diff --git a/.gitmodules b/.gitmodules index 065a4570f..25b5d307b 100644 --- a/.gitmodules +++ b/.gitmodules @@ -30,10 +30,6 @@ path = externals/xbyak url = https://github.com/herumi/xbyak.git shallow = true -[submodule "externals/winpthreads"] - path = externals/winpthreads - url = https://github.com/shadps4-emu/winpthreads.git - shallow = true [submodule "externals/magic_enum"] path = externals/magic_enum url = https://github.com/Neargye/magic_enum.git diff --git a/CMakeLists.txt b/CMakeLists.txt index 3b67bbe74..f989c4dab 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -54,9 +54,9 @@ else() endif() if (ARCHITECTURE STREQUAL "x86_64") - # Target the same CPU architecture as the PS4, to maintain the same level of compatibility. - # Exclude SSE4a as it is only available on AMD CPUs. - add_compile_options(-march=btver2 -mtune=generic -mno-sse4a) + # Target x86-64-v3 CPU architecture as this is a good balance between supporting performance critical + # instructions like AVX2 and maintaining support for older CPUs. + add_compile_options(-march=x86-64-v3) endif() if (APPLE AND ARCHITECTURE STREQUAL "x86_64" AND CMAKE_HOST_SYSTEM_PROCESSOR STREQUAL "arm64") @@ -134,6 +134,7 @@ if (GIT_REMOTE_RESULT OR GIT_REMOTE_NAME STREQUAL "") ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE ) + message("got remote: ${GIT_REMOTE_NAME}") endif() # If running in GitHub Actions and the above fails @@ -177,7 +178,7 @@ if (GIT_REMOTE_RESULT OR GIT_REMOTE_NAME STREQUAL "") set(GIT_BRANCH "${GITHUB_BRANCH}") elseif ("${PR_NUMBER}" STREQUAL "" AND NOT "${GITHUB_REF}" STREQUAL "") set(GIT_BRANCH "${GITHUB_REF}") - else() + elseif("${GIT_BRANCH}" STREQUAL "") message("couldn't find branch") set(GIT_BRANCH "detached-head") endif() @@ -186,8 +187,8 @@ else() string(FIND "${GIT_REMOTE_NAME}" "/" INDEX) if (INDEX GREATER -1) string(SUBSTRING "${GIT_REMOTE_NAME}" 0 "${INDEX}" GIT_REMOTE_NAME) - else() - # If no remote is present (only a branch name), default to origin + elseif("${GIT_REMOTE_NAME}" STREQUAL "") + message("reset to origin") set(GIT_REMOTE_NAME "origin") endif() endif() @@ -202,7 +203,7 @@ execute_process( # Set Version set(EMULATOR_VERSION_MAJOR "0") -set(EMULATOR_VERSION_MINOR "8") +set(EMULATOR_VERSION_MINOR "9") set(EMULATOR_VERSION_PATCH "1") set_source_files_properties(src/shadps4.rc PROPERTIES COMPILE_DEFINITIONS "EMULATOR_VERSION_MAJOR=${EMULATOR_VERSION_MAJOR};EMULATOR_VERSION_MINOR=${EMULATOR_VERSION_MINOR};EMULATOR_VERSION_PATCH=${EMULATOR_VERSION_PATCH}") @@ -226,7 +227,7 @@ find_package(SDL3 3.1.2 CONFIG) find_package(stb MODULE) find_package(toml11 4.2.0 CONFIG) find_package(tsl-robin-map 1.3.0 CONFIG) -find_package(VulkanHeaders 1.4.309 CONFIG) +find_package(VulkanHeaders 1.4.314 CONFIG) find_package(VulkanMemoryAllocator 3.1.0 CONFIG) find_package(xbyak 7.07 CONFIG) find_package(xxHash 0.8.2 MODULE) @@ -239,13 +240,6 @@ if (APPLE) endif() list(POP_BACK CMAKE_MODULE_PATH) -# Note: Windows always has these functions through winpthreads -include(CheckSymbolExists) -check_symbol_exists(pthread_mutex_timedlock "pthread.h" HAVE_PTHREAD_MUTEX_TIMEDLOCK) -if(HAVE_PTHREAD_MUTEX_TIMEDLOCK OR WIN32) - add_compile_options(-DHAVE_PTHREAD_MUTEX_TIMEDLOCK) -endif() - if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "AppleClang") # libc++ requires -fexperimental-library to enable std::jthread and std::stop_token support. include(CheckCXXSymbolExists) @@ -416,6 +410,7 @@ set(SYSTEM_LIBS src/core/libraries/system/commondialog.cpp src/core/libraries/save_data/save_memory.h src/core/libraries/save_data/savedata.cpp src/core/libraries/save_data/savedata.h + src/core/libraries/save_data/savedata_error.h src/core/libraries/save_data/dialog/savedatadialog.cpp src/core/libraries/save_data/dialog/savedatadialog.h src/core/libraries/save_data/dialog/savedatadialog_ui.cpp @@ -610,6 +605,10 @@ set(CAMERA_LIBS src/core/libraries/camera/camera.cpp src/core/libraries/camera/camera_error.h ) +set(COMPANION_LIBS src/core/libraries/companion/companion_httpd.cpp + src/core/libraries/companion/companion_httpd.h + src/core/libraries/companion/companion_error.h +) set(DEV_TOOLS src/core/devtools/layer.cpp src/core/devtools/layer.h src/core/devtools/options.cpp @@ -627,6 +626,8 @@ set(DEV_TOOLS src/core/devtools/layer.cpp src/core/devtools/widget/imgui_memory_editor.h src/core/devtools/widget/memory_map.cpp src/core/devtools/widget/memory_map.h + src/core/devtools/widget/module_list.cpp + src/core/devtools/widget/module_list.h src/core/devtools/widget/reg_popup.cpp src/core/devtools/widget/reg_popup.h src/core/devtools/widget/reg_view.cpp @@ -679,6 +680,8 @@ set(COMMON src/common/logging/backend.cpp src/common/polyfill_thread.h src/common/rdtsc.cpp src/common/rdtsc.h + src/common/recursive_lock.cpp + src/common/recursive_lock.h src/common/sha1.h src/common/signal_context.h src/common/signal_context.cpp @@ -774,6 +777,7 @@ set(CORE src/core/aerolib/stubs.cpp ${VDEC_LIB} ${VR_LIBS} ${CAMERA_LIBS} + ${COMPANION_LIBS} ${DEV_TOOLS} src/core/debug_state.cpp src/core/debug_state.h @@ -868,6 +872,7 @@ set(SHADER_RECOMPILER src/shader_recompiler/exception.h src/shader_recompiler/ir/passes/shared_memory_barrier_pass.cpp src/shader_recompiler/ir/passes/shared_memory_to_storage_pass.cpp src/shader_recompiler/ir/passes/ssa_rewrite_pass.cpp + src/shader_recompiler/ir/abstract_syntax_list.cpp src/shader_recompiler/ir/abstract_syntax_list.h src/shader_recompiler/ir/attribute.cpp src/shader_recompiler/ir/attribute.h @@ -1160,7 +1165,7 @@ if (ENABLE_QT_GUI) endif() if (WIN32) - target_link_libraries(shadps4 PRIVATE mincore winpthreads) + target_link_libraries(shadps4 PRIVATE mincore) if (MSVC) # MSVC likes putting opinions on what people can use, disable: diff --git a/dist/net.shadps4.shadPS4.metainfo.xml b/dist/net.shadps4.shadPS4.metainfo.xml index 9f7b4f9c5..493dc0df6 100644 --- a/dist/net.shadps4.shadPS4.metainfo.xml +++ b/dist/net.shadps4.shadPS4.metainfo.xml @@ -37,7 +37,10 @@ Game - + + https://github.com/shadps4-emu/shadPS4/releases/tag/v.0.9.0 + + https://github.com/shadps4-emu/shadPS4/releases/tag/v.0.8.0 diff --git a/documents/Quickstart/Quickstart.md b/documents/Quickstart/Quickstart.md index 62df95e71..e2145ebbd 100644 --- a/documents/Quickstart/Quickstart.md +++ b/documents/Quickstart/Quickstart.md @@ -21,9 +21,9 @@ SPDX-License-Identifier: GPL-2.0-or-later - A processor with at least 4 cores and 6 threads - Above 2.5 GHz frequency -- A CPU supporting the following instruction sets: MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, AVX, F16C, CLMUL, AES, BMI1, MOVBE, XSAVE, ABM +- A CPU supporting the x86-64-v3 baseline. - **Intel**: Haswell generation or newer - - **AMD**: Jaguar generation or newer + - **AMD**: Excavator generation or newer - **Apple**: Rosetta 2 on macOS 15.4 or newer ### GPU @@ -55,4 +55,4 @@ To configure the emulator, you can go through the interface and go to "settings" You can also configure the emulator by editing the `config.toml` file located in the `user` folder created after the application is started (Mostly useful if you are using the SDL version). Some settings may be related to more technical development and debugging.\ -For more information on this, see [**Debugging**](https://github.com/shadps4-emu/shadPS4/blob/main/documents/Debugging/Debugging.md#configuration). \ No newline at end of file +For more information on this, see [**Debugging**](https://github.com/shadps4-emu/shadPS4/blob/main/documents/Debugging/Debugging.md#configuration). diff --git a/externals/CMakeLists.txt b/externals/CMakeLists.txt index b92e13795..89b0fbfdd 100644 --- a/externals/CMakeLists.txt +++ b/externals/CMakeLists.txt @@ -137,12 +137,6 @@ if (NOT TARGET Zydis::Zydis) add_subdirectory(zydis) endif() -# Winpthreads -if (WIN32) - add_subdirectory(winpthreads) - target_include_directories(winpthreads INTERFACE winpthreads/include) -endif() - # sirit add_subdirectory(sirit) if (WIN32) diff --git a/externals/MoltenVK/MoltenVK b/externals/MoltenVK/MoltenVK index 87a8e8b13..3a0b07a24 160000 --- a/externals/MoltenVK/MoltenVK +++ b/externals/MoltenVK/MoltenVK @@ -1 +1 @@ -Subproject commit 87a8e8b13d4ad8835367fea1ebad1896d0460946 +Subproject commit 3a0b07a24a4a681ffe70b461b1f4333b2729e2ef diff --git a/externals/MoltenVK/SPIRV-Cross b/externals/MoltenVK/SPIRV-Cross index 791877574..969e75f7c 160000 --- a/externals/MoltenVK/SPIRV-Cross +++ b/externals/MoltenVK/SPIRV-Cross @@ -1 +1 @@ -Subproject commit 7918775748c5e2f5c40d9918ce68825035b5a1e1 +Subproject commit 969e75f7cc0718774231d029f9d52fa87d4ae1b2 diff --git a/externals/sirit b/externals/sirit index 09a1416ab..6b450704f 160000 --- a/externals/sirit +++ b/externals/sirit @@ -1 +1 @@ -Subproject commit 09a1416ab1b59ddfebd2618412f118f2004f3b2c +Subproject commit 6b450704f6fedb9413d0c89a9eb59d028eb1e6c0 diff --git a/externals/vulkan-headers b/externals/vulkan-headers index 5ceb9ed48..9c77de5c3 160000 --- a/externals/vulkan-headers +++ b/externals/vulkan-headers @@ -1 +1 @@ -Subproject commit 5ceb9ed481e58e705d0d9b5326537daedd06b97d +Subproject commit 9c77de5c3dd216f28e407eec65ed9c0a296c1f74 diff --git a/externals/winpthreads b/externals/winpthreads deleted file mode 160000 index f35b0948d..000000000 --- a/externals/winpthreads +++ /dev/null @@ -1 +0,0 @@ -Subproject commit f35b0948d36a736e6a2d052ae295a3ffde09703f diff --git a/src/common/config.cpp b/src/common/config.cpp index 111c0cfa9..6bccd0f37 100644 --- a/src/common/config.cpp +++ b/src/common/config.cpp @@ -154,7 +154,7 @@ bool GetLoadGameSizeEnabled() { std::filesystem::path GetSaveDataPath() { if (save_data_path.empty()) { - return Common::FS::GetUserPath(Common::FS::PathType::SaveDataDir); + return Common::FS::GetUserPath(Common::FS::PathType::UserDir) / "savedata"; } return save_data_path; } diff --git a/src/common/elf_info.h b/src/common/elf_info.h index 062cee012..02b845cb5 100644 --- a/src/common/elf_info.h +++ b/src/common/elf_info.h @@ -71,6 +71,7 @@ class ElfInfo { PSFAttributes psf_attributes{}; std::filesystem::path splash_path{}; + std::filesystem::path game_folder{}; public: static constexpr u32 FW_15 = 0x1500000; @@ -123,6 +124,10 @@ public: [[nodiscard]] const std::filesystem::path& GetSplashPath() const { return splash_path; } + + [[nodiscard]] const std::filesystem::path& GetGameFolder() const { + return game_folder; + } }; } // namespace Common diff --git a/src/common/logging/filter.cpp b/src/common/logging/filter.cpp index 1b605e9ed..231cbf849 100644 --- a/src/common/logging/filter.cpp +++ b/src/common/logging/filter.cpp @@ -139,6 +139,7 @@ bool ParseFilterRule(Filter& instance, Iterator begin, Iterator end) { SUB(Lib, Hmd) \ SUB(Lib, SigninDialog) \ SUB(Lib, Camera) \ + SUB(Lib, CompanionHttpd) \ CLS(Frontend) \ CLS(Render) \ SUB(Render, Vulkan) \ diff --git a/src/common/logging/types.h b/src/common/logging/types.h index 5746b648e..e4eae59af 100644 --- a/src/common/logging/types.h +++ b/src/common/logging/types.h @@ -106,6 +106,7 @@ enum class Class : u8 { Lib_Hmd, ///< The LibSceHmd implementation. Lib_SigninDialog, ///< The LibSigninDialog implementation. Lib_Camera, ///< The LibCamera implementation. + Lib_CompanionHttpd, ///< The LibCompanionHttpd implementation. Frontend, ///< Emulator UI Render, ///< Video Core Render_Vulkan, ///< Vulkan backend diff --git a/src/common/path_util.cpp b/src/common/path_util.cpp index 1a6ff9ec8..3270c24dd 100644 --- a/src/common/path_util.cpp +++ b/src/common/path_util.cpp @@ -128,7 +128,6 @@ static auto UserPaths = [] { create_path(PathType::LogDir, user_dir / LOG_DIR); create_path(PathType::ScreenshotsDir, user_dir / SCREENSHOTS_DIR); create_path(PathType::ShaderDir, user_dir / SHADER_DIR); - create_path(PathType::SaveDataDir, user_dir / SAVEDATA_DIR); create_path(PathType::GameDataDir, user_dir / GAMEDATA_DIR); create_path(PathType::TempDataDir, user_dir / TEMPDATA_DIR); create_path(PathType::SysModuleDir, user_dir / SYSMODULES_DIR); diff --git a/src/common/path_util.h b/src/common/path_util.h index 2fd9b1588..b8053a229 100644 --- a/src/common/path_util.h +++ b/src/common/path_util.h @@ -18,7 +18,6 @@ enum class PathType { LogDir, // Where log files are stored. ScreenshotsDir, // Where screenshots are stored. ShaderDir, // Where shaders are stored. - SaveDataDir, // Where guest save data is stored. TempDataDir, // Where game temp data is stored. GameDataDir, // Where game data is stored. SysModuleDir, // Where system modules are stored. @@ -36,7 +35,6 @@ constexpr auto PORTABLE_DIR = "user"; constexpr auto LOG_DIR = "log"; constexpr auto SCREENSHOTS_DIR = "screenshots"; constexpr auto SHADER_DIR = "shader"; -constexpr auto SAVEDATA_DIR = "savedata"; constexpr auto GAMEDATA_DIR = "data"; constexpr auto TEMPDATA_DIR = "temp"; constexpr auto SYSMODULES_DIR = "sys_modules"; diff --git a/src/common/recursive_lock.cpp b/src/common/recursive_lock.cpp new file mode 100644 index 000000000..2471a2ee0 --- /dev/null +++ b/src/common/recursive_lock.cpp @@ -0,0 +1,37 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#include +#include "common/assert.h" +#include "common/recursive_lock.h" + +namespace Common::Detail { + +struct RecursiveLockState { + RecursiveLockType type; + int count; +}; + +thread_local std::unordered_map g_recursive_locks; + +bool IncrementRecursiveLock(void* mutex, RecursiveLockType type) { + auto& state = g_recursive_locks[mutex]; + if (state.count == 0) { + ASSERT(state.type == RecursiveLockType::None); + state.type = type; + } + ASSERT(state.type == type); + return state.count++ == 0; +} + +bool DecrementRecursiveLock(void* mutex, RecursiveLockType type) { + auto& state = g_recursive_locks[mutex]; + ASSERT(state.type == type && state.count > 0); + if (--state.count == 0) { + g_recursive_locks.erase(mutex); + return true; + } + return false; +} + +} // namespace Common::Detail diff --git a/src/common/recursive_lock.h b/src/common/recursive_lock.h new file mode 100644 index 000000000..5a5fc6658 --- /dev/null +++ b/src/common/recursive_lock.h @@ -0,0 +1,67 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#pragma once + +#include +#include +#include + +namespace Common { + +namespace Detail { + +enum class RecursiveLockType { None, Shared, Exclusive }; + +bool IncrementRecursiveLock(void* mutex, RecursiveLockType type); +bool DecrementRecursiveLock(void* mutex, RecursiveLockType type); + +} // namespace Detail + +template +class RecursiveScopedLock { +public: + explicit RecursiveScopedLock(MutexType& mutex) : m_mutex(mutex), m_locked(false) { + if (Detail::IncrementRecursiveLock(&m_mutex, Detail::RecursiveLockType::Exclusive)) { + m_locked = true; + m_lock.emplace(m_mutex); + } + } + + ~RecursiveScopedLock() { + Detail::DecrementRecursiveLock(&m_mutex, Detail::RecursiveLockType::Exclusive); + if (m_locked) { + m_lock.reset(); + } + } + +private: + MutexType& m_mutex; + std::optional> m_lock; + bool m_locked = false; +}; + +template +class RecursiveSharedLock { +public: + explicit RecursiveSharedLock(MutexType& mutex) : m_mutex(mutex), m_locked(false) { + if (Detail::IncrementRecursiveLock(&m_mutex, Detail::RecursiveLockType::Shared)) { + m_locked = true; + m_lock.emplace(m_mutex); + } + } + + ~RecursiveSharedLock() { + Detail::DecrementRecursiveLock(&m_mutex, Detail::RecursiveLockType::Shared); + if (m_locked) { + m_lock.reset(); + } + } + +private: + MutexType& m_mutex; + std::optional> m_lock; + bool m_locked = false; +}; + +} // namespace Common \ No newline at end of file diff --git a/src/common/slot_vector.h b/src/common/slot_vector.h index d4ac51361..2f693fb28 100644 --- a/src/common/slot_vector.h +++ b/src/common/slot_vector.h @@ -14,6 +14,9 @@ namespace Common { struct SlotId { static constexpr u32 INVALID_INDEX = std::numeric_limits::max(); + SlotId() noexcept = default; + constexpr SlotId(u32 index) noexcept : index(index) {} + constexpr auto operator<=>(const SlotId&) const noexcept = default; constexpr explicit operator bool() const noexcept { @@ -28,6 +31,63 @@ class SlotVector { constexpr static std::size_t InitialCapacity = 2048; public: + template + class Iterator { + public: + using iterator_category = std::forward_iterator_tag; + using value_type = ValueType; + using difference_type = std::ptrdiff_t; + using pointer = Pointer; + using reference = Reference; + + Iterator(SlotVector& vector_, SlotId index_) : vector(vector_), slot(index_) { + AdvanceToValid(); + } + + reference operator*() const { + return vector[slot]; + } + + pointer operator->() const { + return &vector[slot]; + } + + Iterator& operator++() { + ++slot.index; + AdvanceToValid(); + return *this; + } + + Iterator operator++(int) { + Iterator temp = *this; + ++(*this); + return temp; + } + + bool operator==(const Iterator& other) const { + return slot == other.slot; + } + + bool operator!=(const Iterator& other) const { + return !(*this == other); + } + + private: + void AdvanceToValid() { + while (slot < vector.values_capacity && !vector.ReadStorageBit(slot.index)) { + ++slot.index; + } + } + + SlotVector& vector; + SlotId slot; + }; + + using iterator = Iterator; + using const_iterator = Iterator; + using reverse_iterator = std::reverse_iterator; + using const_reverse_iterator = std::reverse_iterator; + SlotVector() { Reserve(InitialCapacity); } @@ -60,7 +120,7 @@ public: } template - [[nodiscard]] SlotId insert(Args&&... args) noexcept { + SlotId insert(Args&&... args) noexcept { const u32 index = FreeValueIndex(); new (&values[index].object) T(std::forward(args)...); SetStorageBit(index); @@ -78,6 +138,54 @@ public: return values_capacity - free_list.size(); } + iterator begin() noexcept { + return iterator(*this, 0); + } + + const_iterator begin() const noexcept { + return const_iterator(*this, 0); + } + + const_iterator cbegin() const noexcept { + return begin(); + } + + iterator end() noexcept { + return iterator(*this, values_capacity); + } + + const_iterator end() const noexcept { + return const_iterator(*this, values_capacity); + } + + const_iterator cend() const noexcept { + return end(); + } + + reverse_iterator rbegin() noexcept { + return reverse_iterator(end()); + } + + const_reverse_iterator rbegin() const noexcept { + return const_reverse_iterator(end()); + } + + const_reverse_iterator crbegin() const noexcept { + return rbegin(); + } + + reverse_iterator rend() noexcept { + return reverse_iterator(begin()); + } + + const_reverse_iterator rend() const noexcept { + return const_reverse_iterator(begin()); + } + + const_reverse_iterator crend() const noexcept { + return rend(); + } + private: struct NonTrivialDummy { NonTrivialDummy() noexcept {} diff --git a/src/common/thread.cpp b/src/common/thread.cpp index 9ef1e86d8..982041ebb 100644 --- a/src/common/thread.cpp +++ b/src/common/thread.cpp @@ -2,6 +2,7 @@ // SPDX-FileCopyrightText: 2014 Citra Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later +#include #include #include @@ -104,14 +105,24 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) { SetThreadPriority(handle, windows_priority); } -static void AccurateSleep(std::chrono::nanoseconds duration) { +bool AccurateSleep(const std::chrono::nanoseconds duration, std::chrono::nanoseconds* remaining, + const bool interruptible) { + const auto begin_sleep = std::chrono::high_resolution_clock::now(); + LARGE_INTEGER interval{ .QuadPart = -1 * (duration.count() / 100u), }; HANDLE timer = ::CreateWaitableTimer(NULL, TRUE, NULL); SetWaitableTimer(timer, &interval, 0, NULL, NULL, 0); - WaitForSingleObject(timer, INFINITE); + const auto ret = WaitForSingleObjectEx(timer, INFINITE, interruptible); ::CloseHandle(timer); + + if (remaining) { + const auto end_sleep = std::chrono::high_resolution_clock::now(); + const auto sleep_time = end_sleep - begin_sleep; + *remaining = duration > sleep_time ? duration - sleep_time : std::chrono::nanoseconds(0); + } + return ret == WAIT_OBJECT_0; } #else @@ -134,8 +145,24 @@ void SetCurrentThreadPriority(ThreadPriority new_priority) { pthread_setschedparam(this_thread, scheduling_type, ¶ms); } -static void AccurateSleep(std::chrono::nanoseconds duration) { - std::this_thread::sleep_for(duration); +bool AccurateSleep(const std::chrono::nanoseconds duration, std::chrono::nanoseconds* remaining, + const bool interruptible) { + timespec request = { + .tv_sec = duration.count() / 1'000'000'000, + .tv_nsec = duration.count() % 1'000'000'000, + }; + timespec remain; + int ret; + while ((ret = nanosleep(&request, &remain)) < 0 && errno == EINTR) { + if (interruptible) { + break; + } + request = remain; + } + if (remaining) { + *remaining = std::chrono::nanoseconds(remain.tv_sec * 1'000'000'000 + remain.tv_nsec); + } + return ret == 0 || errno != EINTR; } #endif @@ -196,9 +223,9 @@ AccurateTimer::AccurateTimer(std::chrono::nanoseconds target_interval) : target_interval(target_interval) {} void AccurateTimer::Start() { - auto begin_sleep = std::chrono::high_resolution_clock::now(); + const auto begin_sleep = std::chrono::high_resolution_clock::now(); if (total_wait.count() > 0) { - AccurateSleep(total_wait); + AccurateSleep(total_wait, nullptr, false); } start_time = std::chrono::high_resolution_clock::now(); total_wait -= std::chrono::duration_cast(start_time - begin_sleep); diff --git a/src/common/thread.h b/src/common/thread.h index 92cc0c59d..5bd83d35c 100644 --- a/src/common/thread.h +++ b/src/common/thread.h @@ -25,6 +25,9 @@ void SetCurrentThreadName(const char* name); void SetThreadName(void* thread, const char* name); +bool AccurateSleep(std::chrono::nanoseconds duration, std::chrono::nanoseconds* remaining, + bool interruptible); + class AccurateTimer { std::chrono::nanoseconds target_interval{}; std::chrono::nanoseconds total_wait{}; diff --git a/src/core/devtools/layer.cpp b/src/core/devtools/layer.cpp index a93178de5..5380d3be9 100644 --- a/src/core/devtools/layer.cpp +++ b/src/core/devtools/layer.cpp @@ -17,6 +17,7 @@ #include "widget/frame_dump.h" #include "widget/frame_graph.h" #include "widget/memory_map.h" +#include "widget/module_list.h" #include "widget/shader_list.h" extern std::unique_ptr presenter; @@ -40,6 +41,7 @@ static bool just_opened_options = false; static Widget::MemoryMapViewer memory_map; static Widget::ShaderList shader_list; +static Widget::ModuleList module_list; // clang-format off static std::string help_text = @@ -108,6 +110,9 @@ void L::DrawMenuBar() { if (MenuItem("Memory map")) { memory_map.open = true; } + if (MenuItem("Module list")) { + module_list.open = true; + } ImGui::EndMenu(); } @@ -256,6 +261,9 @@ void L::DrawAdvanced() { if (shader_list.open) { shader_list.Draw(); } + if (module_list.open) { + module_list.Draw(); + } } void L::DrawSimple() { diff --git a/src/core/devtools/widget/module_list.cpp b/src/core/devtools/widget/module_list.cpp new file mode 100644 index 000000000..73afe3462 --- /dev/null +++ b/src/core/devtools/widget/module_list.cpp @@ -0,0 +1,55 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#include "module_list.h" + +#include + +#include "common.h" +#include "core/debug_state.h" +#include "imgui/imgui_std.h" + +using namespace ImGui; + +namespace Core::Devtools::Widget { +void ModuleList::Draw() { + SetNextWindowSize({550.0f, 600.0f}, ImGuiCond_FirstUseEver); + if (!Begin("Module List", &open)) { + End(); + return; + } + + if (BeginTable("ModuleTable", 3, + ImGuiTableFlags_Borders | ImGuiTableFlags_Resizable | ImGuiTableFlags_Sortable | + ImGuiTableFlags_RowBg)) { + TableSetupColumn("Modulname", ImGuiTableColumnFlags_WidthStretch); + TableHeadersRow(); + + std::scoped_lock lock(modules_mutex); + for (const auto& module : modules) { + TableNextRow(); + + TableSetColumnIndex(0); + TextUnformatted(module.name.c_str()); + + TableSetColumnIndex(1); + if (module.is_sys_module) { + TextColored({0.2f, 0.6f, 0.8f, 1.0f}, "System Module"); + } else { + TextColored({0.8f, 0.4f, 0.2f, 1.0f}, "Game Module"); + } + + TableSetColumnIndex(2); + if (module.is_lle) { + TextColored({0.4f, 0.7f, 0.4f, 1.0f}, "LLE"); + } else { + TextColored({0.7f, 0.4f, 0.5f, 1.0f}, "HLE"); + } + } + EndTable(); + } + + End(); +} + +} // namespace Core::Devtools::Widget \ No newline at end of file diff --git a/src/core/devtools/widget/module_list.h b/src/core/devtools/widget/module_list.h new file mode 100644 index 000000000..4c961919e --- /dev/null +++ b/src/core/devtools/widget/module_list.h @@ -0,0 +1,82 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#pragma once + +#include +#include +#include +#include +#include +#include "common/elf_info.h" +#include "common/path_util.h" + +namespace Core::Devtools::Widget { + +class ModuleList { +public: + ModuleList() = default; + ~ModuleList() = default; + + void Draw(); + bool open = false; + + static bool IsSystemModule(const std::filesystem::path& path) { + const auto sys_modules_path = Common::FS::GetUserPath(Common::FS::PathType::SysModuleDir); + + const auto abs_path = std::filesystem::absolute(path).lexically_normal(); + const auto abs_sys_path = std::filesystem::absolute(sys_modules_path).lexically_normal(); + + const auto path_str = abs_path.string(); + const auto sys_path_str = abs_sys_path.string(); + + return path_str.starts_with(sys_path_str); + } + + static bool IsSystemModule(const std::string& name) { + const auto game_modules_path = Common::ElfInfo::Instance().GetGameFolder() / "sce_module"; + const auto prx_path = game_modules_path / name; + + if (!std::filesystem::exists(prx_path)) { + return true; + } + return false; + } + + static void AddModule(const std::string& name, std::filesystem::path path) { + if (name == "eboot.bin") { + return; + } + std::scoped_lock lock(modules_mutex); + modules.push_back({name, IsSystemModule(path), true}); + } + + static void AddModule(std::string name) { + name = name + ".prx"; + std::scoped_lock lock(modules_mutex); + + bool is_sys_module = IsSystemModule(name); + bool is_lle = false; + auto it = std::find_if(modules.begin(), modules.end(), + [&name, is_sys_module, is_lle](const ModuleInfo& entry) { + return entry.name == name && !entry.is_lle; + }); + + if (it == modules.end()) { + modules.push_back({name, is_sys_module, is_lle}); + } + } + +private: + struct ModuleInfo { + std::string name; + bool is_sys_module; + bool is_lle; + }; + + static inline std::mutex modules_mutex; + + static inline std::vector modules; +}; + +} // namespace Core::Devtools::Widget \ No newline at end of file diff --git a/src/core/libraries/companion/companion_error.h b/src/core/libraries/companion/companion_error.h new file mode 100644 index 000000000..2d1a3833c --- /dev/null +++ b/src/core/libraries/companion/companion_error.h @@ -0,0 +1,20 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#pragma once + +// companion_httpd error codes +constexpr int ORBIS_COMPANION_HTTPD_ERROR_UNKNOWN = 0x80E40001; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_FATAL = 0x80E40002; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_NOMEM = 0x80E40003; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_INVALID_PARAM = 0x80E40004; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_INVALID_OPERATION = 0x80E40005; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_NOT_INITIALIZED = 0x80E40006; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_ALREADY_INITIALIZED = 0x80E40007; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_NO_EVENT = 0x80E40008; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_NOT_GENERATE_RESPONSE = 0x80E40009; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_ALREADY_STARTED = 0x80E4000A; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_NOT_STARTED = 0x80E4000B; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_ALREADY_REGISTERED = 0x80E4000; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_NOT_CONNECTED = 0x80E4000D; +constexpr int ORBIS_COMPANION_HTTPD_ERROR_USER_NOT_FOUND = 0x80E4000E; diff --git a/src/core/libraries/companion/companion_httpd.cpp b/src/core/libraries/companion/companion_httpd.cpp new file mode 100644 index 000000000..39081fa4e --- /dev/null +++ b/src/core/libraries/companion/companion_httpd.cpp @@ -0,0 +1,142 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#include "common/logging/log.h" +#include "companion_error.h" +#include "core/libraries/companion/companion_httpd.h" +#include "core/libraries/error_codes.h" +#include "core/libraries/libs.h" + +namespace Libraries::CompanionHttpd { + +s32 PS4_SYSV_ABI sceCompanionHttpdAddHeader(const char* key, const char* value, + OrbisCompanionHttpdResponse* response) { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI +sceCompanionHttpdGet2ndScreenStatus(Libraries::UserService::OrbisUserServiceUserId) { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdGetEvent(OrbisCompanionHttpdEvent* pEvent) { + pEvent->event = ORBIS_COMPANION_HTTPD_EVENT_DISCONNECT; // disconnected + LOG_DEBUG(Lib_CompanionHttpd, "device disconnected"); + return ORBIS_COMPANION_HTTPD_ERROR_NO_EVENT; // No events to obtain +} + +s32 PS4_SYSV_ABI +sceCompanionHttpdGetUserId(u32 addr, Libraries::UserService::OrbisUserServiceUserId* userId) { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdInitialize() { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdInitialize2() { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdOptParamInitialize() { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdRegisterRequestBodyReceptionCallback( + OrbisCompanionHttpdRequestBodyReceptionCallback function, void* param) { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI +sceCompanionHttpdRegisterRequestCallback(OrbisCompanionHttpdRequestCallback function, void* param) { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdRegisterRequestCallback2( + OrbisCompanionHttpdRequestCallback function, void* param) { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdSetBody(const char* body, u64 bodySize, + OrbisCompanionHttpdResponse* response) { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdSetStatus(s32 status, OrbisCompanionHttpdResponse* response) { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdStart() { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdStop() { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdTerminate() { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdUnregisterRequestBodyReceptionCallback() { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceCompanionHttpdUnregisterRequestCallback() { + LOG_ERROR(Lib_CompanionHttpd, "(STUBBED) called"); + return ORBIS_OK; +} + +void RegisterlibSceCompanionHttpd(Core::Loader::SymbolsResolver* sym) { + LIB_FUNCTION("8pWltDG7h6A", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdAddHeader); + LIB_FUNCTION("B-QBMeFdNgY", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdGet2ndScreenStatus); + LIB_FUNCTION("Vku4big+IYM", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdGetEvent); + LIB_FUNCTION("0SySxcuVNG0", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdGetUserId); + LIB_FUNCTION("ykNpWs3ktLY", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdInitialize); + LIB_FUNCTION("OA6FbORefbo", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdInitialize2); + LIB_FUNCTION("r-2-a0c7Kfc", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdOptParamInitialize); + LIB_FUNCTION("fHNmij7kAUM", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdRegisterRequestBodyReceptionCallback); + LIB_FUNCTION("OaWw+IVEdbI", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdRegisterRequestCallback); + LIB_FUNCTION("-0c9TCTwnGs", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdRegisterRequestCallback2); + LIB_FUNCTION("h3OvVxzX4qM", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdSetBody); + LIB_FUNCTION("w7oz0AWHpT4", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdSetStatus); + LIB_FUNCTION("k7F0FcDM-Xc", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdStart); + LIB_FUNCTION("0SCgzfVQHpo", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdStop); + LIB_FUNCTION("+-du9tWgE9s", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdTerminate); + LIB_FUNCTION("ZSHiUfYK+QI", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdUnregisterRequestBodyReceptionCallback); + LIB_FUNCTION("xweOi2QT-BE", "libSceCompanionHttpd", 1, "libSceCompanionHttpd", 1, 1, + sceCompanionHttpdUnregisterRequestCallback); +}; + +} // namespace Libraries::CompanionHttpd \ No newline at end of file diff --git a/src/core/libraries/companion/companion_httpd.h b/src/core/libraries/companion/companion_httpd.h new file mode 100644 index 000000000..b6d441653 --- /dev/null +++ b/src/core/libraries/companion/companion_httpd.h @@ -0,0 +1,91 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#pragma once + +#include "common/types.h" +#include "core/libraries/network/net.h" +#include "core/libraries/system/userservice.h" + +namespace Core::Loader { +class SymbolsResolver; +} + +namespace Libraries::CompanionHttpd { + +// OrbisCompanionHttpdEvent event codes +constexpr int ORBIS_COMPANION_HTTPD_EVENT_CONNECT = 0x10000001; +constexpr int ORBIS_COMPANION_HTTPD_EVENT_DISCONNECT = 0x10000002; + +struct OrbisCompanionHttpdHeader { + char* key; + char* value; + struct OrbisCompanionHttpdHeader* header; +}; + +struct OrbisCompanionHttpdRequest { + s32 method; + char* url; + OrbisCompanionHttpdHeader* header; + char* body; + u64 bodySize; +}; + +struct OrbisCompanionHttpdResponse { + s32 status; + OrbisCompanionHttpdHeader* header; + char* body; + u64 bodySize; +}; + +using OrbisCompanionHttpdRequestBodyReceptionCallback = + PS4_SYSV_ABI s32 (*)(s32 event, Libraries::UserService::OrbisUserServiceUserId userId, + const OrbisCompanionHttpdRequest* httpRequest, void* param); + +using OrbisCompanionHttpdRequestCallback = + PS4_SYSV_ABI s32 (*)(Libraries::UserService::OrbisUserServiceUserId userId, + const OrbisCompanionHttpdRequest* httpRequest, + OrbisCompanionHttpdResponse* httpResponse, void* param); + +struct OrbisCompanionUtilDeviceInfo { + Libraries::UserService::OrbisUserServiceUserId userId; + Libraries::Net::OrbisNetSockaddrIn addr; + char reserved[236]; +}; + +struct OrbisCompanionHttpdEvent { + s32 event; + union { + OrbisCompanionUtilDeviceInfo deviceInfo; + Libraries::UserService::OrbisUserServiceUserId userId; + char reserved[256]; + } data; +}; + +s32 PS4_SYSV_ABI sceCompanionHttpdAddHeader(const char* key, const char* value, + OrbisCompanionHttpdResponse* response); +s32 PS4_SYSV_ABI +sceCompanionHttpdGet2ndScreenStatus(Libraries::UserService::OrbisUserServiceUserId userId); +s32 PS4_SYSV_ABI sceCompanionHttpdGetEvent(OrbisCompanionHttpdEvent* pEvent); +s32 PS4_SYSV_ABI sceCompanionHttpdGetUserId(u32 addr, + Libraries::UserService::OrbisUserServiceUserId* userId); +s32 PS4_SYSV_ABI sceCompanionHttpdInitialize(); +s32 PS4_SYSV_ABI sceCompanionHttpdInitialize2(); +s32 PS4_SYSV_ABI sceCompanionHttpdOptParamInitialize(); +s32 PS4_SYSV_ABI sceCompanionHttpdRegisterRequestBodyReceptionCallback( + OrbisCompanionHttpdRequestBodyReceptionCallback function, void* param); +s32 PS4_SYSV_ABI +sceCompanionHttpdRegisterRequestCallback(OrbisCompanionHttpdRequestCallback function, void* param); +s32 PS4_SYSV_ABI +sceCompanionHttpdRegisterRequestCallback2(OrbisCompanionHttpdRequestCallback function, void* param); +s32 PS4_SYSV_ABI sceCompanionHttpdSetBody(const char* body, u64 bodySize, + OrbisCompanionHttpdResponse* response); +s32 PS4_SYSV_ABI sceCompanionHttpdSetStatus(s32 status, OrbisCompanionHttpdResponse* response); +s32 PS4_SYSV_ABI sceCompanionHttpdStart(); +s32 PS4_SYSV_ABI sceCompanionHttpdStop(); +s32 PS4_SYSV_ABI sceCompanionHttpdTerminate(); +s32 PS4_SYSV_ABI sceCompanionHttpdUnregisterRequestBodyReceptionCallback(); +s32 PS4_SYSV_ABI sceCompanionHttpdUnregisterRequestCallback(); + +void RegisterlibSceCompanionHttpd(Core::Loader::SymbolsResolver* sym); +} // namespace Libraries::CompanionHttpd \ No newline at end of file diff --git a/src/core/libraries/gnmdriver/gnmdriver.cpp b/src/core/libraries/gnmdriver/gnmdriver.cpp index f2f40e0e3..9cf340050 100644 --- a/src/core/libraries/gnmdriver/gnmdriver.cpp +++ b/src/core/libraries/gnmdriver/gnmdriver.cpp @@ -179,7 +179,7 @@ s32 PS4_SYSV_ABI sceGnmComputeWaitOnAddress(u32* cmdbuf, u32 size, uintptr_t add auto* wait_reg_mem = reinterpret_cast(cmdbuf); wait_reg_mem->header = PM4Type3Header{PM4ItOpcode::WaitRegMem, 5}; wait_reg_mem->raw = (is_mem << 4u) | (cmp_func & 7u); - wait_reg_mem->poll_addr_lo = u32(addr & addr_mask); + wait_reg_mem->poll_addr_lo_raw = u32(addr & addr_mask); wait_reg_mem->poll_addr_hi = u32(addr >> 32u); wait_reg_mem->ref = ref; wait_reg_mem->mask = mask; diff --git a/src/core/libraries/kernel/equeue.cpp b/src/core/libraries/kernel/equeue.cpp index a4916208a..958019cd3 100644 --- a/src/core/libraries/kernel/equeue.cpp +++ b/src/core/libraries/kernel/equeue.cpp @@ -12,12 +12,25 @@ namespace Libraries::Kernel { +extern boost::asio::io_context io_context; +extern void KernelSignalRequest(); + +static constexpr auto HrTimerSpinlockThresholdUs = 1200u; + // Events are uniquely identified by id and filter. bool EqueueInternal::AddEvent(EqueueEvent& event) { std::scoped_lock lock{m_mutex}; event.time_added = std::chrono::steady_clock::now(); + if (event.event.filter == SceKernelEvent::Filter::Timer || + event.event.filter == SceKernelEvent::Filter::HrTimer) { + // HrTimer events are offset by the threshold of time at the end that we spinlock for + // greater accuracy. + const auto offset = + event.event.filter == SceKernelEvent::Filter::HrTimer ? HrTimerSpinlockThresholdUs : 0u; + event.timer_interval = std::chrono::microseconds(event.event.data - offset); + } const auto& it = std::ranges::find(m_events, event); if (it != m_events.cend()) { @@ -29,6 +42,47 @@ bool EqueueInternal::AddEvent(EqueueEvent& event) { return true; } +bool EqueueInternal::ScheduleEvent(u64 id, s16 filter, + void (*callback)(SceKernelEqueue, const SceKernelEvent&)) { + std::scoped_lock lock{m_mutex}; + + const auto& it = std::ranges::find_if(m_events, [id, filter](auto& ev) { + return ev.event.ident == id && ev.event.filter == filter; + }); + if (it == m_events.cend()) { + return false; + } + + const auto& event = *it; + ASSERT(event.event.filter == SceKernelEvent::Filter::Timer || + event.event.filter == SceKernelEvent::Filter::HrTimer); + + if (!it->timer) { + it->timer = std::make_unique(io_context, event.timer_interval); + } else { + // If the timer already exists we are scheduling a reoccurrence after the next period. + // Set the expiration time to the previous occurrence plus the period. + it->timer->expires_at(it->timer->expiry() + event.timer_interval); + } + + it->timer->async_wait( + [this, event_data = event.event, callback](const boost::system::error_code& ec) { + if (ec) { + if (ec != boost::system::errc::operation_canceled) { + LOG_ERROR(Kernel_Event, "Timer callback error: {}", ec.message()); + } else { + // Timer was cancelled (removed) before it triggered + LOG_DEBUG(Kernel_Event, "Timer cancelled"); + } + return; + } + callback(this, event_data); + }); + KernelSignalRequest(); + + return true; +} + bool EqueueInternal::RemoveEvent(u64 id, s16 filter) { bool has_found = false; std::scoped_lock lock{m_mutex}; @@ -152,18 +206,14 @@ int EqueueInternal::WaitForSmallTimer(SceKernelEvent* ev, int num, u32 micros) { return count; } -extern boost::asio::io_context io_context; -extern void KernelSignalRequest(); +bool EqueueInternal::EventExists(u64 id, s16 filter) { + std::scoped_lock lock{m_mutex}; -static constexpr auto HrTimerSpinlockThresholdUs = 1200u; + const auto& it = std::ranges::find_if(m_events, [id, filter](auto& ev) { + return ev.event.ident == id && ev.event.filter == filter; + }); -static void SmallTimerCallback(const boost::system::error_code& error, SceKernelEqueue eq, - SceKernelEvent kevent) { - static EqueueEvent event; - event.event = kevent; - event.event.data = HrTimerSpinlockThresholdUs; - eq->AddSmallTimer(event); - eq->TriggerEvent(kevent.ident, SceKernelEvent::Filter::HrTimer, kevent.udata); + return it != m_events.cend(); } int PS4_SYSV_ABI sceKernelCreateEqueue(SceKernelEqueue* eq, const char* name) { @@ -243,6 +293,14 @@ int PS4_SYSV_ABI sceKernelWaitEqueue(SceKernelEqueue eq, SceKernelEvent* ev, int return ORBIS_OK; } +static void HrTimerCallback(SceKernelEqueue eq, const SceKernelEvent& kevent) { + static EqueueEvent event; + event.event = kevent; + event.event.data = HrTimerSpinlockThresholdUs; + eq->AddSmallTimer(event); + eq->TriggerEvent(kevent.ident, SceKernelEvent::Filter::HrTimer, kevent.udata); +} + s32 PS4_SYSV_ABI sceKernelAddHRTimerEvent(SceKernelEqueue eq, int id, timespec* ts, void* udata) { if (eq == nullptr) { return ORBIS_KERNEL_ERROR_EBADF; @@ -273,17 +331,10 @@ s32 PS4_SYSV_ABI sceKernelAddHRTimerEvent(SceKernelEqueue eq, int id, timespec* return eq->AddSmallTimer(event) ? ORBIS_OK : ORBIS_KERNEL_ERROR_ENOMEM; } - event.timer = std::make_unique( - io_context, std::chrono::microseconds(total_us - HrTimerSpinlockThresholdUs)); - - event.timer->async_wait(std::bind(SmallTimerCallback, std::placeholders::_1, eq, event.event)); - - if (!eq->AddEvent(event)) { + if (!eq->AddEvent(event) || + !eq->ScheduleEvent(id, SceKernelEvent::Filter::HrTimer, HrTimerCallback)) { return ORBIS_KERNEL_ERROR_ENOMEM; } - - KernelSignalRequest(); - return ORBIS_OK; } @@ -300,6 +351,57 @@ int PS4_SYSV_ABI sceKernelDeleteHRTimerEvent(SceKernelEqueue eq, int id) { } } +static void TimerCallback(SceKernelEqueue eq, const SceKernelEvent& kevent) { + if (eq->EventExists(kevent.ident, kevent.filter)) { + eq->TriggerEvent(kevent.ident, SceKernelEvent::Filter::Timer, kevent.udata); + + if (!(kevent.flags & SceKernelEvent::Flags::OneShot)) { + // Reschedule the event for its next period. + eq->ScheduleEvent(kevent.ident, kevent.filter, TimerCallback); + } + } +} + +int PS4_SYSV_ABI sceKernelAddTimerEvent(SceKernelEqueue eq, int id, SceKernelUseconds usec, + void* udata) { + if (eq == nullptr) { + return ORBIS_KERNEL_ERROR_EBADF; + } + + EqueueEvent event{}; + event.event.ident = static_cast(id); + event.event.filter = SceKernelEvent::Filter::Timer; + event.event.flags = SceKernelEvent::Flags::Add; + event.event.fflags = 0; + event.event.data = usec; + event.event.udata = udata; + + if (eq->EventExists(event.event.ident, event.event.filter)) { + eq->RemoveEvent(id, SceKernelEvent::Filter::Timer); + LOG_DEBUG(Kernel_Event, + "Timer event already exists, removing it: queue name={}, queue id={}", + eq->GetName(), event.event.ident); + } + + LOG_DEBUG(Kernel_Event, "Added timing event: queue name={}, queue id={}, usec={}, pointer={:x}", + eq->GetName(), event.event.ident, usec, reinterpret_cast(udata)); + + if (!eq->AddEvent(event) || + !eq->ScheduleEvent(id, SceKernelEvent::Filter::Timer, TimerCallback)) { + return ORBIS_KERNEL_ERROR_ENOMEM; + } + return ORBIS_OK; +} + +int PS4_SYSV_ABI sceKernelDeleteTimerEvent(SceKernelEqueue eq, int id) { + if (eq == nullptr) { + return ORBIS_KERNEL_ERROR_EBADF; + } + + return eq->RemoveEvent(id, SceKernelEvent::Filter::Timer) ? ORBIS_OK + : ORBIS_KERNEL_ERROR_ENOENT; +} + int PS4_SYSV_ABI sceKernelAddUserEvent(SceKernelEqueue eq, int id) { if (eq == nullptr) { return ORBIS_KERNEL_ERROR_EBADF; @@ -380,6 +482,8 @@ void RegisterEventQueue(Core::Loader::SymbolsResolver* sym) { LIB_FUNCTION("WDszmSbWuDk", "libkernel", 1, "libkernel", 1, 1, sceKernelAddUserEventEdge); LIB_FUNCTION("R74tt43xP6k", "libkernel", 1, "libkernel", 1, 1, sceKernelAddHRTimerEvent); LIB_FUNCTION("J+LF6LwObXU", "libkernel", 1, "libkernel", 1, 1, sceKernelDeleteHRTimerEvent); + LIB_FUNCTION("57ZK+ODEXWY", "libkernel", 1, "libkernel", 1, 1, sceKernelAddTimerEvent); + LIB_FUNCTION("YWQFUyXIVdU", "libkernel", 1, "libkernel", 1, 1, sceKernelDeleteTimerEvent); LIB_FUNCTION("F6e0kwo4cnk", "libkernel", 1, "libkernel", 1, 1, sceKernelTriggerUserEvent); LIB_FUNCTION("LJDwdSNTnDg", "libkernel", 1, "libkernel", 1, 1, sceKernelDeleteUserEvent); LIB_FUNCTION("mJ7aghmgvfc", "libkernel", 1, "libkernel", 1, 1, sceKernelGetEventId); diff --git a/src/core/libraries/kernel/equeue.h b/src/core/libraries/kernel/equeue.h index 2bd7ef510..a0367c66a 100644 --- a/src/core/libraries/kernel/equeue.h +++ b/src/core/libraries/kernel/equeue.h @@ -21,6 +21,9 @@ namespace Libraries::Kernel { class EqueueInternal; struct EqueueEvent; +using SceKernelUseconds = u32; +using SceKernelEqueue = EqueueInternal*; + struct SceKernelEvent { enum Filter : s16 { None = 0, @@ -77,6 +80,7 @@ struct EqueueEvent { SceKernelEvent event; void* data = nullptr; std::chrono::steady_clock::time_point time_added; + std::chrono::microseconds timer_interval; std::unique_ptr timer; void ResetTriggerState() { @@ -133,6 +137,8 @@ public: } bool AddEvent(EqueueEvent& event); + bool ScheduleEvent(u64 id, s16 filter, + void (*callback)(SceKernelEqueue, const SceKernelEvent&)); bool RemoveEvent(u64 id, s16 filter); int WaitForEvents(SceKernelEvent* ev, int num, u32 micros); bool TriggerEvent(u64 ident, s16 filter, void* trigger_data); @@ -152,6 +158,8 @@ public: int WaitForSmallTimer(SceKernelEvent* ev, int num, u32 micros); + bool EventExists(u64 id, s16 filter); + private: std::string m_name; std::mutex m_mutex; @@ -160,9 +168,6 @@ private: std::condition_variable m_cond; }; -using SceKernelUseconds = u32; -using SceKernelEqueue = EqueueInternal*; - u64 PS4_SYSV_ABI sceKernelGetEventData(const SceKernelEvent* ev); void RegisterEventQueue(Core::Loader::SymbolsResolver* sym); diff --git a/src/core/libraries/kernel/kernel.cpp b/src/core/libraries/kernel/kernel.cpp index c7eafe799..180850217 100644 --- a/src/core/libraries/kernel/kernel.cpp +++ b/src/core/libraries/kernel/kernel.cpp @@ -108,6 +108,9 @@ void SetPosixErrno(int e) { case EACCES: g_posix_errno = POSIX_EACCES; break; + case EFAULT: + g_posix_errno = POSIX_EFAULT; + break; case EINVAL: g_posix_errno = POSIX_EINVAL; break; diff --git a/src/core/libraries/kernel/memory.cpp b/src/core/libraries/kernel/memory.cpp index dd0e07302..cb41a664a 100644 --- a/src/core/libraries/kernel/memory.cpp +++ b/src/core/libraries/kernel/memory.cpp @@ -209,6 +209,19 @@ int PS4_SYSV_ABI sceKernelMapDirectMemory(void** addr, u64 len, int prot, int fl "anon"); } +s32 PS4_SYSV_ABI sceKernelMapDirectMemory2(void** addr, u64 len, s32 type, s32 prot, s32 flags, + s64 phys_addr, u64 alignment) { + LOG_INFO(Kernel_Vmm, "called, redirected to sceKernelMapNamedDirectMemory"); + const s32 ret = + sceKernelMapNamedDirectMemory(addr, len, prot, flags, phys_addr, alignment, "anon"); + + if (ret == 0) { + auto* memory = Core::Memory::Instance(); + memory->SetDirectMemoryType(phys_addr, type); + } + return ret; +} + s32 PS4_SYSV_ABI sceKernelMapNamedFlexibleMemory(void** addr_in_out, std::size_t len, int prot, int flags, const char* name) { @@ -290,6 +303,12 @@ int PS4_SYSV_ABI sceKernelGetDirectMemoryType(u64 addr, int* directMemoryTypeOut directMemoryEndOut); } +int PS4_SYSV_ABI sceKernelIsStack(void* addr, void** start, void** end) { + LOG_DEBUG(Kernel_Vmm, "called, addr = {}", fmt::ptr(addr)); + auto* memory = Core::Memory::Instance(); + return memory->IsStack(std::bit_cast(addr), start, end); +} + s32 PS4_SYSV_ABI sceKernelBatchMap(OrbisKernelBatchMapEntry* entries, int numEntries, int* numEntriesOut) { return sceKernelBatchMap2(entries, numEntries, numEntriesOut, @@ -636,8 +655,10 @@ void RegisterMemory(Core::Loader::SymbolsResolver* sym) { LIB_FUNCTION("7oxv3PPCumo", "libkernel", 1, "libkernel", 1, 1, sceKernelReserveVirtualRange); LIB_FUNCTION("BC+OG5m9+bw", "libkernel", 1, "libkernel", 1, 1, sceKernelGetDirectMemoryType); LIB_FUNCTION("pO96TwzOm5E", "libkernel", 1, "libkernel", 1, 1, sceKernelGetDirectMemorySize); + LIB_FUNCTION("yDBwVAolDgg", "libkernel", 1, "libkernel", 1, 1, sceKernelIsStack); LIB_FUNCTION("NcaWUxfMNIQ", "libkernel", 1, "libkernel", 1, 1, sceKernelMapNamedDirectMemory); LIB_FUNCTION("L-Q3LEjIbgA", "libkernel", 1, "libkernel", 1, 1, sceKernelMapDirectMemory); + LIB_FUNCTION("BQQniolj9tQ", "libkernel", 1, "libkernel", 1, 1, sceKernelMapDirectMemory2); LIB_FUNCTION("WFcfL2lzido", "libkernel", 1, "libkernel", 1, 1, sceKernelQueryMemoryProtection); LIB_FUNCTION("BHouLQzh0X0", "libkernel", 1, "libkernel", 1, 1, sceKernelDirectMemoryQuery); LIB_FUNCTION("MBuItvba6z8", "libkernel", 1, "libkernel", 1, 1, sceKernelReleaseDirectMemory); diff --git a/src/core/libraries/kernel/memory.h b/src/core/libraries/kernel/memory.h index 3e2bf8de5..92e158a00 100644 --- a/src/core/libraries/kernel/memory.h +++ b/src/core/libraries/kernel/memory.h @@ -158,6 +158,7 @@ void PS4_SYSV_ABI _sceKernelRtldSetApplicationHeapAPI(void* func[]); int PS4_SYSV_ABI sceKernelGetDirectMemoryType(u64 addr, int* directMemoryTypeOut, void** directMemoryStartOut, void** directMemoryEndOut); +int PS4_SYSV_ABI sceKernelIsStack(void* addr, void** start, void** end); s32 PS4_SYSV_ABI sceKernelBatchMap(OrbisKernelBatchMapEntry* entries, int numEntries, int* numEntriesOut); diff --git a/src/core/libraries/kernel/threads/event_flag.cpp b/src/core/libraries/kernel/threads/event_flag.cpp index 24ddcb927..91b17bd49 100644 --- a/src/core/libraries/kernel/threads/event_flag.cpp +++ b/src/core/libraries/kernel/threads/event_flag.cpp @@ -315,7 +315,7 @@ int PS4_SYSV_ABI sceKernelPollEventFlag(OrbisKernelEventFlag ef, u64 bitPattern, auto result = ef->Poll(bitPattern, wait, clear, pResultPat); if (result != ORBIS_OK && result != ORBIS_KERNEL_ERROR_EBUSY) { - LOG_ERROR(Kernel_Event, "returned {}", result); + LOG_DEBUG(Kernel_Event, "returned {:#x}", result); } return result; @@ -361,7 +361,7 @@ int PS4_SYSV_ABI sceKernelWaitEventFlag(OrbisKernelEventFlag ef, u64 bitPattern, u32 result = ef->Wait(bitPattern, wait, clear, pResultPat, pTimeout); if (result != ORBIS_OK && result != ORBIS_KERNEL_ERROR_ETIMEDOUT) { - LOG_ERROR(Kernel_Event, "returned {:#x}", result); + LOG_DEBUG(Kernel_Event, "returned {:#x}", result); } return result; diff --git a/src/core/libraries/kernel/time.cpp b/src/core/libraries/kernel/time.cpp index b7e4c1756..2fe74d0a3 100644 --- a/src/core/libraries/kernel/time.cpp +++ b/src/core/libraries/kernel/time.cpp @@ -5,24 +5,23 @@ #include "common/assert.h" #include "common/native_clock.h" +#include "common/thread.h" #include "core/libraries/kernel/kernel.h" #include "core/libraries/kernel/orbis_error.h" +#include "core/libraries/kernel/posix_error.h" #include "core/libraries/kernel/time.h" #include "core/libraries/libs.h" #ifdef _WIN64 -#include #include - #include "common/ntapi.h" - #else #if __APPLE__ #include #endif +#include #include #include -#include #include #endif @@ -52,88 +51,116 @@ u64 PS4_SYSV_ABI sceKernelReadTsc() { return clock->GetUptime(); } -int PS4_SYSV_ABI sceKernelUsleep(u32 microseconds) { -#ifdef _WIN64 - const auto start_time = std::chrono::high_resolution_clock::now(); - auto total_wait_time = std::chrono::microseconds(microseconds); +static s32 posix_nanosleep_impl(const OrbisKernelTimespec* rqtp, OrbisKernelTimespec* rmtp, + const bool interruptible) { + if (!rqtp || rqtp->tv_sec < 0 || rqtp->tv_nsec < 0 || rqtp->tv_nsec >= 1'000'000'000) { + SetPosixErrno(EINVAL); + return -1; + } + const auto duration = std::chrono::nanoseconds(rqtp->tv_sec * 1'000'000'000 + rqtp->tv_nsec); + std::chrono::nanoseconds remain; + const auto uninterrupted = Common::AccurateSleep(duration, &remain, interruptible); + if (rmtp) { + rmtp->tv_sec = remain.count() / 1'000'000'000; + rmtp->tv_nsec = remain.count() % 1'000'000'000; + } + if (!uninterrupted) { + SetPosixErrno(EINTR); + return -1; + } + return 0; +} - while (total_wait_time.count() > 0) { - auto wait_time = std::chrono::ceil(total_wait_time).count(); - u64 res = SleepEx(static_cast(wait_time), true); - if (res == WAIT_IO_COMPLETION) { - auto elapsedTime = std::chrono::high_resolution_clock::now() - start_time; - auto elapsedMicroseconds = - std::chrono::duration_cast(elapsedTime).count(); - total_wait_time = std::chrono::microseconds(microseconds - elapsedMicroseconds); - } else { - break; - } +s32 PS4_SYSV_ABI posix_nanosleep(const OrbisKernelTimespec* rqtp, OrbisKernelTimespec* rmtp) { + return posix_nanosleep_impl(rqtp, rmtp, true); +} + +s32 PS4_SYSV_ABI sceKernelNanosleep(const OrbisKernelTimespec* rqtp, OrbisKernelTimespec* rmtp) { + if (const auto ret = posix_nanosleep_impl(rqtp, rmtp, false); ret < 0) { + return ErrnoToSceKernelError(*__Error()); + } + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI posix_usleep(u32 microseconds) { + const OrbisKernelTimespec ts = { + .tv_sec = microseconds / 1'000'000, + .tv_nsec = (microseconds % 1'000'000) * 1'000, + }; + return posix_nanosleep(&ts, nullptr); +} + +s32 PS4_SYSV_ABI sceKernelUsleep(u32 microseconds) { + const OrbisKernelTimespec ts = { + .tv_sec = microseconds / 1'000'000, + .tv_nsec = (microseconds % 1'000'000) * 1'000, + }; + return sceKernelNanosleep(&ts, nullptr); +} + +u32 PS4_SYSV_ABI posix_sleep(u32 seconds) { + const OrbisKernelTimespec ts = { + .tv_sec = seconds, + .tv_nsec = 0, + }; + OrbisKernelTimespec rm; + if (const auto ret = posix_nanosleep(&ts, &rm); ret < 0) { + return *__Error() == POSIX_EINTR ? rm.tv_sec + (rm.tv_nsec == 0 ? 0 : 1) : seconds; + } + return 0; +} + +s32 PS4_SYSV_ABI sceKernelSleep(u32 seconds) { + return sceKernelUsleep(seconds * 1'000'000); +} + +s32 PS4_SYSV_ABI posix_clock_gettime(u32 clock_id, OrbisKernelTimespec* ts) { + if (ts == nullptr) { + SetPosixErrno(EFAULT); + return -1; } - return 0; -#else - timespec start; - timespec remain; - start.tv_sec = microseconds / 1000000; - start.tv_nsec = (microseconds % 1000000) * 1000; - timespec* requested = &start; - int ret = 0; - do { - ret = nanosleep(requested, &remain); - requested = &remain; - } while (ret != 0); - return ret; -#endif -} + if (clock_id == ORBIS_CLOCK_PROCTIME) { + const auto us = sceKernelGetProcessTime(); + ts->tv_sec = static_cast(us / 1'000'000); + ts->tv_nsec = static_cast((us % 1'000'000) * 1000); + return 0; + } + if (clock_id == ORBIS_CLOCK_EXT_NETWORK || clock_id == ORBIS_CLOCK_EXT_DEBUG_NETWORK || + clock_id == ORBIS_CLOCK_EXT_AD_NETWORK || clock_id == ORBIS_CLOCK_EXT_RAW_NETWORK) { + LOG_ERROR(Lib_Kernel, "Unsupported clock type {}, using CLOCK_MONOTONIC", clock_id); + clock_id = ORBIS_CLOCK_MONOTONIC; + } -int PS4_SYSV_ABI posix_usleep(u32 microseconds) { - return sceKernelUsleep(microseconds); -} - -u32 PS4_SYSV_ABI sceKernelSleep(u32 seconds) { - std::this_thread::sleep_for(std::chrono::seconds(seconds)); - return 0; -} - -#ifdef _WIN64 -#ifndef CLOCK_REALTIME -#define CLOCK_REALTIME 0 -#endif -#ifndef CLOCK_MONOTONIC -#define CLOCK_MONOTONIC 1 -#endif -#ifndef CLOCK_PROCESS_CPUTIME_ID -#define CLOCK_PROCESS_CPUTIME_ID 2 -#endif -#ifndef CLOCK_THREAD_CPUTIME_ID -#define CLOCK_THREAD_CPUTIME_ID 3 -#endif -#ifndef CLOCK_REALTIME_COARSE -#define CLOCK_REALTIME_COARSE 5 -#endif -#ifndef CLOCK_MONOTONIC_COARSE -#define CLOCK_MONOTONIC_COARSE 6 -#endif - -#define DELTA_EPOCH_IN_100NS 116444736000000000ULL - -static u64 FileTimeTo100Ns(FILETIME& ft) { - return *reinterpret_cast(&ft); -} - -static s32 clock_gettime(u32 clock_id, struct timespec* ts) { +#ifdef _WIN32 + static const auto FileTimeTo100Ns = [](FILETIME& ft) { return *reinterpret_cast(&ft); }; switch (clock_id) { - case CLOCK_REALTIME: - case CLOCK_REALTIME_COARSE: { + case ORBIS_CLOCK_REALTIME: + case ORBIS_CLOCK_REALTIME_PRECISE: { FILETIME ft; - GetSystemTimeAsFileTime(&ft); - const u64 ns = FileTimeTo100Ns(ft) - DELTA_EPOCH_IN_100NS; + GetSystemTimePreciseAsFileTime(&ft); + static constexpr u64 DeltaEpochIn100ns = 116444736000000000ULL; + const u64 ns = FileTimeTo100Ns(ft) - DeltaEpochIn100ns; ts->tv_sec = ns / 10'000'000; ts->tv_nsec = (ns % 10'000'000) * 100; return 0; } - case CLOCK_MONOTONIC: - case CLOCK_MONOTONIC_COARSE: { + case ORBIS_CLOCK_SECOND: + case ORBIS_CLOCK_REALTIME_FAST: { + FILETIME ft; + GetSystemTimeAsFileTime(&ft); + static constexpr u64 DeltaEpochIn100ns = 116444736000000000ULL; + const u64 ns = FileTimeTo100Ns(ft) - DeltaEpochIn100ns; + ts->tv_sec = ns / 10'000'000; + ts->tv_nsec = (ns % 10'000'000) * 100; + return 0; + } + case ORBIS_CLOCK_UPTIME: + case ORBIS_CLOCK_UPTIME_PRECISE: + case ORBIS_CLOCK_MONOTONIC: + case ORBIS_CLOCK_MONOTONIC_PRECISE: + case ORBIS_CLOCK_UPTIME_FAST: + case ORBIS_CLOCK_MONOTONIC_FAST: { static LARGE_INTEGER pf = [] { LARGE_INTEGER res{}; QueryPerformanceFrequency(&pf); @@ -141,43 +168,53 @@ static s32 clock_gettime(u32 clock_id, struct timespec* ts) { }(); LARGE_INTEGER pc{}; - QueryPerformanceCounter(&pc); + if (!QueryPerformanceCounter(&pc)) { + SetPosixErrno(EFAULT); + return -1; + } ts->tv_sec = pc.QuadPart / pf.QuadPart; ts->tv_nsec = ((pc.QuadPart % pf.QuadPart) * 1000'000'000) / pf.QuadPart; return 0; } - case CLOCK_PROCESS_CPUTIME_ID: { + case ORBIS_CLOCK_THREAD_CPUTIME_ID: { FILETIME ct, et, kt, ut; - if (!GetProcessTimes(GetCurrentProcess(), &ct, &et, &kt, &ut)) { - return EFAULT; + if (!GetThreadTimes(GetCurrentThread(), &ct, &et, &kt, &ut)) { + SetPosixErrno(EFAULT); + return -1; } const u64 ns = FileTimeTo100Ns(ut) + FileTimeTo100Ns(kt); ts->tv_sec = ns / 10'000'000; ts->tv_nsec = (ns % 10'000'000) * 100; return 0; } - case CLOCK_THREAD_CPUTIME_ID: { + case ORBIS_CLOCK_VIRTUAL: { FILETIME ct, et, kt, ut; - if (!GetThreadTimes(GetCurrentThread(), &ct, &et, &kt, &ut)) { - return EFAULT; + if (!GetProcessTimes(GetCurrentProcess(), &ct, &et, &kt, &ut)) { + SetPosixErrno(EFAULT); + return -1; } - const u64 ns = FileTimeTo100Ns(ut) + FileTimeTo100Ns(kt); + const u64 ns = FileTimeTo100Ns(ut); + ts->tv_sec = ns / 10'000'000; + ts->tv_nsec = (ns % 10'000'000) * 100; + return 0; + } + case ORBIS_CLOCK_PROF: { + FILETIME ct, et, kt, ut; + if (!GetProcessTimes(GetCurrentProcess(), &ct, &et, &kt, &ut)) { + SetPosixErrno(EFAULT); + return -1; + } + const u64 ns = FileTimeTo100Ns(kt); ts->tv_sec = ns / 10'000'000; ts->tv_nsec = (ns % 10'000'000) * 100; return 0; } default: - return EINVAL; + SetPosixErrno(EFAULT); + return -1; } -} -#endif - -int PS4_SYSV_ABI orbis_clock_gettime(s32 clock_id, struct OrbisKernelTimespec* ts) { - if (ts == nullptr) { - return ORBIS_KERNEL_ERROR_EFAULT; - } - - clockid_t pclock_id = CLOCK_MONOTONIC; +#else + clockid_t pclock_id; switch (clock_id) { case ORBIS_CLOCK_REALTIME: case ORBIS_CLOCK_REALTIME_PRECISE: @@ -185,7 +222,7 @@ int PS4_SYSV_ABI orbis_clock_gettime(s32 clock_id, struct OrbisKernelTimespec* t break; case ORBIS_CLOCK_SECOND: case ORBIS_CLOCK_REALTIME_FAST: -#ifndef __APPLE__ +#ifdef CLOCK_REALTIME_COARSE pclock_id = CLOCK_REALTIME_COARSE; #else pclock_id = CLOCK_REALTIME; @@ -199,7 +236,7 @@ int PS4_SYSV_ABI orbis_clock_gettime(s32 clock_id, struct OrbisKernelTimespec* t break; case ORBIS_CLOCK_UPTIME_FAST: case ORBIS_CLOCK_MONOTONIC_FAST: -#ifndef __APPLE__ +#ifdef CLOCK_MONOTONIC_COARSE pclock_id = CLOCK_MONOTONIC_COARSE; #else pclock_id = CLOCK_MONOTONIC; @@ -208,196 +245,226 @@ int PS4_SYSV_ABI orbis_clock_gettime(s32 clock_id, struct OrbisKernelTimespec* t case ORBIS_CLOCK_THREAD_CPUTIME_ID: pclock_id = CLOCK_THREAD_CPUTIME_ID; break; - case ORBIS_CLOCK_PROCTIME: { - const auto us = sceKernelGetProcessTime(); - ts->tv_sec = us / 1'000'000; - ts->tv_nsec = (us % 1'000'000) * 1000; - return 0; - } case ORBIS_CLOCK_VIRTUAL: { -#ifdef _WIN64 - FILETIME ct, et, kt, ut; - if (!GetProcessTimes(GetCurrentProcess(), &ct, &et, &kt, &ut)) { - return EFAULT; - } - const u64 ns = FileTimeTo100Ns(ut); - ts->tv_sec = ns / 10'000'000; - ts->tv_nsec = (ns % 10'000'000) * 100; -#else - struct rusage ru; + rusage ru; const auto res = getrusage(RUSAGE_SELF, &ru); if (res < 0) { - return res; + SetPosixErrno(EFAULT); + return -1; } ts->tv_sec = ru.ru_utime.tv_sec; ts->tv_nsec = ru.ru_utime.tv_usec * 1000; -#endif return 0; } case ORBIS_CLOCK_PROF: { -#ifdef _WIN64 - FILETIME ct, et, kt, ut; - if (!GetProcessTimes(GetCurrentProcess(), &ct, &et, &kt, &ut)) { - return EFAULT; - } - const u64 ns = FileTimeTo100Ns(kt); - ts->tv_sec = ns / 10'000'000; - ts->tv_nsec = (ns % 10'000'000) * 100; -#else - struct rusage ru; + rusage ru; const auto res = getrusage(RUSAGE_SELF, &ru); if (res < 0) { - return res; + SetPosixErrno(EFAULT); + return -1; } ts->tv_sec = ru.ru_stime.tv_sec; ts->tv_nsec = ru.ru_stime.tv_usec * 1000; -#endif return 0; } - case ORBIS_CLOCK_EXT_NETWORK: - case ORBIS_CLOCK_EXT_DEBUG_NETWORK: - case ORBIS_CLOCK_EXT_AD_NETWORK: - case ORBIS_CLOCK_EXT_RAW_NETWORK: - pclock_id = CLOCK_MONOTONIC; - LOG_ERROR(Lib_Kernel, "unsupported = {} using CLOCK_MONOTONIC", clock_id); - break; default: - return EINVAL; + SetPosixErrno(EFAULT); + return -1; } timespec t{}; - int result = clock_gettime(pclock_id, &t); + const auto result = clock_gettime(pclock_id, &t); ts->tv_sec = t.tv_sec; ts->tv_nsec = t.tv_nsec; - return result; -} - -int PS4_SYSV_ABI sceKernelClockGettime(s32 clock_id, OrbisKernelTimespec* tp) { - const auto res = orbis_clock_gettime(clock_id, tp); - if (res < 0) { - return ErrnoToSceKernelError(res); + if (result < 0) { + SetPosixErrno(errno); + return -1; } - return ORBIS_OK; -} - -int PS4_SYSV_ABI posix_nanosleep(const OrbisKernelTimespec* rqtp, OrbisKernelTimespec* rmtp) { - const auto* request = reinterpret_cast(rqtp); - auto* remain = reinterpret_cast(rmtp); - return nanosleep(request, remain); -} - -int PS4_SYSV_ABI sceKernelNanosleep(const OrbisKernelTimespec* rqtp, OrbisKernelTimespec* rmtp) { - if (!rqtp || !rmtp) { - return ORBIS_KERNEL_ERROR_EFAULT; - } - - if (rqtp->tv_sec < 0 || rqtp->tv_nsec < 0) { - return ORBIS_KERNEL_ERROR_EINVAL; - } - - return posix_nanosleep(rqtp, rmtp); -} - -int PS4_SYSV_ABI sceKernelGettimeofday(OrbisKernelTimeval* tp) { - if (!tp) { - return ORBIS_KERNEL_ERROR_EFAULT; - } - -#ifdef _WIN64 - FILETIME filetime; - GetSystemTimePreciseAsFileTime(&filetime); - - constexpr u64 UNIX_TIME_START = 0x295E9648864000; - constexpr u64 TICKS_PER_SECOND = 1000000; - - u64 ticks = filetime.dwHighDateTime; - ticks <<= 32; - ticks |= filetime.dwLowDateTime; - ticks /= 10; - ticks -= UNIX_TIME_START; - - tp->tv_sec = ticks / TICKS_PER_SECOND; - tp->tv_usec = ticks % TICKS_PER_SECOND; -#else - timeval tv; - gettimeofday(&tv, nullptr); - tp->tv_sec = tv.tv_sec; - tp->tv_usec = tv.tv_usec; + return 0; #endif +} + +s32 PS4_SYSV_ABI sceKernelClockGettime(const u32 clock_id, OrbisKernelTimespec* ts) { + if (const auto ret = posix_clock_gettime(clock_id, ts); ret < 0) { + return ErrnoToSceKernelError(*__Error()); + } return ORBIS_OK; } -int PS4_SYSV_ABI gettimeofday(OrbisKernelTimeval* tp, OrbisKernelTimezone* tz) { - // FreeBSD docs mention that the kernel generally does not track these values - // and they are usually returned as zero. - if (tz) { - tz->tz_minuteswest = 0; - tz->tz_dsttime = 0; - } - return sceKernelGettimeofday(tp); -} - -s32 PS4_SYSV_ABI sceKernelGettimezone(OrbisKernelTimezone* tz) { -#ifdef _WIN64 - ASSERT(tz); - static int tzflag = 0; - if (!tzflag) { - _tzset(); - tzflag++; - } - tz->tz_minuteswest = _timezone / 60; - tz->tz_dsttime = _daylight; -#else - struct timezone tzz; - struct timeval tv; - gettimeofday(&tv, &tzz); - tz->tz_dsttime = tzz.tz_dsttime; - tz->tz_minuteswest = tzz.tz_minuteswest; -#endif - return ORBIS_OK; -} - -int PS4_SYSV_ABI posix_clock_getres(u32 clock_id, OrbisKernelTimespec* res) { +s32 PS4_SYSV_ABI posix_clock_getres(u32 clock_id, OrbisKernelTimespec* res) { if (res == nullptr) { - return ORBIS_KERNEL_ERROR_EFAULT; + SetPosixErrno(EFAULT); + return -1; } - clockid_t pclock_id = CLOCK_REALTIME; + + if (clock_id == ORBIS_CLOCK_EXT_NETWORK || clock_id == ORBIS_CLOCK_EXT_DEBUG_NETWORK || + clock_id == ORBIS_CLOCK_EXT_AD_NETWORK || clock_id == ORBIS_CLOCK_EXT_RAW_NETWORK) { + LOG_ERROR(Lib_Kernel, "Unsupported clock type {}, using CLOCK_MONOTONIC", clock_id); + clock_id = ORBIS_CLOCK_MONOTONIC; + } + +#ifdef _WIN32 + switch (clock_id) { + case ORBIS_CLOCK_SECOND: + case ORBIS_CLOCK_REALTIME_FAST: { + DWORD timeAdjustment; + DWORD timeIncrement; + BOOL isTimeAdjustmentDisabled; + if (!GetSystemTimeAdjustment(&timeAdjustment, &timeIncrement, &isTimeAdjustmentDisabled)) { + SetPosixErrno(EFAULT); + return -1; + } + res->tv_sec = 0; + res->tv_nsec = timeIncrement * 100; + return 0; + } + case ORBIS_CLOCK_REALTIME: + case ORBIS_CLOCK_REALTIME_PRECISE: + case ORBIS_CLOCK_UPTIME: + case ORBIS_CLOCK_UPTIME_PRECISE: + case ORBIS_CLOCK_MONOTONIC: + case ORBIS_CLOCK_MONOTONIC_PRECISE: + case ORBIS_CLOCK_UPTIME_FAST: + case ORBIS_CLOCK_MONOTONIC_FAST: { + LARGE_INTEGER pf; + if (!QueryPerformanceFrequency(&pf)) { + SetPosixErrno(EFAULT); + return -1; + } + res->tv_sec = 0; + res->tv_nsec = + std::max(static_cast((1000000000 + (pf.QuadPart >> 1)) / pf.QuadPart), 1); + return 0; + } + default: + UNREACHABLE(); + } +#else + clockid_t pclock_id; switch (clock_id) { case ORBIS_CLOCK_REALTIME: case ORBIS_CLOCK_REALTIME_PRECISE: - case ORBIS_CLOCK_REALTIME_FAST: pclock_id = CLOCK_REALTIME; break; case ORBIS_CLOCK_SECOND: + case ORBIS_CLOCK_REALTIME_FAST: +#ifdef CLOCK_REALTIME_COARSE + pclock_id = CLOCK_REALTIME_COARSE; +#else + pclock_id = CLOCK_REALTIME; +#endif + break; + case ORBIS_CLOCK_UPTIME: + case ORBIS_CLOCK_UPTIME_PRECISE: case ORBIS_CLOCK_MONOTONIC: case ORBIS_CLOCK_MONOTONIC_PRECISE: - case ORBIS_CLOCK_MONOTONIC_FAST: pclock_id = CLOCK_MONOTONIC; break; + case ORBIS_CLOCK_UPTIME_FAST: + case ORBIS_CLOCK_MONOTONIC_FAST: +#ifdef CLOCK_MONOTONIC_COARSE + pclock_id = CLOCK_MONOTONIC_COARSE; +#else + pclock_id = CLOCK_MONOTONIC; +#endif + break; default: UNREACHABLE(); } timespec t{}; - int result = clock_getres(pclock_id, &t); + const auto result = clock_getres(pclock_id, &t); res->tv_sec = t.tv_sec; res->tv_nsec = t.tv_nsec; - if (result == 0) { - return ORBIS_OK; + if (result < 0) { + SetPosixErrno(errno); + return -1; } - return ORBIS_KERNEL_ERROR_EINVAL; + return 0; +#endif } -int PS4_SYSV_ABI sceKernelConvertLocaltimeToUtc(time_t param_1, int64_t param_2, time_t* seconds, - OrbisKernelTimezone* timezone, int* dst_seconds) { +s32 PS4_SYSV_ABI sceKernelClockGetres(const u32 clock_id, OrbisKernelTimespec* res) { + if (const auto ret = posix_clock_getres(clock_id, res); ret < 0) { + return ErrnoToSceKernelError(*__Error()); + } + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI posix_gettimeofday(OrbisKernelTimeval* tp, OrbisKernelTimezone* tz) { +#ifdef _WIN64 + if (tp) { + FILETIME filetime; + GetSystemTimePreciseAsFileTime(&filetime); + + constexpr u64 UNIX_TIME_START = 0x295E9648864000; + constexpr u64 TICKS_PER_SECOND = 1000000; + + u64 ticks = filetime.dwHighDateTime; + ticks <<= 32; + ticks |= filetime.dwLowDateTime; + ticks /= 10; + ticks -= UNIX_TIME_START; + + tp->tv_sec = ticks / TICKS_PER_SECOND; + tp->tv_usec = ticks % TICKS_PER_SECOND; + } + if (tz) { + static int tzflag = 0; + if (!tzflag) { + _tzset(); + tzflag++; + } + tz->tz_minuteswest = _timezone / 60; + tz->tz_dsttime = _daylight; + } + return 0; +#else + struct timezone tzz; + timeval tv; + const auto ret = gettimeofday(&tv, &tzz); + if (tp) { + tp->tv_sec = tv.tv_sec; + tp->tv_usec = tv.tv_usec; + } + if (tz) { + tz->tz_dsttime = tzz.tz_dsttime; + tz->tz_minuteswest = tzz.tz_minuteswest; + } + if (ret < 0) { + SetPosixErrno(errno); + return -1; + } + return 0; +#endif +} + +s32 PS4_SYSV_ABI sceKernelGettimeofday(OrbisKernelTimeval* tp) { + if (const auto ret = posix_gettimeofday(tp, nullptr); ret < 0) { + return ErrnoToSceKernelError(*__Error()); + } + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceKernelGettimezone(OrbisKernelTimezone* tz) { + if (const auto ret = posix_gettimeofday(nullptr, tz); ret < 0) { + return ErrnoToSceKernelError(*__Error()); + } + return ORBIS_OK; +} + +s32 PS4_SYSV_ABI sceKernelConvertLocaltimeToUtc(time_t param_1, int64_t param_2, time_t* seconds, + OrbisKernelTimezone* timezone, s32* dst_seconds) { LOG_INFO(Kernel, "called"); if (timezone) { sceKernelGettimezone(timezone); param_1 -= (timezone->tz_minuteswest + timezone->tz_dsttime) * 60; - if (seconds) + if (seconds) { *seconds = param_1; - if (dst_seconds) + } + if (dst_seconds) { *dst_seconds = timezone->tz_dsttime * 60; + } } else { return ORBIS_KERNEL_ERROR_EINVAL; } @@ -415,7 +482,7 @@ Common::NativeClock* GetClock() { } // namespace Dev -int PS4_SYSV_ABI sceKernelConvertUtcToLocaltime(time_t time, time_t* local_time, +s32 PS4_SYSV_ABI sceKernelConvertUtcToLocaltime(time_t time, time_t* local_time, struct OrbisTimesec* st, u64* dst_sec) { LOG_TRACE(Kernel, "Called"); #ifdef __APPLE__ @@ -444,28 +511,35 @@ int PS4_SYSV_ABI sceKernelConvertUtcToLocaltime(time_t time, time_t* local_time, void RegisterTime(Core::Loader::SymbolsResolver* sym) { clock = std::make_unique(); initial_ptc = clock->GetUptime(); + + // POSIX + LIB_FUNCTION("yS8U2TGCe1A", "libkernel", 1, "libkernel", 1, 1, posix_nanosleep); + LIB_FUNCTION("yS8U2TGCe1A", "libScePosix", 1, "libkernel", 1, 1, posix_nanosleep); + LIB_FUNCTION("QcteRwbsnV0", "libkernel", 1, "libkernel", 1, 1, posix_usleep); + LIB_FUNCTION("QcteRwbsnV0", "libScePosix", 1, "libkernel", 1, 1, posix_usleep); + LIB_FUNCTION("0wu33hunNdE", "libkernel", 1, "libkernel", 1, 1, posix_sleep); + LIB_FUNCTION("0wu33hunNdE", "libScePosix", 1, "libkernel", 1, 1, posix_sleep); + LIB_FUNCTION("lLMT9vJAck0", "libkernel", 1, "libkernel", 1, 1, posix_clock_gettime); + LIB_FUNCTION("lLMT9vJAck0", "libScePosix", 1, "libkernel", 1, 1, posix_clock_gettime); + LIB_FUNCTION("smIj7eqzZE8", "libkernel", 1, "libkernel", 1, 1, posix_clock_getres); + LIB_FUNCTION("smIj7eqzZE8", "libScePosix", 1, "libkernel", 1, 1, posix_clock_getres); + LIB_FUNCTION("n88vx3C5nW8", "libkernel", 1, "libkernel", 1, 1, posix_gettimeofday); + LIB_FUNCTION("n88vx3C5nW8", "libScePosix", 1, "libkernel", 1, 1, posix_gettimeofday); + + // Orbis LIB_FUNCTION("4J2sUJmuHZQ", "libkernel", 1, "libkernel", 1, 1, sceKernelGetProcessTime); LIB_FUNCTION("fgxnMeTNUtY", "libkernel", 1, "libkernel", 1, 1, sceKernelGetProcessTimeCounter); LIB_FUNCTION("BNowx2l588E", "libkernel", 1, "libkernel", 1, 1, sceKernelGetProcessTimeCounterFrequency); LIB_FUNCTION("-2IRUCO--PM", "libkernel", 1, "libkernel", 1, 1, sceKernelReadTsc); LIB_FUNCTION("1j3S3n-tTW4", "libkernel", 1, "libkernel", 1, 1, sceKernelGetTscFrequency); - LIB_FUNCTION("ejekcaNQNq0", "libkernel", 1, "libkernel", 1, 1, sceKernelGettimeofday); - LIB_FUNCTION("n88vx3C5nW8", "libkernel", 1, "libkernel", 1, 1, gettimeofday); - LIB_FUNCTION("n88vx3C5nW8", "libScePosix", 1, "libkernel", 1, 1, gettimeofday); LIB_FUNCTION("QvsZxomvUHs", "libkernel", 1, "libkernel", 1, 1, sceKernelNanosleep); LIB_FUNCTION("1jfXLRVzisc", "libkernel", 1, "libkernel", 1, 1, sceKernelUsleep); - LIB_FUNCTION("QcteRwbsnV0", "libkernel", 1, "libkernel", 1, 1, posix_usleep); - LIB_FUNCTION("QcteRwbsnV0", "libScePosix", 1, "libkernel", 1, 1, posix_usleep); LIB_FUNCTION("-ZR+hG7aDHw", "libkernel", 1, "libkernel", 1, 1, sceKernelSleep); - LIB_FUNCTION("0wu33hunNdE", "libScePosix", 1, "libkernel", 1, 1, sceKernelSleep); - LIB_FUNCTION("yS8U2TGCe1A", "libkernel", 1, "libkernel", 1, 1, posix_nanosleep); - LIB_FUNCTION("yS8U2TGCe1A", "libScePosix", 1, "libkernel", 1, 1, posix_nanosleep); LIB_FUNCTION("QBi7HCK03hw", "libkernel", 1, "libkernel", 1, 1, sceKernelClockGettime); + LIB_FUNCTION("wRYVA5Zolso", "libkernel", 1, "libkernel", 1, 1, sceKernelClockGetres); + LIB_FUNCTION("ejekcaNQNq0", "libkernel", 1, "libkernel", 1, 1, sceKernelGettimeofday); LIB_FUNCTION("kOcnerypnQA", "libkernel", 1, "libkernel", 1, 1, sceKernelGettimezone); - LIB_FUNCTION("lLMT9vJAck0", "libkernel", 1, "libkernel", 1, 1, orbis_clock_gettime); - LIB_FUNCTION("lLMT9vJAck0", "libScePosix", 1, "libkernel", 1, 1, orbis_clock_gettime); - LIB_FUNCTION("smIj7eqzZE8", "libScePosix", 1, "libkernel", 1, 1, posix_clock_getres); LIB_FUNCTION("0NTHN1NKONI", "libkernel", 1, "libkernel", 1, 1, sceKernelConvertLocaltimeToUtc); LIB_FUNCTION("-o5uEDpN+oY", "libkernel", 1, "libkernel", 1, 1, sceKernelConvertUtcToLocaltime); } diff --git a/src/core/libraries/kernel/time.h b/src/core/libraries/kernel/time.h index 407b6f9ed..c80de7bc4 100644 --- a/src/core/libraries/kernel/time.h +++ b/src/core/libraries/kernel/time.h @@ -75,14 +75,14 @@ u64 PS4_SYSV_ABI sceKernelGetProcessTime(); u64 PS4_SYSV_ABI sceKernelGetProcessTimeCounter(); u64 PS4_SYSV_ABI sceKernelGetProcessTimeCounterFrequency(); u64 PS4_SYSV_ABI sceKernelReadTsc(); -int PS4_SYSV_ABI sceKernelClockGettime(s32 clock_id, OrbisKernelTimespec* tp); +s32 PS4_SYSV_ABI sceKernelClockGettime(u32 clock_id, OrbisKernelTimespec* tp); s32 PS4_SYSV_ABI sceKernelGettimezone(OrbisKernelTimezone* tz); -int PS4_SYSV_ABI sceKernelConvertLocaltimeToUtc(time_t param_1, int64_t param_2, time_t* seconds, - OrbisKernelTimezone* timezone, int* dst_seconds); +s32 PS4_SYSV_ABI sceKernelConvertLocaltimeToUtc(time_t param_1, int64_t param_2, time_t* seconds, + OrbisKernelTimezone* timezone, s32* dst_seconds); -int PS4_SYSV_ABI sceKernelConvertUtcToLocaltime(time_t time, time_t* local_time, OrbisTimesec* st, +s32 PS4_SYSV_ABI sceKernelConvertUtcToLocaltime(time_t time, time_t* local_time, OrbisTimesec* st, u64* dst_sec); -int PS4_SYSV_ABI sceKernelUsleep(u32 microseconds); +s32 PS4_SYSV_ABI sceKernelUsleep(u32 microseconds); void RegisterTime(Core::Loader::SymbolsResolver* sym); diff --git a/src/core/libraries/libs.cpp b/src/core/libraries/libs.cpp index 5ef4b259d..2ab46d3a0 100644 --- a/src/core/libraries/libs.cpp +++ b/src/core/libraries/libs.cpp @@ -9,6 +9,7 @@ #include "core/libraries/audio3d/audio3d.h" #include "core/libraries/avplayer/avplayer.h" #include "core/libraries/camera/camera.h" +#include "core/libraries/companion/companion_httpd.h" #include "core/libraries/disc_map/disc_map.h" #include "core/libraries/game_live_streaming/gamelivestreaming.h" #include "core/libraries/gnmdriver/gnmdriver.h" @@ -124,6 +125,7 @@ void InitHLELibs(Core::Loader::SymbolsResolver* sym) { Libraries::Ulobjmgr::RegisterlibSceUlobjmgr(sym); Libraries::SigninDialog::RegisterlibSceSigninDialog(sym); Libraries::Camera::RegisterlibSceCamera(sym); + Libraries::CompanionHttpd::RegisterlibSceCompanionHttpd(sym); } } // namespace Libraries diff --git a/src/core/libraries/ngs2/ngs2.cpp b/src/core/libraries/ngs2/ngs2.cpp index 743be5fd6..9bb73536c 100644 --- a/src/core/libraries/ngs2/ngs2.cpp +++ b/src/core/libraries/ngs2/ngs2.cpp @@ -380,8 +380,7 @@ s32 PS4_SYSV_ABI sceNgs2GeomApply(const OrbisNgs2GeomListenerWork* listener, s32 PS4_SYSV_ABI sceNgs2PanInit(OrbisNgs2PanWork* work, const float* aSpeakerAngle, float unitAngle, u32 numSpeakers) { - LOG_ERROR(Lib_Ngs2, "aSpeakerAngle = {}, unitAngle = {}, numSpeakers = {}", *aSpeakerAngle, - unitAngle, numSpeakers); + LOG_ERROR(Lib_Ngs2, "unitAngle = {}, numSpeakers = {}", unitAngle, numSpeakers); return ORBIS_OK; } diff --git a/src/core/libraries/np_trophy/np_trophy.cpp b/src/core/libraries/np_trophy/np_trophy.cpp index a951d5655..6de84bd93 100644 --- a/src/core/libraries/np_trophy/np_trophy.cpp +++ b/src/core/libraries/np_trophy/np_trophy.cpp @@ -206,6 +206,10 @@ s32 PS4_SYSV_ABI sceNpTrophyDestroyHandle(OrbisNpTrophyHandle handle) { if (handle == ORBIS_NP_TROPHY_INVALID_HANDLE) return ORBIS_NP_TROPHY_ERROR_INVALID_HANDLE; + if (handle >= trophy_handles.size()) { + LOG_ERROR(Lib_NpTrophy, "Invalid handle {}", handle); + return ORBIS_NP_TROPHY_ERROR_INVALID_HANDLE; + } if (!trophy_handles.is_allocated({static_cast(handle)})) { return ORBIS_NP_TROPHY_ERROR_INVALID_HANDLE; } diff --git a/src/core/libraries/save_data/dialog/savedatadialog_ui.cpp b/src/core/libraries/save_data/dialog/savedatadialog_ui.cpp index edb5caa07..05df67eeb 100644 --- a/src/core/libraries/save_data/dialog/savedatadialog_ui.cpp +++ b/src/core/libraries/save_data/dialog/savedatadialog_ui.cpp @@ -155,7 +155,7 @@ SaveDialogState::SaveDialogState(const OrbisSaveDataDialogParam& param) { if (item->focusPos != FocusPos::DIRNAME) { this->focus_pos = item->focusPos; - } else { + } else if (item->focusPosDirName != nullptr) { this->focus_pos = item->focusPosDirName->data.to_string(); } this->style = item->itemStyle; diff --git a/src/core/libraries/save_data/save_memory.cpp b/src/core/libraries/save_data/save_memory.cpp index 13e122c60..ab3ce2d4f 100644 --- a/src/core/libraries/save_data/save_memory.cpp +++ b/src/core/libraries/save_data/save_memory.cpp @@ -10,15 +10,14 @@ #include #include -#include - -#include "common/assert.h" +#include "boost/icl/concept/interval.hpp" #include "common/elf_info.h" #include "common/logging/log.h" #include "common/path_util.h" #include "common/singleton.h" #include "common/thread.h" #include "core/file_sys/fs.h" +#include "core/libraries/system/msgdialog_ui.h" #include "save_instance.h" using Common::FS::IOFile; @@ -35,11 +34,12 @@ namespace Libraries::SaveData::SaveMemory { static Core::FileSys::MntPoints* g_mnt = Common::Singleton::Instance(); struct SlotData { - OrbisUserServiceUserId user_id; + OrbisUserServiceUserId user_id{}; std::string game_serial; std::filesystem::path folder_path; PSF sfo; std::vector memory_cache; + size_t memory_cache_size{}; }; static std::mutex g_slot_mtx; @@ -97,7 +97,8 @@ std::filesystem::path GetSavePath(OrbisUserServiceUserId user_id, u32 slot_id, return SaveInstance::MakeDirSavePath(user_id, Common::ElfInfo::Instance().GameSerial(), dir); } -size_t SetupSaveMemory(OrbisUserServiceUserId user_id, u32 slot_id, std::string_view game_serial) { +size_t SetupSaveMemory(OrbisUserServiceUserId user_id, u32 slot_id, std::string_view game_serial, + size_t memory_size) { std::lock_guard lck{g_slot_mtx}; const auto save_dir = GetSavePath(user_id, slot_id, game_serial); @@ -109,6 +110,7 @@ size_t SetupSaveMemory(OrbisUserServiceUserId user_id, u32 slot_id, std::string_ .folder_path = save_dir, .sfo = {}, .memory_cache = {}, + .memory_cache_size = memory_size, }; SaveInstance::SetupDefaultParamSFO(data.sfo, GetSaveDir(slot_id), std::string{game_serial}); @@ -196,9 +198,9 @@ void ReadMemory(u32 slot_id, void* buf, size_t buf_size, int64_t offset) { auto& data = g_attached_slots[slot_id]; auto& memory = data.memory_cache; if (memory.empty()) { // Load file + memory.resize(data.memory_cache_size); IOFile f{data.folder_path / FilenameSaveDataMemory, Common::FS::FileAccessMode::Read}; if (f.IsOpen()) { - memory.resize(f.GetSize()); f.Seek(0); f.ReadSpan(std::span{memory}); } @@ -222,5 +224,4 @@ void WriteMemory(u32 slot_id, void* buf, size_t buf_size, int64_t offset) { Backup::NewRequest(data.user_id, data.game_serial, GetSaveDir(slot_id), Backup::OrbisSaveDataEventType::__DO_NOT_SAVE); } - } // namespace Libraries::SaveData::SaveMemory \ No newline at end of file diff --git a/src/core/libraries/save_data/save_memory.h b/src/core/libraries/save_data/save_memory.h index 681865634..7765b04cd 100644 --- a/src/core/libraries/save_data/save_memory.h +++ b/src/core/libraries/save_data/save_memory.h @@ -4,13 +4,13 @@ #pragma once #include -#include "save_backup.h" +#include "core/libraries/save_data/save_backup.h" class PSF; namespace Libraries::SaveData { using OrbisUserServiceUserId = s32; -} +} // namespace Libraries::SaveData namespace Libraries::SaveData::SaveMemory { @@ -22,7 +22,8 @@ void PersistMemory(u32 slot_id, bool lock = true); std::string_view game_serial); // returns the size of the save memory if exists -size_t SetupSaveMemory(OrbisUserServiceUserId user_id, u32 slot_id, std::string_view game_serial); +size_t SetupSaveMemory(OrbisUserServiceUserId user_id, u32 slot_id, std::string_view game_serial, + size_t memory_size); // Write the icon. Set buf to null to read the standard icon. void SetIcon(u32 slot_id, void* buf = nullptr, size_t buf_size = 0); diff --git a/src/core/libraries/save_data/savedata.cpp b/src/core/libraries/save_data/savedata.cpp index e9ad77d69..b25ebde6c 100644 --- a/src/core/libraries/save_data/savedata.cpp +++ b/src/core/libraries/save_data/savedata.cpp @@ -5,10 +5,10 @@ #include #include -#include #include #include "common/assert.h" +#include "common/config.h" #include "common/cstring.h" #include "common/elf_info.h" #include "common/enum.h" @@ -20,7 +20,9 @@ #include "core/libraries/error_codes.h" #include "core/libraries/libs.h" #include "core/libraries/save_data/savedata.h" +#include "core/libraries/save_data/savedata_error.h" #include "core/libraries/system/msgdialog.h" +#include "core/libraries/system/msgdialog_ui.h" #include "save_backup.h" #include "save_instance.h" #include "save_memory.h" @@ -33,27 +35,6 @@ using Common::ElfInfo; namespace Libraries::SaveData { -enum class Error : u32 { - OK = 0, - USER_SERVICE_NOT_INITIALIZED = 0x80960002, - PARAMETER = 0x809F0000, - NOT_INITIALIZED = 0x809F0001, - OUT_OF_MEMORY = 0x809F0002, - BUSY = 0x809F0003, - NOT_MOUNTED = 0x809F0004, - EXISTS = 0x809F0007, - NOT_FOUND = 0x809F0008, - NO_SPACE_FS = 0x809F000A, - INTERNAL = 0x809F000B, - MOUNT_FULL = 0x809F000C, - BAD_MOUNTED = 0x809F000D, - BROKEN = 0x809F000F, - INVALID_LOGIN_USER = 0x809F0011, - MEMORY_NOT_READY = 0x809F0012, - BACKUP_BUSY = 0x809F0013, - BUSY_FOR_SAVING = 0x809F0016, -}; - enum class OrbisSaveDataSaveDataMemoryOption : u32 { NONE = 0, SET_PARAM = 1 << 0, @@ -336,7 +317,9 @@ static std::array, 16> g_mount_slots; static void initialize() { g_initialized = true; - g_game_serial = ElfInfo::Instance().GameSerial(); + g_game_serial = Common::Singleton::Instance() + ->GetString("INSTALL_DIR_SAVEDATA") + .value_or(ElfInfo::Instance().GameSerial()); g_fw_ver = ElfInfo::Instance().FirmwareVer(); Backup::StartThread(); } @@ -456,7 +439,7 @@ static Error saveDataMount(const OrbisSaveDataMount2* mount_info, LOG_INFO(Lib_SaveData, "called with invalid block size"); } - const auto root_save = Common::FS::GetUserPath(Common::FS::PathType::SaveDataDir); + const auto root_save = Config::GetSaveDataPath(); fs::create_directories(root_save); const auto available = fs::space(root_save).available; @@ -1593,8 +1576,8 @@ Error PS4_SYSV_ABI sceSaveDataSetupSaveDataMemory2(const OrbisSaveDataMemorySetu } try { - size_t existed_size = - SaveMemory::SetupSaveMemory(setupParam->userId, slot_id, g_game_serial); + size_t existed_size = SaveMemory::SetupSaveMemory(setupParam->userId, slot_id, + g_game_serial, setupParam->memorySize); if (existed_size == 0) { // Just created if (g_fw_ver >= ElfInfo::FW_45 && setupParam->initParam != nullptr) { auto& sfo = SaveMemory::GetParamSFO(slot_id); diff --git a/src/core/libraries/save_data/savedata_error.h b/src/core/libraries/save_data/savedata_error.h new file mode 100644 index 000000000..ef347e855 --- /dev/null +++ b/src/core/libraries/save_data/savedata_error.h @@ -0,0 +1,27 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#pragma once + +namespace Libraries::SaveData { +enum class Error : u32 { + OK = 0, + USER_SERVICE_NOT_INITIALIZED = 0x80960002, + PARAMETER = 0x809F0000, + NOT_INITIALIZED = 0x809F0001, + OUT_OF_MEMORY = 0x809F0002, + BUSY = 0x809F0003, + NOT_MOUNTED = 0x809F0004, + EXISTS = 0x809F0007, + NOT_FOUND = 0x809F0008, + NO_SPACE_FS = 0x809F000A, + INTERNAL = 0x809F000B, + MOUNT_FULL = 0x809F000C, + BAD_MOUNTED = 0x809F000D, + BROKEN = 0x809F000F, + INVALID_LOGIN_USER = 0x809F0011, + MEMORY_NOT_READY = 0x809F0012, + BACKUP_BUSY = 0x809F0013, + BUSY_FOR_SAVING = 0x809F0016, +}; +} // namespace Libraries::SaveData diff --git a/src/core/linker.cpp b/src/core/linker.cpp index eced87968..3e6d8c22e 100644 --- a/src/core/linker.cpp +++ b/src/core/linker.cpp @@ -12,6 +12,7 @@ #include "common/thread.h" #include "core/aerolib/aerolib.h" #include "core/aerolib/stubs.h" +#include "core/devtools/widget/module_list.h" #include "core/libraries/kernel/memory.h" #include "core/libraries/kernel/threads.h" #include "core/linker.h" @@ -147,6 +148,9 @@ s32 Linker::LoadModule(const std::filesystem::path& elf_name, bool is_dynamic) { num_static_modules += !is_dynamic; m_modules.emplace_back(std::move(module)); + + Core::Devtools::Widget::ModuleList::AddModule(elf_name.filename().string(), elf_name); + return m_modules.size() - 1; } @@ -325,6 +329,9 @@ bool Linker::Resolve(const std::string& name, Loader::SymbolType sym_type, Modul } if (record) { *return_info = *record; + + Core::Devtools::Widget::ModuleList::AddModule(sr.library); + return true; } diff --git a/src/core/memory.cpp b/src/core/memory.cpp index 6438670d3..ca6a0d6cd 100644 --- a/src/core/memory.cpp +++ b/src/core/memory.cpp @@ -362,11 +362,8 @@ int MemoryManager::MapMemory(void** out_addr, VAddr virtual_addr, size_t size, M return ORBIS_KERNEL_ERROR_ENOMEM; } - // When virtual addr is zero, force it to virtual_base. The guest cannot pass Fixed - // flag so we will take the branch that searches for free (or reserved) mappings. - virtual_addr = (virtual_addr == 0) ? impl.SystemManagedVirtualBase() : virtual_addr; - alignment = alignment > 0 ? alignment : 16_KB; - VAddr mapped_addr = alignment > 0 ? Common::AlignUp(virtual_addr, alignment) : virtual_addr; + // Limit the minumum address to SystemManagedVirtualBase to prevent hardware-specific issues. + VAddr mapped_addr = (virtual_addr == 0) ? impl.SystemManagedVirtualBase() : virtual_addr; // Fixed mapping means the virtual address must exactly match the provided one. if (True(flags & MemoryMapFlags::Fixed)) { @@ -376,24 +373,29 @@ int MemoryManager::MapMemory(void** out_addr, VAddr virtual_addr, size_t size, M // To account for this, unmap any reserved areas within this mapping range first. auto unmap_addr = mapped_addr; auto unmap_size = size; - while (!vma.IsMapped() && unmap_addr < mapped_addr + size && remaining_size < size) { + // If flag NoOverwrite is provided, don't overwrite mapped VMAs. + // When it isn't provided, VMAs can be overwritten regardless of if they're mapped. + while ((False(flags & MemoryMapFlags::NoOverwrite) || !vma.IsMapped()) && + unmap_addr < mapped_addr + size && remaining_size < size) { auto unmapped = UnmapBytesFromEntry(unmap_addr, vma, unmap_size); unmap_addr += unmapped; unmap_size -= unmapped; vma = FindVMA(unmap_addr)->second; } - // This should return SCE_KERNEL_ERROR_ENOMEM but rarely happens. vma = FindVMA(mapped_addr)->second; remaining_size = vma.base + vma.size - mapped_addr; - ASSERT_MSG(!vma.IsMapped() && remaining_size >= size, - "Memory region {:#x} to {:#x} isn't free enough to map region {:#x} to {:#x}", - vma.base, vma.base + vma.size, virtual_addr, virtual_addr + size); + if (vma.IsMapped() || remaining_size < size) { + LOG_ERROR(Kernel_Vmm, "Unable to map {:#x} bytes at address {:#x}", size, mapped_addr); + return ORBIS_KERNEL_ERROR_ENOMEM; + } } // Find the first free area starting with provided virtual address. if (False(flags & MemoryMapFlags::Fixed)) { - mapped_addr = SearchFree(mapped_addr, size, alignment); + // Provided address needs to be aligned before we can map. + alignment = alignment > 0 ? alignment : 16_KB; + mapped_addr = SearchFree(Common::AlignUp(mapped_addr, alignment), size, alignment); if (mapped_addr == -1) { // No suitable memory areas to map to return ORBIS_KERNEL_ERROR_ENOMEM; @@ -783,6 +785,19 @@ int MemoryManager::DirectQueryAvailable(PAddr search_start, PAddr search_end, si return ORBIS_OK; } +s32 MemoryManager::SetDirectMemoryType(s64 phys_addr, s32 memory_type) { + std::scoped_lock lk{mutex}; + + auto& dmem_area = FindDmemArea(phys_addr)->second; + + ASSERT_MSG(phys_addr <= dmem_area.GetEnd() && !dmem_area.is_free, + "Direct memory area is not mapped"); + + dmem_area.memory_type = memory_type; + + return ORBIS_OK; +} + void MemoryManager::NameVirtualRange(VAddr virtual_addr, size_t size, std::string_view name) { auto it = FindVMA(virtual_addr); @@ -949,4 +964,33 @@ int MemoryManager::GetDirectMemoryType(PAddr addr, int* directMemoryTypeOut, return ORBIS_OK; } +int MemoryManager::IsStack(VAddr addr, void** start, void** end) { + auto vma_handle = FindVMA(addr); + if (vma_handle == vma_map.end()) { + return ORBIS_KERNEL_ERROR_EINVAL; + } + + const VirtualMemoryArea& vma = vma_handle->second; + if (!vma.Contains(addr, 0) || vma.IsFree()) { + return ORBIS_KERNEL_ERROR_EACCES; + } + + auto stack_start = 0ul; + auto stack_end = 0ul; + if (vma.type == VMAType::Stack) { + stack_start = vma.base; + stack_end = vma.base + vma.size; + } + + if (start != nullptr) { + *start = reinterpret_cast(stack_start); + } + + if (end != nullptr) { + *end = reinterpret_cast(stack_end); + } + + return ORBIS_OK; +} + } // namespace Core diff --git a/src/core/memory.h b/src/core/memory.h index 4920aa397..883b48854 100644 --- a/src/core/memory.h +++ b/src/core/memory.h @@ -219,10 +219,14 @@ public: int GetDirectMemoryType(PAddr addr, int* directMemoryTypeOut, void** directMemoryStartOut, void** directMemoryEndOut); + s32 SetDirectMemoryType(s64 phys_addr, s32 memory_type); + void NameVirtualRange(VAddr virtual_addr, size_t size, std::string_view name); void InvalidateMemory(VAddr addr, u64 size) const; + int IsStack(VAddr addr, void** start, void** end); + private: VMAHandle FindVMA(VAddr target) { return std::prev(vma_map.upper_bound(target)); diff --git a/src/core/tls.h b/src/core/tls.h index e9e2b9e6a..470553d85 100644 --- a/src/core/tls.h +++ b/src/core/tls.h @@ -53,7 +53,7 @@ template ReturnType ExecuteGuest(PS4_SYSV_ABI ReturnType (*func)(FuncArgs...), CallArgs&&... args) { EnsureThreadInitialized(); // clear stack to avoid trash from EnsureThreadInitialized - ClearStack<13_KB>(); + ClearStack<12_KB>(); return func(std::forward(args)...); } diff --git a/src/emulator.cpp b/src/emulator.cpp index ebb34054b..2ad8446ab 100644 --- a/src/emulator.cpp +++ b/src/emulator.cpp @@ -25,6 +25,7 @@ #include "common/polyfill_thread.h" #include "common/scm_rev.h" #include "common/singleton.h" +#include "core/devtools/widget/module_list.h" #include "core/file_format/psf.h" #include "core/file_format/trp.h" #include "core/file_sys/fs.h" @@ -188,18 +189,28 @@ void Emulator::Run(const std::filesystem::path& file, const std::vector", id, title, app_version); std::string window_title = ""; - if (Common::g_is_release) { - window_title = fmt::format("shadPS4 v{} | {}", Common::g_version, game_title); - } else { - std::string remote_url(Common::g_scm_remote_url); - std::string remote_host; - try { - remote_host = remote_url.substr(19, remote_url.rfind('/') - 19); - } catch (...) { - remote_host = "unknown"; + std::string remote_url(Common::g_scm_remote_url); + std::string remote_host; + try { + if (*remote_url.rbegin() == '/') { + remote_url.pop_back(); } + remote_host = remote_url.substr(19, remote_url.rfind('/') - 19); + } catch (...) { + remote_host = "unknown"; + } + if (Common::g_is_release) { + if (remote_host == "shadps4-emu" || remote_url.length() == 0) { + window_title = fmt::format("shadPS4 v{} | {}", Common::g_version, game_title); + } else { + window_title = + fmt::format("shadPS4 {}/v{} | {}", remote_host, Common::g_version, game_title); + } + } else { if (remote_host == "shadps4-emu" || remote_url.length() == 0) { window_title = fmt::format("shadPS4 v{} {} {} | {}", Common::g_version, Common::g_scm_branch, Common::g_scm_desc, game_title); diff --git a/src/images/KBM.png b/src/images/KBM.png index 37f52d549..feab9fa0f 100644 Binary files a/src/images/KBM.png and b/src/images/KBM.png differ diff --git a/src/qt_gui/game_info.h b/src/qt_gui/game_info.h index 09e5a4557..723142e1c 100644 --- a/src/qt_gui/game_info.h +++ b/src/qt_gui/game_info.h @@ -79,6 +79,11 @@ public: if (const auto play_time = psf.GetString("PLAY_TIME"); play_time.has_value()) { game.play_time = *play_time; } + if (const auto save_dir = psf.GetString("INSTALL_DIR_SAVEDATA"); save_dir.has_value()) { + game.save_dir = *save_dir; + } else { + game.save_dir = game.serial; + } } return game; } diff --git a/src/qt_gui/game_list_utils.h b/src/qt_gui/game_list_utils.h index 804f0e4b7..e19cf364e 100644 --- a/src/qt_gui/game_list_utils.h +++ b/src/qt_gui/game_list_utils.h @@ -25,6 +25,7 @@ struct GameInfo { std::string version = "Unknown"; std::string region = "Unknown"; std::string fw = "Unknown"; + std::string save_dir = "Unknown"; std::string play_time = "Unknown"; CompatibilityEntry compatibility = CompatibilityEntry{CompatibilityStatus::Unknown}; diff --git a/src/qt_gui/gui_context_menus.h b/src/qt_gui/gui_context_menus.h index 2fd4588d2..46a40c5cd 100644 --- a/src/qt_gui/gui_context_menus.h +++ b/src/qt_gui/gui_context_menus.h @@ -156,11 +156,9 @@ public: } if (selected == openSaveDataFolder) { - QString userPath; - Common::FS::PathToQString(userPath, - Common::FS::GetUserPath(Common::FS::PathType::UserDir)); - QString saveDataPath = - userPath + "/savedata/1/" + QString::fromStdString(m_games[itemID].serial); + QString saveDataPath; + Common::FS::PathToQString(saveDataPath, + Config::GetSaveDataPath() / "1" / m_games[itemID].save_dir); QDir(saveDataPath).mkpath(saveDataPath); QDesktopServices::openUrl(QUrl::fromLocalFile(saveDataPath)); } @@ -485,8 +483,7 @@ public: dlc_path, Config::getAddonInstallDir() / Common::FS::PathFromQString(folder_path).parent_path().filename()); Common::FS::PathToQString(save_data_path, - Common::FS::GetUserPath(Common::FS::PathType::UserDir) / - "savedata/1" / m_games[itemID].serial); + Config::GetSaveDataPath() / "1" / m_games[itemID].save_dir); Common::FS::PathToQString(trophy_data_path, Common::FS::GetUserPath(Common::FS::PathType::MetaDataDir) / diff --git a/src/qt_gui/kbm_gui.ui b/src/qt_gui/kbm_gui.ui index c8d63cd00..109423aa8 100644 --- a/src/qt_gui/kbm_gui.ui +++ b/src/qt_gui/kbm_gui.ui @@ -825,6 +825,9 @@ 0 + + 48 + Qt::FocusPolicy::NoFocus @@ -841,6 +844,9 @@ 0 + + 48 + Qt::FocusPolicy::NoFocus @@ -983,8 +989,8 @@ - 500 - 200 + 424 + 250 diff --git a/src/qt_gui/main_window.cpp b/src/qt_gui/main_window.cpp index 8eeec3536..1966aa52b 100644 --- a/src/qt_gui/main_window.cpp +++ b/src/qt_gui/main_window.cpp @@ -55,16 +55,23 @@ bool MainWindow::Init() { // show ui setMinimumSize(720, 405); std::string window_title = ""; - if (Common::g_is_release) { - window_title = fmt::format("shadPS4 v{}", Common::g_version); - } else { - std::string remote_url(Common::g_scm_remote_url); - std::string remote_host; - try { - remote_host = remote_url.substr(19, remote_url.rfind('/') - 19); - } catch (...) { - remote_host = "unknown"; + std::string remote_url(Common::g_scm_remote_url); + std::string remote_host; + try { + if (*remote_url.rbegin() == '/') { + remote_url.pop_back(); } + remote_host = remote_url.substr(19, remote_url.rfind('/') - 19); + } catch (...) { + remote_host = "unknown"; + } + if (Common::g_is_release) { + if (remote_host == "shadps4-emu" || remote_url.length() == 0) { + window_title = fmt::format("shadPS4 v{}", Common::g_version); + } else { + window_title = fmt::format("shadPS4 {}/v{}", remote_host, Common::g_version); + } + } else { if (remote_host == "shadps4-emu" || remote_url.length() == 0) { window_title = fmt::format("shadPS4 v{} {} {}", Common::g_version, Common::g_scm_branch, Common::g_scm_desc); diff --git a/src/qt_gui/translations/ar_SA.ts b/src/qt_gui/translations/ar_SA.ts index e434b3259..26e768720 100644 --- a/src/qt_gui/translations/ar_SA.ts +++ b/src/qt_gui/translations/ar_SA.ts @@ -1347,10 +1347,6 @@ Game List قائمة الألعاب - - * Unsupported Vulkan Version - * إصدار Vulkan غير مدعوم - Download Cheats For All Installed Games تحميل الشفرات لجميع الألعاب المثبتة @@ -2051,6 +2047,10 @@ Nightly: نُسخ تحتوي على أحدث الميزات، لكنها أقل Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. افتح مجلد الصور/الأصوات الخاصة بالجوائز المخصصة:\nيمكنك إضافة صور مخصصة للجوائز وصوت مرفق.\nأضف الملفات إلى مجلد custom_trophy بالأسماء التالية:\ntrophy.wav أو trophy.mp3، bronze.png، gold.png، platinum.png، silver.png\nملاحظة: الصوت سيعمل فقط في الإصدارات التي تستخدم QT. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/da_DK.ts b/src/qt_gui/translations/da_DK.ts index 131a989e1..1023c584b 100644 --- a/src/qt_gui/translations/da_DK.ts +++ b/src/qt_gui/translations/da_DK.ts @@ -1347,10 +1347,6 @@ Game List Spiloversigt - - * Unsupported Vulkan Version - * Ikke understøttet Vulkan-version - Download Cheats For All Installed Games Hent snyd til alle installerede spil @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/de_DE.ts b/src/qt_gui/translations/de_DE.ts index c7a18dd99..1d44eb717 100644 --- a/src/qt_gui/translations/de_DE.ts +++ b/src/qt_gui/translations/de_DE.ts @@ -1347,10 +1347,6 @@ Game List Spieleliste - - * Unsupported Vulkan Version - * Nicht unterstützte Vulkan-Version - Download Cheats For All Installed Games Cheats für alle installierten Spiele herunterladen @@ -2054,6 +2050,10 @@ Fügen Sie die Dateien dem Ordner custom_trophy mit folgenden Namen hinzu:\n trophy.wav ODER trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\n Hinweis: Der Sound funktioniert nur in Qt-Versionen. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/el_GR.ts b/src/qt_gui/translations/el_GR.ts index c91e0c731..765185c9e 100644 --- a/src/qt_gui/translations/el_GR.ts +++ b/src/qt_gui/translations/el_GR.ts @@ -1347,10 +1347,6 @@ Game List Λίστα παιχνιδιών - - * Unsupported Vulkan Version - * Μη υποστηριζόμενη έκδοση Vulkan - Download Cheats For All Installed Games Λήψη Cheats για όλα τα εγκατεστημένα παιχνίδια @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/en_US.ts b/src/qt_gui/translations/en_US.ts index 780f089e8..cc854120f 100644 --- a/src/qt_gui/translations/en_US.ts +++ b/src/qt_gui/translations/en_US.ts @@ -1347,10 +1347,6 @@ Game List Game List - - * Unsupported Vulkan Version - * Unsupported Vulkan Version - Download Cheats For All Installed Games Download Cheats For All Installed Games @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/es_ES.ts b/src/qt_gui/translations/es_ES.ts index 035aac6a3..e73386c96 100644 --- a/src/qt_gui/translations/es_ES.ts +++ b/src/qt_gui/translations/es_ES.ts @@ -1347,10 +1347,6 @@ Game List Lista de Juegos - - * Unsupported Vulkan Version - * Versión de Vulkan no soportada - Download Cheats For All Installed Games Descargar trucos para todos los juegos instalados @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Abre la carpeta de trofeos/sonidos personalizados:\nPuedes añadir imágenes y un audio personalizados a los trofeos.\nAñade los archivos a custom_trophy con los siguientes nombres:\ntrophy.wav o trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNota: El sonido sólo funcionará en versiones QT. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/fa_IR.ts b/src/qt_gui/translations/fa_IR.ts index 552a0ff23..b9c2282fa 100644 --- a/src/qt_gui/translations/fa_IR.ts +++ b/src/qt_gui/translations/fa_IR.ts @@ -1347,10 +1347,6 @@ Game List لیست بازی - - * Unsupported Vulkan Version - شما پشتیبانی نمیشود Vulkan ورژن * - Download Cheats For All Installed Games دانلود چیت برای همه بازی ها @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/fi_FI.ts b/src/qt_gui/translations/fi_FI.ts index 44c668560..c77c63b3e 100644 --- a/src/qt_gui/translations/fi_FI.ts +++ b/src/qt_gui/translations/fi_FI.ts @@ -1347,10 +1347,6 @@ Game List Pelilista - - * Unsupported Vulkan Version - * Ei Tuettu Vulkan-versio - Download Cheats For All Installed Games Lataa Huijaukset Kaikille Asennetuille Peleille @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/fr_FR.ts b/src/qt_gui/translations/fr_FR.ts index 13e1be9f5..75e424ad0 100644 --- a/src/qt_gui/translations/fr_FR.ts +++ b/src/qt_gui/translations/fr_FR.ts @@ -1347,10 +1347,6 @@ Game List Liste de jeux - - * Unsupported Vulkan Version - * Version de Vulkan non prise en charge - Download Cheats For All Installed Games Télécharger les Cheats pour tous les jeux installés @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Ouvrez le dossier des images/sons des trophées personnalisés:\nVous pouvez ajouter des images personnalisées aux trophées et aux sons.\nAjoutez les fichiers à custom_trophy avec les noms suivants:\ntrophy.wav OU trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote : Le son ne fonctionnera que dans les versions QT. + + * Unsupported Vulkan Version + * Version de Vulkan non prise en charge + TrophyViewer diff --git a/src/qt_gui/translations/hu_HU.ts b/src/qt_gui/translations/hu_HU.ts index 58857d0d7..e396cc4f5 100644 --- a/src/qt_gui/translations/hu_HU.ts +++ b/src/qt_gui/translations/hu_HU.ts @@ -1347,10 +1347,6 @@ Game List Játéklista - - * Unsupported Vulkan Version - * Nem támogatott Vulkan verzió - Download Cheats For All Installed Games Csalások letöltése minden telepített játékhoz @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/id_ID.ts b/src/qt_gui/translations/id_ID.ts index de19824f7..b4fe48637 100644 --- a/src/qt_gui/translations/id_ID.ts +++ b/src/qt_gui/translations/id_ID.ts @@ -1347,10 +1347,6 @@ Game List Daftar game - - * Unsupported Vulkan Version - * Versi Vulkan Tidak Didukung - Download Cheats For All Installed Games Unduh Cheat Untuk Semua Game Yang Terpasang @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/it_IT.ts b/src/qt_gui/translations/it_IT.ts index 8c9e53611..a1aa0bb3f 100644 --- a/src/qt_gui/translations/it_IT.ts +++ b/src/qt_gui/translations/it_IT.ts @@ -1347,10 +1347,6 @@ Game List Elenco giochi - - * Unsupported Vulkan Version - * Versione Vulkan non supportata - Download Cheats For All Installed Games Scarica Trucchi per tutti i giochi installati @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Apri la cartella personalizzata delle immagini/suoni trofei:\nÈ possibile aggiungere immagini personalizzate ai trofei e un audio.\nAggiungi i file a custom_trophy con i seguenti nomi:\ntrophy.wav OPPURE trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNota: Il suono funzionerà solo nelle versioni QT. + + * Unsupported Vulkan Version + * Versione Vulkan non supportata + TrophyViewer diff --git a/src/qt_gui/translations/ja_JP.ts b/src/qt_gui/translations/ja_JP.ts index 146caa515..7bd7fed8a 100644 --- a/src/qt_gui/translations/ja_JP.ts +++ b/src/qt_gui/translations/ja_JP.ts @@ -1347,10 +1347,6 @@ Game List ゲームリスト - - * Unsupported Vulkan Version - * サポートされていないVulkanバージョン - Download Cheats For All Installed Games すべてのインストール済みゲームのチートをダウンロード @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/ko_KR.ts b/src/qt_gui/translations/ko_KR.ts index b79959d38..6e5b232a0 100644 --- a/src/qt_gui/translations/ko_KR.ts +++ b/src/qt_gui/translations/ko_KR.ts @@ -7,15 +7,15 @@ AboutDialog About shadPS4 - About shadPS4 + shadPS4에 관하여 shadPS4 is an experimental open-source emulator for the PlayStation 4. - shadPS4 is an experimental open-source emulator for the PlayStation 4. + shadPS4는 PlayStation 4용 실험적인 오픈 소스 에뮬레이터입니다. This software should not be used to play games you have not legally obtained. - This software should not be used to play games you have not legally obtained. + 이 소프트웨어는 합법적으로 얻지 않은 게임을 플레이하는 데 사용되어서는 안 됩니다. @@ -26,238 +26,238 @@ Cheats/Patches are experimental.\nUse with caution.\n\nDownload cheats individually by selecting the repository and clicking the download button.\nIn the Patches tab, you can download all patches at once, choose which ones you want to use, and save your selection.\n\nSince we do not develop the Cheats/Patches,\nplease report issues to the cheat author.\n\nCreated a new cheat? Visit:\n - Cheats/Patches are experimental.\nUse with caution.\n\nDownload cheats individually by selecting the repository and clicking the download button.\nIn the Patches tab, you can download all patches at once, choose which ones you want to use, and save your selection.\n\nSince we do not develop the Cheats/Patches,\nplease report issues to the cheat author.\n\nCreated a new cheat? Visit:\n + 치트/패치는 실험적인 기능입니다.\n사용 시 주의하시기 바랍니다.\n\n치트를 개별적으로 다운로드하려면, 저장소를 선택한 후 다운로드 버튼을 클릭하세요.\n패치 탭에서는 모든 패치를 한 번에 다운로드할 수 있으며, 원하는 항목을 선택하고 저장할 수 있습니다.\n\n치트/패치는 저희가 개발한 것이 아니므로,\n문제가 발생하면 해당 치트 제작자에게 문의해 주세요.\n\n새로운 치트를 만들었나요? 방문해 주세요:\n No Image Available - No Image Available + 사용 가능한 이미지 없음 Serial: - Serial: + 시리얼: Version: - Version: + 버전: Size: - Size: + 사이즈: Select Cheat File: - Select Cheat File: + 치트 파일 선택: Repository: - Repository: + 저장소: Download Cheats - Download Cheats + 치트 다운로드 Delete File - Delete File + 파일 삭제 No files selected. - No files selected. + 파일 선택되지 않음. You can delete the cheats you don't want after downloading them. - You can delete the cheats you don't want after downloading them. + 다운로드한 후 원하지 않는 치트는 삭제할 수 있습니다. Do you want to delete the selected file?\n%1 - Do you want to delete the selected file?\n%1 + 선택한 파일을 삭제하시겠습니까?\n%1 Select Patch File: - Select Patch File: + 패치 파일 선택: Download Patches - Download Patches + 패치 다운로드 Save - Save + 저장 Cheats - Cheats + 치트 Patches - Patches + 패치 Error - Error + 오류 No patch selected. - No patch selected. + 패치 선택되지 않음. Unable to open files.json for reading. - Unable to open files.json for reading. + Files.json을 읽기 위해 열 수 없습니다. No patch file found for the current serial. - No patch file found for the current serial. + 현재 시리얼에 해당하는 패치 파일을 찾을 수 없습니다. Unable to open the file for reading. - Unable to open the file for reading. + 파일을 읽기 위해 열 수 없습니다. Unable to open the file for writing. - Unable to open the file for writing. + 파일을 쓰기 위해 열 수 없습니다. Failed to parse XML: - Failed to parse XML: + XML 구문 분석에 실패했습니다: Success - Success + 성공 Options saved successfully. - Options saved successfully. + 옵션이 성공적으로 저장되었습니다. Invalid Source - Invalid Source + 잘못된 출처 The selected source is invalid. - The selected source is invalid. + 선택한 출처가 올바르지 않습니다. File Exists - File Exists + 파일이 이미 존재합니다 File already exists. Do you want to replace it? - File already exists. Do you want to replace it? + 파일이 이미 존재합니다. 덮어쓰시겠습니까? Failed to save file: - Failed to save file: + 파일 저장 실패: Failed to download file: - Failed to download file: + 파일 다운로드 실패: Cheats Not Found - Cheats Not Found + 치트 찾을 수 없음 No Cheats found for this game in this version of the selected repository,try another repository or a different version of the game. - No Cheats found for this game in this version of the selected repository,try another repository or a different version of the game. + 선택한 저장소의 이 버전에서 해당 게임에 대한 치트를 찾을 수 없습니다. 다른 저장소나 게임의 다른 버전을 시도해 보세요. Cheats Downloaded Successfully - Cheats Downloaded Successfully + 치트가 성공적으로 다운로드되었습니다 You have successfully downloaded the cheats for this version of the game from the selected repository. You can try downloading from another repository, if it is available it will also be possible to use it by selecting the file from the list. - You have successfully downloaded the cheats for this version of the game from the selected repository. You can try downloading from another repository, if it is available it will also be possible to use it by selecting the file from the list. + 선택한 저장소에서 이 게임 버전의 치트를 성공적으로 다운로드했습니다. 다른 저장소에서 다운로드할 수 있는 경우, 목록에서 파일을 선택하여 사용할 수도 있습니다. Failed to save: - Failed to save: + 저장 실패: Failed to download: - Failed to download: + 다운로드 실패: Download Complete - Download Complete + 다운로드 완료 Patches Downloaded Successfully! All Patches available for all games have been downloaded, there is no need to download them individually for each game as happens in Cheats. If the patch does not appear, it may be that it does not exist for the specific serial and version of the game. - Patches Downloaded Successfully! All Patches available for all games have been downloaded, there is no need to download them individually for each game as happens in Cheats. If the patch does not appear, it may be that it does not exist for the specific serial and version of the game. + 패치가 성공적으로 다운로드되었습니다! 모든 게임에 적용 가능한 모든 패치가 다운로드되었으므로, 치트처럼 각 게임마다 개별적으로 다운로드할 필요가 없습니다. 만약 패치가 나타나지 않는다면, 해당 게임의 특정 시리얼 및 버전에 해당 패치가 없기 때문일 수 있습니다. Failed to parse JSON data from HTML. - Failed to parse JSON data from HTML. + HTML에서 JSON 데이터를 구문 분석하는 데 실패했습니다. Failed to retrieve HTML page. - Failed to retrieve HTML page. + HTML 페이지를 가져오지 못했습니다. The game is in version: %1 - The game is in version: %1 + 게임 버전: %1 The downloaded patch only works on version: %1 - The downloaded patch only works on version: %1 + 다운로드한 패치는 버전 %1 에서만 작동합니다. You may need to update your game. - You may need to update your game. + 게임을 업데이트해야 할 수도 있습니다. Incompatibility Notice - Incompatibility Notice + 호환성 경고 Failed to open file: - Failed to open file: + 파일을 열지 못했습니다: XML ERROR: - XML ERROR: + XML 오류: Failed to open files.json for writing - Failed to open files.json for writing + files.json 파일을 쓰기 위해 열지 못했습니다 Author: - Author: + 제작자: Directory does not exist: - Directory does not exist: + 디렉터리가 존재하지 않습니다: Failed to open files.json for reading. - Failed to open files.json for reading. + files.json 파일을 읽기 위해 열지 못했습니다. Name: - Name: + 이름: Can't apply cheats before the game is started - Can't apply cheats before the game is started + 게임이 시작되기 전에 치트를 적용할 수 없습니다 Close - Close + 닫기 CheckUpdate Auto Updater - Auto Updater + 자동 업데이트 Error - Error + 오류 Network error: - Network error: + 네트워크 오류: The Auto Updater allows up to 60 update checks per hour.\nYou have reached this limit. Please try again later. @@ -265,91 +265,91 @@ Failed to parse update information. - Failed to parse update information. + 업데이트 정보 구문 분석에 실패했습니다. No pre-releases found. - No pre-releases found. + 사전 릴리스가 없습니다. Invalid release data. - Invalid release data. + 잘못된 릴리스 데이터입니다. No download URL found for the specified asset. - No download URL found for the specified asset. + 지정된 자산에 대한 다운로드 URL을 찾을 수 없습니다. Your version is already up to date! - Your version is already up to date! + 버전이 이미 최신입니다! Update Available - Update Available + 업데이트 가능 Update Channel - Update Channel + 업데이트 채널 Current Version - Current Version + 현재 버전 Latest Version - Latest Version + 최신 버전 Do you want to update? - Do you want to update? + 업데이트하시겠습니까? Show Changelog - Show Changelog + 변경 사항 보기 Check for Updates at Startup - Check for Updates at Startup + 시작할 때 업데이트 확인 Update - Update + 업데이트 No - No + 아니요 Hide Changelog - Hide Changelog + 변경 사항 숨기기 Changes - Changes + 변경 사항 Network error occurred while trying to access the URL - Network error occurred while trying to access the URL + URL에 접근하는 동안 네트워크 오류가 발생했습니다 Download Complete - Download Complete + 다운로드 완료 The update has been downloaded, press OK to install. - The update has been downloaded, press OK to install. + 업데이트가 다운로드 되었습니다. 설치하려면 확인을 눌러주세요. Failed to save the update file at - Failed to save the update file at + 업데이트 파일을 다음 위치에 저장하지 못했습니다 Starting Update... - Starting Update... + 업데이트를 시작합니다... Failed to create the update script file - Failed to create the update script file + 업데이트 스크립트 파일을 생성하지 못했습니다 @@ -407,83 +407,83 @@ ControlSettings Configure Controls - Configure Controls + 컨트롤 설정 D-Pad - D-Pad + D-패드 Up - Up + 위 Left - Left + 왼쪽 Right - Right + 오른쪽 Down - Down + 아래 Left Stick Deadzone (def:2 max:127) - Left Stick Deadzone (def:2 max:127) + 왼쪽 스틱 데드존 (기본값:2 최대값:127) Left Deadzone - Left Deadzone + 왼쪽 데드존 Left Stick - Left Stick + 왼쪽 스틱 Config Selection - Config Selection + 설정 선택 Common Config - Common Config + 공통 설정 Use per-game configs - Use per-game configs + 게임 별 설정 사용 L1 / LB - L1 / LB + L1 / LB L2 / LT - L2 / LT + L2 / LT Back - Back + 뒤로 R1 / RB - R1 / RB + R1 / RB R2 / RT - R2 / RT + R2 / RT L3 - L3 + L3 Options / Start - Options / Start + 옵션 / 시작 R3 - R3 + R3 Face Buttons @@ -1347,10 +1347,6 @@ Game List Game List - - * Unsupported Vulkan Version - * Unsupported Vulkan Version - Download Cheats For All Installed Games Download Cheats For All Installed Games @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/lt_LT.ts b/src/qt_gui/translations/lt_LT.ts index 03ff5a003..a0b047dbb 100644 --- a/src/qt_gui/translations/lt_LT.ts +++ b/src/qt_gui/translations/lt_LT.ts @@ -1347,10 +1347,6 @@ Game List Žaidimų sąrašas - - * Unsupported Vulkan Version - * Nepalaikoma Vulkan versija - Download Cheats For All Installed Games Atsisiųsti sukčiavimus visiems įdiegtiems žaidimams @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/nb_NO.ts b/src/qt_gui/translations/nb_NO.ts index e937287fd..cb209cfb1 100644 --- a/src/qt_gui/translations/nb_NO.ts +++ b/src/qt_gui/translations/nb_NO.ts @@ -1253,11 +1253,11 @@ List View - Liste-visning + Listevisning Grid View - Rute-visning + Rutenettvisning Elf Viewer @@ -1347,10 +1347,6 @@ Game List Spilliste - - * Unsupported Vulkan Version - * Ustøttet Vulkan-versjon - Download Cheats For All Installed Games Last ned juks for alle installerte spill @@ -1760,7 +1756,7 @@ Log Filter:\nFilters the log to only print specific information.\nExamples: "Core:Trace" "Lib.Pad:Debug Common.Filesystem:Error" "*:Critical"\nLevels: Trace, Debug, Info, Warning, Error, Critical - in this order, a specific level silences all levels preceding it in the list and logs every level after it. - Loggfilter:\nFiltrerer loggen for å kun skrive ut spesifikk informasjon.\nEksempler: "Core:Trace" "Lib.Pad:Debug Common.Filesystem:Error" "*:Critical" \nNivåer: Trace, Debug, Info, Warning, Error, Critical - i denne rekkefølgen, et spesifikt nivå demper alle tidligere nivåer i lista og loggfører alle nivåer etter det. + Loggfilter:\nFiltrerer loggen for å kun skrive ut spesifikk informasjon.\nEksempler: «Core:Trace» «Lib.Pad:Debug Common.Filesystem:Error» «*:Critical» \nNivåer: Trace, Debug, Info, Warning, Error, Critical - i denne rekkefølgen, et spesifikt nivå demper alle tidligere nivåer i lista og loggfører alle nivåer etter det. Update:\nRelease: Official versions released every month that may be very outdated, but are more reliable and tested.\nNightly: Development versions that have all the latest features and fixes, but may contain bugs and are less stable. @@ -1792,7 +1788,7 @@ Display Compatibility Data:\nDisplays game compatibility information in table view. Enable "Update Compatibility On Startup" to get up-to-date information. - Vis kompatibilitets-data:\nViser informasjon om spillkompatibilitet i tabellvisning. Bruk "Oppdater kompatibilitets-data ved oppstart" for oppdatert informasjon. + Vis kompatibilitets-data:\nViser informasjon om spillkompatibilitet i tabellvisning. Bruk «Oppdater database ved oppstart» for oppdatert informasjon. Update Compatibility On Startup:\nAutomatically update the compatibility database when shadPS4 starts. @@ -1832,7 +1828,7 @@ Graphics Device:\nOn multiple GPU systems, select the GPU the emulator will use from the drop down list,\nor select "Auto Select" to automatically determine it. - Grafikkenhet:\nSystemer med flere GPU-er, kan emulatoren velge hvilken enhet som skal brukes fra rullegardinlista,\neller velg "Velg automatisk". + Grafikkenhet:\nSystemer med flere GPU-er, kan emulatoren velge hvilken enhet som skal brukes fra rullegardinlista,\neller bruk «Velg automatisk». Width/Height:\nSets the size of the emulator window at launch, which can be resized during gameplay.\nThis is different from the in-game resolution. @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Åpne mappa med tilpassede bilder og lyder for trofé:\nDu kan legge til tilpassede bilder til trofeer og en lyd.\nLegg filene til custom_trophy med følgende navn:\ntrophy.wav ELLER trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nMerk: Lyden avspilles kun i Qt-versjonen. + + * Unsupported Vulkan Version + *Ustøttet Vulkan-versjon + TrophyViewer diff --git a/src/qt_gui/translations/nl_NL.ts b/src/qt_gui/translations/nl_NL.ts index 2d75b74eb..ec676d360 100644 --- a/src/qt_gui/translations/nl_NL.ts +++ b/src/qt_gui/translations/nl_NL.ts @@ -1347,10 +1347,6 @@ Game List Lijst met spellen - - * Unsupported Vulkan Version - * Niet ondersteunde Vulkan-versie - Download Cheats For All Installed Games Download cheats voor alle geïnstalleerde spellen @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/pl_PL.ts b/src/qt_gui/translations/pl_PL.ts index bd59a1894..c1343cd2e 100644 --- a/src/qt_gui/translations/pl_PL.ts +++ b/src/qt_gui/translations/pl_PL.ts @@ -1347,10 +1347,6 @@ Game List Lista gier - - * Unsupported Vulkan Version - * Nieobsługiwana wersja Vulkan - Download Cheats For All Installed Games Pobierz kody do wszystkich zainstalowanych gier @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Otwórz niestandardowy folder obrazów/dźwięków:\nMożesz dodać własne obrazy dla trofeów i ich dźwięki.\nDodaj pliki do custom_trophy o następujących nazwach:\ntrophy.wav LUB trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nUwaga: Dźwięki działają tylko w wersji QT. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/pt_BR.ts b/src/qt_gui/translations/pt_BR.ts index 584d6dc19..34d31f240 100644 --- a/src/qt_gui/translations/pt_BR.ts +++ b/src/qt_gui/translations/pt_BR.ts @@ -1347,10 +1347,6 @@ Game List Lista de Jogos - - * Unsupported Vulkan Version - * Versão Vulkan não suportada - Download Cheats For All Installed Games Baixar Trapaças para Todos os Jogos Instalados @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Abrir a pasta de imagens e sons de troféus personalizados:\nVocê pode adicionar imagens personalizadas aos troféus e um áudio.\nAdicione os arquivos na pasta custom_trophy com os seguintes nomes:\ntrophy.wav OU trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nObservação: O som funcionará apenas em versões Qt. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/pt_PT.ts b/src/qt_gui/translations/pt_PT.ts index 70a73afe7..a543d0ec3 100644 --- a/src/qt_gui/translations/pt_PT.ts +++ b/src/qt_gui/translations/pt_PT.ts @@ -543,7 +543,7 @@ Unable to Save - Não é possível salvar + Não foi possível guardar Cannot bind axis values more than once @@ -551,7 +551,7 @@ Save - Salvar + Guardar Apply @@ -559,7 +559,7 @@ Restore Defaults - Restaurar o Padrão + Restaurar Predefinições Cancel @@ -570,11 +570,11 @@ EditorDialog Edit Keyboard + Mouse and Controller input bindings - Editar comandos do Teclado + Mouse e do Controle + Editar configurações de entrada do Teclado + Rato e do Comando Use Per-Game configs - Use uma configuração para cada jogo + Utilizar configurações por jogo Error @@ -582,19 +582,19 @@ Could not open the file for reading - Não foi possível abrir o arquivo para ler + Não foi possível abrir o ficheiro para leitura Could not open the file for writing - Não foi possível abrir o arquivo para escrever + Não foi possível abrir o ficheiro para escrita Save Changes - Salvar mudanças + Guardar as alterações Do you want to save changes? - Salvar as mudanças? + Pretende guardar as alterações? Help @@ -610,7 +610,7 @@ Reset to Default - Resetar ao Padrão + Repor para o Padrão @@ -1150,7 +1150,7 @@ Unable to Save - Não é possível salvar + Não foi possível guardar Cannot bind any unique input more than once @@ -1166,11 +1166,11 @@ Mousewheel cannot be mapped to stick outputs - Roda do rato não pode ser mapeada para saídas empates + Roda do rato não pode ser mapeada para saídas dos manípulos Save - Salvar + Guardar Apply @@ -1178,7 +1178,7 @@ Restore Defaults - Restaurar Definições + Restaurar Predefinições Cancel @@ -1347,10 +1347,6 @@ Game List Lista de Jogos - - * Unsupported Vulkan Version - * Versão do Vulkan não suportada - Download Cheats For All Installed Games Transferir Cheats para Todos os Jogos Instalados @@ -1409,43 +1405,43 @@ Play - Play + Reproduzir Pause - Pause + Pausa Stop - Stop + Parar Restart - Restart + Reiniciar Full Screen - Full Screen + Ecrã Inteiro Controllers - Controllers + Comandos Keyboard - Keyboard + Teclado Refresh List - Refresh List + Atualizar Lista Resume - Resume + Continuar Show Labels Under Icons - Show Labels Under Icons + Mostrar Etiquetas Debaixo dos Ícones @@ -2032,7 +2028,7 @@ Cannot create portable user folder - Não é possível criar pasta de utilizador portátil + Não foi possível criar pasta de utilizador portátil %1 already exists @@ -2048,7 +2044,11 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. - Abra a pasta de imagens/sons de troféus personalizados:\nPoderá adicionar imagens personalizadas aos troféus e um áudio.\nAdicione os arquivos na pasta custom_trophy com os seguintes nomes:\ntrophy.mp3 ou trophy.wav, bronze.png, gold.png, platinum.png, silver.png\nObservação: O som funcionará apenas nas versões Qt. + Abra a pasta de imagens/sons de troféus personalizados:\nPoderá adicionar imagens personalizadas aos troféus e um áudio.\nAdicione os ficheiros na pasta custom_trophy com os seguintes nomes:\ntrophy.mp3 ou trophy.wav, bronze.png, gold.png, platinum.png, silver.png\nObservação: O som funcionará apenas nas versões Qt. + + + * Unsupported Vulkan Version + * Versão do Vulkan não suportada diff --git a/src/qt_gui/translations/ro_RO.ts b/src/qt_gui/translations/ro_RO.ts index 78dd79c53..a05bb06a8 100644 --- a/src/qt_gui/translations/ro_RO.ts +++ b/src/qt_gui/translations/ro_RO.ts @@ -1347,10 +1347,6 @@ Game List Lista jocurilor - - * Unsupported Vulkan Version - * Versiune Vulkan nesuportată - Download Cheats For All Installed Games Descarcă Cheats pentru toate jocurile instalate @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/ru_RU.ts b/src/qt_gui/translations/ru_RU.ts index 0f16efc2c..a56127ece 100644 --- a/src/qt_gui/translations/ru_RU.ts +++ b/src/qt_gui/translations/ru_RU.ts @@ -1347,10 +1347,6 @@ Game List Список игр - - * Unsupported Vulkan Version - * Неподдерживаемая версия Vulkan - Download Cheats For All Installed Games Скачать читы для всех установленных игр @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Открыть папку с пользовательскими изображениями/звуками трофеев:\nВы можете добавить пользовательские изображения к трофеям и аудио.\nДобавьте файлы в custom_trophy со следующими именами:\ntrophy.wav ИЛИ trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nПримечание: звук будет работать только в QT-версии. + + * Unsupported Vulkan Version + * Неподдерживаемая версия Vulkan + TrophyViewer diff --git a/src/qt_gui/translations/sl_SI.ts b/src/qt_gui/translations/sl_SI.ts index ab61a5d3a..47c5f5534 100644 --- a/src/qt_gui/translations/sl_SI.ts +++ b/src/qt_gui/translations/sl_SI.ts @@ -1347,10 +1347,6 @@ Game List Game List - - * Unsupported Vulkan Version - * Unsupported Vulkan Version - Download Cheats For All Installed Games Download Cheats For All Installed Games @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/sq_AL.ts b/src/qt_gui/translations/sq_AL.ts index 50314a9b2..c554a283a 100644 --- a/src/qt_gui/translations/sq_AL.ts +++ b/src/qt_gui/translations/sq_AL.ts @@ -1347,10 +1347,6 @@ Game List Lista e lojërave - - * Unsupported Vulkan Version - * Version i pambështetur i Vulkan - Download Cheats For All Installed Games Shkarko mashtrime për të gjitha lojërat e instaluara @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Hap dosjen e imazheve/tingujve të trofeve të personalizuar:\nMund të shtosh imazhe të personalizuara për trofetë dhe një audio.\nShto skedarët në dosjen custom_trophy me emrat që vijojnë:\ntrophy.wav ose trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nShënim: Tingulli do të punojë vetëm në versionet QT. + + * Unsupported Vulkan Version + * Version i pambështetur i Vulkan + TrophyViewer diff --git a/src/qt_gui/translations/sv_SE.ts b/src/qt_gui/translations/sv_SE.ts index ce0da785c..b8fab701c 100644 --- a/src/qt_gui/translations/sv_SE.ts +++ b/src/qt_gui/translations/sv_SE.ts @@ -1347,10 +1347,6 @@ Game List Spellista - - * Unsupported Vulkan Version - * Vulkan-versionen stöds inte - Download Cheats For All Installed Games Hämta fusk för alla installerade spel @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Öppna mappen för anpassade trofébilder/ljud:\nDu kan lägga till egna bilder till troféerna och ett ljud.\nLägg till filerna i custom_trophy med följande namn:\ntrophy.wav ELLER trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nObservera: Ljudet fungerar endast i QT-versioner. + + * Unsupported Vulkan Version + * Versionen av Vulkan stöds inte + TrophyViewer diff --git a/src/qt_gui/translations/tr_TR.ts b/src/qt_gui/translations/tr_TR.ts index f5f7b65e5..9539ca139 100644 --- a/src/qt_gui/translations/tr_TR.ts +++ b/src/qt_gui/translations/tr_TR.ts @@ -1347,10 +1347,6 @@ Game List Oyun Listesi - - * Unsupported Vulkan Version - * Desteklenmeyen Vulkan Sürümü - Download Cheats For All Installed Games Tüm Yüklenmiş Oyunlar İçin Hileleri İndir @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Özel kupa görüntüleri/sesleri klasörünü aç:\nKupalara özel görüntüler ve sesler ekleyebilirsiniz.\nDosyaları aşağıdaki adlarla custom_trophy'ye ekleyin:\ntrophy.wav ya da trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNot: Ses yalnızca QT sürümlerinde çalışacaktır. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/uk_UA.ts b/src/qt_gui/translations/uk_UA.ts index 3eb88bcab..8b83ae62f 100644 --- a/src/qt_gui/translations/uk_UA.ts +++ b/src/qt_gui/translations/uk_UA.ts @@ -1347,10 +1347,6 @@ Game List Список ігор - - * Unsupported Vulkan Version - * Непідтримувана версія Vulkan - Download Cheats For All Installed Games Завантажити чити для усіх встановлених ігор @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Відкрити папку користувацьких зображень трофеїв/звуків:\nВи можете додати користувацькі зображення до трофеїв та звук.\nДодайте файли до теки custom_trophy з такими назвами:\ntrophy.wav АБО trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nПримітка: Звук буде працювати лише у версіях ShadPS4 з графічним інтерфейсом. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/vi_VN.ts b/src/qt_gui/translations/vi_VN.ts index c657888bf..1e26f626c 100644 --- a/src/qt_gui/translations/vi_VN.ts +++ b/src/qt_gui/translations/vi_VN.ts @@ -1347,10 +1347,6 @@ Game List Danh sách trò chơi - - * Unsupported Vulkan Version - * Phiên bản Vulkan không được hỗ trợ - Download Cheats For All Installed Games Tải xuống cheat cho tất cả các trò chơi đã cài đặt @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/qt_gui/translations/zh_CN.ts b/src/qt_gui/translations/zh_CN.ts index 120310810..2bc635c41 100644 --- a/src/qt_gui/translations/zh_CN.ts +++ b/src/qt_gui/translations/zh_CN.ts @@ -1347,10 +1347,6 @@ Game List 游戏列表 - - * Unsupported Vulkan Version - * 不支持的 Vulkan 版本 - Download Cheats For All Installed Games 下载所有已安装游戏的作弊码 @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. 打开自定义奖杯图像/声音文件夹：\n您可以自定义奖杯图像和声音。\n将文件添加到 custom_trophy 文件夹中，文件名如下：\ntrophy.wav 或 trophy.mp3、bronze.png、gold.png、platinum.png、silver.png。\n注意：自定义声音只能在 QT 版本中生效。 + + * Unsupported Vulkan Version + * 不支持的 Vulkan 版本 + TrophyViewer diff --git a/src/qt_gui/translations/zh_TW.ts b/src/qt_gui/translations/zh_TW.ts index bd051651d..320f73c83 100644 --- a/src/qt_gui/translations/zh_TW.ts +++ b/src/qt_gui/translations/zh_TW.ts @@ -1347,10 +1347,6 @@ Game List 遊戲列表 - - * Unsupported Vulkan Version - * 不支援的 Vulkan 版本 - Download Cheats For All Installed Games 下載所有已安裝遊戲的金手指 @@ -2050,6 +2046,10 @@ Open the custom trophy images/sounds folder:\nYou can add custom images to the trophies and an audio.\nAdd the files to custom_trophy with the following names:\ntrophy.wav OR trophy.mp3, bronze.png, gold.png, platinum.png, silver.png\nNote: The sound will only work in QT versions. 開啟自訂獎盃影像/聲音資料夾：\n您可以將自訂影像新增至獎盃和音訊。 \n將檔案加入 custom_tropy，名稱如下：\ntropy.wav OR trophy.mp3、bronze.png、gold.png、platinum.png、silver.png\n注意：聲音僅在 QT 版本中有效。 + + * Unsupported Vulkan Version + * Unsupported Vulkan Version + TrophyViewer diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.cpp b/src/shader_recompiler/backend/spirv/emit_spirv.cpp index 9ebb842cc..f2e6279f4 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv.cpp @@ -154,6 +154,7 @@ void Traverse(EmitContext& ctx, const IR::Program& program) { for (IR::Inst& inst : node.data.block->Instructions()) { EmitInst(ctx, &inst); } + ctx.first_to_last_label_map[label.value] = ctx.last_label; break; } case IR::AbstractSyntaxNode::Type::If: { @@ -298,6 +299,10 @@ void SetupCapabilities(const Info& info, const Profile& profile, EmitContext& ct if (stage == LogicalStage::TessellationControl || stage == LogicalStage::TessellationEval) { ctx.AddCapability(spv::Capability::Tessellation); } + if (info.dma_types != IR::Type::Void) { + ctx.AddCapability(spv::Capability::PhysicalStorageBufferAddresses); + ctx.AddExtension("SPV_KHR_physical_storage_buffer"); + } } void DefineEntryPoint(const Info& info, EmitContext& ctx, Id main) { @@ -387,7 +392,7 @@ void SetupFloatMode(EmitContext& ctx, const Profile& profile, const RuntimeInfo& void PatchPhiNodes(const IR::Program& program, EmitContext& ctx) { auto inst{program.blocks.front()->begin()}; size_t block_index{0}; - ctx.PatchDeferredPhi([&](size_t phi_arg) { + ctx.PatchDeferredPhi([&](u32 phi_arg, Id first_parent) { if (phi_arg == 0) { ++inst; if (inst == program.blocks[block_index]->end() || @@ -398,7 +403,9 @@ void PatchPhiNodes(const IR::Program& program, EmitContext& ctx) { } while (inst->GetOpcode() != IR::Opcode::Phi); } } - return ctx.Def(inst->Arg(phi_arg)); + const Id arg = ctx.Def(inst->Arg(phi_arg)); + const Id parent = ctx.first_to_last_label_map[first_parent.value]; + return std::make_pair(arg, parent); }); } } // Anonymous namespace diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp index c3799fb4b..d7c73ca8f 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_atomic.cpp @@ -60,7 +60,7 @@ Id BufferAtomicU32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id address = ctx.OpIAdd(ctx.U32[1], address, buffer.offset); } const Id index = ctx.OpShiftRightLogical(ctx.U32[1], address, ctx.ConstU32(2u)); - const auto [id, pointer_type] = buffer[EmitContext::BufferAlias::U32]; + const auto [id, pointer_type] = buffer[EmitContext::PointerType::U32]; const Id ptr = ctx.OpAccessChain(pointer_type, id, ctx.u32_zero_value, index); const auto [scope, semantics]{AtomicArgs(ctx)}; return BufferAtomicU32BoundsCheck(ctx, index, buffer.size_dwords, [&] { @@ -257,7 +257,7 @@ Id EmitImageAtomicExchange32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id co Id EmitDataAppend(EmitContext& ctx, u32 gds_addr, u32 binding) { const auto& buffer = ctx.buffers[binding]; - const auto [id, pointer_type] = buffer[EmitContext::BufferAlias::U32]; + const auto [id, pointer_type] = buffer[EmitContext::PointerType::U32]; const Id ptr = ctx.OpAccessChain(pointer_type, id, ctx.u32_zero_value, ctx.ConstU32(gds_addr)); const auto [scope, semantics]{AtomicArgs(ctx)}; return ctx.OpAtomicIIncrement(ctx.U32[1], ptr, scope, semantics); @@ -265,7 +265,7 @@ Id EmitDataAppend(EmitContext& ctx, u32 gds_addr, u32 binding) { Id EmitDataConsume(EmitContext& ctx, u32 gds_addr, u32 binding) { const auto& buffer = ctx.buffers[binding]; - const auto [id, pointer_type] = buffer[EmitContext::BufferAlias::U32]; + const auto [id, pointer_type] = buffer[EmitContext::PointerType::U32]; const Id ptr = ctx.OpAccessChain(pointer_type, id, ctx.u32_zero_value, ctx.ConstU32(gds_addr)); const auto [scope, semantics]{AtomicArgs(ctx)}; return ctx.OpAtomicIDecrement(ctx.U32[1], ptr, scope, semantics); diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp index 83e8afd78..658d4759f 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp @@ -2,6 +2,7 @@ // SPDX-License-Identifier: GPL-2.0-or-later #include "common/assert.h" +#include "common/logging/log.h" #include "shader_recompiler/backend/spirv/emit_spirv_instructions.h" #include "shader_recompiler/backend/spirv/spirv_emit_context.h" #include "shader_recompiler/ir/attribute.h" @@ -160,32 +161,37 @@ void EmitGetGotoVariable(EmitContext&) { UNREACHABLE_MSG("Unreachable instruction"); } -using BufferAlias = EmitContext::BufferAlias; +using PointerType = EmitContext::PointerType; -Id EmitReadConst(EmitContext& ctx, IR::Inst* inst) { +Id EmitReadConst(EmitContext& ctx, IR::Inst* inst, Id addr, Id offset) { const u32 flatbuf_off_dw = inst->Flags(); - const auto& srt_flatbuf = ctx.buffers.back(); - ASSERT(srt_flatbuf.binding >= 0 && flatbuf_off_dw > 0 && - srt_flatbuf.buffer_type == BufferType::ReadConstUbo); - const auto [id, pointer_type] = srt_flatbuf[BufferAlias::U32]; - const Id ptr{ - ctx.OpAccessChain(pointer_type, id, ctx.u32_zero_value, ctx.ConstU32(flatbuf_off_dw))}; - return ctx.OpLoad(ctx.U32[1], ptr); + // We can only provide a fallback for immediate offsets. + if (flatbuf_off_dw == 0) { + return ctx.OpFunctionCall(ctx.U32[1], ctx.read_const_dynamic, addr, offset); + } else { + return ctx.OpFunctionCall(ctx.U32[1], ctx.read_const, addr, offset, + ctx.ConstU32(flatbuf_off_dw)); + } } -Id EmitReadConstBuffer(EmitContext& ctx, u32 handle, Id index) { +template +Id ReadConstBuffer(EmitContext& ctx, u32 handle, Id index) { const auto& buffer = ctx.buffers[handle]; index = ctx.OpIAdd(ctx.U32[1], index, buffer.offset_dwords); - const auto [id, pointer_type] = buffer[BufferAlias::U32]; + const auto [id, pointer_type] = buffer[type]; + const auto value_type = type == PointerType::U32 ? ctx.U32[1] : ctx.F32[1]; const Id ptr{ctx.OpAccessChain(pointer_type, id, ctx.u32_zero_value, index)}; - const Id result{ctx.OpLoad(ctx.U32[1], ptr)}; + const Id result{ctx.OpLoad(value_type, ptr)}; if (Sirit::ValidId(buffer.size_dwords)) { const Id in_bounds = ctx.OpULessThan(ctx.U1[1], index, buffer.size_dwords); - return ctx.OpSelect(ctx.U32[1], in_bounds, result, ctx.u32_zero_value); - } else { - return result; + return ctx.OpSelect(value_type, in_bounds, result, ctx.u32_zero_value); } + return result; +} + +Id EmitReadConstBuffer(EmitContext& ctx, u32 handle, Id index) { + return ReadConstBuffer(ctx, handle, index); } Id EmitReadStepRate(EmitContext& ctx, int rate_idx) { @@ -244,7 +250,7 @@ Id EmitGetAttribute(EmitContext& ctx, IR::Attribute attr, u32 comp, Id index) { ctx.OpUDiv(ctx.U32[1], ctx.OpLoad(ctx.U32[1], ctx.instance_id), step_rate), ctx.ConstU32(param.num_components)), ctx.ConstU32(comp)); - return EmitReadConstBuffer(ctx, param.buffer_handle, offset); + return ReadConstBuffer(ctx, param.buffer_handle, offset); } Id result; @@ -430,7 +436,7 @@ static Id EmitLoadBufferBoundsCheck(EmitContext& ctx, Id index, Id buffer_size, return result; } -template +template static Id EmitLoadBufferB32xN(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { const auto flags = inst->Flags(); const auto& spv_buffer = ctx.buffers[handle]; @@ -438,7 +444,7 @@ static Id EmitLoadBufferB32xN(EmitContext& ctx, IR::Inst* inst, u32 handle, Id a address = ctx.OpIAdd(ctx.U32[1], address, spv_buffer.offset); } const Id index = ctx.OpShiftRightLogical(ctx.U32[1], address, ctx.ConstU32(2u)); - const auto& data_types = alias == BufferAlias::U32 ? ctx.U32 : ctx.F32; + const auto& data_types = alias == PointerType::U32 ? ctx.U32 : ctx.F32; const auto [id, pointer_type] = spv_buffer[alias]; boost::container::static_vector ids; @@ -449,7 +455,7 @@ static Id EmitLoadBufferB32xN(EmitContext& ctx, IR::Inst* inst, u32 handle, Id a if (!flags.typed) { // Untyped loads have bounds checking per-component. ids.push_back(EmitLoadBufferBoundsCheck<1>(ctx, index_i, spv_buffer.size_dwords, - result_i, alias == BufferAlias::F32)); + result_i, alias == PointerType::F32)); } else { ids.push_back(result_i); } @@ -459,7 +465,7 @@ static Id EmitLoadBufferB32xN(EmitContext& ctx, IR::Inst* inst, u32 handle, Id a if (flags.typed) { // Typed loads have single bounds check for the whole load. return EmitLoadBufferBoundsCheck(ctx, index, spv_buffer.size_dwords, result, - alias == BufferAlias::F32); + alias == PointerType::F32); } return result; } @@ -469,7 +475,7 @@ Id EmitLoadBufferU8(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { if (Sirit::ValidId(spv_buffer.offset)) { address = ctx.OpIAdd(ctx.U32[1], address, spv_buffer.offset); } - const auto [id, pointer_type] = spv_buffer[BufferAlias::U8]; + const auto [id, pointer_type] = spv_buffer[PointerType::U8]; const Id ptr{ctx.OpAccessChain(pointer_type, id, ctx.u32_zero_value, address)}; const Id result{ctx.OpUConvert(ctx.U32[1], ctx.OpLoad(ctx.U8, ptr))}; return EmitLoadBufferBoundsCheck<1>(ctx, address, spv_buffer.size, result, false); @@ -480,7 +486,7 @@ Id EmitLoadBufferU16(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { if (Sirit::ValidId(spv_buffer.offset)) { address = ctx.OpIAdd(ctx.U32[1], address, spv_buffer.offset); } - const auto [id, pointer_type] = spv_buffer[BufferAlias::U16]; + const auto [id, pointer_type] = spv_buffer[PointerType::U16]; const Id index = ctx.OpShiftRightLogical(ctx.U32[1], address, ctx.ConstU32(1u)); const Id ptr{ctx.OpAccessChain(pointer_type, id, ctx.u32_zero_value, index)}; const Id result{ctx.OpUConvert(ctx.U32[1], ctx.OpLoad(ctx.U16, ptr))}; @@ -488,35 +494,35 @@ Id EmitLoadBufferU16(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { } Id EmitLoadBufferU32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { - return EmitLoadBufferB32xN<1, BufferAlias::U32>(ctx, inst, handle, address); + return EmitLoadBufferB32xN<1, PointerType::U32>(ctx, inst, handle, address); } Id EmitLoadBufferU32x2(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { - return EmitLoadBufferB32xN<2, BufferAlias::U32>(ctx, inst, handle, address); + return EmitLoadBufferB32xN<2, PointerType::U32>(ctx, inst, handle, address); } Id EmitLoadBufferU32x3(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { - return EmitLoadBufferB32xN<3, BufferAlias::U32>(ctx, inst, handle, address); + return EmitLoadBufferB32xN<3, PointerType::U32>(ctx, inst, handle, address); } Id EmitLoadBufferU32x4(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { - return EmitLoadBufferB32xN<4, BufferAlias::U32>(ctx, inst, handle, address); + return EmitLoadBufferB32xN<4, PointerType::U32>(ctx, inst, handle, address); } Id EmitLoadBufferF32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { - return EmitLoadBufferB32xN<1, BufferAlias::F32>(ctx, inst, handle, address); + return EmitLoadBufferB32xN<1, PointerType::F32>(ctx, inst, handle, address); } Id EmitLoadBufferF32x2(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { - return EmitLoadBufferB32xN<2, BufferAlias::F32>(ctx, inst, handle, address); + return EmitLoadBufferB32xN<2, PointerType::F32>(ctx, inst, handle, address); } Id EmitLoadBufferF32x3(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { - return EmitLoadBufferB32xN<3, BufferAlias::F32>(ctx, inst, handle, address); + return EmitLoadBufferB32xN<3, PointerType::F32>(ctx, inst, handle, address); } Id EmitLoadBufferF32x4(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { - return EmitLoadBufferB32xN<4, BufferAlias::F32>(ctx, inst, handle, address); + return EmitLoadBufferB32xN<4, PointerType::F32>(ctx, inst, handle, address); } Id EmitLoadBufferFormatF32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address) { @@ -546,7 +552,7 @@ void EmitStoreBufferBoundsCheck(EmitContext& ctx, Id index, Id buffer_size, auto emit_func(); } -template +template static void EmitStoreBufferB32xN(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { const auto flags = inst->Flags(); @@ -555,7 +561,7 @@ static void EmitStoreBufferB32xN(EmitContext& ctx, IR::Inst* inst, u32 handle, I address = ctx.OpIAdd(ctx.U32[1], address, spv_buffer.offset); } const Id index = ctx.OpShiftRightLogical(ctx.U32[1], address, ctx.ConstU32(2u)); - const auto& data_types = alias == BufferAlias::U32 ? ctx.U32 : ctx.F32; + const auto& data_types = alias == PointerType::U32 ? ctx.U32 : ctx.F32; const auto [id, pointer_type] = spv_buffer[alias]; auto store = [&] { @@ -586,7 +592,7 @@ void EmitStoreBufferU8(EmitContext& ctx, IR::Inst*, u32 handle, Id address, Id v if (Sirit::ValidId(spv_buffer.offset)) { address = ctx.OpIAdd(ctx.U32[1], address, spv_buffer.offset); } - const auto [id, pointer_type] = spv_buffer[BufferAlias::U8]; + const auto [id, pointer_type] = spv_buffer[PointerType::U8]; const Id ptr{ctx.OpAccessChain(pointer_type, id, ctx.u32_zero_value, address)}; const Id result{ctx.OpUConvert(ctx.U8, value)}; EmitStoreBufferBoundsCheck<1>(ctx, address, spv_buffer.size, [&] { ctx.OpStore(ptr, result); }); @@ -597,7 +603,7 @@ void EmitStoreBufferU16(EmitContext& ctx, IR::Inst*, u32 handle, Id address, Id if (Sirit::ValidId(spv_buffer.offset)) { address = ctx.OpIAdd(ctx.U32[1], address, spv_buffer.offset); } - const auto [id, pointer_type] = spv_buffer[BufferAlias::U16]; + const auto [id, pointer_type] = spv_buffer[PointerType::U16]; const Id index = ctx.OpShiftRightLogical(ctx.U32[1], address, ctx.ConstU32(1u)); const Id ptr{ctx.OpAccessChain(pointer_type, id, ctx.u32_zero_value, index)}; const Id result{ctx.OpUConvert(ctx.U16, value)}; @@ -606,35 +612,35 @@ void EmitStoreBufferU16(EmitContext& ctx, IR::Inst*, u32 handle, Id address, Id } void EmitStoreBufferU32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { - EmitStoreBufferB32xN<1, BufferAlias::U32>(ctx, inst, handle, address, value); + EmitStoreBufferB32xN<1, PointerType::U32>(ctx, inst, handle, address, value); } void EmitStoreBufferU32x2(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { - EmitStoreBufferB32xN<2, BufferAlias::U32>(ctx, inst, handle, address, value); + EmitStoreBufferB32xN<2, PointerType::U32>(ctx, inst, handle, address, value); } void EmitStoreBufferU32x3(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { - EmitStoreBufferB32xN<3, BufferAlias::U32>(ctx, inst, handle, address, value); + EmitStoreBufferB32xN<3, PointerType::U32>(ctx, inst, handle, address, value); } void EmitStoreBufferU32x4(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { - EmitStoreBufferB32xN<4, BufferAlias::U32>(ctx, inst, handle, address, value); + EmitStoreBufferB32xN<4, PointerType::U32>(ctx, inst, handle, address, value); } void EmitStoreBufferF32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { - EmitStoreBufferB32xN<1, BufferAlias::F32>(ctx, inst, handle, address, value); + EmitStoreBufferB32xN<1, PointerType::F32>(ctx, inst, handle, address, value); } void EmitStoreBufferF32x2(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { - EmitStoreBufferB32xN<2, BufferAlias::F32>(ctx, inst, handle, address, value); + EmitStoreBufferB32xN<2, PointerType::F32>(ctx, inst, handle, address, value); } void EmitStoreBufferF32x3(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { - EmitStoreBufferB32xN<3, BufferAlias::F32>(ctx, inst, handle, address, value); + EmitStoreBufferB32xN<3, PointerType::F32>(ctx, inst, handle, address, value); } void EmitStoreBufferF32x4(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { - EmitStoreBufferB32xN<4, BufferAlias::F32>(ctx, inst, handle, address, value); + EmitStoreBufferB32xN<4, PointerType::F32>(ctx, inst, handle, address, value); } void EmitStoreBufferFormatF32(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address, Id value) { diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_floating_point.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_floating_point.cpp index 347c4cb0a..01c51e399 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_floating_point.cpp +++ b/src/shader_recompiler/backend/spirv/emit_spirv_floating_point.cpp @@ -154,7 +154,7 @@ Id EmitFPRecip32(EmitContext& ctx, Id value) { } Id EmitFPRecip64(EmitContext& ctx, Id value) { - return ctx.OpFDiv(ctx.F64[1], ctx.Constant(ctx.F64[1], 1.0f), value); + return ctx.OpFDiv(ctx.F64[1], ctx.Constant(ctx.F64[1], f64{1.0}), value); } Id EmitFPRecipSqrt32(EmitContext& ctx, Id value) { diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h b/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h index 269f372d5..09f9732bf 100644 --- a/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h +++ b/src/shader_recompiler/backend/spirv/emit_spirv_instructions.h @@ -61,7 +61,7 @@ void EmitSetVectorRegister(EmitContext& ctx); void EmitSetGotoVariable(EmitContext& ctx); void EmitGetGotoVariable(EmitContext& ctx); void EmitSetScc(EmitContext& ctx); -Id EmitReadConst(EmitContext& ctx, IR::Inst* inst); +Id EmitReadConst(EmitContext& ctx, IR::Inst* inst, Id addr, Id offset); Id EmitReadConstBuffer(EmitContext& ctx, u32 handle, Id index); Id EmitLoadBufferU8(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address); Id EmitLoadBufferU16(EmitContext& ctx, IR::Inst* inst, u32 handle, Id address); diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp index 2640030df..68bfcc0d0 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp @@ -7,6 +7,7 @@ #include "shader_recompiler/frontend/fetch_shader.h" #include "shader_recompiler/runtime_info.h" #include "video_core/amdgpu/types.h" +#include "video_core/buffer_cache/buffer_cache.h" #include #include @@ -70,6 +71,12 @@ EmitContext::EmitContext(const Profile& profile_, const RuntimeInfo& runtime_inf Bindings& binding_) : Sirit::Module(profile_.supported_spirv), info{info_}, runtime_info{runtime_info_}, profile{profile_}, stage{info.stage}, l_stage{info.l_stage}, binding{binding_} { + if (info.dma_types != IR::Type::Void) { + SetMemoryModel(spv::AddressingModel::PhysicalStorageBuffer64, spv::MemoryModel::GLSL450); + } else { + SetMemoryModel(spv::AddressingModel::Logical, spv::MemoryModel::GLSL450); + } + AddCapability(spv::Capability::Shader); DefineArithmeticTypes(); DefineInterfaces(); @@ -137,9 +144,13 @@ void EmitContext::DefineArithmeticTypes() { true_value = ConstantTrue(U1[1]); false_value = ConstantFalse(U1[1]); + u8_one_value = Constant(U8, 1U); + u8_zero_value = Constant(U8, 0U); u32_one_value = ConstU32(1U); u32_zero_value = ConstU32(0U); f32_zero_value = ConstF32(0.0f); + u64_one_value = Constant(U64, 1ULL); + u64_zero_value = Constant(U64, 0ULL); pi_x2 = ConstF32(2.0f * float{std::numbers::pi}); @@ -157,6 +168,35 @@ void EmitContext::DefineArithmeticTypes() { if (info.uses_fp64) { frexp_result_f64 = Name(TypeStruct(F64[1], S32[1]), "frexp_result_f64"); } + + if (True(info.dma_types & IR::Type::F64)) { + physical_pointer_types[PointerType::F64] = + TypePointer(spv::StorageClass::PhysicalStorageBuffer, F64[1]); + } + if (True(info.dma_types & IR::Type::U64)) { + physical_pointer_types[PointerType::U64] = + TypePointer(spv::StorageClass::PhysicalStorageBuffer, U64); + } + if (True(info.dma_types & IR::Type::F32)) { + physical_pointer_types[PointerType::F32] = + TypePointer(spv::StorageClass::PhysicalStorageBuffer, F32[1]); + } + if (True(info.dma_types & IR::Type::U32)) { + physical_pointer_types[PointerType::U32] = + TypePointer(spv::StorageClass::PhysicalStorageBuffer, U32[1]); + } + if (True(info.dma_types & IR::Type::F16)) { + physical_pointer_types[PointerType::F16] = + TypePointer(spv::StorageClass::PhysicalStorageBuffer, F16[1]); + } + if (True(info.dma_types & IR::Type::U16)) { + physical_pointer_types[PointerType::U16] = + TypePointer(spv::StorageClass::PhysicalStorageBuffer, U16); + } + if (True(info.dma_types & IR::Type::U8)) { + physical_pointer_types[PointerType::U8] = + TypePointer(spv::StorageClass::PhysicalStorageBuffer, U8); + } } void EmitContext::DefineInterfaces() { @@ -195,9 +235,10 @@ EmitContext::SpirvAttribute EmitContext::GetAttributeInfo(AmdGpu::NumberFormat f } Id EmitContext::GetBufferSize(const u32 sharp_idx) { - const auto& srt_flatbuf = buffers.back(); - ASSERT(srt_flatbuf.buffer_type == BufferType::ReadConstUbo); - const auto [id, pointer_type] = srt_flatbuf[BufferAlias::U32]; + // Can this be done with memory access? Like we do now with ReadConst + const auto& srt_flatbuf = buffers[flatbuf_index]; + ASSERT(srt_flatbuf.buffer_type == BufferType::Flatbuf); + const auto [id, pointer_type] = srt_flatbuf[PointerType::U32]; const auto rsrc1{ OpLoad(U32[1], OpAccessChain(pointer_type, id, u32_zero_value, ConstU32(sharp_idx + 1)))}; @@ -690,8 +731,14 @@ EmitContext::BufferSpv EmitContext::DefineBuffer(bool is_storage, bool is_writte case Shader::BufferType::GdsBuffer: Name(id, "gds_buffer"); break; - case Shader::BufferType::ReadConstUbo: - Name(id, "srt_flatbuf_ubo"); + case Shader::BufferType::Flatbuf: + Name(id, "srt_flatbuf"); + break; + case Shader::BufferType::BdaPagetable: + Name(id, "bda_pagetable"); + break; + case Shader::BufferType::FaultBuffer: + Name(id, "fault_buffer"); break; case Shader::BufferType::SharedMemory: Name(id, "ssbo_shmem"); @@ -705,35 +752,53 @@ EmitContext::BufferSpv EmitContext::DefineBuffer(bool is_storage, bool is_writte }; void EmitContext::DefineBuffers() { - if (!profile.supports_robust_buffer_access && !info.has_readconst) { - // In case ReadConstUbo has not already been bound by IR and is needed + if (!profile.supports_robust_buffer_access && + info.readconst_types == Info::ReadConstType::None) { + // In case Flatbuf has not already been bound by IR and is needed // to query buffer sizes, bind it now. info.buffers.push_back({ .used_types = IR::Type::U32, - .inline_cbuf = AmdGpu::Buffer::Null(), - .buffer_type = BufferType::ReadConstUbo, + // We can't guarantee that flatbuf will not grow past UBO + // limit if there are a lot of ReadConsts. (We could specialize) + .inline_cbuf = AmdGpu::Buffer::Placeholder(std::numeric_limits::max()), + .buffer_type = BufferType::Flatbuf, }); + // In the future we may want to read buffer sizes from GPU memory if available. + // info.readconst_types |= Info::ReadConstType::Immediate; } for (const auto& desc : info.buffers) { const auto buf_sharp = desc.GetSharp(info); const bool is_storage = desc.IsStorage(buf_sharp, profile); + // Set indexes for special buffers. + if (desc.buffer_type == BufferType::Flatbuf) { + flatbuf_index = buffers.size(); + } else if (desc.buffer_type == BufferType::BdaPagetable) { + bda_pagetable_index = buffers.size(); + } else if (desc.buffer_type == BufferType::FaultBuffer) { + fault_buffer_index = buffers.size(); + } + // Define aliases depending on the shader usage. auto& spv_buffer = buffers.emplace_back(binding.buffer++, desc.buffer_type); + if (True(desc.used_types & IR::Type::U64)) { + spv_buffer[PointerType::U64] = + DefineBuffer(is_storage, desc.is_written, 3, desc.buffer_type, U64); + } if (True(desc.used_types & IR::Type::U32)) { - spv_buffer[BufferAlias::U32] = + spv_buffer[PointerType::U32] = DefineBuffer(is_storage, desc.is_written, 2, desc.buffer_type, U32[1]); } if (True(desc.used_types & IR::Type::F32)) { - spv_buffer[BufferAlias::F32] = + spv_buffer[PointerType::F32] = DefineBuffer(is_storage, desc.is_written, 2, desc.buffer_type, F32[1]); } if (True(desc.used_types & IR::Type::U16)) { - spv_buffer[BufferAlias::U16] = + spv_buffer[PointerType::U16] = DefineBuffer(is_storage, desc.is_written, 1, desc.buffer_type, U16); } if (True(desc.used_types & IR::Type::U8)) { - spv_buffer[BufferAlias::U8] = + spv_buffer[PointerType::U8] = DefineBuffer(is_storage, desc.is_written, 0, desc.buffer_type, U8); } ++binding.unified; @@ -1003,6 +1068,101 @@ Id EmitContext::DefineUfloatM5ToFloat32(u32 mantissa_bits, const std::string_vie return func; } +Id EmitContext::DefineGetBdaPointer() { + const auto caching_pagebits{ + Constant(U64, static_cast(VideoCore::BufferCache::CACHING_PAGEBITS))}; + const auto caching_pagemask{Constant(U64, VideoCore::BufferCache::CACHING_PAGESIZE - 1)}; + + const auto func_type{TypeFunction(U64, U64)}; + const auto func{OpFunction(U64, spv::FunctionControlMask::MaskNone, func_type)}; + const auto address{OpFunctionParameter(U64)}; + Name(func, "get_bda_pointer"); + AddLabel(); + + const auto fault_label{OpLabel()}; + const auto available_label{OpLabel()}; + const auto merge_label{OpLabel()}; + + // Get page BDA + const auto page{OpShiftRightLogical(U64, address, caching_pagebits)}; + const auto page32{OpUConvert(U32[1], page)}; + const auto& bda_buffer{buffers[bda_pagetable_index]}; + const auto [bda_buffer_id, bda_pointer_type] = bda_buffer[PointerType::U64]; + const auto bda_ptr{OpAccessChain(bda_pointer_type, bda_buffer_id, u32_zero_value, page32)}; + const auto bda{OpLoad(U64, bda_ptr)}; + + // Check if page is GPU cached + const auto is_fault{OpIEqual(U1[1], bda, u64_zero_value)}; + OpSelectionMerge(merge_label, spv::SelectionControlMask::MaskNone); + OpBranchConditional(is_fault, fault_label, available_label); + + // First time acces, mark as fault + AddLabel(fault_label); + const auto& fault_buffer{buffers[fault_buffer_index]}; + const auto [fault_buffer_id, fault_pointer_type] = fault_buffer[PointerType::U8]; + const auto page_div8{OpShiftRightLogical(U32[1], page32, ConstU32(3U))}; + const auto page_mod8{OpBitwiseAnd(U32[1], page32, ConstU32(7U))}; + const auto page_mask{OpShiftLeftLogical(U8, u8_one_value, page_mod8)}; + const auto fault_ptr{ + OpAccessChain(fault_pointer_type, fault_buffer_id, u32_zero_value, page_div8)}; + const auto fault_value{OpLoad(U8, fault_ptr)}; + const auto fault_value_masked{OpBitwiseOr(U8, fault_value, page_mask)}; + OpStore(fault_ptr, fault_value_masked); + + // Return null pointer + const auto fallback_result{u64_zero_value}; + OpBranch(merge_label); + + // Value is available, compute address + AddLabel(available_label); + const auto offset_in_bda{OpBitwiseAnd(U64, address, caching_pagemask)}; + const auto addr{OpIAdd(U64, bda, offset_in_bda)}; + OpBranch(merge_label); + + // Merge + AddLabel(merge_label); + const auto result{OpPhi(U64, addr, available_label, fallback_result, fault_label)}; + OpReturnValue(result); + OpFunctionEnd(); + return func; +} + +Id EmitContext::DefineReadConst(bool dynamic) { + const auto func_type{!dynamic ? TypeFunction(U32[1], U32[2], U32[1], U32[1]) + : TypeFunction(U32[1], U32[2], U32[1])}; + const auto func{OpFunction(U32[1], spv::FunctionControlMask::MaskNone, func_type)}; + const auto base{OpFunctionParameter(U32[2])}; + const auto offset{OpFunctionParameter(U32[1])}; + const auto flatbuf_offset{!dynamic ? OpFunctionParameter(U32[1]) : Id{}}; + Name(func, dynamic ? "read_const_dynamic" : "read_const"); + AddLabel(); + + const auto base_lo{OpUConvert(U64, OpCompositeExtract(U32[1], base, 0))}; + const auto base_hi{OpUConvert(U64, OpCompositeExtract(U32[1], base, 1))}; + const auto base_shift{OpShiftLeftLogical(U64, base_hi, ConstU32(32U))}; + const auto base_addr{OpBitwiseOr(U64, base_lo, base_shift)}; + const auto offset_bytes{OpShiftLeftLogical(U32[1], offset, ConstU32(2U))}; + const auto addr{OpIAdd(U64, base_addr, OpUConvert(U64, offset_bytes))}; + + const auto result = EmitMemoryRead(U32[1], addr, [&]() { + if (dynamic) { + return u32_zero_value; + } else { + const auto& flatbuf_buffer{buffers[flatbuf_index]}; + ASSERT(flatbuf_buffer.binding >= 0 && + flatbuf_buffer.buffer_type == BufferType::Flatbuf); + const auto [flatbuf_buffer_id, flatbuf_pointer_type] = flatbuf_buffer[PointerType::U32]; + const auto ptr{OpAccessChain(flatbuf_pointer_type, flatbuf_buffer_id, u32_zero_value, + flatbuf_offset)}; + return OpLoad(U32[1], ptr); + } + }); + + OpReturnValue(result); + OpFunctionEnd(); + return func; +} + void EmitContext::DefineFunctions() { if (info.uses_pack_10_11_11) { f32_to_uf11 = DefineFloat32ToUfloatM5(6, "f32_to_uf11"); @@ -1012,6 +1172,18 @@ void EmitContext::DefineFunctions() { uf11_to_f32 = DefineUfloatM5ToFloat32(6, "uf11_to_f32"); uf10_to_f32 = DefineUfloatM5ToFloat32(5, "uf10_to_f32"); } + if (info.dma_types != IR::Type::Void) { + get_bda_pointer = DefineGetBdaPointer(); + } + + if (True(info.readconst_types & Info::ReadConstType::Immediate)) { + LOG_DEBUG(Render_Recompiler, "Shader {:#x} uses immediate ReadConst", info.pgm_hash); + read_const = DefineReadConst(false); + } + if (True(info.readconst_types & Info::ReadConstType::Dynamic)) { + LOG_DEBUG(Render_Recompiler, "Shader {:#x} uses dynamic ReadConst", info.pgm_hash); + read_const_dynamic = DefineReadConst(true); + } } } // namespace Shader::Backend::SPIRV diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.h b/src/shader_recompiler/backend/spirv/spirv_emit_context.h index 38d55e0e4..a2e0d2f47 100644 --- a/src/shader_recompiler/backend/spirv/spirv_emit_context.h +++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.h @@ -4,6 +4,7 @@ #pragma once #include +#include #include #include "shader_recompiler/backend/bindings.h" @@ -41,6 +42,17 @@ public: Bindings& binding); ~EmitContext(); + enum class PointerType : u32 { + U8, + U16, + F16, + U32, + F32, + U64, + F64, + NumAlias, + }; + Id Def(const IR::Value& value); void DefineBufferProperties(); @@ -133,12 +145,72 @@ public: return ConstantComposite(type, constituents); } + inline Id AddLabel() { + last_label = Module::AddLabel(); + return last_label; + } + + inline Id AddLabel(Id label) { + last_label = Module::AddLabel(label); + return last_label; + } + + PointerType PointerTypeFromType(Id type) { + if (type.value == U8.value) + return PointerType::U8; + if (type.value == U16.value) + return PointerType::U16; + if (type.value == F16[1].value) + return PointerType::F16; + if (type.value == U32[1].value) + return PointerType::U32; + if (type.value == F32[1].value) + return PointerType::F32; + if (type.value == U64.value) + return PointerType::U64; + if (type.value == F64[1].value) + return PointerType::F64; + UNREACHABLE_MSG("Unknown type for pointer"); + } + + Id EmitMemoryRead(Id type, Id address, auto&& fallback) { + const Id available_label = OpLabel(); + const Id fallback_label = OpLabel(); + const Id merge_label = OpLabel(); + + const Id addr = OpFunctionCall(U64, get_bda_pointer, address); + const Id is_available = OpINotEqual(U1[1], addr, u64_zero_value); + OpSelectionMerge(merge_label, spv::SelectionControlMask::MaskNone); + OpBranchConditional(is_available, available_label, fallback_label); + + // Available + AddLabel(available_label); + const auto pointer_type = PointerTypeFromType(type); + const Id pointer_type_id = physical_pointer_types[pointer_type]; + const Id addr_ptr = OpConvertUToPtr(pointer_type_id, addr); + const Id result = OpLoad(type, addr_ptr, spv::MemoryAccessMask::Aligned, 4u); + OpBranch(merge_label); + + // Fallback + AddLabel(fallback_label); + const Id fallback_result = fallback(); + OpBranch(merge_label); + + // Merge + AddLabel(merge_label); + const Id final_result = + OpPhi(type, fallback_result, fallback_label, result, available_label); + return final_result; + } + Info& info; const RuntimeInfo& runtime_info; const Profile& profile; Stage stage; LogicalStage l_stage{}; + Id last_label{}; + Id void_id{}; Id U8{}; Id S8{}; @@ -161,9 +233,13 @@ public: Id true_value{}; Id false_value{}; + Id u8_one_value{}; + Id u8_zero_value{}; Id u32_one_value{}; Id u32_zero_value{}; Id f32_zero_value{}; + Id u64_one_value{}; + Id u64_zero_value{}; Id shared_u8{}; Id shared_u16{}; @@ -231,14 +307,6 @@ public: bool is_storage = false; }; - enum class BufferAlias : u32 { - U8, - U16, - U32, - F32, - NumAlias, - }; - struct BufferSpv { Id id; Id pointer_type; @@ -252,22 +320,40 @@ public: Id size; Id size_shorts; Id size_dwords; - std::array aliases; + std::array aliases; - const BufferSpv& operator[](BufferAlias alias) const { + const BufferSpv& operator[](PointerType alias) const { return aliases[u32(alias)]; } - BufferSpv& operator[](BufferAlias alias) { + BufferSpv& operator[](PointerType alias) { return aliases[u32(alias)]; } }; + struct PhysicalPointerTypes { + std::array types; + + const Id& operator[](PointerType type) const { + return types[u32(type)]; + } + + Id& operator[](PointerType type) { + return types[u32(type)]; + } + }; + Bindings& binding; boost::container::small_vector buf_type_ids; boost::container::small_vector buffers; boost::container::small_vector images; boost::container::small_vector samplers; + PhysicalPointerTypes physical_pointer_types; + std::unordered_map first_to_last_label_map; + + size_t flatbuf_index{}; + size_t bda_pagetable_index{}; + size_t fault_buffer_index{}; Id sampler_type{}; Id sampler_pointer_type{}; @@ -292,6 +378,11 @@ public: Id uf10_to_f32{}; Id f32_to_uf10{}; + Id get_bda_pointer{}; + + Id read_const{}; + Id read_const_dynamic{}; + private: void DefineArithmeticTypes(); void DefineInterfaces(); @@ -312,6 +403,10 @@ private: Id DefineFloat32ToUfloatM5(u32 mantissa_bits, std::string_view name); Id DefineUfloatM5ToFloat32(u32 mantissa_bits, std::string_view name); + Id DefineGetBdaPointer(); + + Id DefineReadConst(bool dynamic); + Id GetBufferSize(u32 sharp_idx); }; diff --git a/src/shader_recompiler/frontend/translate/scalar_memory.cpp b/src/shader_recompiler/frontend/translate/scalar_memory.cpp index 89426e080..3c6fd3968 100644 --- a/src/shader_recompiler/frontend/translate/scalar_memory.cpp +++ b/src/shader_recompiler/frontend/translate/scalar_memory.cpp @@ -39,21 +39,22 @@ void Translator::EmitScalarMemory(const GcnInst& inst) { void Translator::S_LOAD_DWORD(int num_dwords, const GcnInst& inst) { const auto& smrd = inst.control.smrd; - const u32 dword_offset = [&] -> u32 { + const IR::ScalarReg sbase{inst.src[0].code * 2}; + const IR::U32 dword_offset = [&] -> IR::U32 { if (smrd.imm) { - return smrd.offset; + return ir.Imm32(smrd.offset); } if (smrd.offset == SQ_SRC_LITERAL) { - return inst.src[1].code; + return ir.Imm32(inst.src[1].code); } - UNREACHABLE(); + return ir.ShiftRightLogical(ir.GetScalarReg(IR::ScalarReg(smrd.offset)), ir.Imm32(2)); }(); - const IR::ScalarReg sbase{inst.src[0].code * 2}; const IR::Value base = ir.CompositeConstruct(ir.GetScalarReg(sbase), ir.GetScalarReg(sbase + 1)); IR::ScalarReg dst_reg{inst.dst[0].code}; for (u32 i = 0; i < num_dwords; i++) { - ir.SetScalarReg(dst_reg++, ir.ReadConst(base, ir.Imm32(dword_offset + i))); + IR::U32 index = ir.IAdd(dword_offset, ir.Imm32(i)); + ir.SetScalarReg(dst_reg + i, ir.ReadConst(base, index)); } } @@ -75,7 +76,7 @@ void Translator::S_BUFFER_LOAD_DWORD(int num_dwords, const GcnInst& inst) { IR::ScalarReg dst_reg{inst.dst[0].code}; for (u32 i = 0; i < num_dwords; i++) { const IR::U32 index = ir.IAdd(dword_offset, ir.Imm32(i)); - ir.SetScalarReg(dst_reg++, ir.ReadConstBuffer(vsharp, index)); + ir.SetScalarReg(dst_reg + i, ir.ReadConstBuffer(vsharp, index)); } } diff --git a/src/shader_recompiler/frontend/translate/vector_alu.cpp b/src/shader_recompiler/frontend/translate/vector_alu.cpp index 3ce86c131..6171cca07 100644 --- a/src/shader_recompiler/frontend/translate/vector_alu.cpp +++ b/src/shader_recompiler/frontend/translate/vector_alu.cpp @@ -989,13 +989,22 @@ void Translator::V_CMP_NE_U64(const GcnInst& inst) { } }; const IR::U1 src0{get_src(inst.src[0])}; - ASSERT(inst.src[1].field == OperandField::ConstZero); // src0 != 0 + auto op = [&inst, this](auto x) { + switch (inst.src[1].field) { + case OperandField::ConstZero: + return x; + case OperandField::SignedConstIntNeg: + return ir.LogicalNot(x); + default: + UNREACHABLE_MSG("unhandled V_CMP_NE_U64 source argument {}", u32(inst.src[1].field)); + } + }; switch (inst.dst[1].field) { case OperandField::VccLo: - ir.SetVcc(src0); + ir.SetVcc(op(src0)); break; case OperandField::ScalarGPR: - ir.SetThreadBitScalarReg(IR::ScalarReg(inst.dst[1].code), src0); + ir.SetThreadBitScalarReg(IR::ScalarReg(inst.dst[1].code), op(src0)); break; default: UNREACHABLE(); diff --git a/src/shader_recompiler/info.h b/src/shader_recompiler/info.h index ba28d7e43..d349d7827 100644 --- a/src/shader_recompiler/info.h +++ b/src/shader_recompiler/info.h @@ -41,7 +41,9 @@ constexpr u32 NUM_TEXTURE_TYPES = 7; enum class BufferType : u32 { Guest, - ReadConstUbo, + Flatbuf, + BdaPagetable, + FaultBuffer, GdsBuffer, SharedMemory, }; @@ -215,11 +217,18 @@ struct Info { bool stores_tess_level_outer{}; bool stores_tess_level_inner{}; bool translation_failed{}; - bool has_readconst{}; u8 mrt_mask{0u}; bool has_fetch_shader{false}; u32 fetch_shader_sgpr_base{0u}; + enum class ReadConstType { + None = 0, + Immediate = 1 << 0, + Dynamic = 1 << 1, + }; + ReadConstType readconst_types{}; + IR::Type dma_types{IR::Type::Void}; + explicit Info(Stage stage_, LogicalStage l_stage_, ShaderParams params) : stage{stage_}, l_stage{l_stage_}, pgm_hash{params.hash}, pgm_base{params.Base()}, user_data{params.user_data} {} @@ -277,6 +286,7 @@ struct Info { sizeof(tess_constants)); } }; +DECLARE_ENUM_FLAG_OPERATORS(Info::ReadConstType); constexpr AmdGpu::Buffer BufferResource::GetSharp(const Info& info) const noexcept { return inline_cbuf ? inline_cbuf : info.ReadUdSharp(sharp_idx); diff --git a/src/shader_recompiler/ir/abstract_syntax_list.cpp b/src/shader_recompiler/ir/abstract_syntax_list.cpp new file mode 100644 index 000000000..0d967ac11 --- /dev/null +++ b/src/shader_recompiler/ir/abstract_syntax_list.cpp @@ -0,0 +1,44 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#include "abstract_syntax_list.h" + +namespace Shader::IR { + +std::string DumpASLNode(const AbstractSyntaxNode& node, + const std::map& block_to_index, + const std::map& inst_to_index) { + switch (node.type) { + case AbstractSyntaxNode::Type::Block: + return fmt::format("Block: ${}", block_to_index.at(node.data.block)); + case AbstractSyntaxNode::Type::If: + return fmt::format("If: cond = %{}, body = ${}, merge = ${}", + inst_to_index.at(node.data.if_node.cond.Inst()), + block_to_index.at(node.data.if_node.body), + block_to_index.at(node.data.if_node.merge)); + case AbstractSyntaxNode::Type::EndIf: + return fmt::format("EndIf: merge = ${}", block_to_index.at(node.data.end_if.merge)); + case AbstractSyntaxNode::Type::Loop: + return fmt::format("Loop: body = ${}, continue = ${}, merge = ${}", + block_to_index.at(node.data.loop.body), + block_to_index.at(node.data.loop.continue_block), + block_to_index.at(node.data.loop.merge)); + case AbstractSyntaxNode::Type::Repeat: + return fmt::format("Repeat: cond = %{}, header = ${}, merge = ${}", + inst_to_index.at(node.data.repeat.cond.Inst()), + block_to_index.at(node.data.repeat.loop_header), + block_to_index.at(node.data.repeat.merge)); + case AbstractSyntaxNode::Type::Break: + return fmt::format("Break: cond = %{}, merge = ${}, skip = ${}", + inst_to_index.at(node.data.break_node.cond.Inst()), + block_to_index.at(node.data.break_node.merge), + block_to_index.at(node.data.break_node.skip)); + case AbstractSyntaxNode::Type::Return: + return "Return"; + case AbstractSyntaxNode::Type::Unreachable: + return "Unreachable"; + }; + UNREACHABLE(); +} + +} // namespace Shader::IR \ No newline at end of file diff --git a/src/shader_recompiler/ir/abstract_syntax_list.h b/src/shader_recompiler/ir/abstract_syntax_list.h index 313a23abc..a620baccb 100644 --- a/src/shader_recompiler/ir/abstract_syntax_list.h +++ b/src/shader_recompiler/ir/abstract_syntax_list.h @@ -3,6 +3,7 @@ #pragma once +#include #include #include "shader_recompiler/ir/value.h" @@ -53,4 +54,8 @@ struct AbstractSyntaxNode { }; using AbstractSyntaxList = std::vector; +std::string DumpASLNode(const AbstractSyntaxNode& node, + const std::map& block_to_index, + const std::map& inst_to_index); + } // namespace Shader::IR diff --git a/src/shader_recompiler/ir/passes/shader_info_collection_pass.cpp b/src/shader_recompiler/ir/passes/shader_info_collection_pass.cpp index f53a0f4d4..d4759b32e 100644 --- a/src/shader_recompiler/ir/passes/shader_info_collection_pass.cpp +++ b/src/shader_recompiler/ir/passes/shader_info_collection_pass.cpp @@ -2,6 +2,7 @@ // SPDX-License-Identifier: GPL-2.0-or-later #include "shader_recompiler/ir/program.h" +#include "video_core/buffer_cache/buffer_cache.h" namespace Shader::Optimization { @@ -79,14 +80,21 @@ void Visit(Info& info, const IR::Inst& inst) { info.uses_lane_id = true; break; case IR::Opcode::ReadConst: - if (!info.has_readconst) { + if (info.readconst_types == Info::ReadConstType::None) { info.buffers.push_back({ .used_types = IR::Type::U32, - .inline_cbuf = AmdGpu::Buffer::Null(), - .buffer_type = BufferType::ReadConstUbo, + // We can't guarantee that flatbuf will not grow past UBO + // limit if there are a lot of ReadConsts. (We could specialize) + .inline_cbuf = AmdGpu::Buffer::Placeholder(std::numeric_limits::max()), + .buffer_type = BufferType::Flatbuf, }); - info.has_readconst = true; } + if (inst.Flags() != 0) { + info.readconst_types |= Info::ReadConstType::Immediate; + } else { + info.readconst_types |= Info::ReadConstType::Dynamic; + } + info.dma_types |= IR::Type::U32; break; case IR::Opcode::PackUfloat10_11_11: info.uses_pack_10_11_11 = true; @@ -105,6 +113,21 @@ void CollectShaderInfoPass(IR::Program& program) { Visit(program.info, inst); } } + + if (program.info.dma_types != IR::Type::Void) { + program.info.buffers.push_back({ + .used_types = IR::Type::U64, + .inline_cbuf = AmdGpu::Buffer::Placeholder(VideoCore::BufferCache::BDA_PAGETABLE_SIZE), + .buffer_type = BufferType::BdaPagetable, + .is_written = true, + }); + program.info.buffers.push_back({ + .used_types = IR::Type::U8, + .inline_cbuf = AmdGpu::Buffer::Placeholder(VideoCore::BufferCache::FAULT_BUFFER_SIZE), + .buffer_type = BufferType::FaultBuffer, + .is_written = true, + }); + } } } // namespace Shader::Optimization diff --git a/src/shader_recompiler/ir/program.cpp b/src/shader_recompiler/ir/program.cpp index 7728a3ccb..f2f6e34fa 100644 --- a/src/shader_recompiler/ir/program.cpp +++ b/src/shader_recompiler/ir/program.cpp @@ -6,13 +6,30 @@ #include +#include "common/config.h" +#include "common/io_file.h" +#include "common/path_util.h" #include "shader_recompiler/ir/basic_block.h" #include "shader_recompiler/ir/program.h" #include "shader_recompiler/ir/value.h" namespace Shader::IR { -std::string DumpProgram(const Program& program) { +void DumpProgram(const Program& program, const Info& info, const std::string& type) { + using namespace Common::FS; + + if (!Config::dumpShaders()) { + return; + } + + const auto dump_dir = GetUserPath(PathType::ShaderDir) / "dumps"; + if (!std::filesystem::exists(dump_dir)) { + std::filesystem::create_directories(dump_dir); + } + const auto ir_filename = + fmt::format("{}_{:#018x}.{}irprogram.txt", info.stage, info.pgm_hash, type); + const auto ir_file = IOFile{dump_dir / ir_filename, FileAccessMode::Write, FileType::TextFile}; + size_t index{0}; std::map inst_to_index; std::map block_to_index; @@ -21,11 +38,20 @@ std::string DumpProgram(const Program& program) { block_to_index.emplace(block, index); ++index; } - std::string ret; + for (const auto& block : program.blocks) { - ret += IR::DumpBlock(*block, block_to_index, inst_to_index, index) + '\n'; + std::string s = IR::DumpBlock(*block, block_to_index, inst_to_index, index) + '\n'; + ir_file.WriteString(s); + } + + const auto asl_filename = fmt::format("{}_{:#018x}.{}asl.txt", info.stage, info.pgm_hash, type); + const auto asl_file = + IOFile{dump_dir / asl_filename, FileAccessMode::Write, FileType::TextFile}; + + for (const auto& node : program.syntax_list) { + std::string s = IR::DumpASLNode(node, block_to_index, inst_to_index) + '\n'; + asl_file.WriteString(s); } - return ret; } } // namespace Shader::IR diff --git a/src/shader_recompiler/ir/program.h b/src/shader_recompiler/ir/program.h index 84a1a2d40..3ffd4dc96 100644 --- a/src/shader_recompiler/ir/program.h +++ b/src/shader_recompiler/ir/program.h @@ -21,6 +21,6 @@ struct Program { Info& info; }; -[[nodiscard]] std::string DumpProgram(const Program& program); +void DumpProgram(const Program& program, const Info& info, const std::string& type = ""); } // namespace Shader::IR diff --git a/src/shader_recompiler/ir/reinterpret.h b/src/shader_recompiler/ir/reinterpret.h index 99819cbb9..2a18f394a 100644 --- a/src/shader_recompiler/ir/reinterpret.h +++ b/src/shader_recompiler/ir/reinterpret.h @@ -46,6 +46,10 @@ inline F32 ApplyReadNumberConversion(IREmitter& ir, const F32& value, const IR::F32 max = ir.Imm32(float(std::numeric_limits::max())); return ir.FPDiv(left, max); } + case AmdGpu::NumberConversion::Uint32ToUnorm: { + const auto float_val = ir.ConvertUToF(32, 32, ir.BitCast(value)); + return ir.FPDiv(float_val, ir.Imm32(static_cast(std::numeric_limits::max()))); + } default: UNREACHABLE(); } @@ -92,6 +96,12 @@ inline F32 ApplyWriteNumberConversion(IREmitter& ir, const F32& value, const IR::U32 raw = ir.ConvertFToS(32, ir.FPDiv(left, ir.Imm32(2.f))); return ir.BitCast(raw); } + case AmdGpu::NumberConversion::Uint32ToUnorm: { + const auto clamped = ir.FPClamp(value, ir.Imm32(0.f), ir.Imm32(1.f)); + const auto unnormalized = + ir.FPMul(clamped, ir.Imm32(static_cast(std::numeric_limits::max()))); + return ir.BitCast(U32{ir.ConvertFToU(32, unnormalized)}); + } default: UNREACHABLE(); } diff --git a/src/shader_recompiler/recompiler.cpp b/src/shader_recompiler/recompiler.cpp index 3e0bd98d2..9f92857d6 100644 --- a/src/shader_recompiler/recompiler.cpp +++ b/src/shader_recompiler/recompiler.cpp @@ -85,6 +85,8 @@ IR::Program TranslateProgram(std::span code, Pools& pools, Info& info Shader::Optimization::ConstantPropagationPass(program.post_order_blocks); Shader::Optimization::CollectShaderInfoPass(program); + Shader::IR::DumpProgram(program, info); + return program; } diff --git a/src/video_core/amdgpu/liverpool.cpp b/src/video_core/amdgpu/liverpool.cpp index 598288085..706e94c2b 100644 --- a/src/video_core/amdgpu/liverpool.cpp +++ b/src/video_core/amdgpu/liverpool.cpp @@ -133,6 +133,7 @@ void Liverpool::Process(std::stop_token stoken) { VideoCore::EndCapture(); if (rasterizer) { + rasterizer->ProcessFaults(); rasterizer->Flush(); } submit_done = false; @@ -584,7 +585,16 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span dcb, std::span(header); + const auto* event = reinterpret_cast(header); + LOG_DEBUG(Render_Vulkan, + "Encountered EventWrite: event_type = {}, event_index = {}", + magic_enum::enum_name(event->event_type.Value()), + magic_enum::enum_name(event->event_index.Value())); + if (event->event_type.Value() == EventType::SoVgtStreamoutFlush) { + // TODO: handle proper synchronization, for now signal that update is done + // immediately + regs.cp_strmout_cntl.offset_update_done = 1; + } break; } case PM4ItOpcode::EventWriteEos: { @@ -696,10 +706,10 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span dcb, std::spanAddress(); if (vo_port->IsVoLabel(wait_addr) && num_submits == mapped_queues[GfxQueueId].submits.size()) { - vo_port->WaitVoLabel([&] { return wait_reg_mem->Test(); }); + vo_port->WaitVoLabel([&] { return wait_reg_mem->Test(regs.reg_array); }); break; } - while (!wait_reg_mem->Test()) { + while (!wait_reg_mem->Test(regs.reg_array)) { YIELD_GFX(); } break; @@ -732,6 +742,16 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span dcb, std::span(header); + LOG_WARNING(Render_Vulkan, + "Unimplemented IT_STRMOUT_BUFFER_UPDATE, update_memory = {}, " + "source_select = {}, buffer_select = {}", + strmout->update_memory.Value(), + magic_enum::enum_name(strmout->source_select.Value()), + strmout->buffer_select.Value()); + break; + } default: UNREACHABLE_MSG("Unknown PM4 type 3 opcode {:#x} with count {}", static_cast(opcode), count); @@ -866,8 +886,9 @@ Liverpool::Task Liverpool::ProcessCompute(const u32* acb, u32 acb_dwords, u32 vq } case PM4ItOpcode::SetQueueReg: { const auto* set_data = reinterpret_cast(header); - UNREACHABLE_MSG("Encountered compute SetQueueReg: vqid = {}, reg_offset = {:#x}", - set_data->vqid.Value(), set_data->reg_offset.Value()); + LOG_WARNING(Render, "Encountered compute SetQueueReg: vqid = {}, reg_offset = {:#x}", + set_data->vqid.Value(), set_data->reg_offset.Value()); + break; } case PM4ItOpcode::DispatchDirect: { const auto* dispatch_direct = reinterpret_cast(header); @@ -934,7 +955,7 @@ Liverpool::Task Liverpool::ProcessCompute(const u32* acb, u32 acb_dwords, u32 vq case PM4ItOpcode::WaitRegMem: { const auto* wait_reg_mem = reinterpret_cast(header); ASSERT(wait_reg_mem->engine.Value() == PM4CmdWaitRegMem::Engine::Me); - while (!wait_reg_mem->Test()) { + while (!wait_reg_mem->Test(regs.reg_array)) { YIELD_ASC(vqid); } break; diff --git a/src/video_core/amdgpu/liverpool.h b/src/video_core/amdgpu/liverpool.h index c4bebd05f..a62141099 100644 --- a/src/video_core/amdgpu/liverpool.h +++ b/src/video_core/amdgpu/liverpool.h @@ -1175,6 +1175,14 @@ struct Liverpool { BitField<22, 2, u32> onchip; }; + union StreamOutControl { + u32 raw; + struct { + u32 offset_update_done : 1; + u32 : 31; + }; + }; + union StreamOutConfig { u32 raw; struct { @@ -1378,7 +1386,9 @@ struct Liverpool { AaConfig aa_config; INSERT_PADDING_WORDS(0xA318 - 0xA2F8 - 1); ColorBuffer color_buffers[NumColorBuffers]; - INSERT_PADDING_WORDS(0xC242 - 0xA390); + INSERT_PADDING_WORDS(0xC03F - 0xA390); + StreamOutControl cp_strmout_cntl; + INSERT_PADDING_WORDS(0xC242 - 0xC040); PrimitiveType primitive_type; INSERT_PADDING_WORDS(0xC24C - 0xC243); u32 num_indices; @@ -1668,6 +1678,7 @@ static_assert(GFX6_3D_REG_INDEX(color_buffers[0].base_address) == 0xA318); static_assert(GFX6_3D_REG_INDEX(color_buffers[0].pitch) == 0xA319); static_assert(GFX6_3D_REG_INDEX(color_buffers[0].slice) == 0xA31A); static_assert(GFX6_3D_REG_INDEX(color_buffers[7].base_address) == 0xA381); +static_assert(GFX6_3D_REG_INDEX(cp_strmout_cntl) == 0xC03F); static_assert(GFX6_3D_REG_INDEX(primitive_type) == 0xC242); static_assert(GFX6_3D_REG_INDEX(num_instances) == 0xC24D); static_assert(GFX6_3D_REG_INDEX(vgt_tf_memory_base) == 0xc250); diff --git a/src/video_core/amdgpu/pm4_cmds.h b/src/video_core/amdgpu/pm4_cmds.h index cd175f6c9..58ecda93e 100644 --- a/src/video_core/amdgpu/pm4_cmds.h +++ b/src/video_core/amdgpu/pm4_cmds.h @@ -246,6 +246,46 @@ struct PM4CmdNop { }; }; +enum class SourceSelect : u32 { + BufferOffset = 0, + VgtStrmoutBufferFilledSize = 1, + SrcAddress = 2, + None = 3, +}; + +struct PM4CmdStrmoutBufferUpdate { + PM4Type3Header header; + union { + BitField<0, 1, u32> update_memory; + BitField<1, 2, SourceSelect> source_select; + BitField<8, 2, u32> buffer_select; + u32 control; + }; + union { + BitField<2, 30, u32> dst_address_lo; + BitField<0, 2, u32> swap_dst; + }; + u32 dst_address_hi; + union { + u32 buffer_offset; + BitField<2, 30, u32> src_address_lo; + BitField<0, 2, u32> swap_src; + }; + u32 src_address_hi; + + template + T DstAddress() const { + ASSERT(update_memory.Value() == 1); + return reinterpret_cast(dst_address_lo.Value() | u64(dst_address_hi & 0xFFFF) << 32); + } + + template + T SrcAddress() const { + ASSERT(source_select.Value() == SourceSelect::SrcAddress); + return reinterpret_cast(src_address_lo.Value() | u64(src_address_hi & 0xFFFF) << 32); + } +}; + struct PM4CmdDrawIndexOffset2 { PM4Type3Header header; u32 max_size; ///< Maximum number of indices @@ -303,6 +343,80 @@ static u64 GetGpuClock64() { return static_cast(ticks); } +// VGT_EVENT_INITIATOR.EVENT_TYPE +enum class EventType : u32 { + SampleStreamoutStats1 = 1, + SampleStreamoutStats2 = 2, + SampleStreamoutStats3 = 3, + CacheFlushTs = 4, + ContextDone = 5, + CacheFlush = 6, + CsPartialFlush = 7, + VgtStreamoutSync = 8, + VgtStreamoutReset = 10, + EndOfPipeIncrDe = 11, + EndOfPipeIbEnd = 12, + RstPixCnt = 13, + VsPartialFlush = 15, + PsPartialFlush = 16, + FlushHsOutput = 17, + FlushLsOutput = 18, + CacheFlushAndInvTsEvent = 20, + ZpassDone = 21, + CacheFlushAndInvEvent = 22, + PerfcounterStart = 23, + PerfcounterStop = 24, + PipelineStatStart = 25, + PipelineStatStop = 26, + PerfcounterSample = 27, + FlushEsOutput = 28, + FlushGsOutput = 29, + SamplePipelineStat = 30, + SoVgtStreamoutFlush = 31, + SampleStreamoutStats = 32, + ResetVtxCnt = 33, + VgtFlush = 36, + ScSendDbVpz = 39, + BottomOfPipeTs = 40, + DbCacheFlushAndInv = 42, + FlushAndInvDbDataTs = 43, + FlushAndInvDbMeta = 44, + FlushAndInvCbDataTs = 45, + FlushAndInvCbMeta = 46, + CsDone = 47, + PsDone = 48, + FlushAndInvCbPixelData = 49, + ThreadTraceStart = 51, + ThreadTraceStop = 52, + ThreadTraceFlush = 54, + ThreadTraceFinish = 55, + PixelPipeStatControl = 56, + PixelPipeStatDump = 57, + PixelPipeStatReset = 58, +}; + +enum class EventIndex : u32 { + Other = 0, + ZpassDone = 1, + SamplePipelineStat = 2, + SampleStreamoutStatSx = 3, + CsVsPsPartialFlush = 4, + EopReserved = 5, + EosReserved = 6, + CacheFlush = 7, +}; + +struct PM4CmdEventWrite { + PM4Type3Header header; + union { + u32 event_control; + BitField<0, 6, EventType> event_type; ///< Event type written to VGT_EVENT_INITIATOR + BitField<8, 4, EventIndex> event_index; ///< Event index + BitField<20, 1, u32> inv_l2; ///< Send WBINVL2 op to the TC L2 cache when EVENT_INDEX = 0111 + }; + u32 address[]; +}; + struct PM4CmdEventWriteEop { PM4Type3Header header; union { @@ -474,7 +588,12 @@ struct PM4CmdWaitRegMem { BitField<8, 1, Engine> engine; u32 raw; }; - u32 poll_addr_lo; + union { + BitField<0, 16, u32> reg; + BitField<2, 30, u32> poll_addr_lo; + BitField<0, 2, u32> swap; + u32 poll_addr_lo_raw; + }; u32 poll_addr_hi; u32 ref; u32 mask; @@ -482,31 +601,36 @@ struct PM4CmdWaitRegMem { template T Address() const { - return std::bit_cast((uintptr_t(poll_addr_hi) << 32) | poll_addr_lo); + return std::bit_cast((uintptr_t(poll_addr_hi) << 32) | (poll_addr_lo << 2)); } - bool Test() const { + u32 Reg() const { + return reg.Value(); + } + + bool Test(const std::array& regs) const { + u32 value = mem_space.Value() == MemSpace::Memory ? *Address() : regs[Reg()]; switch (function.Value()) { case Function::Always: { return true; } case Function::LessThan: { - return (*Address() & mask) < ref; + return (value & mask) < ref; } case Function::LessThanEqual: { - return (*Address() & mask) <= ref; + return (value & mask) <= ref; } case Function::Equal: { - return (*Address() & mask) == ref; + return (value & mask) == ref; } case Function::NotEqual: { - return (*Address() & mask) != ref; + return (value & mask) != ref; } case Function::GreaterThanEqual: { - return (*Address() & mask) >= ref; + return (value & mask) >= ref; } case Function::GreaterThan: { - return (*Address() & mask) > ref; + return (value & mask) > ref; } case Function::Reserved: [[fallthrough]]; diff --git a/src/video_core/amdgpu/resource.h b/src/video_core/amdgpu/resource.h index 9060074fb..89ac04f9a 100644 --- a/src/video_core/amdgpu/resource.h +++ b/src/video_core/amdgpu/resource.h @@ -37,6 +37,13 @@ struct Buffer { return buffer; } + static constexpr Buffer Placeholder(u32 size) { + Buffer buffer{}; + buffer.base_address = 1; + buffer.num_records = size; + return buffer; + } + bool Valid() const { return type == 0u; } diff --git a/src/video_core/amdgpu/types.h b/src/video_core/amdgpu/types.h index ab0df689e..f7536f7e2 100644 --- a/src/video_core/amdgpu/types.h +++ b/src/video_core/amdgpu/types.h @@ -197,8 +197,9 @@ enum class NumberConversion : u32 { UintToUscaled = 1, SintToSscaled = 2, UnormToUbnorm = 3, - Sint8ToSnormNz = 5, - Sint16ToSnormNz = 6, + Sint8ToSnormNz = 4, + Sint16ToSnormNz = 5, + Uint32ToUnorm = 6, }; struct CompMapping { @@ -286,6 +287,17 @@ inline DataFormat RemapDataFormat(const DataFormat format) { inline NumberFormat RemapNumberFormat(const NumberFormat format, const DataFormat data_format) { switch (format) { + case NumberFormat::Unorm: { + switch (data_format) { + case DataFormat::Format32: + case DataFormat::Format32_32: + case DataFormat::Format32_32_32: + case DataFormat::Format32_32_32_32: + return NumberFormat::Uint; + default: + return format; + } + } case NumberFormat::Uscaled: return NumberFormat::Uint; case NumberFormat::Sscaled: @@ -341,6 +353,17 @@ inline CompMapping RemapSwizzle(const DataFormat format, const CompMapping swizz inline NumberConversion MapNumberConversion(const NumberFormat num_fmt, const DataFormat data_fmt) { switch (num_fmt) { + case NumberFormat::Unorm: { + switch (data_fmt) { + case DataFormat::Format32: + case DataFormat::Format32_32: + case DataFormat::Format32_32_32: + case DataFormat::Format32_32_32_32: + return NumberConversion::Uint32ToUnorm; + default: + return NumberConversion::None; + } + } case NumberFormat::Uscaled: return NumberConversion::UintToUscaled; case NumberFormat::Sscaled: diff --git a/src/video_core/buffer_cache/buffer.cpp b/src/video_core/buffer_cache/buffer.cpp index 15ef746cd..15bf0d81e 100644 --- a/src/video_core/buffer_cache/buffer.cpp +++ b/src/video_core/buffer_cache/buffer.cpp @@ -70,8 +70,11 @@ UniqueBuffer::~UniqueBuffer() { void UniqueBuffer::Create(const vk::BufferCreateInfo& buffer_ci, MemoryUsage usage, VmaAllocationInfo* out_alloc_info) { + const bool with_bda = bool(buffer_ci.usage & vk::BufferUsageFlagBits::eShaderDeviceAddress); + const VmaAllocationCreateFlags bda_flag = + with_bda ? VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT : 0; const VmaAllocationCreateInfo alloc_ci = { - .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | MemoryUsageVmaFlags(usage), + .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT | bda_flag | MemoryUsageVmaFlags(usage), .usage = MemoryUsageVma(usage), .requiredFlags = 0, .preferredFlags = MemoryUsagePreferredVmaFlags(usage), @@ -86,6 +89,15 @@ void UniqueBuffer::Create(const vk::BufferCreateInfo& buffer_ci, MemoryUsage usa ASSERT_MSG(result == VK_SUCCESS, "Failed allocating buffer with error {}", vk::to_string(vk::Result{result})); buffer = vk::Buffer{unsafe_buffer}; + + if (with_bda) { + vk::BufferDeviceAddressInfo bda_info{ + .buffer = buffer, + }; + auto bda_result = device.getBufferAddress(bda_info); + ASSERT_MSG(bda_result != 0, "Failed to get buffer device address"); + bda_addr = bda_result; + } } Buffer::Buffer(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_, MemoryUsage usage_, diff --git a/src/video_core/buffer_cache/buffer.h b/src/video_core/buffer_cache/buffer.h index 188b4b2ca..530968787 100644 --- a/src/video_core/buffer_cache/buffer.h +++ b/src/video_core/buffer_cache/buffer.h @@ -68,6 +68,7 @@ struct UniqueBuffer { VmaAllocator allocator; VmaAllocation allocation; vk::Buffer buffer{}; + vk::DeviceAddress bda_addr = 0; }; class Buffer { @@ -115,6 +116,11 @@ public: return buffer; } + vk::DeviceAddress BufferDeviceAddress() const noexcept { + ASSERT_MSG(buffer.bda_addr != 0, "Can't get BDA from a non BDA buffer"); + return buffer.bda_addr; + } + std::optional GetBarrier( vk::Flags dst_acess_mask, vk::PipelineStageFlagBits2 dst_stage, u32 offset = 0) { diff --git a/src/video_core/buffer_cache/buffer_cache.cpp b/src/video_core/buffer_cache/buffer_cache.cpp index c993ef3e5..4717a5ff8 100644 --- a/src/video_core/buffer_cache/buffer_cache.cpp +++ b/src/video_core/buffer_cache/buffer_cache.cpp @@ -3,13 +3,17 @@ #include #include "common/alignment.h" +#include "common/debug.h" #include "common/scope_exit.h" #include "common/types.h" #include "video_core/amdgpu/liverpool.h" #include "video_core/buffer_cache/buffer_cache.h" +#include "video_core/host_shaders/fault_buffer_process_comp.h" #include "video_core/renderer_vulkan/vk_graphics_pipeline.h" #include "video_core/renderer_vulkan/vk_instance.h" +#include "video_core/renderer_vulkan/vk_rasterizer.h" #include "video_core/renderer_vulkan/vk_scheduler.h" +#include "video_core/renderer_vulkan/vk_shader_util.h" #include "video_core/texture_cache/texture_cache.h" namespace VideoCore { @@ -17,17 +21,26 @@ namespace VideoCore { static constexpr size_t DataShareBufferSize = 64_KB; static constexpr size_t StagingBufferSize = 512_MB; static constexpr size_t UboStreamBufferSize = 128_MB; +static constexpr size_t DownloadBufferSize = 128_MB; +static constexpr size_t MaxPageFaults = 1024; BufferCache::BufferCache(const Vulkan::Instance& instance_, Vulkan::Scheduler& scheduler_, - AmdGpu::Liverpool* liverpool_, TextureCache& texture_cache_, - PageManager& tracker_) - : instance{instance_}, scheduler{scheduler_}, liverpool{liverpool_}, + Vulkan::Rasterizer& rasterizer_, AmdGpu::Liverpool* liverpool_, + TextureCache& texture_cache_, PageManager& tracker_) + : instance{instance_}, scheduler{scheduler_}, rasterizer{rasterizer_}, liverpool{liverpool_}, texture_cache{texture_cache_}, tracker{tracker_}, staging_buffer{instance, scheduler, MemoryUsage::Upload, StagingBufferSize}, stream_buffer{instance, scheduler, MemoryUsage::Stream, UboStreamBufferSize}, + download_buffer(instance, scheduler, MemoryUsage::Download, DownloadBufferSize), gds_buffer{instance, scheduler, MemoryUsage::Stream, 0, AllFlags, DataShareBufferSize}, - memory_tracker{&tracker} { + bda_pagetable_buffer{instance, scheduler, MemoryUsage::DeviceLocal, + 0, AllFlags, BDA_PAGETABLE_SIZE}, + fault_buffer(instance, scheduler, MemoryUsage::DeviceLocal, 0, AllFlags, FAULT_BUFFER_SIZE), + memory_tracker{tracker} { Vulkan::SetObjectName(instance.GetDevice(), gds_buffer.Handle(), "GDS Buffer"); + Vulkan::SetObjectName(instance.GetDevice(), bda_pagetable_buffer.Handle(), + "BDA Page Table Buffer"); + Vulkan::SetObjectName(instance.GetDevice(), fault_buffer.Handle(), "Fault Buffer"); // Ensure the first slot is used for the null buffer const auto null_id = @@ -35,15 +48,93 @@ BufferCache::BufferCache(const Vulkan::Instance& instance_, Vulkan::Scheduler& s ASSERT(null_id.index == 0); const vk::Buffer& null_buffer = slot_buffers[null_id].buffer; Vulkan::SetObjectName(instance.GetDevice(), null_buffer, "Null Buffer"); + + // Prepare the fault buffer parsing pipeline + boost::container::static_vector bindings{ + { + .binding = 0, + .descriptorType = vk::DescriptorType::eStorageBuffer, + .descriptorCount = 1, + .stageFlags = vk::ShaderStageFlagBits::eCompute, + }, + { + .binding = 1, + .descriptorType = vk::DescriptorType::eStorageBuffer, + .descriptorCount = 1, + .stageFlags = vk::ShaderStageFlagBits::eCompute, + }, + }; + + const vk::DescriptorSetLayoutCreateInfo desc_layout_ci = { + .flags = vk::DescriptorSetLayoutCreateFlagBits::ePushDescriptorKHR, + .bindingCount = static_cast(bindings.size()), + .pBindings = bindings.data(), + }; + auto [desc_layout_result, desc_layout] = + instance.GetDevice().createDescriptorSetLayoutUnique(desc_layout_ci); + ASSERT_MSG(desc_layout_result == vk::Result::eSuccess, + "Failed to create descriptor set layout: {}", vk::to_string(desc_layout_result)); + fault_process_desc_layout = std::move(desc_layout); + + const auto& module = Vulkan::Compile(HostShaders::FAULT_BUFFER_PROCESS_COMP, + vk::ShaderStageFlagBits::eCompute, instance.GetDevice()); + Vulkan::SetObjectName(instance.GetDevice(), module, "Fault Buffer Parser"); + + const vk::SpecializationMapEntry specialization_map_entry = { + .constantID = 0, + .offset = 0, + .size = sizeof(u32), + }; + + const vk::SpecializationInfo specialization_info = { + .mapEntryCount = 1, + .pMapEntries = &specialization_map_entry, + .dataSize = sizeof(u32), + .pData = &CACHING_PAGEBITS, + }; + + const vk::PipelineShaderStageCreateInfo shader_ci = { + .stage = vk::ShaderStageFlagBits::eCompute, + .module = module, + .pName = "main", + .pSpecializationInfo = &specialization_info, + }; + + const vk::PipelineLayoutCreateInfo layout_info = { + .setLayoutCount = 1U, + .pSetLayouts = &(*fault_process_desc_layout), + }; + auto [layout_result, layout] = instance.GetDevice().createPipelineLayoutUnique(layout_info); + ASSERT_MSG(layout_result == vk::Result::eSuccess, "Failed to create pipeline layout: {}", + vk::to_string(layout_result)); + fault_process_pipeline_layout = std::move(layout); + + const vk::ComputePipelineCreateInfo pipeline_info = { + .stage = shader_ci, + .layout = *fault_process_pipeline_layout, + }; + auto [pipeline_result, pipeline] = + instance.GetDevice().createComputePipelineUnique({}, pipeline_info); + ASSERT_MSG(pipeline_result == vk::Result::eSuccess, "Failed to create compute pipeline: {}", + vk::to_string(pipeline_result)); + fault_process_pipeline = std::move(pipeline); + Vulkan::SetObjectName(instance.GetDevice(), *fault_process_pipeline, + "Fault Buffer Parser Pipeline"); + + instance.GetDevice().destroyShaderModule(module); } BufferCache::~BufferCache() = default; -void BufferCache::InvalidateMemory(VAddr device_addr, u64 size) { +void BufferCache::InvalidateMemory(VAddr device_addr, u64 size, bool unmap) { const bool is_tracked = IsRegionRegistered(device_addr, size); if (is_tracked) { // Mark the page as CPU modified to stop tracking writes. memory_tracker.MarkRegionAsCpuModified(device_addr, size); + + if (unmap) { + return; + } } } @@ -69,20 +160,20 @@ void BufferCache::DownloadBufferMemory(Buffer& buffer, VAddr device_addr, u64 si if (total_size_bytes == 0) { return; } - const auto [staging, offset] = staging_buffer.Map(total_size_bytes); + const auto [download, offset] = download_buffer.Map(total_size_bytes); for (auto& copy : copies) { // Modify copies to have the staging offset in mind copy.dstOffset += offset; } - staging_buffer.Commit(); + download_buffer.Commit(); scheduler.EndRendering(); const auto cmdbuf = scheduler.CommandBuffer(); - cmdbuf.copyBuffer(buffer.buffer, staging_buffer.Handle(), copies); + cmdbuf.copyBuffer(buffer.buffer, download_buffer.Handle(), copies); scheduler.Finish(); for (const auto& copy : copies) { const VAddr copy_device_addr = buffer.CpuAddr() + copy.srcOffset; const u64 dst_offset = copy.dstOffset - offset; - std::memcpy(std::bit_cast(copy_device_addr), staging + dst_offset, copy.size); + std::memcpy(std::bit_cast(copy_device_addr), download + dst_offset, copy.size); } } @@ -206,58 +297,37 @@ void BufferCache::InlineData(VAddr address, const void* value, u32 num_bytes, bo memcpy(std::bit_cast(address), value, num_bytes); return; } - scheduler.EndRendering(); - const Buffer* buffer = [&] { + Buffer* buffer = [&] { if (is_gds) { return &gds_buffer; } const BufferId buffer_id = FindBuffer(address, num_bytes); return &slot_buffers[buffer_id]; }(); - 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(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); + InlineDataBuffer(*buffer, address, value, num_bytes); +} + +void BufferCache::WriteData(VAddr address, const void* value, u32 num_bytes, bool is_gds) { + ASSERT_MSG(address % 4 == 0, "GDS offset must be dword aligned"); + if (!is_gds && !IsRegionRegistered(address, num_bytes)) { + memcpy(std::bit_cast(address), value, num_bytes); + return; } - cmdbuf.pipelineBarrier2(vk::DependencyInfo{ - .dependencyFlags = vk::DependencyFlagBits::eByRegion, - .bufferMemoryBarrierCount = 1, - .pBufferMemoryBarriers = &post_barrier, - }); + Buffer* buffer = [&] { + if (is_gds) { + return &gds_buffer; + } + const BufferId buffer_id = FindBuffer(address, num_bytes); + return &slot_buffers[buffer_id]; + }(); + WriteDataBuffer(*buffer, address, value, num_bytes); } std::pair BufferCache::ObtainBuffer(VAddr device_addr, u32 size, bool is_written, bool is_texel_buffer, BufferId buffer_id) { // For small uniform buffers that have not been modified by gpu // use device local stream buffer to reduce renderpass breaks. + // Maybe we want to modify the threshold now that the page size is 16KB? static constexpr u64 StreamThreshold = CACHING_PAGESIZE; const bool is_gpu_dirty = memory_tracker.IsRegionGpuModified(device_addr, size); if (!is_written && size <= StreamThreshold && !is_gpu_dirty) { @@ -280,7 +350,7 @@ std::pair BufferCache::ObtainBuffer(VAddr device_addr, u32 size, b std::pair BufferCache::ObtainViewBuffer(VAddr gpu_addr, u32 size, bool prefer_gpu) { // Check if any buffer contains the full requested range. const u64 page = gpu_addr >> CACHING_PAGEBITS; - const BufferId buffer_id = page_table[page]; + const BufferId buffer_id = page_table[page].buffer_id; if (buffer_id) { Buffer& buffer = slot_buffers[buffer_id]; if (buffer.IsInBounds(gpu_addr, size)) { @@ -300,24 +370,8 @@ std::pair BufferCache::ObtainViewBuffer(VAddr gpu_addr, u32 size, } bool BufferCache::IsRegionRegistered(VAddr addr, size_t size) { - const VAddr end_addr = addr + size; - const u64 page_end = Common::DivCeil(end_addr, CACHING_PAGESIZE); - for (u64 page = addr >> CACHING_PAGEBITS; page < page_end;) { - const BufferId buffer_id = page_table[page]; - if (!buffer_id) { - ++page; - continue; - } - std::shared_lock lk{mutex}; - Buffer& buffer = slot_buffers[buffer_id]; - const VAddr buf_start_addr = buffer.CpuAddr(); - const VAddr buf_end_addr = buf_start_addr + buffer.SizeBytes(); - if (buf_start_addr < end_addr && addr < buf_end_addr) { - return true; - } - page = Common::DivCeil(buf_end_addr, CACHING_PAGESIZE); - } - return false; + // Check if we are missing some edge case here + return buffer_ranges.Intersects(addr, size); } bool BufferCache::IsRegionCpuModified(VAddr addr, size_t size) { @@ -333,7 +387,7 @@ BufferId BufferCache::FindBuffer(VAddr device_addr, u32 size) { return NULL_BUFFER_ID; } const u64 page = device_addr >> CACHING_PAGEBITS; - const BufferId buffer_id = page_table[page]; + const BufferId buffer_id = page_table[page].buffer_id; if (!buffer_id) { return CreateBuffer(device_addr, size); } @@ -379,7 +433,7 @@ BufferCache::OverlapResult BufferCache::ResolveOverlaps(VAddr device_addr, u32 w } for (; device_addr >> CACHING_PAGEBITS < Common::DivCeil(end, CACHING_PAGESIZE); device_addr += CACHING_PAGESIZE) { - const BufferId overlap_id = page_table[device_addr >> CACHING_PAGEBITS]; + const BufferId overlap_id = page_table[device_addr >> CACHING_PAGEBITS].buffer_id; if (!overlap_id) { continue; } @@ -480,11 +534,21 @@ BufferId BufferCache::CreateBuffer(VAddr device_addr, u32 wanted_size) { const OverlapResult overlap = ResolveOverlaps(device_addr, wanted_size); const u32 size = static_cast(overlap.end - overlap.begin); const BufferId new_buffer_id = [&] { - std::scoped_lock lk{mutex}; + std::scoped_lock lk{slot_buffers_mutex}; return slot_buffers.insert(instance, scheduler, MemoryUsage::DeviceLocal, overlap.begin, - AllFlags, size); + AllFlags | vk::BufferUsageFlagBits::eShaderDeviceAddress, size); }(); auto& new_buffer = slot_buffers[new_buffer_id]; + boost::container::small_vector bda_addrs; + const u64 start_page = overlap.begin >> CACHING_PAGEBITS; + const u64 size_pages = size >> CACHING_PAGEBITS; + bda_addrs.reserve(size_pages); + for (u64 i = 0; i < size_pages; ++i) { + vk::DeviceAddress addr = new_buffer.BufferDeviceAddress() + (i << CACHING_PAGEBITS); + bda_addrs.push_back(addr); + } + WriteDataBuffer(bda_pagetable_buffer, start_page * sizeof(vk::DeviceAddress), bda_addrs.data(), + bda_addrs.size() * sizeof(vk::DeviceAddress)); const size_t size_bytes = new_buffer.SizeBytes(); const auto cmdbuf = scheduler.CommandBuffer(); scheduler.EndRendering(); @@ -496,6 +560,129 @@ BufferId BufferCache::CreateBuffer(VAddr device_addr, u32 wanted_size) { return new_buffer_id; } +void BufferCache::ProcessFaultBuffer() { + // Run fault processing shader + const auto [mapped, offset] = download_buffer.Map(MaxPageFaults * sizeof(u64)); + vk::BufferMemoryBarrier2 fault_buffer_barrier{ + .srcStageMask = vk::PipelineStageFlagBits2::eAllCommands, + .srcAccessMask = vk::AccessFlagBits2::eShaderWrite, + .dstStageMask = vk::PipelineStageFlagBits2::eComputeShader, + .dstAccessMask = vk::AccessFlagBits2::eShaderRead, + .buffer = fault_buffer.Handle(), + .offset = 0, + .size = FAULT_BUFFER_SIZE, + }; + vk::BufferMemoryBarrier2 download_barrier{ + .srcStageMask = vk::PipelineStageFlagBits2::eTransfer, + .srcAccessMask = vk::AccessFlagBits2::eTransferWrite, + .dstStageMask = vk::PipelineStageFlagBits2::eComputeShader, + .dstAccessMask = vk::AccessFlagBits2::eShaderRead | vk::AccessFlagBits2::eShaderWrite, + .buffer = download_buffer.Handle(), + .offset = offset, + .size = MaxPageFaults * sizeof(u64), + }; + std::array barriers{fault_buffer_barrier, download_barrier}; + vk::DescriptorBufferInfo fault_buffer_info{ + .buffer = fault_buffer.Handle(), + .offset = 0, + .range = FAULT_BUFFER_SIZE, + }; + vk::DescriptorBufferInfo download_info{ + .buffer = download_buffer.Handle(), + .offset = offset, + .range = MaxPageFaults * sizeof(u64), + }; + boost::container::small_vector writes{ + { + .dstSet = VK_NULL_HANDLE, + .dstBinding = 0, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = vk::DescriptorType::eStorageBuffer, + .pBufferInfo = &fault_buffer_info, + }, + { + .dstSet = VK_NULL_HANDLE, + .dstBinding = 1, + .dstArrayElement = 0, + .descriptorCount = 1, + .descriptorType = vk::DescriptorType::eStorageBuffer, + .pBufferInfo = &download_info, + }, + }; + download_buffer.Commit(); + scheduler.EndRendering(); + const auto cmdbuf = scheduler.CommandBuffer(); + cmdbuf.fillBuffer(download_buffer.Handle(), offset, MaxPageFaults * sizeof(u64), 0); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 2, + .pBufferMemoryBarriers = barriers.data(), + }); + cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, *fault_process_pipeline); + cmdbuf.pushDescriptorSetKHR(vk::PipelineBindPoint::eCompute, *fault_process_pipeline_layout, 0, + writes); + constexpr u32 num_threads = CACHING_NUMPAGES / 32; // 1 bit per page, 32 pages per workgroup + constexpr u32 num_workgroups = Common::DivCeil(num_threads, 64u); + cmdbuf.dispatch(num_workgroups, 1, 1); + + // Reset fault buffer + const vk::BufferMemoryBarrier2 reset_pre_barrier = { + .srcStageMask = vk::PipelineStageFlagBits2::eComputeShader, + .srcAccessMask = vk::AccessFlagBits2::eShaderRead, + .dstStageMask = vk::PipelineStageFlagBits2::eTransfer, + .dstAccessMask = vk::AccessFlagBits2::eTransferWrite, + .buffer = fault_buffer.Handle(), + .offset = 0, + .size = FAULT_BUFFER_SIZE, + }; + const vk::BufferMemoryBarrier2 reset_post_barrier = { + .srcStageMask = vk::PipelineStageFlagBits2::eTransfer, + .srcAccessMask = vk::AccessFlagBits2::eTransferWrite, + .dstStageMask = vk::PipelineStageFlagBits2::eAllCommands, + .dstAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite, + .buffer = fault_buffer.Handle(), + .offset = 0, + .size = FAULT_BUFFER_SIZE, + }; + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &reset_pre_barrier, + }); + cmdbuf.fillBuffer(fault_buffer.buffer, 0, FAULT_BUFFER_SIZE, 0); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &reset_post_barrier, + }); + + // Defer creating buffers + scheduler.DeferOperation([this, mapped]() { + // Create the fault buffers batched + boost::icl::interval_set fault_ranges; + const u64* fault_ptr = std::bit_cast(mapped); + const u32 fault_count = static_cast(*(fault_ptr++)); + for (u32 i = 0; i < fault_count; ++i) { + const VAddr fault = *(fault_ptr++); + const VAddr fault_end = fault + CACHING_PAGESIZE; // This can be adjusted + fault_ranges += + boost::icl::interval_set::interval_type::right_open(fault, fault_end); + LOG_INFO(Render_Vulkan, "Accessed non-GPU cached memory at {:#x}", fault); + } + for (const auto& range : fault_ranges) { + const VAddr start = range.lower(); + const VAddr end = range.upper(); + const u64 page_start = start >> CACHING_PAGEBITS; + const u64 page_end = Common::DivCeil(end, CACHING_PAGESIZE); + // Buffer size is in 32 bits + ASSERT_MSG((range.upper() - range.lower()) <= std::numeric_limits::max(), + "Buffer size is too large"); + CreateBuffer(start, static_cast(end - start)); + } + }); +} + void BufferCache::Register(BufferId buffer_id) { ChangeRegister(buffer_id); } @@ -514,11 +701,16 @@ void BufferCache::ChangeRegister(BufferId buffer_id) { const u64 page_end = Common::DivCeil(device_addr_end, CACHING_PAGESIZE); for (u64 page = page_begin; page != page_end; ++page) { if constexpr (insert) { - page_table[page] = buffer_id; + page_table[page].buffer_id = buffer_id; } else { - page_table[page] = BufferId{}; + page_table[page].buffer_id = BufferId{}; } } + if constexpr (insert) { + buffer_ranges.Add(buffer.CpuAddr(), buffer.SizeBytes(), buffer_id); + } else { + buffer_ranges.Subtract(buffer.CpuAddr(), buffer.SizeBytes()); + } } void BufferCache::SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size, @@ -697,6 +889,138 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr, return true; } +void BufferCache::SynchronizeBuffersInRange(VAddr device_addr, u64 size) { + if (device_addr == 0) { + return; + } + VAddr device_addr_end = device_addr + size; + ForEachBufferInRange(device_addr, size, [&](BufferId buffer_id, Buffer& buffer) { + RENDERER_TRACE; + VAddr start = std::max(buffer.CpuAddr(), device_addr); + VAddr end = std::min(buffer.CpuAddr() + buffer.SizeBytes(), device_addr_end); + u32 size = static_cast(end - start); + SynchronizeBuffer(buffer, start, size, false); + }); +} + +void BufferCache::MemoryBarrier() { + // Vulkan doesn't know which buffer we access in a shader if we use + // BufferDeviceAddress. We need a full memory barrier. + // For now, we only read memory using BDA. If we want to write to it, + // we might need to change this. + scheduler.EndRendering(); + const auto cmdbuf = scheduler.CommandBuffer(); + vk::MemoryBarrier2 barrier = { + .srcStageMask = vk::PipelineStageFlagBits2::eTransfer, + .srcAccessMask = vk::AccessFlagBits2::eMemoryWrite, + .dstStageMask = vk::PipelineStageFlagBits2::eAllCommands, + .dstAccessMask = vk::AccessFlagBits2::eMemoryRead, + }; + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .memoryBarrierCount = 1, + .pMemoryBarriers = &barrier, + }); +} + +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(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, + .dstOffset = buffer.Offset(address), + .size = num_bytes, + }; + vk::Buffer src_buffer = staging_buffer.Handle(); + if (num_bytes < StagingBufferSize) { + const auto [staging, offset] = staging_buffer.Map(num_bytes); + std::memcpy(staging, value, num_bytes); + copy.srcOffset = offset; + staging_buffer.Commit(); + } else { + // For large one time transfers use a temporary host buffer. + // RenderDoc can lag quite a bit if the stream buffer is too large. + Buffer temp_buffer{ + instance, scheduler, MemoryUsage::Upload, 0, vk::BufferUsageFlagBits::eTransferSrc, + num_bytes}; + src_buffer = temp_buffer.Handle(); + u8* const staging = temp_buffer.mapped_data.data(); + std::memcpy(staging, value, num_bytes); + scheduler.DeferOperation([buffer = std::move(temp_buffer)]() mutable {}); + } + 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 | vk::AccessFlagBits2::eMemoryWrite, + .buffer = buffer.Handle(), + .offset = buffer.Offset(address), + .size = num_bytes, + }; + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &pre_barrier, + }); + cmdbuf.copyBuffer(src_buffer, buffer.Handle(), copy); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &post_barrier, + }); +} + void BufferCache::DeleteBuffer(BufferId buffer_id) { Buffer& buffer = slot_buffers[buffer_id]; Unregister(buffer_id); diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 71a6bed2a..2d6551a7f 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -38,14 +38,22 @@ class TextureCache; class BufferCache { public: - static constexpr u32 CACHING_PAGEBITS = 12; + static constexpr u32 CACHING_PAGEBITS = 14; static constexpr u64 CACHING_PAGESIZE = u64{1} << CACHING_PAGEBITS; - static constexpr u64 DEVICE_PAGESIZE = 4_KB; + static constexpr u64 DEVICE_PAGESIZE = 16_KB; + static constexpr u64 CACHING_NUMPAGES = u64{1} << (40 - CACHING_PAGEBITS); + + static constexpr u64 BDA_PAGETABLE_SIZE = CACHING_NUMPAGES * sizeof(vk::DeviceAddress); + static constexpr u64 FAULT_BUFFER_SIZE = CACHING_NUMPAGES / 8; // Bit per page + + struct PageData { + BufferId buffer_id{}; + }; struct Traits { - using Entry = BufferId; + using Entry = PageData; static constexpr size_t AddressSpaceBits = 40; - static constexpr size_t FirstLevelBits = 14; + static constexpr size_t FirstLevelBits = 16; static constexpr size_t PageBits = CACHING_PAGEBITS; }; using PageTable = MultiLevelPageTable; @@ -59,8 +67,8 @@ public: public: explicit BufferCache(const Vulkan::Instance& instance, Vulkan::Scheduler& scheduler, - AmdGpu::Liverpool* liverpool, TextureCache& texture_cache, - PageManager& tracker); + Vulkan::Rasterizer& rasterizer_, AmdGpu::Liverpool* liverpool, + TextureCache& texture_cache, PageManager& tracker); ~BufferCache(); /// Returns a pointer to GDS device local buffer. @@ -73,13 +81,23 @@ public: return stream_buffer; } + /// Retrieves the device local DBA page table buffer. + [[nodiscard]] Buffer* GetBdaPageTableBuffer() noexcept { + return &bda_pagetable_buffer; + } + + /// Retrieves the fault buffer. + [[nodiscard]] Buffer* GetFaultBuffer() noexcept { + return &fault_buffer; + } + /// Retrieves the buffer with the specified id. [[nodiscard]] Buffer& GetBuffer(BufferId id) { return slot_buffers[id]; } /// Invalidates any buffer in the logical page range. - void InvalidateMemory(VAddr device_addr, u64 size); + void InvalidateMemory(VAddr device_addr, u64 size, bool unmap); /// Binds host vertex buffers for the current draw. void BindVertexBuffers(const Vulkan::GraphicsPipeline& pipeline); @@ -87,9 +105,12 @@ public: /// Bind host index buffer for the current draw. void BindIndexBuffer(u32 index_offset); - /// Writes a value to GPU buffer. + /// 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); + /// Writes a value to GPU buffer. (uses staging buffer to temporarily store the data) + void WriteData(VAddr address, const void* value, u32 num_bytes, bool is_gds); + /// Obtains a buffer for the specified region. [[nodiscard]] std::pair ObtainBuffer(VAddr gpu_addr, u32 size, bool is_written, bool is_texel_buffer = false, @@ -108,24 +129,29 @@ public: /// Return true when a CPU region is modified from the GPU [[nodiscard]] bool IsRegionGpuModified(VAddr addr, size_t size); - [[nodiscard]] BufferId FindBuffer(VAddr device_addr, u32 size); + /// Return buffer id for the specified region + BufferId FindBuffer(VAddr device_addr, u32 size); + + /// Processes the fault buffer. + void ProcessFaultBuffer(); + + /// Synchronizes all buffers in the specified range. + void SynchronizeBuffersInRange(VAddr device_addr, u64 size); + + /// Synchronizes all buffers neede for DMA. + void SynchronizeDmaBuffers(); + + /// Record memory barrier. Used for buffers when accessed via BDA. + void MemoryBarrier(); private: template void ForEachBufferInRange(VAddr device_addr, u64 size, Func&& func) { - const u64 page_end = Common::DivCeil(device_addr + size, CACHING_PAGESIZE); - for (u64 page = device_addr >> CACHING_PAGEBITS; page < page_end;) { - const BufferId buffer_id = page_table[page]; - if (!buffer_id) { - ++page; - continue; - } - Buffer& buffer = slot_buffers[buffer_id]; - func(buffer_id, buffer); - - const VAddr end_addr = buffer.CpuAddr() + buffer.SizeBytes(); - page = Common::DivCeil(end_addr, CACHING_PAGESIZE); - } + buffer_ranges.ForEachInRange(device_addr, size, + [&](u64 page_start, u64 page_end, BufferId id) { + Buffer& buffer = slot_buffers[id]; + func(id, buffer); + }); } void DownloadBufferMemory(Buffer& buffer, VAddr device_addr, u64 size); @@ -134,7 +160,7 @@ private: void JoinOverlap(BufferId new_buffer_id, BufferId overlap_id, bool accumulate_stream_score); - [[nodiscard]] BufferId CreateBuffer(VAddr device_addr, u32 wanted_size); + BufferId CreateBuffer(VAddr device_addr, u32 wanted_size); void Register(BufferId buffer_id); @@ -147,21 +173,33 @@ 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 DeleteBuffer(BufferId buffer_id); const Vulkan::Instance& instance; Vulkan::Scheduler& scheduler; + Vulkan::Rasterizer& rasterizer; AmdGpu::Liverpool* liverpool; TextureCache& texture_cache; PageManager& tracker; StreamBuffer staging_buffer; StreamBuffer stream_buffer; + StreamBuffer download_buffer; Buffer gds_buffer; - std::shared_mutex mutex; + Buffer bda_pagetable_buffer; + Buffer fault_buffer; + std::shared_mutex slot_buffers_mutex; Common::SlotVector slot_buffers; RangeSet gpu_modified_ranges; + SplitRangeMap buffer_ranges; MemoryTracker memory_tracker; PageTable page_table; + vk::UniqueDescriptorSetLayout fault_process_desc_layout; + vk::UniquePipeline fault_process_pipeline; + vk::UniquePipelineLayout fault_process_pipeline_layout; }; } // namespace VideoCore diff --git a/src/video_core/buffer_cache/memory_tracker_base.h b/src/video_core/buffer_cache/memory_tracker_base.h index d9166b11c..c60aa9c80 100644 --- a/src/video_core/buffer_cache/memory_tracker_base.h +++ b/src/video_core/buffer_cache/memory_tracker_base.h @@ -7,6 +7,7 @@ #include #include #include +#include "common/debug.h" #include "common/types.h" #include "video_core/buffer_cache/word_manager.h" @@ -19,11 +20,11 @@ public: static constexpr size_t MANAGER_POOL_SIZE = 32; public: - explicit MemoryTracker(PageManager* tracker_) : tracker{tracker_} {} + explicit MemoryTracker(PageManager& tracker_) : tracker{&tracker_} {} ~MemoryTracker() = default; /// Returns true if a region has been modified from the CPU - [[nodiscard]] bool IsRegionCpuModified(VAddr query_cpu_addr, u64 query_size) noexcept { + bool IsRegionCpuModified(VAddr query_cpu_addr, u64 query_size) noexcept { return IteratePages( query_cpu_addr, query_size, [](RegionManager* manager, u64 offset, size_t size) { return manager->template IsRegionModified(offset, size); @@ -31,7 +32,7 @@ public: } /// Returns true if a region has been modified from the GPU - [[nodiscard]] bool IsRegionGpuModified(VAddr query_cpu_addr, u64 query_size) noexcept { + bool IsRegionGpuModified(VAddr query_cpu_addr, u64 query_size) noexcept { return IteratePages( query_cpu_addr, query_size, [](RegionManager* manager, u64 offset, size_t size) { return manager->template IsRegionModified(offset, size); @@ -57,8 +58,7 @@ public: } /// Call 'func' for each CPU modified range and unmark those pages as CPU modified - template - void ForEachUploadRange(VAddr query_cpu_range, u64 query_size, Func&& func) { + void ForEachUploadRange(VAddr query_cpu_range, u64 query_size, auto&& func) { IteratePages(query_cpu_range, query_size, [&func](RegionManager* manager, u64 offset, size_t size) { manager->template ForEachModifiedRange( @@ -67,17 +67,12 @@ public: } /// Call 'func' for each GPU modified range and unmark those pages as GPU modified - template - void ForEachDownloadRange(VAddr query_cpu_range, u64 query_size, Func&& func) { + template + void ForEachDownloadRange(VAddr query_cpu_range, u64 query_size, auto&& func) { IteratePages(query_cpu_range, query_size, [&func](RegionManager* manager, u64 offset, size_t size) { - if constexpr (clear) { - manager->template ForEachModifiedRange( - manager->GetCpuAddr() + offset, size, func); - } else { - manager->template ForEachModifiedRange( - manager->GetCpuAddr() + offset, size, func); - } + manager->template ForEachModifiedRange( + manager->GetCpuAddr() + offset, size, func); }); } @@ -91,6 +86,7 @@ private: */ template bool IteratePages(VAddr cpu_address, size_t size, Func&& func) { + RENDERER_TRACE; using FuncReturn = typename std::invoke_result::type; static constexpr bool BOOL_BREAK = std::is_same_v; std::size_t remaining_size{size}; diff --git a/src/video_core/buffer_cache/range_set.h b/src/video_core/buffer_cache/range_set.h index 2abf6e524..5c8e78c7c 100644 --- a/src/video_core/buffer_cache/range_set.h +++ b/src/video_core/buffer_cache/range_set.h @@ -3,7 +3,10 @@ #pragma once +#include #include +#include +#include #include #include #include @@ -38,6 +41,22 @@ struct RangeSet { m_ranges_set.subtract(interval); } + void Clear() { + m_ranges_set.clear(); + } + + bool Contains(VAddr base_address, size_t size) const { + const VAddr end_address = base_address + size; + IntervalType interval{base_address, end_address}; + return boost::icl::contains(m_ranges_set, interval); + } + + bool Intersects(VAddr base_address, size_t size) const { + const VAddr end_address = base_address + size; + IntervalType interval{base_address, end_address}; + return boost::icl::intersects(m_ranges_set, interval); + } + template void ForEach(Func&& func) const { if (m_ranges_set.empty()) { @@ -77,14 +96,29 @@ struct RangeSet { } } + template + void ForEachNotInRange(VAddr base_addr, size_t size, Func&& func) const { + const VAddr end_addr = base_addr + size; + ForEachInRange(base_addr, size, [&](VAddr range_addr, VAddr range_end) { + if (size_t gap_size = range_addr - base_addr; gap_size != 0) { + func(base_addr, gap_size); + } + base_addr = range_end; + }); + if (base_addr != end_addr) { + func(base_addr, end_addr - base_addr); + } + } + IntervalSet m_ranges_set; }; +template class RangeMap { public: using IntervalMap = - boost::icl::interval_map; using IntervalType = typename IntervalMap::interval_type; @@ -99,7 +133,7 @@ public: RangeMap(RangeMap&& other); RangeMap& operator=(RangeMap&& other); - void Add(VAddr base_address, size_t size, u64 value) { + void Add(VAddr base_address, size_t size, const T& value) { const VAddr end_address = base_address + size; IntervalType interval{base_address, end_address}; m_ranges_map.add({interval, value}); @@ -111,6 +145,35 @@ public: m_ranges_map -= interval; } + void Clear() { + m_ranges_map.clear(); + } + + bool Contains(VAddr base_address, size_t size) const { + const VAddr end_address = base_address + size; + IntervalType interval{base_address, end_address}; + return boost::icl::contains(m_ranges_map, interval); + } + + bool Intersects(VAddr base_address, size_t size) const { + const VAddr end_address = base_address + size; + IntervalType interval{base_address, end_address}; + return boost::icl::intersects(m_ranges_map, interval); + } + + template + void ForEach(Func&& func) const { + if (m_ranges_map.empty()) { + return; + } + + for (const auto& [interval, value] : m_ranges_map) { + const VAddr inter_addr_end = interval.upper(); + const VAddr inter_addr = interval.lower(); + func(inter_addr, inter_addr_end, value); + } + } + template void ForEachInRange(VAddr base_addr, size_t size, Func&& func) const { if (m_ranges_map.empty()) { @@ -140,7 +203,111 @@ public: template void ForEachNotInRange(VAddr base_addr, size_t size, Func&& func) const { const VAddr end_addr = base_addr + size; - ForEachInRange(base_addr, size, [&](VAddr range_addr, VAddr range_end, u64) { + ForEachInRange(base_addr, size, [&](VAddr range_addr, VAddr range_end, const T&) { + if (size_t gap_size = range_addr - base_addr; gap_size != 0) { + func(base_addr, gap_size); + } + base_addr = range_end; + }); + if (base_addr != end_addr) { + func(base_addr, end_addr - base_addr); + } + } + +private: + IntervalMap m_ranges_map; +}; + +template +class SplitRangeMap { +public: + using IntervalMap = boost::icl::split_interval_map< + VAddr, T, boost::icl::total_absorber, std::less, boost::icl::inplace_identity, + boost::icl::inter_section, ICL_INTERVAL_INSTANCE(ICL_INTERVAL_DEFAULT, VAddr, std::less), + RangeSetsAllocator>; + using IntervalType = typename IntervalMap::interval_type; + +public: + SplitRangeMap() = default; + ~SplitRangeMap() = default; + + SplitRangeMap(SplitRangeMap const&) = delete; + SplitRangeMap& operator=(SplitRangeMap const&) = delete; + + SplitRangeMap(SplitRangeMap&& other); + SplitRangeMap& operator=(SplitRangeMap&& other); + + void Add(VAddr base_address, size_t size, const T& value) { + const VAddr end_address = base_address + size; + IntervalType interval{base_address, end_address}; + m_ranges_map.add({interval, value}); + } + + void Subtract(VAddr base_address, size_t size) { + const VAddr end_address = base_address + size; + IntervalType interval{base_address, end_address}; + m_ranges_map -= interval; + } + + void Clear() { + m_ranges_map.clear(); + } + + bool Contains(VAddr base_address, size_t size) const { + const VAddr end_address = base_address + size; + IntervalType interval{base_address, end_address}; + return boost::icl::contains(m_ranges_map, interval); + } + + bool Intersects(VAddr base_address, size_t size) const { + const VAddr end_address = base_address + size; + IntervalType interval{base_address, end_address}; + return boost::icl::intersects(m_ranges_map, interval); + } + + template + void ForEach(Func&& func) const { + if (m_ranges_map.empty()) { + return; + } + + for (const auto& [interval, value] : m_ranges_map) { + const VAddr inter_addr_end = interval.upper(); + const VAddr inter_addr = interval.lower(); + func(inter_addr, inter_addr_end, value); + } + } + + template + void ForEachInRange(VAddr base_addr, size_t size, Func&& func) const { + if (m_ranges_map.empty()) { + return; + } + const VAddr start_address = base_addr; + const VAddr end_address = start_address + size; + const IntervalType search_interval{start_address, end_address}; + auto it = m_ranges_map.lower_bound(search_interval); + if (it == m_ranges_map.end()) { + return; + } + auto end_it = m_ranges_map.upper_bound(search_interval); + for (; it != end_it; it++) { + VAddr inter_addr_end = it->first.upper(); + VAddr inter_addr = it->first.lower(); + if (inter_addr_end > end_address) { + inter_addr_end = end_address; + } + if (inter_addr < start_address) { + inter_addr = start_address; + } + func(inter_addr, inter_addr_end, it->second); + } + } + + template + void ForEachNotInRange(VAddr base_addr, size_t size, Func&& func) const { + const VAddr end_addr = base_addr + size; + ForEachInRange(base_addr, size, [&](VAddr range_addr, VAddr range_end, const T&) { if (size_t gap_size = range_addr - base_addr; gap_size != 0) { func(base_addr, gap_size); } diff --git a/src/video_core/buffer_cache/word_manager.h b/src/video_core/buffer_cache/word_manager.h index 5ad724f96..51a912c62 100644 --- a/src/video_core/buffer_cache/word_manager.h +++ b/src/video_core/buffer_cache/word_manager.h @@ -10,8 +10,10 @@ #ifdef __linux__ #include "common/adaptive_mutex.h" -#endif +#else #include "common/spin_lock.h" +#endif +#include "common/debug.h" #include "common/types.h" #include "video_core/page_manager.h" @@ -56,7 +58,7 @@ public: return cpu_addr; } - static u64 ExtractBits(u64 word, size_t page_start, size_t page_end) { + static constexpr u64 ExtractBits(u64 word, size_t page_start, size_t page_end) { constexpr size_t number_bits = sizeof(u64) * 8; const size_t limit_page_end = number_bits - std::min(page_end, number_bits); u64 bits = (word >> page_start) << page_start; @@ -64,7 +66,7 @@ public: return bits; } - static std::pair GetWordPage(VAddr address) { + static constexpr std::pair GetWordPage(VAddr address) { const size_t converted_address = static_cast(address); const size_t word_number = converted_address / BYTES_PER_WORD; const size_t amount_pages = converted_address % BYTES_PER_WORD; @@ -73,6 +75,7 @@ public: template void IterateWords(size_t offset, size_t size, Func&& func) const { + RENDERER_TRACE; using FuncReturn = std::invoke_result_t; static constexpr bool BOOL_BREAK = std::is_same_v; const size_t start = static_cast(std::max(static_cast(offset), 0LL)); @@ -104,13 +107,13 @@ public: } } - template - void IteratePages(u64 mask, Func&& func) const { + void IteratePages(u64 mask, auto&& func) const { + RENDERER_TRACE; size_t offset = 0; while (mask != 0) { const size_t empty_bits = std::countr_zero(mask); offset += empty_bits; - mask = mask >> empty_bits; + mask >>= empty_bits; const size_t continuous_bits = std::countr_one(mask); func(offset, continuous_bits); @@ -155,8 +158,9 @@ public: * @param size Size in bytes of the CPU range to loop over * @param func Function to call for each turned off region */ - template - void ForEachModifiedRange(VAddr query_cpu_range, s64 size, Func&& func) { + template + void ForEachModifiedRange(VAddr query_cpu_range, s64 size, auto&& func) { + RENDERER_TRACE; std::scoped_lock lk{lock}; static_assert(type != Type::Untracked); @@ -170,6 +174,7 @@ public: (pending_pointer - pending_offset) * BYTES_PER_PAGE); }; IterateWords(offset, size, [&](size_t index, u64 mask) { + RENDERER_TRACE; if constexpr (type == Type::GPU) { mask &= ~untracked[index]; } @@ -177,14 +182,13 @@ public: if constexpr (clear) { if constexpr (type == Type::CPU) { UpdateProtection(index, untracked[index], mask); - } - state_words[index] &= ~mask; - if constexpr (type == Type::CPU) { untracked[index] &= ~mask; } + state_words[index] &= ~mask; } const size_t base_offset = index * PAGES_PER_WORD; IteratePages(word, [&](size_t pages_offset, size_t pages_size) { + RENDERER_TRACE; const auto reset = [&]() { pending_offset = base_offset + pages_offset; pending_pointer = base_offset + pages_offset + pages_size; @@ -245,11 +249,13 @@ private: */ template void UpdateProtection(u64 word_index, u64 current_bits, u64 new_bits) const { + RENDERER_TRACE; + constexpr s32 delta = add_to_tracker ? 1 : -1; u64 changed_bits = (add_to_tracker ? current_bits : ~current_bits) & new_bits; VAddr addr = cpu_addr + word_index * BYTES_PER_WORD; IteratePages(changed_bits, [&](size_t offset, size_t size) { - tracker->UpdatePagesCachedCount(addr + offset * BYTES_PER_PAGE, size * BYTES_PER_PAGE, - add_to_tracker ? 1 : -1); + tracker->UpdatePageWatchers(addr + offset * BYTES_PER_PAGE, + size * BYTES_PER_PAGE); }); } diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index 3001bf773..d52afe738 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -11,6 +11,7 @@ set(SHADER_FILES detilers/micro_32bpp.comp detilers/micro_64bpp.comp detilers/micro_8bpp.comp + fault_buffer_process.comp fs_tri.vert fsr.comp post_process.frag diff --git a/src/video_core/host_shaders/fault_buffer_process.comp b/src/video_core/host_shaders/fault_buffer_process.comp new file mode 100644 index 000000000..a712cf441 --- /dev/null +++ b/src/video_core/host_shaders/fault_buffer_process.comp @@ -0,0 +1,42 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#version 450 +#extension GL_ARB_gpu_shader_int64 : enable + +layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in; + +layout(std430, binding = 0) buffer input_buf { + uint fault_buffer[]; +}; + +layout(std430, binding = 1) buffer output_buf { + uint64_t download_buffer[]; +}; + +// Overlap for 32 bit atomics +layout(std430, binding = 1) buffer output_buf32 { + uint download_buffer32[]; +}; + +layout(constant_id = 0) const uint CACHING_PAGEBITS = 0; + +void main() { + uint id = gl_GlobalInvocationID.x; + uint word = fault_buffer[id]; + if (word == 0u) { + return; + } + // 1 page per bit + uint base_bit = id * 32u; + while (word != 0u) { + uint bit = findLSB(word); + word &= word - 1; + uint page = base_bit + bit; + uint store_index = atomicAdd(download_buffer32[0], 1u) + 1u; + // It is very unlikely, but should we check for overflow? + if (store_index < 1024u) { // only support 1024 page faults + download_buffer[store_index] = uint64_t(page) << CACHING_PAGEBITS; + } + } +} diff --git a/src/video_core/page_manager.cpp b/src/video_core/page_manager.cpp index 47ed9e543..39c03e7da 100644 --- a/src/video_core/page_manager.cpp +++ b/src/video_core/page_manager.cpp @@ -1,11 +1,9 @@ // SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later -#include -#include -#include "common/alignment.h" +#include #include "common/assert.h" -#include "common/error.h" +#include "common/debug.h" #include "common/signal_context.h" #include "core/memory.h" #include "core/signals.h" @@ -15,23 +13,60 @@ #ifndef _WIN64 #include #ifdef ENABLE_USERFAULTFD +#include #include #include #include #include +#include "common/error.h" #endif #else #include #endif +#ifdef __linux__ +#include "common/adaptive_mutex.h" +#else +#include "common/spin_lock.h" +#endif + namespace VideoCore { -constexpr size_t PAGESIZE = 4_KB; -constexpr size_t PAGEBITS = 12; +constexpr size_t PAGE_SIZE = 4_KB; +constexpr size_t PAGE_BITS = 12; -#ifdef ENABLE_USERFAULTFD struct PageManager::Impl { - Impl(Vulkan::Rasterizer* rasterizer_) : rasterizer{rasterizer_} { + struct PageState { + u8 num_watchers{}; + + Core::MemoryPermission Perm() const noexcept { + return num_watchers == 0 ? Core::MemoryPermission::ReadWrite + : Core::MemoryPermission::Read; + } + + template + u8 AddDelta() { + if constexpr (delta == 1) { + return ++num_watchers; + } else { + ASSERT_MSG(num_watchers > 0, "Not enough watchers"); + return --num_watchers; + } + } + }; + + struct UpdateProtectRange { + VAddr addr; + u64 size; + Core::MemoryPermission perms; + }; + + static constexpr size_t ADDRESS_BITS = 40; + static constexpr size_t NUM_ADDRESS_PAGES = 1ULL << (40 - PAGE_BITS); + inline static Vulkan::Rasterizer* rasterizer; +#ifdef ENABLE_USERFAULTFD + Impl(Vulkan::Rasterizer* rasterizer_) { + rasterizer = rasterizer_; uffd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK | UFFD_USER_MODE_ONLY); ASSERT_MSG(uffd != -1, "{}", Common::GetLastErrorMsg()); @@ -63,7 +98,8 @@ struct PageManager::Impl { ASSERT_MSG(ret != -1, "Uffdio unregister failed"); } - void Protect(VAddr address, size_t size, bool allow_write) { + void Protect(VAddr address, size_t size, Core::MemoryPermission perms) { + bool allow_write = True(perms & Core::MemoryPermission::Write); uffdio_writeprotect wp; wp.range.start = address; wp.range.len = size; @@ -118,12 +154,9 @@ struct PageManager::Impl { } } - Vulkan::Rasterizer* rasterizer; std::jthread ufd_thread; int uffd; -}; #else -struct PageManager::Impl { Impl(Vulkan::Rasterizer* rasterizer_) { rasterizer = rasterizer_; @@ -141,12 +174,11 @@ struct PageManager::Impl { // No-op } - void Protect(VAddr address, size_t size, bool allow_write) { + void Protect(VAddr address, size_t size, Core::MemoryPermission perms) { + RENDERER_TRACE; auto* memory = Core::Memory::Instance(); auto& impl = memory->GetAddressSpace(); - impl.Protect(address, size, - allow_write ? Core::MemoryPermission::ReadWrite - : Core::MemoryPermission::Read); + impl.Protect(address, size, perms); } static bool GuestFaultSignalHandler(void* context, void* fault_address) { @@ -157,23 +189,82 @@ struct PageManager::Impl { return false; } - inline static Vulkan::Rasterizer* rasterizer; -}; #endif + template + void UpdatePageWatchers(VAddr addr, u64 size) { + RENDERER_TRACE; + boost::container::small_vector update_ranges; + { + std::scoped_lock lk(lock); + + size_t page = addr >> PAGE_BITS; + auto perms = cached_pages[page].Perm(); + u64 range_begin = 0; + u64 range_bytes = 0; + + const auto release_pending = [&] { + if (range_bytes > 0) { + RENDERER_TRACE; + // Add pending (un)protect action + update_ranges.push_back({range_begin << PAGE_BITS, range_bytes, perms}); + range_bytes = 0; + } + }; + + // Iterate requested pages + const u64 page_end = Common::DivCeil(addr + size, PAGE_SIZE); + const u64 aligned_addr = page << PAGE_BITS; + const u64 aligned_end = page_end << PAGE_BITS; + ASSERT_MSG(rasterizer->IsMapped(aligned_addr, aligned_end - aligned_addr), + "Attempted to track non-GPU memory at address {:#x}, size {:#x}.", + aligned_addr, aligned_end - aligned_addr); + + for (; page != page_end; ++page) { + PageState& state = cached_pages[page]; + + // Apply the change to the page state + const u8 new_count = state.AddDelta(); + + // If the protection changed add pending (un)protect action + if (auto new_perms = state.Perm(); new_perms != perms) [[unlikely]] { + release_pending(); + perms = new_perms; + } + + // If the page must be (un)protected, add it to the pending range + if ((new_count == 0 && delta < 0) || (new_count == 1 && delta > 0)) { + if (range_bytes == 0) { + range_begin = page; + } + range_bytes += PAGE_SIZE; + } else { + release_pending(); + } + } + + // Add pending (un)protect action + release_pending(); + } + + // Flush deferred protects + for (const auto& range : update_ranges) { + Protect(range.addr, range.size, range.perms); + } + } + + std::array cached_pages{}; +#ifdef __linux__ + Common::AdaptiveMutex lock; +#else + Common::SpinLock lock; +#endif +}; PageManager::PageManager(Vulkan::Rasterizer* rasterizer_) - : impl{std::make_unique(rasterizer_)}, rasterizer{rasterizer_} {} + : impl{std::make_unique(rasterizer_)} {} PageManager::~PageManager() = default; -VAddr PageManager::GetPageAddr(VAddr addr) { - return Common::AlignDown(addr, PAGESIZE); -} - -VAddr PageManager::GetNextPageAddr(VAddr addr) { - return Common::AlignUp(addr + 1, PAGESIZE); -} - void PageManager::OnGpuMap(VAddr address, size_t size) { impl->OnMap(address, size); } @@ -182,41 +273,12 @@ void PageManager::OnGpuUnmap(VAddr address, size_t size) { impl->OnUnmap(address, size); } -void PageManager::UpdatePagesCachedCount(VAddr addr, u64 size, s32 delta) { - static constexpr u64 PageShift = 12; - - std::scoped_lock lk{lock}; - const u64 num_pages = ((addr + size - 1) >> PageShift) - (addr >> PageShift) + 1; - const u64 page_start = addr >> PageShift; - const u64 page_end = page_start + num_pages; - - const auto pages_interval = - decltype(cached_pages)::interval_type::right_open(page_start, page_end); - if (delta > 0) { - cached_pages.add({pages_interval, delta}); - } - - const auto& range = cached_pages.equal_range(pages_interval); - for (const auto& [range, count] : boost::make_iterator_range(range)) { - const auto interval = range & pages_interval; - const VAddr interval_start_addr = boost::icl::first(interval) << PageShift; - const VAddr interval_end_addr = boost::icl::last_next(interval) << PageShift; - const u32 interval_size = interval_end_addr - interval_start_addr; - ASSERT_MSG(rasterizer->IsMapped(interval_start_addr, interval_size), - "Attempted to track non-GPU memory at address {:#x}, size {:#x}.", - interval_start_addr, interval_size); - if (delta > 0 && count == delta) { - impl->Protect(interval_start_addr, interval_size, false); - } else if (delta < 0 && count == -delta) { - impl->Protect(interval_start_addr, interval_size, true); - } else { - ASSERT(count >= 0); - } - } - - if (delta < 0) { - cached_pages.add({pages_interval, delta}); - } +template +void PageManager::UpdatePageWatchers(VAddr addr, u64 size) const { + impl->UpdatePageWatchers(addr, size); } +template void PageManager::UpdatePageWatchers<1>(VAddr addr, u64 size) const; +template void PageManager::UpdatePageWatchers<-1>(VAddr addr, u64 size) const; + } // namespace VideoCore diff --git a/src/video_core/page_manager.h b/src/video_core/page_manager.h index f6bae9641..98dd099af 100644 --- a/src/video_core/page_manager.h +++ b/src/video_core/page_manager.h @@ -4,11 +4,7 @@ #pragma once #include -#include -#ifdef __linux__ -#include "common/adaptive_mutex.h" -#endif -#include "common/spin_lock.h" +#include "common/alignment.h" #include "common/types.h" namespace Vulkan { @@ -18,6 +14,9 @@ class Rasterizer; namespace VideoCore { class PageManager { + static constexpr size_t PAGE_BITS = 12; + static constexpr size_t PAGE_SIZE = 1ULL << PAGE_BITS; + public: explicit PageManager(Vulkan::Rasterizer* rasterizer); ~PageManager(); @@ -28,22 +27,23 @@ public: /// Unregister a range of gpu memory that was unmapped. void OnGpuUnmap(VAddr address, size_t size); - /// Increase/decrease the number of surface in pages touching the specified region - void UpdatePagesCachedCount(VAddr addr, u64 size, s32 delta); + /// Updates watches in the pages touching the specified region. + template + void UpdatePageWatchers(VAddr addr, u64 size) const; - static VAddr GetPageAddr(VAddr addr); - static VAddr GetNextPageAddr(VAddr addr); + /// Returns page aligned address. + static constexpr VAddr GetPageAddr(VAddr addr) { + return Common::AlignDown(addr, PAGE_SIZE); + } + + /// Returns address of the next page. + static constexpr VAddr GetNextPageAddr(VAddr addr) { + return Common::AlignUp(addr + 1, PAGE_SIZE); + } private: struct Impl; std::unique_ptr impl; - Vulkan::Rasterizer* rasterizer; - boost::icl::interval_map cached_pages; -#ifdef PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP - Common::AdaptiveMutex lock; -#else - Common::SpinLock lock; -#endif }; } // namespace VideoCore diff --git a/src/video_core/renderdoc.cpp b/src/video_core/renderdoc.cpp index b082fd1ca..4cf2ddd53 100644 --- a/src/video_core/renderdoc.cpp +++ b/src/video_core/renderdoc.cpp @@ -121,6 +121,7 @@ void SetOutputDir(const std::filesystem::path& path, const std::string& prefix) if (!rdoc_api) { return; } + LOG_WARNING(Common, "RenderDoc capture path: {}", (path / prefix).string()); rdoc_api->SetCaptureFilePathTemplate(fmt::UTF((path / prefix).u8string()).data.data()); } diff --git a/src/video_core/renderer_vulkan/vk_instance.cpp b/src/video_core/renderer_vulkan/vk_instance.cpp index e31b95844..9584329f0 100644 --- a/src/video_core/renderer_vulkan/vk_instance.cpp +++ b/src/video_core/renderer_vulkan/vk_instance.cpp @@ -147,6 +147,7 @@ Instance::Instance(Frontend::WindowSDL& window, s32 physical_device_index, available_extensions = GetSupportedExtensions(physical_device); format_properties = GetFormatProperties(physical_device); properties = physical_device.getProperties(); + memory_properties = physical_device.getMemoryProperties(); CollectDeviceParameters(); ASSERT_MSG(properties.apiVersion >= TargetVulkanApiVersion, "Vulkan {}.{} is required, but only {}.{} is supported by device!", @@ -375,6 +376,7 @@ bool Instance::CreateDevice() { .separateDepthStencilLayouts = vk12_features.separateDepthStencilLayouts, .hostQueryReset = vk12_features.hostQueryReset, .timelineSemaphore = vk12_features.timelineSemaphore, + .bufferDeviceAddress = vk12_features.bufferDeviceAddress, }, vk::PhysicalDeviceVulkan13Features{ .robustImageAccess = vk13_features.robustImageAccess, @@ -505,6 +507,7 @@ void Instance::CreateAllocator() { }; const VmaAllocatorCreateInfo allocator_info = { + .flags = VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT, .physicalDevice = physical_device, .device = *device, .pVulkanFunctions = &functions, diff --git a/src/video_core/renderer_vulkan/vk_instance.h b/src/video_core/renderer_vulkan/vk_instance.h index 573473869..30848e8b7 100644 --- a/src/video_core/renderer_vulkan/vk_instance.h +++ b/src/video_core/renderer_vulkan/vk_instance.h @@ -286,6 +286,11 @@ public: return vk12_props; } + /// Returns the memory properties of the physical device. + const vk::PhysicalDeviceMemoryProperties& GetMemoryProperties() const noexcept { + return memory_properties; + } + /// Returns true if shaders can declare the ClipDistance attribute bool IsShaderClipDistanceSupported() const { return features.shaderClipDistance; @@ -335,6 +340,7 @@ private: vk::PhysicalDevice physical_device; vk::UniqueDevice device; vk::PhysicalDeviceProperties properties; + vk::PhysicalDeviceMemoryProperties memory_properties; vk::PhysicalDeviceVulkan11Properties vk11_props; vk::PhysicalDeviceVulkan12Properties vk12_props; vk::PhysicalDevicePushDescriptorPropertiesKHR push_descriptor_props; diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index e7b42a34b..4bdb08bf2 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -36,7 +36,7 @@ static Shader::PushData MakeUserData(const AmdGpu::Liverpool::Regs& regs) { Rasterizer::Rasterizer(const Instance& instance_, Scheduler& scheduler_, AmdGpu::Liverpool* liverpool_) : instance{instance_}, scheduler{scheduler_}, page_manager{this}, - buffer_cache{instance, scheduler, liverpool_, texture_cache, page_manager}, + buffer_cache{instance, scheduler, *this, liverpool_, texture_cache, page_manager}, texture_cache{instance, scheduler, buffer_cache, page_manager}, liverpool{liverpool_}, memory{Core::Memory::Instance()}, pipeline_cache{instance, scheduler, liverpool} { if (!Config::nullGpu()) { @@ -439,6 +439,13 @@ void Rasterizer::Finish() { scheduler.Finish(); } +void Rasterizer::ProcessFaults() { + if (fault_process_pending) { + fault_process_pending = false; + buffer_cache.ProcessFaultBuffer(); + } +} + bool Rasterizer::BindResources(const Pipeline* pipeline) { if (IsComputeMetaClear(pipeline)) { return false; @@ -449,6 +456,8 @@ bool Rasterizer::BindResources(const Pipeline* pipeline) { buffer_infos.clear(); image_infos.clear(); + bool uses_dma = false; + // Bind resource buffers and textures. Shader::Backend::Bindings binding{}; Shader::PushData push_data = MakeUserData(liverpool->regs); @@ -459,9 +468,28 @@ bool Rasterizer::BindResources(const Pipeline* pipeline) { stage->PushUd(binding, push_data); BindBuffers(*stage, binding, push_data); BindTextures(*stage, binding); + + uses_dma |= stage->dma_types != Shader::IR::Type::Void; } pipeline->BindResources(set_writes, buffer_barriers, push_data); + + if (uses_dma && !fault_process_pending) { + // We only use fault buffer for DMA right now. + { + // TODO: GPU might have written to memory (for example with EVENT_WRITE_EOP) + // we need to account for that and synchronize. + Common::RecursiveSharedLock lock{mapped_ranges_mutex}; + for (auto& range : mapped_ranges) { + buffer_cache.SynchronizeBuffersInRange(range.lower(), + range.upper() - range.lower()); + } + } + buffer_cache.MemoryBarrier(); + } + + fault_process_pending |= uses_dma; + return true; } @@ -520,12 +548,18 @@ void Rasterizer::BindBuffers(const Shader::Info& stage, Shader::Backend::Binding if (desc.buffer_type == Shader::BufferType::GdsBuffer) { const auto* gds_buf = buffer_cache.GetGdsBuffer(); buffer_infos.emplace_back(gds_buf->Handle(), 0, gds_buf->SizeBytes()); - } else if (desc.buffer_type == Shader::BufferType::ReadConstUbo) { + } else if (desc.buffer_type == Shader::BufferType::Flatbuf) { auto& vk_buffer = buffer_cache.GetStreamBuffer(); const u32 ubo_size = stage.flattened_ud_buf.size() * sizeof(u32); const u64 offset = vk_buffer.Copy(stage.flattened_ud_buf.data(), ubo_size, instance.UniformMinAlignment()); buffer_infos.emplace_back(vk_buffer.Handle(), offset, ubo_size); + } else if (desc.buffer_type == Shader::BufferType::BdaPagetable) { + const auto* bda_buffer = buffer_cache.GetBdaPageTableBuffer(); + buffer_infos.emplace_back(bda_buffer->Handle(), 0, bda_buffer->SizeBytes()); + } else if (desc.buffer_type == Shader::BufferType::FaultBuffer) { + const auto* fault_buffer = buffer_cache.GetFaultBuffer(); + buffer_infos.emplace_back(fault_buffer->Handle(), 0, fault_buffer->SizeBytes()); } else if (desc.buffer_type == Shader::BufferType::SharedMemory) { auto& lds_buffer = buffer_cache.GetStreamBuffer(); const auto& cs_program = liverpool->GetCsRegs(); @@ -925,7 +959,7 @@ bool Rasterizer::InvalidateMemory(VAddr addr, u64 size) { // Not GPU mapped memory, can skip invalidation logic entirely. return false; } - buffer_cache.InvalidateMemory(addr, size); + buffer_cache.InvalidateMemory(addr, size, false); texture_cache.InvalidateMemory(addr, size); return true; } @@ -937,24 +971,24 @@ bool Rasterizer::IsMapped(VAddr addr, u64 size) { } const auto range = decltype(mapped_ranges)::interval_type::right_open(addr, addr + size); - std::shared_lock lock{mapped_ranges_mutex}; + Common::RecursiveSharedLock lock{mapped_ranges_mutex}; return boost::icl::contains(mapped_ranges, range); } void Rasterizer::MapMemory(VAddr addr, u64 size) { { - std::unique_lock lock{mapped_ranges_mutex}; + std::scoped_lock lock{mapped_ranges_mutex}; mapped_ranges += decltype(mapped_ranges)::interval_type::right_open(addr, addr + size); } page_manager.OnGpuMap(addr, size); } void Rasterizer::UnmapMemory(VAddr addr, u64 size) { - buffer_cache.InvalidateMemory(addr, size); + buffer_cache.InvalidateMemory(addr, size, true); texture_cache.UnmapMemory(addr, size); page_manager.OnGpuUnmap(addr, size); { - std::unique_lock lock{mapped_ranges_mutex}; + std::scoped_lock lock{mapped_ranges_mutex}; mapped_ranges -= decltype(mapped_ranges)::interval_type::right_open(addr, addr + size); } } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h index 4e0ed0996..fb9ca4bbe 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.h +++ b/src/video_core/renderer_vulkan/vk_rasterizer.h @@ -4,7 +4,7 @@ #pragma once #include - +#include "common/recursive_lock.h" #include "video_core/buffer_cache/buffer_cache.h" #include "video_core/page_manager.h" #include "video_core/renderer_vulkan/vk_pipeline_cache.h" @@ -65,11 +65,21 @@ public: void CpSync(); u64 Flush(); void Finish(); + void ProcessFaults(); PipelineCache& GetPipelineCache() { return pipeline_cache; } + template + void ForEachMappedRangeInRange(VAddr addr, u64 size, Func&& func) { + const auto range = decltype(mapped_ranges)::interval_type::right_open(addr, addr + size); + Common::RecursiveSharedLock lock{mapped_ranges_mutex}; + for (const auto& mapped_range : (mapped_ranges & range)) { + func(mapped_range); + } + } + private: RenderState PrepareRenderState(u32 mrt_mask); void BeginRendering(const GraphicsPipeline& pipeline, RenderState& state); @@ -100,6 +110,8 @@ private: bool IsComputeMetaClear(const Pipeline* pipeline); private: + friend class VideoCore::BufferCache; + const Instance& instance; Scheduler& scheduler; VideoCore::PageManager page_manager; @@ -126,6 +138,7 @@ private: boost::container::static_vector buffer_bindings; using ImageBindingInfo = std::pair; boost::container::static_vector image_bindings; + bool fault_process_pending{false}; }; } // namespace Vulkan diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp index 8d4188a22..e75a69924 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.cpp +++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp @@ -70,6 +70,11 @@ void Scheduler::Flush(SubmitInfo& info) { SubmitExecution(info); } +void Scheduler::Flush() { + SubmitInfo info{}; + Flush(info); +} + void Scheduler::Finish() { // When finishing, we need to wait for the submission to have executed on the device. const u64 presubmit_tick = CurrentTick(); @@ -85,6 +90,15 @@ void Scheduler::Wait(u64 tick) { Flush(info); } master_semaphore.Wait(tick); + + // CAUTION: This can introduce unexpected variation in the wait time. + // We don't currently sync the GPU, and some games are very sensitive to this. + // If this becomes a problem, it can be commented out. + // Idealy we would implement proper gpu sync. + while (!pending_ops.empty() && pending_ops.front().gpu_tick <= tick) { + pending_ops.front().callback(); + pending_ops.pop(); + } } void Scheduler::AllocateWorkerCommandBuffers() { diff --git a/src/video_core/renderer_vulkan/vk_scheduler.h b/src/video_core/renderer_vulkan/vk_scheduler.h index 7709e1d41..c30fc6e0e 100644 --- a/src/video_core/renderer_vulkan/vk_scheduler.h +++ b/src/video_core/renderer_vulkan/vk_scheduler.h @@ -307,6 +307,10 @@ public: /// and increments the scheduler timeline semaphore. void Flush(SubmitInfo& info); + /// Sends the current execution context to the GPU + /// and increments the scheduler timeline semaphore. + void Flush(); + /// Sends the current execution context to the GPU and waits for it to complete. void Finish(); diff --git a/src/video_core/texture_cache/texture_cache.cpp b/src/video_core/texture_cache/texture_cache.cpp index 82f4d6413..ddb4ea799 100644 --- a/src/video_core/texture_cache/texture_cache.cpp +++ b/src/video_core/texture_cache/texture_cache.cpp @@ -222,14 +222,23 @@ std::tuple TextureCache::ResolveOverlap(const ImageInfo& imag -1, -1}; } - ImageId new_image_id{}; - if (image_info.type == tex_cache_image.info.type) { - ASSERT(image_info.resources > tex_cache_image.info.resources); - new_image_id = ExpandImage(image_info, cache_image_id); - } else { - UNREACHABLE(); + if (image_info.type == tex_cache_image.info.type && + image_info.resources > tex_cache_image.info.resources) { + // Size and resources are greater, expand the image. + return {ExpandImage(image_info, cache_image_id), -1, -1}; } - return {new_image_id, -1, -1}; + + if (image_info.tiling_mode != tex_cache_image.info.tiling_mode) { + // Size is greater but resources are not, because the tiling mode is different. + // Likely this memory address is being reused for a different image with a different + // tiling mode. + if (safe_to_delete) { + FreeImage(cache_image_id); + } + return {merged_image_id, -1, -1}; + } + + UNREACHABLE_MSG("Encountered unresolvable image overlap with equal memory address."); } // Right overlap, the image requested is a possible subresource of the image from cache. @@ -538,10 +547,16 @@ void TextureCache::RefreshImage(Image& image, Vulkan::Scheduler* custom_schedule image.mip_hashes[m] = hash; } + auto mip_pitch = static_cast(mip.pitch); + auto mip_height = static_cast(mip.height); + + auto image_extent_width = mip_pitch ? std::min(mip_pitch, width) : width; + auto image_extent_height = mip_height ? std::min(mip_height, height) : height; + image_copy.push_back({ .bufferOffset = mip.offset, - .bufferRowLength = static_cast(mip.pitch), - .bufferImageHeight = static_cast(mip.height), + .bufferRowLength = mip_pitch, + .bufferImageHeight = mip_height, .imageSubresource{ .aspectMask = image.aspect_mask & ~vk::ImageAspectFlagBits::eStencil, .mipLevel = m, @@ -549,7 +564,7 @@ void TextureCache::RefreshImage(Image& image, Vulkan::Scheduler* custom_schedule .layerCount = num_layers, }, .imageOffset = {0, 0, 0}, - .imageExtent = {width, height, depth}, + .imageExtent = {image_extent_width, image_extent_height, depth}, }); } @@ -672,7 +687,7 @@ void TextureCache::TrackImage(ImageId image_id) { // Re-track the whole image image.track_addr = image_begin; image.track_addr_end = image_end; - tracker.UpdatePagesCachedCount(image_begin, image.info.guest_size, 1); + tracker.UpdatePageWatchers<1>(image_begin, image.info.guest_size); } else { if (image_begin < image.track_addr) { TrackImageHead(image_id); @@ -695,7 +710,7 @@ void TextureCache::TrackImageHead(ImageId image_id) { ASSERT(image.track_addr != 0 && image_begin < image.track_addr); const auto size = image.track_addr - image_begin; image.track_addr = image_begin; - tracker.UpdatePagesCachedCount(image_begin, size, 1); + tracker.UpdatePageWatchers<1>(image_begin, size); } void TextureCache::TrackImageTail(ImageId image_id) { @@ -711,7 +726,7 @@ void TextureCache::TrackImageTail(ImageId image_id) { const auto addr = image.track_addr_end; const auto size = image_end - image.track_addr_end; image.track_addr_end = image_end; - tracker.UpdatePagesCachedCount(addr, size, 1); + tracker.UpdatePageWatchers<1>(addr, size); } void TextureCache::UntrackImage(ImageId image_id) { @@ -724,7 +739,7 @@ void TextureCache::UntrackImage(ImageId image_id) { image.track_addr = 0; image.track_addr_end = 0; if (size != 0) { - tracker.UpdatePagesCachedCount(addr, size, -1); + tracker.UpdatePageWatchers<-1>(addr, size); } } @@ -743,7 +758,7 @@ void TextureCache::UntrackImageHead(ImageId image_id) { // Cehck its hash later. MarkAsMaybeDirty(image_id, image); } - tracker.UpdatePagesCachedCount(image_begin, size, -1); + tracker.UpdatePageWatchers<-1>(image_begin, size); } void TextureCache::UntrackImageTail(ImageId image_id) { @@ -762,7 +777,7 @@ void TextureCache::UntrackImageTail(ImageId image_id) { // Cehck its hash later. MarkAsMaybeDirty(image_id, image); } - tracker.UpdatePagesCachedCount(addr, size, -1); + tracker.UpdatePageWatchers<-1>(addr, size); } void TextureCache::DeleteImage(ImageId image_id) {