
+    auto IsUnreserved = [](unsigned char c) -> bool {
+        return (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') || (c >= '0' && c <= '9') ||
+               c == '-' || c == '_' || c == '.' || c == '~';
+    };
+
+    u64 needed = 0;
+    const char* src = in;
+    while (*src) {
+        unsigned char c = static_cast<unsigned char>(*src);
+        if (IsUnreserved(c)) {
+            needed++;
+        } else {
+            needed += 3; // %XX format
+        }
+        src++;
+    }
+    needed++; // null terminator
+
+    if (require) {
+        *require = needed;
+    }
+
+    if (!out) {
+        return ORBIS_OK;
+    }
+
+    if (prepare < needed) {
+        LOG_ERROR(Lib_Http, "Buffer too small: need {} but only {} available", needed, prepare);
+        return ORBIS_HTTP_ERROR_OUT_OF_MEMORY;
+    }
+
+    static const char hex_chars[] = "0123456789ABCDEF";
+    src = in;
+    char* dst = out;
+    while (*src) {
+        unsigned char c = static_cast<unsigned char>(*src);
+        if (IsUnreserved(c)) {
+            *dst++ = *src;
+        } else {
+            *dst++ = '%';
+            *dst++ = hex_chars[(c >> 4) & 0x0F];
+            *dst++ = hex_chars[c & 0x0F];
+        }
+        src++;
+    }
+    *dst = '\0';
+
     return ORBIS_OK;
 }
 
@@ -1072,12 +1125,163 @@ int PS4_SYSV_ABI sceHttpUriParse(OrbisHttpUriElement* out, const char* srcUri, v
 }
 
 int PS4_SYSV_ABI sceHttpUriSweepPath(char* dst, const char* src, u64 srcSize) {
-    LOG_ERROR(Lib_Http, "(STUBBED) called");
+    LOG_TRACE(Lib_Http, "called");
+
+    if (!dst || !src) {
+        LOG_ERROR(Lib_Http, "Invalid parameters");
+        return ORBIS_HTTP_ERROR_INVALID_VALUE;
+    }
+
+    if (srcSize == 0) {
+        dst[0] = '\0';
+        return ORBIS_OK;
+    }
+
+    u64 len = 0;
+    while (len < srcSize && src[len] != '\0') {
+        len++;
+    }
+
+    for (u64 i = 0; i < len; i++) {
+        dst[i] = src[i];
+    }
+    dst[len] = '\0';
+
+    char* read = dst;
+    char* write = dst;
+
+    while (*read) {
+        if (read[0] == '.' && read[1] == '.' && read[2] == '/') {
+            read += 3;
+            continue;
+        }
+
+        if (read[0] == '.' && read[1] == '/') {
+            read += 2;
+            continue;
+        }
+
+        if (read[0] == '/' && read[1] == '.' && read[2] == '/') {
+            read += 2;
+            continue;
+        }
+
+        if (read[0] == '/' && read[1] == '.' && read[2] == '\0') {
+            if (write == dst) {
+                *write++ = '/';
+            }
+            break;
+        }
+
+        bool is_dotdot_mid = (read[0] == '/' && read[1] == '.' && read[2] == '.' && read[3] == '/');
+        bool is_dotdot_end =
+            (read[0] == '/' && read[1] == '.' && read[2] == '.' && read[3] == '\0');
+
+        if (is_dotdot_mid || is_dotdot_end) {
+            if (write > dst) {
+                if (*(write - 1) == '/') {
+                    write--;
+                }
+                while (write > dst && *(write - 1) != '/') {
+                    write--;
+                }
+
+                if (is_dotdot_mid && write > dst) {
+                    write--;
+                }
+            }
+
+            if (is_dotdot_mid) {
+                read += 3;
+            } else {
+                break;
+            }
+            continue;
+        }
+
+        if ((read[0] == '.' && read[1] == '\0') ||
+            (read[0] == '.' && read[1] == '.' && read[2] == '\0')) {
+            break;
+        }
+
+        if (read[0] == '/') {
+            *write++ = *read++;
+        }
+        while (*read && *read != '/') {
+            *write++ = *read++;
+        }
+    }
+
+    *write = '\0';
     return ORBIS_OK;
 }
 
 int PS4_SYSV_ABI sceHttpUriUnescape(char* out, u64* require, u64 prepare, const char* in) {
-    LOG_ERROR(Lib_Http, "(STUBBED) called");
+    LOG_TRACE(Lib_Http, "called");
+
+    if (!in) {
+        LOG_ERROR(Lib_Http, "Invalid input string");
+        return ORBIS_HTTP_ERROR_INVALID_VALUE;
+    }
+
+    // Locale-independent hex digit check
+    auto IsHex = [](char c) -> bool {
+        return (c >= '0' && c <= '9') || (c >= 'A' && c <= 'F') || (c >= 'a' && c <= 'f');
+    };
+
+    // Convert hex char to int value
+    auto HexToInt = [](char c) -> int {
+        if (c >= '0' && c <= '9')
+            return c - '0';
+        if (c >= 'A' && c <= 'F')
+            return c - 'A' + 10;
+        if (c >= 'a' && c <= 'f')
+            return c - 'a' + 10;
+        return 0;
+    };
+
+    // Check for valid percent-encoded sequence (%XX)
+    auto IsValidPercentSequence = [&](const char* s) -> bool {
+        return s[0] == '%' && s[1] != '\0' && s[2] != '\0' && IsHex(s[1]) && IsHex(s[2]);
+    };
+
+    u64 needed = 0;
+    const char* src = in;
+    while (*src) {
+        if (IsValidPercentSequence(src)) {
+            src += 3;
+        } else {
+            src++;
+        }
+        needed++;
+    }
+    needed++; // null terminator
+
+    if (require) {
+        *require = needed;
+    }
+
+    if (!out) {
+        return ORBIS_OK;
+    }
+
+    if (prepare < needed) {
+        LOG_ERROR(Lib_Http, "Buffer too small: need {} but only {} available", needed, prepare);
+        return ORBIS_HTTP_ERROR_OUT_OF_MEMORY;
+    }
+
+    src = in;
+    char* dst = out;
+    while (*src) {
+        if (IsValidPercentSequence(src)) {
+            *dst++ = static_cast<char>((HexToInt(src[1]) << 4) | HexToInt(src[2]));
+            src += 3;
+        } else {
+            *dst++ = *src++;
+        }
+    }
+    *dst = '\0';
+
     return ORBIS_OK;
 }
 

src/core/libraries/network/http.h

@@ -148,7 +148,7 @@ int PS4_SYSV_ABI sceHttpUnsetEpoll();
 int PS4_SYSV_ABI sceHttpUriBuild(char* out, u64* require, u64 prepare,
                                  const OrbisHttpUriElement* srcElement, u32 option);
 int PS4_SYSV_ABI sceHttpUriCopy();
-int PS4_SYSV_ABI sceHttpUriEscape();
+int PS4_SYSV_ABI sceHttpUriEscape(char* out, u64* require, u64 prepare, const char* in);
 int PS4_SYSV_ABI sceHttpUriMerge(char* mergedUrl, char* url, char* relativeUri, u64* require,
                                  u64 prepare, u32 option);
 int PS4_SYSV_ABI sceHttpUriParse(OrbisHttpUriElement* out, const char* srcUri, void* pool,

src/core/libraries/np/np_commerce.cpp

@@ -0,0 +1,129 @@
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "common/logging/log.h"
+#include "core/libraries/error_codes.h"
+#include "core/libraries/libs.h"
+#include "core/libraries/system/commondialog.h"
+
+namespace Libraries::Np::NpCommerce {
+
+using CommonDialog::Error;
+using CommonDialog::Result;
+using CommonDialog::Status;
+
+static Status g_dialog_status = Status::NONE;
+static Result g_dialog_result = Result::OK;
+
+s32 PS4_SYSV_ABI sceNpCommerceDialogClose() {
+    LOG_INFO(Lib_NpCommerce, "called");
+    if (g_dialog_status == Status::NONE) {
+        return static_cast<s32>(Error::NOT_INITIALIZED);
+    }
+    if (g_dialog_status != Status::FINISHED) {
+        return static_cast<s32>(Error::NOT_FINISHED);
+    }
+    g_dialog_status = Status::INITIALIZED;
+    return ORBIS_OK;
+}
+
+s32 PS4_SYSV_ABI sceNpCommerceDialogGetResult(s32* result) {
+    LOG_INFO(Lib_NpCommerce, "called");
+    if (result == nullptr) {
+        return static_cast<s32>(Error::ARG_NULL);
+    }
+    if (g_dialog_status != Status::FINISHED) {
+        return static_cast<s32>(Error::NOT_FINISHED);
+    }
+    *result = static_cast<s32>(g_dialog_result);
+    return ORBIS_OK;
+}
+
+s8 PS4_SYSV_ABI sceNpCommerceDialogGetStatus() {
+    LOG_DEBUG(Lib_NpCommerce, "called, status = {}", static_cast<u32>(g_dialog_status));
+    return static_cast<s8>(g_dialog_status);
+}
+
+s32 PS4_SYSV_ABI sceNpCommerceDialogInitialize() {
+    LOG_INFO(Lib_NpCommerce, "called");
+    if (g_dialog_status != Status::NONE) {
+        return static_cast<s32>(Error::ALREADY_INITIALIZED);
+    }
+    g_dialog_status = Status::INITIALIZED;
+    return ORBIS_OK;
+}
+
+s32 PS4_SYSV_ABI sceNpCommerceDialogInitializeInternal() {
+    LOG_INFO(Lib_NpCommerce, "called");
+    return sceNpCommerceDialogInitialize();
+}
+
+s16 PS4_SYSV_ABI sceNpCommerceDialogOpen(s64 check) {
+    LOG_INFO(Lib_NpCommerce, "called, check = {}", check);
+    if (g_dialog_status != Status::INITIALIZED) {
+        LOG_WARNING(Lib_NpCommerce, "Dialog not initialized");
+        return ORBIS_OK;
+    }
+
+    g_dialog_status = Status::FINISHED;
+    g_dialog_result = Result::USER_CANCELED;
+    return ORBIS_OK;
+}
+
+s32 PS4_SYSV_ABI sceNpCommerceDialogTerminate() {
+    LOG_INFO(Lib_NpCommerce, "called");
+    if (g_dialog_status == Status::NONE) {
+        return static_cast<s32>(Error::NOT_INITIALIZED);
+    }
+    if (g_dialog_status == Status::RUNNING) {
+        return static_cast<s32>(Error::NOT_FINISHED);
+    }
+    g_dialog_status = Status::NONE;
+    return ORBIS_OK;
+}
+
+s32 PS4_SYSV_ABI sceNpCommerceDialogUpdateStatus() {
+    LOG_DEBUG(Lib_NpCommerce, "called, status = {}", static_cast<u32>(g_dialog_status));
+    return static_cast<s32>(g_dialog_status);
+}
+
+s32 PS4_SYSV_ABI sceNpCommerceHidePsStoreIcon() {
+    LOG_ERROR(Lib_NpCommerce, "(STUBBED) called");
+    return ORBIS_OK;
+}
+
+s32 PS4_SYSV_ABI sceNpCommerceSetPsStoreIconLayout(s32 layout) {
+    LOG_ERROR(Lib_NpCommerce, "(STUBBED) called");
+    return ORBIS_OK;
+}
+
+s32 PS4_SYSV_ABI sceNpCommerceShowPsStoreIcon(s16 icon) {
+    LOG_ERROR(Lib_NpCommerce, "(STUBBED) called");
+    return ORBIS_OK;
+}
+
+void RegisterLib(Core::Loader::SymbolsResolver* sym) {
+    LIB_FUNCTION("NU3ckGHMFXo", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceDialogClose);
+    LIB_FUNCTION("r42bWcQbtZY", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceDialogGetResult);
+    LIB_FUNCTION("CCbC+lqqvF0", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceDialogGetStatus);
+    LIB_FUNCTION("0aR2aWmQal4", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceDialogInitialize);
+    LIB_FUNCTION("9ZiLXAGG5rg", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceDialogInitializeInternal);
+    LIB_FUNCTION("DfSCDRA3EjY", "libSceNpCommerce", 1, "libSceNpCommerce", sceNpCommerceDialogOpen);
+    LIB_FUNCTION("m-I92Ab50W8", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceDialogTerminate);
+    LIB_FUNCTION("LR5cwFMMCVE", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceDialogUpdateStatus);
+    LIB_FUNCTION("dsqCVsNM0Zg", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceHidePsStoreIcon);
+    LIB_FUNCTION("uKTDW8hk-ts", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceSetPsStoreIconLayout);
+    LIB_FUNCTION("DHmwsa6S8Tc", "libSceNpCommerce", 1, "libSceNpCommerce",
+                 sceNpCommerceShowPsStoreIcon);
+};
+
+} // namespace Libraries::Np::NpCommerce

src/core/libraries/np/np_commerce.h

@@ -0,0 +1,16 @@
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "common/types.h"
+
+namespace Core::Loader {
+class SymbolsResolver;
+}
+
+namespace Libraries::Np::NpCommerce {
+
+void RegisterLib(Core::Loader::SymbolsResolver* sym);
+
+} // namespace Libraries::Np::NpCommerce

src/core/libraries/system/systemservice.cpp

@@ -1,6 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <cstdlib>
 #include "common/config.h"
 #include "common/logging/log.h"
 #include "common/singleton.h"
@@ -1874,6 +1875,10 @@ int PS4_SYSV_ABI sceSystemServiceLoadExec(const char* path, const char* argv[])
     auto emu = Common::Singleton<Core::Emulator>::Instance();
     auto mnt = Common::Singleton<Core::FileSys::MntPoints>::Instance();
     auto hostPath = mnt->GetHostPath(std::string_view(path));
+    if (hostPath.empty()) {
+        LOG_INFO(Lib_SystemService, "Restart called with invalid file '{}', exiting.", path);
+        std::quick_exit(0);
+    }
     std::vector<std::string> args;
     if (argv != nullptr) {
         for (const char** ptr = argv; *ptr != nullptr; ptr++) {

src/core/memory.cpp

@@ -587,6 +587,10 @@ s32 MemoryManager::MapFile(void** out_addr, VAddr virtual_addr, u64 size, Memory
         // On real hardware, GPU file mmaps cause a full system crash due to an internal error.
         ASSERT_MSG(false, "Files cannot be mapped to GPU memory");
     }
+    if (True(prot & MemoryProt::CpuExec)) {
+        // On real hardware, execute permissions are silently removed.
+        prot &= ~MemoryProt::CpuExec;
+    }
 
     // Add virtual memory area
     auto& new_vma = CarveVMA(mapped_addr, size)->second;
@@ -793,10 +797,9 @@ s32 MemoryManager::QueryProtection(VAddr addr, void** start, void** end, u32* pr
 s64 MemoryManager::ProtectBytes(VAddr addr, VirtualMemoryArea& vma_base, u64 size,
                                 MemoryProt prot) {
     const auto start_in_vma = addr - vma_base.base;
-    const auto adjusted_size =
+    const auto adjusted_size = std::min<u64>(vma_base.size - start_in_vma, size);
-        vma_base.size - start_in_vma < size ? vma_base.size - start_in_vma : size;
 
-    if (vma_base.type == VMAType::Free) {
+    if (vma_base.type == VMAType::Free || vma_base.type == VMAType::PoolReserved) {
         // On PS4, protecting freed memory does nothing.
         return adjusted_size;
     }
@@ -828,8 +831,9 @@ s64 MemoryManager::ProtectBytes(VAddr addr, VirtualMemoryArea& vma_base, u64 siz
         perms |= Core::MemoryPermission::ReadWrite;
     }
 
-    if (vma_base.type == VMAType::Direct || vma_base.type == VMAType::Pooled) {
+    if (vma_base.type == VMAType::Direct || vma_base.type == VMAType::Pooled ||
-        // On PS4, execute permissions are hidden from direct memory mappings.
+        vma_base.type == VMAType::File) {
+        // On PS4, execute permissions are hidden from direct memory and file mappings.
         // Tests show that execute permissions still apply, so handle this after reading perms.
         prot &= ~MemoryProt::CpuExec;
     }
@@ -837,6 +841,12 @@ s64 MemoryManager::ProtectBytes(VAddr addr, VirtualMemoryArea& vma_base, u64 siz
     // Change protection
     vma_base.prot = prot;
 
+    if (vma_base.type == VMAType::Reserved) {
+        // On PS4, protections change vma_map, but don't apply.
+        // Return early to avoid protecting memory that isn't mapped in address space.
+        return adjusted_size;
+    }
+
     impl.Protect(addr, size, perms);
 
     return adjusted_size;
@@ -853,22 +863,20 @@ s32 MemoryManager::Protect(VAddr addr, u64 size, MemoryProt prot) {
     // Ensure the range to modify is valid
     ASSERT_MSG(IsValidMapping(addr, size), "Attempted to access invalid address {:#x}", addr);
 
-    // Validate protection flags
+    // Appropriately restrict flags.
-    constexpr static MemoryProt valid_flags =
+    constexpr static MemoryProt flag_mask =
-        MemoryProt::NoAccess | MemoryProt::CpuRead | MemoryProt::CpuWrite | MemoryProt::CpuExec |
+        MemoryProt::CpuReadWrite | MemoryProt::CpuExec | MemoryProt::GpuReadWrite;
-        MemoryProt::GpuRead | MemoryProt::GpuWrite | MemoryProt::GpuReadWrite;
+    MemoryProt valid_flags = prot & flag_mask;
-
-    MemoryProt invalid_flags = prot & ~valid_flags;
-    if (invalid_flags != MemoryProt::NoAccess) {
-        LOG_ERROR(Kernel_Vmm, "Invalid protection flags");
-        return ORBIS_KERNEL_ERROR_EINVAL;
-    }
 
     // Protect all VMAs between addr and addr + size.
     s64 protected_bytes = 0;
     while (protected_bytes < size) {
         auto it = FindVMA(addr + protected_bytes);
         auto& vma_base = it->second;
+        if (vma_base.base > addr + protected_bytes) {
+            // Account for potential gaps in memory map.
+            protected_bytes += vma_base.base - (addr + protected_bytes);
+        }
         auto result = ProtectBytes(addr + protected_bytes, vma_base, size - protected_bytes, prot);
         if (result < 0) {
             // ProtectBytes returned an error, return it
@@ -904,13 +912,21 @@ s32 MemoryManager::VirtualQuery(VAddr addr, s32 flags,
     const auto& vma = it->second;
     info->start = vma.base;
     info->end = vma.base + vma.size;
-    info->offset = vma.type == VMAType::Flexible ? 0 : vma.phys_base;
+    info->offset = 0;
     info->protection = static_cast<s32>(vma.prot);
     info->is_flexible = vma.type == VMAType::Flexible ? 1 : 0;
     info->is_direct = vma.type == VMAType::Direct ? 1 : 0;
     info->is_stack = vma.type == VMAType::Stack ? 1 : 0;
     info->is_pooled = vma.type == VMAType::PoolReserved || vma.type == VMAType::Pooled ? 1 : 0;
     info->is_committed = vma.IsMapped() ? 1 : 0;
+    if (vma.type == VMAType::Direct || vma.type == VMAType::Pooled) {
+        // Offset is only assigned for direct and pooled mappings.
+        info->offset = vma.phys_base;
+    }
+    if (vma.type == VMAType::Reserved || vma.type == VMAType::PoolReserved) {
+        // Protection is hidden from reserved mappings.
+        info->protection = 0;
+    }
 
     strncpy(info->name, vma.name.data(), ::Libraries::Kernel::ORBIS_KERNEL_MAXIMUM_NAME_LENGTH);
 

src/core/tls.cpp

@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <mutex>
@@ -165,10 +165,7 @@ void SetTcbBase(void* image_address) {
 }
 
 Tcb* GetTcbBase() {
-    void* tcb = nullptr;
+    return Libraries::Kernel::g_curthread->tcb;
-    const int ret = syscall(SYS_arch_prctl, ARCH_GET_GS, &tcb);
-    ASSERT_MSG(ret == 0, "Failed to get GS base: errno {}", errno);
-    return static_cast<Tcb*>(tcb);
 }
 
 #else

src/core/tls.h

@@ -1,10 +1,13 @@
-// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
 
 #include <cstring>
 #include "common/types.h"
+#ifdef _WIN32
+#include <malloc.h>
+#endif
 
 namespace Xbyak {
 class CodeGenerator;

src/emulator.cpp

@@ -42,6 +42,7 @@
 #include "core/linker.h"
 #include "core/memory.h"
 #include "emulator.h"
+#include "video_core/cache_storage.h"
 #include "video_core/renderdoc.h"
 
 #ifdef _WIN32
@@ -58,10 +59,11 @@ Frontend::WindowSDL* g_window = nullptr;
 namespace Core {
 
 Emulator::Emulator() {
-    // Initialize NT API functions and set high priority
+    // Initialize NT API functions, set high priority and disable WER
 #ifdef _WIN32
     Common::NtApi::Initialize();
     SetPriorityClass(GetCurrentProcess(), ABOVE_NORMAL_PRIORITY_CLASS);
+    SetErrorMode(SetErrorMode(0) | SEM_NOGPFAULTERRORBOX);
     // need to init this in order for winsock2 to work
     WORD versionWanted = MAKEWORD(2, 2);
     WSADATA wsaData;
@@ -386,6 +388,7 @@ void Emulator::Run(std::filesystem::path file, std::vector<std::string> args,
     }
 
     UpdatePlayTime(id);
+    Storage::DataBase::Instance().Close();
 
     std::quick_exit(0);
 }

src/input/input_handler.cpp

@@ -106,6 +106,7 @@ auto output_array = std::array{
     ControllerOutput(HOTKEY_RELOAD_INPUTS),
     ControllerOutput(HOTKEY_TOGGLE_MOUSE_TO_JOYSTICK),
     ControllerOutput(HOTKEY_TOGGLE_MOUSE_TO_GYRO),
+    ControllerOutput(HOTKEY_TOGGLE_MOUSE_TO_TOUCHPAD),
     ControllerOutput(HOTKEY_RENDERDOC),
 
     ControllerOutput(SDL_GAMEPAD_BUTTON_INVALID, SDL_GAMEPAD_AXIS_INVALID),
@@ -579,6 +580,9 @@ void ControllerOutput::FinalizeUpdate() {
         case HOTKEY_TOGGLE_MOUSE_TO_GYRO:
             PushSDLEvent(SDL_EVENT_MOUSE_TO_GYRO);
             break;
+        case HOTKEY_TOGGLE_MOUSE_TO_TOUCHPAD:
+            PushSDLEvent(SDL_EVENT_MOUSE_TO_TOUCHPAD);
+            break;
         case HOTKEY_RENDERDOC:
             PushSDLEvent(SDL_EVENT_RDOC_CAPTURE);
             break;
@@ -773,6 +777,9 @@ void ActivateOutputsFromInputs() {
         it.ResetUpdate();
     }
 
+    // Check for input blockers
+    ApplyMouseInputBlockers();
+
     // Iterate over all inputs, and update their respecive outputs accordingly
     for (auto& it : connections) {
         it.output->AddUpdate(it.ProcessBinding());

src/input/input_handler.h

@@ -34,9 +34,10 @@
 #define SDL_EVENT_RELOAD_INPUTS SDL_EVENT_USER + 5
 #define SDL_EVENT_MOUSE_TO_JOYSTICK SDL_EVENT_USER + 6
 #define SDL_EVENT_MOUSE_TO_GYRO SDL_EVENT_USER + 7
-#define SDL_EVENT_RDOC_CAPTURE SDL_EVENT_USER + 8
+#define SDL_EVENT_MOUSE_TO_TOUCHPAD SDL_EVENT_USER + 8
-#define SDL_EVENT_QUIT_DIALOG SDL_EVENT_USER + 9
+#define SDL_EVENT_RDOC_CAPTURE SDL_EVENT_USER + 9
-#define SDL_EVENT_MOUSE_WHEEL_OFF SDL_EVENT_USER + 10
+#define SDL_EVENT_QUIT_DIALOG SDL_EVENT_USER + 10
+#define SDL_EVENT_MOUSE_WHEEL_OFF SDL_EVENT_USER + 11
 
 #define LEFTJOYSTICK_HALFMODE 0x00010000
 #define RIGHTJOYSTICK_HALFMODE 0x00020000
@@ -52,7 +53,8 @@
 #define HOTKEY_RELOAD_INPUTS 0xf0000005
 #define HOTKEY_TOGGLE_MOUSE_TO_JOYSTICK 0xf0000006
 #define HOTKEY_TOGGLE_MOUSE_TO_GYRO 0xf0000007
-#define HOTKEY_RENDERDOC 0xf0000008
+#define HOTKEY_TOGGLE_MOUSE_TO_TOUCHPAD 0xf0000008
+#define HOTKEY_RENDERDOC 0xf0000009
 
 #define SDL_UNMAPPED UINT32_MAX - 1
 
@@ -141,6 +143,7 @@ const std::map<std::string, u32> string_to_cbutton_map = {
     {"hotkey_reload_inputs", HOTKEY_RELOAD_INPUTS},
     {"hotkey_toggle_mouse_to_joystick", HOTKEY_TOGGLE_MOUSE_TO_JOYSTICK},
     {"hotkey_toggle_mouse_to_gyro", HOTKEY_TOGGLE_MOUSE_TO_GYRO},
+    {"hotkey_toggle_mouse_to_touchpad", HOTKEY_TOGGLE_MOUSE_TO_TOUCHPAD},
     {"hotkey_renderdoc_capture", HOTKEY_RENDERDOC},
 };
 

src/input/input_mouse.cpp

@@ -6,12 +6,19 @@
 #include "common/assert.h"
 #include "common/types.h"
 #include "input/controller.h"
+#include "input/input_handler.h"
 #include "input_mouse.h"
 
+#include <common/singleton.h>
+#include <emulator.h>
 #include "SDL3/SDL.h"
 
+extern Frontend::WindowSDL* g_window;
+
 namespace Input {
 
+extern std::list<std::pair<InputEvent, bool>> pressed_keys;
+
 int mouse_joystick_binding = 0;
 float mouse_deadzone_offset = 0.5, mouse_speed = 1, mouse_speed_offset = 0.1250;
 bool mouse_gyro_roll_mode = false;
@@ -80,7 +87,6 @@ void EmulateJoystick(GameController* controller, u32 interval) {
 
 constexpr float constant_down_accel[3] = {0.0f, 10.0f, 0.0f};
 void EmulateGyro(GameController* controller, u32 interval) {
-    // LOG_INFO(Input, "todo gyro");
     float d_x = 0, d_y = 0;
     SDL_GetRelativeMouseState(&d_x, &d_y);
     controller->Acceleration(1, constant_down_accel);
@@ -92,6 +98,31 @@ void EmulateGyro(GameController* controller, u32 interval) {
     controller->Gyro(1, gyro_from_mouse);
 }
 
+void EmulateTouchpad(GameController* controller, u32 interval) {
+    float x, y;
+    SDL_MouseButtonFlags mouse_buttons = SDL_GetMouseState(&x, &y);
+    controller->SetTouchpadState(0, (mouse_buttons & SDL_BUTTON_LMASK) != 0,
+                                 std::clamp(x / g_window->GetWidth(), 0.0f, 1.0f),
+                                 std::clamp(y / g_window->GetHeight(), 0.0f, 1.0f));
+    controller->CheckButton(0, Libraries::Pad::OrbisPadButtonDataOffset::TouchPad,
+                            (mouse_buttons & SDL_BUTTON_RMASK) != 0);
+}
+
+void ApplyMouseInputBlockers() {
+    switch (mouse_mode) {
+    case MouseMode::Touchpad:
+        for (auto& k : pressed_keys) {
+            if (k.first.input.sdl_id == SDL_BUTTON_LEFT ||
+                k.first.input.sdl_id == SDL_BUTTON_RIGHT) {
+                k.second = true;
+            }
+        }
+        break;
+    default:
+        break;
+    }
+}
+
 Uint32 MousePolling(void* param, Uint32 id, Uint32 interval) {
     auto* controller = (GameController*)param;
     switch (mouse_mode) {
@@ -101,6 +132,9 @@ Uint32 MousePolling(void* param, Uint32 id, Uint32 interval) {
     case MouseMode::Gyro:
         EmulateGyro(controller, interval);
         break;
+    case MouseMode::Touchpad:
+        EmulateTouchpad(controller, interval);
+        break;
 
     default:
         break;

src/input/input_mouse.h

@@ -12,6 +12,7 @@ enum MouseMode {
     Off = 0,
     Joystick,
     Gyro,
+    Touchpad,
 };
 
 bool ToggleMouseModeTo(MouseMode m);
@@ -22,6 +23,8 @@ void SetMouseGyroRollMode(bool mode);
 void EmulateJoystick(GameController* controller, u32 interval);
 void EmulateGyro(GameController* controller, u32 interval);
 
+void ApplyMouseInputBlockers();
+
 // Polls the mouse for changes
 Uint32 MousePolling(void* param, Uint32 id, Uint32 interval);
 

src/main.cpp

@@ -1,6 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include <SDL3/SDL_messagebox.h>
 #include "functional"
 #include "iostream"
 #include "string"
@@ -182,6 +183,10 @@ int main(int argc, char* argv[]) {
          }}};
 
     if (argc == 1) {
+        if (!SDL_ShowSimpleMessageBox(
+                SDL_MESSAGEBOX_INFORMATION, "shadPS4",
+                "This is a CLI application. Please use the QTLauncher for a GUI.", nullptr))
+            std::cerr << "Could not display SDL message box! Error: " << SDL_GetError() << "\n";
         int dummy = 0; // one does not simply pass 0 directly
         arg_map.at("-h")(dummy);
         return -1;

src/sdl_window.cpp

@@ -457,6 +457,11 @@ void WindowSDL::WaitEvent() {
         SDL_SetWindowRelativeMouseMode(this->GetSDLWindow(),
                                        Input::ToggleMouseModeTo(Input::MouseMode::Gyro));
         break;
+    case SDL_EVENT_MOUSE_TO_TOUCHPAD:
+        SDL_SetWindowRelativeMouseMode(this->GetSDLWindow(),
+                                       Input::ToggleMouseModeTo(Input::MouseMode::Touchpad));
+        SDL_SetWindowRelativeMouseMode(this->GetSDLWindow(), false);
+        break;
     case SDL_EVENT_RDOC_CAPTURE:
         VideoCore::TriggerCapture();
         break;

src/shader_recompiler/frontend/fetch_shader.cpp

@@ -51,7 +51,7 @@ std::optional<FetchShaderData> ParseFetchShader(const Shader::Info& info) {
     }
 
     const auto* code = GetFetchShaderCode(info, info.fetch_shader_sgpr_base);
-    FetchShaderData data{.code = code};
+    FetchShaderData data{};
     GcnCodeSlice code_slice(code, code + std::numeric_limits<u32>::max());
     GcnDecodeContext decoder;
 

src/shader_recompiler/frontend/fetch_shader.h

@@ -8,6 +8,10 @@
 #include "common/types.h"
 #include "shader_recompiler/info.h"
 
+namespace Serialization {
+struct Archive;
+}
+
 namespace Shader::Gcn {
 
 struct VertexAttribute {
@@ -50,7 +54,6 @@ struct VertexAttribute {
 };
 
 struct FetchShaderData {
-    const u32* code;
     u32 size = 0;
     std::vector<VertexAttribute> attributes;
     s8 vertex_offset_sgpr = -1;   ///< SGPR of vertex offset from VADDR
@@ -60,6 +63,9 @@ struct FetchShaderData {
         return attributes == other.attributes && vertex_offset_sgpr == other.vertex_offset_sgpr &&
                instance_offset_sgpr == other.instance_offset_sgpr;
     }
+
+    void Serialize(Serialization::Archive& ar) const;
+    bool Deserialize(Serialization::Archive& buffer);
 };
 
 const u32* GetFetchShaderCode(const Info& info, u32 sgpr_base);

src/shader_recompiler/frontend/format.cpp

@@ -1837,11 +1837,17 @@ constexpr std::array<InstFormat, 71> InstructionFormatVOP1 = {{
     // 22 = V_CVT_F64_U32
     {InstClass::VectorConv, InstCategory::VectorALU, 1, 1, ScalarType::Uint32, ScalarType::Float64},
     // 23 = V_TRUNC_F64
-    {InstClass::VectorConv, InstCategory::VectorALU, 1, 1, ScalarType::Float64,
+    {InstClass::VectorFpRound64, InstCategory::VectorALU, 1, 1, ScalarType::Float64,
+     ScalarType::Float64},
+    // 24 = V_CEIL_F64
+    {InstClass::VectorFpRound64, InstCategory::VectorALU, 1, 1, ScalarType::Float64,
+     ScalarType::Float64},
+    // 25 = V_RNDNE_F64
+    {InstClass::VectorFpRound64, InstCategory::VectorALU, 1, 1, ScalarType::Float64,
+     ScalarType::Float64},
+    // 26 = V_FLOOR_F64
+    {InstClass::VectorFpRound64, InstCategory::VectorALU, 1, 1, ScalarType::Float64,
      ScalarType::Float64},
-    {},
-    {},
-    {},
     {},
     {},
     {},

src/shader_recompiler/frontend/structured_control_flow.cpp

@@ -596,9 +596,8 @@ public:
                   IR::AbstractSyntaxList& syntax_list_, std::span<const GcnInst> inst_list_,
                   Info& info_, const RuntimeInfo& runtime_info_, const Profile& profile_)
         : stmt_pool{stmt_pool_}, inst_pool{inst_pool_}, block_pool{block_pool_},
-          syntax_list{syntax_list_}, inst_list{inst_list_}, info{info_},
+          syntax_list{syntax_list_}, inst_list{inst_list_}, runtime_info{runtime_info_},
-          runtime_info{runtime_info_}, profile{profile_},
+          profile{profile_}, translator{info_, runtime_info_, profile_} {
-          translator{info_, runtime_info_, profile_} {
         Visit(root_stmt, nullptr, nullptr);
 
         IR::Block* first_block = syntax_list.front().data.block;
@@ -782,7 +781,7 @@ private:
         }
     }
 
-    IR::Block* MergeBlock(Statement& parent, Statement& stmt) {
+    IR::Block* MergeBlock(Statement& parent, Statement& stmt) const {
         Statement* merge_stmt{TryFindForwardBlock(stmt)};
         if (!merge_stmt) {
             // Create a merge block we can visit later
@@ -798,7 +797,6 @@ private:
     IR::AbstractSyntaxList& syntax_list;
     const Block dummy_flow_block{.is_dummy = true};
     std::span<const GcnInst> inst_list;
-    Info& info;
     const RuntimeInfo& runtime_info;
     const Profile& profile;
     Translator translator;

src/shader_recompiler/frontend/translate/translate.cpp

@@ -560,7 +560,8 @@ void Translator::EmitFetch(const GcnInst& inst) {
         }
         const auto filename = fmt::format("vs_{:#018x}.fetch.bin", info.pgm_hash);
         const auto file = IOFile{dump_dir / filename, FileAccessMode::Create};
-        file.WriteRaw<u8>(fetch_data->code, fetch_data->size);
+        const auto* code = GetFetchShaderCode(info, code_sgpr_base);
+        file.WriteRaw<u8>(code, fetch_data->size);
     }
 
     for (const auto& attrib : fetch_data->attributes) {

src/shader_recompiler/frontend/translate/translate.h

@@ -201,6 +201,7 @@ public:
     void V_CVT_F32_F64(const GcnInst& inst);
     void V_CVT_F64_F32(const GcnInst& inst);
     void V_CVT_F32_UBYTE(u32 index, const GcnInst& inst);
+    void V_FLOOR_F64(const GcnInst& inst);
     void V_FRACT_F32(const GcnInst& inst);
     void V_TRUNC_F32(const GcnInst& inst);
     void V_CEIL_F32(const GcnInst& inst);

src/shader_recompiler/frontend/translate/vector_alu.cpp

@@ -142,6 +142,8 @@ void Translator::EmitVectorAlu(const GcnInst& inst) {
         return V_CVT_F32_UBYTE(2, inst);
     case Opcode::V_CVT_F32_UBYTE3:
         return V_CVT_F32_UBYTE(3, inst);
+    case Opcode::V_FLOOR_F64:
+        return V_FLOOR_F64(inst);
     case Opcode::V_FRACT_F32:
         return V_FRACT_F32(inst);
     case Opcode::V_TRUNC_F32:
@@ -806,6 +808,11 @@ void Translator::V_CVT_F32_UBYTE(u32 index, const GcnInst& inst) {
     SetDst(inst.dst[0], ir.ConvertUToF(32, 32, byte));
 }
 
+void Translator::V_FLOOR_F64(const GcnInst& inst) {
+    const IR::F64 src0{GetSrc64<IR::F64>(inst.src[0])};
+    SetDst64(inst.dst[0], ir.FPFloor(src0));
+}
+
 void Translator::V_FRACT_F32(const GcnInst& inst) {
     const IR::F32 src0{GetSrc<IR::F32>(inst.src[0])};
     SetDst(inst.dst[0], ir.FPFract(src0));
@@ -1043,20 +1050,25 @@ void Translator::V_CMP_U32(ConditionOp op, bool is_signed, bool set_exec, const
 }
 
 void Translator::V_CMP_U64(ConditionOp op, bool is_signed, bool set_exec, const GcnInst& inst) {
-    const IR::U64 src0{GetSrc64(inst.src[0])};
+    ASSERT(inst.src[1].field == OperandField::ConstZero);
-    const IR::U64 src1{GetSrc64(inst.src[1])};
+    const IR::U1 src0 = [&] {
+        switch (inst.src[0].field) {
+        case OperandField::ScalarGPR:
+            return ir.GetThreadBitScalarReg(IR::ScalarReg(inst.src[0].code));
+        case OperandField::VccLo:
+            return ir.GetVcc();
+        default:
+            UNREACHABLE_MSG("src0 = {}", u32(inst.src[0].field));
+        }
+    }();
     const IR::U1 result = [&] {
         switch (op) {
         case ConditionOp::EQ:
-            return ir.IEqual(src0, src1);
+            return ir.LogicalNot(src0);
         case ConditionOp::LG: // NE
-            return ir.INotEqual(src0, src1);
+            return src0;
         case ConditionOp::GT:
-            if (src1.IsImmediate() && src1.U64() == 0) {
+            return ir.GroupAny(ir.GetThreadBitScalarReg(IR::ScalarReg(inst.src[0].code)));
-                ASSERT(inst.src[0].field == OperandField::ScalarGPR);
-                return ir.GroupAny(ir.GetThreadBitScalarReg(IR::ScalarReg(inst.src[0].code)));
-            }
-            return ir.IGreaterThan(src0, src1, is_signed);
         default:
             UNREACHABLE_MSG("Unsupported V_CMP_U64 condition operation: {}", u32(op));
         }

src/shader_recompiler/info.h

@@ -19,6 +19,10 @@
 #include "shader_recompiler/resource.h"
 #include "shader_recompiler/runtime_info.h"
 
+namespace Serialization {
+struct Archive;
+}
+
 namespace Shader {
 
 enum class Qualifier : u8 {
@@ -34,7 +38,49 @@ enum class Qualifier : u8 {
 /**
  * Contains general information generated by the shader recompiler for an input program.
  */
-struct Info {
+struct InfoPersistent {
+    BufferResourceList buffers;
+    ImageResourceList images;
+    SamplerResourceList samplers;
+    FMaskResourceList fmasks;
+
+    struct UserDataMask {
+        void Set(IR::ScalarReg reg) noexcept {
+            mask |= 1 << static_cast<u32>(reg);
+        }
+
+        u32 Index(IR::ScalarReg reg) const noexcept {
+            const u32 reg_mask = (1 << static_cast<u32>(reg)) - 1;
+            return std::popcount(mask & reg_mask);
+        }
+
+        u32 NumRegs() const noexcept {
+            return std::popcount(mask);
+        }
+
+        u32 mask;
+    };
+    UserDataMask ud_mask{};
+    u32 fetch_shader_sgpr_base{};
+
+    u64 pgm_hash{};
+
+    s32 tess_consts_dword_offset = -1;
+    IR::ScalarReg tess_consts_ptr_base = IR::ScalarReg::Max;
+    Stage stage;
+    LogicalStage l_stage;
+
+    u8 mrt_mask{};
+    bool has_fetch_shader{};
+    bool has_bitwise_xor{};
+    bool uses_dma{};
+
+    InfoPersistent() = default;
+    InfoPersistent(Stage stage_, LogicalStage l_stage_, u64 pgm_hash_)
+        : stage{stage_}, l_stage{l_stage_}, pgm_hash{pgm_hash_} {}
+};
+
+struct Info : InfoPersistent {
     struct AttributeFlags {
         bool Get(IR::Attribute attrib, u32 comp = 0) const {
             return flags[Index(attrib)] & (1 << comp);
@@ -58,56 +104,32 @@ struct Info {
 
         std::array<u8, IR::NumAttributes> flags;
     };
-    AttributeFlags loads{};
-    AttributeFlags stores{};
 
-    struct UserDataMask {
+    enum class ReadConstType {
-        void Set(IR::ScalarReg reg) noexcept {
+        None = 0,
-            mask |= 1 << static_cast<u32>(reg);
+        Immediate = 1 << 0,
-        }
+        Dynamic = 1 << 1,
-
-        u32 Index(IR::ScalarReg reg) const noexcept {
-            const u32 reg_mask = (1 << static_cast<u32>(reg)) - 1;
-            return std::popcount(mask & reg_mask);
-        }
-
-        u32 NumRegs() const noexcept {
-            return std::popcount(mask);
-        }
-
-        u32 mask;
     };
-    UserDataMask ud_mask{};
-
-    CopyShaderData gs_copy_data;
-    u32 uses_patches{};
-
-    BufferResourceList buffers;
-    ImageResourceList images;
-    SamplerResourceList samplers;
-    FMaskResourceList fmasks;
-
-    PersistentSrtInfo srt_info;
-    std::vector<u32> flattened_ud_buf;
 
     struct Interpolation {
         Qualifier primary;
         Qualifier auxiliary;
     };
-    std::array<Interpolation, IR::NumParams> fs_interpolation{};
-
-    IR::ScalarReg tess_consts_ptr_base = IR::ScalarReg::Max;
-    s32 tess_consts_dword_offset = -1;
 
     std::span<const u32> user_data;
-    Stage stage;
+    std::vector<u32> flattened_ud_buf;
-    LogicalStage l_stage;
+    PersistentSrtInfo srt_info;
+
+    AttributeFlags loads{};
+    AttributeFlags stores{};
+
+    ReadConstType readconst_types{};
+    CopyShaderData gs_copy_data;
+    u32 uses_patches{};
 
-    u64 pgm_hash{};
     VAddr pgm_base;
     bool has_storage_images{};
     bool has_discard{};
-    bool has_bitwise_xor{};
     bool has_image_gather{};
     bool has_image_query{};
     bool uses_buffer_atomic_float_min_max{};
@@ -125,20 +147,12 @@ struct Info {
     bool stores_tess_level_outer{};
     bool stores_tess_level_inner{};
     bool translation_failed{};
-    u8 mrt_mask{0u};
-    bool has_fetch_shader{false};
-    u32 fetch_shader_sgpr_base{0u};
 
-    enum class ReadConstType {
+    std::array<Interpolation, IR::NumParams> fs_interpolation{};
-        None = 0,
-        Immediate = 1 << 0,
-        Dynamic = 1 << 1,
-    };
-    ReadConstType readconst_types{};
-    bool uses_dma{};
 
-    explicit Info(Stage stage_, LogicalStage l_stage_, ShaderParams params)
+    Info() = default;
-        : stage{stage_}, l_stage{l_stage_}, pgm_hash{params.hash}, pgm_base{params.Base()},
+    Info(Stage stage_, LogicalStage l_stage_, ShaderParams params)
+        : InfoPersistent(stage_, l_stage_, params.hash), pgm_base{params.Base()},
           user_data{params.user_data} {}
 
     template <typename T>
@@ -192,6 +206,9 @@ struct Info {
                reinterpret_cast<TessellationDataConstantBuffer*>(tess_constants_addr),
                sizeof(tess_constants));
     }
+
+    void Serialize(Serialization::Archive& ar) const;
+    bool Deserialize(Serialization::Archive& ar);
 };
 DECLARE_ENUM_FLAG_OPERATORS(Info::ReadConstType);
 

src/shader_recompiler/ir/passes/flatten_extended_userdata_pass.cpp

@@ -28,6 +28,17 @@ using namespace Xbyak::util;
 static Xbyak::CodeGenerator g_srt_codegen(32_MB);
 static const u8* g_srt_codegen_start = nullptr;
 
+namespace Shader {
+
+PFN_SrtWalker RegisterWalkerCode(const u8* ptr, size_t size) {
+    const auto func_addr = (PFN_SrtWalker)g_srt_codegen.getCurr();
+    g_srt_codegen.db(ptr, size);
+    g_srt_codegen.ready();
+    return func_addr;
+}
+
+} // namespace Shader
+
 namespace {
 
 static void DumpSrtProgram(const Shader::Info& info, const u8* code, size_t codesize) {
@@ -215,9 +226,12 @@ static void GenerateSrtProgram(Info& info, PassInfo& pass_info) {
     c.ret();
     c.ready();
 
+    info.srt_info.walker_func_size =
+        c.getCurr() - reinterpret_cast<const u8*>(info.srt_info.walker_func);
+
     if (Config::dumpShaders()) {
-        size_t codesize = c.getCurr() - reinterpret_cast<const u8*>(info.srt_info.walker_func);
+        DumpSrtProgram(info, reinterpret_cast<const u8*>(info.srt_info.walker_func),
-        DumpSrtProgram(info, reinterpret_cast<const u8*>(info.srt_info.walker_func), codesize);
+                       info.srt_info.walker_func_size);
     }
 
     info.srt_info.flattened_bufsize_dw = pass_info.dst_off_dw;

src/shader_recompiler/ir/passes/hull_shader_transform.cpp

@@ -363,7 +363,7 @@ static IR::F32 ReadTessControlPointAttribute(IR::U32 addr, const u32 stride, IR:
 
 } // namespace
 
-void HullShaderTransform(IR::Program& program, RuntimeInfo& runtime_info) {
+void HullShaderTransform(IR::Program& program, const RuntimeInfo& runtime_info) {
     const Info& info = program.info;
 
     for (IR::Block* block : program.blocks) {
@@ -561,8 +561,8 @@ void HullShaderTransform(IR::Program& program, RuntimeInfo& runtime_info) {
     }
 }
 
-void DomainShaderTransform(IR::Program& program, RuntimeInfo& runtime_info) {
+void DomainShaderTransform(const IR::Program& program, const RuntimeInfo& runtime_info) {
-    Info& info = program.info;
+    const Info& info = program.info;
 
     for (IR::Block* block : program.blocks) {
         for (IR::Inst& inst : block->Instructions()) {

src/shader_recompiler/ir/passes/ir_passes.h

@@ -24,8 +24,8 @@ void LowerBufferFormatToRaw(IR::Program& program);
 void LowerFp64ToFp32(IR::Program& program);
 void RingAccessElimination(const IR::Program& program, const RuntimeInfo& runtime_info);
 void TessellationPreprocess(IR::Program& program, RuntimeInfo& runtime_info);
-void HullShaderTransform(IR::Program& program, RuntimeInfo& runtime_info);
+void HullShaderTransform(IR::Program& program, const RuntimeInfo& runtime_info);
-void DomainShaderTransform(IR::Program& program, RuntimeInfo& runtime_info);
+void DomainShaderTransform(const IR::Program& program, const RuntimeInfo& runtime_info);
 void SharedMemoryBarrierPass(IR::Program& program, const RuntimeInfo& runtime_info,
                              const Profile& profile);
 void SharedMemorySimplifyPass(IR::Program& program, const Profile& profile);

src/shader_recompiler/ir/passes/resource_tracking_pass.cpp

@@ -498,7 +498,8 @@ void PatchBufferSharp(IR::Block& block, IR::Inst& inst, Info& info, Descriptors&
         // buffer_load_format_xyz v[8:10], v1, s[32:35], 0 ...
         // is used to define an inline buffer resource
         std::array<u64, 2> raw;
-        raw[0] = info.pgm_base + (handle->Arg(0).U32() | u64(handle->Arg(1).U32()) << 32);
+        // Keep relative address, we'll do fixup of the address at buffer fetch later
+        raw[0] = (handle->Arg(0).U32() | u64(handle->Arg(1).U32()) << 32);
         raw[1] = handle->Arg(2).U32() | u64(handle->Arg(3).U32()) << 32;
         const auto buffer = std::bit_cast<AmdGpu::Buffer>(raw);
         buffer_binding = descriptors.Add(BufferResource{

src/shader_recompiler/ir/passes/srt.h

@@ -7,9 +7,14 @@
 #include <boost/container/small_vector.hpp>
 #include "common/types.h"
 
+namespace Serialization {
+struct Archive;
+}
+
 namespace Shader {
 
 using PFN_SrtWalker = void PS4_SYSV_ABI (*)(const u32* /*user_data*/, u32* /*flat_dst*/);
+PFN_SrtWalker RegisterWalkerCode(const u8* ptr, size_t size);
 
 struct PersistentSrtInfo {
     // Special case when fetch shader uses step rates.
@@ -20,7 +25,11 @@ struct PersistentSrtInfo {
     };
 
     PFN_SrtWalker walker_func{};
+    size_t walker_func_size{};
     u32 flattened_bufsize_dw = 16; // NumUserDataRegs
+
+    void Serialize(Serialization::Archive& ar) const;
+    bool Deserialize(Serialization::Archive& ar);
 };
 
 } // namespace Shader

src/shader_recompiler/profile.h

@@ -8,6 +8,10 @@
 namespace Shader {
 
 struct Profile {
+    u64 max_ubo_size{};
+    u32 max_viewport_width{};
+    u32 max_viewport_height{};
+    u32 max_shared_memory_size{};
     u32 supported_spirv{0x00010000};
     u32 subgroup_size{};
     bool support_int8{};
@@ -37,10 +41,7 @@ struct Profile {
     bool needs_lds_barriers{};
     bool needs_buffer_offsets{};
     bool needs_unorm_fixup{};
-    u64 max_ubo_size{};
+    bool _pad0{};
-    u32 max_viewport_width{};
-    u32 max_viewport_height{};
-    u32 max_shared_memory_size{};
 };
 
 } // namespace Shader

src/shader_recompiler/recompiler.cpp

@@ -29,7 +29,7 @@ IR::BlockList GenerateBlocks(const IR::AbstractSyntaxList& syntax_list) {
     return blocks;
 }
 
-IR::Program TranslateProgram(std::span<const u32> code, Pools& pools, Info& info,
+IR::Program TranslateProgram(const std::span<const u32>& code, Pools& pools, Info& info,
                              RuntimeInfo& runtime_info, const Profile& profile) {
     // Ensure first instruction is expected.
     constexpr u32 token_mov_vcchi = 0xBEEB03FF;
@@ -55,8 +55,8 @@ IR::Program TranslateProgram(std::span<const u32> code, Pools& pools, Info& info
     Gcn::CFG cfg{gcn_block_pool, program.ins_list};
 
     // Structurize control flow graph and create program.
-    program.syntax_list = Shader::Gcn::BuildASL(pools.inst_pool, pools.block_pool, cfg,
+    program.syntax_list =
-                                                program.info, runtime_info, profile);
+        Shader::Gcn::BuildASL(pools.inst_pool, pools.block_pool, cfg, info, runtime_info, profile);
     program.blocks = GenerateBlocks(program.syntax_list);
     program.post_order_blocks = Shader::IR::PostOrder(program.syntax_list.front());
 

src/shader_recompiler/recompiler.h

@@ -27,7 +27,8 @@ struct Pools {
     }
 };
 
-[[nodiscard]] IR::Program TranslateProgram(std::span<const u32> code, Pools& pools, Info& info,
+[[nodiscard]] IR::Program TranslateProgram(const std::span<const u32>& code, Pools& pools,
-                                           RuntimeInfo& runtime_info, const Profile& profile);
+                                           Info& info, RuntimeInfo& runtime_info,
+                                           const Profile& profile);
 
 } // namespace Shader

src/shader_recompiler/resource.h

@@ -53,8 +53,15 @@ struct BufferResource {
     }
 
     constexpr AmdGpu::Buffer GetSharp(const auto& info) const noexcept {
-        const auto buffer =
+        AmdGpu::Buffer buffer{};
-            inline_cbuf ? inline_cbuf : info.template ReadUdSharp<AmdGpu::Buffer>(sharp_idx);
+        if (inline_cbuf) {
+            buffer = inline_cbuf;
+            if (inline_cbuf.base_address > 1) {
+                buffer.base_address += info.pgm_base; // address fixup
+            }
+        } else {
+            buffer = info.template ReadUdSharp<AmdGpu::Buffer>(sharp_idx);
+        }
         if (!buffer.Valid()) {
             LOG_DEBUG(Render, "Encountered invalid buffer sharp");
             return AmdGpu::Buffer::Null();

src/shader_recompiler/runtime_info.h

@@ -159,7 +159,8 @@ struct GeometryRuntimeInfo {
         return num_outputs == other.num_outputs && outputs == other.outputs && num_invocations &&
                other.num_invocations && output_vertices == other.output_vertices &&
                in_primitive == other.in_primitive &&
-               std::ranges::equal(out_primitive, other.out_primitive);
+               std::ranges::equal(out_primitive, other.out_primitive) &&
+               vs_copy_hash == other.vs_copy_hash;
     }
 };
 

src/shader_recompiler/specialization.h

@@ -79,8 +79,8 @@ struct SamplerSpecialization {
 struct StageSpecialization {
     static constexpr size_t MaxStageResources = 128;
 
-    const Shader::Info* info;
+    const Info* info{};
-    RuntimeInfo runtime_info;
+    RuntimeInfo runtime_info{};
     std::bitset<MaxStageResources> bitset{};
     std::optional<Gcn::FetchShaderData> fetch_shader_data{};
     boost::container::small_vector<VsAttribSpecialization, 32> vs_attribs;
@@ -90,6 +90,7 @@ struct StageSpecialization {
     boost::container::small_vector<SamplerSpecialization, 16> samplers;
     Backend::Bindings start{};
 
+    StageSpecialization() = default;
     StageSpecialization(const Info& info_, RuntimeInfo runtime_info_, const Profile& profile_,
                         Backend::Bindings start_)
         : info{&info_}, runtime_info{runtime_info_}, start{start_} {
@@ -158,7 +159,7 @@ struct StageSpecialization {
         // Initialize runtime_info fields that rely on analysis in tessellation passes
         if (info->l_stage == LogicalStage::TessellationControl ||
             info->l_stage == LogicalStage::TessellationEval) {
-            Shader::TessellationDataConstantBuffer tess_constants;
+            TessellationDataConstantBuffer tess_constants{};
             info->ReadTessConstantBuffer(tess_constants);
             if (info->l_stage == LogicalStage::TessellationControl) {
                 runtime_info.hs_info.InitFromTessConstants(tess_constants);
@@ -192,21 +193,43 @@ struct StageSpecialization {
         }
     }
 
+    [[nodiscard]] bool Valid() const {
+        return info != nullptr;
+    }
+
     bool operator==(const StageSpecialization& other) const {
-        if (start != other.start) {
+        if (!Valid()) {
             return false;
         }
+
+        if (vs_attribs != other.vs_attribs) {
+            return false;
+        }
+
         if (runtime_info != other.runtime_info) {
             return false;
         }
+
         if (fetch_shader_data != other.fetch_shader_data) {
             return false;
         }
-        for (u32 i = 0; i < vs_attribs.size(); i++) {
+
-            if (vs_attribs[i] != other.vs_attribs[i]) {
+        if (fmasks != other.fmasks) {
-                return false;
+            return false;
-            }
         }
+
+        // For VS which only generates geometry and doesn't have any inputs, its start
+        // bindings still may change as they depend on previously processed FS. The check below
+        // handles this case and prevents generation of redundant permutations. This is also safe
+        // for other types of shaders with no bindings.
+        if (bitset.none() && other.bitset.none()) {
+            return true;
+        }
+
+        if (start != other.start) {
+            return false;
+        }
+
         u32 binding{};
         for (u32 i = 0; i < buffers.size(); i++) {
             if (other.bitset[binding++] && buffers[i] != other.buffers[i]) {
@@ -218,11 +241,7 @@ struct StageSpecialization {
                 return false;
             }
         }
-        for (u32 i = 0; i < fmasks.size(); i++) {
+
-            if (other.bitset[binding++] && fmasks[i] != other.fmasks[i]) {
-                return false;
-            }
-        }
         for (u32 i = 0; i < samplers.size(); i++) {
             if (samplers[i] != other.samplers[i]) {
                 return false;
@@ -230,6 +249,9 @@ struct StageSpecialization {
         }
         return true;
     }
+
+    void Serialize(Serialization::Archive& ar) const;
+    bool Deserialize(Serialization::Archive& ar);
 };
 
 } // namespace Shader

src/video_core/amdgpu/liverpool.cpp

@@ -655,8 +655,8 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
                     break;
                 }
                 if (dma_data->src_sel == DmaDataSrc::Data && dma_data->dst_sel == DmaDataDst::Gds) {
-                    rasterizer->InlineData(dma_data->dst_addr_lo, &dma_data->data, sizeof(u32),
+                    rasterizer->FillBuffer(dma_data->dst_addr_lo, dma_data->NumBytes(),
-                                           true);
+                                           dma_data->data, true);
                 } else if ((dma_data->src_sel == DmaDataSrc::Memory ||
                             dma_data->src_sel == DmaDataSrc::MemoryUsingL2) &&
                            dma_data->dst_sel == DmaDataDst::Gds) {
@@ -665,8 +665,8 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
                 } else if (dma_data->src_sel == DmaDataSrc::Data &&
                            (dma_data->dst_sel == DmaDataDst::Memory ||
                             dma_data->dst_sel == DmaDataDst::MemoryUsingL2)) {
-                    rasterizer->InlineData(dma_data->DstAddress<VAddr>(), &dma_data->data,
+                    rasterizer->FillBuffer(dma_data->DstAddress<VAddr>(), dma_data->NumBytes(),
-                                           sizeof(u32), false);
+                                           dma_data->data, false);
                 } else if (dma_data->src_sel == DmaDataSrc::Gds &&
                            (dma_data->dst_sel == DmaDataDst::Memory ||
                             dma_data->dst_sel == DmaDataDst::MemoryUsingL2)) {
@@ -898,7 +898,8 @@ Liverpool::Task Liverpool::ProcessCompute(std::span<const u32> acb, u32 vqid) {
                 break;
             }
             if (dma_data->src_sel == DmaDataSrc::Data && dma_data->dst_sel == DmaDataDst::Gds) {
-                rasterizer->InlineData(dma_data->dst_addr_lo, &dma_data->data, sizeof(u32), true);
+                rasterizer->FillBuffer(dma_data->dst_addr_lo, dma_data->NumBytes(), dma_data->data,
+                                       true);
             } else if ((dma_data->src_sel == DmaDataSrc::Memory ||
                         dma_data->src_sel == DmaDataSrc::MemoryUsingL2) &&
                        dma_data->dst_sel == DmaDataDst::Gds) {
@@ -907,8 +908,8 @@ Liverpool::Task Liverpool::ProcessCompute(std::span<const u32> acb, u32 vqid) {
             } else if (dma_data->src_sel == DmaDataSrc::Data &&
                        (dma_data->dst_sel == DmaDataDst::Memory ||
                         dma_data->dst_sel == DmaDataDst::MemoryUsingL2)) {
-                rasterizer->InlineData(dma_data->DstAddress<VAddr>(), &dma_data->data, sizeof(u32),
+                rasterizer->FillBuffer(dma_data->DstAddress<VAddr>(), dma_data->NumBytes(),
-                                       false);
+                                       dma_data->data, false);
             } else if (dma_data->src_sel == DmaDataSrc::Gds &&
                        (dma_data->dst_sel == DmaDataDst::Memory ||
                         dma_data->dst_sel == DmaDataDst::MemoryUsingL2)) {

src/video_core/amdgpu/pixel_format.h

@@ -79,10 +79,10 @@ enum class NumberFormat : u32 {
     Ubscaled = 13,
 };
 
-enum class NumberClass {
+enum class NumberClass : u8 {
-    Float,
+    Float = 0,
-    Sint,
+    Sint = 1,
-    Uint,
+    Uint = 2,
 };
 
 enum class CompSwizzle : u8 {

src/video_core/buffer_cache/buffer_cache.cpp

@@ -323,14 +323,13 @@ void BufferCache::BindIndexBuffer(u32 index_offset) {
     cmdbuf.bindIndexBuffer(vk_buffer->Handle(), offset, index_type);
 }
 
-void BufferCache::InlineData(VAddr address, const void* value, u32 num_bytes, bool is_gds) {
+void BufferCache::FillBuffer(VAddr address, u32 num_bytes, u32 value, bool is_gds) {
     ASSERT_MSG(address % 4 == 0, "GDS offset must be dword aligned");
     if (!is_gds) {
-        if (!memory->TryWriteBacking(std::bit_cast<void*>(address), value, num_bytes)) {
+        texture_cache.ClearMeta(address);
-            std::memcpy(std::bit_cast<void*>(address), value, num_bytes);
+        if (!IsRegionGpuModified(address, num_bytes)) {
-            return;
+            u32* buffer = std::bit_cast<u32*>(address);
-        }
+            std::fill(buffer, buffer + num_bytes / sizeof(u32), value);
-        if (!IsRegionRegistered(address, num_bytes)) {
             return;
         }
     }
@@ -338,10 +337,10 @@ void BufferCache::InlineData(VAddr address, const void* value, u32 num_bytes, bo
         if (is_gds) {
             return &gds_buffer;
         }
-        const BufferId buffer_id = FindBuffer(address, num_bytes);
+        const auto [buffer, offset] = ObtainBuffer(address, num_bytes, true);
-        return &slot_buffers[buffer_id];
+        return buffer;
     }();
-    InlineDataBuffer(*buffer, address, value, num_bytes);
+    buffer->Fill(buffer->Offset(address), num_bytes, value);
 }
 
 void BufferCache::CopyBuffer(VAddr dst, VAddr src, u32 num_bytes, bool dst_gds, bool src_gds) {
@@ -853,49 +852,6 @@ void BufferCache::SynchronizeBuffersInRange(VAddr device_addr, u64 size) {
     });
 }
 
-void BufferCache::InlineDataBuffer(Buffer& buffer, VAddr address, const void* value,
-                                   u32 num_bytes) {
-    scheduler.EndRendering();
-    const auto cmdbuf = scheduler.CommandBuffer();
-    const vk::BufferMemoryBarrier2 pre_barrier = {
-        .srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
-        .srcAccessMask = vk::AccessFlagBits2::eMemoryRead,
-        .dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
-        .dstAccessMask = vk::AccessFlagBits2::eTransferWrite,
-        .buffer = buffer.Handle(),
-        .offset = buffer.Offset(address),
-        .size = num_bytes,
-    };
-    const vk::BufferMemoryBarrier2 post_barrier = {
-        .srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
-        .srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
-        .dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
-        .dstAccessMask = vk::AccessFlagBits2::eMemoryRead,
-        .buffer = buffer.Handle(),
-        .offset = buffer.Offset(address),
-        .size = num_bytes,
-    };
-    cmdbuf.pipelineBarrier2(vk::DependencyInfo{
-        .dependencyFlags = vk::DependencyFlagBits::eByRegion,
-        .bufferMemoryBarrierCount = 1,
-        .pBufferMemoryBarriers = &pre_barrier,
-    });
-    // vkCmdUpdateBuffer can only copy up to 65536 bytes at a time.
-    static constexpr u32 UpdateBufferMaxSize = 65536;
-    const auto dst_offset = buffer.Offset(address);
-    for (u32 offset = 0; offset < num_bytes; offset += UpdateBufferMaxSize) {
-        const auto* update_src = static_cast<const u8*>(value) + offset;
-        const auto update_dst = dst_offset + offset;
-        const auto update_size = std::min(num_bytes - offset, UpdateBufferMaxSize);
-        cmdbuf.updateBuffer(buffer.Handle(), update_dst, update_size, update_src);
-    }
-    cmdbuf.pipelineBarrier2(vk::DependencyInfo{
-        .dependencyFlags = vk::DependencyFlagBits::eByRegion,
-        .bufferMemoryBarrierCount = 1,
-        .pBufferMemoryBarriers = &post_barrier,
-    });
-}
-
 void BufferCache::WriteDataBuffer(Buffer& buffer, VAddr address, const void* value, u32 num_bytes) {
     vk::BufferCopy copy = {
         .srcOffset = 0,

src/video_core/buffer_cache/buffer_cache.h

@@ -132,7 +132,7 @@ public:
     void BindIndexBuffer(u32 index_offset);
 
     /// Writes a value to GPU buffer. (uses command buffer to temporarily store the data)
-    void InlineData(VAddr address, const void* value, u32 num_bytes, bool is_gds);
+    void FillBuffer(VAddr address, u32 num_bytes, u32 value, bool is_gds);
 
     /// Performs buffer to buffer data copy on the GPU.
     void CopyBuffer(VAddr dst, VAddr src, u32 num_bytes, bool dst_gds, bool src_gds);
@@ -210,8 +210,6 @@ private:
 
     bool SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr, u32 size);
 
-    void InlineDataBuffer(Buffer& buffer, VAddr address, const void* value, u32 num_bytes);
-
     void WriteDataBuffer(Buffer& buffer, VAddr address, const void* value, u32 num_bytes);
 
     void TouchBuffer(const Buffer& buffer);

src/video_core/cache_storage.cpp

@@ -0,0 +1,264 @@
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "common/config.h"
+#include "common/elf_info.h"
+#include "common/io_file.h"
+#include "common/polyfill_thread.h"
+#include "common/thread.h"
+
+#include "video_core/cache_storage.h"
+#include "video_core/renderer_vulkan/vk_instance.h"
+#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+
+#include <miniz.h>
+
+#include <condition_variable>
+#include <functional>
+#include <future>
+#include <mutex>
+#include <queue>
+
+namespace {
+
+std::mutex submit_mutex{};
+u32 num_requests{};
+std::condition_variable_any request_cv{};
+std::queue<std::packaged_task<void()>> req_queue{};
+std::mutex m_request{};
+
+mz_zip_archive zip_ar{};
+bool ar_is_read_only{true};
+
+} // namespace
+
+namespace Storage {
+
+void ProcessIO(const std::stop_token& stoken) {
+    Common::SetCurrentThreadName("shadPS4:PipelineCacheIO");
+
+    while (!stoken.stop_requested()) {
+        {
+            std::unique_lock lk{submit_mutex};
+            Common::CondvarWait(request_cv, lk, stoken, [&] { return num_requests; });
+        }
+
+        if (stoken.stop_requested()) {
+            break;
+        }
+
+        while (num_requests) {
+            std::packaged_task<void()> request{};
+            {
+                std::scoped_lock lock{m_request};
+                if (req_queue.empty()) {
+                    continue;
+                }
+                request = std::move(req_queue.front());
+                req_queue.pop();
+            }
+
+            if (request.valid()) {
+                request();
+                request.get_future().wait();
+            }
+
+            --num_requests;
+        }
+    }
+}
+
+constexpr std::string GetBlobFileExtension(BlobType type) {
+    switch (type) {
+    case BlobType::ShaderMeta: {
+        return "meta";
+    }
+    case BlobType::ShaderBinary: {
+        return "spv";
+    }
+    case BlobType::PipelineKey: {
+        return "key";
+    }
+    case BlobType::ShaderProfile: {
+        return "bin";
+    }
+    default:
+        UNREACHABLE();
+    }
+}
+
+void DataBase::Open() {
+    if (opened) {
+        return;
+    }
+
+    const auto& game_info = Common::ElfInfo::Instance();
+
+    using namespace Common::FS;
+    if (Config::isPipelineCacheArchived()) {
+        mz_zip_zero_struct(&zip_ar);
+
+        cache_path = GetUserPath(PathType::CacheDir) /
+                     std::filesystem::path{game_info.GameSerial()}.replace_extension(".zip");
+
+        if (!mz_zip_reader_init_file(&zip_ar, cache_path.string().c_str(),
+                                     MZ_ZIP_FLAG_READ_ALLOW_WRITING) ||
+            !mz_zip_validate_archive(&zip_ar, 0)) {
+            LOG_INFO(Render, "Cache archive {} is not found or archive is corrupted",
+                     cache_path.string().c_str());
+            mz_zip_reader_end(&zip_ar);
+            mz_zip_writer_init_file(&zip_ar, cache_path.string().c_str(), 0);
+        }
+    } else {
+        cache_path = GetUserPath(PathType::CacheDir) / game_info.GameSerial();
+        if (!std::filesystem::exists(cache_path)) {
+            std::filesystem::create_directories(cache_path);
+        }
+    }
+
+    io_worker = std::jthread{ProcessIO};
+    opened = true;
+}
+
+void DataBase::Close() {
+    if (!IsOpened()) {
+        return;
+    }
+
+    io_worker.request_stop();
+    io_worker.join();
+
+    if (Config::isPipelineCacheArchived()) {
+        mz_zip_writer_finalize_archive(&zip_ar);
+        mz_zip_writer_end(&zip_ar);
+    }
+
+    LOG_INFO(Render, "Cache dumped");
+}
+
+template <typename T>
+bool WriteVector(const BlobType type, std::filesystem::path&& path_, std::vector<T>&& v) {
+    {
+        auto request = std::packaged_task<void()>{[=]() {
+            auto path{path_};
+            path.replace_extension(GetBlobFileExtension(type));
+            if (Config::isPipelineCacheArchived()) {
+                ASSERT_MSG(!ar_is_read_only,
+                           "The archive is read-only. Did you forget to call `FinishPreload`?");
+                if (!mz_zip_writer_add_mem(&zip_ar, path.string().c_str(), v.data(),
+                                           v.size() * sizeof(T), MZ_BEST_COMPRESSION)) {
+                    LOG_ERROR(Render, "Failed to add {} to the archive", path.string().c_str());
+                }
+            } else {
+                using namespace Common::FS;
+                const auto file = IOFile{path, FileAccessMode::Create};
+                file.Write(v);
+            }
+        }};
+        std::scoped_lock lock{m_request};
+        req_queue.emplace(std::move(request));
+    }
+
+    std::scoped_lock lk{submit_mutex};
+    ++num_requests;
+    request_cv.notify_one();
+    return true;
+}
+
+template <typename T>
+void LoadVector(BlobType type, std::filesystem::path& path, std::vector<T>& v) {
+    using namespace Common::FS;
+    path.replace_extension(GetBlobFileExtension(type));
+    if (Config::isPipelineCacheArchived()) {
+        int index{-1};
+        index = mz_zip_reader_locate_file(&zip_ar, path.string().c_str(), nullptr, 0);
+        if (index < 0) {
+            LOG_WARNING(Render, "File {} is not found in the archive", path.string().c_str());
+            return;
+        }
+        mz_zip_archive_file_stat stat{};
+        mz_zip_reader_file_stat(&zip_ar, index, &stat);
+        v.resize(stat.m_uncomp_size / sizeof(T));
+        mz_zip_reader_extract_to_mem(&zip_ar, index, v.data(), stat.m_uncomp_size, 0);
+    } else {
+        const auto file = IOFile{path, FileAccessMode::Read};
+        v.resize(file.GetSize() / sizeof(T));
+        file.Read(v);
+    }
+}
+
+bool DataBase::Save(BlobType type, const std::string& name, std::vector<u8>&& data) {
+    if (!opened) {
+        return false;
+    }
+
+    auto path = Config::isPipelineCacheArchived() ? std::filesystem::path{name} : cache_path / name;
+    return WriteVector(type, std::move(path), std::move(data));
+}
+
+bool DataBase::Save(BlobType type, const std::string& name, std::vector<u32>&& data) {
+    if (!opened) {
+        return false;
+    }
+
+    auto path = Config::isPipelineCacheArchived() ? std::filesystem::path{name} : cache_path / name;
+    return WriteVector(type, std::move(path), std::move(data));
+}
+
+void DataBase::Load(BlobType type, const std::string& name, std::vector<u8>& data) {
+    if (!opened) {
+        return;
+    }
+
+    auto path = Config::isPipelineCacheArchived() ? std::filesystem::path{name} : cache_path / name;
+    return LoadVector(type, path, data);
+}
+
+void DataBase::Load(BlobType type, const std::string& name, std::vector<u32>& data) {
+    if (!opened) {
+        return;
+    }
+
+    auto path = Config::isPipelineCacheArchived() ? std::filesystem::path{name} : cache_path / name;
+    return LoadVector(type, path, data);
+}
+
+void DataBase::ForEachBlob(BlobType type, const std::function<void(std::vector<u8>&& data)>& func) {
+    const auto& ext = GetBlobFileExtension(type);
+    if (Config::isPipelineCacheArchived()) {
+        const auto num_files = mz_zip_reader_get_num_files(&zip_ar);
+        for (int index = 0; index < num_files; ++index) {
+            std::array<char, MZ_ZIP_MAX_ARCHIVE_FILENAME_SIZE> file_name{};
+            file_name.fill(0);
+            mz_zip_reader_get_filename(&zip_ar, index, file_name.data(), file_name.size());
+            if (std::string{file_name.data()}.ends_with(ext)) {
+                mz_zip_archive_file_stat stat{};
+                mz_zip_reader_file_stat(&zip_ar, index, &stat);
+                std::vector<u8> data(stat.m_uncomp_size);
+                mz_zip_reader_extract_to_mem(&zip_ar, index, data.data(), data.size(), 0);
+                func(std::move(data));
+            }
+        }
+    } else {
+        for (const auto& file_name : std::filesystem::directory_iterator{cache_path}) {
+            if (file_name.path().extension().string().ends_with(ext)) {
+                using namespace Common::FS;
+                const auto& file = IOFile{file_name, FileAccessMode::Read};
+                if (file.IsOpen()) {
+                    std::vector<u8> data(file.GetSize());
+                    file.Read(data);
+                    func(std::move(data));
+                }
+            }
+        }
+    }
+}
+
+void DataBase::FinishPreload() {
+    if (Config::isPipelineCacheArchived()) {
+        mz_zip_writer_init_from_reader(&zip_ar, cache_path.string().c_str());
+        ar_is_read_only = false;
+    }
+}
+
+} // namespace Storage

src/video_core/cache_storage.h

@@ -0,0 +1,50 @@
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "common/path_util.h"
+#include "common/singleton.h"
+#include "common/types.h"
+
+#include <functional>
+#include <thread>
+#include <vector>
+
+namespace Storage {
+
+enum class BlobType : u32 {
+    ShaderMeta,
+    ShaderBinary,
+    PipelineKey,
+    ShaderProfile,
+};
+
+class DataBase {
+public:
+    static DataBase& Instance() {
+        return *Common::Singleton<DataBase>::Instance();
+    }
+
+    void Open();
+    void Close();
+    [[nodiscard]] bool IsOpened() const {
+        return opened;
+    }
+    void FinishPreload();
+
+    bool Save(BlobType type, const std::string& name, std::vector<u8>&& data);
+    bool Save(BlobType type, const std::string& name, std::vector<u32>&& data);
+
+    void Load(BlobType type, const std::string& name, std::vector<u8>& data);
+    void Load(BlobType type, const std::string& name, std::vector<u32>& data);
+
+    void ForEachBlob(BlobType type, const std::function<void(std::vector<u8>&& data)>& func);
+
+private:
+    std::jthread io_worker{};
+    std::filesystem::path cache_path{};
+    bool opened{};
+};
+
+} // namespace Storage

src/video_core/renderer_vulkan/vk_compute_pipeline.cpp

@@ -13,7 +13,8 @@ namespace Vulkan {
 ComputePipeline::ComputePipeline(const Instance& instance, Scheduler& scheduler,
                                  DescriptorHeap& desc_heap, const Shader::Profile& profile,
                                  vk::PipelineCache pipeline_cache, ComputePipelineKey compute_key_,
-                                 const Shader::Info& info_, vk::ShaderModule module)
+                                 const Shader::Info& info_, vk::ShaderModule module,
+                                 SerializationSupport& sdata, bool preloading /*=false*/)
     : Pipeline{instance, scheduler, desc_heap, profile, pipeline_cache, true},
       compute_key{compute_key_} {
     auto& info = stages[int(Shader::LogicalStage::Compute)];
@@ -29,7 +30,11 @@ ComputePipeline::ComputePipeline(const Instance& instance, Scheduler& scheduler,
     u32 binding{};
     boost::container::small_vector<vk::DescriptorSetLayoutBinding, 32> bindings;
     for (const auto& buffer : info->buffers) {
-        const auto sharp = buffer.GetSharp(*info);
+        // During deserialization, we don't have access to the UD to fetch sharp data. To address
+        // this properly we need to track shaprs or portion of them in `sdata`, but since we're
+        // interested only in "is storage" flag (which is not even effective atm), we can take a
+        // shortcut there.
+        const auto sharp = preloading ? AmdGpu::Buffer{} : buffer.GetSharp(*info);
         bindings.push_back({
             .binding = binding++,
             .descriptorType = buffer.IsStorage(sharp) ? vk::DescriptorType::eStorageBuffer

src/video_core/renderer_vulkan/vk_compute_pipeline.h

@@ -11,6 +11,10 @@ class BufferCache;
 class TextureCache;
 } // namespace VideoCore
 
+namespace Serialization {
+struct Archive;
+}
+
 namespace Vulkan {
 
 class Instance;
@@ -26,14 +30,24 @@ struct ComputePipelineKey {
     friend bool operator!=(const ComputePipelineKey& lhs, const ComputePipelineKey& rhs) {
         return !(lhs == rhs);
     }
+
+    void Serialize(Serialization::Archive& ar) const;
+    bool Deserialize(Serialization::Archive& ar);
 };
 
 class ComputePipeline : public Pipeline {
 public:
+    struct SerializationSupport {
+        u32 dummy{};
+
+        void Serialize(Serialization::Archive& ar) const;
+        bool Deserialize(Serialization::Archive& ar);
+    };
+
     ComputePipeline(const Instance& instance, Scheduler& scheduler, DescriptorHeap& desc_heap,
                     const Shader::Profile& profile, vk::PipelineCache pipeline_cache,
                     ComputePipelineKey compute_key, const Shader::Info& info,
-                    vk::ShaderModule module);
+                    vk::ShaderModule module, SerializationSupport& sdata, bool preloading);
     ~ComputePipeline();
 
 private:

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -41,12 +41,12 @@ GraphicsPipeline::GraphicsPipeline(
     vk::PipelineCache pipeline_cache, std::span<const Shader::Info*, MaxShaderStages> infos,
     std::span<const Shader::RuntimeInfo, MaxShaderStages> runtime_infos,
     std::optional<const Shader::Gcn::FetchShaderData> fetch_shader_,
-    std::span<const vk::ShaderModule> modules)
+    std::span<const vk::ShaderModule> modules, SerializationSupport& sdata, bool preloading)
     : Pipeline{instance, scheduler, desc_heap, profile, pipeline_cache}, key{key_},
       fetch_shader{std::move(fetch_shader_)} {
     const vk::Device device = instance.GetDevice();
     std::ranges::copy(infos, stages.begin());
-    BuildDescSetLayout();
+    BuildDescSetLayout(preloading);
     const auto debug_str = GetDebugString();
 
     const vk::PushConstantRange push_constants = {
@@ -68,41 +68,35 @@ GraphicsPipeline::GraphicsPipeline(
     pipeline_layout = std::move(layout);
     SetObjectName(device, *pipeline_layout, "Graphics PipelineLayout {}", debug_str);
 
-    VertexInputs<vk::VertexInputAttributeDescription> vertex_attributes;
+    if (!preloading) {
-    VertexInputs<vk::VertexInputBindingDescription> vertex_bindings;
+        VertexInputs<AmdGpu::Buffer> guest_buffers;
-    VertexInputs<vk::VertexInputBindingDivisorDescriptionEXT> divisors;
+        if (!instance.IsVertexInputDynamicState()) {
-    VertexInputs<AmdGpu::Buffer> guest_buffers;
+            const auto& vs_info = runtime_infos[u32(Shader::LogicalStage::Vertex)].vs_info;
-    if (!instance.IsVertexInputDynamicState()) {
+            GetVertexInputs(sdata.vertex_attributes, sdata.vertex_bindings, sdata.divisors,
-        const auto& vs_info = runtime_infos[u32(Shader::LogicalStage::Vertex)].vs_info;
+                            guest_buffers, vs_info.step_rate_0, vs_info.step_rate_1);
-        GetVertexInputs(vertex_attributes, vertex_bindings, divisors, guest_buffers,
+        }
-                        vs_info.step_rate_0, vs_info.step_rate_1);
     }
 
     const vk::PipelineVertexInputDivisorStateCreateInfo divisor_state = {
-        .vertexBindingDivisorCount = static_cast<u32>(divisors.size()),
+        .vertexBindingDivisorCount = static_cast<u32>(sdata.divisors.size()),
-        .pVertexBindingDivisors = divisors.data(),
+        .pVertexBindingDivisors = sdata.divisors.data(),
     };
 
     const vk::PipelineVertexInputStateCreateInfo vertex_input_info = {
-        .pNext = divisors.empty() ? nullptr : &divisor_state,
+        .pNext = sdata.divisors.empty() ? nullptr : &divisor_state,
-        .vertexBindingDescriptionCount = static_cast<u32>(vertex_bindings.size()),
+        .vertexBindingDescriptionCount = static_cast<u32>(sdata.vertex_bindings.size()),
-        .pVertexBindingDescriptions = vertex_bindings.data(),
+        .pVertexBindingDescriptions = sdata.vertex_bindings.data(),
-        .vertexAttributeDescriptionCount = static_cast<u32>(vertex_attributes.size()),
+        .vertexAttributeDescriptionCount = static_cast<u32>(sdata.vertex_attributes.size()),
-        .pVertexAttributeDescriptions = vertex_attributes.data(),
+        .pVertexAttributeDescriptions = sdata.vertex_attributes.data(),
     };
 
     const auto topology = LiverpoolToVK::PrimitiveType(key.prim_type);
     const vk::PipelineInputAssemblyStateCreateInfo input_assembly = {
         .topology = topology,
-        // Avoid warning spam on all pipelines about unsupported restart disable, if not supported.
-        // However, must be false for list topologies to avoid validation errors.
-        .primitiveRestartEnable =
-            !instance.IsPrimitiveRestartDisableSupported() && !IsPrimitiveTopologyList(topology),
     };
 
     const bool is_rect_list = key.prim_type == AmdGpu::PrimitiveType::RectList;
     const bool is_quad_list = key.prim_type == AmdGpu::PrimitiveType::QuadList;
-    const auto& fs_info = runtime_infos[u32(Shader::LogicalStage::Fragment)].fs_info;
     const vk::PipelineTessellationStateCreateInfo tessellation_state = {
         .patchControlPoints = is_rect_list ? 3U : (is_quad_list ? 4U : key.patch_control_points),
     };
@@ -132,12 +126,15 @@ GraphicsPipeline::GraphicsPipeline(
         raster_chain.unlink<vk::PipelineRasterizationDepthClipStateCreateInfoEXT>();
     }
 
-    const vk::PipelineMultisampleStateCreateInfo multisampling = {
+    if (!preloading) {
-        .rasterizationSamples = LiverpoolToVK::NumSamples(
+        const auto& fs_info = runtime_infos[u32(Shader::LogicalStage::Fragment)].fs_info;
-            key.num_samples, instance.GetColorSampleCounts() & instance.GetDepthSampleCounts()),
+        sdata.multisampling = {
-        .sampleShadingEnable =
+            .rasterizationSamples = LiverpoolToVK::NumSamples(
-            fs_info.addr_flags.persp_sample_ena || fs_info.addr_flags.linear_sample_ena,
+                key.num_samples, instance.GetColorSampleCounts() & instance.GetDepthSampleCounts()),
-    };
+            .sampleShadingEnable =
+                fs_info.addr_flags.persp_sample_ena || fs_info.addr_flags.linear_sample_ena,
+        };
+    }
 
     const vk::PipelineViewportDepthClipControlCreateInfoEXT clip_control = {
         .negativeOneToOne = key.clip_space == AmdGpu::ClipSpace::MinusWToW,
@@ -156,12 +153,9 @@ GraphicsPipeline::GraphicsPipeline(
         vk::DynamicState::eStencilCompareMask, vk::DynamicState::eStencilWriteMask,
         vk::DynamicState::eStencilOp,          vk::DynamicState::eCullMode,
         vk::DynamicState::eFrontFace,          vk::DynamicState::eRasterizerDiscardEnable,
-        vk::DynamicState::eLineWidth,
+        vk::DynamicState::eLineWidth,          vk::DynamicState::ePrimitiveRestartEnable,
     };
 
-    if (instance.IsPrimitiveRestartDisableSupported()) {
-        dynamic_states.push_back(vk::DynamicState::ePrimitiveRestartEnable);
-    }
     if (instance.IsDepthBoundsSupported()) {
         dynamic_states.push_back(vk::DynamicState::eDepthBoundsTestEnable);
         dynamic_states.push_back(vk::DynamicState::eDepthBounds);
@@ -171,7 +165,7 @@ GraphicsPipeline::GraphicsPipeline(
     }
     if (instance.IsVertexInputDynamicState()) {
         dynamic_states.push_back(vk::DynamicState::eVertexInputEXT);
-    } else if (!vertex_bindings.empty()) {
+    } else if (!sdata.vertex_bindings.empty()) {
         dynamic_states.push_back(vk::DynamicState::eVertexInputBindingStride);
     }
 
@@ -207,10 +201,13 @@ GraphicsPipeline::GraphicsPipeline(
         });
     } else if (is_rect_list || is_quad_list) {
         const auto type = is_quad_list ? AuxShaderType::QuadListTCS : AuxShaderType::RectListTCS;
-        auto tcs = Shader::Backend::SPIRV::EmitAuxilaryTessShader(type, fs_info);
+        if (!preloading) {
+            const auto& fs_info = runtime_infos[u32(Shader::LogicalStage::Fragment)].fs_info;
+            sdata.tcs = Shader::Backend::SPIRV::EmitAuxilaryTessShader(type, fs_info);
+        }
         shader_stages.emplace_back(vk::PipelineShaderStageCreateInfo{
             .stage = vk::ShaderStageFlagBits::eTessellationControl,
-            .module = CompileSPV(tcs, instance.GetDevice()),
+            .module = CompileSPV(sdata.tcs, instance.GetDevice()),
             .pName = "main",
         });
     }
@@ -222,11 +219,14 @@ GraphicsPipeline::GraphicsPipeline(
             .pName = "main",
         });
     } else if (is_rect_list || is_quad_list) {
-        auto tes =
+        if (!preloading) {
-            Shader::Backend::SPIRV::EmitAuxilaryTessShader(AuxShaderType::PassthroughTES, fs_info);
+            const auto& fs_info = runtime_infos[u32(Shader::LogicalStage::Fragment)].fs_info;
+            sdata.tes = Shader::Backend::SPIRV::EmitAuxilaryTessShader(
+                AuxShaderType::PassthroughTES, fs_info);
+        }
         shader_stages.emplace_back(vk::PipelineShaderStageCreateInfo{
             .stage = vk::ShaderStageFlagBits::eTessellationEvaluation,
-            .module = CompileSPV(tes, instance.GetDevice()),
+            .module = CompileSPV(sdata.tes, instance.GetDevice()),
             .pName = "main",
         });
     }
@@ -367,7 +367,7 @@ GraphicsPipeline::GraphicsPipeline(
         .pTessellationState = &tessellation_state,
         .pViewportState = &viewport_info,
         .pRasterizationState = &raster_chain.get(),
-        .pMultisampleState = &multisampling,
+        .pMultisampleState = &sdata.multisampling,
         .pColorBlendState = &color_blending,
         .pDynamicState = &dynamic_info,
         .layout = *pipeline_layout,
@@ -435,7 +435,7 @@ template void GraphicsPipeline::GetVertexInputs(
     VertexInputs<vk::VertexInputBindingDivisorDescriptionEXT>& divisors,
     VertexInputs<AmdGpu::Buffer>& guest_buffers, u32 step_rate_0, u32 step_rate_1) const;
 
-void GraphicsPipeline::BuildDescSetLayout() {
+void GraphicsPipeline::BuildDescSetLayout(bool preloading) {
     boost::container::small_vector<vk::DescriptorSetLayoutBinding, 32> bindings;
     u32 binding{};
 
@@ -445,7 +445,9 @@ void GraphicsPipeline::BuildDescSetLayout() {
         }
         const auto stage_bit = LogicalStageToStageBit[u32(stage->l_stage)];
         for (const auto& buffer : stage->buffers) {
-            const auto sharp = buffer.GetSharp(*stage);
+            const auto sharp =
+                preloading ? AmdGpu::Buffer{}
+                           : buffer.GetSharp(*stage); // See for the comment in compute PL creation
             bindings.push_back({
                 .binding = binding++,
                 .descriptorType = buffer.IsStorage(sharp) ? vk::DescriptorType::eStorageBuffer

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -63,17 +63,33 @@ struct GraphicsPipelineKey {
     bool operator==(const GraphicsPipelineKey& key) const noexcept {
         return std::memcmp(this, &key, sizeof(key)) == 0;
     }
+
+    void Serialize(Serialization::Archive& ar) const;
+    bool Deserialize(Serialization::Archive& ar);
 };
 
 class GraphicsPipeline : public Pipeline {
 public:
+    struct SerializationSupport {
+        VertexInputs<vk::VertexInputAttributeDescription> vertex_attributes{};
+        VertexInputs<vk::VertexInputBindingDescription> vertex_bindings{};
+        VertexInputs<vk::VertexInputBindingDivisorDescriptionEXT> divisors{};
+        vk::PipelineMultisampleStateCreateInfo multisampling{};
+        std::vector<u32> tcs{};
+        std::vector<u32> tes{};
+
+        void Serialize(Serialization::Archive& ar) const;
+        bool Deserialize(Serialization::Archive& ar);
+    };
+
     GraphicsPipeline(const Instance& instance, Scheduler& scheduler, DescriptorHeap& desc_heap,
                      const Shader::Profile& profile, const GraphicsPipelineKey& key,
                      vk::PipelineCache pipeline_cache,
                      std::span<const Shader::Info*, MaxShaderStages> stages,
                      std::span<const Shader::RuntimeInfo, MaxShaderStages> runtime_infos,
                      std::optional<const Shader::Gcn::FetchShaderData> fetch_shader,
-                     std::span<const vk::ShaderModule> modules);
+                     std::span<const vk::ShaderModule> modules, SerializationSupport& sdata,
+                     bool preloading);
     ~GraphicsPipeline();
 
     const std::optional<const Shader::Gcn::FetchShaderData>& GetFetchShader() const noexcept {
@@ -92,7 +108,7 @@ public:
                          u32 step_rate_1) const;
 
 private:
-    void BuildDescSetLayout();
+    void BuildDescSetLayout(bool preloading);
 
 private:
     GraphicsPipelineKey key;

src/video_core/renderer_vulkan/vk_instance.h

@@ -411,11 +411,6 @@ public:
                properties.limits.framebufferStencilSampleCounts;
     }
 
-    /// Returns whether disabling primitive restart is supported.
-    bool IsPrimitiveRestartDisableSupported() const {
-        return driver_id != vk::DriverId::eMoltenvk;
-    }
-
     /// Returns true if logic ops are supported by the device.
     bool IsLogicOpSupported() const {
         return features.logicOp;

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -13,9 +13,10 @@
 #include "shader_recompiler/recompiler.h"
 #include "shader_recompiler/runtime_info.h"
 #include "video_core/amdgpu/liverpool.h"
+#include "video_core/cache_storage.h"
 #include "video_core/renderer_vulkan/liverpool_to_vk.h"
 #include "video_core/renderer_vulkan/vk_instance.h"
-#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+#include "video_core/renderer_vulkan/vk_pipeline_serialization.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_shader_util.h"
 
@@ -223,6 +224,13 @@ PipelineCache::PipelineCache(const Instance& instance_, Scheduler& scheduler_,
       desc_heap{instance, scheduler.GetMasterSemaphore(), DescriptorHeapSizes} {
     const auto& vk12_props = instance.GetVk12Properties();
     profile = Shader::Profile{
+        // When binding a UBO, we calculate its size considering the offset in the larger buffer
+        // cache underlying resource. In some cases, it may produce sizes exceeding the system
+        // maximum allowed UBO range, so we need to reduce the threshold to prevent issues.
+        .max_ubo_size = instance.UniformMaxSize() - instance.UniformMinAlignment(),
+        .max_viewport_width = instance.GetMaxViewportWidth(),
+        .max_viewport_height = instance.GetMaxViewportHeight(),
+        .max_shared_memory_size = instance.MaxComputeSharedMemorySize(),
         .supported_spirv = SpirvVersion1_6,
         .subgroup_size = instance.SubgroupSize(),
         .support_int8 = instance.IsShaderInt8Supported(),
@@ -258,14 +266,10 @@ PipelineCache::PipelineCache(const Instance& instance_, Scheduler& scheduler_,
                               instance.GetDriverID() == vk::DriverId::eMoltenvk,
         .needs_buffer_offsets = instance.StorageMinAlignment() > 4,
         .needs_unorm_fixup = instance.GetDriverID() == vk::DriverId::eMoltenvk,
-        // When binding a UBO, we calculate its size considering the offset in the larger buffer
-        // cache underlying resource. In some cases, it may produce sizes exceeding the system
-        // maximum allowed UBO range, so we need to reduce the threshold to prevent issues.
-        .max_ubo_size = instance.UniformMaxSize() - instance.UniformMinAlignment(),
-        .max_viewport_width = instance.GetMaxViewportWidth(),
-        .max_viewport_height = instance.GetMaxViewportHeight(),
-        .max_shared_memory_size = instance.MaxComputeSharedMemorySize(),
     };
+
+    WarmUp();
+
     auto [cache_result, cache] = instance.GetDevice().createPipelineCacheUnique({});
     ASSERT_MSG(cache_result == vk::Result::eSuccess, "Failed to create pipeline cache: {}",
                vk::to_string(cache_result));
@@ -283,9 +287,14 @@ const GraphicsPipeline* PipelineCache::GetGraphicsPipeline() {
         const auto pipeline_hash = std::hash<GraphicsPipelineKey>{}(graphics_key);
         LOG_INFO(Render_Vulkan, "Compiling graphics pipeline {:#x}", pipeline_hash);
 
-        it.value() = std::make_unique<GraphicsPipeline>(instance, scheduler, desc_heap, profile,
+        GraphicsPipeline::SerializationSupport sdata{};
-                                                        graphics_key, *pipeline_cache, infos,
+        it.value() = std::make_unique<GraphicsPipeline>(
-                                                        runtime_infos, fetch_shader, modules);
+            instance, scheduler, desc_heap, profile, graphics_key, *pipeline_cache, infos,
+            runtime_infos, fetch_shader, modules, sdata, false);
+
+        RegisterPipelineData(graphics_key, pipeline_hash, sdata);
+        ++num_new_pipelines;
+
         if (Config::collectShadersForDebug()) {
             for (auto stage = 0; stage < MaxShaderStages; ++stage) {
                 if (infos[stage]) {
@@ -294,6 +303,7 @@ const GraphicsPipeline* PipelineCache::GetGraphicsPipeline() {
                 }
             }
         }
+        fetch_shader.reset();
     }
     return it->second.get();
 }
@@ -307,9 +317,13 @@ const ComputePipeline* PipelineCache::GetComputePipeline() {
         const auto pipeline_hash = std::hash<ComputePipelineKey>{}(compute_key);
         LOG_INFO(Render_Vulkan, "Compiling compute pipeline {:#x}", pipeline_hash);
 
-        it.value() =
+        ComputePipeline::SerializationSupport sdata{};
-            std::make_unique<ComputePipeline>(instance, scheduler, desc_heap, profile,
+        it.value() = std::make_unique<ComputePipeline>(instance, scheduler, desc_heap, profile,
-                                              *pipeline_cache, compute_key, *infos[0], modules[0]);
+                                                       *pipeline_cache, compute_key, *infos[0],
+                                                       modules[0], sdata, false);
+        RegisterPipelineData(compute_key, sdata);
+        ++num_new_pipelines;
+
         if (Config::collectShadersForDebug()) {
             auto& m = modules[0];
             module_related_pipelines[m].emplace_back(compute_key);
@@ -445,6 +459,7 @@ bool PipelineCache::RefreshGraphicsStages() {
     };
 
     infos.fill(nullptr);
+    modules.fill(nullptr);
     bind_stage(Stage::Fragment, LogicalStage::Fragment);
 
     const auto* fs_info = infos[static_cast<u32>(LogicalStage::Fragment)];
@@ -515,7 +530,7 @@ bool PipelineCache::RefreshComputeKey() {
 }
 
 vk::ShaderModule PipelineCache::CompileModule(Shader::Info& info, Shader::RuntimeInfo& runtime_info,
-                                              std::span<const u32> code, size_t perm_idx,
+                                              const std::span<const u32>& code, size_t perm_idx,
                                               Shader::Backend::Bindings& binding) {
     LOG_INFO(Render_Vulkan, "Compiling {} shader {:#x} {}", info.stage, info.pgm_hash,
              perm_idx != 0 ? "(permutation)" : "");
@@ -536,6 +551,8 @@ vk::ShaderModule PipelineCache::CompileModule(Shader::Info& info, Shader::Runtim
         module = CompileSPV(spv, instance.GetDevice());
     }
 
+    RegisterShaderBinary(std::move(spv), info.pgm_hash, perm_idx);
+
     const auto name = GetShaderName(info.stage, info.pgm_hash, perm_idx);
     Vulkan::SetObjectName(instance.GetDevice(), module, name);
     if (Config::collectShadersForDebug()) {
@@ -546,7 +563,7 @@ vk::ShaderModule PipelineCache::CompileModule(Shader::Info& info, Shader::Runtim
 }
 
 PipelineCache::Result PipelineCache::GetProgram(Stage stage, LogicalStage l_stage,
-                                                Shader::ShaderParams params,
+                                                const Shader::ShaderParams& params,
                                                 Shader::Backend::Bindings& binding) {
     auto runtime_info = BuildRuntimeInfo(stage, l_stage);
     auto [it_pgm, new_program] = program_cache.try_emplace(params.hash);
@@ -555,32 +572,42 @@ PipelineCache::Result PipelineCache::GetProgram(Stage stage, LogicalStage l_stag
         auto& program = it_pgm.value();
         auto start = binding;
         const auto module = CompileModule(program->info, runtime_info, params.code, 0, binding);
-        const auto spec = Shader::StageSpecialization(program->info, runtime_info, profile, start);
+        auto spec = Shader::StageSpecialization(program->info, runtime_info, profile, start);
+        const auto perm_hash = HashCombine(params.hash, 0);
+
+        RegisterShaderMeta(program->info, spec.fetch_shader_data, spec, perm_hash, 0);
         program->AddPermut(module, std::move(spec));
-        return std::make_tuple(&program->info, module, spec.fetch_shader_data,
+        return std::make_tuple(&program->info, module, program->modules[0].spec.fetch_shader_data,
-                               HashCombine(params.hash, 0));
+                               perm_hash);
     }
-    it_pgm.value()->info.user_data = params.user_data;
 
     auto& program = it_pgm.value();
     auto& info = program->info;
+    info.pgm_base = params.Base(); // Needs to be actualized for inline cbuffer address fixup
+    info.user_data = params.user_data;
     info.RefreshFlatBuf();
-    const auto spec = Shader::StageSpecialization(info, runtime_info, profile, binding);
+    auto spec = Shader::StageSpecialization(info, runtime_info, profile, binding);
+
     size_t perm_idx = program->modules.size();
+    u64 perm_hash = HashCombine(params.hash, perm_idx);
+
     vk::ShaderModule module{};
 
     const auto it = std::ranges::find(program->modules, spec, &Program::Module::spec);
     if (it == program->modules.end()) {
         auto new_info = Shader::Info(stage, l_stage, params);
         module = CompileModule(new_info, runtime_info, params.code, perm_idx, binding);
+
+        RegisterShaderMeta(info, spec.fetch_shader_data, spec, perm_hash, perm_idx);
         program->AddPermut(module, std::move(spec));
     } else {
         info.AddBindings(binding);
         module = it->module;
         perm_idx = std::distance(program->modules.begin(), it);
+        perm_hash = HashCombine(params.hash, perm_idx);
     }
-    return std::make_tuple(&info, module, spec.fetch_shader_data,
+    return std::make_tuple(&program->info, module,
-                           HashCombine(params.hash, perm_idx));
+                           program->modules[perm_idx].spec.fetch_shader_data, perm_hash);
 }
 
 std::optional<vk::ShaderModule> PipelineCache::ReplaceShader(vk::ShaderModule module,
@@ -654,5 +681,4 @@ std::optional<std::vector<u32>> PipelineCache::GetShaderPatch(u64 hash, Shader::
     file.Read(code);
     return code;
 }
-
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_pipeline_cache.h

@@ -23,6 +23,10 @@ namespace AmdGpu {
 class Liverpool;
 }
 
+namespace Serialization {
+struct Archive;
+}
+
 namespace Shader {
 struct Info;
 }
@@ -38,17 +42,25 @@ struct Program {
         vk::ShaderModule module;
         Shader::StageSpecialization spec;
     };
-    using ModuleList = boost::container::small_vector<Module, 8>;
+    static constexpr size_t MaxPermutations = 8;
+    using ModuleList = boost::container::small_vector<Module, MaxPermutations>;
 
     Shader::Info info;
-    ModuleList modules;
+    ModuleList modules{};
 
-    explicit Program(Shader::Stage stage, Shader::LogicalStage l_stage, Shader::ShaderParams params)
+    Program() = default;
+    Program(Shader::Stage stage, Shader::LogicalStage l_stage, Shader::ShaderParams params)
         : info{stage, l_stage, params} {}
 
-    void AddPermut(vk::ShaderModule module, const Shader::StageSpecialization&& spec) {
+    void AddPermut(vk::ShaderModule module, Shader::StageSpecialization&& spec) {
         modules.emplace_back(module, std::move(spec));
     }
+
+    void InsertPermut(vk::ShaderModule module, Shader::StageSpecialization&& spec,
+                      size_t perm_idx) {
+        modules.resize(std::max(modules.size(), perm_idx + 1)); // <-- beware of realloc
+        modules[perm_idx] = {module, std::move(spec)};
+    }
 };
 
 class PipelineCache {
@@ -57,6 +69,13 @@ public:
                            AmdGpu::Liverpool* liverpool);
     ~PipelineCache();
 
+    void WarmUp();
+    void Sync();
+
+    bool LoadComputePipeline(Serialization::Archive& ar);
+    bool LoadGraphicsPipeline(Serialization::Archive& ar);
+    bool LoadPipelineStage(Serialization::Archive& ar, size_t stage);
+
     const GraphicsPipeline* GetGraphicsPipeline();
 
     const ComputePipeline* GetComputePipeline();
@@ -64,7 +83,7 @@ public:
     using Result = std::tuple<const Shader::Info*, vk::ShaderModule,
                               std::optional<Shader::Gcn::FetchShaderData>, u64>;
     Result GetProgram(Shader::Stage stage, Shader::LogicalStage l_stage,
-                      Shader::ShaderParams params, Shader::Backend::Bindings& binding);
+                      const Shader::ShaderParams& params, Shader::Backend::Bindings& binding);
 
     std::optional<vk::ShaderModule> ReplaceShader(vk::ShaderModule module,
                                                   std::span<const u32> spv_code);
@@ -86,10 +105,14 @@ private:
     std::optional<std::vector<u32>> GetShaderPatch(u64 hash, Shader::Stage stage, size_t perm_idx,
                                                    std::string_view ext);
     vk::ShaderModule CompileModule(Shader::Info& info, Shader::RuntimeInfo& runtime_info,
-                                   std::span<const u32> code, size_t perm_idx,
+                                   const std::span<const u32>& code, size_t perm_idx,
                                    Shader::Backend::Bindings& binding);
     const Shader::RuntimeInfo& BuildRuntimeInfo(Shader::Stage stage, Shader::LogicalStage l_stage);
 
+    [[nodiscard]] bool IsPipelineCacheDirty() const {
+        return num_new_pipelines > 0;
+    }
+
 private:
     const Instance& instance;
     Scheduler& scheduler;
@@ -108,6 +131,7 @@ private:
     std::optional<Shader::Gcn::FetchShaderData> fetch_shader{};
     GraphicsPipelineKey graphics_key{};
     ComputePipelineKey compute_key{};
+    u32 num_new_pipelines{}; // new pipelines added to the cache since the game start
 
     // Only if Config::collectShadersForDebug()
     tsl::robin_map<vk::ShaderModule,

src/video_core/renderer_vulkan/vk_pipeline_serialization.cpp

@@ -0,0 +1,480 @@
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "common/config.h"
+#include "common/serdes.h"
+#include "shader_recompiler/frontend/fetch_shader.h"
+#include "shader_recompiler/info.h"
+#include "video_core/cache_storage.h"
+#include "video_core/renderer_vulkan/vk_instance.h"
+#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+#include "video_core/renderer_vulkan/vk_shader_util.h"
+
+namespace Serialization {
+/* You should increment versions below once corresponding serialization scheme is changed. */
+static constexpr u32 ShaderBinaryVersion = 1u;
+static constexpr u32 ShaderMetaVersion = 1u;
+static constexpr u32 PipelineKeyVersion = 1u;
+} // namespace Serialization
+
+namespace Vulkan {
+
+void RegisterPipelineData(const ComputePipelineKey& key,
+                          ComputePipeline::SerializationSupport& sdata) {
+    if (!Storage::DataBase::Instance().IsOpened()) {
+        return;
+    }
+
+    Serialization::Archive ar{};
+    Serialization::Writer pldata{ar};
+
+    pldata.Write(Serialization::PipelineKeyVersion);
+    pldata.Write(u32{1}); // compute
+
+    key.Serialize(ar);
+    sdata.Serialize(ar);
+
+    Storage::DataBase::Instance().Save(Storage::BlobType::PipelineKey,
+                                       fmt::format("c_{:#018x}", key.value), ar.TakeOff());
+}
+
+void RegisterPipelineData(const GraphicsPipelineKey& key, u64 hash,
+                          GraphicsPipeline::SerializationSupport& sdata) {
+    if (!Storage::DataBase::Instance().IsOpened()) {
+        return;
+    }
+
+    Serialization::Archive ar{};
+    Serialization::Writer pldata{ar};
+
+    pldata.Write(Serialization::PipelineKeyVersion);
+    pldata.Write(u32{0}); // graphics
+
+    key.Serialize(ar);
+    sdata.Serialize(ar);
+
+    Storage::DataBase::Instance().Save(Storage::BlobType::PipelineKey,
+                                       fmt::format("g_{:#018x}", hash), ar.TakeOff());
+}
+
+void RegisterShaderMeta(const Shader::Info& info,
+                        const std::optional<Shader::Gcn::FetchShaderData>& fetch_shader_data,
+                        const Shader::StageSpecialization& spec, size_t perm_hash,
+                        size_t perm_idx) {
+    if (!Storage::DataBase::Instance().IsOpened()) {
+        return;
+    }
+
+    Serialization::Archive ar;
+    Serialization::Writer meta{ar};
+
+    meta.Write(Serialization::ShaderMetaVersion);
+    meta.Write(Serialization::ShaderBinaryVersion);
+
+    meta.Write(perm_hash);
+    meta.Write(perm_idx);
+
+    spec.Serialize(ar);
+    info.Serialize(ar);
+
+    Storage::DataBase::Instance().Save(Storage::BlobType::ShaderMeta,
+                                       fmt::format("{:#018x}", perm_hash), ar.TakeOff());
+}
+
+void RegisterShaderBinary(std::vector<u32>&& spv, u64 pgm_hash, size_t perm_idx) {
+    if (!Storage::DataBase::Instance().IsOpened()) {
+        return;
+    }
+
+    Storage::DataBase::Instance().Save(Storage::BlobType::ShaderBinary,
+                                       fmt::format("{:#018x}_{}", pgm_hash, perm_idx),
+                                       std::move(spv));
+}
+
+bool LoadShaderMeta(Serialization::Archive& ar, Shader::Info& info,
+                    std::optional<Shader::Gcn::FetchShaderData>& fetch_shader_data,
+                    Shader::StageSpecialization& spec, size_t& perm_idx) {
+    Serialization::Reader meta{ar};
+
+    u32 meta_version{};
+    meta.Read(meta_version);
+    if (meta_version != Serialization::ShaderMetaVersion) {
+        return false;
+    }
+
+    u32 binary_version{};
+    meta.Read(binary_version);
+    if (binary_version != Serialization::ShaderBinaryVersion) {
+        return false;
+    }
+
+    u64 perm_hash_ar{};
+    meta.Read(perm_hash_ar);
+    meta.Read(perm_idx);
+
+    spec.Deserialize(ar);
+    info.Deserialize(ar);
+
+    fetch_shader_data = spec.fetch_shader_data;
+    return true;
+}
+
+void ComputePipelineKey::Serialize(Serialization::Archive& ar) const {
+    Serialization::Writer key{ar};
+    key.Write(value);
+}
+
+bool ComputePipelineKey::Deserialize(Serialization::Archive& ar) {
+    Serialization::Reader key{ar};
+    key.Read(value);
+    return true;
+}
+
+void ComputePipeline::SerializationSupport::Serialize(Serialization::Archive& ar) const {
+    // Nothing here yet
+    return;
+}
+
+bool ComputePipeline::SerializationSupport::Deserialize(Serialization::Archive& ar) {
+    // Nothing here yet
+    return true;
+}
+
+bool PipelineCache::LoadComputePipeline(Serialization::Archive& ar) {
+    compute_key.Deserialize(ar);
+
+    ComputePipeline::SerializationSupport sdata{};
+    sdata.Deserialize(ar);
+
+    std::vector<u8> meta_blob;
+    Storage::DataBase::Instance().Load(Storage::BlobType::ShaderMeta,
+                                       fmt::format("{:#018x}", compute_key.value), meta_blob);
+    if (meta_blob.empty()) {
+        return false;
+    }
+
+    Serialization::Archive meta_ar{std::move(meta_blob)};
+
+    if (!LoadPipelineStage(meta_ar, 0)) {
+        return false;
+    }
+
+    const auto [it, is_new] = compute_pipelines.try_emplace(compute_key);
+    ASSERT(is_new);
+
+    it.value() =
+        std::make_unique<ComputePipeline>(instance, scheduler, desc_heap, profile, *pipeline_cache,
+                                          compute_key, *infos[0], modules[0], sdata, true);
+
+    infos.fill(nullptr);
+    modules.fill(nullptr);
+
+    return true;
+}
+
+void GraphicsPipelineKey::Serialize(Serialization::Archive& ar) const {
+    Serialization::Writer key{ar};
+
+    key.Write(this, sizeof(*this));
+}
+
+bool GraphicsPipelineKey::Deserialize(Serialization::Archive& ar) {
+    Serialization::Reader key{ar};
+
+    key.Read(this, sizeof(*this));
+    return true;
+}
+
+void GraphicsPipeline::SerializationSupport::Serialize(Serialization::Archive& ar) const {
+    Serialization::Writer sdata{ar};
+
+    sdata.Write(&vertex_attributes, sizeof(vertex_attributes));
+    sdata.Write(&vertex_bindings, sizeof(vertex_bindings));
+    sdata.Write(&divisors, sizeof(divisors));
+    sdata.Write(multisampling);
+    sdata.Write(tcs);
+    sdata.Write(tes);
+}
+
+bool GraphicsPipeline::SerializationSupport::Deserialize(Serialization::Archive& ar) {
+    Serialization::Reader sdata{ar};
+
+    sdata.Read(&vertex_attributes, sizeof(vertex_attributes));
+    sdata.Read(&vertex_bindings, sizeof(vertex_bindings));
+    sdata.Read(&divisors, sizeof(divisors));
+    sdata.Read(multisampling);
+    sdata.Read(tcs);
+    sdata.Read(tes);
+    return true;
+}
+
+bool PipelineCache::LoadGraphicsPipeline(Serialization::Archive& ar) {
+    graphics_key.Deserialize(ar);
+
+    GraphicsPipeline::SerializationSupport sdata{};
+    sdata.Deserialize(ar);
+
+    for (int stage_idx = 0; stage_idx < MaxShaderStages; ++stage_idx) {
+        const auto& hash = graphics_key.stage_hashes[stage_idx];
+        if (!hash) {
+            continue;
+        }
+
+        std::vector<u8> meta_blob;
+        Storage::DataBase::Instance().Load(Storage::BlobType::ShaderMeta,
+                                           fmt::format("{:#018x}", hash), meta_blob);
+        if (meta_blob.empty()) {
+            return false;
+        }
+
+        Serialization::Archive meta_ar{std::move(meta_blob)};
+
+        if (!LoadPipelineStage(meta_ar, stage_idx)) {
+            return false;
+        }
+    }
+
+    const auto [it, is_new] = graphics_pipelines.try_emplace(graphics_key);
+    ASSERT(is_new);
+
+    it.value() = std::make_unique<GraphicsPipeline>(
+        instance, scheduler, desc_heap, profile, graphics_key, *pipeline_cache, infos,
+        runtime_infos, fetch_shader, modules, sdata, true);
+
+    infos.fill(nullptr);
+    modules.fill(nullptr);
+    fetch_shader.reset();
+
+    return true;
+}
+
+bool PipelineCache::LoadPipelineStage(Serialization::Archive& ar, size_t stage) {
+    auto program = std::make_unique<Program>();
+    Shader::StageSpecialization spec{};
+    spec.info = &program->info;
+    size_t perm_idx{};
+    if (!LoadShaderMeta(ar, program->info, fetch_shader, spec, perm_idx)) {
+        return false;
+    }
+
+    std::vector<u32> spv{};
+    Storage::DataBase::Instance().Load(Storage::BlobType::ShaderBinary,
+                                       fmt::format("{:#018x}_{}", program->info.pgm_hash, perm_idx),
+                                       spv);
+    if (spv.empty()) {
+        return false;
+    }
+
+    // Permutation hash depends on shader variation index. To prevent collisions, we need insert it
+    // at the exact position rather than append
+
+    vk::ShaderModule module{};
+
+    auto [it_pgm, new_program] = program_cache.try_emplace(program->info.pgm_hash);
+    if (new_program) {
+        module = CompileSPV(spv, instance.GetDevice());
+        it_pgm.value() = std::move(program);
+    } else {
+        const auto& it = std::ranges::find(it_pgm.value()->modules, spec, &Program::Module::spec);
+        if (it != it_pgm.value()->modules.end()) {
+            // If the permutation is already preloaded, make sure it has the same permutation index
+            const auto idx = std::distance(it_pgm.value()->modules.begin(), it);
+            ASSERT_MSG(perm_idx == idx, "Permutation {} is already inserted at {}! ({}_{:x})",
+                       perm_idx, idx, program->info.stage, program->info.pgm_hash);
+            module = it->module;
+        } else {
+            module = CompileSPV(spv, instance.GetDevice());
+        }
+    }
+    it_pgm.value()->InsertPermut(module, std::move(spec), perm_idx);
+
+    infos[stage] = &it_pgm.value()->info;
+    modules[stage] = module;
+
+    return true;
+}
+
+void PipelineCache::WarmUp() {
+    if (!Config::isPipelineCacheEnabled()) {
+        return;
+    }
+
+    Storage::DataBase::Instance().Open();
+
+    // Check if cache is compatible
+    std::vector<u8> profile_data{};
+    Storage::DataBase::Instance().Load(Storage::BlobType::ShaderProfile, "profile", profile_data);
+    if (profile_data.empty()) {
+        Storage::DataBase::Instance().FinishPreload();
+
+        profile_data.resize(sizeof(profile));
+        std::memcpy(profile_data.data(), &profile, sizeof(profile));
+        Storage::DataBase::Instance().Save(Storage::BlobType::ShaderProfile, "profile",
+                                           std::move(profile_data));
+        return;
+    }
+    if (std::memcmp(profile_data.data(), &profile, sizeof(profile)) != 0) {
+        LOG_WARNING(Render,
+                    "Pipeline cache isn't compatible with current system. Ignoring the cache");
+        return;
+    }
+
+    u32 num_pipelines{};
+    u32 num_total_pipelines{};
+
+    Storage::DataBase::Instance().ForEachBlob(
+        Storage::BlobType::PipelineKey, [&](std::vector<u8>&& data) {
+            ++num_total_pipelines;
+
+            Serialization::Archive ar{std::move(data)};
+            Serialization::Reader pldata{ar};
+
+            u32 version{};
+            pldata.Read(version);
+            if (version != Serialization::PipelineKeyVersion) {
+                return;
+            }
+
+            u32 is_compute{};
+            pldata.Read(is_compute);
+
+            bool result{};
+            if (is_compute) {
+                result = LoadComputePipeline(ar);
+            } else {
+                result = LoadGraphicsPipeline(ar);
+            }
+
+            if (result) {
+                ++num_pipelines;
+            }
+        });
+
+    LOG_INFO(Render, "Preloaded {} pipelines", num_pipelines);
+    if (num_total_pipelines > num_pipelines) {
+        LOG_WARNING(Render, "{} stale pipelines were found. Consider re-generating the cache",
+                    num_total_pipelines - num_pipelines);
+    }
+
+    Storage::DataBase::Instance().FinishPreload();
+}
+
+void PipelineCache::Sync() {
+    Storage::DataBase::Instance().Close();
+}
+
+} // namespace Vulkan
+
+namespace Shader {
+
+void Info::Serialize(Serialization::Archive& ar) const {
+    Serialization::Writer info{ar};
+
+    info.Write(this, sizeof(InfoPersistent));
+    info.Write(flattened_ud_buf);
+    srt_info.Serialize(ar);
+}
+
+bool Info::Deserialize(Serialization::Archive& ar) {
+    Serialization::Reader info{ar};
+
+    info.Read(this, sizeof(Shader::InfoPersistent));
+    info.Read(flattened_ud_buf);
+
+    return srt_info.Deserialize(ar);
+}
+
+void Gcn::FetchShaderData::Serialize(Serialization::Archive& ar) const {
+    Serialization::Writer fetch{ar};
+    ar.Grow(6 + attributes.size() * sizeof(VertexAttribute));
+
+    fetch.Write(size);
+    fetch.Write(vertex_offset_sgpr);
+    fetch.Write(instance_offset_sgpr);
+    fetch.Write(attributes);
+}
+
+bool Gcn::FetchShaderData::Deserialize(Serialization::Archive& ar) {
+    Serialization::Reader fetch{ar};
+
+    fetch.Read(size);
+    fetch.Read(vertex_offset_sgpr);
+    fetch.Read(instance_offset_sgpr);
+    fetch.Read(attributes);
+
+    return true;
+}
+
+void PersistentSrtInfo::Serialize(Serialization::Archive& ar) const {
+    Serialization::Writer srt{ar};
+
+    srt.Write(this, sizeof(*this));
+    if (walker_func_size) {
+        srt.Write(reinterpret_cast<void*>(walker_func), walker_func_size);
+    }
+}
+
+bool PersistentSrtInfo::Deserialize(Serialization::Archive& ar) {
+    Serialization::Reader srt{ar};
+
+    srt.Read(this, sizeof(*this));
+
+    if (walker_func_size) {
+        walker_func = RegisterWalkerCode(ar.CurrPtr(), walker_func_size);
+        ar.Advance(walker_func_size);
+    }
+
+    return true;
+}
+
+void StageSpecialization::Serialize(Serialization::Archive& ar) const {
+    Serialization::Writer spec{ar};
+
+    spec.Write(start);
+    spec.Write(runtime_info);
+
+    spec.Write(bitset.to_string());
+
+    if (fetch_shader_data) {
+        spec.Write(sizeof(*fetch_shader_data));
+        fetch_shader_data->Serialize(ar);
+    } else {
+        spec.Write(size_t{0});
+    }
+
+    spec.Write(vs_attribs);
+    spec.Write(buffers);
+    spec.Write(images);
+    spec.Write(fmasks);
+    spec.Write(samplers);
+}
+
+bool StageSpecialization::Deserialize(Serialization::Archive& ar) {
+    Serialization::Reader spec{ar};
+
+    spec.Read(start);
+    spec.Read(runtime_info);
+
+    std::string bits{};
+    spec.Read(bits);
+    bitset = std::bitset<MaxStageResources>(bits);
+
+    u64 fetch_data_size{};
+    spec.Read(fetch_data_size);
+
+    if (fetch_data_size) {
+        Gcn::FetchShaderData fetch_data;
+        fetch_data.Deserialize(ar);
+        fetch_shader_data = fetch_data;
+    }
+
+    spec.Read(vs_attribs);
+    spec.Read(buffers);
+    spec.Read(images);
+    spec.Read(fmasks);
+    spec.Read(samplers);
+
+    return true;
+}
+
+} // namespace Shader

src/video_core/renderer_vulkan/vk_pipeline_serialization.h

@@ -0,0 +1,21 @@
+// SPDX-FileCopyrightText: Copyright 2025 shadPS4 Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "shader_recompiler/frontend/fetch_shader.h"
+#include "video_core/renderer_vulkan/vk_pipeline_cache.h"
+#include "video_core/renderer_vulkan/vk_shader_util.h"
+
+namespace Vulkan {
+
+void RegisterPipelineData(const ComputePipelineKey& key,
+                          ComputePipeline::SerializationSupport& sdata);
+void RegisterPipelineData(const GraphicsPipelineKey& key, u64 hash,
+                          GraphicsPipeline::SerializationSupport& sdata);
+void RegisterShaderMeta(const Shader::Info& info,
+                        const std::optional<Shader::Gcn::FetchShaderData>& fetch_shader_data,
+                        const Shader::StageSpecialization& spec, size_t perm_hash, size_t perm_idx);
+void RegisterShaderBinary(std::vector<u32>&& spv, u64 pgm_hash, size_t perm_idx);
+
+} // namespace Vulkan

src/video_core/renderer_vulkan/vk_rasterizer.cpp

@@ -983,8 +983,8 @@ void Rasterizer::DepthStencilCopy(bool is_depth, bool is_stencil) {
     ScopeMarkerEnd();
 }
 
-void Rasterizer::InlineData(VAddr address, const void* value, u32 num_bytes, bool is_gds) {
+void Rasterizer::FillBuffer(VAddr address, u32 num_bytes, u32 value, bool is_gds) {
-    buffer_cache.InlineData(address, value, num_bytes, is_gds);
+    buffer_cache.FillBuffer(address, num_bytes, value, is_gds);
 }
 
 void Rasterizer::CopyBuffer(VAddr dst, VAddr src, u32 num_bytes, bool dst_gds, bool src_gds) {

src/video_core/renderer_vulkan/vk_rasterizer.h

@@ -55,7 +55,7 @@ public:
     void ScopedMarkerInsertColor(const std::string_view& str, const u32 color,
                                  bool from_guest = false);
 
-    void InlineData(VAddr address, const void* value, u32 num_bytes, bool is_gds);
+    void FillBuffer(VAddr address, u32 num_bytes, u32 value, bool is_gds);
     void CopyBuffer(VAddr dst, VAddr src, u32 num_bytes, bool dst_gds, bool src_gds);
     u32 ReadDataFromGds(u32 gsd_offset);
     bool InvalidateMemory(VAddr addr, u64 size);

src/video_core/renderer_vulkan/vk_scheduler.cpp

@@ -3,6 +3,7 @@
 
 #include "common/assert.h"
 #include "common/debug.h"
+#include "common/thread.h"
 #include "imgui/renderer/texture_manager.h"
 #include "video_core/renderer_vulkan/vk_instance.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
@@ -17,6 +18,8 @@ Scheduler::Scheduler(const Instance& instance)
     profiler_scope = reinterpret_cast<tracy::VkCtxScope*>(std::malloc(sizeof(tracy::VkCtxScope)));
 #endif
     AllocateWorkerCommandBuffers();
+    priority_pending_ops_thread =
+        std::jthread(std::bind_front(&Scheduler::PriorityPendingOpsThread, this));
 }
 
 Scheduler::~Scheduler() {
@@ -167,6 +170,32 @@ void Scheduler::SubmitExecution(SubmitInfo& info) {
     PopPendingOperations();
 }
 
+void Scheduler::PriorityPendingOpsThread(std::stop_token stoken) {
+    Common::SetCurrentThreadName("shadPS4:GpuSchedPriorityPendingOpsRunner");
+
+    while (!stoken.stop_requested()) {
+        PendingOp op;
+        {
+            std::unique_lock lk(priority_pending_ops_mutex);
+            priority_pending_ops_cv.wait(lk, stoken,
+                                         [this] { return !priority_pending_ops.empty(); });
+            if (stoken.stop_requested()) {
+                break;
+            }
+
+            op = std::move(priority_pending_ops.front());
+            priority_pending_ops.pop();
+        }
+
+        master_semaphore.Wait(op.gpu_tick);
+        if (stoken.stop_requested()) {
+            break;
+        }
+
+        op.callback();
+    }
+}
+
 void DynamicState::Commit(const Instance& instance, const vk::CommandBuffer& cmdbuf) {
     if (dirty_state.viewports) {
         dirty_state.viewports = false;
@@ -290,9 +319,7 @@ void DynamicState::Commit(const Instance& instance, const vk::CommandBuffer& cmd
     }
     if (dirty_state.primitive_restart_enable) {
         dirty_state.primitive_restart_enable = false;
-        if (instance.IsPrimitiveRestartDisableSupported()) {
+        cmdbuf.setPrimitiveRestartEnable(primitive_restart_enable);
-            cmdbuf.setPrimitiveRestartEnable(primitive_restart_enable);
-        }
     }
     if (dirty_state.rasterizer_discard_enable) {
         dirty_state.rasterizer_discard_enable = false;

src/video_core/renderer_vulkan/vk_scheduler.h

@@ -5,6 +5,7 @@
 
 #include <condition_variable>
 #include <mutex>
+#include <thread>
 #include <queue>
 
 #include "common/unique_function.h"
@@ -401,10 +402,21 @@ public:
     }
 
     /// Defers an operation until the gpu has reached the current cpu tick.
+    /// Will be run when submitting or calling PopPendingOperations.
     void DeferOperation(Common::UniqueFunction<void>&& func) {
         pending_ops.emplace(std::move(func), CurrentTick());
     }
 
+    /// Defers an operation until the gpu has reached the current cpu tick.
+    /// Runs as soon as possible in another thread.
+    void DeferPriorityOperation(Common::UniqueFunction<void>&& func) {
+        {
+            std::unique_lock lk(priority_pending_ops_mutex);
+            priority_pending_ops.emplace(std::move(func), CurrentTick());
+        }
+        priority_pending_ops_cv.notify_one();
+    }
+
     static std::mutex submit_mutex;
 
 private:
@@ -412,6 +424,8 @@ private:
 
     void SubmitExecution(SubmitInfo& info);
 
+    void PriorityPendingOpsThread(std::stop_token stoken);
+
 private:
     const Instance& instance;
     MasterSemaphore master_semaphore;
@@ -424,6 +438,10 @@ private:
         u64 gpu_tick;
     };
     std::queue<PendingOp> pending_ops;
+    std::queue<PendingOp> priority_pending_ops;
+    std::mutex priority_pending_ops_mutex;
+    std::condition_variable_any priority_pending_ops_cv;
+    std::jthread priority_pending_ops_thread;
     RenderState render_state;
     bool is_rendering = false;
     tracy::VkCtxScope* profiler_scope{};

src/video_core/texture_cache/texture_cache.cpp

@@ -52,9 +52,6 @@ TextureCache::TextureCache(const Vulkan::Instance& instance_, Vulkan::Scheduler&
         std::max<u64>(std::min(device_local_memory - min_vacancy_critical, min_spacing_critical),
                       DEFAULT_CRITICAL_GC_MEMORY));
     trigger_gc_memory = static_cast<u64>((device_local_memory - mem_threshold) / 2);
-
-    downloaded_images_thread =
-        std::jthread([&](const std::stop_token& token) { DownloadedImagesThread(token); });
 }
 
 TextureCache::~TextureCache() = default;
@@ -88,11 +85,12 @@ ImageId TextureCache::GetNullImage(const vk::Format format) {
 
 void TextureCache::ProcessDownloadImages() {
     for (const ImageId image_id : download_images) {
-        DownloadImageMemory(image_id);
+        DownloadImageMemory<true>(image_id);
     }
     download_images.clear();
 }
 
+template <bool priority>
 void TextureCache::DownloadImageMemory(ImageId image_id) {
     Image& image = slot_images[image_id];
     if (False(image.flags & ImageFlagBits::GpuModified)) {
@@ -136,33 +134,20 @@ void TextureCache::DownloadImageMemory(ImageId image_id) {
     image.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead, {});
     tile_manager.TileImage(image, buffer_copies, mapping.Buffer()->Handle(), mapping.Offset(),
                            copy_size);
-    {
-        std::unique_lock lock(downloaded_images_mutex);
-        downloaded_images_queue.emplace(scheduler.CurrentTick(), image_addr, mapping.Data(),
-                                        image_size);
-        downloaded_images_cv.notify_one();
-    }
-}
 
-void TextureCache::DownloadedImagesThread(const std::stop_token& token) {
+    const auto operation = [this, device_addr = image.info.guest_address, download = mapping.Data(),
-    auto* memory = Core::Memory::Instance();
+                            image_size] {
-    while (!token.stop_requested()) {
+        Core::Memory::Instance()->TryWriteBacking(std::bit_cast<u8*>(device_addr), download,
-        DownloadedImage image;
+                                                  image_size);
-        {
+        if constexpr (!priority) {
-            std::unique_lock lock{downloaded_images_mutex};
+            buffer_cache.InvalidateMemory(device_addr, image_size, false);
-            downloaded_images_cv.wait(lock, token,
-                                      [this] { return !downloaded_images_queue.empty(); });
-            if (token.stop_requested()) {
-                break;
-            }
-            image = downloaded_images_queue.front();
-            downloaded_images_queue.pop();
         }
+    };
 
-        scheduler.GetMasterSemaphore()->Wait(image.tick);
+    if constexpr (priority) {
-        memory->TryWriteBacking(std::bit_cast<u8*>(image.device_addr), image.download,
+        scheduler.DeferPriorityOperation(operation);
-                                image.download_size);
+    } else {
-        buffer_cache.InvalidateMemory(image.device_addr, image.download_size, false);
+        scheduler.DeferOperation(operation);
     }
 }
 
@@ -967,7 +952,7 @@ void TextureCache::RunGarbageCollector() {
     }
 
     for (const auto& image_id : download_pending) {
-        DownloadImageMemory(image_id);
+        DownloadImageMemory<false>(image_id);
         DeleteImage(image_id);
     }
 
@@ -976,6 +961,7 @@ void TextureCache::RunGarbageCollector() {
         // of the image are requested before they are downloaded in which case
         // outdated buffer cache contents are used instead.
         scheduler.Finish();
+        scheduler.PopPendingOperations();
     }
 }
 

src/video_core/texture_cache/texture_cache.h

@@ -318,16 +318,6 @@ private:
     Common::LeastRecentlyUsedCache<ImageId, u64> lru_cache;
     PageTable page_table;
     std::mutex mutex;
-    struct DownloadedImage {
-        u64 tick;
-        VAddr device_addr;
-        void* download;
-        size_t download_size;
-    };
-    std::queue<DownloadedImage> downloaded_images_queue;
-    std::mutex downloaded_images_mutex;
-    std::condition_variable_any downloaded_images_cv;
-    std::jthread downloaded_images_thread;
     struct MetaDataInfo {
         enum class Type {
             CMask,