diff --git a/.github/ISSUE_TEMPLATE/game-bug-report.yaml b/.github/ISSUE_TEMPLATE/game-bug-report.yaml index 407ee2fe3..2d984b697 100644 --- a/.github/ISSUE_TEMPLATE/game-bug-report.yaml +++ b/.github/ISSUE_TEMPLATE/game-bug-report.yaml @@ -13,7 +13,11 @@ body: **Please do not make support requests on GitHub. Our issue tracker is for tracking bugs and feature requests only. If you have a support request or are unsure about the nature of your issue please contact us on [discord](https://discord.gg/bFJxfftGW6).** - You can also check the [Game Compatibility Repository](https://github.com/shadps4-emu/shadps4-game-compatibility) for the information about the status of the game. + This repository does not provide support for modded games. You should perform and test a clean game installation before submitting an issue. + + This repository does not provide support for game patches. If you are having issues with patches please refer to [Cheats and Patches Repository](https://github.com/shadps4-emu/ps4_cheats). + + Before submitting an issue please check [Game Compatibility Repository](https://github.com/shadps4-emu/shadps4-game-compatibility) for the information about the status of the game. Please make an effort to make sure your issue isn't already reported. @@ -21,15 +25,15 @@ body: - type: checkboxes id: checklist attributes: - label: Checklist + label: Checklist (we expect you to perform these steps before opening the issue) options: - label: I have searched for a similar issue in this repository and did not find one. required: true - label: I am using an official build obtained from [releases](https://github.com/shadps4-emu/shadPS4/releases) or updated one of those builds using its in-app updater. required: true - - label: I have re-dumped the game and performed a clean install without mods. + - label: I have re-dumped the game and performed a clean install without mods and the issue is still present. required: true - - label: I have disabled all patches and cheats. + - label: I have disabled all patches and cheats and the issue is still present. required: true - label: I have all the required [system modules](https://github.com/shadps4-emu/shadps4-game-compatibility?tab=readme-ov-file#informations) installed. required: true diff --git a/src/common/config.cpp b/src/common/config.cpp index deef0fa88..246644e2d 100644 --- a/src/common/config.cpp +++ b/src/common/config.cpp @@ -67,6 +67,7 @@ static int cursorHideTimeout = 5; // 5 seconds (default) static bool separateupdatefolder = false; static bool compatibilityData = false; static bool checkCompatibilityOnStartup = false; +static std::string trophyKey; // Gui std::vector settings_install_dirs = {}; @@ -91,6 +92,14 @@ std::string emulator_language = "en"; // Language u32 m_language = 1; // english +std::string getTrophyKey() { + return trophyKey; +} + +void setTrophyKey(std::string key) { + trophyKey = key; +} + bool isNeoMode() { return isNeo; } @@ -652,6 +661,11 @@ void load(const std::filesystem::path& path) { m_language = toml::find_or(settings, "consoleLanguage", 1); } + + if (data.contains("Keys")) { + const toml::value& keys = data.at("Keys"); + trophyKey = toml::find_or(keys, "TrophyKey", ""); + } } void save(const std::filesystem::path& path) { @@ -712,6 +726,8 @@ void save(const std::filesystem::path& path) { data["Debug"]["DebugDump"] = isDebugDump; data["Debug"]["CollectShader"] = isShaderDebug; + data["Keys"]["TrophyKey"] = trophyKey; + std::vector install_dirs; for (const auto& dirString : settings_install_dirs) { install_dirs.emplace_back(std::string{fmt::UTF(dirString.u8string()).data}); diff --git a/src/common/config.h b/src/common/config.h index 701aadb12..9d943008b 100644 --- a/src/common/config.h +++ b/src/common/config.h @@ -15,6 +15,9 @@ void load(const std::filesystem::path& path); void save(const std::filesystem::path& path); void saveMainWindow(const std::filesystem::path& path); +std::string getTrophyKey(); +void setTrophyKey(std::string key); + bool isNeoMode(); bool isFullscreenMode(); bool getPlayBGM(); diff --git a/src/core/crypto/crypto.cpp b/src/core/crypto/crypto.cpp index 00f1dea46..4020edfd8 100644 --- a/src/core/crypto/crypto.cpp +++ b/src/core/crypto/crypto.cpp @@ -2,6 +2,7 @@ // SPDX-License-Identifier: GPL-2.0-or-later #include + #include "crypto.h" CryptoPP::RSA::PrivateKey Crypto::key_pkg_derived_key3_keyset_init() { @@ -137,17 +138,20 @@ void Crypto::aesCbcCfb128DecryptEntry(std::span ivkey, } } -void Crypto::decryptEFSM(std::span NPcommID, +void Crypto::decryptEFSM(std::span trophyKey, + std::span NPcommID, std::span efsmIv, std::span ciphertext, std::span decrypted) { - std::vector TrophyIV = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}; // step 1: Encrypt NPcommID CryptoPP::CBC_Mode::Encryption encrypt; + std::vector trophyIv(16, 0); std::vector trpKey(16); + encrypt.SetKeyWithIV(trophyKey.data(), trophyKey.size(), trophyIv.data()); encrypt.ProcessData(trpKey.data(), NPcommID.data(), 16); + // step 2: decrypt efsm. CryptoPP::CBC_Mode::Decryption decrypt; decrypt.SetKeyWithIV(trpKey.data(), trpKey.size(), efsmIv.data()); diff --git a/src/core/crypto/crypto.h b/src/core/crypto/crypto.h index 83249bd7d..b5d8104b5 100644 --- a/src/core/crypto/crypto.h +++ b/src/core/crypto/crypto.h @@ -32,7 +32,8 @@ public: void aesCbcCfb128DecryptEntry(std::span ivkey, std::span ciphertext, std::span decrypted); - void decryptEFSM(std::span, std::span efsmIv, + void decryptEFSM(std::span trophyKey, + std::span NPcommID, std::span efsmIv, std::span ciphertext, std::span decrypted); void PfsGenCryptoKey(std::span ekpfs, std::span seed, diff --git a/src/core/devtools/widget/reg_popup.cpp b/src/core/devtools/widget/reg_popup.cpp index 2727e1745..fae620901 100644 --- a/src/core/devtools/widget/reg_popup.cpp +++ b/src/core/devtools/widget/reg_popup.cpp @@ -66,7 +66,7 @@ void RegPopup::DrawColorBuffer(const AmdGpu::Liverpool::ColorBuffer& buffer) { "GetColorSliceSize()", buffer.GetColorSliceSize(), "GetTilingMode()", buffer.GetTilingMode(), "IsTiled()", buffer.IsTiled(), - "NumFormat()", buffer.NumFormat() + "NumFormat()", buffer.GetNumberFmt() ); // clang-format on diff --git a/src/core/file_format/trp.cpp b/src/core/file_format/trp.cpp index 2ca88c778..d25c93c3f 100644 --- a/src/core/file_format/trp.cpp +++ b/src/core/file_format/trp.cpp @@ -1,6 +1,7 @@ // SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project // SPDX-License-Identifier: GPL-2.0-or-later +#include "common/config.h" #include "common/logging/log.h" #include "common/path_util.h" #include "trp.h" @@ -33,12 +34,29 @@ static void removePadding(std::vector& vec) { } } +static void hexToBytes(const char* hex, unsigned char* dst) { + for (size_t i = 0; hex[i] != 0; i++) { + const unsigned char value = (hex[i] < 0x3A) ? (hex[i] - 0x30) : (hex[i] - 0x37); + dst[i / 2] |= ((i % 2) == 0) ? (value << 4) : (value); + } +} + bool TRP::Extract(const std::filesystem::path& trophyPath, const std::string titleId) { std::filesystem::path gameSysDir = trophyPath / "sce_sys/trophy/"; if (!std::filesystem::exists(gameSysDir)) { LOG_CRITICAL(Common_Filesystem, "Game sce_sys directory doesn't exist"); return false; } + + const auto user_key_str = Config::getTrophyKey(); + if (user_key_str.size() != 32) { + LOG_CRITICAL(Common_Filesystem, "Trophy decryption key is not specified"); + return false; + } + + std::array user_key{}; + hexToBytes(user_key_str.c_str(), user_key.data()); + for (int index = 0; const auto& it : std::filesystem::directory_iterator(gameSysDir)) { if (it.is_regular_file()) { GetNPcommID(trophyPath, index); @@ -97,7 +115,7 @@ bool TRP::Extract(const std::filesystem::path& trophyPath, const std::string tit return false; } file.Read(ESFM); - crypto.decryptEFSM(np_comm_id, esfmIv, ESFM, XML); // decrypt + crypto.decryptEFSM(user_key, np_comm_id, esfmIv, ESFM, XML); // decrypt removePadding(XML); std::string xml_name = entry.entry_name; size_t pos = xml_name.find("ESFM"); diff --git a/src/core/libraries/audio/audioout.cpp b/src/core/libraries/audio/audioout.cpp index d69454c39..f0ad59c3b 100644 --- a/src/core/libraries/audio/audioout.cpp +++ b/src/core/libraries/audio/audioout.cpp @@ -3,13 +3,13 @@ #include #include -#include +#include +#include #include #include "common/assert.h" #include "common/config.h" #include "common/logging/log.h" -#include "common/polyfill_thread.h" #include "common/thread.h" #include "core/libraries/audio/audioout.h" #include "core/libraries/audio/audioout_backend.h" @@ -18,7 +18,7 @@ namespace Libraries::AudioOut { -std::shared_mutex ports_mutex; +std::mutex port_open_mutex{}; std::array ports_out{}; static std::unique_ptr audio; @@ -93,17 +93,20 @@ int PS4_SYSV_ABI sceAudioOutClose(s32 handle) { return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; } - std::scoped_lock lock(ports_mutex); + std::unique_lock open_lock{port_open_mutex}; auto& port = ports_out.at(handle - 1); - if (!port.impl) { - return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; + { + std::unique_lock lock{port.mutex}; + if (!port.IsOpen()) { + return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; + } + std::free(port.output_buffer); + port.output_buffer = nullptr; + port.output_ready = false; + port.impl = nullptr; } - + // Stop outside of port lock scope to prevent deadlocks. port.output_thread.Stop(); - std::free(port.output_buffer); - port.output_buffer = nullptr; - port.output_ready = false; - port.impl = nullptr; return ORBIS_OK; } @@ -172,35 +175,34 @@ int PS4_SYSV_ABI sceAudioOutGetPortState(s32 handle, OrbisAudioOutPortState* sta return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; } - std::scoped_lock lock(ports_mutex); - const auto& port = ports_out.at(handle - 1); - if (!port.impl) { - return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; + auto& port = ports_out.at(handle - 1); + { + std::unique_lock lock{port.mutex}; + if (!port.IsOpen()) { + return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; + } + switch (port.type) { + case OrbisAudioOutPort::Main: + case OrbisAudioOutPort::Bgm: + case OrbisAudioOutPort::Voice: + state->output = 1; + state->channel = port.format_info.num_channels > 2 ? 2 : port.format_info.num_channels; + break; + case OrbisAudioOutPort::Personal: + case OrbisAudioOutPort::Padspk: + state->output = 4; + state->channel = 1; + break; + case OrbisAudioOutPort::Aux: + state->output = 0; + state->channel = 0; + break; + default: + UNREACHABLE(); + } + state->rerouteCounter = 0; + state->volume = 127; } - - state->rerouteCounter = 0; - state->volume = 127; - - switch (port.type) { - case OrbisAudioOutPort::Main: - case OrbisAudioOutPort::Bgm: - case OrbisAudioOutPort::Voice: - state->output = 1; - state->channel = port.format_info.num_channels > 2 ? 2 : port.format_info.num_channels; - break; - case OrbisAudioOutPort::Personal: - case OrbisAudioOutPort::Padspk: - state->output = 4; - state->channel = 1; - break; - case OrbisAudioOutPort::Aux: - state->output = 0; - state->channel = 0; - break; - default: - UNREACHABLE(); - } - return ORBIS_OK; } @@ -279,15 +281,16 @@ static void AudioOutputThread(PortOut* port, const std::stop_token& stop) { while (true) { timer.Start(); { - std::unique_lock lock{port->output_mutex}; - Common::CondvarWait(port->output_cv, lock, stop, [&] { return port->output_ready; }); - if (stop.stop_requested()) { - break; + std::unique_lock lock{port->mutex}; + if (port->output_cv.wait(lock, stop, [&] { return port->output_ready; })) { + port->impl->Output(port->output_buffer); + port->output_ready = false; } - port->impl->Output(port->output_buffer); - port->output_ready = false; } port->output_cv.notify_one(); + if (stop.stop_requested()) { + break; + } timer.End(); } } @@ -332,27 +335,30 @@ s32 PS4_SYSV_ABI sceAudioOutOpen(UserService::OrbisUserServiceUserId user_id, return ORBIS_AUDIO_OUT_ERROR_INVALID_FORMAT; } - std::scoped_lock lock{ports_mutex}; + std::unique_lock open_lock{port_open_mutex}; const auto port = - std::ranges::find_if(ports_out, [&](const PortOut& p) { return p.impl == nullptr; }); + std::ranges::find_if(ports_out, [&](const PortOut& p) { return !p.IsOpen(); }); if (port == ports_out.end()) { LOG_ERROR(Lib_AudioOut, "Audio ports are full"); return ORBIS_AUDIO_OUT_ERROR_PORT_FULL; } - port->type = port_type; - port->format_info = GetFormatInfo(format); - port->sample_rate = sample_rate; - port->buffer_frames = length; - port->volume.fill(SCE_AUDIO_OUT_VOLUME_0DB); + { + std::unique_lock port_lock(port->mutex); - port->impl = audio->Open(*port); + port->type = port_type; + port->format_info = GetFormatInfo(format); + port->sample_rate = sample_rate; + port->buffer_frames = length; + port->volume.fill(SCE_AUDIO_OUT_VOLUME_0DB); - port->output_buffer = std::malloc(port->BufferSize()); - port->output_ready = false; - port->output_thread.Run( - [port](const std::stop_token& stop) { AudioOutputThread(&*port, stop); }); + port->impl = audio->Open(*port); + port->output_buffer = std::malloc(port->BufferSize()); + port->output_ready = false; + port->output_thread.Run( + [port](const std::stop_token& stop) { AudioOutputThread(&*port, stop); }); + } return std::distance(ports_out.begin(), port) + 1; } @@ -367,14 +373,13 @@ s32 PS4_SYSV_ABI sceAudioOutOutput(s32 handle, void* ptr) { } auto& port = ports_out.at(handle - 1); - if (!port.impl) { - return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; - } - { - std::unique_lock lock{port.output_mutex}; + std::unique_lock lock{port.mutex}; + if (!port.IsOpen()) { + return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; + } port.output_cv.wait(lock, [&] { return !port.output_ready; }); - if (ptr != nullptr) { + if (ptr != nullptr && port.IsOpen()) { std::memcpy(port.output_buffer, ptr, port.BufferSize()); port.output_ready = true; } @@ -488,19 +493,19 @@ s32 PS4_SYSV_ABI sceAudioOutSetVolume(s32 handle, s32 flag, s32* vol) { return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; } - std::scoped_lock lock(ports_mutex); auto& port = ports_out.at(handle - 1); - if (!port.impl) { - return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; - } - - for (int i = 0; i < port.format_info.num_channels; i++, flag >>= 1u) { - if (flag & 0x1u) { - port.volume[i] = vol[i]; + { + std::unique_lock lock{port.mutex}; + if (!port.IsOpen()) { + return ORBIS_AUDIO_OUT_ERROR_INVALID_PORT; } + for (int i = 0; i < port.format_info.num_channels; i++, flag >>= 1u) { + if (flag & 0x1u) { + port.volume[i] = vol[i]; + } + } + port.impl->SetVolume(port.volume); } - - port.impl->SetVolume(port.volume); return ORBIS_OK; } diff --git a/src/core/libraries/audio/audioout.h b/src/core/libraries/audio/audioout.h index 4f7378dcd..5eafb43a1 100644 --- a/src/core/libraries/audio/audioout.h +++ b/src/core/libraries/audio/audioout.h @@ -3,7 +3,9 @@ #pragma once +#include #include +#include #include "common/bit_field.h" #include "core/libraries/kernel/threads.h" @@ -74,10 +76,10 @@ struct AudioFormatInfo { }; struct PortOut { + std::mutex mutex; std::unique_ptr impl{}; void* output_buffer; - std::mutex output_mutex; std::condition_variable_any output_cv; bool output_ready; Kernel::Thread output_thread{}; @@ -88,6 +90,10 @@ struct PortOut { u32 buffer_frames; std::array volume; + [[nodiscard]] bool IsOpen() const { + return impl != nullptr; + } + [[nodiscard]] u32 BufferSize() const { return buffer_frames * format_info.FrameSize(); } diff --git a/src/core/libraries/audio/sdl_audio.cpp b/src/core/libraries/audio/sdl_audio.cpp index 59d2d5cfb..762a9f682 100644 --- a/src/core/libraries/audio/sdl_audio.cpp +++ b/src/core/libraries/audio/sdl_audio.cpp @@ -14,7 +14,7 @@ namespace Libraries::AudioOut { class SDLPortBackend : public PortBackend { public: explicit SDLPortBackend(const PortOut& port) - : frame_size(port.format_info.FrameSize()), buffer_size(port.BufferSize()) { + : frame_size(port.format_info.FrameSize()), guest_buffer_size(port.BufferSize()) { // We want the latency for delivering frames out to be as small as possible, // so set the sample frames hint to the number of frames per buffer. const auto samples_num_str = std::to_string(port.buffer_frames); @@ -33,7 +33,7 @@ public: LOG_ERROR(Lib_AudioOut, "Failed to create SDL audio stream: {}", SDL_GetError()); return; } - queue_threshold = CalculateQueueThreshold(); + CalculateQueueThreshold(); if (!SDL_SetAudioStreamInputChannelMap(stream, port.format_info.channel_layout.data(), port.format_info.num_channels)) { LOG_ERROR(Lib_AudioOut, "Failed to configure SDL audio stream channel map: {}", @@ -71,9 +71,9 @@ public: queue_threshold); SDL_ClearAudioStream(stream); // Recalculate the threshold in case this happened because of a device change. - queue_threshold = CalculateQueueThreshold(); + CalculateQueueThreshold(); } - if (!SDL_PutAudioStreamData(stream, ptr, static_cast(buffer_size))) { + if (!SDL_PutAudioStreamData(stream, ptr, static_cast(guest_buffer_size))) { LOG_ERROR(Lib_AudioOut, "Failed to output to SDL audio stream: {}", SDL_GetError()); } } @@ -91,7 +91,7 @@ public: } private: - [[nodiscard]] u32 CalculateQueueThreshold() const { + void CalculateQueueThreshold() { SDL_AudioSpec discard; int sdl_buffer_frames; if (!SDL_GetAudioDeviceFormat(SDL_GetAudioStreamDevice(stream), &discard, @@ -100,13 +100,22 @@ private: SDL_GetError()); sdl_buffer_frames = 0; } - return std::max(buffer_size, sdl_buffer_frames * frame_size) * 4; + const auto sdl_buffer_size = sdl_buffer_frames * frame_size; + const auto new_threshold = std::max(guest_buffer_size, sdl_buffer_size) * 4; + if (host_buffer_size != sdl_buffer_size || queue_threshold != new_threshold) { + host_buffer_size = sdl_buffer_size; + queue_threshold = new_threshold; + LOG_INFO(Lib_AudioOut, + "SDL audio buffers: guest = {} bytes, host = {} bytes, threshold = {} bytes", + guest_buffer_size, host_buffer_size, queue_threshold); + } } u32 frame_size; - u32 buffer_size; - u32 queue_threshold; - SDL_AudioStream* stream; + u32 guest_buffer_size; + u32 host_buffer_size{}; + u32 queue_threshold{}; + SDL_AudioStream* stream{}; }; std::unique_ptr SDLAudioOut::Open(PortOut& port) { diff --git a/src/core/libraries/kernel/memory.cpp b/src/core/libraries/kernel/memory.cpp index 7d326cbbf..8deefb496 100644 --- a/src/core/libraries/kernel/memory.cpp +++ b/src/core/libraries/kernel/memory.cpp @@ -505,6 +505,41 @@ int PS4_SYSV_ABI posix_munmap(void* addr, size_t len) { return result; } +static constexpr int MAX_PRT_APERTURES = 3; +static constexpr VAddr PRT_AREA_START_ADDR = 0x1000000000; +static constexpr size_t PRT_AREA_SIZE = 0xec00000000; +static std::array, MAX_PRT_APERTURES> PrtApertures{}; + +int PS4_SYSV_ABI sceKernelSetPrtAperture(int id, VAddr address, size_t size) { + if (id < 0 || id >= MAX_PRT_APERTURES) { + return ORBIS_KERNEL_ERROR_EINVAL; + } + + if (address < PRT_AREA_START_ADDR || address + size > PRT_AREA_START_ADDR + PRT_AREA_SIZE) { + return ORBIS_KERNEL_ERROR_EINVAL; + } + + if (address % 4096 != 0) { + return ORBIS_KERNEL_ERROR_EINVAL; + } + + LOG_WARNING(Kernel_Vmm, + "PRT aperture id = {}, address = {:#x}, size = {:#x} is set but not used", id, + address, size); + + PrtApertures[id] = {address, size}; + return ORBIS_OK; +} + +int PS4_SYSV_ABI sceKernelGetPrtAperture(int id, VAddr* address, size_t* size) { + if (id < 0 || id >= MAX_PRT_APERTURES) { + return ORBIS_KERNEL_ERROR_EINVAL; + } + + std::tie(*address, *size) = PrtApertures[id]; + return ORBIS_OK; +} + void RegisterMemory(Core::Loader::SymbolsResolver* sym) { LIB_FUNCTION("rTXw65xmLIA", "libkernel", 1, "libkernel", 1, 1, sceKernelAllocateDirectMemory); LIB_FUNCTION("B+vc2AO2Zrc", "libkernel", 1, "libkernel", 1, 1, @@ -551,6 +586,10 @@ void RegisterMemory(Core::Loader::SymbolsResolver* sym) { LIB_FUNCTION("BPE9s9vQQXo", "libScePosix", 1, "libkernel", 1, 1, posix_mmap); LIB_FUNCTION("UqDGjXA5yUM", "libkernel", 1, "libkernel", 1, 1, posix_munmap); LIB_FUNCTION("UqDGjXA5yUM", "libScePosix", 1, "libkernel", 1, 1, posix_munmap); + + // PRT memory management + LIB_FUNCTION("BohYr-F7-is", "libkernel", 1, "libkernel", 1, 1, sceKernelSetPrtAperture); + LIB_FUNCTION("L0v2Go5jOuM", "libkernel", 1, "libkernel", 1, 1, sceKernelGetPrtAperture); } } // namespace Libraries::Kernel diff --git a/src/core/libraries/kernel/threads.h b/src/core/libraries/kernel/threads.h index ad1393599..409136968 100644 --- a/src/core/libraries/kernel/threads.h +++ b/src/core/libraries/kernel/threads.h @@ -55,6 +55,9 @@ public: stop.request_stop(); Join(); } + thread = nullptr; + func = nullptr; + stop = std::stop_source{}; } static void* PS4_SYSV_ABI RunWrapper(void* arg) { diff --git a/src/core/libraries/libs.cpp b/src/core/libraries/libs.cpp index c30c2d7c3..49cd54a5b 100644 --- a/src/core/libraries/libs.cpp +++ b/src/core/libraries/libs.cpp @@ -47,6 +47,8 @@ #include "core/libraries/videodec/videodec.h" #include "core/libraries/videodec/videodec2.h" #include "core/libraries/videoout/video_out.h" +#include "fiber/fiber.h" +#include "jpeg/jpegenc.h" namespace Libraries { @@ -93,6 +95,8 @@ void InitHLELibs(Core::Loader::SymbolsResolver* sym) { Libraries::Videodec::RegisterlibSceVideodec(sym); Libraries::RazorCpu::RegisterlibSceRazorCpu(sym); Libraries::Move::RegisterlibSceMove(sym); + Libraries::Fiber::RegisterlibSceFiber(sym); + Libraries::JpegEnc::RegisterlibSceJpegEnc(sym); } } // namespace Libraries diff --git a/src/core/libraries/np_manager/np_manager.cpp b/src/core/libraries/np_manager/np_manager.cpp index ec9cc6bf5..87d752c69 100644 --- a/src/core/libraries/np_manager/np_manager.cpp +++ b/src/core/libraries/np_manager/np_manager.cpp @@ -972,11 +972,8 @@ int PS4_SYSV_ABI sceNpGetGamePresenceStatusA() { } int PS4_SYSV_ABI sceNpGetNpId(OrbisUserServiceUserId user_id, OrbisNpId* np_id) { - LOG_INFO(Lib_NpManager, "user_id {}", user_id); - const auto name = Config::getUserName(); - std::memset(np_id, 0, sizeof(OrbisNpId)); - name.copy(np_id->handle.data, sizeof(np_id->handle.data)); - return ORBIS_OK; + LOG_DEBUG(Lib_NpManager, "user_id {}", user_id); + return ORBIS_NP_ERROR_SIGNED_OUT; } int PS4_SYSV_ABI sceNpGetNpReachabilityState() { @@ -986,10 +983,7 @@ int PS4_SYSV_ABI sceNpGetNpReachabilityState() { int PS4_SYSV_ABI sceNpGetOnlineId(s32 user_id, OrbisNpOnlineId* online_id) { LOG_DEBUG(Lib_NpManager, "user_id {}", user_id); - const auto name = Config::getUserName(); - std::memset(online_id, 0, sizeof(OrbisNpOnlineId)); - name.copy(online_id->data, sizeof(online_id->data)); - return ORBIS_OK; + return ORBIS_NP_ERROR_SIGNED_OUT; } int PS4_SYSV_ABI sceNpGetParentalControlInfo() { diff --git a/src/core/libraries/pad/pad.cpp b/src/core/libraries/pad/pad.cpp index 98f086dd9..7eb628a90 100644 --- a/src/core/libraries/pad/pad.cpp +++ b/src/core/libraries/pad/pad.cpp @@ -286,6 +286,7 @@ int PS4_SYSV_ABI scePadOutputReport() { } int PS4_SYSV_ABI scePadRead(s32 handle, OrbisPadData* pData, s32 num) { + LOG_TRACE(Lib_Pad, "called"); int connected_count = 0; bool connected = false; Input::State states[64]; @@ -304,16 +305,15 @@ int PS4_SYSV_ABI scePadRead(s32 handle, OrbisPadData* pData, s32 num) { pData[i].rightStick.y = states[i].axes[static_cast(Input::Axis::RightY)]; pData[i].analogButtons.l2 = states[i].axes[static_cast(Input::Axis::TriggerLeft)]; pData[i].analogButtons.r2 = states[i].axes[static_cast(Input::Axis::TriggerRight)]; - pData[i].orientation.x = 0.0f; - pData[i].orientation.y = 0.0f; - pData[i].orientation.z = 0.0f; - pData[i].orientation.w = 1.0f; - pData[i].acceleration.x = 0.0f; - pData[i].acceleration.y = 0.0f; - pData[i].acceleration.z = 0.0f; - pData[i].angularVelocity.x = 0.0f; - pData[i].angularVelocity.y = 0.0f; - pData[i].angularVelocity.z = 0.0f; + pData[i].acceleration.x = states[i].acceleration.x; + pData[i].acceleration.y = states[i].acceleration.y; + pData[i].acceleration.z = states[i].acceleration.z; + pData[i].angularVelocity.x = states[i].angularVelocity.x; + pData[i].angularVelocity.y = states[i].angularVelocity.y; + pData[i].angularVelocity.z = states[i].angularVelocity.z; + Input::GameController::CalculateOrientation(pData[i].acceleration, pData[i].angularVelocity, + 1.0f / controller->accel_poll_rate, + pData[i].orientation); pData[i].touchData.touchNum = (states[i].touchpad[0].state ? 1 : 0) + (states[i].touchpad[1].state ? 1 : 0); pData[i].touchData.touch[0].x = states[i].touchpad[0].x; @@ -352,6 +352,7 @@ int PS4_SYSV_ABI scePadReadHistory() { } int PS4_SYSV_ABI scePadReadState(s32 handle, OrbisPadData* pData) { + LOG_TRACE(Lib_Pad, "called"); if (handle == ORBIS_PAD_ERROR_DEVICE_NO_HANDLE) { return ORBIS_PAD_ERROR_INVALID_HANDLE; } @@ -367,16 +368,15 @@ int PS4_SYSV_ABI scePadReadState(s32 handle, OrbisPadData* pData) { pData->rightStick.y = state.axes[static_cast(Input::Axis::RightY)]; pData->analogButtons.l2 = state.axes[static_cast(Input::Axis::TriggerLeft)]; pData->analogButtons.r2 = state.axes[static_cast(Input::Axis::TriggerRight)]; - pData->orientation.x = 0; - pData->orientation.y = 0; - pData->orientation.z = 0; - pData->orientation.w = 1; - pData->acceleration.x = 0.0f; - pData->acceleration.y = 0.0f; - pData->acceleration.z = 0.0f; - pData->angularVelocity.x = 0.0f; - pData->angularVelocity.y = 0.0f; - pData->angularVelocity.z = 0.0f; + pData->acceleration.x = state.acceleration.x; + pData->acceleration.y = state.acceleration.y; + pData->acceleration.z = state.acceleration.z; + pData->angularVelocity.x = state.angularVelocity.x; + pData->angularVelocity.y = state.angularVelocity.y; + pData->angularVelocity.z = state.angularVelocity.z; + Input::GameController::CalculateOrientation(pData->acceleration, pData->angularVelocity, + 1.0f / controller->accel_poll_rate, + pData->orientation); pData->touchData.touchNum = (state.touchpad[0].state ? 1 : 0) + (state.touchpad[1].state ? 1 : 0); pData->touchData.touch[0].x = state.touchpad[0].x; @@ -498,6 +498,8 @@ int PS4_SYSV_ABI scePadSetLoginUserNumber() { int PS4_SYSV_ABI scePadSetMotionSensorState(s32 handle, bool bEnable) { LOG_ERROR(Lib_Pad, "(STUBBED) called"); return ORBIS_OK; + // it's already handled by the SDL backend and will be on no matter what + // (assuming the controller supports it) } int PS4_SYSV_ABI scePadSetProcessFocus() { diff --git a/src/core/libraries/save_data/savedata.cpp b/src/core/libraries/save_data/savedata.cpp index 66899fb34..b573ded1e 100644 --- a/src/core/libraries/save_data/savedata.cpp +++ b/src/core/libraries/save_data/savedata.cpp @@ -2,6 +2,7 @@ // SPDX-License-Identifier: GPL-2.0-or-later #include +#include #include #include @@ -1139,10 +1140,6 @@ Error PS4_SYSV_ABI sceSaveDataGetSaveDataMemory2(OrbisSaveDataMemoryGet2* getPar LOG_INFO(Lib_SaveData, "called without save memory initialized"); return Error::MEMORY_NOT_READY; } - if (SaveMemory::IsSaving()) { - LOG_TRACE(Lib_SaveData, "called while saving"); - return Error::BUSY_FOR_SAVING; - } LOG_DEBUG(Lib_SaveData, "called"); auto data = getParam->data; if (data != nullptr) { @@ -1502,8 +1499,14 @@ Error PS4_SYSV_ABI sceSaveDataSetSaveDataMemory2(const OrbisSaveDataMemorySet2* return Error::MEMORY_NOT_READY; } if (SaveMemory::IsSaving()) { - LOG_TRACE(Lib_SaveData, "called while saving"); - return Error::BUSY_FOR_SAVING; + int count = 0; + while (++count < 100 && SaveMemory::IsSaving()) { // try for more 10 seconds + std::this_thread::sleep_for(chrono::milliseconds(100)); + } + if (SaveMemory::IsSaving()) { + LOG_TRACE(Lib_SaveData, "called while saving"); + return Error::BUSY_FOR_SAVING; + } } LOG_DEBUG(Lib_SaveData, "called"); auto data = setParam->data; @@ -1584,8 +1587,8 @@ Error PS4_SYSV_ABI sceSaveDataSetupSaveDataMemory2(const OrbisSaveDataMemorySetu } else { SaveMemory::SetIcon(nullptr, 0); } + SaveMemory::TriggerSaveWithoutEvent(); } - SaveMemory::TriggerSaveWithoutEvent(); if (g_fw_ver >= ElfInfo::FW_45 && result != nullptr) { result->existedMemorySize = existed_size; } diff --git a/src/core/libraries/usbd/usbd.cpp b/src/core/libraries/usbd/usbd.cpp index c0e1b7ea8..fdfa50b23 100644 --- a/src/core/libraries/usbd/usbd.cpp +++ b/src/core/libraries/usbd/usbd.cpp @@ -10,327 +10,327 @@ namespace Libraries::Usbd { int PS4_SYSV_ABI sceUsbdAllocTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdAttachKernelDriver() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdBulkTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdCancelTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdCheckConnected() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdClaimInterface() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdClearHalt() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdClose() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdControlTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdControlTransferGetData() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdControlTransferGetSetup() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdDetachKernelDriver() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdEventHandlerActive() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdEventHandlingOk() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdExit() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdFillBulkTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdFillControlSetup() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdFillControlTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdFillInterruptTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdFillIsoTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdFreeConfigDescriptor() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdFreeDeviceList() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdFreeTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetActiveConfigDescriptor() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetBusNumber() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetConfigDescriptor() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetConfigDescriptorByValue() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetConfiguration() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetDescriptor() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetDevice() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetDeviceAddress() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetDeviceDescriptor() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetDeviceList() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetDeviceSpeed() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetIsoPacketBuffer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetMaxIsoPacketSize() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetMaxPacketSize() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetStringDescriptor() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdGetStringDescriptorAscii() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdHandleEvents() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdHandleEventsLocked() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdHandleEventsTimeout() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_DEBUG(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdInit() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return 0x80240005; // Skip } int PS4_SYSV_ABI sceUsbdInterruptTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdKernelDriverActive() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdLockEvents() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdLockEventWaiters() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdOpen() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdOpenDeviceWithVidPid() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdRefDevice() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdReleaseInterface() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdResetDevice() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdSetConfiguration() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdSetInterfaceAltSetting() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdSetIsoPacketLengths() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdSubmitTransfer() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdTryLockEvents() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdUnlockEvents() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdUnlockEventWaiters() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdUnrefDevice() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI sceUsbdWaitForEvent() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI Func_65F6EF33E38FFF50() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI Func_97F056BAD90AADE7() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI Func_C55104A33B35B264() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } int PS4_SYSV_ABI Func_D56B43060720B1E0() { - LOG_ERROR(Lib_Usbd, "(STUBBED)called"); + LOG_ERROR(Lib_Usbd, "(STUBBED) called"); return ORBIS_OK; } diff --git a/src/core/libraries/videoout/video_out.cpp b/src/core/libraries/videoout/video_out.cpp index f36de6ade..78a2b11a4 100644 --- a/src/core/libraries/videoout/video_out.cpp +++ b/src/core/libraries/videoout/video_out.cpp @@ -52,8 +52,7 @@ s32 PS4_SYSV_ABI sceVideoOutAddFlipEvent(Kernel::SceKernelEqueue eq, s32 handle, Kernel::EqueueEvent event{}; event.event.ident = u64(OrbisVideoOutEventId::Flip); event.event.filter = Kernel::SceKernelEvent::Filter::VideoOut; - // The library only sets EV_ADD but kernel driver forces EV_CLEAR - event.event.flags = Kernel::SceKernelEvent::Flags::Clear; + event.event.flags = Kernel::SceKernelEvent::Flags::Add; event.event.udata = udata; event.event.fflags = 0; event.event.data = 0; @@ -79,8 +78,7 @@ s32 PS4_SYSV_ABI sceVideoOutAddVblankEvent(Kernel::SceKernelEqueue eq, s32 handl Kernel::EqueueEvent event{}; event.event.ident = u64(OrbisVideoOutEventId::Vblank); event.event.filter = Kernel::SceKernelEvent::Filter::VideoOut; - // The library only sets EV_ADD but kernel driver forces EV_CLEAR - event.event.flags = Kernel::SceKernelEvent::Flags::Clear; + event.event.flags = Kernel::SceKernelEvent::Flags::Add; event.event.udata = udata; event.event.fflags = 0; event.event.data = 0; diff --git a/src/emulator.cpp b/src/emulator.cpp index dbe21a141..4f0c61236 100644 --- a/src/emulator.cpp +++ b/src/emulator.cpp @@ -282,16 +282,14 @@ void Emulator::Run(const std::filesystem::path& file) { } void Emulator::LoadSystemModules(const std::filesystem::path& file, std::string game_serial) { - constexpr std::array ModulesToLoad{ + constexpr std::array ModulesToLoad{ {{"libSceNgs2.sprx", &Libraries::Ngs2::RegisterlibSceNgs2}, - {"libSceFiber.sprx", &Libraries::Fiber::RegisterlibSceFiber}, {"libSceUlt.sprx", nullptr}, {"libSceJson.sprx", nullptr}, {"libSceJson2.sprx", nullptr}, {"libSceLibcInternal.sprx", &Libraries::LibcInternal::RegisterlibSceLibcInternal}, {"libSceDiscMap.sprx", &Libraries::DiscMap::RegisterlibSceDiscMap}, {"libSceRtc.sprx", &Libraries::Rtc::RegisterlibSceRtc}, - {"libSceJpegEnc.sprx", &Libraries::JpegEnc::RegisterlibSceJpegEnc}, {"libSceCesCs.sprx", nullptr}, {"libSceFont.sprx", nullptr}, {"libSceFontFt.sprx", nullptr}, diff --git a/src/input/controller.cpp b/src/input/controller.cpp index 3927b096f..daef9c940 100644 --- a/src/input/controller.cpp +++ b/src/input/controller.cpp @@ -2,6 +2,7 @@ // SPDX-License-Identifier: GPL-2.0-or-later #include +#include "common/logging/log.h" #include "core/libraries/kernel/time.h" #include "core/libraries/pad/pad.h" #include "input/controller.h" @@ -116,6 +117,103 @@ void GameController::Axis(int id, Input::Axis axis, int value) { AddState(state); } +void GameController::Gyro(int id, const float gyro[3]) { + std::scoped_lock lock{m_mutex}; + auto state = GetLastState(); + state.time = Libraries::Kernel::sceKernelGetProcessTime(); + + // Update the angular velocity (gyro data) + state.angularVelocity.x = gyro[0]; // X-axis + state.angularVelocity.y = gyro[1]; // Y-axis + state.angularVelocity.z = gyro[2]; // Z-axis + + AddState(state); +} +void GameController::Acceleration(int id, const float acceleration[3]) { + std::scoped_lock lock{m_mutex}; + auto state = GetLastState(); + state.time = Libraries::Kernel::sceKernelGetProcessTime(); + + // Update the acceleration values + state.acceleration.x = acceleration[0]; // X-axis + state.acceleration.y = acceleration[1]; // Y-axis + state.acceleration.z = acceleration[2]; // Z-axis + + AddState(state); +} + +// Stolen from +// https://github.com/xioTechnologies/Open-Source-AHRS-With-x-IMU/blob/master/x-IMU%20IMU%20and%20AHRS%20Algorithms/x-IMU%20IMU%20and%20AHRS%20Algorithms/AHRS/MahonyAHRS.cs +float eInt[3] = {0.0f, 0.0f, 0.0f}; // Integral error terms +const float Kp = 50.0f; // Proportional gain +const float Ki = 1.0f; // Integral gain +Libraries::Pad::OrbisFQuaternion o = {1, 0, 0, 0}; +void GameController::CalculateOrientation(Libraries::Pad::OrbisFVector3& acceleration, + Libraries::Pad::OrbisFVector3& angularVelocity, + float deltaTime, + Libraries::Pad::OrbisFQuaternion& orientation) { + float ax = acceleration.x, ay = acceleration.y, az = acceleration.z; + float gx = angularVelocity.x, gy = angularVelocity.y, gz = angularVelocity.z; + + float q1 = o.w, q2 = o.x, q3 = o.y, q4 = o.z; + + // Normalize accelerometer measurement + float norm = std::sqrt(ax * ax + ay * ay + az * az); + if (norm == 0.0f) + return; // Handle NaN + norm = 1.0f / norm; + ax *= norm; + ay *= norm; + az *= norm; + + // Estimated direction of gravity + float vx = 2.0f * (q2 * q4 - q1 * q3); + float vy = 2.0f * (q1 * q2 + q3 * q4); + float vz = q1 * q1 - q2 * q2 - q3 * q3 + q4 * q4; + + // Error is cross product between estimated direction and measured direction of gravity + float ex = (ay * vz - az * vy); + float ey = (az * vx - ax * vz); + float ez = (ax * vy - ay * vx); + if (Ki > 0.0f) { + eInt[0] += ex * deltaTime; // Accumulate integral error + eInt[1] += ey * deltaTime; + eInt[2] += ez * deltaTime; + } else { + eInt[0] = eInt[1] = eInt[2] = 0.0f; // Prevent integral wind-up + } + + // Apply feedback terms + gx += Kp * ex + Ki * eInt[0]; + gy += Kp * ey + Ki * eInt[1]; + gz += Kp * ez + Ki * eInt[2]; + + //// Integrate rate of change of quaternion + // float pa = q2, pb = q3, pc = q4; + // q1 += (-q2 * gx - q3 * gy - q4 * gz) * (0.5f * deltaTime); + // q2 += (pa * gx + pb * gz - pc * gy) * (0.5f * deltaTime); + // q3 += (pb * gy - pa * gz + pc * gx) * (0.5f * deltaTime); + // q4 += (pc * gz + pa * gy - pb * gx) * (0.5f * deltaTime); + q1 += (-q2 * gx - q3 * gy - q4 * gz) * (0.5f * deltaTime); + q2 += (q1 * gx + q3 * gz - q4 * gy) * (0.5f * deltaTime); + q3 += (q1 * gy - q2 * gz + q4 * gx) * (0.5f * deltaTime); + q4 += (q1 * gz + q2 * gy - q3 * gx) * (0.5f * deltaTime); + + // Normalize quaternion + norm = std::sqrt(q1 * q1 + q2 * q2 + q3 * q3 + q4 * q4); + norm = 1.0f / norm; + orientation.w = q1 * norm; + orientation.x = q2 * norm; + orientation.y = q3 * norm; + orientation.z = q4 * norm; + o.w = q1 * norm; + o.x = q2 * norm; + o.y = q3 * norm; + o.z = q4 * norm; + LOG_DEBUG(Lib_Pad, "Calculated orientation: {:.2f} {:.2f} {:.2f} {:.2f}", orientation.x, + orientation.y, orientation.z, orientation.w); +} + void GameController::SetLightBarRGB(u8 r, u8 g, u8 b) { if (m_sdl_gamepad != nullptr) { SDL_SetGamepadLED(m_sdl_gamepad, r, g, b); @@ -149,6 +247,18 @@ void GameController::TryOpenSDLController() { int gamepad_count; SDL_JoystickID* gamepads = SDL_GetGamepads(&gamepad_count); m_sdl_gamepad = gamepad_count > 0 ? SDL_OpenGamepad(gamepads[0]) : nullptr; + if (SDL_SetGamepadSensorEnabled(m_sdl_gamepad, SDL_SENSOR_GYRO, true)) { + gyro_poll_rate = SDL_GetGamepadSensorDataRate(m_sdl_gamepad, SDL_SENSOR_GYRO); + LOG_INFO(Input, "Gyro initialized, poll rate: {}", gyro_poll_rate); + } else { + LOG_ERROR(Input, "Failed to initialize gyro controls for gamepad"); + } + if (SDL_SetGamepadSensorEnabled(m_sdl_gamepad, SDL_SENSOR_ACCEL, true)) { + accel_poll_rate = SDL_GetGamepadSensorDataRate(m_sdl_gamepad, SDL_SENSOR_ACCEL); + LOG_INFO(Input, "Accel initialized, poll rate: {}", accel_poll_rate); + } else { + LOG_ERROR(Input, "Failed to initialize accel controls for gamepad"); + } SDL_free(gamepads); SetLightBarRGB(0, 0, 255); diff --git a/src/input/controller.h b/src/input/controller.h index d425fb46c..c6fc02c24 100644 --- a/src/input/controller.h +++ b/src/input/controller.h @@ -33,6 +33,9 @@ struct State { u64 time = 0; int axes[static_cast(Axis::AxisMax)] = {128, 128, 128, 128, 0, 0}; TouchpadEntry touchpad[2] = {{false, 0, 0}, {false, 0, 0}}; + Libraries::Pad::OrbisFVector3 acceleration = {0.0f, 0.0f, 0.0f}; + Libraries::Pad::OrbisFVector3 angularVelocity = {0.0f, 0.0f, 0.0f}; + Libraries::Pad::OrbisFQuaternion orientation = {0.0f, 0.0f, 0.0f, 1.0f}; }; inline int GetAxis(int min, int max, int value) { @@ -53,12 +56,21 @@ public: void CheckButton(int id, Libraries::Pad::OrbisPadButtonDataOffset button, bool isPressed); void AddState(const State& state); void Axis(int id, Input::Axis axis, int value); + void Gyro(int id, const float gyro[3]); + void Acceleration(int id, const float acceleration[3]); void SetLightBarRGB(u8 r, u8 g, u8 b); bool SetVibration(u8 smallMotor, u8 largeMotor); void SetTouchpadState(int touchIndex, bool touchDown, float x, float y); void TryOpenSDLController(); u32 Poll(); + float gyro_poll_rate; + float accel_poll_rate; + static void CalculateOrientation(Libraries::Pad::OrbisFVector3& acceleration, + Libraries::Pad::OrbisFVector3& angularVelocity, + float deltaTime, + Libraries::Pad::OrbisFQuaternion& orientation); + private: struct StateInternal { bool obtained = false; diff --git a/src/qt_gui/check_update.cpp b/src/qt_gui/check_update.cpp index edd55b804..e3e019144 100644 --- a/src/qt_gui/check_update.cpp +++ b/src/qt_gui/check_update.cpp @@ -14,6 +14,7 @@ #include #include #include +#include #include #include #include @@ -24,11 +25,9 @@ #include #include #include -#include #include "check_update.h" using namespace Common::FS; -namespace fs = std::filesystem; CheckUpdate::CheckUpdate(const bool showMessage, QWidget* parent) : QDialog(parent), networkManager(new QNetworkAccessManager(this)) { @@ -254,7 +253,11 @@ void CheckUpdate::setupUI(const QString& downloadUrl, const QString& latestDate, connect(noButton, &QPushButton::clicked, this, [this]() { close(); }); autoUpdateCheckBox->setChecked(Config::autoUpdate()); +#if (QT_VERSION < QT_VERSION_CHECK(6, 7, 0)) connect(autoUpdateCheckBox, &QCheckBox::stateChanged, this, [](int state) { +#else + connect(autoUpdateCheckBox, &QCheckBox::checkStateChanged, this, [](Qt::CheckState state) { +#endif const auto user_dir = Common::FS::GetUserPath(Common::FS::PathType::UserDir); Config::setAutoUpdate(state == Qt::Checked); Config::save(user_dir / "config.toml"); diff --git a/src/qt_gui/settings_dialog.cpp b/src/qt_gui/settings_dialog.cpp index 0c1375c7f..6d76a5318 100644 --- a/src/qt_gui/settings_dialog.cpp +++ b/src/qt_gui/settings_dialog.cpp @@ -3,22 +3,24 @@ #include #include +#include #include +#include -#include #include "common/config.h" +#include "common/version.h" #include "qt_gui/compatibility_info.h" #ifdef ENABLE_DISCORD_RPC #include "common/discord_rpc_handler.h" +#include "common/singleton.h" #endif #ifdef ENABLE_UPDATER #include "check_update.h" #endif #include +#include "background_music_player.h" #include "common/logging/backend.h" #include "common/logging/filter.h" -#include "common/logging/formatter.h" -#include "main_window.h" #include "settings_dialog.h" #include "ui_settings_dialog.h" QStringList languageNames = {"Arabic", @@ -130,8 +132,13 @@ SettingsDialog::SettingsDialog(std::span physical_devices, // GENERAL TAB { #ifdef ENABLE_UPDATER +#if (QT_VERSION < QT_VERSION_CHECK(6, 7, 0)) connect(ui->updateCheckBox, &QCheckBox::stateChanged, this, [](int state) { Config::setAutoUpdate(state == Qt::Checked); }); +#else + connect(ui->updateCheckBox, &QCheckBox::checkStateChanged, this, + [](Qt::CheckState state) { Config::setAutoUpdate(state == Qt::Checked); }); +#endif connect(ui->updateComboBox, &QComboBox::currentTextChanged, this, [](const QString& channel) { Config::setUpdateChannel(channel.toStdString()); }); @@ -150,7 +157,12 @@ SettingsDialog::SettingsDialog(std::span physical_devices, emit CompatibilityChanged(); }); +#if (QT_VERSION < QT_VERSION_CHECK(6, 7, 0)) connect(ui->enableCompatibilityCheckBox, &QCheckBox::stateChanged, this, [this](int state) { +#else + connect(ui->enableCompatibilityCheckBox, &QCheckBox::checkStateChanged, this, + [this](Qt::CheckState state) { +#endif Config::setCompatibilityEnabled(state); emit CompatibilityChanged(); }); @@ -201,6 +213,8 @@ SettingsDialog::SettingsDialog(std::span physical_devices, ui->showSplashCheckBox->installEventFilter(this); ui->discordRPCCheckbox->installEventFilter(this); ui->userName->installEventFilter(this); + ui->label_Trophy->installEventFilter(this); + ui->trophyKeyLineEdit->installEventFilter(this); ui->logTypeGroupBox->installEventFilter(this); ui->logFilter->installEventFilter(this); #ifdef ENABLE_UPDATER @@ -295,6 +309,9 @@ void SettingsDialog::LoadValuesFromConfig() { QString::fromStdString(toml::find_or(data, "General", "logFilter", ""))); ui->userNameLineEdit->setText( QString::fromStdString(toml::find_or(data, "General", "userName", "shadPS4"))); + ui->trophyKeyLineEdit->setText( + QString::fromStdString(toml::find_or(data, "Keys", "TrophyKey", ""))); + ui->trophyKeyLineEdit->setEchoMode(QLineEdit::Password); ui->debugDump->setChecked(toml::find_or(data, "Debug", "DebugDump", false)); ui->vkValidationCheckBox->setChecked(toml::find_or(data, "Vulkan", "validation", false)); ui->vkSyncValidationCheckBox->setChecked( @@ -358,7 +375,7 @@ void SettingsDialog::InitializeEmulatorLanguages() { idx++; } - connect(ui->emulatorLanguageComboBox, qOverload(&QComboBox::currentIndexChanged), this, + connect(ui->emulatorLanguageComboBox, &QComboBox::currentIndexChanged, this, &SettingsDialog::OnLanguageChanged); } @@ -407,6 +424,10 @@ void SettingsDialog::updateNoteTextEdit(const QString& elementName) { text = tr("discordRPCCheckbox"); } else if (elementName == "userName") { text = tr("userName"); + } else if (elementName == "label_Trophy") { + text = tr("TrophyKey"); + } else if (elementName == "trophyKeyLineEdit") { + text = tr("TrophyKey"); } else if (elementName == "logTypeGroupBox") { text = tr("logTypeGroupBox"); } else if (elementName == "logFilter") { @@ -517,6 +538,7 @@ void SettingsDialog::UpdateSettings() { Config::setLogType(ui->logTypeComboBox->currentText().toStdString()); Config::setLogFilter(ui->logFilterLineEdit->text().toStdString()); Config::setUserName(ui->userNameLineEdit->text().toStdString()); + Config::setTrophyKey(ui->trophyKeyLineEdit->text().toStdString()); Config::setCursorState(ui->hideCursorComboBox->currentIndex()); Config::setCursorHideTimeout(ui->idleTimeoutSpinBox->value()); Config::setGpuId(ui->graphicsAdapterBox->currentIndex() - 1); @@ -578,4 +600,4 @@ void SettingsDialog::ResetInstallFolders() { } Config::setGameInstallDirs(settings_install_dirs_config); } -} \ No newline at end of file +} diff --git a/src/qt_gui/settings_dialog.ui b/src/qt_gui/settings_dialog.ui index af1edb0dd..2e7e3db37 100644 --- a/src/qt_gui/settings_dialog.ui +++ b/src/qt_gui/settings_dialog.ui @@ -11,8 +11,8 @@ 0 0 - 950 - 780 + 970 + 670 @@ -67,8 +67,8 @@ 0 0 - 822 - 487 + 946 + 536 @@ -77,87 +77,7 @@ 0 - - - - - - Logger - - - - - - - 0 - - - 0 - - - 0 - - - 0 - - - - - Log Type - - - - - - - async - - - - - sync - - - - - - - - - - - - - - 6 - - - 0 - - - - - - - Log Filter - - - - - - - - - - - - - - - - - - + @@ -194,7 +114,7 @@ - + @@ -268,10 +188,10 @@ - + - -1 + 6 QLayout::SizeConstraint::SetDefaultConstraint @@ -436,138 +356,8 @@ - - - - - - 0 - 0 - - - - - 0 - 0 - - - - GUI Settings - - - - 1 - - - 11 - - - - - Disable Trophy Pop-ups - - - - - - - - 0 - 0 - - - - Play title music - - - - - - - 1 - - - 0 - - - - - Qt::Orientation::Vertical - - - QSizePolicy::Policy::Fixed - - - - 20 - 13 - - - - - - - - - 0 - 0 - - - - - 16777215 - 16777215 - - - - Volume - - - - - - - Set the volume of the background music. - - - 100 - - - 10 - - - 20 - - - 50 - - - Qt::Orientation::Horizontal - - - false - - - false - - - QSlider::TickPosition::NoTicks - - - 10 - - - - - - - - - - - - + @@ -638,6 +428,160 @@ + + + + + + + 0 + 0 + + + + + 0 + 0 + + + + GUI Settings + + + + 1 + + + 11 + + + + + + 0 + 0 + + + + Play title music + + + + + + + 1 + + + 0 + + + + + + 0 + 0 + + + + + 16777215 + 16777215 + + + + Volume + + + + + + + Set the volume of the background music. + + + 100 + + + 10 + + + 20 + + + 50 + + + Qt::Orientation::Horizontal + + + false + + + false + + + QSlider::TickPosition::NoTicks + + + 10 + + + + + + + 6 + + + 0 + + + + + + + Trophy + + + + + + Disable Trophy Pop-ups + + + + + + + Trophy Key + + + + + + + + 0 + 0 + + + + + + + + + + + + + + + + + + @@ -655,8 +599,8 @@ 0 0 - 396 - 222 + 926 + 536 @@ -946,8 +890,8 @@ 0 0 - 536 - 192 + 926 + 536 @@ -1197,8 +1141,8 @@ 0 0 - 146 - 215 + 926 + 536 @@ -1211,18 +1155,25 @@ - - - Remove + + + 0 - - - - - - Add... - - + + + + Add... + + + + + + + Remove + + + + @@ -1263,71 +1214,145 @@ 0 0 - 288 - 163 + 926 + 536 - + + + 0 + + + 0 + - - - true - - - General - - - Qt::AlignmentFlag::AlignLeading|Qt::AlignmentFlag::AlignLeft|Qt::AlignmentFlag::AlignTop - - - - - - Enable Debug Dumping - - - - - - - Qt::Orientation::Vertical - - - QSizePolicy::Policy::MinimumExpanding - - - - 0 - 0 - - - - - - - - Enable Vulkan Validation Layers - - - - - - - Enable Vulkan Synchronization Validation - - - - - - - Enable RenderDoc Debugging - - - - - + + + + + true + + + General + + + Qt::AlignmentFlag::AlignLeading|Qt::AlignmentFlag::AlignLeft|Qt::AlignmentFlag::AlignTop + + + + + + Enable Debug Dumping + + + + + + + Enable Vulkan Validation Layers + + + + + + + Enable Vulkan Synchronization Validation + + + + + + + Enable RenderDoc Debugging + + + + + + + + + + + + + + Logger + + + + + + + 0 + + + 0 + + + 0 + + + 0 + + + + + Log Type + + + + + + + async + + + + + sync + + + + + + + + + + + + + + 6 + + + 0 + + + + + + + Log Filter + + + + + + + + + + + + + + + + diff --git a/src/qt_gui/translations/ar.ts b/src/qt_gui/translations/ar.ts index 1f65db04a..e851f59a7 100644 --- a/src/qt_gui/translations/ar.ts +++ b/src/qt_gui/translations/ar.ts @@ -537,6 +537,16 @@ Username اسم المستخدم + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName اسم المستخدم:\nيضبط اسم حساب PS4، الذي قد يتم عرضه في بعض الألعاب. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/da_DK.ts b/src/qt_gui/translations/da_DK.ts index 943e2d092..41319c7ff 100644 --- a/src/qt_gui/translations/da_DK.ts +++ b/src/qt_gui/translations/da_DK.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Brugernavn:\nIndstiller PS4-kontoens navn, som kan blive vist i nogle spil. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/de.ts b/src/qt_gui/translations/de.ts index cbbef8215..62897fe24 100644 --- a/src/qt_gui/translations/de.ts +++ b/src/qt_gui/translations/de.ts @@ -537,6 +537,16 @@ Username Benutzername + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Benutzername:\nLegt den Namen des PS4-Kontos fest, der in einigen Spielen angezeigt werden kann. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/el.ts b/src/qt_gui/translations/el.ts index 8737f5216..43ed81c33 100644 --- a/src/qt_gui/translations/el.ts +++ b/src/qt_gui/translations/el.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Όνομα Χρήστη:\nΟρίζει το όνομα του λογαριασμού PS4, το οποίο μπορεί να εμφανιστεί σε ορισμένα παιχνίδια. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/en.ts b/src/qt_gui/translations/en.ts index 692aa527e..293b5fae7 100644 --- a/src/qt_gui/translations/en.ts +++ b/src/qt_gui/translations/en.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Username:\nSets the PS4's account username, which may be displayed by some games. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/es_ES.ts b/src/qt_gui/translations/es_ES.ts index 70be2253d..096e104e3 100644 --- a/src/qt_gui/translations/es_ES.ts +++ b/src/qt_gui/translations/es_ES.ts @@ -537,6 +537,16 @@ Username Nombre de usuario + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Nombre de Usuario:\nEstablece el nombre de usuario de la cuenta de PS4, que puede ser mostrado por algunos juegos. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/fa_IR.ts b/src/qt_gui/translations/fa_IR.ts index 54187cf9b..7b93c6769 100644 --- a/src/qt_gui/translations/fa_IR.ts +++ b/src/qt_gui/translations/fa_IR.ts @@ -537,6 +537,16 @@ Username نام کاربری + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName نام کاربری:\nنام کاربری حساب PS4 را تنظیم می‌کند که ممکن است توسط برخی بازی‌ها نمایش داده شود. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/fi.ts b/src/qt_gui/translations/fi.ts index bdc1eb703..cdf331796 100644 --- a/src/qt_gui/translations/fi.ts +++ b/src/qt_gui/translations/fi.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Käyttäjänimi:\nAsettaa PS4-tilin käyttäjänimen, joka voi näkyä joissakin peleissä. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/fr.ts b/src/qt_gui/translations/fr.ts index 19b0f9358..441eaddb1 100644 --- a/src/qt_gui/translations/fr.ts +++ b/src/qt_gui/translations/fr.ts @@ -537,6 +537,16 @@ Username Nom d'utilisateur + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Nom d'utilisateur:\nDéfinit le nom d'utilisateur du compte PS4, qui peut être affiché par certains jeux. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/hu_HU.ts b/src/qt_gui/translations/hu_HU.ts index bc337f2cd..f6b853e4b 100644 --- a/src/qt_gui/translations/hu_HU.ts +++ b/src/qt_gui/translations/hu_HU.ts @@ -537,6 +537,16 @@ Username Felhasználónév + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Felhasználónév:\nBeállítja a PS4 fiók felhasználónevét, amelyet egyes játékok megjeleníthetnek. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/id.ts b/src/qt_gui/translations/id.ts index 7a0bf5d05..bee61083c 100644 --- a/src/qt_gui/translations/id.ts +++ b/src/qt_gui/translations/id.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Nama Pengguna:\nMenetapkan nama pengguna akun PS4, yang mungkin ditampilkan oleh beberapa permainan. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/it.ts b/src/qt_gui/translations/it.ts index 1391fbc55..9e375a45e 100644 --- a/src/qt_gui/translations/it.ts +++ b/src/qt_gui/translations/it.ts @@ -537,6 +537,16 @@ Username Nome Utente + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Nome Utente:\nImposta il nome utente dell'account PS4, che potrebbe essere visualizzato da alcuni giochi. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/ja_JP.ts b/src/qt_gui/translations/ja_JP.ts index 58f213e03..409900ade 100644 --- a/src/qt_gui/translations/ja_JP.ts +++ b/src/qt_gui/translations/ja_JP.ts @@ -537,6 +537,16 @@ Username ユーザー名 + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName ユーザー名:\nPS4のアカウントユーザー名を設定します。これは、一部のゲームで表示される場合があります。 + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/ko_KR.ts b/src/qt_gui/translations/ko_KR.ts index 75a1b53cf..ab6404a7e 100644 --- a/src/qt_gui/translations/ko_KR.ts +++ b/src/qt_gui/translations/ko_KR.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Username:\nSets the PS4's account username, which may be displayed by some games. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/lt_LT.ts b/src/qt_gui/translations/lt_LT.ts index 092521fdf..0b9c5b542 100644 --- a/src/qt_gui/translations/lt_LT.ts +++ b/src/qt_gui/translations/lt_LT.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Vartotojo vardas:\nNustato PS4 paskyros vartotojo vardą, kuris gali būti rodomas kai kuriuose žaidimuose. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/nb.ts b/src/qt_gui/translations/nb.ts index cc41573db..4d3c4f5af 100644 --- a/src/qt_gui/translations/nb.ts +++ b/src/qt_gui/translations/nb.ts @@ -537,6 +537,16 @@ Username Brukernavn + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Brukernavn:\nAngir brukernavnet for PS4-kontoen, som kan vises av enkelte spill. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/nl.ts b/src/qt_gui/translations/nl.ts index 5cd4a4224..0cb890186 100644 --- a/src/qt_gui/translations/nl.ts +++ b/src/qt_gui/translations/nl.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Gebruikersnaam:\nStelt de gebruikersnaam van het PS4-account in, die door sommige games kan worden weergegeven. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/pl_PL.ts b/src/qt_gui/translations/pl_PL.ts index b85393bb0..1aed08394 100644 --- a/src/qt_gui/translations/pl_PL.ts +++ b/src/qt_gui/translations/pl_PL.ts @@ -537,6 +537,16 @@ Username Nazwa użytkownika + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Nazwa użytkownika:\nUstala nazwę użytkownika konta PS4, która może być wyświetlana w niektórych grach. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/pt_BR.ts b/src/qt_gui/translations/pt_BR.ts index 8ab8db093..cce66c105 100644 --- a/src/qt_gui/translations/pt_BR.ts +++ b/src/qt_gui/translations/pt_BR.ts @@ -537,6 +537,16 @@ Username Nome de usuário + + + Trophy Key + Trophy Key + + + + Trophy + Troféus + Logger @@ -1236,6 +1246,11 @@ userName Nome de usuário:\nDefine o nome de usuário da conta PS4 que pode ser exibido por alguns jogos. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/ro_RO.ts b/src/qt_gui/translations/ro_RO.ts index 00547d6ba..63df2ff80 100644 --- a/src/qt_gui/translations/ro_RO.ts +++ b/src/qt_gui/translations/ro_RO.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Nume utilizator:\nSetează numele de utilizator al contului PS4, care poate fi afișat de unele jocuri. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/ru_RU.ts b/src/qt_gui/translations/ru_RU.ts index 505a05a3e..88eff1aeb 100644 --- a/src/qt_gui/translations/ru_RU.ts +++ b/src/qt_gui/translations/ru_RU.ts @@ -537,6 +537,16 @@ Username Имя пользователя + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Имя пользователя:\nУстановите имя пользователя аккаунта PS4. Это может отображаться в некоторых играх. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/sq.ts b/src/qt_gui/translations/sq.ts index 0c318f4f7..1df2a40e2 100644 --- a/src/qt_gui/translations/sq.ts +++ b/src/qt_gui/translations/sq.ts @@ -537,6 +537,16 @@ Username Përdoruesi + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Përdoruesi:\nPërcakton emrin e përdoruesit të llogarisë PS4, i cili mund të shfaqet nga disa lojra. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/tr_TR.ts b/src/qt_gui/translations/tr_TR.ts index 2845af462..a03a48660 100644 --- a/src/qt_gui/translations/tr_TR.ts +++ b/src/qt_gui/translations/tr_TR.ts @@ -537,6 +537,16 @@ Username Kullanıcı Adı + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Kullanıcı Adı:\nBazı oyunlar tarafından gösterilebilen PS4 hesabının kullanıcı adını ayarlar. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/uk_UA.ts b/src/qt_gui/translations/uk_UA.ts index 8abfca435..7e0a58ffb 100644 --- a/src/qt_gui/translations/uk_UA.ts +++ b/src/qt_gui/translations/uk_UA.ts @@ -537,6 +537,16 @@ Username Ім'я користувача + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Ім'я користувача:\nВстановіть ім'я користувача акаунта PS4. Це може відображатися в деяких іграх. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/vi_VN.ts b/src/qt_gui/translations/vi_VN.ts index 7d0e9a2cd..997c3d3f9 100644 --- a/src/qt_gui/translations/vi_VN.ts +++ b/src/qt_gui/translations/vi_VN.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName Tên người dùng:\nChọn tên người dùng của tài khoản PS4, có thể được một số trò chơi hiển thị. + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/zh_CN.ts b/src/qt_gui/translations/zh_CN.ts index 32b838fac..fecb8857f 100644 --- a/src/qt_gui/translations/zh_CN.ts +++ b/src/qt_gui/translations/zh_CN.ts @@ -537,6 +537,16 @@ Username 用户名 + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName 用户名：\n设置 PS4 帐户的用户名，某些游戏中可能会显示此名称。 + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/qt_gui/translations/zh_TW.ts b/src/qt_gui/translations/zh_TW.ts index 3d27267b6..293ed81a6 100644 --- a/src/qt_gui/translations/zh_TW.ts +++ b/src/qt_gui/translations/zh_TW.ts @@ -537,6 +537,16 @@ Username Username + + + Trophy Key + Trophy Key + + + + Trophy + Trophy + Logger @@ -1236,6 +1246,11 @@ userName 用戶名:\n設定PS4帳號的用戶名，某些遊戲中可能會顯示。 + + + TrophyKey + Trophy Key:\nKey used to decrypt trophies. Must be obtained from your jailbroken console.\nMust contain only hex characters. + logTypeGroupBox diff --git a/src/sdl_window.cpp b/src/sdl_window.cpp index 50c3e93ee..d694b0939 100644 --- a/src/sdl_window.cpp +++ b/src/sdl_window.cpp @@ -161,6 +161,20 @@ void WindowSDL::WaitEvent() { case SDL_EVENT_GAMEPAD_TOUCHPAD_MOTION: OnGamepadEvent(&event); break; + // i really would have appreciated ANY KIND OF DOCUMENTATION ON THIS + // AND IT DOESN'T EVEN USE PROPER ENUMS + case SDL_EVENT_GAMEPAD_SENSOR_UPDATE: + switch ((SDL_SensorType)event.gsensor.sensor) { + case SDL_SENSOR_GYRO: + controller->Gyro(0, event.gsensor.data); + break; + case SDL_SENSOR_ACCEL: + controller->Acceleration(0, event.gsensor.data); + break; + default: + break; + } + break; case SDL_EVENT_QUIT: is_open = false; break; diff --git a/src/shader_recompiler/frontend/translate/data_share.cpp b/src/shader_recompiler/frontend/translate/data_share.cpp index 4408cae28..62c0423dd 100644 --- a/src/shader_recompiler/frontend/translate/data_share.cpp +++ b/src/shader_recompiler/frontend/translate/data_share.cpp @@ -205,7 +205,6 @@ void Translator::DS_WRITE(int bit_size, bool is_signed, bool is_pair, bool strid addr, ir.Imm32((u32(inst.control.ds.offset1) << 8u) + u32(inst.control.ds.offset0))); ir.WriteShared(bit_size, ir.GetVectorReg(data0), addr0); } - emit_ds_read_barrier = true; } void Translator::DS_SWIZZLE_B32(const GcnInst& inst) { @@ -222,11 +221,6 @@ void Translator::DS_SWIZZLE_B32(const GcnInst& inst) { void Translator::DS_READ(int bit_size, bool is_signed, bool is_pair, bool stride64, const GcnInst& inst) { - if (emit_ds_read_barrier && profile.needs_lds_barriers) { - ir.Barrier(); - emit_ds_read_barrier = false; - } - const IR::U32 addr{ir.GetVectorReg(IR::VectorReg(inst.src[0].code))}; IR::VectorReg dst_reg{inst.dst[0].code}; if (is_pair) { diff --git a/src/shader_recompiler/frontend/translate/translate.h b/src/shader_recompiler/frontend/translate/translate.h index e8584ec2f..9da0844e4 100644 --- a/src/shader_recompiler/frontend/translate/translate.h +++ b/src/shader_recompiler/frontend/translate/translate.h @@ -308,7 +308,6 @@ private: const RuntimeInfo& runtime_info; const Profile& profile; bool opcode_missing = false; - bool emit_ds_read_barrier = false; }; void Translate(IR::Block* block, u32 block_base, std::span inst_list, Info& info, diff --git a/src/shader_recompiler/ir/passes/shared_memory_barrier_pass.cpp b/src/shader_recompiler/ir/passes/shared_memory_barrier_pass.cpp index aad8fb148..ec7d7e986 100644 --- a/src/shader_recompiler/ir/passes/shared_memory_barrier_pass.cpp +++ b/src/shader_recompiler/ir/passes/shared_memory_barrier_pass.cpp @@ -8,6 +8,54 @@ namespace Shader::Optimization { +static void EmitBarrierInBlock(IR::Block* block) { + // This is inteded to insert a barrier when shared memory write and read + // occur in the same basic block. Also checks if branch depth is zero as + // we don't want to insert barrier in potentially divergent code. + bool emit_barrier_on_write = false; + bool emit_barrier_on_read = false; + const auto emit_barrier = [block](bool& emit_cond, IR::Inst& inst) { + if (emit_cond) { + IR::IREmitter ir{*block, IR::Block::InstructionList::s_iterator_to(inst)}; + ir.Barrier(); + emit_cond = false; + } + }; + for (IR::Inst& inst : block->Instructions()) { + if (inst.GetOpcode() == IR::Opcode::LoadSharedU32 || + inst.GetOpcode() == IR::Opcode::LoadSharedU64) { + emit_barrier(emit_barrier_on_read, inst); + emit_barrier_on_write = true; + } + if (inst.GetOpcode() == IR::Opcode::WriteSharedU32 || + inst.GetOpcode() == IR::Opcode::WriteSharedU64) { + emit_barrier(emit_barrier_on_write, inst); + emit_barrier_on_read = true; + } + } +} + +static void EmitBarrierInMergeBlock(const IR::AbstractSyntaxNode::Data& data) { + // Insert a barrier after divergent conditional blocks. + // This avoids potential softlocks and crashes when some threads + // initialize shared memory and others read from it. + const IR::U1 cond = data.if_node.cond; + const auto insert_barrier = + IR::BreadthFirstSearch(cond, [](IR::Inst* inst) -> std::optional { + if (inst->GetOpcode() == IR::Opcode::GetAttributeU32 && + inst->Arg(0).Attribute() == IR::Attribute::LocalInvocationId) { + return true; + } + return std::nullopt; + }); + if (insert_barrier) { + IR::Block* const merge = data.if_node.merge; + auto insert_point = std::ranges::find_if_not(merge->Instructions(), IR::IsPhi); + IR::IREmitter ir{*merge, insert_point}; + ir.Barrier(); + } +} + void SharedMemoryBarrierPass(IR::Program& program, const Profile& profile) { if (!program.info.uses_shared || !profile.needs_lds_barriers) { return; @@ -19,27 +67,12 @@ void SharedMemoryBarrierPass(IR::Program& program, const Profile& profile) { --branch_depth; continue; } - if (node.type != Type::If) { + if (node.type == Type::If && branch_depth++ == 0) { + EmitBarrierInMergeBlock(node.data); continue; } - u32 curr_depth = branch_depth++; - if (curr_depth != 0) { - continue; - } - const IR::U1 cond = node.data.if_node.cond; - const auto insert_barrier = - IR::BreadthFirstSearch(cond, [](IR::Inst* inst) -> std::optional { - if (inst->GetOpcode() == IR::Opcode::GetAttributeU32 && - inst->Arg(0).Attribute() == IR::Attribute::LocalInvocationId) { - return true; - } - return std::nullopt; - }); - if (insert_barrier) { - IR::Block* const merge = node.data.if_node.merge; - auto insert_point = std::ranges::find_if_not(merge->Instructions(), IR::IsPhi); - IR::IREmitter ir{*merge, insert_point}; - ir.Barrier(); + if (node.type == Type::Block && branch_depth == 0) { + EmitBarrierInBlock(node.data.block); } } } diff --git a/src/video_core/amdgpu/liverpool.cpp b/src/video_core/amdgpu/liverpool.cpp index 2926bcc69..985f3c652 100644 --- a/src/video_core/amdgpu/liverpool.cpp +++ b/src/video_core/amdgpu/liverpool.cpp @@ -815,7 +815,8 @@ Liverpool::Task Liverpool::ProcessCompute(std::span acb, u32 vqid) { } if (rasterizer && (cs_program.dispatch_initiator & 1)) { const auto cmd_address = reinterpret_cast(header); - rasterizer->ScopeMarkerBegin(fmt::format("acb[{}]:{}:Dispatch", vqid, cmd_address)); + rasterizer->ScopeMarkerBegin( + fmt::format("acb[{}]:{}:DispatchIndirect", vqid, cmd_address)); rasterizer->DispatchDirect(); rasterizer->ScopeMarkerEnd(); } diff --git a/src/video_core/amdgpu/liverpool.h b/src/video_core/amdgpu/liverpool.h index f1607f03e..d2d1aab3c 100644 --- a/src/video_core/amdgpu/liverpool.h +++ b/src/video_core/amdgpu/liverpool.h @@ -889,11 +889,11 @@ struct Liverpool { return !info.linear_general; } - [[nodiscard]] DataFormat DataFormat() const { + [[nodiscard]] DataFormat GetDataFmt() const { return RemapDataFormat(info.format); } - [[nodiscard]] NumberFormat NumFormat() const { + [[nodiscard]] NumberFormat GetNumberFmt() const { // There is a small difference between T# and CB number types, account for it. return RemapNumberFormat(info.number_type == NumberFormat::SnormNz ? NumberFormat::Srgb diff --git a/src/video_core/amdgpu/resource.h b/src/video_core/amdgpu/resource.h index 4de25adbf..208f7f380 100644 --- a/src/video_core/amdgpu/resource.h +++ b/src/video_core/amdgpu/resource.h @@ -79,21 +79,23 @@ inline NumberFormat RemapNumberFormat(const NumberFormat format) { inline CompMapping RemapComponents(const DataFormat format, const CompMapping components) { switch (format) { - case DataFormat::Format11_11_10: - return { - .r = components.b, - .g = components.g, - .b = components.r, - .a = components.a, - }; + case DataFormat::Format11_11_10: { + CompMapping result; + result.r = components.b; + result.g = components.g; + result.b = components.r; + result.a = components.a; + return result; + } case DataFormat::Format10_10_10_2: - case DataFormat::Format5_5_5_1: - return { - .r = components.a, - .g = components.b, - .b = components.g, - .a = components.r, - }; + case DataFormat::Format5_5_5_1: { + CompMapping result; + result.r = components.a; + result.g = components.b; + result.b = components.g; + result.a = components.r; + return result; + } default: return components; } diff --git a/src/video_core/buffer_cache/buffer_cache.cpp b/src/video_core/buffer_cache/buffer_cache.cpp index 0088ea4fa..3e43b4fbc 100644 --- a/src/video_core/buffer_cache/buffer_cache.cpp +++ b/src/video_core/buffer_cache/buffer_cache.cpp @@ -54,18 +54,10 @@ BufferCache::BufferCache(const Vulkan::Instance& instance_, Vulkan::Scheduler& s BufferCache::~BufferCache() = default; void BufferCache::InvalidateMemory(VAddr device_addr, u64 size) { - std::scoped_lock lk{mutex}; const bool is_tracked = IsRegionRegistered(device_addr, size); - if (!is_tracked) { - return; - } - // Mark the page as CPU modified to stop tracking writes. - SCOPE_EXIT { + if (is_tracked) { + // Mark the page as CPU modified to stop tracking writes. memory_tracker.MarkRegionAsCpuModified(device_addr, size); - }; - if (!memory_tracker.IsRegionGpuModified(device_addr, size)) { - // Page has not been modified by the GPU, nothing to do. - return; } } @@ -267,7 +259,16 @@ void BufferCache::InlineData(VAddr address, const void* value, u32 num_bytes, bo const BufferId buffer_id = FindBuffer(address, num_bytes); return &slot_buffers[buffer_id]; }(); - const vk::BufferMemoryBarrier2 buf_barrier = { + 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, @@ -279,9 +280,14 @@ void BufferCache::InlineData(VAddr address, const void* value, u32 num_bytes, bo cmdbuf.pipelineBarrier2(vk::DependencyInfo{ .dependencyFlags = vk::DependencyFlagBits::eByRegion, .bufferMemoryBarrierCount = 1, - .pBufferMemoryBarriers = &buf_barrier, + .pBufferMemoryBarriers = &pre_barrier, + }); + cmdbuf.updateBuffer(buffer->Handle(), buffer->Offset(address), num_bytes, value); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &post_barrier, }); - cmdbuf.updateBuffer(buffer->Handle(), buf_barrier.offset, num_bytes, value); } std::pair BufferCache::ObtainHostUBO(std::span data) { @@ -346,6 +352,7 @@ bool BufferCache::IsRegionRegistered(VAddr addr, size_t size) { ++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(); @@ -472,21 +479,48 @@ void BufferCache::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id, }; scheduler.EndRendering(); const auto cmdbuf = scheduler.CommandBuffer(); - static constexpr vk::MemoryBarrier READ_BARRIER{ - .srcAccessMask = vk::AccessFlagBits::eMemoryWrite, - .dstAccessMask = vk::AccessFlagBits::eTransferRead | vk::AccessFlagBits::eTransferWrite, + const std::array pre_barriers = { + vk::BufferMemoryBarrier2{ + .srcStageMask = vk::PipelineStageFlagBits2::eAllCommands, + .srcAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite, + .dstStageMask = vk::PipelineStageFlagBits2::eTransfer, + .dstAccessMask = vk::AccessFlagBits2::eTransferRead, + .buffer = overlap.Handle(), + .offset = 0, + .size = overlap.SizeBytes(), + }, }; - static constexpr vk::MemoryBarrier WRITE_BARRIER{ - .srcAccessMask = vk::AccessFlagBits::eTransferWrite, - .dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite, + const std::array post_barriers = { + vk::BufferMemoryBarrier2{ + .srcStageMask = vk::PipelineStageFlagBits2::eTransfer, + .srcAccessMask = vk::AccessFlagBits2::eTransferRead, + .dstStageMask = vk::PipelineStageFlagBits2::eAllCommands, + .dstAccessMask = vk::AccessFlagBits2::eMemoryWrite, + .buffer = overlap.Handle(), + .offset = 0, + .size = overlap.SizeBytes(), + }, + vk::BufferMemoryBarrier2{ + .srcStageMask = vk::PipelineStageFlagBits2::eTransfer, + .srcAccessMask = vk::AccessFlagBits2::eTransferWrite, + .dstStageMask = vk::PipelineStageFlagBits2::eAllCommands, + .dstAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite, + .buffer = new_buffer.Handle(), + .offset = dst_base_offset, + .size = overlap.SizeBytes(), + }, }; - cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands, - vk::PipelineStageFlagBits::eTransfer, vk::DependencyFlagBits::eByRegion, - READ_BARRIER, {}, {}); - cmdbuf.copyBuffer(overlap.buffer, new_buffer.buffer, copy); - cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, - vk::PipelineStageFlagBits::eAllCommands, - vk::DependencyFlagBits::eByRegion, WRITE_BARRIER, {}, {}); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = pre_barriers.data(), + }); + cmdbuf.copyBuffer(overlap.Handle(), new_buffer.Handle(), copy); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = static_cast(post_barriers.size()), + .pBufferMemoryBarriers = post_barriers.data(), + }); DeleteBuffer(overlap_id); } @@ -496,8 +530,11 @@ BufferId BufferCache::CreateBuffer(VAddr device_addr, u32 wanted_size) { wanted_size = static_cast(device_addr_end - device_addr); const OverlapResult overlap = ResolveOverlaps(device_addr, wanted_size); const u32 size = static_cast(overlap.end - overlap.begin); - const BufferId new_buffer_id = slot_buffers.insert( - instance, scheduler, MemoryUsage::DeviceLocal, overlap.begin, AllFlags, size); + const BufferId new_buffer_id = [&] { + std::scoped_lock lk{mutex}; + return slot_buffers.insert(instance, scheduler, MemoryUsage::DeviceLocal, overlap.begin, + AllFlags, size); + }(); auto& new_buffer = slot_buffers[new_buffer_id]; const size_t size_bytes = new_buffer.SizeBytes(); const auto cmdbuf = scheduler.CommandBuffer(); @@ -537,10 +574,8 @@ void BufferCache::ChangeRegister(BufferId buffer_id) { void BufferCache::SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size, bool is_texel_buffer) { - std::scoped_lock lk{mutex}; boost::container::small_vector copies; u64 total_size_bytes = 0; - u64 largest_copy = 0; VAddr buffer_start = buffer.CpuAddr(); memory_tracker.ForEachUploadRange(device_addr, size, [&](u64 device_addr_out, u64 range_size) { copies.push_back(vk::BufferCopy{ @@ -549,7 +584,6 @@ void BufferCache::SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size, .size = range_size, }); total_size_bytes += range_size; - largest_copy = std::max(largest_copy, range_size); }); SCOPE_EXIT { if (is_texel_buffer) { @@ -590,21 +624,35 @@ void BufferCache::SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size, } scheduler.EndRendering(); const auto cmdbuf = scheduler.CommandBuffer(); - static constexpr vk::MemoryBarrier READ_BARRIER{ - .srcAccessMask = vk::AccessFlagBits::eMemoryWrite, - .dstAccessMask = vk::AccessFlagBits::eTransferRead | vk::AccessFlagBits::eTransferWrite, + 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 = 0, + .size = buffer.SizeBytes(), }; - static constexpr vk::MemoryBarrier WRITE_BARRIER{ - .srcAccessMask = vk::AccessFlagBits::eTransferWrite, - .dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite, + 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 = 0, + .size = buffer.SizeBytes(), }; - cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands, - vk::PipelineStageFlagBits::eTransfer, vk::DependencyFlagBits::eByRegion, - READ_BARRIER, {}, {}); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &pre_barrier, + }); cmdbuf.copyBuffer(src_buffer, buffer.buffer, copies); - cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer, - vk::PipelineStageFlagBits::eAllCommands, - vk::DependencyFlagBits::eByRegion, WRITE_BARRIER, {}, {}); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &post_barrier, + }); } bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr, u32 size) { @@ -654,10 +702,42 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr, } if (!copies.empty()) { scheduler.EndRendering(); - image.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead, {}); + 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 = max_offset - size, + .size = size, + }; + 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 = max_offset - size, + .size = size, + }; + auto barriers = image.GetBarriers(vk::ImageLayout::eTransferSrcOptimal, + vk::AccessFlagBits2::eTransferRead, + vk::PipelineStageFlagBits2::eTransfer, {}); const auto cmdbuf = scheduler.CommandBuffer(); - cmdbuf.copyImageToBuffer(image.image, vk::ImageLayout::eTransferSrcOptimal, buffer.buffer, + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &pre_barrier, + .imageMemoryBarrierCount = static_cast(barriers.size()), + .pImageMemoryBarriers = barriers.data(), + }); + cmdbuf.copyImageToBuffer(image.image, vk::ImageLayout::eTransferSrcOptimal, buffer.Handle(), copies); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &post_barrier, + }); } return true; } diff --git a/src/video_core/buffer_cache/buffer_cache.h b/src/video_core/buffer_cache/buffer_cache.h index 0c70fa10b..c367795f1 100644 --- a/src/video_core/buffer_cache/buffer_cache.h +++ b/src/video_core/buffer_cache/buffer_cache.h @@ -3,7 +3,7 @@ #pragma once -#include +#include #include #include #include @@ -157,7 +157,7 @@ private: StreamBuffer staging_buffer; StreamBuffer stream_buffer; Buffer gds_buffer; - std::mutex mutex; + std::shared_mutex mutex; Common::SlotVector slot_buffers; RangeSet gpu_modified_ranges; vk::BufferView null_buffer_view; diff --git a/src/video_core/buffer_cache/memory_tracker_base.h b/src/video_core/buffer_cache/memory_tracker_base.h index ae61b55f2..d9166b11c 100644 --- a/src/video_core/buffer_cache/memory_tracker_base.h +++ b/src/video_core/buffer_cache/memory_tracker_base.h @@ -15,13 +15,8 @@ namespace VideoCore { class MemoryTracker { public: static constexpr size_t MAX_CPU_PAGE_BITS = 40; - static constexpr size_t HIGHER_PAGE_BITS = 22; - static constexpr size_t HIGHER_PAGE_SIZE = 1ULL << HIGHER_PAGE_BITS; - static constexpr size_t HIGHER_PAGE_MASK = HIGHER_PAGE_SIZE - 1ULL; static constexpr size_t NUM_HIGH_PAGES = 1ULL << (MAX_CPU_PAGE_BITS - HIGHER_PAGE_BITS); static constexpr size_t MANAGER_POOL_SIZE = 32; - static constexpr size_t WORDS_STACK_NEEDED = HIGHER_PAGE_SIZE / BYTES_PER_WORD; - using Manager = WordManager; public: explicit MemoryTracker(PageManager* tracker_) : tracker{tracker_} {} @@ -30,7 +25,7 @@ public: /// Returns true if a region has been modified from the CPU [[nodiscard]] bool IsRegionCpuModified(VAddr query_cpu_addr, u64 query_size) noexcept { return IteratePages( - query_cpu_addr, query_size, [](Manager* manager, u64 offset, size_t size) { + query_cpu_addr, query_size, [](RegionManager* manager, u64 offset, size_t size) { return manager->template IsRegionModified(offset, size); }); } @@ -38,52 +33,34 @@ public: /// Returns true if a region has been modified from the GPU [[nodiscard]] bool IsRegionGpuModified(VAddr query_cpu_addr, u64 query_size) noexcept { return IteratePages( - query_cpu_addr, query_size, [](Manager* manager, u64 offset, size_t size) { + query_cpu_addr, query_size, [](RegionManager* manager, u64 offset, size_t size) { return manager->template IsRegionModified(offset, size); }); } /// Mark region as CPU modified, notifying the device_tracker about this change void MarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 query_size) { - IteratePages(dirty_cpu_addr, query_size, - [](Manager* manager, u64 offset, size_t size) { - manager->template ChangeRegionState( - manager->GetCpuAddr() + offset, size); - }); - } - - /// Unmark region as CPU modified, notifying the device_tracker about this change - void UnmarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 query_size) { - IteratePages(dirty_cpu_addr, query_size, - [](Manager* manager, u64 offset, size_t size) { - manager->template ChangeRegionState( - manager->GetCpuAddr() + offset, size); - }); + IteratePages(dirty_cpu_addr, query_size, + [](RegionManager* manager, u64 offset, size_t size) { + manager->template ChangeRegionState( + manager->GetCpuAddr() + offset, size); + }); } /// Mark region as modified from the host GPU void MarkRegionAsGpuModified(VAddr dirty_cpu_addr, u64 query_size) noexcept { - IteratePages(dirty_cpu_addr, query_size, - [](Manager* manager, u64 offset, size_t size) { - manager->template ChangeRegionState( - manager->GetCpuAddr() + offset, size); - }); - } - - /// Unmark region as modified from the host GPU - void UnmarkRegionAsGpuModified(VAddr dirty_cpu_addr, u64 query_size) noexcept { - IteratePages(dirty_cpu_addr, query_size, - [](Manager* manager, u64 offset, size_t size) { - manager->template ChangeRegionState( - manager->GetCpuAddr() + offset, size); - }); + IteratePages(dirty_cpu_addr, query_size, + [](RegionManager* manager, u64 offset, size_t size) { + manager->template ChangeRegionState( + manager->GetCpuAddr() + offset, size); + }); } /// 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) { IteratePages(query_cpu_range, query_size, - [&func](Manager* manager, u64 offset, size_t size) { + [&func](RegionManager* manager, u64 offset, size_t size) { manager->template ForEachModifiedRange( manager->GetCpuAddr() + offset, size, func); }); @@ -93,7 +70,7 @@ public: template void ForEachDownloadRange(VAddr query_cpu_range, u64 query_size, Func&& func) { IteratePages(query_cpu_range, query_size, - [&func](Manager* manager, u64 offset, size_t size) { + [&func](RegionManager* manager, u64 offset, size_t size) { if constexpr (clear) { manager->template ForEachModifiedRange( manager->GetCpuAddr() + offset, size, func); @@ -114,7 +91,7 @@ private: */ template bool IteratePages(VAddr cpu_address, size_t size, Func&& func) { - using FuncReturn = typename std::invoke_result::type; + using FuncReturn = typename std::invoke_result::type; static constexpr bool BOOL_BREAK = std::is_same_v; std::size_t remaining_size{size}; std::size_t page_index{cpu_address >> HIGHER_PAGE_BITS}; @@ -155,7 +132,7 @@ private: manager_pool.emplace_back(); auto& last_pool = manager_pool.back(); for (size_t i = 0; i < MANAGER_POOL_SIZE; i++) { - std::construct_at(&last_pool[i], tracker, 0, HIGHER_PAGE_SIZE); + std::construct_at(&last_pool[i], tracker, 0); free_managers.push_back(&last_pool[i]); } } @@ -167,9 +144,9 @@ private: } PageManager* tracker; - std::deque> manager_pool; - std::vector free_managers; - std::array top_tier{}; + std::deque> manager_pool; + std::vector free_managers; + std::array top_tier{}; }; } // namespace VideoCore diff --git a/src/video_core/buffer_cache/word_manager.h b/src/video_core/buffer_cache/word_manager.h index ae85d1eb1..7ad33d7a6 100644 --- a/src/video_core/buffer_cache/word_manager.h +++ b/src/video_core/buffer_cache/word_manager.h @@ -3,10 +3,12 @@ #pragma once -#include +#include +#include #include #include -#include "common/div_ceil.h" + +#include "common/spin_lock.h" #include "common/types.h" #include "video_core/page_manager.h" @@ -16,135 +18,32 @@ constexpr u64 PAGES_PER_WORD = 64; constexpr u64 BYTES_PER_PAGE = 4_KB; constexpr u64 BYTES_PER_WORD = PAGES_PER_WORD * BYTES_PER_PAGE; +constexpr u64 HIGHER_PAGE_BITS = 22; +constexpr u64 HIGHER_PAGE_SIZE = 1ULL << HIGHER_PAGE_BITS; +constexpr u64 HIGHER_PAGE_MASK = HIGHER_PAGE_SIZE - 1ULL; +constexpr u64 NUM_REGION_WORDS = HIGHER_PAGE_SIZE / BYTES_PER_WORD; + enum class Type { CPU, GPU, Untracked, }; -/// Vector tracking modified pages tightly packed with small vector optimization -template -struct WordsArray { - /// Returns the pointer to the words state - [[nodiscard]] const u64* Pointer(bool is_short) const noexcept { - return is_short ? stack.data() : heap; - } +using WordsArray = std::array; - /// Returns the pointer to the words state - [[nodiscard]] u64* Pointer(bool is_short) noexcept { - return is_short ? stack.data() : heap; - } - - std::array stack{}; ///< Small buffers storage - u64* heap; ///< Not-small buffers pointer to the storage -}; - -template -struct Words { - explicit Words() = default; - explicit Words(u64 size_bytes_) : size_bytes{size_bytes_} { - num_words = Common::DivCeil(size_bytes, BYTES_PER_WORD); - if (IsShort()) { - cpu.stack.fill(~u64{0}); - gpu.stack.fill(0); - untracked.stack.fill(~u64{0}); - } else { - // Share allocation between CPU and GPU pages and set their default values - u64* const alloc = new u64[num_words * 3]; - cpu.heap = alloc; - gpu.heap = alloc + num_words; - untracked.heap = alloc + num_words * 2; - std::fill_n(cpu.heap, num_words, ~u64{0}); - std::fill_n(gpu.heap, num_words, 0); - std::fill_n(untracked.heap, num_words, ~u64{0}); - } - // Clean up tailing bits - const u64 last_word_size = size_bytes % BYTES_PER_WORD; - const u64 last_local_page = Common::DivCeil(last_word_size, BYTES_PER_PAGE); - const u64 shift = (PAGES_PER_WORD - last_local_page) % PAGES_PER_WORD; - const u64 last_word = (~u64{0} << shift) >> shift; - cpu.Pointer(IsShort())[NumWords() - 1] = last_word; - untracked.Pointer(IsShort())[NumWords() - 1] = last_word; - } - - ~Words() { - Release(); - } - - Words& operator=(Words&& rhs) noexcept { - Release(); - size_bytes = rhs.size_bytes; - num_words = rhs.num_words; - cpu = rhs.cpu; - gpu = rhs.gpu; - untracked = rhs.untracked; - rhs.cpu.heap = nullptr; - return *this; - } - - Words(Words&& rhs) noexcept - : size_bytes{rhs.size_bytes}, num_words{rhs.num_words}, cpu{rhs.cpu}, gpu{rhs.gpu}, - untracked{rhs.untracked} { - rhs.cpu.heap = nullptr; - } - - Words& operator=(const Words&) = delete; - Words(const Words&) = delete; - - /// Returns true when the buffer fits in the small vector optimization - [[nodiscard]] bool IsShort() const noexcept { - return num_words <= stack_words; - } - - /// Returns the number of words of the buffer - [[nodiscard]] size_t NumWords() const noexcept { - return num_words; - } - - /// Release buffer resources - void Release() { - if (!IsShort()) { - // CPU written words is the base for the heap allocation - delete[] cpu.heap; - } - } - - template - std::span Span() noexcept { - if constexpr (type == Type::CPU) { - return std::span(cpu.Pointer(IsShort()), num_words); - } else if constexpr (type == Type::GPU) { - return std::span(gpu.Pointer(IsShort()), num_words); - } else if constexpr (type == Type::Untracked) { - return std::span(untracked.Pointer(IsShort()), num_words); - } - } - - template - std::span Span() const noexcept { - if constexpr (type == Type::CPU) { - return std::span(cpu.Pointer(IsShort()), num_words); - } else if constexpr (type == Type::GPU) { - return std::span(gpu.Pointer(IsShort()), num_words); - } else if constexpr (type == Type::Untracked) { - return std::span(untracked.Pointer(IsShort()), num_words); - } - } - - u64 size_bytes = 0; - size_t num_words = 0; - WordsArray cpu; - WordsArray gpu; - WordsArray untracked; -}; - -template -class WordManager { +/** + * Allows tracking CPU and GPU modification of pages in a contigious 4MB virtual address region. + * Information is stored in bitsets for spacial locality and fast update of single pages. + */ +class RegionManager { public: - explicit WordManager(PageManager* tracker_, VAddr cpu_addr_, u64 size_bytes) - : tracker{tracker_}, cpu_addr{cpu_addr_}, words{size_bytes} {} - - explicit WordManager() = default; + explicit RegionManager(PageManager* tracker_, VAddr cpu_addr_) + : tracker{tracker_}, cpu_addr{cpu_addr_} { + cpu.fill(~u64{0}); + gpu.fill(0); + untracked.fill(~u64{0}); + } + explicit RegionManager() = default; void SetCpuAddress(VAddr new_cpu_addr) { cpu_addr = new_cpu_addr; @@ -175,12 +74,12 @@ public: static constexpr bool BOOL_BREAK = std::is_same_v; const size_t start = static_cast(std::max(static_cast(offset), 0LL)); const size_t end = static_cast(std::max(static_cast(offset + size), 0LL)); - if (start >= SizeBytes() || end <= start) { + if (start >= HIGHER_PAGE_SIZE || end <= start) { return; } auto [start_word, start_page] = GetWordPage(start); auto [end_word, end_page] = GetWordPage(end + BYTES_PER_PAGE - 1ULL); - const size_t num_words = NumWords(); + constexpr size_t num_words = NUM_REGION_WORDS; start_word = std::min(start_word, num_words); end_word = std::min(end_word, num_words); const size_t diff = end_word - start_word; @@ -225,21 +124,21 @@ public: */ template void ChangeRegionState(u64 dirty_addr, u64 size) noexcept(type == Type::GPU) { - std::span state_words = words.template Span(); - [[maybe_unused]] std::span untracked_words = words.template Span(); + std::scoped_lock lk{lock}; + std::span state_words = Span(); IterateWords(dirty_addr - cpu_addr, size, [&](size_t index, u64 mask) { if constexpr (type == Type::CPU) { - NotifyPageTracker(index, untracked_words[index], mask); + UpdateProtection(index, untracked[index], mask); } if constexpr (enable) { state_words[index] |= mask; if constexpr (type == Type::CPU) { - untracked_words[index] |= mask; + untracked[index] |= mask; } } else { state_words[index] &= ~mask; if constexpr (type == Type::CPU) { - untracked_words[index] &= ~mask; + untracked[index] &= ~mask; } } }); @@ -255,10 +154,10 @@ public: */ template void ForEachModifiedRange(VAddr query_cpu_range, s64 size, Func&& func) { + std::scoped_lock lk{lock}; static_assert(type != Type::Untracked); - std::span state_words = words.template Span(); - [[maybe_unused]] std::span untracked_words = words.template Span(); + std::span state_words = Span(); const size_t offset = query_cpu_range - cpu_addr; bool pending = false; size_t pending_offset{}; @@ -269,16 +168,16 @@ public: }; IterateWords(offset, size, [&](size_t index, u64 mask) { if constexpr (type == Type::GPU) { - mask &= ~untracked_words[index]; + mask &= ~untracked[index]; } const u64 word = state_words[index] & mask; if constexpr (clear) { if constexpr (type == Type::CPU) { - NotifyPageTracker(index, untracked_words[index], mask); + UpdateProtection(index, untracked[index], mask); } state_words[index] &= ~mask; if constexpr (type == Type::CPU) { - untracked_words[index] &= ~mask; + untracked[index] &= ~mask; } } const size_t base_offset = index * PAGES_PER_WORD; @@ -315,13 +214,11 @@ public: [[nodiscard]] bool IsRegionModified(u64 offset, u64 size) const noexcept { static_assert(type != Type::Untracked); - const std::span state_words = words.template Span(); - [[maybe_unused]] const std::span untracked_words = - words.template Span(); + const std::span state_words = Span(); bool result = false; IterateWords(offset, size, [&](size_t index, u64 mask) { if constexpr (type == Type::GPU) { - mask &= ~untracked_words[index]; + mask &= ~untracked[index]; } const u64 word = state_words[index] & mask; if (word != 0) { @@ -333,44 +230,7 @@ public: return result; } - /// Returns the number of words of the manager - [[nodiscard]] size_t NumWords() const noexcept { - return words.NumWords(); - } - - /// Returns the size in bytes of the manager - [[nodiscard]] u64 SizeBytes() const noexcept { - return words.size_bytes; - } - - /// Returns true when the buffer fits in the small vector optimization - [[nodiscard]] bool IsShort() const noexcept { - return words.IsShort(); - } - private: - template - u64* Array() noexcept { - if constexpr (type == Type::CPU) { - return words.cpu.Pointer(IsShort()); - } else if constexpr (type == Type::GPU) { - return words.gpu.Pointer(IsShort()); - } else if constexpr (type == Type::Untracked) { - return words.untracked.Pointer(IsShort()); - } - } - - template - const u64* Array() const noexcept { - if constexpr (type == Type::CPU) { - return words.cpu.Pointer(IsShort()); - } else if constexpr (type == Type::GPU) { - return words.gpu.Pointer(IsShort()); - } else if constexpr (type == Type::Untracked) { - return words.untracked.Pointer(IsShort()); - } - } - /** * Notify tracker about changes in the CPU tracking state of a word in the buffer * @@ -381,7 +241,7 @@ private: * @tparam add_to_tracker True when the tracker should start tracking the new pages */ template - void NotifyPageTracker(u64 word_index, u64 current_bits, u64 new_bits) const { + void UpdateProtection(u64 word_index, u64 current_bits, u64 new_bits) const { 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) { @@ -390,9 +250,34 @@ private: }); } + template + std::span Span() noexcept { + if constexpr (type == Type::CPU) { + return cpu; + } else if constexpr (type == Type::GPU) { + return gpu; + } else if constexpr (type == Type::Untracked) { + return untracked; + } + } + + template + std::span Span() const noexcept { + if constexpr (type == Type::CPU) { + return cpu; + } else if constexpr (type == Type::GPU) { + return gpu; + } else if constexpr (type == Type::Untracked) { + return untracked; + } + } + + Common::SpinLock lock; PageManager* tracker; VAddr cpu_addr = 0; - Words words; + WordsArray cpu; + WordsArray gpu; + WordsArray untracked; }; } // namespace VideoCore diff --git a/src/video_core/host_shaders/CMakeLists.txt b/src/video_core/host_shaders/CMakeLists.txt index c2a3b53fd..44761545d 100644 --- a/src/video_core/host_shaders/CMakeLists.txt +++ b/src/video_core/host_shaders/CMakeLists.txt @@ -7,6 +7,7 @@ set(SHADER_FILES detile_m32x1.comp detile_m32x2.comp detile_m32x4.comp + detile_macro8x1.comp detile_macro32x1.comp detile_macro32x2.comp fs_tri.vert diff --git a/src/video_core/host_shaders/detile_m32x1.comp b/src/video_core/host_shaders/detile_m32x1.comp index 802f5f531..cdc8d0018 100644 --- a/src/video_core/host_shaders/detile_m32x1.comp +++ b/src/video_core/host_shaders/detile_m32x1.comp @@ -20,7 +20,7 @@ layout(push_constant) uniform image_info { } info; // Inverse morton LUT, small enough to fit into K$ -uint rmort[16] = { +const uint rmort[16] = { 0x11011000, 0x31213020, 0x13031202, 0x33233222, 0x51415040, 0x71617060, diff --git a/src/video_core/host_shaders/detile_m32x2.comp b/src/video_core/host_shaders/detile_m32x2.comp index 90063a185..c128ba5a1 100644 --- a/src/video_core/host_shaders/detile_m32x2.comp +++ b/src/video_core/host_shaders/detile_m32x2.comp @@ -20,7 +20,7 @@ layout(push_constant) uniform image_info { } info; // Inverse morton LUT, small enough to fit into K$ -uint rmort[16] = { +const uint rmort[16] = { 0x11011000, 0x31213020, 0x13031202, 0x33233222, 0x51415040, 0x71617060, diff --git a/src/video_core/host_shaders/detile_m32x4.comp b/src/video_core/host_shaders/detile_m32x4.comp index e1b988172..a09a0b4c4 100644 --- a/src/video_core/host_shaders/detile_m32x4.comp +++ b/src/video_core/host_shaders/detile_m32x4.comp @@ -20,7 +20,7 @@ layout(push_constant) uniform image_info { } info; // Inverse morton LUT, small enough to fit into K$ -uint rmort[16] = { +const uint rmort[16] = { 0x11011000, 0x31213020, 0x13031202, 0x33233222, 0x51415040, 0x71617060, diff --git a/src/video_core/host_shaders/detile_m8x1.comp b/src/video_core/host_shaders/detile_m8x1.comp index 39d0aaeb1..ecf706450 100644 --- a/src/video_core/host_shaders/detile_m8x1.comp +++ b/src/video_core/host_shaders/detile_m8x1.comp @@ -48,4 +48,4 @@ void main() { uint dw_ofs_x = target_tile_x * 2 + col; // 2 = uints uint dw_ofs_y = (target_tile_y * MICRO_TILE_DIM + row) * tiles_per_pitch * 2; // 2 = uints out_data[dw_ofs_x + dw_ofs_y] = dst_tx; -} \ No newline at end of file +} diff --git a/src/video_core/host_shaders/detile_m8x2.comp b/src/video_core/host_shaders/detile_m8x2.comp index 3f8e5ab33..909a14acc 100644 --- a/src/video_core/host_shaders/detile_m8x2.comp +++ b/src/video_core/host_shaders/detile_m8x2.comp @@ -25,7 +25,7 @@ layout(push_constant) uniform image_info { #define TEXELS_PER_ELEMENT 2 // Inverse morton LUT, small enough to fit into K$ -uint rmort[16] = { +const uint rmort[16] = { 0x11011000, 0x31213020, 0x13031202, 0x33233222, 0x51415040, 0x71617060, diff --git a/src/video_core/host_shaders/detile_macro32x1.comp b/src/video_core/host_shaders/detile_macro32x1.comp index 086fbcfb5..ecac47d1c 100644 --- a/src/video_core/host_shaders/detile_macro32x1.comp +++ b/src/video_core/host_shaders/detile_macro32x1.comp @@ -21,46 +21,46 @@ layout(push_constant) uniform image_info { } info; // Each LUT is 64 bytes, so should fit into K$ given tiled slices locality -const uint lut_32bpp[][64] = { +const uint lut_32bpp[][16] = { { - 0x00, 0x01, 0x04, 0x05, 0x40, 0x41, 0x44, 0x45, - 0x02, 0x03, 0x06, 0x07, 0x42, 0x43, 0x46, 0x47, - 0x10, 0x11, 0x14, 0x15, 0x50, 0x51, 0x54, 0x55, - 0x12, 0x13, 0x16, 0x17, 0x52, 0x53, 0x56, 0x57, - 0x80, 0x81, 0x84, 0x85, 0xc0, 0xc1, 0xc4, 0xc5, - 0x82, 0x83, 0x86, 0x87, 0xc2, 0xc3, 0xc6, 0xc7, - 0x90, 0x91, 0x94, 0x95, 0xd0, 0xd1, 0xd4, 0xd5, - 0x92, 0x93, 0x96, 0x97, 0xd2, 0xd3, 0xd6, 0xd7, + 0x05040100, 0x45444140, + 0x07060302, 0x47464342, + 0x15141110, 0x55545150, + 0x17161312, 0x57565352, + 0x85848180, 0xc5c4c1c0, + 0x87868382, 0xc7c6c3c2, + 0x95949190, 0xd5d4d1d0, + 0x97969392, 0xd7d6d3d2, }, { - 0x08, 0x09, 0x0c, 0x0d, 0x48, 0x49, 0x4c, 0x4d, - 0x0a, 0x0b, 0x0e, 0x0f, 0x4a, 0x4b, 0x4e, 0x4f, - 0x18, 0x19, 0x1c, 0x1d, 0x58, 0x59, 0x5c, 0x5d, - 0x1a, 0x1b, 0x1e, 0x1f, 0x5a, 0x5b, 0x5e, 0x5f, - 0x88, 0x89, 0x8c, 0x8d, 0xc8, 0xc9, 0xcc, 0xcd, - 0x8a, 0x8b, 0x8e, 0x8f, 0xca, 0xcb, 0xce, 0xcf, - 0x98, 0x99, 0x9c, 0x9d, 0xd8, 0xd9, 0xdc, 0xdd, - 0x9a, 0x9b, 0x9e, 0x9f, 0xda, 0xdb, 0xde, 0xdf, + 0x0d0c0908, 0x4d4c4948, + 0x0f0e0b0a, 0x4f4e4b4a, + 0x1d1c1918, 0x5d5c5958, + 0x1f1e1b1a, 0x5f5e5b5a, + 0x8d8c8988, 0xcdccc9c8, + 0x8f8e8b8a, 0xcfcecbca, + 0x9d9c9998, 0xdddcd9d8, + 0x9f9e9b9a, 0xdfdedbda, }, { - 0x20, 0x21, 0x24, 0x25, 0x60, 0x61, 0x64, 0x65, - 0x22, 0x23, 0x26, 0x27, 0x62, 0x63, 0x66, 0x67, - 0x30, 0x31, 0x34, 0x35, 0x70, 0x71, 0x74, 0x75, - 0x32, 0x33, 0x36, 0x37, 0x72, 0x73, 0x76, 0x77, - 0xa0, 0xa1, 0xa4, 0xa5, 0xe0, 0xe1, 0xe4, 0xe5, - 0xa2, 0xa3, 0xa6, 0xa7, 0xe2, 0xe3, 0xe6, 0xe7, - 0xb0, 0xb1, 0xb4, 0xb5, 0xf0, 0xf1, 0xf4, 0xf5, - 0xb2, 0xb3, 0xb6, 0xb7, 0xf2, 0xf3, 0xf6, 0xf7, + 0x25242120, 0x65646160, + 0x27262322, 0x67666362, + 0x35343130, 0x75747170, + 0x37363332, 0x77767372, + 0xa5a4a1a0, 0xe5e4e1e0, + 0xa7a6a3a2, 0xe7e6e3e2, + 0xb5b4b1b0, 0xf5f4f1f0, + 0xb7b6b3b2, 0xf7f6f3f2, }, { - 0x28, 0x29, 0x2c, 0x2d, 0x68, 0x69, 0x6c, 0x6d, - 0x2a, 0x2b, 0x2e, 0x2f, 0x6a, 0x6b, 0x6e, 0x6f, - 0x38, 0x39, 0x3c, 0x3d, 0x78, 0x79, 0x7c, 0x7d, - 0x3a, 0x3b, 0x3e, 0x3f, 0x7a, 0x7b, 0x7e, 0x7f, - 0xa8, 0xa9, 0xac, 0xad, 0xe8, 0xe9, 0xec, 0xed, - 0xaa, 0xab, 0xae, 0xaf, 0xea, 0xeb, 0xee, 0xef, - 0xb8, 0xb9, 0xbc, 0xbd, 0xf8, 0xf9, 0xfc, 0xfd, - 0xba, 0xbb, 0xbe, 0xbf, 0xfa, 0xfb, 0xfe, 0xff, + 0x2d2c2928, 0x6d6c6968, + 0x2f2e2b2a, 0x6f6e6b6a, + 0x3d3c3938, 0x7d7c7978, + 0x3f3e3b3a, 0x7f7e7b7a, + 0xadaca9a8, 0xedece9e8, + 0xafaeabaa, 0xefeeebea, + 0xbdbcb9b8, 0xfdfcf9f8, + 0xbfbebbba, 0xfffefbfa, } }; @@ -77,7 +77,9 @@ void main() { uint col = bitfieldExtract(x, 0, 3); uint row = bitfieldExtract(y, 0, 3); uint lut = bitfieldExtract(z, 0, 2); - uint idx = lut_32bpp[lut][col + row * MICRO_TILE_DIM]; + uint idx_dw = lut_32bpp[lut][(col + row * MICRO_TILE_DIM) >> 2u]; + uint byte_ofs = gl_LocalInvocationID.x & 3u; + uint idx = bitfieldExtract(idx_dw >> (8 * byte_ofs), 0, 8); uint slice_offs = (z >> 2u) * info.c1 * MICRO_TILE_SZ; uint tile_row = y / MICRO_TILE_DIM; diff --git a/src/video_core/host_shaders/detile_macro32x2.comp b/src/video_core/host_shaders/detile_macro32x2.comp index 296311c7a..986acc963 100644 --- a/src/video_core/host_shaders/detile_macro32x2.comp +++ b/src/video_core/host_shaders/detile_macro32x2.comp @@ -20,46 +20,46 @@ layout(push_constant) uniform image_info { uint c1; } info; -const uint lut_64bpp[][64] = { +const uint lut_64bpp[][16] = { { - 0x00, 0x01, 0x08, 0x09, 0x40, 0x41, 0x48, 0x49, - 0x02, 0x03, 0x0a, 0x0b, 0x42, 0x43, 0x4a, 0x4b, - 0x10, 0x11, 0x18, 0x19, 0x50, 0x51, 0x58, 0x59, - 0x12, 0x13, 0x1a, 0x1b, 0x52, 0x53, 0x5a, 0x5b, - 0x80, 0x81, 0x88, 0x89, 0xc0, 0xc1, 0xc8, 0xc9, - 0x82, 0x83, 0x8a, 0x8b, 0xc2, 0xc3, 0xca, 0xcb, - 0x90, 0x91, 0x98, 0x99, 0xd0, 0xd1, 0xd8, 0xd9, - 0x92, 0x93, 0x9a, 0x9b, 0xd2, 0xd3, 0xda, 0xdb, + 0x09080100, 0x49484140, + 0x0b0a0302, 0x4a4b4342, + 0x19181110, 0x59585150, + 0x1b1a1312, 0x5a5b5352, + 0x89888180, 0xc9c8c1c0, + 0x8b8a8382, 0xcacbc3c2, + 0x99989190, 0xd9d8d1d0, + 0x9b9a9392, 0xdbdad3d2, }, { - 0x04, 0x05, 0x0c, 0x0d, 0x44, 0x45, 0x4c, 0x4d, - 0x06, 0x07, 0x0e, 0x0f, 0x46, 0x47, 0x4e, 0x4f, - 0x14, 0x15, 0x1c, 0x1d, 0x54, 0x55, 0x5c, 0x5d, - 0x16, 0x17, 0x1e, 0x1f, 0x56, 0x57, 0x5e, 0x5f, - 0x84, 0x85, 0x8c, 0x8d, 0xc4, 0xc5, 0xcc, 0xcd, - 0x86, 0x87, 0x8e, 0x8f, 0xc6, 0xc7, 0xce, 0xcf, - 0x94, 0x95, 0x9c, 0x9d, 0xd4, 0xd5, 0xdc, 0xdd, - 0x96, 0x97, 0x9e, 0x9f, 0xd6, 0xd7, 0xde, 0xdf, + 0x0d0c0504, 0x4d4c4544, + 0x0f0e0706, 0x4f4e4746, + 0x1d1c1514, 0x5d5c5554, + 0x1f1e1716, 0x5f5e5756, + 0x8d8c8584, 0xcdccc5c4, + 0x8f8e8786, 0xcfcec7c6, + 0x9d9c9594, 0xdddcd5d4, + 0x9f9e9796, 0xdfded7d6, }, { - 0x20, 0x21, 0x28, 0x29, 0x60, 0x61, 0x68, 0x69, - 0x22, 0x23, 0x2a, 0x2b, 0x62, 0x63, 0x6a, 0x6b, - 0x30, 0x31, 0x38, 0x39, 0x70, 0x71, 0x78, 0x79, - 0x32, 0x33, 0x3a, 0x3b, 0x72, 0x73, 0x7a, 0x7b, - 0xa0, 0xa1, 0xa8, 0xa9, 0xe0, 0xe1, 0xe8, 0xe9, - 0xa2, 0xa3, 0xaa, 0xab, 0xe2, 0xe3, 0xea, 0xeb, - 0xb0, 0xb1, 0xb8, 0xb9, 0xf0, 0xf1, 0xf8, 0xf9, - 0xb2, 0xb3, 0xba, 0xbb, 0xf2, 0xf3, 0xfa, 0xfb, + 0x29282120, 0x69686160, + 0x2b2a2322, 0x6b6a6362, + 0x39383130, 0x79787170, + 0x3b3a3332, 0x7b7a7372, + 0xa9a8a1a0, 0xe9e8e1e0, + 0xabaaa3a2, 0xebeae3e2, + 0xb9b8b1b0, 0xf9f8f1f0, + 0xbbbab3b2, 0xfbfaf3f2, }, { - 0x24, 0x25, 0x2c, 0x2d, 0x64, 0x65, 0x6c, 0x6d, - 0x26, 0x27, 0x2e, 0x2f, 0x66, 0x67, 0x6e, 0x6f, - 0x34, 0x35, 0x3c, 0x3d, 0x74, 0x75, 0x7c, 0x7d, - 0x36, 0x37, 0x3e, 0x3f, 0x76, 0x77, 0x7e, 0x7f, - 0xa4, 0xa5, 0xac, 0xad, 0xe4, 0xe5, 0xec, 0xed, - 0xa6, 0xa7, 0xae, 0xaf, 0xe6, 0xe7, 0xee, 0xef, - 0xb4, 0xb5, 0xbc, 0xbd, 0xf4, 0xf5, 0xfc, 0xfd, - 0xb6, 0xb7, 0xbe, 0xbf, 0xf6, 0xf7, 0xfe, 0xff, + 0x2d2c2524, 0x6d6c6564, + 0x2f2e2726, 0x6f6e6766, + 0x3d3c3534, 0x7d7c7574, + 0x3f3e3736, 0x7f7e7776, + 0xadaca5a4, 0xedece5e4, + 0xafaea7a6, 0xefeee7e6, + 0xbdbcb5b4, 0xfdfcf5f4, + 0xbfbeb7b6, 0xfffef7f6, }, }; @@ -76,7 +76,9 @@ void main() { uint col = bitfieldExtract(x, 0, 3); uint row = bitfieldExtract(y, 0, 3); uint lut = bitfieldExtract(z, 0, 2); - uint idx = lut_64bpp[lut][col + row * MICRO_TILE_DIM]; + uint idx_dw = lut_64bpp[lut][(col + row * MICRO_TILE_DIM) >> 2u]; + uint byte_ofs = gl_LocalInvocationID.x & 3u; + uint idx = bitfieldExtract(idx_dw >> (8 * byte_ofs), 0, 8); uint slice_offs = (z >> 2u) * info.c1 * MICRO_TILE_SZ; uint tile_row = y / MICRO_TILE_DIM; @@ -85,7 +87,7 @@ void main() { uint offs = slice_offs + tile_offs + (idx * BPP / 8); uint p0 = in_data[(offs >> 2) + 0]; - uint p1 = in_data[(offs >> 2) + 1]; + uint p1 = in_data[(offs >> 2) + 1]; out_data[2 * gl_GlobalInvocationID.x + 0] = p0; - out_data[2 * gl_GlobalInvocationID.x + 1] = p1; + out_data[2 * gl_GlobalInvocationID.x + 1] = p1; } diff --git a/src/video_core/host_shaders/detile_macro8x1.comp b/src/video_core/host_shaders/detile_macro8x1.comp new file mode 100644 index 000000000..cddc8af5b --- /dev/null +++ b/src/video_core/host_shaders/detile_macro8x1.comp @@ -0,0 +1,101 @@ +// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project +// SPDX-License-Identifier: GPL-2.0-or-later + +#version 450 + +layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in; + +layout(std430, binding = 0) buffer input_buf { + uint in_data[]; +}; +layout(std430, binding = 1) buffer output_buf { + uint out_data[]; +}; + +layout(push_constant) uniform image_info { + uint num_levels; + uint pitch; + uint height; + uint c0; + uint c1; +} info; + +const uint lut_8bpp[][16] = { + { + 0x05040100, 0x45444140, + 0x07060302, 0x47464342, + 0x0d0c0908, 0x4d4c4948, + 0x0f0e0b0a, 0x4f4e4b4a, + 0x85848180, 0xc5c4c1c0, + 0x87868382, 0xc7c6c3c2, + 0x8d8c8988, 0xcdccc9c8, + 0x8f8e8b8a, 0xcfcecbca, + }, + { + 0x15141110, 0x55545150, + 0x17161312, 0x57565352, + 0x1d1c1918, 0x5d5c5958, + 0x1f1e1b1a, 0x5f5e5b5a, + 0x95949190, 0xd5d4d1d0, + 0x97969392, 0xd7d6d3d2, + 0x9d9c9998, 0xdddcd9d8, + 0x9f9e9b9a, 0xdfdedbda, + }, + { + 0x25242120, 0x65646160, + 0x27262322, 0x67666362, + 0x2d2c2928, 0x6d6c6968, + 0x2f2e2b2a, 0x6f6e6b6a, + 0xa5a4a1a0, 0xe5e4e1e0, + 0xa7a6a3a2, 0xe7e6e3e2, + 0xadaca9a8, 0xedece9e8, + 0xafaeabaa, 0xefeeebea, + }, + { + 0x35343130, 0x75747170, + 0x37363332, 0x77767372, + 0x3d3c3938, 0x7d7c7978, + 0x3f3e3b3a, 0x7f7e7b7a, + 0xb5b4b1b0, 0xf5f4f1f0, + 0xb7b6b3b2, 0xf7f6f3f2, + 0xbdbcb9b8, 0xfdfcf9f8, + 0xbfbebbba, 0xfffefbfa, + }, +}; + +#define MICRO_TILE_DIM (8) +#define MICRO_TILE_SZ (256) +#define TEXELS_PER_ELEMENT (1) +#define BPP (8) + +shared uint scratch[16]; + +void main() { + uint slot = gl_LocalInvocationID.x >> 2u; + atomicAnd(scratch[slot], 0); + + uint x = gl_GlobalInvocationID.x % info.pitch; + uint y = (gl_GlobalInvocationID.x / info.pitch) % info.height; + uint z = gl_GlobalInvocationID.x / (info.pitch * info.height); + + uint col = bitfieldExtract(x, 0, 3); + uint row = bitfieldExtract(y, 0, 3); + uint lut = bitfieldExtract(z, 0, 2); + uint idx_dw = lut_8bpp[lut][(col + row * MICRO_TILE_DIM) >> 2u]; + uint byte_ofs = (gl_LocalInvocationID.x & 3u) * 8; + uint idx = bitfieldExtract(idx_dw >> byte_ofs, 0, 8); + + uint slice_offs = (z >> 2u) * info.c1 * MICRO_TILE_SZ; + uint tile_row = y / MICRO_TILE_DIM; + uint tile_column = x / MICRO_TILE_DIM; + uint tile_offs = ((tile_row * info.c0) + tile_column) * MICRO_TILE_SZ; + uint offs = (slice_offs + tile_offs) + (idx * BPP / 8); + + uint p0 = in_data[offs >> 2u]; + uint byte = bitfieldExtract(p0 >> (offs * 8), 0, 8); + atomicOr(scratch[slot], byte << byte_ofs); + + if (byte_ofs == 0) { + out_data[gl_GlobalInvocationID.x >> 2u] = scratch[slot]; + } +} diff --git a/src/video_core/multi_level_page_table.h b/src/video_core/multi_level_page_table.h index 527476f3b..7f3205e1a 100644 --- a/src/video_core/multi_level_page_table.h +++ b/src/video_core/multi_level_page_table.h @@ -39,6 +39,15 @@ public: return &(*first_level_map[l1_page])[l2_page]; } + [[nodiscard]] const Entry* find(size_t page) const { + const size_t l1_page = page >> SecondLevelBits; + const size_t l2_page = page & (NumEntriesPerL1Page - 1); + if (!first_level_map[l1_page]) { + return nullptr; + } + return &(*first_level_map[l1_page])[l2_page]; + } + [[nodiscard]] const Entry& operator[](size_t page) const { const size_t l1_page = page >> SecondLevelBits; const size_t l2_page = page & (NumEntriesPerL1Page - 1); diff --git a/src/video_core/page_manager.cpp b/src/video_core/page_manager.cpp index 556555c25..47ed9e543 100644 --- a/src/video_core/page_manager.cpp +++ b/src/video_core/page_manager.cpp @@ -185,7 +185,7 @@ void PageManager::OnGpuUnmap(VAddr address, size_t size) { void PageManager::UpdatePagesCachedCount(VAddr addr, u64 size, s32 delta) { static constexpr u64 PageShift = 12; - std::scoped_lock lk{mutex}; + 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; diff --git a/src/video_core/page_manager.h b/src/video_core/page_manager.h index 29a946a8f..f44307f92 100644 --- a/src/video_core/page_manager.h +++ b/src/video_core/page_manager.h @@ -4,8 +4,8 @@ #pragma once #include -#include #include +#include "common/spin_lock.h" #include "common/types.h" namespace Vulkan { @@ -35,8 +35,8 @@ private: struct Impl; std::unique_ptr impl; Vulkan::Rasterizer* rasterizer; - std::mutex mutex; boost::icl::interval_map cached_pages; + Common::SpinLock lock; }; } // namespace VideoCore diff --git a/src/video_core/renderer_vulkan/liverpool_to_vk.cpp b/src/video_core/renderer_vulkan/liverpool_to_vk.cpp index c41b760ba..690d26cfc 100644 --- a/src/video_core/renderer_vulkan/liverpool_to_vk.cpp +++ b/src/video_core/renderer_vulkan/liverpool_to_vk.cpp @@ -691,16 +691,40 @@ std::span SurfaceFormats() { return formats; } +// Table 8.13 Data and Image Formats [Sea Islands Series Instruction Set Architecture] +static const size_t amd_gpu_data_format_bit_size = 6; // All values are under 64 +static const size_t amd_gpu_number_format_bit_size = 4; // All values are under 16 + +static size_t GetSurfaceFormatTableIndex(AmdGpu::DataFormat data_format, + AmdGpu::NumberFormat num_format) { + DEBUG_ASSERT(u32(data_format) < 1 << amd_gpu_data_format_bit_size); + DEBUG_ASSERT(u32(num_format) < 1 << amd_gpu_number_format_bit_size); + size_t result = static_cast(num_format) | + (static_cast(data_format) << amd_gpu_number_format_bit_size); + return result; +} + +static auto surface_format_table = []() constexpr { + std::array + result; + for (auto& entry : result) { + entry = vk::Format::eUndefined; + } + for (const auto& supported_format : SurfaceFormats()) { + result[GetSurfaceFormatTableIndex(supported_format.data_format, + supported_format.number_format)] = + supported_format.vk_format; + } + return result; +}(); + vk::Format SurfaceFormat(AmdGpu::DataFormat data_format, AmdGpu::NumberFormat num_format) { - const auto& formats = SurfaceFormats(); - const auto format = - std::find_if(formats.begin(), formats.end(), [&](const SurfaceFormatInfo& format_info) { - return format_info.data_format == data_format && - format_info.number_format == num_format; - }); - ASSERT_MSG(format != formats.end(), "Unknown data_format={} and num_format={}", - static_cast(data_format), static_cast(num_format)); - return format->vk_format; + vk::Format result = surface_format_table[GetSurfaceFormatTableIndex(data_format, num_format)]; + bool found = + result != vk::Format::eUndefined || data_format == AmdGpu::DataFormat::FormatInvalid; + ASSERT_MSG(found, "Unknown data_format={} and num_format={}", static_cast(data_format), + static_cast(num_format)); + return result; } static constexpr DepthFormatInfo CreateDepthFormatInfo( @@ -746,8 +770,8 @@ vk::Format DepthFormat(DepthBuffer::ZFormat z_format, DepthBuffer::StencilFormat vk::ClearValue ColorBufferClearValue(const AmdGpu::Liverpool::ColorBuffer& color_buffer) { const auto comp_swizzle = color_buffer.Swizzle(); - const auto format = color_buffer.DataFormat(); - const auto number_type = color_buffer.NumFormat(); + const auto format = color_buffer.GetDataFmt(); + const auto number_type = color_buffer.GetNumberFmt(); const auto& c0 = color_buffer.clear_word0; const auto& c1 = color_buffer.clear_word1; diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp index cd1b42b05..ba069dae1 100644 --- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp +++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp @@ -328,8 +328,8 @@ bool PipelineCache::RefreshGraphicsKey() { } key.color_formats[remapped_cb] = - LiverpoolToVK::SurfaceFormat(col_buf.DataFormat(), col_buf.NumFormat()); - key.color_num_formats[remapped_cb] = col_buf.NumFormat(); + LiverpoolToVK::SurfaceFormat(col_buf.GetDataFmt(), col_buf.GetNumberFmt()); + key.color_num_formats[remapped_cb] = col_buf.GetNumberFmt(); key.color_swizzles[remapped_cb] = col_buf.Swizzle(); } diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp index 4384cdbea..6e628239b 100644 --- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp +++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp @@ -562,6 +562,12 @@ void Rasterizer::BindBuffers(const Shader::Info& stage, Shader::Backend::Binding push_data.AddOffset(binding.buffer, adjust); buffer_infos.emplace_back(vk_buffer->Handle(), offset_aligned, vsharp.GetSize() + adjust); + if (auto barrier = + vk_buffer->GetBarrier(desc.is_written ? vk::AccessFlagBits2::eShaderWrite + : vk::AccessFlagBits2::eShaderRead, + vk::PipelineStageFlagBits2::eAllCommands)) { + buffer_barriers.emplace_back(*barrier); + } } set_writes.push_back({ @@ -600,7 +606,7 @@ void Rasterizer::BindBuffers(const Shader::Info& stage, Shader::Backend::Binding if (auto barrier = vk_buffer->GetBarrier(desc.is_written ? vk::AccessFlagBits2::eShaderWrite : vk::AccessFlagBits2::eShaderRead, - vk::PipelineStageFlagBits2::eComputeShader)) { + vk::PipelineStageFlagBits2::eAllCommands)) { buffer_barriers.emplace_back(*barrier); } if (desc.is_written) { diff --git a/src/video_core/texture_cache/image_info.cpp b/src/video_core/texture_cache/image_info.cpp index 0559f1be3..adc72c21f 100644 --- a/src/video_core/texture_cache/image_info.cpp +++ b/src/video_core/texture_cache/image_info.cpp @@ -265,9 +265,9 @@ ImageInfo::ImageInfo(const AmdGpu::Liverpool::ColorBuffer& buffer, const AmdGpu::Liverpool::CbDbExtent& hint /*= {}*/) noexcept { props.is_tiled = buffer.IsTiled(); tiling_mode = buffer.GetTilingMode(); - pixel_format = LiverpoolToVK::SurfaceFormat(buffer.DataFormat(), buffer.NumFormat()); + pixel_format = LiverpoolToVK::SurfaceFormat(buffer.GetDataFmt(), buffer.GetNumberFmt()); num_samples = buffer.NumSamples(); - num_bits = NumBits(buffer.DataFormat()); + num_bits = NumBits(buffer.GetDataFmt()); type = vk::ImageType::e2D; size.width = hint.Valid() ? hint.width : buffer.Pitch(); size.height = hint.Valid() ? hint.height : buffer.Height(); diff --git a/src/video_core/texture_cache/image_view.cpp b/src/video_core/texture_cache/image_view.cpp index a9ae41dd1..68b116558 100644 --- a/src/video_core/texture_cache/image_view.cpp +++ b/src/video_core/texture_cache/image_view.cpp @@ -76,7 +76,8 @@ ImageViewInfo::ImageViewInfo(const AmdGpu::Liverpool::ColorBuffer& col_buffer) n range.base.layer = col_buffer.view.slice_start; range.extent.layers = col_buffer.NumSlices() - range.base.layer; type = range.extent.layers > 1 ? vk::ImageViewType::e2DArray : vk::ImageViewType::e2D; - format = Vulkan::LiverpoolToVK::SurfaceFormat(col_buffer.DataFormat(), col_buffer.NumFormat()); + format = + Vulkan::LiverpoolToVK::SurfaceFormat(col_buffer.GetDataFmt(), col_buffer.GetNumberFmt()); } ImageViewInfo::ImageViewInfo(const AmdGpu::Liverpool::DepthBuffer& depth_buffer, diff --git a/src/video_core/texture_cache/texture_cache.cpp b/src/video_core/texture_cache/texture_cache.cpp index 897d6f67e..291e1da7c 100644 --- a/src/video_core/texture_cache/texture_cache.cpp +++ b/src/video_core/texture_cache/texture_cache.cpp @@ -542,31 +542,62 @@ void TextureCache::RefreshImage(Image& image, Vulkan::Scheduler* custom_schedule sched_ptr->EndRendering(); const auto cmdbuf = sched_ptr->CommandBuffer(); - image.Transit(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite, {}, - cmdbuf); - const VAddr image_addr = image.info.guest_address; const size_t image_size = image.info.guest_size_bytes; const auto [vk_buffer, buf_offset] = buffer_cache.ObtainViewBuffer(image_addr, image_size, is_gpu_dirty); + // The obtained buffer may be written by a shader so we need to emit a barrier to prevent RAW // hazard if (auto barrier = vk_buffer->GetBarrier(vk::AccessFlagBits2::eTransferRead, vk::PipelineStageFlagBits2::eTransfer)) { - const auto dependencies = vk::DependencyInfo{ + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ .dependencyFlags = vk::DependencyFlagBits::eByRegion, .bufferMemoryBarrierCount = 1, .pBufferMemoryBarriers = &barrier.value(), - }; - cmdbuf.pipelineBarrier2(dependencies); + }); } - const auto [buffer, offset] = tile_manager.TryDetile(vk_buffer->Handle(), buf_offset, image); + const auto [buffer, offset] = + tile_manager.TryDetile(vk_buffer->Handle(), buf_offset, image.info); for (auto& copy : image_copy) { copy.bufferOffset += offset; } + const vk::BufferMemoryBarrier2 pre_barrier{ + .srcStageMask = vk::PipelineStageFlagBits2::eAllCommands, + .srcAccessMask = vk::AccessFlagBits2::eMemoryWrite, + .dstStageMask = vk::PipelineStageFlagBits2::eTransfer, + .dstAccessMask = vk::AccessFlagBits2::eTransferRead, + .buffer = buffer, + .offset = offset, + .size = image_size, + }; + 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, + .offset = offset, + .size = image_size, + }; + const auto image_barriers = + image.GetBarriers(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite, + vk::PipelineStageFlagBits2::eTransfer, {}); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &pre_barrier, + .imageMemoryBarrierCount = static_cast(image_barriers.size()), + .pImageMemoryBarriers = image_barriers.data(), + }); cmdbuf.copyBufferToImage(buffer, image.image, vk::ImageLayout::eTransferDstOptimal, image_copy); + cmdbuf.pipelineBarrier2(vk::DependencyInfo{ + .dependencyFlags = vk::DependencyFlagBits::eByRegion, + .bufferMemoryBarrierCount = 1, + .pBufferMemoryBarriers = &post_barrier, + }); image.flags &= ~ImageFlagBits::Dirty; } diff --git a/src/video_core/texture_cache/tile_manager.cpp b/src/video_core/texture_cache/tile_manager.cpp index de108843b..0e550c7dc 100644 --- a/src/video_core/texture_cache/tile_manager.cpp +++ b/src/video_core/texture_cache/tile_manager.cpp @@ -4,6 +4,7 @@ #include "video_core/renderer_vulkan/vk_instance.h" #include "video_core/renderer_vulkan/vk_scheduler.h" #include "video_core/renderer_vulkan/vk_shader_util.h" +#include "video_core/texture_cache/image_info.h" #include "video_core/texture_cache/image_view.h" #include "video_core/texture_cache/tile_manager.h" @@ -14,6 +15,7 @@ #include "video_core/host_shaders/detile_m8x2_comp.h" #include "video_core/host_shaders/detile_macro32x1_comp.h" #include "video_core/host_shaders/detile_macro32x2_comp.h" +#include "video_core/host_shaders/detile_macro8x1_comp.h" #include #include @@ -32,6 +34,7 @@ static vk::Format DemoteImageFormatForDetiling(vk::Format format) { case vk::Format::eR5G5B5A1UnormPack16: case vk::Format::eR8G8Unorm: case vk::Format::eR16Sfloat: + case vk::Format::eR16Uint: case vk::Format::eR16Unorm: case vk::Format::eD16Unorm: return vk::Format::eR8G8Uint; @@ -85,10 +88,10 @@ static vk::Format DemoteImageFormatForDetiling(vk::Format format) { return format; } -const DetilerContext* TileManager::GetDetiler(const Image& image) const { - const auto format = DemoteImageFormatForDetiling(image.info.pixel_format); +const DetilerContext* TileManager::GetDetiler(const ImageInfo& info) const { + const auto format = DemoteImageFormatForDetiling(info.pixel_format); - switch (image.info.tiling_mode) { + switch (info.tiling_mode) { case AmdGpu::TilingMode::Texture_MicroTiled: switch (format) { case vk::Format::eR8Uint: @@ -106,6 +109,8 @@ const DetilerContext* TileManager::GetDetiler(const Image& image) const { } case AmdGpu::TilingMode::Texture_Volume: switch (format) { + case vk::Format::eR8Uint: + return &detilers[DetilerType::Macro8x1]; case vk::Format::eR32Uint: return &detilers[DetilerType::Macro32x1]; case vk::Format::eR32G32Uint: @@ -131,8 +136,8 @@ TileManager::TileManager(const Vulkan::Instance& instance, Vulkan::Scheduler& sc static const std::array detiler_shaders{ HostShaders::DETILE_M8X1_COMP, HostShaders::DETILE_M8X2_COMP, HostShaders::DETILE_M32X1_COMP, HostShaders::DETILE_M32X2_COMP, - HostShaders::DETILE_M32X4_COMP, HostShaders::DETILE_MACRO32X1_COMP, - HostShaders::DETILE_MACRO32X2_COMP, + HostShaders::DETILE_M32X4_COMP, HostShaders::DETILE_MACRO8X1_COMP, + HostShaders::DETILE_MACRO32X1_COMP, HostShaders::DETILE_MACRO32X2_COMP, }; boost::container::static_vector bindings{ @@ -257,23 +262,23 @@ void TileManager::FreeBuffer(ScratchBuffer buffer) { } std::pair TileManager::TryDetile(vk::Buffer in_buffer, u32 in_offset, - Image& image) { - if (!image.info.props.is_tiled) { + const ImageInfo& info) { + if (!info.props.is_tiled) { return {in_buffer, in_offset}; } - const auto* detiler = GetDetiler(image); + const auto* detiler = GetDetiler(info); if (!detiler) { - if (image.info.tiling_mode != AmdGpu::TilingMode::Texture_MacroTiled && - image.info.tiling_mode != AmdGpu::TilingMode::Display_MacroTiled && - image.info.tiling_mode != AmdGpu::TilingMode::Depth_MacroTiled) { + if (info.tiling_mode != AmdGpu::TilingMode::Texture_MacroTiled && + info.tiling_mode != AmdGpu::TilingMode::Display_MacroTiled && + info.tiling_mode != AmdGpu::TilingMode::Depth_MacroTiled) { LOG_ERROR(Render_Vulkan, "Unsupported tiled image: {} ({})", - vk::to_string(image.info.pixel_format), NameOf(image.info.tiling_mode)); + vk::to_string(info.pixel_format), NameOf(info.tiling_mode)); } return {in_buffer, in_offset}; } - const u32 image_size = image.info.guest_size_bytes; + const u32 image_size = info.guest_size_bytes; // Prepare output buffer auto out_buffer = AllocBuffer(image_size, true); @@ -316,22 +321,20 @@ std::pair TileManager::TryDetile(vk::Buffer in_buffer, u32 in_o set_writes); DetilerParams params; - params.num_levels = image.info.resources.levels; - params.pitch0 = image.info.pitch >> (image.info.props.is_block ? 2u : 0u); - params.height = image.info.size.height; - if (image.info.tiling_mode == AmdGpu::TilingMode::Texture_Volume) { - ASSERT(image.info.resources.levels == 1); - ASSERT(image.info.num_bits >= 32); - const auto tiles_per_row = image.info.pitch / 8u; - const auto tiles_per_slice = tiles_per_row * ((image.info.size.height + 7u) / 8u); + params.num_levels = info.resources.levels; + params.pitch0 = info.pitch >> (info.props.is_block ? 2u : 0u); + params.height = info.size.height; + if (info.tiling_mode == AmdGpu::TilingMode::Texture_Volume) { + ASSERT(info.resources.levels == 1); + const auto tiles_per_row = info.pitch / 8u; + const auto tiles_per_slice = tiles_per_row * ((info.size.height + 7u) / 8u); params.sizes[0] = tiles_per_row; params.sizes[1] = tiles_per_slice; } else { - - ASSERT(image.info.resources.levels <= 14); + ASSERT(info.resources.levels <= 14); std::memset(¶ms.sizes, 0, sizeof(params.sizes)); - for (int m = 0; m < image.info.resources.levels; ++m) { - params.sizes[m] = image.info.mips_layout[m].size * image.info.resources.layers + + for (int m = 0; m < info.resources.levels; ++m) { + params.sizes[m] = info.mips_layout[m].size * info.resources.layers + (m > 0 ? params.sizes[m - 1] : 0); } } @@ -340,20 +343,9 @@ std::pair TileManager::TryDetile(vk::Buffer in_buffer, u32 in_o ¶ms); ASSERT((image_size % 64) == 0); - const auto bpp = image.info.num_bits * (image.info.props.is_block ? 16u : 1u); + const auto bpp = info.num_bits * (info.props.is_block ? 16u : 1u); const auto num_tiles = image_size / (64 * (bpp / 8)); cmdbuf.dispatch(num_tiles, 1, 1); - - const vk::BufferMemoryBarrier post_barrier{ - .srcAccessMask = vk::AccessFlagBits::eShaderWrite, - .dstAccessMask = vk::AccessFlagBits::eTransferRead, - .buffer = out_buffer.first, - .size = image_size, - }; - cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eComputeShader, - vk::PipelineStageFlagBits::eTransfer, vk::DependencyFlagBits::eByRegion, - {}, post_barrier, {}); - return {out_buffer.first, 0}; } diff --git a/src/video_core/texture_cache/tile_manager.h b/src/video_core/texture_cache/tile_manager.h index 72860bca0..bcf5accd3 100644 --- a/src/video_core/texture_cache/tile_manager.h +++ b/src/video_core/texture_cache/tile_manager.h @@ -5,11 +5,11 @@ #include "common/types.h" #include "video_core/buffer_cache/buffer.h" -#include "video_core/texture_cache/image.h" namespace VideoCore { class TextureCache; +struct ImageInfo; enum DetilerType : u32 { Micro8x1, @@ -18,6 +18,7 @@ enum DetilerType : u32 { Micro32x2, Micro32x4, + Macro8x1, Macro32x1, Macro32x2, @@ -36,14 +37,15 @@ public: TileManager(const Vulkan::Instance& instance, Vulkan::Scheduler& scheduler); ~TileManager(); - std::pair TryDetile(vk::Buffer in_buffer, u32 in_offset, Image& image); + std::pair TryDetile(vk::Buffer in_buffer, u32 in_offset, + const ImageInfo& info); ScratchBuffer AllocBuffer(u32 size, bool is_storage = false); void Upload(ScratchBuffer buffer, const void* data, size_t size); void FreeBuffer(ScratchBuffer buffer); private: - const DetilerContext* GetDetiler(const Image& image) const; + const DetilerContext* GetDetiler(const ImageInfo& info) const; private: const Vulkan::Instance& instance;