Initial Linux Aarch64 support

* Update asmjit dependency (aarch64 branch)
* Disable USE_DISCORD_RPC by default
* Dump some JIT objects in rpcs3 cache dir
* Add SIGILL handler for all platforms
* Fix resetting zeroing denormals in thread pool
* Refactor most v128:: utils into global gv_** functions
* Refactor PPU interpreter (incomplete), remove "precise"
* - Instruction specializations with multiple accuracy flags
* - Adjust calling convention for speed
* - Removed precise/fast setting, replaced with static
* - Started refactoring interpreters for building at runtime JIT
*   (I got tired of poor compiler optimizations)
* - Expose some accuracy settings (SAT, NJ, VNAN, FPCC)
* - Add exec_bytes PPU thread variable (akin to cycle count)
* PPU LLVM: fix VCTUXS+VCTSXS instruction NaN results
* SPU interpreter: remove "precise" for now (extremely non-portable)
* - As with PPU, settings changed to static/dynamic for interpreters.
* - Precise options will be implemented later
* Fix termination after fatal error dialog

3rdparty/discord-rpc/CMakeLists.txt

@@ -2,7 +2,7 @@
 add_library(3rdparty_discordRPC INTERFACE)
 
 # We don't want Discord Rich Presence on the BSDs and other OSes
-if (USE_DISCORD_RPC AND (WIN32 OR CMAKE_SYSTEM MATCHES "Linux" OR APPLE))
+if (USE_DISCORD_RPC AND (WIN32 OR CMAKE_SYSTEM MATCHES "Linux" OR APPLE) AND COMPILER_X86)
 	if (WIN32 AND NOT MSVC)
 		ExternalProject_Add(discordRPC
 				GIT_REPOSITORY https://github.com/discordapp/discord-rpc

3rdparty/llvm.cmake

@@ -1,8 +1,10 @@
 if(WITH_LLVM)
+	CHECK_CXX_COMPILER_FLAG("-msse -msse2 -mcx16" COMPILER_X86)
+	CHECK_CXX_COMPILER_FLAG("-march=armv8-a+lse" COMPILER_ARM)
+
 	if(BUILD_LLVM_SUBMODULE)
 		message(STATUS "LLVM will be built from the submodule.")
 
-		set(LLVM_TARGETS_TO_BUILD "X86" CACHE INTERNAL "")
 		option(LLVM_BUILD_RUNTIME OFF)
 		option(LLVM_BUILD_TOOLS OFF)
 		option(LLVM_INCLUDE_BENCHMARKS OFF)
@@ -61,7 +63,15 @@ if(WITH_LLVM)
 		endif()
 	endif()
 
-	set(LLVM_LIBS LLVMMCJIT LLVMX86CodeGen LLVMX86AsmParser)
+	set(LLVM_LIBS LLVMMCJIT)
+
+	if(COMPILER_X86)
+		set(LLVM_LIBS ${LLVM_LIBS} LLVMX86CodeGen LLVMX86AsmParser)
+	endif()
+
+	if(COMPILER_ARM)
+		set(LLVM_LIBS ${LLVM_LIBS} LLVMX86CodeGen LLVMX86AsmParser LLVMARMCodeGen LLVMARMAsmParser)
+	endif()
 
 	if(WIN32 OR CMAKE_SYSTEM MATCHES "Linux")
 		set(LLVM_LIBS ${LLVM_LIBS} LLVMIntelJITEvents)

CMakeLists.txt

@@ -17,7 +17,7 @@ option(WITH_LLVM "Enable usage of LLVM library" ON)
 option(BUILD_LLVM_SUBMODULE "Build LLVM from git submodule" ON)
 option(USE_FAUDIO "FAudio audio backend" ON)
 option(USE_LIBEVDEV "libevdev-based joystick support" ON)
-option(USE_DISCORD_RPC "Discord rich presence integration" ON)
+option(USE_DISCORD_RPC "Discord rich presence integration" OFF)
 option(USE_SYSTEM_ZLIB "Prefer system ZLIB instead of the builtin one" ON)
 option(USE_VULKAN "Vulkan render backend" ON)
 option(USE_PRECOMPILED_HEADERS "Use precompiled headers" OFF)

Utilities/JIT.cpp

@@ -18,6 +18,12 @@ LOG_CHANNEL(jit_log, "JIT");
 
 void jit_announce(uptr func, usz size, std::string_view name)
 {
+	if (!size)
+	{
+		jit_log.error("Empty function announced: %s (%p)", name, func);
+		return;
+	}
+
 #ifdef __linux__
 	static const fs::file s_map(fmt::format("/tmp/perf-%d.map", getpid()), fs::rewrite + fs::append);
 
@@ -124,15 +130,31 @@ void* jit_runtime_base::_add(asmjit::CodeHolder* code) noexcept
 {
 	ensure(!code->flatten());
 	ensure(!code->resolveUnresolvedLinks());
-	usz codeSize = ensure(code->codeSize());
+	usz codeSize = code->codeSize();
+	if (!codeSize)
+		return nullptr;
+
 	auto p = ensure(this->_alloc(codeSize, 64));
 	ensure(!code->relocateToBase(uptr(p)));
 
-	asmjit::VirtMem::ProtectJitReadWriteScope rwScope(p, codeSize);
+	{
+		asmjit::VirtMem::ProtectJitReadWriteScope rwScope(p, codeSize);
 
-	for (asmjit::Section* section : code->_sections)
+		for (asmjit::Section* section : code->_sections)
+		{
+			std::memcpy(p + section->offset(), section->data(), section->bufferSize());
+		}
+	}
+
+	if (!dump_name.empty())
 	{
-		std::memcpy(p + section->offset(), section->data(), section->bufferSize());
+		// If directory ASMJIT doesn't exist, nothing will be written
+		fs::file dump(fmt::format("%s/ASMJIT/%s", fs::get_cache_dir(), dump_name), fs::rewrite);
+
+		if (dump)
+		{
+			dump.write(p, codeSize);
+		}
 	}
 
 	return p;
@@ -349,8 +371,9 @@ static u64 make_null_function(const std::string& name)
 		using namespace asmjit;
 
 		// Build a "null" function that contains its name
-		const auto func = build_function_asm<void (*)()>("NULL", [&](x86::Assembler& c, auto& args)
+		const auto func = build_function_asm<void (*)()>("NULL", [&](native_asm& c, auto& args)
 		{
+#if defined(ARCH_X64)
 			Label data = c.newLabel();
 			c.lea(args[0], x86::qword_ptr(data, 0));
 			c.jmp(Imm(&null));
@@ -362,6 +385,7 @@ static u64 make_null_function(const std::string& name)
 				c.db(ch);
 			c.db(0);
 			c.align(AlignMode::kData, 16);
+#endif
 		});
 
 		func_ptr = reinterpret_cast<u64>(func);

Utilities/JIT.h

@@ -22,10 +22,17 @@
 #pragma GCC diagnostic ignored "-Wredundant-decls"
 #pragma GCC diagnostic ignored "-Wnon-virtual-dtor"
 #pragma GCC diagnostic ignored "-Weffc++"
-#ifndef __clang__
+#ifdef __clang__
+#pragma GCC diagnostic ignored "-Wdeprecated-anon-enum-enum-conversion"
+#pragma GCC diagnostic ignored "-Wcast-qual"
+#else
 #pragma GCC diagnostic ignored "-Wduplicated-branches"
+#pragma GCC diagnostic ignored "-Wdeprecated-enum-enum-conversion"
 #endif
 #include <asmjit/asmjit.h>
+#if defined(ARCH_ARM64)
+#include <asmjit/a64.h>
+#endif
 #pragma GCC diagnostic pop
 #endif
 
@@ -36,6 +43,14 @@
 #include <string_view>
 #include <unordered_map>
 
+#if defined(ARCH_X64)
+using native_asm = asmjit::x86::Assembler;
+using native_args = std::array<asmjit::x86::Gp, 4>;
+#elif defined(ARCH_ARM64)
+using native_asm = asmjit::a64::Assembler;
+using native_args = std::array<asmjit::a64::Gp, 4>;
+#endif
+
 void jit_announce(uptr func, usz size, std::string_view name);
 
 void jit_announce(auto* func, usz size, std::string_view name)
@@ -62,6 +77,8 @@ struct jit_runtime_base
 	const asmjit::Environment& environment() const noexcept;
 	void* _add(asmjit::CodeHolder* code) noexcept;
 	virtual uchar* _alloc(usz size, usz align) noexcept = 0;
+
+	std::string_view dump_name;
 };
 
 // ASMJIT runtime for emitting code in a single 2G region
@@ -167,11 +184,39 @@ namespace asmjit
 		}
 	}
 
+	inline void build_init_args_from_ghc(native_asm& c, native_args& args)
+	{
+#if defined(ARCH_X64)
+		// TODO: handle case when args don't overlap with r13/rbp/r12/rbx
+		c.mov(args[0], x86::r13);
+		c.mov(args[1], x86::rbp);
+		c.mov(args[2], x86::r12);
+		c.mov(args[3], x86::rbx);
+#else
+		static_cast<void>(c);
+		static_cast<void>(args);
+#endif
+	}
+
+	inline void build_init_ghc_args(native_asm& c, native_args& args)
+	{
+#if defined(ARCH_X64)
+		// TODO: handle case when args don't overlap with r13/rbp/r12/rbx
+		c.mov(x86::r13, args[0]);
+		c.mov(x86::rbp, args[1]);
+		c.mov(x86::r12, args[2]);
+		c.mov(x86::rbx, args[3]);
+#else
+		static_cast<void>(c);
+		static_cast<void>(args);
+#endif
+	}
+
 	using imm_ptr = Imm;
 }
 
 // Build runtime function with asmjit::X86Assembler
-template <typename FT, typename F>
+template <typename FT, typename Asm = native_asm, typename F>
 inline FT build_function_asm(std::string_view name, F&& builder)
 {
 	using namespace asmjit;
@@ -181,7 +226,8 @@ inline FT build_function_asm(std::string_view name, F&& builder)
 	CodeHolder code;
 	code.init(rt.environment());
 
-	std::array<x86::Gp, 4> args;
+#if defined(ARCH_X64)
+	native_args args;
 #ifdef _WIN32
 	args[0] = x86::rcx;
 	args[1] = x86::rdx;
@@ -193,16 +239,27 @@ inline FT build_function_asm(std::string_view name, F&& builder)
 	args[2] = x86::rdx;
 	args[3] = x86::rcx;
 #endif
+#elif defined(ARCH_ARM64)
+	native_args args;
+	args[0] = a64::x0;
+	args[1] = a64::x1;
+	args[2] = a64::x2;
+	args[3] = a64::x3;
+#endif
 
-	x86::Assembler compiler(&code);
+	Asm compiler(&code);
 	compiler.addEncodingOptions(EncodingOptions::kOptimizedAlign);
-	builder(std::ref(compiler), args);
+	if constexpr (std::is_invocable_v<F, Asm&, native_args&>)
+		builder(compiler, args);
+	else
+		builder(compiler);
+	rt.dump_name = name;
 	const auto result = rt._add(&code);
 	jit_announce(result, code.codeSize(), name);
 	return reinterpret_cast<FT>(uptr(result));
 }
 
-#ifdef __APPLE__
+#if !defined(ARCH_X64) || defined(__APPLE__)
 template <typename FT, usz = 4096>
 class built_function
 {
@@ -213,9 +270,23 @@ class built_function
 
 	built_function& operator=(const built_function&) = delete;
 
-	template <typename F>
-	built_function(std::string_view name, F&& builder)
-		: m_func(ensure(build_function_asm<FT>(name, std::forward<F>(builder))))
+	template <typename F> requires (std::is_invocable_v<F, native_asm&, native_args&>)
+	built_function(std::string_view name, F&& builder,
+		u32 line = __builtin_LINE(),
+		u32 col = __builtin_COLUMN(),
+		const char* file = __builtin_FILE(),
+		const char* func = __builtin_FUNCTION())
+		: m_func(ensure(build_function_asm<FT>(name, std::forward<F>(builder)), const_str(), line, col, file, func))
+	{
+	}
+
+	template <typename F> requires (std::is_invocable_v<F>)
+	built_function(std::string_view, F&& getter,
+		u32 line = __builtin_LINE(),
+		u32 col = __builtin_COLUMN(),
+		const char* file = __builtin_FILE(),
+		const char* func = __builtin_FUNCTION())
+		: m_func(ensure(getter(), const_str(), line, col, file, func))
 	{
 	}
 
@@ -251,7 +322,8 @@ class built_function
 		CodeHolder code;
 		code.init(rt.environment());
 
-		std::array<x86::Gp, 4> args;
+#if defined(ARCH_X64)
+		native_args args;
 	#ifdef _WIN32
 		args[0] = x86::rcx;
 		args[1] = x86::rdx;
@@ -263,10 +335,18 @@ class built_function
 		args[2] = x86::rdx;
 		args[3] = x86::rcx;
 	#endif
+#elif defined(ARCH_ARM64)
+		native_args args;
+		args[0] = a64::x0;
+		args[1] = a64::x1;
+		args[2] = a64::x2;
+		args[3] = a64::x3;
+#endif
 
-		x86::Assembler compiler(&code);
+		native_asm compiler(&code);
 		compiler.addEncodingOptions(EncodingOptions::kOptimizedAlign);
-		builder(std::ref(compiler), args);
+		builder(compiler, args);
+		rt.dump_name = name;
 		jit_announce(rt._add(&code), code.codeSize(), name);
 	}
 

Utilities/StrFmt.h

@@ -239,7 +239,7 @@ struct fmt_class_string<T, void>
 	static void format(std::string& out, u64 arg)
 	{
 		const auto& obj = get_object(arg);
-	
+
 		void format_byte_array(std::string&, const uchar*, usz);
 		format_byte_array(out, reinterpret_cast<const uchar*>(std::data(obj)), std::size(obj));
 	}