Merge branch 'master' into vsh_fix

3rdparty/CMakeLists.txt

@@ -44,7 +44,7 @@ add_subdirectory(libpng EXCLUDE_FROM_ALL)
 
 # pugixml
 if (USE_SYSTEM_PUGIXML)
-	pkg_check_modules(PUGIXML REQUIRED IMPORTED_TARGET pugixml>=1.11)
+	pkg_check_modules(PUGIXML REQUIRED IMPORTED_TARGET pugixml>=1.15)
 	add_library(pugixml INTERFACE)
 	target_link_libraries(pugixml INTERFACE PkgConfig::PUGIXML)
 else()

3rdparty/libpng/libpng

@@ -1 +1 @@
-Subproject commit f5e92d76973a7a53f517579bc95d61483bf108c0
+Subproject commit 51f5bd68b9b806d2c92b4318164d28b49357da31

3rdparty/pugixml

@@ -1 +1 @@
-Subproject commit db78afc2b7d8f043b4bc6b185635d949ea2ed2a8
+Subproject commit ee86beb30e4973f5feffe3ce63bfa4fbadf72f38

3rdparty/wolfssl/CMakeLists.txt

@@ -22,6 +22,5 @@ else()
 
 	add_subdirectory(wolfssl EXCLUDE_FROM_ALL)
 
-	target_compile_definitions(wolfssl PUBLIC WOLFSSL_DES_ECB HAVE_WRITE_DUP)
+	target_compile_definitions(wolfssl PUBLIC WOLFSSL_DES_ECB HAVE_WRITE_DUP FP_MAX_BITS=8192 WOLFSSL_NO_OPTIONS_H)
-	target_compile_definitions(wolfssl PUBLIC FP_MAX_BITS=8192)
 endif()

Utilities/File.cpp

@@ -21,7 +21,7 @@ using namespace std::literals::string_literals;
 #include <cwchar>
 #include <Windows.h>
 
-static std::unique_ptr<wchar_t[]> to_wchar(const std::string& source)
+static std::unique_ptr<wchar_t[]> to_wchar(std::string_view source)
 {
 	// String size + null terminator
 	const usz buf_size = source.size() + 1;
@@ -44,7 +44,7 @@ static std::unique_ptr<wchar_t[]> to_wchar(const std::string& source)
 		std::memcpy(buffer.get() + 32768 + 4, L"UNC\\", 4 * sizeof(wchar_t));
 	}
 
-	ensure(MultiByteToWideChar(CP_UTF8, 0, source.c_str(), size, buffer.get() + 32768 + (unc ? 8 : 4), size)); // "to_wchar"
+	ensure(MultiByteToWideChar(CP_UTF8, 0, source.data(), size, buffer.get() + 32768 + (unc ? 8 : 4), size)); // "to_wchar"
 
 	// Canonicalize wide path (replace '/', ".", "..", \\ repetitions, etc)
 	ensure(GetFullPathNameW(buffer.get() + 32768, 32768, buffer.get(), nullptr) - 1 < 32768 - 1); // "to_wchar"
@@ -2021,7 +2021,7 @@ std::string fs::get_executable_dir()
 	return s_exe_dir;
 }
 
-const std::string& fs::get_config_dir()
+const std::string& fs::get_config_dir([[maybe_unused]] bool get_config_subdirectory)
 {
 	// Use magic static
 	static const std::string s_dir = []
@@ -2103,6 +2103,14 @@ const std::string& fs::get_config_dir()
 		return dir;
 	}();
 
+#ifdef _WIN32
+	if (get_config_subdirectory)
+	{
+		static const std::string subdir = s_dir + "config/";
+		return subdir;
+	}
+#endif
+
 	return s_dir;
 }
 
@@ -2144,6 +2152,16 @@ const std::string& fs::get_cache_dir()
 	return s_dir;
 }
 
+const std::string& fs::get_log_dir()
+{
+#ifdef _WIN32
+	static const std::string s_dir = fs::get_config_dir() + "log/";
+	return s_dir;
+#else
+	return fs::get_cache_dir();
+#endif
+}
+
 const std::string& fs::get_temp_dir()
 {
 	static const std::string s_dir = []

Utilities/File.h

@@ -599,12 +599,15 @@ namespace fs
 	// Get executable containing directory
 	std::string get_executable_dir();
 
-	// Get configuration directory
+	// Get configuration directory. Set get_config_subdirectory to true to get the nested config dir on windows.
-	const std::string& get_config_dir();
+	const std::string& get_config_dir(bool get_config_subdirectory = false);
 
 	// Get common cache directory
 	const std::string& get_cache_dir();
 
+	// Get common log directory
+	const std::string& get_log_dir();
+
 	// Temporary directory
 	const std::string& get_temp_dir();
 

Utilities/Thread.h

@@ -33,7 +33,8 @@ enum class thread_state : u32
 	aborting = 1, // The thread has been joined in the destructor or explicitly aborted
 	errored = 2, // Set after the emergency_exit call
 	finished = 3,  // Final state, always set at the end of thread execution
-	mask = 3
+	mask = 3,
+	destroying_context = 7, // Special value assigned to destroy data explicitly before the destructor
 };
 
 template <class Context>
@@ -702,14 +703,17 @@ public:
 			thread::m_sync.notify_all();
 		}
 
-		if (s == thread_state::finished)
+		if (s == thread_state::finished || s == thread_state::destroying_context)
 		{
 			// This participates in emulation stopping, use destruction-alike semantics
 			thread::join(true);
+		}
 
+		if (s == thread_state::destroying_context)
+		{
 			if constexpr (std::is_assignable_v<Context&, thread_state>)
 			{
-				static_cast<Context&>(*this) = thread_state::finished;
+				static_cast<Context&>(*this) = thread_state::destroying_context;
 			}
 		}
 

Utilities/bin_patch.cpp

@@ -125,19 +125,15 @@ patch_engine::patch_engine()
 
 std::string patch_engine::get_patch_config_path()
 {
-#ifdef _WIN32
+	const std::string config_dir = fs::get_config_dir(true);
-	const std::string config_dir = fs::get_config_dir() + "config/";
 	const std::string patch_path = config_dir + "patch_config.yml";
-
+#ifdef _WIN32
 	if (!fs::create_path(config_dir))
 	{
 		patch_log.error("Could not create path: %s (%s)", patch_path, fs::g_tls_error);
 	}
-
-	return patch_path;
-#else
-	return fs::get_config_dir() + "patch_config.yml";
 #endif
+	return patch_path;
 }
 
 std::string patch_engine::get_patches_path()
@@ -1449,6 +1445,8 @@ static usz apply_modification(std::vector<u32>& applied, patch_engine::patch_inf
 
 void patch_engine::apply(std::vector<u32>& applied_total, const std::string& name, std::function<u8*(u32, u32)> mem_translate, u32 filesz, u32 min_addr)
 {
+	// applied_total may be non-empty, do not clear it
+
 	if (!m_map.contains(name))
 	{
 		return;
@@ -1597,6 +1595,9 @@ void patch_engine::apply(std::vector<u32>& applied_total, const std::string& nam
 			}
 		}
 	}
+
+	// Ensure consistent order
+	std::sort(applied_total.begin(), applied_total.end());
 }
 
 void patch_engine::unload(const std::string& name)

Utilities/cheat_info.h

@@ -14,6 +14,7 @@ enum class cheat_type : u8
 	signed_16_cheat,
 	signed_32_cheat,
 	signed_64_cheat,
+	float_32_cheat,
 	max
 };
 

Utilities/rXml.cpp

@@ -49,12 +49,11 @@ std::string rXmlNode::GetName()
 	return {};
 }
 
-std::string rXmlNode::GetAttribute(const std::string& name)
+std::string rXmlNode::GetAttribute(std::string_view name)
 {
 	if (handle)
 	{
-		const auto pred = [&name](const pugi::xml_attribute& attr) { return (name == attr.name()); };
+		if (const pugi::xml_attribute attr = handle.attribute(name))
-		if (const pugi::xml_attribute attr = handle.find_attribute(pred))
 		{
 			if (const pugi::char_t* value = attr.value())
 			{
@@ -86,7 +85,7 @@ rXmlDocument::rXmlDocument()
 {
 }
 
-pugi::xml_parse_result rXmlDocument::Read(const std::string& data)
+pugi::xml_parse_result rXmlDocument::Read(std::string_view data)
 {
 	if (handle)
 	{

Utilities/rXml.h

@@ -23,7 +23,7 @@ struct rXmlNode
 	std::shared_ptr<rXmlNode> GetChildren();
 	std::shared_ptr<rXmlNode> GetNext();
 	std::string GetName();
-	std::string GetAttribute(const std::string& name);
+	std::string GetAttribute(std::string_view name);
 	std::string GetNodeContent();
 
 	pugi::xml_node handle{};
@@ -34,7 +34,7 @@ struct rXmlDocument
 	rXmlDocument();
 	rXmlDocument(const rXmlDocument& other) = delete;
 	rXmlDocument &operator=(const rXmlDocument& other) = delete;
-	pugi::xml_parse_result Read(const std::string& data);
+	pugi::xml_parse_result Read(std::string_view data);
 	virtual std::shared_ptr<rXmlNode> GetRoot();
 
 	pugi::xml_document handle{};

rpcs3/CMakeLists.txt

@@ -164,7 +164,7 @@ elseif(WIN32)
         COMMAND ${CMAKE_COMMAND} -E copy_directory ${CMAKE_SOURCE_DIR}/bin/GuiConfigs $<TARGET_FILE_DIR:rpcs3>/GuiConfigs
         COMMAND ${CMAKE_COMMAND} -E copy_directory ${CMAKE_SOURCE_DIR}/bin/git $<TARGET_FILE_DIR:rpcs3>/git
         COMMAND "${WINDEPLOYQT_EXECUTABLE}" --no-compiler-runtime --no-opengl-sw --no-patchqt
-            --no-translations --no-quick --no-system-d3d-compiler --no-quick-import
+            --no-translations --no-system-d3d-compiler --no-quick-import
             --plugindir "$<IF:$<CXX_COMPILER_ID:MSVC>,$<TARGET_FILE_DIR:rpcs3>/plugins,$<TARGET_FILE_DIR:rpcs3>/share/qt6/plugins>"
             --verbose 0 "$<TARGET_FILE:rpcs3>")
 endif()

rpcs3/Emu/CMakeLists.txt

@@ -172,6 +172,7 @@ target_link_libraries(rpcs3_emu
 
 # Cell
 target_sources(rpcs3_emu PRIVATE
+    Cell/ErrorCodes.cpp
     Cell/MFC.cpp
     Cell/PPUAnalyser.cpp
     Cell/PPUDisAsm.cpp
@@ -557,6 +558,7 @@ target_sources(rpcs3_emu PRIVATE
     RSX/Program/CgBinaryFragmentProgram.cpp
     RSX/Program/CgBinaryVertexProgram.cpp
     RSX/Program/FragmentProgramDecompiler.cpp
+    RSX/Program/FragmentProgramRegister.cpp
     RSX/Program/GLSLCommon.cpp
     RSX/Program/ProgramStateCache.cpp
     RSX/Program/program_util.cpp
@@ -593,6 +595,7 @@ if(TARGET 3rdparty_vulkan)
         RSX/VK/VKAsyncScheduler.cpp
         RSX/VK/VKCommandStream.cpp
         RSX/VK/VKCommonDecompiler.cpp
+        RSX/VK/VKCommonPipelineLayout.cpp
         RSX/VK/VKCompute.cpp
         RSX/VK/VKDMA.cpp
         RSX/VK/VKDraw.cpp

rpcs3/Emu/Cell/ErrorCodes.cpp

@@ -0,0 +1,91 @@
+#include "stdafx.h"
+#include "ErrorCodes.h"
+#include "PPUThread.h"
+#include "SPUThread.h"
+
+LOG_CHANNEL(sys_log, "SYS");
+
+bool g_log_all_errors = false;
+
+s32 error_code::error_report(s32 result, const logs::message* channel, const char* fmt, const fmt_type_info* sup, const u64* args)
+{
+	static thread_local std::string g_tls_error_str;
+	static thread_local std::unordered_map<std::string, usz> g_tls_error_stats;
+
+	if (!channel)
+	{
+		channel = &sys_log.error;
+	}
+
+	if (!sup && !args)
+	{
+		if (!fmt)
+		{
+			// Report and clean error state
+			for (auto&& pair : g_tls_error_stats)
+			{
+				if (pair.second > 3)
+				{
+					channel->operator()("Stat: %s [x%u]", pair.first, pair.second);
+				}
+			}
+
+			g_tls_error_stats.clear();
+			return 0;
+		}
+	}
+
+	ensure(fmt);
+
+	const char* func = "Unknown function";
+
+	if (auto ppu = get_current_cpu_thread<ppu_thread>())
+	{
+		if (auto current = ppu->current_function)
+		{
+			func = current;
+		}
+	}
+	else if (auto spu = get_current_cpu_thread<spu_thread>())
+	{
+		if (auto current = spu->current_func; current && spu->start_time)
+		{
+			func = current;
+		}
+	}
+
+	// Format log message (use preallocated buffer)
+	g_tls_error_str.clear();
+
+	fmt::append(g_tls_error_str, "'%s' failed with 0x%08x", func, result);
+
+	// Add spacer between error and fmt if necessary
+	if (fmt[0] != ' ')
+		g_tls_error_str += " : ";
+
+	fmt::raw_append(g_tls_error_str, fmt, sup, args);
+
+	// Update stats and check log threshold
+
+	if (g_log_all_errors) [[unlikely]]
+	{
+		if (!g_tls_error_stats.empty())
+		{
+			// Report and clean error state
+			error_report(0, nullptr, nullptr, nullptr, nullptr);
+		}
+
+		channel->operator()("%s", g_tls_error_str);
+	}
+	else
+	{
+		const auto stat = ++g_tls_error_stats[g_tls_error_str];
+
+		if (stat <= 3)
+		{
+			channel->operator()("%s [%u]", g_tls_error_str, stat);
+		}
+	}
+
+	return result;
+}

rpcs3/Emu/Cell/Modules/cellAtracXdec.cpp

@@ -270,35 +270,75 @@ void AtracXdecContext::exec(ppu_thread& ppu)
 		decoder.init_avcodec();
 	}
 
+	switch (savestate)
+	{
+	case atracxdec_state::initial: break;
+	case atracxdec_state::waiting_for_cmd: goto label1_wait_for_cmd_state;
+	case atracxdec_state::checking_run_thread_1: goto label2_check_run_thread_1_state;
+	case atracxdec_state::executing_cmd: goto label3_execute_cmd_state;
+	case atracxdec_state::waiting_for_output: goto label4_wait_for_output_state;
+	case atracxdec_state::checking_run_thread_2: goto label5_check_run_thread_2_state;
+	case atracxdec_state::decoding: goto label6_decode_state;
+	}
+
 	for (;;cmd_counter++)
 	{
 		cellAtracXdec.trace("Command counter: %llu, waiting for next command...", cmd_counter);
 
-		if (!skip_getting_command)
+		for (;;)
 		{
-			lv2_obj::sleep(ppu);
+			savestate = atracxdec_state::initial;
-			std::lock_guard lock{queue_mutex};
 
-			while (cmd_queue.empty() && !ppu.is_stopped())
+			ensure(sys_mutex_lock(ppu, queue_mutex, 0) == CELL_OK);
+
+			if (ppu.state & cpu_flag::again)
 			{
-				lv2_obj::sleep(ppu);
+				return;
-				queue_not_empty.wait(queue_mutex, 20000);
 			}
 
-			if (ppu.is_stopped())
+			if (!cmd_queue.empty())
+			{
+				break;
+			}
+
+			savestate = atracxdec_state::waiting_for_cmd;
+			label1_wait_for_cmd_state:
+
+			ensure(sys_cond_wait(ppu, queue_not_empty, 0) == CELL_OK);
+
+			if (ppu.state & cpu_flag::again)
 			{
-				ppu.state += cpu_flag::again;
 				return;
 			}
 
+			ensure(sys_mutex_unlock(ppu, queue_mutex) == CELL_OK);
+		}
+
 		cmd_queue.pop(cmd);
 
-			if (!run_thread)
+		ensure(sys_mutex_unlock(ppu, queue_mutex) == CELL_OK);
+
+		savestate = atracxdec_state::checking_run_thread_1;
+		label2_check_run_thread_1_state:
+
+		ensure(sys_mutex_lock(ppu, run_thread_mutex, 0) == CELL_OK);
+
+		if (ppu.state & cpu_flag::again)
 		{
 			return;
 		}
+
+		if (!run_thread)
+		{
+			ensure(sys_mutex_unlock(ppu, run_thread_mutex) == CELL_OK);
+			return;
 		}
 
+		ensure(sys_mutex_unlock(ppu, run_thread_mutex) == CELL_OK);
+
+		savestate = atracxdec_state::executing_cmd;
+		label3_execute_cmd_state:
+
 		cellAtracXdec.trace("Command type: %d", static_cast<u32>(cmd.type.get()));
 
 		switch (cmd.type)
@@ -327,8 +367,6 @@ void AtracXdecContext::exec(ppu_thread& ppu)
 		}
 		case AtracXdecCmdType::end_seq:
 		{
-			skip_getting_command = true;
-
 			// Block savestate creation during callbacks
 			std::unique_lock savestate_lock{g_fxo->get<hle_locks_t>(), std::try_to_lock};
 
@@ -338,41 +376,59 @@ void AtracXdecContext::exec(ppu_thread& ppu)
 				return;
 			}
 
-			skip_getting_command = false;
-
 			// Doesn't do anything else
 			notify_seq_done.cbFunc(ppu, notify_seq_done.cbArg);
 			break;
 		}
 		case AtracXdecCmdType::decode_au:
 		{
-			skip_getting_command = true;
-
 			ensure(!!cmd.au_start_addr); // Not checked on LLE
 
 			cellAtracXdec.trace("Waiting for output to be consumed...");
 
-			lv2_obj::sleep(ppu);
+			ensure(sys_mutex_lock(ppu, output_mutex, 0) == CELL_OK);
-			std::unique_lock output_mutex_lock{output_mutex};
 
-			while (output_locked && !ppu.is_stopped())
+			if (ppu.state & cpu_flag::again)
 			{
-				lv2_obj::sleep(ppu);
+				return;
-				output_consumed.wait(output_mutex, 20000);
 			}
 
-			if (ppu.is_stopped())
+			while (output_locked)
+			{
+				savestate = atracxdec_state::waiting_for_output;
+				label4_wait_for_output_state:
+
+				ensure(sys_cond_wait(ppu, output_consumed, 0) == CELL_OK);
+
+				if (ppu.state & cpu_flag::again)
+				{
+					return;
+				}
+			}
+
+			cellAtracXdec.trace("Output consumed");
+
+			savestate = atracxdec_state::checking_run_thread_2;
+			label5_check_run_thread_2_state:
+
+			ensure(sys_mutex_lock(ppu, run_thread_mutex, 0) == CELL_OK);
+
+			if (ppu.state & cpu_flag::again)
 			{
-				ppu.state += cpu_flag::again;
 				return;
 			}
 
 			if (!run_thread)
 			{
+				ensure(sys_mutex_unlock(ppu, run_thread_mutex) == CELL_OK);
+				ensure(sys_mutex_unlock(ppu, output_mutex) == CELL_OK);
 				return;
 			}
 
-			cellAtracXdec.trace("Output consumed");
+			ensure(sys_mutex_unlock(ppu, run_thread_mutex) == CELL_OK);
+
+			savestate = atracxdec_state::decoding;
+			label6_decode_state:
 
 			u32 error = CELL_OK;
 
@@ -578,14 +634,12 @@ void AtracXdecContext::exec(ppu_thread& ppu)
 				return;
 			}
 
-			skip_getting_command = false;
-
 			// au_done and pcm_out callbacks are always called after a decode command, even if an error occurred
 			// The output always has to be consumed as well
 			notify_au_done.cbFunc(ppu, cmd.pcm_handle, notify_au_done.cbArg);
 
 			output_locked = true;
-			output_mutex_lock.unlock();
+			ensure(sys_mutex_unlock(ppu, output_mutex) == CELL_OK);
 
 			const u32 output_size = decoded_samples_num * (decoder.bw_pcm & 0x7fu) * decoder.nch_out;
 
@@ -614,29 +668,46 @@ void AtracXdecContext::exec(ppu_thread& ppu)
 template <AtracXdecCmdType type>
 error_code AtracXdecContext::send_command(ppu_thread& ppu, auto&&... args)
 {
-	ppu.state += cpu_flag::wait;
+	auto& savestate = *ppu.optional_savestate_state;
+	const bool signal = savestate.try_read<bool>().second;
+	savestate.clear();
 
+	if (!signal)
 	{
-		std::lock_guard lock{queue_mutex};
+		ensure(sys_mutex_lock(ppu, queue_mutex, 0) == CELL_OK);
+
+		if (ppu.state & cpu_flag::again)
+		{
+			return {};
+		}
 
 		if constexpr (type == AtracXdecCmdType::close)
 		{
 			// Close command is only sent if the queue is empty on LLE
 			if (!cmd_queue.empty())
 			{
+				ensure(sys_mutex_unlock(ppu, queue_mutex) == CELL_OK);
 				return {};
 			}
 		}
 
 		if (cmd_queue.full())
 		{
+			ensure(sys_mutex_unlock(ppu, queue_mutex) == CELL_OK);
 			return CELL_ADEC_ERROR_ATX_BUSY;
 		}
 
 		cmd_queue.emplace(std::forward<AtracXdecCmdType>(type), std::forward<decltype(args)>(args)...);
+
+		ensure(sys_mutex_unlock(ppu, queue_mutex) == CELL_OK);
 	}
 
-	queue_not_empty.notify_one();
+	ensure(sys_cond_signal(ppu, queue_not_empty) == CELL_OK);
+
+	if (ppu.state & cpu_flag::again)
+	{
+		savestate(true);
+	}
 
 	return CELL_OK;
 }
@@ -699,6 +770,29 @@ error_code _CellAdecCoreOpOpenExt_atracx(ppu_thread& ppu, vm::ptr<AtracXdecConte
 	write_to_ptr(handle.get_ptr(), AtracXdecContext(notifyAuDone, notifyAuDoneArg, notifyPcmOut, notifyPcmOutArg, notifyError, notifyErrorArg, notifySeqDone, notifySeqDoneArg,
 		vm::bptr<u8>::make(handle.addr() + utils::align(static_cast<u32>(sizeof(AtracXdecContext)), 0x80) + ATXDEC_SPURS_STRUCTS_SIZE)));
 
+	const vm::var<sys_mutex_attribute_t> mutex_attr{{ SYS_SYNC_PRIORITY, SYS_SYNC_NOT_RECURSIVE, SYS_SYNC_NOT_PROCESS_SHARED, SYS_SYNC_NOT_ADAPTIVE, 0, 0, 0, { "_atd001"_u64 } }};
+	const vm::var<sys_cond_attribute_t> cond_attr{{ SYS_SYNC_NOT_PROCESS_SHARED, 0, 0, { "_atd002"_u64 } }};
+
+	ensure(sys_mutex_create(ppu, handle.ptr(&AtracXdecContext::queue_mutex), mutex_attr) == CELL_OK);
+	ensure(sys_cond_create(ppu, handle.ptr(&AtracXdecContext::queue_not_empty), handle->queue_mutex, cond_attr) == CELL_OK);
+
+	mutex_attr->name_u64 = "_atd003"_u64;
+	cond_attr->name_u64 = "_atd004"_u64;
+
+	ensure(sys_mutex_create(ppu, handle.ptr(&AtracXdecContext::run_thread_mutex), mutex_attr) == CELL_OK);
+	ensure(sys_cond_create(ppu, handle.ptr(&AtracXdecContext::run_thread_cond), handle->run_thread_mutex, cond_attr) == CELL_OK);
+
+	mutex_attr->name_u64 = "_atd005"_u64;
+	cond_attr->name_u64 = "_atd006"_u64;
+
+	ensure(sys_mutex_create(ppu, handle.ptr(&AtracXdecContext::output_mutex), mutex_attr) == CELL_OK);
+	ensure(sys_cond_create(ppu, handle.ptr(&AtracXdecContext::output_consumed), handle->output_mutex, cond_attr) == CELL_OK);
+
+	ensure(sys_mutex_lock(ppu, handle->output_mutex, 0) == CELL_OK);
+	handle->output_locked = false;
+	ensure(sys_cond_signal(ppu, handle->output_consumed) == CELL_OK);
+	ensure(sys_mutex_unlock(ppu, handle->output_mutex) == CELL_OK);
+
 	const vm::var<char[]> _name = vm::make_str("HLE ATRAC3plus decoder");
 	const auto entry = g_fxo->get<ppu_function_manager>().func_addr(FIND_FUNC(atracXdecEntry));
 	ppu_execute<&sys_ppu_thread_create>(ppu, handle.ptr(&AtracXdecContext::thread_id), entry, handle.addr(), +res->ppuThreadPriority, +res->ppuThreadStackSize, SYS_PPU_THREAD_CREATE_JOINABLE, +_name);
@@ -725,29 +819,32 @@ error_code _CellAdecCoreOpClose_atracx(ppu_thread& ppu, vm::ptr<AtracXdecContext
 		return {};
 	}
 
-	ppu.state += cpu_flag::wait;
-
 	cellAtracXdec.notice("_CellAdecCoreOpClose_atracx(handle=*0x%x)", handle);
 
 	ensure(!!handle); // Not checked on LLE
 
+	ensure(sys_mutex_lock(ppu, handle->run_thread_mutex, 0) == CELL_OK);
 	handle->run_thread = false;
+	ensure(sys_mutex_unlock(ppu, handle->run_thread_mutex) == CELL_OK);
+
 	handle->send_command<AtracXdecCmdType::close>(ppu);
 
-	{
+	ensure(sys_mutex_lock(ppu, handle->output_mutex, 0) == CELL_OK);
-		std::lock_guard lock{handle->output_mutex};
 	handle->output_locked = false;
-	}
+	ensure(sys_mutex_unlock(ppu, handle->output_mutex) == CELL_OK);
+	ensure(sys_cond_signal(ppu, handle->output_consumed) == CELL_OK);
 
-	handle->output_consumed.notify_one();
+	vm::var<u64> thread_ret;
+	ensure(sys_ppu_thread_join(ppu, static_cast<u32>(handle->thread_id), +thread_ret) == CELL_OK);
 
-	if (vm::var<u64> ret; sys_ppu_thread_join(ppu, static_cast<u32>(handle->thread_id), +ret) != CELL_OK)
+	error_code ret = sys_cond_destroy(ppu, handle->queue_not_empty);
-	{
+	ret = ret ? ret : sys_cond_destroy(ppu, handle->run_thread_cond);
-		// Other thread already closed the decoder
+	ret = ret ? ret : sys_cond_destroy(ppu, handle->output_consumed);
-		return CELL_ADEC_ERROR_FATAL;
+	ret = ret ? ret : sys_mutex_destroy(ppu, handle->queue_mutex);
-	}
+	ret = ret ? ret : sys_mutex_destroy(ppu, handle->run_thread_mutex);
+	ret = ret ? ret : sys_mutex_destroy(ppu, handle->output_mutex);
 
-	return CELL_OK;
+	return ret != CELL_OK ? static_cast<error_code>(CELL_ADEC_ERROR_FATAL) : CELL_OK;
 }
 
 error_code _CellAdecCoreOpStartSeq_atracx(ppu_thread& ppu, vm::ptr<AtracXdecContext> handle, vm::cptr<CellAdecParamAtracX> atracxParam)
@@ -808,15 +905,35 @@ error_code _CellAdecCoreOpRealign_atracx(vm::ptr<AtracXdecContext> handle, vm::p
 
 error_code _CellAdecCoreOpReleasePcm_atracx(ppu_thread& ppu, vm::ptr<AtracXdecContext> handle, s32 pcmHandle, vm::cptr<void> outBuffer)
 {
-	ppu.state += cpu_flag::wait;
-
 	cellAtracXdec.trace("_CellAdecCoreOpReleasePcm_atracx(handle=*0x%x, pcmHandle=%d, outBuffer=*0x%x)", handle, pcmHandle, outBuffer);
 
 	ensure(!!handle); // Not checked on LLE
 
-	std::lock_guard lock{handle->output_mutex};
+	auto& savestate = *ppu.optional_savestate_state;
+	const bool signal = savestate.try_read<bool>().second;
+	savestate.clear();
+
+	if (!signal)
+	{
+		ensure(sys_mutex_lock(ppu, handle->output_mutex, 0) == CELL_OK);
+
+		if (ppu.state & cpu_flag::again)
+		{
+			return {};
+		}
+
 		handle->output_locked = false;
-	handle->output_consumed.notify_one();
+	}
+
+	ensure(sys_cond_signal(ppu, handle->output_consumed) == CELL_OK);
+
+	if (ppu.state & cpu_flag::again)
+	{
+		savestate(true);
+		return {};
+	}
+
+	ensure(sys_mutex_unlock(ppu, handle->output_mutex) == CELL_OK);
 
 	return CELL_OK;
 }

rpcs3/Emu/Cell/Modules/cellAtracXdec.h

@@ -20,7 +20,6 @@ constexpr int averror_eof = AVERROR_EOF; // Workaround for old-style-cast error
 #pragma GCC diagnostic pop
 #endif
 
-#include "Utilities/cond.h"
 #include "cellPamf.h"
 #include "cellAdec.h"
 
@@ -215,16 +214,28 @@ struct AtracXdecDecoder
 
 CHECK_SIZE(AtracXdecDecoder, 0xa8);
 
+// HLE exclusive, for savestates
+enum class atracxdec_state : u8
+{
+	initial,
+	waiting_for_cmd,
+	checking_run_thread_1,
+	executing_cmd,
+	waiting_for_output,
+	checking_run_thread_2,
+	decoding
+};
+
 struct AtracXdecContext
 {
 	be_t<u64> thread_id; // sys_ppu_thread_t
 
-	shared_mutex queue_mutex;      // sys_mutex_t
+	be_t<u32> queue_mutex;     // sys_mutex_t
-	cond_variable queue_not_empty; // sys_cond_t
+	be_t<u32> queue_not_empty; // sys_cond_t
 	AdecCmdQueue<AtracXdecCmd> cmd_queue;
 
-	shared_mutex output_mutex;     // sys_mutex_t
+	be_t<u32> output_mutex;    // sys_mutex_t
-	cond_variable output_consumed; // sys_cond_t
+	be_t<u32> output_consumed; // sys_cond_t
 	be_t<u32> output_locked = false;
 
 	be_t<u32> run_thread_mutex; // sys_mutex_t
@@ -241,7 +252,7 @@ struct AtracXdecContext
 	// HLE exclusive
 	u64 cmd_counter = 0;         // For debugging
 	AtracXdecCmd cmd;            // For savestates; if savestate was created while processing a decode command, we need to save the current command
-	b8 skip_getting_command = false; // For savestates; skips getting a new command from the queue
+	atracxdec_state savestate{}; // For savestates
 	b8 skip_next_frame;          // Needed to emulate behavior of LLE SPU program, it doesn't output the first frame after a sequence reset or error
 
 	u8 spurs_stuff[58]; // 120 bytes on LLE, pointers to CellSpurs, CellSpursTaskset, etc.

rpcs3/Emu/Cell/Modules/cellCamera.cpp

@@ -148,9 +148,44 @@ void camera_context::save(utils::serial& ar)
 		return;
 	}
 
-	GET_OR_USE_SERIALIZATION_VERSION(ar.is_writing(), cellCamera);
+	const s32 version = GET_OR_USE_SERIALIZATION_VERSION(ar.is_writing(), cellCamera);
 
 	ar(notify_data_map, start_timestamp_us, read_mode, is_streaming, is_attached, is_open, info, attr, frame_num);
+
+	if (ar.is_writing() || version >= 2)
+	{
+		ar(is_attached_dirty);
+	}
+
+	if (!ar.is_writing())
+	{
+		if (is_open)
+		{
+			if (!open_camera())
+			{
+				cellCamera.error("Failed to open camera while loading savestate");
+			}
+			else if (is_streaming && !start_camera())
+			{
+				cellCamera.error("Failed to start camera while loading savestate");
+			}
+		}
+	}
+}
+
+gem_camera_shared::gem_camera_shared(utils::serial& ar)
+{
+	save(ar);
+}
+
+void gem_camera_shared::save(utils::serial& ar)
+{
+	const s32 version = GET_OR_USE_SERIALIZATION_VERSION(ar.is_writing(), cellCamera);
+
+	if (ar.is_writing() || version >= 2)
+	{
+		ar(frame_timestamp_us, width, height, size, format);
+	}
 }
 
 static bool check_dev_num(s32 dev_num)
@@ -435,7 +470,6 @@ error_code cellCameraInit()
 		g_camera.attr[CELL_CAMERA_USBLOAD] = { 4 };
 		break;
 	}
-
 	case fake_camera_type::eyetoy2:
 	{
 		g_camera.attr[CELL_CAMERA_SATURATION] = { 64 };
@@ -455,7 +489,6 @@ error_code cellCameraInit()
 		g_camera.attr[CELL_CAMERA_AGCHIGH] = { 64 };
 		break;
 	}
-
 	case fake_camera_type::uvc1_1:
 	{
 		g_camera.attr[CELL_CAMERA_DEVICEID] = { 0x5ca, 0x18d0 }; // KBCR-S01MU
@@ -463,14 +496,14 @@ error_code cellCameraInit()
 		g_camera.attr[CELL_CAMERA_NUMFRAME] = { 1 }; // Amount of supported resolutions
 		break;
 	}
-
 	default:
 		cellCamera.todo("Trying to init cellCamera with un-researched camera type.");
+		break;
 	}
 
 	// TODO: Some other default attributes? Need to check the actual behaviour on a real PS3.
 
-	g_camera.is_attached = true;
+	g_camera.is_attached = g_cfg.io.camera != camera_handler::null;
 	g_camera.init = 1;
 	return CELL_OK;
 }
@@ -816,8 +849,8 @@ s32 cellCameraIsAttached(s32 dev_num)
 		// normally should be attached immediately after event queue is registered, but just to be sure
 		if (!is_attached)
 		{
-			g_camera.send_attach_state(true);
+			g_camera.is_attached = is_attached = true;
-			is_attached = g_camera.is_attached;
+			g_camera.is_attached_dirty = true;
 		}
 	}
 
@@ -1606,9 +1639,15 @@ void camera_context::operator()()
 {
 	while (thread_ctrl::state() != thread_state::aborting && !Emu.IsStopped())
 	{
+		// send ATTACH event
+		if (init && is_attached_dirty && !Emu.IsPaused())
+		{
+			send_attach_state(is_attached);
+		}
+
 		const s32 fps = info.framerate;
 
-		if (!fps || Emu.IsPaused() || g_cfg.io.camera == camera_handler::null)
+		if (!init || !fps || Emu.IsPaused() || g_cfg.io.camera == camera_handler::null)
 		{
 			thread_ctrl::wait_for(1000); // hack
 			continue;
@@ -1783,6 +1822,7 @@ void camera_context::reset_state()
 	read_mode = CELL_CAMERA_READ_FUNCCALL;
 	is_streaming = false;
 	is_attached = false;
+	is_attached_dirty = false;
 	is_open = false;
 	info.framerate = 0;
 	std::memset(&attr, 0, sizeof(attr));
@@ -1828,6 +1868,7 @@ void camera_context::send_attach_state(bool attached)
 
 	// We're not expected to send any events for attaching/detaching
 	is_attached = attached;
+	is_attached_dirty = false;
 }
 
 void camera_context::set_attr(s32 attrib, u32 arg1, u32 arg2)
@@ -1862,15 +1903,13 @@ void camera_context::set_attr(s32 attrib, u32 arg1, u32 arg2)
 
 void camera_context::add_queue(u64 key, u64 source, u64 flag)
 {
-	std::lock_guard lock(mutex);
 	{
 		std::lock_guard lock_data_map(mutex_notify_data_map);
 
 		notify_data_map[key] = { source, flag };
 	}
 
-	// send ATTACH event - HACKY
+	is_attached_dirty = true;
-	send_attach_state(is_attached);
 }
 
 void camera_context::remove_queue(u64 key)
@@ -1897,7 +1936,8 @@ bool camera_context::on_handler_state(camera_handler_base::camera_handler_state
 	{
 		if (is_attached)
 		{
-			send_attach_state(false);
+			is_attached = false;
+			is_attached_dirty = true;
 		}
 		if (handler)
 		{
@@ -1938,7 +1978,8 @@ bool camera_context::on_handler_state(camera_handler_base::camera_handler_state
 		if (!is_attached)
 		{
 			cellCamera.warning("Camera handler not attached. Sending attach event...", static_cast<int>(state));
-			send_attach_state(true);
+			is_attached = true;
+			is_attached_dirty = true;
 		}
 		break;
 	}

rpcs3/Emu/Cell/Modules/cellCamera.h

@@ -427,6 +427,7 @@ public:
 	atomic_t<u8> read_mode{CELL_CAMERA_READ_FUNCCALL};
 	atomic_t<bool> is_streaming{false};
 	atomic_t<bool> is_attached{false};
+	atomic_t<bool> is_attached_dirty{false};
 	atomic_t<bool> is_open{false};
 
 	CellCameraInfoEx info{};
@@ -471,6 +472,13 @@ using camera_thread = named_thread<camera_context>;
 /// Shared data between cellGem and cellCamera
 struct gem_camera_shared
 {
+	gem_camera_shared() {}
+	gem_camera_shared(utils::serial& ar);
+
+	void save(utils::serial& ar);
+
+	SAVESTATE_INIT_POS(7);
+
 	atomic_t<u64> frame_timestamp_us{}; // latest read timestamp from cellCamera (cellCameraRead(Ex))
 	atomic_t<s32> width{640};
 	atomic_t<s32> height{480};

rpcs3/Emu/Cell/Modules/cellPad.cpp

@@ -196,7 +196,7 @@ bool cellPad_NotifyStateChange(usz index, u64 /*state*/, bool locked, bool is_bl
 		return true;
 	}
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[index];
 
@@ -268,7 +268,7 @@ error_code cellPadInit(ppu_thread& ppu, u32 max_connect)
 	config.port_setting.fill(CELL_PAD_SETTING_PRESS_OFF | CELL_PAD_SETTING_SENSOR_OFF);
 	config.reported_info = {};
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 
 	for (usz i = 0; i < config.get_max_connect(); ++i)
@@ -336,7 +336,7 @@ error_code cellPadClearBuf(u32 port_no)
 	if (port_no >= config.get_max_connect())
 		return CELL_PAD_ERROR_NO_DEVICE;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -351,7 +351,7 @@ error_code cellPadClearBuf(u32 port_no)
 void pad_get_data(u32 port_no, CellPadData* data, bool get_periph_data = false)
 {
 	auto& config = g_fxo->get<pad_info>();
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pad = handler->GetPads()[port_no];
 	const PadInfo& rinfo = handler->GetInfo();
 
@@ -709,7 +709,7 @@ error_code cellPadGetData(u32 port_no, vm::ptr<CellPadData> data)
 	if (port_no >= config.get_max_connect())
 		return CELL_PAD_ERROR_NO_DEVICE;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -740,7 +740,7 @@ error_code cellPadPeriphGetInfo(vm::ptr<CellPadPeriphInfo> info)
 	if (!info)
 		return CELL_PAD_ERROR_INVALID_PARAMETER;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const PadInfo& rinfo = handler->GetInfo();
 
 	std::memset(info.get_ptr(), 0, sizeof(CellPadPeriphInfo));
@@ -795,7 +795,7 @@ error_code cellPadPeriphGetData(u32 port_no, vm::ptr<CellPadPeriphData> data)
 	if (port_no >= config.get_max_connect())
 		return CELL_PAD_ERROR_NO_DEVICE;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -827,7 +827,7 @@ error_code cellPadGetRawData(u32 port_no, vm::ptr<CellPadData> data)
 	if (port_no >= config.get_max_connect())
 		return CELL_PAD_ERROR_NO_DEVICE;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -891,7 +891,7 @@ error_code cellPadSetActDirect(u32 port_no, vm::ptr<CellPadActParam> param)
 	if (port_no >= config.get_max_connect())
 		return CELL_PAD_ERROR_NO_DEVICE;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -923,7 +923,7 @@ error_code cellPadGetInfo(vm::ptr<CellPadInfo> info)
 
 	std::memset(info.get_ptr(), 0, sizeof(CellPadInfo));
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const PadInfo& rinfo = handler->GetInfo();
 	info->max_connect = config.max_connect;
 	info->system_info = rinfo.system_info;
@@ -968,7 +968,7 @@ error_code cellPadGetInfo2(vm::ptr<CellPadInfo2> info)
 
 	std::memset(info.get_ptr(), 0, sizeof(CellPadInfo2));
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const PadInfo& rinfo = handler->GetInfo();
 	info->max_connect = config.get_max_connect(); // Here it is forcibly clamped
 	info->system_info = rinfo.system_info;
@@ -1018,7 +1018,7 @@ error_code cellPadGetCapabilityInfo(u32 port_no, vm::ptr<CellPadCapabilityInfo>
 	if (port_no >= config.get_max_connect())
 		return CELL_PAD_ERROR_NO_DEVICE;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -1074,7 +1074,7 @@ error_code cellPadInfoPressMode(u32 port_no)
 	if (port_no >= config.get_max_connect())
 		return CELL_PAD_ERROR_NO_DEVICE;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -1101,7 +1101,7 @@ error_code cellPadInfoSensorMode(u32 port_no)
 	if (port_no >= config.get_max_connect())
 		return CELL_PAD_ERROR_NO_DEVICE;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -1129,7 +1129,7 @@ error_code cellPadSetPressMode(u32 port_no, u32 mode)
 	if (port_no >= CELL_PAD_MAX_PORT_NUM)
 		return CELL_OK;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -1163,7 +1163,7 @@ error_code cellPadSetSensorMode(u32 port_no, u32 mode)
 	if (port_no >= CELL_PAD_MAX_PORT_NUM)
 		return CELL_OK;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 	const auto& pad = pads[port_no];
 
@@ -1190,7 +1190,7 @@ error_code cellPadLddRegisterController()
 	if (!config.max_connect)
 		return CELL_PAD_ERROR_UNINITIALIZED;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 
 	const s32 handle = handler->AddLddPad();
 
@@ -1215,7 +1215,7 @@ error_code cellPadLddDataInsert(s32 handle, vm::ptr<CellPadData> data)
 	if (!config.max_connect)
 		return CELL_PAD_ERROR_UNINITIALIZED;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	auto& pads = handler->GetPads();
 
 	if (handle < 0 || static_cast<u32>(handle) >= pads.size() || !data) // data == NULL stalls on decr
@@ -1240,7 +1240,7 @@ error_code cellPadLddGetPortNo(s32 handle)
 	if (!config.max_connect)
 		return CELL_PAD_ERROR_UNINITIALIZED;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	auto& pads = handler->GetPads();
 
 	if (handle < 0 || static_cast<u32>(handle) >= pads.size())
@@ -1264,7 +1264,7 @@ error_code cellPadLddUnregisterController(s32 handle)
 	if (!config.max_connect)
 		return CELL_PAD_ERROR_UNINITIALIZED;
 
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads = handler->GetPads();
 
 	if (handle < 0 || static_cast<u32>(handle) >= pads.size())

rpcs3/Emu/Cell/Modules/sys_prx_.cpp

@@ -221,7 +221,7 @@ error_code sys_prx_get_module_info(ppu_thread& ppu, u32 id, u64 flags, vm::ptr<s
 	opt->info = info;
 
 	// Call the syscall
-	return _sys_prx_get_module_info(ppu, id, 0, opt);
+	return _sys_prx_get_module_info(ppu, id, flags, opt);
 }
 
 error_code sys_prx_get_module_id_by_name(ppu_thread& ppu, vm::cptr<char> name, u64 flags, vm::ptr<sys_prx_get_module_id_by_name_option_t> pOpt)

rpcs3/Emu/Cell/PPUAnalyser.h

@@ -96,6 +96,7 @@ struct ppu_module : public Type
 	std::vector<ppu_segment> segs{};
 	std::vector<ppu_segment> secs{};
 	std::vector<ppu_function> funcs{};
+	std::vector<u32> applied_patches;
 	std::deque<std::shared_ptr<void>> allocations;
 	std::map<u32, u32> addr_to_seg_index;
 
@@ -185,7 +186,6 @@ struct main_ppu_module : public ppu_module<T>
 {
 	u32 elf_entry{};
 	u32 seg0_code_end{};
-	std::vector<u32> applied_patches;
 
 	// Disable inherited savestate ordering
 	void save(utils::serial&) = delete;

rpcs3/Emu/Cell/PPUModule.cpp

@@ -1816,6 +1816,9 @@ shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, bool virtual_load, c
 		prx->module_info_version[1] = lib_info->version[1];
 		prx->module_info_attributes = lib_info->attributes;
 
+		prx->imports_start = lib_info->imports_start;
+		prx->imports_end = lib_info->imports_end;
+
 		prx->exports_start = lib_info->exports_start;
 		prx->exports_end = lib_info->exports_end;
 
@@ -1947,6 +1950,7 @@ shared_ptr<lv2_prx> ppu_load_prx(const ppu_prx_object& elf, bool virtual_load, c
 		ppu_check_patch_spu_images(*prx, seg);
 	}
 
+	prx->applied_patches = applied;
 	prx->analyse(toc, 0, end, applied, exported_funcs);
 
 	if (!ar && !virtual_load)

rpcs3/Emu/Cell/PPUThread.cpp

@@ -4898,6 +4898,30 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
 				sha1_update(&ctx, ensure(info.get_ptr<const u8>(func.addr)), func.size);
 			}
 
+			if (!workload.empty() && fpos >= info.funcs.size())
+			{
+				// Hash the entire function grouped addresses for the integrity of the symbol resolver function
+				// Potentially occuring during patches
+				// Avoid doing it for files with a single module such as most PRX
+
+				std::vector<be_t<u32>> addrs;
+
+				for (const ppu_function& func : info.funcs)
+				{
+					if (func.size == 0)
+					{
+						continue;
+					}
+
+					addrs.emplace_back(func.addr - reloc);
+				}
+
+				// Hash its size too
+				addrs.emplace_back(::size32(addrs));
+
+				sha1_update(&ctx, reinterpret_cast<const u8*>(addrs.data()), addrs.size() * sizeof(be_t<u32>));
+			}
+
 			if (false)
 			{
 				const be_t<u64> forced_upd = 3;

rpcs3/Emu/Cell/SPUThread.cpp

@@ -5481,7 +5481,7 @@ bool spu_thread::reservation_check(u32 addr, u32 hash, atomic_t<u64, 64>* range_
 
 usz spu_thread::register_cache_line_waiter(u32 addr)
 {
-	const u64 value = u64{compute_rdata_hash32(rdata)} << 32 | raddr;
+	const u64 value = u64{compute_rdata_hash32(rdata)} << 32 | addr;
 
 	for (usz i = 0; i < std::size(g_spu_waiters_by_value); i++)
 	{

rpcs3/Emu/Cell/lv2/sys_config.cpp

@@ -122,7 +122,7 @@ void lv2_config::remove_service_event(u32 id)
 
 lv2_config_service_event& lv2_config_service_event::operator=(thread_state s) noexcept
 {
-	if (s == thread_state::finished)
+	if (s == thread_state::destroying_context && !m_destroyed.exchange(true))
 	{
 		if (auto global = g_fxo->try_get<lv2_config>())
 		{
@@ -133,6 +133,23 @@ lv2_config_service_event& lv2_config_service_event::operator=(thread_state s) no
 	return *this;
 }
 
+lv2_config_service_event::~lv2_config_service_event() noexcept
+{
+	operator=(thread_state::destroying_context);
+}
+
+lv2_config::~lv2_config() noexcept
+{
+	for (auto& [key, event] : events)
+	{
+		if (event)
+		{
+			// Avoid collision with lv2_config_service_event destructor
+			event->m_destroyed = true;
+		}
+	}
+}
+
 // LV2 Config Service Listener
 bool lv2_config_service_listener::check_service(const lv2_config_service& service) const
 {

rpcs3/Emu/Cell/lv2/sys_config.h

@@ -161,6 +161,8 @@ public:
 
 		return null_ptr;
 	}
+
+	~lv2_config() noexcept;
 };
 
 /*
@@ -276,7 +278,7 @@ public:
 
 	// Utilities
 	usz get_size() const { return sizeof(sys_config_service_event_t)-1 + data.size(); }
-	shared_ptr<lv2_config_service> get_shared_ptr () const { return idm::get_unlocked<lv2_config_service>(idm_id); }
+	shared_ptr<lv2_config_service> get_shared_ptr () const { return stx::make_shared_from_this<lv2_config_service>(this); }
 	u32 get_id() const { return idm_id; }
 };
 
@@ -342,7 +344,7 @@ public:
 
 	// Utilities
 	u32 get_id() const { return idm_id; }
-	shared_ptr<lv2_config_service_listener> get_shared_ptr() const { return idm::get_unlocked<lv2_config_service_listener>(idm_id); }
+	shared_ptr<lv2_config_service_listener> get_shared_ptr() const { return stx::make_shared_from_this<lv2_config_service_listener>(this); }
 };
 
 /*
@@ -360,6 +362,10 @@ class lv2_config_service_event
 		return g_fxo->get<service_event_id>().next_id++;
 	}
 
+	atomic_t<bool> m_destroyed = false;
+
+	friend class lv2_config;
+
 public:
 	const u32 id;
 
@@ -391,8 +397,7 @@ public:
 
 	// Destructor
 	lv2_config_service_event& operator=(thread_state s) noexcept;
-
+	~lv2_config_service_event() noexcept;
-	~lv2_config_service_event() noexcept = default;
 
 	// Notify queue that this event exists
 	bool notify() const;

rpcs3/Emu/Cell/lv2/sys_event.cpp

@@ -420,10 +420,8 @@ error_code sys_event_queue_tryreceive(ppu_thread& ppu, u32 equeue_id, vm::ptr<sy
 	while (count < size && !queue->events.empty())
 	{
 		auto& dest = events[count++];
-		const auto event = queue->events.front();
+		std::tie(dest.source, dest.data1, dest.data2, dest.data3) = queue->events.front();
 		queue->events.pop_front();
-
-		std::tie(dest.source, dest.data1, dest.data2, dest.data3) = event;
 	}
 
 	lock.unlock();

rpcs3/Emu/Cell/lv2/sys_net/lv2_socket.cpp

@@ -178,7 +178,7 @@ void lv2_socket::queue_wake(ppu_thread* ppu)
 
 lv2_socket& lv2_socket::operator=(thread_state s) noexcept
 {
-	if (s == thread_state::finished)
+	if (s == thread_state::destroying_context)
 	{
 		close();
 	}

rpcs3/Emu/Cell/lv2/sys_overlay.h

@@ -11,7 +11,6 @@ struct lv2_overlay final : ppu_module<lv2_obj>
 
 	u32 entry{};
 	u32 seg0_code_end{};
-	std::vector<u32> applied_patches;
 
 	lv2_overlay() = default;
 	lv2_overlay(utils::serial&){}

rpcs3/Emu/Cell/lv2/sys_process.cpp

@@ -271,7 +271,7 @@ error_code _sys_process_get_paramsfo(vm::ptr<char> buffer)
 {
 	sys_process.warning("_sys_process_get_paramsfo(buffer=0x%x)", buffer);
 
-	if (!Emu.GetTitleID().length())
+	if (Emu.GetTitleID().empty())
 	{
 		return CELL_ENOENT;
 	}
@@ -498,6 +498,9 @@ void lv2_exitspawn(ppu_thread& ppu, std::vector<std::string>& argv, std::vector<
 		};
 
 		signal_system_cache_can_stay();
+
+		// Make sure we keep the game window opened
+		Emu.SetContinuousMode(true);
 		Emu.Kill(false);
 	});
 

rpcs3/Emu/Cell/lv2/sys_prx.cpp

@@ -1034,7 +1034,7 @@ error_code _sys_prx_get_module_info(ppu_thread& ppu, u32 id, u64 flags, vm::ptr<
 		return CELL_EFAULT;
 	}
 
-	if (pOpt->info->size != pOpt->info.size())
+	if (pOpt->info->size != pOpt->info.size() && pOpt->info_v2->size != pOpt->info_v2.size())
 	{
 		return CELL_EINVAL;
 	}
@@ -1072,6 +1072,14 @@ error_code _sys_prx_get_module_info(ppu_thread& ppu, u32 id, u64 flags, vm::ptr<
 		pOpt->info->segments_num = i;
 	}
 
+	if (pOpt->info_v2->size == pOpt->info_v2.size())
+	{
+		pOpt->info_v2->exports_addr = prx->exports_start;
+		pOpt->info_v2->exports_size = prx->exports_end - prx->exports_start;
+		pOpt->info_v2->imports_addr = prx->imports_start;
+		pOpt->info_v2->imports_size = prx->imports_end - prx->imports_start;
+	}
+
 	return CELL_OK;
 }
 
@@ -1079,11 +1087,30 @@ error_code _sys_prx_get_module_id_by_name(ppu_thread& ppu, vm::cptr<char> name,
 {
 	ppu.state += cpu_flag::wait;
 
-	sys_prx.todo("_sys_prx_get_module_id_by_name(name=%s, flags=%d, pOpt=*0x%x)", name, flags, pOpt);
+	sys_prx.warning("_sys_prx_get_module_id_by_name(name=%s, flags=%d, pOpt=*0x%x)", name, flags, pOpt);
 
-	//if (realName == "?") ...
+	std::string module_name;
+	if (!vm::read_string(name.addr(), 28, module_name))
+	{
+		return CELL_EINVAL;
+	}
 
-	return not_an_error(CELL_PRX_ERROR_UNKNOWN_MODULE);
+	const auto [prx, id] = idm::select<lv2_obj, lv2_prx>([&](u32 id, lv2_prx& prx) -> u32
+	{
+		if (strncmp(module_name.c_str(), prx.module_info_name, sizeof(prx.module_info_name)) == 0)
+		{
+			return id;
+		}
+
+		return 0;
+	});
+
+	if (!id)
+	{
+		return CELL_PRX_ERROR_UNKNOWN_MODULE;
+	}
+
+	return not_an_error(id);
 }
 
 error_code _sys_prx_get_module_id_by_address(ppu_thread& ppu, u32 addr)

rpcs3/Emu/Cell/lv2/sys_prx.h

@@ -76,10 +76,22 @@ struct sys_prx_module_info_t
 	be_t<u32> segments_num; // 0x44
 };
 
+struct sys_prx_module_info_v2_t : sys_prx_module_info_t
+{
+	be_t<u32> exports_addr; // 0x48
+	be_t<u32> exports_size; // 0x4C
+	be_t<u32> imports_addr; // 0x50
+	be_t<u32> imports_size; // 0x54
+};
+
 struct sys_prx_module_info_option_t
 {
 	be_t<u64> size; // 0x10
+	union
+	{
 		vm::bptr<sys_prx_module_info_t> info;
+		vm::bptr<sys_prx_module_info_v2_t> info_v2;
+	};
 };
 
 struct sys_prx_start_module_option_t
@@ -192,6 +204,9 @@ struct lv2_prx final : ppu_module<lv2_obj>
 	u8 module_info_version[2]{};
 	be_t<u16> module_info_attributes{};
 
+	u32 imports_start = umax;
+	u32 imports_end = 0;
+
 	u32 exports_start = umax;
 	u32 exports_end = 0;
 

rpcs3/Emu/IPC_config.cpp

@@ -25,7 +25,7 @@ void cfg_ipc::load()
 void cfg_ipc::save() const
 {
 #ifdef _WIN32
-	const std::string path_to_cfg = fs::get_config_dir() + "config/";
+	const std::string path_to_cfg = fs::get_config_dir(true);
 	if (!fs::create_path(path_to_cfg))
 	{
 		IPC.error("Could not create path: %s", path_to_cfg);
@@ -42,11 +42,7 @@ void cfg_ipc::save() const
 
 std::string cfg_ipc::get_path()
 {
-#ifdef _WIN32
+	return fs::get_config_dir(true) + "ipc.yml";
-	return fs::get_config_dir() + "config/ipc.yml";
-#else
-	return fs::get_config_dir() + "ipc.yml";
-#endif
 }
 
 bool cfg_ipc::get_server_enabled() const

rpcs3/Emu/IdManager.h

@@ -805,8 +805,8 @@ public:
 		{
 			if (ptr)
 			{
-				constexpr thread_state finished{3};
+				constexpr thread_state destroying_context{7};
-				*static_cast<Get*>(ptr.get()) = finished;
+				*static_cast<Get*>(ptr.get()) = destroying_context;
 			}
 		}
 
@@ -837,8 +837,8 @@ public:
 		{
 			if (ptr)
 			{
-				constexpr thread_state finished{3};
+				constexpr thread_state destroying_context{7};
-				*static_cast<Get*>(ptr.get()) = finished;
+				*static_cast<Get*>(ptr.get()) = destroying_context;
 			}
 		}
 

rpcs3/Emu/Io/Buzz.cpp

@@ -146,7 +146,7 @@ void usb_device_buzz::interrupt_transfer(u32 buf_size, u8* buf, u32 /*endpoint*/
 	}
 
 	std::lock_guard lock(pad::g_pad_mutex);
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads   = handler->GetPads();
 	ensure(pads.size() > m_last_controller);
 	ensure(g_cfg_buzz.players.size() > m_last_controller);
@@ -161,7 +161,7 @@ void usb_device_buzz::interrupt_transfer(u32 buf_size, u8* buf, u32 /*endpoint*/
 		}
 
 		const auto& cfg = g_cfg_buzz.players[i];
-		cfg->handle_input(pad, true, [&buf, &index](buzz_btn btn, u16 /*value*/, bool pressed)
+		cfg->handle_input(pad, true, [&buf, &index](buzz_btn btn, pad_button /*pad_btn*/, u16 /*value*/, bool pressed, bool& /*abort*/)
 			{
 				if (!pressed)
 					return;

rpcs3/Emu/Io/GHLtar.cpp

@@ -138,7 +138,7 @@ void usb_device_ghltar::interrupt_transfer(u32 buf_size, u8* buf, u32 /*endpoint
 	}
 
 	std::lock_guard lock(pad::g_pad_mutex);
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pad    = ::at32(handler->GetPads(), m_controller_index);
 
 	if (!(pad->m_port_status & CELL_PAD_STATUS_CONNECTED))
@@ -147,7 +147,7 @@ void usb_device_ghltar::interrupt_transfer(u32 buf_size, u8* buf, u32 /*endpoint
 	}
 
 	const auto& cfg = ::at32(g_cfg_ghltar.players, m_controller_index);
-	cfg->handle_input(pad, true, [&buf](ghltar_btn btn, u16 value, bool pressed)
+	cfg->handle_input(pad, true, [&buf](ghltar_btn btn, pad_button /*pad_btn*/, u16 value, bool pressed, bool& /*abort*/)
 		{
 			if (!pressed)
 				return;

rpcs3/Emu/Io/GameTablet.cpp

@@ -195,7 +195,7 @@ void usb_device_gametablet::interrupt_transfer(u32 buf_size, u8* buf, u32 /*endp
 
 	{
 		std::lock_guard lock(pad::g_pad_mutex);
-		const auto gamepad_handler = pad::get_current_handler();
+		const auto gamepad_handler = pad::get_pad_thread();
 		const auto& pads = gamepad_handler->GetPads();
 		const auto& pad = ::at32(pads, m_controller_index);
 		if (pad->m_port_status & CELL_PAD_STATUS_CONNECTED)

rpcs3/Emu/Io/GunCon3.cpp

@@ -227,7 +227,7 @@ void usb_device_guncon3::interrupt_transfer(u32 buf_size, u8* buf, u32 endpoint,
 		return;
 	}
 
-	const auto input_callback = [&gc](guncon3_btn btn, u16 value, bool pressed)
+	const auto input_callback = [&gc](guncon3_btn btn, pad_button /*pad_button*/, u16 value, bool pressed, bool& /*abort*/)
 	{
 		if (!pressed)
 			return;
@@ -255,7 +255,7 @@ void usb_device_guncon3::interrupt_transfer(u32 buf_size, u8* buf, u32 endpoint,
 
 	{
 		std::lock_guard lock(pad::g_pad_mutex);
-		const auto gamepad_handler = pad::get_current_handler();
+		const auto gamepad_handler = pad::get_pad_thread();
 		const auto& pads = gamepad_handler->GetPads();
 		const auto& pad = ::at32(pads, m_controller_index);
 		if (pad->m_port_status & CELL_PAD_STATUS_CONNECTED)

rpcs3/Emu/Io/MouseHandler.cpp

@@ -31,12 +31,12 @@ void MouseHandlerBase::save(utils::serial& ar)
 	ar(inited ? m_info.max_connect : 0);
 }
 
-bool MouseHandlerBase::is_time_for_update(double elapsed_time)
+bool MouseHandlerBase::is_time_for_update(double elapsed_time_ms)
 {
 	steady_clock::time_point now = steady_clock::now();
-	double elapsed = (now - last_update).count() / 1000'000.;
+	const double elapsed_ms = (now - last_update).count() / 1'000'000.;
 
-	if (elapsed > elapsed_time)
+	if (elapsed_ms > elapsed_time_ms)
 	{
 		last_update = now;
 		return true;

rpcs3/Emu/Io/MouseHandler.h

@@ -127,7 +127,7 @@ protected:
 	std::vector<Mouse> m_mice;
 	steady_clock::time_point last_update{};
 
-	bool is_time_for_update(double elapsed_time = 10.0); // 4-10 ms, let's use 10 for now
+	bool is_time_for_update(double elapsed_time_ms = 10.0); // 4-10 ms, let's use 10 for now
 
 public:
 	shared_mutex mutex;

rpcs3/Emu/Io/Null/NullPadHandler.h

@@ -48,6 +48,7 @@ public:
 
 		cfg->pressure_intensity_button.def = "";
 		cfg->analog_limiter_button.def = "";
+		cfg->orientation_reset_button.def = "";
 
 		// Apply defaults
 		cfg->from_default();

rpcs3/Emu/Io/PadHandler.cpp

@@ -2,6 +2,7 @@
 #include "PadHandler.h"
 #include "Emu/system_utils.hpp"
 #include "Emu/system_config.h"
+#include "Emu/Cell/timers.hpp"
 #include "Input/pad_thread.h"
 #include "Input/product_info.h"
 
@@ -494,6 +495,12 @@ bool PadHandlerBase::bindPadToDevice(std::shared_ptr<Pad> pad)
 		pad->m_analog_limiter_button_index = static_cast<s32>(pad->m_buttons.size()) - 1;
 	}
 
+	if (b_has_orientation)
+	{
+		pad->m_buttons.emplace_back(special_button_offset, mapping[button::orientation_reset_button], special_button_value::orientation_reset);
+		pad->m_orientation_reset_button_index = static_cast<s32>(pad->m_buttons.size()) - 1;
+	}
+
 	pad->m_buttons.emplace_back(CELL_PAD_BTN_OFFSET_DIGITAL1, mapping[button::up], CELL_PAD_CTRL_UP);
 	pad->m_buttons.emplace_back(CELL_PAD_BTN_OFFSET_DIGITAL1, mapping[button::down], CELL_PAD_CTRL_DOWN);
 	pad->m_buttons.emplace_back(CELL_PAD_BTN_OFFSET_DIGITAL1, mapping[button::left], CELL_PAD_CTRL_LEFT);
@@ -600,6 +607,11 @@ std::array<std::set<u32>, PadHandlerBase::button::button_count> PadHandlerBase::
 		mapping[button::analog_limiter_button] = FindKeyCodes<u32, u32>(button_list, cfg->analog_limiter_button);
 	}
 
+	if (b_has_orientation)
+	{
+		mapping[button::orientation_reset_button] = FindKeyCodes<u32, u32>(button_list, cfg->orientation_reset_button);
+	}
+
 	return mapping;
 }
 
@@ -739,6 +751,8 @@ void PadHandlerBase::process()
 		if (!device || !pad)
 			continue;
 
+		pad->move_data.orientation_enabled = b_has_orientation && device->config && device->config->orientation_enabled.get();
+
 		const connection status = update_connection(device);
 
 		switch (status)
@@ -754,6 +768,11 @@ void PadHandlerBase::process()
 
 				last_connection_status[i] = true;
 				connected_devices++;
+
+				if (b_has_orientation)
+				{
+					device->reset_orientation();
+				}
 			}
 
 			if (status == connection::no_data)
@@ -790,6 +809,11 @@ void PadHandlerBase::process()
 
 				last_connection_status[i] = false;
 				connected_devices--;
+
+				if (b_has_orientation)
+				{
+					device->reset_orientation();
+				}
 			}
 			continue;
 		}
@@ -797,6 +821,142 @@ void PadHandlerBase::process()
 
 		get_mapping(m_bindings[i]);
 		get_extended_info(m_bindings[i]);
+		get_orientation(m_bindings[i]);
 		apply_pad_data(m_bindings[i]);
 	}
 }
+
+void PadHandlerBase::set_raw_orientation(ps_move_data& move_data, f32 accel_x, f32 accel_y, f32 accel_z, f32 gyro_x, f32 gyro_y, f32 gyro_z)
+{
+	if (!move_data.orientation_enabled)
+	{
+		move_data.reset_sensors();
+		return;
+	}
+
+	// This function expects DS3 sensor accel values in linear velocity (m/s²) and gyro values in angular velocity (degree/s)
+	// The default position is flat on the ground, pointing forward.
+	// The accelerometers constantly measure G forces.
+	// The gyros measure changes in orientation and will reset when the device isn't moved anymore.
+	move_data.accelerometer_x = -accel_x;      // move_data: Increases if the device is rolled to the left
+	move_data.accelerometer_y = accel_z;       // move_data: Increases if the device is pitched upwards
+	move_data.accelerometer_z = accel_y;       // move_data: Increases if the device is moved upwards
+	move_data.gyro_x = degree_to_rad(-gyro_x); // move_data: Increases if the device is pitched upwards
+	move_data.gyro_y = degree_to_rad(gyro_z);  // move_data: Increases if the device is rolled to the right
+	move_data.gyro_z = degree_to_rad(-gyro_y); // move_data: Increases if the device is yawed to the left
+}
+
+void PadHandlerBase::set_raw_orientation(Pad& pad)
+{
+	if (!pad.move_data.orientation_enabled)
+	{
+		pad.move_data.reset_sensors();
+		return;
+	}
+
+	// acceleration (linear velocity in m/s²)
+	const f32 accel_x = (pad.m_sensors[0].m_value - 512) / static_cast<f32>(MOTION_ONE_G);
+	const f32 accel_y = (pad.m_sensors[1].m_value - 512) / static_cast<f32>(MOTION_ONE_G);
+	const f32 accel_z = (pad.m_sensors[2].m_value - 512) / static_cast<f32>(MOTION_ONE_G);
+
+	// gyro (angular velocity in degree/s)
+	constexpr f32 gyro_x = 0.0f;
+	const f32 gyro_y = (pad.m_sensors[3].m_value - 512) / (123.f / 90.f);
+	constexpr f32 gyro_z = 0.0f;
+
+	set_raw_orientation(pad.move_data, accel_x, accel_y, accel_z, gyro_x, gyro_y, gyro_z);
+}
+
+void PadHandlerBase::get_orientation(const pad_ensemble& binding) const
+{
+	if (!b_has_orientation) return;
+
+	const auto& pad = binding.pad;
+	const auto& device = binding.device;
+	if (!pad || !device) return;
+
+	if (pad->move_data.calibration_requested)
+	{
+		device->reset_orientation();
+		pad->move_data.quaternion = ps_move_data::default_quaternion;
+		pad->move_data.calibration_succeeded = true;
+		return;
+	}
+
+	if (!pad->move_data.orientation_enabled || pad->get_orientation_reset_button_active())
+	{
+		// This can be called extensively in quick succession, so let's just reset the pointer instead of creating a new object.
+		device->ahrs.reset();
+		pad->move_data.quaternion = ps_move_data::default_quaternion;
+		return;
+	}
+
+	device->update_orientation(pad->move_data);
+}
+
+void PadDevice::reset_orientation()
+{
+	// Initialize Fusion
+	ahrs = std::make_shared<FusionAhrs>();
+	FusionAhrsInitialise(ahrs.get());
+	ahrs->settings.convention = FusionConvention::FusionConventionEnu;
+	ahrs->settings.gain = 0.0f; // If gain is set, the algorithm tries to adjust the orientation over time.
+	FusionAhrsSetSettings(ahrs.get(), &ahrs->settings);
+	FusionAhrsReset(ahrs.get());
+}
+
+void PadDevice::update_orientation(ps_move_data& move_data)
+{
+	if (!ahrs)
+	{
+		reset_orientation();
+	}
+
+	// Get elapsed time since last update
+	const u64 now_us = get_system_time();
+	const float elapsed_sec = (last_ahrs_update_time_us == 0) ? 0.0f : ((now_us - last_ahrs_update_time_us) / 1'000'000.0f);
+	last_ahrs_update_time_us = now_us;
+
+	// The ps move handler's axis may differ from the Fusion axis, so we have to map them correctly.
+	// Don't ask how the axis work. It's basically been trial and error.
+	ensure(ahrs->settings.convention == FusionConvention::FusionConventionEnu); // East-North-Up
+
+	const FusionVector accelerometer{
+		.axis {
+			.x = -move_data.accelerometer_x,
+			.y = +move_data.accelerometer_y,
+			.z = +move_data.accelerometer_z
+		}
+	};
+
+	const FusionVector gyroscope{
+		.axis {
+			.x = +PadHandlerBase::rad_to_degree(move_data.gyro_x),
+			.y = +PadHandlerBase::rad_to_degree(move_data.gyro_z),
+			.z = -PadHandlerBase::rad_to_degree(move_data.gyro_y)
+		}
+	};
+
+	FusionVector magnetometer {};
+
+	if (move_data.magnetometer_enabled)
+	{
+		magnetometer = FusionVector{
+			.axis {
+				.x = move_data.magnetometer_x,
+				.y = move_data.magnetometer_y,
+				.z = move_data.magnetometer_z
+			}
+		};
+	}
+
+	// Update Fusion
+	FusionAhrsUpdate(ahrs.get(), gyroscope, accelerometer, magnetometer, elapsed_sec);
+
+	// Get quaternion
+	const FusionQuaternion quaternion = FusionAhrsGetQuaternion(ahrs.get());
+	move_data.quaternion[0] = quaternion.array[1];
+	move_data.quaternion[1] = quaternion.array[2];
+	move_data.quaternion[2] = quaternion.array[3];
+	move_data.quaternion[3] = quaternion.array[0];
+}

rpcs3/Emu/Io/PadHandler.h

@@ -5,6 +5,15 @@
 #include "pad_config_types.h"
 #include "util/types.hpp"
 
+#ifndef _MSC_VER
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wold-style-cast"
+#endif
+#include "3rdparty/fusion/fusion/Fusion/Fusion.h"
+#ifndef _MSC_VER
+#pragma GCC diagnostic pop
+#endif
+
 #include <cmath>
 #include <functional>
 #include <string>
@@ -38,6 +47,12 @@ public:
 	};
 	color color_override{};
 	bool color_override_active{};
+
+	std::shared_ptr<FusionAhrs> ahrs; // Used to calculate quaternions from sensor data
+	u64 last_ahrs_update_time_us = 0; // Last ahrs update
+
+	void update_orientation(ps_move_data& move_data);
+	void reset_orientation();
 };
 
 struct pad_ensemble
@@ -125,6 +140,7 @@ protected:
 
 		pressure_intensity_button,
 		analog_limiter_button,
+		orientation_reset_button,
 
 		button_count
 	};
@@ -153,6 +169,7 @@ protected:
 	bool b_has_config = false;
 	bool b_has_pressure_intensity_button = true;
 	bool b_has_analog_limiter_button = true;
+	bool b_has_orientation = false;
 
 	std::array<cfg_pad, MAX_GAMEPADS> m_pad_configs;
 	std::vector<pad_ensemble> m_bindings;
@@ -301,6 +318,7 @@ public:
 	bool has_battery_led() const { return b_has_battery_led; }
 	bool has_pressure_intensity_button() const { return b_has_pressure_intensity_button; }
 	bool has_analog_limiter_button() const { return b_has_analog_limiter_button; }
+	bool has_orientation() const { return b_has_orientation; }
 
 	u16 NormalizeStickInput(u16 raw_value, s32 threshold, s32 multiplier, bool ignore_threshold = false) const;
 	void convert_stick_values(u16& x_out, u16& y_out, s32 x_in, s32 y_in, u32 deadzone, u32 anti_deadzone, u32 padsquircling) const;
@@ -323,6 +341,18 @@ public:
 	virtual void get_motion_sensors(const std::string& pad_id, const motion_callback& callback, const motion_fail_callback& fail_callback, motion_preview_values preview_values, const std::array<AnalogSensor, 4>& sensors);
 	virtual std::unordered_map<u32, std::string> get_motion_axis_list() const { return {}; }
 
+	static constexpr f32 PI = 3.14159265f;
+
+	static f32 degree_to_rad(f32 degree)
+	{
+		return degree * PI / 180.0f;
+	}
+
+	static f32 rad_to_degree(f32 radians)
+	{
+		return radians * 180.0f / PI;
+	};
+
 private:
 	virtual std::shared_ptr<PadDevice> get_device(const std::string& /*device*/) { return nullptr; }
 	virtual bool get_is_left_trigger(const std::shared_ptr<PadDevice>& /*device*/, u64 /*keyCode*/) { return false; }
@@ -336,10 +366,15 @@ private:
 	virtual std::unordered_map<u64, u16> get_button_values(const std::shared_ptr<PadDevice>& /*device*/) { return {}; }
 	virtual pad_preview_values get_preview_values(const std::unordered_map<u64, u16>& /*data*/) { return {}; }
 
+	void get_orientation(const pad_ensemble& binding) const;
+
 protected:
 	virtual std::array<std::set<u32>, PadHandlerBase::button::button_count> get_mapped_key_codes(const std::shared_ptr<PadDevice>& device, const cfg_pad* cfg);
 	virtual void get_mapping(const pad_ensemble& binding);
 	void TranslateButtonPress(const std::shared_ptr<PadDevice>& device, u64 keyCode, bool& pressed, u16& val, bool use_stick_multipliers, bool ignore_stick_threshold = false, bool ignore_trigger_threshold = false);
 	void init_configs();
 	cfg_pad* get_config(const std::string& pad_id);
+
+	static void set_raw_orientation(ps_move_data& move_data, f32 accel_x, f32 accel_y, f32 accel_z, f32 gyro_x, f32 gyro_y, f32 gyro_z);
+	static void set_raw_orientation(Pad& pad);
 };

rpcs3/Emu/Io/TopShotElite.cpp

@@ -280,7 +280,7 @@ void usb_device_topshotelite::interrupt_transfer(u32 buf_size, u8* buf, u32 /*en
 	}
 
 	bool up = false, right = false, down = false, left = false;
-	const auto input_callback = [&ts, &up, &down, &left, &right](topshotelite_btn btn, u16 value, bool pressed)
+	const auto input_callback = [&ts, &up, &down, &left, &right](topshotelite_btn btn, pad_button /*pad_button*/, u16 value, bool pressed, bool& /*abort*/)
 	{
 		if (!pressed)
 			return;
@@ -315,7 +315,7 @@ void usb_device_topshotelite::interrupt_transfer(u32 buf_size, u8* buf, u32 /*en
 
 	{
 		std::lock_guard lock(pad::g_pad_mutex);
-		const auto gamepad_handler = pad::get_current_handler();
+		const auto gamepad_handler = pad::get_pad_thread();
 		const auto& pads = gamepad_handler->GetPads();
 		const auto& pad = ::at32(pads, m_controller_index);
 		if (pad->m_port_status & CELL_PAD_STATUS_CONNECTED)

rpcs3/Emu/Io/TopShotFearmaster.cpp

@@ -308,7 +308,7 @@ void usb_device_topshotfearmaster::interrupt_transfer(u32 buf_size, u8* buf, u32
 	}
 
 	bool up = false, right = false, down = false, left = false;
-	const auto input_callback = [&ts, &up, &down, &left, &right](topshotfearmaster_btn btn, u16 value, bool pressed)
+	const auto input_callback = [&ts, &up, &down, &left, &right](topshotfearmaster_btn btn, pad_button /*pad_button*/, u16 value, bool pressed, bool& /*abort*/)
 	{
 		if (!pressed)
 			return;
@@ -339,7 +339,7 @@ void usb_device_topshotfearmaster::interrupt_transfer(u32 buf_size, u8* buf, u32
 
 	{
 		std::lock_guard lock(pad::g_pad_mutex);
-		const auto gamepad_handler = pad::get_current_handler();
+		const auto gamepad_handler = pad::get_pad_thread();
 		const auto& pads = gamepad_handler->GetPads();
 		const auto& pad = ::at32(pads, m_controller_index);
 		if (pad->m_port_status & CELL_PAD_STATUS_CONNECTED)

rpcs3/Emu/Io/Turntable.cpp

@@ -151,7 +151,7 @@ void usb_device_turntable::interrupt_transfer(u32 buf_size, u8* buf, u32 /*endpo
 	// All other bufs are always 0x00
 
 	std::lock_guard lock(pad::g_pad_mutex);
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 	const auto& pads   = handler->GetPads();
 	const auto& pad    = ::at32(pads, m_controller_index);
 
@@ -159,7 +159,7 @@ void usb_device_turntable::interrupt_transfer(u32 buf_size, u8* buf, u32 /*endpo
 		return;
 
 	const auto& cfg = ::at32(g_cfg_turntable.players, m_controller_index);
-	cfg->handle_input(pad, true, [&buf](turntable_btn btn, u16 value, bool pressed)
+	cfg->handle_input(pad, true, [&buf](turntable_btn btn, pad_button /*pad_btn*/, u16 value, bool pressed, bool& /*abort*/)
 		{
 			if (!pressed)
 				return;

rpcs3/Emu/Io/camera_config.cpp

@@ -8,11 +8,7 @@ cfg_camera g_cfg_camera;
 
 cfg_camera::cfg_camera()
 	: cfg::node()
-#ifdef _WIN32
+	, path(fs::get_config_dir(true) + "camera.yml")
-	, path(fs::get_config_dir() + "config/camera.yml")
-#else
-	, path(fs::get_config_dir() + "camera.yml")
-#endif
 {
 }
 

rpcs3/Emu/Io/emulated_pad_config.h

@@ -88,7 +88,7 @@ public:
 		button_map.clear();
 	}
 
-	void handle_input(std::shared_ptr<Pad> pad, bool press_only, const std::function<void(T, u16, bool)>& func) const
+	void handle_input(std::shared_ptr<Pad> pad, bool press_only, const std::function<void(T, pad_button, u16, bool, bool&)>& func) const
 	{
 		if (!pad)
 			return;
@@ -97,19 +97,25 @@ public:
 		{
 			if (button.m_pressed || !press_only)
 			{
-				handle_input(func, button.m_offset, button.m_outKeyCode, button.m_value, button.m_pressed, true);
+				if (handle_input(func, button.m_offset, button.m_outKeyCode, button.m_value, button.m_pressed, true))
+				{
+					return;
+				}
 			}
 		}
 
 		for (const AnalogStick& stick : pad->m_sticks)
 		{
-			handle_input(func, stick.m_offset, get_axis_keycode(stick.m_offset, stick.m_value), stick.m_value, true, true);
+			if (handle_input(func, stick.m_offset, get_axis_keycode(stick.m_offset, stick.m_value), stick.m_value, true, true))
+			{
+				return;
+			}
 		}
 	}
 
-	void handle_input(const Mouse& mouse, const std::function<void(T, u16, bool)>& func) const
+	void handle_input(const Mouse& mouse, const std::function<void(T, pad_button, u16, bool, bool&)>& func) const
 	{
-		for (int i = 0; i < 7; i++)
+		for (int i = 0; i < 8; i++)
 		{
 			const MouseButtonCodes cell_code = get_mouse_button_code(i);
 			if ((mouse.buttons & cell_code))
@@ -117,7 +123,11 @@ public:
 				const pad_button button = static_cast<pad_button>(static_cast<int>(pad_button::mouse_button_1) + i);
 				const u32 offset = pad_button_offset(button);
 				const u32 keycode = pad_button_keycode(button);
-				handle_input(func, offset, keycode, 255, true, true);
+
+				if (handle_input(func, offset, keycode, 255, true, true))
+				{
+					return;
+				}
 			}
 		}
 	}
@@ -163,10 +173,12 @@ protected:
 		return empty_set;
 	}
 
-	void handle_input(const std::function<void(T, u16, bool)>& func, u32 offset, u32 keycode, u16 value, bool pressed, bool check_axis) const
+	bool handle_input(const std::function<void(T, pad_button, u16, bool, bool&)>& func, u32 offset, u32 keycode, u16 value, bool pressed, bool check_axis) const
 	{
 		m_mutex.lock();
 
+		bool abort = false;
+
 		const auto& btns = find_button(offset, keycode);
 		if (btns.empty())
 		{
@@ -180,24 +192,26 @@ protected:
 				case CELL_PAD_BTN_OFFSET_ANALOG_LEFT_Y:
 				case CELL_PAD_BTN_OFFSET_ANALOG_RIGHT_X:
 				case CELL_PAD_BTN_OFFSET_ANALOG_RIGHT_Y:
-					handle_input(func, offset, static_cast<u32>(axis_direction::both), value, pressed, false);
+					abort = handle_input(func, offset, static_cast<u32>(axis_direction::both), value, pressed, false);
 					break;
 				default:
 					break;
 				}
 			}
-			return;
+			return abort;
 		}
 
 		for (const auto& btn : btns)
 		{
 			if (btn && func)
 			{
-				func(btn->btn_id(), value, pressed);
+				func(btn->btn_id(), btn->get(), value, pressed, abort);
+				if (abort) break;
 			}
 		}
 
 		m_mutex.unlock();
+		return abort;
 	}
 };
 
@@ -221,7 +235,7 @@ struct emulated_pads_config : cfg::node
 		m_mutex.lock();
 
 		bool result = false;
-		const std::string cfg_name = fmt::format("%sconfig/%s.yml", fs::get_config_dir(), cfg_id);
+		const std::string cfg_name = fmt::format("%s%s.yml", fs::get_config_dir(true), cfg_id);
 		cfg_log.notice("Loading %s config: %s", cfg_id, cfg_name);
 
 		from_default();
@@ -258,7 +272,7 @@ struct emulated_pads_config : cfg::node
 	{
 		std::lock_guard lock(m_mutex);
 
-		const std::string cfg_name = fmt::format("%sconfig/%s.yml", fs::get_config_dir(), cfg_id);
+		const std::string cfg_name = fmt::format("%s%s.yml", fs::get_config_dir(true), cfg_id);
 		cfg_log.notice("Saving %s config to '%s'", cfg_id, cfg_name);
 
 		if (!fs::create_path(fs::get_parent_dir(cfg_name)))

rpcs3/Emu/Io/gem_config.h

@@ -4,7 +4,7 @@
 
 #include <array>
 
-enum class gem_btn
+enum class gem_btn : u32
 {
 	start,
 	select,
@@ -17,6 +17,18 @@ enum class gem_btn
 	x_axis,
 	y_axis,
 
+	combo_begin,
+	combo = combo_begin,
+	combo_start,
+	combo_select,
+	combo_triangle,
+	combo_circle,
+	combo_cross,
+	combo_square,
+	combo_move,
+	combo_t,
+	combo_end = combo_t,
+
 	count
 };
 
@@ -41,6 +53,34 @@ struct cfg_fake_gems final : public emulated_pads_config<cfg_fake_gem, 4>
 	cfg_fake_gems() : emulated_pads_config<cfg_fake_gem, 4>("gem") {};
 };
 
+struct cfg_mouse_gem final : public emulated_pad_config<gem_btn>
+{
+	cfg_mouse_gem(node* owner, const std::string& name) : emulated_pad_config(owner, name) {}
+
+	cfg_pad_btn<gem_btn> start{ this, "Start", gem_btn::start, pad_button::mouse_button_6 };
+	cfg_pad_btn<gem_btn> select{ this, "Select", gem_btn::select, pad_button::mouse_button_7 };
+	cfg_pad_btn<gem_btn> triangle{ this, "Triangle", gem_btn::triangle, pad_button::mouse_button_8 };
+	cfg_pad_btn<gem_btn> circle{ this, "Circle", gem_btn::circle, pad_button::mouse_button_4 };
+	cfg_pad_btn<gem_btn> cross{ this, "Cross", gem_btn::cross, pad_button::mouse_button_5 };
+	cfg_pad_btn<gem_btn> square{ this, "Square", gem_btn::square, pad_button::mouse_button_3 };
+	cfg_pad_btn<gem_btn> move{ this, "Move", gem_btn::move, pad_button::mouse_button_2 };
+	cfg_pad_btn<gem_btn> t{ this, "T", gem_btn::t, pad_button::mouse_button_1 };
+	cfg_pad_btn<gem_btn> combo{ this, "Combo", gem_btn::combo, pad_button::pad_button_max_enum };
+	cfg_pad_btn<gem_btn> combo_start{ this, "Combo Start", gem_btn::combo_start, pad_button::pad_button_max_enum };
+	cfg_pad_btn<gem_btn> combo_select{ this, "Combo Select", gem_btn::combo_select, pad_button::pad_button_max_enum };
+	cfg_pad_btn<gem_btn> combo_triangle{ this, "Combo Triangle", gem_btn::combo_triangle, pad_button::pad_button_max_enum };
+	cfg_pad_btn<gem_btn> combo_circle{ this, "Combo Circle", gem_btn::combo_circle, pad_button::pad_button_max_enum };
+	cfg_pad_btn<gem_btn> combo_cross{ this, "Combo Cross", gem_btn::combo_cross, pad_button::pad_button_max_enum };
+	cfg_pad_btn<gem_btn> combo_square{ this, "Combo Square", gem_btn::combo_square, pad_button::pad_button_max_enum };
+	cfg_pad_btn<gem_btn> combo_move{ this, "Combo Move", gem_btn::combo_move, pad_button::pad_button_max_enum };
+	cfg_pad_btn<gem_btn> combo_t{ this, "Combo T", gem_btn::combo_t, pad_button::pad_button_max_enum };
+};
+
+struct cfg_mouse_gems final : public emulated_pads_config<cfg_mouse_gem, 4>
+{
+	cfg_mouse_gems() : emulated_pads_config<cfg_mouse_gem, 4>("gem_mouse") {};
+};
+
 struct cfg_gem final : public emulated_pad_config<gem_btn>
 {
 	cfg_gem(node* owner, const std::string& name) : emulated_pad_config(owner, name) {}
@@ -62,3 +102,4 @@ struct cfg_gems final : public emulated_pads_config<cfg_gem, 4>
 
 extern cfg_gems g_cfg_gem_real;
 extern cfg_fake_gems g_cfg_gem_fake;
+extern cfg_mouse_gems g_cfg_gem_mouse;

rpcs3/Emu/Io/mouse_config.cpp

@@ -4,11 +4,7 @@
 #include "Utilities/File.h"
 
 mouse_config::mouse_config()
-#ifdef _WIN32
+	: cfg_name(fs::get_config_dir(true) + "config_mouse.yml")
-	: cfg_name(fs::get_config_dir() + "config/config_mouse.yml")
-#else
-	: cfg_name(fs::get_config_dir() + "config_mouse.yml")
-#endif
 {
 }
 

rpcs3/Emu/Io/pad_config.cpp

@@ -32,6 +32,20 @@ std::string cfg_pad::get_buttons(std::vector<std::string> vec)
 	return fmt::merge(vec, ",");
 }
 
+u8 cfg_pad::get_large_motor_speed(const std::array<VibrateMotor, 2>& motor_speed) const
+{
+	const u8 idx = switch_vibration_motors ? 1 : 0;
+	const f32 multiplier = multiplier_vibration_motor_large / 100.0f;
+	return static_cast<u8>(std::clamp(motor_speed[idx].m_value * multiplier, 0.0f, 255.0f));
+}
+
+u8 cfg_pad::get_small_motor_speed(const std::array<VibrateMotor, 2>& motor_speed) const
+{
+	const u8 idx = switch_vibration_motors ? 0 : 1;
+	const f32 multiplier = multiplier_vibration_motor_small / 100.0f;
+	return static_cast<u8>(std::clamp(motor_speed[idx].m_value * multiplier, 0.0f, 255.0f));
+}
+
 bool cfg_input::load(const std::string& title_id, const std::string& config_file, bool strict)
 {
 	input_log.notice("Loading pad config (title_id='%s', config_file='%s', strict=%d)", title_id, config_file, strict);

rpcs3/Emu/Io/pad_config.h

@@ -28,6 +28,9 @@ struct cfg_pad final : cfg::node
 	static std::vector<std::string> get_buttons(const std::string& str);
 	static std::string get_buttons(std::vector<std::string> vec);
 
+	u8 get_large_motor_speed(const std::array<VibrateMotor, 2>& motor_speed) const;
+	u8 get_small_motor_speed(const std::array<VibrateMotor, 2>& motor_speed) const;
+
 	cfg::string ls_left{ this, "Left Stick Left", "" };
 	cfg::string ls_down{ this, "Left Stick Down", "" };
 	cfg::string ls_right{ this, "Left Stick Right", "" };
@@ -66,6 +69,9 @@ struct cfg_pad final : cfg::node
 	cfg_sensor motion_sensor_z{ this, "Motion Sensor Z" };
 	cfg_sensor motion_sensor_g{ this, "Motion Sensor G" };
 
+	cfg::string orientation_reset_button{ this, "Orientation Reset Button", "" };
+	cfg::_bool orientation_enabled{ this, "Orientation Enabled", false };
+
 	cfg::string pressure_intensity_button{ this, "Pressure Intensity Button", "" };
 	cfg::uint<0, 100> pressure_intensity{ this, "Pressure Intensity Percent", 50 };
 	cfg::_bool pressure_intensity_toggle_mode{ this, "Pressure Intensity Toggle Mode", false };
@@ -93,8 +99,8 @@ struct cfg_pad final : cfg::node
 	cfg::uint<0, 100> led_battery_indicator_brightness{ this, "LED battery indicator brightness", 50 };
 	cfg::_bool player_led_enabled{ this, "Player LED enabled", true };
 
-	cfg::_bool enable_vibration_motor_large{ this, "Enable Large Vibration Motor", true };
+	cfg::uint<0, 200> multiplier_vibration_motor_large{ this, "Large Vibration Motor Multiplier", 100 };
-	cfg::_bool enable_vibration_motor_small{ this, "Enable Small Vibration Motor", true };
+	cfg::uint<0, 200> multiplier_vibration_motor_small{ this, "Small Vibration Motor Multiplier", 100 };
 	cfg::_bool switch_vibration_motors{ this, "Switch Vibration Motors", false };
 
 	cfg::_enum<mouse_movement_mode> mouse_move_mode{ this, "Mouse Movement Mode", mouse_movement_mode::relative };

rpcs3/Emu/Io/pad_types.cpp

@@ -39,7 +39,7 @@ void fmt_class_string<pad_button>::format(std::string& out, u64 arg)
 		case pad_button::rs_right: return "Right Stick Right";
 		case pad_button::rs_x: return "Right Stick X-Axis";
 		case pad_button::rs_y: return "Right Stick Y-Axis";
-		case pad_button::pad_button_max_enum: return "MAX_ENUM";
+		case pad_button::pad_button_max_enum: return "";
 		case pad_button::mouse_button_1: return "Mouse Button 1";
 		case pad_button::mouse_button_2: return "Mouse Button 2";
 		case pad_button::mouse_button_3: return "Mouse Button 3";
@@ -159,6 +159,20 @@ u32 get_axis_keycode(u32 offset, u16 value)
 	}
 }
 
+void ps_move_data::reset_sensors()
+{
+	quaternion = default_quaternion;
+	accelerometer_x = 0.0f;
+	accelerometer_y = 0.0f;
+	accelerometer_z = 0.0f;
+	gyro_x = 0.0f;
+	gyro_y = 0.0f;
+	gyro_z = 0.0f;
+	magnetometer_x = 0.0f;
+	magnetometer_y = 0.0f;
+	magnetometer_z = 0.0f;
+}
+
 bool Pad::get_pressure_intensity_button_active(bool is_toggle_mode, u32 player_id)
 {
 	if (m_pressure_intensity_button_index < 0)
@@ -238,3 +252,13 @@ bool Pad::get_analog_limiter_button_active(bool is_toggle_mode, u32 player_id)
 
 	return analog_limiter_button.m_pressed;
 }
+
+bool Pad::get_orientation_reset_button_active()
+{
+	if (m_orientation_reset_button_index < 0)
+	{
+		return false;
+	}
+
+	return m_buttons[m_orientation_reset_button_index].m_pressed;
+}

rpcs3/Emu/Io/pad_types.h

@@ -365,7 +365,8 @@ constexpr u32 special_button_offset = 666; // Must not conflict with other CELL
 enum special_button_value
 {
 	pressure_intensity,
-	analog_limiter
+	analog_limiter,
+	orientation_reset
 };
 
 struct Button
@@ -470,8 +471,10 @@ struct ps_move_data
 	bool calibration_succeeded = false;
 
 	bool magnetometer_enabled = false;
+	bool orientation_enabled = false;
 
-	std::array<f32, 4> quaternion { 1.0f, 0.0f, 0.0f, 0.0f }; // quaternion orientation (x,y,z,w) of controller relative to default (facing the camera with buttons up)
+	static constexpr std::array<f32, 4> default_quaternion { 1.0f, 0.0f, 0.0f, 0.0f };
+	std::array<f32, 4> quaternion = default_quaternion; // quaternion orientation (x,y,z,w) of controller relative to default (facing the camera with buttons up)
 	f32 accelerometer_x = 0.0f; // linear velocity in m/s²
 	f32 accelerometer_y = 0.0f; // linear velocity in m/s²
 	f32 accelerometer_z = 0.0f; // linear velocity in m/s²
@@ -482,6 +485,8 @@ struct ps_move_data
 	f32 magnetometer_y = 0.0f;
 	f32 magnetometer_z = 0.0f;
 	s16 temperature = 0;
+
+	void reset_sensors();
 };
 
 struct Pad
@@ -512,6 +517,9 @@ struct Pad
 	bool m_analog_limiter_enabled_last{}; // only used in keyboard_pad_handler
 	bool get_analog_limiter_button_active(bool is_toggle_mode, u32 player_id);
 
+	s32 m_orientation_reset_button_index{-1}; // Special button index. -1 if not set.
+	bool get_orientation_reset_button_active();
+
 	// Cable State:   0 - 1  plugged in ?
 	u8 m_cable_state{0};
 

rpcs3/Emu/Io/rb3drums_config.cpp

@@ -8,13 +8,7 @@ cfg_rb3drums g_cfg_rb3drums;
 
 cfg_rb3drums::cfg_rb3drums()
 	: cfg::node()
-#ifdef _WIN32
+	, path(fs::get_config_dir(true) + "rb3drums.yml")
-	  ,
-	  path(fs::get_config_dir() + "config/rb3drums.yml")
-#else
-	  ,
-	  path(fs::get_config_dir() + "rb3drums.yml")
-#endif
 {
 }
 

rpcs3/Emu/Io/recording_config.cpp

@@ -8,11 +8,7 @@ cfg_recording g_cfg_recording;
 
 cfg_recording::cfg_recording()
 	: cfg::node()
-#ifdef _WIN32
+	, path(fs::get_config_dir(true) + "recording.yml")
-	, path(fs::get_config_dir() + "config/recording.yml")
-#else
-	, path(fs::get_config_dir() + "recording.yml")
-#endif
 {
 }
 

rpcs3/Emu/Io/usio.cpp

@@ -189,7 +189,7 @@ void usb_device_usio::save_backup()
 void usb_device_usio::translate_input_taiko()
 {
 	std::lock_guard lock(pad::g_pad_mutex);
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 
 	std::vector<u8> input_buf(0x60);
 	constexpr le_t<u16> c_hit = 0x1800;
@@ -203,7 +203,7 @@ void usb_device_usio::translate_input_taiko()
 		if (const auto& pad = ::at32(handler->GetPads(), pad_number); (pad->m_port_status & CELL_PAD_STATUS_CONNECTED) && is_input_allowed())
 		{
 			const auto& cfg = ::at32(g_cfg_usio.players, pad_number);
-			cfg->handle_input(pad, false, [&](usio_btn btn, u16 /*value*/, bool pressed)
+			cfg->handle_input(pad, false, [&](usio_btn btn, pad_button /*pad_btn*/, u16 /*value*/, bool pressed, bool& /*abort*/)
 			{
 				switch (btn)
 				{
@@ -273,7 +273,7 @@ void usb_device_usio::translate_input_taiko()
 void usb_device_usio::translate_input_tekken()
 {
 	std::lock_guard lock(pad::g_pad_mutex);
-	const auto handler = pad::get_current_handler();
+	const auto handler = pad::get_pad_thread();
 
 	std::vector<u8> input_buf(0x180);
 	le_t<u64> digital_input[2]{};
@@ -288,7 +288,7 @@ void usb_device_usio::translate_input_tekken()
 		if (const auto& pad = ::at32(handler->GetPads(), pad_number); (pad->m_port_status & CELL_PAD_STATUS_CONNECTED) && is_input_allowed())
 		{
 			const auto& cfg = ::at32(g_cfg_usio.players, pad_number);
-			cfg->handle_input(pad, false, [&](usio_btn btn, u16 /*value*/, bool pressed)
+			cfg->handle_input(pad, false, [&](usio_btn btn, pad_button /*pad_btn*/, u16 /*value*/, bool pressed, bool& /*abort*/)
 			{
 				switch (btn)
 				{

rpcs3/Emu/NP/np_handler.cpp

@@ -59,11 +59,7 @@ namespace np
 {
 	std::string get_players_history_path()
 	{
-#ifdef _WIN32
+		return fs::get_config_dir(true) + "players_history.yml";
-		return fs::get_config_dir() + "config/players_history.yml";
-#else
-		return fs::get_config_dir() + "players_history.yml";
-#endif
 	}
 
 	std::map<std::string, player_history> load_players_history()
@@ -1440,7 +1436,7 @@ namespace np
 	void np_handler::save_players_history()
 	{
 #ifdef _WIN32
-		const std::string path_to_cfg = fs::get_config_dir() + "config/";
+		const std::string path_to_cfg = fs::get_config_dir(true);
 		if (!fs::create_path(path_to_cfg))
 		{
 			nph_log.error("Could not create path: %s", path_to_cfg);

rpcs3/Emu/NP/rpcn_config.cpp

@@ -34,7 +34,7 @@ void cfg_rpcn::load()
 void cfg_rpcn::save() const
 {
 #ifdef _WIN32
-	const std::string path_to_cfg = fs::get_config_dir() + "config/";
+	const std::string path_to_cfg = fs::get_config_dir(true);
 	if (!fs::create_path(path_to_cfg))
 	{
 		rpcn_log.error("Could not create path: %s", path_to_cfg);
@@ -51,11 +51,7 @@ void cfg_rpcn::save() const
 
 std::string cfg_rpcn::get_path()
 {
-#ifdef _WIN32
+	return fs::get_config_dir(true) + "rpcn.yml";
-	return fs::get_config_dir() + "config/rpcn.yml";
-#else
-	return fs::get_config_dir() + "rpcn.yml";
-#endif
 }
 
 std::string cfg_rpcn::generate_npid()

rpcs3/Emu/NP/upnp_config.cpp

@@ -24,7 +24,7 @@ void cfg_upnp::load()
 void cfg_upnp::save() const
 {
 #ifdef _WIN32
-	const std::string path_to_cfg = fs::get_config_dir() + "config/";
+	const std::string path_to_cfg = fs::get_config_dir(true);
 	if (!fs::create_path(path_to_cfg))
 	{
 		upnp_cfg_log.error("Could not create path: %s", path_to_cfg);
@@ -51,9 +51,5 @@ void cfg_upnp::set_device_url(std::string_view url)
 
 std::string cfg_upnp::get_path()
 {
-#ifdef _WIN32
+	return fs::get_config_dir(true) + "upnp.yml";
-	return fs::get_config_dir() + "config/upnp.yml";
-#else
-	return fs::get_config_dir() + "upnp.yml";
-#endif
 }

rpcs3/Emu/RSX/Common/simple_array.hpp

@@ -427,5 +427,17 @@ namespace rsx
 			}
 			return result;
 		}
+
+		template <typename F, typename U>
+			requires std::is_invocable_r_v<U, F, const U&, const Ty&>
+		U reduce(U initial_value, F&& reducer) const
+		{
+			U accumulate = initial_value;
+			for (auto it = begin(); it != end(); ++it)
+			{
+				accumulate = reducer(accumulate, *it);
+			}
+			return accumulate;
+		}
 	};
 }

rpcs3/Emu/RSX/Common/surface_store.cpp

@@ -18,7 +18,21 @@ namespace rsx
 			case surface_target::surfaces_a_b_c: return{ 0, 1, 2 };
 			case surface_target::surfaces_a_b_c_d: return{ 0, 1, 2, 3 };
 			}
-			fmt::throw_exception("Wrong color_target");
+			fmt::throw_exception("Invalid color target %d", static_cast<int>(color_target));
+		}
+
+		u8 get_mrt_buffers_count(surface_target color_target)
+		{
+			switch (color_target)
+			{
+			case surface_target::none: return 0;
+			case surface_target::surface_a: return 1;
+			case surface_target::surface_b: return 1;
+			case surface_target::surfaces_a_b: return 2;
+			case surface_target::surfaces_a_b_c: return 3;
+			case surface_target::surfaces_a_b_c_d: return 4;
+			}
+			fmt::throw_exception("Invalid color target %d", static_cast<int>(color_target));
 		}
 
 		usz get_aligned_pitch(surface_color_format format, u32 width)

rpcs3/Emu/RSX/Common/surface_store.h

@@ -17,6 +17,7 @@ namespace rsx
 	namespace utility
 	{
 		std::vector<u8> get_rtt_indexes(surface_target color_target);
+		u8 get_mrt_buffers_count(surface_target color_target);
 		usz get_aligned_pitch(surface_color_format format, u32 width);
 		usz get_packed_pitch(surface_color_format format, u32 width);
 	}

rpcs3/Emu/RSX/Core/RSXDrawCommands.cpp

@@ -735,7 +735,7 @@ namespace rsx
 		utils::stream_vector(dst + 4, 0u, fog_mode, std::bit_cast<u32>(wpos_scale), std::bit_cast<u32>(wpos_bias));
 	}
 
-	void draw_command_processor::fill_constants_instancing_buffer(rsx::io_buffer& indirection_table_buf, rsx::io_buffer& constants_data_array_buffer, const VertexProgramBase& prog) const
+	void draw_command_processor::fill_constants_instancing_buffer(rsx::io_buffer& indirection_table_buf, rsx::io_buffer& constants_data_array_buffer, const VertexProgramBase* prog) const
 	{
 		auto& draw_call = REGS(m_ctx)->current_draw_clause;
 
@@ -745,8 +745,9 @@ namespace rsx
 		// Temp indirection table. Used to track "running" updates.
 		rsx::simple_array<u32> instancing_indirection_table;
 		// indirection table size
-		const auto reloc_table = prog.has_indexed_constants ? decltype(prog.constant_ids){} : prog.constant_ids;
+		const auto full_reupload = !prog || prog->has_indexed_constants;
-		const auto redirection_table_size = prog.has_indexed_constants ? 468u : ::size32(prog.constant_ids);
+		const auto reloc_table = full_reupload ? decltype(prog->constant_ids){} : prog->constant_ids;
+		const auto redirection_table_size = full_reupload ? 468u : ::size32(prog->constant_ids);
 		instancing_indirection_table.resize(redirection_table_size);
 
 		// Temp constants data
@@ -787,9 +788,9 @@ namespace rsx
 				continue;
 			}
 
-			const int translated_offset = prog.has_indexed_constants
+			const int translated_offset = full_reupload
 				? instance_config.patch_load_offset
-				: prog.TranslateConstantsRange(instance_config.patch_load_offset, instance_config.patch_load_count);
+				: prog->translate_constants_range(instance_config.patch_load_offset, instance_config.patch_load_count);
 
 			if (translated_offset >= 0)
 			{
@@ -809,14 +810,14 @@ namespace rsx
 				continue;
 			}
 
-			ensure(!prog.has_indexed_constants);
+			ensure(!full_reupload);
 
 			// Sparse update. Update records individually instead of bulk
 			// FIXME: Range batching optimization
 			const auto load_end = instance_config.patch_load_offset + instance_config.patch_load_count;
 			for (u32 i = 0; i < redirection_table_size; ++i)
 			{
-				const auto read_index = prog.constant_ids[i];
+				const auto read_index = prog->constant_ids[i];
 				if (read_index < instance_config.patch_load_offset || read_index >= load_end)
 				{
 					// Reading outside "hot" range.

rpcs3/Emu/RSX/Core/RSXDrawCommands.h

@@ -105,6 +105,6 @@ namespace rsx
 
 		// Fill instancing buffers. A single iobuf is used for both. 256byte alignment enforced to allow global bind
 		// Returns offsets to the index redirection lookup table and constants field array
-		void fill_constants_instancing_buffer(rsx::io_buffer& indirection_table_buf, rsx::io_buffer& constants_data_array_buffer, const VertexProgramBase& prog) const;
+		void fill_constants_instancing_buffer(rsx::io_buffer& indirection_table_buf, rsx::io_buffer& constants_data_array_buffer, const VertexProgramBase* prog) const;
 	};
 }

rpcs3/Emu/RSX/GL/GLGSRender.cpp

@@ -39,7 +39,7 @@ u64 GLGSRender::get_cycles()
 
 GLGSRender::GLGSRender(utils::serial* ar) noexcept : GSRender(ar)
 {
-	m_shaders_cache = std::make_unique<gl::shader_cache>(m_prog_buffer, "opengl", "v1.94");
+	m_shaders_cache = std::make_unique<gl::shader_cache>(m_prog_buffer, "opengl", "v1.95");
 
 	if (g_cfg.video.disable_vertex_cache)
 		m_vertex_cache = std::make_unique<gl::null_vertex_cache>();
@@ -52,6 +52,14 @@ GLGSRender::GLGSRender(utils::serial* ar) noexcept : GSRender(ar)
 	backend_config.supports_normalized_barycentrics = true;
 }
 
+GLGSRender::~GLGSRender()
+{
+	if (m_frame)
+	{
+		m_frame->reset();
+	}
+}
+
 extern CellGcmContextData current_context;
 
 void GLGSRender::set_viewport()
@@ -870,7 +878,7 @@ void GLGSRender::load_program_env()
 		}
 	}
 
-	if (update_fragment_constants && !update_instruction_buffers)
+	if (update_fragment_constants && !m_shader_interpreter.is_interpreter(m_program))
 	{
 		// Fragment constants
 		auto mapping = m_fragment_constants_buffer->alloc_from_heap(fragment_constants_size, m_uniform_buffer_offset_align);
@@ -970,12 +978,23 @@ void GLGSRender::load_program_env()
 		}
 	}
 
-	m_graphics_state.clear(
+	rsx::flags32_t handled_flags =
 		rsx::pipeline_state::fragment_state_dirty |
 		rsx::pipeline_state::vertex_state_dirty |
 		rsx::pipeline_state::transform_constants_dirty |
-		rsx::pipeline_state::fragment_constants_dirty |
+		rsx::pipeline_state::fragment_texture_state_dirty;
-		rsx::pipeline_state::fragment_texture_state_dirty);
+
+	if (update_fragment_constants && !m_shader_interpreter.is_interpreter(m_program))
+	{
+		handled_flags |= rsx::pipeline_state::fragment_constants_dirty;
+	}
+
+	m_graphics_state.clear(handled_flags);
+}
+
+bool GLGSRender::is_current_program_interpreted() const
+{
+	return m_program && m_shader_interpreter.is_interpreter(m_program);
 }
 
 void GLGSRender::upload_transform_constants(const rsx::io_buffer& buffer)
@@ -1026,13 +1045,19 @@ void GLGSRender::update_vertex_env(const gl::vertex_upload_info& upload_info)
 
 void GLGSRender::patch_transform_constants(rsx::context* ctx, u32 index, u32 count)
 {
-	if (!m_vertex_prog)
+	if (!m_program || !m_vertex_prog)
 	{
 		// Shouldn't be reachable, but handle it correctly anyway
 		m_graphics_state |= rsx::pipeline_state::transform_constants_dirty;
 		return;
 	}
 
+	if (!m_vertex_prog->overlaps_constants_range(index, count))
+	{
+		// Nothing meaningful to us
+		return;
+	}
+
 	std::pair<u32, u32> data_range {};
 	void* data_source = nullptr;
 	const auto bound_range = m_transform_constants_buffer->bound_range();
@@ -1046,7 +1071,7 @@ void GLGSRender::patch_transform_constants(rsx::context* ctx, u32 index, u32 cou
 		data_range = { bound_range.first + byte_offset, byte_count};
 		data_source = &REGS(ctx)->transform_constants[index];
 	}
-	else if (auto xform_id = m_vertex_prog->TranslateConstantsRange(index, count); xform_id >= 0)
+	else if (auto xform_id = m_vertex_prog->translate_constants_range(index, count); xform_id >= 0)
 	{
 		const auto write_offset = xform_id * 16;
 		const auto byte_count = count * 16;

rpcs3/Emu/RSX/GL/GLGSRender.h

@@ -159,6 +159,7 @@ public:
 
 	GLGSRender(utils::serial* ar) noexcept;
 	GLGSRender() noexcept : GLGSRender(nullptr) {}
+	virtual ~GLGSRender();
 
 private:
 
@@ -205,6 +206,9 @@ public:
 	// GRAPH backend
 	void patch_transform_constants(rsx::context* ctx, u32 index, u32 count) override;
 
+	// Misc
+	bool is_current_program_interpreted() const override;
+
 protected:
 	void clear_surface(u32 arg) override;
 	void begin() override;

rpcs3/Emu/RSX/GL/GLShaderInterpreter.cpp

@@ -347,7 +347,7 @@ namespace gl
 		return data;
 	}
 
-	bool shader_interpreter::is_interpreter(const glsl::program* program)
+	bool shader_interpreter::is_interpreter(const glsl::program* program) const
 	{
 		return (program == &m_current_interpreter->prog);
 	}

rpcs3/Emu/RSX/GL/GLShaderInterpreter.h

@@ -84,6 +84,6 @@ namespace gl
 		void update_fragment_textures(const std::array<std::unique_ptr<rsx::sampled_image_descriptor_base>, 16>& descriptors, u16 reference_mask, u32* out);
 
 		glsl::program* get(const interpreter::program_metadata& fp_metadata);
-		bool is_interpreter(const glsl::program* program);
+		bool is_interpreter(const glsl::program* program) const;
 	};
 }

rpcs3/Emu/RSX/GSFrameBase.h

@@ -14,6 +14,7 @@ public:
 	virtual ~GSFrameBase() = default;
 
 	virtual void close() = 0;
+	virtual void reset() = 0;
 	virtual bool shown() = 0;
 	virtual void hide() = 0;
 	virtual void show() = 0;

rpcs3/Emu/RSX/GSRender.cpp

@@ -18,7 +18,7 @@ GSRender::~GSRender()
 {
 	m_context = nullptr;
 
-	if (m_frame)
+	if (m_frame && !m_continuous_mode)
 	{
 		m_frame->close();
 	}
@@ -38,8 +38,11 @@ void GSRender::on_exit()
 	rsx::thread::on_exit();
 
 	if (m_frame)
+	{
+		if (!m_continuous_mode)
 		{
 			m_frame->hide();
+		}
 		m_frame->delete_context(m_context);
 		m_context = nullptr;
 	}

rpcs3/Emu/RSX/GSRender.h

@@ -21,12 +21,15 @@ class GSRender : public rsx::thread
 protected:
 	GSFrameBase* m_frame;
 	draw_context_t m_context = nullptr;
+	bool m_continuous_mode = false;
 
 public:
 	~GSRender() override;
 
 	GSRender(utils::serial* ar) noexcept;
 
+	void set_continuous_mode(bool continuous_mode) { m_continuous_mode = continuous_mode; }
+
 	void on_init_thread() override;
 	void on_exit() override;
 

rpcs3/Emu/RSX/Overlays/HomeMenu/overlay_home_menu_main_menu.cpp

@@ -114,6 +114,8 @@ namespace rsx
 
 				Emu.CallFromMainThread([]()
 				{
+					// Make sure we keep the game window opened
+					Emu.SetContinuousMode(true);
 					Emu.Restart(false);
 				});
 				return page_navigation::exit;

rpcs3/Emu/RSX/Overlays/HomeMenu/overlay_home_menu_savestate.cpp

@@ -26,6 +26,9 @@ namespace rsx
 					if (!suspend_mode)
 					{
 						Emu.after_kill_callback = []() { Emu.Restart(); };
+
+						// Make sure we keep the game window opened
+						Emu.SetContinuousMode(true);
 					}
 					Emu.Kill(false, true);
 				});

rpcs3/Emu/RSX/Overlays/Trophies/overlay_trophy_list_dialog.cpp

@@ -265,7 +265,6 @@ namespace rsx
 			fade_animation.end = color4f(1.f);
 			fade_animation.active = true;
 
-			this->on_close = std::move(on_close);
 			visible = true;
 
 			const auto notify = std::make_shared<atomic_t<u32>>(0);

rpcs3/Emu/RSX/Overlays/overlays.cpp

@@ -193,7 +193,7 @@ namespace rsx
 
 				// Get gamepad input
 				std::lock_guard lock(pad::g_pad_mutex);
-				const auto handler = pad::get_current_handler();
+				const auto handler = pad::get_pad_thread();
 				const PadInfo& rinfo = handler->GetInfo();
 
 				const bool ignore_gamepad_input = (!rinfo.now_connect || !input::g_pads_intercepted);

rpcs3/Emu/RSX/Program/FragmentProgramDecompiler.cpp

@@ -22,6 +22,15 @@ namespace rsx
 
 using namespace rsx::fragment_program;
 
+// SIMD vector lanes
+enum VectorLane : u8
+{
+	X = 0,
+	Y = 1,
+	Z = 2,
+	W = 3,
+};
+
 FragmentProgramDecompiler::FragmentProgramDecompiler(const RSXFragmentProgram &prog, u32& size)
 	: m_size(size)
 	, m_prog(prog)
@@ -141,8 +150,7 @@ void FragmentProgramDecompiler::SetDst(std::string code, u32 flags)
 		AddCode(m_parr.AddParam(PF_PARAM_NONE, getFloatTypeName(4), "cc" + std::to_string(src0.cond_mod_reg_index)) + "$m = " + dest + ";");
 	}
 
-	u32 reg_index = dst.fp16 ? dst.dest_reg >> 1 : dst.dest_reg;
+	const u32 reg_index = dst.fp16 ? (dst.dest_reg >> 1) : dst.dest_reg;
-
 	ensure(reg_index < temp_registers.size());
 
 	if (dst.opcode == RSX_FP_OPCODE_MOV &&
@@ -754,14 +762,26 @@ std::string FragmentProgramDecompiler::BuildCode()
 	const std::string init_value = float4_type + "(0.)";
 	std::array<std::string, 4> output_register_names;
 	std::array<u32, 4> ouput_register_indices = { 0, 2, 3, 4 };
-	bool shader_is_valid = false;
+
+	// Holder for any "cleanup" before exiting main
+	std::stringstream main_epilogue;
 
 	// Check depth export
 	if (m_ctrl & CELL_GCM_SHADER_CONTROL_DEPTH_EXPORT)
 	{
 		// Hw tests show that the depth export register is default-initialized to 0 and not wpos.z!!
 		m_parr.AddParam(PF_PARAM_NONE, getFloatTypeName(4), "r1", init_value);
-		shader_is_valid = (!!temp_registers[1].h1_writes);
+
+		auto& r1 = temp_registers[1];
+		if (r1.requires_gather(VectorLane::Z))
+		{
+			// r1.zw was not written to
+			properties.has_gather_op = true;
+			main_epilogue << "	r1.z = " << float4_type << r1.gather_r() << ".z;\n";
+
+			// Emit debug warning. Useful to diagnose regressions, but should be removed in future.
+			rsx_log.warning("ROP reads from shader depth without writing to it. Final value will be gathered.");
+		}
 	}
 
 	// Add the color output registers. They are statically written to and have guaranteed initialization (except r1.z which == wpos.z)
@@ -775,28 +795,49 @@ std::string FragmentProgramDecompiler::BuildCode()
 		output_register_names = { "h0", "h4", "h6", "h8" };
 	}
 
-	for (int n = 0; n < 4; ++n)
+	for (u32 n = 0; n < 4; ++n)
 	{
-		if (!m_parr.HasParam(PF_PARAM_NONE, float4_type, output_register_names[n]))
+		const auto& reg_name = output_register_names[n];
+		if (!m_parr.HasParam(PF_PARAM_NONE, float4_type, reg_name))
 		{
-			m_parr.AddParam(PF_PARAM_NONE, float4_type, output_register_names[n], init_value);
+			m_parr.AddParam(PF_PARAM_NONE, float4_type, reg_name, init_value);
+		}
+
+		if (n >= m_prog.mrt_buffers_count)
+		{
+			// Skip gather
 			continue;
 		}
 
 		const auto block_index = ouput_register_indices[n];
-		shader_is_valid |= (!!temp_registers[block_index].h0_writes);
+		auto& r = temp_registers[block_index];
+
+		if (fp16_out)
+		{
+			// Check if we need a split/extract op
+			if (r.requires_split(0))
+			{
+				main_epilogue << "	" << reg_name << " = " << float4_type << r.split_h0() << ";\n";
+
+				// Emit debug warning. Useful to diagnose regressions, but should be removed in future.
+				rsx_log.warning("ROP reads from %s without writing to it. Final value will be extracted from the 32-bit register.", reg_name);
 			}
 
-	if (!shader_is_valid)
+			continue;
-	{
-		properties.has_no_output = true;
-
-		if (!properties.has_discard_op)
-		{
-			// NOTE: Discard operation overrides output
-			rsx_log.warning("Shader does not write to any output register and will be NOPed");
-			main = "/*" + main + "*/";
 		}
+
+		if (!r.requires_gather128())
+		{
+			// Nothing to do
+			continue;
+		}
+
+		// We need to gather the data from existing registers
+		main_epilogue << "	" << reg_name << " = " << float4_type << r.gather_r() << ";\n";
+		properties.has_gather_op = true;
+
+		// Emit debug warning. Useful to diagnose regressions, but should be removed in future.
+		rsx_log.warning("ROP reads from %s without writing to it. Final value will be gathered.", reg_name);
 	}
 
 	if (properties.has_dynamic_register_load)
@@ -822,6 +863,9 @@ std::string FragmentProgramDecompiler::BuildCode()
 		OS << "#endif\n";
 		OS << "	discard;\n";
 		OS << "}\n";
+
+		// Don't consume any args
+		m_parr.Clear();
 		return OS.str();
 	}
 
@@ -1019,6 +1063,12 @@ std::string FragmentProgramDecompiler::BuildCode()
 
 	insertMainStart(OS);
 	OS << main << std::endl;
+
+	if (const auto epilogue = main_epilogue.str(); !epilogue.empty())
+	{
+		OS << "	// Epilogue\n";
+		OS << epilogue << std::endl;
+	}
 	insertMainEnd(OS);
 
 	return OS.str();
@@ -1360,12 +1410,12 @@ std::string FragmentProgramDecompiler::Decompile()
 
 		switch (opcode)
 		{
-		case RSX_FP_OPCODE_NOP: break;
+		case RSX_FP_OPCODE_NOP:
+			break;
 		case RSX_FP_OPCODE_KIL:
 			properties.has_discard_op = true;
 			AddFlowOp("_kill()");
 			break;
-
 		default:
 			int prev_force_unit = forced_unit;
 

rpcs3/Emu/RSX/Program/FragmentProgramDecompiler.h

@@ -1,116 +1,10 @@
 #pragma once
 #include "ShaderParam.h"
+#include "FragmentProgramRegister.h"
 #include "RSXFragmentProgram.h"
+
 #include <sstream>
 
-// Helper for GPR occupancy tracking
-struct temp_register
-{
-	bool aliased_r0 = false;
-	bool aliased_h0 = false;
-	bool aliased_h1 = false;
-	bool last_write_half[4] = { false, false, false, false };
-
-	u32 real_index = -1;
-
-	u32 h0_writes = 0u;   // Number of writes to the first 64-bits of the register
-	u32 h1_writes = 0u;   // Number of writes to the last 64-bits of the register
-
-	void tag(u32 index, bool half_register, bool x, bool y, bool z, bool w)
-	{
-		if (half_register)
-		{
-			if (index & 1)
-			{
-				if (x) last_write_half[2] = true;
-				if (y) last_write_half[2] = true;
-				if (z) last_write_half[3] = true;
-				if (w) last_write_half[3] = true;
-
-				aliased_h1 = true;
-				h1_writes++;
-			}
-			else
-			{
-				if (x) last_write_half[0] = true;
-				if (y) last_write_half[0] = true;
-				if (z) last_write_half[1] = true;
-				if (w) last_write_half[1] = true;
-
-				aliased_h0 = true;
-				h0_writes++;
-			}
-		}
-		else
-		{
-			if (x) last_write_half[0] = false;
-			if (y) last_write_half[1] = false;
-			if (z) last_write_half[2] = false;
-			if (w) last_write_half[3] = false;
-
-			aliased_r0 = true;
-
-			h0_writes++;
-			h1_writes++;
-		}
-
-		if (real_index == umax)
-		{
-			if (half_register)
-				real_index = index >> 1;
-			else
-				real_index = index;
-		}
-	}
-
-	bool requires_gather(u8 channel) const
-	{
-		//Data fetched from the single precision register requires merging of the two half registers
-		ensure(channel < 4);
-		if (aliased_h0 && channel < 2)
-		{
-			return last_write_half[channel];
-		}
-
-		if (aliased_h1 && channel > 1)
-		{
-			return last_write_half[channel];
-		}
-
-		return false;
-	}
-
-	bool requires_split(u32 /*index*/) const
-	{
-		//Data fetched from any of the two half registers requires sync with the full register
-		if (!(last_write_half[0] || last_write_half[1]) && aliased_r0)
-		{
-			//r0 has been written to
-			//TODO: Check for specific elements in real32 register
-			return true;
-		}
-
-		return false;
-	}
-
-	std::string gather_r() const
-	{
-		std::string h0 = "h" + std::to_string(real_index << 1);
-		std::string h1 = "h" + std::to_string(real_index << 1 | 1);
-		std::string reg = "r" + std::to_string(real_index);
-		std::string ret = "//Invalid gather";
-
-		if (aliased_h0 && aliased_h1)
-			ret = "(gather(" + h0 + ", " + h1 + "))";
-		else if (aliased_h0)
-			ret = "(gather(" + h0 + "), " + reg + ".zw)";
-		else if (aliased_h1)
-			ret = "(" + reg + ".xy, gather(" + h1 + "))";
-
-		return ret;
-	}
-};
-
 /**
  * This class is used to translate RSX Fragment program to GLSL/HLSL code
  * Backend with text based shader can subclass this class and implement :
@@ -157,7 +51,7 @@ class FragmentProgramDecompiler
 
 	bool m_is_valid_ucode = true;
 
-	std::array<temp_register, 64> temp_registers;
+	std::array<rsx::MixedPrecisionRegister, 64> temp_registers;
 
 	std::string GetMask() const;
 

rpcs3/Emu/RSX/Program/FragmentProgramRegister.cpp

@@ -0,0 +1,196 @@
+#include "stdafx.h"
+#include "FragmentProgramRegister.h"
+
+namespace rsx
+{
+	MixedPrecisionRegister::MixedPrecisionRegister()
+	{
+		std::fill(content_mask.begin(), content_mask.end(), data_type_bits::undefined);
+	}
+
+	void MixedPrecisionRegister::tag_h0(bool x, bool y, bool z, bool w)
+	{
+		if (x) content_mask[0] = data_type_bits::f16;
+		if (y) content_mask[1] = data_type_bits::f16;
+		if (z) content_mask[2] = data_type_bits::f16;
+		if (w) content_mask[3] = data_type_bits::f16;
+	}
+
+	void MixedPrecisionRegister::tag_h1(bool x, bool y, bool z, bool w)
+	{
+		if (x) content_mask[4] = data_type_bits::f16;
+		if (y) content_mask[5] = data_type_bits::f16;
+		if (z) content_mask[6] = data_type_bits::f16;
+		if (w) content_mask[7] = data_type_bits::f16;
+	}
+
+	void MixedPrecisionRegister::tag_r(bool x, bool y, bool z, bool w)
+	{
+		if (x) content_mask[0] = content_mask[1] = data_type_bits::f32;
+		if (y) content_mask[2] = content_mask[3] = data_type_bits::f32;
+		if (z) content_mask[4] = content_mask[5] = data_type_bits::f32;
+		if (w) content_mask[6] = content_mask[7] = data_type_bits::f32;
+	}
+
+	void MixedPrecisionRegister::tag(u32 index, bool is_fp16, bool x, bool y, bool z, bool w)
+	{
+		if (file_index == umax)
+		{
+			// First-time use. Initialize...
+			const u32 real_index = is_fp16 ? (index >> 1) : index;
+			file_index = real_index;
+		}
+
+		if (is_fp16)
+		{
+			ensure((index / 2) == file_index);
+
+			if (index & 1)
+			{
+				tag_h1(x, y, z, w);
+				return;
+			}
+
+			tag_h0(x, y, z, w);
+			return;
+		}
+
+		tag_r(x, y, z, w);
+	}
+
+	std::string MixedPrecisionRegister::gather_r() const
+	{
+		const auto half_index = file_index << 1;
+		const std::string reg = "r" + std::to_string(file_index);
+		const std::string gather_half_regs[] = {
+			"gather(h" + std::to_string(half_index) + ")",
+			"gather(h" + std::to_string(half_index + 1) + ")"
+		};
+
+		std::string outputs[4];
+		for (int ch = 0; ch < 4; ++ch)
+		{
+			// FIXME: This approach ignores mixed register bits. Not ideal!!!!
+			const auto channel0 = content_mask[ch * 2];
+			const auto is_fp16_ch = channel0 == content_mask[ch * 2 + 1] && channel0 == data_type_bits::f16;
+			outputs[ch] = is_fp16_ch ? gather_half_regs[ch / 2] : reg;
+		}
+
+		// Grouping. Only replace relevant bits...
+		if (outputs[0] == outputs[1]) outputs[0] = "";
+		if (outputs[2] == outputs[3]) outputs[2] = "";
+
+		// Assemble
+		bool group = false;
+		std::string result = "";
+		constexpr std::string_view swz_mask = "xyzw";
+
+		for (int ch = 0; ch < 4; ++ch)
+		{
+			if (outputs[ch].empty())
+			{
+				group = true;
+				continue;
+			}
+
+			if (!result.empty())
+			{
+				result += ", ";
+			}
+
+			if (group)
+			{
+				ensure(ch > 0);
+				group = false;
+
+				if (outputs[ch] == reg)
+				{
+					result += reg + "." + swz_mask[ch - 1] + swz_mask[ch];
+					continue;
+				}
+
+				result += outputs[ch];
+				continue;
+			}
+
+			const int subch = outputs[ch] == reg ? ch : (ch % 2); // Avoid .xyxy.z and other such ugly swizzles
+			result += outputs[ch] + "." + swz_mask[subch];
+		}
+
+		// Optimize dual-gather (128-bit gather) to use special function
+		const std::string double_gather = gather_half_regs[0] + ", " + gather_half_regs[1];
+		if (result == double_gather)
+		{
+			result = "gather(h" + std::to_string(half_index) + ", h" + std::to_string(half_index + 1) + ")";
+		}
+
+		return "(" + result + ")";
+	}
+
+	std::string MixedPrecisionRegister::fetch_halfreg(u32 word_index) const
+	{
+		// Reads half-word 0 (H16x4) from a full real (R32x4) register
+		constexpr std::string_view swz_mask = "xyzw";
+		const std::string reg = "r" + std::to_string(file_index);
+		const std::string hreg = "h" + std::to_string(file_index * 2 + word_index);
+
+		const std::string word0_bits = "floatBitsToUint(" + reg + "." + swz_mask[word_index * 2] + ")";
+		const std::string word1_bits = "floatBitsToUint(" + reg + "." + swz_mask[word_index * 2 + 1] + ")";
+		const std::string words[] = {
+			"unpackHalf2x16(" + word0_bits + ")",
+			"unpackHalf2x16(" + word1_bits + ")"
+		};
+
+		// Assemble
+		std::string outputs[4];
+
+		ensure(word_index <= 1);
+		const int word_offset = word_index * 4;
+		for (int ch = 0; ch < 4; ++ch)
+		{
+			outputs[ch] = content_mask[ch + word_offset] == data_type_bits::f32
+				? words[ch / 2]
+				: hreg;
+		}
+
+		// Grouping. Only replace relevant bits...
+		if (outputs[0] == outputs[1]) outputs[0] = "";
+		if (outputs[2] == outputs[3]) outputs[2] = "";
+
+		// Assemble
+		bool group = false;
+		std::string result = "";
+
+		for (int ch = 0; ch < 4; ++ch)
+		{
+			if (outputs[ch].empty())
+			{
+				group = true;
+				continue;
+			}
+
+			if (!result.empty())
+			{
+				result += ", ";
+			}
+
+			if (group)
+			{
+				ensure(ch > 0);
+				group = false;
+				result += outputs[ch];
+
+				if (outputs[ch] == hreg)
+				{
+					result += std::string(".") + swz_mask[ch - 1] + swz_mask[ch];
+				}
+				continue;
+			}
+
+			const int subch = outputs[ch] == hreg ? ch : (ch % 2); // Avoid .xyxy.z and other such ugly swizzles
+			result += outputs[ch] + "." + swz_mask[subch];
+		}
+
+		return "(" + result + ")";
+	}
+}

rpcs3/Emu/RSX/Program/FragmentProgramRegister.h

@@ -0,0 +1,111 @@
+#pragma once
+
+#include <util/types.hpp>
+
+namespace rsx
+{
+	class MixedPrecisionRegister
+	{
+		enum data_type_bits
+		{
+			undefined = 0,
+			f16 = 1,
+			f32 = 2
+		};
+
+		std::array<data_type_bits, 8> content_mask; // Content details for each half-word
+		u32 file_index = umax;
+
+		void tag_h0(bool x, bool y, bool z, bool w);
+
+		void tag_h1(bool x, bool y, bool z, bool w);
+
+		void tag_r(bool x, bool y, bool z, bool w);
+
+		std::string fetch_halfreg(u32 word_index) const;
+
+	public:
+		MixedPrecisionRegister();
+
+		void tag(u32 index, bool is_fp16, bool x, bool y, bool z, bool w);
+
+		std::string gather_r() const;
+
+		std::string split_h0() const
+		{
+			return fetch_halfreg(0);
+		}
+
+		std::string split_h1() const
+		{
+			return fetch_halfreg(1);
+		}
+
+		// Getters
+
+		// Return true if all values are unwritten to (undefined)
+		bool floating() const
+		{
+			return file_index == umax;
+		}
+
+		// Return true if the first half register is all undefined
+		bool floating_h0() const
+		{
+			return content_mask[0] == content_mask[1] &&
+				content_mask[1] == content_mask[2] &&
+				content_mask[2] == content_mask[3] &&
+				content_mask[3] == data_type_bits::undefined;
+		}
+
+		// Return true if the second half register is all undefined
+		bool floating_h1() const
+		{
+			return content_mask[4] == content_mask[5] &&
+				content_mask[5] == content_mask[6] &&
+				content_mask[6] == content_mask[7] &&
+				content_mask[7] == data_type_bits::undefined;
+		}
+
+		// Return true if any of the half-words are 16-bit
+		bool requires_gather(u8 channel) const
+		{
+			// Data fetched from the single precision register requires merging of the two half registers
+			const auto channel_offset = channel * 2;
+			ensure(channel_offset <= 6);
+
+			return (content_mask[channel_offset] == data_type_bits::f16 || content_mask[channel_offset + 1] == data_type_bits::f16);
+		}
+
+		// Return true if the entire 128-bit register is filled with 2xfp16x4 data words
+		bool requires_gather128() const
+		{
+			// Full 128-bit check
+			for (const auto& ch : content_mask)
+			{
+				if (ch == data_type_bits::f16)
+				{
+					return true;
+				}
+			}
+
+			return false;
+		}
+
+		// Return true if the half-register is polluted with fp32 data
+		bool requires_split(u32 word_index) const
+		{
+			const u32 content_offset = word_index * 4;
+			for (u32 i = 0; i < 4; ++i)
+			{
+				if (content_mask[content_offset + i] == data_type_bits::f32)
+				{
+					return true;
+				}
+			}
+
+			return false;
+		}
+	};
+}
+

rpcs3/Emu/RSX/Program/GLSLInterpreter/FragmentInterpreter.glsl

@@ -86,8 +86,11 @@ layout(location=0) in vec4 in_regs[16];
 #define CELL_GCM_SHADER_CONTROL_DEPTH_EXPORT 0xe
 #define CELL_GCM_SHADER_CONTROL_32_BITS_EXPORTS 0x40
 
-#define GET_BITS(word, offset, count) bitfieldExtract(inst.words[word], offset, count)
+#define GET_BITS(bitfield, offset, count) bitfieldExtract(bitfield, offset, count)
-#define TEST_BIT(word, offset) (GET_BITS(word, offset, 1) > 0)
+#define TEST_BIT(bitfield, offset) (GET_BITS(bitfield, offset, 1) > 0)
+
+#define GET_INST_BITS(word, offset, count) GET_BITS(inst.words[word], offset, count)
+#define TEST_INST_BIT(word, offset) (GET_INST_BITS(word, offset, 1) > 0)
 
 #define select mix
 #define reg_mov(d, s, m) d = select(d, s, m)
@@ -174,7 +177,7 @@ int counter = 0;
 
 vec4 read_src(const in int index)
 {
-	ur0 = GET_BITS(index + 1, 0, 2);
+	ur0 = GET_INST_BITS(index + 1, 0, 2);
 
 	switch (ur0)
 	{
@@ -183,14 +186,14 @@ vec4 read_src(const in int index)
 		switch(index)
 		{
 		case 0:
-			ur1 = GET_BITS(1, 2, 6); break;
+			ur1 = GET_INST_BITS(1, 2, 6); break;
 		case 1:
-			ur1 = GET_BITS(2, 2, 6); break;
+			ur1 = GET_INST_BITS(2, 2, 6); break;
 		case 2:
-			ur1 = GET_BITS(3, 2, 6); break;
+			ur1 = GET_INST_BITS(3, 2, 6); break;
 		}
 
-		if (TEST_BIT(index + 1, 8))
+		if (TEST_INST_BIT(index + 1, 8))
 		{
 			vr0 = regs16[ur1];
 		}
@@ -202,7 +205,7 @@ vec4 read_src(const in int index)
 	}
 	case RSX_FP_REGISTER_TYPE_INPUT:
 	{
-		ur1 = GET_BITS(0, 13, 4);
+		ur1 = GET_INST_BITS(0, 13, 4);
 		switch (ur1)
 		{
 		case 0:
@@ -235,27 +238,27 @@ vec4 read_src(const in int index)
 	}
 	}
 
-	ur1 = GET_BITS(index + 1, 9, 8);
+	ur1 = GET_INST_BITS(index + 1, 9, 8);
 	vr0 = shuffle(vr0, ur1);
 
 	// abs
 	if (index == 0)
 	{
-		if (TEST_BIT(1, 29)) vr0 = abs(vr0);
+		if (TEST_INST_BIT(1, 29)) vr0 = abs(vr0);
 	}
 	else
 	{
 		ur1 = index + 1;
-		if (TEST_BIT(ur1, 18)) vr0 = abs(vr0);
+		if (TEST_INST_BIT(ur1, 18)) vr0 = abs(vr0);
 	}
 
 	// neg
-	return (TEST_BIT(index + 1, 17))? -vr0 : vr0;
+	return (TEST_INST_BIT(index + 1, 17))? -vr0 : vr0;
 }
 
 vec4 read_cond()
 {
-	return shuffle(cc[GET_BITS(1, 31, 1)], GET_BITS(1, 21, 8));
+	return shuffle(cc[GET_INST_BITS(1, 31, 1)], GET_INST_BITS(1, 21, 8));
 }
 
 bvec4 decode_cond(const in uint mode, const in vec4 cond)
@@ -283,7 +286,7 @@ bvec4 decode_cond(const in uint mode, const in vec4 cond)
 
 bool check_cond()
 {
-	ur0 = GET_BITS(1, 18, 3);
+	ur0 = GET_INST_BITS(1, 18, 3);
 	if (ur0 == 0x7)
 	{
 		return true;
@@ -351,14 +354,14 @@ vec3 _texcoord_xform(const in vec3 coord, const in sampler_info params)
 
 vec4 _texture(in vec4 coord, float bias)
 {
-	ur0 = GET_BITS(0, 17, 4);
+	ur0 = GET_INST_BITS(0, 17, 4);
 	if (!IS_TEXTURE_RESIDENT(ur0))
 	{
 		return vr_zero;
 	}
 
 	ur1 = ur0 + ur0;
-	const uint type = bitfieldExtract(texture_control, int(ur1), 2);
+	const uint type = GET_BITS(texture_control, int(ur1), 2);
 
 	switch (type)
 	{
@@ -380,7 +383,7 @@ vec4 _texture(in vec4 coord, float bias)
 		break;
 	}
 
-	if (TEST_BIT(0, 21))
+	if (TEST_INST_BIT(0, 21))
 	{
 		vr0 = vr0 * 2. - 1.;
 	}
@@ -390,14 +393,14 @@ vec4 _texture(in vec4 coord, float bias)
 
 vec4 _textureLod(in vec4 coord, float lod)
 {
-	ur0 = GET_BITS(0, 17, 4);
+	ur0 = GET_INST_BITS(0, 17, 4);
 	if (!IS_TEXTURE_RESIDENT(ur0))
 	{
 		return vr_zero;
 	}
 
 	ur1 = ur0 + ur0;
-	const uint type = bitfieldExtract(texture_control, int(ur1), 2);
+	const uint type = GET_BITS(texture_control, int(ur1), 2);
 
 	switch (type)
 	{
@@ -419,7 +422,7 @@ vec4 _textureLod(in vec4 coord, float lod)
 		break;
 	}
 
-	if (TEST_BIT(0, 21))
+	if (TEST_INST_BIT(0, 21))
 	{
 		// Normal-expand, v = 2v - 1
 		vr0 += vr0;
@@ -436,27 +439,27 @@ void write_dst(const in vec4 value)
 	uvr0 = uvec4(uint(1 << 9), uint(1 << 10), uint(1 << 11), uint(1 << 12));
 	bvr0 = bvec4(uvr0 & inst.words.xxxx);
 
-	if (TEST_BIT(0, 8)) // SET COND
+	if (TEST_INST_BIT(0, 8)) // SET COND
 	{
-		ur0 = GET_BITS(1, 30, 1);
+		ur0 = GET_INST_BITS(1, 30, 1);
 		reg_mov(cc[ur0], value, bvr0);
 	}
 
-	if (TEST_BIT(0, 30)) // NO DEST
+	if (TEST_INST_BIT(0, 30)) // NO DEST
 	{
 		return;
 	}
 
-	ur1 = GET_BITS(2, 28, 3);
+	ur1 = GET_INST_BITS(2, 28, 3);
 	sr0 = modifier_scale[ur1];
 	vr0 = value * sr0;
 
-	if (TEST_BIT(0, 31)) // SAT
+	if (TEST_INST_BIT(0, 31)) // SAT
 	{
 		vr0 = clamp(vr0, 0, 1);
 	}
 
-	ur0 = GET_BITS(1, 18, 3);
+	ur0 = GET_INST_BITS(1, 18, 3);
 	if (ur0 != 0x7)
 	{
 		vr1 = read_cond();
@@ -464,8 +467,8 @@ void write_dst(const in vec4 value)
 		bvr0 = bvec4(uvec4(bvr0) & uvec4(bvr1));
 	}
 
-	ur1 = GET_BITS(0, 1, 6);
+	ur1 = GET_INST_BITS(0, 1, 6);
-	if (TEST_BIT(0, 7))
+	if (TEST_INST_BIT(0, 7))
 	{
 		reg_mov(regs16[ur1], vr0, bvr0);
 	}
@@ -481,7 +484,7 @@ void initialize()
 	// NOTE: Register count is the number of 'full' registers that will be consumed. Hardware seems to do some renaming.
 	// NOTE: Attempting to zero-initialize all the registers will slow things to a crawl!
 
-	uint register_count = bitfieldExtract(shader_control, 24, 6);
+	uint register_count = GET_BITS(shader_control, 24, 6);
 	ur0 = 0, ur1 = 0;
 	while (register_count > 0)
 	{
@@ -587,11 +590,11 @@ void main()
 			((fp_instructions[ip] << 8) & uvec4(0xFF00FF00)) |
 			((fp_instructions[ip] >> 8) & uvec4(0x00FF00FF));
 
-		inst.opcode = GET_BITS(0, 24, 6);
+		inst.opcode = GET_INST_BITS(0, 24, 6);
-		inst.end = TEST_BIT(0, 0);
+		inst.end = TEST_INST_BIT(0, 0);
 
 #ifdef WITH_FLOW_CTRL
-		if (TEST_BIT(2, 31))
+		if (TEST_INST_BIT(2, 31))
 		{
 			// Flow control
 			switch (inst.opcode | (1 << 6))
@@ -623,8 +626,8 @@ void main()
 			case RSX_FP_OPCODE_REP:
 				if (check_cond())
 				{
-					counter = int(GET_BITS(2, 2, 8) - GET_BITS(2, 10, 8));
+					counter = int(GET_INST_BITS(2, 2, 8) - GET_INST_BITS(2, 10, 8));
-					counter /= int(GET_BITS(2, 19, 8));
+					counter /= int(GET_INST_BITS(2, 19, 8));
 					loop_start_addr = ip + 1;
 					loop_end_addr = int(inst.words.w >> 2);
 				}

rpcs3/Emu/RSX/Program/GLSLSnippets/RSXProg/RSXFragmentTextureMSAAOps.glsl

@@ -34,7 +34,8 @@ vec2 texture2DMSCoord(const in vec2 coords, const in uint flags)
 		return coords;
 	}
 
-	const vec2 wrapped_coords = mod(coords, vec2(1.0));
+	const vec2 wrapped_coords_raw = mod(coords, vec2(1.0));
+	const vec2 wrapped_coords = mod(wrapped_coords_raw + vec2(1.0), vec2(1.0));
 	const bvec2 wrap_control_mask = bvec2(uvec2(flags) & uvec2(WRAP_S_MASK, WRAP_T_MASK));
 	return _select(coords, wrapped_coords, wrap_control_mask);
 }

rpcs3/Emu/RSX/Program/ProgramStateCache.cpp

@@ -340,12 +340,16 @@ vertex_program_utils::vertex_program_metadata vertex_program_utils::analyse_vert
 
 usz vertex_program_storage_hash::operator()(const RSXVertexProgram &program) const
 {
-	usz hash = vertex_program_utils::get_vertex_program_ucode_hash(program);
+	const usz ucode_hash = vertex_program_utils::get_vertex_program_ucode_hash(program);
-	hash ^= program.ctrl;
+	const u32 state_params[] =
-	hash ^= program.output_mask;
+	{
-	hash ^= program.texture_state.texture_dimensions;
+		program.ctrl,
-	hash ^= program.texture_state.multisampled_textures;
+		program.output_mask,
-	return hash;
+		program.texture_state.texture_dimensions,
+		program.texture_state.multisampled_textures,
+	};
+	const usz metadata_hash = rpcs3::hash_array(state_params);
+	return rpcs3::hash64(ucode_hash, metadata_hash);
 }
 
 bool vertex_program_compare::operator()(const RSXVertexProgram &binary1, const RSXVertexProgram &binary2) const
@@ -541,24 +545,33 @@ usz fragment_program_utils::get_fragment_program_ucode_hash(const RSXFragmentPro
 
 usz fragment_program_storage_hash::operator()(const RSXFragmentProgram& program) const
 {
-	usz hash = fragment_program_utils::get_fragment_program_ucode_hash(program);
+	const usz ucode_hash = fragment_program_utils::get_fragment_program_ucode_hash(program);
-	hash ^= program.ctrl;
+	const u32 state_params[] =
-	hash ^= +program.two_sided_lighting;
+	{
-	hash ^= program.texture_state.texture_dimensions;
+		program.ctrl,
-	hash ^= program.texture_state.shadow_textures;
+		program.two_sided_lighting ? 1u : 0u,
-	hash ^= program.texture_state.redirected_textures;
+		program.texture_state.texture_dimensions,
-	hash ^= program.texture_state.multisampled_textures;
+		program.texture_state.shadow_textures,
-	hash ^= program.texcoord_control_mask;
+		program.texture_state.redirected_textures,
-
+		program.texture_state.multisampled_textures,
-	return hash;
+		program.texcoord_control_mask,
+		program.mrt_buffers_count
+	};
+	const usz metadata_hash = rpcs3::hash_array(state_params);
+	return rpcs3::hash64(ucode_hash, metadata_hash);
 }
 
 bool fragment_program_compare::operator()(const RSXFragmentProgram& binary1, const RSXFragmentProgram& binary2) const
 {
-	if (binary1.ctrl != binary2.ctrl || binary1.texture_state != binary2.texture_state ||
+	if (binary1.ucode_length != binary2.ucode_length ||
+		binary1.ctrl != binary2.ctrl ||
+		binary1.texture_state != binary2.texture_state ||
 		binary1.texcoord_control_mask != binary2.texcoord_control_mask ||
-		binary1.two_sided_lighting != binary2.two_sided_lighting)
+		binary1.two_sided_lighting != binary2.two_sided_lighting ||
+		binary1.mrt_buffers_count != binary2.mrt_buffers_count)
+	{
 		return false;
+	}
 
 	const void* instBuffer1 = binary1.get_data();
 	const void* instBuffer2 = binary2.get_data();
@@ -569,7 +582,9 @@ bool fragment_program_compare::operator()(const RSXFragmentProgram& binary1, con
 		const auto inst2 = v128::loadu(instBuffer2, instIndex);
 
 		if (inst1._u ^ inst2._u)
+		{
 			return false;
+		}
 
 		instIndex++;
 		// Skip constants
@@ -578,10 +593,12 @@ bool fragment_program_compare::operator()(const RSXFragmentProgram& binary1, con
 			fragment_program_utils::is_constant(inst1._u32[3]))
 			instIndex++;
 
-		bool end = ((inst1._u32[0] >> 8) & 0x1) && ((inst2._u32[0] >> 8) & 0x1);
+		const bool end = ((inst1._u32[0] >> 8) & 0x1) && ((inst2._u32[0] >> 8) & 0x1);
 		if (end)
+		{
 			return true;
 		}
+	}
 }
 
 namespace rsx

rpcs3/Emu/RSX/Program/ProgramStateCache.h

@@ -295,8 +295,8 @@ public:
 		Args&& ...args
 	)
 	{
-		const auto &vp_search = search_vertex_program(vertexShader);
+		const auto& vp_search = search_vertex_program(vertexShader);
-		const auto &fp_search = search_fragment_program(fragmentShader);
+		const auto& fp_search = search_fragment_program(fragmentShader);
 
 		const bool already_existing_fragment_program = std::get<1>(fp_search);
 		const bool already_existing_vertex_program = std::get<1>(vp_search);
@@ -385,7 +385,13 @@ public:
 
 	void fill_fragment_constants_buffer(std::span<f32> dst_buffer, const fragment_program_type& fragment_program, const RSXFragmentProgram& rsx_prog, bool sanitize = false) const
 	{
-		ensure((dst_buffer.size_bytes() >= ::narrow<int>(fragment_program.FragmentConstantOffsetCache.size()) * 16u));
+		if (dst_buffer.size_bytes() < (fragment_program.FragmentConstantOffsetCache.size() * 16))
+		{
+			// This can happen if CELL alters the shader after it has been loaded by RSX.
+			rsx_log.error("Insufficient constants buffer size passed to fragment program! Corrupt shader?");
+			return;
+		}
+
 		rsx::write_fragment_constants_to_buffer(dst_buffer, rsx_prog, fragment_program.FragmentConstantOffsetCache, sanitize);
 	}
 

rpcs3/Emu/RSX/Program/RSXFragmentProgram.h

@@ -300,8 +300,10 @@ struct RSXFragmentProgram
 	u32 ucode_length = 0;
 	u32 total_length = 0;
 	u32 ctrl = 0;
-	bool two_sided_lighting = false;
 	u32 texcoord_control_mask = 0;
+	u32 mrt_buffers_count = 0;
+
+	bool two_sided_lighting = false;
 
 	rsx::fragment_program_texture_state texture_state;
 	rsx::fragment_program_texture_config texture_params;

rpcs3/Emu/RSX/Program/RSXVertexProgram.h

@@ -223,10 +223,10 @@ struct RSXVertexProgram
 {
 	std::vector<u32> data;
 	rsx::vertex_program_texture_state texture_state;
-	u32 ctrl;
+	u32 ctrl = 0;
-	u32 output_mask;
+	u32 output_mask = 0;
-	u32 base_address;
+	u32 base_address = 0;
-	u32 entry;
+	u32 entry = 0;
 	std::bitset<rsx::max_vertex_program_instructions> instruction_mask;
 	std::set<u32> jump_table;
 

rpcs3/Emu/RSX/Program/ShaderParam.h

@@ -227,6 +227,14 @@ struct ParamArray
 
 		return name;
 	}
+
+	void Clear()
+	{
+		for (auto& param : params)
+		{
+			param.clear();
+		}
+	}
 };
 
 class ShaderVariable

rpcs3/Emu/RSX/Program/program_util.cpp

@@ -110,7 +110,7 @@ namespace rsx
 			multisampled_textures == other.multisampled_textures;
 	}
 
-	int VertexProgramBase::TranslateConstantsRange(int first_index, int count) const
+	int VertexProgramBase::translate_constants_range(int first_index, int count) const
 	{
 		// The constant ids should be sorted, so just find the first one and check for continuity
 		int index = -1;
@@ -157,4 +157,31 @@ namespace rsx
 		// OOB or partial match
 		return -1;
 	}
+
+	bool VertexProgramBase::overlaps_constants_range(int first_index, int count) const
+	{
+		if (has_indexed_constants)
+		{
+			return true;
+		}
+
+		const int last_index = first_index + count - 1;
+
+		// Early rejection test
+		if (constant_ids.empty() || first_index > constant_ids.back() || last_index < first_index)
+		{
+			return false;
+		}
+
+		// Check for any hits
+		for (auto& idx : constant_ids)
+		{
+			if (idx >= first_index && idx <= last_index)
+			{
+				return true;
+			}
+		}
+
+		return false;
+	}
 }

rpcs3/Emu/RSX/Program/program_util.h

@@ -67,6 +67,9 @@ namespace rsx
 		// Translates an incoming range of constants against our mapping.
 		// If there is no linear mapping available, return -1, otherwise returns the translated index of the first slot
 		// TODO: Move this somewhere else during refactor
-		int TranslateConstantsRange(int first_index, int count) const;
+		int translate_constants_range(int first_index, int count) const;
+
+		// Returns true if this program consumes any constants in the range [first, first + count - 1]
+		bool overlaps_constants_range(int first_index, int count) const;
 	};
 }

rpcs3/Emu/RSX/RSXThread.cpp

@@ -1719,7 +1719,7 @@ namespace rsx
 
 			for (uint i = 0; i < mrt_buffers.size(); ++i)
 			{
-				if (rsx::method_registers.color_write_enabled(i))
+				if (m_ctx->register_state->color_write_enabled(i))
 				{
 					const auto real_index = mrt_buffers[i];
 					m_framebuffer_layout.color_write_enabled[real_index] = true;
@@ -1727,6 +1727,14 @@ namespace rsx
 				}
 			}
 
+			if (::size32(mrt_buffers) != current_fragment_program.mrt_buffers_count &&
+				!m_graphics_state.test(rsx::pipeline_state::fragment_program_dirty) &&
+				!is_current_program_interpreted())
+			{
+				// Notify that we should recompile the FS
+				m_graphics_state |= rsx::pipeline_state::fragment_program_state_dirty;
+			}
+
 			return any_found;
 		};
 
@@ -2038,9 +2046,10 @@ namespace rsx
 
 		m_graphics_state.clear(rsx::pipeline_state::fragment_program_dirty);
 
-		current_fragment_program.ctrl = rsx::method_registers.shader_control() & (CELL_GCM_SHADER_CONTROL_32_BITS_EXPORTS | CELL_GCM_SHADER_CONTROL_DEPTH_EXPORT);
+		current_fragment_program.ctrl = m_ctx->register_state->shader_control() & (CELL_GCM_SHADER_CONTROL_32_BITS_EXPORTS | CELL_GCM_SHADER_CONTROL_DEPTH_EXPORT);
-		current_fragment_program.texcoord_control_mask = rsx::method_registers.texcoord_control_mask();
+		current_fragment_program.texcoord_control_mask = m_ctx->register_state->texcoord_control_mask();
-		current_fragment_program.two_sided_lighting = rsx::method_registers.two_side_light_en();
+		current_fragment_program.two_sided_lighting = m_ctx->register_state->two_side_light_en();
+		current_fragment_program.mrt_buffers_count = rsx::utility::get_mrt_buffers_count(m_ctx->register_state->surface_color_target());
 
 		if (method_registers.current_draw_clause.classify_mode() == primitive_class::polygon)
 		{

rpcs3/Emu/RSX/RSXThread.h

@@ -436,6 +436,8 @@ namespace rsx
 
 		bool is_current_vertex_program_instanced() const { return !!(current_vertex_program.ctrl & RSX_SHADER_CONTROL_INSTANCED_CONSTANTS); }
 
+		virtual bool is_current_program_interpreted() const { return false; }
+
 	public:
 		void reset();
 		void init(u32 ctrlAddress);

rpcs3/Emu/RSX/VK/VKCommonPipelineLayout.cpp

@@ -0,0 +1,160 @@
+#include "stdafx.h"
+#include "vkutils/device.h"
+#include "vkutils/descriptors.h"
+#include "VKCommonPipelineLayout.h"
+#include "VKHelpers.h"
+
+#include "Emu/RSX/Common/simple_array.hpp"
+
+namespace vk
+{
+	rsx::simple_array<VkDescriptorSetLayoutBinding> get_common_binding_table()
+	{
+		const auto& binding_table = vk::get_current_renderer()->get_pipeline_binding_table();
+		rsx::simple_array<VkDescriptorSetLayoutBinding> bindings(binding_table.instancing_constants_buffer_slot + 1);
+
+		u32 idx = 0;
+
+		// Vertex stream, one stream for cacheable data, one stream for transient data
+		for (int i = 0; i < 3; i++)
+		{
+			bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
+			bindings[idx].descriptorCount = 1;
+			bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
+			bindings[idx].binding = binding_table.vertex_buffers_first_bind_slot + i;
+			bindings[idx].pImmutableSamplers = nullptr;
+			idx++;
+		}
+
+		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
+		bindings[idx].descriptorCount = 1;
+		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
+		bindings[idx].binding = binding_table.fragment_constant_buffers_bind_slot;
+		bindings[idx].pImmutableSamplers = nullptr;
+
+		idx++;
+
+		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
+		bindings[idx].descriptorCount = 1;
+		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
+		bindings[idx].binding = binding_table.fragment_state_bind_slot;
+		bindings[idx].pImmutableSamplers = nullptr;
+
+		idx++;
+
+		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
+		bindings[idx].descriptorCount = 1;
+		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
+		bindings[idx].binding = binding_table.fragment_texture_params_bind_slot;
+		bindings[idx].pImmutableSamplers = nullptr;
+
+		idx++;
+
+		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
+		bindings[idx].descriptorCount = 1;
+		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
+		bindings[idx].binding = binding_table.vertex_constant_buffers_bind_slot;
+		bindings[idx].pImmutableSamplers = nullptr;
+
+		idx++;
+
+		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
+		bindings[idx].descriptorCount = 1;
+		bindings[idx].stageFlags = VK_SHADER_STAGE_ALL_GRAPHICS;
+		bindings[idx].binding = binding_table.vertex_params_bind_slot;
+		bindings[idx].pImmutableSamplers = nullptr;
+
+		idx++;
+
+		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
+		bindings[idx].descriptorCount = 1;
+		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
+		bindings[idx].binding = binding_table.conditional_render_predicate_slot;
+		bindings[idx].pImmutableSamplers = nullptr;
+
+		idx++;
+
+		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
+		bindings[idx].descriptorCount = 1;
+		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
+		bindings[idx].binding = binding_table.rasterizer_env_bind_slot;
+		bindings[idx].pImmutableSamplers = nullptr;
+
+		idx++;
+
+		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
+		bindings[idx].descriptorCount = 1;
+		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
+		bindings[idx].binding = binding_table.instancing_lookup_table_bind_slot;
+		bindings[idx].pImmutableSamplers = nullptr;
+
+		idx++;
+
+		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
+		bindings[idx].descriptorCount = 1;
+		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
+		bindings[idx].binding = binding_table.instancing_constants_buffer_slot;
+		bindings[idx].pImmutableSamplers = nullptr;
+
+		idx++;
+
+		return bindings;
+	}
+
+	std::tuple<VkPipelineLayout, VkDescriptorSetLayout> get_common_pipeline_layout(VkDevice dev)
+	{
+		const auto& binding_table = vk::get_current_renderer()->get_pipeline_binding_table();
+		auto bindings = get_common_binding_table();
+		u32 idx = ::size32(bindings);
+
+		bindings.resize(binding_table.total_descriptor_bindings);
+
+		for (auto binding = binding_table.textures_first_bind_slot;
+			binding < binding_table.vertex_textures_first_bind_slot;
+			binding++)
+		{
+			bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
+			bindings[idx].descriptorCount = 1;
+			bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
+			bindings[idx].binding = binding;
+			bindings[idx].pImmutableSamplers = nullptr;
+			idx++;
+		}
+
+		for (int i = 0; i < rsx::limits::vertex_textures_count; i++)
+		{
+			bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
+			bindings[idx].descriptorCount = 1;
+			bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
+			bindings[idx].binding = binding_table.vertex_textures_first_bind_slot + i;
+			bindings[idx].pImmutableSamplers = nullptr;
+			idx++;
+		}
+
+		ensure(idx == binding_table.total_descriptor_bindings);
+
+		std::array<VkPushConstantRange, 1> push_constants;
+		push_constants[0].offset = 0;
+		push_constants[0].size = 16;
+		push_constants[0].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
+
+		if (vk::emulate_conditional_rendering())
+		{
+			// Conditional render toggle
+			push_constants[0].size = 20;
+		}
+
+		const auto set_layout = vk::descriptors::create_layout(bindings);
+
+		VkPipelineLayoutCreateInfo layout_info = {};
+		layout_info.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
+		layout_info.setLayoutCount = 1;
+		layout_info.pSetLayouts = &set_layout;
+		layout_info.pushConstantRangeCount = 1;
+		layout_info.pPushConstantRanges = push_constants.data();
+
+		VkPipelineLayout result;
+		CHECK_RESULT(vkCreatePipelineLayout(dev, &layout_info, nullptr, &result));
+		return std::make_tuple(result, set_layout);
+	}
+}

rpcs3/Emu/RSX/VK/VKCommonPipelineLayout.h

@@ -0,0 +1,14 @@
+#pragma once
+
+#include "vkutils/shared.h"
+#include "Emu/RSX/Common/simple_array.hpp"
+
+namespace vk
+{
+	// Grab standard layout for decompiled RSX programs. Also used by the interpreter.
+	// FIXME: This generates a bloated monstrosity that needs to die.
+	std::tuple<VkPipelineLayout, VkDescriptorSetLayout> get_common_pipeline_layout(VkDevice dev);
+
+	// Returns the standard binding layout without texture slots. Those have special handling depending on the consumer.
+	rsx::simple_array<VkDescriptorSetLayoutBinding> get_common_binding_table();
+}

rpcs3/Emu/RSX/VK/VKGSRender.cpp

@@ -5,6 +5,7 @@
 #include "VKAsyncScheduler.h"
 #include "VKCommandStream.h"
 #include "VKCommonDecompiler.h"
+#include "VKCommonPipelineLayout.h"
 #include "VKCompute.h"
 #include "VKGSRender.h"
 #include "VKHelpers.h"
@@ -401,148 +402,6 @@ namespace vk
 	}
 }
 
-namespace
-{
-	std::tuple<VkPipelineLayout, VkDescriptorSetLayout> get_shared_pipeline_layout(VkDevice dev)
-	{
-		const auto& binding_table = vk::get_current_renderer()->get_pipeline_binding_table();
-		rsx::simple_array<VkDescriptorSetLayoutBinding> bindings(binding_table.total_descriptor_bindings);
-
-		u32 idx = 0;
-
-		// Vertex stream, one stream for cacheable data, one stream for transient data
-		for (int i = 0; i < 3; i++)
-		{
-			bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
-			bindings[idx].descriptorCount = 1;
-			bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-			bindings[idx].binding = binding_table.vertex_buffers_first_bind_slot + i;
-			bindings[idx].pImmutableSamplers = nullptr;
-			idx++;
-		}
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
-		bindings[idx].binding = binding_table.fragment_constant_buffers_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
-		bindings[idx].binding = binding_table.fragment_state_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
-		bindings[idx].binding = binding_table.fragment_texture_params_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-		bindings[idx].binding = binding_table.vertex_constant_buffers_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_ALL_GRAPHICS;
-		bindings[idx].binding = binding_table.vertex_params_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-		bindings[idx].binding = binding_table.conditional_render_predicate_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
-		bindings[idx].binding = binding_table.rasterizer_env_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-		bindings[idx].binding = binding_table.instancing_lookup_table_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-		bindings[idx].binding = binding_table.instancing_constants_buffer_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		for (auto binding = binding_table.textures_first_bind_slot;
-			 binding < binding_table.vertex_textures_first_bind_slot;
-			 binding++)
-		{
-			bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
-			bindings[idx].descriptorCount = 1;
-			bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
-			bindings[idx].binding = binding;
-			bindings[idx].pImmutableSamplers = nullptr;
-			idx++;
-		}
-
-		for (int i = 0; i < rsx::limits::vertex_textures_count; i++)
-		{
-			bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
-			bindings[idx].descriptorCount = 1;
-			bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-			bindings[idx].binding = binding_table.vertex_textures_first_bind_slot + i;
-			bindings[idx].pImmutableSamplers = nullptr;
-			idx++;
-		}
-
-		ensure(idx == binding_table.total_descriptor_bindings);
-
-		std::array<VkPushConstantRange, 1> push_constants;
-		push_constants[0].offset = 0;
-		push_constants[0].size = 16;
-		push_constants[0].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-
-		if (vk::emulate_conditional_rendering())
-		{
-			// Conditional render toggle
-			push_constants[0].size = 20;
-		}
-
-		const auto set_layout = vk::descriptors::create_layout(bindings);
-
-		VkPipelineLayoutCreateInfo layout_info = {};
-		layout_info.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
-		layout_info.setLayoutCount = 1;
-		layout_info.pSetLayouts = &set_layout;
-		layout_info.pushConstantRangeCount = 1;
-		layout_info.pPushConstantRanges = push_constants.data();
-
-		VkPipelineLayout result;
-		CHECK_RESULT(vkCreatePipelineLayout(dev, &layout_info, nullptr, &result));
-		return std::make_tuple(result, set_layout);
-	}
-}
-
 u64 VKGSRender::get_cycles()
 {
 	return thread_ctrl::get_cycles(static_cast<named_thread<VKGSRender>&>(*this));
@@ -633,7 +492,7 @@ VKGSRender::VKGSRender(utils::serial* ar) noexcept : GSRender(ar)
 	m_secondary_cb_list.create(m_secondary_command_buffer_pool, vk::command_buffer::access_type_hint::all);
 
 	//Precalculated stuff
-	std::tie(m_pipeline_layout, m_descriptor_layouts) = get_shared_pipeline_layout(*m_device);
+	std::tie(m_pipeline_layout, m_descriptor_layouts) = vk::get_common_pipeline_layout(*m_device);
 
 	//Occlusion
 	m_occlusion_query_manager = std::make_unique<vk::query_pool_manager>(*m_device, VK_QUERY_TYPE_OCCLUSION, OCCLUSION_MAX_POOL_SIZE);
@@ -688,10 +547,10 @@ VKGSRender::VKGSRender(utils::serial* ar) noexcept : GSRender(ar)
 	}
 
 	// Initialize optional allocation information with placeholders
-	m_vertex_env_buffer_info = { m_vertex_env_ring_info.heap->value, 0, 32 };
+	m_vertex_env_buffer_info = { m_vertex_env_ring_info.heap->value, 0, 16 };
-	m_vertex_constants_buffer_info = { m_transform_constants_ring_info.heap->value, 0, 32 };
+	m_vertex_constants_buffer_info = { m_transform_constants_ring_info.heap->value, 0, 16 };
-	m_fragment_env_buffer_info = { m_fragment_env_ring_info.heap->value, 0, 32 };
+	m_fragment_env_buffer_info = { m_fragment_env_ring_info.heap->value, 0, 16 };
-	m_fragment_texture_params_buffer_info = { m_fragment_texture_params_ring_info.heap->value, 0, 32 };
+	m_fragment_texture_params_buffer_info = { m_fragment_texture_params_ring_info.heap->value, 0, 16 };
 	m_raster_env_buffer_info = { m_raster_env_ring_info.heap->value, 0, 128 };
 
 	const auto limits = m_device->gpu().get_limits();
@@ -730,7 +589,7 @@ VKGSRender::VKGSRender(utils::serial* ar) noexcept : GSRender(ar)
 	else
 		m_vertex_cache = std::make_unique<vk::weak_vertex_cache>();
 
-	m_shaders_cache = std::make_unique<vk::shader_cache>(*m_prog_buffer, "vulkan", "v1.94");
+	m_shaders_cache = std::make_unique<vk::shader_cache>(*m_prog_buffer, "vulkan", "v1.95");
 
 	for (u32 i = 0; i < m_swapchain->get_swap_image_count(); ++i)
 	{
@@ -2192,7 +2051,7 @@ void VKGSRender::load_program_env()
 			return std::make_pair(m_instancing_buffer_ring_info.map(constants_data_table_offset, size), size);
 		});
 
-		m_draw_processor.fill_constants_instancing_buffer(indirection_table_buf, constants_array_buf, *m_vertex_prog);
+		m_draw_processor.fill_constants_instancing_buffer(indirection_table_buf, constants_array_buf, m_vertex_prog);
 		m_instancing_buffer_ring_info.unmap();
 
 		m_instancing_indirection_buffer_info = { m_instancing_buffer_ring_info.heap->value, indirection_table_offset, indirection_table_buf.size() };
@@ -2219,7 +2078,7 @@ void VKGSRender::load_program_env()
 		}
 	}
 
-	if (update_fragment_constants && !update_instruction_buffers)
+	if (update_fragment_constants && !m_shader_interpreter.is_interpreter(m_program))
 	{
 		check_heap_status(VK_HEAP_CHECK_FRAGMENT_CONSTANTS_STORAGE);
 
@@ -2350,9 +2209,9 @@ void VKGSRender::load_program_env()
 	}
 
 	// Clear flags
-	u32 handled_flags = rsx::pipeline_state::fragment_state_dirty |
+	rsx::flags32_t handled_flags =
+		rsx::pipeline_state::fragment_state_dirty |
 		rsx::pipeline_state::vertex_state_dirty |
-		rsx::pipeline_state::fragment_constants_dirty |
 		rsx::pipeline_state::fragment_texture_state_dirty;
 
 	if (!update_instancing_data)
@@ -2360,9 +2219,19 @@ void VKGSRender::load_program_env()
 		handled_flags |= rsx::pipeline_state::transform_constants_dirty;
 	}
 
+	if (update_fragment_constants && !m_shader_interpreter.is_interpreter(m_program))
+	{
+		handled_flags |= rsx::pipeline_state::fragment_constants_dirty;
+	}
+
 	m_graphics_state.clear(handled_flags);
 }
 
+bool VKGSRender::is_current_program_interpreted() const
+{
+	return m_program && m_shader_interpreter.is_interpreter(m_program);
+}
+
 void VKGSRender::upload_transform_constants(const rsx::io_buffer& buffer)
 {
 	const usz transform_constants_size = (!m_vertex_prog || m_vertex_prog->has_indexed_constants) ? 8192 : m_vertex_prog->constant_ids.size() * 16;
@@ -2433,13 +2302,19 @@ void VKGSRender::update_vertex_env(u32 id, const vk::vertex_upload_info& vertex_
 
 void VKGSRender::patch_transform_constants(rsx::context* ctx, u32 index, u32 count)
 {
-	if (!m_vertex_prog)
+	if (!m_program || !m_vertex_prog)
 	{
 		// Shouldn't be reachable, but handle it correctly anyway
 		m_graphics_state |= rsx::pipeline_state::transform_constants_dirty;
 		return;
 	}
 
+	if (!m_vertex_prog->overlaps_constants_range(index, count))
+	{
+		// Nothing meaningful to us
+		return;
+	}
+
 	// Hot-patching transform constants mid-draw (instanced draw)
 	std::pair<VkDeviceSize, VkDeviceSize> data_range;
 	void* data_source = nullptr;
@@ -2453,7 +2328,7 @@ void VKGSRender::patch_transform_constants(rsx::context* ctx, u32 index, u32 cou
 		data_range = { m_vertex_constants_buffer_info.offset + byte_offset, byte_count };
 		data_source = &REGS(ctx)->transform_constants[index];
 	}
-	else if (auto xform_id = m_vertex_prog->TranslateConstantsRange(index, count); xform_id >= 0)
+	else if (auto xform_id = m_vertex_prog->translate_constants_range(index, count); xform_id >= 0)
 	{
 		const auto write_offset = xform_id * 16;
 		const auto byte_count = count * 16;

rpcs3/Emu/RSX/VK/VKGSRender.h

@@ -288,6 +288,9 @@ public:
 	// GRAPH backend
 	void patch_transform_constants(rsx::context* ctx, u32 index, u32 count) override;
 
+	// Misc
+	bool is_current_program_interpreted() const override;
+
 protected:
 	void clear_surface(u32 mask) override;
 	void begin() override;

rpcs3/Emu/RSX/VK/VKPipelineCompiler.cpp

@@ -63,7 +63,7 @@ namespace vk
 			const std::vector<glsl::program_input>& vs_inputs, const std::vector<glsl::program_input>& fs_inputs)
 	{
 		VkPipeline pipeline;
-		CHECK_RESULT(vkCreateGraphicsPipelines(*m_device, nullptr, 1, &create_info, NULL, &pipeline));
+		CHECK_RESULT(vkCreateGraphicsPipelines(*m_device, VK_NULL_HANDLE, 1, &create_info, nullptr, &pipeline));
 		auto result = std::make_unique<vk::glsl::program>(*m_device, pipeline, pipe_layout, vs_inputs, fs_inputs);
 		result->link();
 		return result;

rpcs3/Emu/RSX/VK/VKResolveHelper.h

@@ -387,7 +387,7 @@ namespace vk
 
 	struct stencilonly_unresolve : depth_resolve_base
 	{
-		VkClearRect region{};
+		VkClearRect clear_region{};
 		VkClearAttachment clear_info{};
 
 		stencilonly_unresolve()
@@ -402,8 +402,8 @@ namespace vk
 			renderpass_config.set_depth_mask(false);
 
 			clear_info.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
-			region.baseArrayLayer = 0;
+			clear_region.baseArrayLayer = 0;
-			region.layerCount = 1;
+			clear_region.layerCount = 1;
 
 			static_parameters_width = 3;
 
@@ -425,7 +425,7 @@ namespace vk
 
 		void emit_geometry(vk::command_buffer& cmd) override
 		{
-			vkCmdClearAttachments(cmd, 1, &clear_info, 1, &region);
+			vkCmdClearAttachments(cmd, 1, &clear_info, 1, &clear_region);
 
 			for (s32 write_mask = 0x1; write_mask <= 0x80; write_mask <<= 1)
 			{
@@ -444,8 +444,8 @@ namespace vk
 
 			auto stencil_view = resolve_image->get_view(rsx::default_remap_vector.with_encoding(VK_REMAP_IDENTITY), VK_IMAGE_ASPECT_STENCIL_BIT);
 
-			region.rect.extent.width = resolve_image->width();
+			clear_region.rect.extent.width = msaa_image->width();
-			region.rect.extent.height = resolve_image->height();
+			clear_region.rect.extent.height = msaa_image->height();
 
 			overlay_pass::run(
 				cmd,

rpcs3/Emu/RSX/VK/VKShaderInterpreter.cpp

@@ -1,5 +1,6 @@
 #include "stdafx.h"
 #include "VKShaderInterpreter.h"
+#include "VKCommonPipelineLayout.h"
 #include "VKVertexProgram.h"
 #include "VKFragmentProgram.h"
 #include "VKGSRender.h"
@@ -233,77 +234,12 @@ namespace vk
 	std::pair<VkDescriptorSetLayout, VkPipelineLayout> shader_interpreter::create_layout(VkDevice dev)
 	{
 		const auto& binding_table = vk::get_current_renderer()->get_pipeline_binding_table();
-		rsx::simple_array<VkDescriptorSetLayoutBinding> bindings(binding_table.total_descriptor_bindings);
+		auto bindings = get_common_binding_table();
+		u32 idx = ::size32(bindings);
 
-		u32 idx = 0;
+		bindings.resize(binding_table.total_descriptor_bindings);
-
-		// Vertex stream, one stream for cacheable data, one stream for transient data. Third stream contains vertex layout info
-		for (int i = 0; i < 3; i++)
-		{
-			bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
-			bindings[idx].descriptorCount = 1;
-			bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-			bindings[idx].binding = binding_table.vertex_buffers_first_bind_slot + i;
-			bindings[idx].pImmutableSamplers = nullptr;
-			idx++;
-		}
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
-		bindings[idx].binding = binding_table.fragment_constant_buffers_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
-		bindings[idx].binding = binding_table.fragment_state_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
-		bindings[idx].binding = binding_table.fragment_texture_params_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-		bindings[idx].binding = binding_table.vertex_constant_buffers_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_ALL_GRAPHICS;
-		bindings[idx].binding = binding_table.vertex_params_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
-		bindings[idx].binding = binding_table.conditional_render_predicate_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
-
-		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
-		bindings[idx].descriptorCount = 1;
-		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
-		bindings[idx].binding = binding_table.rasterizer_env_bind_slot;
-		bindings[idx].pImmutableSamplers = nullptr;
-
-		idx++;
 
+		// Texture 1D array
 		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
 		bindings[idx].descriptorCount = 16;
 		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
@@ -313,6 +249,7 @@ namespace vk
 		m_fragment_textures_start = bindings[idx].binding;
 		idx++;
 
+		// Texture 2D array
 		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
 		bindings[idx].descriptorCount = 16;
 		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
@@ -321,6 +258,7 @@ namespace vk
 
 		idx++;
 
+		// Texture 3D array
 		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
 		bindings[idx].descriptorCount = 16;
 		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
@@ -329,6 +267,7 @@ namespace vk
 
 		idx++;
 
+		// Texture CUBE array
 		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
 		bindings[idx].descriptorCount = 16;
 		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
@@ -337,6 +276,7 @@ namespace vk
 
 		idx++;
 
+		// Vertex texture array (2D only)
 		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
 		bindings[idx].descriptorCount = 4;
 		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
@@ -345,6 +285,7 @@ namespace vk
 
 		idx++;
 
+		// Vertex program ucode block
 		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
 		bindings[idx].descriptorCount = 1;
 		bindings[idx].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
@@ -354,6 +295,7 @@ namespace vk
 		m_vertex_instruction_start = bindings[idx].binding;
 		idx++;
 
+		// Fragment program ucode block
 		bindings[idx].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
 		bindings[idx].descriptorCount = 1;
 		bindings[idx].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
@@ -364,6 +306,22 @@ namespace vk
 		idx++;
 		bindings.resize(idx);
 
+		// Compile descriptor pool sizes
+		const u32 num_ubo = bindings.reduce(0, FN(x + (y.descriptorType == VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER ? y.descriptorCount : 0)));
+		const u32 num_texel_buffers = bindings.reduce(0, FN(x + (y.descriptorType == VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER ? y.descriptorCount : 0)));
+		const u32 num_combined_image_sampler = bindings.reduce(0, FN(x + (y.descriptorType == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER ? y.descriptorCount : 0)));
+		const u32 num_ssbo = bindings.reduce(0, FN(x + (y.descriptorType == VK_DESCRIPTOR_TYPE_STORAGE_BUFFER ? y.descriptorCount : 0)));
+
+		ensure(num_ubo > 0 && num_texel_buffers > 0 && num_combined_image_sampler > 0 && num_ssbo > 0);
+
+		m_descriptor_pool_sizes =
+		{
+			{ VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER , num_ubo },
+			{ VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER , num_texel_buffers },
+			{ VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER , num_combined_image_sampler },
+			{ VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, num_ssbo }
+		};
+
 		std::array<VkPushConstantRange, 1> push_constants;
 		push_constants[0].offset = 0;
 		push_constants[0].size = 16;
@@ -392,16 +350,7 @@ namespace vk
 	void shader_interpreter::create_descriptor_pools(const vk::render_device& dev)
 	{
 		const auto max_draw_calls = dev.get_descriptor_max_draw_calls();
-
+		m_descriptor_pool.create(dev, m_descriptor_pool_sizes, max_draw_calls);
-		rsx::simple_array<VkDescriptorPoolSize> sizes =
-		{
-			{ VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER , 6 },
-			{ VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER , 3 },
-			{ VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER , 68 },
-			{ VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, 3 }
-		};
-
-		m_descriptor_pool.create(dev, sizes, max_draw_calls);
 	}
 
 	void shader_interpreter::init(const vk::render_device& dev)
@@ -410,6 +359,7 @@ namespace vk
 		std::tie(m_shared_descriptor_layout, m_shared_pipeline_layout) = create_layout(dev);
 		create_descriptor_pools(dev);
 
+		rsx_log.notice("Building global vertex program interpreter...");
 		build_vs();
 		// TODO: Seed the cache
 	}
@@ -449,6 +399,7 @@ namespace vk
 		}
 		else
 		{
+			rsx_log.notice("Compiling FS...");
 			fs = build_fs(compiler_opt);
 		}
 
@@ -463,15 +414,17 @@ namespace vk
 		shader_stages[1].module = fs->get_handle();
 		shader_stages[1].pName = "main";
 
-		std::vector<VkDynamicState> dynamic_state_descriptors;
+		std::vector<VkDynamicState> dynamic_state_descriptors =
-		dynamic_state_descriptors.push_back(VK_DYNAMIC_STATE_VIEWPORT);
+		{
-		dynamic_state_descriptors.push_back(VK_DYNAMIC_STATE_SCISSOR);
+			VK_DYNAMIC_STATE_VIEWPORT,
-		dynamic_state_descriptors.push_back(VK_DYNAMIC_STATE_LINE_WIDTH);
+			VK_DYNAMIC_STATE_SCISSOR,
-		dynamic_state_descriptors.push_back(VK_DYNAMIC_STATE_BLEND_CONSTANTS);
+			VK_DYNAMIC_STATE_LINE_WIDTH,
-		dynamic_state_descriptors.push_back(VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK);
+			VK_DYNAMIC_STATE_BLEND_CONSTANTS,
-		dynamic_state_descriptors.push_back(VK_DYNAMIC_STATE_STENCIL_WRITE_MASK);
+			VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
-		dynamic_state_descriptors.push_back(VK_DYNAMIC_STATE_STENCIL_REFERENCE);
+			VK_DYNAMIC_STATE_STENCIL_WRITE_MASK,
-		dynamic_state_descriptors.push_back(VK_DYNAMIC_STATE_DEPTH_BIAS);
+			VK_DYNAMIC_STATE_STENCIL_REFERENCE,
+			VK_DYNAMIC_STATE_DEPTH_BIAS
+		};
 
 		if (vk::get_current_renderer()->get_depth_bounds_support())
 		{
@@ -502,6 +455,9 @@ namespace vk
 		VkPipelineColorBlendStateCreateInfo cs = properties.state.cs;
 		cs.pAttachments = properties.state.att_state;
 
+		VkPipelineTessellationStateCreateInfo ts = {};
+		ts.sType = VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO;
+
 		VkGraphicsPipelineCreateInfo info = {};
 		info.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO;
 		info.pVertexInputState = &vi;
@@ -511,6 +467,7 @@ namespace vk
 		info.pMultisampleState = &ms;
 		info.pViewportState = &vp;
 		info.pDepthStencilState = &properties.state.ds;
+		info.pTessellationState = &ts;
 		info.stageCount = 2;
 		info.pStages = shader_stages;
 		info.pDynamicState = &dynamic_state_info;

rpcs3/Emu/RSX/VK/VKShaderInterpreter.h

@@ -41,6 +41,7 @@ namespace vk
 
 		std::unordered_map<pipeline_key, std::unique_ptr<glsl::program>, key_hasher> m_program_cache;
 		std::unordered_map<u64, std::unique_ptr<glsl::shader>> m_fs_cache;
+		rsx::simple_array<VkDescriptorPoolSize> m_descriptor_pool_sizes;
 		vk::descriptor_pool m_descriptor_pool;
 
 		u32 m_vertex_instruction_start = 0;