Merge PR #716: Fix Unicode paths on Windows

Preserve native Windows paths for filesystem operations, command-line options,
and game icons. Complete Unicode handling in recursive file listings and file
logging, restore unmounting by host directory, and extend regression coverage.

Original implementation by Augusto Domingos (@AghastyGD); retain all five of
his PR commits and include the reviewed follow-up fixes.

Closes #714

Co-authored-by: Augusto Domingos <augustodomingosvasco@gmail.com>

.coderabbit.yaml

@@ -0,0 +1,37 @@
+# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
+
+# Keep inherited review profiles, automatic reviews, tools, and pre-merge checks.
+inheritance: true
+
+reviews:
+  # Trim supplemental output and keep the remaining walkthrough collapsed.
+  high_level_summary: false
+  review_status: false
+  review_details: false
+  collapse_walkthrough: true
+  changed_files_summary: false
+  sequence_diagrams: false
+  estimate_code_review_effort: false
+  related_issues: false
+  related_prs: false
+  in_progress_fortune: false
+  poem: false
+  enable_prompt_for_ai_agents: false
+
+  # Hide standard finishing-touch controls; this also disables their commands.
+  finishing_touches:
+    docstrings:
+      enabled: false
+    unit_tests:
+      enabled: false
+    simplify:
+      enabled: false
+    autofix:
+      enabled: false
+    fix_ci:
+      enabled: false
+    resolve_merge_conflict:
+      enabled: false
+
+chat:
+  art: false

.vscode/settings.json

@@ -2,15 +2,17 @@
     "cmake.sourceDirectory": "${workspaceFolder}",
     "cmake.buildDirectory": "${workspaceFolder}/_Build/vscode-clang",
     "cmake.generator": "Ninja",
+    "cmake.useVsDeveloperEnvironment": "always",
     "cmake.environment": {
         "CC": "clang-cl",
-        "CXX": "clang-cl",
-        "PATH": "${env:ProgramFiles}/Microsoft Visual Studio/18/Community/VC/Tools/Llvm/x64/bin;${env:ProgramFiles}/Microsoft Visual Studio/18/Community/Common7/IDE/CommonExtensions/Microsoft/CMake/Ninja;${env:PATH}"
+        "CXX": "clang-cl"
     },
     "cmake.configureSettings": {
         "CMAKE_C_COMPILER": "clang-cl",
         "CMAKE_CXX_COMPILER": "clang-cl",
         "CMAKE_PREFIX_PATH": "C:/Qt/6.11.0/msvc2022_64"
-    }
+    },
+    "C_Cpp.default.intelliSenseMode": "windows-clang-x64",
+    "C_Cpp.default.compileCommands": "${workspaceFolder}/_Build/vscode-clang/compile_commands.json",
+    "C_Cpp.default.cppStandard": "c++20"
 }
-

3rdparty/CMakeLists.txt

@@ -86,6 +86,9 @@ set(SPDLOG_BUILD_WARNINGS OFF CACHE BOOL "" FORCE)
 set(SPDLOG_INSTALL OFF CACHE BOOL "" FORCE)
 set(SPDLOG_FMT_EXTERNAL ON CACHE BOOL "" FORCE)
 set(SPDLOG_NO_EXCEPTIONS ON CACHE BOOL "" FORCE)
+if(WIN32)
+	set(SPDLOG_WCHAR_FILENAMES ON CACHE BOOL "" FORCE)
+endif()
 add_subdirectory(spdlog EXCLUDE_FROM_ALL)
 
 if (NOT TARGET FFmpeg::ffmpeg)

CMakeLists.txt

@@ -450,6 +450,12 @@ add_executable(audio_out2_port_tests EXCLUDE_FROM_ALL
 target_link_libraries(audio_out2_port_tests common fmt::fmt)
 target_include_directories(audio_out2_port_tests PRIVATE ${inc_headers})
 
+add_executable(save_data_memory_tests EXCLUDE_FROM_ALL
+	"${KYTY_TESTS_DIR}/SaveDataMemoryTests.cpp"
+)
+target_link_libraries(save_data_memory_tests common fmt::fmt)
+target_include_directories(save_data_memory_tests PRIVATE ${inc_headers})
+
 add_executable(event_queue_lifetime_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/EventQueueLifetimeTests.cpp"
 	"${KYTY_SOURCE_DIR}/kernel/eventQueue.cpp"
@@ -534,6 +540,7 @@ if(BUILD_TESTING)
 	add_test(NAME kernel_file_system COMMAND $<TARGET_FILE:kernel_file_system_tests>)
 	add_test(NAME sync_on_address COMMAND $<TARGET_FILE:sync_on_address_tests>)
 	add_test(NAME audio_out2_port COMMAND $<TARGET_FILE:audio_out2_port_tests>)
+	add_test(NAME save_data_memory COMMAND $<TARGET_FILE:save_data_memory_tests>)
 	add_test(NAME shader_recompiler_compute COMMAND $<TARGET_FILE:shader_recompiler_compute_tests>)
 	add_test(NAME shader_recompiler_alignbyte
 		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --alignbyte-only)
@@ -601,6 +608,21 @@ if(BUILD_TESTING)
 	)
 endif()
 
+# Test Darwin signal-context restoration for SSE4a instructions.
+if(APPLE AND (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR
+		(NOT CMAKE_OSX_ARCHITECTURES AND CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|AMD64)$")))
+	add_executable(macos_sse4a_tests EXCLUDE_FROM_ALL
+		"${KYTY_TESTS_DIR}/macosSse4aTests.cpp"
+		"${KYTY_SOURCE_DIR}/loader/x64InstructionEmulator.cpp"
+	)
+	target_link_libraries(macos_sse4a_tests common fmt::fmt Zydis::Zydis)
+	target_include_directories(macos_sse4a_tests PRIVATE ${inc_headers})
+	if(BUILD_TESTING)
+		add_test(NAME macos_sse4a COMMAND $<TARGET_FILE:macos_sse4a_tests>)
+		set_tests_properties(macos_sse4a PROPERTIES TIMEOUT 15)
+		add_dependencies(kyty_tests macos_sse4a_tests)
+	endif()
+endif()
 
 add_executable(kyty_emulator "${KYTY_SOURCE_DIR}/main.cpp" ${kyty_emulator_src})
 configure_macos_guest_address_space(kyty_emulator)

README.md

@@ -137,6 +137,7 @@ the Vulkan/SPIR-V validation rules.
 - Visual Studio 2022 or Build Tools 2022 with the **Desktop development with C++** workload and
   **C++ Clang tools for Windows** component
 - Qt 6 for MSVC 2022 64-bit, including Concurrent, Network, and Widgets
+- [glslang](https://github.com/KhronosGroup/glslang/releases) (`glslangValidator`) on `PATH`
 
 The Microsoft C++ compiler (`cl.exe`) is not supported; use `clang-cl`.
 

src/common/byteBuffer.h

@@ -1,166 +0,0 @@
-#ifndef KYTY_COMMON_BYTEBUFFER_H_
-#define KYTY_COMMON_BYTEBUFFER_H_
-
-#include "common/assert.h"
-#include "common/common.h"
-
-#include <algorithm>
-#include <cstddef> // IWYU pragma: export
-#include <cstring>
-#include <utility>
-#include <vector>
-
-namespace Common {
-
-using Byte = std::byte;
-
-class ByteBuffer {
-public:
-	using DataType = std::vector<Byte>;
-
-	static constexpr uint32_t INVALID_INDEX = static_cast<uint32_t>(-1);
-
-	using iterator       = DataType::iterator;
-	using const_iterator = DataType::const_iterator;
-
-	ByteBuffer()                  = default;
-	ByteBuffer(const ByteBuffer&) = default;
-	ByteBuffer(ByteBuffer&&)      = default;
-
-	ByteBuffer& operator=(const ByteBuffer&) = default;
-	ByteBuffer& operator=(ByteBuffer&&)      = default;
-
-	explicit ByteBuffer(uint32_t size): m_data(size) {}
-
-	ByteBuffer(std::initializer_list<uint8_t> list) {
-		for (const uint8_t& e: list) {
-			m_data.emplace_back(static_cast<Byte>(e));
-		}
-	}
-
-	explicit ByteBuffer(const void* buf, uint32_t size)
-	    : ByteBuffer(size) // @suppress("Ambiguous problem")
-	{
-		EXIT_IF(buf == nullptr && size != 0);
-
-		if (size != 0) {
-			std::memcpy(m_data.data(), buf, size);
-		}
-	}
-
-	[[nodiscard]] uint32_t Size() const { return static_cast<uint32_t>(m_data.size()); }
-	[[nodiscard]] bool     IsEmpty() const { return m_data.empty(); }
-
-	void Clear() { m_data.clear(); }
-
-	void Add(const Byte& val) { m_data.push_back(val); }
-	void Add(Byte&& val) { m_data.push_back(std::move(val)); }
-
-	void Add(const Byte* val, uint32_t num) {
-		EXIT_IF(val == nullptr && num != 0);
-		m_data.insert(m_data.end(), val, val + num);
-	}
-
-	void Add(const ByteBuffer& v) {
-		if (this == &v) {
-			const ByteBuffer copy(v);
-			Add(copy.GetDataConst(), copy.Size());
-			return;
-		}
-		Add(v.GetDataConst(), v.Size());
-	}
-
-	void InsertAt(uint32_t index, const Byte& val) {
-		EXIT_IF(index > Size());
-		m_data.insert(m_data.begin() + index, val);
-	}
-
-	void InsertAt(uint32_t index, Byte&& val) {
-		EXIT_IF(index > Size());
-		m_data.insert(m_data.begin() + index, std::move(val));
-	}
-
-	[[nodiscard]] uint32_t Find(const Byte& t) const {
-		const auto it = std::find(m_data.begin(), m_data.end(), t);
-		return (it == m_data.end() ? INVALID_INDEX
-		                           : static_cast<uint32_t>(std::distance(m_data.begin(), it)));
-	}
-
-	[[nodiscard]] bool Contains(const Byte& t) const { return Find(t) != INVALID_INDEX; }
-
-	bool Remove(const Byte& t) {
-		const auto it = std::find(m_data.begin(), m_data.end(), t);
-		if (it == m_data.end()) {
-			return false;
-		}
-		m_data.erase(it);
-		return true;
-	}
-
-	bool RemoveAt(uint32_t index, uint32_t count = 1) {
-		if (index >= Size() || count == 0) {
-			return false;
-		}
-
-		const uint32_t erase_count = std::min(count, Size() - index);
-		m_data.erase(m_data.begin() + index, m_data.begin() + index + erase_count);
-		return true;
-	}
-
-	[[nodiscard]] bool IndexValid(uint32_t index) const { return index < Size(); }
-
-	Byte& operator[](uint32_t index) {
-		EXIT_IF(index >= Size());
-		return m_data[index];
-	}
-
-	const Byte& operator[](uint32_t index) const {
-		EXIT_IF(index >= Size());
-		return m_data[index];
-	}
-
-	[[nodiscard]] const Byte& At(uint32_t index) const {
-		EXIT_IF(index >= Size());
-		return m_data[index];
-	}
-
-	Byte*                     GetData() { return m_data.data(); }
-	[[nodiscard]] const Byte* GetData() const { return m_data.data(); }
-	[[nodiscard]] const Byte* GetDataConst() const { return m_data.data(); }
-
-	void Sort() {
-		std::sort(m_data.begin(), m_data.end(), [](Byte a, Byte b) {
-			return std::to_integer<uint8_t>(a) < std::to_integer<uint8_t>(b);
-		});
-	}
-
-	template <typename OP>
-	void Sort(OP&& comp_func) {
-		std::sort(m_data.begin(), m_data.end(), std::forward<OP>(comp_func));
-	}
-
-	bool operator==(const ByteBuffer& s) const { return m_data == s.m_data; }
-	bool operator!=(const ByteBuffer& s) const { return !(*this == s); }
-
-	iterator begin() { return m_data.begin(); } // NOLINT(readability-identifier-naming)
-	iterator end() { return m_data.end(); }     // NOLINT(readability-identifier-naming)
-	[[nodiscard]] const_iterator begin() const {
-		return m_data.begin();
-	} // NOLINT(readability-identifier-naming)
-	[[nodiscard]] const_iterator end() const {
-		return m_data.end();
-	} // NOLINT(readability-identifier-naming)
-	[[nodiscard]] const_iterator cbegin() const {
-		return m_data.cbegin();
-	} // NOLINT(readability-identifier-naming)
-	[[nodiscard]] const_iterator cend() const {
-		return m_data.cend();
-	} // NOLINT(readability-identifier-naming)
-
-private:
-	DataType m_data;
-};
-
-} // namespace Common
-
-#endif /* KYTY_COMMON_BYTEBUFFER_H_ */

src/common/debug.cpp

@@ -24,8 +24,4 @@ std::string Debug::GetLinker() {
 #endif
 }
 
-std::string Debug::GetBitness() {
-	return "64";
-}
-
 } // namespace Common

src/common/debug.h

@@ -10,7 +10,6 @@ namespace Common {
 namespace Debug {
 std::string GetCompiler();
 std::string GetLinker();
-std::string GetBitness();
 } // namespace Debug
 
 } // namespace Common

src/common/file.cpp

@@ -136,10 +136,6 @@ bool File::OpenInMem(void* buf, uint32_t buf_size) {
 	return true;
 }
 
-bool File::OpenInMem(ByteBuffer& buf) {
-	return OpenInMem(buf.GetData(), buf.Size());
-}
-
 bool File::CreateInMem() {
 	EXIT_IF(m_p->f != nullptr);
 
@@ -249,88 +245,78 @@ void File::Printf(const char* format, ...) {
 	Write(s.data(), static_cast<uint32_t>(s.size()));
 }
 
+static std::filesystem::path WithoutTrailingSeparator(
+	const std::filesystem::path& path) {
+		if (!path.empty() && !path.has_filename() &&
+			path != path.root_path()) {
+			return path.parent_path();
+		}
+
+		return path;
+}
+
+
 bool File::IsDirectoryExisting(const std::filesystem::path& path) {
-	auto path_str = PathToGenericString(path);
-	return SysFileIsDirectoryExisting(
-	    Common::EndsWith(path_str, "/") ? Common::RemoveLast(path_str, 1) : path_str);
+	return SysFileIsDirectoryExisting(WithoutTrailingSeparator(path));
 }
 
 bool File::IsFileExisting(const std::filesystem::path& name) {
 	return SysFileIsFileExisting(name);
 }
 
-bool File::CreateDirectory(
-    const std::filesystem::path& path) // @suppress("Member declaration not found")
-{
-	auto path_str = PathToGenericString(path);
-	return SysFileCreateDirectory(Common::EndsWith(path_str, "/") ? Common::RemoveLast(path_str, 1)
-	                                                              : path_str);
+bool File::CreateDirectory(const std::filesystem::path& path) {
+    return SysFileCreateDirectory(
+        WithoutTrailingSeparator(path));
 }
 
 bool File::DeleteDirectory(const std::filesystem::path& path) {
-	auto path_str = PathToGenericString(path);
-	return SysFileDeleteDirectory(Common::EndsWith(path_str, "/") ? Common::RemoveLast(path_str, 1)
-	                                                              : path_str);
+    return SysFileDeleteDirectory(
+        WithoutTrailingSeparator(path));
 }
 
 bool File::CreateDirectories(const std::filesystem::path& path) {
-	std::string real_path = Common::ReplaceChar(PathToGenericString(path), '\\', '/');
+    const auto normalized_path =
+        WithoutTrailingSeparator(path);
 
-	std::vector<std::string> list = Common::Split(real_path, "/");
+    auto current_path = normalized_path.root_path();
 
-	std::string p;
+    for (const auto& component:
+         normalized_path.relative_path()) {
+        current_path /= component;
 
-	for (uint32_t si = 0; si < list.size(); si++) {
-		const std::string& s = list[si];
+        if (IsDirectoryExisting(current_path)) {
+            continue;
+        }
 
-		if (si != 0 || Common::StartsWith(real_path, "/")) {
-			p += "/";
-		}
+        if (!CreateDirectory(current_path)) {
+            return false;
+        }
+    }
 
-		p += s;
-
-		if (IsDirectoryExisting(p)) {
-			continue;
-		}
-
-		if (!CreateDirectory(p)) // @suppress("Invalid arguments")
-		{
-			return false;
-		}
-	}
-
-	return true;
+    return true;
 }
 
 bool File::DeleteDirectories(const std::filesystem::path& path) {
-	std::string real_path = Common::ReplaceChar(PathToGenericString(path), '\\', '/');
+    const auto normalized_path =
+        WithoutTrailingSeparator(path);
 
-	std::vector<std::string> list = Common::Split(real_path, "/");
+    std::vector<std::filesystem::path> directories;
+    auto current_path = normalized_path.root_path();
 
-	std::string              p;
-	std::vector<std::string> list2;
+    for (const auto& component:
+         normalized_path.relative_path()) {
+        current_path /= component;
+        directories.push_back(current_path);
+    }
 
-	for (uint32_t si = 0; si < list.size(); si++) {
-		const std::string& s = list[si];
+    for (auto it = directories.rbegin();
+         it != directories.rend(); ++it) {
+        if (!DeleteDirectory(*it)) {
+            return false;
+        }
+    }
 
-		if (si != 0 || Common::StartsWith(real_path, "/")) {
-			p += "/";
-		}
-
-		p += s;
-
-		list2.push_back(p);
-	}
-
-	uint32_t num = list2.size();
-
-	for (uint32_t si = num - 1; si < num; si--) {
-		if (!DeleteDirectory(list2[si])) {
-			return false;
-		}
-	}
-
-	return true;
+    return true;
 }
 
 bool File::DeleteFile(
@@ -345,7 +331,7 @@ bool File::Flush() {
 	return SysFileFlush(*m_p->f);
 }
 
-ByteBuffer File::ReadWholeBuffer() {
+std::vector<std::byte> File::ReadWholeBuffer() {
 	EXIT_IF(IsInvalid());
 	EXIT_IF(Tell() != 0);
 
@@ -353,25 +339,14 @@ ByteBuffer File::ReadWholeBuffer() {
 
 	EXIT_IF((s >> 32u) != 0);
 
-	ByteBuffer buf(s);
+	const auto            read_size = static_cast<uint32_t>(s);
+	std::vector<std::byte> buf(read_size);
 
-	Read(buf.GetData(), s);
+	Read(buf.data(), read_size);
 
 	return buf;
 }
 
-ByteBuffer File::Read(uint32_t size) {
-	ByteBuffer buf(size);
-	uint32_t   b = 0;
-	Read(buf.GetData(), size, &b);
-	buf.RemoveAt(b, size - b);
-	return buf;
-}
-
-void File::Write(const ByteBuffer& buf, uint32_t* bytes_written) {
-	Write(buf.GetDataConst(), buf.Size(), bytes_written);
-}
-
 DateTime File::GetLastAccessTimeUTC(const std::filesystem::path& name) {
 	SysTimeStruct t {};
 	SysFileToSystemTimeUtc(SysFileGetLastAccessTimeUtc(name), t);
@@ -475,13 +450,7 @@ std::vector<File::FindInfo> File::FindFiles(const std::filesystem::path& path) {
 
 	SysFileFindFiles(path, files);
 
-	auto     path_str = PathToGenericString(path);
-	uint32_t len      = static_cast<uint32_t>(path_str.size());
-
-	if (!Common::EndsWith(path_str, "/") && !Common::EndsWith(path_str, "\\")) {
-		len++;
-	}
-
+	const auto normalized_path = path.lexically_normal();
 	std::vector<File::FindInfo> ret;
 	ret.reserve(files.size());
 
@@ -489,7 +458,7 @@ std::vector<File::FindInfo> File::FindFiles(const std::filesystem::path& path) {
 		File::FindInfo r {};
 
 		r.path_with_name     = f.path_with_name;
-		r.rel_path_with_name = Common::Mid(PathToGenericString(f.path_with_name), len);
+		r.rel_path_with_name = f.path_with_name.lexically_normal().lexically_relative(normalized_path);
 		r.size               = f.size;
 
 		SysTimeStruct at {};

src/common/file.h

@@ -1,10 +1,10 @@
 #ifndef KYTY_COMMON_FILE_H_
 #define KYTY_COMMON_FILE_H_
 
-#include "common/byteBuffer.h"
 #include "common/common.h"
 #include "common/dateTime.h"
 
+#include <cstddef>
 #include <filesystem>
 #include <memory>
 #include <string>
@@ -51,7 +51,6 @@ public:
 	bool Create(const std::filesystem::path& name);
 	bool Open(const std::filesystem::path& name, Mode mode);
 	bool OpenInMem(void* buf, uint32_t buf_size);
-	bool OpenInMem(ByteBuffer& buf); // NOLINT(google-runtime-references)
 	bool CreateInMem();
 
 	void Close();
@@ -73,14 +72,12 @@ public:
 
 	void GetLastAccessAndWriteTimeUTC(DateTime* access, DateTime* write);
 
-	void       Read(void* data, uint32_t size, uint32_t* bytes_read = nullptr);
-	ByteBuffer Read(uint32_t size);
-	void       Write(const void* data, uint32_t size, uint32_t* bytes_written = nullptr);
-	void       Write(const ByteBuffer& buf, uint32_t* bytes_written = nullptr);
+	void Read(void* data, uint32_t size, uint32_t* bytes_read = nullptr);
+	void Write(const void* data, uint32_t size, uint32_t* bytes_written = nullptr);
 
 	void Printf(const char* format, ...) KYTY_FORMAT_PRINTF(2, 3);
 
-	ByteBuffer ReadWholeBuffer();
+	std::vector<std::byte> ReadWholeBuffer();
 
 	static uint64_t Size(const std::filesystem::path& name);
 

src/common/hash.h

@@ -54,103 +54,6 @@ inline uint32_t hash(const void* key, uint32_t key_len) {
 	return hash;
 }
 
-inline uint32_t hash8(uint8_t key) {
-	uint32_t hash = 0;
-
-	uint8_t* ptr = &key;
-
-	hash += ptr[0];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-
-	hash += (hash << 3u);
-	hash ^= (hash >> 11u);
-	hash += (hash << 15u);
-
-	return hash;
-}
-
-inline uint32_t hash16(uint16_t key) {
-	uint32_t hash = 0;
-
-	auto* ptr = reinterpret_cast<uint8_t*>(&key);
-
-	hash += ptr[0];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[1];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-
-	hash += (hash << 3u);
-	hash ^= (hash >> 11u);
-	hash += (hash << 15u);
-
-	return hash;
-}
-
-inline uint32_t hash32(uint32_t key) {
-	uint32_t hash = 0;
-
-	auto* ptr = reinterpret_cast<uint8_t*>(&key);
-
-	hash += ptr[0];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[1];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[2];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[3];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-
-	hash += (hash << 3u);
-	hash ^= (hash >> 11u);
-	hash += (hash << 15u);
-
-	return hash;
-}
-
-inline uint32_t hash64(uint64_t key) {
-	uint32_t hash = 0;
-
-	auto* ptr = reinterpret_cast<uint8_t*>(&key);
-
-	hash += ptr[0];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[1];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[2];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[3];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[4];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[5];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[6];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-	hash += ptr[7];
-	hash += (hash << 10u);
-	hash ^= (hash >> 6u);
-
-	hash += (hash << 3u);
-	hash ^= (hash >> 11u);
-	hash += (hash << 15u);
-
-	return hash;
-}
-
 } // namespace Common
 
 #endif /* KYTY_COMMON_HASH_H_ */

src/common/logging/log.cpp

@@ -3,7 +3,6 @@
 
 #include "common/assert.h"
 #include "common/emulatorConfig.h"
-#include "common/stringUtils.h"
 
 #include <cstdio>
 #include <filesystem>
@@ -47,7 +46,7 @@ std::shared_ptr<spdlog::logger> MakeFileLogger(std::string                  name
 	}
 
 	auto sink =
-	    std::make_shared<spdlog::sinks::basic_file_sink_mt>(Common::PathToString(path), true);
+	    std::make_shared<spdlog::sinks::basic_file_sink_mt>(path.native(), true);
 	return MakeLogger(std::move(name), std::move(sink));
 }
 

src/common/magicEnum.h

@@ -1,42 +0,0 @@
-#ifndef KYTY_COMMON_MAGICENUM_H_
-#define KYTY_COMMON_MAGICENUM_H_
-
-#include "common/common.h"
-#include "common/stringUtils.h"
-#include "magic_enum.hpp" // IWYU pragma: export
-
-namespace Common {
-
-template <typename E>
-inline std::string EnumName(E v) {
-	auto str = magic_enum::enum_name(v);
-	return std::string(str.data(), str.length());
-}
-
-template <typename E>
-inline std::string EnumName8(E v) {
-	auto str = magic_enum::enum_name(v);
-	return std::string(str.data(), str.length());
-}
-
-template <typename E>
-inline E EnumValue(const std::string& str, E default_value) {
-	auto v = magic_enum::enum_cast<E>(str.c_str());
-	if (v.has_value()) {
-		return v.value();
-	}
-	return default_value;
-}
-
-#define KYTY_ENUM_RANGE(e, mx, mn)                                                                 \
-	namespace magic_enum::customize {                                                              \
-	template <>                                                                                    \
-	struct enum_range<e> {                                                                         \
-		static constexpr int min = (mx);                                                           \
-		static constexpr int max = (mn);                                                           \
-	};                                                                                             \
-	}
-
-} // namespace Common
-
-#endif /* KYTY_COMMON_MAGICENUM_H_ */

src/common/platform/sysFileIO.h

@@ -52,7 +52,6 @@ uint64_t    SysFileTell(sys_file_t& f);                    // NOLINT(google-runt
 bool        SysFileTruncate(sys_file_t& f, uint64_t size); // NOLINT(google-runtime-references)
 bool        SysFileUnlink(sys_file_t&                  f,
                           const std::filesystem::path& name); // NOLINT(google-runtime-references)
-void        SysFileWrite(uint32_t n, sys_file_t& f);          // NOLINT(google-runtime-references)
 uint64_t    SysFileSize(const std::filesystem::path& file_name);
 bool        SysFileIsError(sys_file_t& f); // NOLINT(google-runtime-references)
 bool        SysFileIsDirectoryExisting(const std::filesystem::path& path);

src/common/platform/sysLinuxFileIO.cpp

@@ -11,6 +11,7 @@
 
 #include <cerrno>
 #include <cstdlib>
+#include <cstring>
 #include <dirent.h>
 #include <fcntl.h>
 #include <filesystem>
@@ -151,10 +152,6 @@ void SysFileWrite(const void* data, uint32_t size, sys_file_t& f, uint32_t* byte
 	}
 }
 
-void SysFileWrite(uint32_t n, sys_file_t& f) {
-	SysFileWrite(&n, 4, f);
-}
-
 sys_file_t* SysFileCreate(const std::filesystem::path& file_name) {
 	auto* ret = new sys_file_t;
 

src/common/platform/sysLinuxVirtual.cpp

@@ -5,7 +5,6 @@
 #else
 
 #include "common/assert.h"
-#include "common/platform/sysVirtual.h"
 #include "common/virtualMemory.h"
 
 #include <atomic>
@@ -28,12 +27,12 @@
 #define KYTY_FIXED_NOREPLACE
 #endif
 
-namespace Common {
+namespace Common::VirtualMemory {
 
 static pthread_mutex_t              g_virtual_mutex {};
 static std::map<uintptr_t, size_t>* g_allocs = nullptr;
 
-void SysVirtualInit() {
+void Init() {
 	pthread_mutexattr_t attr {};
 
 	pthread_mutexattr_init(&attr);
@@ -48,19 +47,19 @@ void SysVirtualInit() {
 	g_allocs = new std::map<uintptr_t, size_t>;
 }
 
-static int get_protection_flag(VirtualMemory::Mode mode) {
+static int get_protection_flag(Mode mode) {
 	int protect = PROT_NONE;
 	switch (mode) {
-		case VirtualMemory::Mode::Read: protect = PROT_READ; break;
-		case VirtualMemory::Mode::Write:
-		case VirtualMemory::Mode::ReadWrite: protect = PROT_READ | PROT_WRITE; break; // NOLINT
-		case VirtualMemory::Mode::Execute: protect = PROT_EXEC; break;
-		case VirtualMemory::Mode::ExecuteRead: protect = PROT_EXEC | PROT_READ; break; // NOLINT
-		case VirtualMemory::Mode::ExecuteWrite:
-		case VirtualMemory::Mode::ExecuteReadWrite:
+		case Mode::Read: protect = PROT_READ; break;
+		case Mode::Write:
+		case Mode::ReadWrite: protect = PROT_READ | PROT_WRITE; break; // NOLINT
+		case Mode::Execute: protect = PROT_EXEC; break;
+		case Mode::ExecuteRead: protect = PROT_EXEC | PROT_READ; break; // NOLINT
+		case Mode::ExecuteWrite:
+		case Mode::ExecuteReadWrite:
 			protect = PROT_EXEC | PROT_WRITE | PROT_READ;
 			break; // NOLINT
-		case VirtualMemory::Mode::NoAccess:
+		case Mode::NoAccess:
 		default: protect = PROT_NONE; break;
 	}
 	return protect;
@@ -130,7 +129,7 @@ static void* map_anonymous(uintptr_t addr, size_t size, int protect, int flags)
 	return mmap(nullptr, size, protect, flags, -1, 0); // NOLINT
 }
 
-uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
 	EXIT_IF(g_allocs == nullptr);
 
 	auto addr = static_cast<uintptr_t>(address);
@@ -154,8 +153,7 @@ static uintptr_t align_up(uintptr_t addr, uint64_t alignment) {
 	return (addr + alignment - 1) & ~(alignment - 1);
 }
 
-uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
-                                uint64_t alignment) {
+uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t alignment) {
 	if (alignment == 0) {
 		return 0;
 	}
@@ -224,7 +222,7 @@ uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::
 	}
 
 	if (ptr == MAP_FAILED) {
-		return SysVirtualAllocAligned(address, size, mode, alignment << 1u);
+		return AllocAligned(address, size, mode, alignment << 1u);
 	}
 
 	pthread_mutex_lock(&g_virtual_mutex);
@@ -281,7 +279,7 @@ static bool is_mapped(void* ptr, size_t length) {
 }
 #endif
 
-bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
 	EXIT_IF(g_allocs == nullptr);
 
 	auto addr    = static_cast<uintptr_t>(address);
@@ -318,15 +316,15 @@ bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode m
 	return false;
 }
 
-bool SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
-	return SysVirtualProtect(address, size, mode);
+bool Commit(uint64_t address, uint64_t size, Mode mode) {
+	return Protect(address, size, mode);
 }
 
-uint64_t SysVirtualReserve(uint64_t address, uint64_t size) {
-	return SysVirtualReserveAligned(address, size, 1);
+uint64_t Reserve(uint64_t address, uint64_t size) {
+	return ReserveAligned(address, size, 1);
 }
 
-uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
+uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
 	if (alignment == 0) {
 		return 0;
 	}
@@ -395,7 +393,7 @@ uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alig
 	}
 
 	if (ptr == MAP_FAILED) {
-		return SysVirtualReserveAligned(address, size, alignment << 1u);
+		return ReserveAligned(address, size, alignment << 1u);
 	}
 
 	pthread_mutex_lock(&g_virtual_mutex);
@@ -405,7 +403,7 @@ uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alig
 	return ret_addr;
 }
 
-bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
+bool ReserveFixed(uint64_t address, uint64_t size) {
 	EXIT_IF(g_allocs == nullptr);
 
 	auto addr = static_cast<uintptr_t>(address);
@@ -441,9 +439,9 @@ bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
 	return false;
 }
 
-bool SysVirtualDecommit(uint64_t address, uint64_t size) {
+bool Decommit(uint64_t address, uint64_t size) {
 	// Drop physical pages while preserving the reservation.
-	if (!SysVirtualProtect(address, size, VirtualMemory::Mode::NoAccess)) {
+	if (!Protect(address, size, Mode::NoAccess)) {
 		return false;
 	}
 
@@ -467,7 +465,7 @@ bool SysVirtualDecommit(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool SysVirtualFree(uint64_t address) {
+bool Free(uint64_t address) {
 	EXIT_IF(g_allocs == nullptr);
 	size_t size = 0;
 
@@ -487,7 +485,7 @@ bool SysVirtualFree(uint64_t address) {
 	return munmap(reinterpret_cast<void*>(addr), size) == 0;
 }
 
-bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
+bool FreeRange(uint64_t address, uint64_t size) {
 	EXIT_IF(g_allocs == nullptr);
 	if (size == 0 || (address & 0xfffu) != 0 || (size & 0xfffu) != 0) {
 		return false;
@@ -543,7 +541,7 @@ bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool Protect(uint64_t address, uint64_t size, Mode mode) {
 	const auto addr       = static_cast<uintptr_t>(address);
 	const auto page_start = addr >> 12u;
 	const auto page_end   = (addr + size - 1) >> 12u;
@@ -551,10 +549,10 @@ bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode
 	                get_protection_flag(mode)) == 0;
 }
 
-bool SysVirtualFlushInstructionCache(uint64_t /*address*/, uint64_t /*size*/) {
+bool FlushInstructionCache(uint64_t /*address*/, uint64_t /*size*/) {
 	return true;
 }
 
-} // namespace Common
+} // namespace Common::VirtualMemory
 
 #endif

src/common/platform/sysSwapByteOrder.h

@@ -1,53 +0,0 @@
-#ifndef KYTY_COMMON_PLATFORM_SYSSWAPBYTEORDER_H_
-#define KYTY_COMMON_PLATFORM_SYSSWAPBYTEORDER_H_
-
-#include "common/common.h"
-
-#include <type_traits>
-
-inline uint16_t SwapByteOrder16(uint16_t value) {
-	uint16_t hi = value << 8u;
-	uint16_t lo = value >> 8u;
-	return hi | lo;
-}
-
-inline uint32_t SwapByteOrder32(uint32_t value) {
-	return __builtin_bswap32(value);
-}
-
-inline uint64_t SwapByteOrder64(uint64_t value) {
-	return __builtin_bswap64(value);
-}
-
-template <typename T>
-inline void SwapByteOrder(T& x) {
-	if (sizeof(x) == 2) {
-		if (std::is_signed_v<T>) {
-			x = static_cast<std::make_signed_t<T>>(
-			    SwapByteOrder16(static_cast<std::make_unsigned_t<uint16_t>>(x)));
-		} else {
-			x = SwapByteOrder16(x);
-		}
-	}
-	if (sizeof(x) == 4) {
-		if (std::is_signed_v<T>) {
-			x = static_cast<std::make_signed_t<T>>(
-			    SwapByteOrder32(static_cast<std::make_unsigned_t<uint32_t>>(x)));
-		} else {
-			x = SwapByteOrder32(x);
-		}
-	}
-	if (sizeof(x) == 8) {
-		if (std::is_signed_v<T>) {
-			x = static_cast<std::make_signed_t<T>>(
-			    SwapByteOrder64(static_cast<std::make_unsigned_t<uint64_t>>(x)));
-		} else {
-			x = SwapByteOrder64(x);
-		}
-	}
-}
-
-template <typename T>
-inline void NoSwapByteOrder(T& x) {}
-
-#endif /* KYTY_COMMON_PLATFORM_SYSSWAPBYTEORDER_H_ */

src/common/platform/sysTimer.h

@@ -151,10 +151,6 @@ inline void SysTimeTToSystem(time_t t, SysTimeStruct& s) {
 	SysFileToSystemTimeUtc(ft, s);
 }
 
-inline time_t SysTimegm(struct tm* tm) {
-	return timegm(tm);
-}
-
 inline void SysSystemToFileTimeUtc(const SysTimeStruct& f, SysFileTimeStruct& t) {
 	struct tm i {};
 
@@ -165,7 +161,7 @@ inline void SysSystemToFileTimeUtc(const SysTimeStruct& f, SysFileTimeStruct& t)
 	i.tm_min  = f.Minute;
 	i.tm_sec  = f.Second;
 
-	t.is_invalid = (f.is_invalid || (t.time = SysTimegm(&i)) == static_cast<time_t>(-1));
+	t.is_invalid = (f.is_invalid || (t.time = ::timegm(&i)) == static_cast<time_t>(-1));
 }
 
 // Retrieves the current local date and time.

src/common/platform/sysVirtual.h

@@ -1,27 +0,0 @@
-#ifndef KYTY_COMMON_PLATFORM_SYSVIRTUAL_H_
-#define KYTY_COMMON_PLATFORM_SYSVIRTUAL_H_
-
-#include "common/common.h"
-#include "common/virtualMemory.h"
-
-namespace Common {
-
-void SysVirtualInit();
-
-uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
-uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
-                                uint64_t alignment);
-bool     SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
-bool     SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
-uint64_t SysVirtualReserve(uint64_t address, uint64_t size);
-uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alignment);
-bool     SysVirtualReserveFixed(uint64_t address, uint64_t size);
-bool     SysVirtualDecommit(uint64_t address, uint64_t size);
-bool     SysVirtualFree(uint64_t address);
-bool     SysVirtualFreeRange(uint64_t address, uint64_t size);
-bool     SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode);
-bool     SysVirtualFlushInstructionCache(uint64_t address, uint64_t size);
-
-} // namespace Common
-
-#endif /* KYTY_COMMON_PLATFORM_SYSVIRTUAL_H_ */

src/common/platform/sysWindowsFileIO.cpp

@@ -14,6 +14,7 @@
 #include "common/stringUtils.h"
 
 #include <cstdlib>
+#include <cstring>
 #include <vector>
 
 // NOLINTNEXTLINE(readability-identifier-naming)
@@ -135,10 +136,6 @@ void SysFileWrite(const void* data, uint32_t size, sys_file_t& f, uint32_t* byte
 	}
 }
 
-void SysFileWrite(uint32_t n, sys_file_t& f) {
-	SysFileWrite(&n, 4, f);
-}
-
 sys_file_t* SysFileCreate(const std::filesystem::path& file_name) {
 	auto* ret = new sys_file_t;
 
@@ -480,18 +477,10 @@ bool SysFileSetLastAccessAndWriteTimeUtc(const std::filesystem::path& name,
 }
 
 void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_find_t>& out) {
-	std::string real_path = Common::ReplaceChar(Common::PathToGenericString(path), '\\', '/');
-	if (!Common::EndsWith(real_path, "/")) {
-		real_path += "/";
-	}
+	const auto pattern = path / L"*";
 
-	std::string pattern = real_path + "*";
-
-	HANDLE           h = nullptr;
-	WIN32_FIND_DATAW data;
-
-	auto wide_pattern = std::filesystem::path(pattern).wstring();
-	h                 = FindFirstFileW(wide_pattern.c_str(), &data);
+	WIN32_FIND_DATAW data {};
+	HANDLE h = FindFirstFileW(pattern.c_str(), &data);
 
 	if (h == INVALID_HANDLE_VALUE) {
 		return;
@@ -499,18 +488,19 @@ void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_fi
 
 	do {
 		std::filesystem::path file_name(data.cFileName);
-		auto                  file_name_str = Common::PathToString(file_name);
 
-		if (file_name_str == "." || file_name_str == "..") {
+		if (file_name == L"." || file_name == L"..") {
 			continue;
 		}
 
+		const auto full_path = path / file_name;
+
 		if ((data.dwFileAttributes & static_cast<DWORD>(FILE_ATTRIBUTE_DIRECTORY)) != 0u) {
-			SysFileFindFiles(std::filesystem::path(real_path) / file_name, out);
+			SysFileFindFiles(full_path, out);
 		} else {
 			sys_file_find_t r {};
 
-			r.path_with_name              = std::filesystem::path(real_path) / file_name;
+			r.path_with_name              = full_path;
 			r.size                        = (static_cast<uint64_t>(data.nFileSizeHigh) << 32u) +
 			                                static_cast<uint64_t>(data.nFileSizeLow);
 			r.last_access_time.is_invalid = false;
@@ -518,7 +508,7 @@ void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_fi
 			r.last_write_time.is_invalid  = false;
 			r.last_write_time.time        = data.ftLastWriteTime;
 
-			out.push_back(r);
+			out.push_back(std::move(r));
 		}
 
 	} while (FindNextFileW(h, &data) != 0);
@@ -527,18 +517,10 @@ void SysFileFindFiles(const std::filesystem::path& path, std::vector<sys_file_fi
 }
 
 void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entry_t>& out) {
-	std::string real_path = Common::ReplaceChar(Common::PathToGenericString(path), '\\', '/');
-	if (!Common::EndsWith(real_path, "/")) {
-		real_path += "/";
-	}
+	const auto pattern = path / L"*";
 
-	std::string pattern = real_path + "*";
-
-	HANDLE           h = nullptr;
-	WIN32_FIND_DATAW data;
-
-	auto wide_pattern = std::filesystem::path(pattern).wstring();
-	h                 = FindFirstFileW(wide_pattern.c_str(), &data);
+	WIN32_FIND_DATAW data {};
+	HANDLE h = FindFirstFileW(pattern.c_str(), &data);
 
 	if (h == INVALID_HANDLE_VALUE) {
 		return;
@@ -552,7 +534,7 @@ void SysFileGetDents(const std::filesystem::path& path, std::vector<sys_dir_entr
 		r.is_file = ((data.dwFileAttributes & static_cast<DWORD>(FILE_ATTRIBUTE_DIRECTORY)) == 0u);
 		r.name    = Common::PathToString(file_name);
 
-		out.push_back(r);
+		out.push_back(std::move(r));
 
 	} while (FindNextFileW(h, &data) != 0);
 

src/common/platform/sysWindowsVirtual.cpp

@@ -10,7 +10,6 @@
 #include <windows.h> // IWYU pragma: keep
 
 #include "common/assert.h"
-#include "common/platform/sysVirtual.h"
 #include "common/virtualMemory.h"
 
 // IWYU pragma: no_include <basetsd.h>
@@ -22,33 +21,33 @@
 // IWYU pragma: no_include <winerror.h>
 // IWYU pragma: no_include <wtypes.h>
 
-namespace Common {
+namespace Common::VirtualMemory {
 
-static DWORD GetProtectionFlag(VirtualMemory::Mode mode) {
+static DWORD GetProtectionFlag(Mode mode) {
 	DWORD protect = PAGE_NOACCESS;
 	switch (mode) {
-		case VirtualMemory::Mode::Read: protect = PAGE_READONLY; break;
+		case Mode::Read: protect = PAGE_READONLY; break;
 
-		case VirtualMemory::Mode::Write:
-		case VirtualMemory::Mode::ReadWrite: protect = PAGE_READWRITE; break;
+		case Mode::Write:
+		case Mode::ReadWrite: protect = PAGE_READWRITE; break;
 
-		case VirtualMemory::Mode::Execute: protect = PAGE_EXECUTE; break;
+		case Mode::Execute: protect = PAGE_EXECUTE; break;
 
-		case VirtualMemory::Mode::ExecuteRead: protect = PAGE_EXECUTE_READ; break;
+		case Mode::ExecuteRead: protect = PAGE_EXECUTE_READ; break;
 
-		case VirtualMemory::Mode::ExecuteWrite:
-		case VirtualMemory::Mode::ExecuteReadWrite: protect = PAGE_EXECUTE_READWRITE; break;
+		case Mode::ExecuteWrite:
+		case Mode::ExecuteReadWrite: protect = PAGE_EXECUTE_READWRITE; break;
 
-		case VirtualMemory::Mode::NoAccess:
+		case Mode::NoAccess:
 		default: protect = PAGE_NOACCESS; break;
 	}
 	return protect;
 }
 
-void SysVirtualInit() {}
+void Init() {}
 
-uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
-	auto ptr = (address == 0 ? SysVirtualAllocAligned(address, size, mode, 1)
+uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
+	auto ptr = (address == 0 ? AllocAligned(address, size, mode, 1)
 	                         : reinterpret_cast<uintptr_t>(VirtualAlloc(
 	                               reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                               static_cast<DWORD>(MEM_COMMIT) | static_cast<DWORD>(MEM_RESERVE),
@@ -59,7 +58,7 @@ uint64_t SysVirtualAlloc(uint64_t address, uint64_t size, VirtualMemory::Mode mo
 		if (err != ERROR_INVALID_ADDRESS) {
 			printf("VirtualAlloc() failed: 0x%08" PRIx32 "\n", err);
 		} else {
-			return SysVirtualAllocAligned(address, size, mode, 1);
+			return AllocAligned(address, size, mode, 1);
 		}
 	}
 	return ptr;
@@ -80,8 +79,7 @@ static uint64_t AlignUp(uint64_t addr, uint64_t alignment) {
 	return (addr + alignment - 1) & ~(alignment - 1);
 }
 
-uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::Mode mode,
-                                uint64_t alignment) {
+uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t alignment) {
 	if (alignment == 0) {
 		printf("VirtualAlloc2 failed: 0x%08" PRIx32 "\n", static_cast<uint32_t>(GetLastError()));
 		return 0;
@@ -135,13 +133,13 @@ uint64_t SysVirtualAllocAligned(uint64_t address, uint64_t size, VirtualMemory::
 			printf("VirtualAlloc2(alignment = 0x%016" PRIx64 ") failed: 0x%08" PRIx32 "\n",
 			       alignment, err);
 		} else {
-			return SysVirtualAllocAligned(address, size, mode, alignment << 1u);
+			return AllocAligned(address, size, mode, alignment << 1u);
 		}
 	}
 	return ptr;
 }
 
-bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
 	auto ptr = reinterpret_cast<uintptr_t>(VirtualAlloc(
 	    reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	    static_cast<DWORD>(MEM_COMMIT) | static_cast<DWORD>(MEM_RESERVE), GetProtectionFlag(mode)));
@@ -161,7 +159,7 @@ bool SysVirtualAllocFixed(uint64_t address, uint64_t size, VirtualMemory::Mode m
 	return true;
 }
 
-bool SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool Commit(uint64_t address, uint64_t size, Mode mode) {
 	auto ptr = reinterpret_cast<uintptr_t>(
 	    VirtualAlloc(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                 static_cast<DWORD>(MEM_COMMIT), GetProtectionFlag(mode)));
@@ -179,8 +177,8 @@ bool SysVirtualCommit(uint64_t address, uint64_t size, VirtualMemory::Mode mode)
 	return true;
 }
 
-uint64_t SysVirtualReserve(uint64_t address, uint64_t size) {
-	auto ptr = (address == 0 ? SysVirtualReserveAligned(address, size, 1)
+uint64_t Reserve(uint64_t address, uint64_t size) {
+	auto ptr = (address == 0 ? ReserveAligned(address, size, 1)
 	                         : reinterpret_cast<uintptr_t>(VirtualAlloc(
 	                               reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                               static_cast<DWORD>(MEM_RESERVE), PAGE_NOACCESS)));
@@ -190,13 +188,13 @@ uint64_t SysVirtualReserve(uint64_t address, uint64_t size) {
 		if (err != ERROR_INVALID_ADDRESS) {
 			printf("VirtualAlloc(MEM_RESERVE) failed: 0x%08" PRIx32 "\n", err);
 		} else {
-			return SysVirtualReserveAligned(address, size, 1);
+			return ReserveAligned(address, size, 1);
 		}
 	}
 	return ptr;
 }
 
-uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
+uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
 	if (alignment == 0) {
 		printf("VirtualAlloc2(MEM_RESERVE) failed: 0x%08" PRIx32 "\n",
 		       static_cast<uint32_t>(GetLastError()));
@@ -250,13 +248,13 @@ uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alig
 			       "\n",
 			       alignment, err);
 		} else {
-			return SysVirtualReserveAligned(address, size, alignment << 1u);
+			return ReserveAligned(address, size, alignment << 1u);
 		}
 	}
 	return ptr;
 }
 
-bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
+bool ReserveFixed(uint64_t address, uint64_t size) {
 	auto ptr = reinterpret_cast<uintptr_t>(
 	    VirtualAlloc(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                 static_cast<DWORD>(MEM_RESERVE), PAGE_NOACCESS));
@@ -276,7 +274,7 @@ bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool SysVirtualDecommit(uint64_t address, uint64_t size) {
+bool Decommit(uint64_t address, uint64_t size) {
 	if (VirtualFree(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                MEM_DECOMMIT) == 0) {
 		printf("VirtualFree(MEM_DECOMMIT) failed: 0x%08" PRIx32 "\n",
@@ -286,7 +284,7 @@ bool SysVirtualDecommit(uint64_t address, uint64_t size) {
 	return true;
 }
 
-bool SysVirtualFree(uint64_t address) {
+bool Free(uint64_t address) {
 	if (VirtualFree(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), 0, MEM_RELEASE) ==
 	    0) {
 		printf("VirtualFree() failed: 0x%08" PRIx32 "\n", static_cast<uint32_t>(GetLastError()));
@@ -295,7 +293,7 @@ bool SysVirtualFree(uint64_t address) {
 	return true;
 }
 
-bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
+bool FreeRange(uint64_t address, uint64_t size) {
 	if (address == 0 || size == 0) {
 		return false;
 	}
@@ -313,10 +311,10 @@ bool SysVirtualFreeRange(uint64_t address, uint64_t size) {
 		}
 		current = next;
 	}
-	return current - address == size && SysVirtualFree(address);
+	return current - address == size && Free(address);
 }
 
-bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode) {
+bool Protect(uint64_t address, uint64_t size, Mode mode) {
 	DWORD old_protect = 0;
 	if (VirtualProtect(reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)), size,
 	                   GetProtectionFlag(mode), &old_protect) == 0) {
@@ -326,7 +324,7 @@ bool SysVirtualProtect(uint64_t address, uint64_t size, VirtualMemory::Mode mode
 	return true;
 }
 
-bool SysVirtualFlushInstructionCache(uint64_t address, uint64_t size) {
+bool FlushInstructionCache(uint64_t address, uint64_t size) {
 	if (::FlushInstructionCache(GetCurrentProcess(),
 	                            reinterpret_cast<LPVOID>(static_cast<uintptr_t>(address)),
 	                            size) == 0) {
@@ -337,6 +335,6 @@ bool SysVirtualFlushInstructionCache(uint64_t address, uint64_t size) {
 	return true;
 }
 
-} // namespace Common
+} // namespace Common::VirtualMemory
 
 #endif

src/common/stringUtils.h

@@ -1,7 +1,6 @@
 #ifndef KYTY_COMMON_STRING_UTILS_H_
 #define KYTY_COMMON_STRING_UTILS_H_
 
-#include "common/byteBuffer.h"
 #include "common/common.h"
 
 #include <algorithm>
@@ -9,7 +8,6 @@
 #include <charconv>
 #include <codecvt>
 #include <filesystem>
-#include <fmt/format.h>
 #include <locale>
 #include <string>
 #include <string_view>
@@ -18,8 +16,6 @@
 
 namespace Common {
 
-constexpr uint32_t FIND_INVALID_INDEX = static_cast<uint32_t>(-1);
-
 inline std::string PathToString(const std::filesystem::path& path) {
 #if defined(__cpp_char8_t)
 	auto u8 = path.u8string();
@@ -38,8 +34,12 @@ inline std::string PathToGenericString(const std::filesystem::path& path) {
 #endif
 }
 
-inline bool IsSpace(char c) {
-	return std::isspace(static_cast<unsigned char>(c)) != 0;
+inline std::filesystem::path PathFromUtf8(std::string_view text) {
+#if defined(__cpp_char8_t)
+	return std::filesystem::path(std::u8string(text.begin(), text.end()));
+#else
+	return std::filesystem::u8path(text.begin(), text.end());
+#endif
 }
 
 inline bool EqualNoCase(std::string_view lhs, std::string_view rhs) {
@@ -63,65 +63,6 @@ inline std::string ToLower(std::string_view text) {
 	return ret;
 }
 
-inline bool ContainsStr(std::string_view text, std::string_view needle) {
-	return text.find(needle) != std::string_view::npos;
-}
-
-inline bool StartsWith(std::string_view text, std::string_view prefix) {
-	return text.starts_with(prefix);
-}
-
-inline bool StartsWith(std::string_view text, char prefix) {
-	return !text.empty() && text.front() == prefix;
-}
-
-inline bool EndsWith(std::string_view text, std::string_view suffix) {
-	return text.ends_with(suffix);
-}
-
-inline bool EndsWith(std::string_view text, char suffix) {
-	return !text.empty() && text.back() == suffix;
-}
-
-inline bool ContainsChar(std::string_view text, char needle) {
-	return text.find(needle) != std::string_view::npos;
-}
-
-inline bool IndexValid(std::string_view text, uint32_t index) {
-	return index < text.size();
-}
-
-inline uint32_t FindIndex(std::string_view text, std::string_view needle, size_t from = 0) {
-	const auto pos = text.find(needle, from);
-	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
-}
-
-inline uint32_t FindIndex(std::string_view text, char needle, size_t from = 0) {
-	const auto pos = text.find(needle, from);
-	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
-}
-
-inline uint32_t FindLastIndex(std::string_view text, std::string_view needle,
-                              size_t from = FIND_INVALID_INDEX) {
-	if (needle.empty()) {
-		return 0;
-	}
-	if (from == FIND_INVALID_INDEX || from > text.size()) {
-		from = text.size();
-	}
-	const auto pos = text.rfind(needle, from);
-	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
-}
-
-inline uint32_t FindLastIndex(std::string_view text, char needle,
-                              size_t from = FIND_INVALID_INDEX) {
-	if (from == FIND_INVALID_INDEX || from > text.size()) {
-		from = text.size();
-	}
-	const auto pos = text.rfind(needle, from);
-	return pos == std::string_view::npos ? FIND_INVALID_INDEX : static_cast<uint32_t>(pos);
-}
-
 inline std::string Mid(std::string_view text, size_t first, size_t count = std::string_view::npos) {
 	if (first >= text.size()) {
 		return {};
@@ -129,49 +70,18 @@ inline std::string Mid(std::string_view text, size_t first, size_t count = std::
 	return std::string(text.substr(first, count));
 }
 
-inline std::string Left(std::string_view text, size_t count) {
-	return Mid(text, 0, count);
-}
-
-inline std::string Right(std::string_view text, size_t count) {
-	if (count >= text.size()) {
-		return std::string(text);
-	}
-	return Mid(text, text.size() - count, count);
-}
-
 inline std::string RemoveLast(std::string_view text, size_t count) {
 	if (count >= text.size()) {
 		return {};
 	}
-	return Left(text, text.size() - count);
+	return std::string(text.substr(0, text.size() - count));
 }
 
 inline std::string RemoveFirst(std::string_view text, size_t count) {
 	if (count >= text.size()) {
 		return {};
 	}
-	return Right(text, text.size() - count);
-}
-
-inline std::string TrimRight(std::string_view text) {
-	size_t end = text.size();
-	while (end > 0 && IsSpace(text[end - 1])) {
-		end--;
-	}
-	return std::string(text.substr(0, end));
-}
-
-inline std::string TrimLeft(std::string_view text) {
-	size_t first = 0;
-	while (first < text.size() && IsSpace(text[first])) {
-		first++;
-	}
-	return std::string(text.substr(first));
-}
-
-inline std::string Trim(std::string_view text) {
-	return TrimRight(TrimLeft(text));
+	return std::string(text.substr(count));
 }
 
 inline std::string ReplaceChar(std::string_view text, char old_char, char new_char) {
@@ -204,16 +114,6 @@ inline std::string FilenameWithoutDirectory(std::string_view text) {
 	return pos == std::string_view::npos ? std::string(text) : std::string(text.substr(pos + 1));
 }
 
-inline std::string FilenameWithoutExtension(std::string_view text) {
-	const auto pos = text.find_last_of('.');
-	return pos == std::string_view::npos ? std::string(text) : std::string(text.substr(0, pos));
-}
-
-inline std::string ExtensionWithoutFilename(std::string_view text) {
-	const auto pos = text.find_last_of('.');
-	return pos == std::string_view::npos ? std::string() : std::string(text.substr(pos));
-}
-
 inline std::string FixFilenameSlash(std::string_view text) {
 	return ReplaceChar(text, '\\', '/');
 }
@@ -257,65 +157,12 @@ inline std::vector<std::string> Split(std::string_view text, char sep, bool keep
 	return Split(text, std::string_view(&sep, 1), keep_empty);
 }
 
-inline std::string Concat(const std::vector<std::string>& list, std::string_view sep) {
-	std::string ret;
-	for (size_t i = 0; i < list.size(); i++) {
-		if (i != 0) {
-			ret += sep;
-		}
-		ret += list[i];
-	}
-	return ret;
-}
-
-inline std::string Concat(const std::vector<std::string>& list, char sep) {
-	return Concat(list, std::string_view(&sep, 1));
-}
-
-inline uint32_t ToUint32(std::string_view text, int base = 10) {
-	uint32_t value = 0;
-	std::from_chars(text.data(), text.data() + text.size(), value, base);
-	return value;
-}
-
-inline uint64_t ToUint64(std::string_view text, int base = 10) {
-	uint64_t value = 0;
-	std::from_chars(text.data(), text.data() + text.size(), value, base);
-	return value;
-}
-
 inline int32_t ToInt32(std::string_view text, int base = 10) {
 	int32_t value = 0;
 	std::from_chars(text.data(), text.data() + text.size(), value, base);
 	return value;
 }
 
-inline int64_t ToInt64(std::string_view text, int base = 10) {
-	int64_t value = 0;
-	std::from_chars(text.data(), text.data() + text.size(), value, base);
-	return value;
-}
-
-inline double ToDouble(std::string_view text) {
-	return std::stod(std::string(text));
-}
-
-inline float ToFloat(std::string_view text) {
-	return std::stof(std::string(text));
-}
-
-inline std::string SafeCsv(std::string_view text) {
-	const bool add_space =
-	    !text.empty() && (text.front() == '+' || text.front() == '=' || text.front() == '-');
-	if (text.find('"') != std::string_view::npos || text.find(';') != std::string_view::npos ||
-	    text.find('+') != std::string_view::npos || text.find('=') != std::string_view::npos ||
-	    text.find('-') != std::string_view::npos) {
-		return "\"" + std::string(add_space ? " " : "") +
-		       ReplaceStr(std::string(text), "\"", "\"\"") + "\"";
-	}
-	return std::string(text);
-}
-
 inline std::string Utf16ToUtf8(std::u16string_view utf16) {
 	if (utf16.empty()) {
 		return {};
@@ -324,23 +171,6 @@ inline std::string Utf16ToUtf8(std::u16string_view utf16) {
 	return convert.to_bytes(utf16.data(), utf16.data() + utf16.size());
 }
 
-inline ByteBuffer HexToBin(std::string_view text) {
-	ByteBuffer ret;
-	for (size_t i = 0; i + 1 < text.size(); i += 2) {
-		ret.Add(static_cast<Byte>((ToUint32(text.substr(i, 2), 16) & 0xffu)));
-	}
-	return ret;
-}
-
-inline std::string HexFromBin(const ByteBuffer& bin) {
-	std::string ret;
-	ret.reserve(bin.Size() * 2);
-	for (uint32_t i = 0; i < bin.Size(); i++) {
-		ret += fmt::format("{:02X}", std::to_integer<uint8_t>(bin.At(i)));
-	}
-	return ret;
-}
-
 } // namespace Common
 
 #endif /* KYTY_COMMON_STRING_UTILS_H_ */

src/common/threads.cpp

@@ -253,10 +253,6 @@ void Thread::Detach() {
 	m_thread->m_thread.detach();
 }
 
-void Thread::Sleep(uint32_t millis) {
-	std::this_thread::sleep_for(std::chrono::milliseconds(millis));
-}
-
 void Thread::SleepMicro(uint32_t micros) {
 #ifdef KYTY_WIN_CS
 	SleepHighResolution100ns(static_cast<uint64_t>(micros) * 10);

src/common/threads.h

@@ -31,7 +31,6 @@ public:
 	// The id is unique and can't be reused by another thread.
 	[[nodiscard]] int GetUniqueId() const;
 
-	static void Sleep(uint32_t millis);
 	static void SleepMicro(uint32_t micros);
 	static void SleepNano(uint64_t nanos);
 	static bool IsMainThread();

src/common/timer.cpp

@@ -62,23 +62,6 @@ double Timer::GetTimeS() const {
 	return (static_cast<double>(current_time - m_StartTime)) / static_cast<double>(m_Frequency);
 }
 
-// return time in ticks
-uint64_t Timer::GetTicks() const {
-	if (m_is_paused) {
-		return (m_PauseTime - m_StartTime);
-	}
-
-	uint64_t current_time = 0;
-	SysQueryPerformanceCounter(&current_time);
-
-	return (current_time - m_StartTime);
-}
-
-// return ticks frequency
-uint64_t Timer::GetFrequency() const {
-	return m_Frequency;
-}
-
 uint64_t Timer::QueryPerformanceFrequency() {
 	uint64_t ret = 0;
 	SysQueryPerformanceFrequency(&ret);

src/common/timer.h

@@ -24,12 +24,6 @@ public:
 	// return time in seconds
 	[[nodiscard]] double GetTimeS() const;
 
-	// return time in ticks
-	[[nodiscard]] uint64_t GetTicks() const;
-
-	// return ticks frequency
-	[[nodiscard]] uint64_t GetFrequency() const;
-
 	KYTY_CLASS_NO_COPY(Timer);
 
 	[[nodiscard]] static uint64_t QueryPerformanceFrequency();

src/common/virtualMemory.cpp

@@ -1,63 +0,0 @@
-#include "common/virtualMemory.h"
-
-#include "common/platform/sysVirtual.h"
-
-namespace Common {
-
-namespace VirtualMemory {
-
-void Init() {
-	SysVirtualInit();
-}
-
-uint64_t Alloc(uint64_t address, uint64_t size, Mode mode) {
-	return SysVirtualAlloc(address, size, mode);
-}
-
-uint64_t AllocAligned(uint64_t address, uint64_t size, Mode mode, uint64_t alignment) {
-	return SysVirtualAllocAligned(address, size, mode, alignment);
-}
-
-bool AllocFixed(uint64_t address, uint64_t size, Mode mode) {
-	return SysVirtualAllocFixed(address, size, mode);
-}
-
-bool Commit(uint64_t address, uint64_t size, Mode mode) {
-	return SysVirtualCommit(address, size, mode);
-}
-
-uint64_t Reserve(uint64_t address, uint64_t size) {
-	return SysVirtualReserve(address, size);
-}
-
-uint64_t ReserveAligned(uint64_t address, uint64_t size, uint64_t alignment) {
-	return SysVirtualReserveAligned(address, size, alignment);
-}
-
-bool ReserveFixed(uint64_t address, uint64_t size) {
-	return SysVirtualReserveFixed(address, size);
-}
-
-bool Decommit(uint64_t address, uint64_t size) {
-	return SysVirtualDecommit(address, size);
-}
-
-bool Free(uint64_t address) {
-	return SysVirtualFree(address);
-}
-
-bool FreeRange(uint64_t address, uint64_t size) {
-	return SysVirtualFreeRange(address, size);
-}
-
-bool Protect(uint64_t address, uint64_t size, Mode mode) {
-	return SysVirtualProtect(address, size, mode);
-}
-
-bool FlushInstructionCache(uint64_t address, uint64_t size) {
-	return SysVirtualFlushInstructionCache(address, size);
-}
-
-} // namespace VirtualMemory
-
-} // namespace Common

src/graphics/guest_gpu/graphicsRun.cpp

@@ -386,8 +386,12 @@ void CommandProcessor::WriteReferenceClock(uint64_t dst_address, uint32_t num_by
 	}
 	const auto value = Sync::ReadReferenceClock();
 	std::memcpy(reinterpret_cast<void*>(dst_address), &value, num_bytes);
-	LOGF("\t copy_data reference clock: dst=0x%016" PRIx64 " value=0x%016" PRIx64 " size=%u\n",
-	     dst_address, value, num_bytes);
+	static std::atomic<uint32_t> clock_log_count {0};
+	if (clock_log_count.fetch_add(1) < 64) {
+		LOGF("\t copy_data reference clock: dst=0x%016" PRIx64 " value=0x%016" PRIx64
+		     " size=%u\n",
+		     dst_address, value, num_bytes);
+	}
 }
 
 void CommandProcessor::DmaData(uint8_t engine, uint8_t dst_sel, uint8_t dst_cache_policy,

src/graphics/guest_gpu/tile.cpp

@@ -4,15 +4,12 @@
 #include "common/assert.h"
 #include "common/logging/log.h"
 #include "common/profiler.h"
-#include "common/stringUtils.h"
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/guest_gpu/gpu_format.h"
 
 #include <algorithm>
 #include <array>
 #include <bit>
-#include <fmt/format.h>
-#include <vector>
 
 namespace Libs::Graphics {
 
@@ -1322,13 +1319,8 @@ void TileGetTextureSize(Prospero::BufferFormat format, uint32_t width, uint32_t
 		return;
 	}
 	if (total_size != nullptr && total_size->size == 0) {
-		std::vector<std::string> list;
-		list.push_back(fmt::format("format = {}", static_cast<uint32_t>(format)));
-		list.push_back(fmt::format("width  = {}", width));
-		list.push_back(fmt::format("height = {}", height));
-		list.push_back(fmt::format("levels = {}", levels));
-		list.push_back(fmt::format("tile   = {}", static_cast<uint32_t>(tile)));
-		EXIT("unknown format:\n%s\n", Common::Concat(list, '\n').c_str());
+		EXIT("unknown format:\nformat = %u\nwidth  = %u\nheight = %u\nlevels = %u\ntile   = %u\n",
+		     static_cast<uint32_t>(format), width, height, levels, static_cast<uint32_t>(tile));
 	}
 }
 

src/graphics/host_gpu/regionManager.h

@@ -16,6 +16,7 @@
 #include <windows.h>
 #undef min
 #undef max
+#undef MemoryBarrier
 #elif defined(__APPLE__)
 #include <pthread.h>
 #elif defined(__linux__)

src/graphics/host_gpu/renderer/pipeline/pipelineCache.cpp

@@ -593,6 +593,21 @@ PipelineCache::GraphicsPrograms PipelineCache::GetGraphicsPrograms(
 	ShaderParams pixel_params;
 	if (pixel_active) {
 		pixel_params = PrepareProgram(pixel_regs, sh, target_export_mapping, pixel_info);
+		const auto& blend          = context.GetBlendControl(0);
+		const auto  is_dual_source = [](uint8_t factor) {
+			return factor >= static_cast<uint8_t>(Prospero::BlendFactor::kSrc1Color) &&
+			       factor <= static_cast<uint8_t>(Prospero::BlendFactor::kOneMinusSrc1Alpha);
+		};
+		pixel_info.dual_source_blending =
+		    blend.enable && !context.GetRenderTarget(0).info.blend_bypass &&
+		    (is_dual_source(blend.color_srcblend) || is_dual_source(blend.color_destblend) ||
+		     (blend.separate_alpha_blend &&
+		      (is_dual_source(blend.alpha_srcblend) || is_dual_source(blend.alpha_destblend))));
+		if (pixel_info.dual_source_blending) {
+			// MRT1 supplies a second blend source for the same render target as MRT0.
+			pixel_info.target_output_mode[1]    = pixel_info.target_output_mode[0];
+			pixel_info.target_export_mapping[1] = pixel_info.target_export_mapping[0];
+		}
 	}
 	if (context.GetClipControl().clip_disable) {
 		const auto& viewport = context.GetScreenViewport().viewports[0];

src/graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.cpp

@@ -46,8 +46,8 @@ vk::PipelineStageFlags ShaderPipelineStages(vk::ShaderStageFlags stages) {
 	return result;
 }
 
-VulkanMemoryBarrier MakeShaderWriteDependency() {
-	VulkanMemoryBarrier barrier {};
+vk::MemoryBarrier MakeShaderWriteDependency() {
+	vk::MemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eShaderWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite |
 	                        vk::AccessFlagBits::eVertexAttributeRead |
@@ -58,15 +58,15 @@ VulkanMemoryBarrier MakeShaderWriteDependency() {
 	return barrier;
 }
 
-VulkanMemoryBarrier MakeShaderAccessDependency() {
-	VulkanMemoryBarrier barrier {};
+vk::MemoryBarrier MakeShaderAccessDependency() {
+	vk::MemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
 	return barrier;
 }
 
-VulkanMemoryBarrier MakeShaderWriteHazardDependency() {
-	VulkanMemoryBarrier barrier {};
+vk::MemoryBarrier MakeShaderWriteHazardDependency() {
+	vk::MemoryBarrier barrier {};
 	barrier.srcAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite;
 	barrier.dstAccessMask = vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite;
 	return barrier;

src/graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h

@@ -10,9 +10,9 @@ struct ShaderStageRuntime;
 
 vk::ShaderStageFlagBits NativeShaderStage(ShaderType stage);
 vk::PipelineStageFlags  ShaderPipelineStages(vk::ShaderStageFlags stages);
-VulkanMemoryBarrier     MakeShaderAccessDependency();
-VulkanMemoryBarrier     MakeShaderWriteHazardDependency();
-VulkanMemoryBarrier     MakeShaderWriteDependency();
+vk::MemoryBarrier       MakeShaderAccessDependency();
+vk::MemoryBarrier       MakeShaderWriteHazardDependency();
+vk::MemoryBarrier       MakeShaderWriteDependency();
 vk::BufferMemoryBarrier MakeGdsDependency(vk::Buffer buffer);
 bool HasShaderBufferWrites(const ShaderStageRuntime& runtime);
 void ShaderAccessBarrier(vk::CommandBuffer vk_buffer, vk::PipelineStageFlags source_stages);

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -225,7 +225,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, CommandBuffer& buffer,
 		return;
 	}
 
-	if (!ShaderAddressValid(sh_ctx.GetCs().cs_regs.data_addr)) {
+	if (sh_ctx.GetCs().cs_regs.data_addr == 0) {
 		return;
 	}
 

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -421,7 +421,7 @@ static void SetGraphicsDynamicParams(const CommandBuffer& buffer, vk::CommandBuf
 static bool DrawHasValidVertexShader(const HW::Shader& sh_ctx) {
 
 	const auto& vs = sh_ctx.GetVs();
-	return ShaderAddressValid(vs.es_regs.data_addr);
+	return vs.es_regs.data_addr != 0;
 }
 
 static bool PixelShaderHasDepthOrCoverageSideEffects(const HW::ShaderRegisters& sh_regs) {
@@ -608,7 +608,7 @@ static bool DrawHasActivePixelShader(const CommandBuffer& buffer) {
 	const auto& ctx              = buffer.GetRegisters();
 	const auto& sh_regs          = ctx.GetShaderRegisters();
 	const bool  has_color_output = (ctx.GetRenderTargetMask() & sh_regs.m_cbShaderMask) != 0;
-	return ShaderAddressValid(buffer.GetShaders().GetPs().ps_regs.data_addr) &&
+	return buffer.GetShaders().GetPs().ps_regs.data_addr != 0 &&
 	       (has_color_output || PixelShaderHasDepthOrCoverageSideEffects(sh_regs));
 }
 

src/graphics/host_gpu/vulkanCommon.h

@@ -24,8 +24,6 @@
 
 namespace Libs::Graphics {
 
-using VulkanMemoryBarrier = vk::MemoryBarrier;
-
 vk::Format  VulkanFormat(Prospero::BufferFormat guest_format);
 void        RequireVulkanSuccess(vk::Result result, const char* operation);
 vk::ShaderModule CompileSPV(std::span<const uint32_t> code, vk::Device device);

src/graphics/presentation/window/swapchain.cpp

@@ -22,9 +22,6 @@
 
 // IWYU pragma: no_include <intrin.h>
 
-#define KYTY_ENABLE_DEBUG_PRINTF
-#define KYTY_DBG_INPUT
-
 namespace Libs::Graphics {
 
 struct Presenter::Frame {

src/graphics/presentation/window/vulkanWindow.cpp

@@ -49,7 +49,6 @@
 // IWYU pragma: no_include <intrin.h>
 
 #define KYTY_ENABLE_DEBUG_PRINTF
-#define KYTY_DBG_INPUT
 
 namespace Libs::Graphics {
 
@@ -611,10 +610,11 @@ static vk::Device VulkanCreateDevice(GraphicContext& graphics,
 		     graphics.mesh_shader_properties.maxMeshOutputPrimitives,
 		     graphics.mesh_shader_properties.maxMeshSharedMemorySize);
 	}
-	// VulkanFindPhysicalDevice already checked the required creation features. These two
+	// VulkanFindPhysicalDevice already checked the required creation features. These
 	// requirements are specific to this creation path and are not part of device selection.
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.shaderInt64 != VK_TRUE);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.vertexPipelineStoresAndAtomics != VK_TRUE);
+	EXIT_NOT_IMPLEMENTED(supported_features2.features.dualSrcBlend != VK_TRUE);
 	vk::PhysicalDeviceFeatures device_features {};
 	device_features.fragmentStoresAndAtomics = VK_TRUE;
 	device_features.samplerAnisotropy        = VK_TRUE;
@@ -625,6 +625,7 @@ static vk::Device VulkanCreateDevice(GraphicContext& graphics,
 	device_features.shaderStorageImageWriteWithoutFormat = VK_TRUE;
 	device_features.shaderImageGatherExtended            = VK_TRUE;
 	device_features.independentBlend                     = VK_TRUE;
+	device_features.dualSrcBlend                         = VK_TRUE;
 	device_features.tessellationShader                   = VK_TRUE;
 	device_features.sampleRateShading                    = VK_TRUE;
 	device_features.depthBiasClamp                       = VK_TRUE;

src/graphics/presentation/window/window.cpp

@@ -27,6 +27,7 @@
 #include "common/systemInfo.h"
 #include "common/threads.h"
 #include "common/timer.h"
+#include "common/stringUtils.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
@@ -45,6 +46,7 @@
 #include <string>
 #include <vector>
 #include <vulkan/vk_platform.h>
+#include <filesystem>
 
 #define STB_IMAGE_IMPLEMENTATION
 #define STBI_NO_SIMD
@@ -52,7 +54,6 @@
 
 // IWYU pragma: no_include <intrin.h>
 
-#define KYTY_ENABLE_DEBUG_PRINTF
 #define KYTY_DBG_INPUT
 
 namespace Libs::Graphics {
@@ -180,12 +181,6 @@ struct EventDisplay {
 	DisplayOrientation orientation;
 };
 
-constexpr uint32_t KYTY_SDL_BUTTON_LMASK  = SDL_BUTTON_LMASK;  // NOLINT(hicpp-signed-bitwise)
-constexpr uint32_t KYTY_SDL_BUTTON_MMASK  = SDL_BUTTON_MMASK;  // NOLINT(hicpp-signed-bitwise)
-constexpr uint32_t KYTY_SDL_BUTTON_RMASK  = SDL_BUTTON_RMASK;  // NOLINT(hicpp-signed-bitwise)
-constexpr uint32_t KYTY_SDL_BUTTON_X1MASK = SDL_BUTTON_X1MASK; // NOLINT(hicpp-signed-bitwise)
-constexpr uint32_t KYTY_SDL_BUTTON_X2MASK = SDL_BUTTON_X2MASK; // NOLINT(hicpp-signed-bitwise)
-
 namespace {
 
 std::unique_ptr<WindowContext> g_window;
@@ -193,7 +188,6 @@ std::unique_ptr<WindowContext> g_window;
 } // namespace
 
 constexpr const char* KYTY_SDL_WINDOW_CAPTION = "Game";
-constexpr int KYTY_SDL_WINDOWPOS_CENTERED = SDL_WINDOWPOS_CENTERED; /*NOLINT(hicpp-signed-bitwise)*/
 
 static void SetPause(WindowLoopState& game, bool flag) {
 	LOGF("Pause: %s\n", flag ? "true" : "false");
@@ -614,11 +608,11 @@ void WindowContext::ProcessEvent(double time_s) {
 
 			mb.down              = false;
 			mb.up                = false;
-			mb.left              = ((event->motion.state & KYTY_SDL_BUTTON_LMASK) != 0u);
-			mb.middle            = ((event->motion.state & KYTY_SDL_BUTTON_MMASK) != 0u);
-			mb.right             = ((event->motion.state & KYTY_SDL_BUTTON_RMASK) != 0u);
-			mb.x1                = ((event->motion.state & KYTY_SDL_BUTTON_X1MASK) != 0u);
-			mb.x2                = ((event->motion.state & KYTY_SDL_BUTTON_X2MASK) != 0u);
+			mb.left              = ((event->motion.state & SDL_BUTTON_LMASK) != 0u);
+			mb.middle            = ((event->motion.state & SDL_BUTTON_MMASK) != 0u);
+			mb.right             = ((event->motion.state & SDL_BUTTON_RMASK) != 0u);
+			mb.x1                = ((event->motion.state & SDL_BUTTON_X1MASK) != 0u);
+			mb.x2                = ((event->motion.state & SDL_BUTTON_X2MASK) != 0u);
 			mb.touch             = (event->motion.which == SDL_TOUCH_MOUSEID);
 			mb.pressed           = false;
 			mb.released          = false;
@@ -713,6 +707,20 @@ void WindowContext::ProcessEvent(double time_s) {
 			}
 			break;
 
+		case SDL_CONTROLLERSENSORUPDATE: {
+			const auto& sensor = event->csensor;
+			if (sensor.sensor == SDL_SENSOR_ACCEL || sensor.sensor == SDL_SENSOR_GYRO) {
+				Controller::SetSensor(sensor.which,
+				                      sensor.sensor == SDL_SENSOR_ACCEL ? Controller::Sensor::Accel
+				                                                        : Controller::Sensor::Gyro,
+				                      sensor.data,
+				                      sensor.timestamp_us != 0
+				                          ? sensor.timestamp_us
+				                          : static_cast<uint64_t>(sensor.timestamp) * 1000);
+			}
+			break;
+		}
+
 		case SDL_CONTROLLERDEVICEADDED:
 		case SDL_CONTROLLERDEVICEREMOVED:
 		case SDL_CONTROLLERDEVICEREMAPPED: {
@@ -854,8 +862,8 @@ static void WindowCreate(WindowContext& context) {
 		window_flags |= static_cast<uint32_t>(SDL_WINDOW_BORDERLESS);
 	}
 #endif
-	context.window = SDL_CreateWindow(KYTY_SDL_WINDOW_CAPTION, KYTY_SDL_WINDOWPOS_CENTERED,
-	                                  KYTY_SDL_WINDOWPOS_CENTERED, width, height, window_flags);
+	context.window = SDL_CreateWindow(KYTY_SDL_WINDOW_CAPTION, SDL_WINDOWPOS_CENTERED,
+	                                  SDL_WINDOWPOS_CENTERED, width, height, window_flags);
 
 	if (context.window == nullptr) {
 		EXIT("%s\n", SDL_GetError());
@@ -939,10 +947,10 @@ struct WindowIcon {
 	}
 };
 
-static void WindowLoadPngIcon(const std::string& path, WindowIcon* icon) {
+static void WindowLoadPngIcon(const std::filesystem::path& path, WindowIcon* icon) {
 	Common::File f;
 	if (!f.Open(path, Common::File::Mode::Read)) {
-		EXIT("Can't open icon file %s\n", path.c_str());
+		EXIT("Can't open icon file %s\n", Common::PathToString(path).c_str());
 	}
 
 	int width  = 0;
@@ -968,7 +976,7 @@ void WindowContext::UpdateIcon() {
 	static bool       icon_loaded = false;
 
 	if (!icon_loaded) {
-		std::string icon_path;
+		std::filesystem::path icon_path;
 		if (Loader::SystemContentGetIconPath(&icon_path)) {
 			WindowLoadPngIcon(icon_path, &icon);
 		}

src/graphics/shader/recompiler/ShaderRecompiler.cpp

@@ -105,8 +105,6 @@ void ClearEmbeddedFetchVectorLanes(EmbeddedFetchVectorLanes* lanes, uint32_t reg
 	lanes->erase(first, last);
 }
 
-using EmbeddedFetchData = Frontend::EmbeddedFetchPlan;
-
 bool IsDecodedSgpr(const Decoder::Operand& op) {
 	return op.kind == Decoder::OperandKind::Sgpr || op.kind == Decoder::OperandKind::VccLo ||
 	       op.kind == Decoder::OperandKind::VccHi;
@@ -210,13 +208,12 @@ int BufferTableAttribFromOffset(uint32_t raw_offset, int dword) {
 	return static_cast<int>((raw_offset + static_cast<uint32_t>(dword) * 4u) / 16u);
 }
 
-EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded,
-                                            const ShaderVertexInputInfo* input_info,
-                                            uint32_t user_data_base, uint32_t user_data_count,
-                                            uint32_t wave_size) {
+Frontend::EmbeddedFetchPlan DetectEmbeddedVertexFetch(
+    const Decoder::Program& decoded, const ShaderVertexInputInfo* input_info,
+    uint32_t user_data_base, uint32_t user_data_count, uint32_t wave_size) {
 	const uint32_t    vertex_index_reg   = input_info->logical_stage == ShaderType::Local ? 2u : 5u;
 	const uint32_t    instance_index_reg = input_info->logical_stage == ShaderType::Local ? 5u : 8u;
-	EmbeddedFetchData data;
+	Frontend::EmbeddedFetchPlan data;
 	data.loads.reserve(input_info->resources_num);
 	int32_t vertex_offset_candidate   = -1;
 	int32_t instance_offset_candidate = -1;
@@ -560,7 +557,7 @@ TranslateResult TranslateProgram(std::span<const uint32_t> code, const CompileOp
 		     static_cast<uint64_t>(cfg.natural_loops.size()), phase_ms());
 	}
 
-	EmbeddedFetchData embedded_fetch;
+	Frontend::EmbeddedFetchPlan embedded_fetch;
 	if ((options.stage == ShaderType::Vertex || options.stage == ShaderType::Local) &&
 	    options.input_info.vertex != nullptr && options.input_info.vertex->fetch_embedded) {
 		embedded_fetch = DetectEmbeddedVertexFetch(

src/graphics/shader/recompiler/backend/spirv/spirvEmitterFlow.cpp

@@ -138,7 +138,7 @@ uint32_t EmitDppWriteCondition(ValueEmitContext& ctx, const IR::DppMoveFlags& fl
 	state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), masks_ok, bank_ok, row_ok);
 	uint32_t writable = masks_ok;
 	if (!flags.bound_control) {
-		const auto target  = EmitDppTargetLane(state, flags.control);
+		const auto target  = EmitDppTargetLane(state, flags);
 		const auto bounded = state.builder.AllocateId();
 		state.builder.AddFunction(spv::OpLogicalAnd, TypeBool(state), bounded, writable,
 		                          target.valid);
@@ -611,7 +611,7 @@ uint32_t EmitUndefU1(EmitterState& state, const IR::Inst& inst) {
 uint32_t EmitDppMoveU32(ValueEmitContext& ctx, const IR::Inst& inst) {
 	auto&      state    = ctx.state;
 	const auto flags    = inst.Flags<IR::DppMoveFlags>();
-	const auto target   = EmitDppTargetLane(state, flags.control);
+	const auto target   = EmitDppTargetLane(state, flags);
 	const auto shuffled = ctx.Shuffle(inst, 0, target.lane);
 	if (flags.fetch_inactive) {
 		return shuffled;

src/graphics/shader/recompiler/backend/spirv/spirvEmitterHelpers.cpp

@@ -2,24 +2,26 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 
-DppTargetLane EmitDppQuadPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control) {
-	const auto quad_base = state.builder.AllocateId();
+DppTargetLane EmitDppPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control,
+                                    uint32_t lane_bits) {
+	const auto lane_mask  = (1u << lane_bits) - 1u;
+	const auto group_base = state.builder.AllocateId();
 	const auto lane      = state.builder.AllocateId();
 	const auto shift     = state.builder.AllocateId();
 	const auto selected0 = state.builder.AllocateId();
 	const auto selected  = state.builder.AllocateId();
 	const auto target    = state.builder.AllocateId();
-	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), quad_base, subid,
-	                          ConstantU32(state, 0xfffffffcu));
+	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), group_base, subid,
+	                          ConstantU32(state, ~lane_mask));
 	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), lane, subid,
-	                          ConstantU32(state, 3));
-	state.builder.AddFunction(spv::OpShiftLeftLogical, TypeU32(state), shift, lane,
-	                          ConstantU32(state, 1));
+	                          ConstantU32(state, lane_mask));
+	state.builder.AddFunction(spv::OpIMul, TypeU32(state), shift, lane,
+	                          ConstantU32(state, lane_bits));
 	state.builder.AddFunction(spv::OpShiftRightLogical, TypeU32(state), selected0,
 	                          ConstantU32(state, control), shift);
 	state.builder.AddFunction(spv::OpBitwiseAnd, TypeU32(state), selected, selected0,
-	                          ConstantU32(state, 3));
-	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, quad_base, selected);
+	                          ConstantU32(state, lane_mask));
+	state.builder.AddFunction(spv::OpBitwiseOr, TypeU32(state), target, group_base, selected);
 	return {target, ConstantBool(state, true)};
 }
 
@@ -89,10 +91,14 @@ DppTargetLane EmitDppMirrorTargetLane(EmitterState& state, uint32_t subid, bool
 	return {target, ConstantBool(state, true)};
 }
 
-DppTargetLane EmitDppTargetLane(EmitterState& state, uint32_t control) {
+DppTargetLane EmitDppTargetLane(EmitterState& state, const IR::DppMoveFlags& flags) {
 	const auto subid = EmitSubgroupLocalInvocationId(state);
+	const auto control = flags.control;
+	if (flags.dpp8) {
+		return EmitDppPermTargetLane(state, subid, control, 3u);
+	}
 	if (control <= 0xffu) {
-		return EmitDppQuadPermTargetLane(state, subid, control);
+		return EmitDppPermTargetLane(state, subid, control, 2u);
 	}
 	if (control >= 0x101u && control <= 0x10fu) {
 		return EmitDppRowShiftTargetLane(state, subid, control & 0xfu, true);

src/graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h

@@ -308,8 +308,6 @@ uint32_t ConstantI32(EmitterState& state, int32_t value);
 
 uint32_t ConstantF32(EmitterState& state, uint32_t bits);
 
-uint32_t FloatBits(float value);
-
 uint32_t ConstantF32Value(EmitterState& state, float value);
 
 uint32_t ConstantBool(EmitterState& state, bool value);
@@ -329,7 +327,8 @@ void     EmitMeshAllocate(ValueEmitContext& ctx, const IR::Inst& inst);
 uint32_t MeshOutputPointer(EmitterState& state, IR::StageOutputKind kind, uint32_t index = 0);
 uint32_t MeshPrimitivePointer(EmitterState& state);
 
-DppTargetLane EmitDppQuadPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control);
+DppTargetLane EmitDppPermTargetLane(EmitterState& state, uint32_t subid, uint32_t control,
+                                    uint32_t lane_bits);
 
 DppTargetLane EmitDppRowShiftTargetLane(EmitterState& state, uint32_t subid, uint32_t amount,
                                         bool left);
@@ -338,7 +337,7 @@ DppTargetLane EmitDppRowRotateRightTargetLane(EmitterState& state, uint32_t subi
 
 DppTargetLane EmitDppMirrorTargetLane(EmitterState& state, uint32_t subid, bool half_row);
 
-DppTargetLane EmitDppTargetLane(EmitterState& state, uint32_t control);
+DppTargetLane EmitDppTargetLane(EmitterState& state, const IR::DppMoveFlags& flags);
 
 uint32_t EmitSubgroupLocalInvocationId(EmitterState& state);
 

src/graphics/shader/recompiler/backend/spirv/spirvEmitterMemory.cpp

@@ -514,31 +514,6 @@ void StoreWord(ValueEmitContext& ctx, const IR::Inst& inst, IR::MemoryInfo mem)
 	});
 }
 
-void FormattedStorePrepared(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem,
-                            uint32_t component, const MemoryResourceAccess& resource,
-                            uint32_t data) {
-	const auto info = Format::GetFormatInfo(BufferFormat(ctx, mem));
-	if (info.type == Format::ComponentType::Unknown) {
-		StoreWordPrepared(ctx, inst, RebaseRawComponent(mem, component), resource, data);
-		return;
-	}
-	if (component >= info.component_count) return;
-	const auto bits          = info.component_bits[component];
-	const auto component_mem = RebaseFormattedComponent(mem, info, component);
-	if (bits == 8u || bits == 16u) {
-		StoreSubwordPrepared(ctx, inst, component_mem, resource, bits, data);
-	} else {
-		StoreWordPrepared(ctx, inst, component_mem, resource, data);
-	}
-}
-
-void FormattedStore(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem) {
-	EmitIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
-		const auto resource = PrepareMemoryResourceAccess(ctx.state, mem);
-		FormattedStorePrepared(ctx, inst, mem, 0u, resource, ctx.Arg(inst, inst.NumArgs() - 2));
-	});
-}
-
 spv::Op SpirvAtomicOpcode(IR::ValueOpcode opcode) {
 	switch (opcode) {
 		case IR::ValueOpcode::BufferAtomicCmpSwap32: return spv::OpAtomicCompareExchange;
@@ -729,6 +704,11 @@ void StoreFormattedInBounds(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
                             uint32_t data) {
 	if (component >= plan.info.component_count) return;
 	const auto bits = plan.info.component_bits[component];
+	if (plan.info.type == Format::ComponentType::Snorm && bits == 16u) {
+		const auto value = EmitBitCastF32U32(ctx.state, data);
+		data = EmitPackSnorm2x16(
+		    ctx.state, EmitCompositeConstructF32x2(ctx.state, value, ConstantF32Value(ctx.state, 0.0f)));
+	}
 	if (bits == 8u || bits == 16u) {
 		StoreSubwordInBounds(ctx, mem, plan.resource, plan.addresses[component],
 		                     plan.indices[component], bits, data);
@@ -737,6 +717,23 @@ void StoreFormattedInBounds(ValueEmitContext& ctx, const IR::MemoryInfo& mem,
 	}
 }
 
+void FormattedStore(ValueEmitContext& ctx, const IR::Inst& inst, const IR::MemoryInfo& mem) {
+	EmitIfCondition(ctx.state, ctx.Arg(inst, inst.NumArgs() - 1), [&]() {
+		const auto resource = PrepareMemoryResourceAccess(ctx.state, mem);
+		const auto data     = ctx.Arg(inst, inst.NumArgs() - 2);
+		const auto info     = Format::GetFormatInfo(BufferFormat(ctx, mem));
+		if (info.type == Format::ComponentType::Unknown) {
+			StoreWordPrepared(ctx, inst, RebaseRawComponent(mem, 0u), resource, data);
+			return;
+		}
+		const auto plan = PrepareFormattedMemory(ctx, inst, mem, resource, info, 1u,
+		                                         FormattedAccess::Store);
+		EmitIfCondition(ctx.state, plan.in_bounds, [&]() {
+			StoreFormattedInBounds(ctx, mem, plan, 0u, data);
+		});
+	});
+}
+
 uint32_t LoadWideBuffer(ValueEmitContext& ctx, const IR::Inst& inst, uint32_t components) {
 	auto& state = ctx.state;
 	return EmitValueOrDefaultIfCondition(

src/graphics/shader/recompiler/backend/spirv/spirvEmitterModule.cpp

@@ -1,6 +1,7 @@
 #include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
 
 #include <algorithm>
+#include <bit>
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
 
@@ -256,14 +257,8 @@ uint32_t ConstantF32(EmitterState& state, uint32_t bits) {
 	return state.builder.Constant(spv::OpConstant, TypeF32(state), bits);
 }
 
-uint32_t FloatBits(float value) {
-	uint32_t bits = 0;
-	std::memcpy(&bits, &value, sizeof(bits));
-	return bits;
-}
-
 uint32_t ConstantF32Value(EmitterState& state, float value) {
-	return ConstantF32(state, FloatBits(value));
+	return ConstantF32(state, std::bit_cast<uint32_t>(value));
 }
 
 uint32_t ConstantBool(EmitterState& state, bool value) {
@@ -558,8 +553,17 @@ void DefineOutputs(EmitterState& state) {
 				const auto type = uint_output ? TypeU32Vector(state, 4) : TypeF32Vector(state, 4);
 				binding.variable_id = DefineInterfaceVariable(state, type, spv::StorageClassOutput,
 				                                              binding.debug_name.c_str());
+				const bool dual_source = binding.kind == IR::StageOutputKind::Mrt &&
+				                         state.program.stage == ShaderType::Pixel &&
+				                         state.input_info.pixel->dual_source_blending;
+				EXIT_NOT_IMPLEMENTED(dual_source && binding.index > 1);
 				state.builder.AddAnnotation(spv::OpDecorate, binding.variable_id,
-				                            spv::DecorationLocation, binding.location);
+				                            spv::DecorationLocation,
+				                            dual_source ? 0u : binding.location);
+				if (dual_source) {
+					state.builder.AddAnnotation(spv::OpDecorate, binding.variable_id,
+					                            spv::DecorationIndex, binding.index);
+				}
 				break;
 			}
 		}

src/graphics/shader/recompiler/frontend/decode/ImageOps.cpp

@@ -3,6 +3,7 @@
 #include "graphics/shader/recompiler/frontend/decode/OpcodeTable.h"
 
 #include <algorithm>
+#include <bit>
 
 namespace Libs::Graphics::ShaderRecompiler::Decoder {
 namespace {
@@ -203,18 +204,6 @@ constexpr auto MIMG_SAMPLE_OPS = Detail::MakeOpcodeTable<0x100>(MIMG_SAMPLE_OPCO
 constexpr auto MIMG_GATHER_OPS = Detail::MakeOpcodeTable<0x100>(MIMG_GATHER_OPCODE_LIST);
 constexpr auto MIMG_ATOMIC_OPS = Detail::MakeOpcodeTable<0x100>(MIMG_ATOMIC_OPCODE_LIST);
 
-const MimgSampleInfo* LookupSample(uint32_t opcode) {
-	return Detail::FindOpcode(MIMG_SAMPLE_OPS, opcode);
-}
-
-const MimgGatherInfo* LookupGather(uint32_t opcode) {
-	return Detail::FindOpcode(MIMG_GATHER_OPS, opcode);
-}
-
-const Detail::OpcodeMap* LookupAtomic(uint32_t opcode) {
-	return Detail::FindOpcode(MIMG_ATOMIC_OPS, opcode);
-}
-
 Opcode DecodeMimgOpcode(uint32_t opcode, const MimgSampleInfo* sample, const MimgGatherInfo* gather,
                         const Detail::OpcodeMap* atomic) {
 	if (sample != nullptr) {
@@ -280,10 +269,6 @@ uint32_t CountDmaskComponents(uint32_t dmask) {
 	return count != 0 ? count : 1u;
 }
 
-bool IsSingleDmaskBit(uint32_t dmask) {
-	return dmask != 0u && (dmask & (dmask - 1u)) == 0u;
-}
-
 } // namespace
 
 ImageAddressComponent ImageAddressComponentLayout(uint32_t flags, uint32_t component) {
@@ -332,9 +317,9 @@ void DecodeMimg(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	const bool     r128   = ((word0 >> 15u) & 0x1u) != 0u;
 	const bool     a16    = ((word1 >> 30u) & 0x1u) != 0u;
 	const bool     d16    = ((word1 >> 31u) & 0x1u) != 0u;
-	const auto*    sample = LookupSample(opcode);
-	const auto*    gather = LookupGather(opcode);
-	const auto*    atomic = LookupAtomic(opcode);
+	const auto*    sample = Detail::FindOpcode(MIMG_SAMPLE_OPS, opcode);
+	const auto*    gather = Detail::FindOpcode(MIMG_GATHER_OPS, opcode);
+	const auto*    atomic = Detail::FindOpcode(MIMG_ATOMIC_OPS, opcode);
 
 	inst.pc                 = pc;
 	inst.word_count         = word_count;
@@ -364,7 +349,7 @@ void DecodeMimg(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	if (inst.opcode == Opcode::UNSUPPORTED) {
 		SetUnsupported(inst, Family::MIMG, opcode, "MIMG opcode is not implemented");
 	}
-	if (gather != nullptr && !IsSingleDmaskBit(inst.dmask)) {
+	if (gather != nullptr && !std::has_single_bit(inst.dmask)) {
 		SetUnsupported(inst, Family::MIMG, opcode,
 		               "MIMG image gather requires exactly one dmask bit");
 	}
@@ -382,7 +367,7 @@ void DecodeMimg(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 }
 
 const char* MimgSampleOpcodeName(uint32_t opcode) {
-	const auto* sample = LookupSample(opcode);
+	const auto* sample = Detail::FindOpcode(MIMG_SAMPLE_OPS, opcode);
 	return sample != nullptr ? sample->name : nullptr;
 }
 

src/graphics/shader/recompiler/frontend/decode/ScalarAluOps.cpp

@@ -24,7 +24,8 @@ constexpr OpcodeMap SOP2_OPCODE_LIST[] = {
     {0x1cu, Opcode::S_XNOR_B32},        {0x1du, Opcode::S_XNOR_B64},
     {0x1eu, Opcode::S_LSHL_B32},        {0x1fu, Opcode::S_LSHL_B64},
     {0x20u, Opcode::S_LSHR_B32},        {0x21u, Opcode::S_LSHR_B64},
-    {0x22u, Opcode::S_ASHR_I32},        {0x24u, Opcode::S_BFM_B32},
+    {0x22u, Opcode::S_ASHR_I32},        {0x23u, Opcode::S_ASHR_I64},
+    {0x24u, Opcode::S_BFM_B32},
     {0x25u, Opcode::S_BFM_B64},         {0x26u, Opcode::S_MUL_I32},
     {0x27u, Opcode::S_BFE_U32},         {0x28u, Opcode::S_BFE_I32},
     {0x29u, Opcode::S_BFE_U64},         {0x2cu, Opcode::S_ABSDIFF_I32},

src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.cpp

@@ -14,10 +14,6 @@
 namespace Libs::Graphics::ShaderRecompiler::Decoder {
 namespace {
 
-uint32_t FloatBits(float value) {
-	return std::bit_cast<uint32_t>(value);
-}
-
 bool HasLiteral(const Instruction& inst) {
 	return inst.src0.kind == OperandKind::LiteralConstant ||
 	       inst.src1.kind == OperandKind::LiteralConstant ||
@@ -242,7 +238,7 @@ void DecodeScalarSource(uint32_t code, uint32_t pc, Operand& operand) {
 	if (code >= 240u && code <= 247u) {
 		constexpr float values[] = {0.5f, -0.5f, 1.0f, -1.0f, 2.0f, -2.0f, 4.0f, -4.0f};
 		operand.kind             = OperandKind::FloatInlineConstant;
-		operand.value            = FloatBits(values[code - 240u]);
+		operand.value            = std::bit_cast<uint32_t>(values[code - 240u]);
 		return;
 	}
 	if (code >= 256u && code <= 511u) {
@@ -260,7 +256,7 @@ void DecodeScalarSource(uint32_t code, uint32_t pc, Operand& operand) {
 		case 239u: operand.kind = OperandKind::PopsExitingWaveId; return;
 		case 248u:
 			operand.kind      = OperandKind::FloatInlineConstant;
-			operand.value = FloatBits(0.15915494309189535f);
+			operand.value = std::bit_cast<uint32_t>(0.15915494309189535f);
 			return;
 		case 251u: operand.kind = OperandKind::VccZ; return;
 		case 252u: operand.kind = OperandKind::ExecZ; return;
@@ -476,8 +472,9 @@ std::string OperandToString(const Operand& operand) {
 		text += ".clamp";
 	}
 	if (operand.dpp) {
-		text += fmt::format(".dpp(ctrl=0x{:x},fi={},bc={})", operand.dpp_ctrl,
-		                    operand.dpp_fetch_inactive ? 1u : 0u, operand.dpp_bound_ctrl ? 1u : 0u);
+		text += fmt::format(".{}(ctrl=0x{:x},fi={},bc={})", operand.dpp8 ? "dpp8" : "dpp",
+		                    operand.dpp_ctrl, operand.dpp_fetch_inactive ? 1u : 0u,
+		                    operand.dpp_bound_ctrl ? 1u : 0u);
 	}
 	return text;
 }

src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.h

@@ -2,8 +2,8 @@
 #define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_SHADERDECODER_H_
 
 #include "common/common.h"
-#include "common/magicEnum.h"
 
+#include <magic_enum.hpp>
 #include <span>
 #include <string_view>
 #include <vector>
@@ -106,6 +106,7 @@ enum class Opcode {
 	S_LSHR_B32,
 	S_LSHR_B64,
 	S_ASHR_I32,
+	S_ASHR_I64,
 	S_MUL_I32,
 	S_MUL_HI_U32,
 	S_MUL_HI_I32,
@@ -227,6 +228,7 @@ enum class Opcode {
 	V_CVT_PKRTZ_F16_F32,
 	V_CVT_PK_U8_F32,
 	V_MAD_F32,
+	V_MAD_I16,
 	V_MAD_I32_I24,
 	V_MAD_U32_U24,
 	V_MAD_U64_U32,
@@ -299,6 +301,7 @@ enum class Opcode {
 	V_SUBREV_NC_U32,
 	V_ADD_NC_U16,
 	V_SUB_NC_U16,
+	V_MUL_LO_U16,
 	V_MAX_U16,
 	V_MAX_I16,
 	V_MIN_U16,
@@ -377,7 +380,9 @@ enum class Opcode {
 	V_CMP_GT_F16,
 	V_CMP_LG_F16,
 	V_CMP_GE_F16,
+	V_CMP_NGT_F16,
 	V_CMP_NEQ_F16,
+	V_CMP_NLT_F16,
 	V_CMPX_LT_F16,
 	V_CMPX_EQ_F16,
 	V_CMPX_LE_F16,
@@ -665,6 +670,7 @@ struct Operand {
 	bool     absolute           = false;
 	bool     clamp              = false;
 	bool     dpp                = false;
+	bool     dpp8               = false;
 };
 
 struct Instruction {

src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp

@@ -83,10 +83,6 @@ constexpr Vop2OpcodeInfo VOP2_OPCODE_LIST[] = {
 
 constexpr auto VOP2_OPS = Detail::MakeOpcodeTable<0x40>(VOP2_OPCODE_LIST);
 
-constexpr Opcode LookupVop2Opcode(uint32_t encoding) {
-	return Detail::LookupOpcode(VOP2_OPS, encoding);
-}
-
 constexpr OpcodeMap VOP1_OPCODE_LIST[] = {
     {0x00u, Opcode::V_NOP},
     {0x01u, Opcode::V_MOV_B32},
@@ -250,7 +246,8 @@ constexpr VopcOpcodeInfo VOPC_OPCODE_LIST[] = {
     {0xf5u, Opcode::V_CMPX_NE_U64, false}, {0xc9u, Opcode::V_CMP_LT_F16},
     {0xcau, Opcode::V_CMP_EQ_F16},         {0xcbu, Opcode::V_CMP_LE_F16},
     {0xccu, Opcode::V_CMP_GT_F16},         {0xcdu, Opcode::V_CMP_LG_F16},
-    {0xceu, Opcode::V_CMP_GE_F16},         {0xedu, Opcode::V_CMP_NEQ_F16},
+    {0xceu, Opcode::V_CMP_GE_F16},         {0xebu, Opcode::V_CMP_NGT_F16},
+    {0xedu, Opcode::V_CMP_NEQ_F16},        {0xeeu, Opcode::V_CMP_NLT_F16},
     {0xd9u, Opcode::V_CMPX_LT_F16},        {0xdau, Opcode::V_CMPX_EQ_F16},
     {0xdbu, Opcode::V_CMPX_LE_F16},        {0xdcu, Opcode::V_CMPX_GT_F16},
     {0xdeu, Opcode::V_CMPX_GE_F16},        {0xfbu, Opcode::V_CMPX_NGT_F16},
@@ -298,6 +295,7 @@ constexpr OpcodeMap VOP3_OPCODE_LIST[] = {
     {0x300u, Opcode::V_LSHRREV_B64},
     {0x303u, Opcode::V_ADD_NC_U16},
     {0x304u, Opcode::V_SUB_NC_U16},
+    {0x305u, Opcode::V_MUL_LO_U16},
     {0x307u, Opcode::V_LSHRREV_B16},
     {0x308u, Opcode::V_ASHRREV_I16},
     {0x309u, Opcode::V_MAX_U16},
@@ -314,6 +312,7 @@ constexpr OpcodeMap VOP3_OPCODE_LIST[] = {
     {0x345u, Opcode::V_XAD_U32},
     {0x346u, Opcode::V_LSHL_ADD_U32},
     {0x347u, Opcode::V_ADD_LSHL_U32},
+    {0x35eu, Opcode::V_MAD_I16},
     {0x360u, Opcode::V_READLANE_B32},
     {0x361u, Opcode::V_WRITELANE_B32},
     {0x362u, Opcode::V_LDEXP_F32},
@@ -374,7 +373,7 @@ Opcode LookupVop3Opcode(uint32_t opcode) {
 		if (IsUnsupportedVop3EncodedVop2Alias(opcode - 0x100u)) {
 			return Opcode::UNSUPPORTED;
 		}
-		return LookupVop2Opcode(opcode - 0x100u);
+		return Detail::LookupOpcode(VOP2_OPS, opcode - 0x100u);
 	}
 	if (opcode >= 0x180u && opcode <= 0x1ffu) {
 		return Detail::LookupOpcode(VOP3_ENCODED_VOP1_OPS, opcode - 0x180u);
@@ -426,13 +425,14 @@ bool IsNativeVop3F16TernaryOpcode(Opcode opcode) {
 }
 
 bool IsNativeVop3I16TernaryOpcode(Opcode opcode) {
-	return opcode == Opcode::V_MED3_I16;
+	return opcode == Opcode::V_MED3_I16 || opcode == Opcode::V_MAD_I16;
 }
 
 bool IsNativeVop3B16BinaryOpcode(Opcode opcode) {
 	switch (opcode) {
 		case Opcode::V_ADD_NC_U16:
 		case Opcode::V_SUB_NC_U16:
+		case Opcode::V_MUL_LO_U16:
 		case Opcode::V_MAX_U16:
 		case Opcode::V_MAX_I16:
 		case Opcode::V_MIN_U16:
@@ -685,10 +685,15 @@ void DecodeVop1Sdwa(uint32_t pc, std::span<const uint32_t> code, uint32_t word_i
 	ReadLiteralOperands(code, word_index, inst);
 }
 
-void ApplyDppModifier(Operand& operand, uint32_t modifier) {
+void ApplyDppModifier(Operand& operand, uint32_t modifier, uint32_t encoding) {
+	operand.dpp = true;
+	if (encoding == 233u) {
+		operand.dpp8     = true;
+		operand.dpp_ctrl = modifier >> 8u;
+		return;
+	}
 	operand.negate             = ((modifier >> 20u) & 0x1u) != 0u;
 	operand.absolute           = ((modifier >> 21u) & 0x1u) != 0u;
-	operand.dpp                = true;
 	operand.dpp_ctrl           = (modifier >> 8u) & 0x1ffu;
 	operand.dpp_fetch_inactive = ((modifier >> 18u) & 0x1u) != 0u;
 	operand.dpp_bound_ctrl     = ((modifier >> 19u) & 0x1u) != 0u;
@@ -709,7 +714,7 @@ void DecodeVop1Dpp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_in
 		DecodeVectorGpr(vdst, inst.dst);
 	}
 	DecodeScalarSource(src0 + 256u, pc, inst.src0);
-	ApplyDppModifier(inst.src0, modifier);
+	ApplyDppModifier(inst.src0, modifier, code[word_index] & 0x1ffu);
 	inst.src_count = 1;
 
 	if (!IsVop1FloatSourceOpcode(inst.opcode) && (inst.src0.negate || inst.src0.absolute)) {
@@ -824,7 +829,9 @@ bool IsVopcFloatCompareOpcode(Opcode opcode) {
 		case Opcode::V_CMP_GT_F16:
 		case Opcode::V_CMP_LG_F16:
 		case Opcode::V_CMP_GE_F16:
+		case Opcode::V_CMP_NGT_F16:
 		case Opcode::V_CMP_NEQ_F16:
+		case Opcode::V_CMP_NLT_F16:
 		case Opcode::V_CMPX_LT_F16:
 		case Opcode::V_CMPX_EQ_F16:
 		case Opcode::V_CMPX_LE_F16:
@@ -1046,7 +1053,7 @@ void DecodeVop2Dpp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_in
 	DecodeVectorGpr(vdst, inst.dst);
 	DecodeVectorGpr(vsrc1, inst.src1);
 	DecodeScalarSource(src0 + 256u, pc, inst.src0);
-	ApplyDppModifier(inst.src0, modifier);
+	ApplyDppModifier(inst.src0, modifier, code[word_index] & 0x1ffu);
 	inst.src1.negate       = ((modifier >> 22u) & 0x1u) != 0u;
 	inst.src1.absolute     = ((modifier >> 23u) & 0x1u) != 0u;
 	const bool packed_fmac = inst.opcode == Opcode::V_PK_FMAC_F16;
@@ -1155,7 +1162,7 @@ void DecodeVopcDpp(uint32_t pc, std::span<const uint32_t> code, uint32_t word_in
 	DecodeVectorGpr(vsrc1, inst.src1);
 	DecodeScalarSource(src0 + 256u, pc, inst.src0);
 	inst.dst.kind = IsVopcCompareExec(inst.opcode) ? OperandKind::ExecLo : OperandKind::VccLo;
-	ApplyDppModifier(inst.src0, modifier);
+	ApplyDppModifier(inst.src0, modifier, code[word_index] & 0x1ffu);
 	inst.src1.negate   = ((modifier >> 22u) & 0x1u) != 0u;
 	inst.src1.absolute = ((modifier >> 23u) & 0x1u) != 0u;
 	inst.src_count     = 2;
@@ -1171,6 +1178,7 @@ uint32_t NativeVop3SourceCount(Opcode opcode) {
 		case Opcode::V_MUL_I32_I24:
 		case Opcode::V_ADD_NC_U16:
 		case Opcode::V_SUB_NC_U16:
+		case Opcode::V_MUL_LO_U16:
 		case Opcode::V_MAX_U16:
 		case Opcode::V_MAX_I16:
 		case Opcode::V_MIN_U16:
@@ -1355,7 +1363,8 @@ bool SupportsNativeVop3ResultModifiers(Opcode opcode) {
 }
 
 bool SupportsNativeVop3Clamp(Opcode opcode) {
-	return SupportsNativeVop3ResultModifiers(opcode) || UsesInexactClampControl(opcode);
+	return SupportsNativeVop3ResultModifiers(opcode) || UsesInexactClampControl(opcode) ||
+	       opcode == Opcode::V_MAD_I16;
 }
 
 bool HasUnsupportedNativeVop3Modifiers(Opcode opcode, bool permlane, bool mad_mix,
@@ -1376,7 +1385,7 @@ bool HasUnsupportedNativeVop3Modifiers(Opcode opcode, bool permlane, bool mad_mi
 		return opcode != Opcode::V_FMA_F16 && (clamp != 0u || omod != 0u);
 	}
 	if (IsNativeVop3I16TernaryOpcode(opcode)) {
-		return abs != 0u || clamp != 0u || omod != 0u || neg != 0u;
+		return abs != 0u || (clamp != 0u && !clamp_modifier) || omod != 0u || neg != 0u;
 	}
 	if (IsNativeVop3B16BinaryOpcode(opcode)) {
 		return abs != 0u || clamp != 0u || omod != 0u || neg != 0u;
@@ -1468,7 +1477,7 @@ void DecodeVop2(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 	inst.pc        = pc;
 	inst.family    = Family::VOP2;
 	inst.opcode_id = opcode;
-	inst.opcode    = LookupVop2Opcode(opcode);
+	inst.opcode    = Detail::LookupOpcode(VOP2_OPS, opcode);
 	SetRawWords(inst, code, word_index, 1);
 
 	if (inst.opcode == Opcode::UNSUPPORTED) {
@@ -1510,6 +1519,7 @@ void DecodeVop1(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index
 		return;
 	}
 	switch (src0) {
+		case 233u: DecodeVop1Dpp(pc, code, word_index, opcode, vdst, inst); return;
 		case 249u: DecodeVop1Sdwa(pc, code, word_index, opcode, vdst, inst); return;
 		case 250u: DecodeVop1Dpp(pc, code, word_index, opcode, vdst, inst); return;
 		default: break;

src/graphics/shader/recompiler/frontend/translate/Convert.cpp

@@ -166,9 +166,9 @@ void Translator::V_CVT_PK_U8_F32(const Decoder::Instruction& inst) {
 }
 
 void Translator::V_PACK_B32_F16(const Decoder::Instruction& inst) {
-	const auto low = ReadF16LaneBits(inst.src0, false);
+	const auto low = Read16LaneBits(inst.src0, false);
 	const auto high =
-	    ir.ShiftLeftLogical(ReadF16LaneBits(inst.src1, false), IR::U32(IR::Value(16u)));
+	    ir.ShiftLeftLogical(Read16LaneBits(inst.src1, false), IR::U32(IR::Value(16u)));
 	WriteOperand(DestinationOperand(inst), ir.BitwiseOr(low, high));
 }
 

src/graphics/shader/recompiler/frontend/translate/Integer.cpp

@@ -22,6 +22,18 @@ bool Translator::Integer16Binary(const Decoder::Instruction& inst, IR::ValueOpco
 	return true;
 }
 
+bool Translator::V_MAD_I16(const Decoder::Instruction& inst) {
+	const auto lhs    = ReadU16AsU32(inst.src0, true);
+	const auto rhs    = ReadU16AsU32(inst.src1, true);
+	auto       result = ir.IAdd(ir.IMul(lhs, rhs), ReadU16AsU32(inst.src2, true));
+	if (inst.dst.clamp) {
+		result = IR::U32(ir.Emit(IR::ValueOpcode::SMax32, {result, IR::Value(0xffff8000u)}));
+		result = IR::U32(ir.Emit(IR::ValueOpcode::SMin32, {result, IR::Value(0x7fffu)}));
+	}
+	Write16Bits(DestinationOperand(inst), ir.BitwiseAnd(result, IR::U32(IR::Value(0xffffu))));
+	return true;
+}
+
 bool Translator::V_MED3_I16(const Decoder::Instruction& inst) {
 	const auto result = IR::U32(ir.Emit(
 	    IR::ValueOpcode::SMedTri32, {ReadU16AsU32(inst.src0, true), ReadU16AsU32(inst.src1, true),
@@ -168,6 +180,21 @@ bool Translator::SimpleInteger(const Decoder::Instruction& inst, IR::ValueOpcode
 	return true;
 }
 
+bool Translator::S_ASHR_I64(const Decoder::Instruction& inst) {
+	// Signed 64-bit sources sign-extend literals; generic B64 operands zero-extend them.
+	const auto source =
+	    inst.src0.kind == Decoder::OperandKind::LiteralConstant
+	        ? ir.ConstructU64(
+	              ReadU32(inst.src0),
+	              IR::U32(IR::Value((inst.src0.value & 0x80000000u) ? 0xffffffffu : 0u)))
+	        : ReadU64(inst.src0);
+	const auto result =
+	    ir.Emit(IR::ValueOpcode::ShiftRightArithmetic64, {source, ReadU32(inst.src1)});
+	WriteOperand(inst.dst, result);
+	ir.SetScc(IR::U1(ir.Emit(IR::ValueOpcode::INotEqual64, {result, IR::Value(uint64_t {0})})));
+	return true;
+}
+
 bool Translator::ComposedIntegerBinary(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
                                        bool negate_rhs, bool negate_result, bool update_scc) {
 	const auto lhs = ReadU32(inst.src0);

src/graphics/shader/recompiler/frontend/translate/Memory.cpp

@@ -34,6 +34,7 @@ Decoder::Operand OffsetDecodedRegister(const Decoder::Operand& operand, uint32_t
 	result.negate_hi          = false;
 	result.absolute           = false;
 	result.dpp_ctrl           = 0;
+	result.dpp8               = false;
 	result.dpp_row_mask       = 0xf;
 	result.dpp_bank_mask      = 0xf;
 	result.explicit_sdwa_dst  = false;

src/graphics/shader/recompiler/frontend/translate/Scalar.cpp

@@ -179,6 +179,7 @@ bool Translator::EmitScalar(const Decoder::Instruction& inst) {
 		case O::S_LSHR_B64:
 			return SimpleInteger(inst, IR::ValueOpcode::ShiftRightLogical64, IR::Type::U64, false,
 			                     false, true);
+		case O::S_ASHR_I64: return S_ASHR_I64(inst);
 
 		case O::S_ANDN2_B32:
 			return ComposedIntegerBinary(inst, IR::ValueOpcode::BitwiseAnd32, true, false, true);

src/graphics/shader/recompiler/frontend/translate/Translate.cpp

@@ -9,6 +9,17 @@
 
 namespace Libs::Graphics::ShaderRecompiler::Frontend {
 
+static IR::DppMoveFlags DppFlags(const Decoder::Operand& operand) {
+	return {
+	    .control        = operand.dpp_ctrl,
+	    .row_mask       = operand.dpp_row_mask,
+	    .bank_mask      = operand.dpp_bank_mask,
+	    .fetch_inactive = operand.dpp_fetch_inactive,
+	    .bound_control  = operand.dpp_bound_ctrl,
+	    .dpp8           = operand.dpp8,
+	};
+}
+
 const Decoder::Operand& Translator::SourceAt(const Decoder::Instruction& inst, uint32_t index) {
 	switch (index) {
 		case 0: return inst.src0;
@@ -30,6 +41,7 @@ Decoder::Operand Translator::DestinationOperand(const Decoder::Instruction& inst
 			continue;
 		}
 		destination.dpp                = true;
+		destination.dpp8               = source.dpp8;
 		destination.dpp_ctrl           = source.dpp_ctrl;
 		destination.dpp_row_mask       = source.dpp_row_mask;
 		destination.dpp_bank_mask      = source.dpp_bank_mask;
@@ -106,6 +118,7 @@ Decoder::Operand Translator::PlainOperand(const Decoder::Operand& operand) {
 	result.dpp_fetch_inactive = false;
 	result.dpp_bound_ctrl     = false;
 	result.dpp                = false;
+	result.dpp8               = false;
 	return result;
 }
 
@@ -167,14 +180,8 @@ IR::U32 Translator::ReadScalarCode(uint32_t code) {
 
 IR::U32 Translator::ApplyBitSourceModifiers(const Decoder::Operand& operand, IR::U32 value) {
 	if (operand.dpp) {
-		const IR::DppMoveFlags flags {
-		    .control        = static_cast<uint16_t>(operand.dpp_ctrl),
-		    .row_mask       = static_cast<uint8_t>(operand.dpp_row_mask),
-		    .bank_mask      = static_cast<uint8_t>(operand.dpp_bank_mask),
-		    .fetch_inactive = operand.dpp_fetch_inactive,
-		    .bound_control  = operand.dpp_bound_ctrl,
-		};
-		value = IR::U32(ir.Emit(IR::ValueOpcode::DppMoveU32, {value, ir.GetExec()}, flags));
+		value =
+		    IR::U32(ir.Emit(IR::ValueOpcode::DppMoveU32, {value, ir.GetExec()}, DppFlags(operand)));
 	}
 	if (operand.sdwa_sel != 6u) {
 		uint32_t offset = 0;
@@ -301,15 +308,8 @@ void Translator::WriteRawU32(const Decoder::Operand& operand, IR::U32 value) {
 			const auto reg = static_cast<IR::VectorReg>(operand.reg);
 			const auto old = ir.GetVectorReg(reg);
 			if (operand.dpp) {
-				const IR::DppMoveFlags flags {
-				    .control        = static_cast<uint16_t>(operand.dpp_ctrl),
-				    .row_mask       = static_cast<uint8_t>(operand.dpp_row_mask),
-				    .bank_mask      = static_cast<uint8_t>(operand.dpp_bank_mask),
-				    .fetch_inactive = operand.dpp_fetch_inactive,
-				    .bound_control  = operand.dpp_bound_ctrl,
-				};
-				value = IR::U32(
-				    ir.Emit(IR::ValueOpcode::DppUpdateU32, {value, old, ir.GetExec()}, flags));
+				value = IR::U32(ir.Emit(IR::ValueOpcode::DppUpdateU32, {value, old, ir.GetExec()},
+				                        DppFlags(operand)));
 			} else {
 				value = ir.Select(ir.GetExec(), value, old);
 			}
@@ -541,10 +541,7 @@ IR::U32 Translator::ReadU16LaneRaw(const Decoder::Operand& operand, bool high_la
 
 IR::U32 Translator::ReadU16LaneAsU32(const Decoder::Operand& operand, bool high_lane,
                                      bool sign_extend) {
-	auto value = ReadU16LaneRaw(operand, high_lane);
-	if (high_lane ? operand.negate_hi : operand.negate) {
-		value = ir.BitwiseAnd(ir.ISub(IR::U32(IR::Value(0u)), value), IR::U32(IR::Value(0xffffu)));
-	}
+	auto value = Read16LaneBits(operand, high_lane);
 	if (sign_extend || operand.sdwa_sext) {
 		value = IR::U32(
 		    ir.Emit(IR::ValueOpcode::BitFieldSExtract, {value, IR::Value(0u), IR::Value(16u)}));
@@ -556,12 +553,13 @@ IR::U32 Translator::ReadU16AsU32(const Decoder::Operand& operand, bool sign_exte
 	return ReadU16LaneAsU32(operand, false, sign_extend);
 }
 
-IR::U32 Translator::ReadF16LaneBits(const Decoder::Operand& operand, bool high_lane) {
+IR::U32 Translator::Read16LaneBits(const Decoder::Operand& operand, bool high_lane) {
 	auto value = ReadU16LaneRaw(operand, high_lane);
 	if (operand.absolute) {
 		value = ir.BitwiseAnd(value, IR::U32(IR::Value(0x7fffu)));
 	}
 	if (high_lane ? operand.negate_hi : operand.negate) {
+		// RDNA2 source NEG flips the sign bit, including packed integer operands.
 		value = ir.BitwiseXor(value, IR::U32(IR::Value(0x8000u)));
 	}
 	return value;

src/graphics/shader/recompiler/frontend/translate/Translator.h

@@ -44,7 +44,7 @@ private:
 	IR::U32 ReadU16LaneRaw(const Decoder::Operand& operand, bool high_lane);
 	IR::U32 ReadU16LaneAsU32(const Decoder::Operand& operand, bool high_lane, bool sign_extend);
 	IR::U32 ReadU16AsU32(const Decoder::Operand& operand, bool sign_extend);
-	IR::U32 ReadF16LaneBits(const Decoder::Operand& operand, bool high_lane);
+	IR::U32 Read16LaneBits(const Decoder::Operand& operand, bool high_lane);
 	std::array<IR::U32, 2> ExtractU64(IR::U64 value);
 	void    WriteU32Pair(const Decoder::Operand& operand, const std::array<IR::U32, 2>& value);
 	IR::U32 ConditionBit(const Decoder::Operand& operand);
@@ -164,6 +164,7 @@ private:
 	bool FloatCube(const Decoder::Instruction& inst, uint32_t result_kind);
 	bool Integer16Shift(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool arithmetic);
 	bool Integer16Binary(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool sign);
+	bool V_MAD_I16(const Decoder::Instruction& inst);
 	bool V_MED3_I16(const Decoder::Instruction& inst);
 	bool PackedInteger16Shift(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
 	                          bool arithmetic);
@@ -176,6 +177,7 @@ private:
 	                     bool negate_result);
 	bool SimpleInteger(const Decoder::Instruction& inst, IR::ValueOpcode opcode, IR::Type type,
 	                   bool reverse, bool mask_shift_count, bool update_scc);
+	bool S_ASHR_I64(const Decoder::Instruction& inst);
 	bool ComposedIntegerBinary(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
 	                           bool negate_rhs, bool negate_result, bool update_scc);
 	bool V_AND_OR_B32(const Decoder::Instruction& inst);

src/graphics/shader/recompiler/frontend/translate/Vector.cpp

@@ -264,6 +264,9 @@ bool Translator::EmitVector(const Decoder::Instruction& inst) {
 		case O::V_CMPX_LE_F16:
 			EmitFloatCompare(inst, IR::ValueOpcode::FPOrdLessThanEqual32, true, true);
 			return true;
+		case O::V_CMP_NGT_F16:
+			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordLessThanEqual32, true, false);
+			return true;
 		case O::V_CMPX_NGT_F16:
 			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordLessThanEqual32, true, true);
 			return true;
@@ -273,6 +276,9 @@ bool Translator::EmitVector(const Decoder::Instruction& inst) {
 		case O::V_CMPX_NEQ_F16:
 			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordNotEqual32, true, true);
 			return true;
+		case O::V_CMP_NLT_F16:
+			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordGreaterThanEqual32, true, false);
+			return true;
 		case O::V_CMPX_NLT_F16:
 			EmitFloatCompare(inst, IR::ValueOpcode::FPUnordGreaterThanEqual32, true, true);
 			return true;
@@ -321,6 +327,8 @@ bool Translator::EmitVector(const Decoder::Instruction& inst) {
 		case O::V_ADD_NC_I16: return Integer16Binary(inst, IR::ValueOpcode::IAdd32, false);
 		case O::V_SUB_NC_U16:
 		case O::V_SUB_NC_I16: return Integer16Binary(inst, IR::ValueOpcode::ISub32, false);
+		case O::V_MUL_LO_U16: return Integer16Binary(inst, IR::ValueOpcode::IMul32, false);
+		case O::V_MAD_I16: return V_MAD_I16(inst);
 		case O::V_MED3_I16: return V_MED3_I16(inst);
 		case O::V_MIN_I16: return Integer16Binary(inst, IR::ValueOpcode::SMin32, true);
 		case O::V_MAX_I16: return Integer16Binary(inst, IR::ValueOpcode::SMax32, true);

src/graphics/shader/recompiler/ir/opcodes/ValueOpcodes.h

@@ -34,11 +34,12 @@ struct ImageOpcodeInfo {
 };
 
 struct DppMoveFlags {
-	uint16_t control        = 0;
-	uint8_t  row_mask       = 0xf;
-	uint8_t  bank_mask      = 0xf;
+	uint32_t control   : 24 = 0;
+	uint32_t row_mask  : 4  = 0xf;
+	uint32_t bank_mask : 4  = 0xf;
 	bool     fetch_inactive = false;
 	bool     bound_control  = false;
+	bool     dpp8           = false;
 };
 static_assert(sizeof(DppMoveFlags) <= sizeof(uint64_t));
 static_assert(std::is_trivially_copyable_v<DppMoveFlags>);

src/graphics/shader/recompiler/ir/passes/SrtWalker.cpp

@@ -479,8 +479,6 @@ private:
 		return std::bit_cast<float>(static_cast<uint32_t>(bits));
 	}
 
-	static uint64_t Float32Bits(float value) { return std::bit_cast<uint32_t>(value); }
-
 	bool EvaluateWide(Value value, uint64_t& result) {
 		value = value.Resolve();
 		if (value.IsImmediate()) {
@@ -490,7 +488,7 @@ private:
 				case Type::U16: result = value.U16(); return true;
 				case Type::U32: result = value.U32(); return true;
 				case Type::U64: result = value.U64(); return true;
-				case Type::F32: result = Float32Bits(value.F32Value()); return true;
+				case Type::F32: result = std::bit_cast<uint32_t>(value.F32Value()); return true;
 				default: return false;
 			}
 		}
@@ -733,7 +731,7 @@ private:
 				return false;
 			case ValueOpcode::ConvertF32U32:
 				if (Arg(inst, 0, a)) {
-					result = Float32Bits(static_cast<float>(static_cast<uint32_t>(a)));
+					result = std::bit_cast<uint32_t>(static_cast<float>(static_cast<uint32_t>(a)));
 					return true;
 				}
 				return false;
@@ -750,13 +748,13 @@ private:
 				return false;
 			case ValueOpcode::FPMul32:
 				if (binary()) {
-					result = Float32Bits(Float32(a) * Float32(b));
+					result = std::bit_cast<uint32_t>(Float32(a) * Float32(b));
 					return true;
 				}
 				return false;
 			case ValueOpcode::FPTrunc32:
 				if (Arg(inst, 0, a)) {
-					result = Float32Bits(std::trunc(Float32(a)));
+					result = std::bit_cast<uint32_t>(std::trunc(Float32(a)));
 					return true;
 				}
 				return false;

src/graphics/shader/rectListShader.cpp

@@ -60,12 +60,12 @@ public:
 	    : parameters(parameters_) {
 		builder.AddMemoryModel(spv::AddressingModelLogical, spv::MemoryModelGLSL450);
 
-		void_type       = Type(spv::OpTypeVoid);
-		uint_type       = Type(spv::OpTypeInt, 32u, 0u);
-		int_type        = Type(spv::OpTypeInt, 32u, 1u);
-		float_type      = Type(spv::OpTypeFloat, 32u);
-		vec4_float_type = Type(spv::OpTypeVector, float_type, 4u);
-		function_type   = Type(spv::OpTypeFunction, void_type);
+		void_type       = builder.Type(spv::OpTypeVoid);
+		uint_type       = builder.Type(spv::OpTypeInt, 32u, 0u);
+		int_type        = builder.Type(spv::OpTypeInt, 32u, 1u);
+		float_type      = builder.Type(spv::OpTypeFloat, 32u);
+		vec4_float_type = builder.Type(spv::OpTypeVector, float_type, 4u);
+		function_type   = builder.Type(spv::OpTypeFunction, void_type);
 
 		per_vertex_type = builder.DecoratedType(
 		    spv::OpTypeStruct,
@@ -76,12 +76,12 @@ public:
 		ptr_input_vec4_float  = Pointer(spv::StorageClassInput, vec4_float_type);
 		ptr_output_vec4_float = Pointer(spv::StorageClassOutput, vec4_float_type);
 		if (model == spv::ExecutionModelTessellationControl) {
-			bool_type        = Type(spv::OpTypeBool);
-			vec2_bool_type   = Type(spv::OpTypeVector, bool_type, 2u);
-			vec2_float_type  = Type(spv::OpTypeVector, float_type, 2u);
+			bool_type        = builder.Type(spv::OpTypeBool);
+			vec2_bool_type   = builder.Type(spv::OpTypeVector, bool_type, 2u);
+			vec2_float_type  = builder.Type(spv::OpTypeVector, float_type, 2u);
 			ptr_output_float = Pointer(spv::StorageClassOutput, float_type);
 		} else {
-			vec3_float_type = Type(spv::OpTypeVector, float_type, 3u);
+			vec3_float_type = builder.Type(spv::OpTypeVector, float_type, 3u);
 			ptr_input_float = Pointer(spv::StorageClassInput, float_type);
 		}
 	}
@@ -161,8 +161,8 @@ public:
 			Store(Access(ptr_output_vec4_float, outputs[i], invocation), value);
 		}
 
-		Emit(spv::OpReturn);
-		Emit(spv::OpFunctionEnd);
+		builder.AddFunction(spv::OpReturn);
+		builder.AddFunction(spv::OpFunctionEnd);
 		return builder.Build();
 	}
 
@@ -184,27 +184,22 @@ public:
 			      Load(vec4_float_type, Access(ptr_input_vec4_float, inputs[i], index)));
 		}
 
-		Emit(spv::OpReturn);
-		Emit(spv::OpFunctionEnd);
+		builder.AddFunction(spv::OpReturn);
+		builder.AddFunction(spv::OpFunctionEnd);
 		return builder.Build();
 	}
 
 private:
-	template <typename... Args>
-	uint32_t Type(spv::Op opcode, Args... operands) {
-		return builder.Type(opcode, operands...);
-	}
-
 	uint32_t Constant(uint32_t type, uint32_t value) {
 		return builder.Constant(spv::OpConstant, type, value);
 	}
 
 	uint32_t Pointer(spv::StorageClass storage, uint32_t type) {
-		return Type(spv::OpTypePointer, storage, type);
+		return builder.Type(spv::OpTypePointer, storage, type);
 	}
 
 	uint32_t Array(uint32_t type, uint32_t size) {
-		return Type(spv::OpTypeArray, type, Uint(size));
+		return builder.Type(spv::OpTypeArray, type, Uint(size));
 	}
 
 	template <typename... Args>
@@ -221,11 +216,6 @@ private:
 		return id;
 	}
 
-	template <typename... Args>
-	void Emit(spv::Op opcode, Args... operands) {
-		builder.AddFunction(opcode, operands...);
-	}
-
 	template <typename... Args>
 	uint32_t Access(uint32_t pointer_type, uint32_t base, Args... indices) {
 		return Result(spv::OpAccessChain, pointer_type, base, indices...);
@@ -233,7 +223,7 @@ private:
 
 	uint32_t Load(uint32_t type, uint32_t pointer) { return Result(spv::OpLoad, type, pointer); }
 
-	void Store(uint32_t pointer, uint32_t value) { Emit(spv::OpStore, pointer, value); }
+	void Store(uint32_t pointer, uint32_t value) { builder.AddFunction(spv::OpStore, pointer, value); }
 
 	uint32_t Int(uint32_t value) { return Constant(int_type, value); }
 

src/graphics/shader/shader.cpp

@@ -4,7 +4,6 @@
 #include "common/common.h"
 #include "common/emulatorConfig.h"
 #include "common/logging/log.h"
-#include "common/magicEnum.h"
 #include "common/profiler.h"
 #include "common/stringUtils.h"
 #include "graphics/guest_gpu/gpu_defs.h"
@@ -722,6 +721,7 @@ void BuildStageStaticKey(const ShaderPixelInputInfo& info, std::vector<uint32_t>
 	key.push_back(static_cast<uint32_t>(info.ps_depth_export_enable));
 	key.push_back(static_cast<uint32_t>(info.ps_sample_mask_export_enable));
 	key.push_back(static_cast<uint32_t>(info.ps_early_z));
+	key.push_back(static_cast<uint32_t>(info.dual_source_blending));
 	key.insert(key.end(), std::begin(info.target_output_mode), std::end(info.target_output_mode));
 	for (uint32_t base = 0; base < info.target_export_mapping.size(); base += 4u) {
 		uint32_t packed = 0;
@@ -1006,8 +1006,4 @@ void ShaderDbgDumpInputInfo(const ShaderComputeInputInfo& info) {
 	     info.group_id[1] ? "true" : "false", info.group_id[2] ? "true" : "false");
 }
 
-bool ShaderAddressValid(uint64_t addr) {
-	return reinterpret_cast<const uint32_t*>(addr) != nullptr;
-}
-
 } // namespace Libs::Graphics

src/graphics/shader/shader.h

@@ -175,6 +175,7 @@ struct ShaderPixelInputInfo {
 	bool                                           ps_depth_export_enable       = false;
 	bool                                           ps_sample_mask_export_enable = false;
 	bool                                           ps_sample_shading            = false;
+	bool                                           dual_source_blending         = false;
 	bool                                           ps_early_z                   = false;
 	bool                                           ps_execute_on_noop           = false;
 	ShaderStageRuntime                             stage;
@@ -303,7 +304,6 @@ void ShaderMapUserData(uint64_t addr, const ShaderMappedData& data);
 void     ShaderDbgDumpInputInfo(const ShaderVertexInputInfo& info);
 void     ShaderDbgDumpInputInfo(const ShaderPixelInputInfo& info);
 void     ShaderDbgDumpInputInfo(const ShaderComputeInputInfo& info);
-bool ShaderAddressValid(uint64_t addr);
 
 } // namespace Libs::Graphics
 

src/kernel/fileSystem.cpp

@@ -251,16 +251,15 @@ void FileDescriptors::CloseAll() {
 void MountPoints::Mount(const std::filesystem::path& folder, const std::string& point) {
 	Common::LockGuard lock(m_mutex);
 
-	auto folder_str = Common::FixDirectorySlash(Common::PathToGenericString(folder));
 	auto point_str  = Common::FixDirectorySlash(point);
 
 	Umount(point_str);
 
 	MountPair p;
-	p.dir   = folder_str;
+	p.dir   = folder;
 	p.point = point_str;
 
-	m_mount_pairs.push_back(p);
+	m_mount_pairs.push_back(std::move(p));
 }
 
 void MountPoints::Umount(const std::string& folder_or_point) {
@@ -270,7 +269,7 @@ void MountPoints::Umount(const std::string& folder_or_point) {
 
 	const auto it = std::find_if(
 	    m_mount_pairs.begin(), m_mount_pairs.end(), [&folder_or_point_str](const MountPair& p) {
-		    return Common::PathToGenericString(p.dir) == folder_or_point_str ||
+		    return Common::FixDirectorySlash(Common::PathToGenericString(p.dir)) == folder_or_point_str ||
 		           p.point == folder_or_point_str;
 	    });
 	if (it != m_mount_pairs.end()) {
@@ -334,23 +333,26 @@ std::filesystem::path MountPoints::ResolvePath(const std::string& mounted_name)
 
 	// Match the entire guest path so a mount root works with or without a trailing slash.
 	const auto mounted_path = Common::FixDirectorySlash(mounted_name);
-	const auto it = std::find_if(
+	const auto it           = std::find_if(
 	    m_mount_pairs.begin(), m_mount_pairs.end(),
-	    [&mounted_path](const MountPair& p) { return Common::StartsWith(mounted_path, p.point); });
+	    [&mounted_path](const MountPair& p) { return mounted_path.starts_with(p.point); });
 	if (it != m_mount_pairs.end()) {
 		const auto& p = *it;
 		auto rel_path = Common::RemoveFirst(Common::FixFilenameSlash(mounted_name), p.point.size());
-		while (Common::StartsWith(rel_path, '/')) {
+		while (rel_path.starts_with('/')) {
 			rel_path = Common::RemoveFirst(rel_path, 1);
 		}
+
+		const auto native_rel_path = Common::PathFromUtf8(rel_path);
+
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 		if (HasWindowsForbiddenFilenameCharacter(rel_path)) {
 			::printf("FileSystem: Windows-incompatible guest filename: %s\n",
 			         mounted_name.c_str());
 		}
-		return p.dir / rel_path;
+		return p.dir / native_rel_path;
 #else
-		return ResolvePathIgnoringCase(p.dir / rel_path);
+		return ResolvePathIgnoringCase(p.dir / native_rel_path);
 #endif
 	}
 

src/kernel/memory.cpp

@@ -2,7 +2,6 @@
 
 #include "common/assert.h"
 #include "common/logging/log.h"
-#include "common/magicEnum.h"
 #include "common/stringUtils.h"
 #include "common/threads.h"
 #include "common/virtualMemory.h"
@@ -14,9 +13,11 @@
 #include <algorithm>
 #include <array>
 #include <atomic>
+#include <bit>
 #include <cstddef>
 #include <cstdlib>
 #include <cstring>
+#include <magic_enum.hpp>
 #include <map>
 #include <memory>
 #include <mutex>
@@ -2324,8 +2325,8 @@ int32_t KYTY_SYSV_ABI KernelMapNamedFlexibleMemory(void** addr_in_out, size_t le
 	     "\t flags    = 0x%08" PRIx32 "\n"
 	     "\t name     = %s\n"
 	     "\t gpu_mode = %s\n",
-	     in_addr, out_addr, len, Common::EnumName(mode).c_str(), static_cast<uint32_t>(flags), name,
-	     Common::EnumName(gpu_mode).c_str());
+	     in_addr, out_addr, len, magic_enum::enum_name(mode), static_cast<uint32_t>(flags), name,
+	     magic_enum::enum_name(gpu_mode));
 
 	MapGpuRange(out_addr, len);
 
@@ -3041,8 +3042,8 @@ int KYTY_SYSV_ABI KernelMapDirectMemory(void** addr, size_t len, int prot, int f
 	     "\t shared   = %s\n"
 	     "\t reason   = %s\n",
 	     in_addr, out_addr, static_cast<uint64_t>(direct_memory_start), len,
-	     Common::EnumName(mode).c_str(), static_cast<uint32_t>(flags), alignment,
-	     Common::EnumName(gpu_mode).c_str(), shared_backing ? "yes" : "no", shared_reason);
+	     magic_enum::enum_name(mode), static_cast<uint32_t>(flags), alignment,
+	     magic_enum::enum_name(gpu_mode), shared_backing ? "yes" : "no", shared_reason);
 
 	if (out_addr == 0) {
 		if (consumed_reservation) {
@@ -3305,7 +3306,7 @@ static bool ReplaceFixedRangeWithReserved(uint64_t start, uint64_t size) {
 			           "\t reserve-fixed replace: backend unmap failed at 0x%016" PRIx64
 			           ", size=0x%016" PRIx64 ", type=%s\n",
 			           chunk.range.start, chunk.range.size,
-			           Common::EnumName(chunk.range.type).c_str());
+			           magic_enum::enum_name(chunk.range.type));
 			if (!restore_chunks()) {
 				EXIT("reserve-fixed backend-unmap rollback failed\n");
 			}
@@ -3769,7 +3770,7 @@ int KYTY_SYSV_ABI KernelMprotect(const void* addr, size_t len, int prot) {
 	}
 	g_virtual_ranges->Protect(aligned_addr, aligned_len, prot);
 
-	LOGF("\t prot: %s -> %s\n", Common::EnumName(old_mode).c_str(), Common::EnumName(mode).c_str());
+	LOGF("\t prot: %s -> %s\n", magic_enum::enum_name(old_mode), magic_enum::enum_name(mode));
 
 	return OK;
 }
@@ -3879,10 +3880,6 @@ static bool IsAligned(uint64_t value, uint64_t alignment) {
 	return alignment == 0 || (value & (alignment - 1u)) == 0;
 }
 
-static bool IsPowerOfTwo(uint64_t value) {
-	return value != 0 && (value & (value - 1u)) == 0;
-}
-
 static void MemoryPoolSubtractCommitted(uint64_t len) {
 	auto current = g_memory_pool_committed.load(std::memory_order_relaxed);
 	while (current != 0) {
@@ -3904,7 +3901,7 @@ int KYTY_SYSV_ABI KernelMemoryPoolExpand(int64_t search_start, int64_t search_en
 	if (search_start < 0 || search_end <= search_start || len == 0 ||
 	    (len & (POOL_PAGE_SIZE - 1u)) != 0 || phys_addr_out == nullptr ||
 	    (alignment != 0 &&
-	     (!IsPowerOfTwo(alignment) || (alignment & (POOL_PAGE_SIZE - 1u)) != 0))) {
+	     (!std::has_single_bit(alignment) || (alignment & (POOL_PAGE_SIZE - 1u)) != 0))) {
 		return KERNEL_ERROR_EINVAL;
 	}
 	if (static_cast<uint64_t>(search_end - search_start) < len) {
@@ -3953,7 +3950,7 @@ int KYTY_SYSV_ABI KernelMemoryPoolReserve(void* addr_in, size_t len, size_t alig
 		return KERNEL_ERROR_EINVAL;
 	}
 	if (alignment != 0 &&
-	    (!IsPowerOfTwo(alignment) || !IsAligned(alignment, POOL_RESERVE_ALIGNMENT))) {
+	    (!std::has_single_bit(alignment) || !IsAligned(alignment, POOL_RESERVE_ALIGNMENT))) {
 		return KERNEL_ERROR_EINVAL;
 	}
 

src/kernel/pthread.cpp

@@ -149,7 +149,7 @@ static uint64_t KernelGetTscFrequencyNative() {
 		}
 
 		KernelReadTscNative();
-		Common::Thread::Sleep(1);
+		std::this_thread::sleep_for(std::chrono::milliseconds(1));
 		KernelReadTscNative();
 
 		const auto host_start = Common::Timer::QueryPerformanceCounter();
@@ -993,7 +993,7 @@ static void FreeDetachedThreads(void* /*arg*/) {
 	auto* pthread_pool = g_pthread_context->GetPthreadPool();
 
 	while (true) {
-		Common::Thread::Sleep(10000);
+		std::this_thread::sleep_for(std::chrono::milliseconds(10000));
 		pthread_pool->FreeDetachedThreads();
 	}
 }
@@ -1202,13 +1202,6 @@ static int32_t GetDstSeconds() {
 #endif
 }
 
-#if KYTY_PLATFORM != KYTY_PLATFORM_WINDOWS
-static void sec_to_timeval(KernelTimeval* ts, double sec) {
-	ts->tv_sec  = static_cast<int64_t>(sec);
-	ts->tv_usec = static_cast<int64_t>((sec - static_cast<double>(ts->tv_sec)) * 1000000.0);
-}
-#endif
-
 static bool GetPosixClockId(KernelClockid clock_id, clockid_t* out) {
 	EXIT_IF(out == nullptr);
 

src/launcher/forms/configuration_edit_dialog.ui

@@ -329,12 +329,4 @@
   </customwidget>
  </customwidgets>
  <resources/>
- <connections>
-  <connection>
-   <sender>cancel_button</sender>
-   <signal>clicked()</signal>
-   <receiver>ConfigurationEditDialog</receiver>
-   <slot>reject()</slot>
-  </connection>
- </connections>
 </ui>

src/launcher/include/common.h

@@ -1,20 +0,0 @@
-#ifndef LAUNCHER_INCLUDE_COMMON_H_
-#define LAUNCHER_INCLUDE_COMMON_H_
-
-// IWYU pragma: begin_exports
-#include <QArgument>
-#include <QObject>
-
-#include <cinttypes>
-#include <cstdint>
-#include <cstdio>
-// IWYU pragma: end_exports
-
-#define KYTY_QT_CLASS_NO_COPY(name)                                                                \
-public:                                                                                            \
-	name(const name&)                = delete; /* NOLINT(bugprone-macro-parentheses) */            \
-	name& operator=(const name&)     = delete; /* NOLINT(bugprone-macro-parentheses) */            \
-	name(name&&) noexcept            = delete; /* NOLINT(bugprone-macro-parentheses) */            \
-	name& operator=(name&&) noexcept = delete; /* NOLINT(bugprone-macro-parentheses) */
-
-#endif /* LAUNCHER_INCLUDE_COMMON_H_ */

src/launcher/include/compatibilityDatabase.h

@@ -1,7 +1,6 @@
 #ifndef COMPATIBILITY_DATABASE_H
 #define COMPATIBILITY_DATABASE_H
 
-#include "common.h"
 #include "configuration.h"
 
 #include <QMap>
@@ -15,7 +14,6 @@ struct CompatibilityEntry {
 
 class CompatibilityDatabase: public QObject {
 	Q_OBJECT
-	KYTY_QT_CLASS_NO_COPY(CompatibilityDatabase);
 
 public:
 	explicit CompatibilityDatabase(bool local, QObject* parent = nullptr);

src/launcher/include/configuration.h

@@ -1,13 +1,13 @@
 #ifndef LAUNCHER_INCLUDE_CONFIGURATION_H_
 #define LAUNCHER_INCLUDE_CONFIGURATION_H_
 
-#include "common.h"
 #include "common/emulatorConfig.h"
 
 #include <QByteArray>
 #include <QChar>
 #include <QMetaEnum>
 #include <QMetaType>
+#include <QObject>
 #include <QSettings>
 #include <QString>
 #include <QStringList>

src/launcher/include/configurationEditDialog.h

@@ -1,8 +1,6 @@
 #ifndef CONFIGURATION_EDIT_DIALOG_H
 #define CONFIGURATION_EDIT_DIALOG_H
 
-#include "common.h"
-
 #include <QDialog>
 #include <QString>
 #include <QStringList>
@@ -22,7 +20,6 @@ class ConfigurationEditDialog;
 
 class ConfigurationEditDialog: public QDialog {
 	Q_OBJECT
-	KYTY_QT_CLASS_NO_COPY(ConfigurationEditDialog);
 
 public:
 	explicit ConfigurationEditDialog(Configuration& info, QWidget* parent = nullptr);
@@ -31,7 +28,6 @@ public:
 	static void WriteSettings(QSettings& s);
 	static void ReadSettings(QSettings& s);
 
-	void                      SetTitle(const QString& str);
 	void                      SetGameDirectories(const QStringList& dirs);
 	[[nodiscard]] QStringList GetGameDirectories() const;
 
@@ -55,8 +51,6 @@ protected:
 
 	/*slots:*/
 
-	void update_info();
-	void adjust_size();
 	void save();
 	void clear();
 	void add_game_directory();

src/launcher/include/configurationItem.h

@@ -1,8 +1,6 @@
 #ifndef CONFIGURATION_ITEM_H
 #define CONFIGURATION_ITEM_H
 
-#include "common.h"
-
 #include <QObject>
 #include <QTreeWidgetItem>
 
@@ -23,8 +21,6 @@ public:
 	void Update();
 	bool operator<(const QTreeWidgetItem& other) const override;
 
-	KYTY_QT_CLASS_NO_COPY(ConfigurationItem);
-
 	Configuration&                     GetInfo() { return *m_info; }
 	[[nodiscard]] const Configuration& GetInfo() const { return *m_info; }
 	// void                                     SetInfo(Configuration* info);

src/launcher/include/configurationListWidget.h

@@ -1,7 +1,6 @@
 #ifndef CONFIGURATION_LIST_WIDGET_H
 #define CONFIGURATION_LIST_WIDGET_H
 
-#include "common.h"
 #include "configuration.h"
 
 #include <QMap>
@@ -14,7 +13,6 @@ class CompatibilityDatabase;
 class QEvent;
 class QTreeWidgetItem;
 class QPoint;
-class MainDialog;
 
 namespace Ui {
 class ConfigurationListWidget;
@@ -22,27 +20,20 @@ class ConfigurationListWidget;
 
 class ConfigurationListWidget: public QWidget {
 	Q_OBJECT
-	KYTY_QT_CLASS_NO_COPY(ConfigurationListWidget);
 
 public:
 	explicit ConfigurationListWidget(QWidget* parent = nullptr);
 	~ConfigurationListWidget() override;
 
-	void               SetRunEnabled(bool flag) { m_run_enabled = flag; }
-	[[nodiscard]] bool IsRunEnabled() const { return m_run_enabled; }
+	void SetRunEnabled(bool flag) { m_run_enabled = flag; }
 
 	[[nodiscard]] const ConfigurationItem* GetSelectedItem() const { return m_selected_item; }
 	ConfigurationItem*                     GetSelectedItem() { return m_selected_item; }
 
-	void        SetMainDialog(MainDialog* main_dialog) { m_main_dialog = main_dialog; }
-	MainDialog* GetMainDialog() { return m_main_dialog; }
-
 	[[nodiscard]] const QString&     GetSettingsFile() const { return m_settings_file; }
-	[[nodiscard]] const QStringList& GetGameDirectories() const { return m_game_dirs; }
 	[[nodiscard]] const QStringList& GetHostInputMapping() const {
 		return m_global_info.host_input_mapping;
 	}
-	[[nodiscard]] bool CanViewSelectedTrophies() const;
 
 	bool EnsureGameDirectory();
 	void ScanGameDirectory();
@@ -66,8 +57,6 @@ protected slots:
 	void delete_configuartion();
 	void edit_global_settings();
 	void edit_input_mapping();
-	void run_configuration();
-	void list_currentItemChanged(QTreeWidgetItem* current, QTreeWidgetItem* previous);
 	void list_itemDoubleClicked(QTreeWidgetItem* witem, int column);
 	void show_context_menu(const QPoint& pos);
 	void open_game_folder();
@@ -84,7 +73,6 @@ private:
 	ConfigurationItem*            m_selected_item = nullptr;
 	bool                          m_run_enabled   = true;
 	Ui::ConfigurationListWidget*  m_ui            = nullptr;
-	MainDialog*                   m_main_dialog   = nullptr;
 	QString                       m_settings_file;
 	QStringList                   m_game_dirs;
 	Configuration                 m_global_info;

src/launcher/include/mainDialog.h

@@ -1,8 +1,6 @@
 #ifndef MAIN_DIALOG_H
 #define MAIN_DIALOG_H
 
-#include "common.h"
-
 #include <QDialog>
 #include <QString>
 
@@ -15,11 +13,9 @@ class QResizeEvent;
 class Configuration;
 class MainDialog: public QDialog {
 	Q_OBJECT
-	KYTY_QT_CLASS_NO_COPY(MainDialog);
 
 signals:
 	void Start();
-	void Quit();
 	void Resize();
 
 public:

src/launcher/include/mandatoryLineEdit.h

@@ -1,9 +1,8 @@
 #ifndef MANDATORY_LINEEDIT_H
 #define MANDATORY_LINEEDIT_H
 
-#include "common.h"
-
 #include <QLineEdit>
+#include <QObject>
 #include <QString>
 
 class QPaintEvent;

src/launcher/include/patchesDialog.h

@@ -1,8 +1,6 @@
 #ifndef PATCHES_DIALOG_H
 #define PATCHES_DIALOG_H
 
-#include "common.h"
-
 #include <QDialog>
 #include <QString>
 
@@ -16,8 +14,6 @@ public:
 	explicit PatchesDialog(const Configuration& game, QWidget* parent = nullptr);
 	~PatchesDialog() override = default;
 
-	KYTY_QT_CLASS_NO_COPY(PatchesDialog);
-
 	[[nodiscard]] static bool    IsSupportedTitleId(const QString& title_id);
 	[[nodiscard]] static QString PatchPlanPath(const QString& title_id);
 

src/launcher/include/trophyViewerDialog.h

@@ -1,8 +1,6 @@
 #ifndef TROPHY_VIEWER_DIALOG_H
 #define TROPHY_VIEWER_DIALOG_H
 
-#include "common.h"
-
 #include <QDialog>
 
 class QTabWidget;
@@ -10,8 +8,6 @@ class QWidget;
 
 class Configuration;
 class TrophyViewerDialog: public QDialog {
-	KYTY_QT_CLASS_NO_COPY(TrophyViewerDialog);
-
 public:
 	explicit TrophyViewerDialog(QWidget* parent = nullptr);
 

src/launcher/src/configurationEditDialog.cpp

@@ -108,6 +108,7 @@ ConfigurationEditDialog::ConfigurationEditDialog(Configuration& info, QWidget* p
 	InitGameDirectories();
 
 	connect(m_ui->ok_button, &QPushButton::clicked, this, &ConfigurationEditDialog::save);
+	connect(m_ui->cancel_button, &QPushButton::clicked, this, &QDialog::reject);
 	connect(m_ui->clear_button, &QPushButton::clicked, this, &ConfigurationEditDialog::clear);
 	connect(m_ui->comboBox_shader_log_direction, &QComboBox::currentTextChanged, this,
 	        [this](const QString& text) {
@@ -213,7 +214,7 @@ void ConfigurationEditDialog::Init(const Configuration& info) {
 #endif
 	QVulkanInstance instance;
 	instance.setApiVersion(QVersionNumber(1, 3, 0));
-#if !defined(__APPLE__)
+#if !defined(__APPLE__) && QT_VERSION >= QT_VERSION_CHECK(6, 6, 0)
 	instance.setFlags(QVulkanInstance::NoPortabilityDrivers);
 #endif
 	if (instance.create()) {
@@ -306,10 +307,6 @@ void ConfigurationEditDialog::InitGameDirectories() {
 	update_game_directory_buttons();
 }
 
-void ConfigurationEditDialog::SetTitle(const QString& str) {
-	setWindowTitle(str);
-}
-
 void ConfigurationEditDialog::SetGameDirectories(const QStringList& dirs) {
 	m_show_game_dirs = true;
 	m_game_dirs_list->clear();
@@ -400,14 +397,6 @@ static void UpdateInfo(Configuration& info, Ui::ConfigurationEditDialog& ui) {
 	info.profiler_enabled = ui.checkBox_profiler->isChecked();
 }
 
-void ConfigurationEditDialog::update_info() {
-	UpdateInfo(m_info, *m_ui);
-}
-
-void ConfigurationEditDialog::adjust_size() {
-	this->adjustSize();
-}
-
 void ConfigurationEditDialog::save() {
 	if (MandatoryLineEdit::FindEmpty(this)) {
 		QMessageBox::critical(this, tr("Save failed"), tr("Please fill all mandatory fields"));
@@ -427,7 +416,7 @@ void ConfigurationEditDialog::save() {
 		return;
 	}
 
-	update_info();
+	UpdateInfo(m_info, *m_ui);
 
 	emit accept();
 }

src/launcher/src/configurationItem.cpp

@@ -77,10 +77,6 @@ void SetStatus(QComboBox* combo, Configuration::GameStatus status) {
 	combo->setCurrentIndex(index >= 0 ? index : 0);
 }
 
-QIcon StandardIcon(QStyle::StandardPixmap icon) {
-	return QApplication::style()->standardIcon(icon);
-}
-
 QString GetPathText(const Configuration& info) {
 	return !info.game_path.isEmpty() ? info.game_path : info.basedir;
 }
@@ -278,11 +274,11 @@ void ConfigurationItem::UpdateIcon() {
 	}
 
 	if (m_running) {
-		setIcon(NameColumn, StandardIcon(QStyle::SP_MediaPlay));
+		setIcon(NameColumn, QApplication::style()->standardIcon(QStyle::SP_MediaPlay));
 	} else if (m_info->custom_settings) {
-		setIcon(NameColumn, StandardIcon(QStyle::SP_FileIcon));
+		setIcon(NameColumn, QApplication::style()->standardIcon(QStyle::SP_FileIcon));
 	} else {
-		setIcon(NameColumn, StandardIcon(QStyle::SP_ComputerIcon));
+		setIcon(NameColumn, QApplication::style()->standardIcon(QStyle::SP_ComputerIcon));
 	}
 }
 

src/launcher/src/configurationListWidget.cpp

@@ -1,6 +1,5 @@
 #include "configurationListWidget.h"
 
-#include "common.h"
 #include "compatibilityDatabase.h"
 #include "configuration.h"
 #include "configurationEditDialog.h"
@@ -232,7 +231,7 @@ ConfigurationListWidget::ConfigurationListWidget(QWidget* parent)
 	connect(m_ui->delete_button, &QToolButton::clicked, this,
 	        &ConfigurationListWidget::delete_configuartion);
 	connect(m_ui->cfgs_list, &QTreeWidget::currentItemChanged, this,
-	        &ConfigurationListWidget::list_currentItemChanged);
+	        &ConfigurationListWidget::SelectItem);
 	connect(m_ui->cfgs_list, &QTreeWidget::itemDoubleClicked, this,
 	        &ConfigurationListWidget::list_itemDoubleClicked);
 	connect(m_ui->cfgs_list, &QTreeWidget::customContextMenuRequested, this,
@@ -676,7 +675,7 @@ void ConfigurationListWidget::edit_configuration() {
 	}
 
 	ConfigurationEditDialog dlg(item->GetInfo(), this);
-	dlg.SetTitle(tr("Edit game settings"));
+	dlg.setWindowTitle(tr("Edit game settings"));
 
 	if (dlg.exec() == QDialog::Accepted) {
 		item->GetInfo().custom_settings = true;
@@ -708,7 +707,7 @@ void ConfigurationListWidget::edit_global_settings() {
 	info.name = tr("Global settings");
 
 	ConfigurationEditDialog dlg(info, this);
-	dlg.SetTitle(tr("Global settings"));
+	dlg.setWindowTitle(tr("Global settings"));
 	dlg.SetGameDirectories(m_game_dirs);
 
 	if (dlg.exec() == QDialog::Accepted) {
@@ -731,15 +730,6 @@ void ConfigurationListWidget::ClearCustomSettings(ConfigurationItem* item) {
 	delete m_custom_infos.take(item->GetInfo().game_path);
 }
 
-void ConfigurationListWidget::run_configuration() {
-	emit Run();
-}
-
-bool ConfigurationListWidget::CanViewSelectedTrophies() const {
-	return m_selected_item != nullptr &&
-	       TrophyViewerDialog::HasTrophyData(&m_selected_item->GetInfo());
-}
-
 void ConfigurationListWidget::ViewTrophies() {
 	auto* current = static_cast<ConfigurationItem*>(m_ui->cfgs_list->currentItem());
 	auto* item    = current != nullptr ? current : m_selected_item;
@@ -852,11 +842,6 @@ void ConfigurationListWidget::SelectItem(QTreeWidgetItem* witem) {
 	emit Select();
 }
 
-void ConfigurationListWidget::list_currentItemChanged(QTreeWidgetItem* current,
-                                                      QTreeWidgetItem* /*previous*/) {
-	SelectItem(current);
-}
-
 void ConfigurationListWidget::list_itemDoubleClicked(QTreeWidgetItem* witem, int /*column*/) {
 	SelectItem(witem);
 	if (m_run_enabled) {
@@ -877,7 +862,7 @@ void ConfigurationListWidget::show_context_menu(const QPoint& pos) {
 	const bool has_trophy_data =
 	    item != nullptr && TrophyViewerDialog::HasTrophyData(&item->GetInfo());
 
-	QAction* action_run = menu.addAction(tr("Run"), this, SLOT(run_configuration()));
+	QAction* action_run = menu.addAction(tr("Run"), this, SIGNAL(Run()));
 	QAction* action_open_folder =
 	    menu.addAction(style()->standardIcon(QStyle::SP_DirOpenIcon), tr("Open game folder"), this,
 	                   SLOT(open_game_folder()));

src/launcher/src/main.cpp

@@ -2,8 +2,6 @@
 #include "launcherTheme.h"
 
 #include <QApplication>
-#include <QArgument>
-#include <QObject>
 
 int main(int argc, char* argv[]) {
 	QApplication a(argc, argv);
@@ -11,8 +9,6 @@ int main(int argc, char* argv[]) {
 
 	MainDialog w;
 
-	QObject::connect(&a, &QApplication::aboutToQuit, &w, &MainDialog::Quit);
-
 	w.emit Start();
 
 	w.show();

src/launcher/src/mainDialog.cpp

@@ -1,6 +1,5 @@
 #include "mainDialog.h"
 
-#include "common.h"
 #include "configuration.h"
 #include "configurationItem.h"
 #include "configurationListWidget.h"
@@ -17,6 +16,7 @@
 #include <QIODevice>
 #include <QLabel>
 #include <QMessageBox>
+#include <QObject>
 #include <QPointer>
 #include <QProcess>
 #include <QRadioButton>
@@ -27,6 +27,8 @@
 #include <QVariant>
 #include <QtCore>
 
+#include <cstdint>
+
 #include "ui_main_dialog.h"
 
 #if defined(_WIN32)
@@ -58,20 +60,8 @@ constexpr char SETTINGS_MAIN_DIALOG[]        = "MainDialog";
 constexpr char SETTINGS_MAIN_LAST_GEOMETRY[] = "geometry";
 constexpr char SETTINGS_CHECK_UPDATES[]       = "check_updates_on_startup";
 
-class DetachableProcess: public QProcess {
-	Q_OBJECT;
-
-public:
-	explicit DetachableProcess(QObject* parent = nullptr): QProcess(parent) {}
-	void Detach() {
-		this->waitForStarted();
-		setProcessState(QProcess::NotRunning);
-	}
-};
-
 class MainDialogPrivate: public QObject {
 	Q_OBJECT
-	KYTY_QT_CLASS_NO_COPY(MainDialogPrivate);
 
 public:
 	explicit MainDialogPrivate(QObject* parent = nullptr): QObject(parent) {}
@@ -99,7 +89,7 @@ private:
 	UpdateChecker*  m_update_checker = nullptr;
 	QString         m_interpreter;
 
-	/*DetachableProcess*/ QProcess m_process;
+	QProcess m_process;
 
 	QPointer<ConfigurationItem> m_running_item;
 };
@@ -121,7 +111,6 @@ void MainDialogPrivate::Setup(MainDialog* main_dialog) {
 
 	m_main_dialog = main_dialog;
 	m_update_checker = new UpdateChecker(main_dialog);
-	m_ui->widget->SetMainDialog(main_dialog);
 	m_ui->check_updates_on_startup->setChecked(g_check_updates_on_startup);
 	m_ui->check_updates_link->setVisible(UpdateChecker::IsSupported());
 	m_ui->check_updates_on_startup->setVisible(UpdateChecker::IsSupported());
@@ -154,8 +143,6 @@ void MainDialogPrivate::Setup(MainDialog* main_dialog) {
 		        Update();
 	        });
 
-	// connect(main_dialog, &MainDialog::Quit, [=]() { m_process.Detach(); });
-
 	m_ui->label_settings_file->setText(tr("Settings file: ") + m_ui->widget->GetSettingsFile());
 
 	m_main_dialog->restoreGeometry(g_last_geometry);

src/launcher/src/patchesDialog.cpp

@@ -3,6 +3,7 @@
 #include "configuration.h"
 
 #include <QCoreApplication>
+#include <QDebug>
 #include <QDialogButtonBox>
 #include <QDir>
 #include <QFile>
@@ -129,8 +130,11 @@ void PatchesDialog::Save() {
 	document.setObject(root);
 
 	QSaveFile output(path);
-	if (output.open(QIODevice::WriteOnly) && output.write(document.toJson()) >= 0 &&
-	    output.commit()) {
-		m_status->setText(tr("Cheat selection saved."));
+	if (!output.open(QIODevice::WriteOnly) || output.write(document.toJson()) < 0 ||
+	    !output.commit()) {
+		qWarning() << "Could not save cheat file:" << output.errorString();
+		m_status->setText(tr("Could not save cheat file: %1").arg(output.errorString()));
+		return;
 	}
+	m_status->setText(tr("Cheat selection saved."));
 }

src/launcher/src/trophyViewerDialog.cpp

@@ -20,6 +20,7 @@
 #include <QLabel>
 #include <QMap>
 #include <QMessageBox>
+#include <QObject>
 #include <QPixmap>
 #include <QRegularExpression>
 #include <QSet>
@@ -32,8 +33,9 @@
 #include <QVBoxLayout>
 #include <QVariant>
 #include <QWidget>
+#include <QtEndian>
 
-#include <limits>
+#include <utility>
 
 namespace {
 
@@ -83,21 +85,6 @@ static bool CanRead(const QByteArray& data, qsizetype offset, qsizetype size) {
 	return offset >= 0 && size >= 0 && offset <= data.size() && size <= data.size() - offset;
 }
 
-static quint32 ReadBe32(const QByteArray& data, qsizetype offset) {
-	const auto* p = reinterpret_cast<const uchar*>(data.constData() + offset);
-	return (static_cast<quint32>(p[0]) << 24u) | (static_cast<quint32>(p[1]) << 16u) |
-	       (static_cast<quint32>(p[2]) << 8u) | static_cast<quint32>(p[3]);
-}
-
-static quint64 ReadBe64(const QByteArray& data, qsizetype offset) {
-	return (static_cast<quint64>(ReadBe32(data, offset)) << 32u) |
-	       static_cast<quint64>(ReadBe32(data, offset + 4));
-}
-
-static bool FitsQSizeType(quint64 value) {
-	return value <= static_cast<quint64>(std::numeric_limits<qsizetype>::max());
-}
-
 static QString ReadFixedString(const QByteArray& data, qsizetype offset, qsizetype max_size) {
 	if (!CanRead(data, offset, max_size)) {
 		return {};
@@ -125,33 +112,33 @@ static bool ReadUcp(const QString& file_name, QMap<QString, QByteArray>& files,
 		return false;
 	}
 
-	const auto magic = ReadBe32(data, 0x00);
+	const auto magic = qFromBigEndian<quint32>(data.constData() + 0x00);
 	if (magic != UCP_MAGIC) {
 		error = QObject::tr("%1 has an invalid trophy package magic.")
 		            .arg(QFileInfo(file_name).fileName());
 		return false;
 	}
 
-	const auto version = ReadBe32(data, 0x04);
+	const auto version = qFromBigEndian<quint32>(data.constData() + 0x04);
 	if (version != UCP_VERSION) {
 		error = QObject::tr("%1 uses unsupported trophy package version %2.")
 		            .arg(QFileInfo(file_name).fileName(), QString::number(version));
 		return false;
 	}
 
-	const auto declared_size = ReadBe64(data, 0x08);
+	const auto declared_size = qFromBigEndian<quint64>(data.constData() + 0x08);
 	if (declared_size > static_cast<quint64>(data.size())) {
 		error = QObject::tr("%1 is truncated.").arg(QFileInfo(file_name).fileName());
 		return false;
 	}
 
-	const auto file_count = ReadBe32(data, 0x10);
-	const auto toc_offset = static_cast<quint64>(ReadBe32(data, 0x14));
+	const auto file_count = qFromBigEndian<quint32>(data.constData() + 0x10);
+	const auto toc_offset = static_cast<quint64>(qFromBigEndian<quint32>(data.constData() + 0x14));
 	const auto data_size  = static_cast<quint64>(data.size());
 
 	const quint64 table_size = UCP_TOC_SKIP + static_cast<quint64>(file_count) * UCP_ENTRY_LEN;
 	if (toc_offset > data_size || table_size > data_size - toc_offset ||
-	    !FitsQSizeType(toc_offset + table_size)) {
+	    !std::in_range<qsizetype>(toc_offset + table_size)) {
 		error = QObject::tr("%1 has an invalid table of contents.")
 		            .arg(QFileInfo(file_name).fileName());
 		return false;
@@ -162,14 +149,14 @@ static bool ReadUcp(const QString& file_name, QMap<QString, QByteArray>& files,
 		                                                 static_cast<quint64>(i) * UCP_ENTRY_LEN);
 		UcpEntry   entry;
 		entry.name   = ReadFixedString(data, entry_offset, UCP_NAME_LEN).trimmed();
-		entry.offset = ReadBe64(data, entry_offset + 0x20);
-		entry.size   = ReadBe64(data, entry_offset + 0x28);
+		entry.offset = qFromBigEndian<quint64>(data.constData() + entry_offset + 0x20);
+		entry.size   = qFromBigEndian<quint64>(data.constData() + entry_offset + 0x28);
 
 		if (entry.name.isEmpty()) {
 			continue;
 		}
 		if (entry.offset > data_size || entry.size > data_size - entry.offset ||
-		    !FitsQSizeType(entry.offset) || !FitsQSizeType(entry.size)) {
+		    !std::in_range<qsizetype>(entry.offset) || !std::in_range<qsizetype>(entry.size)) {
 			error = QObject::tr("%1 has an invalid entry for %2.")
 			            .arg(QFileInfo(file_name).fileName(), entry.name);
 			return false;

src/libs/ajm.cpp

@@ -849,8 +849,9 @@ int KYTY_SYSV_ABI AjmMemoryUnregister(uint32_t context, void* ptr) {
 int KYTY_SYSV_ABI AjmBatchInitialize(void* buffer, size_t size, AjmBatchInfo* info) {
 	PRINT_NAME();
 
-	EXIT_NOT_IMPLEMENTED(buffer == nullptr);
-	EXIT_NOT_IMPLEMENTED(info == nullptr);
+	if (buffer == nullptr || info == nullptr) {
+		return AJM_ERROR_INVALID_PARAMETER;
+	}
 
 	info->buffer           = buffer;
 	info->offset           = 0;

src/libs/ajm/aac_decoder.h

@@ -38,9 +38,9 @@ static constexpr uint32_t AJM_DEC_M4AAC_MAX_SAMPLING_FREQ_INDEX = 11;
 
 class AjmAacDecoder final: public AjmDecoder {
 public:
-	AjmAacDecoder(uint32_t channels, uint32_t sample_rate, AjmSampleEncoding encoding,
+	AjmAacDecoder(uint32_t max_channels, uint32_t sample_rate, AjmSampleEncoding encoding,
 	              uint64_t flags)
-	    : AjmDecoder(channels, sample_rate, encoding), m_flags(flags) {
+	    : AjmDecoder(0, sample_rate, encoding), m_max_channels(max_channels), m_flags(flags) {
 		m_codec = avcodec_find_decoder(AV_CODEC_ID_AAC);
 		if (m_codec == nullptr) {
 			LOGF("AJM AAC: avcodec_find_decoder failed\n");
@@ -94,6 +94,11 @@ public:
 		    ((m_flags & AJM_INSTANCE_FLAG_DEC_M4AAC_ENABLE_NONDELAY_OUTPUT) != 0 ? 0u : 2u);
 		m_total_decoded_samples = 0;
 		m_bitrate               = 0;
+		m_channels              = 0;
+		m_raw_configured        = false;
+		if (m_config_number == AJM_DEC_M4AAC_CONFIG_NUMBER_RAW) {
+			m_sample_rate = SamplingRateFromIndex(m_sampling_freq_index);
+		}
 		if (m_is_initialized) {
 			OpenDecoder();
 		}
@@ -189,26 +194,6 @@ private:
 			return;
 		}
 
-		if (m_config_number == AJM_DEC_M4AAC_CONFIG_NUMBER_RAW) {
-			const auto sample_index =
-			    std::min(m_sampling_freq_index, AJM_DEC_M4AAC_MAX_SAMPLING_FREQ_INDEX);
-			const auto channels = std::clamp<uint32_t>(m_channels, 1u, AJM_DEC_M4AAC_MAX_CHANNELS);
-			constexpr uint32_t audio_object_type = 2;
-
-			auto* extradata = static_cast<uint8_t*>(av_mallocz(2 + AV_INPUT_BUFFER_PADDING_SIZE));
-			if (extradata == nullptr) {
-				Release();
-				return;
-			}
-
-			extradata[0] = static_cast<uint8_t>((audio_object_type << 3u) | (sample_index >> 1u));
-			extradata[1] = static_cast<uint8_t>(((sample_index & 0x1u) << 7u) |
-			                                    (std::min(channels, 7u) << 3u));
-			m_codec_context->extradata      = extradata;
-			m_codec_context->extradata_size = 2;
-			SetFormat(channels, SamplingRateFromIndex(sample_index), m_sample_encoding);
-		}
-
 		if (avcodec_open2(m_codec_context, m_codec, nullptr) < 0) {
 			LOGF("AJM AAC: decoder initialization failed\n");
 			Release();
@@ -222,6 +207,10 @@ private:
 		}
 
 		const auto channels = static_cast<uint32_t>(std::max(frame->ch_layout.nb_channels, 1));
+		if (channels > m_max_channels) {
+			result->result = AJM_RESULT_TOO_MANY_CHANNELS;
+			return false;
+		}
 		const auto bpf      = channels * AjmBytesPerSample(m_sample_encoding);
 		if (bpf == 0) {
 			result->result = AJM_RESULT_INVALID_PARAMETER;
@@ -306,6 +295,25 @@ private:
 			return false;
 		}
 		std::memcpy(packet->data, packet_data, static_cast<size_t>(packet_size));
+		if (m_config_number == AJM_DEC_M4AAC_CONFIG_NUMBER_RAW && !m_raw_configured) {
+			// Only indexed stereo starts with a channel pair; the instance limit is not
+			// the stream's channel count. A leading SCE remains ambiguous.
+			const auto channels = (packet_data[0] >> 5u) == 1u
+			                          ? 2u
+			                          : std::clamp(m_max_channels, 1u, AJM_DEC_M4AAC_MAX_CHANNELS);
+			auto* extradata = av_packet_new_side_data(packet, AV_PKT_DATA_NEW_EXTRADATA, 2);
+			if (extradata == nullptr) {
+				av_packet_free(&packet);
+				result->result = AJM_RESULT_CODEC_ERROR | AJM_RESULT_FATAL;
+				return false;
+			}
+			constexpr uint32_t audio_object_type = 2;
+			extradata[0] = static_cast<uint8_t>((audio_object_type << 3u) |
+			                                    (m_sampling_freq_index >> 1u));
+			extradata[1] = static_cast<uint8_t>(((m_sampling_freq_index & 1u) << 7u) |
+			                                    (std::min(channels, 7u) << 3u));
+			m_raw_configured = true;
+		}
 
 		int rc = avcodec_send_packet(m_codec_context, packet);
 		av_packet_free(&packet);
@@ -353,12 +361,14 @@ private:
 
 	const AVCodec*  m_codec               = nullptr;
 	AVCodecContext* m_codec_context       = nullptr;
+	const uint32_t  m_max_channels;
 	uint64_t        m_flags               = 0;
 	uint32_t        m_config_number       = AJM_DEC_M4AAC_CONFIG_NUMBER_ADTS;
 	uint32_t        m_sampling_freq_index = 3;
 	uint32_t        m_skip_frames         = 2;
 	uint32_t        m_bitrate             = 0;
 	bool            m_is_initialized      = false;
+	bool            m_raw_configured      = false;
 };
 
 } // namespace Libs::Audio::Ajm

src/libs/ajm/decoder.h

@@ -72,6 +72,7 @@ constexpr int32_t AJM_RESULT_INVALID_DATA       = 0x00000002;
 constexpr int32_t AJM_RESULT_INVALID_PARAMETER  = 0x00000004;
 constexpr int32_t AJM_RESULT_PARTIAL_INPUT      = 0x00000008;
 constexpr int32_t AJM_RESULT_NOT_ENOUGH_ROOM    = 0x00000010;
+constexpr int32_t AJM_RESULT_TOO_MANY_CHANNELS  = 0x00000040;
 constexpr int32_t AJM_RESULT_CODEC_ERROR        = 0x40000000;
 constexpr int32_t AJM_RESULT_FATAL              = static_cast<int32_t>(0x80000000u);
 

src/libs/audio.cpp

@@ -5,11 +5,11 @@
 #include "common/common.h"
 #include "common/emulatorConfig.h"
 #include "common/logging/log.h"
-#include "common/magicEnum.h"
 #include "common/stringUtils.h"
 #include "common/threads.h"
 #include "kernel/pthread.h"
 #include "kernel/semaphore.h"
+#include "libs/ajm/atrac9_decoder.h"
 #include "libs/audio_internal.h"
 #include "libs/errno.h"
 #include "libs/libs.h"
@@ -19,7 +19,10 @@
 #include <atomic>
 #include <cstdlib>
 #include <cstring>
+#include <deque>
 #include <limits>
+#include <magic_enum.hpp>
+#include <memory>
 #include <vector>
 
 #include "libatrac9.h"
@@ -800,8 +803,7 @@ int KYTY_SYSV_ABI AudioOutOpen(int user_id, int type, int index, uint32_t len, u
 		default:;
 	}
 
-	LOGF("\t param   = %u (format=%u, %s)\n", param, format_param,
-	     Common::EnumName(format).c_str());
+	LOGF("\t param   = %u (format=%u, %s)\n", param, format_param, magic_enum::enum_name(format));
 
 	EXIT_NOT_IMPLEMENTED(format == Audio::Format::Unknown);
 
@@ -966,7 +968,7 @@ static int OpenPort(int user_id, int type, int index, uint32_t len, uint32_t fre
 		default: return AUDIO_IN_ERROR_INVALID_PARAM;
 	}
 
-	LOGF("\t param   = %u (%s)\n", param, Common::EnumName(format).c_str());
+	LOGF("\t param   = %u (%s)\n", param, magic_enum::enum_name(format));
 
 	EXIT_IF(g_audio == nullptr);
 
@@ -1825,26 +1827,146 @@ enum class Ngs2RackType {
 	CustomSampler,
 };
 
+struct Ngs2UserFxContext {
+	void*     common;
+	void*     param;
+	void*     work;
+	uintptr_t user_data;
+	uint32_t  max_voices;
+	uint32_t  voice_index;
+	uint64_t  reserved[4] {};
+};
+
+struct Ngs2UserFxProcessContext {
+	float**     channels;
+	void*       common;
+	const void* param;
+	void*       work;
+	void*       state;
+	uintptr_t   user_data;
+	uint32_t    flags;
+	uint32_t    input_channels;
+	uint32_t    output_channels;
+	uint32_t    grain_samples;
+	uint32_t    sample_rate;
+	uint32_t    reserved {};
+	uint64_t    reserved2[4] {};
+};
+
+struct Ngs2UserFxOption {
+	Ngs2CustomModuleOption header;
+	int                    KYTY_SYSV_ABI (*setup)(Ngs2UserFxContext*);
+	int                    KYTY_SYSV_ABI (*cleanup)(Ngs2UserFxContext*);
+	uintptr_t              control;
+	int                    KYTY_SYSV_ABI (*process)(Ngs2UserFxProcessContext*);
+	size_t                 common_size;
+	size_t                 param_size;
+	size_t                 work_size;
+	uintptr_t              user_data;
+};
+
 struct Ngs2RackInternal {
-	Ngs2Internal*         ngs  = nullptr;
-	Ngs2RackInternal*     next = nullptr;
-	Ngs2RackType          type = Ngs2RackType::Sampler;
-	Ngs2RackOptionUnion   option;
-	Ngs2ContextBufferInfo buffer_info;
-	Ngs2BufferAllocator   allocator;
+	Ngs2Internal*                        ngs  = nullptr;
+	Ngs2RackInternal*                    next = nullptr;
+	Ngs2RackType                         type = Ngs2RackType::Sampler;
+	Ngs2RackOptionUnion                  option;
+	Ngs2ContextBufferInfo                buffer_info;
+	Ngs2BufferAllocator                  allocator;
+	std::array<std::vector<uint8_t>, 24> common;
+	std::array<Ngs2UserFxOption, 24>     fx {};
 };
 
 enum class Ngs2VoicePlayState { Empty, Playing, Paused, Stopped };
-
 enum class Ngs2VoicePlayEvent { None, Play, Pause, Resume, Stop, StopImm, Kill };
 
 struct Ngs2VoiceInternal {
-	Ngs2VoicePlayEvent event          = Ngs2VoicePlayEvent::None;
-	Ngs2VoicePlayState state          = Ngs2VoicePlayState::Empty;
-	Ngs2RackInternal*  rack           = nullptr;
-	uintptr_t          callback       = 0;
-	uintptr_t          callback_data  = 0;
-	uint32_t           callback_flags = 0;
+	struct Port {
+		Ngs2VoiceInternal* dest   = nullptr;
+		uint32_t           input  = 0;
+		float              volume = 1.0f;
+		int32_t            matrix = -1;
+	};
+	struct Block {
+		const uint8_t*    data;
+		Ngs2WaveformBlock info;
+		uint32_t          cursor         = 0;
+		size_t            data_cursor    = 0;
+		uint32_t          skip_remaining = info.num_skip_samples;
+	};
+	struct Module {
+		std::vector<uint8_t> param, work, state;
+		uint32_t             flags = 1;
+	};
+	Ngs2VoicePlayEvent              event          = Ngs2VoicePlayEvent::None;
+	Ngs2VoicePlayState              state          = Ngs2VoicePlayState::Empty;
+	Ngs2RackInternal*               rack           = nullptr;
+	uintptr_t                       callback       = 0;
+	uintptr_t                       callback_data  = 0;
+	uint32_t                        callback_flags = 0;
+	std::deque<Block>               blocks;
+	std::vector<Port>               ports;
+	std::vector<std::vector<float>> matrices;
+	std::vector<Module>             modules;
+	std::vector<float>              samples;
+	std::vector<float>              output_matrix;
+	uint32_t                        channels        = 0;
+	uint32_t                        output_id       = 0;
+	bool                            rendering       = false;
+	bool                            rendered        = false;
+	bool                            has_samples     = false;
+
+	std::unique_ptr<Ajm::AjmAt9Decoder> decoder;
+	std::vector<float>                 decoded_frame;
+	bool                               accepts_blocks = true;
+
+	void SetupSampler(const Ngs2WaveformFormat& format) {
+		EXIT_NOT_IMPLEMENTED(format.sample_rate != rack->ngs->option.sample_rate ||
+		                     format.frame_margin != 0 || format.frame_offset != 0);
+		channels = format.num_channels;
+		blocks.clear();
+		decoder.reset();
+		decoded_frame.clear();
+		accepts_blocks = true;
+		if (format.waveform_type == NGS2_WAVEFORM_TYPE_ATRAC9) {
+			decoder = std::make_unique<Ajm::AjmAt9Decoder>(channels, format.sample_rate,
+			                                               Ajm::AjmSampleEncoding::Float, 0);
+			const auto result =
+			    decoder->Initialize(&format.config_data, sizeof(format.config_data));
+			EXIT_NOT_IMPLEMENTED(result.result != OK);
+			EXIT_NOT_IMPLEMENTED(result.format.channel_num != channels ||
+			                     result.format.sampling_frequency != format.sample_rate);
+			Ajm::AjmSidebandDecAt9CodecInfo info {};
+			decoder->WriteCodecInfo(&info, sizeof(info), result);
+			EXIT_NOT_IMPLEMENTED(info.frame_samples > std::numeric_limits<uint16_t>::max());
+			decoded_frame.resize(info.frame_samples * channels);
+		} else {
+			EXIT_NOT_IMPLEMENTED(format.waveform_type != 0x12);
+		}
+	}
+
+	void SetMatrix(uint32_t index, const float* levels, uint32_t count) {
+		EXIT_NOT_IMPLEMENTED(index >= matrices.size());
+		matrices[index].assign(levels, levels + count);
+	}
+	void SetVolume(uint32_t index, float volume) {
+		EXIT_NOT_IMPLEMENTED(index >= ports.size());
+		ports[index].volume = volume;
+	}
+	void SetPortMatrix(uint32_t index, int32_t matrix) {
+		EXIT_NOT_IMPLEMENTED(index >= ports.size());
+		ports[index].matrix = matrix;
+	}
+	void SetEvent(uint32_t id) {
+		switch (id) {
+			case 1: event = Ngs2VoicePlayEvent::Play; break;
+			case 2: event = Ngs2VoicePlayEvent::Stop; break;
+			case 4: event = Ngs2VoicePlayEvent::StopImm; break;
+			case 8: event = Ngs2VoicePlayEvent::Kill; break;
+			case 16: event = Ngs2VoicePlayEvent::Pause; break;
+			case 32: event = Ngs2VoicePlayEvent::Resume; break;
+			default: EXIT("unknown event_id: 0x%08" PRIx32 "\n", id);
+		}
+	}
 };
 
 struct Ngs2VoiceParamHeader {
@@ -1932,6 +2054,9 @@ static Ngs2RackInternal*    g_racks_list   = nullptr;
 static std::atomic_uint32_t g_next_ngs_uid = 1;
 static Common::Mutex        g_racks_mutex;
 
+static_assert(sizeof(Ngs2UserFxContext) == 72);
+static_assert(sizeof(Ngs2UserFxProcessContext) == 104);
+static_assert(sizeof(Ngs2UserFxOption) == 72);
 static_assert(sizeof(Ngs2SystemOption) == 144);
 static_assert(sizeof(Ngs2SystemInfo) == 184);
 static_assert(sizeof(Ngs2RackOption) == 176);
@@ -2273,6 +2398,35 @@ int KYTY_SYSV_ABI Ngs2SystemDestroy(uintptr_t system_handle, Ngs2ContextBufferIn
 	return OK;
 }
 
+static void Ngs2InitModules(Ngs2RackInternal& rack) {
+	if (!Ngs2RackIsCustom(rack.type)) {
+		return;
+	}
+	const auto& custom = rack.option.custom_sampler.custom_rack_option;
+	EXIT_NOT_IMPLEMENTED(custom.num_buffers != 1);
+	auto* voices = reinterpret_cast<Ngs2VoiceInternal*>(&rack + 1);
+	for (uint32_t m = 0; m < custom.num_modules; ++m) {
+		if (custom.module[m].module_id != 0x1f) {
+			continue;
+		}
+		auto& fx = rack.fx[m];
+		fx       = *reinterpret_cast<const Ngs2UserFxOption*>(custom.module[m].option);
+		rack.common[m].resize(fx.common_size);
+		for (uint32_t i = 0; i < rack.option.common.max_voices; ++i) {
+			auto& module = voices[i].modules[m];
+			module.param.resize(fx.param_size);
+			module.work.resize(fx.work_size);
+			module.state.resize(custom.module[m].state_size);
+			if (fx.setup != nullptr) {
+				Ngs2UserFxContext context {
+				    rack.common[m].data(), module.param.data(),           module.work.data(),
+				    fx.user_data,          rack.option.common.max_voices, i};
+				EXIT_NOT_IMPLEMENTED(fx.setup(&context) != OK);
+			}
+		}
+	}
+}
+
 int KYTY_SYSV_ABI Ngs2RackCreate(uintptr_t system_handle, uint32_t rack_id,
                                  const Ngs2RackOption*        option,
                                  const Ngs2ContextBufferInfo* buffer_info, uintptr_t* handle) {
@@ -2314,7 +2468,7 @@ int KYTY_SYSV_ABI Ngs2RackCreate(uintptr_t system_handle, uint32_t rack_id,
 	     static_cast<uint64_t>(buffer_info->host_buffer_size));
 
 	auto* ngs    = reinterpret_cast<Ngs2Internal*>(system_handle);
-	auto* rack   = static_cast<Ngs2RackInternal*>(buffer_info->host_buffer);
+	auto* rack   = new (buffer_info->host_buffer) Ngs2RackInternal {};
 	auto* voices = reinterpret_cast<Ngs2VoiceInternal*>(rack + 1);
 
 	Common::LockGuard lock(ngs->mutex);
@@ -2361,7 +2515,7 @@ int KYTY_SYSV_ABI Ngs2RackCreate(uintptr_t system_handle, uint32_t rack_id,
 		default: EXIT("unknown rack_id: 0x%" PRIx32 "\n", rack_id);
 	}
 
-	LOGF("\t type                   = %s\n", Common::EnumName(rack->type).c_str());
+	LOGF("\t type                   = %s\n", magic_enum::enum_name(rack->type));
 
 	rack->allocator   = Ngs2BufferAllocator();
 	rack->buffer_info = *buffer_info;
@@ -2374,10 +2528,15 @@ int KYTY_SYSV_ABI Ngs2RackCreate(uintptr_t system_handle, uint32_t rack_id,
 	}
 
 	for (uint32_t i = 0; i < option->max_voices; i++) {
-		voices[i].rack  = rack;
-		voices[i].event = Ngs2VoicePlayEvent::None;
-		voices[i].state = Ngs2VoicePlayState::Empty;
+		auto& voice = *std::construct_at(voices + i);
+		voice.rack  = rack;
+		voice.ports.resize(option->max_ports);
+		voice.matrices.resize(option->max_matrices);
+		if (Ngs2RackIsCustom(rack->type)) {
+			voice.modules.resize(rack->option.custom_sampler.custom_rack_option.num_modules);
+		}
 	}
+	Ngs2InitModules(*rack);
 
 	*handle = reinterpret_cast<uintptr_t>(rack);
 
@@ -2407,22 +2566,9 @@ int KYTY_SYSV_ABI Ngs2RackCreateWithAllocator(uintptr_t system_handle, uint32_t
 
 	EXIT_NOT_IMPLEMENTED(option->size < sizeof(Ngs2RackOption));
 
-	LOGF("\t rack_id                = 0x%" PRIx32 "\n"
-	     "\t name                   = %.64s\n"
-	     "\t flags                  = %u\n"
-	     "\t max_grain_samples      = %u\n"
-	     "\t max_voices             = %u\n"
-	     "\t max_input_delay_blocks = %u\n"
-	     "\t max_matrices           = %u\n"
-	     "\t max_ports              = %u\n"
-	     "\t max_voice_channels     = %u\n"
-	     "\t max_output_channels    = %u\n"
-	     "\t alloc_handler          = 0x%016" PRIx64 "\n"
+	LOGF("\t alloc_handler          = 0x%016" PRIx64 "\n"
 	     "\t free_handler           = 0x%016" PRIx64 "\n"
 	     "\t user_data              = 0x%016" PRIx64 "\n",
-	     rack_id, option->name, option->flags, option->max_grain_samples, option->max_voices,
-	     option->max_input_delay_blocks, option->max_matrices, option->max_ports,
-	     option->max_voice_channels, option->max_output_channels,
 	     reinterpret_cast<uint64_t>(allocator->alloc_handler),
 	     reinterpret_cast<uint64_t>(allocator->free_handler),
 	     static_cast<uint64_t>(allocator->user_data));
@@ -2478,26 +2624,41 @@ int KYTY_SYSV_ABI Ngs2RackDestroy(uintptr_t rack_handle, Ngs2ContextBufferInfo*
 	if (ngs == nullptr) {
 		return ERROR_INVALID_RACK_HANDLE;
 	}
-
 	Ngs2ContextBufferInfo context_buffer;
 	Ngs2BufferAllocator   allocator;
 	{
 		Common::LockGuard lock(ngs->mutex);
-		Common::LockGuard racks_lock(g_racks_mutex);
-
-		auto** link = &g_racks_list;
-		while (*link != nullptr && *link != rack) {
-			link = &(*link)->next;
+		{
+			Common::LockGuard racks_lock(g_racks_mutex);
+			auto**            link = &g_racks_list;
+			while (*link != nullptr && *link != rack) {
+				link = &(*link)->next;
+			}
+			if (*link == nullptr) {
+				return ERROR_INVALID_RACK_HANDLE;
+			}
+			*link = rack->next;
 		}
-		if (*link == nullptr) {
-			return ERROR_INVALID_RACK_HANDLE;
+		auto* voices = reinterpret_cast<Ngs2VoiceInternal*>(rack + 1);
+		for (uint32_t i = 0; i < rack->option.common.max_voices; ++i) {
+			for (size_t m = 0; m < voices[i].modules.size(); ++m) {
+				const auto& fx = rack->fx[m];
+				if (fx.cleanup != nullptr) {
+					auto&             module = voices[i].modules[m];
+					Ngs2UserFxContext context {rack->common[m].data(),
+					                           module.param.data(),
+					                           module.work.data(),
+					                           fx.user_data,
+					                           rack->option.common.max_voices,
+					                           i};
+					EXIT_NOT_IMPLEMENTED(fx.cleanup(&context) != OK);
+				}
+			}
+			std::destroy_at(voices + i);
 		}
-
-		*link          = rack->next;
 		context_buffer = rack->buffer_info;
 		allocator      = rack->allocator;
-		rack->ngs      = nullptr;
-		rack->next     = nullptr;
+		std::destroy_at(rack);
 	}
 
 	if (allocator.free_handler != nullptr) {
@@ -2540,75 +2701,231 @@ int KYTY_SYSV_ABI Ngs2RackUnlock(uintptr_t rack_handle) {
 	return OK;
 }
 
+static void Ngs2ApplyEvent(Ngs2VoiceInternal& voice) {
+	switch (voice.event) {
+		case Ngs2VoicePlayEvent::None: break;
+		case Ngs2VoicePlayEvent::Play:
+			if (voice.state == Ngs2VoicePlayState::Empty) {
+				voice.state = Ngs2VoicePlayState::Playing;
+			}
+			break;
+		case Ngs2VoicePlayEvent::Pause:
+			if (voice.state == Ngs2VoicePlayState::Playing) {
+				voice.state = Ngs2VoicePlayState::Paused;
+			}
+			break;
+		case Ngs2VoicePlayEvent::Resume:
+			if (voice.state == Ngs2VoicePlayState::Paused) {
+				voice.state = Ngs2VoicePlayState::Playing;
+			}
+			break;
+		case Ngs2VoicePlayEvent::Stop:
+			if (voice.state == Ngs2VoicePlayState::Playing) {
+				voice.state = Ngs2VoicePlayState::Stopped;
+			}
+			break;
+		case Ngs2VoicePlayEvent::StopImm:
+		case Ngs2VoicePlayEvent::Kill: voice.state = Ngs2VoicePlayState::Empty; break;
+	}
+	voice.event = Ngs2VoicePlayEvent::None;
+}
+
+static void Ngs2ConsumeSamples(Ngs2VoiceInternal& voice, uint32_t grain) {
+	uint32_t output = 0;
+	while (output < grain && !voice.blocks.empty()) {
+		auto&          block   = voice.blocks.front();
+		auto           count   = std::min(grain - output, block.info.num_samples - block.cursor);
+		const int16_t* pcm     = nullptr;
+		const float*   decoded = nullptr;
+		if (voice.decoder != nullptr) {
+			const auto skip =
+			    std::min(block.skip_remaining,
+			             static_cast<uint32_t>(voice.decoded_frame.size() / voice.channels));
+			Ajm::AjmGaplessState gapless;
+			gapless.Set({block.info.num_samples - block.cursor, static_cast<uint16_t>(skip), 0},
+			            true);
+			const auto result = voice.decoder->Decode(
+			    block.data + block.data_cursor, block.info.data_size - block.data_cursor,
+			    voice.decoded_frame.data(), voice.decoded_frame.size() * sizeof(float), false,
+			    &gapless);
+			EXIT_NOT_IMPLEMENTED(result.result != OK || result.input_consumed == 0);
+			block.data_cursor += result.input_consumed;
+			block.skip_remaining -= skip - gapless.current.skip_samples;
+			const auto decoded_samples =
+			    static_cast<uint32_t>(result.output_written / (sizeof(float) * voice.channels));
+			if (decoded_samples == 0) {
+				continue;
+			}
+			EXIT_NOT_IMPLEMENTED(decoded_samples > count);
+			count   = decoded_samples;
+			decoded = voice.decoded_frame.data();
+		} else {
+			pcm = reinterpret_cast<const int16_t*>(block.data) +
+			      (block.info.num_skip_samples + block.cursor) * voice.channels;
+		}
+		for (uint32_t c = 0; c < voice.channels; ++c) {
+			for (uint32_t i = 0; i < count; ++i) {
+				const auto index = i * voice.channels + c;
+				voice.samples[c * grain + output + i] =
+				    decoded != nullptr ? decoded[index] : pcm[index] / 32768.0f;
+			}
+		}
+		output += count;
+		block.cursor += count;
+		if (block.cursor == block.info.num_samples) {
+			struct CallbackInfo {
+				uintptr_t   data, voice;
+				uint32_t    flag, reserved;
+				uintptr_t   user;
+				const void* block_data;
+				size_t      size;
+				uint32_t    repeats, attributes;
+			} info {voice.callback_data,
+			        reinterpret_cast<uintptr_t>(&voice),
+			        1,
+			        0,
+			        block.info.user_data,
+			        block.data,
+			        block.info.data_size,
+			        0,
+			        0};
+			static_assert(sizeof(CallbackInfo) == 56);
+			voice.blocks.pop_front();
+			if (voice.blocks.empty() && !voice.accepts_blocks) {
+				voice.state = Ngs2VoicePlayState::Empty;
+			}
+			if (voice.callback != 0 && (voice.callback_flags & 1u) != 0) {
+				reinterpret_cast<void KYTY_SYSV_ABI (*)(const CallbackInfo*)>(voice.callback)(
+				    &info);
+			}
+		}
+	}
+	voice.has_samples = output != 0;
+}
+
+static void Ngs2RenderVoice(Ngs2VoiceInternal& voice, const std::vector<Ngs2VoiceInternal*>& voices,
+                            uint32_t grain) {
+	if (voice.rendered) {
+		return;
+	}
+	EXIT_NOT_IMPLEMENTED(voice.rendering);
+	voice.rendering = true;
+	voice.samples.assign(grain * voice.channels, 0.0f);
+	voice.has_samples = false;
+	if (voice.state == Ngs2VoicePlayState::Playing) {
+		if (voice.rack->type == Ngs2RackType::CustomSampler) {
+			Ngs2ConsumeSamples(voice, grain);
+		}
+		for (auto* source: voices) {
+			for (const auto& port: source->ports) {
+				if (port.dest != &voice || port.volume == 0.0f) {
+					continue;
+				}
+				Ngs2RenderVoice(*source, voices, grain);
+				if (!source->has_samples) {
+					continue;
+				}
+				EXIT_NOT_IMPLEMENTED(port.input != 0);
+				const auto* matrix = port.matrix < 0 ? nullptr : &source->matrices.at(port.matrix);
+				for (uint32_t dst = 0; dst < voice.channels; ++dst) {
+					for (uint32_t src = 0; src < source->channels; ++src) {
+						const float level =
+						    port.volume * (matrix == nullptr
+						                       ? (src == dst ? 1.0f : 0.0f)
+						                       : matrix->at(dst * source->channels + src));
+						for (uint32_t i = 0; i < grain; ++i) {
+							voice.samples[dst * grain + i] +=
+							    source->samples[src * grain + i] * level;
+						}
+					}
+				}
+				voice.has_samples = true;
+			}
+		}
+		if (voice.has_samples) {
+			const auto& custom = voice.rack->option.custom_sampler.custom_rack_option;
+			for (size_t m = 0; m < voice.modules.size(); ++m) {
+				EXIT_NOT_IMPLEMENTED(custom.module[m].module_id != 0x1f);
+				EXIT_NOT_IMPLEMENTED(custom.module[m].source_buffer_id != 0 ||
+				                     custom.module[m].dest_buffer_id != 0);
+				auto&               module = voice.modules[m];
+				const auto&         fx     = voice.rack->fx[m];
+				std::vector<float*> channels(voice.channels);
+				for (uint32_t c = 0; c < voice.channels; ++c) {
+					channels[c] = voice.samples.data() + c * grain;
+				}
+				Ngs2UserFxProcessContext context {channels.data(),
+				                                  voice.rack->common[m].data(),
+				                                  module.param.data(),
+				                                  module.work.data(),
+				                                  module.state.data(),
+				                                  fx.user_data,
+				                                  module.flags,
+				                                  voice.channels,
+				                                  voice.channels,
+				                                  grain,
+				                                  voice.rack->ngs->option.sample_rate};
+				EXIT_NOT_IMPLEMENTED(fx.process(&context) != OK);
+				module.flags = 0;
+			}
+		}
+	}
+	voice.rendering = false;
+	voice.rendered  = true;
+}
+
 int KYTY_SYSV_ABI Ngs2SystemRender(uintptr_t system_handle, const Ngs2RenderBufferInfo* buffer_info,
                                    uint32_t num_buffer_info) {
-	static std::atomic_uint32_t render_log_count = 0;
-	const auto log_index     = render_log_count.fetch_add(1, std::memory_order_relaxed);
-	const bool log_this_call = (log_index < 16 || (log_index % 600) == 0);
-
-	if (log_this_call) {
-		PRINT_NAME();
-		LOGF("\t call_count      = %" PRIu32 "\n", log_index + 1);
-	}
-
-	EXIT_NOT_IMPLEMENTED(buffer_info == nullptr);
-	EXIT_NOT_IMPLEMENTED(system_handle == 0);
-	EXIT_NOT_IMPLEMENTED(num_buffer_info == 0);
-
+	EXIT_NOT_IMPLEMENTED(buffer_info == nullptr || system_handle == 0 || num_buffer_info == 0);
 	auto* ngs = reinterpret_cast<Ngs2Internal*>(system_handle);
-
 	Common::LockGuard lock(ngs->mutex);
-
-	for (uint32_t i = 0; i < num_buffer_info; i++) {
+	std::vector<Ngs2VoiceInternal*> voices;
+	{
+		Common::LockGuard racks_lock(g_racks_mutex);
+		for (auto* rack = g_racks_list; rack != nullptr; rack = rack->next) {
+			if (rack->ngs != ngs) {
+				continue;
+			}
+			auto* items = reinterpret_cast<Ngs2VoiceInternal*>(rack + 1);
+			for (uint32_t i = 0; i < rack->option.common.max_voices; ++i) {
+				Ngs2ApplyEvent(items[i]);
+				items[i].rendered = false;
+				if (items[i].channels != 0) {
+					voices.push_back(items + i);
+				}
+			}
+		}
+	}
+	for (uint32_t i = 0; i < num_buffer_info; ++i) {
 		if (buffer_info[i].buffer != nullptr && buffer_info[i].buffer_size != 0) {
 			std::memset(buffer_info[i].buffer, 0, buffer_info[i].buffer_size);
 		}
 	}
-
-	Common::LockGuard racks_lock(g_racks_mutex);
-	for (auto* rack = g_racks_list; rack != nullptr; rack = rack->next) {
-		if (rack->ngs == ngs) {
-			auto* voices = reinterpret_cast<Ngs2VoiceInternal*>(rack + 1);
-
-			for (uint32_t i = 0; i < rack->option.common.max_voices; i++) {
-				auto& voice = voices[i];
-				switch (voice.event) {
-					case Ngs2VoicePlayEvent::None:
-						if (voice.state == Ngs2VoicePlayState::Playing ||
-						    voice.state == Ngs2VoicePlayState::Stopped) {
-							voice.state = Ngs2VoicePlayState::Empty;
-						}
-						break;
-					case Ngs2VoicePlayEvent::Play:
-						if (voice.state == Ngs2VoicePlayState::Empty) {
-							voice.state = Ngs2VoicePlayState::Playing;
-						}
-						break;
-					case Ngs2VoicePlayEvent::Pause:
-						if (voice.state == Ngs2VoicePlayState::Playing) {
-							voice.state = Ngs2VoicePlayState::Paused;
-						}
-						break;
-					case Ngs2VoicePlayEvent::Resume:
-						if (voice.state == Ngs2VoicePlayState::Paused) {
-							voice.state = Ngs2VoicePlayState::Playing;
-						}
-						break;
-					case Ngs2VoicePlayEvent::Stop:
-						if (voice.state == Ngs2VoicePlayState::Playing) {
-							voice.state = Ngs2VoicePlayState::Stopped;
-						}
-						break;
-					case Ngs2VoicePlayEvent::StopImm:
-					case Ngs2VoicePlayEvent::Kill: voice.state = Ngs2VoicePlayState::Empty; break;
+	const auto grain = ngs->option.num_grain_samples;
+	for (auto* voice: voices) {
+		Ngs2RenderVoice(*voice, voices, grain);
+		if (voice->rack->type != Ngs2RackType::Mastering || !voice->has_samples) {
+			continue;
+		}
+		EXIT_NOT_IMPLEMENTED(voice->output_id >= num_buffer_info);
+		const auto& output = buffer_info[voice->output_id];
+		EXIT_NOT_IMPLEMENTED(output.buffer == nullptr);
+		EXIT_NOT_IMPLEMENTED(output.waveform_type != 0x18 ||
+		                     output.num_channels != voice->channels);
+		EXIT_NOT_IMPLEMENTED(output.buffer_size < grain * output.num_channels * sizeof(float));
+		auto* pcm = static_cast<float*>(output.buffer);
+		for (uint32_t dst = 0; dst < output.num_channels; ++dst) {
+			for (uint32_t src = 0; src < voice->channels; ++src) {
+				const float level = voice->output_matrix.empty()
+				                        ? (src == dst ? 1.0f : 0.0f)
+				                        : voice->output_matrix.at(src * output.num_channels + dst);
+				for (uint32_t i = 0; i < grain; ++i) {
+					pcm[i * output.num_channels + dst] += voice->samples[src * grain + i] * level;
 				}
-				voice.event = Ngs2VoicePlayEvent::None;
 			}
 		}
 	}
-
-	ngs->render_count++;
-
+	++ngs->render_count;
 	return OK;
 }
 
@@ -2945,6 +3262,7 @@ int KYTY_SYSV_ABI Ngs2VoiceControl(uintptr_t voice_handle, const Ngs2VoiceParamH
 					case 0x0001: {
 						EXIT_NOT_IMPLEMENTED(param->size != sizeof(Ngs2VoiceMatrixLevelsParam));
 						const auto* ml = reinterpret_cast<const Ngs2VoiceMatrixLevelsParam*>(param);
+						voice->SetMatrix(ml->matrix_id, ml->levels, ml->num_levels);
 						LOGF("\t matrix_id  = %u\n"
 						     "\t num_levels = %u\n"
 						     "\t levels     = 0x%016" PRIx64 "\n",
@@ -2955,6 +3273,7 @@ int KYTY_SYSV_ABI Ngs2VoiceControl(uintptr_t voice_handle, const Ngs2VoiceParamH
 						EXIT_NOT_IMPLEMENTED(param->size != sizeof(Ngs2VoicePortVolumeParam));
 						const auto* volume =
 						    reinterpret_cast<const Ngs2VoicePortVolumeParam*>(param);
+						voice->SetVolume(volume->port, volume->level);
 						LOGF("\t port  = %u\n"
 						     "\t level = %f\n",
 						     volume->port, volume->level);
@@ -2963,6 +3282,7 @@ int KYTY_SYSV_ABI Ngs2VoiceControl(uintptr_t voice_handle, const Ngs2VoiceParamH
 					case 0x0003: {
 						EXIT_NOT_IMPLEMENTED(param->size != sizeof(Ngs2VoicePortMatrixParam));
 						const auto* pm = reinterpret_cast<const Ngs2VoicePortMatrixParam*>(param);
+						voice->SetPortMatrix(pm->port, pm->matrix_id);
 						LOGF("\t port      = %u\n"
 						     "\t matrix_id = %d\n",
 						     pm->port, pm->matrix_id);
@@ -2979,6 +3299,10 @@ int KYTY_SYSV_ABI Ngs2VoiceControl(uintptr_t voice_handle, const Ngs2VoiceParamH
 					case 0x0005: {
 						EXIT_NOT_IMPLEMENTED(param->size != sizeof(Ngs2VoicePatchParam));
 						const auto* patch = reinterpret_cast<const Ngs2VoicePatchParam*>(param);
+						EXIT_NOT_IMPLEMENTED(patch->port >= voice->ports.size());
+						voice->ports[patch->port].dest =
+						    reinterpret_cast<Ngs2VoiceInternal*>(patch->dest_handle);
+						voice->ports[patch->port].input = patch->dest_input_id;
 						LOGF("\t connect->port          = %u\n"
 						     "\t connect->dest_input_id = %u\n"
 						     "\t connect->dest_handle   = 0x%016" PRIx64 "\n",
@@ -2988,15 +3312,7 @@ int KYTY_SYSV_ABI Ngs2VoiceControl(uintptr_t voice_handle, const Ngs2VoiceParamH
 					case 0x0006: {
 						EXIT_NOT_IMPLEMENTED(param->size != sizeof(Ngs2VoiceEventParam));
 						const auto* event = reinterpret_cast<const Ngs2VoiceEventParam*>(param);
-						switch (event->event_id) {
-							case 0x0001: voice->event = Ngs2VoicePlayEvent::Play; break;
-							case 0x0002: voice->event = Ngs2VoicePlayEvent::Stop; break;
-							case 0x0004: voice->event = Ngs2VoicePlayEvent::StopImm; break;
-							case 0x0008: voice->event = Ngs2VoicePlayEvent::Kill; break;
-							case 0x0010: voice->event = Ngs2VoicePlayEvent::Pause; break;
-							case 0x0020: voice->event = Ngs2VoicePlayEvent::Resume; break;
-							default: EXIT("unknown event_id: 0x%08" PRIx32 "\n", event->event_id);
-						}
+						voice->SetEvent(event->event_id);
 						LOGF("\t event = %u\n", event->event_id);
 						break;
 					}
@@ -3021,24 +3337,89 @@ int KYTY_SYSV_ABI Ngs2VoiceControl(uintptr_t voice_handle, const Ngs2VoiceParamH
 			case 0x1000: EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::Sampler); break;
 			case 0x2000: EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::Submixer); break;
 			case 0x2001: EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::Reverb); break;
-			case 0x3000: EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::Mastering); break;
-			case 0x4000: {
-				EXIT_NOT_IMPLEMENTED(!Ngs2RackIsCustom(voice->rack->type));
-				auto cid       = param->id & 0xffffu;
-				auto module_id = (cid >> 8u) & 0xffu;
-				auto ctl_id    = (cid >> 5u) & 0x7u;
-				auto module_no = cid & 0x1fu;
-				LOGF("\t custom module_id = 0x%02" PRIx32 ", ctl_id = 0x%" PRIx32
-				     ", module_no = %" PRIu32 "\n",
-				     module_id, ctl_id, module_no);
+			case 0x3000: {
+				EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::Mastering);
+				const auto* values = reinterpret_cast<const uint32_t*>(param + 1);
+				switch (param->id & 0xffffu) {
+					case 0: voice->channels = values[0]; break;
+					case 1: {
+						const auto* matrix =
+						    reinterpret_cast<const Ngs2VoiceMatrixLevelsParam*>(param);
+						EXIT_NOT_IMPLEMENTED(matrix->matrix_id != 5);
+						voice->output_matrix.assign(matrix->levels,
+						                            matrix->levels + matrix->num_levels);
+						break;
+					}
+					case 5: voice->output_id = values[0]; break;
+					default: EXIT("unsupported mastering control: 0x%08" PRIx32 "\n", param->id);
+				}
 				break;
 			}
-			case 0x4001:
+			case 0x4000: {
+				EXIT_NOT_IMPLEMENTED(!Ngs2RackIsCustom(voice->rack->type));
+				const auto index = param->id & 0x1fu;
+				EXIT_NOT_IMPLEMENTED((param->id & 0xffffffe0u) != 0x40001f00u ||
+				                     index >= voice->modules.size());
+				struct FxParam {
+					Ngs2VoiceParamHeader header;
+					const void*          data;
+					size_t               size;
+				};
+				const auto& fx     = *reinterpret_cast<const FxParam*>(param);
+				auto&       module = voice->modules[index];
+				EXIT_NOT_IMPLEMENTED(voice->rack->fx[index].control != 0 ||
+				                     fx.size != module.param.size());
+				std::memcpy(module.param.data(), fx.data, fx.size);
+				module.flags |= 2;
+				break;
+			}
+			case 0x4001: {
 				EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::CustomSampler);
+				switch (param->id & 0xffffu) {
+					case 0: {
+						const auto& format =
+						    *reinterpret_cast<const Ngs2WaveformFormat*>(param + 1);
+						voice->SetupSampler(format);
+						break;
+					}
+					case 1: {
+						struct BlocksParam {
+							Ngs2VoiceParamHeader     header;
+							const uint8_t*           data;
+							uint32_t                 flags, count;
+							const Ngs2WaveformBlock* blocks;
+						};
+						const auto& blocks = *reinterpret_cast<const BlocksParam*>(param);
+						EXIT_NOT_IMPLEMENTED(!voice->accepts_blocks ||
+						                     (blocks.flags != 0 && blocks.flags != 0x11));
+						voice->accepts_blocks = (blocks.flags & 1u) != 0;
+						for (uint32_t i = 0; i < blocks.count; ++i) {
+							const auto& block = blocks.blocks[i];
+							EXIT_NOT_IMPLEMENTED(block.num_repeats != 0 || block.num_samples == 0);
+							EXIT_NOT_IMPLEMENTED(
+							    voice->decoder == nullptr &&
+							    (uint64_t(block.num_skip_samples) + block.num_samples) *
+							            voice->channels * 2 >
+							        block.data_size);
+							voice->blocks.push_back({blocks.data + block.data_offset, block});
+						}
+						break;
+					}
+					case 5:
+						EXIT_NOT_IMPLEMENTED(*reinterpret_cast<const float*>(param + 1) != 1.0f);
+						break;
+					default: EXIT("unsupported sampler control: 0x%08" PRIx32 "\n", param->id);
+				}
 				break;
-			case 0x4002:
-				EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::CustomSubmixer);
+			}
+			case 0x4002: {
+				EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::CustomSubmixer ||
+				                     (param->id & 0xffffu) != 0);
+				const auto* channels   = reinterpret_cast<const uint32_t*>(param + 1);
+				EXIT_NOT_IMPLEMENTED(channels[0] != channels[1]);
+				voice->channels = channels[0];
 				break;
+			}
 			case 0x4003:
 				EXIT_NOT_IMPLEMENTED(voice->rack->type != Ngs2RackType::CustomMastering);
 				break;
@@ -3056,14 +3437,44 @@ int KYTY_SYSV_ABI Ngs2VoiceControl(uintptr_t voice_handle, const Ngs2VoiceParamH
 }
 
 int KYTY_SYSV_ABI Ngs2VoiceRunCommands(uintptr_t voice_handle, const void* commands,
-                                       uint32_t num_commands, uint32_t flags) {
-	PRINT_NAME();
-
-	(void)voice_handle;
-	(void)commands;
-	(void)num_commands;
-	(void)flags;
-
+                                       size_t num_commands) {
+	struct Command {
+		uint32_t id;
+		uint8_t  flags, type;
+		uint16_t count;
+		union {
+			uint32_t     u;
+			int32_t      i;
+			float        f;
+			const float* levels;
+		} value;
+	};
+	auto*             voice = reinterpret_cast<Ngs2VoiceInternal*>(voice_handle);
+	Common::LockGuard lock(voice->rack->ngs->mutex);
+	const auto*       params = static_cast<const Command*>(commands);
+	for (size_t i = 0; i < num_commands; ++i) {
+		const auto& p     = params[i];
+		const auto  index = p.id >> 24;
+		switch (p.id & 0xffffffu) {
+			case 2:
+				EXIT_NOT_IMPLEMENTED(p.type != 4);
+				voice->SetEvent(p.value.u);
+				break;
+			case 5:
+				EXIT_NOT_IMPLEMENTED(p.type != 0x11);
+				voice->SetMatrix(index, p.value.levels, p.count);
+				break;
+			case 6:
+				EXIT_NOT_IMPLEMENTED(p.type != 1);
+				voice->SetVolume(index, p.value.f);
+				break;
+			case 7:
+				EXIT_NOT_IMPLEMENTED(p.type != 3);
+				voice->SetPortMatrix(index, p.value.i);
+				break;
+			default: EXIT("unknown command: 0x%08" PRIx32 "\n", p.id);
+		}
+	}
 	return OK;
 }
 
@@ -3120,7 +3531,7 @@ int KYTY_SYSV_ABI Ngs2VoiceGetState(uintptr_t voice_handle, Ngs2VoiceState* stat
 			sampler->waveform_data       = nullptr;
 			break;
 		}
-		default: EXIT("unknown type: %s\n", Common::EnumName(voice->rack->type).c_str());
+		default: EXIT("unknown type: %s\n", magic_enum::enum_name(voice->rack->type));
 	}
 
 	return OK;

src/libs/audio.h

@@ -336,7 +336,7 @@ int KYTY_SYSV_ABI Ngs2RackGetVoiceHandle(uintptr_t rack_handle, uint32_t voice_i
                                          uintptr_t* handle);
 int KYTY_SYSV_ABI Ngs2VoiceControl(uintptr_t voice_handle, const Ngs2VoiceParamHeader* param_list);
 int KYTY_SYSV_ABI Ngs2VoiceRunCommands(uintptr_t voice_handle, const void* commands,
-                                       uint32_t num_commands, uint32_t flags);
+                                       size_t num_commands);
 int KYTY_SYSV_ABI Ngs2VoiceGetState(uintptr_t voice_handle, Ngs2VoiceState* state,
                                     size_t state_size);
 int KYTY_SYSV_ABI Ngs2VoiceGetStateFlags(uintptr_t voice_handle, uint32_t* state_flags);

src/libs/avPlayer.cpp

@@ -1,6 +1,5 @@
 #include "common/common.h"
 #include "common/logging/log.h"
-#include "common/magicEnum.h"
 #include "common/stringUtils.h"
 #include "kernel/fileSystem.h"
 #include "kernel/pthread.h"
@@ -16,6 +15,7 @@
 #include <cstdio>
 #include <cstring>
 #include <deque>
+#include <magic_enum.hpp>
 #include <memory>
 #include <mutex>
 #include <optional>
@@ -117,6 +117,13 @@ struct AvPlayerEventReplacement {
 	void*                 object_pointer = nullptr;
 	AvPlayerEventCallback event_callback = nullptr;
 };
+static void emit_event(AvPlayerEventReplacement event, int32_t id, void* data = nullptr) {
+	if (event.event_callback != nullptr) {
+		LOGF("\t event_id = %d\n", id);
+		event.event_callback(event.object_pointer, id, 0, data);
+	}
+}
+
 struct AvPlayerUri {
 	const char* name   = nullptr;
 	uint32_t    length = 0;
@@ -639,9 +646,9 @@ private:
 
 class Source {
 public:
-	Source(AvPlayerMemAllocator m, AvPlayerFileReplacement f, int32_t video_buffers, bool vdec2,
-	       uint32_t demux_buffer_size)
-	    : mem(m), file(f),
+	Source(AvPlayerMemAllocator m, AvPlayerFileReplacement f, AvPlayerEventReplacement e,
+	       int32_t video_buffers, bool vdec2, uint32_t demux_buffer_size)
+	    : mem(m), file(f), event(e),
 	      max_video_buffers(std::clamp(video_buffers <= 0 ? 2 : video_buffers, 2, 16)),
 	      use_vdec2(vdec2),
 	      video_packets(
@@ -817,21 +824,26 @@ public:
 		return result;
 	}
 	int Stop() {
-		std::scoped_lock lifecycle_lock(lifecycle_mutex);
-		bool             was_playing_video = false;
+		bool                     was_playing_video = false;
+		AvPlayerEventReplacement stop_event;
 		{
-			std::lock_guard lock(mutex);
-			was_playing_video = !stopped && video_id.has_value();
-			stopped           = true;
-		}
-		StopWorkers();
-		{
-			std::lock_guard lock(mutex);
-			ResetNoLock();
+			std::scoped_lock lifecycle_lock(lifecycle_mutex);
+			{
+				std::lock_guard lock(mutex);
+				was_playing_video = !stopped && video_id.has_value();
+				stopped           = true;
+			}
+			StopWorkers();
+			{
+				std::lock_guard lock(mutex);
+				ResetNoLock();
+			}
+			stop_event = TakeStopEvent();
 		}
 		if (was_playing_video) {
 			::printf("AvPlayer video stopped\n");
 		}
+		emit_event(stop_event, AVPLAYER_EVENT_STATE_STOP);
 		return 0;
 	}
 	void Pause() {
@@ -860,9 +872,7 @@ public:
 	}
 	bool Active() const {
 		std::lock_guard lock(mutex);
-		return !stopped && !pipeline_failed &&
-		       (!demux_eof || !video_done || !audio_done || !video_frames.Empty() ||
-		        !audio_frames.Empty());
+		return !stopped && !pipeline_failed && !DrainedNoLock();
 	}
 	uint64_t CurrentTime() const {
 		std::lock_guard lock(mutex);
@@ -945,6 +955,7 @@ public:
 		current_video = std::move(*frame);
 		*out          = current_video->info;
 		RecordLoopBoundary(*current_video);
+		completion_changed.notify_all();
 		if (deliver_seek_frame) {
 			seek_video_frame_pending = false;
 		}
@@ -976,6 +987,7 @@ public:
 		            current_audio->info.details.audio.language_code, 4);
 		last_audio_ts = out->time_stamp;
 		RecordLoopBoundary(*current_audio);
+		completion_changed.notify_all();
 		return true;
 	}
 	std::optional<int32_t> TakeWarning() {
@@ -988,6 +1000,13 @@ public:
 	}
 
 private:
+	bool DrainedNoLock() const {
+		return demux_eof && video_done && audio_done && video_frames.Empty() &&
+		       audio_frames.Empty();
+	}
+	AvPlayerEventReplacement TakeStopEvent() {
+		return stop_event_sent.exchange(true) ? AvPlayerEventReplacement {} : event;
+	}
 	void RecordLoopBoundary(const ReadyFrame& frame) {
 		if (frame.timestamp_offset > last_output_loop_offset) {
 			last_output_loop_offset = frame.timestamp_offset;
@@ -1002,23 +1021,17 @@ private:
 		streamer.reset();
 	}
 	void ResetNoLock(bool retain_output_buffers = false) {
-		std::optional<ReadyFrame>                delivered_video;
-		std::optional<ReadyFrame>                delivered_audio;
-		std::deque<std::unique_ptr<GuestBuffer>> in_flight_video;
-		if (retain_output_buffers) {
-			delivered_video = std::move(current_video);
-			delivered_audio = std::move(current_audio);
-			in_flight_video = std::move(retired_video);
-		}
 		video_packets.Clear();
 		audio_packets.Clear();
 		video_frames.Clear();
 		audio_frames.Clear();
 		video_buffers.Clear();
 		audio_buffers.Clear();
-		retired_video.clear();
-		current_video.reset();
-		current_audio.reset();
+		if (!retain_output_buffers) {
+			retired_video.clear();
+			current_video.reset();
+			current_audio.reset();
+		}
 		if (sws != nullptr) {
 			sws_freeContext(sws);
 			sws = nullptr;
@@ -1042,9 +1055,6 @@ private:
 		last_audio_ts            = 0;
 		last_output_loop_offset  = 0;
 		pending_loop_warnings    = 0;
-		current_video            = std::move(delivered_video);
-		current_audio            = std::move(delivered_audio);
-		retired_video            = std::move(in_flight_video);
 	}
 	void AutoEnable() {
 		for (uint32_t i = 0; i < fmt->nb_streams; i++) {
@@ -1174,6 +1184,7 @@ private:
 		return true;
 	}
 	bool StartWorkers() {
+		std::unique_lock lock(mutex);
 		worker_stop     = false;
 		interrupt_io    = false;
 		pipeline_failed = false;
@@ -1183,20 +1194,26 @@ private:
 		if ((video_id && !CreateWorker(video_thread, VideoDecoderEntry, "AvPlayerVideoDecoder")) ||
 		    (audio_id && !CreateWorker(audio_thread, AudioDecoderEntry, "AvPlayerAudioDecoder")) ||
 		    !CreateWorker(demux_thread, DemuxEntry, "AvPlayerDemuxer")) {
+			lock.unlock();
 			StopWorkers();
 			return false;
 		}
+		stop_event_sent = false;
 		return true;
 	}
 	void NotifyWorkers() {
+		completion_changed.notify_all();
 		video_packets.Notify();
 		audio_packets.Notify();
 		video_buffers.Notify();
 		audio_buffers.Notify();
 	}
 	void StopWorkers() {
-		worker_stop  = true;
-		interrupt_io = true;
+		{
+			std::lock_guard lock(mutex);
+			worker_stop  = true;
+			interrupt_io = true;
+		}
 		NotifyWorkers();
 		if (demux_thread != nullptr) {
 			LibKernel::PthreadJoin(demux_thread, nullptr);
@@ -1212,9 +1229,12 @@ private:
 		}
 	}
 	void FailPipeline() {
-		pipeline_failed = true;
-		worker_stop     = true;
-		interrupt_io    = true;
+		{
+			std::lock_guard lock(mutex);
+			pipeline_failed = true;
+			worker_stop     = true;
+			interrupt_io    = true;
+		}
 		NotifyWorkers();
 	}
 	void Demux() {
@@ -1269,6 +1289,13 @@ private:
 		demux_eof = true;
 		video_packets.Notify();
 		audio_packets.Notify();
+		std::unique_lock lock(mutex);
+		completion_changed.wait(lock, [&] { return worker_stop || DrainedNoLock(); });
+		if (!worker_stop) {
+			const auto stop_event = TakeStopEvent();
+			lock.unlock();
+			emit_event(stop_event, AVPLAYER_EVENT_STATE_STOP);
+		}
 	}
 	void VideoDecoder() {
 		Decoder(video_packets, video_ctx, video_buffers, video_frames, video_done,
@@ -1316,7 +1343,11 @@ private:
 				break;
 			}
 		}
-		done = true;
+		{
+			std::lock_guard lock(mutex);
+			done = true;
+		}
+		completion_changed.notify_all();
 	}
 	bool DecodePacket(AVCodecContext* codec, AVPacket* packet, uint64_t timestamp_offset,
 	                  WorkQueue<std::unique_ptr<GuestBuffer>>& buffers,
@@ -1607,6 +1638,7 @@ private:
 	}
 	AvPlayerMemAllocator                     mem;
 	AvPlayerFileReplacement                  file;
+	AvPlayerEventReplacement                 event;
 	int                                      max_video_buffers = 2;
 	bool                                     use_vdec2         = false;
 	AvPlayerSourceType                       source_type       = AvPlayerSourceUnknown;
@@ -1640,6 +1672,8 @@ private:
 	std::atomic_bool                         demux_eof {true};
 	std::atomic_bool                         video_done {true};
 	std::atomic_bool                         audio_done {true};
+	std::atomic_bool                         stop_event_sent {true};
+	std::condition_variable                  completion_changed;
 	std::mutex                               lifecycle_mutex;
 	mutable std::mutex                       mutex;
 	bool                                     stopped                  = true;
@@ -1673,19 +1707,13 @@ struct AvPlayerInternal {
 static bool valid_allocators(const AvPlayerMemAllocator& m) {
 	return m.allocate && m.deallocate && m.allocate_texture && m.deallocate_texture;
 }
-static void emit_event(AvPlayerInternal* h, int32_t id, void* data = nullptr) {
-	if (h != nullptr && h->event.event_callback != nullptr) {
-		LOGF("\t event_id = %d\n", id);
-		h->event.event_callback(h->event.object_pointer, id, 0, data);
-	}
-}
 static void pump_warnings(AvPlayerInternal* h) {
 	if (h == nullptr || h->source == nullptr) {
 		return;
 	}
 	while (auto w = h->source->TakeWarning()) {
 		int32_t warning = *w;
-		emit_event(h, AVPLAYER_EVENT_WARNING_ID, &warning);
+		emit_event(h->event, AVPLAYER_EVENT_WARNING_ID, &warning);
 	}
 }
 static AvPlayerInternal* create_player(const AvPlayerMemAllocator&     mem,
@@ -1713,17 +1741,17 @@ static int add_source(AvPlayerInternal* h, const std::string& filename, AvPlayer
 	}
 	bool vdec2 =
 	    h->post_init.video_decoder_init.decoder_type.video_type == AvPlayerVideoDecoderSoftware2;
-	auto s = std::make_unique<Source>(h->mem, h->file, h->video_buffers, vdec2,
+	auto s = std::make_unique<Source>(h->mem, h->file, h->event, h->video_buffers, vdec2,
 	                                  h->post_init.demux_video_buffer_size);
 	if (auto rc = s->Init(filename, type); rc < 0) {
 		return rc;
 	}
 	h->source = std::move(s);
-	emit_event(h, AVPLAYER_EVENT_STATE_READY);
+	emit_event(h->event, AVPLAYER_EVENT_STATE_READY);
 	if (h->auto_start) {
 		auto rc = h->source->Start();
 		if (rc == 0) {
-			emit_event(h, AVPLAYER_EVENT_STATE_PLAY);
+			emit_event(h->event, AVPLAYER_EVENT_STATE_PLAY);
 		}
 		return rc;
 	}
@@ -1758,7 +1786,7 @@ static void log_init_common(const AvPlayerMemAllocator& mem, const AvPlayerFileR
 	     "= %d\n\t base_priority                   "
 	     "      = %u\n\t auto_start                            = %u\n\t default_language           "
 	     "           = %s\n",
-	     Common::EnumName(debug_level).c_str(), buffers, priority, auto_start,
+	     magic_enum::enum_name(debug_level), buffers, priority, auto_start,
 	     language == nullptr ? "(null)" : language);
 }
 
@@ -1872,7 +1900,7 @@ int KYTY_SYSV_ABI AvPlayerStart(AvPlayerInternal* h) {
 	}
 	auto rc = h->source->Start();
 	if (rc == 0) {
-		emit_event(h, AVPLAYER_EVENT_STATE_PLAY);
+		emit_event(h->event, AVPLAYER_EVENT_STATE_PLAY);
 	}
 	return rc;
 }
@@ -1887,7 +1915,7 @@ int KYTY_SYSV_ABI AvPlayerStartEx(AvPlayerInternal* h, const void* start_info_ex
 	        : static_cast<const AvPlayerStartInfoEx*>(start_info_ex)->start_time_milliseconds;
 	auto rc = h->source->Start(ms);
 	if (rc == 0) {
-		emit_event(h, AVPLAYER_EVENT_STATE_PLAY);
+		emit_event(h->event, AVPLAYER_EVENT_STATE_PLAY);
 	}
 	return rc;
 }
@@ -1896,9 +1924,7 @@ int KYTY_SYSV_ABI AvPlayerStop(AvPlayerInternal* h) {
 	if (h == nullptr || h->source == nullptr) {
 		return AVPLAYER_ERROR_INVALID_PARAMS;
 	}
-	auto rc = h->source->Stop();
-	emit_event(h, AVPLAYER_EVENT_STATE_STOP);
-	return rc;
+	return h->source->Stop();
 }
 int KYTY_SYSV_ABI AvPlayerPause(AvPlayerInternal* h) {
 	PRINT_NAME();
@@ -1906,7 +1932,7 @@ int KYTY_SYSV_ABI AvPlayerPause(AvPlayerInternal* h) {
 		return AVPLAYER_ERROR_INVALID_PARAMS;
 	}
 	h->source->Pause();
-	emit_event(h, AVPLAYER_EVENT_STATE_PAUSE);
+	emit_event(h->event, AVPLAYER_EVENT_STATE_PAUSE);
 	return 0;
 }
 int KYTY_SYSV_ABI AvPlayerResume(AvPlayerInternal* h) {
@@ -1915,7 +1941,7 @@ int KYTY_SYSV_ABI AvPlayerResume(AvPlayerInternal* h) {
 		return AVPLAYER_ERROR_INVALID_PARAMS;
 	}
 	h->source->Resume();
-	emit_event(h, AVPLAYER_EVENT_STATE_PLAY);
+	emit_event(h->event, AVPLAYER_EVENT_STATE_PLAY);
 	return 0;
 }
 int KYTY_SYSV_ABI AvPlayerSetLooping(AvPlayerInternal* h, Bool loop) {
@@ -2017,7 +2043,7 @@ int KYTY_SYSV_ABI AvPlayerJumpToTime(AvPlayerInternal* h, uint64_t time_ms) {
 	auto rc = h->source->Jump(time_ms);
 	if (rc == 0) {
 		int32_t w = AVPLAYER_WARNING_JUMP_COMPLETE;
-		emit_event(h, AVPLAYER_EVENT_WARNING_ID, &w);
+		emit_event(h->event, AVPLAYER_EVENT_WARNING_ID, &w);
 	}
 	return rc;
 }

src/libs/controller.cpp

@@ -13,6 +13,8 @@
 #include "libs/padData.h"
 
 #include <algorithm>
+#include <array>
+#include <cmath>
 #include <cstring>
 #include <vector>
 
@@ -87,6 +89,9 @@ struct ControllerState {
 	uint32_t buttons                               = 0;
 	int      axes[static_cast<int>(Axis::AxisMax)] = {128, 128, 128, 128, 0, 0};
 	Touch    touch[2];
+	std::array<float, 3> accel {0.0f, 1.0f, 0.0f};
+	std::array<float, 3> gyro {};
+	std::array<float, 4> orientation {0.0f, 0.0f, 0.0f, 1.0f};
 };
 
 class GameController {
@@ -101,6 +106,9 @@ public:
 	void Axis(int id, Axis axis, int value);
 	void RightStick(int id, int x, int y);
 	void TouchPad(int id, int finger, bool down, float x, float y);
+	void Motion(int id, Sensor sensor, const float* data, uint64_t time_us);
+	void SetMotionSensorState(bool enable);
+	void ResetOrientation();
 	void ResetInputState();
 	void ReleaseHostPads();
 	void GetConnectionInfo(bool* flag, int* count);
@@ -121,6 +129,8 @@ private:
 	int              m_active_id       = -1;
 	bool             m_connected       = false;
 	int              m_connected_count = 0;
+	bool             m_motion_enabled  = true;
+	uint64_t         m_gyro_time       = 0;
 	ControllerState  m_state;
 	ControllerState  m_states[STATES_MAX];
 	bool             m_obtained[STATES_MAX] {};
@@ -144,7 +154,16 @@ static void pad_fill_data(PadData* data, const ControllerState& state, bool conn
 	data->right_stick_y     = state.axes[static_cast<int>(Axis::RightY)];
 	data->analog_buttons_l2 = state.axes[static_cast<int>(Axis::TriggerLeft)];
 	data->analog_buttons_r2 = state.axes[static_cast<int>(Axis::TriggerRight)];
-	data->orientation_w     = 1.0f;
+	data->acceleration_x     = state.accel[0];
+	data->acceleration_y     = state.accel[1];
+	data->acceleration_z     = state.accel[2];
+	data->angular_velocity_x = state.gyro[0];
+	data->angular_velocity_y = state.gyro[1];
+	data->angular_velocity_z = state.gyro[2];
+	data->orientation_x      = state.orientation[0];
+	data->orientation_y      = state.orientation[1];
+	data->orientation_z      = state.orientation[2];
+	data->orientation_w      = state.orientation[3];
 	for (const auto& touch: state.touch) {
 		if (touch.down) {
 			auto& output = data->touch_data.touch[data->touch_data.touch_num++];
@@ -273,6 +292,18 @@ void GameController::Connect(int id) {
 
 	m_connected_ids.push_back(id);
 
+	if (id != HOST_INPUT_CONTROLLER_ID) {
+		if (auto* pad = SDL_GameControllerFromInstanceID(static_cast<SDL_JoystickID>(id));
+		    pad != nullptr) {
+			for (auto sensor: {SDL_SENSOR_ACCEL, SDL_SENSOR_GYRO}) {
+				if (SDL_GameControllerHasSensor(pad, sensor) &&
+				    SDL_GameControllerSetSensorEnabled(pad, sensor, SDL_TRUE) != 0) {
+					LOGF("\t enabling controller sensor failed: %s\n", SDL_GetError());
+				}
+			}
+		}
+	}
+
 	CheckActive();
 }
 
@@ -311,6 +342,7 @@ void GameController::CheckActive() {
 	m_active_id     = new_active_id;
 	m_connected     = new_connected;
 	m_state         = {};
+	m_gyro_time     = 0;
 	m_states_num    = 0;
 	m_first_state   = 0;
 	m_next_touch_id = 1;
@@ -402,9 +434,81 @@ void GameController::TouchPad(int id, int finger, bool down, float x, float y) {
 	}
 }
 
+void GameController::Motion(int id, Sensor sensor, const float* data, uint64_t time_us) {
+	Common::LockGuard lock(m_mutex);
+	if (id != m_active_id || !m_motion_enabled) {
+		return;
+	}
+
+	// Convert acceleration to G; angular velocity is already in rad/s.
+	if (sensor == Sensor::Accel) {
+		for (int i = 0; i < 3; i++) {
+			m_state.accel[i] = data[i] / SDL_STANDARD_GRAVITY;
+		}
+	} else {
+		std::copy_n(data, 3, m_state.gyro.begin());
+		// Do not extrapolate a single sample across lost reports (e.g. loss of window focus).
+		constexpr uint64_t max_gyro_interval_us = 100000;
+		if (m_gyro_time != 0 && time_us > m_gyro_time &&
+		    time_us - m_gyro_time <= max_gyro_interval_us) {
+			const float dt = static_cast<float>(time_us - m_gyro_time) * 0.000001f;
+			const float speed =
+			    std::sqrt(data[0] * data[0] + data[1] * data[1] + data[2] * data[2]);
+			const float half_angle = speed * dt * 0.5f;
+			const float scale      = speed > 0.0f ? std::sin(half_angle) / speed : 0.0f;
+			const float x          = data[0] * scale;
+			const float y          = data[1] * scale;
+			const float z          = data[2] * scale;
+			const float w          = std::cos(half_angle);
+			const auto  q          = m_state.orientation;
+			// Accumulate body-local rotation relative to connection / orientation reset.
+			m_state.orientation = {q[3] * x + q[0] * w + q[1] * z - q[2] * y,
+			                       q[3] * y - q[0] * z + q[1] * w + q[2] * x,
+			                       q[3] * z + q[0] * y - q[1] * x + q[2] * w,
+			                       q[3] * w - q[0] * x - q[1] * y - q[2] * z};
+			float length        = 0.0f;
+			for (float value: m_state.orientation) {
+				length += value * value;
+			}
+			length = std::sqrt(length);
+			for (float& value: m_state.orientation) {
+				value /= length;
+			}
+		}
+		m_gyro_time = time_us;
+	}
+	m_state.time = LibKernel::KernelGetProcessTime();
+	AddState();
+}
+
+void GameController::SetMotionSensorState(bool enable) {
+	Common::LockGuard lock(m_mutex);
+	if (m_motion_enabled != enable) {
+		m_motion_enabled = enable;
+		m_gyro_time      = 0;
+		m_state.accel    = {0.0f, 1.0f, 0.0f};
+		m_state.gyro     = {};
+		m_state.time     = LibKernel::KernelGetProcessTime();
+		AddState();
+	}
+}
+
+void GameController::ResetOrientation() {
+	Common::LockGuard lock(m_mutex);
+	m_state.orientation = {0.0f, 0.0f, 0.0f, 1.0f};
+	m_gyro_time         = 0;
+	m_state.time        = LibKernel::KernelGetProcessTime();
+	AddState();
+}
+
 void GameController::ResetInputState() {
 	Common::LockGuard lock(m_mutex);
-	m_state         = {};
+	m_state.buttons = 0;
+	std::fill_n(m_state.axes, 4, 128);
+	std::fill_n(m_state.axes + 4, 2, 0);
+	for (auto& touch: m_state.touch) {
+		touch = {};
+	}
 	m_state.time    = LibKernel::KernelGetProcessTime();
 	m_states_num    = 0;
 	m_first_state   = 0;
@@ -580,6 +684,10 @@ void SetTouchPad(int id, int finger, bool down, float x, float y) {
 	g_controller->TouchPad(id, finger, down, x, y);
 }
 
+void SetSensor(int id, Sensor sensor, const float* data, uint64_t time_us) {
+	g_controller->Motion(id, sensor, data, time_us);
+}
+
 void ResetInputState() {
 	g_controller->ResetInputState();
 }
@@ -646,6 +754,7 @@ int KYTY_SYSV_ABI PadSetMotionSensorState(int handle, bool enable) {
 	}
 
 	LOGF("\t enable = %s\n", (enable ? "true" : "false"));
+	g_controller->SetMotionSensorState(enable);
 
 	return OK;
 }
@@ -669,6 +778,7 @@ int KYTY_SYSV_ABI PadResetOrientation(int handle) {
 		return PAD_ERROR_INVALID_HANDLE;
 	}
 
+	g_controller->ResetOrientation();
 	return OK;
 }
 
@@ -702,6 +812,20 @@ int KYTY_SYSV_ABI PadGetControllerInformation(int handle, PadControllerInformati
 	return OK;
 }
 
+int KYTY_SYSV_ABI PadIsRemoteController(int handle, bool* is_remote) {
+	PRINT_NAME();
+
+	if (handle != 1) {
+		return PAD_ERROR_INVALID_HANDLE;
+	}
+	if (is_remote == nullptr) {
+		return PAD_ERROR_INVALID_ARG;
+	}
+
+	*is_remote = false;
+	return OK;
+}
+
 int KYTY_SYSV_ABI PadReadState(int handle, PadData* data) {
 	PRINT_NAME();
 

src/libs/controller.h

@@ -46,6 +46,8 @@ enum class Axis {
 	AxisMax
 };
 
+enum class Sensor { Accel, Gyro };
+
 struct PadControllerInformation;
 struct PadData;
 struct PadVibrationParam;
@@ -69,6 +71,7 @@ void SetButton(int id, uint32_t button, bool down);
 void SetAxis(int id, Axis axis, int value);
 void SetRightStick(int id, int x, int y);
 void SetTouchPad(int id, int finger, bool down, float x, float y);
+void SetSensor(int id, Sensor sensor, const float* data, uint64_t time_us);
 void ResetInputState();
 
 int KYTY_SYSV_ABI PadInit();
@@ -78,6 +81,7 @@ int KYTY_SYSV_ABI PadSetMotionSensorState(int handle, bool enable);
 int KYTY_SYSV_ABI PadSetAngularVelocityDeadbandState(int handle, bool enable);
 int KYTY_SYSV_ABI PadResetOrientation(int handle);
 int KYTY_SYSV_ABI PadGetControllerInformation(int handle, PadControllerInformation* info);
+int KYTY_SYSV_ABI PadIsRemoteController(int handle, bool* is_remote);
 int KYTY_SYSV_ABI PadReadState(int handle, PadData* data);
 int KYTY_SYSV_ABI PadRead(int handle, PadData* data, int num);
 int KYTY_SYSV_ABI PadSetVibration(int handle, const PadVibrationParam* param);

src/libs/guestPrintf.cpp

@@ -527,7 +527,12 @@ static int kyty_printf_internal(bool sn, char* sn_s, size_t sn_n, const char* fo
 			} else if (*format == '*') {
 				// const int prec = (int)va_arg(va, int);
 				const int prec = VaArg_int(va_list);
-				precision      = prec > 0 ? static_cast<unsigned int>(prec) : 0U;
+				if (prec < 0) {
+					// A negative dynamic precision is treated as omitted.
+					flags &= ~FLAGS_PRECISION;
+				} else {
+					precision = static_cast<unsigned int>(prec);
+				}
 				format++;
 			}
 		}

src/libs/ime.cpp

@@ -7,6 +7,7 @@
 #include <algorithm>
 #include <atomic>
 #include <cmath>
+#include <cstdint>
 #include <cstring>
 #include <deque>
 #include <mutex>
@@ -18,35 +19,31 @@
 namespace Libs::Ime {
 namespace {
 
-constexpr int Error(uint32_t value) {
-	return static_cast<int32_t>(value);
-}
-
-constexpr int ERROR_BUSY                    = Error(0x80bc0001);
-constexpr int ERROR_NOT_OPENED              = Error(0x80bc0002);
-constexpr int ERROR_CONNECTION_FAILED       = Error(0x80bc0004);
-constexpr int ERROR_EVENT_OVERFLOW          = Error(0x80bc0007);
-constexpr int ERROR_INVALID_TEXT            = Error(0x80bc0006);
-constexpr int ERROR_INVALID_USER_ID         = Error(0x80bc0010);
-constexpr int ERROR_INVALID_TYPE            = Error(0x80bc0011);
-constexpr int ERROR_INVALID_LANGUAGES       = Error(0x80bc0012);
-constexpr int ERROR_INVALID_ENTER_LABEL     = Error(0x80bc0013);
-constexpr int ERROR_INVALID_INPUT_METHOD    = Error(0x80bc0014);
-constexpr int ERROR_INVALID_OPTION          = Error(0x80bc0015);
-constexpr int ERROR_INVALID_MAX_TEXT_LENGTH = Error(0x80bc0016);
-constexpr int ERROR_INVALID_TEXT_BUFFER     = Error(0x80bc0017);
-constexpr int ERROR_INVALID_POSX            = Error(0x80bc0018);
-constexpr int ERROR_INVALID_POSY            = Error(0x80bc0019);
-constexpr int ERROR_INVALID_HALIGN          = Error(0x80bc001a);
-constexpr int ERROR_INVALID_VALIGN          = Error(0x80bc001b);
-constexpr int ERROR_INVALID_EXTENDED        = Error(0x80bc001c);
-constexpr int ERROR_INVALID_WORK            = Error(0x80bc0020);
-constexpr int ERROR_INVALID_HANDLER         = Error(0x80bc0022);
-constexpr int ERROR_NO_RESOURCE_ID          = Error(0x80bc0023);
-constexpr int ERROR_INVALID_MODE            = Error(0x80bc0024);
-constexpr int ERROR_INVALID_PARAM           = Error(0x80bc0030);
-constexpr int ERROR_INVALID_ADDRESS         = Error(0x80bc0031);
-constexpr int ERROR_INVALID_RESERVED        = Error(0x80bc0032);
+constexpr int ERROR_BUSY                    = static_cast<int32_t>(0x80bc0001);
+constexpr int ERROR_NOT_OPENED              = static_cast<int32_t>(0x80bc0002);
+constexpr int ERROR_CONNECTION_FAILED       = static_cast<int32_t>(0x80bc0004);
+constexpr int ERROR_EVENT_OVERFLOW          = static_cast<int32_t>(0x80bc0007);
+constexpr int ERROR_INVALID_TEXT            = static_cast<int32_t>(0x80bc0006);
+constexpr int ERROR_INVALID_USER_ID         = static_cast<int32_t>(0x80bc0010);
+constexpr int ERROR_INVALID_TYPE            = static_cast<int32_t>(0x80bc0011);
+constexpr int ERROR_INVALID_LANGUAGES       = static_cast<int32_t>(0x80bc0012);
+constexpr int ERROR_INVALID_ENTER_LABEL     = static_cast<int32_t>(0x80bc0013);
+constexpr int ERROR_INVALID_INPUT_METHOD    = static_cast<int32_t>(0x80bc0014);
+constexpr int ERROR_INVALID_OPTION          = static_cast<int32_t>(0x80bc0015);
+constexpr int ERROR_INVALID_MAX_TEXT_LENGTH = static_cast<int32_t>(0x80bc0016);
+constexpr int ERROR_INVALID_TEXT_BUFFER     = static_cast<int32_t>(0x80bc0017);
+constexpr int ERROR_INVALID_POSX            = static_cast<int32_t>(0x80bc0018);
+constexpr int ERROR_INVALID_POSY            = static_cast<int32_t>(0x80bc0019);
+constexpr int ERROR_INVALID_HALIGN          = static_cast<int32_t>(0x80bc001a);
+constexpr int ERROR_INVALID_VALIGN          = static_cast<int32_t>(0x80bc001b);
+constexpr int ERROR_INVALID_EXTENDED        = static_cast<int32_t>(0x80bc001c);
+constexpr int ERROR_INVALID_WORK            = static_cast<int32_t>(0x80bc0020);
+constexpr int ERROR_INVALID_HANDLER         = static_cast<int32_t>(0x80bc0022);
+constexpr int ERROR_NO_RESOURCE_ID          = static_cast<int32_t>(0x80bc0023);
+constexpr int ERROR_INVALID_MODE            = static_cast<int32_t>(0x80bc0024);
+constexpr int ERROR_INVALID_PARAM           = static_cast<int32_t>(0x80bc0030);
+constexpr int ERROR_INVALID_ADDRESS         = static_cast<int32_t>(0x80bc0031);
+constexpr int ERROR_INVALID_RESERVED        = static_cast<int32_t>(0x80bc0032);
 
 constexpr uint32_t VALID_OPTIONS           = 0x00007bff;
 constexpr uint32_t VALID_EXTENDED_OPTIONS  = 0x00004fde;

src/libs/imeDialog.cpp

@@ -5,6 +5,7 @@
 #include <algorithm>
 #include <atomic>
 #include <cmath>
+#include <cstdint>
 #include <cstring>
 #include <mutex>
 #include <vector>
@@ -13,31 +14,27 @@ namespace Libs::Dialog::ImeDialog {
 
 namespace {
 
-constexpr int Error(uint32_t value) {
-	return static_cast<int32_t>(value);
-}
-
-constexpr int ERROR_BUSY                    = Error(0x80bc0001);
-constexpr int ERROR_INVALID_USER_ID         = Error(0x80bc0010);
-constexpr int ERROR_INVALID_TYPE            = Error(0x80bc0011);
-constexpr int ERROR_INVALID_LANGUAGES       = Error(0x80bc0012);
-constexpr int ERROR_INVALID_ENTER_LABEL     = Error(0x80bc0013);
-constexpr int ERROR_INVALID_INPUT_METHOD    = Error(0x80bc0014);
-constexpr int ERROR_INVALID_OPTION          = Error(0x80bc0015);
-constexpr int ERROR_INVALID_MAX_TEXT_LENGTH = Error(0x80bc0016);
-constexpr int ERROR_INVALID_TEXT_BUFFER     = Error(0x80bc0017);
-constexpr int ERROR_INVALID_POSX            = Error(0x80bc0018);
-constexpr int ERROR_INVALID_POSY            = Error(0x80bc0019);
-constexpr int ERROR_INVALID_HALIGN          = Error(0x80bc001a);
-constexpr int ERROR_INVALID_VALIGN          = Error(0x80bc001b);
-constexpr int ERROR_INVALID_EXTENDED        = Error(0x80bc001c);
-constexpr int ERROR_INVALID_PARAM           = Error(0x80bc0030);
-constexpr int ERROR_INVALID_ADDRESS         = Error(0x80bc0031);
-constexpr int ERROR_INVALID_RESERVED        = Error(0x80bc0032);
-constexpr int ERROR_INVALID_TITLE           = Error(0x80bc0101);
-constexpr int ERROR_NOT_RUNNING             = Error(0x80bc0105);
-constexpr int ERROR_NOT_FINISHED            = Error(0x80bc0106);
-constexpr int ERROR_NOT_IN_USE              = Error(0x80bc0107);
+constexpr int ERROR_BUSY                    = static_cast<int32_t>(0x80bc0001);
+constexpr int ERROR_INVALID_USER_ID         = static_cast<int32_t>(0x80bc0010);
+constexpr int ERROR_INVALID_TYPE            = static_cast<int32_t>(0x80bc0011);
+constexpr int ERROR_INVALID_LANGUAGES       = static_cast<int32_t>(0x80bc0012);
+constexpr int ERROR_INVALID_ENTER_LABEL     = static_cast<int32_t>(0x80bc0013);
+constexpr int ERROR_INVALID_INPUT_METHOD    = static_cast<int32_t>(0x80bc0014);
+constexpr int ERROR_INVALID_OPTION          = static_cast<int32_t>(0x80bc0015);
+constexpr int ERROR_INVALID_MAX_TEXT_LENGTH = static_cast<int32_t>(0x80bc0016);
+constexpr int ERROR_INVALID_TEXT_BUFFER     = static_cast<int32_t>(0x80bc0017);
+constexpr int ERROR_INVALID_POSX            = static_cast<int32_t>(0x80bc0018);
+constexpr int ERROR_INVALID_POSY            = static_cast<int32_t>(0x80bc0019);
+constexpr int ERROR_INVALID_HALIGN          = static_cast<int32_t>(0x80bc001a);
+constexpr int ERROR_INVALID_VALIGN          = static_cast<int32_t>(0x80bc001b);
+constexpr int ERROR_INVALID_EXTENDED        = static_cast<int32_t>(0x80bc001c);
+constexpr int ERROR_INVALID_PARAM           = static_cast<int32_t>(0x80bc0030);
+constexpr int ERROR_INVALID_ADDRESS         = static_cast<int32_t>(0x80bc0031);
+constexpr int ERROR_INVALID_RESERVED        = static_cast<int32_t>(0x80bc0032);
+constexpr int ERROR_INVALID_TITLE           = static_cast<int32_t>(0x80bc0101);
+constexpr int ERROR_NOT_RUNNING             = static_cast<int32_t>(0x80bc0105);
+constexpr int ERROR_NOT_FINISHED            = static_cast<int32_t>(0x80bc0106);
+constexpr int ERROR_NOT_IN_USE              = static_cast<int32_t>(0x80bc0107);
 
 constexpr uint32_t VALID_OPTIONS           = 0x00007bff;
 constexpr uint32_t VALID_EXTENDED_OPTIONS  = 0x00005fde;

src/libs/libAppContent.cpp

@@ -51,17 +51,17 @@ static bool TryGetTitleIdFromContentId(std::string* title_id) {
 		return false;
 	}
 
-	const uint32_t dash = Common::FindIndex(content_id, '-');
-	if (!Common::IndexValid(content_id, dash)) {
+	const auto dash = content_id.find('-');
+	if (dash == std::string::npos) {
 		return false;
 	}
 
-	const uint32_t underscore = Common::FindIndex(content_id, '_', dash + 1);
-	if (!Common::IndexValid(content_id, underscore) || underscore <= dash + 1) {
+	const auto underscore = content_id.find('_', dash + 1);
+	if (underscore == std::string::npos || underscore <= dash + 1) {
 		return false;
 	}
 
-	*title_id = Common::Mid(content_id, dash + 1, underscore - dash - 1);
+	*title_id = content_id.substr(dash + 1, underscore - dash - 1);
 	return !title_id->empty();
 }
 

src/libs/libAudio.cpp

@@ -1052,7 +1052,7 @@ static int32_t KYTY_SYSV_ABI AudiodecDecode(int32_t handle, AudiodecCtrl* ctrl)
 		ctrl->pPcmItem->uiPcmSize = static_cast<uint32_t>(result.output_written);
 		audiodec_fill_info(ctrl, decoder);
 		auto* info = static_cast<AudiodecM4aacInfo*>(ctrl->pBsiInfo);
-		info->iResult = (result.result & AJM_RESULT_CODEC_ERROR) != 0
+		info->iResult = (result.result & (AJM_RESULT_CODEC_ERROR | AJM_RESULT_TOO_MANY_CHANNELS)) != 0
 		                    ? AUDIODEC_M4AAC_RESULT_DECODE_ERROR
 		                    : (result.result & AJM_RESULT_PARTIAL_INPUT) != 0
 		                          ? AUDIODEC_M4AAC_RESULT_INSUFFICIENT_DATA : 0;

src/libs/libDbgAsan.cpp

@@ -11,6 +11,7 @@
 #include "loader/symbolDatabase.h"
 
 #include <algorithm>
+#include <cstring>
 #include <vector>
 
 namespace Libs {

src/libs/libHmd2.cpp

@@ -1,3 +1,4 @@
+#include "common/assert.h"
 #include "common/emulatorConfig.h"
 #include "graphics/guest_gpu/tile.h"
 #include "libs/agc.h"

src/libs/libKernel.cpp

@@ -24,6 +24,7 @@
 #include <algorithm>
 #include <array>
 #include <atomic>
+#include <bit>
 #include <chrono>
 #include <cstddef>
 #include <cstdio>
@@ -68,10 +69,6 @@ using KernelModule                   = int32_t;
 using get_thread_atexit_count_func_t = KYTY_SYSV_ABI int (*)(KernelModule);
 using thread_atexit_report_func_t    = KYTY_SYSV_ABI void (*)(KernelModule);
 
-static uint32_t sha1_rol(uint32_t value, uint32_t bits) {
-	return (value << bits) | (value >> (32u - bits));
-}
-
 static std::array<uint8_t, 20> sha1_digest(const uint8_t* data, size_t size) {
 	uint32_t h0 = 0x67452301u;
 	uint32_t h1 = 0xefcdab89u;
@@ -99,7 +96,7 @@ static std::array<uint8_t, 20> sha1_digest(const uint8_t* data, size_t size) {
 			       (static_cast<uint32_t>(msg[j + 2]) << 8u) | static_cast<uint32_t>(msg[j + 3]);
 		}
 		for (int i = 16; i < 80; i++) {
-			w[i] = sha1_rol(w[i - 3] ^ w[i - 8] ^ w[i - 14] ^ w[i - 16], 1);
+			w[i] = std::rotl(w[i - 3] ^ w[i - 8] ^ w[i - 14] ^ w[i - 16], 1);
 		}
 
 		uint32_t a = h0;
@@ -125,10 +122,10 @@ static std::array<uint8_t, 20> sha1_digest(const uint8_t* data, size_t size) {
 				k = 0xca62c1d6u;
 			}
 
-			const auto temp = sha1_rol(a, 5) + f + e + k + w[i];
+			const auto temp = std::rotl(a, 5) + f + e + k + w[i];
 			e               = d;
 			d               = c;
-			c               = sha1_rol(b, 30);
+			c               = std::rotl(b, 30);
 			b               = a;
 			a               = temp;
 		}

src/libs/libNet.cpp

@@ -8,22 +8,13 @@
 #include "loader/symbolDatabase.h"
 
 #include <cctype>
+#include <cstddef>
 #include <cstring>
 #include <map>
 #include <mutex>
 #include <string>
 #include <vector>
 
-// NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
-#define NET_CALL(func)                                                                             \
-	[&]() {                                                                                        \
-		auto result = func;                                                                        \
-		if (result < 0) {                                                                          \
-			*GetNetErrorAddr() = result;                                                           \
-		}                                                                                          \
-		return result;                                                                             \
-	}()
-
 namespace Libs {
 
 namespace Network::Net {
@@ -43,8 +34,15 @@ KYTY_SYSV_ABI int* GetNetErrorAddr() {
 	return &g_net_errno;
 }
 
+static int FinishNetCall(int result) {
+	if (result < 0) {
+		g_net_errno = result - Network::NET_ERROR_EPERM + 1;
+	}
+	return result;
+}
+
 static int KYTY_SYSV_ABI NetInit() {
-	return NET_CALL(Net::NetInit());
+	return FinishNetCall(Net::NetInit());
 }
 
 static int PosixToNetError(int error) {
@@ -66,13 +64,7 @@ static int PosixToNetError(int error) {
 }
 
 static int FinishSocketCall(int result) {
-	if (result >= 0) {
-		return result;
-	}
-
-	const int net_error = PosixToNetError(*Posix::GetErrorAddr());
-	*GetNetErrorAddr()  = net_error;
-	return net_error;
+	return FinishNetCall(result < 0 ? PosixToNetError(*Posix::GetErrorAddr()) : result);
 }
 
 int KYTY_SYSV_ABI NetSocket(const char* name, int family, int type, int protocol) {
@@ -102,48 +94,48 @@ int KYTY_SYSV_ABI NetGetsockname(int s, void* addr, uint32_t* addrlen) {
 }
 
 int KYTY_SYSV_ABI NetPoolCreate(const char* name, int size, int flags) {
-	return NET_CALL(Net::NetPoolCreate(name, size, flags));
+	return FinishNetCall(Net::NetPoolCreate(name, size, flags));
 }
 
 int KYTY_SYSV_ABI NetPoolDestroy(int memid) {
-	return NET_CALL(Net::NetPoolDestroy(memid));
+	return FinishNetCall(Net::NetPoolDestroy(memid));
 }
 
 int KYTY_SYSV_ABI NetResolverCreate(const char* name, int memid, int flags) {
-	return NET_CALL(Net::NetResolverCreate(name, memid, flags));
+	return FinishNetCall(Net::NetResolverCreate(name, memid, flags));
 }
 
 int KYTY_SYSV_ABI NetResolverDestroy(int rid) {
-	return NET_CALL(Net::NetResolverDestroy(rid));
+	return FinishNetCall(Net::NetResolverDestroy(rid));
 }
 
 int KYTY_SYSV_ABI NetResolverStartNtoa(int rid, const char* hostname, void* addr, int timeout,
                                        int retry, int flags) {
-	return NET_CALL(Net::NetResolverStartNtoa(rid, hostname, addr, timeout, retry, flags));
+	return FinishNetCall(Net::NetResolverStartNtoa(rid, hostname, addr, timeout, retry, flags));
 }
 
 int KYTY_SYSV_ABI NetInetPton(int af, const char* src, void* dst) {
-	return NET_CALL(Net::NetInetPton(af, src, dst));
+	return FinishNetCall(Net::NetInetPton(af, src, dst));
 }
 
 const char* KYTY_SYSV_ABI NetInetNtop(int af, const void* src, char* dst, uint32_t size) {
 	const char* result = Net::NetInetNtop(af, src, dst, size);
 	if (result == nullptr) {
-		*GetNetErrorAddr() = PosixToNetError(*Posix::GetErrorAddr());
+		FinishNetCall(PosixToNetError(*Posix::GetErrorAddr()));
 	}
 	return result;
 }
 
 int KYTY_SYSV_ABI NetEtherNtostr(const Net::NetEtherAddr* n, char* str, size_t len) {
-	return NET_CALL(Net::NetEtherNtostr(n, str, len));
+	return FinishNetCall(Net::NetEtherNtostr(n, str, len));
 }
 
 int KYTY_SYSV_ABI NetGetMacAddress(Net::NetEtherAddr* addr, int flags) {
-	return NET_CALL(Net::NetGetMacAddress(addr, flags));
+	return FinishNetCall(Net::NetGetMacAddress(addr, flags));
 }
 
 int KYTY_SYSV_ABI NetGetSockInfo(int s, void* info, int n, int flags) {
-	return NET_CALL(Net::NetGetSockInfo(s, info, n, flags));
+	return FinishNetCall(Net::NetGetSockInfo(s, info, n, flags));
 }
 
 int KYTY_SYSV_ABI NetEpollCreate(const char* name, int flags) {
@@ -163,7 +155,7 @@ int KYTY_SYSV_ABI NetEpollDestroy(int eid) {
 }
 
 int KYTY_SYSV_ABI NetSocketClose(int s) {
-	return NET_CALL(Net::SocketClose(s));
+	return FinishNetCall(Net::SocketClose(s));
 }
 
 int KYTY_SYSV_ABI NetSetsockopt(int s, int level, int optname, const void* optval,
@@ -1465,7 +1457,7 @@ LIB_DEFINE(InitNet_1_NpManager) {
 	LIB_FUNC("Ec63y59l9tw", NpManager::NpSetNpTitleId);
 	LIB_FUNC("A2CQ3kgSopQ", NpManager::NpSetContentRestriction);
 	LIB_FUNC("VfRSmPmj8Q8", NpManager::NpRegisterStateCallback);
-	LIB_FUNC("qQJfO8HAiaY", NpManager::NpRegisterStateCallback);
+	LIB_FUNC("qQJfO8HAiaY", NpManager::NpRegisterStateCallbackA);
 	LIB_FUNC("M3wFXbYQtAA", NpManager::NpUnregisterStateCallback);
 	LIB_FUNC("uFJpaKNBAj4", NpManager::NpRegisterGamePresenceCallback);
 	LIB_FUNC("GImICnh+boA", NpManager::NpRegisterPlusEventCallback);
@@ -3118,16 +3110,6 @@ static std::map<int64_t, NpWebApi2Request>& NpWebApi2Requests() {
 	return requests;
 }
 
-[[maybe_unused]] static std::string NpWebApi2MakeResponse(const NpWebApi2Request& request) {
-	if (request.api_group.find("sessionManager") != std::string::npos ||
-	    request.path.find("sessions") != std::string::npos ||
-	    request.path.find("Sessions") != std::string::npos) {
-		return R"({"gameSessions":[],"playerSessions":[]})";
-	}
-
-	return "{}";
-}
-
 static int KYTY_SYSV_ABI NpWebApi2Initialize(int lib_http_ctx_id, size_t pool_size) {
 	PRINT_NAME();
 
@@ -3224,12 +3206,10 @@ NpWebApi2SendRequest(int64_t request_id, const void* data, size_t data_size,
 		return NP_WEBAPI2_ERROR_REQUEST_NOT_FOUND;
 	}
 
-	// request->second.response    = NpWebApi2MakeResponse(request->second);
 	request->second.response.clear();
 	request->second.read_offset = 0;
 
 	if (response_info_option != nullptr) {
-		// response_info_option->http_status        = 200;
 		response_info_option->http_status        = 0;
 		response_info_option->response_data_size = request->second.response.size();
 		if (response_info_option->error_object != nullptr &&
@@ -3238,7 +3218,6 @@ NpWebApi2SendRequest(int64_t request_id, const void* data, size_t data_size,
 		}
 	}
 
-	// return 0;
 	return NP_WEBAPI2_ERROR_NOT_SIGNED_IN;
 }
 
@@ -3743,6 +3722,21 @@ namespace LibGameLiveStreaming {
 
 LIB_VERSION("GameLiveStreaming", 1, "GameLiveStreaming", 1, 1);
 
+constexpr int GAME_LIVE_STREAMING_ERROR_INVALID_PARAM = -2136997886;
+
+struct GameLiveStreamingStatus2 {
+	int32_t  user_id;
+	bool     is_on_air;
+	uint32_t spectator_count;
+	uint32_t text_message_count;
+	uint32_t command_message_count;
+	uint8_t  reserved[52];
+};
+
+static_assert(sizeof(GameLiveStreamingStatus2) == 72);
+static_assert(offsetof(GameLiveStreamingStatus2, is_on_air) == 4);
+static_assert(offsetof(GameLiveStreamingStatus2, spectator_count) == 8);
+
 static int KYTY_SYSV_ABI GameLiveStreamingInitialize(size_t heap_size) {
 	PRINT_NAME();
 
@@ -3757,9 +3751,34 @@ static int KYTY_SYSV_ABI GameLiveStreamingTerminate() {
 	return 0;
 }
 
+static int KYTY_SYSV_ABI GameLiveStreamingGetCurrentStatus2(GameLiveStreamingStatus2* status) {
+	PRINT_NAME();
+
+	if (status == nullptr) {
+		return GAME_LIVE_STREAMING_ERROR_INVALID_PARAM;
+	}
+
+	std::memset(status, 0, sizeof(*status));
+	status->user_id = -1;
+	return OK;
+}
+
+static int KYTY_SYSV_ABI GameLiveStreamingGetSocialFeedbackMessagesCount(int type, uint32_t* count) {
+	PRINT_NAME();
+
+	if (type < 1 || type > 3 || count == nullptr) {
+		return GAME_LIVE_STREAMING_ERROR_INVALID_PARAM;
+	}
+
+	*count = 0;
+	return OK;
+}
+
 LIB_DEFINE(InitPlatform_1_GameLiveStreaming) {
 	LIB_FUNC("kvYEw2lBndk", LibGameLiveStreaming::GameLiveStreamingInitialize);
 	LIB_FUNC("9yK6Fk8mKOQ", LibGameLiveStreaming::GameLiveStreamingTerminate);
+	LIB_FUNC("lK8dLBNp9OE", LibGameLiveStreaming::GameLiveStreamingGetCurrentStatus2);
+	LIB_FUNC("yeQKjHETi40", LibGameLiveStreaming::GameLiveStreamingGetSocialFeedbackMessagesCount);
 }
 
 } // namespace LibGameLiveStreaming