Merge pull request #9198 from liamwhite/arm64

Initial ARM64 support

15 files changed, 66 insertions, 34 deletions

diff --git a/CMakeLists.txt b/CMakeLists.txt
index c6fc5dd9e..1d13dc74e 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -133,13 +133,13 @@ if (NOT ENABLE_GENERIC)
     if (MSVC)
         detect_architecture("_M_AMD64" x86_64)
         detect_architecture("_M_IX86" x86)
-        detect_architecture("_M_ARM" ARM)
-        detect_architecture("_M_ARM64" ARM64)
+        detect_architecture("_M_ARM" arm)
+        detect_architecture("_M_ARM64" arm64)
     else()
         detect_architecture("__x86_64__" x86_64)
         detect_architecture("__i386__" x86)
-        detect_architecture("__arm__" ARM)
-        detect_architecture("__aarch64__" ARM64)
+        detect_architecture("__arm__" arm)
+        detect_architecture("__aarch64__" arm64)
     endif()
 endif()
 
diff --git a/externals/CMakeLists.txt b/externals/CMakeLists.txt
index e80fd124e..7f0a6d069 100644
--- a/externals/CMakeLists.txt
+++ b/externals/CMakeLists.txt
@@ -7,15 +7,14 @@ include(DownloadExternals)
 
 # xbyak
 if (ARCHITECTURE_x86 OR ARCHITECTURE_x86_64)
-    add_library(xbyak INTERFACE)
-    file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/xbyak/include)
-    file(COPY ${CMAKE_CURRENT_SOURCE_DIR}/xbyak/xbyak DESTINATION ${CMAKE_CURRENT_BINARY_DIR}/xbyak/include)
-    target_include_directories(xbyak SYSTEM INTERFACE ${CMAKE_CURRENT_BINARY_DIR}/xbyak/include)
-    target_compile_definitions(xbyak INTERFACE XBYAK_NO_OP_NAMES)
+    add_subdirectory(xbyak)
 endif()
 
 # Dynarmic
-if (ARCHITECTURE_x86_64)
+if (ARCHITECTURE_x86_64 OR ARCHITECTURE_arm64)
+    if (ARCHITECTURE_arm64)
+       set(DYNARMIC_FRONTENDS "A32")
+    endif()
     set(DYNARMIC_NO_BUNDLED_FMT ON)
     set(DYNARMIC_IGNORE_ASSERTS ON CACHE BOOL "" FORCE)
     add_subdirectory(dynarmic)
diff --git a/externals/dynarmic b/externals/dynarmic
-Subproject 2d4602a6516c67d547000d4c80bcc5f74976abd
+Subproject 424fdb5c5026ec5bdd7553271190397f63fb503
diff --git a/externals/xbyak b/externals/xbyak
-Subproject c306b8e5786eeeb87b8925a8af5c3bf057ff5a9
+Subproject 348e3e548ebac06d243e5881caec8440e249f65
diff --git a/src/audio_core/CMakeLists.txt b/src/audio_core/CMakeLists.txt
index 0a1f3bf18..8e3a8f5a8 100644
--- a/src/audio_core/CMakeLists.txt
+++ b/src/audio_core/CMakeLists.txt
@@ -217,7 +217,7 @@ else()
 endif()
 
 target_link_libraries(audio_core PUBLIC common core)
-if (ARCHITECTURE_x86_64)
+if (ARCHITECTURE_x86_64 OR ARCHITECTURE_arm64)
     target_link_libraries(audio_core PRIVATE dynarmic)
 endif()
 
diff --git a/src/common/common_funcs.h b/src/common/common_funcs.h
index e1e2a90fc..0dad9338a 100644
--- a/src/common/common_funcs.h
+++ b/src/common/common_funcs.h
@@ -31,8 +31,10 @@
 
 #ifndef _MSC_VER
 
-#ifdef ARCHITECTURE_x86_64
+#if defined(ARCHITECTURE_x86_64)
 #define Crash() __asm__ __volatile__("int $3")
+#elif defined(ARCHITECTURE_arm64)
+#define Crash() __asm__ __volatile__("brk #0")
 #else
 #define Crash() exit(1)
 #endif
diff --git a/src/common/host_memory.cpp b/src/common/host_memory.cpp
index 7f9659612..909f6cf3f 100644
--- a/src/common/host_memory.cpp
+++ b/src/common/host_memory.cpp
@@ -359,6 +359,12 @@ public:
             }
         });
 
+        long page_size = sysconf(_SC_PAGESIZE);
+        if (page_size != 0x1000) {
+            LOG_CRITICAL(HW_Memory, "page size {:#x} is incompatible with 4K paging", page_size);
+            throw std::bad_alloc{};
+        }
+
         // Backing memory initialization
 #if defined(__FreeBSD__) && __FreeBSD__ < 13
         // XXX Drop after FreeBSD 12.* reaches EOL on 2024-06-30
diff --git a/src/core/CMakeLists.txt b/src/core/CMakeLists.txt
index f6e082c36..f67f1ce92 100644
--- a/src/core/CMakeLists.txt
+++ b/src/core/CMakeLists.txt
@@ -497,10 +497,6 @@ add_library(core STATIC
     hle/service/hid/irsensor/processor_base.h
     hle/service/hid/irsensor/tera_plugin_processor.cpp
     hle/service/hid/irsensor/tera_plugin_processor.h
-    hle/service/jit/jit_context.cpp
-    hle/service/jit/jit_context.h
-    hle/service/jit/jit.cpp
-    hle/service/jit/jit.h
     hle/service/lbl/lbl.cpp
     hle/service/lbl/lbl.h
     hle/service/ldn/lan_discovery.cpp
@@ -805,14 +801,18 @@ if (ENABLE_WEB_SERVICE)
     target_link_libraries(core PRIVATE web_service)
 endif()
 
-if (ARCHITECTURE_x86_64)
+if (ARCHITECTURE_x86_64 OR ARCHITECTURE_arm64)
     target_sources(core PRIVATE
-        arm/dynarmic/arm_dynarmic_32.cpp
-        arm/dynarmic/arm_dynarmic_32.h
         arm/dynarmic/arm_dynarmic_64.cpp
         arm/dynarmic/arm_dynarmic_64.h
+        arm/dynarmic/arm_dynarmic_32.cpp
+        arm/dynarmic/arm_dynarmic_32.h
         arm/dynarmic/arm_dynarmic_cp15.cpp
         arm/dynarmic/arm_dynarmic_cp15.h
+        hle/service/jit/jit_context.cpp
+        hle/service/jit/jit_context.h
+        hle/service/jit/jit.cpp
+        hle/service/jit/jit.h
     )
     target_link_libraries(core PRIVATE dynarmic)
 endif()
diff --git a/src/core/arm/dynarmic/arm_dynarmic_32.cpp b/src/core/arm/dynarmic/arm_dynarmic_32.cpp
index 287ba102e..227e06ea1 100644
--- a/src/core/arm/dynarmic/arm_dynarmic_32.cpp
+++ b/src/core/arm/dynarmic/arm_dynarmic_32.cpp
@@ -301,6 +301,11 @@ std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable*
         }
     }
 
+#ifdef ARCHITECTURE_arm64
+    // TODO: remove when fixed in dynarmic
+    config.optimizations &= ~Dynarmic::OptimizationFlag::BlockLinking;
+#endif
+
     return std::make_unique<Dynarmic::A32::Jit>(config);
 }
 
diff --git a/src/core/arm/dynarmic/arm_dynarmic_cp15.cpp b/src/core/arm/dynarmic/arm_dynarmic_cp15.cpp
index 200efe4db..5a4eba3eb 100644
--- a/src/core/arm/dynarmic/arm_dynarmic_cp15.cpp
+++ b/src/core/arm/dynarmic/arm_dynarmic_cp15.cpp
@@ -52,12 +52,16 @@ CallbackOrAccessOneWord DynarmicCP15::CompileSendOneWord(bool two, unsigned opc1
         case 4:
             // CP15_DATA_SYNC_BARRIER
             return Callback{
-                [](Dynarmic::A32::Jit*, void*, std::uint32_t, std::uint32_t) -> std::uint64_t {
-#ifdef _MSC_VER
+                [](void*, std::uint32_t, std::uint32_t) -> std::uint64_t {
+#if defined(_MSC_VER) && defined(ARCHITECTURE_x86_64)
                     _mm_mfence();
                     _mm_lfence();
-#else
+#elif defined(ARCHITECTURE_x86_64)
                     asm volatile("mfence\n\tlfence\n\t" : : : "memory");
+#elif defined(ARCHITECTURE_arm64)
+                    asm volatile("dsb sy\n\t" : : : "memory");
+#else
+#error Unsupported architecture
 #endif
                     return 0;
                 },
@@ -66,11 +70,15 @@ CallbackOrAccessOneWord DynarmicCP15::CompileSendOneWord(bool two, unsigned opc1
         case 5:
             // CP15_DATA_MEMORY_BARRIER
             return Callback{
-                [](Dynarmic::A32::Jit*, void*, std::uint32_t, std::uint32_t) -> std::uint64_t {
-#ifdef _MSC_VER
+                [](void*, std::uint32_t, std::uint32_t) -> std::uint64_t {
+#if defined(_MSC_VER) && defined(ARCHITECTURE_x86_64)
                     _mm_mfence();
-#else
+#elif defined(ARCHITECTURE_x86_64)
                     asm volatile("mfence\n\t" : : : "memory");
+#elif defined(ARCHITECTURE_arm64)
+                    asm volatile("dmb sy\n\t" : : : "memory");
+#else
+#error Unsupported architecture
 #endif
                     return 0;
                 },
@@ -115,7 +123,7 @@ CallbackOrAccessOneWord DynarmicCP15::CompileGetOneWord(bool two, unsigned opc1,
 CallbackOrAccessTwoWords DynarmicCP15::CompileGetTwoWords(bool two, unsigned opc, CoprocReg CRm) {
     if (!two && opc == 0 && CRm == CoprocReg::C14) {
         // CNTPCT
-        const auto callback = [](Dynarmic::A32::Jit*, void* arg, u32, u32) -> u64 {
+        const auto callback = [](void* arg, u32, u32) -> u64 {
             const auto& parent_arg = *static_cast<ARM_Dynarmic_32*>(arg);
             return parent_arg.system.CoreTiming().GetClockTicks();
         };
diff --git a/src/core/arm/exclusive_monitor.cpp b/src/core/arm/exclusive_monitor.cpp
index 2db0b035d..20550faeb 100644
--- a/src/core/arm/exclusive_monitor.cpp
+++ b/src/core/arm/exclusive_monitor.cpp
@@ -1,7 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2018 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
-#ifdef ARCHITECTURE_x86_64
+#if defined(ARCHITECTURE_x86_64) || defined(ARCHITECTURE_arm64)
 #include "core/arm/dynarmic/arm_exclusive_monitor.h"
 #endif
 #include "core/arm/exclusive_monitor.h"
@@ -13,7 +13,7 @@ ExclusiveMonitor::~ExclusiveMonitor() = default;
 
 std::unique_ptr<Core::ExclusiveMonitor> MakeExclusiveMonitor(Memory::Memory& memory,
                                                              std::size_t num_cores) {
-#ifdef ARCHITECTURE_x86_64
+#if defined(ARCHITECTURE_x86_64) || defined(ARCHITECTURE_arm64)
     return std::make_unique<Core::DynarmicExclusiveMonitor>(memory, num_cores);
 #else
     // TODO(merry): Passthrough exclusive monitor
diff --git a/src/core/hle/kernel/physical_core.cpp b/src/core/hle/kernel/physical_core.cpp
index d4375962f..3044922ac 100644
--- a/src/core/hle/kernel/physical_core.cpp
+++ b/src/core/hle/kernel/physical_core.cpp
@@ -12,7 +12,7 @@ namespace Kernel {
 
 PhysicalCore::PhysicalCore(std::size_t core_index_, Core::System& system_, KScheduler& scheduler_)
     : core_index{core_index_}, system{system_}, scheduler{scheduler_} {
-#ifdef ARCHITECTURE_x86_64
+#if defined(ARCHITECTURE_x86_64) || defined(ARCHITECTURE_arm64)
     // TODO(bunnei): Initialization relies on a core being available. We may later replace this with
     // a 32-bit instance of Dynarmic. This should be abstracted out to a CPU manager.
     auto& kernel = system.Kernel();
@@ -26,7 +26,7 @@ PhysicalCore::PhysicalCore(std::size_t core_index_, Core::System& system_, KSche
 PhysicalCore::~PhysicalCore() = default;
 
 void PhysicalCore::Initialize([[maybe_unused]] bool is_64_bit) {
-#ifdef ARCHITECTURE_x86_64
+#if defined(ARCHITECTURE_x86_64) || defined(ARCHITECTURE_arm64)
     auto& kernel = system.Kernel();
     if (!is_64_bit) {
         // We already initialized a 64-bit core, replace with a 32-bit one.
diff --git a/src/video_core/CMakeLists.txt b/src/video_core/CMakeLists.txt
index 106991969..d7f7d336c 100644
--- a/src/video_core/CMakeLists.txt
+++ b/src/video_core/CMakeLists.txt
@@ -73,8 +73,6 @@ add_library(video_core STATIC
     macro/macro_hle.h
     macro/macro_interpreter.cpp
     macro/macro_interpreter.h
-    macro/macro_jit_x64.cpp
-    macro/macro_jit_x64.h
     fence_manager.h
     gpu.cpp
     gpu.h
@@ -245,7 +243,7 @@ add_library(video_core STATIC
 create_target_directory_groups(video_core)
 
 target_link_libraries(video_core PUBLIC common core)
-target_link_libraries(video_core PUBLIC glad shader_recompiler xbyak)
+target_link_libraries(video_core PUBLIC glad shader_recompiler)
 
 if (YUZU_USE_BUNDLED_FFMPEG AND NOT WIN32)
     add_dependencies(video_core ffmpeg-build)
@@ -282,8 +280,19 @@ else()
 
         -Wno-sign-conversion
     )
+
+    # xbyak
+    set_source_files_properties(macro/macro_jit_x64.cpp PROPERTIES COMPILE_OPTIONS "-Wno-conversion;-Wno-shadow")
 endif()
 
 if (ARCHITECTURE_x86_64)
+    target_sources(video_core PRIVATE
+        macro/macro_jit_x64.cpp
+        macro/macro_jit_x64.h
+    )
+    target_link_libraries(video_core PUBLIC xbyak)
+endif()
+
+if (ARCHITECTURE_x86_64 OR ARCHITECTURE_arm64)
     target_link_libraries(video_core PRIVATE dynarmic)
 endif()
diff --git a/src/video_core/macro/macro.cpp b/src/video_core/macro/macro.cpp
index f61d5998e..505d81c1e 100644
--- a/src/video_core/macro/macro.cpp
+++ b/src/video_core/macro/macro.cpp
@@ -16,7 +16,10 @@
 #include "video_core/macro/macro.h"
 #include "video_core/macro/macro_hle.h"
 #include "video_core/macro/macro_interpreter.h"
+
+#ifdef ARCHITECTURE_x86_64
 #include "video_core/macro/macro_jit_x64.h"
+#endif
 
 namespace Tegra {
 
diff --git a/src/yuzu/CMakeLists.txt b/src/yuzu/CMakeLists.txt
index 239f12382..5cc1fbf32 100644
--- a/src/yuzu/CMakeLists.txt
+++ b/src/yuzu/CMakeLists.txt
@@ -385,6 +385,6 @@ if (NOT APPLE)
     target_compile_definitions(yuzu PRIVATE HAS_OPENGL)
 endif()
 
-if (ARCHITECTURE_x86_64)
+if (ARCHITECTURE_x86_64 OR ARCHITECTURE_arm64)
     target_link_libraries(yuzu PRIVATE dynarmic)
 endif()