Docs: Update for release FEX-2301

Create a new ARM64 Emitter and move JIT over to it.

CMakeLists.txt

@@ -7,7 +7,7 @@ CHECK_INCLUDE_FILES ("gdb/jit-reader.h" HAVE_GDB_JIT_READER_H)
 option(BUILD_TESTS "Build unit tests to ensure sanity" TRUE)
 option(BUILD_FEX_LINUX_TESTS "Build FEXLinuxTests, requires x86 compiler" FALSE)
 option(BUILD_THUNKS "Build thunks" FALSE)
-option(BUILD_CLANG_THUNKS "Build thunks with clang" FALSE)
+option(ENABLE_CLANG_THUNKS "Build thunks with clang" FALSE)
 option(ENABLE_CLANG_FORMAT "Run clang format over the source" FALSE)
 option(ENABLE_IWYU "Enables include what you use program" FALSE)
 option(ENABLE_LTO "Enable LTO with compilation" TRUE)
@@ -29,11 +29,25 @@ option(ENABLE_INTERPRETER "Enables FEX's Interpreter" FALSE)
 option(ENABLE_CCACHE "Enables ccache for compile caching" TRUE)
 option(ENABLE_TERMUX_BUILD "Forces building for Termux on a non-Termux build machine" FALSE)
 option(ENABLE_VIXL_SIMULATOR "Forces the FEX JIT to use the VIXL simulator" FALSE)
+option(ENABLE_VIXL_DISASSEMBLER "Enables debug disassembler output with VIXL" FALSE)
+option(ENABLE_FEXCORE_PROFILER "Enables use of the FEXCore timeline profiling capabilities" FALSE)
+set (FEXCORE_PROFILER_BACKEND "gpuvis" CACHE STRING "Set which backend you want to use for the FEXCore profiler")
 
 set (X86_32_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/toolchain_x86_32.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting i686")
 set (X86_64_TOOLCHAIN_FILE "${CMAKE_CURRENT_SOURCE_DIR}/toolchain_x86_64.cmake" CACHE FILEPATH "Toolchain file for the (cross-)compiler targeting x86_64")
 set (DATA_DIRECTORY "${CMAKE_INSTALL_PREFIX}/share/fex-emu" CACHE PATH "global data directory")
 
+if (ENABLE_FEXCORE_PROFILER)
+  add_definitions(-DENABLE_FEXCORE_PROFILER=1)
+  string(TOUPPER "${FEXCORE_PROFILER_BACKEND}" FEXCORE_PROFILER_BACKEND)
+
+  if (FEXCORE_PROFILER_BACKEND STREQUAL "GPUVIS")
+    add_definitions(-DFEXCORE_PROFILER_BACKEND=1)
+  else()
+    message(FATAL_ERROR "Unknown FEXCore profiler backend ${FEXCORE_PROFILER_BACKEND}")
+  endif()
+endif()
+
 # uninstall target
 if(NOT TARGET uninstall)
   configure_file(

Data/ThunksDB.json

@@ -6,10 +6,10 @@
         "X11"
       ],
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libGL.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libGL.so.1",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libGL.so.1.2.0",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libGL.so.1.7.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libGL.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libGL.so.1",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libGL.so.1.2.0",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libGL.so.1.7.0"
       ]
     },
     "GLESv2": {
@@ -18,17 +18,17 @@
         "X11"
       ],
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libGLESv2.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libGLESv2.so.2",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libGLESv2.so.2.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libGLESv2.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libGLESv2.so.2",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libGLESv2.so.2.0.0"
       ]
     },
     "X11": {
       "Library": "libX11-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libX11.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libX11.so.6",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libX11.so.6.4.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libX11.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libX11.so.6",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libX11.so.6.4.0"
       ]
     },
     "Vulkan": {
@@ -37,8 +37,8 @@
         "xcb"
       ],
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libvulkan.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libvulkan.so.1",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libvulkan.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libvulkan.so.1",
         "@HOME@/.local/share/Steam/ubuntu12_32/steam-runtime/pinned_libs_64/libvulkan.so.1"
       ],
       "Comment": [
@@ -48,129 +48,129 @@
     "xcb": {
       "Library": "libxcb-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb.so.1",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb.so.1.1.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb.so.1",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb.so.1.1.0"
       ]
     },
     "xcb-dri2": {
-      "Library": "libxcb_dri2-guest.so",
+      "Library": "libxcb-dri2-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri2.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri2.so.0.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-dri2.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-dri2.so.0",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-dri2.so.0.0.0"
       ]
     },
     "xcb-dri3": {
-      "Library": "libxcb_dri3-guest.so",
+      "Library": "libxcb-dri3-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri3.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-dri3.so.0.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-dri3.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-dri3.so.0",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-dri3.so.0.0.0"
       ]
     },
     "xcb-xfixes": {
-      "Library": "libxcb_xfixes-guest.so",
+      "Library": "libxcb-xfixes-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-xfixes.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-xfixes.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-xfixes.so.0",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-xfixes.so.0.0.0"
       ]
     },
     "xcb-shm": {
-      "Library": "libxcb_shm-guest.so",
+      "Library": "libxcb-shm-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-shm.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-shm.so.0",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-shm.so.0.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-shm.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-shm.so.0",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-shm.so.0.0.0"
       ]
     },
     "xcb-sync": {
-      "Library": "libxcb_sync-guest.so",
+      "Library": "libxcb-sync-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-sync.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-sync.so.1",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-sync.so.1.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-sync.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-sync.so.1",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-sync.so.1.0.0"
       ]
     },
     "xcb-randr": {
-      "Library": "libxcb_randr-guest.so",
+      "Library": "libxcb-randr-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-randr.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-randr.so.0",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-randr.so.0.1.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-randr.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-randr.so.0",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-randr.so.0.1.0"
       ]
     },
     "xcb-present": {
-      "Library": "libxcb_present-guest.so",
+      "Library": "libxcb-present-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-present.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-present.so.0",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-present.so.0.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-present.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-present.so.0",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-present.so.0.0.0"
       ]
     },
     "xcb-glx": {
-      "Library": "libxcb_glx-guest.so",
+      "Library": "libxcb-glx-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-glx.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-glx.so.0",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxcb-glx.so.0.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-glx.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-glx.so.0",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxcb-glx.so.0.0.0"
       ]
     },
     "xshmfence": {
-      "Library": "libshmfence-guest.so",
+      "Library": "libxshmfence-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxshmfence.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxshmfence.so.1",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libxshmfence.so.1.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxshmfence.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxshmfence.so.1",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libxshmfence.so.1.0.0"
       ]
     },
     "drm": {
       "Library": "libdrm-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libdrm.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libdrm.so.2",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libdrm.so.2.4.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libdrm.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libdrm.so.2",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libdrm.so.2.4.0"
       ]
     },
     "asound": {
       "Library": "libasound-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libasound.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libasound.so.2",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libasound.so.2.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libasound.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libasound.so.2",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libasound.so.2.0.0"
       ]
     },
     "Xrender": {
       "Library": "libXrender-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libXrender.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libXrender.so.1",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libXrender.so.1.3.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libXrender.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libXrender.so.1",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libXrender.so.1.3.0"
       ]
     },
     "Xext": {
       "Library": "libXext-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libXext.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libXext.so.6",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libXext.so.6.4.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libXext.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libXext.so.6",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libXext.so.6.4.0"
       ]
     },
     "Xfixes": {
       "Library": "libXfixes-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libXfixes.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libXfixes.so.3",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libXfixes.so.3.1.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libXfixes.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libXfixes.so.3",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libXfixes.so.3.1.0"
       ]
     },
     "OpenCL": {
       "Library" : "libOpenCL-guest.so",
       "Overlay": [
-        "@PREFIX_LIB@/x86_64-linux-gnu/libOpenCL.so",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libOpenCL.so.1",
-        "@PREFIX_LIB@/x86_64-linux-gnu/libOpenCL.so.1.0.0"
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libOpenCL.so",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libOpenCL.so.1",
+        "@PREFIX_LIB@/@PREFIX_ARCH@-linux-gnu/libOpenCL.so.1.0.0"
       ]
     },
     "":{}

External/FEXCore/CMakeLists.txt

@@ -82,3 +82,7 @@ add_subdirectory(Source/)
 install (DIRECTORY include/FEXCore ${CMAKE_BINARY_DIR}/include/FEXCore
   DESTINATION include
   COMPONENT Development)
+
+if (BUILD_TESTS)
+  add_subdirectory(unittests/)
+endif()

External/FEXCore/Source/CMakeLists.txt

@@ -135,7 +135,6 @@ set (SRCS
   Interface/IR/Passes/PhiValidation.cpp
   Interface/IR/Passes/RedundantFlagCalculationElimination.cpp
   Interface/IR/Passes/DeadStoreElimination.cpp
-  Interface/IR/Passes/StaticRegisterAllocationPass.cpp
   Interface/IR/Passes/RegisterAllocationPass.cpp
   Interface/IR/Passes/SyscallOptimization.cpp
   Utils/Allocator.cpp
@@ -143,6 +142,7 @@ set (SRCS
   Utils/NetStream.cpp
   Utils/Telemetry.cpp
   Utils/Threads.cpp
+  Utils/Profiler.cpp
   )
 
 if (ENABLE_INTERPRETER)
@@ -182,6 +182,10 @@ if (ENABLE_VIXL_SIMULATOR)
   list(APPEND DEFINES -DVIXL_SIMULATOR=1 -DVIXL_INCLUDE_SIMULATOR_AARCH64=1)
 endif()
 
+if (ENABLE_VIXL_DISASSEMBLER)
+  list(APPEND DEFINES -DVIXL_DISASSEMBLER=1)
+endif()
+
 if (ENABLE_JIT_X86_64)
   list(APPEND SRCS
     Interface/Core/JIT/x86_64/JIT.cpp

External/FEXCore/Source/Interface/Config/Config.cpp

@@ -211,10 +211,12 @@ namespace JSON {
   static std::map<FEXCore::Config::LayerType, std::unique_ptr<FEXCore::Config::Layer>> ConfigLayers;
   static FEXCore::Config::Layer *Meta{};
 
-  constexpr std::array<FEXCore::Config::LayerType, 7> LoadOrder = {
+  constexpr std::array<FEXCore::Config::LayerType, 9> LoadOrder = {
     FEXCore::Config::LayerType::LAYER_GLOBAL_MAIN,
     FEXCore::Config::LayerType::LAYER_MAIN,
+    FEXCore::Config::LayerType::LAYER_GLOBAL_STEAM_APP,
     FEXCore::Config::LayerType::LAYER_GLOBAL_APP,
+    FEXCore::Config::LayerType::LAYER_LOCAL_STEAM_APP,
     FEXCore::Config::LayerType::LAYER_LOCAL_APP,
     FEXCore::Config::LayerType::LAYER_ARGUMENTS,
     FEXCore::Config::LayerType::LAYER_ENVIRONMENT,
@@ -629,7 +631,7 @@ namespace JSON {
 
   class AppLoader final : public FEXCore::Config::OptionMapper {
   public:
-    explicit AppLoader(const std::string& Filename, bool Global);
+    explicit AppLoader(const std::string& Filename, FEXCore::Config::LayerType Type);
     void Load();
 
   private:
@@ -681,8 +683,10 @@ namespace JSON {
     });
   }
 
-  AppLoader::AppLoader(const std::string& Filename, bool Global)
-    : FEXCore::Config::OptionMapper(Global ? FEXCore::Config::LayerType::LAYER_GLOBAL_APP : FEXCore::Config::LayerType::LAYER_LOCAL_APP) {
+  AppLoader::AppLoader(const std::string& Filename, FEXCore::Config::LayerType Type)
+    : FEXCore::Config::OptionMapper(Type) {
+    const bool Global = Type == FEXCore::Config::LayerType::LAYER_GLOBAL_STEAM_APP ||
+                        Type == FEXCore::Config::LayerType::LAYER_GLOBAL_APP;
     Config = FEXCore::Config::GetApplicationConfig(Filename, Global);
 
     // Immediately load so we can reload the meta layer
@@ -754,8 +758,8 @@ namespace JSON {
     }
   }
 
-  std::unique_ptr<FEXCore::Config::Layer> CreateAppLayer(const std::string& Filename, bool Global) {
-    return std::make_unique<FEXCore::Config::AppLoader>(Filename, Global);
+  std::unique_ptr<FEXCore::Config::Layer> CreateAppLayer(const std::string& Filename, FEXCore::Config::LayerType Type) {
+    return std::make_unique<FEXCore::Config::AppLoader>(Filename, Type);
   }
 
   std::unique_ptr<FEXCore::Config::Layer> CreateEnvironmentLayer(char *const _envp[]) {

External/FEXCore/Source/Interface/Config/Config.json.in

@@ -16,10 +16,11 @@
       },
       "Multiblock": {
         "Type": "bool",
-        "Default": "true",
+        "Default": "false",
         "ShortArg": "m",
         "Desc": [
-          "Controls multiblock code compilation"
+          "Controls multiblock code compilation",
+          "Can cause long JIT compilation times and stutter"
         ]
       },
       "MaxInst": {
@@ -90,6 +91,13 @@
           "Folder to find the guest-side thunking libraries."
         ]
       },
+      "ThunkHostLibs32": {
+        "Type": "str",
+        "Default": "@CMAKE_INSTALL_PREFIX@/lib/fex-emu/HostThunks_32/",
+        "Desc": [
+          "Folder to find the 32-bit host-side thunking libraries."
+        ]
+      },
       "ThunkGuestLibs32": {
         "Type": "str",
         "Default": "@CMAKE_INSTALL_PREFIX@/share/fex-emu/GuestThunks_32/",

External/FEXCore/Source/Interface/Context/Context.h

@@ -105,6 +105,7 @@ namespace FEXCore::Context {
       FEX_CONFIG_OPT(MaxInstPerBlock, MAXINST);
       FEX_CONFIG_OPT(RootFSPath, ROOTFS);
       FEX_CONFIG_OPT(ThunkHostLibsPath, THUNKHOSTLIBS);
+      FEX_CONFIG_OPT(ThunkHostLibsPath32, THUNKHOSTLIBS32);
       FEX_CONFIG_OPT(ThunkConfigFile, THUNKCONFIG);
       FEX_CONFIG_OPT(DumpIR, DUMPIR);
       FEX_CONFIG_OPT(StaticRegisterAllocation, SRA);

External/FEXCore/Source/Interface/Core/ArchHelpers/Arm64.cpp

@@ -1,7 +1,6 @@
 #include "Interface/Core/ArchHelpers/Arm64.h"
 #include "Interface/Core/ArchHelpers/MContext.h"
-
-#include <aarch64/cpu-aarch64.h>
+#include "Interface/Core/ArchHelpers/CodeEmitter/Buffer.h"
 
 #include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/LogManager.h>
@@ -572,7 +571,7 @@ bool HandleAtomicVectorStore(void *_ucontext, void *_info, uint32_t Instr) {
       PC[1] = STP;
       PC[2] = DMB;
       // Back up one instruction and have another go
-      vixl::aarch64::CPU::EnsureIAndDCacheCoherency(&PC[0], 16);
+      FEXCore::ARMEmitter::Buffer::ClearICache(&PC[0], 16);
       return true;
     }
   }
@@ -2311,7 +2310,7 @@ bool HandleSIGBUS(bool ParanoidTSO, int Signal, void *info, void *ucontext) {
       return false;
     }
 
-    vixl::aarch64::CPU::EnsureIAndDCacheCoherency(&PC[-1], 16);
+    FEXCore::ARMEmitter::Buffer::ClearICache(&PC[-1], 16);
     return true;
   }
   return false;

External/FEXCore/Source/Interface/Core/ArchHelpers/Arm64Emitter.cpp

@@ -1,4 +1,5 @@
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Context/Context.h"
 #include "Interface/HLE/Thunks/Thunks.h"
@@ -20,29 +21,24 @@ namespace FEXCore::CPU {
 
 // We want vixl to not allocate a default buffer. Jit and dispatcher will manually create one.
 Arm64Emitter::Arm64Emitter(FEXCore::Context::Context *ctx, size_t size)
-  : vixl::aarch64::Assembler(size, vixl::aarch64::PositionDependentCode)
+  : Emitter(size ? (uint8_t*)FEXCore::Allocator::mmap(nullptr, size, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0) : nullptr, size)
   , EmitterCTX {ctx} {
   CPU.SetUp();
-
-#ifdef VIXL_SIMULATOR
-  auto Features = vixl::CPUFeatures::All();
-#else
-  auto Features = vixl::CPUFeatures::InferFromOS();
-  if (ctx->HostFeatures.SupportsAtomics) {
-    // Hypervisor can hide this on the c630?
-    Features.Combine(vixl::CPUFeatures::Feature::kLORegions);
-  }
-#endif
-
-  SetCPUFeatures(Features);
 }
 
-void Arm64Emitter::LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant, bool NOPPad) {
-  bool Is64Bit = Reg.IsX();
+Arm64Emitter::~Arm64Emitter() {
+  auto BufferSize = GetBufferSize();
+  if (BufferSize) {
+    FEXCore::Allocator::munmap(GetBufferBase(), BufferSize);
+  }
+}
+
+void Arm64Emitter::LoadConstant(ARMEmitter::Size s, ARMEmitter::Register Reg, uint64_t Constant, bool NOPPad) {
+  bool Is64Bit = s == ARMEmitter::Size::i64Bit;
   int Segments = Is64Bit ? 4 : 2;
 
   if (Is64Bit && ((~Constant)>> 16) == 0) {
-    movn(Reg, (~Constant) & 0xFFFF);
+    movn(s, Reg, (~Constant) & 0xFFFF);
 
     if (NOPPad) {
       nop(); nop(); nop();
@@ -89,17 +85,17 @@ void Arm64Emitter::LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant,
       else {
         // Need to use ADRP + ADD
         adrp(Reg, AlignedOffset >> 12);
-        add(Reg, Reg, Constant & 0xFFF);
+        add(s, Reg, Reg, Constant & 0xFFF);
         NumMoves = 2;
       }
     }
   }
   else {
-    movz(Reg, (Constant) & 0xFFFF, 0);
+    movz(s, Reg, (Constant) & 0xFFFF, 0);
     for (int i = 1; i < Segments; ++i) {
       uint16_t Part = (Constant >> (i * 16)) & 0xFFFF;
       if (Part) {
-        movk(Reg, Part, i * 16);
+        movk(s, Reg, Part, i * 16);
         ++NumMoves;
       }
     }
@@ -115,126 +111,143 @@ void Arm64Emitter::LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant,
 void Arm64Emitter::PushCalleeSavedRegisters() {
   // We need to save pairs of registers
   // We save r19-r30
-  MemOperand PairOffset(sp, -16, PreIndex);
-  const std::array<std::pair<vixl::aarch64::Register, vixl::aarch64::Register>, 6> CalleeSaved = {{
-    {x19, x20},
-    {x21, x22},
-    {x23, x24},
-    {x25, x26},
-    {x27, x28},
-    {x29, x30},
+  const std::array<std::pair<ARMEmitter::XRegister, ARMEmitter::XRegister>, 6> CalleeSaved = {{
+    {ARMEmitter::XReg::x19, ARMEmitter::XReg::x20},
+    {ARMEmitter::XReg::x21, ARMEmitter::XReg::x22},
+    {ARMEmitter::XReg::x23, ARMEmitter::XReg::x24},
+    {ARMEmitter::XReg::x25, ARMEmitter::XReg::x26},
+    {ARMEmitter::XReg::x27, ARMEmitter::XReg::x28},
+    {ARMEmitter::XReg::x29, ARMEmitter::XReg::x30},
   }};
 
   for (auto &RegPair : CalleeSaved) {
-    stp(RegPair.first, RegPair.second, PairOffset);
+    stp<ARMEmitter::IndexType::PRE>(RegPair.first, RegPair.second, ARMEmitter::Reg::rsp, -16);
   }
 
   // Additionally we need to store the lower 64bits of v8-v15
   // Here's a fun thing, we can use two ST4 instructions to store everything
   // We just need a single sub to sp before that
   const std::array<
-    std::tuple<vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister>, 2> FPRs = {{
-    {v8, v9, v10, v11},
-    {v12, v13, v14, v15},
+    std::tuple<ARMEmitter::DRegister,
+               ARMEmitter::DRegister,
+               ARMEmitter::DRegister,
+               ARMEmitter::DRegister>, 2> FPRs = {{
+    {ARMEmitter::DReg::d8, ARMEmitter::DReg::d9, ARMEmitter::DReg::d10, ARMEmitter::DReg::d11},
+    {ARMEmitter::DReg::d12, ARMEmitter::DReg::d13, ARMEmitter::DReg::d14, ARMEmitter::DReg::d15},
   }};
 
   uint32_t VectorSaveSize = sizeof(uint64_t) * 8;
-  sub(sp, sp, VectorSaveSize);
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, VectorSaveSize);
   // SP supporting move
   // We just saved x19 so it is safe
-  add(x19, sp, 0);
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r19, ARMEmitter::Reg::rsp, 0);
 
-  MemOperand QuadOffset(x19, 32, PostIndex);
   for (auto &RegQuad : FPRs) {
-    st4(std::get<0>(RegQuad).D(),
-        std::get<1>(RegQuad).D(),
-        std::get<2>(RegQuad).D(),
-        std::get<3>(RegQuad).D(),
+    st4(ARMEmitter::SubRegSize::i64Bit,
+        std::get<0>(RegQuad),
+        std::get<1>(RegQuad),
+        std::get<2>(RegQuad),
+        std::get<3>(RegQuad),
         0,
-        QuadOffset);
+        ARMEmitter::Reg::r19,
+        32);
   }
 }
 
 void Arm64Emitter::PopCalleeSavedRegisters() {
   const std::array<
-    std::tuple<vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister,
-               vixl::aarch64::VRegister>, 2> FPRs = {{
-    {v12, v13, v14, v15},
-    {v8, v9, v10, v11},
+    std::tuple<ARMEmitter::DRegister,
+               ARMEmitter::DRegister,
+               ARMEmitter::DRegister,
+               ARMEmitter::DRegister>, 2> FPRs = {{
+    {ARMEmitter::DReg::d12, ARMEmitter::DReg::d13, ARMEmitter::DReg::d14, ARMEmitter::DReg::d15},
+    {ARMEmitter::DReg::d8, ARMEmitter::DReg::d9, ARMEmitter::DReg::d10, ARMEmitter::DReg::d11},
   }};
 
-  MemOperand QuadOffset(sp, 32, PostIndex);
   for (auto &RegQuad : FPRs) {
-    ld4(std::get<0>(RegQuad).D(),
-        std::get<1>(RegQuad).D(),
-        std::get<2>(RegQuad).D(),
-        std::get<3>(RegQuad).D(),
+    ld4(ARMEmitter::SubRegSize::i64Bit,
+        std::get<0>(RegQuad),
+        std::get<1>(RegQuad),
+        std::get<2>(RegQuad),
+        std::get<3>(RegQuad),
         0,
-        QuadOffset);
+        ARMEmitter::Reg::rsp,
+        32);
   }
 
-  MemOperand PairOffset(sp, 16, PostIndex);
-  const std::array<std::pair<vixl::aarch64::Register, vixl::aarch64::Register>, 6> CalleeSaved = {{
-    {x29, x30},
-    {x27, x28},
-    {x25, x26},
-    {x23, x24},
-    {x21, x22},
-    {x19, x20},
+  const std::array<std::pair<ARMEmitter::XRegister, ARMEmitter::XRegister>, 6> CalleeSaved = {{
+    {ARMEmitter::XReg::x29, ARMEmitter::XReg::x30},
+    {ARMEmitter::XReg::x27, ARMEmitter::XReg::x28},
+    {ARMEmitter::XReg::x25, ARMEmitter::XReg::x26},
+    {ARMEmitter::XReg::x23, ARMEmitter::XReg::x24},
+    {ARMEmitter::XReg::x21, ARMEmitter::XReg::x22},
+    {ARMEmitter::XReg::x19, ARMEmitter::XReg::x20},
   }};
 
   for (auto &RegPair : CalleeSaved) {
-    ldp(RegPair.first, RegPair.second, PairOffset);
+    ldp<ARMEmitter::IndexType::POST>(RegPair.first, RegPair.second, ARMEmitter::Reg::rsp, 16);
   }
 }
 
-
 void Arm64Emitter::SpillStaticRegs(bool FPRs, uint32_t GPRSpillMask, uint32_t FPRSpillMask) {
-  if (StaticRegisterAllocation()) {
-    for (size_t i = 0; i < SRA64.size(); i+=2) {
-      auto Reg1 = SRA64[i];
-      auto Reg2 = SRA64[i+1];
-      if (((1U << Reg1.GetCode()) & GPRSpillMask) &&
-          ((1U << Reg2.GetCode()) & GPRSpillMask)) {
-        stp(Reg1, Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
-      }
-      else if (((1U << Reg1.GetCode()) & GPRSpillMask)) {
-        str(Reg1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
-      }
-      else if (((1U << Reg2.GetCode()) & GPRSpillMask)) {
-        str(Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1])));
-      }
+  if (!StaticRegisterAllocation()) {
+    return;
+  }
+
+  for (size_t i = 0; i < SRA64.size(); i+=2) {
+    auto Reg1 = SRA64[i];
+    auto Reg2 = SRA64[i+1];
+    if (((1U << Reg1.Idx()) & GPRSpillMask) &&
+        ((1U << Reg2.Idx()) & GPRSpillMask)) {
+      stp<ARMEmitter::IndexType::OFFSET>(Reg1.X(), Reg2.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i]));
     }
+    else if (((1U << Reg1.Idx()) & GPRSpillMask)) {
+      str(Reg1.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i]));
+    }
+    else if (((1U << Reg2.Idx()) & GPRSpillMask)) {
+      str(Reg2.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1]));
+    }
+  }
 
-    if (FPRs) {
-      if (EmitterCTX->HostFeatures.SupportsAVX) {
-        for (size_t i = 0; i < SRAFPR.size(); i++) {
-          const auto Reg = SRAFPR[i];
+  if (FPRs) {
+    if (EmitterCTX->HostFeatures.SupportsAVX) {
+      for (size_t i = 0; i < SRAFPR.size(); i++) {
+        const auto Reg = SRAFPR[i];
 
-          if (((1U << Reg.GetCode()) & FPRSpillMask) != 0) {
-            mov(TMP4, offsetof(Core::CpuStateFrame, State.xmm.avx.data[i][0]));
-            st1b(Reg.Z().VnB(), PRED_TMP_32B, SVEMemOperand(STATE, TMP4));
-          }
+        if (((1U << Reg.Idx()) & FPRSpillMask) != 0) {
+          mov(ARMEmitter::Size::i64Bit, TMP4.R(), offsetof(Core::CpuStateFrame, State.xmm.avx.data[i][0]));
+          st1b<ARMEmitter::SubRegSize::i8Bit>(Reg, PRED_TMP_32B, STATE.R(), TMP4.R());
         }
-      } else {
+      }
+    } else {
+      if (GPRSpillMask && FPRSpillMask == ~0U) {
+        // Optimize the common case where we can spill four registers per instruction
+        auto TmpReg = SRA64[__builtin_ffs(GPRSpillMask)];
+
+        // Load the sse offset in to the temporary register
+        add(ARMEmitter::Size::i64Bit, TmpReg, STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[0][0]));
+        for (size_t i = 0; i < SRAFPR.size(); i += 4) {
+          const auto Reg1 = SRAFPR[i];
+          const auto Reg2 = SRAFPR[i + 1];
+          const auto Reg3 = SRAFPR[i + 2];
+          const auto Reg4 = SRAFPR[i + 3];
+          st1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), TmpReg, 64);
+        }
+      }
+      else {
         for (size_t i = 0; i < SRAFPR.size(); i += 2) {
           const auto Reg1 = SRAFPR[i];
           const auto Reg2 = SRAFPR[i + 1];
 
-          if (((1U << Reg1.GetCode()) & FPRSpillMask) &&
-              ((1U << Reg2.GetCode()) & FPRSpillMask)) {
-            stp(Reg1.Q(), Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0])));
+          if (((1U << Reg1.Idx()) & FPRSpillMask) &&
+              ((1U << Reg2.Idx()) & FPRSpillMask)) {
+            stp<ARMEmitter::IndexType::OFFSET>(Reg1.Q(), Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
           }
-          else if (((1U << Reg1.GetCode()) & FPRSpillMask)) {
-            str(Reg1.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0])));
+          else if (((1U << Reg1.Idx()) & FPRSpillMask)) {
+            str(Reg1.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
           }
-          else if (((1U << Reg2.GetCode()) & FPRSpillMask)) {
-            str(Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i+1][0])));
+          else if (((1U << Reg2.Idx()) & FPRSpillMask)) {
+            str(Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i+1][0]));
           }
         }
       }
@@ -243,126 +256,137 @@ void Arm64Emitter::SpillStaticRegs(bool FPRs, uint32_t GPRSpillMask, uint32_t FP
 }
 
 void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRFillMask) {
-  if (StaticRegisterAllocation()) {
-    for (size_t i = 0; i < SRA64.size(); i+=2) {
-      auto Reg1 = SRA64[i];
-      auto Reg2 = SRA64[i+1];
-      if (((1U << Reg1.GetCode()) & GPRFillMask) &&
-          ((1U << Reg2.GetCode()) & GPRFillMask)) {
-        ldp(Reg1, Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
-      }
-      else if (((1U << Reg1.GetCode()) & GPRFillMask)) {
-        ldr(Reg1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
-      }
-      else if (((1U << Reg2.GetCode()) & GPRFillMask)) {
-        ldr(Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1])));
-      }
-    }
+  if (!StaticRegisterAllocation()) {
+    return;
+  }
 
-    if (FPRs) {
-      if (EmitterCTX->HostFeatures.SupportsAVX) {
-        // Set up predicate registers.
-        // We don't bother spilling these in SpillStaticRegs,
-        // since all that matters is we restore them on a fill.
-        // It's not a concern if they get trounced by something else.
-        ptrue(PRED_TMP_16B.VnB(), SVE_VL16);
-        ptrue(PRED_TMP_32B.VnB(), SVE_VL32);
+  if (FPRs) {
+    if (EmitterCTX->HostFeatures.SupportsAVX) {
+      // Set up predicate registers.
+      // We don't bother spilling these in SpillStaticRegs,
+      // since all that matters is we restore them on a fill.
+      // It's not a concern if they get trounced by something else.
+      ptrue<ARMEmitter::SubRegSize::i8Bit>(PRED_TMP_16B, ARMEmitter::PredicatePattern::SVE_VL16);
+      ptrue<ARMEmitter::SubRegSize::i8Bit>(PRED_TMP_32B, ARMEmitter::PredicatePattern::SVE_VL32);
 
-        for (size_t i = 0; i < SRAFPR.size(); i++) {
-          const auto Reg = SRAFPR[i];
-
-          if (((1U << Reg.GetCode()) & FPRFillMask) != 0) {
-            mov(TMP4, offsetof(Core::CpuStateFrame, State.xmm.avx.data[i][0]));
-            ld1b(Reg.Z().VnB(), PRED_TMP_32B.Zeroing(), SVEMemOperand(STATE, TMP4));
-          }
+      for (size_t i = 0; i < SRAFPR.size(); i++) {
+        const auto Reg = SRAFPR[i];
+        if (((1U << Reg.Idx()) & FPRFillMask) != 0) {
+          mov(ARMEmitter::Size::i64Bit, TMP4.R(), offsetof(Core::CpuStateFrame, State.xmm.avx.data[i][0]));
+          ld1b<ARMEmitter::SubRegSize::i8Bit>(Reg, PRED_TMP_32B, STATE.R(), TMP4.R());
         }
-      } else {
+      }
+    } else {
+      if (GPRFillMask && FPRFillMask == ~0U) {
+        // Optimize the common case where we can fill four registers per instruction.
+        // Use one of the filling static registers before we fill it.
+        auto TmpReg = SRA64[__builtin_ffs(GPRFillMask)];
+
+        // Load the sse offset in to the temporary register
+        add(ARMEmitter::Size::i64Bit, TmpReg, STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[0][0]));
+        for (size_t i = 0; i < SRAFPR.size(); i += 4) {
+          const auto Reg1 = SRAFPR[i];
+          const auto Reg2 = SRAFPR[i + 1];
+          const auto Reg3 = SRAFPR[i + 2];
+          const auto Reg4 = SRAFPR[i + 3];
+          ld1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), TmpReg, 64);
+        }
+      }
+      else {
         for (size_t i = 0; i < SRAFPR.size(); i += 2) {
           const auto Reg1 = SRAFPR[i];
           const auto Reg2 = SRAFPR[i + 1];
 
-          if (((1U << Reg1.GetCode()) & FPRFillMask) &&
-              ((1U << Reg2.GetCode()) & FPRFillMask)) {
-            ldp(Reg1.Q(), Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0])));
+          if (((1U << Reg1.Idx()) & FPRFillMask) &&
+              ((1U << Reg2.Idx()) & FPRFillMask)) {
+            ldp<ARMEmitter::IndexType::OFFSET>(Reg1.Q(), Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
           }
-          else if (((1U << Reg1.GetCode()) & FPRFillMask)) {
-            ldr(Reg1.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0])));
+          else if (((1U << Reg1.Idx()) & FPRFillMask)) {
+            ldr(Reg1.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
           }
-          else if (((1U << Reg2.GetCode()) & FPRFillMask)) {
-            ldr(Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i+1][0])));
+          else if (((1U << Reg2.Idx()) & FPRFillMask)) {
+            ldr(Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i+1][0]));
           }
         }
       }
     }
   }
+
+  for (size_t i = 0; i < SRA64.size(); i+=2) {
+    auto Reg1 = SRA64[i];
+    auto Reg2 = SRA64[i+1];
+    if (((1U << Reg1.Idx()) & GPRFillMask) &&
+        ((1U << Reg2.Idx()) & GPRFillMask)) {
+      ldp<ARMEmitter::IndexType::OFFSET>(Reg1.X(), Reg2.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i]));
+    }
+    else if ((1U << Reg1.Idx()) & GPRFillMask) {
+      ldr(Reg1.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i]));
+    }
+    else if ((1U << Reg2.Idx()) & GPRFillMask) {
+      ldr(Reg2.X(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1]));
+    }
+  }
 }
 
-void Arm64Emitter::PushDynamicRegsAndLR() {
+void Arm64Emitter::PushDynamicRegsAndLR(FEXCore::ARMEmitter::Register TmpReg) {
   const auto CanUseSVE = EmitterCTX->HostFeatures.SupportsAVX;
-  const auto GPRSize = (RA64.size() + 1) * Core::CPUState::GPR_REG_SIZE;
+  const auto GPRSize = 1 * Core::CPUState::GPR_REG_SIZE;
   const auto FPRRegSize = CanUseSVE ? Core::CPUState::XMM_AVX_REG_SIZE
                                     : Core::CPUState::XMM_SSE_REG_SIZE;
   const auto FPRSize = RAFPR.size() * FPRRegSize;
   const uint64_t SPOffset = AlignUp(GPRSize + FPRSize, 16);
 
-  sub(sp, sp, SPOffset);
-  int i = 0;
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, SPOffset);
+
+  // rsp capable move
+  add(ARMEmitter::Size::i64Bit, TmpReg, ARMEmitter::Reg::rsp, 0);
 
   if (CanUseSVE) {
-    for (const auto& RA : RAFPR) {
-      mov(TMP4, i * 8);
-      st1b(RA.Z().VnB(), PRED_TMP_32B, SVEMemOperand(sp, TMP4));
-      i += 4;
+    for (size_t i = 0; i < RAFPR.size(); i += 4) {
+      const auto Reg1 = RAFPR[i];
+      const auto Reg2 = RAFPR[i + 1];
+      const auto Reg3 = RAFPR[i + 2];
+      const auto Reg4 = RAFPR[i + 3];
+      st4b(Reg1, Reg2, Reg3, Reg4, PRED_TMP_32B, TmpReg, 0);
+      add(ARMEmitter::Size::i64Bit, TmpReg, TmpReg, 32 * 4);
     }
   } else {
-    for (const auto& RA : RAFPR) {
-      str(RA.Q(), MemOperand(sp, i * 8));
-      i += 2;
+    static_assert(RAFPR.size() % 4 == 0, "Needs to have multiple of 4 FPRs for RA");
+    for (size_t i = 0; i < RAFPR.size(); i += 4) {
+      const auto Reg1 = RAFPR[i];
+      const auto Reg2 = RAFPR[i + 1];
+      const auto Reg3 = RAFPR[i + 2];
+      const auto Reg4 = RAFPR[i + 3];
+      st1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), TmpReg, 64);
     }
   }
 
-#if 0 // All GPRs should be caller saved
-  for (const auto& RA : RA64) {
-    str(RA, MemOperand(sp, i * 8));
-    i++;
-  }
-#endif
-
-  str(lr, MemOperand(sp, i * 8));
+  str(ARMEmitter::XReg::lr, TmpReg, 0);
 }
 
 void Arm64Emitter::PopDynamicRegsAndLR() {
   const auto CanUseSVE = EmitterCTX->HostFeatures.SupportsAVX;
-  const auto GPRSize = (RA64.size() + 1) * Core::CPUState::GPR_REG_SIZE;
-  const auto FPRRegSize = CanUseSVE ? Core::CPUState::XMM_AVX_REG_SIZE
-                                    : Core::CPUState::XMM_SSE_REG_SIZE;
-  const auto FPRSize = RAFPR.size() * FPRRegSize;
-  const uint64_t SPOffset = AlignUp(GPRSize + FPRSize, 16);
-  int i = 0;
 
   if (CanUseSVE) {
-    for (const auto& RA : RAFPR) {
-      mov(TMP4, i * 8);
-      ld1b(RA.Z().VnB(), PRED_TMP_32B.Zeroing(), SVEMemOperand(sp, TMP4));
-      i += 4;
+    for (size_t i = 0; i < RAFPR.size(); i += 4) {
+      const auto Reg1 = RAFPR[i];
+      const auto Reg2 = RAFPR[i + 1];
+      const auto Reg3 = RAFPR[i + 2];
+      const auto Reg4 = RAFPR[i + 3];
+      ld4b(Reg1, Reg2, Reg3, Reg4, PRED_TMP_32B, ARMEmitter::Reg::rsp);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 32 * 4);
     }
   } else {
-    for (const auto& RA : RAFPR) {
-      ldr(RA.Q(), MemOperand(sp, i * 8));
-      i += 2;
+    for (size_t i = 0; i < RAFPR.size(); i += 4) {
+      const auto Reg1 = RAFPR[i];
+      const auto Reg2 = RAFPR[i + 1];
+      const auto Reg3 = RAFPR[i + 2];
+      const auto Reg4 = RAFPR[i + 3];
+      ld1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), ARMEmitter::Reg::rsp, 64);
     }
   }
 
-#if 0 // All GPRs should be caller saved
-  for (const auto& RA : RA64) {
-    ldr(RA, MemOperand(sp, i * 8));
-    i++;
-  }
-#endif
-
-  ldr(lr, MemOperand(sp, i * 8));
-
-  add(sp, sp, SPOffset);
+  ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
 }
 
 void Arm64Emitter::Align16B() {

External/FEXCore/Source/Interface/Core/ArchHelpers/Arm64Emitter.h

@@ -1,5 +1,9 @@
 #pragma once
 
+#include "FEXCore/Utils/EnumUtils.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Registers.h"
+
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/ObjectCache/Relocations.h"
 
@@ -8,6 +12,9 @@
 #include <aarch64/cpu-aarch64.h>
 #include <aarch64/operands-aarch64.h>
 #include <platform-vixl.h>
+#ifdef VIXL_DISASSEMBLER
+#include <aarch64/disasm-aarch64.h>
+#endif
 #ifdef VIXL_SIMULATOR
 #include <aarch64/simulator-aarch64.h>
 #include <aarch64/simulator-constants-aarch64.h>
@@ -21,77 +28,70 @@
 #include <utility>
 
 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-
 // All but x29 are caller saved
-const std::array<aarch64::Register, 16> SRA64 = {
-  x4, x5, x6, x7, x8, x9, x10, x11,
-  x12, x18, x17, x16, x15, x14, x13, x29
+constexpr std::array<FEXCore::ARMEmitter::Register, 16> SRA64 = {
+  FEXCore::ARMEmitter::Reg::r4, FEXCore::ARMEmitter::Reg::r5, FEXCore::ARMEmitter::Reg::r6, FEXCore::ARMEmitter::Reg::r7, FEXCore::ARMEmitter::Reg::r8, FEXCore::ARMEmitter::Reg::r9, FEXCore::ARMEmitter::Reg::r10, FEXCore::ARMEmitter::Reg::r11,
+  FEXCore::ARMEmitter::Reg::r12, FEXCore::ARMEmitter::Reg::r18, FEXCore::ARMEmitter::Reg::r17, FEXCore::ARMEmitter::Reg::r16, FEXCore::ARMEmitter::Reg::r15, FEXCore::ARMEmitter::Reg::r14, FEXCore::ARMEmitter::Reg::r13, FEXCore::ARMEmitter::Reg::r29
 };
 
 // All are callee saved
-const std::array<aarch64::Register, 9> RA64 = {
-  x20, x21, x22, x23, x24, x25, x26, x27,
-  x19
+constexpr std::array<FEXCore::ARMEmitter::Register, 9> RA64 = {
+  FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21, FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23, FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25, FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27,
+  FEXCore::ARMEmitter::Reg::r19
 };
 
-const std::array<std::pair<aarch64::Register, aarch64::Register>, 4>  RA64Pair = {{
-  {x20, x21},
-  {x22, x23},
-  {x24, x25},
-  {x26, x27},
-}};
-
-const std::array<std::pair<aarch64::Register, aarch64::Register>, 4> RA32Pair = {{
-  {w20, w21},
-  {w22, w23},
-  {w24, w25},
-  {w26, w27},
+constexpr std::array<std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register>, 4>  RA64Pair = {{
+  {FEXCore::ARMEmitter::Reg::r20, FEXCore::ARMEmitter::Reg::r21},
+  {FEXCore::ARMEmitter::Reg::r22, FEXCore::ARMEmitter::Reg::r23},
+  {FEXCore::ARMEmitter::Reg::r24, FEXCore::ARMEmitter::Reg::r25},
+  {FEXCore::ARMEmitter::Reg::r26, FEXCore::ARMEmitter::Reg::r27},
 }};
 
 // All are caller saved
-const std::array<aarch64::VRegister, 16> SRAFPR = {
-  v16, v17, v18, v19, v20, v21, v22, v23,
-  v24, v25, v26, v27, v28, v29, v30, v31
+constexpr std::array<FEXCore::ARMEmitter::VRegister, 16> SRAFPR = {
+  FEXCore::ARMEmitter::VReg::v16, FEXCore::ARMEmitter::VReg::v17, FEXCore::ARMEmitter::VReg::v18, FEXCore::ARMEmitter::VReg::v19, FEXCore::ARMEmitter::VReg::v20, FEXCore::ARMEmitter::VReg::v21, FEXCore::ARMEmitter::VReg::v22, FEXCore::ARMEmitter::VReg::v23,
+  FEXCore::ARMEmitter::VReg::v24, FEXCore::ARMEmitter::VReg::v25, FEXCore::ARMEmitter::VReg::v26, FEXCore::ARMEmitter::VReg::v27, FEXCore::ARMEmitter::VReg::v28, FEXCore::ARMEmitter::VReg::v29, FEXCore::ARMEmitter::VReg::v30, FEXCore::ARMEmitter::VReg::v31
 };
 
 //  v8..v15 = (lower 64bits) Callee saved
-const std::array<aarch64::VRegister, 12> RAFPR = {
-/*v0,  v1,  v2,  v3,*/v4,  v5,  v6,  v7,  // v0 ~ v3 are used as temps
-  v8,  v9,  v10, v11, v12, v13, v14, v15
+constexpr std::array<FEXCore::ARMEmitter::VRegister, 12> RAFPR = {
+/*FEXCore::ARMEmitter::VReg::v0,  FEXCore::ARMEmitter::VReg::v1,  FEXCore::ARMEmitter::VReg::v2,  FEXCore::ARMEmitter::VReg::v3,*/FEXCore::ARMEmitter::VReg::v4,  FEXCore::ARMEmitter::VReg::v5,  FEXCore::ARMEmitter::VReg::v6,  FEXCore::ARMEmitter::VReg::v7,  // FEXCore::ARMEmitter::VReg::v0 ~ FEXCore::ARMEmitter::VReg::v3 are used as temps
+  FEXCore::ARMEmitter::VReg::v8,  FEXCore::ARMEmitter::VReg::v9,  FEXCore::ARMEmitter::VReg::v10, FEXCore::ARMEmitter::VReg::v11, FEXCore::ARMEmitter::VReg::v12, FEXCore::ARMEmitter::VReg::v13, FEXCore::ARMEmitter::VReg::v14, FEXCore::ARMEmitter::VReg::v15
 };
 
 // Contains the address to the currently available CPU state
-#define STATE x28
+constexpr auto STATE = FEXCore::ARMEmitter::XReg::x28;
 
 // GPR temporaries. Only x3 can be used across spill boundaries
 // so if these ever need to change, be very careful about that.
-#define TMP1 x0
-#define TMP2 x1
-#define TMP3 x2
-#define TMP4 x3
+constexpr auto TMP1 = FEXCore::ARMEmitter::XReg::x0;
+constexpr auto TMP2 = FEXCore::ARMEmitter::XReg::x1;
+constexpr auto TMP3 = FEXCore::ARMEmitter::XReg::x2;
+constexpr auto TMP4 = FEXCore::ARMEmitter::XReg::x3;
 
 // Vector temporaries
-#define VTMP1 v1
-#define VTMP2 v2
-#define VTMP3 v3
+constexpr auto VTMP1 = FEXCore::ARMEmitter::VReg::v0;
+constexpr auto VTMP2 = FEXCore::ARMEmitter::VReg::v1;
+constexpr auto VTMP3 = FEXCore::ARMEmitter::VReg::v2;
+constexpr auto VTMP4 = FEXCore::ARMEmitter::VReg::v3;
 
 // Predicate register temporaries (used when AVX support is enabled)
 // PRED_TMP_16B indicates a predicate register that indicates the first 16 bytes set to 1.
 // PRED_TMP_32B indicates a predicate register that indicates the first 32 bytes set to 1.
-#define PRED_TMP_16B p6
-#define PRED_TMP_32B p7
+constexpr FEXCore::ARMEmitter::PRegister PRED_TMP_16B = FEXCore::ARMEmitter::PReg::p6;
+constexpr FEXCore::ARMEmitter::PRegister PRED_TMP_32B = FEXCore::ARMEmitter::PReg::p7;
 
 // This class contains common emitter utility functions that can
 // be used by both Arm64 JIT and ARM64 Dispatcher
-class Arm64Emitter : public vixl::aarch64::Assembler {
+class Arm64Emitter : public FEXCore::ARMEmitter::Emitter {
 protected:
   Arm64Emitter(FEXCore::Context::Context *ctx, size_t size);
+  ~Arm64Emitter();
 
   FEXCore::Context::Context *EmitterCTX;
   vixl::aarch64::CPU CPU;
-  void LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant, bool NOPPad = false);
+  void LoadConstant(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register Reg, uint64_t Constant, bool NOPPad = false);
+
 
   // NOTE: These functions WILL clobber the register TMP4 if AVX support is enabled
   //       and FPRs are being spilled or filled. If only GPRs are spilled/filled, then
@@ -105,13 +105,14 @@ protected:
   // We can't guarantee only the lower 64bits are used so flush everything
   static constexpr uint32_t CALLER_FPR_MASK = ~0U;
 
-  void PushDynamicRegsAndLR();
+  void PushDynamicRegsAndLR(FEXCore::ARMEmitter::Register TmpReg);
   void PopDynamicRegsAndLR();
 
   void PushCalleeSavedRegisters();
   void PopCalleeSavedRegisters();
 
   void Align16B();
+
 #ifdef VIXL_SIMULATOR
   // Generates a vixl simulator runtime call.
   //
@@ -123,61 +124,64 @@ protected:
   // 2) Simulator wrapper handler
   // 3) Function to call
   // 4) Style of the function call (Call versus tail-call)
+
   template<typename R, typename... P>
   void GenerateRuntimeCall(R (*Function)(P...)) {
     uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
-      &(Simulator::RuntimeCallStructHelper<R, P...>::Wrapper));
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<R, P...>::Wrapper));
 
     uintptr_t FunctionAddress = reinterpret_cast<uintptr_t>(Function);
 
-    hlt(kRuntimeCallOpcode);
+    hlt(vixl::aarch64::kRuntimeCallOpcode);
 
     // Simulator wrapper address pointer.
-    dc(SimulatorWrapperAddress);
+    dc64(SimulatorWrapperAddress);
 
     // Runtime function address to call
-    dc(FunctionAddress);
+    dc64(FunctionAddress);
 
     // Call type
-    dc32(kCallRuntime);
+    dc32(vixl::aarch64::kCallRuntime);
   }
 
   template<typename R, typename... P>
-  void GenerateIndirectRuntimeCall(vixl::aarch64::Register Reg) {
+  void GenerateIndirectRuntimeCall(ARMEmitter::Register Reg) {
     uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
-      &(Simulator::RuntimeCallStructHelper<R, P...>::Wrapper));
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<R, P...>::Wrapper));
 
-    hlt(kIndirectRuntimeCallOpcode);
+    hlt(vixl::aarch64::kIndirectRuntimeCallOpcode);
 
     // Simulator wrapper address pointer.
-    dc(SimulatorWrapperAddress);
+    dc64(SimulatorWrapperAddress);
 
     // Register that contains the function to call
-    dc(Reg.GetCode());
+    dc32(Reg.Idx());
 
     // Call type
-    dc32(kCallRuntime);
+    dc32(vixl::aarch64::kCallRuntime);
   }
 
   template<>
-  void GenerateIndirectRuntimeCall<float, __uint128_t>(vixl::aarch64::Register Reg) {
+  void GenerateIndirectRuntimeCall<float, __uint128_t>(ARMEmitter::Register Reg) {
     uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
-      &(Simulator::RuntimeCallStructHelper<float, __uint128_t>::Wrapper));
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<float, __uint128_t>::Wrapper));
 
-    hlt(kIndirectRuntimeCallOpcode);
+    hlt(vixl::aarch64::kIndirectRuntimeCallOpcode);
 
     // Simulator wrapper address pointer.
-    dc(SimulatorWrapperAddress);
+    dc64(SimulatorWrapperAddress);
 
     // Register that contains the function to call
-    dc(Reg.GetCode());
+    dc32(Reg.Idx());
 
     // Call type
-    dc32(kCallRuntime);
+    dc32(vixl::aarch64::kCallRuntime);
   }
 
 #endif
-
+#ifdef VIXL_DISASSEMBLER
+  vixl::aarch64::PrintDisassembler Disasm {stderr};
+#endif
   FEX_CONFIG_OPT(StaticRegisterAllocation, SRA);
 };
 

External/FEXCore/Source/Interface/Core/ArchHelpers/CodeEmitter/ALUOps.inl

@@ -0,0 +1,912 @@
+/* ALU instruction emitters.
+ *
+ * Almost all of these operations have `ARMEmitter::Size` as their first argument.
+ * This allows both 32-bit and 64-bit selection of how that instruction is going to operate.
+ *
+ * Some emitter operations explicitly use `XRegister` or `WRegister`.
+ * This is usually due to the instruction only supporting one operating size.
+ * Although in some cases is a minor convenience without any performance implications.
+ *
+ * FEX-Emu ALU operations usually have a 32-bit or 64-bit operating size encoded in the IR operation,
+ * This allows FEX to use a single helper function which decodes to both handlers.
+ */
+public:
+  // PC relative
+  void adr(FEXCore::ARMEmitter::Register rd, uint32_t Imm) {
+    constexpr uint32_t Op = 0b0001'0000 << 24;
+    DataProcessing_PCRel_Imm(Op, rd, Imm);
+  }
+
+  void adr(FEXCore::ARMEmitter::Register rd, BackwardLabel const* Label) {
+    int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+    LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575, "Unscaled offset too large");
+
+    constexpr uint32_t Op = 0b0001'0000 << 24;
+    DataProcessing_PCRel_Imm(Op, rd, Imm);
+  }
+  void adr(FEXCore::ARMEmitter::Register rd, ForwardLabel *Label) {
+    Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::ADR });
+    constexpr uint32_t Op = 0b0001'0000 << 24;
+    DataProcessing_PCRel_Imm(Op, rd, 0);
+  }
+
+  void adr(FEXCore::ARMEmitter::Register rd, BiDirectionalLabel *Label) {
+    if (Label->Backward.Location) {
+      adr(rd, &Label->Backward);
+    }
+    else {
+      adr(rd, &Label->Forward);
+    }
+  }
+
+  void adrp(FEXCore::ARMEmitter::Register rd, uint32_t Imm) {
+    constexpr uint32_t Op = 0b1001'0000 << 24;
+    DataProcessing_PCRel_Imm(Op, rd, Imm);
+  }
+
+  void adrp(FEXCore::ARMEmitter::Register rd, BackwardLabel const* Label) {
+    int64_t Imm = reinterpret_cast<int64_t>(Label->Location) - (GetCursorAddress<int64_t>() & ~0xFFFLL);
+    LOGMAN_THROW_A_FMT(Imm >= -4294967296 && Imm <= 4294963200 && (Imm & 0xFFF) == 0, "Unscaled offset too large");
+
+    constexpr uint32_t Op = 0b1001'0000 << 24;
+    DataProcessing_PCRel_Imm(Op, rd, Imm);
+  }
+  void adrp(FEXCore::ARMEmitter::Register rd, ForwardLabel *Label) {
+    Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::ADRP });
+    constexpr uint32_t Op = 0b1001'0000 << 24;
+    DataProcessing_PCRel_Imm(Op, rd, 0);
+  }
+
+  void adrp(FEXCore::ARMEmitter::Register rd, BiDirectionalLabel *Label) {
+    if (Label->Backward.Location) {
+      adrp(rd, &Label->Backward);
+    }
+    else {
+      adrp(rd, &Label->Forward);
+    }
+  }
+
+  // Add/subtract immediate
+  void add(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t Imm, bool LSL12 = false) {
+    constexpr uint32_t Op = 0b0001'0001'0 << 23;
+    DataProcessing_AddSub_Imm(Op, s, rd, rn, Imm, LSL12);
+  }
+
+  void adds(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t Imm, bool LSL12 = false) {
+    constexpr uint32_t Op = 0b0011'0001'0 << 23;
+    DataProcessing_AddSub_Imm(Op, s, rd, rn, Imm, LSL12);
+  }
+
+  void sub(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t Imm, bool LSL12 = false) {
+    constexpr uint32_t Op = 0b0101'0001'0 << 23;
+    DataProcessing_AddSub_Imm(Op, s, rd, rn, Imm, LSL12);
+  }
+
+  void cmp(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rn, uint32_t Imm, bool LSL12 = false) {
+    constexpr uint32_t Op = 0b0111'0001'0 << 23;
+    DataProcessing_AddSub_Imm(Op, s, FEXCore::ARMEmitter::Reg::rsp, rn, Imm, LSL12);
+  }
+
+  void subs(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t Imm, bool LSL12 = false) {
+    constexpr uint32_t Op = 0b0111'0001'0 << 23;
+    DataProcessing_AddSub_Imm(Op, s, rd, rn, Imm, LSL12);
+  }
+
+  // Logical immediate
+  void and_(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint64_t Imm) {
+    uint32_t n, immr, imms;
+    [[maybe_unused]] const auto IsImm = vixl::aarch64::Assembler::IsImmLogical(Imm,
+                           RegSizeInBits(s),
+                           &n,
+                           &imms,
+                           &immr);
+    LOGMAN_THROW_A_FMT(IsImm, "Couldn't encode immediate to logical op");
+    and_(s, rd, rn, n, immr, imms);
+  }
+
+  void bic(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint64_t Imm) {
+    and_(s, rd, rn, ~Imm);
+  }
+
+  void ands(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint64_t Imm) {
+    uint32_t n, immr, imms;
+    [[maybe_unused]] const auto IsImm = vixl::aarch64::Assembler::IsImmLogical(Imm,
+                           RegSizeInBits(s),
+                           &n,
+                           &imms,
+                           &immr);
+    LOGMAN_THROW_A_FMT(IsImm, "Couldn't encode immediate to logical op");
+    ands(s, rd, rn, n, immr, imms);
+  }
+
+  void bics(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint64_t Imm) {
+    ands(s, rd, rn, ~Imm);
+  }
+
+  void orr(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint64_t Imm) {
+    uint32_t n, immr, imms;
+    [[maybe_unused]] const auto IsImm = vixl::aarch64::Assembler::IsImmLogical(Imm,
+                           RegSizeInBits(s),
+                           &n,
+                           &imms,
+                           &immr);
+    LOGMAN_THROW_A_FMT(IsImm, "Couldn't encode immediate to logical op");
+    orr(s, rd, rn, n, immr, imms);
+  }
+
+  void eor(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint64_t Imm) {
+    uint32_t n, immr, imms;
+    [[maybe_unused]] const auto IsImm = vixl::aarch64::Assembler::IsImmLogical(Imm,
+                           RegSizeInBits(s),
+                           &n,
+                           &imms,
+                           &immr);
+    LOGMAN_THROW_A_FMT(IsImm, "Couldn't encode immediate to logical op");
+    eor(s, rd, rn, n, immr, imms);
+  }
+
+  // Move wide immediate
+  void movn(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, uint32_t Imm, uint32_t Offset = 0) {
+    LOGMAN_THROW_A_FMT((Imm & 0xFFFF0000U) == 0, "Upper bits of move wide not valid");
+    LOGMAN_THROW_A_FMT((Offset % 16) == 0, "Offset must be 16bit aligned");
+
+    constexpr uint32_t Op = 0b001'0010'100 << 21;
+    DataProcessing_MoveWide(Op, s, rd, Imm, Offset >> 4);
+  }
+  void mov(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, uint32_t Imm) {
+    movz(s, rd, Imm, 0);
+  }
+  void mov(FEXCore::ARMEmitter::XRegister rd, uint32_t Imm) {
+    movz(FEXCore::ARMEmitter::Size::i64Bit, rd.R(), Imm, 0);
+  }
+  void mov(FEXCore::ARMEmitter::WRegister rd, uint32_t Imm) {
+    movz(FEXCore::ARMEmitter::Size::i32Bit, rd.R(), Imm, 0);
+  }
+
+  void movz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, uint32_t Imm, uint32_t Offset = 0) {
+    LOGMAN_THROW_A_FMT((Imm & 0xFFFF0000U) == 0, "Upper bits of move wide not valid");
+    LOGMAN_THROW_A_FMT((Offset % 16) == 0, "Offset must be 16bit aligned");
+
+    constexpr uint32_t Op = 0b101'0010'100 << 21;
+    DataProcessing_MoveWide(Op, s, rd, Imm, Offset >> 4);
+  }
+  void movk(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, uint32_t Imm, uint32_t Offset = 0) {
+    LOGMAN_THROW_A_FMT((Imm & 0xFFFF0000U) == 0, "Upper bits of move wide not valid");
+    LOGMAN_THROW_A_FMT((Offset % 16) == 0, "Offset must be 16bit aligned");
+
+    constexpr uint32_t Op = 0b111'0010'100 << 21;
+    DataProcessing_MoveWide(Op, s, rd, Imm, Offset >> 4);
+  }
+
+  void movn(FEXCore::ARMEmitter::XRegister rd, uint32_t Imm, uint32_t Offset = 0) {
+    movn(FEXCore::ARMEmitter::Size::i64Bit, rd.R(), Imm, Offset);
+  }
+  void movz(FEXCore::ARMEmitter::XRegister rd, uint32_t Imm, uint32_t Offset = 0) {
+    movz(FEXCore::ARMEmitter::Size::i64Bit, rd.R(), Imm, Offset);
+  }
+  void movk(FEXCore::ARMEmitter::XRegister rd, uint32_t Imm, uint32_t Offset = 0) {
+    movk(FEXCore::ARMEmitter::Size::i64Bit, rd.R(), Imm, Offset);
+  }
+  void movn(FEXCore::ARMEmitter::WRegister rd, uint32_t Imm, uint32_t Offset = 0) {
+    movn(FEXCore::ARMEmitter::Size::i32Bit, rd.R(), Imm, Offset);
+  }
+  void movz(FEXCore::ARMEmitter::WRegister rd, uint32_t Imm, uint32_t Offset = 0) {
+    movz(FEXCore::ARMEmitter::Size::i32Bit, rd.R(), Imm, Offset);
+  }
+  void movk(FEXCore::ARMEmitter::WRegister rd, uint32_t Imm, uint32_t Offset = 0) {
+    movk(FEXCore::ARMEmitter::Size::i32Bit, rd.R(), Imm, Offset);
+  }
+
+  // Bitfield
+  void sxtb(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    sbfm(s, rd, rn, 0, 7);
+  }
+  void sxth(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    sbfm(s, rd, rn, 0, 15);
+  }
+  void sxtw(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn) {
+    sbfm(ARMEmitter::Size::i64Bit, rd, rn, 0, 31);
+  }
+  void sbfx(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t lsb, uint32_t width) {
+    LOGMAN_THROW_A_FMT(width > 0, "sbfx needs width > 0");
+    LOGMAN_THROW_A_FMT((lsb + width) <= RegSizeInBits(s), "Tried to sbfx a region larger than the register");
+    sbfm(s, rd, rn, lsb, lsb + width - 1);
+  }
+  void asr(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t shift) {
+    LOGMAN_THROW_A_FMT(shift <= RegSizeInBits(s), "Tried to asr a region larger than the register");
+    sbfm(s, rd, rn, shift, RegSizeInBits(s) - 1);
+  }
+
+  void uxtb(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    ubfm(s, rd, rn, 0, 7);
+  }
+  void uxth(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    ubfm(s, rd, rn, 0, 15);
+  }
+  void uxtw(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    ubfm(s, rd, rn, 0, 31);
+  }
+
+  void ubfm(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t immr, uint32_t imms) {
+    constexpr uint32_t Op = 0b0101'0011'00 << 22;
+    DataProcessing_Logical_Imm(Op, s, rd, rn, s == ARMEmitter::Size::i64Bit, immr, imms);
+  }
+
+  void lsl(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t shift) {
+    const auto RegSize = RegSizeInBits(s);
+    LOGMAN_THROW_A_FMT(shift < RegSize, "Tried to asr a region larger than the register");
+    ubfm(s, rd, rn, (RegSize - shift) % RegSize, RegSize - shift - 1);
+  }
+  void lsr(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t shift) {
+    const auto RegSize = RegSizeInBits(s);
+    LOGMAN_THROW_A_FMT(shift < RegSize, "Tried to asr a region larger than the register");
+    ubfm(s, rd, rn, shift, RegSize - 1);
+  }
+  void ubfx(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t lsb, uint32_t width) {
+    LOGMAN_THROW_A_FMT(width > 0, "ubfx needs width > 0");
+    LOGMAN_THROW_A_FMT((lsb + width) <= RegSizeInBits(s), "Tried to ubfx a region larger than the register");
+    ubfm(s, rd, rn, lsb, lsb + width - 1);
+  }
+
+  void bfi(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t lsb, uint32_t width) {
+    const auto RegSize = RegSizeInBits(s);
+    LOGMAN_THROW_A_FMT(width > 0, "sbfx needs width > 0");
+    LOGMAN_THROW_A_FMT((lsb + width) <= RegSize, "Tried to sbfx a region larger than the register");
+    bfm(s, rd, rn, (RegSize - lsb) & (RegSize - 1), width - 1);
+  }
+
+  // Extract
+  void extr(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, uint32_t Imm) {
+    constexpr uint32_t Op = 0b001'0011'100 << 21;
+    LOGMAN_THROW_A_FMT(Imm < RegSizeInBits(s), "Tried to extr a region larger than the register");
+    DataProcessing_Extract(Op, s, rd, rn, rm, Imm);
+  }
+
+  void ror(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t Imm) {
+    LOGMAN_THROW_A_FMT(Imm < RegSizeInBits(s), "Tried to extr a region larger than the register");
+    extr(s, rd, rn, rn, Imm);
+  }
+
+  // Data processing - 2 source
+  void udiv(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0000'10U << 10);
+    DataProcessing_2Source(Op, s, rd, rn, rm);
+  }
+  void sdiv(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0000'11U << 10);
+    DataProcessing_2Source(Op, s, rd, rn, rm);
+  }
+
+  void lslv(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0010'00U << 10);
+    DataProcessing_2Source(Op, s, rd, rn, rm);
+  }
+  void lsrv(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0010'01U << 10);
+    DataProcessing_2Source(Op, s, rd, rn, rm);
+  }
+  void asrv(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0010'10U << 10);
+    DataProcessing_2Source(Op, s, rd, rn, rm);
+  }
+  void rorv(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0010'11U << 10);
+    DataProcessing_2Source(Op, s, rd, rn, rm);
+  }
+  void crc32b(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0100'00U << 10);
+    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
+  }
+  void crc32h(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0100'01U << 10);
+    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
+  }
+  void crc32w(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0100'10U << 10);
+    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
+  }
+  void crc32cb(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0101'00U << 10);
+    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
+  }
+  void crc32ch(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0101'01U << 10);
+    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
+  }
+  void crc32cw(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0101'10U << 10);
+    DataProcessing_2Source(Op, ARMEmitter::Size::i32Bit, rd, rn, rm);
+  }
+  void subp(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0000'00U << 10);
+    DataProcessing_2Source(Op, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm);
+  }
+  void irg(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0001'00U << 10);
+    DataProcessing_2Source(Op, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm);
+  }
+  void gmi(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0001'01U << 10);
+    DataProcessing_2Source(Op, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm);
+  }
+  void pacga(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0011'00U << 10);
+    DataProcessing_2Source(Op, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm);
+  }
+  void crc32x(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0100'11U << 10);
+    DataProcessing_2Source(Op, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm);
+  }
+  void crc32cx(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm) {
+    constexpr uint32_t Op = (0b001'1010'110U << 21) |
+                        (0b0101'11U << 10);
+    DataProcessing_2Source(Op, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm);
+  }
+  void subps(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm) {
+    constexpr uint32_t Op = (0b011'1010'110U << 21) |
+                        (0b0000'00U << 10);
+    DataProcessing_2Source(Op, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm);
+  }
+
+  // Data processing - 1 source
+  void rbit(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    constexpr uint32_t Op = (0b101'1010'110U << 21) |
+                        (0b0'0000U << 16) |
+                        (0b0000'00U << 10);
+    DataProcessing_1Source(Op, s, rd, rn);
+  }
+  void rev16(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    constexpr uint32_t Op = (0b101'1010'110U << 21) |
+                        (0b0'0000U << 16) |
+                        (0b0000'01U << 10);
+    DataProcessing_1Source(Op, s, rd, rn);
+  }
+  void rev(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn) {
+    constexpr uint32_t Op = (0b101'1010'110U << 21) |
+                        (0b0'0000U << 16) |
+                        (0b0000'10U << 10);
+    DataProcessing_1Source(Op, FEXCore::ARMEmitter::Size::i32Bit, rd, rn);
+  }
+  void rev32(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn) {
+    constexpr uint32_t Op = (0b101'1010'110U << 21) |
+                        (0b0'0000U << 16) |
+                        (0b0000'10U << 10);
+    DataProcessing_1Source(Op, FEXCore::ARMEmitter::Size::i64Bit, rd, rn);
+  }
+  void clz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    constexpr uint32_t Op = (0b101'1010'110U << 21) |
+                        (0b0'0000U << 16) |
+                        (0b0001'00U << 10);
+    DataProcessing_1Source(Op, s, rd, rn);
+  }
+  void cls(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    constexpr uint32_t Op = (0b101'1010'110U << 21) |
+                        (0b0'0000U << 16) |
+                        (0b0001'01U << 10);
+    DataProcessing_1Source(Op, s, rd, rn);
+  }
+  void rev(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn) {
+    constexpr uint32_t Op = (0b101'1010'110U << 21) |
+                        (0b0'0000U << 16) |
+                        (0b0000'11U << 10);
+    DataProcessing_1Source(Op, FEXCore::ARMEmitter::Size::i64Bit, rd, rn);
+  }
+  void rev(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    uint32_t Op = (0b101'1010'110U << 21) |
+                        (0b0'0000U << 16) |
+                        (0b0000'10U << 10) |
+                        (s == ARMEmitter::Size::i64Bit ? (1U << 10) : 0);
+    DataProcessing_1Source(Op, s, rd, rn);
+  }
+
+
+  // TODO: PAUTH
+
+  // Logical - shifted register
+  void mov(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn) {
+    orr(s, rd, FEXCore::ARMEmitter::Reg::zr, rn, ARMEmitter::ShiftType::LSL, 0);
+  }
+  void mov(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn) {
+    orr(FEXCore::ARMEmitter::Size::i64Bit, rd.R(), FEXCore::ARMEmitter::Reg::zr, rn.R(), ARMEmitter::ShiftType::LSL, 0);
+  }
+  void mov(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn) {
+    orr(FEXCore::ARMEmitter::Size::i32Bit, rd.R(), FEXCore::ARMEmitter::Reg::zr, rn.R(), ARMEmitter::ShiftType::LSL, 0);
+  }
+
+  void mvn(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    orn(s, rd, FEXCore::ARMEmitter::Reg::zr, rn, Shift, amt);
+  }
+
+  void and_(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    constexpr uint32_t Op = 0b000'1010'000U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void ands(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    constexpr uint32_t Op = 0b110'1010'000U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void bic(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    constexpr uint32_t Op = 0b000'1010'001U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void bics(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    constexpr uint32_t Op = 0b110'1010'001U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void orr(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    constexpr uint32_t Op = 0b010'1010'000U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+
+  void orn(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    constexpr uint32_t Op = 0b010'1010'001U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void eor(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    constexpr uint32_t Op = 0b100'1010'000U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void eon(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    constexpr uint32_t Op = 0b100'1010'001U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+
+  // AddSub - shifted register
+  void add(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    add(ARMEmitter::Size::i64Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
+  }
+  void adds(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    adds(ARMEmitter::Size::i64Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
+  }
+  void sub(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    sub(ARMEmitter::Size::i64Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
+  }
+  void neg(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    sub(rd, FEXCore::ARMEmitter::XReg::zr, rm, Shift, amt);
+  }
+  void cmp(FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    subs(ARMEmitter::Size::i64Bit, FEXCore::ARMEmitter::Reg::rsp, rn.R(), rm.R(), Shift, amt);
+  }
+  void subs(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    subs(ARMEmitter::Size::i64Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
+  }
+  void negs(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    subs(rd, FEXCore::ARMEmitter::XReg::zr, rm, Shift, amt);
+  }
+
+  void add(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    add(ARMEmitter::Size::i32Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
+  }
+  void adds(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    adds(ARMEmitter::Size::i32Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
+  }
+  void sub(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    sub(ARMEmitter::Size::i32Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
+  }
+  void neg(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    sub(rd, FEXCore::ARMEmitter::WReg::zr, rm, Shift, amt);
+  }
+  void cmp(FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    subs(ARMEmitter::Size::i32Bit, FEXCore::ARMEmitter::Reg::rsp, rn.R(), rm.R(), Shift, amt);
+  }
+  void subs(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    subs(ARMEmitter::Size::i32Bit, rd.R(), rn.R(), rm.R(), Shift, amt);
+  }
+  void negs(FEXCore::ARMEmitter::WRegister rd, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    subs(rd, FEXCore::ARMEmitter::WReg::zr, rm, Shift, amt);
+  }
+
+  void add(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    LOGMAN_THROW_AA_FMT(Shift != FEXCore::ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
+    constexpr uint32_t Op = 0b000'1011'000U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void adds(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    LOGMAN_THROW_AA_FMT(Shift != FEXCore::ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
+    constexpr uint32_t Op = 0b010'1011'000U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void sub(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    LOGMAN_THROW_AA_FMT(Shift != FEXCore::ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
+    constexpr uint32_t Op = 0b100'1011'000U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void neg(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    sub(s, rd, FEXCore::ARMEmitter::Reg::zr, rm, Shift, amt);
+  }
+  void cmp(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    subs(s, FEXCore::ARMEmitter::Reg::zr, rn, rm, Shift, amt);
+  }
+
+  void subs(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    LOGMAN_THROW_AA_FMT(Shift != FEXCore::ARMEmitter::ShiftType::ROR, "Doesn't support ROR");
+    constexpr uint32_t Op = 0b110'1011'000U << 21;
+    DataProcessing_Shifted_Reg(Op, s, rd, rn, rm, Shift, amt);
+  }
+  void negs(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift = FEXCore::ARMEmitter::ShiftType::LSL, uint32_t amt = 0) {
+    subs(s, rd, FEXCore::ARMEmitter::Reg::zr, rm, Shift, amt);
+  }
+
+  // AddSub - extended register
+  void add(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
+    LOGMAN_THROW_AA_FMT(Shift <= 4, "Shift amount is too large");
+    constexpr uint32_t Op = 0b000'1011'001U << 21;
+    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, Option, Shift);
+  }
+  void adds(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
+    constexpr uint32_t Op = 0b010'1011'001U << 21;
+    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, Option, Shift);
+  }
+  void sub(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
+    constexpr uint32_t Op = 0b100'1011'001U << 21;
+    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, Option, Shift);
+  }
+  void subs(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
+    constexpr uint32_t Op = 0b110'1011'001U << 21;
+    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, Option, Shift);
+  }
+  void cmp(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ExtendedType Option, uint32_t Shift = 0) {
+    constexpr uint32_t Op = 0b110'1011'001U << 21;
+    DataProcessing_Extended_Reg(Op, s, FEXCore::ARMEmitter::Reg::zr, rn, rm, Option, Shift);
+  }
+
+  // AddSub - with carry
+  void adc(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = 0b0001'1010'000U << 21;
+    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, FEXCore::ARMEmitter::ExtendedType::UXTB, 0);
+  }
+  void adcs(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = 0b0011'1010'000U << 21;
+    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, FEXCore::ARMEmitter::ExtendedType::UXTB, 0);
+  }
+  void sbc(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = 0b0101'1010'000U << 21;
+    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, FEXCore::ARMEmitter::ExtendedType::UXTB, 0);
+  }
+  void sbcs(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    constexpr uint32_t Op = 0b0111'1010'000U << 21;
+    DataProcessing_Extended_Reg(Op, s, rd, rn, rm, FEXCore::ARMEmitter::ExtendedType::UXTB, 0);
+  }
+  // Rotate right into flags
+  // TODO
+  // Evaluate into flags
+  // TODO
+  // Conditional compare - register
+  void ccmn(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::StatusFlags flags, FEXCore::ARMEmitter::Condition Cond) {
+    constexpr uint32_t Op = 0b0011'1010'010 << 21;
+    ConditionalCompare(Op, 0, 0b00, 0, s, rn, rm, flags, Cond);
+  }
+  void ccmp(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::StatusFlags flags, FEXCore::ARMEmitter::Condition Cond) {
+    constexpr uint32_t Op = 0b0011'1010'010 << 21;
+    ConditionalCompare(Op, 1, 0b00, 0, s, rn, rm, flags, Cond);
+  }
+
+  // Conditional compare - immediate
+  void ccmn(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rn, uint32_t rm, FEXCore::ARMEmitter::StatusFlags flags, FEXCore::ARMEmitter::Condition Cond) {
+    LOGMAN_THROW_A_FMT((rm & ~0b1'1111) == 0, "Comparison imm too large");
+    constexpr uint32_t Op = 0b0011'1010'010 << 21;
+    ConditionalCompare(Op, 0, 0b10, 0, s, rn, rm, flags, Cond);
+  }
+  void ccmp(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rn, uint32_t rm, FEXCore::ARMEmitter::StatusFlags flags, FEXCore::ARMEmitter::Condition Cond) {
+    LOGMAN_THROW_A_FMT((rm & ~0b1'1111) == 0, "Comparison imm too large");
+    constexpr uint32_t Op = 0b0011'1010'010 << 21;
+    ConditionalCompare(Op, 1, 0b10, 0, s, rn, rm, flags, Cond);
+  }
+
+  // Conditional select
+  void csel(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::Condition Cond) {
+    constexpr uint32_t Op = 0b0001'1010'100 << 21;
+    ConditionalCompare(Op, 0, 0b00, s, rd, rn, rm, Cond);
+  }
+  void cset(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Condition Cond) {
+    constexpr uint32_t Op = 0b0001'1010'100 << 21;
+    ConditionalCompare(Op, 0, 0b01, s, rd, FEXCore::ARMEmitter::Reg::zr, FEXCore::ARMEmitter::Reg::zr, static_cast<FEXCore::ARMEmitter::Condition>(FEXCore::ToUnderlying(Cond) ^ FEXCore::ToUnderlying(FEXCore::ARMEmitter::Condition::CC_NE)));
+  }
+  void csinc(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::Condition Cond) {
+    constexpr uint32_t Op = 0b0001'1010'100 << 21;
+    ConditionalCompare(Op, 0, 0b01, s, rd, rn, rm, Cond);
+  }
+  void csinv(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::Condition Cond) {
+    constexpr uint32_t Op = 0b0001'1010'100 << 21;
+    ConditionalCompare(Op, 1, 0b00, s, rd, rn, rm, Cond);
+  }
+  void csneg(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::Condition Cond) {
+    constexpr uint32_t Op = 0b0001'1010'100 << 21;
+    ConditionalCompare(Op, 1, 0b01, s, rd, rn, rm, Cond);
+  }
+
+  // Data processing - 3 source
+  void madd(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::Register ra) {
+    constexpr uint32_t Op = 0b001'1011'000U << 21;
+    DataProcessing_3Source(Op, 0, s, rd, rn, rm, ra);
+  }
+  void mul(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    madd(s, rd, rn, rm, FEXCore::ARMEmitter::Reg::zr);
+  }
+  void msub(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::Register ra) {
+    constexpr uint32_t Op = 0b001'1011'000U << 21;
+    DataProcessing_3Source(Op, 1, s, rd, rn, rm, ra);
+  }
+  void mneg(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    msub(s, rd, rn, rm, FEXCore::ARMEmitter::Reg::zr);
+  }
+  void smaddl(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::XRegister ra) {
+    constexpr uint32_t Op = 0b001'1011'001U << 21;
+    DataProcessing_3Source(Op, 0, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm, ra);
+  }
+  void smull(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    smaddl(rd, rn, rm, FEXCore::ARMEmitter::Reg::zr);
+  }
+  void smsubl(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::XRegister ra) {
+    constexpr uint32_t Op = 0b001'1011'001U << 21;
+    DataProcessing_3Source(Op, 1, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm, ra);
+  }
+  void smnegl(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    smsubl(rd, rn, rm, FEXCore::ARMEmitter::Reg::zr);
+  }
+  void smulh(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm) {
+    constexpr uint32_t Op = 0b001'1011'010U << 21;
+    DataProcessing_3Source(Op, 0, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm, FEXCore::ARMEmitter::Reg::zr);
+  }
+  void umaddl(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::XRegister ra) {
+    constexpr uint32_t Op = 0b001'1011'101U << 21;
+    DataProcessing_3Source(Op, 0, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm, ra);
+  }
+  void umull(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    umaddl(rd, rn, rm, FEXCore::ARMEmitter::Reg::zr);
+  }
+  void umsubl(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm, FEXCore::ARMEmitter::XRegister ra) {
+    constexpr uint32_t Op = 0b001'1011'101U << 21;
+    DataProcessing_3Source(Op, 1, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm, ra);
+  }
+  void umnegl(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::WRegister rn, FEXCore::ARMEmitter::WRegister rm) {
+    umsubl(rd, rn, rm, FEXCore::ARMEmitter::Reg::zr);
+  }
+  void umulh(FEXCore::ARMEmitter::XRegister rd, FEXCore::ARMEmitter::XRegister rn, FEXCore::ARMEmitter::XRegister rm) {
+    constexpr uint32_t Op = 0b001'1011'110U << 21;
+    DataProcessing_3Source(Op, 0, FEXCore::ARMEmitter::Size::i64Bit, rd, rn, rm, FEXCore::ARMEmitter::Reg::zr);
+  }
+
+private:
+  void and_(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t n, uint32_t immr, uint32_t imms) {
+    constexpr uint32_t Op = 0b001'0010'00 << 22;
+    DataProcessing_Logical_Imm(Op, s, rd, rn, n, immr, imms);
+  }
+  void ands(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t n, uint32_t immr, uint32_t imms) {
+    constexpr uint32_t Op = 0b111'0010'00 << 22;
+    DataProcessing_Logical_Imm(Op, s, rd, rn, n, immr, imms);
+  }
+  void orr(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t n, uint32_t immr, uint32_t imms) {
+    constexpr uint32_t Op = 0b011'0010'00 << 22;
+    DataProcessing_Logical_Imm(Op, s, rd, rn, n, immr, imms);
+  }
+  void eor(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t n, uint32_t immr, uint32_t imms) {
+    constexpr uint32_t Op = 0b101'0010'00 << 22;
+    DataProcessing_Logical_Imm(Op, s, rd, rn, n, immr, imms);
+  }
+
+  void sbfm(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t immr, uint32_t imms) {
+    constexpr uint32_t Op = 0b0001'0011'00 << 22;
+    DataProcessing_Logical_Imm(Op, s, rd, rn, s == ARMEmitter::Size::i64Bit, immr, imms);
+  }
+  void bfm(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t immr, uint32_t imms) {
+    constexpr uint32_t Op = 0b0011'0011'00 << 22;
+    DataProcessing_Logical_Imm(Op, s, rd, rn, s == ARMEmitter::Size::i64Bit, immr, imms);
+  }
+  // 4.1.64 - Data processing - Immediate
+  void DataProcessing_PCRel_Imm(uint32_t Op, FEXCore::ARMEmitter::Register rd, uint32_t Imm) {
+    // Ensure the immediate is masked.
+    Imm &= 0b1'1111'1111'1111'1111'1111U;
+
+    uint32_t Instr = Op;
+
+    Instr |= (Imm & 0b11) << 29;
+    Instr |= (Imm >> 2) << 5;
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+  void DataProcessing_AddSub_Imm(uint32_t Op, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t Imm, bool LSL12) {
+    bool TooLarge = (Imm & ~0b1111'1111'1111U) != 0;
+    if (TooLarge && !LSL12 && ((Imm >> 12) & ~0b1111'1111'1111U) == 0) {
+      // We can convert an immediate
+      TooLarge = false;
+      LSL12 = true;
+      Imm >>= 12;
+    }
+    LOGMAN_THROW_AA_FMT(TooLarge == false, "Imm amount too large: 0x{:x}", Imm);
+
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= LSL12 << 22;
+    Instr |= Imm << 10;
+    Instr |= Encode_rn(rn);
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+  // Move Wide
+  void DataProcessing_MoveWide(uint32_t Op, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, uint32_t Imm, uint32_t Offset) {
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= Imm << 5;
+    Instr |= Offset << 21;
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+  // Logical immediate
+  void DataProcessing_Logical_Imm(uint32_t Op, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, uint32_t n, uint32_t immr, uint32_t imms) {
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= n << 22;
+    Instr |= immr << 16;
+    Instr |= imms << 10;
+    Instr |= Encode_rn(rn);
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+  void DataProcessing_Extract(uint32_t Op, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, uint32_t Imm) {
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    // Current ARMv8 spec hardcodes SF == N for this class of instructions.
+    // Anythign else is undefined behaviour.
+    const uint32_t N  = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 22) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= N;
+    Instr |= Encode_rm(rm);
+    Instr |= Imm << 10;
+    Instr |= Encode_rn(rn);
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+  // Data-processing - 2 source
+  void DataProcessing_2Source(uint32_t Op, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm) {
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= Encode_rm(rm);
+    Instr |= Encode_rn(rn);
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+  // Data processing - 1 source
+  template<typename T>
+  void DataProcessing_1Source(uint32_t Op, FEXCore::ARMEmitter::Size s, T rd, T rn) {
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= Encode_rn(rn);
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+  // AddSub - shifted register
+  void DataProcessing_Shifted_Reg(uint32_t Op, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ShiftType Shift, uint32_t amt) {
+    LOGMAN_THROW_AA_FMT((amt & ~0b11'1111U) == 0, "Shift amount too large");
+
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= FEXCore::ToUnderlying(Shift) << 22;
+    Instr |= Encode_rm(rm);
+    Instr |= static_cast<uint32_t>(amt) << 10;
+    Instr |= Encode_rn(rn);
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+  // AddSub - extended register
+  void DataProcessing_Extended_Reg(uint32_t Op, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::ExtendedType Option, uint32_t Shift) {
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= Encode_rm(rm);
+    Instr |= FEXCore::ToUnderlying(Option) << 13;
+    Instr |= static_cast<uint32_t>(Shift) << 10;
+    Instr |= Encode_rn(rn);
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+  // Conditional compare - register
+  template<typename T>
+  void ConditionalCompare(uint32_t Op, uint32_t o1, uint32_t o2, uint32_t o3, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rn, T rm, FEXCore::ARMEmitter::StatusFlags flags, FEXCore::ARMEmitter::Condition Cond) {
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= o1 << 30;
+    Instr |= Encode_rm(rm);
+    Instr |= FEXCore::ToUnderlying(Cond) << 12;
+    Instr |= o2 << 10;
+    Instr |= Encode_rn(rn);
+    Instr |= o3 << 4;
+    Instr |= FEXCore::ToUnderlying(flags);
+
+    dc32(Instr);
+  }
+
+  template<typename T>
+  void ConditionalCompare(uint32_t Op, uint32_t o1, uint32_t o2, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, T rm, FEXCore::ARMEmitter::Condition Cond) {
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= o1 << 30;
+    Instr |= Encode_rm(rm);
+    Instr |= FEXCore::ToUnderlying(Cond) << 12;
+    Instr |= o2 << 10;
+    Instr |= Encode_rn(rn);
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+  // Data-processing - 3 source
+  void DataProcessing_3Source(uint32_t Op, uint32_t Op0, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::Register rn, FEXCore::ARMEmitter::Register rm, FEXCore::ARMEmitter::Register ra) {
+    const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+    uint32_t Instr = Op;
+
+    Instr |= SF;
+    Instr |= Encode_rm(rm);
+    Instr |= Op0 << 15;
+    Instr |= Encode_ra(ra);
+    Instr |= Encode_rn(rn);
+    Instr |= Encode_rd(rd);
+
+    dc32(Instr);
+  }
+
+

External/FEXCore/Source/Interface/Core/ArchHelpers/CodeEmitter/BranchOps.inl

@@ -0,0 +1,322 @@
+/* Branch instruction emitters.
+ *
+ * Most of these instructions will use `BackwardLabel`, `ForwardLabel`, or `BiDirectionLabel` to determine where a branch targets.
+ */
+public:
+  // Branches, Exception Generating and System instructions
+  public:
+    // Conditional branch immediate
+    ///< Branch conditional
+    void b(FEXCore::ARMEmitter::Condition Cond, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 0, Cond, Imm);
+    }
+    void b(FEXCore::ARMEmitter::Condition Cond, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 0, Cond, Imm >> 2);
+    }
+    void b(FEXCore::ARMEmitter::Condition Cond, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::BC });
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 0, Cond, 0);
+    }
+
+    void b(FEXCore::ARMEmitter::Condition Cond, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        b(Cond, &Label->Backward);
+      }
+      else {
+        b(Cond, &Label->Forward);
+      }
+    }
+
+    ///< Branch consistent conditional
+    void bc(FEXCore::ARMEmitter::Condition Cond, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 1, Cond, Imm);
+    }
+    void bc(FEXCore::ARMEmitter::Condition Cond, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 1, Cond, Imm >> 2);
+    }
+
+    void bc(FEXCore::ARMEmitter::Condition Cond, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::BC });
+      constexpr uint32_t Op = 0b0101'010 << 25;
+      Branch_Conditional(Op, 0, 1, Cond, 0);
+    }
+
+    void bc(FEXCore::ARMEmitter::Condition Cond, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        bc(Cond, &Label->Backward);
+      }
+      else {
+        bc(Cond, &Label->Forward);
+      }
+    }
+
+    // Unconditional branch register
+    void br(FEXCore::ARMEmitter::Register rn) {
+      constexpr uint32_t Op = 0b1101011 << 25 |
+                              0b0'000 << 21 |   // opc
+                              0b1'1111 << 16 |  // op2
+                              0b0000'00 << 10 | // op3
+                              0b0'0000;         // op4
+
+      UnconditionalBranch(Op, rn);
+    }
+    void blr(FEXCore::ARMEmitter::Register rn) {
+      constexpr uint32_t Op = 0b1101011 << 25 |
+                              0b0'001 << 21 |   // opc
+                              0b1'1111 << 16 |  // op2
+                              0b0000'00 << 10 | // op3
+                              0b0'0000;         // op4
+
+      UnconditionalBranch(Op, rn);
+    }
+    void ret(FEXCore::ARMEmitter::Register rn = FEXCore::ARMEmitter::Reg::r30) {
+      constexpr uint32_t Op = 0b1101011 << 25 |
+                              0b0'010 << 21 |   // opc
+                              0b1'1111 << 16 |  // op2
+                              0b0000'00 << 10 | // op3
+                              0b0'0000;         // op4
+
+      UnconditionalBranch(Op, rn);
+    }
+
+    // Unconditional branch immediate
+    void b(uint32_t Imm) {
+      constexpr uint32_t Op = 0b0001'01 << 26;
+
+      UnconditionalBranch(Op, Imm);
+    }
+    void b(BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      constexpr uint32_t Op = 0b0001'01 << 26;
+
+      UnconditionalBranch(Op, Imm >> 2);
+    }
+    void b(ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::B });
+      constexpr uint32_t Op = 0b0001'01 << 26;
+
+      UnconditionalBranch(Op, 0);
+    }
+
+    void b(BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        b(&Label->Backward);
+      }
+      else {
+        b(&Label->Forward);
+      }
+    }
+
+    void bl(uint32_t Imm) {
+      constexpr uint32_t Op = 0b1001'01 << 26;
+
+      UnconditionalBranch(Op, Imm);
+    }
+
+    void bl(BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+      constexpr uint32_t Op = 0b1001'01 << 26;
+
+      UnconditionalBranch(Op, Imm >> 2);
+    }
+    void bl(ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::B });
+      constexpr uint32_t Op = 0b1001'01 << 26;
+
+      UnconditionalBranch(Op, 0);
+    }
+
+    void bl(BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        bl(&Label->Backward);
+      }
+      else {
+        bl(&Label->Forward);
+      }
+    }
+
+    // Compare and branch
+    void cbz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0011'0100 << 24;
+
+      CompareAndBranch(Op, s, rt, Imm);
+    }
+
+    void cbz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+
+      constexpr uint32_t Op = 0b0011'0100 << 24;
+
+      CompareAndBranch(Op, s, rt, Imm >> 2);
+    }
+
+    void cbz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::BC });
+
+      constexpr uint32_t Op = 0b0011'0100 << 24;
+
+      CompareAndBranch(Op, s, rt, 0);
+    }
+
+    void cbz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        cbz(s, rt, &Label->Backward);
+      }
+      else {
+        cbz(s, rt, &Label->Forward);
+      }
+    }
+
+    void cbnz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0011'0101 << 24;
+
+      CompareAndBranch(Op, s, rt, Imm);
+    }
+
+    void cbnz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+
+      constexpr uint32_t Op = 0b0011'0101 << 24;
+
+      CompareAndBranch(Op, s, rt, Imm >> 2);
+    }
+
+    void cbnz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::BC });
+
+      constexpr uint32_t Op = 0b0011'0101 << 24;
+
+      CompareAndBranch(Op, s, rt, 0);
+    }
+
+    void cbnz(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        cbnz(s, rt, &Label->Backward);
+      }
+      else {
+        cbnz(s, rt, &Label->Forward);
+      }
+    }
+
+    // Test and branch immediate
+    void tbz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0011'0110 << 24;
+
+      TestAndBranch(Op, rt, Bit, Imm);
+    }
+    void tbz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+
+      constexpr uint32_t Op = 0b0011'0110 << 24;
+
+      TestAndBranch(Op, rt, Bit, Imm >> 2);
+    }
+    void tbz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::TEST_BRANCH });
+
+      constexpr uint32_t Op = 0b0011'0110 << 24;
+
+      TestAndBranch(Op, rt, Bit, 0);
+    }
+
+    void tbz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        tbz(rt, Bit, &Label->Backward);
+      }
+      else {
+        tbz(rt, Bit, &Label->Forward);
+      }
+    }
+
+    void tbnz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
+      constexpr uint32_t Op = 0b0011'0111 << 24;
+
+      TestAndBranch(Op, rt, Bit, Imm);
+    }
+    void tbnz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, BackwardLabel const* Label) {
+      int32_t Imm = static_cast<int32_t>(Label->Location - GetCursorAddress<uint8_t*>());
+      LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+
+      constexpr uint32_t Op = 0b0011'0111 << 24;
+
+      TestAndBranch(Op, rt, Bit, Imm >> 2);
+    }
+    void tbnz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, ForwardLabel *Label) {
+      Label->Insts.emplace_back(ForwardLabel::Instructions{ .Location = GetCursorAddress<uint8_t*>(), .Type = ForwardLabel::Instructions::InstType::TEST_BRANCH });
+      constexpr uint32_t Op = 0b0011'0111 << 24;
+
+      TestAndBranch(Op, rt, Bit, 0);
+    }
+
+    void tbnz(FEXCore::ARMEmitter::Register rt, uint32_t Bit, BiDirectionalLabel *Label) {
+      if (Label->Backward.Location) {
+        tbnz(rt, Bit, &Label->Backward);
+      }
+      else {
+        tbnz(rt, Bit, &Label->Forward);
+      }
+    }
+
+private:
+    // Conditional branch immediate
+    void Branch_Conditional(uint32_t Op, uint32_t Op1, uint32_t Op0, FEXCore::ARMEmitter::Condition Cond, uint32_t Imm) {
+      uint32_t Instr = Op;
+
+      Instr |= Op1 << 24;
+      Instr |= (Imm & 0x7'FFFF) << 5;
+      Instr |= Op0 << 4;
+      Instr |= FEXCore::ToUnderlying(Cond);
+
+      dc32(Instr);
+    }
+
+    // Unconditional branch register
+    void UnconditionalBranch(uint32_t Op, FEXCore::ARMEmitter::Register rn) {
+      uint32_t Instr = Op;
+      Instr |= Encode_rn(rn);
+      dc32(Instr);
+    }
+
+    // Unconditional branch - immediate
+    void UnconditionalBranch(uint32_t Op, uint32_t Imm) {
+      uint32_t Instr = Op;
+      Instr |= Imm & 0x3FF'FFFF;
+      dc32(Instr);
+    }
+
+    // Compare and branch
+    void CompareAndBranch(uint32_t Op, FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register rt, uint32_t Imm) {
+      const uint32_t SF = s == FEXCore::ARMEmitter::Size::i64Bit ? (1U << 31) : 0;
+
+      uint32_t Instr = Op;
+
+      Instr |= SF;
+      Instr |= (Imm & 0x7'FFFF) << 5;
+      Instr |= Encode_rt(rt);
+      dc32(Instr);
+    }
+
+    // Test and branch - immediate
+    void TestAndBranch(uint32_t Op, FEXCore::ARMEmitter::Register rt, uint32_t Bit, uint32_t Imm) {
+      uint32_t Instr = Op;
+
+      Instr |= (Bit >> 5) << 31;
+      Instr |= (Bit & 0b1'1111) << 19;
+      Instr |= (Imm & 0x3FFF) << 5;
+      Instr |= Encode_rt(rt);
+      dc32(Instr);
+    }

External/FEXCore/Source/Interface/Core/ArchHelpers/CodeEmitter/Buffer.h

@@ -0,0 +1,100 @@
+#pragma once
+#include <cstddef>
+#include <cstdint>
+#include <cstring>
+
+namespace FEXCore::ARMEmitter {
+  class Buffer {
+    public:
+      Buffer() {
+        SetBuffer(nullptr, 0);
+      }
+
+      Buffer(uint8_t* Base, uint64_t BaseSize) {
+        SetBuffer(Base, BaseSize);
+      }
+
+      void SetBuffer(uint8_t* Base, uint64_t BaseSize) {
+        BufferBase = Base;
+        CurrentOffset = BufferBase;
+        Size = BaseSize;
+      }
+
+      void dc8(uint8_t Data) {
+        decltype(Data) *Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
+        *Memory = Data;
+        CurrentOffset += sizeof(Data);
+      }
+
+      void dc16(uint16_t Data) {
+        decltype(Data) *Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
+        *Memory = Data;
+        CurrentOffset += sizeof(Data);
+      }
+
+      void dc32(uint32_t Data) {
+        decltype(Data) *Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
+        *Memory = Data;
+        CurrentOffset += sizeof(Data);
+      }
+
+      void dc64(uint64_t Data) {
+        decltype(Data) *Memory = reinterpret_cast<decltype(Data)*>(CurrentOffset);
+        *Memory = Data;
+        CurrentOffset += sizeof(Data);
+      }
+      void EmitString(const char *String) {
+        const auto StringLength = strlen(String);
+        memcpy(CurrentOffset, String, StringLength);
+        CurrentOffset += StringLength;
+      }
+
+      void Align() {
+        // Align the buffer to instruction size
+        auto CurrentAlignment = reinterpret_cast<uint64_t>(CurrentOffset) & 0b11;
+        if (!CurrentAlignment) {
+          return;
+        }
+        CurrentOffset += 4 - CurrentAlignment;
+      }
+
+      template<typename T>
+      T GetCursorAddress() const {
+        return reinterpret_cast<T>(CurrentOffset);
+      }
+
+      static void ClearICache(void* Begin, std::size_t Length) {
+        __builtin___clear_cache(static_cast<char*>(Begin), static_cast<char*>(Begin) + Length);
+      }
+
+      size_t GetCursorOffset() const {
+        return static_cast<size_t>(CurrentOffset - BufferBase);
+      }
+
+      uint8_t *GetBufferBase() const {
+        return BufferBase;
+      }
+
+      void CursorIncrement(size_t Size) {
+        CurrentOffset += Size;
+      }
+
+      void SetCursorOffset(size_t Offset) {
+        CurrentOffset = BufferBase + Offset;
+      }
+
+      uint64_t GetBufferSize() const {
+        return Size;
+      }
+
+    protected:
+
+      void ResetBuffer() {
+        CurrentOffset = BufferBase;
+      }
+
+      uint8_t* BufferBase;
+      uint8_t* CurrentOffset;
+      uint64_t Size;
+  };
+}

External/FEXCore/Source/Interface/Core/ArchHelpers/CodeEmitter/Emitter.h

@@ -0,0 +1,721 @@
+#pragma once
+
+#include "Interface/Core/ArchHelpers/CodeEmitter/Buffer.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Registers.h"
+
+#include <FEXCore/Utils/CompilerDefs.h>
+#include <FEXCore/Utils/EnumUtils.h>
+#include <FEXCore/Utils/LogManager.h>
+
+#include <aarch64/assembler-aarch64.h>
+
+#include <cstdint>
+#include <utility>
+#include <type_traits>
+#include <vector>
+
+/*
+ * Welcome to FEX-Emu's custom AArch64 emitter.
+ * This was written specifically to avoid the performance cost of the vixl emitter.
+ *
+ * There are some specific design constraints in this design to target a couple features:
+ *   - High performance
+ *   - Low CPU cache performance hit
+ *   - Significantly reduced code footprint
+ *   - Low number of branches
+ *
+ * These requirements are mostly achieved by removing a bunch of developer conveniences
+ * that vixl provides. The developer needs to take a lot of care to not shoot themselves in the foot.
+ *
+ * Misc design decisions:
+ * - Registers are encoded as basic uint32_t enums.
+ *   - Converting between different registers is zero-cost.
+ *   - Passing around as arguments are as cheap as registers
+ *     - Contrast to vixl where every register requires living on the stack.
+ *   - Registers can get encoded in to instructions with a simple `BFM` instruction.
+ *
+ * - Instructions are very simply emitted, allowing direct inlining most of the time.
+ *   - These are simple enough that multiple back-to-back instructions get optimized to 128-bit load-store operations.
+ *     - Contrast to vixl where pretty much no instruction emitter gets inlined.
+ *
+ * - Instruction emitters are /mostly/ unsized. Most instructions take a size argument first, which gets encoded
+ *   directly in to the instruction.
+ *   - Contrast to vixl where the register arguments are how the instructions determine operating size.
+ *   - Size argument allows FEX to use `CSEL` to select a size at runtime, instead of branching.
+ *   - Some instructions are explicitly sized based on register type. Read comments in the respective `inl` files to
+ *     see why.
+ *     Some scalar/vector operations are an example of this.
+ *
+ * - Almost zero helper functions.
+ *   - Primary exception to this rule is load-store operations. These will use a helper to make
+ *     it easier to select the correct load-store instruction. Mostly because these are a nightmare selecting
+ *     the right instruction.
+ */
+namespace FEXCore::ARMEmitter {
+  /*
+   * This `Size` enum is used for most ALU operations.
+   * These follow the AArch64 encoding style in most cases.
+   */
+  enum class Size : uint32_t {
+    i32Bit = 0,
+    i64Bit,
+  };
+
+  // This allows us to get the `Size` enum in bits.
+  template<Size size>
+  constexpr size_t RegSizeInBits() {
+    constexpr size_t RegSize[] = {
+      32, 64, 128,
+    };
+    return RegSize[FEXCore::ToUnderlying(size)];
+  }
+
+  [[maybe_unused]]
+  static inline size_t RegSizeInBits(Size size) {
+    constexpr size_t RegSize[] = {
+      32, 64, 128,
+    };
+    return RegSize[FEXCore::ToUnderlying(size)];
+  }
+
+  /* This `SubRegSize` enum is used for most ASIMD operations.
+   * These follow the AArch64 encoding style in most cases.
+   */
+  enum class SubRegSize : uint32_t {
+    i8Bit   = 0b00,
+    i16Bit  = 0b01,
+    i32Bit  = 0b10,
+    i64Bit  = 0b11,
+    i128Bit = 0b100,
+  };
+
+  // This allows us to get the `SubRegSize` in bits.
+  template<SubRegSize size>
+  constexpr size_t SubRegSizeInBits() {
+    return (1 << FEXCore::ToUnderlying(size)) * 8;
+  }
+
+  [[maybe_unused]]
+  static inline size_t SubRegSizeInBits(SubRegSize size) {
+    return (1 << FEXCore::ToUnderlying(size)) * 8;
+  }
+
+  /* This `ScalarRegSize` enum is used for most scalar float
+   * operations.
+   *
+   * This is specifically duplicated from `SubRegSize` to have strongly
+   * typed functions.
+   *
+   * `ScalarRegSize` specifically doesn't have `i128Bit` because scalar operations
+   * can't operate at 128-bit.
+   */
+  enum class ScalarRegSize : uint32_t {
+    i8Bit   = 0b00,
+    i16Bit  = 0b01,
+    i32Bit  = 0b10,
+    i64Bit  = 0b11,
+  };
+
+  // This allows us to get the `ScalarRegSize` in bits.
+  template<ScalarRegSize size>
+  constexpr size_t ScalarRegSizeInBits() {
+    return (1 << FEXCore::ToUnderlying(size)) * 8;
+  }
+
+  [[maybe_unused]]
+  static inline size_t ScalarRegSizeInBits(ScalarRegSize size) {
+    return (1 << FEXCore::ToUnderlying(size)) * 8;
+  }
+
+  /* This `VectorRegSizePair` union allows us to have an overlapping type
+   * to select a scalar operation or a vector depending on which operation
+   * we pass in.
+   * Useful in FEX's vector operations that behave as scalar or vector
+   * depending on various factors. But since the operation will have the sa,e
+   * element size, we want to choose the operation more easily
+   */
+  union VectorRegSizePair {
+    ScalarRegSize Scalar;
+    SubRegSize Vector;
+  };
+
+  // This allows us to create a `VectorRegSizePair` union.
+  [[maybe_unused]]
+  static inline VectorRegSizePair ToVectorSizePair(SubRegSize size) {
+    return VectorRegSizePair {.Vector = size};
+  }
+  [[maybe_unused]]
+  static inline VectorRegSizePair ToVectorSizePair(ScalarRegSize size) {
+    return VectorRegSizePair {.Scalar = size};
+  }
+
+  // This `ShiftType` enum is used for ALU shift-register encoded instructions.
+  enum class ShiftType : uint32_t {
+    LSL = 0,
+    LSR,
+    ASR,
+    ROR,
+  };
+
+  // This `ExtendedType` enum is used for ALU extended-register encoded instructions.
+  enum class ExtendedType : uint32_t {
+    UXTB = 0b000,
+    UXTH = 0b001,
+    UXTW = 0b010,
+    UXTX = 0b011,
+    SXTB = 0b100,
+    SXTH = 0b101,
+    SXTW = 0b110,
+    SXTX = 0b111,
+    LSL_32 = UXTW,
+    LSL_64 = UXTX,
+  };
+
+  // This `Condition` enum is used for various conditional instructions.
+  enum class Condition : uint32_t {
+    // Meaning:   Int                    - Float
+    CC_EQ = 0, // Equal                  - Equal
+    CC_NE,     // Not Eq                 - Not Eq or unordered
+    CC_CS,     // Carry set              - Greater than, equal, or unordered
+    CC_CC,     // Carry clear            - Less than
+    CC_MI,     // Minus/Negative         - Less than
+    CC_PL,     // Plus, positive or zero - GT, equal, or unordered
+    CC_VS,     // Overflow               - Unordered
+    CC_VC,     // No Overflow            - Ordered
+    CC_HI,     // Unsigned higher        - GT, or unordered
+    CC_LS,     // Unsigned lower or same - LT or EQ
+    CC_GE,     // Signed GT or EQ        - GT or EQ
+    CC_LT,     // Signed LT              - LT or Unordered
+    CC_GT,     // Signed GT              - GT
+    CC_LE,     // Signed LT or EQ        - LT, EQ, or Unordered
+    CC_AL,     // Always                 - Always
+    CC_NV,     // Always                 - Always
+
+    // Aliases
+    CC_HS = CC_CS,
+    CC_LO = CC_CC,
+  };
+
+  /*
+   * This `StatusFlags` enum is used for conditional compare encoded instructions.
+   * These directly encode to the `nzcv` flags.
+   */
+  enum class StatusFlags : uint32_t {
+    None = 0,
+    Flag_V = 0b0001,
+    Flag_C = 0b0010,
+    Flag_Z = 0b0100,
+    Flag_N = 0b1000,
+
+    Flag_NZCV = Flag_N | Flag_Z | Flag_C | Flag_V,
+  };
+
+
+  /*
+   * This `IndexType` enum is used for load-store instructions.
+   * Not all load-store instructions use this, so the user needs to be careful.
+   */
+  enum class IndexType {
+    POST,
+    OFFSET,
+    PRE,
+
+    UNPRIVILEGED,
+  };
+
+  /* This `SVEMemOperand` class is used for the helper SVE load-store instructions.
+   * Load-store instructions are quite expressive, so having a helper that handles these differences is worth it.
+   */
+  class SVEMemOperand final {
+    public:
+      SVEMemOperand(XRegister rn, XRegister rm = XReg::zr)
+        : rn {rn}
+        , MetaType {
+          .ScalarScalarType {
+            .Header = { .MemType = TYPE_SCALAR_SCALAR },
+            .rm = rm,
+          }
+        } {}
+      SVEMemOperand(XRegister rn, int32_t imm = 0)
+        : rn {rn}
+        , MetaType {
+          .ScalarImmType {
+            .Header = { .MemType = TYPE_SCALAR_IMM },
+            .Imm = imm,
+          }
+        } {}
+
+      Register rn;
+      enum Type {
+        TYPE_SCALAR_SCALAR,
+        TYPE_SCALAR_IMM,
+        TYPE_SCALAR_VECTOR,
+        TYPE_VECTOR_IMM,
+      };
+      struct HeaderStruct {
+        Type MemType;
+      };
+
+      union {
+        HeaderStruct Header;
+        struct {
+          HeaderStruct Header;
+          Register rm;
+        } ScalarScalarType;
+
+        struct {
+          HeaderStruct Header;
+          int32_t Imm;
+        } ScalarImmType;
+
+        struct {
+          HeaderStruct Header;
+          ZRegister zm;
+          // TODO: Implement support for modifier
+        } ScalarVectorType;
+
+        struct {
+          HeaderStruct Header;
+          // rn will be a ZRegister
+          int32_t Imm;
+        } VectorImmType;
+      } MetaType;
+  };
+
+  /* This `ExtendedMemOperand` class is used for the helper load-store instructions.
+   * Load-store instructions are quite expressive, so having a helper that handles these differences is worth it.
+   */
+  class ExtendedMemOperand final {
+    public:
+      ExtendedMemOperand(XRegister rn, XRegister rm = XReg::zr, ExtendedType Option = ExtendedType::LSL_64, uint32_t Shift = 0)
+        : rn {rn}
+        , MetaType {
+          .ExtendedType {
+            .Header = { .MemType = TYPE_EXTENDED },
+            .rm = rm,
+            .Option = Option,
+            .Shift = Shift,
+          }
+        } {}
+      ExtendedMemOperand(XRegister rn, IndexType Index = IndexType::OFFSET, int32_t Imm = 0)
+        : rn {rn}
+        , MetaType {
+          .ImmType {
+            .Header = { .MemType = TYPE_IMM },
+            .Index = Index,
+            .Imm = Imm,
+          }
+        } {}
+
+      Register rn;
+      enum Type {
+        TYPE_EXTENDED,
+        TYPE_IMM,
+      };
+      struct HeaderStruct {
+        Type MemType;
+      };
+      union {
+        HeaderStruct Header;
+        struct {
+          HeaderStruct Header;
+          Register rm;
+          ExtendedType Option;
+          uint32_t Shift;
+        } ExtendedType;
+        struct {
+          HeaderStruct Header;
+          IndexType Index;
+          int32_t Imm;
+        } ImmType;
+      } MetaType;
+  };
+
+  template<uint32_t op0, uint32_t op1, uint32_t CRn, uint32_t CRm, uint32_t op2>
+  constexpr uint32_t GenSystemReg() {
+    return op0 << 19 |
+      op1 << 16 |
+      CRn << 12 |
+      CRm << 8 |
+      op2 << 5;
+  };
+
+  // This `SystemRegister` enum is used for the mrs/msr instructions.
+  enum class SystemRegister : uint32_t {
+    CTR_EL0    = GenSystemReg<0b11, 0b011, 0b0000, 0b0000, 0b001>(),
+    DCZID_EL0  = GenSystemReg<0b11, 0b011, 0b0000, 0b0000, 0b111>(),
+    TPIDR_EL0  = GenSystemReg<0b11, 0b011, 0b1101, 0b0000, 0b010>(),
+    RNDR       = GenSystemReg<0b11, 0b011, 0b0010, 0b0100, 0b000>(),
+    RNDRRS     = GenSystemReg<0b11, 0b011, 0b0010, 0b0100, 0b001>(),
+    NZCV       = GenSystemReg<0b11, 0b011, 0b0100, 0b0010, 0b000>(),
+    FPCR       = GenSystemReg<0b11, 0b011, 0b0100, 0b0100, 0b000>(),
+    CNTFRQ_EL0 = GenSystemReg<0b11, 0b011, 0b1110, 0b0000, 0b000>(),
+    CNTVCT_EL0 = GenSystemReg<0b11, 0b011, 0b1110, 0b0000, 0b010>(),
+  };
+
+  template<uint32_t op1, uint32_t CRm, uint32_t op2>
+  constexpr uint32_t GenDCReg() {
+    return op1 << 16 |
+      CRm << 8 |
+      op2 << 5;
+  };
+
+  // This `DataCacheOperation` enum is used for the dc instruction.
+  enum class DataCacheOperation : uint32_t {
+    IVAC  = GenDCReg<0b000, 0b0110, 0b001>(),
+    ISW   = GenDCReg<0b000, 0b0110, 0b010>(),
+    CSW   = GenDCReg<0b000, 0b1010, 0b010>(),
+    CISW  = GenDCReg<0b000, 0b1110, 0b010>(),
+    ZVA   = GenDCReg<0b011, 0b0100, 0b001>(),
+    CVAC  = GenDCReg<0b011, 0b1010, 0b001>(),
+    CVAU  = GenDCReg<0b011, 0b1011, 0b001>(),
+    CIVAC = GenDCReg<0b011, 0b1110, 0b001>(),
+
+    // MTE2
+    IGVAC  = GenDCReg<0b000, 0b0110, 0b011>(),
+    IGSW   = GenDCReg<0b000, 0b0110, 0b100>(),
+    IGDVAC = GenDCReg<0b000, 0b0110, 0b101>(),
+    IGDSW  = GenDCReg<0b000, 0b0110, 0b110>(),
+    CGSW   = GenDCReg<0b000, 0b1010, 0b100>(),
+    CGDSW  = GenDCReg<0b000, 0b1010, 0b110>(),
+    CIGSW  = GenDCReg<0b000, 0b1110, 0b100>(),
+    CIGDSW = GenDCReg<0b000, 0b1110, 0b110>(),
+
+    // MTE
+    GVA      = GenDCReg<0b011, 0b0100, 0b011>(),
+    GZVA     = GenDCReg<0b011, 0b0100, 0b100>(),
+    CGVAC    = GenDCReg<0b011, 0b1010, 0b011>(),
+    CGDVAC   = GenDCReg<0b011, 0b1010, 0b101>(),
+    CGVAP    = GenDCReg<0b011, 0b1100, 0b011>(),
+    CGDVAP   = GenDCReg<0b011, 0b1100, 0b101>(),
+    CGVADP   = GenDCReg<0b011, 0b1101, 0b011>(),
+    CGDVADP  = GenDCReg<0b011, 0b1101, 0b101>(),
+    CIGVAC   = GenDCReg<0b011, 0b1110, 0b011>(),
+    CIGDVAC  = GenDCReg<0b011, 0b1110, 0b101>(),
+
+    // DPB
+    CVAP = GenDCReg<0b011, 0b1100, 0b001>(),
+
+    // DPB2
+    CVADP = GenDCReg<0b011, 0b1101, 0b001>(),
+  };
+
+  template<uint32_t CRm, uint32_t op2>
+  constexpr uint32_t GenHintBarrierReg() {
+    return CRm << 8 |
+      op2 << 5;
+  }
+
+  // This `HintRegister` enum is used for the hint instruction.
+  enum class HintRegister : uint32_t {
+    NOP   = GenHintBarrierReg<0b0000, 0b000>(),
+    YIELD = GenHintBarrierReg<0b0000, 0b001>(),
+    WFE   = GenHintBarrierReg<0b0000, 0b010>(),
+    WFI   = GenHintBarrierReg<0b0000, 0b011>(),
+    SEV   = GenHintBarrierReg<0b0000, 0b100>(),
+    SEVL  = GenHintBarrierReg<0b0000, 0b101>(),
+    DGH   = GenHintBarrierReg<0b0000, 0b110>(),
+    CSDB  = GenHintBarrierReg<0b0010, 0b100>(),
+  };
+
+  // This `BarrierRegister` enum is used for the various barrier instructions.
+  enum class BarrierRegister : uint32_t {
+    CLREX   = GenHintBarrierReg<0b0000, 0b010>(),
+    TCOMMIT = GenHintBarrierReg<0b0000, 0b011>(),
+    DSB     = GenHintBarrierReg<0b0000, 0b100>(),
+    DMB     = GenHintBarrierReg<0b0000, 0b101>(),
+    ISB     = GenHintBarrierReg<0b0000, 0b110>(),
+    SB      = GenHintBarrierReg<0b0000, 0b111>(),
+  };
+
+  // This `BarrierScope` enum is used for the dsb/dmb instructions.
+  enum class BarrierScope : uint32_t {
+    // Outer shareable
+    OSHLD  = 0b0001,
+    OSHST  = 0b0010,
+    OSH    = 0b0011,
+    // Non shareable
+    NSHLD  = 0b0101,
+    NSHST  = 0b0110,
+    NSH    = 0b0111,
+    // Inner shareable
+    ISHLD  = 0b1001,
+    ISHST  = 0b1010,
+    ISH    = 0b1011,
+    // Full System visibility
+    LD     = 0b1101,
+    ST     = 0b1110,
+    SY     = 0b1111,
+  };
+
+  // This `Prefetch` enum is used for prefetch instructions.
+  enum class Prefetch : uint32_t {
+    // Prefetch for load
+    PLDL1KEEP = 0b00000,
+    PLDL1STRM = 0b00001,
+    PLDL2KEEP = 0b00010,
+    PLDL2STRM = 0b00011,
+    PLDL3KEEP = 0b00100,
+    PLDL3STRM = 0b00101,
+
+    // Preload instructions
+    PLIL1KEEP = 0b01000,
+    PLIL1STRM = 0b01001,
+    PLIL2KEEP = 0b01010,
+    PLIL2STRM = 0b01011,
+    PLIL3KEEP = 0b01100,
+    PLIL3STRM = 0b01101,
+
+    // Preload for store
+    PSTL1KEEP = 0b10000,
+    PSTL1STRM = 0b10001,
+    PSTL2KEEP = 0b10010,
+    PSTL2STRM = 0b10011,
+    PSTL3KEEP = 0b10100,
+    PSTL3STRM = 0b10101,
+  };
+
+  // This `PredicatePattern` enun is used for some SVE instructions.
+  enum class PredicatePattern : uint32_t {
+    SVE_POW2  = 0b00000,
+    SVE_VL1   = 0b00001,
+    SVE_VL2   = 0b00010,
+    SVE_VL3   = 0b00011,
+    SVE_VL4   = 0b00100,
+    SVE_VL5   = 0b00101,
+    SVE_VL6   = 0b00110,
+    SVE_VL7   = 0b00111,
+    SVE_VL8   = 0b01000,
+    SVE_VL16  = 0b01001,
+    SVE_VL32  = 0b01010,
+    SVE_VL64  = 0b01011,
+    SVE_VL128 = 0b01100,
+    SVE_VL256 = 0b01101,
+    SVE_MUL4  = 0b11101,
+    SVE_MUL3  = 0b11110,
+    SVE_ALL   = 0b11111,
+  };
+
+  /* This `BackwardLabel` struct used for retaining a location for PC-Relative instructions.
+   * This is specifically a label for a target that is logically `below` an instruction that uses it.
+   * Which means that a branch would jump backwards.
+   */
+  struct BackwardLabel {
+    uint8_t *Location{};
+  };
+
+  /* This `ForwardLabel` struct used for retaining a location for PC-Relative instructions.
+   * This is specifically a label for a target that is logically `above` an instruction that uses it.
+   * Which means that a branch would jump forwards.
+   *
+   * This can be bound to multiple instructions, so it needs a vector for each bind instruction type.
+   */
+  struct ForwardLabel {
+    struct Instructions {
+      enum class InstType {
+        ADR,
+        ADRP,
+        B,
+        BC,
+        TEST_BRANCH,
+        RELATIVE_LOAD,
+      };
+      uint8_t *Location{};
+      InstType Type;
+    };
+    std::vector<Instructions> Insts{};
+  };
+
+  /* This `BiDirectionalLabel` struct used for retaining a location for PC-Relative instructions.
+   * This is specifically a label for a target that is in either direction of an instruction that uses it.
+   * Which means a branch could jump backwards or forwards depending on situation.
+   */
+  struct BiDirectionalLabel {
+    BackwardLabel Backward;
+    ForwardLabel Forward;
+  };
+
+  // This is an emitter that is designed around the smallest code bloat as possible.
+  // Eschewing most developer convenience in order to keep code as small as possible.
+
+  // Choices:
+  // - Size of ops passed as an argument rather than template to let the compiler use csel instead of branching.
+  // - Registers are unsized so they can be passed in a GPR and not need conversion operations
+  class Emitter : public FEXCore::ARMEmitter::Buffer {
+  public:
+    Emitter() = default;
+
+    Emitter(uint8_t* Base, uint64_t BaseSize)
+      : Buffer (Base, BaseSize) {
+    }
+
+    // Bind a backward label to an address.
+    // Address that is bound is the current emitter location.
+    void Bind(BackwardLabel *Label) {
+      LOGMAN_THROW_AA_FMT(Label->Location == nullptr, "Trying to bind a label twice");
+      Label->Location = GetCursorAddress<uint8_t*>();
+    }
+
+    // Bind a forward label to a location.
+    // This walks all the instructions in the label's vector.
+    // Then backpatching all instructions that have used the label.
+    template<bool WarnAboutEmpty = false>
+    void Bind(ForwardLabel *Label) {
+      if constexpr (WarnAboutEmpty) {
+        LOGMAN_THROW_A_FMT(Label->Insts.empty() == false, "Binding forward label that didn't have any instructions using it");
+      }
+      uint8_t *CurrentAddress = GetCursorAddress<uint8_t*>();
+      for (const auto &Inst : Label->Insts) {
+        // Patch up the instructions
+        switch (Inst.Type) {
+          case ForwardLabel::Instructions::InstType::ADR: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575, "Unscaled offset too large");
+            uint32_t InstMask = 0b11 << 29 | 0b1111'1111'1111'1111'111 << 5;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & 0x3F'FFFF;
+            uint32_t Inst = *Instruction & ~InstMask;
+            Inst |= (Offset & 0b11) << 29;
+            Inst |= (Offset >> 2) << 5;
+            *Instruction = Inst;
+            break;
+          }
+          case ForwardLabel::Instructions::InstType::ADRP: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(Imm >= -4294967296 && Imm <= 4294963200 && (Imm & 0xFFF) == 0, "Unscaled offset too large");
+            Imm >>= 12;
+            uint32_t InstMask = 0b11 << 29 | 0b1111'1111'1111'1111'111 << 5;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & 0x3F'FFFF;
+            uint32_t Inst = *Instruction & ~InstMask;
+            Inst |= (Offset & 0b11) << 29;
+            Inst |= (Offset >> 2) << 5;
+            *Instruction = Inst;
+            break;
+          }
+
+          case ForwardLabel::Instructions::InstType::B: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(Imm >= -134217728 && Imm <= 134217724 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+            Imm >>= 2;
+            uint32_t InstMask = 0x3FF'FFFF;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
+            uint32_t Inst = *Instruction & ~InstMask;
+            Inst |= Offset;
+            *Instruction = Inst;
+
+            break;
+          }
+
+          case ForwardLabel::Instructions::InstType::TEST_BRANCH: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(Imm >= -32768 && Imm <= 32764 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+            Imm >>= 2;
+            uint32_t InstMask = 0x3FFF;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
+            uint32_t Inst = *Instruction & ~(InstMask << 5);
+            Inst |= Offset << 5;
+            *Instruction = Inst;
+
+            break;
+          }
+          case ForwardLabel::Instructions::InstType::BC:
+          case ForwardLabel::Instructions::InstType::RELATIVE_LOAD: {
+            uint32_t *Instruction = reinterpret_cast<uint32_t*>(Inst.Location);
+            int64_t Imm = reinterpret_cast<int64_t>(CurrentAddress) - reinterpret_cast<int64_t>(Instruction);
+            LOGMAN_THROW_A_FMT(Imm >= -1048576 && Imm <= 1048575 && ((Imm & 0b11) == 0), "Unscaled offset too large");
+            Imm >>= 2;
+            uint32_t InstMask = 0x7'FFFF;
+            uint32_t Offset = static_cast<uint32_t>(Imm) & InstMask;
+            uint32_t Inst = *Instruction & ~(InstMask << 5);
+            Inst |= Offset << 5;
+            *Instruction = Inst;
+            break;
+          }
+          default: LOGMAN_MSG_A_FMT("Unexpected inst type in label fixup");
+        }
+      }
+    }
+
+    // Bind a bidirectional location to a location.
+    // Binds both forwards and backwards depending on how the label was used.
+    void Bind(BiDirectionalLabel *Label) {
+      if (!Label->Backward.Location) {
+        Bind(&Label->Backward);
+      }
+      Bind<false>(&Label->Forward);
+    }
+
+  public:
+  // TODO: Implement SME when it matters.
+#include "Interface/Core/ArchHelpers/CodeEmitter/ALUOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/BranchOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/LoadstoreOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/SystemOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/ScalarOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/ASIMDOps.inl"
+#include "Interface/Core/ArchHelpers/CodeEmitter/SVEOps.inl"
+
+  private:
+    template<typename T>
+    uint32_t Encode_ra(T Reg) const {
+      return Reg.Idx() << 10;
+    }
+    uint32_t Encode_ra(uint32_t Reg) const {
+      return Reg << 10;
+    }
+    template<typename T>
+    uint32_t Encode_rt2(T Reg) const {
+      return Reg.Idx() << 10;
+    }
+    template<>
+    uint32_t Encode_rt2(uint32_t Reg) const {
+      return Reg << 10;
+    }
+    template<typename T>
+    uint32_t Encode_rm(T Reg) const {
+      return Reg.Idx() << 16;
+    }
+    uint32_t Encode_rm(uint32_t Reg) const {
+      return Reg << 16;
+    }
+    template<typename T>
+    uint32_t Encode_rs(T Reg) const {
+      return Reg.Idx() << 16;
+    }
+    uint32_t Encode_rs(uint32_t Reg) const {
+      return Reg << 16;
+    }
+    template<typename T>
+    uint32_t Encode_rn(T Reg) const {
+      return Reg.Idx() << 5;
+    }
+    uint32_t Encode_rn(uint32_t Reg) const {
+      return Reg << 5;
+    }
+    template<typename T>
+    uint32_t Encode_rd(T Reg) const {
+      return Reg.Idx();
+    }
+    uint32_t Encode_rd(uint32_t Reg) const {
+      return Reg;
+    }
+    template<typename T>
+    uint32_t Encode_rt(T Reg) const {
+      return Reg.Idx();
+    }
+    template<>
+    uint32_t Encode_rt(Prefetch Reg) const {
+      return FEXCore::ToUnderlying(Reg);
+    }
+    uint32_t Encode_rt(uint32_t Reg) const {
+      return Reg;
+    }
+    template<typename T>
+    uint32_t Encode_pd(T Reg) const {
+      return FEXCore::ToUnderlying(Reg);
+    }
+  };
+}

External/FEXCore/Source/Interface/Core/ArchHelpers/CodeEmitter/SystemOps.inl

@@ -0,0 +1,175 @@
+/* System instruction emitters.
+ *
+ * This is mostly a mashup of various instruction types.
+ * Nothing follows an explicit pattern since they are mostly different.
+ */
+public:
+    // System with result
+    // TODO: SYSL
+    // System Instruction
+    // TODO: AT
+    // TODO: CFP
+    // TODO: CPP
+    void dc(FEXCore::ARMEmitter::DataCacheOperation DCOp, FEXCore::ARMEmitter::Register rt) {
+      constexpr uint32_t Op = 0b1101'0101'0000'1000'0111 << 12;
+      SystemInstruction(Op, 0, FEXCore::ToUnderlying(DCOp), rt);
+    }
+    // TODO: DVP
+    // TODO: IC
+    // TODO: TLBI
+
+    // Exception generation
+    void svc(uint32_t Imm) {
+      ExceptionGeneration(0b000, 0b000, 0b01, Imm);
+    }
+    void hvc(uint32_t Imm) {
+      ExceptionGeneration(0b000, 0b000, 0b10, Imm);
+    }
+    void smc(uint32_t Imm) {
+      ExceptionGeneration(0b000, 0b000, 0b11, Imm);
+    }
+    void brk(uint32_t Imm) {
+      ExceptionGeneration(0b001, 0b000, 0b00, Imm);
+    }
+    void hlt(uint32_t Imm) {
+      ExceptionGeneration(0b010, 0b000, 0b00, Imm);
+    }
+    void tcancel(uint32_t Imm) {
+      ExceptionGeneration(0b011, 0b000, 0b00, Imm);
+    }
+    void dcps1(uint32_t Imm) {
+      ExceptionGeneration(0b101, 0b000, 0b01, Imm);
+    }
+    void dcps2(uint32_t Imm) {
+      ExceptionGeneration(0b101, 0b000, 0b10, Imm);
+    }
+    void dcps3(uint32_t Imm) {
+      ExceptionGeneration(0b101, 0b000, 0b11, Imm);
+    }
+    // System instructions with register argument
+    void wfet(FEXCore::ARMEmitter::Register rt) {
+      SystemInstructionWithReg(0b0000, 0b000, rt);
+    }
+    void wfit(FEXCore::ARMEmitter::Register rt) {
+      SystemInstructionWithReg(0b0000, 0b001, rt);
+    }
+
+    // Hints
+    void nop() {
+      Hint(FEXCore::ARMEmitter::HintRegister::NOP);
+    }
+    void yield() {
+      Hint(FEXCore::ARMEmitter::HintRegister::YIELD);
+    }
+    void wfe() {
+      Hint(FEXCore::ARMEmitter::HintRegister::WFE);
+    }
+    void wfi() {
+      Hint(FEXCore::ARMEmitter::HintRegister::WFI);
+    }
+    void sev() {
+      Hint(FEXCore::ARMEmitter::HintRegister::SEV);
+    }
+    void sevl() {
+      Hint(FEXCore::ARMEmitter::HintRegister::SEVL);
+    }
+    void dgh() {
+      Hint(FEXCore::ARMEmitter::HintRegister::DGH);
+    }
+    void csdb() {
+      Hint(FEXCore::ARMEmitter::HintRegister::CSDB);
+    }
+
+    // Barriers
+    void clrex(uint32_t imm = 15) {
+      LOGMAN_THROW_AA_FMT(imm < 16, "Immediate out of range");
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::CLREX, imm);
+    }
+    void dsb(FEXCore::ARMEmitter::BarrierScope Scope) {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::DSB, FEXCore::ToUnderlying(Scope));
+    }
+    void dmb(FEXCore::ARMEmitter::BarrierScope Scope) {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::DMB, FEXCore::ToUnderlying(Scope));
+    }
+    void isb() {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::ISB, FEXCore::ToUnderlying(FEXCore::ARMEmitter::BarrierScope::SY));
+    }
+    void sb() {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::SB, 0);
+    }
+    void tcommit() {
+      Barrier(FEXCore::ARMEmitter::BarrierRegister::TCOMMIT, 0);
+    }
+
+    // System register move
+    void msr(FEXCore::ARMEmitter::SystemRegister reg, FEXCore::ARMEmitter::Register rt) {
+      constexpr uint32_t Op = 0b1101'0101'0001 << 20;
+      SystemRegisterMove(Op, rt, reg);
+    }
+
+    void mrs(FEXCore::ARMEmitter::Register rd, FEXCore::ARMEmitter::SystemRegister reg) {
+      constexpr uint32_t Op = 0b1101'0101'0011 << 20;
+      SystemRegisterMove(Op, rd, reg);
+    }
+
+private:
+
+    // Exception Generation
+    void ExceptionGeneration(uint32_t opc, uint32_t op2, uint32_t LL, uint32_t Imm) {
+      LOGMAN_THROW_AA_FMT((Imm & 0xFFFF'0000) == 0, "Imm amount too large");
+
+      uint32_t Instr = 0b1101'0100 << 24;
+
+      Instr |= opc << 21;
+      Instr |= Imm << 5;
+      Instr |= op2 << 2;
+      Instr |= LL;
+
+      dc32(Instr);
+    }
+
+    // System instructions with register argument
+    void SystemInstructionWithReg(uint32_t CRm, uint32_t op2, FEXCore::ARMEmitter::Register rt) {
+      uint32_t Instr = 0b1101'0101'0000'0011'0001 << 12;
+
+      Instr |= CRm << 8;
+      Instr |= op2 << 5;
+      Instr |= Encode_rt(rt);
+      dc32(Instr);
+    }
+
+    // Hints
+    void Hint(FEXCore::ARMEmitter::HintRegister Reg) {
+      uint32_t Instr = 0b1101'0101'0000'0011'0010'0000'0001'1111U;
+      Instr |= FEXCore::ToUnderlying(Reg);
+      dc32(Instr);
+    }
+    // Barriers
+    void Barrier(FEXCore::ARMEmitter::BarrierRegister Reg, uint32_t CRm) {
+      uint32_t Instr = 0b1101'0101'0000'0011'0011'0000'0001'1111U;
+      Instr |= CRm << 8;
+      Instr |= FEXCore::ToUnderlying(Reg);
+      dc32(Instr);
+    }
+
+    // System Instruction
+    void SystemInstruction(uint32_t Op, uint32_t L, uint32_t SubOp, FEXCore::ARMEmitter::Register rt) {
+      uint32_t Instr = Op;
+
+      Instr |= L << 21;
+      Instr |= SubOp;
+      Instr |= Encode_rt(rt);
+
+      dc32(Instr);
+    }
+
+    // System register move
+    void SystemRegisterMove(uint32_t Op, FEXCore::ARMEmitter::Register rt, FEXCore::ARMEmitter::SystemRegister reg) {
+      uint32_t Instr = Op;
+
+      Instr |= FEXCore::ToUnderlying(reg);
+      Instr |= Encode_rt(rt);
+
+      dc32(Instr);
+    }
+

External/FEXCore/Source/Interface/Core/ArchHelpers/MContext.h

@@ -3,6 +3,7 @@
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Core/UContext.h>
+#include <FEXCore/Core/X86Enums.h>
 
 #include <signal.h>
 #include <string.h>
@@ -10,7 +11,6 @@
 #include <stdint.h>
 #include <type_traits>
 
-
 namespace FEXCore::ArchHelpers::Context {
 
 enum ContextFlags : uint32_t {
@@ -76,6 +76,7 @@ static inline mcontext_t* GetMContext(void* ucontext) {
 #ifdef _M_ARM_64
 
 constexpr uint32_t FPR_MAGIC = 0x46508001U;
+constexpr uint32_t ESR1_MAGIC = 0x45535201U;
 
 struct HostCTXHeader {
   uint32_t Magic;
@@ -89,6 +90,11 @@ struct HostFPRState {
   __uint128_t FPRs[32];
 };
 
+struct HostESRState {
+  HostCTXHeader Head;
+  uint64_t ESR;
+};
+
 static inline uint64_t GetSp(void* ucontext) {
   return GetMContext(ucontext)->sp;
 }
@@ -129,6 +135,61 @@ static inline __uint128_t GetArmFPR(void* ucontext, uint32_t id) {
   return HostState->FPRs[id];
 }
 
+static inline uint64_t GetArmESR(void* ucontext) {
+  auto MContext = GetMContext(ucontext);
+
+  size_t i = 0;
+  auto HostState = reinterpret_cast<HostCTXHeader*>(&MContext->__reserved[i]);
+  do {
+    if (HostState->Magic == ESR1_MAGIC) {
+      auto ESR = reinterpret_cast<HostESRState*>(HostState);
+      return ESR->ESR;
+    }
+    i += HostState->Size;
+    HostState = reinterpret_cast<HostCTXHeader*>(&MContext->__reserved[i]);
+  } while (HostState->Size != 0);
+
+  return 0;
+}
+
+constexpr static uint64_t ESR1_EC = 0b111111U << 26;
+constexpr static uint64_t ESR1_EC_DataAbort = 0b100100U << 26;
+
+// Write-Not-Read flag
+// When set - Abort is due to a write
+constexpr static uint64_t ESR1_WNR = 1 << 6;
+
+// DFSC - Default Status Code
+// Translation fault - No page mapped
+// Permissions fault - Page mapped but with incorrect permission from access.
+constexpr static uint64_t ESR1_DataAbort_DFSC = 0b111111;
+constexpr static uint64_t ESR1_DataAbort_TranslationFault_EL0 = 0b000111;
+constexpr static uint64_t ESR1_DataAbort_PermissionFault_EL0 = 0b001111;
+constexpr static uint64_t ESR1_DataAbort_Level = 0b11;
+constexpr static uint64_t ESR1_DataAbort_Level_EL3 = 0b00;
+constexpr static uint64_t ESR1_DataAbort_Level_EL2 = 0b01;
+constexpr static uint64_t ESR1_DataAbort_Level_EL1 = 0b10;
+constexpr static uint64_t ESR1_DataAbort_Level_EL0 = 0b11;
+
+static inline uint32_t GetProtectFlags(void* ucontext) {
+  uint64_t ESR = GetArmESR(ucontext);
+  LOGMAN_THROW_A_FMT((ESR & ESR1_EC) == ESR1_EC_DataAbort, "Unknown ESR1 EC type: 0x{:x} != 0x{:x}", ESR & ESR1_EC, ESR1_EC_DataAbort);
+
+  uint32_t ProtectFlags{};
+  if ((ESR & ESR1_DataAbort_Level) == ESR1_DataAbort_Level_EL0) {
+    // Always a user error for us.
+    ProtectFlags |= X86State::X86_PF_USER;
+  }
+
+  if (ESR & ESR1_WNR) {
+    // Fault was due to a write
+    ProtectFlags |= X86State::X86_PF_WRITE;
+  }
+
+  // PF_PROT is not returned to user on x86, so don't return the difference between permission fault and translation fault.
+  return ProtectFlags;
+}
+
 using ContextBackup = ArmContextBackup;
 template <typename T>
 static inline void BackupContext(void* ucontext, T *Backup) {
@@ -222,6 +283,10 @@ static inline __uint128_t GetArmFPR(void* ucontext, uint32_t id) {
   ERROR_AND_DIE_FMT("Not implemented for x86 host");
 }
 
+static inline uint32_t GetProtectFlags(void* ucontext) {
+  return GetMContext(ucontext)->gregs[REG_ERR];
+}
+
 using ContextBackup = X86ContextBackup;
 template <typename T>
 static inline void BackupContext(void* ucontext, T *Backup) {

External/FEXCore/Source/Interface/Core/CPUID.cpp

@@ -421,7 +421,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
 
   Res.ecx =
     (1 <<  0) | // SSE3
-    (1 <<  1) | // PCLMULQDQ
+    (CTX->HostFeatures.SupportsPMULL_128Bit <<  1) | // PCLMULQDQ
     (1 <<  2) | // DS area supports 64bit layout
     (1 <<  3) | // MWait
     (0 <<  4) | // DS-CPL

External/FEXCore/Source/Interface/Core/Core.cpp

@@ -44,6 +44,7 @@ $end_info$
 #include <FEXCore/Utils/Event.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Threads.h>
+#include <FEXCore/Utils/Profiler.h>
 #include <FEXHeaderUtils/Syscalls.h>
 #include <FEXHeaderUtils/TodoDefines.h>
 
@@ -674,6 +675,8 @@ namespace FEXCore::Context {
   }
 
   void Context::ClearCodeCache(FEXCore::Core::InternalThreadState *Thread) {
+    FEXCORE_PROFILE_INSTANT("ClearCodeCache");
+
     {
       // Ensure the Code Object Serialization service has fully serialized this thread's data before clearing the cache
       // Use the thread's object cache ref counter for this
@@ -741,6 +744,8 @@ namespace FEXCore::Context {
   }
 
   Context::GenerateIRResult Context::GenerateIR(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP, bool ExtendedDebugInfo) {
+    FEXCORE_PROFILE_SCOPED("GenerateIR");
+
     Thread->OpDispatcher->ReownOrClaimBuffer();
     Thread->OpDispatcher->ResetWorkingList();
 
@@ -851,22 +856,25 @@ namespace FEXCore::Context {
             Thread->OpDispatcher->_ExitFunction(Thread->OpDispatcher->_EntrypointOffset(Block.Entry - GuestRIP, GPRSize));
           }
 
-          // If we had a dispatch error then leave early
-          if (HadDispatchError) {
-            if (TotalInstructions == 0) {
-              // Couldn't handle any instruction in op dispatcher
-              Thread->OpDispatcher->ResetWorkingList();
-              return { nullptr, nullptr, 0, 0, 0, 0 };
-            }
-            else {
-              const uint8_t GPRSize = GetGPRSize();
+          const bool NeedsBlockEnd = (HadDispatchError && TotalInstructions > 0) ||
+            (Thread->OpDispatcher->NeedsBlockEnder() && i + 1 == InstsInBlock);
 
-              // We had some instructions. Early exit
-              Thread->OpDispatcher->_ExitFunction(Thread->OpDispatcher->_EntrypointOffset(Block.Entry + BlockInstructionsLength - GuestRIP, GPRSize));
-              break;
-            }
+          // If we had a dispatch error then leave early
+          if (HadDispatchError && TotalInstructions == 0) {
+            // Couldn't handle any instruction in op dispatcher
+            Thread->OpDispatcher->ResetWorkingList();
+            return { nullptr, nullptr, 0, 0, 0, 0 };
           }
 
+          if (NeedsBlockEnd) {
+            const uint8_t GPRSize = GetGPRSize();
+
+            // We had some instructions. Early exit
+            Thread->OpDispatcher->_ExitFunction(Thread->OpDispatcher->_EntrypointOffset(Block.Entry + BlockInstructionsLength - GuestRIP, GPRSize));
+            break;
+          }
+
+
           if (Thread->OpDispatcher->FinishOp(DecodedInfo->PC + DecodedInfo->InstSize, i + 1 == InstsInBlock)) {
             break;
           }
@@ -1011,6 +1019,7 @@ namespace FEXCore::Context {
   }
 
   uintptr_t Context::CompileBlock(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
+    FEXCORE_PROFILE_SCOPED("CompileBlock");
     auto Thread = Frame->Thread;
 
     // Invalidate might take a unique lock on this, to guarantee that during invalidation no code gets compiled
@@ -1206,8 +1215,6 @@ namespace FEXCore::Context {
       IsMemoryShared = true;
 
       if (Config.TSOAutoMigration) {
-        LogMan::Msg::IFmt("Migrating to shared memory mode");
-
         std::lock_guard<std::mutex> lkThreads(ThreadCreationMutex);
         LogMan::Throw::AFmt(Threads.size() == 1, "First MarkMemoryShared called must be before creating any threads");
 

External/FEXCore/Source/Interface/Core/Dispatcher/Arm64Dispatcher.cpp

@@ -1,3 +1,4 @@
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/LookupCache.h"
 
 #include "Interface/Core/ArchHelpers/MContext.h"
@@ -30,15 +31,9 @@
 #include <sys/syscall.h>
 #include <unistd.h>
 
-#define STATE_PTR(STATE_TYPE, FIELD) \
-  MemOperand(STATE, offsetof(FEXCore::Core::STATE_TYPE, FIELD))
-
 namespace FEXCore::CPU {
 
-using namespace vixl;
-using namespace vixl::aarch64;
-
-constexpr size_t MAX_DISPATCHER_CODE_SIZE = 8192;
+constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096;
 
 Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &config)
   : FEXCore::CPU::Dispatcher(ctx, config), Arm64Emitter(ctx, MAX_DISPATCHER_CODE_SIZE)
@@ -46,13 +41,18 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
   , Simulator {&Decoder}
 #endif
 {
-
 #ifdef VIXL_SIMULATOR
   // Hardcode a 256-bit vector width if we are running in the simulator.
   Simulator.SetVectorLengthInBits(256);
 #endif
 
-  SetAllowAssembler(true);
+  EmitDispatcher();
+}
+
+void Arm64Dispatcher::EmitDispatcher() {
+#ifdef VIXL_DISASSEMBLER
+  const auto DisasmBegin = GetCursorAddress<const vixl::aarch64::Instruction*>();
+#endif
 
   DispatchPtr = GetCursorAddress<AsmDispatch>();
 
@@ -64,9 +64,9 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
   //    Ptr();
   // }
 
-  Literal l_CTX {reinterpret_cast<uintptr_t>(CTX)};
-  Literal l_Sleep {reinterpret_cast<uint64_t>(SleepThread)};
-  Literal l_CompileBlock {GetCompileBlockPtr()};
+  ARMEmitter::ForwardLabel l_CTX;
+  ARMEmitter::ForwardLabel l_Sleep;
+  ARMEmitter::ForwardLabel l_CompileBlock;
 
   // Push all the register we need to save
   PushCalleeSavedRegisters();
@@ -74,12 +74,12 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
   // Push our memory base to the correct register
   // Move our thread pointer to the correct register
   // This is passed in to parameter 0 (x0)
-  mov(STATE, x0);
+  mov(STATE, ARMEmitter::XReg::x0);
 
   // Save this stack pointer so we can cleanly shutdown the emulation with a long jump
   // regardless of where we were in the stack
-  add(x0, sp, 0);
-  str(x0, STATE_PTR(CpuStateFrame, ReturningStackLocation));
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, ARMEmitter::Reg::rsp, 0);
+  str(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, ReturningStackLocation));
 
   AbsoluteLoopTopAddressFillSRA = GetCursorAddress<uint64_t>();
 
@@ -89,91 +89,97 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
 
   // We want to ensure that we are 16 byte aligned at the top of this loop
   Align16B();
-  aarch64::Label FullLookup{};
-  aarch64::Label CallBlock{};
-  aarch64::Label LoopTop{};
-  aarch64::Label ExitSpillSRA{};
-  aarch64::Label ThreadPauseHandler{};
+  ARMEmitter::BiDirectionalLabel FullLookup{};
+  ARMEmitter::BiDirectionalLabel CallBlock{};
+  ARMEmitter::BackwardLabel LoopTop{};
 
-  bind(&LoopTop);
-  AbsoluteLoopTopAddress = GetLabelAddress<uint64_t>(&LoopTop);
+  Bind(&LoopTop);
+  AbsoluteLoopTopAddress = GetCursorAddress<uint64_t>();
 
   // Load in our RIP
   // Don't modify x2 since it contains our RIP once the block doesn't exist
-  ldr(x2, STATE_PTR(CpuStateFrame, State.rip));
-  auto RipReg = x2;
+
+  auto RipReg = ARMEmitter::XReg::x2;
+  ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));
 
   // L1 Cache
-  ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
+  ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
 
-  and_(x3, RipReg, LookupCache::L1_ENTRIES_MASK);
-  add(x0, x0, Operand(x3, Shift::LSL, 4));
-  ldp(x3, x0, MemOperand(x0));
-  cmp(x0, RipReg);
-  b(&FullLookup, Condition::ne);
+  and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, RipReg.R(), LookupCache::L1_ENTRIES_MASK);
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, ARMEmitter::Reg::r0, ARMEmitter::Reg::r3, ARMEmitter::ShiftType::LSL , 4);
+  ldp<ARMEmitter::IndexType::OFFSET>(ARMEmitter::XReg::x3, ARMEmitter::XReg::x0, ARMEmitter::Reg::r0, 0);
+  cmp(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, RipReg.R());
+  b(ARMEmitter::Condition::CC_NE, &FullLookup);
 
-  br(x3);
+  br(ARMEmitter::Reg::r3);
 
   // L1C check failed, do a full lookup
-  bind(&FullLookup);
+  Bind(&FullLookup);
 
   // This is the block cache lookup routine
   // It matches what is going on it LookupCache.h::FindBlock
-  ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
+  ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
 
   // Mask the address by the virtual address size so we can check for aliases
   uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
   if (std::popcount(VirtualMemorySize) == 1) {
-    and_(x3, RipReg, VirtualMemorySize - 1);
+    and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, RipReg.R(), VirtualMemorySize - 1);
   }
   else {
-    LoadConstant(x3, VirtualMemorySize);
-    and_(x3, RipReg, x3);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, VirtualMemorySize);
+    and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, RipReg.R(), ARMEmitter::Reg::r3);
   }
 
-  aarch64::Label NoBlock;
+#ifdef VIXL_SIMULATOR
+  // VIXL simulator can't run syscalls.
+  constexpr bool SignalSafeCompile = false;
+#else
+  constexpr bool SignalSafeCompile = true;
+#endif
+
+  ARMEmitter::ForwardLabel NoBlock;
+
   {
     // Offset the address and add to our page pointer
-    lsr(x1, x3, 12);
+    lsr(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::r3, 12);
 
     // Load the pointer from the offset
-    ldr(x0, MemOperand(x0, x1, Shift::LSL, 3));
+    ldr(ARMEmitter::XReg::x0, ARMEmitter::Reg::r0, ARMEmitter::Reg::r1, ARMEmitter::ExtendedType::LSL_64, 3);
 
     // If page pointer is zero then we have no block
-    cbz(x0, &NoBlock);
+    cbz(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, &NoBlock);
 
     // Steal the page offset
-    and_(x1, x3, 0x0FFF);
+    and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::r3, 0x0FFF);
 
     // Shift the offset by the size of the block cache entry
-    add(x0, x0, Operand(x1, Shift::LSL, (int)log2(sizeof(FEXCore::LookupCache::LookupCacheEntry))));
+    add(ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, ARMEmitter::XReg::x1, ARMEmitter::ShiftType::LSL, (int)log2(sizeof(FEXCore::LookupCache::LookupCacheEntry)));
 
     // Load the guest address first to ensure it maps to the address we are currently at
     // This fixes aliasing problems
-    ldr(x1, MemOperand(x0, offsetof(FEXCore::LookupCache::LookupCacheEntry, GuestCode)));
-    cmp(x1, RipReg);
-    b(&NoBlock, Condition::ne);
+    ldr(ARMEmitter::XReg::x1, ARMEmitter::Reg::r0, offsetof(FEXCore::LookupCache::LookupCacheEntry, GuestCode));
+    cmp(ARMEmitter::XReg::x1, RipReg);
+    b(ARMEmitter::Condition::CC_NE, &NoBlock);
 
     // Now load the actual host block to execute if we can
-    ldr(x3, MemOperand(x0, offsetof(FEXCore::LookupCache::LookupCacheEntry, HostCode)));
-    cbz(x3, &NoBlock);
+    ldr(ARMEmitter::XReg::x3, ARMEmitter::Reg::r0, offsetof(FEXCore::LookupCache::LookupCacheEntry, HostCode));
+    cbz(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, &NoBlock);
 
     // If we've made it here then we have a real compiled block
     {
       // update L1 cache
-      ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
+      ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
 
-      and_(x1, RipReg, LookupCache::L1_ENTRIES_MASK);
-      add(x0, x0, Operand(x1, Shift::LSL, 4));
-      stp(x3, x2, MemOperand(x0));
+      and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, RipReg.R(), LookupCache::L1_ENTRIES_MASK);
+      add(ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, ARMEmitter::XReg::x1, ARMEmitter::ShiftType::LSL, 4);
+      stp<ARMEmitter::IndexType::OFFSET>(ARMEmitter::XReg::x3, ARMEmitter::XReg::x2, ARMEmitter::Reg::r0);
 
       // Jump to the block
-      br(x3);
+      br(ARMEmitter::Reg::r3);
     }
   }
 
   {
-    bind(&ExitSpillSRA);
     ThreadStopHandlerAddressSpillSRA = GetCursorAddress<uint64_t>();
     if (config.StaticRegisterAllocation)
       SpillStaticRegs();
@@ -187,12 +193,6 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
     ret();
   }
 
-#ifdef VIXL_SIMULATOR
-  // VIXL simulator can't run syscalls.
-  constexpr bool SignalSafeCompile = false;
-#else
-  constexpr bool SignalSafeCompile = true;
-#endif
   {
     ExitFunctionLinkerAddress = GetCursorAddress<uint64_t>();
     if (config.StaticRegisterAllocation)
@@ -207,52 +207,52 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
       // X3: Size of mask, sizeof(uint64_t)
       // X8: Syscall
 
-      LoadConstant(x0, ~0ULL);
-      stp(x0, x0, MemOperand(sp, -16, PreIndex));
-      LoadConstant(x0, SIG_SETMASK);
-      add(x1, sp, 0);
-      add(x2, sp, 0);
-      LoadConstant(x3, 8);
-      LoadConstant(x8, SYS_rt_sigprocmask);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, ~0ULL);
+      stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, ARMEmitter::Reg::rsp, -16);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SIG_SETMASK);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::rsp, 0);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::rsp, 0);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, 8);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, SYS_rt_sigprocmask);
       svc(0);
     }
 
-    mov(x0, STATE);
-    mov(x1, lr);
+    mov(ARMEmitter::XReg::x0, STATE);
+    mov(ARMEmitter::XReg::x1, ARMEmitter::XReg::lr);
 
-    ldr(x2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
+    ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
 #ifdef VIXL_SIMULATOR
-    GenerateIndirectRuntimeCall<uintptr_t, void *, void *>(x2);
+    GenerateIndirectRuntimeCall<uintptr_t, void *, void *>(ARMEmitter::Reg::r2);
 #else
-    blr(x2);
+    blr(ARMEmitter::Reg::r2);
 #endif
 
     if (SignalSafeCompile) {
       // Now restore the signal mask
       // Living in the same location
 
-      mov(x4, x0);
-      LoadConstant(x0, SIG_SETMASK);
-      add(x1, sp, 0);
-      LoadConstant(x2, 0);
-      LoadConstant(x3, 8);
-      LoadConstant(x8, SYS_rt_sigprocmask);
+      mov(ARMEmitter::XReg::x4, ARMEmitter::XReg::x0);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SIG_SETMASK);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::rsp, 0);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, 0);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, 8);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, SYS_rt_sigprocmask);
       svc(0);
 
       // Bring stack back
-      add(sp, sp, 16);
-
-      mov(x0, x4);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);
+      mov(ARMEmitter::XReg::x0, ARMEmitter::XReg::x4);
     }
 
     if (config.StaticRegisterAllocation)
       FillStaticRegs();
-    br(x0);
+
+    br(ARMEmitter::Reg::r0);
   }
 
   // Need to create the block
   {
-    bind(&NoBlock);
+    Bind(&NoBlock);
 
     if (config.StaticRegisterAllocation)
       SpillStaticRegs();
@@ -266,42 +266,42 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
       // X3: Size of mask, sizeof(uint64_t)
       // X8: Syscall
 
-      LoadConstant(x0, ~0ULL);
-      stp(x0, x2, MemOperand(sp, -16, PreIndex));
-      LoadConstant(x0, SIG_SETMASK);
-      add(x1, sp, 0);
-      add(x2, sp, 0);
-      LoadConstant(x3, 8);
-      LoadConstant(x8, SYS_rt_sigprocmask);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, ~0ULL);
+      stp<ARMEmitter::IndexType::PRE>(ARMEmitter::XReg::x0, ARMEmitter::XReg::x2, ARMEmitter::Reg::rsp, -16);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SIG_SETMASK);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::rsp, 0);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::rsp, 0);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, 8);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, SYS_rt_sigprocmask);
       svc(0);
 
       // Reload x2 to bring back RIP
-      ldr(x2, MemOperand(sp, 8, Offset));
+      ldr(ARMEmitter::XReg::x2, ARMEmitter::Reg::rsp, 8);
     }
 
-    ldr(x0, &l_CTX);
-    mov(x1, STATE);
-    ldr(x3, &l_CompileBlock);
+    ldr(ARMEmitter::XReg::x0, &l_CTX);
+    mov(ARMEmitter::XReg::x1, STATE);
+    ldr(ARMEmitter::XReg::x3, &l_CompileBlock);
 
     // X2 contains our guest RIP
 #ifdef VIXL_SIMULATOR
-    GenerateIndirectRuntimeCall<void, void *, uint64_t, void *>(x3);
+    GenerateIndirectRuntimeCall<void, void *, uint64_t, void *>(ARMEmitter::Reg::r3);
 #else
-    blr(x3); // { CTX, Frame, RIP}
+    blr(ARMEmitter::Reg::r3); // { CTX, Frame, RIP}
 #endif
 
     if (SignalSafeCompile) {
       // Now restore the signal mask
       // Living in the same location
-      LoadConstant(x0, SIG_SETMASK);
-      add(x1, sp, 0);
-      LoadConstant(x2, 0);
-      LoadConstant(x3, 8);
-      LoadConstant(x8, SYS_rt_sigprocmask);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SIG_SETMASK);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::rsp, 0);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, 0);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, 8);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, SYS_rt_sigprocmask);
       svc(0);
 
       // Bring stack back
-      add(sp, sp, 16);
+      add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);
     }
 
     if (config.StaticRegisterAllocation)
@@ -321,7 +321,7 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
   {
     // Guest SIGILL handler
     // Needs to be distinct from the SignalHandlerReturnAddress
-    GuestSignal_SIGILL  = GetCursorAddress<uint64_t>();
+    GuestSignal_SIGILL = GetCursorAddress<uint64_t>();
 
     if (config.StaticRegisterAllocation)
       SpillStaticRegs();
@@ -352,8 +352,8 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
     // brk = SIGTRAP
     // ??? = SIGSEGV
     // Force a SIGSEGV by loading zero
-    LoadConstant(x1, 0);
-    ldr(x1, MemOperand(x1));
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, 0);
+    ldr(ARMEmitter::XReg::x1, ARMEmitter::Reg::r1);
   }
 
   {
@@ -361,19 +361,18 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
     if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
-    bind(&ThreadPauseHandler);
     ThreadPauseHandlerAddress = GetCursorAddress<uint64_t>();
     // We are pausing, this means the frontend should be waiting for this thread to idle
     // We will have faulted and jumped to this location at this point
 
     // Call our sleep handler
-    ldr(x0, &l_CTX);
-    mov(x1, STATE);
-    ldr(x2, &l_Sleep);
+    ldr(ARMEmitter::XReg::x0, &l_CTX);
+    mov(ARMEmitter::XReg::x1, STATE);
+    ldr(ARMEmitter::XReg::x2, &l_Sleep);
 #ifdef VIXL_SIMULATOR
-    GenerateIndirectRuntimeCall<void, void *, void *>(x2);
+    GenerateIndirectRuntimeCall<void, void *, void *>(ARMEmitter::Reg::r2);
 #else
-    blr(x2);
+    blr(ARMEmitter::Reg::r2);
 #endif
 
     PauseReturnInstruction = GetCursorAddress<uint64_t>();
@@ -403,27 +402,27 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
     PushCalleeSavedRegisters();
 
     // First thing we need to move the thread state pointer back in to our register
-    mov(STATE, x0);
+    mov(STATE, ARMEmitter::XReg::x0);
 
     // Make sure to adjust the refcounter so we don't clear the cache now
-    ldr(w2, STATE_PTR(CpuStateFrame, SignalHandlerRefCounter));
-    add(w2, w2, 1);
-    str(w2, STATE_PTR(CpuStateFrame, SignalHandlerRefCounter));
+    ldr(ARMEmitter::WReg::w2, STATE_PTR(CpuStateFrame, SignalHandlerRefCounter));
+    add(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, 1);
+    str(ARMEmitter::WReg::w2, STATE_PTR(CpuStateFrame, SignalHandlerRefCounter));
 
     // Now push the callback return trampoline to the guest stack
     // Guest will be misaligned because calling a thunk won't correct the guest's stack once we call the callback from the host
-    LoadConstant(x0, CTX->X86CodeGen.CallbackReturn);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, CTX->X86CodeGen.CallbackReturn);
 
-    ldr(x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
-    sub(x2, x2, 16);
-    str(x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
+    ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
+    sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, 16);
+    str(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
 
     // Store the trampoline to the guest stack
     // Guest stack is now correctly misaligned after a regular call instruction
-    str(x0, MemOperand(x2));
+    str(ARMEmitter::XReg::x0, ARMEmitter::Reg::r2, 0);
 
     // Store RIP to the context state
-    str(x1, STATE_PTR(CpuStateFrame, State.rip));
+    str(ARMEmitter::XReg::x1, STATE_PTR(CpuStateFrame, State.rip));
 
     // load static regs
     if (config.StaticRegisterAllocation)
@@ -436,14 +435,14 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
   {
     LUDIVHandlerAddress = GetCursorAddress<uint64_t>();
 
-    PushDynamicRegsAndLR();
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
     SpillStaticRegs();
 
-    ldr(x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUDIV));
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUDIV));
 #ifdef VIXL_SIMULATOR
-    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(x3);
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
 #else
-    blr(x3);
+    blr(ARMEmitter::Reg::r3);
 #endif
     FillStaticRegs();
 
@@ -458,14 +457,14 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
   {
     LDIVHandlerAddress = GetCursorAddress<uint64_t>();
 
-    PushDynamicRegsAndLR();
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
     SpillStaticRegs();
 
-    ldr(x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LDIV));
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LDIV));
 #ifdef VIXL_SIMULATOR
-    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(x3);
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
 #else
-    blr(x3);
+    blr(ARMEmitter::Reg::r3);
 #endif
     FillStaticRegs();
 
@@ -480,14 +479,14 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
   {
     LUREMHandlerAddress = GetCursorAddress<uint64_t>();
 
-    PushDynamicRegsAndLR();
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
     SpillStaticRegs();
 
-    ldr(x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUREM));
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUREM));
 #ifdef VIXL_SIMULATOR
-    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(x3);
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
 #else
-    blr(x3);
+    blr(ARMEmitter::Reg::r3);
 #endif
     FillStaticRegs();
 
@@ -502,14 +501,15 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
   {
     LREMHandlerAddress = GetCursorAddress<uint64_t>();
 
-    PushDynamicRegsAndLR();
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
     SpillStaticRegs();
 
-    ldr(x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LREM));
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LREM));
+
 #ifdef VIXL_SIMULATOR
-    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(x3);
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
 #else
-    blr(x3);
+    blr(ARMEmitter::Reg::r3);
 #endif
     FillStaticRegs();
 
@@ -521,16 +521,16 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
     ret();
   }
 
-  place(&l_CTX);
-  place(&l_Sleep);
-  place(&l_CompileBlock);
+  Bind(&l_CTX);
+  dc64(reinterpret_cast<uintptr_t>(CTX));
+  Bind(&l_Sleep);
+  dc64(reinterpret_cast<uint64_t>(SleepThread));
+  Bind(&l_CompileBlock);
+  dc64(GetCompileBlockPtr());
 
-
-  FinalizeCode();
   Start = reinterpret_cast<uint64_t>(DispatchPtr);
   End = GetCursorAddress<uint64_t>();
-  vixl::aarch64::CPU::EnsureIAndDCacheCoherency(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr));
-  GetBuffer()->SetExecutable();
+  ClearICache(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr));
 
   if (CTX->Config.BlockJITNaming()) {
     std::string Name = "Dispatch_" + std::to_string(FHU::Syscalls::gettid());
@@ -539,106 +539,100 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const Dispatche
   if (CTX->Config.GlobalJITNaming()) {
     CTX->Symbols.RegisterJITSpace(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr));
   }
+
+#ifdef VIXL_DISASSEMBLER
+  const auto DisasmEnd = GetCursorAddress<const vixl::aarch64::Instruction*>();
+  Disasm.DisassembleBuffer(DisasmBegin, DisasmEnd);
+#endif
 }
 
 #ifdef VIXL_SIMULATOR
 void Arm64Dispatcher::ExecuteDispatch(FEXCore::Core::CpuStateFrame *Frame) {
   Simulator.WriteXRegister(0, reinterpret_cast<int64_t>(Frame));
-  Simulator.RunFrom(reinterpret_cast<Instruction const*>(DispatchPtr));
+  Simulator.RunFrom(reinterpret_cast<vixl::aarch64::Instruction const*>(DispatchPtr));
 }
 
 void Arm64Dispatcher::ExecuteJITCallback(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP) {
   Simulator.WriteXRegister(0, reinterpret_cast<int64_t>(Frame));
   Simulator.WriteXRegister(1, RIP);
-  Simulator.RunFrom(reinterpret_cast<Instruction const*>(CallbackPtr));
+  Simulator.RunFrom(reinterpret_cast<vixl::aarch64::Instruction const*>(CallbackPtr));
 }
 
 #endif
 
-// Used by GenerateGDBPauseCheck, GenerateInterpreterTrampoline, destination buffer is set before use
-static thread_local vixl::aarch64::Assembler emit((uint8_t*)&emit, 1);
-
 size_t Arm64Dispatcher::GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) {
+  FEXCore::ARMEmitter::Emitter emit{CodeBuffer, MaxGDBPauseCheckSize};
 
-  *emit.GetBuffer() = vixl::CodeBuffer(CodeBuffer, MaxGDBPauseCheckSize);
-
-  vixl::CodeBufferCheckScope scope(&emit, MaxGDBPauseCheckSize, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
-
-  aarch64::Label RunBlock;
+  ARMEmitter::ForwardLabel RunBlock;
 
   // If we have a gdb server running then run in a less efficient mode that checks if we need to exit
   // This happens when single stepping
 
   static_assert(sizeof(FEXCore::Context::Context::Config.RunningMode) == 4, "This is expected to be size of 4");
-  emit.ldr(x0, STATE_PTR(CpuStateFrame, Thread)); // Get thread
-  emit.ldr(x0, MemOperand(x0, offsetof(FEXCore::Core::InternalThreadState, CTX))); // Get Context
-  emit.ldr(w0, MemOperand(x0, offsetof(FEXCore::Context::Context, Config.RunningMode)));
+  emit.ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Thread));
+  emit.ldr(ARMEmitter::XReg::x0, ARMEmitter::Reg::r0, offsetof(FEXCore::Core::InternalThreadState, CTX)); // Get Context
+  emit.ldr(ARMEmitter::WReg::w0, ARMEmitter::Reg::r0, offsetof(FEXCore::Context::Context, Config.RunningMode));
 
   // If the value == 0 then we don't need to stop
-  emit.cbz(w0, &RunBlock);
+  emit.cbz(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r0, &RunBlock);
   {
-    Literal l_GuestRIP {GuestRIP};
+    ARMEmitter::ForwardLabel l_GuestRIP;
     // Make sure RIP is syncronized to the context
-    emit.ldr(x0, &l_GuestRIP);
-    emit.str(x0, STATE_PTR(CpuStateFrame, State.rip));
+    emit.ldr(ARMEmitter::XReg::x0, &l_GuestRIP);
+    emit.str(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, State.rip));
 
     // Stop the thread
-    emit.ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.ThreadPauseHandlerSpillSRA));
-    emit.br(x0);
-    emit.place(&l_GuestRIP);
+    emit.ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.ThreadPauseHandlerSpillSRA));
+    emit.br(ARMEmitter::Reg::r0);
+    emit.Bind(&l_GuestRIP);
+    emit.dc64(GuestRIP);
   }
-  emit.bind(&RunBlock);
-  emit.FinalizeCode();
+  emit.Bind(&RunBlock);
 
-  auto UsedBytes = emit.GetBuffer()->GetCursorOffset();
-  vixl::aarch64::CPU::EnsureIAndDCacheCoherency(CodeBuffer, UsedBytes);
+  auto UsedBytes = emit.GetCursorOffset();
+  emit.ClearICache(CodeBuffer, UsedBytes);
   return UsedBytes;
 }
 
 size_t Arm64Dispatcher::GenerateInterpreterTrampoline(uint8_t *CodeBuffer) {
   LOGMAN_THROW_AA_FMT(!config.StaticRegisterAllocation, "GenerateInterpreterTrampoline dispatcher does not support SRA");
 
-  *emit.GetBuffer() = vixl::CodeBuffer(CodeBuffer, MaxInterpreterTrampolineSize);
+  FEXCore::ARMEmitter::Emitter emit{CodeBuffer, MaxInterpreterTrampolineSize};
+  ARMEmitter::ForwardLabel InlineIRData;
 
-  vixl::CodeBufferCheckScope scope(&emit, MaxInterpreterTrampolineSize, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
+  emit.mov(ARMEmitter::XReg::x0, STATE);
+  emit.adr(ARMEmitter::Reg::r1, &InlineIRData);
 
-  aarch64::Label InlineIRData;
+  emit.ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.Interpreter.FragmentExecuter));
+  emit.blr(ARMEmitter::Reg::r3);
 
-  emit.mov(x0, STATE);
-  emit.adr(x1, &InlineIRData);
+  emit.ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+  emit.br(ARMEmitter::Reg::r0);
 
-  emit.ldr(x3, STATE_PTR(CpuStateFrame, Pointers.Interpreter.FragmentExecuter));
-  emit.blr(x3);
+  emit.Bind(&InlineIRData);
 
-  emit.ldr(x0, STATE_PTR(CpuStateFrame, Pointers.Common.DispatcherLoopTop));
-  emit.br(x0);
-
-  emit.bind(&InlineIRData);
-
-  emit.FinalizeCode();
-
-  auto UsedBytes = emit.GetBuffer()->GetCursorOffset();
-  vixl::aarch64::CPU::EnsureIAndDCacheCoherency(CodeBuffer, UsedBytes);
+  auto UsedBytes = emit.GetCursorOffset();
+  emit.ClearICache(CodeBuffer, UsedBytes);
   return UsedBytes;
 }
 
 void Arm64Dispatcher::SpillSRA(FEXCore::Core::InternalThreadState *Thread, void *ucontext, uint32_t IgnoreMask) {
   for (size_t i = 0; i < SRA64.size(); i++) {
-    if (IgnoreMask & (1U << SRA64[i].GetCode())) {
+    if (IgnoreMask & (1U << SRA64[i].Idx())) {
       // Skip this one, it's already spilled
       continue;
     }
-    Thread->CurrentFrame->State.gregs[i] = ArchHelpers::Context::GetArmReg(ucontext, SRA64[i].GetCode());
+    Thread->CurrentFrame->State.gregs[i] = ArchHelpers::Context::GetArmReg(ucontext, SRA64[i].Idx());
   }
 
   if (EmitterCTX->HostFeatures.SupportsAVX) {
     for (size_t i = 0; i < SRAFPR.size(); i++) {
-      auto FPR = ArchHelpers::Context::GetArmFPR(ucontext, SRAFPR[i].GetCode());
+      auto FPR = ArchHelpers::Context::GetArmFPR(ucontext, SRAFPR[i].Idx());
       memcpy(&Thread->CurrentFrame->State.xmm.avx.data[i][0], &FPR, sizeof(__uint128_t));
     }
   } else {
     for (size_t i = 0; i < SRAFPR.size(); i++) {
-      auto FPR = ArchHelpers::Context::GetArmFPR(ucontext, SRAFPR[i].GetCode());
+      auto FPR = ArchHelpers::Context::GetArmFPR(ucontext, SRAFPR[i].Idx());
       memcpy(&Thread->CurrentFrame->State.xmm.sse.data[i][0], &FPR, sizeof(__uint128_t));
     }
   }

External/FEXCore/Source/Interface/Core/Dispatcher/Arm64Dispatcher.h

@@ -15,6 +15,9 @@ namespace FEXCore::Core {
 struct InternalThreadState;
 }
 
+#define STATE_PTR(STATE_TYPE, FIELD) \
+  STATE.R(), offsetof(FEXCore::Core::STATE_TYPE, FIELD)
+
 namespace FEXCore::CPU {
 
 class Arm64Dispatcher final : public Dispatcher, public Arm64Emitter {
@@ -29,6 +32,8 @@ class Arm64Dispatcher final : public Dispatcher, public Arm64Emitter {
   void ExecuteJITCallback(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP) override;
 #endif
 
+  void EmitDispatcher();
+
   protected:
     void SpillSRA(FEXCore::Core::InternalThreadState *Thread, void *ucontext, uint32_t IgnoreMask) override;
 

External/FEXCore/Source/Interface/Core/Dispatcher/Dispatcher.cpp

@@ -212,12 +212,20 @@ void Dispatcher::RestoreThreadState(FEXCore::Core::InternalThreadState *Thread,
         Frame->State.flags[9] = 1;
 
         Frame->State.rip = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EIP];
-        Frame->State.cs = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_CS];
-        Frame->State.ds = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS];
-        Frame->State.es = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ES];
-        Frame->State.fs = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_FS];
-        Frame->State.gs = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_GS];
-        Frame->State.ss = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_SS];
+        Frame->State.cs_idx = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_CS];
+        Frame->State.ds_idx = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS];
+        Frame->State.es_idx = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ES];
+        Frame->State.fs_idx = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_FS];
+        Frame->State.gs_idx = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_GS];
+        Frame->State.ss_idx = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_SS];
+
+        Frame->State.cs_cached = Frame->State.gdt[Frame->State.cs_idx >> 3].base;
+        Frame->State.ds_cached = Frame->State.gdt[Frame->State.ds_idx >> 3].base;
+        Frame->State.es_cached = Frame->State.gdt[Frame->State.es_idx >> 3].base;
+        Frame->State.fs_cached = Frame->State.gdt[Frame->State.fs_idx >> 3].base;
+        Frame->State.gs_cached = Frame->State.gdt[Frame->State.gs_idx >> 3].base;
+        Frame->State.ss_cached = Frame->State.gdt[Frame->State.ss_idx >> 3].base;
+
 #define COPY_REG(x) \
       Frame->State.gregs[X86State::REG_##x] = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_##x];
         COPY_REG(RDI);
@@ -280,15 +288,14 @@ static uint32_t ConvertSignalToTrapNo(int Signal, siginfo_t *HostSigInfo) {
   return Signal;
 }
 
-static uint32_t ConvertSignalToError(int Signal, siginfo_t *HostSigInfo) {
+static uint32_t ConvertSignalToError(void *ucontext, int Signal, siginfo_t *HostSigInfo) {
   switch (Signal) {
     case SIGSEGV:
       if (HostSigInfo->si_code == SEGV_MAPERR ||
           HostSigInfo->si_code == SEGV_ACCERR) {
         // Protection fault
         // Always a user fault for us
-        // XXX: PF_PROT and PF_WRITE
-        return X86State::X86_PF_USER;
+        return ArchHelpers::Context::GetProtectFlags(ucontext);
       }
       break;
   }
@@ -469,7 +476,7 @@ bool Dispatcher::HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, i
       }
       else {
         guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_TRAPNO] = ConvertSignalToTrapNo(Signal, HostSigInfo);
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_ERR] = ConvertSignalToError(Signal, HostSigInfo);
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_ERR] = ConvertSignalToError(ucontext, Signal, HostSigInfo);
       }
       guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_OLDMASK] = 0;
       guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_CR2] = 0;
@@ -565,10 +572,13 @@ bool Dispatcher::HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, i
       auto *xstate = reinterpret_cast<x86::xstate*>(FPStateLocation);
       SetXStateInfo(xstate, IsAVXEnabled);
 
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_GS] = Frame->State.gs;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_FS] = Frame->State.fs;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ES] = Frame->State.es;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS] = Frame->State.ds;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_CS] = Frame->State.cs_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS] = Frame->State.ds_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ES] = Frame->State.es_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_FS] = Frame->State.fs_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_GS] = Frame->State.gs_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_SS] = Frame->State.ss_idx;
+
       if (ContextBackup->FaultToTopAndGeneratedException) {
         guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = Frame->SynchronousFaultData.TrapNo;
         guest_siginfo->si_code = Frame->SynchronousFaultData.si_code;
@@ -578,13 +588,11 @@ bool Dispatcher::HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, i
       else {
         guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = ConvertSignalToTrapNo(Signal, HostSigInfo);
         guest_siginfo->si_code = HostSigInfo->si_code;
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ERR] = ConvertSignalToError(Signal, HostSigInfo);
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ERR] = ConvertSignalToError(ucontext, Signal, HostSigInfo);
       }
       guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EIP] = Frame->State.rip;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_CS] = Frame->State.cs;
       guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EFL] = 0;
       guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_UESP] = 0;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_SS] = Frame->State.ss;
 
 #define COPY_REG(x) \
       guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_##x] = Frame->State.gregs[X86State::REG_##x];

External/FEXCore/Source/Interface/Core/Frontend.cpp

@@ -18,6 +18,7 @@ $end_info$
 #include <FEXCore/HLE/SyscallHandler.h>
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 #include <FEXCore/Utils/Telemetry.h>
 #include <FEXHeaderUtils/TypeDefines.h>
 #include <set>
@@ -450,8 +451,17 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
       DestSize = 2;
     }
     else if (DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_128BIT) {
-      DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_128BIT);
-      DestSize = 16;
+      if (Options.L) {
+        DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_256BIT);
+        DestSize = 32;
+      } else {
+        DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_128BIT);
+        DestSize = 16;
+      }
+    }
+    else if (DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_256BIT) {
+      DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_256BIT);
+      DestSize = 32;
     }
     else if (HasNarrowingDisplacement &&
       (DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_DEF ||
@@ -482,7 +492,14 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
       DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_16BIT);
     }
     else if (SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_128BIT) {
-      DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_128BIT);
+      if (Options.L) {
+        DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_256BIT);
+      } else {
+        DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_128BIT);
+      }
+    }
+    else if (SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_256BIT) {
+      DecodeInst->Flags |= DecodeFlags::GenSizeSrcSize(DecodeFlags::SIZE_256BIT);
     }
     else if (HasNarrowingDisplacement &&
       (SrcSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_DEF ||
@@ -776,6 +793,7 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
     if (Op == 0xC5) { // Two byte VEX
       pp = Byte1 & 0b11;
       options.vvvv = 15 - ((Byte1 & 0b01111000) >> 3);
+      options.L = (Byte1 & 0b100) != 0;
     }
     else { // 0xC4 = Three byte VEX
       const uint8_t Byte2 = ReadByte();
@@ -783,6 +801,7 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
       map_select = Byte1 & 0b11111;
       options.vvvv = 15 - ((Byte2 & 0b01111000) >> 3);
       options.w = (Byte2 & 0b10000000) != 0;
+      options.L = (Byte2 & 0b100) != 0;
       if ((Byte1 & 0b01000000) == 0) {
         LOGMAN_THROW_A_FMT(CTX->Config.Is64BitMode, "VEX.X shouldn't be 0 in 32-bit mode!");
         DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_X;
@@ -1112,6 +1131,35 @@ void Decoder::BranchTargetInMultiblockRange() {
   }
 }
 
+bool Decoder::BranchTargetCanContinue(bool FinalInstruction) const {
+  if (FinalInstruction) {
+    return false;
+  }
+
+  uint64_t TargetRIP = 0;
+  const uint8_t GPRSize = CTX->GetGPRSize();
+
+  if (DecodeInst->OP == 0xE8) { // Call - immediate target
+    const uint64_t NextRIP = DecodeInst->PC + DecodeInst->InstSize;
+    LOGMAN_THROW_A_FMT(DecodeInst->Src[0].IsLiteral(), "Had wrong operand type");
+    TargetRIP = DecodeInst->PC + DecodeInst->InstSize + DecodeInst->Src[0].Data.Literal.Value;
+
+    if (GPRSize == 4) {
+      // If we are running a 32bit guest then wrap around addresses that go above 32bit
+      TargetRIP &= 0xFFFFFFFFU;
+    }
+
+    if (TargetRIP == NextRIP) {
+      // Optimize the case that the instruction is jumping just after itself.
+      // This is a GOT calculation which we can optimize out.
+      // Optimization occurs inside of the OpDispatcher implementation
+      return true;
+    }
+  }
+
+  return false;
+}
+
 const uint8_t *Decoder::AdjustAddrForSpecialRegion(uint8_t const* _InstStream, uint64_t EntryPoint, uint64_t RIP) {
   constexpr uint64_t VSyscall_Base = 0xFFFF'FFFF'FF60'0000ULL;
   constexpr uint64_t VSyscall_End = VSyscall_Base + 0x1000;
@@ -1132,6 +1180,7 @@ const uint8_t *Decoder::AdjustAddrForSpecialRegion(uint8_t const* _InstStream, u
 }
 
 void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC, std::function<void(uint64_t BlockEntry, uint64_t Start, uint64_t Length)> AddContainedCodePage) {
+  FEXCORE_PROFILE_SCOPED("DecodeInstructions");
   Blocks.clear();
   BlocksToDecode.clear();
   HasBlocks.clear();
@@ -1166,7 +1215,6 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC,
   std::set<uint64_t> CodePages = { CurrentCodePage };
 
   AddContainedCodePage(PC, CurrentCodePage, FHU::FEX_PAGE_SIZE);
-  
 
   while (!BlocksToDecode.empty()) {
     auto BlockDecodeIt = BlocksToDecode.begin();
@@ -1236,23 +1284,21 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC,
         CanContinue = true;
       }
 
+      bool FinalInstruction = DecodedSize >= CTX->Config.MaxInstPerBlock ||
+          DecodedSize >= DefaultDecodedBufferSize ||
+          TotalInstructions >= CTX->Config.MaxInstPerBlock;
+
       if (DecodeInst->TableInfo->Flags & FEXCore::X86Tables::InstFlags::FLAGS_SETS_RIP) {
         // If we have multiblock enabled
         // If the branch target is within our multiblock range then we can keep going on
         // We don't want to short circuit this since we want to calculate our ranges still
         BranchTargetInMultiblockRange();
+
+        // Bypass branches if we can continue through them in some cases.
+        CanContinue |= BranchTargetCanContinue(FinalInstruction);
       }
 
-      if (!CanContinue) {
-        break;
-      }
-
-      if (DecodedSize >= CTX->Config.MaxInstPerBlock ||
-          DecodedSize >= DefaultDecodedBufferSize) {
-        break;
-      }
-
-      if (TotalInstructions >= CTX->Config.MaxInstPerBlock) {
+      if (FinalInstruction || !CanContinue) {
         break;
       }
 

External/FEXCore/Source/Interface/Core/Frontend.h

@@ -49,6 +49,7 @@ private:
   struct DecodedHeader {
     uint8_t vvvv; // Encoded operand in a VEX prefix.
     bool w;       // VEX.W bit.
+    bool L;       // VEX.L bit (if set then 256 bit operation, if unset then scalar or 128-bit operation)
   };
 
   FEXCore::Context::Context *CTX;
@@ -57,6 +58,7 @@ private:
   bool DecodeInstruction(uint64_t PC);
 
   void BranchTargetInMultiblockRange();
+  bool BranchTargetCanContinue(bool FinalInstruction) const;
 
   uint8_t ReadByte();
   uint8_t PeekByte(uint8_t Offset) const;

External/FEXCore/Source/Interface/Core/HostFeatures.cpp

@@ -65,6 +65,7 @@ HostFeatures::HostFeatures() {
   SupportsFlushInputsToZero = Features.Has(vixl::CPUFeatures::Feature::kAFP);
   SupportsRCPC = Features.Has(vixl::CPUFeatures::Feature::kRCpc);
   SupportsTSOImm9 = Features.Has(vixl::CPUFeatures::Feature::kRCpcImm);
+  SupportsPMULL_128Bit = Features.Has(vixl::CPUFeatures::Feature::kPmull1Q);
 
   Supports3DNow = true;
   SupportsSSE4A = true;
@@ -127,6 +128,7 @@ HostFeatures::HostFeatures() {
   SupportsSHA = Features.has(Xbyak::util::Cpu::tSHA);
   SupportsBMI1 = Features.has(Xbyak::util::Cpu::tBMI1);
   SupportsBMI2 = Features.has(Xbyak::util::Cpu::tBMI2);
+  SupportsPMULL_128Bit = Features.has(Xbyak::util::Cpu::tPCLMULQDQ);
 
   // xbyak doesn't know how to check for CLZero
   uint32_t eax, ebx, ecx, edx;

External/FEXCore/Source/Interface/Core/Interpreter/ALUOps.cpp

@@ -894,33 +894,51 @@ DEF_OP(Select) {
 }
 
 DEF_OP(VExtractToGPR) {
-  auto Op = IROp->C<IR::IROp_VExtractToGPR>();
+  const auto Op = IROp->C<IR::IROp_VExtractToGPR>();
+  const auto OpSize = IROp->Size;
+
+  constexpr auto AVXRegSize = Core::CPUState::XMM_AVX_REG_SIZE;
+  constexpr auto SSERegSize = Core::CPUState::XMM_SSE_REG_SIZE;
+  constexpr auto SSEBitSize = SSERegSize * 8;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto ElementSizeBits = ElementSize * 8;
+  const auto Shift = ElementSizeBits * Op->Index;
 
   const uint32_t SourceSize = GetOpSize(Data->CurrentIR, Op->Vector);
 
-  LOGMAN_THROW_AA_FMT(IROp->Size <= 16, "OpSize is too large for VExtractToGPR: {}", IROp->Size);
+  LOGMAN_THROW_AA_FMT(OpSize <= AVXRegSize,
+                      "OpSize is too large for VExtractToGPR: {}", OpSize);
 
-  if (SourceSize == 16) {
-    __uint128_t SourceMask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-    uint64_t Shift = Op->Header.ElementSize * Op->Index * 8;
-    if (Op->Header.ElementSize == 8)
+  if (SourceSize >= SSERegSize) {
+    __uint128_t SourceMask = (1ULL << ElementSizeBits) - 1;
+    if (ElementSize == 8) {
       SourceMask = ~0ULL;
+    }
 
-    __uint128_t Src = *GetSrc<__uint128_t*>(Data->SSAData, Op->Vector);
-    Src >>= Shift;
-    Src &= SourceMask;
-    memcpy(GDP, &Src, Op->Header.ElementSize);
+    const auto Src = *GetSrc<InterpVector256*>(Data->SSAData, Op->Vector);
+
+    const auto GetResult = [&] {
+      if (Shift >= SSEBitSize) {
+        const auto NormalizedShift = Shift - SSEBitSize;
+        return (Src.Upper >> NormalizedShift) & SourceMask;
+      } else {
+        return (Src.Lower >> Shift) & SourceMask;
+      }
+    };
+
+    const auto Result = GetResult();
+    memcpy(GDP, &Result, ElementSize);
   }
   else {
-    uint64_t SourceMask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-    uint64_t Shift = Op->Header.ElementSize * Op->Index * 8;
-    if (Op->Header.ElementSize == 8)
+    uint64_t SourceMask = (1ULL << ElementSizeBits) - 1;
+    if (ElementSize == 8) {
       SourceMask = ~0ULL;
+    }
 
-    uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Vector);
-    Src >>= Shift;
-    Src &= SourceMask;
-    GD = Src;
+    const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Vector);
+    const uint64_t Result = (Src >> Shift) & SourceMask;
+    GD = Result;
   }
 }
 

External/FEXCore/Source/Interface/Core/Interpreter/ConversionOps.cpp

@@ -13,22 +13,46 @@ $end_info$
 namespace FEXCore::CPU {
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(VInsGPR) {
-  auto Op = IROp->C<IR::IROp_VInsGPR>();
-  const uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_VInsGPR>();
+  const auto OpSize = IROp->Size;
 
-  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->DestVector);
-  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Src);
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto ElementSizeBits = ElementSize * 8;
+  constexpr auto SSEBitSize = Core::CPUState::XMM_SSE_REG_SIZE * 8;
 
-  uint64_t Offset = Op->DestIdx * Op->Header.ElementSize * 8;
-  __uint128_t Mask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-  if (Op->Header.ElementSize == 8) {
+  const uint64_t Offset = Op->DestIdx * ElementSizeBits;
+  const auto InUpperLane = Offset >= SSEBitSize;
+
+  __uint128_t Mask = (1ULL << ElementSizeBits) - 1;
+  if (ElementSize == 8) {
     Mask = ~0ULL;
   }
-  Src2 = Src2 & Mask;
-  Mask <<= Offset;
+
+  const auto Src1 = *GetSrc<InterpVector256*>(Data->SSAData, Op->DestVector);
+  const auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Src);
+
+  const auto Scalar = Src2 & Mask;
+  const auto ScaledOffset = InUpperLane ? Offset - SSEBitSize
+                                        : Offset;
+
+  // Now shift into place and set all bits but
+  // the ones where we're going to insert our value.
+  Mask <<= ScaledOffset;
   Mask = ~Mask;
-  __uint128_t Dst = Src1 & Mask;
-  Dst |= Src2 << Offset;
+
+  const auto Dst = [&] {
+    if (InUpperLane) {
+      return InterpVector256{
+        .Lower = Src1.Lower,
+        .Upper = (Src1.Upper & Mask) | (Scalar << ScaledOffset),
+      };
+    } else {
+      return InterpVector256{
+        .Lower = (Src1.Lower & Mask) | (Scalar << ScaledOffset),
+        .Upper = Src1.Upper,
+      };
+    }
+  }();
 
   memcpy(GDP, &Dst, OpSize);
 }
@@ -89,63 +113,73 @@ DEF_OP(Vector_SToF) {
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
 
   const auto Func = [](auto a, auto min, auto max) { return a; };
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP(4, float, int32_t, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP(8, double, int64_t, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(Vector_FToZS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToZS>();
+  const auto Op = IROp->C<IR::IROp_Vector_FToZS>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
 
   const auto Func = [](auto a, auto min, auto max) { return std::trunc(a); };
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP(4, int32_t, float, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP(8, int64_t, double, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(Vector_FToS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToS>();
+  const auto Op = IROp->C<IR::IROp_Vector_FToS>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
 
   const auto Func = [](auto a, auto min, auto max) { return std::nearbyint(a); };
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP(4, int32_t, float, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP(8, int64_t, double, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(Vector_FToF) {
-  auto Op = IROp->C<IR::IROp_Vector_FToF>();
+  const auto Op = IROp->C<IR::IROp_Vector_FToF>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t ElementSize = Op->Header.ElementSize;
+  const uint16_t Conv = (ElementSize << 8) | Op->SrcElementSize;
 
   const auto Func = [](auto a, auto min, auto max) { return a; };
   switch (Conv) {
@@ -161,23 +195,26 @@ DEF_OP(Vector_FToF) {
       // Sometimes is used to convert from a 128bit vector register
       // in to a 64bit vector register with different sized elements
       // eg: %ssa5 i32v2 = Vector_FToF %ssa4 i128, #0x8
-      uint8_t Elements = (OpSize << 1) / Op->SrcElementSize;
+      uint8_t Elements = OpSize == 8 ? 2 : OpSize / Op->SrcElementSize;
       DO_VECTOR_1SRC_2TYPE_OP_NOSIZE(float, double, Func, 0, 0)
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unknown Conversion Type : 0x{:04x}", Conv); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Conversion Type : 0x{:04x}", Conv);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(Vector_FToI) {
-  auto Op = IROp->C<IR::IROp_Vector_FToI>();
+  const auto Op = IROp->C<IR::IROp_Vector_FToI>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
   const auto Func_Nearest = [](auto a) { return std::rint(a); };
   const auto Func_Neg = [](auto a) { return std::floor(a); };
   const auto Func_Pos = [](auto a) { return std::ceil(a); };
@@ -186,31 +223,31 @@ DEF_OP(Vector_FToI) {
 
   switch (Op->Round) {
     case FEXCore::IR::Round_Nearest.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
         DO_VECTOR_1SRC_OP(4, float, Func_Nearest)
         DO_VECTOR_1SRC_OP(8, double, Func_Nearest)
       }
     break;
     case FEXCore::IR::Round_Negative_Infinity.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
         DO_VECTOR_1SRC_OP(4, float, Func_Neg)
         DO_VECTOR_1SRC_OP(8, double, Func_Neg)
       }
     break;
     case FEXCore::IR::Round_Positive_Infinity.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
         DO_VECTOR_1SRC_OP(4, float, Func_Pos)
         DO_VECTOR_1SRC_OP(8, double, Func_Pos)
       }
     break;
     case FEXCore::IR::Round_Towards_Zero.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
         DO_VECTOR_1SRC_OP(4, float, Func_Trunc)
         DO_VECTOR_1SRC_OP(8, double, Func_Trunc)
       }
     break;
     case FEXCore::IR::Round_Host.Val:
-      switch (Op->Header.ElementSize) {
+      switch (ElementSize) {
         DO_VECTOR_1SRC_OP(4, float, Func_Host)
         DO_VECTOR_1SRC_OP(8, double, Func_Host)
       }

External/FEXCore/Source/Interface/Core/Interpreter/InterpreterFallbacks.cpp

@@ -146,7 +146,7 @@ void InterpreterOps::FillFallbackIndexPointers(uint64_t *Info) {
 
 }
 
-bool InterpreterOps::GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info) {
+bool InterpreterOps::GetFallbackHandler(IR::IROp_Header const *IROp, FallbackInfo *Info) {
   uint8_t OpSize = IROp->Size;
   switch(IROp->Op) {
     case IR::OP_F80LOADFCW: {

External/FEXCore/Source/Interface/Core/Interpreter/InterpreterOps.cpp

@@ -154,8 +154,6 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(STOREMEM,               StoreMem);
   REGISTER_OP(LOADMEMTSO,             LoadMem);
   REGISTER_OP(STOREMEMTSO,            StoreMem);
-  REGISTER_OP(VLOADMEMELEMENT,        VLoadMemElement);
-  REGISTER_OP(VSTOREMEMELEMENT,       VStoreMemElement);
   REGISTER_OP(CACHELINECLEAR,         CacheLineClear);
   REGISTER_OP(CACHELINEZERO,          CacheLineZero);
 
@@ -245,8 +243,6 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(VINSELEMENT,            VInsElement);
   REGISTER_OP(VDUPELEMENT,            VDupElement);
   REGISTER_OP(VEXTR,                  VExtr);
-  REGISTER_OP(VSLI,                   VSLI);
-  REGISTER_OP(VSRI,                   VSRI);
   REGISTER_OP(VUSHRI,                 VUShrI);
   REGISTER_OP(VSSHRI,                 VSShrI);
   REGISTER_OP(VSHLI,                  VShlI);

External/FEXCore/Source/Interface/Core/Interpreter/InterpreterOps.h

@@ -49,7 +49,7 @@ namespace FEXCore::CPU {
     public:
       static void InterpretIR(FEXCore::Core::CpuStateFrame *Frame, FEXCore::IR::IRListView const *IR);
       static void FillFallbackIndexPointers(uint64_t *Info);
-      static bool GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info);
+      static bool GetFallbackHandler(IR::IROp_Header const *IROp, FallbackInfo *Info);
 
       struct IROpData {
         FEXCore::Core::InternalThreadState *State{};
@@ -181,8 +181,6 @@ namespace FEXCore::CPU {
   DEF_OP(StoreFlag);
   DEF_OP(LoadMem);
   DEF_OP(StoreMem);
-  DEF_OP(VLoadMemElement);
-  DEF_OP(VStoreMemElement);
   DEF_OP(CacheLineClear);
   DEF_OP(CacheLineZero);
 
@@ -265,8 +263,6 @@ namespace FEXCore::CPU {
   DEF_OP(VInsElement);
   DEF_OP(VDupElement);
   DEF_OP(VExtr);
-  DEF_OP(VSLI);
-  DEF_OP(VSRI);
   DEF_OP(VUShrI);
   DEF_OP(VSShrI);
   DEF_OP(VShlI);

External/FEXCore/Source/Interface/Core/Interpreter/MemoryOps.cpp

@@ -25,93 +25,139 @@ static inline void CacheLineFlush(char *Addr) {
 
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(LoadContext) {
-  auto Op = IROp->C<IR::IROp_LoadContext>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_LoadContext>();
+  const auto OpSize = IROp->Size;
+
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Src = ContextPtr + Op->Offset;
 
-  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
-  ContextPtr += Op->Offset;
   #define LOAD_CTX(x, y) \
     case x: { \
-      y const *MemData = reinterpret_cast<y const*>(ContextPtr); \
+      y const *MemData = reinterpret_cast<y const*>(Src); \
       GD = *MemData; \
       break; \
     }
+
   switch (OpSize) {
     LOAD_CTX(1, uint8_t)
     LOAD_CTX(2, uint16_t)
     LOAD_CTX(4, uint32_t)
     LOAD_CTX(8, uint64_t)
-    case 16: {
-      void const *MemData = reinterpret_cast<void const*>(ContextPtr);
+    case 16:
+    case 32: {
+      void const *MemData = reinterpret_cast<void const*>(Src);
       memcpy(GDP, MemData, OpSize);
       break;
     }
-    default:  LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+      break;
   }
   #undef LOAD_CTX
 }
 
 DEF_OP(StoreContext) {
-  auto Op = IROp->C<IR::IROp_StoreContext>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_StoreContext>();
+  const auto OpSize = IROp->Size;
 
-  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
-  ContextPtr += Op->Offset;
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Dst = ContextPtr + Op->Offset;
 
-  void *MemData = reinterpret_cast<void*>(ContextPtr);
+  void *MemData = reinterpret_cast<void*>(Dst);
   void *Src = GetSrc<void*>(Data->SSAData, Op->Value);
   memcpy(MemData, Src, OpSize);
 }
 
 DEF_OP(LoadRegister) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
+  const auto Op = IROp->C<IR::IROp_LoadRegister>();
+  const auto OpSize = IROp->Size;
 
-DEF_OP(StoreRegister) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
-
-DEF_OP(LoadContextIndexed) {
-  auto Op = IROp->C<IR::IROp_LoadContextIndexed>();
-  uint64_t Index = *GetSrc<uint64_t*>(Data->SSAData, Op->Index);
-
-  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
-
-  ContextPtr += Op->BaseOffset;
-  ContextPtr += Index * Op->Stride;
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Src = ContextPtr + Op->Offset;
 
   #define LOAD_CTX(x, y) \
     case x: { \
-      y const *MemData = reinterpret_cast<y const*>(ContextPtr); \
+      y const *MemData = reinterpret_cast<y const*>(Src); \
       GD = *MemData; \
       break; \
     }
-  switch (IROp->Size) {
+
+  switch (OpSize) {
     LOAD_CTX(1, uint8_t)
     LOAD_CTX(2, uint16_t)
     LOAD_CTX(4, uint32_t)
     LOAD_CTX(8, uint64_t)
-    case 16: {
-      void const *MemData = reinterpret_cast<void const*>(ContextPtr);
-      memcpy(GDP, MemData, IROp->Size);
+    case 16:
+    case 32: {
+      void const *MemData = reinterpret_cast<void const*>(Src);
+      memcpy(GDP, MemData, OpSize);
       break;
     }
-    default:  LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", IROp->Size);
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+      break;
+  }
+  #undef LOAD_CTX
+}
+
+DEF_OP(StoreRegister) {
+  const auto Op = IROp->C<IR::IROp_StoreRegister>();
+  const auto OpSize = IROp->Size;
+
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Dst = ContextPtr + Op->Offset;
+
+  void *MemData = reinterpret_cast<void*>(Dst);
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Value);
+  memcpy(MemData, Src, OpSize);
+}
+
+DEF_OP(LoadContextIndexed) {
+  const auto Op = IROp->C<IR::IROp_LoadContextIndexed>();
+  const auto OpSize = IROp->Size;
+
+  const auto Index = *GetSrc<uint64_t*>(Data->SSAData, Op->Index);
+
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Src = ContextPtr + Op->BaseOffset + (Index * Op->Stride);
+
+  #define LOAD_CTX(x, y) \
+    case x: { \
+      y const *MemData = reinterpret_cast<y const*>(Src); \
+      GD = *MemData; \
+      break; \
+    }
+
+  switch (OpSize) {
+    LOAD_CTX(1, uint8_t)
+    LOAD_CTX(2, uint16_t)
+    LOAD_CTX(4, uint32_t)
+    LOAD_CTX(8, uint64_t)
+    case 16:
+    case 32: {
+      void const *MemData = reinterpret_cast<void const*>(Src);
+      memcpy(GDP, MemData, OpSize);
+      break;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", OpSize);
+      break;
   }
   #undef LOAD_CTX
 }
 
 DEF_OP(StoreContextIndexed) {
-  auto Op = IROp->C<IR::IROp_StoreContextIndexed>();
-  uint64_t Index = *GetSrc<uint64_t*>(Data->SSAData, Op->Index);
+  const auto Op = IROp->C<IR::IROp_StoreContextIndexed>();
+  const auto OpSize = IROp->Size;
 
-  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
-  ContextPtr += Op->BaseOffset;
-  ContextPtr += Index * Op->Stride;
+  const auto Index = *GetSrc<uint64_t*>(Data->SSAData, Op->Index);
 
-  void *MemData = reinterpret_cast<void*>(ContextPtr);
+  const auto ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  const auto Dst = ContextPtr + Op->BaseOffset + (Index * Op->Stride);
+
+  void *MemData = reinterpret_cast<void*>(Dst);
   void *Src = GetSrc<void*>(Data->SSAData, Op->Value);
-  memcpy(MemData, Src, IROp->Size);
+  memcpy(MemData, Src, OpSize);
 }
 
 DEF_OP(SpillRegister) {
@@ -144,8 +190,8 @@ DEF_OP(StoreFlag) {
 }
 
 DEF_OP(LoadMem) {
-  auto Op = IROp->C<IR::IROp_LoadMem>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_LoadMem>();
+  const auto OpSize = IROp->Size;
 
   uint8_t const *MemData = *GetSrc<uint8_t const**>(Data->SSAData, Op->Addr);
 
@@ -158,7 +204,8 @@ DEF_OP(LoadMem) {
       case IR::MEM_OFFSET_SXTW.Val: MemData += (int32_t)Offset; break;
     }
   }
-  memset(GDP, 0, 16);
+
+  memset(GDP, 0, Core::CPUState::XMM_AVX_REG_SIZE);
   switch (OpSize) {
     case 1: {
       auto D = reinterpret_cast<const std::atomic<uint8_t>*>(MemData);
@@ -180,16 +227,15 @@ DEF_OP(LoadMem) {
       GD = D->load();
       break;
     }
-
     default:
-      memcpy(GDP, MemData, IROp->Size);
+      memcpy(GDP, MemData, OpSize);
       break;
   }
 }
 
 DEF_OP(StoreMem) {
-  auto Op = IROp->C<IR::IROp_StoreMem>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_StoreMem>();
+  const auto OpSize = IROp->Size;
 
   uint8_t *MemData = *GetSrc<uint8_t **>(Data->SSAData, Op->Addr);
 
@@ -221,41 +267,11 @@ DEF_OP(StoreMem) {
     }
 
     default:
-      memcpy(MemData, GetSrc<void*>(Data->SSAData, Op->Value), IROp->Size);
+      memcpy(MemData, GetSrc<void*>(Data->SSAData, Op->Value), OpSize);
       break;
   }
 }
 
-DEF_OP(VLoadMemElement) {
-  auto Op = IROp->C<IR::IROp_VLoadMemElement>();
-  void const *MemData = *GetSrc<void const**>(Data->SSAData, Op->Value);
-
-  memcpy(GDP, GetSrc<void*>(Data->SSAData, Op->Addr), 16);
-  memcpy(reinterpret_cast<void*>(reinterpret_cast<uintptr_t>(GDP) + (Op->Header.ElementSize * Op->Index)),
-    MemData, Op->Header.ElementSize);
-}
-
-DEF_OP(VStoreMemElement) {
-  #define STORE_DATA(x, y) \
-    case x: { \
-      y *MemData = *GetSrc<y**>(Data->SSAData, Op->Value); \
-      memcpy(MemData, &GetSrc<y*>(Data->SSAData, Op->Addr)[Op->Index], sizeof(y)); \
-      break; \
-    }
-
-  auto Op = IROp->C<IR::IROp_VStoreMemElement>();
-  uint8_t OpSize = IROp->Size;
-
-  switch (OpSize) {
-    STORE_DATA(1, uint8_t)
-    STORE_DATA(2, uint16_t)
-    STORE_DATA(4, uint32_t)
-    STORE_DATA(8, uint64_t)
-    default: LOGMAN_MSG_A_FMT("Unhandled StoreMem size"); break;
-  }
-  #undef STORE_DATA
-}
-
 DEF_OP(CacheLineClear) {
   auto Op = IROp->C<IR::IROp_CacheLineClear>();
 

External/FEXCore/Source/Interface/Core/Interpreter/MiscOps.cpp

@@ -19,13 +19,6 @@ $end_info$
 #include <sys/random.h>
 
 namespace FEXCore::CPU {
-[[noreturn]]
-static void StopThread(FEXCore::Core::InternalThreadState *Thread) {
-  Thread->CTX->StopThread(Thread);
-
-  LOGMAN_MSG_A_FMT("unreachable");
-  FEX_UNREACHABLE;
-}
 
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(Fence) {

External/FEXCore/Source/Interface/Core/Interpreter/VectorOps.cpp

@@ -11,6 +11,7 @@ $end_info$
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Utils/BitUtils.h>
 
+#include <array>
 #include <bit>
 #include <cstdint>
 #include <limits>
@@ -358,23 +359,26 @@ DEF_OP(VAddP) {
 }
 
 DEF_OP(VAddV) {
-  auto Op = IROp->C<IR::IROp_VAddV>();
-  const uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_VAddV>();
+  const auto OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
 
   const auto Func = [](auto current, auto a) { return current + a; };
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_REDUCE_1SRC_OP(1, int8_t, Func, 0)
     DO_VECTOR_REDUCE_1SRC_OP(2, int16_t, Func, 0)
     DO_VECTOR_REDUCE_1SRC_OP(4, int32_t, Func, 0)
     DO_VECTOR_REDUCE_1SRC_OP(8, int64_t, Func, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      return;
   }
-  memcpy(GDP, Tmp, Op->Header.ElementSize);
+  memcpy(GDP, Tmp, ElementSize);
 }
 
 DEF_OP(VUMinV) {
@@ -838,17 +842,18 @@ DEF_OP(VSMax) {
 }
 
 DEF_OP(VZip) {
-  auto Op = IROp->C<IR::IROp_VZip>();
+  const auto Op = IROp->C<IR::IROp_VZip>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->VectorLower);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->VectorUpper);
-  uint8_t Tmp[16];
-  uint8_t Elements = OpSize / Op->Header.ElementSize;
-  uint8_t BaseOffset = IROp->Op == IR::OP_VZIP2 ? (Elements / 2) : 0;
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  uint8_t Elements = OpSize / ElementSize;
+  const uint8_t BaseOffset = IROp->Op == IR::OP_VZIP2 ? (Elements / 2) : 0;
   Elements >>= 1;
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     case 1: {
       auto *Dst_d  = reinterpret_cast<uint8_t*>(Tmp);
       auto *Src1_d = reinterpret_cast<uint8_t*>(Src1);
@@ -889,24 +894,27 @@ DEF_OP(VZip) {
       }
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
 
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VUnZip) {
-  auto Op = IROp->C<IR::IROp_VUnZip>();
+  const auto Op = IROp->C<IR::IROp_VUnZip>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->VectorLower);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->VectorUpper);
-  uint8_t Tmp[16];
-  uint8_t Elements = OpSize / Op->Header.ElementSize;
-  unsigned Start = IROp->Op == IR::OP_VUNZIP ? 0 : 1;
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  uint8_t Elements = OpSize / ElementSize;
+  const unsigned Start = IROp->Op == IR::OP_VUNZIP ? 0 : 1;
   Elements >>= 1;
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     case 1: {
       auto *Dst_d  = reinterpret_cast<uint8_t*>(Tmp);
       auto *Src1_d = reinterpret_cast<uint8_t*>(Src1);
@@ -947,7 +955,9 @@ DEF_OP(VUnZip) {
       }
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
 
   memcpy(GDP, Tmp, OpSize);
@@ -1503,16 +1513,18 @@ DEF_OP(VSShrS) {
 }
 
 DEF_OP(VInsElement) {
-  auto Op = IROp->C<IR::IROp_VInsElement>();
-  const uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_VInsElement>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->DestVector);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->SrcVector);
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
   // Copy src1 in to dest
   memcpy(Tmp, Src1, OpSize);
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     case 1: {
       auto *Dst_d  = reinterpret_cast<uint8_t*>(Tmp);
       auto *Src2_d = reinterpret_cast<uint8_t*>(Src2);
@@ -1537,83 +1549,133 @@ DEF_OP(VInsElement) {
       Dst_d[Op->DestIdx] = Src2_d[Op->SrcIdx];
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    case 16: {
+      auto *Dst_d  = reinterpret_cast<__uint128_t*>(Tmp);
+      auto *Src2_d = reinterpret_cast<__uint128_t*>(Src2);
+      Dst_d[Op->DestIdx] = Src2_d[Op->SrcIdx];
+      break;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   };
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VDupElement) {
-  auto Op = IROp->C<IR::IROp_VDupElement>();
+  const auto Op = IROp->C<IR::IROp_VDupElement>();
   const uint8_t OpSize = IROp->Size;
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
 
-  LOGMAN_THROW_AA_FMT(OpSize <= 16, "OpSize is too large for VDupElement: {}", OpSize);
-  if (OpSize == 16) {
-    __uint128_t SourceMask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-    uint64_t Shift = Op->Header.ElementSize * Op->Index * 8;
-    if (Op->Header.ElementSize == 8)
+  const uint64_t ElementSize = Op->Header.ElementSize;
+  const uint64_t ElementSizeBits = ElementSize * 8;
+  const uint8_t Elements = OpSize / ElementSize;
+
+  constexpr auto AVXRegSize = Core::CPUState::XMM_AVX_REG_SIZE;
+  constexpr auto SSERegSize = Core::CPUState::XMM_SSE_REG_SIZE;
+  constexpr auto SSEBitSize = SSERegSize * 8;
+
+  const auto Is128BitElement = ElementSizeBits == SSEBitSize;
+  const auto Is256Bit = OpSize == AVXRegSize;
+
+  LOGMAN_THROW_AA_FMT(OpSize <= AVXRegSize,
+                      "OpSize is too large for VDupElement: {}", OpSize);
+
+  if (OpSize >= SSERegSize) {
+    __uint128_t SourceMask = (1ULL << ElementSizeBits) - 1;
+    if (ElementSize == 8) {
       SourceMask = ~0ULL;
-
-    __uint128_t Src = *GetSrc<__uint128_t*>(Data->SSAData, Op->Vector);
-    Src >>= Shift;
-    Src &= SourceMask;
-    for (size_t i = 0; i < Elements; ++i) {
-      memcpy(reinterpret_cast<void*>(reinterpret_cast<uintptr_t>(GDP) + (Op->Header.ElementSize * i)),
-        &Src, Op->Header.ElementSize);
     }
-  }
-  else {
-    uint64_t SourceMask = (1ULL << (Op->Header.ElementSize * 8)) - 1;
-    uint64_t Shift = Op->Header.ElementSize * Op->Index * 8;
-    if (Op->Header.ElementSize == 8)
-      SourceMask = ~0ULL;
 
-    uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Vector);
-    Src >>= Shift;
-    Src &= SourceMask;
+    const auto GetResult = [&]() -> __uint128_t {
+      const auto Src = *GetSrc<InterpVector256*>(Data->SSAData, Op->Vector);
+      uint64_t Shift = ElementSizeBits * Op->Index;
+
+      if (Is128BitElement) {
+        if (Shift == 0) {
+          return Src.Lower;
+        } else {
+          return Src.Upper;
+        }
+      } else {
+        // Normalize shift to act on upper uint128_t
+        if (Is256Bit && Shift >= SSEBitSize) {
+          Shift -= SSEBitSize;
+          return (Src.Upper >> Shift) & SourceMask;
+        } else {
+          return (Src.Lower >> Shift) & SourceMask;
+        }
+      }
+    };
+
+    const __uint128_t Result = GetResult();
     for (size_t i = 0; i < Elements; ++i) {
-      memcpy(reinterpret_cast<void*>(reinterpret_cast<uintptr_t>(GDP) + (Op->Header.ElementSize * i)),
-        &Src, Op->Header.ElementSize);
+      auto* Dst = static_cast<uint8_t*>(GDP) + (ElementSize * i);
+      memcpy(Dst, &Result, ElementSize);
+    }
+  } else {
+    const uint64_t Shift = ElementSizeBits * Op->Index;
+    uint64_t SourceMask = (1ULL << ElementSizeBits) - 1;
+    if (ElementSize == 8) {
+      SourceMask = ~0ULL;
+    }
+
+    const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Vector);
+    const uint64_t Result = (Src >> Shift) & SourceMask;
+    for (size_t i = 0; i < Elements; ++i) {
+      auto* Dst = static_cast<uint8_t*>(GDP) + (ElementSize * i);
+      memcpy(Dst, &Result, ElementSize);
     }
   }
 }
 
 DEF_OP(VExtr) {
-  auto Op = IROp->C<IR::IROp_VExtr>();
-  const uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_VExtr>();
+  const auto OpSize = IROp->Size;
+  const auto OpSizeBits = OpSize * 8;
 
-  const auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->VectorLower);
-  const auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->VectorUpper);
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Index = Op->Index;
 
-  uint64_t Offset = Op->Index * Op->Header.ElementSize * 8;
-  __uint128_t Dst{};
-  if (Offset >= (OpSize * 8)) {
-    Offset -= OpSize * 8;
-    Dst = Src1 >> Offset;
+  if (Is256Bit) {
+    const auto ByteIndex = Index * ElementSize;
+    const auto IsUpperVectorZero = ByteIndex >= OpSize;
+    const auto SanitizedByteIndex = IsUpperVectorZero ? ByteIndex - OpSize
+                                                      : ByteIndex;
+
+    const auto Vectors = IsUpperVectorZero
+      ?
+        std::array<InterpVector256, 2>{
+          *GetSrc<InterpVector256*>(Data->SSAData, Op->VectorLower),
+          InterpVector256{},
+        }
+      :
+        std::array<InterpVector256, 2>{
+          *GetSrc<InterpVector256*>(Data->SSAData, Op->VectorUpper),
+          *GetSrc<InterpVector256*>(Data->SSAData, Op->VectorLower),
+        };
+
+    const auto* VectorsPtr = reinterpret_cast<const uint8_t*>(Vectors.data());
+    const auto* SrcPtr = VectorsPtr + SanitizedByteIndex;
+    const auto CopyAmount = std::max(0, int(sizeof(Vectors) - SanitizedByteIndex));
+
+    memcpy(GDP, SrcPtr, CopyAmount);
+  } else {
+    uint64_t Offset = Index * ElementSize * 8;
+
+    const auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->VectorLower);
+    const auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->VectorUpper);
+
+    __uint128_t Dst{};
+    if (Offset >= OpSizeBits) {
+      Offset -= OpSizeBits;
+      Dst = Src1 >> Offset;
+    } else {
+      Dst = (Src1 << (OpSizeBits - Offset)) | (Src2 >> Offset);
+    }
+
+    memcpy(GDP, &Dst, OpSize);
   }
-  else {
-    Dst = (Src1 << (OpSize * 8 - Offset)) | (Src2 >> Offset);
-  }
-
-  memcpy(GDP, &Dst, OpSize);
-}
-
-DEF_OP(VSLI) {
-  auto Op = IROp->C<IR::IROp_VSLI>();
-  const __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Vector);
-  const __uint128_t Src2 = Op->ByteShift * 8;
-
-  const __uint128_t Dst = Op->ByteShift >= sizeof(__uint128_t) ? 0 : Src1 << Src2;
-  memcpy(GDP, &Dst, 16);
-}
-
-DEF_OP(VSRI) {
-  auto Op = IROp->C<IR::IROp_VSRI>();
-  const __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Vector);
-  const __uint128_t Src2 = Op->ByteShift * 8;
-
-  const __uint128_t Dst = Op->ByteShift >= sizeof(__uint128_t) ? 0 : Src1 >> Src2;
-  memcpy(GDP, &Dst, 16);
 }
 
 DEF_OP(VUShrI) {
@@ -1695,205 +1757,235 @@ DEF_OP(VShlI) {
 }
 
 DEF_OP(VUShrNI) {
-  auto Op = IROp->C<IR::IROp_VUShrNI>();
+  const auto Op = IROp->C<IR::IROp_VUShrNI>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t BitShift = Op->BitShift;
-  uint8_t Tmp[16]{};
+  const uint8_t BitShift = Op->BitShift;
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / (Op->Header.ElementSize << 1);
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / (ElementSize << 1);
   const auto Func = [BitShift](auto a, auto min, auto max) {
     return BitShift >= (sizeof(a) * 8) ? 0 : a >> BitShift;
   };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP(1, uint8_t, uint16_t, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP(2, uint16_t, uint32_t, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP(4, uint32_t, uint64_t, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VUShrNI2) {
-  auto Op = IROp->C<IR::IROp_VUShrNI2>();
+  const auto Op = IROp->C<IR::IROp_VUShrNI2>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->VectorLower);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->VectorUpper);
-  uint8_t BitShift = Op->BitShift;
-  uint8_t Tmp[16];
+  const uint8_t BitShift = Op->BitShift;
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
-  const uint8_t Elements = OpSize / (Op->Header.ElementSize << 1);
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / (ElementSize << 1);
   const auto Func = [BitShift](auto a, auto min, auto max) {
     return BitShift >= (sizeof(a) * 8) ? 0 : a >> BitShift;
   };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP_TOP(1, uint8_t, uint16_t, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP_TOP(2, uint16_t, uint32_t, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP_TOP(4, uint32_t, uint64_t, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VSXTL) {
-  auto Op = IROp->C<IR::IROp_VSXTL>();
+  const auto Op = IROp->C<IR::IROp_VSXTL>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
   const auto Func = [](auto a, auto min, auto max) { return a; };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP(2, int16_t, int8_t, Func,  0, 0)
     DO_VECTOR_1SRC_2TYPE_OP(4, int32_t, int16_t, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP(8, int64_t, int32_t, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VSXTL2) {
-  auto Op = IROp->C<IR::IROp_VSXTL2>();
+  const auto Op = IROp->C<IR::IROp_VSXTL2>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
   const auto Func = [](auto a, auto min, auto max) { return a; };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP_TOP_SRC(2, int16_t, int8_t, Func,  0, 0)
     DO_VECTOR_1SRC_2TYPE_OP_TOP_SRC(4, int32_t, int16_t, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP_TOP_SRC(8, int64_t, int32_t, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VUXTL) {
-  auto Op = IROp->C<IR::IROp_VUXTL>();
+  const auto Op = IROp->C<IR::IROp_VUXTL>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
   const auto Func = [](auto a, auto min, auto max) { return a; };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP(2, uint16_t, uint8_t, Func,  0, 0)
     DO_VECTOR_1SRC_2TYPE_OP(4, uint32_t, uint16_t, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP(8, uint64_t, uint32_t, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VUXTL2) {
-  auto Op = IROp->C<IR::IROp_VUXTL2>();
+  const auto Op = IROp->C<IR::IROp_VUXTL2>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
 
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
   const auto Func = [](auto a, auto min, auto max) { return a; };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP_TOP_SRC(2, uint16_t, uint8_t, Func,  0, 0)
     DO_VECTOR_1SRC_2TYPE_OP_TOP_SRC(4, uint32_t, uint16_t, Func, 0, 0)
     DO_VECTOR_1SRC_2TYPE_OP_TOP_SRC(8, uint64_t, uint32_t, Func, 0, 0)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VSQXTN) {
-  auto Op = IROp->C<IR::IROp_VSQXTN>();
+  const auto Op = IROp->C<IR::IROp_VSQXTN>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / (Op->Header.ElementSize << 1);
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / (ElementSize << 1);
   const auto Func = [](auto a, auto min, auto max) {
     return std::max(std::min(a, (decltype(a))max), (decltype(a))min);
   };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP(1, int8_t, int16_t, Func, std::numeric_limits<int8_t>::min(), std::numeric_limits<int8_t>::max())
     DO_VECTOR_1SRC_2TYPE_OP(2, int16_t, int32_t, Func, std::numeric_limits<int16_t>::min(), std::numeric_limits<int16_t>::max())
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VSQXTN2) {
-  auto Op = IROp->C<IR::IROp_VSQXTN2>();
+  const auto Op = IROp->C<IR::IROp_VSQXTN2>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->VectorLower);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->VectorUpper);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / (Op->Header.ElementSize << 1);
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / (ElementSize << 1);
   const auto Func = [](auto a, auto min, auto max) {
     return std::max(std::min(a, (decltype(a))max), (decltype(a))min);
   };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP_TOP(1, int8_t, int16_t, Func, std::numeric_limits<int8_t>::min(), std::numeric_limits<int8_t>::max())
     DO_VECTOR_1SRC_2TYPE_OP_TOP(2, int16_t, int32_t, Func, std::numeric_limits<int16_t>::min(), std::numeric_limits<int16_t>::max())
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VSQXTUN) {
-  auto Op = IROp->C<IR::IROp_VSQXTUN>();
+  const auto Op = IROp->C<IR::IROp_VSQXTUN>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / (Op->Header.ElementSize << 1);
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / (ElementSize << 1);
   const auto Func = [](auto a, auto min, auto max) {
     return std::max(std::min(a, (decltype(a))max), (decltype(a))min);
   };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP(1, uint8_t, int16_t, Func, 0, (1 << 8) - 1)
     DO_VECTOR_1SRC_2TYPE_OP(2, uint16_t, int32_t, Func, 0, (1 << 16) - 1)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VSQXTUN2) {
-  auto Op = IROp->C<IR::IROp_VSQXTUN2>();
+  const auto Op = IROp->C<IR::IROp_VSQXTUN2>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->VectorLower);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->VectorUpper);
-  uint8_t Tmp[16]{};
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  const uint8_t Elements = OpSize / (Op->Header.ElementSize << 1);
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / (ElementSize << 1);
   const auto Func = [](auto a, auto min, auto max) {
     return std::max(std::min(a, (decltype(a))max), (decltype(a))min);
   };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_1SRC_2TYPE_OP_TOP(1, uint8_t, int16_t, Func, 0, (1 << 8) - 1)
     DO_VECTOR_1SRC_2TYPE_OP_TOP(2, uint16_t, int32_t, Func, 0, (1 << 16) - 1)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
@@ -1923,22 +2015,25 @@ DEF_OP(VUMul) {
 }
 
 DEF_OP(VUMull) {
-  auto Op = IROp->C<IR::IROp_VUMull>();
+  const auto Op = IROp->C<IR::IROp_VUMull>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->Vector1);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->Vector2);
 
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
   const auto Func = [](auto a, auto b) { return a * b; };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_2SRC_2TYPE_OP(2, uint16_t, uint8_t, Func)
     DO_VECTOR_2SRC_2TYPE_OP(4, uint32_t, uint16_t, Func)
     DO_VECTOR_2SRC_2TYPE_OP(8, uint64_t, uint32_t, Func)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
@@ -1968,100 +2063,112 @@ DEF_OP(VSMul) {
 }
 
 DEF_OP(VSMull) {
-  auto Op = IROp->C<IR::IROp_VSMull>();
+  const auto Op = IROp->C<IR::IROp_VSMull>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->Vector1);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->Vector2);
 
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
   const auto Func = [](auto a, auto b) { return a * b; };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_2SRC_2TYPE_OP(2, int16_t, int8_t, Func)
     DO_VECTOR_2SRC_2TYPE_OP(4, int32_t, int16_t, Func)
     DO_VECTOR_2SRC_2TYPE_OP(8, int64_t, int32_t, Func)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VUMull2) {
-  auto Op = IROp->C<IR::IROp_VUMull2>();
+  const auto Op = IROp->C<IR::IROp_VUMull2>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->Vector1);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->Vector2);
 
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
   const auto Func = [](auto a, auto b) { return a * b; };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_2SRC_2TYPE_OP_TOP_SRC(2, uint16_t, uint8_t, Func)
     DO_VECTOR_2SRC_2TYPE_OP_TOP_SRC(4, uint32_t, uint16_t, Func)
     DO_VECTOR_2SRC_2TYPE_OP_TOP_SRC(8, uint64_t, uint32_t, Func)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VSMull2) {
-  auto Op = IROp->C<IR::IROp_VSMull2>();
+  const auto Op = IROp->C<IR::IROp_VSMull2>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->Vector1);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->Vector2);
 
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
   const auto Func = [](auto a, auto b) { return a * b; };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_2SRC_2TYPE_OP_TOP_SRC(2, int16_t, int8_t, Func)
     DO_VECTOR_2SRC_2TYPE_OP_TOP_SRC(4, int32_t, int16_t, Func)
     DO_VECTOR_2SRC_2TYPE_OP_TOP_SRC(8, int64_t, int32_t, Func)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, Op->Header.Size);
 }
 
 DEF_OP(VUABDL) {
-  auto Op = IROp->C<IR::IROp_VUABDL>();
+  const auto Op = IROp->C<IR::IROp_VUABDL>();
   const uint8_t OpSize = IROp->Size;
 
   void *Src1 = GetSrc<void*>(Data->SSAData, Op->Vector1);
   void *Src2 = GetSrc<void*>(Data->SSAData, Op->Vector2);
 
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
-  const uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
 
   const auto Func8 = [](auto a, auto b) { return std::abs((int16_t)a - (int16_t)b); };
   const auto Func16 = [](auto a, auto b) { return std::abs((int32_t)a - (int32_t)b); };
   const auto Func32 = [](auto a, auto b) { return std::abs((int64_t)a - (int64_t)b); };
 
-  switch (Op->Header.ElementSize) {
+  switch (ElementSize) {
     DO_VECTOR_2SRC_2TYPE_OP(2, uint16_t, uint8_t, Func8)
     DO_VECTOR_2SRC_2TYPE_OP(4, uint32_t, uint16_t, Func16)
     DO_VECTOR_2SRC_2TYPE_OP(8, uint64_t, uint32_t, Func32)
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
   memcpy(GDP, Tmp, OpSize);
 }
 
 DEF_OP(VTBL1) {
-  auto Op = IROp->C<IR::IROp_VTBL1>();
+  const auto Op = IROp->C<IR::IROp_VTBL1>();
   const uint8_t OpSize = IROp->Size;
 
   const auto *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->VectorTable);
   const auto *Src2 = GetSrc<uint8_t*>(Data->SSAData, Op->VectorIndices);
 
-  uint8_t Tmp[16];
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE];
 
   for (size_t i = 0; i < OpSize; ++i) {
     const uint8_t Index = Src2[i];

External/FEXCore/Source/Interface/Core/JIT/Arm64/Arm64Relocations.cpp

@@ -9,7 +9,7 @@ $end_info$
 #include "Interface/HLE/Thunks/Thunks.h"
 
 namespace FEXCore::CPU {
-    
+
 uint64_t Arm64JITCore::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
   switch (Op) {
     case FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER:
@@ -22,18 +22,18 @@ uint64_t Arm64JITCore::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiter
   return ~0ULL;
 }
 
-void Arm64JITCore::InsertNamedThunkRelocation(vixl::aarch64::Register Reg, const IR::SHA256Sum &Sum) {
+void Arm64JITCore::InsertNamedThunkRelocation(ARMEmitter::Register Reg, const IR::SHA256Sum &Sum) {
   Relocation MoveABI{};
   MoveABI.NamedThunkMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE;
   // Offset is the offset from the entrypoint of the block
   auto CurrentCursor = GetCursorAddress<uint8_t *>();
   MoveABI.NamedThunkMove.Offset = CurrentCursor - GuestEntry;
   MoveABI.NamedThunkMove.Symbol = Sum;
-  MoveABI.NamedThunkMove.RegisterIndex = Reg.GetCode();
+  MoveABI.NamedThunkMove.RegisterIndex = Reg.Idx();
 
   uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Sum));
 
-  LoadConstant(Reg, Pointer, EmitterCTX->Config.CacheObjectCodeCompilation());
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Pointer, EmitterCTX->Config.CacheObjectCodeCompilation());
   Relocations.emplace_back(MoveABI);
 }
 
@@ -41,7 +41,7 @@ Arm64JITCore::NamedSymbolLiteralPair Arm64JITCore::InsertNamedSymbolLiteral(FEXC
   uint64_t Pointer = GetNamedSymbolLiteral(Op);
 
   Arm64JITCore::NamedSymbolLiteralPair Lit {
-    .Lit = Literal(Pointer),
+    .Lit = Pointer,
     .MoveABI = {
       .NamedSymbolLiteral = {
         .Header = {
@@ -60,20 +60,21 @@ void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit) {
   auto CurrentCursor = GetCursorAddress<uint8_t *>();
   Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - GuestEntry;
 
-  place(&Lit.Lit);
+  Bind(&Lit.Loc);
+  dc64(Lit.Lit);
   Relocations.emplace_back(Lit.MoveABI);
 }
 
-void Arm64JITCore::InsertGuestRIPMove(vixl::aarch64::Register Reg, uint64_t Constant) {
+void Arm64JITCore::InsertGuestRIPMove(ARMEmitter::Register Reg, uint64_t Constant) {
   Relocation MoveABI{};
   MoveABI.GuestRIPMove.Header.Type = FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE;
   // Offset is the offset from the entrypoint of the block
   auto CurrentCursor = GetCursorAddress<uint8_t *>();
   MoveABI.GuestRIPMove.Offset = CurrentCursor - GuestEntry;
   MoveABI.GuestRIPMove.GuestRIP = Constant;
-  MoveABI.GuestRIPMove.RegisterIndex = Reg.GetCode();
+  MoveABI.GuestRIPMove.RegisterIndex = Reg.Idx();
 
-  LoadConstant(Reg, Constant, EmitterCTX->Config.CacheObjectCodeCompilation());
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Constant, EmitterCTX->Config.CacheObjectCodeCompilation());
   Relocations.emplace_back(MoveABI);
 }
 
@@ -87,11 +88,10 @@ bool Arm64JITCore::ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uin
       case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {
         uint64_t Pointer = GetNamedSymbolLiteral(Reloc->NamedSymbolLiteral.Symbol);
         // Relocation occurs at the cursorEntry + offset relative to that cursor
-        GetBuffer()->SetCursorOffset(CursorEntry + Reloc->NamedSymbolLiteral.Offset);
+        SetCursorOffset(CursorEntry + Reloc->NamedSymbolLiteral.Offset);
 
         // Generate a literal so we can place it
-        Literal<uint64_t> Lit(Pointer);
-        place(&Lit);
+        dc64(Pointer);
 
         DataIndex += sizeof(Reloc->NamedSymbolLiteral);
         break;
@@ -103,8 +103,8 @@ bool Arm64JITCore::ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uin
         }
 
         // Relocation occurs at the cursorEntry + offset relative to that cursor.
-        GetBuffer()->SetCursorOffset(CursorEntry + Reloc->NamedThunkMove.Offset);
-        LoadConstant(vixl::aarch64::XRegister(Reloc->NamedThunkMove.RegisterIndex), Pointer, true);
+        SetCursorOffset(CursorEntry + Reloc->NamedThunkMove.Offset);
+        LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc->NamedThunkMove.RegisterIndex), Pointer, true);
         DataIndex += sizeof(Reloc->NamedThunkMove);
         break;
       }
@@ -117,8 +117,8 @@ bool Arm64JITCore::ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uin
         }
 
         // Relocation occurs at the cursorEntry + offset relative to that cursor.
-        GetBuffer()->SetCursorOffset(CursorEntry + Reloc->GuestRIPMove.Offset);
-        LoadConstant(vixl::aarch64::XRegister(Reloc->GuestRIPMove.RegisterIndex), Pointer, true);
+        SetCursorOffset(CursorEntry + Reloc->GuestRIPMove.Offset);
+        LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc->GuestRIPMove.RegisterIndex), Pointer, true);
         DataIndex += sizeof(Reloc->GuestRIPMove);
         break;
       }

External/FEXCore/Source/Interface/Core/JIT/Arm64/BranchOps.cpp

@@ -6,6 +6,7 @@ $end_info$
 
 #include "Interface/Context/Context.h"
 #include "FEXCore/IR/IR.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/LookupCache.h"
 
 #include "Interface/Core/JIT/Arm64/JITClass.h"
@@ -17,9 +18,7 @@ $end_info$
 #include <Interface/HLE/Thunks/Thunks.h>
 
 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 
 DEF_OP(SignalReturn) {
   // First we must reset the stack
@@ -27,12 +26,11 @@ DEF_OP(SignalReturn) {
 
   // Now branch to our signal return helper
   // This can't be a direct branch since the code needs to live at a constant location
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SignalReturnHandler)));
-  br(x0);
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SignalReturnHandler));
+  br(ARMEmitter::Reg::r0);
 }
 
 DEF_OP(CallbackReturn) {
-
   // spill back to CTX
   SpillStaticRegs();
 
@@ -41,14 +39,14 @@ DEF_OP(CallbackReturn) {
 
   // We can now lower the ref counter again
 
-  ldr(w2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter)));
-  sub(w2, w2, 1);
-  str(w2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter)));
+  ldr(ARMEmitter::WReg::w2, STATE, offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter));
+  sub(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, 1);
+  str(ARMEmitter::WReg::w2, STATE, offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter));
 
   // We need to adjust an additional 8 bytes to get back to the original "misaligned" RSP state
-  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
-  add(x2, x2, 8);
-  str(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
+  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP]));
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r2, 8);
+  str(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP]));
 
   PopCalleeSavedRegisters();
 
@@ -59,39 +57,41 @@ DEF_OP(CallbackReturn) {
 DEF_OP(ExitFunction) {
   auto Op = IROp->C<IR::IROp_ExitFunction>();
 
-  Label FullLookup;
-
   ResetStack();
 
-  aarch64::Register RipReg;
   uint64_t NewRIP;
 
   if (IsInlineConstant(Op->NewRIP, &NewRIP) || IsInlineEntrypointOffset(Op->NewRIP, &NewRIP)) {
-    Literal l_BranchHost{ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker};
-    Literal l_BranchGuest{NewRIP};
+    ARMEmitter::ForwardLabel l_BranchHost;
+    ARMEmitter::ForwardLabel l_BranchGuest;
 
-    ldr(x0, &l_BranchHost);
-    blr(x0);
+    ldr(ARMEmitter::XReg::x0, &l_BranchHost);
+    blr(ARMEmitter::Reg::r0);
+
+    Bind(&l_BranchHost);
+    dc64(ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker);
+    Bind(&l_BranchGuest);
+    dc64(NewRIP);
 
-    place(&l_BranchHost);
-    place(&l_BranchGuest);
   } else {
-    RipReg = GetReg<RA_64>(Op->NewRIP.ID());
+
+    ARMEmitter::ForwardLabel FullLookup;
+    auto RipReg = GetReg(Op->NewRIP.ID());
 
     // L1 Cache
-    ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.L1Pointer)));
+    ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.L1Pointer));
 
-    and_(x3, RipReg, LookupCache::L1_ENTRIES_MASK);
-    add(x0, x0, Operand(x3, Shift::LSL, 4));
+    and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, RipReg, LookupCache::L1_ENTRIES_MASK);
+    add(ARMEmitter::XReg::x0, ARMEmitter::XReg::x0, ARMEmitter::XReg::x3, ARMEmitter::ShiftType::LSL, 4);
 
-    ldp(x1, x0, MemOperand(x0));
-    cmp(x0, RipReg);
-    b(&FullLookup, Condition::ne);
-    br(x1);
+    ldp<ARMEmitter::IndexType::OFFSET>(ARMEmitter::XReg::x1, ARMEmitter::XReg::x0, ARMEmitter::Reg::r0, 0);
+    cmp(ARMEmitter::XReg::x0, RipReg.X());
+    b(ARMEmitter::Condition::CC_NE, &FullLookup);
+    br(ARMEmitter::Reg::r1);
 
-    bind(&FullLookup);
-    ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop)));
-    str(RipReg, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
+    Bind(&FullLookup);
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+    str(RipReg.X(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip));
     br(TMP1);
   }
 }
@@ -103,66 +103,65 @@ DEF_OP(Jump) {
   PendingTargetLabel = &JumpTargets.try_emplace(Target).first->second;
 }
 
-#define GRCMP(Node) (Op->CompareSize == 4 ? GetReg<RA_32>(Node) : GetReg<RA_64>(Node))
-#define GRFCMP(Node) (Op->CompareSize == 4 ? GetDst(Node).S() : GetDst(Node).D())
-
-static Condition MapBranchCC(IR::CondClassType Cond) {
+static ARMEmitter::Condition MapBranchCC(IR::CondClassType Cond) {
   switch (Cond.Val) {
-  case FEXCore::IR::COND_EQ: return Condition::eq;
-  case FEXCore::IR::COND_NEQ: return Condition::ne;
-  case FEXCore::IR::COND_SGE: return Condition::ge;
-  case FEXCore::IR::COND_SLT: return Condition::lt;
-  case FEXCore::IR::COND_SGT: return Condition::gt;
-  case FEXCore::IR::COND_SLE: return Condition::le;
-  case FEXCore::IR::COND_UGE: return Condition::cs;
-  case FEXCore::IR::COND_ULT: return Condition::cc;
-  case FEXCore::IR::COND_UGT: return Condition::hi;
-  case FEXCore::IR::COND_ULE: return Condition::ls;
-  case FEXCore::IR::COND_FLU: return Condition::lt;
-  case FEXCore::IR::COND_FGE: return Condition::ge;
-  case FEXCore::IR::COND_FLEU:return Condition::le;
-  case FEXCore::IR::COND_FGT: return Condition::gt;
-  case FEXCore::IR::COND_FU:  return Condition::vs;
-  case FEXCore::IR::COND_FNU: return Condition::vc;
+  case FEXCore::IR::COND_EQ:  return ARMEmitter::Condition::CC_EQ;
+  case FEXCore::IR::COND_NEQ: return ARMEmitter::Condition::CC_NE;
+  case FEXCore::IR::COND_SGE: return ARMEmitter::Condition::CC_GE;
+  case FEXCore::IR::COND_SLT: return ARMEmitter::Condition::CC_LT;
+  case FEXCore::IR::COND_SGT: return ARMEmitter::Condition::CC_GT;
+  case FEXCore::IR::COND_SLE: return ARMEmitter::Condition::CC_LE;
+  case FEXCore::IR::COND_UGE: return ARMEmitter::Condition::CC_CS;
+  case FEXCore::IR::COND_ULT: return ARMEmitter::Condition::CC_CC;
+  case FEXCore::IR::COND_UGT: return ARMEmitter::Condition::CC_HI;
+  case FEXCore::IR::COND_ULE: return ARMEmitter::Condition::CC_LS;
+  case FEXCore::IR::COND_FLU: return ARMEmitter::Condition::CC_LT;
+  case FEXCore::IR::COND_FGE: return ARMEmitter::Condition::CC_GE;
+  case FEXCore::IR::COND_FLEU:return ARMEmitter::Condition::CC_LE;
+  case FEXCore::IR::COND_FGT: return ARMEmitter::Condition::CC_GT;
+  case FEXCore::IR::COND_FU:  return ARMEmitter::Condition::CC_VS;
+  case FEXCore::IR::COND_FNU: return ARMEmitter::Condition::CC_VC;
   case FEXCore::IR::COND_VS:
   case FEXCore::IR::COND_VC:
   case FEXCore::IR::COND_MI:
   case FEXCore::IR::COND_PL:
   default:
   LOGMAN_MSG_A_FMT("Unsupported compare type");
-  return Condition::nv;
+  return ARMEmitter::Condition::CC_NV;
   }
 }
 
-
 DEF_OP(CondJump) {
   auto Op = IROp->C<IR::IROp_CondJump>();
 
-  Label *TrueTargetLabel = &JumpTargets.try_emplace(Op->TrueBlock.ID()).first->second;
+  auto TrueTargetLabel = &JumpTargets.try_emplace(Op->TrueBlock.ID()).first->second;
 
   uint64_t Const;
   const bool isConst = IsInlineConstant(Op->Cmp2, &Const);
 
+  const auto Size = Op->CompareSize == 4 ? ARMEmitter::Size::i32Bit : ARMEmitter::Size::i64Bit;
+  const auto SubSize = ARMEmitter::ToVectorSizePair(Op->CompareSize == 4 ? ARMEmitter::SubRegSize::i32Bit : ARMEmitter::SubRegSize::i64Bit);
+
   if (isConst && Const == 0 && Op->Cond.Val == FEXCore::IR::COND_EQ) {
     LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1.ID()), "CondJump: Expected GPR");
-    cbz(GRCMP(Op->Cmp1.ID()), TrueTargetLabel);
+    cbz(Size, GetReg(Op->Cmp1.ID()), TrueTargetLabel);
   } else if (isConst && Const == 0 && Op->Cond.Val == FEXCore::IR::COND_NEQ) {
     LOGMAN_THROW_A_FMT(IsGPR(Op->Cmp1.ID()), "CondJump: Expected GPR");
-    cbnz(GRCMP(Op->Cmp1.ID()), TrueTargetLabel);
+    cbnz(Size, GetReg(Op->Cmp1.ID()), TrueTargetLabel);
   } else {
     if (IsGPR(Op->Cmp1.ID())) {
       if (isConst) {
-        cmp(GRCMP(Op->Cmp1.ID()), Const);
+        cmp(Size, GetReg(Op->Cmp1.ID()), Const);
       } else {
-        cmp(GRCMP(Op->Cmp1.ID()), GRCMP(Op->Cmp2.ID()));
+        cmp(Size, GetReg(Op->Cmp1.ID()), GetReg(Op->Cmp2.ID()));
       }
     } else if (IsFPR(Op->Cmp1.ID())) {
-      fcmp(GRFCMP(Op->Cmp1.ID()), GRFCMP(Op->Cmp2.ID()));
+      fcmp(SubSize.Scalar, GetVReg(Op->Cmp1.ID()), GetVReg(Op->Cmp2.ID()));
     } else {
       LOGMAN_MSG_A_FMT("CondJump: Expected GPR or FPR");
     }
 
-    b(TrueTargetLabel, MapBranchCC(Op->Cond));
+    b(MapBranchCC(Op->Cond), TrueTargetLabel);
   }
 
   PendingTargetLabel = &JumpTargets.try_emplace(Op->FalseBlock.ID()).first->second;
@@ -176,7 +175,7 @@ DEF_OP(Syscall) {
   // X2: Pointer to SyscallArguments
 
   FEXCore::IR::SyscallFlags Flags = Op->Flags;
-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
 
   if ((Flags & FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) != FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) {
     SpillStaticRegs();
@@ -187,23 +186,25 @@ DEF_OP(Syscall) {
   }
 
   uint64_t SPOffset = AlignUp(FEXCore::HLE::SyscallArguments::MAX_ARGS * 8, 16);
-  sub(sp, sp, SPOffset);
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, SPOffset);
   for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS; ++i) {
     if (Op->Header.Args[i].IsInvalid()) continue;
-    str(GetReg<RA_64>(Op->Header.Args[i].ID()), MemOperand(sp, i * 8));
+    str(GetReg(Op->Header.Args[i].ID()).X(), ARMEmitter::Reg::rsp, i * 8);
   }
 
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerObj)));
-  ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerFunc)));
-  mov(x1, STATE);
-  mov(x2, sp);
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerObj));
+  ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerFunc));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, STATE.R());
+
+  // SP supporting move
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::rsp, 0);
 #ifdef VIXL_SIMULATOR
-  GenerateIndirectRuntimeCall<uint64_t, void*, void*, void*>(x3);
+  GenerateIndirectRuntimeCall<uint64_t, void*, void*, void*>(ARMEmitter::Reg::r3);
 #else
-  blr(x3);
+  blr(ARMEmitter::Reg::r3);
 #endif
 
-  add(sp, sp, SPOffset);
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, SPOffset);
 
   if ((Flags & FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) != FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY &&
       (Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
@@ -219,7 +220,7 @@ DEF_OP(Syscall) {
 
   if ((Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
     // Move result to its destination register
-    mov(GetReg<RA_64>(Node), x0);
+    mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
   }
 }
 
@@ -236,30 +237,25 @@ DEF_OP(InlineSyscall) {
   // X6: Arg6 - Doesn't exist in x86-64 land. RA INTERSECT
 
   // One argument is removed from the SyscallArguments::MAX_ARGS since the first argument was syscall number
-  const static std::array<vixl::aarch64::Register, FEXCore::HLE::SyscallArguments::MAX_ARGS-1> RegArgs = {{
-    x0, x1, x2, x3, x4, x5
+  const static std::array<ARMEmitter::XRegister, FEXCore::HLE::SyscallArguments::MAX_ARGS-1> RegArgs = {{
+    ARMEmitter::XReg::x0, ARMEmitter::XReg::x1, ARMEmitter::XReg::x2, ARMEmitter::XReg::x3, ARMEmitter::XReg::x4, ARMEmitter::XReg::x5
   }};
 
   bool Intersects{};
   // We always need to spill x8 since we can't know if it is live at this SSA location
   uint32_t SpillMask = 1U << 8;
-  std::vector<vixl::aarch64::Register> IntersectRegs(FEXCore::HLE::SyscallArguments::MAX_ARGS);
   for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
     if (Op->Header.Args[i].IsInvalid()) break;
 
-    auto Reg = GetReg<RA_64>(Op->Header.Args[i].ID());
-    if (Reg.GetCode() == x8.GetCode() ||
-        Reg.GetCode() == x4.GetCode() ||
-        Reg.GetCode() == x5.GetCode()) {
+    auto Reg = GetReg(Op->Header.Args[i].ID());
+    if (Reg.Idx() == ARMEmitter::Reg::r8.Idx() ||
+        Reg.Idx() == ARMEmitter::Reg::r4.Idx() ||
+        Reg.Idx() == ARMEmitter::Reg::r5.Idx()) {
 
-      SpillMask |= (1U << Reg.GetCode());
+      SpillMask |= (1U << Reg.Idx());
       Intersects = true;
     }
   }
-  // XXX: For some reason spilling only the x4, x5, and x8 registers was causing issues
-  // Come back to this once investigation reveals why it fails the gvisor ioctl test
-  // For now override to all GPRs
-  SpillMask = ~0U;
 
   // Ordering is incredibly important here
   // We must spill any overlapping registers first THEN claim we are in a syscall without invalidating state at all
@@ -271,66 +267,31 @@ DEF_OP(InlineSyscall) {
   // 16bit LoadConstant to be a single instruction
   // We must always spill at least one register (x8) so this value always has a bit set
   // This gives the signal handler a value to check to see if we are in a syscall at all
-  LoadConstant(x0, SpillMask & 0xFFFF);
-  str(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SpillMask & 0xFFFF);
+  str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
 
   // Now that we have claimed to be a syscall we can set up the arguments
+  const auto EmitSize = CTX->Config.Is64BitMode() ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+  const auto EmitSubSize = CTX->Config.Is64BitMode() ? ARMEmitter::SubRegSize::i64Bit : ARMEmitter::SubRegSize::i32Bit;
   if (Intersects) {
     for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
       if (Op->Header.Args[i].IsInvalid()) break;
 
-      if (CTX->Config.Is64BitMode()) {
-        auto Reg = GetReg<RA_64>(Op->Header.Args[i].ID());
-        // In the case of intersection with x4, x5, or x8 then these are currently SRA
-        // for registers RAX, RBX, and RSI. Which have just been spilled
-        // Just load back from the context. Could be slightly smarter but this is fairly uncommon
-        if (Reg.GetCode() == x8.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSI])));
-        }
-        else if (Reg.GetCode() == x4.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RAX])));
-        }
-        else if (Reg.GetCode() == x5.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RBX])));
-        }
+      auto Reg = GetReg(Op->Header.Args[i].ID());
+      // In the case of intersection with x4, x5, or x8 then these are currently SRA
+      // for registers RAX, RBX, and RSI. Which have just been spilled
+      // Just load back from the context. Could be slightly smarter but this is fairly uncommon
+      if (Reg.Idx() == FEXCore::ARMEmitter::Reg::r8.Idx()) {
+        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSI]));
       }
-    }
-
-    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
-      if (Op->Header.Args[i].IsInvalid()) break;
-
-      if (CTX->Config.Is64BitMode()) {
-        auto Reg = GetReg<RA_64>(Op->Header.Args[i].ID());
-        // In the case of intersection with x4, x5, or x8 then these are currently SRA
-        // for registers RAX, RBX, and RSI. Which have just been spilled
-        // Just load back from the context. Could be slightly smarter but this is fairly uncommon
-        if (Reg.GetCode() == x8.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSI])));
-        }
-        else if (Reg.GetCode() == x4.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RAX])));
-        }
-        else if (Reg.GetCode() == x5.GetCode()) {
-          ldr(RegArgs[i], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RBX])));
-        }
-        else {
-          mov(RegArgs[i], Reg);
-        }
+      else if (Reg.Idx() == FEXCore::ARMEmitter::Reg::r4.Idx()) {
+        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RAX]));
+      }
+      else if (Reg.Idx() == FEXCore::ARMEmitter::Reg::r5.Idx()) {
+        ldr(EmitSubSize, RegArgs[i].R(), STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RBX]));
       }
       else {
-        auto Reg = GetReg<RA_32>(Op->Header.Args[i].ID());
-        if (Reg.GetCode() == w8.GetCode()) {
-          ldr(RegArgs[i].W(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSI])));
-        }
-        else if (Reg.GetCode() == w4.GetCode()) {
-          ldr(RegArgs[i].W(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RAX])));
-        }
-        else if (Reg.GetCode() == w5.GetCode()) {
-          ldr(RegArgs[i].W(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RBX])));
-        }
-        else {
-          uxtw(RegArgs[i].W(), Reg);
-        }
+        mov(EmitSize, RegArgs[i].R(), Reg);
       }
     }
   }
@@ -338,16 +299,11 @@ DEF_OP(InlineSyscall) {
     for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
       if (Op->Header.Args[i].IsInvalid()) break;
 
-      if (CTX->Config.Is64BitMode()) {
-        mov(RegArgs[i], GetReg<RA_64>(Op->Header.Args[i].ID()));
-      }
-      else {
-        uxtw(RegArgs[i], GetReg<RA_64>(Op->Header.Args[i].ID()));
-      }
+      mov(EmitSize, RegArgs[i].R(), GetReg(Op->Header.Args[i].ID()));
     }
   }
 
-  LoadConstant(x8, Op->HostSyscallNumber);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, Op->HostSyscallNumber);
   svc(0);
   // On updated signal mask we can receive a signal RIGHT HERE
 
@@ -358,16 +314,11 @@ DEF_OP(InlineSyscall) {
 
     // Now the registers we've spilled are back in their original host registers
     // We can safely claim we are no longer in a syscall
-    str(xzr, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
+    str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
 
     // Result is now in x0
     // Move result to its destination register
-    if (CTX->Config.Is64BitMode()) {
-      mov(GetReg<RA_64>(Node), x0);
-    }
-    else {
-      uxtw(GetReg<RA_64>(Node), x0);
-    }
+    mov(EmitSize, GetReg(Node), ARMEmitter::Reg::r0);
   }
 }
 
@@ -379,16 +330,16 @@ DEF_OP(Thunk) {
 
   SpillStaticRegs(); // spill to ctx before ra64 spill
 
-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
 
-  mov(x0, GetReg<RA_64>(Op->ArgPtr.ID()));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetReg(Op->ArgPtr.ID()));
 
   auto thunkFn = ThreadState->CTX->ThunkHandler->LookupThunk(Op->ThunkNameHash);
-  LoadConstant(x2, (uintptr_t)thunkFn);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, (uintptr_t)thunkFn);
 #ifdef VIXL_SIMULATOR
-  GenerateIndirectRuntimeCall<void, void*, void*>(x2);
+  GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
 #else
-  blr(x2);
+  blr(ARMEmitter::Reg::r2);
 #endif
 
   PopDynamicRegsAndLR();
@@ -402,43 +353,45 @@ DEF_OP(ValidateCode) {
   int len = Op->CodeLength;
   int idx = 0;
 
-  LoadConstant(GetReg<RA_64>(Node), 0);
-  LoadConstant(x0, Entry + Op->Offset);
-  LoadConstant(x1, 1);
+  LoadConstant(ARMEmitter::Size::i64Bit, GetReg(Node), 0);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, Entry + Op->Offset);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, 1);
+
+  const auto Dst = GetReg(Node);
 
   while (len >= 8)
   {
-    ldr(x2, MemOperand(x0, idx));
-    LoadConstant(x3, *(const uint32_t *)(OldCode + idx));
-    cmp(x2, x3);
-    csel(GetReg<RA_64>(Node), GetReg<RA_64>(Node), x1, Condition::eq);
+    ldr(ARMEmitter::XReg::x2, ARMEmitter::Reg::r0, idx);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, *(const uint32_t *)(OldCode + idx));
+    cmp(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r3);
+    csel(ARMEmitter::Size::i64Bit, Dst, Dst, ARMEmitter::Reg::r1, ARMEmitter::Condition::CC_EQ);
     len -= 8;
     idx += 8;
   }
   while (len >= 4)
   {
-    ldr(w2, MemOperand(x0, idx));
-    LoadConstant(w3, *(const uint32_t *)(OldCode + idx));
-    cmp(w2, w3);
-    csel(GetReg<RA_64>(Node), GetReg<RA_64>(Node), x1, Condition::eq);
+    ldr(ARMEmitter::WReg::w2, ARMEmitter::Reg::r0, idx);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, *(const uint32_t *)(OldCode + idx));
+    cmp(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r3);
+    csel(ARMEmitter::Size::i64Bit, Dst, Dst, ARMEmitter::Reg::r1, ARMEmitter::Condition::CC_EQ);
     len -= 4;
     idx += 4;
   }
   while (len >= 2)
   {
-    ldrh(w2, MemOperand(x0, idx));
-    LoadConstant(w3, *(const uint16_t *)(OldCode + idx));
-    cmp(w2, w3);
-    csel(GetReg<RA_64>(Node), GetReg<RA_64>(Node), x1, Condition::eq);
+    ldrh(ARMEmitter::Reg::r2, ARMEmitter::Reg::r0, idx);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, *(const uint16_t *)(OldCode + idx));
+    cmp(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r3);
+    csel(ARMEmitter::Size::i64Bit, Dst, Dst, ARMEmitter::Reg::r1, ARMEmitter::Condition::CC_EQ);
     len -= 2;
     idx += 2;
   }
   while (len >= 1)
   {
-    ldrb(w2, MemOperand(x0, idx));
-    LoadConstant(w3, *(const uint8_t *)(OldCode + idx));
-    cmp(w2, w3);
-    csel(GetReg<RA_64>(Node), GetReg<RA_64>(Node), x1, Condition::eq);
+    ldrb(ARMEmitter::Reg::r2, ARMEmitter::Reg::r0, idx);
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, *(const uint8_t *)(OldCode + idx));
+    cmp(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::r3);
+    csel(ARMEmitter::Size::i64Bit, Dst, Dst, ARMEmitter::Reg::r1, ARMEmitter::Condition::CC_EQ);
     len -= 1;
     idx += 1;
   }
@@ -449,17 +402,17 @@ DEF_OP(ThreadRemoveCodeEntry) {
   // X0: Thread
   // X1: RIP
 
-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
 
-  mov(x0, STATE);
-  LoadConstant(x1, Entry);
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, STATE.R());
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, Entry);
 
-  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadRemoveCodeEntryFromJIT)));
+  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadRemoveCodeEntryFromJIT));
   SpillStaticRegs();
 #ifdef VIXL_SIMULATOR
-  GenerateIndirectRuntimeCall<void, void*, void*>(x2);
+  GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
 #else
-  blr(x2);
+  blr(ARMEmitter::Reg::r2);
 #endif
   FillStaticRegs();
 
@@ -470,30 +423,31 @@ DEF_OP(ThreadRemoveCodeEntry) {
 DEF_OP(CPUID) {
   auto Op = IROp->C<IR::IROp_CPUID>();
 
-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
   SpillStaticRegs();
 
   // x0 = CPUID Handler
   // x1 = CPUID Function
   // x2 = CPUID Leaf
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj)));
-  ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDFunction)));
-  mov(x1, GetReg<RA_64>(Op->Function.ID()));
-  mov(x2, GetReg<RA_64>(Op->Leaf.ID()));
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj));
+  ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDFunction));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, GetReg(Op->Function.ID()));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, GetReg(Op->Leaf.ID()));
 #ifdef VIXL_SIMULATOR
-  GenerateIndirectRuntimeCall<__uint128_t, void*, uint64_t, uint64_t>(x3);
+  GenerateIndirectRuntimeCall<__uint128_t, void*, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
 #else
-  blr(x3);
+  blr(ARMEmitter::Reg::r3);
 #endif
 
   FillStaticRegs();
+
   PopDynamicRegsAndLR();
 
   // Results are in x0, x1
   // Results want to be in a i64v2 vector
-  auto Dst = GetSrcPair<RA_64>(Node);
-  mov(Dst.first,  x0);
-  mov(Dst.second, x1);
+  auto Dst = GetRegPair(Node);
+  mov(ARMEmitter::Size::i64Bit, Dst.first,  ARMEmitter::Reg::r0);
+  mov(ARMEmitter::Size::i64Bit, Dst.second, ARMEmitter::Reg::r1);
 }
 
 #undef DEF_OP

External/FEXCore/Source/Interface/Core/JIT/Arm64/ConversionOps.cpp

@@ -4,91 +4,157 @@ tags: backend|arm64
 $end_info$
 */
 
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/JIT/Arm64/JITClass.h"
 
 namespace FEXCore::CPU {
-
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 DEF_OP(VInsGPR) {
-  auto Op = IROp->C<IR::IROp_VInsGPR>();
-  mov(GetDst(Node), GetSrc(Op->DestVector.ID()));
-  switch (Op->Header.ElementSize) {
-    case 1: {
-      ins(GetDst(Node).V16B(), Op->DestIdx, GetReg<RA_32>(Op->Src.ID()));
-    break;
+  const auto Op = IROp->C<IR::IROp_VInsGPR>();
+  const auto OpSize = IROp->Size;
+
+  const auto DestIdx = Op->DestIdx;
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2 || ElementSize == 1, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit :
+    ElementSize == 1 ? ARMEmitter::SubRegSize::i8Bit : ARMEmitter::SubRegSize::i8Bit;
+  const auto ElementsPer128Bit = 16 / ElementSize;
+
+  const auto Dst = GetVReg(Node);
+  const auto DestVector = GetVReg(Op->DestVector.ID());
+  const auto Src = GetReg(Op->Src.ID());
+
+  if (HostSupportsSVE && Is256Bit) {
+    const auto ElementSizeBits = ElementSize * 8;
+    const auto Offset = ElementSizeBits * DestIdx;
+
+    const auto SSEBitSize = Core::CPUState::XMM_SSE_REG_SIZE * 8;
+    const auto InUpperLane = Offset >= SSEBitSize;
+
+    // This is going to be a little gross. Pls forgive me.
+    // Since SVE has the whole vector length agnostic programming
+    // thing going on, we can't exactly freely insert entries into
+    // arbitrary locations in the vector.
+    //
+    // SVE *does* have INSR, however this only shifts the entire
+    // vector to the left by an element size and inserts a value
+    // at the beginning of the vector. Not *quite* what we need.
+    // (though INSR *is* very useful for other things).
+    //
+    // The idea is (in the case of the upper lane), move the upper
+    // lane down, insert into it and recombine with the lower lane.
+    //
+    // In the case of the lower lane, insert and then recombine with
+    // the upper lane.
+
+    if (InUpperLane) {
+      // Move the upper lane down for the insertion.
+      const auto CompactPred = ARMEmitter::PReg::p0;
+      not_(CompactPred, PRED_TMP_32B.Zeroing(), PRED_TMP_16B);
+      compact(ARMEmitter::SubRegSize::i64Bit, VTMP1.Z(), CompactPred, DestVector);
     }
-    case 2: {
-      ins(GetDst(Node).V8H(), Op->DestIdx, GetReg<RA_32>(Op->Src.ID()));
-    break;
+
+    // Put data in place for destructive SPLICE below.
+    mov(Dst.Z(), DestVector.Z());
+
+    // Inserts the GPR value into the given V register.
+    // Also automatically adjusts the index in the case of using the
+    // moved upper lane.
+    const auto Insert = [&](const FEXCore::ARMEmitter::VRegister& reg, int index) {
+      if (InUpperLane) {
+        index -= ElementsPer128Bit;
+      }
+      ins(SubEmitSize, reg, index, Src);
+    };
+
+    if (InUpperLane) {
+      Insert(VTMP1, DestIdx);
+      splice<ARMEmitter::OpType::Destructive>(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), PRED_TMP_16B, Dst.Z(), VTMP1.Z());
+    } else {
+      Insert(Dst, DestIdx);
+      splice<ARMEmitter::OpType::Destructive>(ARMEmitter::SubRegSize::i64Bit, Dst.Z(), PRED_TMP_16B, Dst.Z(), DestVector.Z());
     }
-    case 4: {
-      ins(GetDst(Node).V4S(), Op->DestIdx, GetReg<RA_32>(Op->Src.ID()));
-    break;
-    }
-    case 8: {
-      ins(GetDst(Node).V2D(), Op->DestIdx, GetReg<RA_64>(Op->Src.ID()));
-    break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+  } else {
+    mov(Dst.Q(), DestVector.Q());
+    ins(SubEmitSize, Dst, DestIdx, Src);
   }
 }
 
 DEF_OP(VCastFromGPR) {
   auto Op = IROp->C<IR::IROp_VCastFromGPR>();
+  auto Dst = GetVReg(Node);
+  auto Src = GetReg(Op->Src.ID());
+
   switch (Op->Header.ElementSize) {
     case 1:
-      uxtb(TMP1.W(), GetReg<RA_32>(Op->Src.ID()));
-      fmov(GetDst(Node).S(), TMP1.W());
+      uxtb(ARMEmitter::Size::i32Bit, TMP1, Src);
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1);
       break;
     case 2:
-      uxth(TMP1.W(), GetReg<RA_32>(Op->Src.ID()));
-      fmov(GetDst(Node).S(), TMP1.W());
+      uxth(ARMEmitter::Size::i32Bit, TMP1, Src);
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), TMP1);
       break;
     case 4:
-      fmov(GetDst(Node).S(), GetReg<RA_32>(Op->Src.ID()).W());
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), Src);
       break;
     case 8:
-      fmov(GetDst(Node).D(), GetReg<RA_64>(Op->Src.ID()).X());
+      fmov(ARMEmitter::Size::i64Bit, Dst.D(), Src);
       break;
     default: LOGMAN_MSG_A_FMT("Unknown castGPR element size: {}", Op->Header.ElementSize);
   }
 }
 
 DEF_OP(Float_FromGPR_S) {
-  auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
-  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
+
+  const uint16_t ElementSize = Op->Header.ElementSize;
+  const uint16_t Conv = (ElementSize << 8) | Op->SrcElementSize;
+
+  auto Dst = GetVReg(Node);
+  auto Src = GetReg(Op->Src.ID());
+
   switch (Conv) {
     case 0x0404: { // Float <- int32_t
-      scvtf(GetDst(Node).S(), GetReg<RA_32>(Op->Src.ID()));
+      scvtf(ARMEmitter::Size::i32Bit, Dst.S(), Src);
       break;
     }
     case 0x0408: { // Float <- int64_t
-      scvtf(GetDst(Node).S(), GetReg<RA_64>(Op->Src.ID()));
+      scvtf(ARMEmitter::Size::i64Bit, Dst.S(), Src);
       break;
     }
     case 0x0804: { // Double <- int32_t
-      scvtf(GetDst(Node).D(), GetReg<RA_32>(Op->Src.ID()));
+      scvtf(ARMEmitter::Size::i32Bit, Dst.D(), Src);
       break;
     }
     case 0x0808: { // Double <- int64_t
-      scvtf(GetDst(Node).D(), GetReg<RA_64>(Op->Src.ID()));
+      scvtf(ARMEmitter::Size::i64Bit, Dst.D(), Src);
       break;
     }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled conversion mask: Mask=0x{:04x}, ElementSize={}, SrcElementSize={}",
+                       Conv, ElementSize, Op->SrcElementSize);
+      break;
   }
 }
 
 DEF_OP(Float_FToF) {
   auto Op = IROp->C<IR::IROp_Float_FToF>();
   const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+
+  auto Dst = GetVReg(Node);
+  auto Src = GetVReg(Op->Scalar.ID());
+
   switch (Conv) {
     case 0x0804: { // Double <- Float
-      fcvt(GetDst(Node).D(), GetSrc(Op->Scalar.ID()).S());
+      fcvt(Dst.D(), Src.S());
       break;
     }
     case 0x0408: { // Float <- Double
-      fcvt(GetDst(Node).S(), GetSrc(Op->Scalar.ID()).D());
+      fcvt(Dst.S(), Src.D());
       break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown FCVT sizes: 0x{:x}", Conv);
@@ -96,116 +162,205 @@ DEF_OP(Float_FToF) {
 }
 
 DEF_OP(Vector_SToF) {
-  auto Op = IROp->C<IR::IROp_Vector_SToF>();
-  switch (Op->Header.ElementSize) {
-    case 4:
-      scvtf(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
-    break;
-    case 8:
-      scvtf(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
-    break;
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_SToF element size: {}", Op->Header.ElementSize);
+  const auto Op = IROp->C<IR::IROp_Vector_SToF>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+  if (HostSupportsSVE && Is256Bit) {
+    const auto Mask = PRED_TMP_32B;
+    scvtf(Dst.Z(), SubEmitSize, Mask.Merging(), Vector.Z(), SubEmitSize);
+  } else {
+    scvtf(SubEmitSize, Dst.Q(), Vector.Q());
   }
 }
 
 DEF_OP(Vector_FToZS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToZS>();
-  switch (Op->Header.ElementSize) {
-    case 4:
-      fcvtzs(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
-    break;
-    case 8:
-      fcvtzs(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
-    break;
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_FToZS element size: {}", Op->Header.ElementSize);
+  const auto Op = IROp->C<IR::IROp_Vector_FToZS>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+  if (HostSupportsSVE && Is256Bit) {
+    const auto Mask = PRED_TMP_32B;
+    fcvtzs(Dst, SubEmitSize, Mask.Merging(), Vector, SubEmitSize);
+  } else {
+    fcvtzs(SubEmitSize, Dst.Q(), Vector.Q());
   }
 }
 
 DEF_OP(Vector_FToS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToS>();
-  switch (Op->Header.ElementSize) {
-    case 4:
-      frinti(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
-      fcvtzs(GetDst(Node).V4S(), GetDst(Node).V4S());
-    break;
-    case 8:
-      frinti(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
-      fcvtzs(GetDst(Node).V2D(), GetDst(Node).V2D());
-    break;
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_FToS element size: {}", Op->Header.ElementSize);
+  const auto Op = IROp->C<IR::IROp_Vector_FToS>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+
+  if (HostSupportsSVE && Is256Bit) {
+    const auto Mask = PRED_TMP_32B;
+    frinti(SubEmitSize, Dst, Mask.Merging(), Vector);
+    fcvtzs(Dst, SubEmitSize, Mask.Merging(), Dst, SubEmitSize);
+  } else {
+    const auto Dst = GetVReg(Node);
+    const auto Vector = GetVReg(Op->Vector.ID());
+    frinti(SubEmitSize, Dst.Q(), Vector.Q());
+    fcvtzs(SubEmitSize, Dst.Q(), Dst.Q());
   }
 }
 
 DEF_OP(Vector_FToF) {
-  auto Op = IROp->C<IR::IROp_Vector_FToF>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const auto Op = IROp->C<IR::IROp_Vector_FToF>();
+  const auto OpSize = IROp->Size;
 
-  switch (Conv) {
-    case 0x0804: { // Double <- Float
-      fcvtl(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2S());
-      break;
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+  const auto Conv = (ElementSize << 8) | Op->SrcElementSize;
+
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+
+  if (HostSupportsSVE && Is256Bit) {
+    // Curiously, FCVTLT and FCVTNT have no bottom variants,
+    // and also interesting is that FCVTLT will iterate the
+    // source vector by accessing each odd element and storing
+    // them consecutively in the destination.
+    //
+    // FCVTNT is somewhat like the opposite. It will read each
+    // consecutive element, but store each result into every odd
+    // element in the destination vector.
+    //
+    // We need to undo the behavior of FCVTNT with UZP2. In the case
+    // of FCVTLT, we instead need to set the vector up with ZIP1, so
+    // that the elements will be processed correctly.
+
+    const auto Mask = PRED_TMP_32B.Merging();
+
+    switch (Conv) {
+      case 0x0402: { // Float <- Half
+        zip1(FEXCore::ARMEmitter::SubRegSize::i16Bit, Dst.Z(), Vector.Z(), Vector.Z());
+        fcvtlt(FEXCore::ARMEmitter::SubRegSize::i32Bit, Dst.Z(), Mask, Dst.Z());
+        break;
+      }
+      case 0x0804: { // Double <- Float
+        zip1(FEXCore::ARMEmitter::SubRegSize::i32Bit, Dst.Z(), Vector.Z(), Vector.Z());
+        fcvtlt(FEXCore::ARMEmitter::SubRegSize::i64Bit, Dst.Z(), Mask, Dst.Z());
+        break;
+      }
+      case 0x0204: { // Half <- Float
+        fcvtnt(FEXCore::ARMEmitter::SubRegSize::i16Bit, Dst, Mask, Vector);
+        uzp2(FEXCore::ARMEmitter::SubRegSize::i16Bit, Dst.Z(), Dst.Z(), Dst.Z());
+        break;
+      }
+      case 0x0408: { // Float <- Double
+        fcvtnt(FEXCore::ARMEmitter::SubRegSize::i32Bit, Dst, Mask, Vector);
+        uzp2(FEXCore::ARMEmitter::SubRegSize::i32Bit, Dst.Z(), Dst.Z(), Dst.Z());
+        break;
+      }
+      default:
+        LOGMAN_MSG_A_FMT("Unknown Vector_FToF Type : 0x{:04x}", Conv);
+        break;
     }
-    case 0x0408: { // Float <- Double
-      fcvtn(GetDst(Node).V2S(), GetSrc(Op->Vector.ID()).V2D());
-      break;
+  } else {
+    switch (Conv) {
+      case 0x0402: // Float <- Half
+      case 0x0804: { // Double <- Float
+        fcvtl(SubEmitSize, Dst.D(), Vector.D());
+        break;
+      }
+      case 0x0204: // Half <- Float
+      case 0x0408: { // Float <- Double
+        fcvtn(SubEmitSize, Dst.D(), Vector.D());
+        break;
+      }
+      default:
+        LOGMAN_MSG_A_FMT("Unknown Vector_FToF Type : 0x{:04x}", Conv);
+        break;
     }
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_FToF Type : 0x{:04x}", Conv); break;
   }
 }
 
 DEF_OP(Vector_FToI) {
-  auto Op = IROp->C<IR::IROp_Vector_FToI>();
-  switch (Op->Round) {
-    case FEXCore::IR::Round_Nearest.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frintn(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
+  const auto Op = IROp->C<IR::IROp_Vector_FToI>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+  LOGMAN_THROW_AA_FMT(ElementSize == 8 || ElementSize == 4 || ElementSize == 2, "Unexpected {} size", __func__);
+
+  const auto SubEmitSize = ElementSize == 8 ? ARMEmitter::SubRegSize::i64Bit :
+    ElementSize == 4 ? ARMEmitter::SubRegSize::i32Bit :
+    ElementSize == 2 ? ARMEmitter::SubRegSize::i16Bit : ARMEmitter::SubRegSize::i16Bit;
+
+  const auto Dst = GetVReg(Node);
+  const auto Vector = GetVReg(Op->Vector.ID());
+
+  if (HostSupportsSVE && Is256Bit) {
+    const auto Mask = PRED_TMP_32B.Merging();
+
+    switch (Op->Round) {
+      case FEXCore::IR::Round_Nearest.Val:
+        frintn(SubEmitSize, Dst.Z(), Mask, Vector.Z());
         break;
-        case 8:
-          frintn(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
+      case FEXCore::IR::Round_Negative_Infinity.Val:
+        frintm(SubEmitSize, Dst.Z(), Mask, Vector.Z());
         break;
-      }
-    break;
-    case FEXCore::IR::Round_Negative_Infinity.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frintm(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
+      case FEXCore::IR::Round_Positive_Infinity.Val:
+        frintp(SubEmitSize, Dst.Z(), Mask, Vector.Z());
         break;
-        case 8:
-          frintm(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
+      case FEXCore::IR::Round_Towards_Zero.Val:
+        frintz(SubEmitSize, Dst.Z(), Mask, Vector.Z());
         break;
-      }
-    break;
-    case FEXCore::IR::Round_Positive_Infinity.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frintp(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
+      case FEXCore::IR::Round_Host.Val:
+        frinti(SubEmitSize, Dst.Z(), Mask, Vector.Z());
         break;
-        case 8:
-          frintp(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
+    }
+  } else {
+    switch (Op->Round) {
+      case FEXCore::IR::Round_Nearest.Val:
+        frinti(SubEmitSize, Dst.Q(), Vector.Q());
         break;
-      }
-    break;
-    case FEXCore::IR::Round_Towards_Zero.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frintz(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
+      case FEXCore::IR::Round_Negative_Infinity.Val:
+        frintm(SubEmitSize, Dst.Q(), Vector.Q());
         break;
-        case 8:
-          frintz(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
+      case FEXCore::IR::Round_Positive_Infinity.Val:
+        frintp(SubEmitSize, Dst.Q(), Vector.Q());
         break;
-      }
-    break;
-    case FEXCore::IR::Round_Host.Val:
-      switch (Op->Header.ElementSize) {
-        case 4:
-          frinti(GetDst(Node).V4S(), GetSrc(Op->Vector.ID()).V4S());
+      case FEXCore::IR::Round_Towards_Zero.Val:
+        frintz(SubEmitSize, Dst.Q(), Vector.Q());
         break;
-        case 8:
-          frinti(GetDst(Node).V2D(), GetSrc(Op->Vector.ID()).V2D());
+      case FEXCore::IR::Round_Host.Val:
+        frinti(SubEmitSize, Dst.Q(), Vector.Q());
         break;
-      }
-    break;
+    }
   }
 }
 

External/FEXCore/Source/Interface/Core/JIT/Arm64/EncryptionOps.cpp

@@ -4,99 +4,104 @@ tags: backend|arm64
 $end_info$
 */
 
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/JIT/Arm64/JITClass.h"
 #include "Interface/IR/Passes/RegisterAllocationPass.h"
 
 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 
 DEF_OP(AESImc) {
   auto Op = IROp->C<IR::IROp_VAESImc>();
-  aesimc(GetDst(Node).V16B(), GetSrc(Op->Vector.ID()).V16B());
+  aesimc(GetVReg(Node), GetVReg(Op->Vector.ID()));
 }
 
 DEF_OP(AESEnc) {
-	auto Op = IROp->C<IR::IROp_VAESEnc>();
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->State.ID()).V16B());
-  aese(VTMP1.V16B(), VTMP2.V16B());
-  aesmc(VTMP1.V16B(), VTMP1.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Key.ID()).V16B());
+  auto Op = IROp->C<IR::IROp_VAESEnc>();
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), GetVReg(Op->State.ID()).Q());
+  aese(VTMP1, VTMP2);
+  aesmc(VTMP1, VTMP1);
+  eor(GetVReg(Node).Q(), VTMP1.Q(), GetVReg(Op->Key.ID()).Q());
 }
 
 DEF_OP(AESEncLast) {
-	auto Op = IROp->C<IR::IROp_VAESEncLast>();
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->State.ID()).V16B());
-  aese(VTMP1.V16B(), VTMP2.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Key.ID()).V16B());
+  auto Op = IROp->C<IR::IROp_VAESEncLast>();
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), GetVReg(Op->State.ID()).Q());
+  aese(VTMP1, VTMP2);
+  eor(GetVReg(Node).Q(), VTMP1.Q(), GetVReg(Op->Key.ID()).Q());
 }
 
 DEF_OP(AESDec) {
-	auto Op = IROp->C<IR::IROp_VAESDec>();
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->State.ID()).V16B());
-  aesd(VTMP1.V16B(), VTMP2.V16B());
-  aesimc(VTMP1.V16B(), VTMP1.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Key.ID()).V16B());
+  auto Op = IROp->C<IR::IROp_VAESDec>();
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), GetVReg(Op->State.ID()).Q());
+  aesd(VTMP1, VTMP2);
+  aesimc(VTMP1, VTMP1);
+  eor(GetVReg(Node).Q(), VTMP1.Q(), GetVReg(Op->Key.ID()).Q());
 }
 
 DEF_OP(AESDecLast) {
-	auto Op = IROp->C<IR::IROp_VAESDecLast>();
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->State.ID()).V16B());
-  aesd(VTMP1.V16B(), VTMP2.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Key.ID()).V16B());
+  auto Op = IROp->C<IR::IROp_VAESDecLast>();
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), GetVReg(Op->State.ID()).Q());
+  aesd(VTMP1, VTMP2);
+  eor(GetVReg(Node).Q(), VTMP1.Q(), GetVReg(Op->Key.ID()).Q());
 }
 
 DEF_OP(AESKeyGenAssist) {
-	auto Op = IROp->C<IR::IROp_VAESKeyGenAssist>();
+  auto Op = IROp->C<IR::IROp_VAESKeyGenAssist>();
 
-  aarch64::Literal ConstantLiteral (0x0C030609'0306090CULL, 0x040B0E01'0B0E0104ULL);
-  aarch64::Label PastConstant;
+  ARMEmitter::ForwardLabel Constant;
+  ARMEmitter::ForwardLabel PastConstant;
 
   // Do a "regular" AESE step
-  eor(VTMP2.V16B(), VTMP2.V16B(), VTMP2.V16B());
-  mov(VTMP1.V16B(), GetSrc(Op->Src.ID()).V16B());
-  aese(VTMP1.V16B(), VTMP2.V16B());
+  eor(VTMP2.Q(), VTMP2.Q(), VTMP2.Q());
+  mov(VTMP1.Q(), GetVReg(Op->Src.ID()).Q());
+  aese(VTMP1, VTMP2);
 
   // Do a table shuffle to undo ShiftRows
-  ldr(VTMP3, &ConstantLiteral);
+  ldr(VTMP3.Q(), &Constant);
 
   // Now EOR in the RCON
   if (Op->RCON) {
-    tbl(VTMP1.V16B(), VTMP1.V16B(), VTMP3.V16B());
+    tbl(VTMP1.Q(), VTMP1.Q(), VTMP3.Q());
 
-    LoadConstant(TMP1.W(), Op->RCON);
-    ins(VTMP2.V4S(), 1, TMP1.W());
-    ins(VTMP2.V4S(), 3, TMP1.W());
-    eor(GetDst(Node).V16B(), VTMP1.V16B(), VTMP2.V16B());
+    LoadConstant(ARMEmitter::Size::i64Bit, TMP1, static_cast<uint64_t>(Op->RCON) << 32);
+    dup(ARMEmitter::SubRegSize::i64Bit, VTMP2.Q(), TMP1);
+    eor(GetVReg(Node).Q(), VTMP1.Q(), VTMP2.Q());
   }
   else {
-    tbl(GetDst(Node).V16B(), VTMP1.V16B(), VTMP3.V16B());
+    tbl(GetVReg(Node).Q(), VTMP1.Q(), VTMP3.Q());
   }
 
   b(&PastConstant);
-  place(&ConstantLiteral);
-  bind(&PastConstant);
+  Bind(&Constant);
+  dc64(0x040B0E01'0B0E0104ULL);
+  dc64(0x0C030609'0306090CULL);
+  Bind(&PastConstant);
 }
 
 DEF_OP(CRC32) {
   auto Op = IROp->C<IR::IROp_CRC32>();
+
+  const auto Dst = GetReg(Node);
+  const auto Src1 = GetReg(Op->Src1.ID());
+  const auto Src2 = GetReg(Op->Src2.ID());
+
   switch (Op->SrcSize) {
     case 1:
-      crc32cb(GetReg<RA_32>(Node), GetReg<RA_32>(Op->Src1.ID()), GetReg<RA_32>(Op->Src2.ID()));
+      crc32cb(Dst.W(), Src1.W(), Src2.W());
       break;
     case 2:
-      crc32ch(GetReg<RA_32>(Node), GetReg<RA_32>(Op->Src1.ID()), GetReg<RA_32>(Op->Src2.ID()));
+      crc32ch(Dst.W(), Src1.W(), Src2.W());
       break;
     case 4:
-      crc32cw(GetReg<RA_32>(Node), GetReg<RA_32>(Op->Src1.ID()), GetReg<RA_32>(Op->Src2.ID()));
+      crc32cw(Dst.W(), Src1.W(), Src2.W());
       break;
     case 8:
-      crc32cx(GetReg<RA_32>(Node), GetReg<RA_32>(Op->Src1.ID()), GetReg<RA_64>(Op->Src2.ID()));
+      crc32cx(Dst, Src1, Src2);
       break;
     default: LOGMAN_MSG_A_FMT("Unknown CRC32 size: {}", Op->SrcSize);
   }
@@ -105,24 +110,24 @@ DEF_OP(CRC32) {
 DEF_OP(PCLMUL) {
   auto Op = IROp->C<IR::IROp_PCLMUL>();
 
-  auto Dst  = GetDst(Node).Q();
-  auto Src1 = GetSrc(Op->Src1.ID()).V2D();
-  auto Src2 = GetSrc(Op->Src2.ID()).V2D();
+  auto Dst  = GetVReg(Node);
+  auto Src1 = GetVReg(Op->Src1.ID());
+  auto Src2 = GetVReg(Op->Src2.ID());
 
   switch (Op->Selector) {
   case 0b00000000:
-    pmull(Dst, Src1, Src2);
+    pmull(ARMEmitter::SubRegSize::i128Bit, Dst.D(), Src1.D(), Src2.D());
     break;
   case 0b00000001:
-    mov(VTMP1.V1D(), Src1, 1);
-    pmull(Dst, VTMP1.V2D(), Src2);
+    dup(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), Src1.Q(), 1);
+    pmull(ARMEmitter::SubRegSize::i128Bit, Dst.D(), VTMP1.D(), Src2.D());
     break;
   case 0b00010000:
-    mov(VTMP1.V1D(), Src2, 1);
-    pmull(Dst, VTMP1.V2D(), Src1);
+    dup(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), Src2.Q(), 1);
+    pmull(ARMEmitter::SubRegSize::i128Bit, Dst.D(), VTMP1.D(), Src1.D());
     break;
   case 0b00010001:
-    pmull2(Dst, Src1, Src2);
+    pmull2(ARMEmitter::SubRegSize::i128Bit, Dst.Q(), Src1.Q(), Src2.Q());
     break;
   default:
     LOGMAN_MSG_A_FMT("Unknown PCLMUL selector: {}", Op->Selector);

External/FEXCore/Source/Interface/Core/JIT/Arm64/FlagOps.cpp

@@ -7,13 +7,10 @@ $end_info$
 #include "Interface/Core/JIT/Arm64/JITClass.h"
 
 namespace FEXCore::CPU {
-
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 DEF_OP(GetHostFlag) {
   auto Op = IROp->C<IR::IROp_GetHostFlag>();
-  ubfx(GetReg<RA_64>(Node), GetReg<RA_64>(Op->Value.ID()), Op->Flag, 1);
+  ubfx(ARMEmitter::Size::i64Bit, GetReg(Node), GetReg(Op->Value.ID()), Op->Flag, 1);
 }
 
 #undef DEF_OP

External/FEXCore/Source/Interface/Core/JIT/Arm64/JIT.cpp

@@ -11,6 +11,7 @@ $end_info$
 */
 
 #include "Interface/Context/Context.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/LookupCache.h"
 
 #include "Interface/Core/ArchHelpers/Arm64.h"
@@ -28,6 +29,7 @@ $end_info$
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/CompilerDefs.h>
 #include <FEXCore/Utils/EnumUtils.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 
@@ -76,10 +78,7 @@ static void PrintVectorValue(uint64_t Value, uint64_t ValueUpper) {
 
 namespace FEXCore::CPU {
 
-using namespace vixl;
-using namespace vixl::aarch64;
-
-void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
+void Arm64JITCore::Op_Unhandled(IR::IROp_Header const *IROp, IR::NodeID Node) {
   FallbackInfo Info;
   if (!InterpreterOps::GetFallbackHandler(IROp, &Info)) {
 #if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
@@ -90,15 +89,15 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
       case FABI_VOID_U16:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
-
-        uxth(w0, GetReg<RA_32>(IROp->Args[0].ID()));
-        ldr(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        PushDynamicRegsAndLR(TMP1);
 
+        const auto Src1 = GetReg(IROp->Args[0].ID());
+        uxth(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r0, Src1);
+        ldr(ARMEmitter::XReg::x1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<void, uint16_t>(x1);
+        GenerateIndirectRuntimeCall<void, uint16_t>(ARMEmitter::Reg::r1);
 #else
-        blr(x1);
+        blr(ARMEmitter::Reg::r1);
 #endif
 
         PopDynamicRegsAndLR();
@@ -110,46 +109,49 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
       case FABI_F80_F32:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
-
-        fmov(v0.S(), GetSrc(IROp->Args[0].ID()).S()) ;
-        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        PushDynamicRegsAndLR(TMP1);
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
+        fmov(ARMEmitter::SReg::s0, Src1.S());
+        ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<__uint128_t, float>(x0);
+        GenerateIndirectRuntimeCall<__uint128_t, float>(ARMEmitter::Reg::r0);
 #else
-        blr(x0);
+        blr(ARMEmitter::Reg::r0);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        eor(GetDst(Node).V16B(), GetDst(Node).V16B(), GetDst(Node).V16B());
-        ins(GetDst(Node).V2D(), 0, x0);
-        ins(GetDst(Node).V8H(), 4, w1);
+        const auto Dst = GetVReg(Node);
+        eor(Dst.Q(), Dst.Q(), Dst.Q());
+        ins(ARMEmitter::SubRegSize::i64Bit, Dst, 0, ARMEmitter::Reg::r0);
+        ins(ARMEmitter::SubRegSize::i16Bit, Dst, 4, ARMEmitter::Reg::r1);
       }
       break;
 
       case FABI_F80_F64:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        mov(v0.D(), GetSrc(IROp->Args[0].ID()).D());
-        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
+        mov(ARMEmitter::DReg::d0, Src1.D());
+        ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<__uint128_t, double>(x0);
+        GenerateIndirectRuntimeCall<__uint128_t, double>(ARMEmitter::Reg::r0);
 #else
-        blr(x0);
+        blr(ARMEmitter::Reg::r0);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        eor(GetDst(Node).V16B(), GetDst(Node).V16B(), GetDst(Node).V16B());
-        ins(GetDst(Node).V2D(), 0, x0);
-        ins(GetDst(Node).V8H(), 4, w1);
+        const auto Dst = GetVReg(Node);
+        eor(Dst.Q(), Dst.Q(), Dst.Q());
+        ins(ARMEmitter::SubRegSize::i64Bit, Dst, 0, ARMEmitter::Reg::r0);
+        ins(ARMEmitter::SubRegSize::i16Bit, Dst, 4, ARMEmitter::Reg::r1);
       }
       break;
 
@@ -157,264 +159,296 @@ void Arm64JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
       case FABI_F80_I32: {
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
+        const auto Src1 = GetReg(IROp->Args[0].ID());
         if (Info.ABI == FABI_F80_I16) {
-          uxth(w0, GetReg<RA_32>(IROp->Args[0].ID()));
+          uxth(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r0, Src1);
         }
         else {
-          mov(w0, GetReg<RA_32>(IROp->Args[0].ID()));
+          mov(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r0, Src1);
         }
-        ldr(x1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        ldr(ARMEmitter::XReg::x1, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<__uint128_t, uint32_t>(x1);
+        GenerateIndirectRuntimeCall<__uint128_t, uint32_t>(ARMEmitter::Reg::r1);
 #else
-        blr(x1);
+        blr(ARMEmitter::Reg::r1);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        eor(GetDst(Node).V16B(), GetDst(Node).V16B(), GetDst(Node).V16B());
-        ins(GetDst(Node).V2D(), 0, x0);
-        ins(GetDst(Node).V8H(), 4, w1);
+        const auto Dst = GetVReg(Node);
+        eor(Dst.Q(), Dst.Q(), Dst.Q());
+        ins(ARMEmitter::SubRegSize::i64Bit, Dst, 0, ARMEmitter::Reg::r0);
+        ins(ARMEmitter::SubRegSize::i16Bit, Dst, 4, ARMEmitter::Reg::r1);
       }
       break;
 
       case FABI_F32_F80:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
-        umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r0, Src1, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r1, Src1, 4);
+
+        ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<float, uint64_t, uint64_t>(x2);
+        GenerateIndirectRuntimeCall<float, uint64_t, uint64_t>(ARMEmitter::Reg::r2);
 #else
-        blr(x2);
+        blr(ARMEmitter::Reg::r2);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        fmov(GetDst(Node).S(), v0.S());
+        const auto Dst = GetVReg(Node);
+        fmov(Dst.S(), ARMEmitter::SReg::s0);
       }
       break;
 
       case FABI_F64_F80:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
-        umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r0, Src1, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r1, Src1, 4);
+
+        ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<double, uint64_t, uint64_t>(x2);
+        GenerateIndirectRuntimeCall<double, uint64_t, uint64_t>(ARMEmitter::Reg::r2);
 #else
-        blr(x2);
+        blr(ARMEmitter::Reg::r2);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        mov(GetDst(Node).D(), v0.D());
+        const auto Dst = GetVReg(Node);
+        mov(Dst.D(), ARMEmitter::DReg::d0);
       }
       break;
 
       case FABI_F64_F64: {
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        mov(v0.D(), GetSrc(IROp->Args[0].ID()).D());
-        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
+
+        mov(ARMEmitter::DReg::d0, Src1.D());
+        ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<double, double>(x0);
+        GenerateIndirectRuntimeCall<double, double>(ARMEmitter::Reg::r0);
 #else
-        blr(x0);
+        blr(ARMEmitter::Reg::r0);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        mov(GetDst(Node).D(), v0.D());
-
+        const auto Dst = GetVReg(Node);
+        mov(Dst.D(), ARMEmitter::DReg::d0);
       }
       break;
 
       case FABI_F64_F64_F64: {
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        mov(v0.D(), GetSrc(IROp->Args[0].ID()).D());
-        mov(v1.D(), GetSrc(IROp->Args[1].ID()).D());
-        ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
+        const auto Src2 = GetVReg(IROp->Args[1].ID());
+
+        mov(ARMEmitter::DReg::d0, Src1.D());
+        mov(ARMEmitter::DReg::d1, Src2.D());
+        ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<double, double, double>(x0);
+        GenerateIndirectRuntimeCall<double, double, double>(ARMEmitter::Reg::r0);
 #else
-        blr(x0);
+        blr(ARMEmitter::Reg::r0);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        mov(GetDst(Node).D(), v0.D());
-
+        const auto Dst = GetVReg(Node);
+        mov(Dst.D(), ARMEmitter::DReg::d0);
       }
       break;
 
       case FABI_I16_F80:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
-        umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r0, Src1, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r1, Src1, 4);
+
+        ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<uint32_t, uint64_t, uint64_t>(x2);
+        GenerateIndirectRuntimeCall<uint32_t, uint64_t, uint64_t>(ARMEmitter::Reg::r2);
 #else
-        blr(x2);
+        blr(ARMEmitter::Reg::r2);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        uxth(GetReg<RA_64>(Node), x0);
+        const auto Dst = GetReg(Node);
+        uxth(ARMEmitter::Size::i64Bit, Dst, ARMEmitter::Reg::r0);
       }
       break;
       case FABI_I32_F80:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
-        umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r0, Src1, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r1, Src1, 4);
+
+        ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<uint32_t, uint64_t, uint64_t>(x2);
+        GenerateIndirectRuntimeCall<uint32_t, uint64_t, uint64_t>(ARMEmitter::Reg::r2);
 #else
-        blr(x2);
+        blr(ARMEmitter::Reg::r2);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        mov(GetReg<RA_32>(Node), w0);
+        const auto Dst = GetReg(Node);
+        mov(ARMEmitter::Size::i32Bit, Dst, ARMEmitter::Reg::r0);
       }
       break;
       case FABI_I64_F80:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
-        umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r0, Src1, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r1, Src1, 4);
+
+        ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t>(x2);
+        GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r2);
 #else
-        blr(x2);
+        blr(ARMEmitter::Reg::r2);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        mov(GetReg<RA_64>(Node), x0);
+        const auto Dst = GetReg(Node);
+        mov(ARMEmitter::Size::i64Bit, Dst, ARMEmitter::Reg::r0);
       }
       break;
       case FABI_I64_F80_F80:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
-        umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
+        const auto Src2 = GetVReg(IROp->Args[1].ID());
 
-        umov(x2, GetSrc(IROp->Args[1].ID()).V2D(), 0);
-        umov(w3, GetSrc(IROp->Args[1].ID()).V8H(), 4);
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r0, Src1, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r1, Src1, 4);
 
-        ldr(x4, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r2, Src2, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r3, Src2, 4);
+
+        ldr(ARMEmitter::XReg::x4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t, uint64_t>(x4);
+        GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r4);
 #else
-        blr(x4);
+        blr(ARMEmitter::Reg::r4);
 #endif
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        mov(GetReg<RA_64>(Node), x0);
+        const auto Dst = GetReg(Node);
+        mov(ARMEmitter::Size::i64Bit, Dst, ARMEmitter::Reg::r0);
       }
       break;
       case FABI_F80_F80:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
-        umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
 
-        ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r0, Src1, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r1, Src1, 4);
+
+        ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<__uint128_t, uint64_t, uint64_t>(x2);
+        GenerateIndirectRuntimeCall<__uint128_t, uint64_t, uint64_t>(ARMEmitter::Reg::r2);
 #else
-        blr(x2);
+        blr(ARMEmitter::Reg::r2);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        eor(GetDst(Node).V16B(), GetDst(Node).V16B(), GetDst(Node).V16B());
-        ins(GetDst(Node).V2D(), 0, x0);
-        ins(GetDst(Node).V8H(), 4, w1);
+        const auto Dst = GetVReg(Node);
+        eor(Dst.Q(), Dst.Q(), Dst.Q());
+        ins(ARMEmitter::SubRegSize::i64Bit, Dst, 0, ARMEmitter::Reg::r0);
+        ins(ARMEmitter::SubRegSize::i16Bit, Dst, 4, ARMEmitter::Reg::r1);
       }
       break;
       case FABI_F80_F80_F80:{
         SpillStaticRegs();
 
-        PushDynamicRegsAndLR();
+        PushDynamicRegsAndLR(TMP1);
 
-        umov(x0, GetSrc(IROp->Args[0].ID()).V2D(), 0);
-        umov(w1, GetSrc(IROp->Args[0].ID()).V8H(), 4);
+        const auto Src1 = GetVReg(IROp->Args[0].ID());
+        const auto Src2 = GetVReg(IROp->Args[1].ID());
 
-        umov(x2, GetSrc(IROp->Args[1].ID()).V2D(), 0);
-        umov(w3, GetSrc(IROp->Args[1].ID()).V8H(), 4);
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r0, Src1, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r1, Src1, 4);
 
-        ldr(x4, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])));
+        umov<ARMEmitter::SubRegSize::i64Bit>(ARMEmitter::Reg::r2, Src2, 0);
+        umov<ARMEmitter::SubRegSize::i16Bit>(ARMEmitter::Reg::r3, Src2, 4);
+
+        ldr(ARMEmitter::XReg::x4, STATE_PTR(CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex]));
 #ifdef VIXL_SIMULATOR
-        GenerateIndirectRuntimeCall<__uint128_t, uint64_t, uint64_t, uint64_t, uint64_t>(x4);
+        GenerateIndirectRuntimeCall<__uint128_t, uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r4);
 #else
-        blr(x4);
+        blr(ARMEmitter::Reg::r4);
 #endif
 
         PopDynamicRegsAndLR();
 
         FillStaticRegs();
 
-        eor(GetDst(Node).V16B(), GetDst(Node).V16B(), GetDst(Node).V16B());
-        ins(GetDst(Node).V2D(), 0, x0);
-        ins(GetDst(Node).V8H(), 4, w1);
+        const auto Dst = GetVReg(Node);
+        eor(Dst.Q(), Dst.Q(), Dst.Q());
+        ins(ARMEmitter::SubRegSize::i64Bit, Dst, 0, ARMEmitter::Reg::r0);
+        ins(ARMEmitter::SubRegSize::i16Bit, Dst, 4, ARMEmitter::Reg::r1);
       }
       break;
-
       case FABI_UNKNOWN:
       default:
 #if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
@@ -434,7 +468,6 @@ static uint64_t Arm64JITCore_ExitFunctionLink(FEXCore::Core::CpuStateFrame *Fram
   auto HostCode = Thread->LookupCache->FindBlock(GuestRip);
 
   if (!HostCode) {
-    //fmt::print("ExitFunctionLink: Aborting, {:X} not in cache\n", GuestRip);
     Frame->State.rip = GuestRip;
     return Frame->Pointers.Common.DispatcherLoopTop;
   }
@@ -443,25 +476,22 @@ static uint64_t Arm64JITCore_ExitFunctionLink(FEXCore::Core::CpuStateFrame *Fram
   auto LinkerAddress = Frame->Pointers.Common.ExitFunctionLinker;
 
   auto offset = HostCode/4 - branch/4;
-  if (IsInt26(offset)) {
+  if (vixl::IsInt26(offset)) {
     // optimal case - can branch directly
     // patch the code
-    vixl::aarch64::Assembler emit((uint8_t*)(branch), 24);
-    vixl::CodeBufferCheckScope scope(&emit, 24, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
+    FEXCore::ARMEmitter::Emitter emit((uint8_t*)(branch), 24);
     emit.b(offset);
-    emit.FinalizeCode();
-    vixl::aarch64::CPU::EnsureIAndDCacheCoherency((void*)branch, 24);
+    FEXCore::ARMEmitter::Emitter::ClearICache((void*)branch, 24);
 
     // Add de-linking handler
     Context::Context::ThreadAddBlockLink(Thread, GuestRip, (uintptr_t)record, [branch, LinkerAddress]{
-      vixl::aarch64::Assembler emit((uint8_t*)(branch), 24);
-      vixl::CodeBufferCheckScope scope(&emit, 24, vixl::CodeBufferCheckScope::kDontReserveBufferSpace, vixl::CodeBufferCheckScope::kNoAssert);
-      Literal l_BranchHost{LinkerAddress};
-      emit.ldr(x0, &l_BranchHost);
-      emit.blr(x0);
-      emit.place(&l_BranchHost);
-      emit.FinalizeCode();
-      vixl::aarch64::CPU::EnsureIAndDCacheCoherency((void*)branch, 24);
+      FEXCore::ARMEmitter::Emitter emit((uint8_t*)(branch), 24);
+      FEXCore::ARMEmitter::ForwardLabel l_BranchHost;
+      emit.ldr(FEXCore::ARMEmitter::XReg::x0, &l_BranchHost);
+      emit.blr(FEXCore::ARMEmitter::Reg::r0);
+      emit.Bind(&l_BranchHost);
+      emit.dc64(LinkerAddress);
+      FEXCore::ARMEmitter::Emitter::ClearICache((void*)branch, 24);
     });
   } else {
     // fallback case - do a soft-er link by patching the pointer
@@ -476,7 +506,7 @@ static uint64_t Arm64JITCore_ExitFunctionLink(FEXCore::Core::CpuStateFrame *Fram
   return HostCode;
 }
 
-void Arm64JITCore::Op_NoOp(IR::IROp_Header *IROp, IR::NodeID Node) {
+void Arm64JITCore::Op_NoOp(IR::IROp_Header const *IROp, IR::NodeID Node) {
 }
 
 Arm64JITCore::Arm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
@@ -487,10 +517,6 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::Intern
 
   RAPass = Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA");
 
-#if DEBUG
-  Decoder.AppendVisitor(&Disasm)
-#endif
-
   uint32_t NumUsedGPRs = NumGPRs;
   uint32_t NumUsedGPRPairs = NumGPRPairs;
   uint32_t UsedRegisterCount = RegisterCount;
@@ -558,7 +584,6 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::Intern
   }
 
   // Must be done after Dispatcher init
-  SetAllowAssembler(true);
   ClearCache();
 }
 
@@ -581,16 +606,15 @@ void Arm64JITCore::InitializeSignalHandlers(FEXCore::Context::Context *CTX) {
 
 void Arm64JITCore::EmitDetectionString() {
   const char JITString[] = "FEXJIT::Arm64JITCore::";
-  auto Buffer = GetBuffer();
-  Buffer->EmitString(JITString);
-  Buffer->Align();
+  EmitString(JITString);
+  Align();
 }
 
 void Arm64JITCore::ClearCache() {
   // Get the backing code buffer
 
   auto CodeBuffer = GetEmptyCodeBuffer();
-  *GetBuffer() = vixl::CodeBuffer(CodeBuffer->Ptr, CodeBuffer->Size);
+  SetBuffer(CodeBuffer->Ptr, CodeBuffer->Size);
   EmitDetectionString();
 }
 
@@ -598,84 +622,6 @@ Arm64JITCore::~Arm64JITCore() {
 
 }
 
-IR::PhysicalRegister Arm64JITCore::GetPhys(IR::NodeID Node) const {
-  auto PhyReg = RAData->GetNodeRegister(Node);
-
-  LOGMAN_THROW_A_FMT(!PhyReg.IsInvalid(), "Couldn't Allocate register for node: ssa{}. Class: {}", Node, PhyReg.Class);
-
-  return PhyReg;
-}
-
-template<>
-aarch64::Register Arm64JITCore::GetReg<Arm64JITCore::RA_32>(IR::NodeID Node) const {
-  auto Reg = GetPhys(Node);
-
-  LOGMAN_THROW_AA_FMT(Reg.Class == IR::GPRFixedClass.Val || Reg.Class == IR::GPRClass.Val, "Unexpected Class: {}", Reg.Class);
-
-  if (Reg.Class == IR::GPRFixedClass.Val) {
-    return SRA64[Reg.Reg].W();
-  } else if (Reg.Class == IR::GPRClass.Val) {
-    return RA64[Reg.Reg].W();
-  }
-
-  FEX_UNREACHABLE;
-}
-
-template<>
-aarch64::Register Arm64JITCore::GetReg<Arm64JITCore::RA_64>(IR::NodeID Node) const {
-  auto Reg = GetPhys(Node);
-
-  LOGMAN_THROW_AA_FMT(Reg.Class == IR::GPRFixedClass.Val || Reg.Class == IR::GPRClass.Val, "Unexpected Class: {}", Reg.Class);
-
-  if (Reg.Class == IR::GPRFixedClass.Val) {
-    return SRA64[Reg.Reg];
-  } else if (Reg.Class == IR::GPRClass.Val) {
-    return RA64[Reg.Reg];
-  }
-
-  FEX_UNREACHABLE;
-}
-
-template<>
-std::pair<aarch64::Register, aarch64::Register> Arm64JITCore::GetSrcPair<Arm64JITCore::RA_32>(IR::NodeID Node) const {
-  uint32_t Reg = GetPhys(Node).Reg;
-  return RA32Pair[Reg];
-}
-
-template<>
-std::pair<aarch64::Register, aarch64::Register> Arm64JITCore::GetSrcPair<Arm64JITCore::RA_64>(IR::NodeID Node) const {
-  uint32_t Reg = GetPhys(Node).Reg;
-  return RA64Pair[Reg];
-}
-
-aarch64::VRegister Arm64JITCore::GetSrc(IR::NodeID Node) const {
-  auto Reg = GetPhys(Node);
-
-  LOGMAN_THROW_AA_FMT(Reg.Class == IR::FPRFixedClass.Val || Reg.Class == IR::FPRClass.Val, "Unexpected Class: {}", Reg.Class);
-
-  if (Reg.Class == IR::FPRFixedClass.Val) {
-    return SRAFPR[Reg.Reg];
-  } else if (Reg.Class == IR::FPRClass.Val) {
-    return RAFPR[Reg.Reg];
-  }
-
-  FEX_UNREACHABLE;
-}
-
-aarch64::VRegister Arm64JITCore::GetDst(IR::NodeID Node) const {
-  auto Reg = GetPhys(Node);
-
-  LOGMAN_THROW_AA_FMT(Reg.Class == IR::FPRFixedClass.Val || Reg.Class == IR::FPRClass.Val, "Unexpected Class: {}", Reg.Class);
-
-  if (Reg.Class == IR::FPRFixedClass.Val) {
-    return SRAFPR[Reg.Reg];
-  } else if (Reg.Class == IR::FPRClass.Val) {
-    return RAFPR[Reg.Reg];
-  }
-
-  FEX_UNREACHABLE;
-}
-
 bool Arm64JITCore::IsInlineConstant(const IR::OrderedNodeWrapper& WNode, uint64_t* Value) const {
   auto OpHeader = IR->GetOp<IR::IROp_Header>(WNode);
 
@@ -713,7 +659,6 @@ FEXCore::IR::RegisterClassType Arm64JITCore::GetRegClass(IR::NodeID Node) const
   return FEXCore::IR::RegisterClassType {GetPhys(Node).Class};
 }
 
-
 bool Arm64JITCore::IsFPR(IR::NodeID Node) const {
   auto Class = GetRegClass(Node);
 
@@ -731,7 +676,8 @@ void *Arm64JITCore::CompileCode(uint64_t Entry,
                                 FEXCore::Core::DebugData *DebugData,
                                 FEXCore::IR::RegisterAllocationData *RAData,
                                 bool GDBEnabled) {
-  using namespace aarch64;
+  FEXCORE_PROFILE_SCOPED("Arm64::CompileCode");
+
   JumpTargets.clear();
   uint32_t SSACount = IR->GetSSACount();
 
@@ -739,9 +685,13 @@ void *Arm64JITCore::CompileCode(uint64_t Entry,
   this->RAData = RAData;
   this->DebugData = DebugData;
 
-  #ifndef NDEBUG
-  LoadConstant(x0, Entry);
-  #endif
+#ifdef VIXL_DISASSEMBLER
+  const auto DisasmBegin = GetCursorAddress<const vixl::aarch64::Instruction*>();
+#endif
+
+#ifndef NDEBUG
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, Entry);
+#endif
 
   this->IR = IR;
 
@@ -775,7 +725,7 @@ void *Arm64JITCore::CompileCode(uint64_t Entry,
 
   if (GDBEnabled) {
     auto GDBSize = CTX->Dispatcher->GenerateGDBPauseCheck(GuestEntry, Entry);
-    GetBuffer()->CursorForward(GDBSize);
+    CursorIncrement(GDBSize);
   }
 
   //LOGMAN_THROW_A_FMT(RAData->HasFullRA(), "Arm64 JIT only works with RA");
@@ -785,11 +735,11 @@ void *Arm64JITCore::CompileCode(uint64_t Entry,
   if (SpillSlots) {
     const auto TotalSpillSlotsSize = SpillSlots * MaxSpillSlotSize;
 
-    if (IsImmAddSub(TotalSpillSlotsSize)) {
-      sub(sp, sp, TotalSpillSlotsSize);
+    if (vixl::aarch64::Assembler::IsImmAddSub(TotalSpillSlotsSize)) {
+      sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, TotalSpillSlotsSize);
     } else {
-      LoadConstant(x0, TotalSpillSlotsSize);
-      sub(sp, sp, x0);
+      LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, TotalSpillSlotsSize);
+      sub(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::rsp, ARMEmitter::XReg::rsp, ARMEmitter::XReg::x0, ARMEmitter::ExtendedType::LSL_64, 0);
     }
   }
 
@@ -814,7 +764,7 @@ void *Arm64JITCore::CompileCode(uint64_t Entry,
       }
       PendingTargetLabel = nullptr;
 
-      bind(&IsTarget->second);
+      Bind(&IsTarget->second);
     }
 
     for (auto [CodeNode, IROp] : IR->GetCode(BlockNode)) {
@@ -840,10 +790,13 @@ void *Arm64JITCore::CompileCode(uint64_t Entry,
   }
   PendingTargetLabel = nullptr;
 
-  FinalizeCode();
-
   auto CodeEnd = GetCursorAddress<uint8_t *>();
-  CPU.EnsureIAndDCacheCoherency(GuestEntry, CodeEnd - GuestEntry);
+  ClearICache(GuestEntry, CodeEnd - GuestEntry);
+
+#ifdef VIXL_DISASSEMBLER
+  const auto DisasmEnd = GetCursorAddress<const vixl::aarch64::Instruction*>();
+  Disasm.DisassembleBuffer(DisasmBegin, DisasmEnd);
+#endif
 
   if (DebugData) {
     DebugData->HostCodeSize = CodeEnd - GuestEntry;
@@ -862,12 +815,12 @@ void Arm64JITCore::ResetStack() {
 
   const auto TotalSpillSlotsSize = SpillSlots * MaxSpillSlotSize;
 
-  if (IsImmAddSub(TotalSpillSlotsSize)) {
-    add(sp, sp, TotalSpillSlotsSize);
+  if (vixl::aarch64::Assembler::IsImmAddSub(TotalSpillSlotsSize)) {
+    add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, TotalSpillSlotsSize);
   } else {
-   // Too big to fit in a 12bit immediate
-   LoadConstant(x0, TotalSpillSlotsSize);
-   add(sp, sp, x0);
+    // Too big to fit in a 12bit immediate
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, TotalSpillSlotsSize);
+    add(ARMEmitter::Size::i64Bit, ARMEmitter::XReg::rsp, ARMEmitter::XReg::rsp, ARMEmitter::XReg::x0, ARMEmitter::ExtendedType::LSL_64, 0);
   }
 }
 

External/FEXCore/Source/Interface/Core/JIT/Arm64/JITClass.h

@@ -8,6 +8,7 @@ $end_info$
 
 #include <FEXCore/IR/RegisterAllocationData.h>
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 
 #include <aarch64/assembler-aarch64.h>
@@ -28,9 +29,6 @@ namespace FEXCore::Core {
 }
 
 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-
 class Arm64JITCore final : public CPUBackend, public Arm64Emitter  {
 public:
   explicit Arm64JITCore(FEXCore::Context::Context *ctx,
@@ -58,12 +56,12 @@ private:
   FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
   const bool HostSupportsSVE{};
 
-  Label *PendingTargetLabel;
+  ARMEmitter::BiDirectionalLabel *PendingTargetLabel;
   FEXCore::Context::Context *CTX;
   FEXCore::IR::IRListView const *IR;
   uint64_t Entry;
 
-  std::map<IR::NodeID, aarch64::Label> JumpTargets;
+  std::map<IR::NodeID, ARMEmitter::BiDirectionalLabel> JumpTargets;
 
   /**
    * @name Register Allocation
@@ -86,38 +84,72 @@ private:
   constexpr static uint8_t RA_64 = 1;
   constexpr static uint8_t RA_FPR = 2;
 
-  template<uint8_t RAType>
-  [[nodiscard]] aarch64::Register GetReg(IR::NodeID Node) const;
+  [[nodiscard]] FEXCore::ARMEmitter::Register GetReg(IR::NodeID Node) const {
+    const auto Reg = GetPhys(Node);
 
-  template<>
-  [[nodiscard]] aarch64::Register GetReg<RA_32>(IR::NodeID Node) const;
-  template<>
-  [[nodiscard]] aarch64::Register GetReg<RA_64>(IR::NodeID Node) const;
+    LOGMAN_THROW_AA_FMT(Reg.Class == IR::GPRFixedClass.Val || Reg.Class == IR::GPRClass.Val, "Unexpected Class: {}", Reg.Class);
 
-  template<uint8_t RAType>
-  [[nodiscard]] std::pair<aarch64::Register, aarch64::Register> GetSrcPair(IR::NodeID Node) const;
+    if (Reg.Class == IR::GPRFixedClass.Val) {
+      return SRA64[Reg.Reg];
+    } else if (Reg.Class == IR::GPRClass.Val) {
+      return RA64[Reg.Reg];
+    }
 
-  template<>
-  [[nodiscard]] std::pair<aarch64::Register, aarch64::Register> GetSrcPair<RA_32>(IR::NodeID Node) const;
-  template<>
-  [[nodiscard]] std::pair<aarch64::Register, aarch64::Register> GetSrcPair<RA_64>(IR::NodeID Node) const;
+    FEX_UNREACHABLE;
+  }
 
-  [[nodiscard]] aarch64::VRegister GetSrc(IR::NodeID Node) const;
-  [[nodiscard]] aarch64::VRegister GetDst(IR::NodeID Node) const;
+  [[nodiscard]] FEXCore::ARMEmitter::VRegister GetVReg(IR::NodeID Node) const {
+    const auto Reg = GetPhys(Node);
+
+    LOGMAN_THROW_AA_FMT(Reg.Class == IR::FPRFixedClass.Val || Reg.Class == IR::FPRClass.Val, "Unexpected Class: {}", Reg.Class);
+
+    if (Reg.Class == IR::FPRFixedClass.Val) {
+      return SRAFPR[Reg.Reg];
+    } else if (Reg.Class == IR::FPRClass.Val) {
+      return RAFPR[Reg.Reg];
+    }
+
+    FEX_UNREACHABLE;
+  }
+
+  [[nodiscard]] std::pair<FEXCore::ARMEmitter::Register, FEXCore::ARMEmitter::Register> GetRegPair(IR::NodeID Node) const {
+    const auto Reg = GetPhys(Node);
+
+    LOGMAN_THROW_AA_FMT(Reg.Class == IR::GPRPairClass.Val, "Unexpected Class: {}", Reg.Class);
+
+    return RA64Pair[Reg.Reg];
+  }
 
   [[nodiscard]] FEXCore::IR::RegisterClassType GetRegClass(IR::NodeID Node) const;
 
-  [[nodiscard]] IR::PhysicalRegister GetPhys(IR::NodeID Node) const;
+  [[nodiscard]] IR::PhysicalRegister GetPhys(IR::NodeID Node) const {
+    auto PhyReg = RAData->GetNodeRegister(Node);
+
+    LOGMAN_THROW_A_FMT(!PhyReg.IsInvalid(), "Couldn't Allocate register for node: ssa{}. Class: {}", Node, PhyReg.Class);
+
+    return PhyReg;
+  }
 
   [[nodiscard]] bool IsFPR(IR::NodeID Node) const;
   [[nodiscard]] bool IsGPR(IR::NodeID Node) const;
 
-  [[nodiscard]] MemOperand GenerateMemOperand(uint8_t AccessSize,
-                                              aarch64::Register Base,
+  [[nodiscard]] FEXCore::ARMEmitter::ExtendedMemOperand GenerateMemOperand(uint8_t AccessSize,
+                                              FEXCore::ARMEmitter::Register Base,
                                               IR::OrderedNodeWrapper Offset,
                                               IR::MemOffsetType OffsetType,
                                               uint8_t OffsetScale);
 
+  // NOTE: Will use TMP1 as a way to encode immediates that happen to fall outside
+  //       the limits of the scalar plus immediate variant of SVE load/stores.
+  //
+  //       TMP1 is safe to use again once this memory operand is used with its
+  //       equivalent loads or stores that this was called for.
+  [[nodiscard]] FEXCore::ARMEmitter::SVEMemOperand GenerateSVEMemOperand(uint8_t AccessSize,
+                                                    FEXCore::ARMEmitter::Register Base,
+                                                    IR::OrderedNodeWrapper Offset,
+                                                    IR::MemOffsetType OffsetType,
+                                                    uint8_t OffsetScale);
+
   [[nodiscard]] bool IsInlineConstant(const IR::OrderedNodeWrapper& Node, uint64_t* Value = nullptr) const;
   [[nodiscard]] bool IsInlineEntrypointOffset(const IR::OrderedNodeWrapper& WNode, uint64_t* Value) const;
 
@@ -151,7 +183,8 @@ private:
      * @brief A literal pair relocation object for named symbol literals
      */
     struct NamedSymbolLiteralPair {
-      Literal<uint64_t> Lit;
+      ARMEmitter::ForwardLabel Loc;
+      uint64_t Lit;
       Relocation MoveABI{};
     };
 
@@ -161,7 +194,7 @@ private:
      * @param Reg - The GPR to move the thunk handler in to
      * @param Sum - The hash of the thunk
      */
-    void InsertNamedThunkRelocation(vixl::aarch64::Register Reg, const IR::SHA256Sum &Sum);
+    void InsertNamedThunkRelocation(ARMEmitter::Register Reg, const IR::SHA256Sum &Sum);
 
     /**
      * @brief Inserts a guest GPR move relocation
@@ -169,7 +202,7 @@ private:
      * @param Reg - The GPR to move the guest RIP in to
      * @param Constant - The guest RIP that will be relocated
      */
-    void InsertGuestRIPMove(vixl::aarch64::Register Reg, uint64_t Constant);
+    void InsertGuestRIPMove(ARMEmitter::Register Reg, uint64_t Constant);
 
     /**
      * @brief Inserts a named symbol as a literal in memory
@@ -202,7 +235,7 @@ private:
   */
   uint8_t *GuestEntry{};
 
-  using OpHandler = void (Arm64JITCore::*)(IR::IROp_Header *IROp, IR::NodeID Node);
+  using OpHandler = void (Arm64JITCore::*)(IR::IROp_Header const *IROp, IR::NodeID Node);
   std::array<OpHandler, IR::IROps::OP_LAST + 1> OpHandlers {};
   void RegisterALUHandlers();
   void RegisterAtomicHandlers();
@@ -214,7 +247,7 @@ private:
   void RegisterMoveHandlers();
   void RegisterVectorHandlers();
   void RegisterEncryptionHandlers();
-#define DEF_OP(x) void Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 
   ///< Unhandled handler
   DEF_OP(Unhandled);
@@ -334,8 +367,6 @@ private:
   DEF_OP(StoreMemTSO);
   DEF_OP(ParanoidLoadMemTSO);
   DEF_OP(ParanoidStoreMemTSO);
-  DEF_OP(VLoadMemElement);
-  DEF_OP(VStoreMemElement);
   DEF_OP(CacheLineClear);
   DEF_OP(CacheLineZero);
 
@@ -419,8 +450,6 @@ private:
   DEF_OP(VInsElement);
   DEF_OP(VDupElement);
   DEF_OP(VExtr);
-  DEF_OP(VSLI);
-  DEF_OP(VSRI);
   DEF_OP(VUShrI);
   DEF_OP(VSShrI);
   DEF_OP(VShlI);

External/FEXCore/Source/Interface/Core/JIT/Arm64/MiscOps.cpp

@@ -5,13 +5,12 @@ $end_info$
 */
 
 #include <syscall.h>
+#include "Interface/Core/ArchHelpers/CodeEmitter/Emitter.h"
 #include "Interface/Core/JIT/Arm64/JITClass.h"
 #include "FEXCore/Debug/InternalThreadState.h"
 
 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 
 DEF_OP(GuestOpcode) {
   auto Op = IROp->C<IR::IROp_GuestOpcode>();
@@ -23,13 +22,13 @@ DEF_OP(Fence) {
   auto Op = IROp->C<IR::IROp_Fence>();
   switch (Op->Fence) {
     case IR::Fence_Load.Val:
-      dmb(FullSystem, BarrierReads);
+      dmb(FEXCore::ARMEmitter::BarrierScope::LD);
       break;
     case IR::Fence_LoadStore.Val:
-      dmb(FullSystem, BarrierAll);
+      dmb(FEXCore::ARMEmitter::BarrierScope::SY);
       break;
     case IR::Fence_Store.Val:
-      dmb(FullSystem, BarrierWrites);
+      dmb(FEXCore::ARMEmitter::BarrierScope::ST);
       break;
     default: LOGMAN_MSG_A_FMT("Unknown Fence: {}", Op->Fence); break;
   }
@@ -41,116 +40,118 @@ DEF_OP(Break) {
   // First we must reset the stack
   ResetStack();
 
-  LoadConstant(w1, 1);
-  strb(w1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.FaultToTopAndGeneratedException)));
-  LoadConstant(w1, Op->Reason.Signal);
-  strb(w1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.Signal)));
-  LoadConstant(w1, Op->Reason.TrapNumber);
-  str(w1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.TrapNo)));
-  LoadConstant(w1, Op->Reason.si_code);
-  str(w1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.si_code)));
-  LoadConstant(x1, Op->Reason.ErrorRegister);
-  str(w1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.err_code)));
+  Core::CpuStateFrame::SynchronousFaultDataStruct State = {
+    .FaultToTopAndGeneratedException = 1,
+    .Signal = Op->Reason.Signal,
+    .TrapNo = Op->Reason.TrapNumber,
+    .si_code = Op->Reason.si_code,
+    .err_code = Op->Reason.ErrorRegister,
+  };
+
+  uint64_t Constant{};
+  memcpy(&Constant, &State, sizeof(State));
+
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, Constant);
+  str(ARMEmitter::XReg::x1, STATE, offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData));
 
   switch (Op->Reason.Signal) {
   case SIGILL:
-    ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGILL)));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGILL));
     br(TMP1);
     break;
   case SIGTRAP:
-    ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP)));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
     br(TMP1);
     break;
   case SIGSEGV:
-    ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGSEGV)));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGSEGV));
     br(TMP1);
     break;
   default:
-    ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP)));
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
     br(TMP1);
     break;
   }
 }
 
 DEF_OP(GetRoundingMode) {
-  auto Dst = GetReg<RA_64>(Node);
-  mrs(Dst, FPCR);
-  lsr(Dst, Dst,  22);
+  auto Dst = GetReg(Node);
+  mrs(Dst, ARMEmitter::SystemRegister::FPCR);
+  lsr(ARMEmitter::Size::i64Bit, Dst, Dst, 22);
 
   // FTZ is already in the correct location
   // Rounding mode is different
-  and_(TMP1, Dst, 0b11);
+  and_(ARMEmitter::Size::i64Bit, TMP1, Dst, 0b11);
 
-  cmp(TMP1, 1);
-  LoadConstant(TMP3, IR::ROUND_MODE_POSITIVE_INFINITY);
-  csel(TMP2, TMP3, xzr, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, 1);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, IR::ROUND_MODE_POSITIVE_INFINITY);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, ARMEmitter::Reg::zr, ARMEmitter::Condition::CC_EQ);
 
-  cmp(TMP1, 2);
-  LoadConstant(TMP3, IR::ROUND_MODE_NEGATIVE_INFINITY);
-  csel(TMP2, TMP3, TMP2, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, 2);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, IR::ROUND_MODE_NEGATIVE_INFINITY);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, TMP2, ARMEmitter::Condition::CC_EQ);
 
-  cmp(TMP1, 3);
-  LoadConstant(TMP3, IR::ROUND_MODE_TOWARDS_ZERO);
-  csel(TMP2, TMP3, TMP2, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, 3);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, IR::ROUND_MODE_TOWARDS_ZERO);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, TMP2, ARMEmitter::Condition::CC_EQ);
 
-  orr(Dst, Dst, TMP2);
+  orr(ARMEmitter::Size::i64Bit, Dst, Dst, TMP2.R());
 
-  bfi(Dst, TMP2, 0, 2);
+  bfi(ARMEmitter::Size::i64Bit, Dst, TMP2, 0, 2);
 }
 
 DEF_OP(SetRoundingMode) {
   auto Op = IROp->C<IR::IROp_SetRoundingMode>();
-  auto Src = GetReg<RA_64>(Op->RoundMode.ID());
+  auto Src = GetReg(Op->RoundMode.ID());
 
   // Setup the rounding flags correctly
-  and_(TMP1, Src, 0b11);
+  and_(ARMEmitter::Size::i64Bit, TMP1, Src, 0b11);
 
-  cmp(TMP1, IR::ROUND_MODE_POSITIVE_INFINITY);
-  LoadConstant(TMP3, 1);
-  csel(TMP2, TMP3, xzr, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, IR::ROUND_MODE_POSITIVE_INFINITY);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, 1);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, ARMEmitter::Reg::zr, ARMEmitter::Condition::CC_EQ);
 
-  cmp(TMP1, IR::ROUND_MODE_NEGATIVE_INFINITY);
-  LoadConstant(TMP3, 2);
-  csel(TMP2, TMP3, TMP2, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, IR::ROUND_MODE_NEGATIVE_INFINITY);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, 2);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, TMP2, ARMEmitter::Condition::CC_EQ);
 
-  cmp(TMP1, IR::ROUND_MODE_TOWARDS_ZERO);
-  LoadConstant(TMP3, 3);
-  csel(TMP2, TMP3, TMP2, vixl::aarch64::Condition::eq);
+  cmp(ARMEmitter::Size::i64Bit, TMP1, IR::ROUND_MODE_TOWARDS_ZERO);
+  LoadConstant(ARMEmitter::Size::i64Bit, TMP3, 3);
+  csel(ARMEmitter::Size::i64Bit, TMP2, TMP3, TMP2, ARMEmitter::Condition::CC_EQ);
 
-  mrs(TMP1, FPCR);
+  mrs(TMP1, ARMEmitter::SystemRegister::FPCR);
 
   // vixl simulator doesn't support anything beyond ties-to-even rounding
 #ifndef VIXL_SIMULATOR
   // Insert the rounding flags
-  bfi(TMP1, TMP2, 22, 2);
+  bfi(ARMEmitter::Size::i64Bit, TMP1, TMP2, 22, 2);
 #endif
 
   // Insert the FTZ flag
-  lsr(TMP2, Src, 2);
-  bfi(TMP1, TMP2, 24, 1);
+  lsr(ARMEmitter::Size::i64Bit, TMP2, Src, 2);
+  bfi(ARMEmitter::Size::i64Bit, TMP1, TMP2, 24, 1);
 
   // Now save the new FPCR
-  msr(FPCR, TMP1);
+  msr(ARMEmitter::SystemRegister::FPCR, TMP1);
 }
 
 DEF_OP(Print) {
   auto Op = IROp->C<IR::IROp_Print>();
 
-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
   SpillStaticRegs();
 
   if (IsGPR(Op->Value.ID())) {
-    mov(x0, GetReg<RA_64>(Op->Value.ID()));
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintValue)));
+    mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetReg(Op->Value.ID()));
+    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintValue));
   }
   else {
-    fmov(x0, GetSrc(Op->Value.ID()).V1D());
-    // Bug in vixl that source vector needs to b V1D rather than V2D?
-    fmov(x1, GetSrc(Op->Value.ID()).V1D(), 1);
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintVectorValue)));
+    fmov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, GetVReg(Op->Value.ID()), false);
+    fmov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, GetVReg(Op->Value.ID()), true);
+    ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintVectorValue));
   }
 
-  blr(x3);
+  blr(ARMEmitter::Reg::r3);
 
   FillStaticRegs();
   PopDynamicRegsAndLR();
@@ -170,27 +171,27 @@ DEF_OP(ProcessorID) {
   // 16bit LoadConstant to be a single instruction
   // We must always spill at least one register (x8) so this value always has a bit set
   // This gives the signal handler a value to check to see if we are in a syscall at all
-  LoadConstant(x0, SpillMask & 0xFFFF);
-  str(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, SpillMask & 0xFFFF);
+  str(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
 
   // Allocate some temporary space for storing the uint32_t CPU and Node IDs
-  sub(sp, sp, 16);
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);
 
   // Load the getcpu syscall number
-  LoadConstant(x8, SYS_getcpu);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, SYS_getcpu);
 
   // CPU pointer in x0
-  add(x0, sp, 0);
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r0, ARMEmitter::Reg::rsp, 0);
   // Node in x1
-  add(x1, sp, 4);
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, ARMEmitter::Reg::rsp, 4);
 
   svc(0);
   // On updated signal mask we can receive a signal RIGHT HERE
 
   // Load the values returned by the kernel
-  ldp(w0, w1, MemOperand(sp));
+  ldp<ARMEmitter::IndexType::OFFSET>(ARMEmitter::WReg::w0, ARMEmitter::WReg::w1, ARMEmitter::Reg::rsp);
   // Deallocate stack space
-  sub(sp, sp, 16);
+  sub(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);
 
   // Now that we are done in the syscall we need to carefully peel back the state
   // First unspill the registers from before
@@ -198,14 +199,13 @@ DEF_OP(ProcessorID) {
 
   // Now the registers we've spilled are back in their original host registers
   // We can safely claim we are no longer in a syscall
-  str(xzr, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo)));
-
+  str(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
 
   // Now store the result in the destination in the expected format
   // uint32_t Res = (node << 12) | cpu;
   // CPU is in w0
   // Node is in w1
-  orr(GetReg<RA_64>(Node), x0, Operand(x1, LSL, 12));
+  orr(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0, ARMEmitter::Reg::r1, ARMEmitter::ShiftType::LSL, 12);
 }
 
 DEF_OP(RDRAND) {
@@ -213,21 +213,21 @@ DEF_OP(RDRAND) {
 
   // Results are in x0, x1
   // Results want to be in a i64v2 vector
-  auto Dst = GetSrcPair<RA_64>(Node);
+  auto Dst = GetRegPair(Node);
 
   if (Op->GetReseeded) {
-    mrs(Dst.first, RNDRRS);
+    mrs(Dst.first, ARMEmitter::SystemRegister::RNDRRS);
   }
   else {
-    mrs(Dst.first, RNDR);
+    mrs(Dst.first, ARMEmitter::SystemRegister::RNDR);
   }
 
   // If the rng number is valid then NZCV is 0b0000, otherwise NZCV is 0b0100
-  cset(Dst.second, Condition::ne);
+  cset(ARMEmitter::Size::i64Bit, Dst.second, ARMEmitter::Condition::CC_NE);
 }
 
 DEF_OP(Yield) {
-  hint(SystemHint::YIELD);
+  yield();
 }
 
 #undef DEF_OP

External/FEXCore/Source/Interface/Core/JIT/Arm64/MoveOps.cpp

@@ -7,64 +7,37 @@ $end_info$
 #include "Interface/Core/JIT/Arm64/JITClass.h"
 
 namespace FEXCore::CPU {
-
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
+#define DEF_OP(x) void Arm64JITCore::Op_##x(IR::IROp_Header const *IROp, IR::NodeID Node)
 DEF_OP(ExtractElementPair) {
   auto Op = IROp->C<IR::IROp_ExtractElementPair>();
-  switch (Op->Header.Size) {
-    case 4: {
-      auto Src = GetSrcPair<RA_32>(Op->Pair.ID());
-      std::array<aarch64::Register, 2> Regs = {Src.first, Src.second};
-      mov (GetReg<RA_32>(Node), Regs[Op->Element]);
-      break;
-    }
-    case 8: {
-      auto Src = GetSrcPair<RA_64>(Op->Pair.ID());
-      std::array<aarch64::Register, 2> Regs = {Src.first, Src.second};
-      mov (GetReg<RA_64>(Node), Regs[Op->Element]);
-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Size"); break;
-  }
+  LOGMAN_THROW_AA_FMT(Op->Header.Size == 4 || Op->Header.Size == 8, "Invalid size");
+  const auto EmitSize = Op->Header.Size == 8 ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
+
+  const auto Src = GetRegPair(Op->Pair.ID());
+  const std::array<ARMEmitter::Register, 2> Regs = {Src.first, Src.second};
+  mov(EmitSize, GetReg(Node), Regs[Op->Element]);
 }
 
 DEF_OP(CreateElementPair) {
   auto Op = IROp->C<IR::IROp_CreateElementPair>();
-  std::pair<aarch64::Register, aarch64::Register> Dst;
-  aarch64::Register RegFirst;
-  aarch64::Register RegSecond;
-  aarch64::Register RegTmp;
+  LOGMAN_THROW_AA_FMT(IROp->ElementSize == 4 || IROp->ElementSize == 8, "Invalid size");
+  std::pair<ARMEmitter::Register, ARMEmitter::Register> Dst = GetRegPair(Node);
+  ARMEmitter::Register RegFirst = GetReg(Op->Lower.ID());
+  ARMEmitter::Register RegSecond = GetReg(Op->Upper.ID());
+  ARMEmitter::Register RegTmp = TMP1.R();
 
-  switch (IROp->ElementSize) {
-    case 4: {
-      Dst = GetSrcPair<RA_32>(Node);
-      RegFirst = GetReg<RA_32>(Op->Lower.ID());
-      RegSecond = GetReg<RA_32>(Op->Upper.ID());
-      RegTmp = w0;
-      break;
-    }
-    case 8: {
-      Dst = GetSrcPair<RA_64>(Node);
-      RegFirst = GetReg<RA_64>(Op->Lower.ID());
-      RegSecond = GetReg<RA_64>(Op->Upper.ID());
-      RegTmp = x0;
-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Size"); break;
-  }
+  const auto EmitSize = IROp->ElementSize == 8 ? ARMEmitter::Size::i64Bit : ARMEmitter::Size::i32Bit;
 
-  if (Dst.first.GetCode() != RegSecond.GetCode()) {
-    mov(Dst.first, RegFirst);
-    mov(Dst.second, RegSecond);
-  } else if (Dst.second.GetCode() != RegFirst.GetCode()) {
-    mov(Dst.second, RegSecond);
-    mov(Dst.first, RegFirst);
+  if (Dst.first.Idx() != RegSecond.Idx()) {
+    mov(EmitSize, Dst.first, RegFirst);
+    mov(EmitSize, Dst.second, RegSecond);
+  } else if (Dst.second.Idx() != RegFirst.Idx()) {
+    mov(EmitSize, Dst.second, RegSecond);
+    mov(EmitSize, Dst.first, RegFirst);
   } else {
-    mov(RegTmp, RegFirst);
-    mov(Dst.second, RegSecond);
-    mov(Dst.first, RegTmp);
+    mov(EmitSize, RegTmp, RegFirst);
+    mov(EmitSize, Dst.second, RegSecond);
+    mov(EmitSize, Dst.first, RegTmp);
   }
 }
 

External/FEXCore/Source/Interface/Core/JIT/x86_64/ALUOps.cpp

@@ -1143,26 +1143,59 @@ DEF_OP(Select) {
 }
 
 DEF_OP(VExtractToGPR) {
-  auto Op = IROp->C<IR::IROp_VExtractToGPR>();
+  const auto Op = IROp->C<IR::IROp_VExtractToGPR>();
 
-  switch (Op->Header.ElementSize) {
+  constexpr auto SSERegSize = Core::CPUState::XMM_SSE_REG_SIZE;
+  constexpr auto SSEBitSize = SSERegSize * 8;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto ElementSizeBits = ElementSize * 8;
+  const auto Offset = ElementSizeBits * Op->Index;
+
+  const auto Is256Bit = Offset >= SSEBitSize;
+
+  const auto Vector = GetSrc(Op->Vector.ID());
+
+  switch (ElementSize) {
     case 1: {
-      pextrb(GetDst<RA_32>(Node), GetSrc(Op->Vector.ID()), Op->Index);
-    break;
+      if (Is256Bit) {
+        vextracti128(xmm15, ToYMM(Vector), 1);
+        pextrb(GetDst<RA_32>(Node), xmm15, Op->Index - 16);
+      } else {
+        pextrb(GetDst<RA_32>(Node), Vector, Op->Index);
+      }
+      break;
     }
     case 2: {
-      pextrw(GetDst<RA_32>(Node), GetSrc(Op->Vector.ID()), Op->Index);
-    break;
+      if (Is256Bit) {
+        vextracti128(xmm15, ToYMM(Vector), 1);
+        pextrw(GetDst<RA_32>(Node), xmm15, Op->Index - 8);
+      } else {
+        pextrw(GetDst<RA_32>(Node), Vector, Op->Index);
+      }
+      break;
     }
     case 4: {
-      pextrd(GetDst<RA_32>(Node), GetSrc(Op->Vector.ID()), Op->Index);
-    break;
+      if (Is256Bit) {
+        vextracti128(xmm15, ToYMM(Vector), 1);
+        pextrd(GetDst<RA_32>(Node), xmm15, Op->Index - 4);
+      } else {
+        pextrd(GetDst<RA_32>(Node), Vector, Op->Index);
+      }
+      break;
     }
     case 8: {
-      pextrq(GetDst<RA_64>(Node), GetSrc(Op->Vector.ID()), Op->Index);
-    break;
+      if (Is256Bit) {
+        vextracti128(xmm15, ToYMM(Vector), 1);
+        pextrq(GetDst<RA_64>(Node), xmm15, Op->Index - 2);
+      } else {
+        pextrq(GetDst<RA_64>(Node), Vector, Op->Index);
+      }
+      break;
     }
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+      break;
   }
 }
 

External/FEXCore/Source/Interface/Core/JIT/x86_64/ConversionOps.cpp

@@ -17,27 +17,76 @@ namespace FEXCore::CPU {
 
 #define DEF_OP(x) void X86JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 DEF_OP(VInsGPR) {
-  auto Op = IROp->C<IR::IROp_VInsGPR>();
-  movapd(GetDst(Node), GetSrc(Op->DestVector.ID()));
+  const auto Op = IROp->C<IR::IROp_VInsGPR>();
+  const auto OpSize = IROp->Size;
 
-  switch (Op->Header.ElementSize) {
-    case 1: {
-      pinsrb(GetDst(Node), GetSrc<RA_32>(Op->Src.ID()), Op->DestIdx);
-      break;
+  const auto Dst = GetDst(Node);
+  const auto DestVector = GetSrc(Op->DestVector.ID());
+
+  const auto DestIdx = Op->DestIdx;
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto ElementSizeBits = ElementSize * 8;
+  const auto Offset = ElementSizeBits * DestIdx;
+
+  constexpr auto SSEBitSize = Core::CPUState::XMM_SSE_REG_SIZE * 8;
+
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+  const auto InUpperLane = Offset >= SSEBitSize;
+
+  if (InUpperLane && !Is256Bit) {
+    LOGMAN_MSG_A_FMT("Attempt to access upper 128-bit lane in 128-bit operation! Offset={}",
+                     Offset);
+    return;
+  }
+
+  if (Is256Bit) {
+    vmovapd(ToYMM(Dst), ToYMM(DestVector));
+  } else {
+    vmovapd(Dst, DestVector);
+  }
+
+  const auto Insert = [&](const Xbyak::Xmm& reg, int index) {
+    switch (ElementSize) {
+      case 1: {
+        if (InUpperLane) {
+          index -= 16;
+        }
+        pinsrb(reg, GetSrc<RA_32>(Op->Src.ID()), index);
+        break;
+      }
+      case 2: {
+        if (InUpperLane) {
+          index -= 8;
+        }
+        pinsrw(reg, GetSrc<RA_32>(Op->Src.ID()), index);
+        break;
+      }
+      case 4: {
+        if (InUpperLane) {
+          index -= 4;
+        }
+        pinsrd(reg, GetSrc<RA_32>(Op->Src.ID()), index);
+        break;
+      }
+      case 8: {
+        if (InUpperLane) {
+          index -= 2;
+        }
+        pinsrq(reg, GetSrc<RA_64>(Op->Src.ID()), index);
+        break;
+      }
+      default:
+        LOGMAN_MSG_A_FMT("Unknown Element Size: {}", ElementSize);
+        break;
     }
-    case 2: {
-      pinsrw(GetDst(Node), GetSrc<RA_32>(Op->Src.ID()), Op->DestIdx);
-      break;
-    }
-    case 4: {
-      pinsrd(GetDst(Node), GetSrc<RA_32>(Op->Src.ID()), Op->DestIdx);
-      break;
-    }
-    case 8: {
-      pinsrq(GetDst(Node), GetSrc<RA_64>(Op->Src.ID()), Op->DestIdx);
-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Element Size: {}", Op->Header.ElementSize); break;
+  };
+
+  if (InUpperLane) {
+    vextracti128(xmm15, ToYMM(Dst), 1);
+    Insert(xmm15, DestIdx);
+    vinserti128(ToYMM(Dst), ToYMM(Dst), xmm15, 1);
+  } else {
+    Insert(Dst, DestIdx);
   }
 }
 
@@ -63,8 +112,10 @@ DEF_OP(VCastFromGPR) {
 }
 
 DEF_OP(Float_FromGPR_S) {
-  auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
-  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
+
+  const uint16_t ElementSize = Op->Header.ElementSize;
+  const uint16_t Conv = (ElementSize << 8) | Op->SrcElementSize;
 
   switch (Conv) {
     case 0x0404: { // Float <- int32_t
@@ -83,6 +134,10 @@ DEF_OP(Float_FromGPR_S) {
       cvtsi2sd(GetDst(Node), GetSrc<RA_64>(Op->Src.ID()));
       break;
     }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled conversion mask: Mask=0x{:04x}, ElementSize={}, SrcElementSize={}",
+                       Conv, ElementSize, Op->SrcElementSize);
+      break;
   }
 }
 
@@ -104,99 +159,196 @@ DEF_OP(Float_FToF) {
 }
 
 DEF_OP(Vector_SToF) {
-  auto Op = IROp->C<IR::IROp_Vector_SToF>();
-  switch (Op->Header.ElementSize) {
+  const auto Op = IROp->C<IR::IROp_Vector_SToF>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  const auto Dst = GetDst(Node);
+  const auto Vector = GetSrc(Op->Vector.ID());
+
+  switch (ElementSize) {
     case 4:
-      cvtdq2ps(GetDst(Node), GetSrc(Op->Vector.ID()));
-    break;
+      if (Is256Bit) {
+        vcvtdq2ps(ToYMM(Dst), ToYMM(Vector));
+      } else {
+        vcvtdq2ps(Dst, Vector);
+      }
+      break;
     case 8:
       // This operation is a bit disgusting in x86
       // There is no vector form of this instruction until AVX512VL + AVX512DQ (vcvtqq2pd)
       // 1) First extract the top 64bits
       // 2) Do a scalar conversion on each
       // 3) Make sure to merge them together at the end
-      pextrq(rax, GetSrc(Op->Vector.ID()), 1);
-      pextrq(rcx, GetSrc(Op->Vector.ID()), 0);
-      cvtsi2sd(GetDst(Node), rcx);
+      pextrq(rax, Vector, 1);
+      pextrq(rcx, Vector, 0);
+      cvtsi2sd(Dst, rcx);
       cvtsi2sd(xmm15, rax);
-      movlhps(GetDst(Node), xmm15);
-    break;
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_SToF element size: {}", Op->Header.ElementSize);
+      if (Is256Bit) {
+        movlhps(Dst, xmm15);
+        vextracti128(xmm15, ToYMM(Vector), 1);
+
+        pextrq(rax, xmm15, 1);
+        pextrq(rcx, xmm15, 0);
+        cvtsi2sd(xmm15, rcx);
+        cvtsi2sd(xmm14, rax);
+        movlhps(xmm15, xmm14);
+
+        vinserti128(ToYMM(Dst), ToYMM(Dst), xmm15, 1);
+      } else {
+        vmovlhps(Dst, Dst, xmm15);
+      }
+      break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Vector_SToF element size: {}", ElementSize);
+      break;
   }
 }
 
 DEF_OP(Vector_FToZS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToZS>();
-  switch (Op->Header.ElementSize) {
+  const auto Op = IROp->C<IR::IROp_Vector_FToZS>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  const auto Dst = GetDst(Node);
+  const auto Vector = GetSrc(Op->Vector.ID());
+
+  switch (ElementSize) {
     case 4:
-      cvttps2dq(GetDst(Node), GetSrc(Op->Vector.ID()));
-    break;
+      if (Is256Bit) {
+        vcvttps2dq(ToYMM(Dst), ToYMM(Vector));
+      } else {
+        vcvttps2dq(Dst, Vector);
+      }
+      break;
     case 8:
-      cvttpd2dq(GetDst(Node), GetSrc(Op->Vector.ID()));
-    break;
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_FToZS element size: {}", Op->Header.ElementSize);
+      if (Is256Bit) {
+        vcvttpd2dq(ToYMM(Dst), ToYMM(Vector));
+      } else {
+        vcvttpd2dq(Dst, Vector);
+      }
+      break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Vector_FToZS element size: {}", ElementSize);
+      break;
   }
 }
 
 DEF_OP(Vector_FToS) {
-  auto Op = IROp->C<IR::IROp_Vector_FToS>();
-  switch (Op->Header.ElementSize) {
+  const auto Op = IROp->C<IR::IROp_Vector_FToS>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  const auto Dst = GetDst(Node);
+  const auto Vector = GetSrc(Op->Vector.ID());
+
+  switch (ElementSize) {
     case 4:
-      cvtps2dq(GetDst(Node), GetSrc(Op->Vector.ID()));
-    break;
+      if (Is256Bit) {
+        vcvtps2dq(ToYMM(Dst), ToYMM(Vector));
+      } else {
+        vcvtps2dq(Dst, Vector);
+      }
+      break;
     case 8:
-      cvtpd2dq(GetDst(Node), GetSrc(Op->Vector.ID()));
-    break;
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_FToS element size: {}", Op->Header.ElementSize);
+      if (Is256Bit) {
+        vcvtpd2dq(ToYMM(Dst), ToYMM(Vector));
+      } else {
+        vcvtpd2dq(Dst, Vector);
+      }
+      break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Vector_FToS element size: {}", ElementSize);
+      break;
   }
 }
 
 DEF_OP(Vector_FToF) {
-  auto Op = IROp->C<IR::IROp_Vector_FToF>();
-  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const auto Op = IROp->C<IR::IROp_Vector_FToF>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+  const auto Conv = (ElementSize << 8) | Op->SrcElementSize;
+
+  const auto Dst = GetDst(Node);
+  const auto Vector = GetSrc(Op->Vector.ID());
 
   switch (Conv) {
     case 0x0804: { // Double <- Float
-      cvtps2pd(GetDst(Node), GetSrc(Op->Vector.ID()));
+      if (Is256Bit) {
+        vcvtps2pd(ToYMM(Dst), Vector);
+      } else {
+        vcvtps2pd(Dst, Vector);
+      }
       break;
     }
     case 0x0408: { // Float <- Double
-      cvtpd2ps(GetDst(Node), GetSrc(Op->Vector.ID()));
+      if (Is256Bit) {
+        vcvtpd2ps(Dst, ToYMM(Vector));
+      } else {
+        vcvtpd2ps(Dst, Vector);
+      }
       break;
     }
-    default: LOGMAN_MSG_A_FMT("Unknown Vector_FToF conversion type : 0x{:04x}", Conv); break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown Vector_FToF conversion type : 0x{:04x}", Conv);
+      break;
   }
 }
 
 DEF_OP(Vector_FToI) {
-  auto Op = IROp->C<IR::IROp_Vector_FToI>();
-  uint8_t RoundMode{};
+  const auto Op = IROp->C<IR::IROp_Vector_FToI>();
+  const auto OpSize = IROp->Size;
 
-  switch (Op->Round) {
-    case FEXCore::IR::Round_Nearest.Val:
-      RoundMode = 0b0000'0'0'00;
-    break;
-    case FEXCore::IR::Round_Negative_Infinity.Val:
-      RoundMode = 0b0000'0'0'01;
-    break;
-    case FEXCore::IR::Round_Positive_Infinity.Val:
-      RoundMode = 0b0000'0'0'10;
-    break;
-    case FEXCore::IR::Round_Towards_Zero.Val:
-      RoundMode = 0b0000'0'0'11;
-    break;
-    case FEXCore::IR::Round_Host.Val:
-      RoundMode = 0b0000'0'1'00;
-    break;
-  }
+  const uint8_t RoundMode = [Op] {
+    switch (Op->Round) {
+      case FEXCore::IR::Round_Nearest.Val:
+        return 0b0000'0'0'00;
+      case FEXCore::IR::Round_Negative_Infinity.Val:
+        return 0b0000'0'0'01;
+      case FEXCore::IR::Round_Positive_Infinity.Val:
+        return 0b0000'0'0'10;
+      case FEXCore::IR::Round_Towards_Zero.Val:
+        return 0b0000'0'0'11;
+      case FEXCore::IR::Round_Host.Val:
+        return 0b0000'0'1'00;
+      default:
+        LOGMAN_MSG_A_FMT("Unhandled rounding mode");
+        return 0;
+    }
+  }();
 
-  switch (Op->Header.ElementSize) {
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto Is256Bit = OpSize == Core::CPUState::XMM_AVX_REG_SIZE;
+
+  const auto Dst = GetDst(Node);
+  const auto Vector = GetSrc(Op->Vector.ID());
+
+  switch (ElementSize) {
     case 4:
-      roundps(GetDst(Node), GetSrc(Op->Vector.ID()), RoundMode);
-    break;
+      if (Is256Bit) {
+        vroundps(ToYMM(Dst), ToYMM(Vector), RoundMode);
+      } else {
+        vroundps(Dst, Vector, RoundMode);
+      }
+      break;
     case 8:
-      roundpd(GetDst(Node), GetSrc(Op->Vector.ID()), RoundMode);
-    break;
+      if (Is256Bit) {
+        vroundpd(ToYMM(Dst), ToYMM(Vector), RoundMode);
+      } else {
+        vroundpd(Dst, Vector, RoundMode);
+      }
+      break;
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled element size: {}", ElementSize);
+      break;
   }
 }
 

External/FEXCore/Source/Interface/Core/JIT/x86_64/JIT.cpp

@@ -27,6 +27,7 @@ $end_info$
 #include <FEXCore/Utils/Allocator.h>
 #include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <algorithm>
 #include <array>
@@ -370,7 +371,7 @@ X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalTh
 
   {
     auto &Common = ThreadState->CurrentFrame->Pointers.Common;
-    
+
     Common.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
     Common.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
     Common.ThreadRemoveCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::ThreadRemoveCodeEntryFromJit);
@@ -582,6 +583,8 @@ std::tuple<X86JITCore::SetCC, X86JITCore::CMovCC, X86JITCore::JCC> X86JITCore::G
 }
 
 void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) {
+
+  FEXCORE_PROFILE_SCOPED("x86::CompileCode");
   JumpTargets.clear();
   uint32_t SSACount = IR->GetSSACount();
 

External/FEXCore/Source/Interface/Core/JIT/x86_64/JITClass.h

@@ -337,6 +337,8 @@ private:
   ///< Memory ops
   DEF_OP(LoadContext);
   DEF_OP(StoreContext);
+  DEF_OP(LoadRegister);
+  DEF_OP(StoreRegister);
   DEF_OP(LoadContextIndexed);
   DEF_OP(StoreContextIndexed);
   DEF_OP(SpillRegister);
@@ -345,8 +347,6 @@ private:
   DEF_OP(StoreFlag);
   DEF_OP(LoadMem);
   DEF_OP(StoreMem);
-  DEF_OP(VLoadMemElement);
-  DEF_OP(VStoreMemElement);
   DEF_OP(CacheLineClear);
   DEF_OP(CacheLineZero);
 
@@ -430,8 +430,6 @@ private:
   DEF_OP(VInsElement);
   DEF_OP(VDupElement);
   DEF_OP(VExtr);
-  DEF_OP(VSLI);
-  DEF_OP(VSRI);
   DEF_OP(VUShrI);
   DEF_OP(VSShrI);
   DEF_OP(VShlI);

External/FEXCore/Source/Interface/Core/JIT/x86_64/MemoryOps.cpp

@@ -21,130 +21,266 @@ namespace FEXCore::CPU {
 #define DEF_OP(x) void X86JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 
 DEF_OP(LoadContext) {
-  auto Op = IROp->C<IR::IROp_LoadContext>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_LoadContext>();
+  const auto OpSize = IROp->Size;
 
   if (Op->Class == IR::GPRClass) {
     switch (OpSize) {
     case 1: {
       movzx(GetDst<RA_32>(Node), byte [STATE + Op->Offset]);
+      break;
     }
-    break;
     case 2: {
       movzx(GetDst<RA_32>(Node), word [STATE + Op->Offset]);
+      break;
     }
-    break;
     case 4: {
       mov(GetDst<RA_32>(Node), dword [STATE + Op->Offset]);
+      break;
     }
-    break;
     case 8: {
       mov(GetDst<RA_64>(Node), qword [STATE + Op->Offset]);
+      break;
     }
-    break;
     case 16: {
       LOGMAN_MSG_A_FMT("Invalid GPR load of size 16");
+      break;
     }
-    break;
-    default:  LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+      break;
     }
   }
   else {
+    const auto Dst = GetDst(Node);
+
     switch (OpSize) {
     case 1: {
       movzx(rax, byte [STATE + Op->Offset]);
-      vmovq(GetDst(Node), rax);
+      vmovq(Dst, rax);
+      break;
     }
-    break;
     case 2: {
       movzx(rax, word [STATE + Op->Offset]);
-      vmovq(GetDst(Node), rax);
+      vmovq(Dst, rax);
+      break;
     }
-    break;
     case 4: {
-      vmovd(GetDst(Node), dword [STATE + Op->Offset]);
+      vmovd(Dst, dword [STATE + Op->Offset]);
+      break;
     }
-    break;
     case 8: {
-      vmovq(GetDst(Node), qword [STATE + Op->Offset]);
+      vmovq(Dst, qword [STATE + Op->Offset]);
+      break;
     }
-    break;
     case 16: {
-      if (Op->Offset % 16 == 0)
-        movaps(GetDst(Node), xword [STATE + Op->Offset]);
-      else
-        movups(GetDst(Node), xword [STATE + Op->Offset]);
+      if (Op->Offset % 16 == 0) {
+        vmovaps(Dst, xword [STATE + Op->Offset]);
+      } else {
+        vmovups(Dst, xword [STATE + Op->Offset]);
+      }
+      break;
     }
-    break;
-    default:  LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+    case 32: {
+      vmovups(ToYMM(Dst), yword [STATE + Op->Offset]);
+      break;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadContext size: {}", OpSize);
+      break;
     }
   }
 }
 
 DEF_OP(StoreContext) {
-  auto Op = IROp->C<IR::IROp_StoreContext>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_StoreContext>();
+  const auto OpSize = IROp->Size;
 
   if (Op->Class == IR::GPRClass) {
     switch (OpSize) {
     case 1: {
       mov(byte [STATE + Op->Offset], GetSrc<RA_8>(Op->Value.ID()));
+      break;
     }
-    break;
-
     case 2: {
       mov(word [STATE + Op->Offset], GetSrc<RA_16>(Op->Value.ID()));
+      break;
     }
-    break;
     case 4: {
       mov(dword [STATE + Op->Offset], GetSrc<RA_32>(Op->Value.ID()));
+      break;
     }
-    break;
     case 8: {
       mov(qword [STATE + Op->Offset], GetSrc<RA_64>(Op->Value.ID()));
+      break;
     }
-    break;
-    case 16:
-      LogMan::Msg::DFmt("Invalid store size of 16");
-    break;
-    default:  LOGMAN_MSG_A_FMT("Unhandled StoreContext size: {}", OpSize);
+    case 16: {
+      LOGMAN_MSG_A_FMT("Invalid store size of 16");
+      break;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled StoreContext size: {}", OpSize);
+      break;
     }
   }
   else {
+    const auto Value = GetSrc(Op->Value.ID());
+
     switch (OpSize) {
     case 1: {
-      pextrb(byte [STATE + Op->Offset], GetSrc(Op->Value.ID()), 0);
+      pextrb(byte [STATE + Op->Offset], Value, 0);
+      break;
     }
-    break;
-
     case 2: {
-      pextrw(word [STATE + Op->Offset], GetSrc(Op->Value.ID()), 0);
+      pextrw(word [STATE + Op->Offset], Value, 0);
+      break;
     }
-    break;
     case 4: {
-      vmovd(dword [STATE + Op->Offset], GetSrc(Op->Value.ID()));
+      vmovd(dword [STATE + Op->Offset], Value);
+      break;
     }
-    break;
     case 8: {
-      vmovq(qword [STATE + Op->Offset], GetSrc(Op->Value.ID()));
+      vmovq(qword [STATE + Op->Offset], Value);
+      break;
     }
-    break;
     case 16: {
-      if (Op->Offset % 16 == 0)
-        movaps(xword [STATE + Op->Offset], GetSrc(Op->Value.ID()));
-      else
-        movups(xword [STATE + Op->Offset], GetSrc(Op->Value.ID()));
+      if (Op->Offset % 16 == 0) {
+        vmovaps(xword [STATE + Op->Offset], Value);
+      } else {
+        vmovups(xword [STATE + Op->Offset], Value);
+      }
+      break;
     }
-    break;
-    default:  LOGMAN_MSG_A_FMT("Unhandled StoreContext size: {}", OpSize);
+    case 32: {
+      vmovups(yword [STATE + Op->Offset], ToYMM(Value));
+      break;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled StoreContext size: {}", OpSize);
+      break;
     }
   }
 }
 
+DEF_OP(LoadRegister) {
+  const auto Op = IROp->C<IR::IROp_LoadRegister>();
+  const auto OpSize = IROp->Size;
+
+  if (Op->Class == IR::GPRClass) {
+    switch (OpSize) {
+    case 1: {
+      movzx(GetSrc<RA_32>(Node), byte [STATE + Op->Offset]);
+      break;
+    }
+    case 2: {
+      movzx(GetSrc<RA_32>(Node), word [STATE + Op->Offset]);
+      break;
+    }
+    case 4: {
+      mov(GetSrc<RA_32>(Node), dword [STATE + Op->Offset]);
+      break;
+    }
+    case 8: {
+      mov(GetSrc<RA_64>(Node), qword [STATE + Op->Offset]);
+      break;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadRegister size: {}", OpSize);
+      break;
+    }
+  }
+  else {
+    const auto Dst = GetSrc(Node);
+
+    switch (OpSize) {
+    case 1: {
+      movzx(rax, byte [STATE + Op->Offset]);
+      vmovq(Dst, rax);
+      break;
+    }
+    case 2: {
+      movzx(rax, word [STATE + Op->Offset]);
+      vmovq(Dst, rax);
+      break;
+    }
+    case 4: {
+      vmovd(Dst, dword [STATE + Op->Offset]);
+      break;
+    }
+    case 8: {
+      vmovq(Dst, qword [STATE + Op->Offset]);
+      break;
+    }
+    case 16: {
+      if (Op->Offset % 16 == 0) {
+        vmovaps(Dst, xword [STATE + Op->Offset]);
+      } else {
+        vmovups(Dst, xword [STATE + Op->Offset]);
+      }
+      break;
+    }
+    case 32: {
+      vmovups(ToYMM(Dst), yword [STATE + Op->Offset]);
+      break;
+    }
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled LoadRegister size: {}", OpSize);
+      break;
+    }
+  }
+}
+
+DEF_OP(StoreRegister) {
+  const auto Op = IROp->C<IR::IROp_StoreRegister>();
+  const auto OpSize = IROp->Size;
+
+  if (Op->Class == IR::GPRClass) {
+    const auto regOffs = Op->Offset & 7;
+
+    switch (OpSize) {
+      case 4:
+        LOGMAN_THROW_AA_FMT(regOffs == 0, "unexpected regOffs");
+        mov(dword [STATE + Op->Offset], GetSrc<RA_32>(Op->Value.ID()));
+        break;
+
+      case 8:
+        LOGMAN_THROW_AA_FMT(regOffs == 0, "unexpected regOffs");
+        mov(qword [STATE + Op->Offset], GetSrc<RA_64>(Op->Value.ID()));
+        break;
+
+      default:
+        LOGMAN_MSG_A_FMT("Unhandled StoreRegister GPR size: {}", OpSize);
+        break;
+    }
+  } else if (Op->Class == IR::FPRClass) {
+    const auto Value = GetSrc(Op->Value.ID());
+    switch (OpSize) {
+      case 16: {
+        if (Op->Offset % 16 == 0) {
+          vmovaps(xword [STATE + Op->Offset], Value);
+        } else {
+          vmovups(xword [STATE + Op->Offset], Value);
+        }
+        break;
+      }
+      case 32: {
+        vmovups(yword [STATE + Op->Offset], ToYMM(Value));
+        break;
+      }
+      default:
+        LOGMAN_MSG_A_FMT("Unhandled StoreContext size: {}", OpSize);
+        break;
+    }
+  } else {
+    LOGMAN_THROW_AA_FMT(false, "Unhandled Op->Class {}", Op->Class);
+  }
+}
+
 DEF_OP(LoadContextIndexed) {
-  auto Op = IROp->C<IR::IROp_LoadContextIndexed>();
-  size_t size = IROp->Size;
-  Reg index = GetSrc<RA_64>(Op->Index.ID());
+  const auto Op = IROp->C<IR::IROp_LoadContextIndexed>();
+  const auto OpSize = IROp->Size;
+
+  const Reg Index = GetSrc<RA_64>(Op->Index.ID());
 
   if (Op->Class == IR::GPRClass) {
     switch (Op->Stride) {
@@ -153,21 +289,21 @@ DEF_OP(LoadContextIndexed) {
     case 4:
     case 8: {
       lea(rax, dword [STATE + Op->BaseOffset]);
-      switch (size) {
+      switch (OpSize) {
       case 1:
-        movzx(GetDst<RA_32>(Node), byte [rax + index * Op->Stride]);
+        movzx(GetDst<RA_32>(Node), byte [rax + Index * Op->Stride]);
         break;
       case 2:
-        movzx(GetDst<RA_32>(Node), word [rax + index * Op->Stride]);
+        movzx(GetDst<RA_32>(Node), word [rax + Index * Op->Stride]);
         break;
       case 4:
-        mov(GetDst<RA_32>(Node),  dword [rax + index * Op->Stride]);
+        mov(GetDst<RA_32>(Node),  dword [rax + Index * Op->Stride]);
         break;
       case 8:
-        mov(GetDst<RA_64>(Node),  qword [rax + index * Op->Stride]);
+        mov(GetDst<RA_64>(Node),  qword [rax + Index * Op->Stride]);
         break;
       default:
-        LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", IROp->Size);
+        LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", OpSize);
         break;
       }
       break;
@@ -186,53 +322,63 @@ DEF_OP(LoadContextIndexed) {
     case 2:
     case 4:
     case 8: {
+      const auto Dst = GetDst(Node);
+
       lea(rax, dword [STATE + Op->BaseOffset]);
-      switch (size) {
+      switch (OpSize) {
       case 1:
-        movzx(eax, byte [rax + index * Op->Stride]);
-        vmovd(GetDst(Node), eax);
+        movzx(eax, byte [rax + Index * Op->Stride]);
+        vmovd(Dst, eax);
         break;
       case 2:
-        movzx(eax, word [rax + index * Op->Stride]);
-        vmovd(GetDst(Node), eax);
+        movzx(eax, word [rax + Index * Op->Stride]);
+        vmovd(Dst, eax);
         break;
       case 4:
-        vmovd(GetDst(Node),  dword [rax + index * Op->Stride]);
+        vmovd(Dst,  dword [rax + Index * Op->Stride]);
         break;
       case 8:
-        vmovq(GetDst(Node),  qword [rax + index * Op->Stride]);
+        vmovq(Dst,  qword [rax + Index * Op->Stride]);
         break;
       default:
-        LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", IROp->Size);
+        LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", OpSize);
         break;
       }
       break;
     }
-    case 16: {
-      mov(rax, index);
-      shl(rax, 4);
+    case 16:
+    case 32: {
+      const auto Dst = GetDst(Node);
+      const auto Shift = Op->Stride == 16 ? 4 : 5;
+
+      mov(rax, Index);
+      shl(rax, Shift);
       lea(rax, dword [rax + Op->BaseOffset]);
-      switch (size) {
+      switch (OpSize) {
       case 1:
-        pinsrb(GetDst(Node), byte [STATE + rax], 0);
+        pinsrb(Dst, byte [STATE + rax], 0);
         break;
       case 2:
-        pinsrw(GetDst(Node), word [STATE + rax], 0);
+        pinsrw(Dst, word [STATE + rax], 0);
         break;
       case 4:
-        vmovd(GetDst(Node), dword [STATE + rax]);
+        vmovd(Dst, dword [STATE + rax]);
         break;
       case 8:
-        vmovq(GetDst(Node), qword [STATE + rax]);
+        vmovq(Dst, qword [STATE + rax]);
         break;
       case 16:
-        if (Op->BaseOffset % 16 == 0)
-          movaps(GetDst(Node), xword [STATE + rax]);
-        else
-          movups(GetDst(Node), xword [STATE + rax]);
+        if (Op->BaseOffset % 16 == 0) {
+          vmovaps(Dst, xword [STATE + rax]);
+        } else {
+          vmovups(Dst, xword [STATE + rax]);
+        }
+        break;
+      case 32:
+        vmovups(ToYMM(Dst), yword [STATE + rax]);
         break;
       default:
-        LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", IROp->Size);
+        LOGMAN_MSG_A_FMT("Unhandled LoadContextIndexed size: {}", OpSize);
         break;
       }
       break;
@@ -245,12 +391,13 @@ DEF_OP(LoadContextIndexed) {
 }
 
 DEF_OP(StoreContextIndexed) {
-  auto Op = IROp->C<IR::IROp_StoreContextIndexed>();
-  Reg index = GetSrc<RA_64>(Op->Index.ID());
-  size_t size = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_StoreContextIndexed>();
+  const auto OpSize = IROp->Size;
+
+  const Reg Index = GetSrc<RA_64>(Op->Index.ID());
 
   if (Op->Class == IR::GPRClass) {
-    auto value = GetSrc<RA_64>(Op->Value.ID());
+    const auto Value = GetSrc<RA_64>(Op->Value.ID());
     lea(rax, dword [STATE + Op->BaseOffset]);
 
     switch (Op->Stride) {
@@ -258,10 +405,10 @@ DEF_OP(StoreContextIndexed) {
     case 2:
     case 4:
     case 8: {
-      if (!(size == 1 || size == 2 || size == 4 || size == 8)) {
-        LOGMAN_MSG_A_FMT("Unhandled StoreContextIndexed size: {}", IROp->Size);
+      if (!(OpSize == 1 || OpSize == 2 || OpSize == 4 || OpSize == 8)) {
+        LOGMAN_MSG_A_FMT("Unhandled StoreContextIndexed size: {}", OpSize);
       }
-      mov(AddressFrame(IROp->Size * 8) [rax + index * Op->Stride], value);
+      mov(AddressFrame(OpSize * 8) [rax + Index * Op->Stride], Value);
       break;
     }
     default:
@@ -270,57 +417,64 @@ DEF_OP(StoreContextIndexed) {
     }
   }
   else {
-    auto value = GetSrc(Op->Value.ID());
+    const auto Value = GetSrc(Op->Value.ID());
     switch (Op->Stride) {
     case 1:
     case 2:
     case 4:
     case 8: {
       lea(rax, dword [STATE + Op->BaseOffset]);
-      switch (size) {
+      switch (OpSize) {
       case 1:
-        pextrb(AddressFrame(IROp->Size * 8) [rax + index * Op->Stride], value, 0);
+        pextrb(AddressFrame(OpSize * 8) [rax + Index * Op->Stride], Value, 0);
         break;
       case 2:
-        pextrw(AddressFrame(IROp->Size * 8) [rax + index * Op->Stride], value, 0);
+        pextrw(AddressFrame(OpSize * 8) [rax + Index * Op->Stride], Value, 0);
         break;
       case 4:
-        vmovd(AddressFrame(IROp->Size * 8) [rax + index * Op->Stride], value);
+        vmovd(AddressFrame(OpSize * 8) [rax + Index * Op->Stride], Value);
         break;
       case 8:
-        vmovq(AddressFrame(IROp->Size * 8) [rax + index * Op->Stride], value);
+        vmovq(AddressFrame(OpSize * 8) [rax + Index * Op->Stride], Value);
         break;
       default:
-        LOGMAN_MSG_A_FMT("Unhandled StoreContextIndexed size: {}", size);
+        LOGMAN_MSG_A_FMT("Unhandled StoreContextIndexed size: {}", OpSize);
         break;
       }
       break;
     }
-    case 16: {
-      mov(rax, index);
-      shl(rax, 4);
+    case 16:
+    case 32: {
+      const auto Shift = Op->Stride == 16 ? 4 : 5;
+
+      mov(rax, Index);
+      shl(rax, Shift);
       lea(rax, dword [rax + Op->BaseOffset]);
-      switch (size) {
+      switch (OpSize) {
       case 1:
-        pextrb(AddressFrame(IROp->Size * 8) [STATE + rax], value, 0);
+        pextrb(AddressFrame(OpSize * 8) [STATE + rax], Value, 0);
         break;
       case 2:
-        pextrw(AddressFrame(IROp->Size * 8) [STATE + rax], value, 0);
+        pextrw(AddressFrame(OpSize * 8) [STATE + rax], Value, 0);
         break;
       case 4:
-        vmovd(AddressFrame(IROp->Size * 8) [STATE + rax], value);
+        vmovd(AddressFrame(OpSize * 8) [STATE + rax], Value);
         break;
       case 8:
-        vmovq(AddressFrame(IROp->Size * 8) [STATE + rax], value);
+        vmovq(AddressFrame(OpSize * 8) [STATE + rax], Value);
         break;
       case 16:
-        if (Op->BaseOffset % 16 == 0)
-          movaps(xword [STATE + rax], value);
-        else
-          movups(xword [STATE + rax], value);
+        if (Op->BaseOffset % 16 == 0) {
+          vmovaps(xword [STATE + rax], Value);
+        } else {
+          vmovups(xword [STATE + rax], Value);
+        }
+        break;
+      case 32:
+        vmovups(yword [STATE + rax], ToYMM(Value));
         break;
       default:
-        LOGMAN_MSG_A_FMT("Unhandled StoreContextIndexed size: {}", size);
+        LOGMAN_MSG_A_FMT("Unhandled StoreContextIndexed size: {}", OpSize);
         break;
       }
       break;
@@ -376,7 +530,7 @@ DEF_OP(SpillRegister) {
         break;
       }
       case 32: {
-        vmovaps(yword [rsp + SlotOffset], ToYMM(Src));
+        vmovups(yword [rsp + SlotOffset], ToYMM(Src));
         break;
       }
       default:
@@ -420,19 +574,19 @@ DEF_OP(FillRegister) {
 
     switch (OpSize) {
       case 4: {
-        movss(Dst, dword [rsp + SlotOffset]);
+        vmovss(Dst, dword [rsp + SlotOffset]);
         break;
       }
       case 8: {
-        movsd(Dst, qword [rsp + SlotOffset]);
+        vmovsd(Dst, qword [rsp + SlotOffset]);
         break;
       }
       case 16: {
-        movaps(Dst, xword [rsp + SlotOffset]);
+        vmovaps(Dst, xword [rsp + SlotOffset]);
         break;
       }
       case 32: {
-        vmovaps(ToYMM(Dst), yword [rsp + SlotOffset]);
+        vmovups(ToYMM(Dst), yword [rsp + SlotOffset]);
         break;
       }
       default:
@@ -482,130 +636,136 @@ Xbyak::RegExp X86JITCore::GenerateModRM(Xbyak::Reg Base, IR::OrderedNodeWrapper
 }
 
 DEF_OP(LoadMem) {
-  auto Op = IROp->C<IR::IROp_LoadMem>();
+  const auto Op = IROp->C<IR::IROp_LoadMem>();
+  const auto OpSize = IROp->Size;
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
-
-  auto MemPtr = GenerateModRM(MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
+  const Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
+  const auto MemPtr = GenerateModRM(MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
 
   if (Op->Class == IR::GPRClass) {
-    auto Dst = GetDst<RA_64>(Node);
+    const auto Dst = GetDst<RA_64>(Node);
 
-    switch (IROp->Size) {
+    switch (OpSize) {
       case 1: {
-        movzx (Dst, byte [MemPtr]);
+        movzx(Dst, byte [MemPtr]);
+        break;
       }
-      break;
       case 2: {
-        movzx (Dst, word [MemPtr]);
+        movzx(Dst, word [MemPtr]);
+        break;
       }
-      break;
       case 4: {
         mov(Dst.cvt32(), dword [MemPtr]);
+        break;
       }
-      break;
       case 8: {
         mov(Dst, qword [MemPtr]);
+        break;
       }
-      break;
-      default:  LOGMAN_MSG_A_FMT("Unhandled LoadMem size: {}", IROp->Size);
+      default:
+        LOGMAN_MSG_A_FMT("Unhandled LoadMem size: {}", OpSize);
+        break;
     }
   }
   else
   {
-    auto Dst = GetDst(Node);
+    const auto Dst = GetDst(Node);
 
-    switch (IROp->Size) {
+    switch (OpSize) {
       case 1: {
         movzx(eax, byte [MemPtr]);
         vmovd(Dst, eax);
+        break;
       }
-      break;
       case 2: {
         movzx(eax, word [MemPtr]);
         vmovd(Dst, eax);
+        break;
       }
-      break;
       case 4: {
         vmovd(Dst, dword [MemPtr]);
+        break;
       }
-      break;
       case 8: {
         vmovq(Dst, qword [MemPtr]);
+        break;
       }
-      break;
       case 16: {
-         if (IROp->Size == Op->Align)
-           movups(GetDst(Node), xword [MemPtr]);
-         else
-           movups(GetDst(Node), xword [MemPtr]);
-         if (MemoryDebug) {
-           movq(rcx, GetDst(Node));
-         }
-       }
-       break;
-      default:  LOGMAN_MSG_A_FMT("Unhandled LoadMem size: {}", IROp->Size);
+        vmovups(Dst, xword [MemPtr]);
+        if (MemoryDebug) {
+          movq(rcx, Dst);
+        }
+        break;
+      }
+      case 32: {
+        vmovups(ToYMM(Dst), yword [MemPtr]);
+        if (MemoryDebug) {
+          movq(rcx, Dst);
+        }
+        break;
+      }
+      default:
+        LOGMAN_MSG_A_FMT("Unhandled LoadMem size: {}", OpSize);
+        break;
     }
   }
 }
 
 DEF_OP(StoreMem) {
-  auto Op = IROp->C<IR::IROp_StoreMem>();
+  const auto Op = IROp->C<IR::IROp_StoreMem>();
+  const auto OpSize = IROp->Size;
 
-  Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
-
-  auto MemPtr = GenerateModRM(MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
+  const Xbyak::Reg MemReg = GetSrc<RA_64>(Op->Addr.ID());
+  const auto MemPtr = GenerateModRM(MemReg, Op->Offset, Op->OffsetType, Op->OffsetScale);
 
   if (Op->Class == IR::GPRClass) {
-    switch (IROp->Size) {
+    switch (OpSize) {
     case 1:
       mov(byte [MemPtr], GetSrc<RA_8>(Op->Value.ID()));
-    break;
+      break;
     case 2:
       mov(word [MemPtr], GetSrc<RA_16>(Op->Value.ID()));
-    break;
+      break;
     case 4:
       mov(dword [MemPtr], GetSrc<RA_32>(Op->Value.ID()));
-    break;
+      break;
     case 8:
       mov(qword [MemPtr], GetSrc<RA_64>(Op->Value.ID()));
-    break;
-    default:  LOGMAN_MSG_A_FMT("Unhandled StoreMem size: {}", IROp->Size);
+      break;
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled StoreMem size: {}", OpSize);
+      break;
     }
   }
   else {
-    switch (IROp->Size) {
+    const auto Value = GetSrc(Op->Value.ID());
+
+    switch (OpSize) {
     case 1:
-      pextrb(byte [MemPtr], GetSrc(Op->Value.ID()), 0);
-    break;
+      pextrb(byte [MemPtr], Value, 0);
+      break;
     case 2:
-      pextrw(word [MemPtr], GetSrc(Op->Value.ID()), 0);
-    break;
+      pextrw(word [MemPtr], Value, 0);
+      break;
     case 4:
-      vmovd(dword [MemPtr], GetSrc(Op->Value.ID()));
-    break;
+      vmovd(dword [MemPtr], Value);
+      break;
     case 8:
-      vmovq(qword [MemPtr], GetSrc(Op->Value.ID()));
-    break;
+      vmovq(qword [MemPtr], Value);
+      break;
     case 16:
-      if (IROp->Size == Op->Align)
-        movups(xword [MemPtr], GetSrc(Op->Value.ID()));
-      else
-        movups(xword [MemPtr], GetSrc(Op->Value.ID()));
-    break;
-    default:  LOGMAN_MSG_A_FMT("Unhandled StoreMem size: {}", IROp->Size);
+      vmovups(xword [MemPtr], Value);
+      break;
+    case 32:
+      vmovups(yword [MemPtr], ToYMM(Value));
+      break;
+    default:
+      LOGMAN_MSG_A_FMT("Unhandled StoreMem size: {}", OpSize);
+      break;
     }
   }
 }
 
-DEF_OP(VLoadMemElement) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
-
-DEF_OP(VStoreMemElement) {
-  LOGMAN_MSG_A_FMT("Unimplemented");
-}
-
 DEF_OP(CacheLineClear) {
   auto Op = IROp->C<IR::IROp_CacheLineClear>();
 
@@ -636,8 +796,8 @@ void X86JITCore::RegisterMemoryHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
   REGISTER_OP(LOADCONTEXT,         LoadContext);
   REGISTER_OP(STORECONTEXT,        StoreContext);
-  REGISTER_OP(LOADREGISTER,        Unhandled); // SRA specific, not supported on this backend
-  REGISTER_OP(STOREREGISTER,       Unhandled);
+  REGISTER_OP(LOADREGISTER,        LoadRegister);
+  REGISTER_OP(STOREREGISTER,       StoreRegister);
   REGISTER_OP(LOADCONTEXTINDEXED,  LoadContextIndexed);
   REGISTER_OP(STORECONTEXTINDEXED, StoreContextIndexed);
   REGISTER_OP(SPILLREGISTER,       SpillRegister);
@@ -648,8 +808,6 @@ void X86JITCore::RegisterMemoryHandlers() {
   REGISTER_OP(STOREMEM,            StoreMem);
   REGISTER_OP(LOADMEMTSO,          LoadMem);
   REGISTER_OP(STOREMEMTSO,         StoreMem);
-  REGISTER_OP(VLOADMEMELEMENT,     VLoadMemElement);
-  REGISTER_OP(VSTOREMEMELEMENT,    VStoreMemElement);
   REGISTER_OP(CACHELINECLEAR,      CacheLineClear);
   REGISTER_OP(CACHELINEZERO,       CacheLineZero);
 #undef REGISTER_OP

External/FEXCore/Source/Interface/Core/JIT/x86_64/MiscOps.cpp

@@ -50,11 +50,19 @@ DEF_OP(Break) {
     add(rsp, SpillSlots * MaxSpillSlotSize);
   }
 
-  mov(byte [STATE + offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.FaultToTopAndGeneratedException)], 1);
-  mov(byte [STATE + offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.Signal)], Op->Reason.Signal);
-  mov(dword [STATE + offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.TrapNo)], Op->Reason.TrapNumber);
-  mov(dword [STATE + offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.err_code)], Op->Reason.ErrorRegister);
-  mov(dword [STATE + offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData.si_code)], Op->Reason.si_code);
+  Core::CpuStateFrame::SynchronousFaultDataStruct State = {
+    .FaultToTopAndGeneratedException = 1,
+    .Signal = Op->Reason.Signal,
+    .TrapNo = Op->Reason.TrapNumber,
+    .si_code = Op->Reason.si_code,
+    .err_code = Op->Reason.ErrorRegister,
+  };
+
+  uint64_t Constant{};
+  memcpy(&Constant, &State, sizeof(State));
+
+  mov(TMP1, Constant);
+  mov(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, SynchronousFaultData)], TMP1);
 
   switch (Op->Reason.Signal) {
   case SIGILL:

External/FEXCore/Source/Interface/Core/LookupCache.cpp

@@ -17,6 +17,10 @@ namespace FEXCore {
 LookupCache::LookupCache(FEXCore::Context::Context *CTX)
   : ctx {CTX} {
 
+  TotalCacheSize = ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE + L1_SIZE;
+  // Setup our PMR map.
+  BlockLinks = BlockLinks_pma.new_object<BlockLinksMapType>();
+
   // Block cache ends up looking like this
   // PageMemoryMap[VirtualMemoryRegion >> 12]
   //       |
@@ -29,46 +33,48 @@ LookupCache::LookupCache(FEXCore::Context::Context *CTX)
   // Allocate a region of memory that we can use to back our block pointers
   // We need one pointer per page of virtual memory
   // At 64GB of virtual memory this will allocate 128MB of virtual memory space
-  PagePointer = reinterpret_cast<uintptr_t>(FEXCore::Allocator::mmap(nullptr, ctx->Config.VirtualMemSize / 4096 * 8, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
+  PagePointer = reinterpret_cast<uintptr_t>(FEXCore::Allocator::mmap(nullptr, TotalCacheSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
 
   // Allocate our memory backing our pages
   // We need 32KB per guest page (One pointer per byte)
   // XXX: We can drop down to 16KB if we store 4byte offsets from the code base
   // We currently limit to 128MB of real memory for caching for the total cache size.
   // Can end up being inefficient if we compile a small number of blocks per page
-  PageMemory = reinterpret_cast<uintptr_t>(FEXCore::Allocator::mmap(nullptr, CODE_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
+  PageMemory = PagePointer + ctx->Config.VirtualMemSize / 4096 * 8;
   LOGMAN_THROW_AA_FMT(PageMemory != -1ULL, "Failed to allocate page memory");
 
   // L1 Cache
-  L1Pointer = reinterpret_cast<uintptr_t>(FEXCore::Allocator::mmap(nullptr, L1_SIZE, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
+  L1Pointer = PageMemory + CODE_SIZE;
   LOGMAN_THROW_AA_FMT(L1Pointer != -1ULL, "Failed to allocate L1Pointer");
 
   VirtualMemSize = ctx->Config.VirtualMemSize;
 }
 
 LookupCache::~LookupCache() {
-  FEXCore::Allocator::munmap(reinterpret_cast<void*>(PagePointer), ctx->Config.VirtualMemSize / 4096 * 8);
-  FEXCore::Allocator::munmap(reinterpret_cast<void*>(PageMemory), CODE_SIZE);
-  FEXCore::Allocator::munmap(reinterpret_cast<void*>(L1Pointer), L1_SIZE);
+  const size_t TotalCacheSize = ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE + L1_SIZE;
+  FEXCore::Allocator::munmap(reinterpret_cast<void*>(PagePointer), TotalCacheSize);
+
+  // No need to free BlockLinks map.
+  // These will get freed when their memory allocators are deallocated.
 }
 
 void LookupCache::ClearL2Cache() {
   std::lock_guard<std::recursive_mutex> lk(WriteLock);
   // Clear out the page memory
-  madvise(reinterpret_cast<void*>(PagePointer), ctx->Config.VirtualMemSize / 4096 * 8, MADV_DONTNEED);
-  madvise(reinterpret_cast<void*>(PageMemory), CODE_SIZE, MADV_DONTNEED);
+  // PagePointer and PageMemory are sequential with each other. Clear both at once.
+  madvise(reinterpret_cast<void*>(PagePointer), ctx->Config.VirtualMemSize / 4096 * 8 + CODE_SIZE, MADV_DONTNEED);
   AllocateOffset = 0;
 }
 
 void LookupCache::ClearCache() {
   std::lock_guard<std::recursive_mutex> lk(WriteLock);
 
-  // Clear L1
-  madvise(reinterpret_cast<void*>(L1Pointer), L1_SIZE, MADV_DONTNEED);
-  // Clear L2
-  ClearL2Cache();
-  // All code is gone, remove links
-  BlockLinks.clear();
+  // Clear L1 and L2 by clearing the full cache.
+  madvise(reinterpret_cast<void*>(PagePointer), TotalCacheSize, MADV_DONTNEED);
+  // Clear the BlockLinks allocator which frees the BlockLinks map implicitly.
+  BlockLinks_mbr.release();
+  // Allocate a new pointer from the BlockLinks pma again.
+  BlockLinks = BlockLinks_pma.new_object<BlockLinksMapType>();
   // All code is gone, clear the block list
   BlockList.clear();
 }

External/FEXCore/Source/Interface/Core/LookupCache.h

@@ -4,10 +4,12 @@
 #include <cstdint>
 #include <functional>
 #include <map>
+#include <memory_resource>
 #include <stddef.h>
 #include <utility>
 #include <vector>
 #include <mutex>
+#include <tsl/robin_map.h>
 
 namespace FEXCore {
 namespace Context {
@@ -17,7 +19,7 @@ namespace Context {
 class LookupCache {
 public:
 
-  struct LookupCacheEntry { 
+  struct LookupCacheEntry {
     uintptr_t HostCode;
     uintptr_t GuestCode;
   };
@@ -54,7 +56,7 @@ public:
         return L1Entry.HostCode;
       }
     }
-    
+
     // Try L3
     auto HostCode = BlockList.find(Address);
 
@@ -62,7 +64,7 @@ public:
       CacheBlockMapping(Address, HostCode->second);
       return HostCode->second;
     }
-      
+
     // Failed to find
     return 0;
   }
@@ -73,7 +75,7 @@ public:
   // Returns true if new pages are marked as containing code
   bool AddBlockExecutableRange(uint64_t Address, uint64_t Start, uint64_t Length) {
     std::lock_guard<std::recursive_mutex> lk(WriteLock);
-    
+
     bool rv = false;
 
     for (auto CurrentPage = Start >> 12, EndPage = (Start + Length -1) >> 12; CurrentPage <= EndPage; CurrentPage++) {
@@ -88,7 +90,7 @@ public:
   // Adds to Guest -> Host code mapping
   void AddBlockMapping(uint64_t Address, void *HostCode) {
     std::lock_guard<std::recursive_mutex> lk(WriteLock);
-    
+
     [[maybe_unused]] auto Inserted = BlockList.emplace(Address, (uintptr_t)HostCode).second;
     LOGMAN_THROW_AA_FMT(Inserted, "Duplicate block mapping added");
 
@@ -104,9 +106,9 @@ public:
     std::lock_guard<std::recursive_mutex> lk(WriteLock);
 
     // Sever any links to this block
-    auto lower = BlockLinks.lower_bound({Address, 0});
-    auto upper = BlockLinks.upper_bound({Address, UINTPTR_MAX});
-    for (auto it = lower; it != upper; it = BlockLinks.erase(it)) {
+    auto lower = BlockLinks->lower_bound({Address, 0});
+    auto upper = BlockLinks->upper_bound({Address, UINTPTR_MAX});
+    for (auto it = lower; it != upper; it = BlockLinks->erase(it)) {
       it->second();
     }
 
@@ -144,7 +146,7 @@ public:
   void AddBlockLink(uint64_t GuestDestination, uintptr_t HostLink, const std::function<void()> &delinker) {
     std::lock_guard<std::recursive_mutex> lk(WriteLock);
 
-    BlockLinks.insert({{GuestDestination, HostLink}, delinker});
+    BlockLinks->insert({{GuestDestination, HostLink}, delinker});
   }
 
   void ClearCache();
@@ -158,7 +160,7 @@ public:
   constexpr static size_t L1_ENTRIES_MASK = L1_ENTRIES - 1;
 
   // This needs to be taken before reads or writes to L2, L3, CodePages, Thread::DebugStore,
-  // and before writes to L1. Concurrent access from a thread that this LookupCache doesn't belong to 
+  // and before writes to L1. Concurrent access from a thread that this LookupCache doesn't belong to
   // may only happen during cross thread invalidation (::Erase).
   // All other operations must be done from the owning thread.
   // Some care is taken so that L1 lookups can be done without locks, and even tearing is unlikely to lead to a crash.
@@ -167,7 +169,7 @@ public:
   std::recursive_mutex WriteLock;
 
 private:
-  void CacheBlockMapping(uint64_t Address, uintptr_t HostCode) { 
+  void CacheBlockMapping(uint64_t Address, uintptr_t HostCode) {
     std::lock_guard<std::recursive_mutex> lk(WriteLock);
 
     // Do L1
@@ -237,9 +239,20 @@ private:
     }
   };
 
+  // Use a monotonic buffer resource to allocate both the std::pmr::map and its members.
+  // This allows us to quickly clear the block link map by clearing the monotonic allocator.
+  // If we had allocated the block link map without the MBR, then clearing the map would require slowly
+  // walking each block member and destructing objects.
+  //
+  // This makes `BlockLinks` look like a raw pointer that could memory leak, but since it is backed by the MBR, it won't.
+  std::pmr::monotonic_buffer_resource BlockLinks_mbr;
+  using BlockLinksMapType = std::pmr::map<BlockLinkTag, std::function<void()>>;
+  std::pmr::polymorphic_allocator<std::byte> BlockLinks_pma {&BlockLinks_mbr};
+  BlockLinksMapType *BlockLinks;
 
-  std::map<BlockLinkTag, std::function<void()>> BlockLinks;
-  std::map<uint64_t, uint64_t> BlockList;
+  tsl::robin_map<uint64_t, uint64_t> BlockList;
+
+  size_t TotalCacheSize;
 
   constexpr static size_t CODE_SIZE = 128 * 1024 * 1024;
   constexpr static size_t SIZE_PER_PAGE = 4096 * sizeof(LookupCacheEntry);

External/FEXCore/Source/Interface/Core/OpcodeDispatcher.h

@@ -76,10 +76,10 @@ public:
   OrderedNode* flagsOpSrcSigned{};
 
   FEXCore::Context::Context *CTX{};
-  
+
   // Used during new op bringup
   bool ShouldDump {false};
-  
+
   struct JumpTargetInfo {
     OrderedNode* BlockEntry;
     bool HaveEmitted;
@@ -153,8 +153,9 @@ public:
   OpDispatchBuilder(FEXCore::Utils::IntrusivePooledAllocator &Allocator);
 
   void ResetWorkingList();
-  void ResetDecodeFailure() { DecodeFailure = false; }
+  void ResetDecodeFailure() { NeedsBlockEnd = DecodeFailure = false; }
   bool HadDecodeFailure() const { return DecodeFailure; }
+  bool NeedsBlockEnder() const { return NeedsBlockEnd; }
 
   void BeginFunction(uint64_t RIP, std::vector<FEXCore::Frontend::Decoder::DecodedBlocks> const *Blocks);
   void Finalize();
@@ -278,6 +279,12 @@ public:
   void NOTOp(OpcodeArgs);
   void XADDOp(OpcodeArgs);
   void PopcountOp(OpcodeArgs);
+  void DAAOp(OpcodeArgs);
+  void DASOp(OpcodeArgs);
+  void AAAOp(OpcodeArgs);
+  void AASOp(OpcodeArgs);
+  void AAMOp(OpcodeArgs);
+  void AADOp(OpcodeArgs);
   void XLATOp(OpcodeArgs);
   template<bool Reseed>
   void RDRANDOp(OpcodeArgs);
@@ -292,6 +299,8 @@ public:
   void WriteSegmentReg(OpcodeArgs);
   void EnterOp(OpcodeArgs);
 
+  void SGDTOp(OpcodeArgs);
+
   // SSE
   void MOVAPSOp(OpcodeArgs);
   void MOVUPSOp(OpcodeArgs);
@@ -314,10 +323,6 @@ public:
 
   void MOVQOp(OpcodeArgs);
   template<size_t ElementSize>
-  void PADDQOp(OpcodeArgs);
-  template<size_t ElementSize>
-  void PSUBQOp(OpcodeArgs);
-  template<size_t ElementSize>
   void MOVMSKOp(OpcodeArgs);
   void MOVMSKOpOne(OpcodeArgs);
   template<size_t ElementSize>
@@ -342,8 +347,6 @@ public:
   void PSLLDQ(OpcodeArgs);
   template<size_t ElementSize>
   void PSRAIOp(OpcodeArgs);
-  template<size_t ElementSize>
-  void PAVGOp(OpcodeArgs);
   void MOVDDUPOp(OpcodeArgs);
   template<size_t DstElementSize>
   void CVTGPR_To_FPR(OpcodeArgs);
@@ -370,15 +373,16 @@ public:
   void VFCMPOp(OpcodeArgs);
   template<size_t ElementSize>
   void SHUFOp(OpcodeArgs);
-  void ANDNOp(OpcodeArgs);
   template<size_t ElementSize>
   void PINSROp(OpcodeArgs);
   void InsertPSOp(OpcodeArgs);
   template<size_t ElementSize>
   void PExtrOp(OpcodeArgs);
 
-  template<size_t ElementSize>
+  template <size_t ElementSize>
   void PSIGN(OpcodeArgs);
+  template <size_t ElementSize>
+  void VPSIGN(OpcodeArgs);
 
   // BMI1 Ops
   void ANDNBMIOp(OpcodeArgs);
@@ -398,6 +402,117 @@ public:
   // ADX Ops
   void ADXOp(OpcodeArgs);
 
+  // AVX Ops
+  template <IROps IROp, size_t ElementSize>
+  void AVXVectorALUOp(OpcodeArgs);
+  template <IROps IROp, size_t ElementSize>
+  void AVXVectorScalarALUOp(OpcodeArgs);
+  template <IROps IROp, size_t ElementSize, bool Scalar>
+  void AVXVectorUnaryOp(OpcodeArgs);
+
+  template <size_t ElementSize, size_t DstElementSize, bool Signed>
+  void AVXExtendVectorElements(OpcodeArgs);
+
+  template <size_t ElementSize, bool Scalar>
+  void AVXVectorRound(OpcodeArgs);
+
+  template <size_t SrcElementSize, bool Narrow, bool HostRoundingMode>
+  void AVXVector_CVT_Float_To_Int(OpcodeArgs);
+
+  template <size_t SrcElementSize, bool Widen>
+  void AVXVector_CVT_Int_To_Float(OpcodeArgs);
+
+  template <size_t ElementSize, bool Scalar>
+  void AVXVFCMPOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VADDSUBPOp(OpcodeArgs);
+
+  void VAESDecOp(OpcodeArgs);
+  void VAESDecLastOp(OpcodeArgs);
+  void VAESEncOp(OpcodeArgs);
+  void VAESEncLastOp(OpcodeArgs);
+  void VAESIMCOp(OpcodeArgs);
+  void VAESKeyGenAssistOp(OpcodeArgs);
+
+  void VANDNOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VBROADCASTOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VDPPOp(OpcodeArgs);
+
+  template <IROps IROp, size_t ElementSize>
+  void VHADDPOp(OpcodeArgs);
+
+  void VINSERTOp(OpcodeArgs);
+  void VINSERTPSOp(OpcodeArgs);
+
+  void VMOVAPS_VMOVAPD_Op(OpcodeArgs);
+  void VMOVUPS_VMOVUPD_Op(OpcodeArgs);
+
+  void VMOVHPOp(OpcodeArgs);
+  void VMOVLPOp(OpcodeArgs);
+
+  void VMOVDDUPOp(OpcodeArgs);
+  void VMOVSHDUPOp(OpcodeArgs);
+  void VMOVSLDUPOp(OpcodeArgs);
+
+  void VMOVVectorNTOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPACKSSOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPACKUSOp(OpcodeArgs);
+
+  void VPERM2Op(OpcodeArgs);
+  void VPERMQOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPERMILImmOp(OpcodeArgs);
+
+  void VPHMINPOSUWOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPHSUBOp(OpcodeArgs);
+
+  void VPMULHRSWOp(OpcodeArgs);
+
+  template <bool Signed>
+  void VPMULHWOp(OpcodeArgs);
+
+  template <size_t ElementSize, bool Signed>
+  void VPMULLOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPSLLOp(OpcodeArgs);
+  void VPSLLDQOp(OpcodeArgs);
+  template <size_t ElementSize>
+  void VPSLLIOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPSRAOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPSRAIOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPSRLDOp(OpcodeArgs);
+  void VPSRLDQOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPUNPCKHOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPUNPCKLOp(OpcodeArgs);
+
+  template <size_t ElementSize>
+  void VPSRLIOp(OpcodeArgs);
+
+  void VZEROOp(OpcodeArgs);
+
   // X87 Ops
   template<size_t width>
   void FLD(OpcodeArgs);
@@ -515,7 +630,7 @@ public:
   void X87FRSTORF64(OpcodeArgs);
   void X87FXAMF64(OpcodeArgs);
   void X87LDENVF64(OpcodeArgs);
-  
+
   template<size_t width, bool Integer, FCOMIFlags whichflags, bool poptwice>
   void FCOMIF64(OpcodeArgs);
 
@@ -540,12 +655,6 @@ public:
   template<bool ToXMM>
   void MOVQ2DQ(OpcodeArgs);
 
-  template<size_t ElementSize, bool Signed>
-  void PADDSOp(OpcodeArgs);
-
-  template<size_t ElementSize, bool Signed>
-  void PSUBSOp(OpcodeArgs);
-
   template<size_t ElementSize>
   void ADDSUBPOp(OpcodeArgs);
 
@@ -570,12 +679,7 @@ public:
 
   void MOVBEOp(OpcodeArgs);
   template<size_t ElementSize>
-  void HADDP(OpcodeArgs);
-  template<size_t ElementSize>
   void HSUBP(OpcodeArgs);
-
-  template<size_t ElementSize>
-  void PHADD(OpcodeArgs);
   template<size_t ElementSize>
   void PHSUB(OpcodeArgs);
 
@@ -632,8 +736,6 @@ public:
 
   void InvalidOp(OpcodeArgs);
 
-#undef OpcodeArgs
-
   void SetPackedRFLAG(bool Lower8, OrderedNode *Src);
   OrderedNode *GetPackedRFLAG(bool Lower8);
 
@@ -642,10 +744,87 @@ public:
   bool HandledLock = false;
 private:
   bool DecodeFailure{false};
+  bool NeedsBlockEnd{false};
   FEXCore::IR::IROp_IRHeader *Current_Header{};
   OrderedNode *Current_HeaderNode{};
 
+  // Opcode helpers for generalizing behavior across VEX and non-VEX variants.
+
+  OrderedNode* ADDSUBPOpImpl(OpcodeArgs, size_t ElementSize,
+                             OrderedNode *Src1, OrderedNode *Src2);
+
+  void AVXVectorALUOpImpl(OpcodeArgs, IROps IROp, size_t ElementSize);
+  void AVXVectorScalarALUOpImpl(OpcodeArgs, IROps IROp, size_t ElementSize);
+  void AVXVectorUnaryOpImpl(OpcodeArgs, IROps IROp, size_t ElementSize, bool Scalar);
+
+  OrderedNode* AESKeyGenAssistImpl(OpcodeArgs);
+  OrderedNode* AESIMCImpl(OpcodeArgs);
+
+  OrderedNode* DPPOpImpl(OpcodeArgs, const X86Tables::DecodedOperand& Src1,
+                         const X86Tables::DecodedOperand& Src2,
+                         const X86Tables::DecodedOperand& Imm, size_t ElementSize);
+
+  OrderedNode* ExtendVectorElementsImpl(OpcodeArgs, size_t ElementSize,
+                                        size_t DstElementSize, bool Signed);
+
+  OrderedNode* InsertPSOpImpl(OpcodeArgs, const X86Tables::DecodedOperand& Src1,
+                              const X86Tables::DecodedOperand& Src2,
+                              const X86Tables::DecodedOperand& Imm);
+
+  OrderedNode* PACKSSOpImpl(OpcodeArgs, size_t ElementSize,
+                            OrderedNode *Src1, OrderedNode *Src2);
+
+  OrderedNode* PACKUSOpImpl(OpcodeArgs, size_t ElementSize,
+                            OrderedNode *Src1, OrderedNode *Src2);
+
+  OrderedNode* PHMINPOSUWOpImpl(OpcodeArgs);
+
+  OrderedNode* PHSUBOpImpl(OpcodeArgs, const X86Tables::DecodedOperand& Src1,
+                           const X86Tables::DecodedOperand& Src2, size_t ElementSize);
+
+  OrderedNode* PMULHRSWOpImpl(OpcodeArgs, OrderedNode *Src1, OrderedNode *Src2);
+
+  OrderedNode* PMULHWOpImpl(OpcodeArgs, bool Signed,
+                            OrderedNode *Src1, OrderedNode *Src2);
+
+  OrderedNode* PMULLOpImpl(OpcodeArgs, size_t ElementSize, bool Signed,
+                           OrderedNode *Src1, OrderedNode *Src2);
+
+  OrderedNode* PSIGNImpl(OpcodeArgs, size_t ElementSize,
+                         OrderedNode *Src1, OrderedNode *Src2);
+
+  OrderedNode* PSLLIImpl(OpcodeArgs, size_t ElementSize,
+                         OrderedNode *Src, uint64_t Shift);
+
+  OrderedNode* PSLLImpl(OpcodeArgs, size_t ElementSize,
+                        OrderedNode *Src, OrderedNode *ShiftVec);
+
+  OrderedNode* PSRAOpImpl(OpcodeArgs, size_t ElementSize,
+                          OrderedNode *Src, OrderedNode *ShiftVec);
+
+  OrderedNode* PSRLDOpImpl(OpcodeArgs, size_t ElementSize,
+                           OrderedNode *Src, OrderedNode *ShiftVec);
+
+  OrderedNode* VFCMPOpImpl(OpcodeArgs, size_t ElementSize, bool Scalar,
+                           OrderedNode *Src1, OrderedNode *Src2, uint8_t CompType);
+
+  void VectorALUOpImpl(OpcodeArgs, IROps IROp, size_t ElementSize);
+  void VectorALUROpImpl(OpcodeArgs, IROps IROp, size_t ElementSize);
+  void VectorScalarALUOpImpl(OpcodeArgs, IROps IROp, size_t ElementSize);
+  void VectorUnaryOpImpl(OpcodeArgs, IROps IROp, size_t ElementSize, bool Scalar);
+  void VectorUnaryDuplicateOpImpl(OpcodeArgs, IROps IROp, size_t ElementSize);
+
+  OrderedNode* VectorRoundImpl(OpcodeArgs, size_t ElementSize,
+                               OrderedNode *Src, uint64_t Mode);
+
+  OrderedNode* Vector_CVT_Float_To_IntImpl(OpcodeArgs, size_t SrcElementSize, bool Narrow, bool HostRoundingMode);
+
+  OrderedNode* Vector_CVT_Int_To_FloatImpl(OpcodeArgs, size_t SrcElementSize, bool Widen);
+
+  #undef OpcodeArgs
+
   OrderedNode *AppendSegmentOffset(OrderedNode *Value, uint32_t Flags, uint32_t DefaultPrefix = 0, bool Override = false);
+  void UpdatePrefixFromSegment(OrderedNode *Segment, uint32_t SegmentReg);
 
   enum class MemoryAccessType {
     // Choose TSO or Non-TSO depending on access type
@@ -657,6 +836,11 @@ private:
     // Non-temporal streaming
     ACCESS_STREAM,
   };
+  OrderedNode *LoadGPRRegister(uint32_t GPR, int8_t Size = -1, uint8_t Offset = 0);
+  OrderedNode *LoadXMMRegister(uint32_t XMM);
+  void StoreGPRRegister(uint32_t GPR, OrderedNode *const Src, int8_t Size = -1, uint8_t Offset = 0);
+  void StoreXMMRegister(uint32_t XMM, OrderedNode *const Src);
+
   OrderedNode *GetRelocatedPC(FEXCore::X86Tables::DecodedOp const& Op, int64_t Offset = 0);
   OrderedNode *LoadSource(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint32_t Flags, int8_t Align, bool LoadData = true, bool ForceLoad = false, MemoryAccessType AccessType = MemoryAccessType::ACCESS_DEFAULT);
   OrderedNode *LoadSource_WithOpSize(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp const& Op, FEXCore::X86Tables::DecodedOperand const& Operand, uint8_t OpSize, uint32_t Flags, int8_t Align, bool LoadData = true, bool ForceLoad = false, MemoryAccessType AccessType = MemoryAccessType::ACCESS_DEFAULT);

External/FEXCore/Source/Interface/Core/OpcodeDispatcher/Crypto.cpp

@@ -35,19 +35,16 @@ void OpDispatchBuilder::SHA1MSG1Op(OpcodeArgs) {
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
 
-  auto W0 = _VExtractToGPR(16, 4, Dest, 3);
-  auto W1 = _VExtractToGPR(16, 4, Dest, 2);
-  auto W2 = _VExtractToGPR(16, 4, Dest, 1);
-  auto W3 = _VExtractToGPR(16, 4, Dest, 0);
-  auto W4 = _VExtractToGPR(16, 4, Src, 3);
-  auto W5 = _VExtractToGPR(16, 4, Src, 2);
+  OrderedNode *NewVec{};
+  NewVec = _VInsElement(16, 4, 3, 1, Dest, Dest);
+  NewVec = _VInsElement(16, 4, 2, 0, NewVec, Dest);
+  NewVec = _VInsElement(16, 4, 1, 3, NewVec, Src);
+  NewVec = _VInsElement(16, 4, 0, 2, NewVec, Src);
 
-  auto D3 = _VInsGPR(16, 4, 3, Dest, _Xor(W2, W0));
-  auto D2 = _VInsGPR(16, 4, 2, D3,   _Xor(W3, W1));
-  auto D1 = _VInsGPR(16, 4, 1, D2,   _Xor(W4, W2));
-  auto D0 = _VInsGPR(16, 4, 0, D1,   _Xor(W5, W3));
+  // [W0, W1, W2, W3] ^ [W2, W3, W4, W5]
+  OrderedNode *Result = _VXor(16, 1, Dest, NewVec);
 
-  StoreResult(FPRClass, Op, D0, -1);
+  StoreResult(FPRClass, Op, Result, -1);
 }
 
 void OpDispatchBuilder::SHA1MSG2Op(OpcodeArgs) {
@@ -221,7 +218,8 @@ void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
 
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
-  OrderedNode *XMM0 = _LoadContext(16, FPRClass, offsetof(FEXCore::Core::CPUState, xmm.avx.data[0]));
+  // Hardcoded to XMM0
+  auto XMM0 = LoadXMMRegister(0);
 
   auto A0 = _VExtractToGPR(16, 4, Src, 3);
   auto B0 = _VExtractToGPR(16, 4, Src, 2);
@@ -263,47 +261,135 @@ void OpDispatchBuilder::SHA256RNDS2Op(OpcodeArgs) {
   StoreResult(FPRClass, Op, Res0, -1);
 }
 
-void OpDispatchBuilder::AESImcOp(OpcodeArgs) {
+OrderedNode* OpDispatchBuilder::AESIMCImpl(OpcodeArgs) {
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
-  auto Res = _VAESImc(Src);
-  StoreResult(FPRClass, Op, Res, -1);
+  return _VAESImc(Src);
+}
+
+void OpDispatchBuilder::AESImcOp(OpcodeArgs) {
+  OrderedNode *Result = AESIMCImpl(Op);
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::VAESIMCOp(OpcodeArgs) {
+  OrderedNode *Mixed = AESIMCImpl(Op);
+  OrderedNode *Result = _VMov(16, Mixed);
+  StoreResult(FPRClass, Op, Result, -1);
 }
 
 void OpDispatchBuilder::AESEncOp(OpcodeArgs) {
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
-  auto Res = _VAESEnc(Dest, Src);
-  StoreResult(FPRClass, Op, Res, -1);
+  OrderedNode *Result = _VAESEnc(Dest, Src);
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::VAESEncOp(OpcodeArgs) {
+  const auto DstSize = GetDstSize(Op);
+  const auto Is128Bit = DstSize == Core::CPUState::XMM_SSE_REG_SIZE;
+
+  // TODO: Handle 256-bit VAESENC.
+  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESENC unimplemented");
+
+  OrderedNode *State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  OrderedNode *Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags, -1);
+  OrderedNode *Result = _VAESEnc(State, Key);
+
+  if (Is128Bit) {
+    Result = _VMov(16, Result);
+  }
+  StoreResult(FPRClass, Op, Result, -1);
 }
 
 void OpDispatchBuilder::AESEncLastOp(OpcodeArgs) {
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
-  auto Res = _VAESEncLast(Dest, Src);
-  StoreResult(FPRClass, Op, Res, -1);
+  OrderedNode *Result = _VAESEncLast(Dest, Src);
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::VAESEncLastOp(OpcodeArgs) {
+  const auto DstSize = GetDstSize(Op);
+  const auto Is128Bit = DstSize == Core::CPUState::XMM_SSE_REG_SIZE;
+
+  // TODO: Handle 256-bit VAESENCLAST.
+  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESENCLAST unimplemented");
+
+  OrderedNode *State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  OrderedNode *Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags, -1);
+  OrderedNode *Result = _VAESEncLast(State, Key);
+
+  if (Is128Bit) {
+    Result = _VMov(16, Result);
+  }
+  StoreResult(FPRClass, Op, Result, -1);
 }
 
 void OpDispatchBuilder::AESDecOp(OpcodeArgs) {
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
-  auto Res = _VAESDec(Dest, Src);
-  StoreResult(FPRClass, Op, Res, -1);
+  OrderedNode *Result = _VAESDec(Dest, Src);
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::VAESDecOp(OpcodeArgs) {
+  const auto DstSize = GetDstSize(Op);
+  const auto Is128Bit = DstSize == Core::CPUState::XMM_SSE_REG_SIZE;
+
+  // TODO: Handle 256-bit VAESDEC.
+  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESDEC unimplemented");
+
+  OrderedNode *State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  OrderedNode *Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags, -1);
+  OrderedNode *Result = _VAESDec(State, Key);
+
+  if (Is128Bit) {
+    Result = _VMov(16, Result);
+  }
+  StoreResult(FPRClass, Op, Result, -1);
 }
 
 void OpDispatchBuilder::AESDecLastOp(OpcodeArgs) {
   OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
   OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
-  auto Res = _VAESDecLast(Dest, Src);
-  StoreResult(FPRClass, Op, Res, -1);
+  OrderedNode *Result = _VAESDecLast(Dest, Src);
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::VAESDecLastOp(OpcodeArgs) {
+  const auto DstSize = GetDstSize(Op);
+  const auto Is128Bit = DstSize == Core::CPUState::XMM_SSE_REG_SIZE;
+
+  // TODO: Handle 256-bit VAESDECLAST.
+  LOGMAN_THROW_A_FMT(Is128Bit, "256-bit VAESDECLAST unimplemented");
+
+  OrderedNode *State = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  OrderedNode *Key = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags, -1);
+  OrderedNode *Result = _VAESDecLast(State, Key);
+
+  if (Is128Bit) {
+    Result = _VMov(16, Result);
+  }
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+OrderedNode* OpDispatchBuilder::AESKeyGenAssistImpl(OpcodeArgs) {
+  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
+  LOGMAN_THROW_A_FMT(Op->Src[1].IsLiteral(), "Src1 needs to be literal here");
+  const uint64_t RCON = Op->Src[1].Data.Literal.Value;
+
+  return _VAESKeyGenAssist(Src, RCON);
 }
 
 void OpDispatchBuilder::AESKeyGenAssist(OpcodeArgs) {
-  OrderedNode *Src = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
-  LOGMAN_THROW_A_FMT(Op->Src[1].IsLiteral(), "Src1 needs to be literal here");
-  uint64_t RCON = Op->Src[1].Data.Literal.Value;
+  OrderedNode *Result = AESKeyGenAssistImpl(Op);
+  StoreResult(FPRClass, Op, Result, -1);
+}
 
-  auto Res = _VAESKeyGenAssist(Src, RCON);
-  StoreResult(FPRClass, Op, Res, -1);
+void OpDispatchBuilder::VAESKeyGenAssistOp(OpcodeArgs) {
+  OrderedNode *Assist = AESKeyGenAssistImpl(Op);
+  OrderedNode *Result = _VMov(16, Assist);
+  StoreResult(FPRClass, Op, Result, -1);
 }
 
 void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
@@ -320,11 +406,18 @@ void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
 void OpDispatchBuilder::VPCLMULQDQOp(OpcodeArgs) {
   LOGMAN_THROW_A_FMT(Op->Src[2].IsLiteral(), "Selector needs to be literal here");
 
+  const auto DstSize = GetDstSize(Op);
+  const auto Is128Bit = DstSize == Core::CPUState::XMM_SSE_REG_SIZE;
+
   OrderedNode *Src1 = LoadSource(FPRClass, Op, Op->Src[0], Op->Flags, -1);
   OrderedNode *Src2 = LoadSource(FPRClass, Op, Op->Src[1], Op->Flags, -1);
   const auto Selector = static_cast<uint8_t>(Op->Src[2].Data.Literal.Value);
 
-  auto Res = _PCLMUL(Src1, Src2, Selector);
+  OrderedNode *Res = _PCLMUL(Src1, Src2, Selector);
+  if (Is128Bit) {
+    Res = _VMov(16, Res);
+  }
+
   StoreResult(FPRClass, Op, Res, -1);
 }
 

External/FEXCore/Source/Interface/Core/OpcodeDispatcher/Flags.cpp

@@ -769,7 +769,11 @@ void OpDispatchBuilder::CalculcateFlags_ShiftLeftImmediate(uint8_t SrcSize, Orde
   // CF
   {
     // Extract the last bit shifted in to CF
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(_Bfe(1, SrcSize * 8 - Shift, Src1));
+    auto OpSize = SrcSize * 8;
+    if (OpSize < Shift) {
+      Shift &= (OpSize - 1);
+    }
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(_Bfe(1, OpSize - Shift, Src1));
   }
 
   // PF
@@ -934,6 +938,7 @@ void OpDispatchBuilder::CalculcateFlags_RotateRight(uint8_t SrcSize, OrderedNode
     auto OldOF = GetRFLAG(FEXCore::X86State::RFLAG_OF_LOC);
 
     // OF is set to the XOR of the new CF bit and the most significant bit of the result
+    // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
     auto NewOF = _Xor(_Bfe(1, OpSize - 2, Res), NewCF);
 
     // If shift == 0, don't update flags
@@ -963,7 +968,9 @@ void OpDispatchBuilder::CalculcateFlags_RotateLeft(uint8_t SrcSize, OrderedNode
   // OF
   {
     auto OldOF = GetRFLAG(FEXCore::X86State::RFLAG_OF_LOC);
-    // OF is set to the XOR of the new CF bit and the most significant bit of the result
+    // OF is the LSB and MSB XOR'd together.
+    // OF is set to the XOR of the new CF bit and the most significant bit of the result.
+    // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
     auto NewOF = _Xor(_Bfe(1, OpSize - 1, Res), NewCF);
 
     auto OF = _Select(FEXCore::IR::COND_EQ, Src2, _Constant(0), OldOF, NewOF);
@@ -977,8 +984,7 @@ void OpDispatchBuilder::CalculcateFlags_RotateRightImmediate(uint8_t SrcSize, Or
   if (Shift == 0) return;
 
   auto OpSize = SrcSize * 8;
-
-  auto NewCF = _Bfe(1, OpSize - Shift, Src1);
+  auto NewCF = _Bfe(1, OpSize - 1, Res);
 
   // CF
   {
@@ -989,8 +995,10 @@ void OpDispatchBuilder::CalculcateFlags_RotateRightImmediate(uint8_t SrcSize, Or
   // OF
   {
     if (Shift == 1) {
-      // OF is set to the XOR of the new CF bit and the most significant bit of the result
-      SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(_Xor(_Bfe(1, OpSize - 1, Res), NewCF));
+      // OF is the top two MSBs XOR'd together
+      // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
+      auto NewOF = _Xor(_Bfe(1, OpSize - 2, Res), NewCF);
+      SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(NewOF);
     }
   }
 }
@@ -1000,17 +1008,22 @@ void OpDispatchBuilder::CalculcateFlags_RotateLeftImmediate(uint8_t SrcSize, Ord
 
   auto OpSize = SrcSize * 8;
 
+  auto NewCF = _Bfe(1, 0, Res);
   // CF
   {
     // Extract the last bit shifted in to CF
-    SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(_Bfe(1, Shift, Src1));
+    SetRFLAG<FEXCore::X86State::RFLAG_CF_LOC>(NewCF);
   }
 
   // OF
   {
     if (Shift == 1) {
-      // OF is the top two MSBs XOR'd together
-      SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(_Xor(_Bfe(1, OpSize - 1, Src1), _Bfe(1, OpSize - 2, Src1)));
+      // OF is the LSB and MSB XOR'd together.
+      // OF is set to the XOR of the new CF bit and the most significant bit of the result.
+      // OF is architecturally only defined for 1-bit rotate, which is why this only happens when the shift is one.
+      auto NewOF = _Xor(_Bfe(1, OpSize - 1, Res), NewCF);
+
+      SetRFLAG<FEXCore::X86State::RFLAG_OF_LOC>(NewOF);
     }
   }
 }

External/FEXCore/Source/Interface/Core/OpcodeDispatcher/X87.cpp

@@ -782,6 +782,12 @@ void OpDispatchBuilder::X87UnaryOp(OpcodeArgs) {
   // Overwrite the op
   result.first->Header.Op = IROp;
 
+  if constexpr (IROp == IR::OP_F80SIN ||
+                IROp == IR::OP_F80COS) {
+    // TODO: ACCURACY: should check source is in range –2^63 to +2^63
+    SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(_Constant(0));
+  }
+
   // Write to ST[TOP]
   _StoreContextIndexed(result, top, 16, MMBaseOffset(), 16, FPRClass);
 }
@@ -809,7 +815,8 @@ void OpDispatchBuilder::X87BinaryOp(OpcodeArgs) {
   // Overwrite the op
   result.first->Header.Op = IROp;
 
-  if constexpr (IROp == IR::OP_F80FPREM) {
+  if constexpr (IROp == IR::OP_F80FPREM ||
+    IROp == IR::OP_F80FPREM1) {
     //TODO: Set C0 to Q2, C3 to Q1, C1 to Q0
     SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(_Constant(0));
   }
@@ -854,6 +861,9 @@ void OpDispatchBuilder::X87SinCos(OpcodeArgs) {
   auto sin = _F80SIN(a);
   auto cos = _F80COS(a);
 
+  // TODO: ACCURACY: should check source is in range –2^63 to +2^63
+  SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(_Constant(0));
+
   // Write to ST[TOP]
   _StoreContextIndexed(sin, orig_top, 16, MMBaseOffset(), 16, FPRClass);
   _StoreContextIndexed(cos, top, 16, MMBaseOffset(), 16, FPRClass);
@@ -900,6 +910,9 @@ void OpDispatchBuilder::X87TAN(OpcodeArgs) {
   OrderedNode *data = _VCastFromGPR(16, 8, low);
   data = _VInsGPR(16, 8, 1, data, high);
 
+  // TODO: ACCURACY: should check source is in range –2^63 to +2^63
+  SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(_Constant(0));
+
   // Write to ST[TOP]
   _StoreContextIndexed(result, orig_top, 16, MMBaseOffset(), 16, FPRClass);
   _StoreContextIndexed(data, top, 16, MMBaseOffset(), 16, FPRClass);

External/FEXCore/Source/Interface/Core/OpcodeDispatcher/X87F64.cpp

@@ -778,6 +778,12 @@ void OpDispatchBuilder::X87UnaryOpF64(OpcodeArgs) {
   // Overwrite the op
   result.first->Header.Op = IROp;
 
+  if constexpr (IROp == IR::OP_F64SIN ||
+                IROp == IR::OP_F64COS) {
+    // TODO: ACCURACY: should check source is in range –2^63 to +2^63
+    SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(_Constant(0));
+  }
+
   // Write to ST[TOP]
   _StoreContextIndexed(result, top, 8, MMBaseOffset(), 16, FPRClass);
 }
@@ -804,7 +810,8 @@ void OpDispatchBuilder::X87BinaryOpF64(OpcodeArgs) {
   // Overwrite the op
   result.first->Header.Op = IROp;
 
-  if constexpr (IROp == IR::OP_F64FPREM) {
+  if constexpr (IROp == IR::OP_F80FPREM ||
+    IROp == IR::OP_F80FPREM1) {
     //TODO: Set C0 to Q2, C3 to Q1, C1 to Q0
     SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(_Constant(0));
   }
@@ -831,6 +838,9 @@ void OpDispatchBuilder::X87SinCosF64(OpcodeArgs) {
   auto sin = _F64SIN(a);
   auto cos = _F64COS(a);
 
+  // TODO: ACCURACY: should check source is in range –2^63 to +2^63
+  SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(_Constant(0));
+
   // Write to ST[TOP]
   _StoreContextIndexed(sin, orig_top, 8, MMBaseOffset(), 16, FPRClass);
   _StoreContextIndexed(cos, top, 8, MMBaseOffset(), 16, FPRClass);
@@ -871,6 +881,9 @@ void OpDispatchBuilder::X87TANF64(OpcodeArgs) {
 
   auto one = _VCastFromGPR(8, 8, _Constant(0x3FF0000000000000));
 
+  // TODO: ACCURACY: should check source is in range –2^63 to +2^63
+  SetRFLAG<FEXCore::X86State::X87FLAG_C2_LOC>(_Constant(0));
+
   // Write to ST[TOP]
   _StoreContextIndexed(result, orig_top, 8, MMBaseOffset(), 16, FPRClass);
   _StoreContextIndexed(one, top, 8, MMBaseOffset(), 16, FPRClass);

External/FEXCore/Source/Interface/Core/X86Tables/BaseTables.cpp

@@ -266,10 +266,10 @@ void InitializeBaseTables(Context::OperatingMode Mode) {
     {0x17, 1, X86InstInfo{"POP SS",   TYPE_INST, GenFlagsSizes(SIZE_16BIT, SIZE_DEF) | FLAGS_DEBUG_MEM_ACCESS,    0, nullptr}},
     {0x1E, 1, X86InstInfo{"PUSH DS",  TYPE_INST, GenFlagsSrcSize(SIZE_16BIT) | FLAGS_DEBUG_MEM_ACCESS,            0, nullptr}},
     {0x1F, 1, X86InstInfo{"POP DS",   TYPE_INST, GenFlagsSizes(SIZE_16BIT, SIZE_DEF) | FLAGS_DEBUG_MEM_ACCESS,    0, nullptr}},
-    {0x27, 1, X86InstInfo{"DAA",      TYPE_INST, FLAGS_NONE,                                                      0, nullptr}},
-    {0x2F, 1, X86InstInfo{"DAS",      TYPE_INST, FLAGS_NONE,                                                      0, nullptr}},
-    {0x37, 1, X86InstInfo{"AAA",      TYPE_INST, FLAGS_NONE,                                                      0, nullptr}},
-    {0x3F, 1, X86InstInfo{"AAS",      TYPE_INST, FLAGS_NONE,                                                      0, nullptr}},
+    {0x27, 1, X86InstInfo{"DAA",      TYPE_INST, GenFlagsDstSize(SIZE_8BIT) | FLAGS_SF_DST_RAX,                   0, nullptr}},
+    {0x2F, 1, X86InstInfo{"DAS",      TYPE_INST, GenFlagsDstSize(SIZE_8BIT) | FLAGS_SF_DST_RAX,                   0, nullptr}},
+    {0x37, 1, X86InstInfo{"AAA",      TYPE_INST, GenFlagsDstSize(SIZE_16BIT) | FLAGS_SF_DST_RAX,                  0, nullptr}},
+    {0x3F, 1, X86InstInfo{"AAS",      TYPE_INST, GenFlagsDstSize(SIZE_16BIT) | FLAGS_SF_DST_RAX,                  0, nullptr}},
 
     {0x40, 8, X86InstInfo{"INC",    TYPE_INST, FLAGS_SF_REX_IN_BYTE,                                              0, nullptr}},
     {0x48, 8, X86InstInfo{"DEC",    TYPE_INST, FLAGS_SF_REX_IN_BYTE,                                              0, nullptr}},
@@ -283,8 +283,8 @@ void InitializeBaseTables(Context::OperatingMode Mode) {
     {0xA1, 1, X86InstInfo{"MOV",    TYPE_INST, FLAGS_SF_DST_RAX | FLAGS_MEM_OFFSET,                               4, nullptr}},
     {0xA3, 1, X86InstInfo{"MOV",    TYPE_INST, FLAGS_SF_SRC_RAX | FLAGS_MEM_OFFSET,                               4, nullptr}},
     {0xCE, 1, X86InstInfo{"INTO",   TYPE_INST, FLAGS_NONE,                                                        0, nullptr}},
-    {0xD4, 1, X86InstInfo{"AAM",    TYPE_INST, FLAGS_NONE,                                                        1, nullptr}},
-    {0xD5, 1, X86InstInfo{"AAD",    TYPE_INST, FLAGS_NONE,                                                        1, nullptr}},
+    {0xD4, 1, X86InstInfo{"AAM",    TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_SF_DST_RAX,                    1, nullptr}},
+    {0xD5, 1, X86InstInfo{"AAD",    TYPE_INST, GenFlagsSameSize(SIZE_8BIT) | FLAGS_SF_DST_RAX,                    1, nullptr}},
     {0xEA, 1, X86InstInfo{"JMPF",   TYPE_INST, FLAGS_NONE,                                                        0, nullptr}},
   };
 

External/FEXCore/Source/Interface/Core/X86Tables/H0F3ATables.cpp

@@ -24,8 +24,8 @@ void InitializeH0F3ATables(Context::OperatingMode Mode) {
     {OPD(0, PF_3A_NONE, 0x0F), 1, X86InstInfo{"PALIGNR",         TYPE_INST, GenFlagsSameSize(SIZE_64BIT)  | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_MMX, 1, nullptr}},
     {OPD(0, PF_3A_66,   0x08), 1, X86InstInfo{"ROUNDPS",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(0, PF_3A_66,   0x09), 1, X86InstInfo{"ROUNDPD",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
-    {OPD(0, PF_3A_66,   0x0A), 1, X86InstInfo{"ROUNDSS",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
-    {OPD(0, PF_3A_66,   0x0B), 1, X86InstInfo{"ROUNDSD",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(0, PF_3A_66,   0x0A), 1, X86InstInfo{"ROUNDSS",         TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(0, PF_3A_66,   0x0B), 1, X86InstInfo{"ROUNDSD",         TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(0, PF_3A_66,   0x0C), 1, X86InstInfo{"BLENDPS",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,           1, nullptr}},
     {OPD(0, PF_3A_66,   0x0D), 1, X86InstInfo{"BLENDPD",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,           1, nullptr}},
     {OPD(0, PF_3A_66,   0x0E), 1, X86InstInfo{"PBLENDW",         TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,           1, nullptr}},

External/FEXCore/Source/Interface/Core/X86Tables/SecondaryGroupTables.cpp

@@ -67,7 +67,7 @@ void InitializeSecondaryGroupTables() {
     {OPD(TYPE_GROUP_6, PF_F2, 7), 1, X86InstInfo{"",        TYPE_INVALID, FLAGS_NONE,    0, nullptr}},
 
     // GROUP 7
-    {OPD(TYPE_GROUP_7, PF_NONE, 0), 1, X86InstInfo{"SGDT", TYPE_UNDEC, FLAGS_MODRM | FLAGS_SF_MOD_DST,         0, nullptr}},
+    {OPD(TYPE_GROUP_7, PF_NONE, 0), 1, X86InstInfo{"SGDT", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST,         0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_NONE, 1), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE, 0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_NONE, 2), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE, 0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_NONE, 3), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE, 0, nullptr}},
@@ -76,7 +76,7 @@ void InitializeSecondaryGroupTables() {
     {OPD(TYPE_GROUP_7, PF_NONE, 6), 1, X86InstInfo{"LMSW", TYPE_PRIV, FLAGS_MODRM,          0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_NONE, 7), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(TYPE_GROUP_7, PF_F3, 0), 1, X86InstInfo{"SGDT", TYPE_UNDEC, FLAGS_MODRM | FLAGS_SF_MOD_DST,           0, nullptr}},
+    {OPD(TYPE_GROUP_7, PF_F3, 0), 1, X86InstInfo{"SGDT", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST,           0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_F3, 1), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_F3, 2), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_F3, 3), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
@@ -85,7 +85,7 @@ void InitializeSecondaryGroupTables() {
     {OPD(TYPE_GROUP_7, PF_F3, 6), 1, X86InstInfo{"LMSW", TYPE_PRIV, FLAGS_MODRM,            0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_F3, 7), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
 
-    {OPD(TYPE_GROUP_7, PF_66, 0), 1, X86InstInfo{"SGDT", TYPE_UNDEC, FLAGS_MODRM | FLAGS_SF_MOD_DST,           0, nullptr}},
+    {OPD(TYPE_GROUP_7, PF_66, 0), 1, X86InstInfo{"SGDT", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST,           0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_66, 1), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_66, 2), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_66, 3), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
@@ -94,7 +94,7 @@ void InitializeSecondaryGroupTables() {
     {OPD(TYPE_GROUP_7, PF_66, 6), 1, X86InstInfo{"LMSW", TYPE_PRIV, FLAGS_MODRM,            0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_66, 7), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
 
-    {OPD(TYPE_GROUP_7, PF_F2, 0), 1, X86InstInfo{"SGDT", TYPE_UNDEC, FLAGS_MODRM | FLAGS_SF_MOD_DST,           0, nullptr}},
+    {OPD(TYPE_GROUP_7, PF_F2, 0), 1, X86InstInfo{"SGDT", TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST,           0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_F2, 1), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_F2, 2), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},
     {OPD(TYPE_GROUP_7, PF_F2, 3), 1, X86InstInfo{"",     TYPE_SECOND_GROUP_MODRM, FLAGS_NONE,   0, nullptr}},

External/FEXCore/Source/Interface/Core/X86Tables/SecondaryTables.cpp

@@ -295,7 +295,7 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
 
     {0x20, 4, X86InstInfo{"",          TYPE_COPY_OTHER, FLAGS_NONE,                                     0, nullptr}},
     {0x24, 6, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                        0, nullptr}},
-    {0x2A, 1, X86InstInfo{"CVTSI2SS",  TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_SRC_GPR, 0, nullptr}},
+    {0x2A, 1, X86InstInfo{"CVTSI2SS",  TYPE_INST, GenFlagsDstSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_SRC_GPR, 0, nullptr}},
     {0x2B, 1, X86InstInfo{"MOVNTSS",   TYPE_INST, GenFlagsSameSize(SIZE_32BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
     {0x2C, 1, X86InstInfo{"CVTTSS2SI", TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
     {0x2D, 1, X86InstInfo{"CVTSS2SI",  TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
@@ -305,18 +305,18 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0x40, 16, X86InstInfo{"",         TYPE_COPY_OTHER, FLAGS_NONE,                                     0, nullptr}},
 
     {0x50, 1, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                        0, nullptr}},
-    {0x51, 1, X86InstInfo{"SQRTSS",    TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x52, 1, X86InstInfo{"RSQRTSS",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x53, 1, X86InstInfo{"RCPSS",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x51, 1, X86InstInfo{"SQRTSS",    TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x52, 1, X86InstInfo{"RSQRTSS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x53, 1, X86InstInfo{"RCPSS",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
     {0x54, 4, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                        0, nullptr}},
-    {0x58, 1, X86InstInfo{"ADDSS",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x59, 1, X86InstInfo{"MULSS",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x5A, 1, X86InstInfo{"CVTSS2SD",  TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,   0, nullptr}},
+    {0x58, 1, X86InstInfo{"ADDSS",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x59, 1, X86InstInfo{"MULSS",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x5A, 1, X86InstInfo{"CVTSS2SD",  TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,   0, nullptr}},
     {0x5B, 1, X86InstInfo{"CVTTPS2DQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,  0, nullptr}},
-    {0x5C, 1, X86InstInfo{"SUBSS",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x5D, 1, X86InstInfo{"MINSS",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x5E, 1, X86InstInfo{"DIVSS",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x5F, 1, X86InstInfo{"MAXSS",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x5C, 1, X86InstInfo{"SUBSS",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x5D, 1, X86InstInfo{"MINSS",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x5E, 1, X86InstInfo{"DIVSS",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x5F, 1, X86InstInfo{"MAXSS",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
 
     {0x60, 8, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                        0, nullptr}},
     {0x68, 7, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                        0, nullptr}},
@@ -353,7 +353,7 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0xD8, 8, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                        0, nullptr}},
 
     {0xE0, 6, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                        0, nullptr}},
-    {0xE6, 1, X86InstInfo{"CVTDQ2PD",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,  0, nullptr}},
+    {0xE6, 1, X86InstInfo{"CVTDQ2PD",  TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,  0, nullptr}},
     {0xE7, 1, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                        0, nullptr}},
     {0xE8, 8, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                        0, nullptr}},
 
@@ -375,7 +375,7 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0x24, 6, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                                         0, nullptr}},
     {0x2A, 1, X86InstInfo{"CVTSI2SD",  TYPE_INST, GenFlagsDstSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_SRC_GPR,                    0, nullptr}},
     {0x2B, 1, X86InstInfo{"MOVNTSD",   TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {0x2C, 1, X86InstInfo{"CVTTSD2SI", TYPE_INST, GenFlagsSrcSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
+    {0x2C, 1, X86InstInfo{"CVTTSD2SI", TYPE_INST, GenFlagsSrcSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
     {0x2D, 1, X86InstInfo{"CVTSD2SI",  TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
     {0x2E, 2, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                                         0, nullptr}},
 
@@ -383,16 +383,16 @@ void InitializeSecondaryTables(Context::OperatingMode Mode) {
     {0x40, 16, X86InstInfo{"",         TYPE_COPY_OTHER, FLAGS_NONE,                                                      0, nullptr}},
 
     {0x50, 1, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                                         0, nullptr}},
-    {0x51, 1, X86InstInfo{"SQRTSD",    TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x51, 1, X86InstInfo{"SQRTSD",    TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {0x52, 6, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                                         0, nullptr}},
-    {0x58, 1, X86InstInfo{"ADDSD",     TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x59, 1, X86InstInfo{"MULSD",     TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x58, 1, X86InstInfo{"ADDSD",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x59, 1, X86InstInfo{"MULSD",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
     {0x5A, 1, X86InstInfo{"CVTSD2SS",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
     {0x5B, 1, X86InstInfo{"",          TYPE_INVALID, FLAGS_NONE,                                                         0, nullptr}},
-    {0x5C, 1, X86InstInfo{"SUBSD",     TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x5D, 1, X86InstInfo{"MINSD",     TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x5E, 1, X86InstInfo{"DIVSD",     TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
-    {0x5F, 1, X86InstInfo{"MAXSD",     TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x5C, 1, X86InstInfo{"SUBSD",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x5D, 1, X86InstInfo{"MINSD",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x5E, 1, X86InstInfo{"DIVSD",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
+    {0x5F, 1, X86InstInfo{"MAXSD",     TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS,                    0, nullptr}},
 
     {0x60, 16, X86InstInfo{"",         TYPE_INVALID, FLAGS_NONE,                                                         0, nullptr}},
 

External/FEXCore/Source/Interface/Core/X86Tables/VEXTables.cpp

@@ -17,71 +17,71 @@ void InitializeVEXTables() {
   static constexpr U16U8InfoStruct VEXTable[] = {
     // Map 0 (Reserved)
     // VEX Map 1
-    {OPD(1, 0b00, 0x10), 1, X86InstInfo{"VMOVUPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x10), 1, X86InstInfo{"VMODUPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x10), 1, X86InstInfo{"VMOVUPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x10), 1, X86InstInfo{"VMOVUPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(1, 0b10, 0x10), 1, X86InstInfo{"VMOVSS",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(1, 0b11, 0x10), 1, X86InstInfo{"VMOVSD",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x11), 1, X86InstInfo{"VMOVUPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x11), 1, X86InstInfo{"VMODUPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x11), 1, X86InstInfo{"VMOVUPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x11), 1, X86InstInfo{"VMOVUPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(1, 0b10, 0x11), 1, X86InstInfo{"VMOVSS",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(1, 0b11, 0x11), 1, X86InstInfo{"VMOVSD",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x12), 1, X86InstInfo{"VMOVLPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x12), 1, X86InstInfo{"VMOVLPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x12), 1, X86InstInfo{"VMOVSLDUP", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x12), 1, X86InstInfo{"VMOVDDUP",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x12), 1, X86InstInfo{"VMOVLPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_XMM_FLAGS | FLAGS_VEX_1ST_SRC, 0, nullptr}},
+    {OPD(1, 0b01, 0x12), 1, X86InstInfo{"VMOVLPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_XMM_FLAGS | FLAGS_VEX_1ST_SRC, 0, nullptr}},
+    {OPD(1, 0b10, 0x12), 1, X86InstInfo{"VMOVSLDUP", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0x12), 1, X86InstInfo{"VMOVDDUP",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x13), 1, X86InstInfo{"VMOVLPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x13), 1, X86InstInfo{"VMOVLPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x13), 1, X86InstInfo{"VMOVLPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x13), 1, X86InstInfo{"VMOVLPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x14), 1, X86InstInfo{"VUNPCKLPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x14), 1, X86InstInfo{"VUNPCKLPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x14), 1, X86InstInfo{"VUNPCKLPS", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x14), 1, X86InstInfo{"VUNPCKLPD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x15), 1, X86InstInfo{"VUNPCKHPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x15), 1, X86InstInfo{"VUNPCKHPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x15), 1, X86InstInfo{"VUNPCKHPS", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x15), 1, X86InstInfo{"VUNPCKHPD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x16), 1, X86InstInfo{"VMOVHPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x16), 1, X86InstInfo{"VMOVHPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x16), 1, X86InstInfo{"VMOVSHDUP", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x16), 1, X86InstInfo{"VMOVHPS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_XMM_FLAGS | FLAGS_VEX_1ST_SRC, 0, nullptr}},
+    {OPD(1, 0b01, 0x16), 1, X86InstInfo{"VMOVHPD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_XMM_FLAGS | FLAGS_VEX_1ST_SRC, 0, nullptr}},
+    {OPD(1, 0b10, 0x16), 1, X86InstInfo{"VMOVSHDUP", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x17), 1, X86InstInfo{"VMOVHPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x17), 1, X86InstInfo{"VMOVHPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x17), 1, X86InstInfo{"VMOVHPS",   TYPE_INST, GenFlagsSizes(SIZE_64BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x17), 1, X86InstInfo{"VMOVHPD",   TYPE_INST, GenFlagsSizes(SIZE_64BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x50), 1, X86InstInfo{"VMOVMSKPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x50), 1, X86InstInfo{"VMOVMSKPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x50), 1, X86InstInfo{"VMOVMSKPS", TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
+    {OPD(1, 0b01, 0x50), 1, X86InstInfo{"VMOVMSKPD", TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x51), 1, X86InstInfo{"VSQRTPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x51), 1, X86InstInfo{"VSQRTPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x51), 1, X86InstInfo{"VSQRTSS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x51), 1, X86InstInfo{"VSQRTSD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x51), 1, X86InstInfo{"VSQRTPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x51), 1, X86InstInfo{"VSQRTPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x51), 1, X86InstInfo{"VSQRTSS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0x51), 1, X86InstInfo{"VSQRTSD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x52), 1, X86InstInfo{"VRSQRTPS",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x52), 1, X86InstInfo{"VRSQRTSS",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x52), 1, X86InstInfo{"VRSQRTPS",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x52), 1, X86InstInfo{"VRSQRTSS",  TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x53), 1, X86InstInfo{"VRCPPS",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x53), 1, X86InstInfo{"VRCPSS",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x53), 1, X86InstInfo{"VRCPPS",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x53), 1, X86InstInfo{"VRCPSS",    TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x54), 1, X86InstInfo{"VANDPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x54), 1, X86InstInfo{"VANDPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x54), 1, X86InstInfo{"VANDPS",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x54), 1, X86InstInfo{"VANDPD",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x55), 1, X86InstInfo{"VANDNPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x55), 1, X86InstInfo{"VANDNPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x55), 1, X86InstInfo{"VANDNPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x55), 1, X86InstInfo{"VANDNPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x56), 1, X86InstInfo{"VORPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x56), 1, X86InstInfo{"VORPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x56), 1, X86InstInfo{"VORPS",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x56), 1, X86InstInfo{"VORPD",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x57), 1, X86InstInfo{"VXORPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x57), 1, X86InstInfo{"VDORPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x57), 1, X86InstInfo{"VXORPS",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x57), 1, X86InstInfo{"VXORPD",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0x60), 1, X86InstInfo{"VPUNPCKLBW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x61), 1, X86InstInfo{"VPUNPCKLWD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x62), 1, X86InstInfo{"VPUNPCKLDQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x63), 1, X86InstInfo{"VPACKSSWB",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x64), 1, X86InstInfo{"VPCMPGTB",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0x65), 1, X86InstInfo{"VPVMPGTW",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0x66), 1, X86InstInfo{"VPVMPGTD",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0x67), 1, X86InstInfo{"VPACKUSWB",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b01, 0x60), 1, X86InstInfo{"VPUNPCKLBW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x61), 1, X86InstInfo{"VPUNPCKLWD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x62), 1, X86InstInfo{"VPUNPCKLDQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x63), 1, X86InstInfo{"VPACKSSWB",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x64), 1, X86InstInfo{"VPCMPGTB",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x65), 1, X86InstInfo{"VPCMPGTW",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x66), 1, X86InstInfo{"VPCMPGTD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x67), 1, X86InstInfo{"VPACKUSWB",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
     {OPD(1, 0b01, 0x70), 1, X86InstInfo{"VPSHUFD",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(1, 0b10, 0x70), 1, X86InstInfo{"VPSHUFHW",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
@@ -91,19 +91,19 @@ void InitializeVEXTables() {
     {OPD(1, 0b01, 0x72), 1, X86InstInfo{"",           TYPE_VEX_GROUP_13, FLAGS_NONE, 0, nullptr}}, // VEX Group 13
     {OPD(1, 0b01, 0x73), 1, X86InstInfo{"",           TYPE_VEX_GROUP_14, FLAGS_NONE, 0, nullptr}}, // VEX Group 14
 
-    {OPD(1, 0b01, 0x74), 1, X86InstInfo{"VPCMPEQB",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0x75), 1, X86InstInfo{"VPCMPEQW",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0x76), 1, X86InstInfo{"VPCMPEQD",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x74), 1, X86InstInfo{"VPCMPEQB",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x75), 1, X86InstInfo{"VPCMPEQW",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x76), 1, X86InstInfo{"VPCMPEQD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x77), 1, X86InstInfo{"VZERO*",     TYPE_INST, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x77), 1, X86InstInfo{"VZERO*",     TYPE_INST, GenFlagsDstSize(SIZE_128BIT), 0, nullptr}},
 
-    {OPD(1, 0b00, 0xC2), 1, X86InstInfo{"VCMPccPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xC2), 1, X86InstInfo{"VCMPccPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0xC2), 1, X86InstInfo{"VCMPccSS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0xC2), 1, X86InstInfo{"VCMPccSD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0xC2), 1, X86InstInfo{"VCMPccPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(1, 0b01, 0xC2), 1, X86InstInfo{"VCMPccPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(1, 0b10, 0xC2), 1, X86InstInfo{"VCMPccSS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(1, 0b11, 0xC2), 1, X86InstInfo{"VCMPccSD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
 
     {OPD(1, 0b01, 0xC4), 1, X86InstInfo{"VPINSRW",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xC5), 1, X86InstInfo{"VPEXTRW",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b01, 0xC5), 1, X86InstInfo{"VPEXTRW",    TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
 
     {OPD(1, 0b00, 0xC6), 1, X86InstInfo{"VSHUFPS",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(1, 0b01, 0xC6), 1, X86InstInfo{"VSHUFPD",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
@@ -112,167 +112,166 @@ void InitializeVEXTables() {
     // This table doesn't state which VEX.pp is for which instruction
     // XXX: Confirm all the above encoding opcodes
 
-    {OPD(1, 0b00, 0x28), 1, X86InstInfo{"VMOVAPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x28), 1, X86InstInfo{"VMOVAPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x28), 1, X86InstInfo{"VMOVAPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x28), 1, X86InstInfo{"VMOVAPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x29), 1, X86InstInfo{"VMOVAPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x29), 1, X86InstInfo{"VMOVAPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x29), 1, X86InstInfo{"VMOVAPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x29), 1, X86InstInfo{"VMOVAPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
 
     {OPD(1, 0b10, 0x2A), 1, X86InstInfo{"VCVTSI2SS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(1, 0b11, 0x2A), 1, X86InstInfo{"VCVTSI2SD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x2B), 1, X86InstInfo{"VMOVNTPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x2B), 1, X86InstInfo{"VMOVNTPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x2B), 1, X86InstInfo{"VMOVNTPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x2B), 1, X86InstInfo{"VMOVNTPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b10, 0x2C), 1, X86InstInfo{"VCVTTSS2SI",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x2C), 1, X86InstInfo{"VCVTTSD2SI",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b10, 0x2C), 1, X86InstInfo{"VCVTTSS2SI",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
+    {OPD(1, 0b11, 0x2C), 1, X86InstInfo{"VCVTTSD2SI",   TYPE_INST, GenFlagsSrcSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
 
-    {OPD(1, 0b10, 0x2D), 1, X86InstInfo{"VCVTSS2SI",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x2D), 1, X86InstInfo{"VCVTSD2SI",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b10, 0x2D), 1, X86InstInfo{"VCVTSS2SI",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
+    {OPD(1, 0b11, 0x2D), 1, X86InstInfo{"VCVTSD2SI",   TYPE_INST, GenFlagsSameSize(SIZE_64BITDEF) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x2E), 1, X86InstInfo{"VUCOMISS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x2E), 1, X86InstInfo{"VUCOMISD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x2E), 1, X86InstInfo{"VUCOMISS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x2E), 1, X86InstInfo{"VUCOMISD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x2F), 1, X86InstInfo{"VUCOMISS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x2F), 1, X86InstInfo{"VUCOMISD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x2F), 1, X86InstInfo{"VCOMISS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x2F), 1, X86InstInfo{"VCOMISD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x58), 1, X86InstInfo{"VADDPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x58), 1, X86InstInfo{"VADDPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x58), 1, X86InstInfo{"VADDSS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x58), 1, X86InstInfo{"VADDSD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x58), 1, X86InstInfo{"VADDPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x58), 1, X86InstInfo{"VADDPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x58), 1, X86InstInfo{"VADDSS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0x58), 1, X86InstInfo{"VADDSD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x59), 1, X86InstInfo{"VMULPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x59), 1, X86InstInfo{"VMULPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x59), 1, X86InstInfo{"VMULSS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x59), 1, X86InstInfo{"VMULSD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x59), 1, X86InstInfo{"VMULPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x59), 1, X86InstInfo{"VMULPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x59), 1, X86InstInfo{"VMULSS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0x59), 1, X86InstInfo{"VMULSD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x5B), 1, X86InstInfo{"VCVTDQ2PS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x5B), 1, X86InstInfo{"VCVTPS2DQ",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x5B), 1, X86InstInfo{"VCVTPS2DQ",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x5B), 1, X86InstInfo{"VCVTDQ2PS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x5B), 1, X86InstInfo{"VCVTPS2DQ",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x5B), 1, X86InstInfo{"VCVTTPS2DQ",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x5C), 1, X86InstInfo{"VSUBPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x5C), 1, X86InstInfo{"VSUBPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x5C), 1, X86InstInfo{"VSUBSS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x5C), 1, X86InstInfo{"VSUBSD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x5C), 1, X86InstInfo{"VSUBPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x5C), 1, X86InstInfo{"VSUBPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x5C), 1, X86InstInfo{"VSUBSS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0x5C), 1, X86InstInfo{"VSUBSD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x5D), 1, X86InstInfo{"VMINPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x5D), 1, X86InstInfo{"VMINPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x5D), 1, X86InstInfo{"VMINSS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x5D), 1, X86InstInfo{"VMINSD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x5D), 1, X86InstInfo{"VMINPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x5D), 1, X86InstInfo{"VMINPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x5D), 1, X86InstInfo{"VMINSS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0x5D), 1, X86InstInfo{"VMINSD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x5E), 1, X86InstInfo{"VDIVPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x5E), 1, X86InstInfo{"VDIVPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x5E), 1, X86InstInfo{"VDIVSS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x5E), 1, X86InstInfo{"VDIVSD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x5E), 1, X86InstInfo{"VDIVPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x5E), 1, X86InstInfo{"VDIVPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x5E), 1, X86InstInfo{"VDIVSS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0x5E), 1, X86InstInfo{"VDIVSD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b00, 0x5F), 1, X86InstInfo{"VMAXPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x5F), 1, X86InstInfo{"VMAXPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0x5F), 1, X86InstInfo{"VMAXSS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x5F), 1, X86InstInfo{"VMAXSD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b00, 0x5F), 1, X86InstInfo{"VMAXPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x5F), 1, X86InstInfo{"VMAXPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x5F), 1, X86InstInfo{"VMAXSS",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0x5F), 1, X86InstInfo{"VMAXSD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-
-    {OPD(1, 0b01, 0x68), 1, X86InstInfo{"VPUNPCKHBW",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x69), 1, X86InstInfo{"VPUNPCKHWD",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x6A), 1, X86InstInfo{"VPUNPCKHDQ",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x6B), 1, X86InstInfo{"VPACKSSDW",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x6C), 1, X86InstInfo{"VPUNPCKLQDQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0x6D), 1, X86InstInfo{"VPUNPCKHQDQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b01, 0x68), 1, X86InstInfo{"VPUNPCKHBW",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x69), 1, X86InstInfo{"VPUNPCKHWD",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x6A), 1, X86InstInfo{"VPUNPCKHDQ",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x6B), 1, X86InstInfo{"VPACKSSDW",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x6C), 1, X86InstInfo{"VPUNPCKLQDQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x6D), 1, X86InstInfo{"VPUNPCKHQDQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(1, 0b01, 0x6E), 1, X86InstInfo{"VMOV*",       TYPE_INST, GenFlagsDstSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_SRC_GPR, 0, nullptr}},
 
-    {OPD(1, 0b01, 0x6F), 1, X86InstInfo{"VMOVDQA",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b10, 0x6F), 1, X86InstInfo{"VMOVDQU",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x6F), 1, X86InstInfo{"VMOVDQA",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x6F), 1, X86InstInfo{"VMOVDQU",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0x7C), 1, X86InstInfo{"VHADDPD",     TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0x7C), 1, X86InstInfo{"VHADDPS",     TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b01, 0x7C), 1, X86InstInfo{"VHADDPD",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0x7C), 1, X86InstInfo{"VHADDPS",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
     {OPD(1, 0b01, 0x7D), 1, X86InstInfo{"VHSUBPD",     TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(1, 0b11, 0x7D), 1, X86InstInfo{"VHSUBPS",     TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(1, 0b01, 0x7E), 1, X86InstInfo{"VMOV*",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b10, 0x7E), 1, X86InstInfo{"VMOVQ",     TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x7E), 1, X86InstInfo{"VMOV*",     TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x7E), 1, X86InstInfo{"VMOVQ",     TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0x7F), 1, X86InstInfo{"VMOVDQA",     TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b10, 0x7F), 1, X86InstInfo{"VMOVDQU",     TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0x7F), 1, X86InstInfo{"VMOVDQA",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0x7F), 1, X86InstInfo{"VMOVDQU",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
 
     {OPD(1, 0b00, 0xAE), 1, X86InstInfo{"",     TYPE_VEX_GROUP_15, FLAGS_NONE, 0, nullptr}}, // VEX Group 15
     {OPD(1, 0b01, 0xAE), 1, X86InstInfo{"",     TYPE_VEX_GROUP_15, FLAGS_NONE, 0, nullptr}}, // VEX Group 15
     {OPD(1, 0b10, 0xAE), 1, X86InstInfo{"",     TYPE_VEX_GROUP_15, FLAGS_NONE, 0, nullptr}}, // VEX Group 15
     {OPD(1, 0b11, 0xAE), 1, X86InstInfo{"",     TYPE_VEX_GROUP_15, FLAGS_NONE, 0, nullptr}}, // VEX Group 15
 
-    {OPD(1, 0b01, 0xD0), 1, X86InstInfo{"VADDSUBPD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0xD0), 1, X86InstInfo{"VADDSUBPS",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b01, 0xD0), 1, X86InstInfo{"VADDSUBPD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0xD0), 1, X86InstInfo{"VADDSUBPS",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0xD1), 1, X86InstInfo{"VPSRLW",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xD2), 1, X86InstInfo{"VPSRLD",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xD3), 1, X86InstInfo{"VPSRLQ",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xD4), 1, X86InstInfo{"VPADDQ",      TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xD5), 1, X86InstInfo{"VPMULLW",     TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xD6), 1, X86InstInfo{"VMOVQ",       TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xD7), 1, X86InstInfo{"VPMOVMSKB",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR | FLAGS_SF_MOD_REG_ONLY, 0, nullptr}},
+    {OPD(1, 0b01, 0xD1), 1, X86InstInfo{"VPSRLW",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xD2), 1, X86InstInfo{"VPSRLD",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xD3), 1, X86InstInfo{"VPSRLQ",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xD4), 1, X86InstInfo{"VPADDQ",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xD5), 1, X86InstInfo{"VPMULLW",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xD6), 1, X86InstInfo{"VMOVQ",       TYPE_INST, GenFlagsSameSize(SIZE_64BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xD7), 1, X86InstInfo{"VPMOVMSKB",   TYPE_UNDEC, FLAGS_MODRM | FLAGS_XMM_FLAGS | FLAGS_SF_DST_GPR | FLAGS_SF_MOD_REG_ONLY, 0, nullptr}},
 
-    {OPD(1, 0b01, 0xD8), 1, X86InstInfo{"VPSUBUSB", TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xD9), 1, X86InstInfo{"VPSUBUSW", TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xDA), 1, X86InstInfo{"VPMINUB",  TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xDB), 1, X86InstInfo{"VPAND",    TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xDC), 1, X86InstInfo{"VPADDUSB", TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xDD), 1, X86InstInfo{"VPADDUSW", TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xDE), 1, X86InstInfo{"VPMAXUB",  TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xDF), 1, X86InstInfo{"VPANDN",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xD8), 1, X86InstInfo{"VPSUBUSB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xD9), 1, X86InstInfo{"VPSUBUSW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xDA), 1, X86InstInfo{"VPMINUB",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xDB), 1, X86InstInfo{"VPAND",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xDC), 1, X86InstInfo{"VPADDUSB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xDD), 1, X86InstInfo{"VPADDUSW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xDE), 1, X86InstInfo{"VPMAXUB",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xDF), 1, X86InstInfo{"VPANDN",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0xE0), 1, X86InstInfo{"VPAVGB",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xE1), 1, X86InstInfo{"VPSRAW",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xE2), 1, X86InstInfo{"VPSRAD",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xE3), 1, X86InstInfo{"VPAVGW",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xE4), 1, X86InstInfo{"VPMULHUW",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xE5), 1, X86InstInfo{"VPMULHW",     TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b01, 0xE0), 1, X86InstInfo{"VPAVGB",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xE1), 1, X86InstInfo{"VPSRAW",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xE2), 1, X86InstInfo{"VPSRAD",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xE3), 1, X86InstInfo{"VPAVGW",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xE4), 1, X86InstInfo{"VPMULHUW",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xE5), 1, X86InstInfo{"VPMULHW",     TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0xE6), 1, X86InstInfo{"VCVTTPD2DQ",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b10, 0xE6), 1, X86InstInfo{"VCVTDQ2PD",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b11, 0xE6), 1, X86InstInfo{"VCVTPD2DQ",   TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b01, 0xE6), 1, X86InstInfo{"VCVTTPD2DQ",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b10, 0xE6), 1, X86InstInfo{"VCVTDQ2PD",   TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b11, 0xE6), 1, X86InstInfo{"VCVTPD2DQ",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0xE7), 1, X86InstInfo{"VMOVNTDQ",    TYPE_INST, FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xE7), 1, X86InstInfo{"VMOVNTDQ",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0xE8), 1, X86InstInfo{"VPSUBSB", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xE9), 1, X86InstInfo{"VPSUBSW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xEA), 1, X86InstInfo{"VPMINSW",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xEB), 1, X86InstInfo{"VPOR",    TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xEC), 1, X86InstInfo{"VPADDSB", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xED), 1, X86InstInfo{"VPADDSW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xEE), 1, X86InstInfo{"VPMAXSW",  TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xEF), 1, X86InstInfo{"VPXOR",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xE8), 1, X86InstInfo{"VPSUBSB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xE9), 1, X86InstInfo{"VPSUBSW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xEA), 1, X86InstInfo{"VPMINSW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xEB), 1, X86InstInfo{"VPOR",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xEC), 1, X86InstInfo{"VPADDSB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xED), 1, X86InstInfo{"VPADDSW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xEE), 1, X86InstInfo{"VPMAXSW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xEF), 1, X86InstInfo{"VPXOR",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b11, 0xF0), 1, X86InstInfo{"VLDDQU",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b11, 0xF0), 1, X86InstInfo{"VLDDQU",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0xF1), 1, X86InstInfo{"VPSLLW",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xF2), 1, X86InstInfo{"VPSLLD",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xF3), 1, X86InstInfo{"VPSLLQ",      TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xF4), 1, X86InstInfo{"VPMULUDQ",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b01, 0xF1), 1, X86InstInfo{"VPSLLW",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xF2), 1, X86InstInfo{"VPSLLD",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xF3), 1, X86InstInfo{"VPSLLQ",      TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xF4), 1, X86InstInfo{"VPMULUDQ",    TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(1, 0b01, 0xF5), 1, X86InstInfo{"VPMADDWD",    TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(1, 0b01, 0xF6), 1, X86InstInfo{"VPSADBW",     TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(1, 0b01, 0xF7), 1, X86InstInfo{"VMASKMOVDQU", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(1, 0b01, 0xF7), 1, X86InstInfo{"VMASKMOVDQU", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_REG_ONLY | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(1, 0b01, 0xF8), 1, X86InstInfo{"VPSUBB", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xF9), 1, X86InstInfo{"VPSUBW", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xFA), 1, X86InstInfo{"VPSUBD", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xFB), 1, X86InstInfo{"VPSUBQ", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xFC), 1, X86InstInfo{"VPADDB", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xFD), 1, X86InstInfo{"VPADDW", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(1, 0b01, 0xFE), 1, X86InstInfo{"VPADDD", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xF8), 1, X86InstInfo{"VPSUBB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xF9), 1, X86InstInfo{"VPSUBW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xFA), 1, X86InstInfo{"VPSUBD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xFB), 1, X86InstInfo{"VPSUBQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xFC), 1, X86InstInfo{"VPADDB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xFD), 1, X86InstInfo{"VPADDW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(1, 0b01, 0xFE), 1, X86InstInfo{"VPADDD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
     // VEX Map 2
-    {OPD(2, 0b01, 0x00), 1, X86InstInfo{"VPSHUFB", TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(2, 0b01, 0x01), 1, X86InstInfo{"VPADDW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x02), 1, X86InstInfo{"VPHADDD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x00), 1, X86InstInfo{"VPSHUFB", TYPE_UNDEC, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x01), 1, X86InstInfo{"VPHADDW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x02), 1, X86InstInfo{"VPHADDD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(2, 0b01, 0x03), 1, X86InstInfo{"VPHADDSW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x04), 1, X86InstInfo{"VPMADDUBSW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x05), 1, X86InstInfo{"VPHSUBW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x06), 1, X86InstInfo{"VPHSUBD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x05), 1, X86InstInfo{"VPHSUBW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x06), 1, X86InstInfo{"VPHSUBD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(2, 0b01, 0x07), 1, X86InstInfo{"VPHSUBSW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(2, 0b01, 0x08), 1, X86InstInfo{"VPSIGNB", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x09), 1, X86InstInfo{"VPSIGNW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x0A), 1, X86InstInfo{"VPSIGND", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x0B), 1, X86InstInfo{"VPMULHRSW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x08), 1, X86InstInfo{"VPSIGNB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x09), 1, X86InstInfo{"VPSIGNW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x0A), 1, X86InstInfo{"VPSIGND", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x0B), 1, X86InstInfo{"VPMULHRSW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(2, 0b01, 0x0C), 1, X86InstInfo{"VPERMILPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x0D), 1, X86InstInfo{"VPERMILPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x0E), 1, X86InstInfo{"VTESTPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
@@ -282,59 +281,59 @@ void InitializeVEXTables() {
     {OPD(2, 0b01, 0x16), 1, X86InstInfo{"VPERMPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x17), 1, X86InstInfo{"VPTEST", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(2, 0b01, 0x18), 1, X86InstInfo{"VBROADCASTSS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x19), 1, X86InstInfo{"VBROADCASTSD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x1A), 1, X86InstInfo{"VBROADCASTF128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x1C), 1, X86InstInfo{"VPABSB", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x1D), 1, X86InstInfo{"VPABSW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x1E), 1, X86InstInfo{"VPABSD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x18), 1, X86InstInfo{"VBROADCASTSS", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x19), 1, X86InstInfo{"VBROADCASTSD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x1A), 1, X86InstInfo{"VBROADCASTF128", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x1C), 1, X86InstInfo{"VPABSB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x1D), 1, X86InstInfo{"VPABSW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x1E), 1, X86InstInfo{"VPABSD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(2, 0b01, 0x20), 1, X86InstInfo{"VPMOVSXBW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x21), 1, X86InstInfo{"VPMOVSXBD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x22), 1, X86InstInfo{"VPMOVSXBQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x23), 1, X86InstInfo{"VPMOVSXWD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x24), 1, X86InstInfo{"VPMOVSXWQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x25), 1, X86InstInfo{"VPMOVSXDQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x20), 1, X86InstInfo{"VPMOVSXBW", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x21), 1, X86InstInfo{"VPMOVSXBD", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x22), 1, X86InstInfo{"VPMOVSXBQ", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_16BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x23), 1, X86InstInfo{"VPMOVSXWD", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x24), 1, X86InstInfo{"VPMOVSXWQ", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x25), 1, X86InstInfo{"VPMOVSXDQ", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(2, 0b01, 0x28), 1, X86InstInfo{"VPMULDQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x29), 1, X86InstInfo{"VPCMPEQQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x2A), 1, X86InstInfo{"VMOVNTDQA", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x2B), 1, X86InstInfo{"VPACKUSDW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x28), 1, X86InstInfo{"VPMULDQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x29), 1, X86InstInfo{"VPCMPEQQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x2A), 1, X86InstInfo{"VMOVNTDQA", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_MEM_ONLY | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x2B), 1, X86InstInfo{"VPACKUSDW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(2, 0b01, 0x2C), 1, X86InstInfo{"VMASKMOVPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x2D), 1, X86InstInfo{"VMASKMOVPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x2E), 1, X86InstInfo{"VMASKMOVPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x2F), 1, X86InstInfo{"VMASKMOVPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(2, 0b01, 0x30), 1, X86InstInfo{"VPMOVZXBW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x31), 1, X86InstInfo{"VPMOVZXBD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x32), 1, X86InstInfo{"VPMOVZXBQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x33), 1, X86InstInfo{"VPMOVZXWD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x34), 1, X86InstInfo{"VPMOVZXWQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x35), 1, X86InstInfo{"VPMOVZXDQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x30), 1, X86InstInfo{"VPMOVZXBW", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x31), 1, X86InstInfo{"VPMOVZXBD", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x32), 1, X86InstInfo{"VPMOVZXBQ", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_16BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x33), 1, X86InstInfo{"VPMOVZXWD", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x34), 1, X86InstInfo{"VPMOVZXWQ", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x35), 1, X86InstInfo{"VPMOVZXDQ", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(2, 0b01, 0x36), 1, X86InstInfo{"VPERMD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x37), 1, X86InstInfo{"VPVMPGTQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x37), 1, X86InstInfo{"VPCMPGTQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(2, 0b01, 0x38), 1, X86InstInfo{"VPMINSB", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x39), 1, X86InstInfo{"VPMINSD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x3A), 1, X86InstInfo{"VPMINUW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x3B), 1, X86InstInfo{"VPMINUD", TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
-    {OPD(2, 0b01, 0x3C), 1, X86InstInfo{"VPMAXSB", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x3D), 1, X86InstInfo{"VPMAXSD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x3E), 1, X86InstInfo{"VPMAXUW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x3F), 1, X86InstInfo{"VPMAXUD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x38), 1, X86InstInfo{"VPMINSB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x39), 1, X86InstInfo{"VPMINSD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x3A), 1, X86InstInfo{"VPMINUW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x3B), 1, X86InstInfo{"VPMINUD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x3C), 1, X86InstInfo{"VPMAXSB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x3D), 1, X86InstInfo{"VPMAXSD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x3E), 1, X86InstInfo{"VPMAXUW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x3F), 1, X86InstInfo{"VPMAXUD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(2, 0b01, 0x40), 1, X86InstInfo{"VPMULLD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0x41), 1, X86InstInfo{"VPHMINPOSUW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0x40), 1, X86InstInfo{"VPMULLD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x41), 1, X86InstInfo{"VPHMINPOSUW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
     {OPD(2, 0b01, 0x45), 1, X86InstInfo{"VPSRLV", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x46), 1, X86InstInfo{"VPSRAVD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x47), 1, X86InstInfo{"VPSLLV", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(2, 0b01, 0x58), 1, X86InstInfo{"VPBROADCASTD", TYPE_INST, FLAGS_MODRM, 0, nullptr}},
-    {OPD(2, 0b01, 0x59), 1, X86InstInfo{"VPBROADCASTQ", TYPE_INST, FLAGS_MODRM, 0, nullptr}},
-    {OPD(2, 0b01, 0x5A), 1, X86InstInfo{"VBBROADCASTI128", TYPE_INST, FLAGS_MODRM, 0, nullptr}},
+    {OPD(2, 0b01, 0x58), 1, X86InstInfo{"VPBROADCASTD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x59), 1, X86InstInfo{"VPBROADCASTQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x5A), 1, X86InstInfo{"VBROADCASTI128", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
-    {OPD(2, 0b01, 0x78), 1, X86InstInfo{"VPBROADCASTB", TYPE_INST, FLAGS_MODRM, 0, nullptr}},
-    {OPD(2, 0b01, 0x79), 1, X86InstInfo{"VPBROADCASTW", TYPE_INST, FLAGS_MODRM, 0, nullptr}},
+    {OPD(2, 0b01, 0x78), 1, X86InstInfo{"VPBROADCASTB", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0x79), 1, X86InstInfo{"VPBROADCASTW", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
 
     {OPD(2, 0b01, 0x8C), 1, X86InstInfo{"VPMASKMOV", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0x8E), 1, X86InstInfo{"VPMASKMOV", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
@@ -380,11 +379,11 @@ void InitializeVEXTables() {
     {OPD(2, 0b01, 0xB6), 1, X86InstInfo{"VFMADDSUB231", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(2, 0b01, 0xB7), 1, X86InstInfo{"VFMSUBADD231", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(2, 0b01, 0xDB), 1, X86InstInfo{"VAESIMC", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0xDC), 1, X86InstInfo{"VAESENC", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0xDD), 1, X86InstInfo{"VAESENCLAST", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0xDE), 1, X86InstInfo{"VAESDEC", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(2, 0b01, 0xDF), 1, X86InstInfo{"VAESDECLAST", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(2, 0b01, 0xDB), 1, X86InstInfo{"VAESIMC", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0xDC), 1, X86InstInfo{"VAESENC", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0xDD), 1, X86InstInfo{"VAESENCLAST", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0xDE), 1, X86InstInfo{"VAESDEC", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
+    {OPD(2, 0b01, 0xDF), 1, X86InstInfo{"VAESDECLAST", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 0, nullptr}},
 
     {OPD(2, 0b00, 0xF2), 1, X86InstInfo{"ANDN", TYPE_INST, FLAGS_MODRM | FLAGS_VEX_1ST_SRC, 0, nullptr}},
 
@@ -406,43 +405,43 @@ void InitializeVEXTables() {
     {OPD(2, 0b11, 0xF7), 1, X86InstInfo{"SHRX", TYPE_INST, FLAGS_MODRM | FLAGS_VEX_2ND_SRC, 0, nullptr}},
 
     // VEX Map 3
-    {OPD(3, 0b01, 0x00), 1, X86InstInfo{"VPERMQ", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x01), 1, X86InstInfo{"VPERMPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x00), 1, X86InstInfo{"VPERMQ", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x01), 1, X86InstInfo{"VPERMPD", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(3, 0b01, 0x02), 1, X86InstInfo{"VPBLENDD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x04), 1, X86InstInfo{"VPERMILPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x05), 1, X86InstInfo{"VPERMILPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x06), 1, X86InstInfo{"VPERM2F128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x04), 1, X86InstInfo{"VPERMILPS", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x05), 1, X86InstInfo{"VPERMILPD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x06), 1, X86InstInfo{"VPERM2F128", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
 
-    {OPD(3, 0b01, 0x08), 1, X86InstInfo{"VROUNDPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x09), 1, X86InstInfo{"VROUNDPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x0A), 1, X86InstInfo{"VROUNDSS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x0B), 1, X86InstInfo{"VROUNDSD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x08), 1, X86InstInfo{"VROUNDPS", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x09), 1, X86InstInfo{"VROUNDPD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x0A), 1, X86InstInfo{"VROUNDSS", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_32BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x0B), 1, X86InstInfo{"VROUNDSD", TYPE_INST, GenFlagsSizes(SIZE_128BIT, SIZE_64BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(3, 0b01, 0x0C), 1, X86InstInfo{"VBLENDPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(3, 0b01, 0x0D), 1, X86InstInfo{"VBLENDPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(3, 0b01, 0x0E), 1, X86InstInfo{"VBLENDW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(3, 0b01, 0x0F), 1, X86InstInfo{"VPALIGNR", TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
 
-    {OPD(3, 0b01, 0x14), 1, X86InstInfo{"VPEXTRB", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x15), 1, X86InstInfo{"VPEXTRW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x16), 1, X86InstInfo{"VPEXTRD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x17), 1, X86InstInfo{"VEXTRACTPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x14), 1, X86InstInfo{"VPEXTRB", TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x15), 1, X86InstInfo{"VPEXTRW", TYPE_INST, GenFlagsSizes(SIZE_16BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x16), 1, X86InstInfo{"VPEXTRD", TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x17), 1, X86InstInfo{"VEXTRACTPS", TYPE_INST, GenFlagsSizes(SIZE_32BIT, SIZE_128BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_DST_GPR | FLAGS_XMM_FLAGS, 1, nullptr}},
 
-    {OPD(3, 0b01, 0x18), 1, X86InstInfo{"VINSERTF128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x18), 1, X86InstInfo{"VINSERTF128", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(3, 0b01, 0x19), 1, X86InstInfo{"VEXTRACTF128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(3, 0b01, 0x1D), 1, X86InstInfo{"VCVTPS2PH", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
     {OPD(3, 0b01, 0x20), 1, X86InstInfo{"VPINSRB", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x21), 1, X86InstInfo{"VINSERTPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x21), 1, X86InstInfo{"VINSERTPS", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(3, 0b01, 0x22), 1, X86InstInfo{"VPINSRD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(3, 0b01, 0x38), 1, X86InstInfo{"VINSERTI128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x38), 1, X86InstInfo{"VINSERTI128", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(3, 0b01, 0x39), 1, X86InstInfo{"VEXTRACTI128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(3, 0b01, 0x40), 1, X86InstInfo{"VDPPS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
-    {OPD(3, 0b01, 0x41), 1, X86InstInfo{"VDPPD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x40), 1, X86InstInfo{"VDPPS", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(3, 0b01, 0x41), 1, X86InstInfo{"VDPPD", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
     {OPD(3, 0b01, 0x42), 1, X86InstInfo{"VMPSADBW", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(3, 0b01, 0x44), 1, X86InstInfo{"VPCLMULQDQ", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
-    {OPD(3, 0b01, 0x46), 1, X86InstInfo{"VPERM2I128", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0x46), 1, X86InstInfo{"VPERM2I128", TYPE_INST, GenFlagsSameSize(SIZE_256BIT) | FLAGS_MODRM | FLAGS_VEX_1ST_SRC | FLAGS_XMM_FLAGS, 1, nullptr}},
 
     {OPD(3, 0b01, 0x48), 1, X86InstInfo{"VPERMILzz2PS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(3, 0b01, 0x49), 1, X86InstInfo{"VPERMILzz2PD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
@@ -478,7 +477,7 @@ void InitializeVEXTables() {
     {OPD(3, 0b01, 0x7E), 1, X86InstInfo{"VFNMSUBSS", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
     {OPD(3, 0b01, 0x7F), 1, X86InstInfo{"VFNMSUBSD", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
 
-    {OPD(3, 0b01, 0xDF), 1, X86InstInfo{"VAESKEYGENASSIST", TYPE_UNDEC, FLAGS_NONE, 0, nullptr}},
+    {OPD(3, 0b01, 0xDF), 1, X86InstInfo{"VAESKEYGENASSIST", TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
 
     {OPD(3, 0b11, 0xF0), 1, X86InstInfo{"RORX", TYPE_INST, FLAGS_MODRM, 1, nullptr}},
 
@@ -488,21 +487,21 @@ void InitializeVEXTables() {
 
 #define OPD(group, pp, opcode) (((group - TYPE_VEX_GROUP_12) << 4) | (pp << 3) | (opcode))
   static constexpr U8U8InfoStruct VEXGroupTable[] = {
-    {OPD(TYPE_VEX_GROUP_12, 1, 0b010), 1, X86InstInfo{"VPSRLW",   TYPE_UNDEC, FLAGS_MODRM, 0, nullptr}},
-    {OPD(TYPE_VEX_GROUP_12, 1, 0b100), 1, X86InstInfo{"VPSRAW",   TYPE_UNDEC, FLAGS_MODRM, 0, nullptr}},
-    {OPD(TYPE_VEX_GROUP_12, 1, 0b110), 1, X86InstInfo{"VPSLLW",   TYPE_UNDEC, FLAGS_MODRM, 0, nullptr}},
+    {OPD(TYPE_VEX_GROUP_12, 1, 0b010), 1, X86InstInfo{"VPSRLW",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(TYPE_VEX_GROUP_12, 1, 0b100), 1, X86InstInfo{"VPSRAW",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(TYPE_VEX_GROUP_12, 1, 0b110), 1, X86InstInfo{"VPSLLW",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
 
-    {OPD(TYPE_VEX_GROUP_13, 1, 0b010), 1, X86InstInfo{"VPSRLD",   TYPE_UNDEC, FLAGS_MODRM, 0, nullptr}},
-    {OPD(TYPE_VEX_GROUP_13, 1, 0b100), 1, X86InstInfo{"VPSRAD",   TYPE_UNDEC, FLAGS_MODRM, 0, nullptr}},
-    {OPD(TYPE_VEX_GROUP_13, 1, 0b110), 1, X86InstInfo{"VPSLLD",   TYPE_UNDEC, FLAGS_MODRM, 0, nullptr}},
+    {OPD(TYPE_VEX_GROUP_13, 1, 0b010), 1, X86InstInfo{"VPSRLD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(TYPE_VEX_GROUP_13, 1, 0b100), 1, X86InstInfo{"VPSRAD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(TYPE_VEX_GROUP_13, 1, 0b110), 1, X86InstInfo{"VPSLLD",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
 
-    {OPD(TYPE_VEX_GROUP_14, 1, 0b010), 1, X86InstInfo{"VPSRLQ",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
-    {OPD(TYPE_VEX_GROUP_14, 1, 0b011), 1, X86InstInfo{"VPSRLDQ",  TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
-    {OPD(TYPE_VEX_GROUP_14, 1, 0b110), 1, X86InstInfo{"VPSLLQ",   TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
-    {OPD(TYPE_VEX_GROUP_14, 1, 0b111), 1, X86InstInfo{"VPSLLDQ",  TYPE_INST, FLAGS_MODRM | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(TYPE_VEX_GROUP_14, 1, 0b010), 1, X86InstInfo{"VPSRLQ",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(TYPE_VEX_GROUP_14, 1, 0b011), 1, X86InstInfo{"VPSRLDQ",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(TYPE_VEX_GROUP_14, 1, 0b110), 1, X86InstInfo{"VPSLLQ",   TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
+    {OPD(TYPE_VEX_GROUP_14, 1, 0b111), 1, X86InstInfo{"VPSLLDQ",  TYPE_INST, GenFlagsSameSize(SIZE_128BIT) | FLAGS_MODRM | FLAGS_VEX_DST | FLAGS_XMM_FLAGS, 1, nullptr}},
 
-    {OPD(TYPE_VEX_GROUP_15, 1, 0b010), 1, X86InstInfo{"VLDMXCSR", TYPE_UNDEC, FLAGS_MODRM, 0, nullptr}},
-    {OPD(TYPE_VEX_GROUP_15, 1, 0b011), 1, X86InstInfo{"VSTMXCSR", TYPE_UNDEC, FLAGS_MODRM, 0, nullptr}},
+    {OPD(TYPE_VEX_GROUP_15, 0, 0b010), 1, X86InstInfo{"VLDMXCSR", TYPE_INST, GenFlagsSameSize(SIZE_32BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY, 0, nullptr}},
+    {OPD(TYPE_VEX_GROUP_15, 0, 0b011), 1, X86InstInfo{"VSTMXCSR", TYPE_INST, GenFlagsSameSize(SIZE_32BIT) | FLAGS_MODRM | FLAGS_SF_MOD_DST | FLAGS_SF_MOD_MEM_ONLY, 0, nullptr}},
 
     {OPD(TYPE_VEX_GROUP_17, 0, 0b001), 1, X86InstInfo{"BLSR",     TYPE_INST, FLAGS_MODRM | FLAGS_VEX_DST, 0, nullptr}},
     {OPD(TYPE_VEX_GROUP_17, 0, 0b010), 1, X86InstInfo{"BLSMSK",   TYPE_INST, FLAGS_MODRM | FLAGS_VEX_DST, 0, nullptr}},

External/FEXCore/Source/Interface/HLE/Thunks/Thunks.cpp

@@ -167,6 +167,8 @@ namespace FEXCore {
          * address to another function. The original callee address is passed
          * to the target function through an implicit argument stored in r11.
          *
+         * For 32-bit the implicit argument is stored in the lower 32-bits of mm0.
+         *
          * The primary use case of this is ensuring that host function pointers
          * returned from thunked APIs can safely be called by the guest.
          */
@@ -199,7 +201,12 @@ namespace FEXCore {
 
                         const uint8_t GPRSize = CTX->GetGPRSize();
 
-                        emit->_StoreContext(GPRSize, IR::GPRClass, emit->_Constant(Entrypoint), offsetof(Core::CPUState, gregs[X86State::REG_R11]));
+                        if (GPRSize == 8) {
+                          emit->_StoreRegister(emit->_Constant(Entrypoint), false, offsetof(Core::CPUState, gregs[X86State::REG_R11]), IR::GPRClass, IR::GPRFixedClass, GPRSize);
+                        }
+                        else {
+                          emit->_StoreRegister(emit->_Constant(Entrypoint), false, offsetof(Core::CPUState, mm[0][0]), IR::GPRClass, IR::GPRFixedClass, GPRSize);
+                        }
                         emit->_ExitFunction(emit->_Constant(GuestThunkEntrypoint));
                     }, CTX->ThunkHandler.get(), (void*)args->target_addr);
 
@@ -263,7 +270,10 @@ namespace FEXCore {
 
             auto Name = Args->Name;
 
-            auto SOName = CTX->Config.ThunkHostLibsPath() + "/" + (const char*)Name + "-host.so";
+            auto SOName = (CTX->Config.Is64BitMode() ?
+                           CTX->Config.ThunkHostLibsPath() :
+                           CTX->Config.ThunkHostLibsPath32())
+                          + "/" + (const char*)Name + "-host.so";
 
             LogMan::Msg::DFmt("LoadLib: {} -> {}", Name, SOName);
 
@@ -431,6 +441,9 @@ namespace FEXCore {
 
     FEX_DEFAULT_VISIBILITY
     void FinalizeHostTrampolineForGuestFunction(HostToGuestTrampolinePtr* TrampolineAddress, void* HostPacker) {
+
+      if (TrampolineAddress == nullptr) return;
+
       auto& Trampoline = GetInstanceInfo(TrampolineAddress);
 
       LOGMAN_THROW_A_FMT(Trampoline.CallCallback == (uintptr_t)&ThunkHandler_impl::CallCallback,

External/FEXCore/Source/Interface/IR/IR.json

@@ -355,7 +355,9 @@
         "DestSize": "ByteSize",
         "EmitValidation": [
           "($Class == GPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8 || #ByteSize == 16)) || $Class == GPRClass"
+          "($Class == FPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8 || #ByteSize == 16 || #ByteSize == 32)) || $Class == GPRClass",
+          "!($Offset >= offsetof(Core::CPUState, gregs[0]) && $Offset < offsetof(Core::CPUState, gregs[16])) && \"Can't LoadContext to GPR\"",
+          "!($Offset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't LoadContext to XMM\""
         ]
       },
 
@@ -370,7 +372,9 @@
         "EmitValidation": [
           "WalkFindRegClass($Value) == $Class",
           "($Class == GPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8 || #ByteSize == 16)) || $Class == GPRClass"
+          "($Class == FPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8 || #ByteSize == 16 || #ByteSize == 32)) || $Class == GPRClass",
+          "!($Offset >= offsetof(Core::CPUState, gregs[0]) && $Offset < offsetof(Core::CPUState, gregs[16])) && \"Can't StoreContext to GPR\"",
+          "!($Offset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $Offset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't StoreContext to XMM\""
         ]
       },
 
@@ -381,7 +385,9 @@
         "DestSize": "ByteSize",
         "EmitValidation": [
           "($Class == GPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8 || #ByteSize == 16)) || $Class == GPRClass"
+          "($Class == FPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8 || #ByteSize == 16 || #ByteSize == 32)) || $Class == GPRClass",
+          "!($BaseOffset >= offsetof(Core::CPUState, gregs[0]) && $BaseOffset < offsetof(Core::CPUState, gregs[16])) && \"Can't LoadContextIndexed to GPR\"",
+          "!($BaseOffset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $BaseOffset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't LoadContextIndexed to XMM\""
         ]
       },
       "StoreContextIndexed SSA:$Value, GPR:$Index, u8:#ByteSize, u32:$BaseOffset, u32:$Stride, RegisterClass:$Class": {
@@ -393,7 +399,9 @@
         "EmitValidation": [
           "WalkFindRegClass($Value) == $Class",
           "($Class == GPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8)) || $Class == FPRClass",
-          "($Class == FPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8 || #ByteSize == 16)) || $Class == GPRClass"
+          "($Class == FPRClass && (#ByteSize == 1 || #ByteSize == 2 || #ByteSize == 4 || #ByteSize == 8 || #ByteSize == 16 || #ByteSize == 32)) || $Class == GPRClass",
+          "!($BaseOffset >= offsetof(Core::CPUState, gregs[0]) && $BaseOffset < offsetof(Core::CPUState, gregs[16])) && \"Can't StoreContextIndexed to GPR\"",
+          "!($BaseOffset >= offsetof(Core::CPUState, xmm.avx.data[0]) && $BaseOffset < offsetof(Core::CPUState, xmm.avx.data[16])) && \"Can't StoreContextIndexed to XMM\""
         ]
       },
 
@@ -471,22 +479,6 @@
         ]
       },
 
-      "FPR = VLoadMemElement u8:#RegisterSize, u8:#ElementSize, FPR:$Value, GPR:$Addr, u8:$Index, u8:$Align{1}": {
-        "Desc": ["Loads an element of size #ElementSize in to $Value from $Addr at $Index"
-                ],
-        "OpClass": "Memory",
-        "DestSize": "RegisterSize",
-        "NumElements": "RegisterSize / ElementSize"
-      },
-
-      "VStoreMemElement u8:#RegisterSize, u8:#ElementSize, FPR:$Value, GPR:$Addr, u8:$Index, u8:$Align": {
-        "Desc": ["Stores an element of size #ElementSize from $Value[$Index] to $Addr"
-                ],
-        "HasSideEffects": true,
-        "DestSize": "ElementSize",
-        "NumElements": "RegisterSize / ElementSize"
-      },
-
       "CacheLineClear GPR:$Addr": {
         "Desc": ["Does a 64 byte cacheline clear at the address specified",
                  "Only clears the data cachelines. Doesn't do any zeroing"
@@ -1022,16 +1014,6 @@
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
       },
-      "FPR = VSLI u8:#RegisterSize, u8:#ElementSize, FPR:$Vector, u8:$ByteShift": {
-        "DestSize": "RegisterSize",
-        "NumElements": "RegisterSize / ElementSize"
-      },
-
-      "FPR = VSRI u8:#RegisterSize, u8:#ElementSize, FPR:$Vector, u8:$ByteShift": {
-        "DestSize": "RegisterSize",
-        "NumElements": "RegisterSize / ElementSize"
-      },
-
       "FPR = VShlI u8:#RegisterSize, u8:#ElementSize, FPR:$Vector, u8:$BitShift": {
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / ElementSize"
@@ -1065,7 +1047,7 @@
       },
       "FPR = VSXTL2 u8:#RegisterSize, u8:#ElementSize, FPR:$Vector": {
         "Desc": ["Sign extends elements from the source element size to the next size up",
-                 "Source elements come from the upper 64bits of the register"
+                 "Source elements come from the upper half of the register"
                 ],
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / (ElementSize << 1)"
@@ -1077,7 +1059,7 @@
       },
       "FPR = VUXTL2 u8:#RegisterSize, u8:#ElementSize, FPR:$Vector": {
         "Desc": ["Zero extends elements from the source element size to the next size up",
-                 "Source elements come from the upper 64bits of the register"
+                 "Source elements come from the upper half of the register"
                 ],
         "DestSize": "RegisterSize",
         "NumElements": "RegisterSize / (ElementSize << 1)"

External/FEXCore/Source/Interface/IR/IRParser.cpp

@@ -162,6 +162,12 @@ class IRParser: public FEXCore::IR::IREmitter {
     else if (Arg == "FPR") {
       return {DecodeFailure::DECODE_OKAY, FEXCore::IR::FPRClass};
     }
+    else if (Arg == "GPRFixed") {
+      return {DecodeFailure::DECODE_OKAY, FEXCore::IR::GPRFixedClass};
+    }
+    else if (Arg == "FPRFixed") {
+      return {DecodeFailure::DECODE_OKAY, FEXCore::IR::FPRFixedClass};
+    }
     else if (Arg == "GPRPair") {
       return {DecodeFailure::DECODE_OKAY, FEXCore::IR::GPRPairClass};
     }

External/FEXCore/Source/Interface/IR/PassManager.cpp

@@ -12,6 +12,7 @@ $end_info$
 #include "Interface/IR/Passes/RegisterAllocationPass.h"
 
 #include <FEXCore/Config/Config.h>
+#include <FEXCore/Utils/Profiler.h>
 
 namespace FEXCore::IR {
 class IREmitter;
@@ -36,15 +37,6 @@ void PassManager::AddDefaultPasses(FEXCore::Context::Context *ctx, bool InlineCo
 
     InsertPass(CreateSyscallOptimization());
     InsertPass(CreatePassDeadCodeElimination());
-
-    // only do SRA if enabled and JIT
-    if (InlineConstants && StaticRegisterAllocation)
-      InsertPass(CreateStaticRegisterAllocationPass(ctx->HostFeatures.SupportsAVX));
-  }
-  else {
-    // only do SRA if enabled and JIT
-    if (InlineConstants && StaticRegisterAllocation)
-      InsertPass(CreateStaticRegisterAllocationPass(ctx->HostFeatures.SupportsAVX));
   }
 
   // If the IR is compacted post-RA then the node indexing gets messed up and the backend isn't able to find the register assigned to a node
@@ -66,6 +58,8 @@ void PassManager::InsertRegisterAllocationPass(bool OptimizeSRA, bool SupportsAV
 }
 
 bool PassManager::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::Run");
+
   bool Changed = false;
   for (auto const &Pass : Passes) {
     Changed |= Pass->Run(IREmit);

External/FEXCore/Source/Interface/IR/Passes.h

@@ -21,7 +21,6 @@ std::unique_ptr<FEXCore::IR::Pass> CreateIRCompaction(FEXCore::Utils::IntrusiveP
 std::unique_ptr<FEXCore::IR::RegisterAllocationPass> CreateRegisterAllocationPass(FEXCore::IR::Pass* CompactionPass,
                                                                                   bool OptimizeSRA,
                                                                                   bool SupportsAVX);
-std::unique_ptr<FEXCore::IR::Pass> CreateStaticRegisterAllocationPass(bool SupportsAVX);
 std::unique_ptr<FEXCore::IR::Pass> CreateLongDivideEliminationPass();
 
 namespace Validation {

External/FEXCore/Source/Interface/IR/Passes/ConstProp.cpp

@@ -20,6 +20,7 @@ $end_info$
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <bit>
 #include <cstdint>
@@ -1028,6 +1029,8 @@ bool ConstProp::ConstantInlining(IREmitter *IREmit, const IRListView& CurrentIR)
 }
 
 bool ConstProp::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::ConstProp");
+
   bool Changed = false;
   auto CurrentIR = IREmit->ViewIR();
   auto OriginalWriteCursor = IREmit->GetWriteCursor();

External/FEXCore/Source/Interface/IR/Passes/DeadCodeElimination.cpp

@@ -9,6 +9,7 @@ $end_info$
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <memory>
 
@@ -22,6 +23,7 @@ private:
 };
 
 bool DeadCodeElimination::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::DCE");
   auto CurrentIR = IREmit->ViewIR();
   int NumRemoved = 0;
 

External/FEXCore/Source/Interface/IR/Passes/DeadContextStoreElimination.cpp

@@ -13,6 +13,7 @@ $end_info$
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <array>
 #include <memory>
@@ -76,24 +77,6 @@ namespace {
     std::vector<ContextMemberInfo> ClassificationInfo;
   };
 
-  constexpr static std::array<LastAccessType, 15> DefaultAccess = {
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_INVALID, // SSE padding in non-AVX case
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-    ACCESS_NONE,
-  };
-
   static void ClassifyContextStruct(ContextInfo *ContextClassificationInfo, bool SupportsAVX) {
     auto ContextClassification = &ContextClassificationInfo->ClassificationInfo;
 
@@ -102,7 +85,7 @@ namespace {
         offsetof(FEXCore::Core::CPUState, rip),
         sizeof(FEXCore::Core::CPUState::rip),
       },
-      DefaultAccess[0],
+      ACCESS_NONE,
       FEXCore::IR::InvalidClass,
     });
 
@@ -112,62 +95,134 @@ namespace {
           offsetof(FEXCore::Core::CPUState, gregs[0]) + sizeof(FEXCore::Core::CPUState::gregs[0]) * i,
           FEXCore::Core::CPUState::GPR_REG_SIZE,
         },
-        DefaultAccess[1],
+        ACCESS_NONE,
         FEXCore::IR::InvalidClass,
       });
     }
 
     ContextClassification->emplace_back(ContextMemberInfo{
       ContextMemberClassification {
-        offsetof(FEXCore::Core::CPUState, es),
-        sizeof(FEXCore::Core::CPUState::es),
+        offsetof(FEXCore::Core::CPUState, es_idx),
+        sizeof(FEXCore::Core::CPUState::es_idx),
       },
-      DefaultAccess[2],
+      ACCESS_NONE,
       FEXCore::IR::InvalidClass,
     });
 
     ContextClassification->emplace_back(ContextMemberInfo{
       ContextMemberClassification {
-        offsetof(FEXCore::Core::CPUState, cs),
-        sizeof(FEXCore::Core::CPUState::cs),
+        offsetof(FEXCore::Core::CPUState, cs_idx),
+        sizeof(FEXCore::Core::CPUState::cs_idx),
       },
-      DefaultAccess[3],
+      ACCESS_NONE,
       FEXCore::IR::InvalidClass,
     });
 
     ContextClassification->emplace_back(ContextMemberInfo{
       ContextMemberClassification {
-        offsetof(FEXCore::Core::CPUState, ss),
-        sizeof(FEXCore::Core::CPUState::ss),
+        offsetof(FEXCore::Core::CPUState, ss_idx),
+        sizeof(FEXCore::Core::CPUState::ss_idx),
       },
-      DefaultAccess[4],
+      ACCESS_NONE,
       FEXCore::IR::InvalidClass,
     });
 
     ContextClassification->emplace_back(ContextMemberInfo{
       ContextMemberClassification {
-        offsetof(FEXCore::Core::CPUState, ds),
-        sizeof(FEXCore::Core::CPUState::ds),
+        offsetof(FEXCore::Core::CPUState, ds_idx),
+        sizeof(FEXCore::Core::CPUState::ds_idx),
       },
-      DefaultAccess[5],
+      ACCESS_NONE,
       FEXCore::IR::InvalidClass,
     });
 
     ContextClassification->emplace_back(ContextMemberInfo{
       ContextMemberClassification {
-        offsetof(FEXCore::Core::CPUState, gs),
-        sizeof(FEXCore::Core::CPUState::gs),
+        offsetof(FEXCore::Core::CPUState, gs_idx),
+        sizeof(FEXCore::Core::CPUState::gs_idx),
       },
-      DefaultAccess[6],
+      ACCESS_NONE,
       FEXCore::IR::InvalidClass,
     });
 
     ContextClassification->emplace_back(ContextMemberInfo{
       ContextMemberClassification {
-        offsetof(FEXCore::Core::CPUState, fs),
-        sizeof(FEXCore::Core::CPUState::fs),
+        offsetof(FEXCore::Core::CPUState, fs_idx),
+        sizeof(FEXCore::Core::CPUState::fs_idx),
       },
-      DefaultAccess[7],
+      ACCESS_NONE,
+      FEXCore::IR::InvalidClass,
+    });
+
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, _pad),
+        sizeof(FEXCore::Core::CPUState::_pad),
+      },
+      ACCESS_INVALID,
+      FEXCore::IR::InvalidClass,
+    });
+
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, es_cached),
+        sizeof(FEXCore::Core::CPUState::es_cached),
+      },
+      ACCESS_NONE,
+      FEXCore::IR::InvalidClass,
+    });
+
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, cs_cached),
+        sizeof(FEXCore::Core::CPUState::cs_cached),
+      },
+      ACCESS_NONE,
+      FEXCore::IR::InvalidClass,
+    });
+
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, ss_cached),
+        sizeof(FEXCore::Core::CPUState::ss_cached),
+      },
+      ACCESS_NONE,
+      FEXCore::IR::InvalidClass,
+    });
+
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, ds_cached),
+        sizeof(FEXCore::Core::CPUState::ds_cached),
+      },
+      ACCESS_NONE,
+      FEXCore::IR::InvalidClass,
+    });
+
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, gs_cached),
+        sizeof(FEXCore::Core::CPUState::gs_cached),
+      },
+      ACCESS_NONE,
+      FEXCore::IR::InvalidClass,
+    });
+
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, fs_cached),
+        sizeof(FEXCore::Core::CPUState::fs_cached),
+      },
+      ACCESS_NONE,
+      FEXCore::IR::InvalidClass,
+    });
+
+    ContextClassification->emplace_back(ContextMemberInfo{
+      ContextMemberClassification {
+        offsetof(FEXCore::Core::CPUState, _pad2),
+        sizeof(FEXCore::Core::CPUState::_pad2),
+      },
+      ACCESS_INVALID,
       FEXCore::IR::InvalidClass,
     });
 
@@ -178,7 +233,7 @@ namespace {
             offsetof(FEXCore::Core::CPUState, xmm.avx.data[0][0]) + FEXCore::Core::CPUState::XMM_AVX_REG_SIZE * i,
             FEXCore::Core::CPUState::XMM_AVX_REG_SIZE,
           },
-          DefaultAccess[8],
+          ACCESS_NONE,
           FEXCore::IR::InvalidClass,
         });
       }
@@ -189,7 +244,7 @@ namespace {
             offsetof(FEXCore::Core::CPUState, xmm.sse.data[0][0]) + FEXCore::Core::CPUState::XMM_SSE_REG_SIZE * i,
             FEXCore::Core::CPUState::XMM_SSE_REG_SIZE,
           },
-          DefaultAccess[8],
+          ACCESS_NONE,
           FEXCore::IR::InvalidClass,
         });
       }
@@ -199,7 +254,7 @@ namespace {
             offsetof(FEXCore::Core::CPUState, xmm.sse.pad[0][0]),
             static_cast<uint16_t>(FEXCore::Core::CPUState::XMM_SSE_REG_SIZE * FEXCore::Core::CPUState::NUM_XMMS),
           },
-          DefaultAccess[9],
+          ACCESS_INVALID,
           FEXCore::IR::InvalidClass,
         });
     }
@@ -210,7 +265,7 @@ namespace {
           offsetof(FEXCore::Core::CPUState, flags[0]) + sizeof(FEXCore::Core::CPUState::flags[0]) * i,
           FEXCore::Core::CPUState::FLAG_SIZE,
         },
-        DefaultAccess[10],
+        ACCESS_NONE,
         FEXCore::IR::InvalidClass,
       });
     }
@@ -221,7 +276,7 @@ namespace {
           offsetof(FEXCore::Core::CPUState, mm[0][0]) + sizeof(FEXCore::Core::CPUState::mm[0]) * i,
           FEXCore::Core::CPUState::MM_REG_SIZE
         },
-        DefaultAccess[11],
+        ACCESS_NONE,
         FEXCore::IR::InvalidClass,
       });
     }
@@ -233,7 +288,7 @@ namespace {
           offsetof(FEXCore::Core::CPUState, gdt[0]) + sizeof(FEXCore::Core::CPUState::gdt[0]) * i,
           sizeof(FEXCore::Core::CPUState::gdt[0]),
         },
-        DefaultAccess[12],
+        ACCESS_NONE,
         FEXCore::IR::InvalidClass,
       });
     }
@@ -244,7 +299,7 @@ namespace {
         offsetof(FEXCore::Core::CPUState, FCW),
         sizeof(FEXCore::Core::CPUState::FCW),
       },
-      DefaultAccess[13],
+      ACCESS_NONE,
       FEXCore::IR::InvalidClass,
     });
 
@@ -254,7 +309,7 @@ namespace {
         offsetof(FEXCore::Core::CPUState, FTW),
         sizeof(FEXCore::Core::CPUState::FTW),
       },
-      DefaultAccess[14],
+      ACCESS_NONE,
       FEXCore::IR::InvalidClass,
     });
 
@@ -288,40 +343,55 @@ namespace {
       ContextClassification->at(Offset).StoreNode = nullptr;
     };
     size_t Offset = 0;
-    SetAccess(Offset++, DefaultAccess[0]);
+    SetAccess(Offset++, ACCESS_NONE);
     for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_GPRS; ++i) {
-      SetAccess(Offset++, DefaultAccess[1]);
+      SetAccess(Offset++, ACCESS_NONE);
     }
 
-    SetAccess(Offset++, DefaultAccess[2]);
-    SetAccess(Offset++, DefaultAccess[3]);
-    SetAccess(Offset++, DefaultAccess[4]);
-    SetAccess(Offset++, DefaultAccess[5]);
-    SetAccess(Offset++, DefaultAccess[6]);
-    SetAccess(Offset++, DefaultAccess[7]);
+    // Segment indexes
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+
+    // Pad
+    SetAccess(Offset++, ACCESS_INVALID);
+
+    // Segments
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
+
+    // Pad2
+    SetAccess(Offset++, ACCESS_INVALID);
 
     for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_XMMS; ++i) {
-      SetAccess(Offset++, DefaultAccess[8]);
+      SetAccess(Offset++, ACCESS_NONE);
     }
 
     if (!SupportsAVX) {
-      SetAccess(Offset++, DefaultAccess[9]);
+      SetAccess(Offset++, ACCESS_NONE);
     }
 
     for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_FLAGS; ++i) {
-      SetAccess(Offset++, DefaultAccess[10]);
+      SetAccess(Offset++, ACCESS_NONE);
     }
 
     for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_MMS; ++i) {
-      SetAccess(Offset++, DefaultAccess[11]);
+      SetAccess(Offset++, ACCESS_NONE);
     }
 
     for (size_t i = 0; i < FEXCore::Core::CPUState::NUM_GDTS; ++i) {
-      SetAccess(Offset++, DefaultAccess[12]);
+      SetAccess(Offset++, ACCESS_NONE);
     }
 
-    SetAccess(Offset++, DefaultAccess[13]);
-    SetAccess(Offset++, DefaultAccess[14]);
+    SetAccess(Offset++, ACCESS_NONE);
+    SetAccess(Offset++, ACCESS_NONE);
   }
 
   struct BlockInfo {
@@ -695,6 +765,7 @@ bool RCLSE::RedundantStoreLoadElimination(FEXCore::IR::IREmitter *IREmit) {
 }
 
 bool RCLSE::Run(FEXCore::IR::IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::RCLSE");
   // XXX: We don't do cross-block optimizations yet
   //CalculateControlFlowInfo(IREmit);
   bool Changed = false;

External/FEXCore/Source/Interface/IR/Passes/DeadStoreElimination.cpp

@@ -12,6 +12,7 @@ $end_info$
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <memory>
 #include <stddef.h>
@@ -154,6 +155,8 @@ struct Info {
  *
  */
 bool DeadStoreElimination::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::DSE");
+
   std::unordered_map<OrderedNode*, Info> InfoMap;
 
   bool Changed = false;

External/FEXCore/Source/Interface/IR/Passes/IRCompaction.cpp

@@ -13,6 +13,7 @@ $end_info$
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <algorithm>
 #include <cstdint>
@@ -52,6 +53,8 @@ IRCompaction::IRCompaction(FEXCore::Utils::IntrusivePooledAllocator &Allocator)
 }
 
 bool IRCompaction::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::IRCompaction");
+
   LocalBuilder.ReownOrClaimBuffer();
 
   auto CurrentIR = IREmit->ViewIR();

External/FEXCore/Source/Interface/IR/Passes/IRValidation.cpp

@@ -14,6 +14,7 @@ $end_info$
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/IR/RegisterAllocationData.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <cstdint>
 #include <memory>
@@ -32,6 +33,8 @@ IRValidation::~IRValidation() {
 }
 
 bool IRValidation::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::IRValidation");
+
   bool HadError = false;
   bool HadWarning = false;
 

External/FEXCore/Source/Interface/IR/Passes/LongDivideRemovalPass.cpp

@@ -9,6 +9,7 @@ $end_info$
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <memory>
 #include <stdint.h>
@@ -53,6 +54,8 @@ bool LongDivideEliminationPass::IsSextOp(IREmitter *IREmit, OrderedNodeWrapper L
 }
 
 bool LongDivideEliminationPass::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::LDE");
+
   bool Changed = false;
   auto CurrentIR = IREmit->ViewIR();
   auto OriginalWriteCursor = IREmit->GetWriteCursor();

External/FEXCore/Source/Interface/IR/Passes/PhiValidation.cpp

@@ -9,6 +9,7 @@ $end_info$
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include "Interface/IR/PassManager.h"
 
@@ -24,6 +25,8 @@ public:
 };
 
 bool PhiValidation::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::PHIValidation");
+
   bool HadError = false;
   auto CurrentIR = IREmit->ViewIR();
 

External/FEXCore/Source/Interface/IR/Passes/RAValidation.cpp

@@ -6,7 +6,7 @@
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/IR/RegisterAllocationData.h>
-
+#include <FEXCore/Utils/Profiler.h>
 
 #include <algorithm>
 #include <deque>
@@ -191,6 +191,8 @@ private:
 bool RAValidation::Run(IREmitter *IREmit) {
   if (!Manager->HasPass("RA")) return false;
 
+  FEXCORE_PROFILE_SCOPED("PassManager::RAValidation");
+
   IR::RegisterAllocationData* RAData = Manager->GetPass<IR::RegisterAllocationPass>("RA")->GetAllocationData();
   BlockExitState.clear();
   // BlocksToVisit will already be empty

External/FEXCore/Source/Interface/IR/Passes/RedundantFlagCalculationElimination.cpp

@@ -8,6 +8,8 @@ $end_info$
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
+#include <FEXCore/Utils/Profiler.h>
+
 #include "Interface/IR/PassManager.h"
 
 #include <array>
@@ -32,6 +34,8 @@ public:
  *
  */
 bool DeadFlagCalculationEliminination::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::DFE");
+
   std::array<OrderedNode*, 32> LastValidFlagStores{};
 
   bool Changed = false;

External/FEXCore/Source/Interface/IR/Passes/RegisterAllocationPass.cpp

@@ -15,6 +15,8 @@ $end_info$
 #include <FEXCore/Utils/BucketList.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
+#include <FEXCore/Utils/Profiler.h>
+
 #include <FEXHeaderUtils/TypeDefines.h>
 
 #include <algorithm>
@@ -1527,6 +1529,7 @@ namespace {
   }
 
   bool ConstrainedRAPass::Run(IREmitter *IREmit) {
+    FEXCORE_PROFILE_SCOPED("PassManager::RA");
     bool Changed = false;
 
     auto IR = IREmit->ViewIR();

External/FEXCore/Source/Interface/IR/Passes/StaticRegisterAllocationPass.cpp

@@ -1,128 +0,0 @@
-/*
-$info$
-tags: ir|opts
-desc: Replaces Load/StoreContext with Load/StoreReg for SRA regs
-$end_info$
-*/
-
-#include "Interface/IR/PassManager.h"
-#include <FEXCore/Core/CoreState.h>
-#include <FEXCore/IR/IR.h>
-#include <FEXCore/IR/IREmitter.h>
-#include <FEXCore/IR/IntrusiveIRList.h>
-#include <FEXCore/Utils/LogManager.h>
-
-#include <memory>
-#include <stddef.h>
-#include <stdint.h>
-
-namespace FEXCore::IR {
-
-class StaticRegisterAllocationPass final : public FEXCore::IR::Pass {
-public:
-  explicit StaticRegisterAllocationPass(bool SupportsAVX_) : SupportsAVX{SupportsAVX_} {}
-
-  bool Run(IREmitter *IREmit) override;
-
-private:
-  bool SupportsAVX;
-
-  bool IsStaticAllocGpr(uint32_t Offset, RegisterClassType Class) const {
-    const auto begin = offsetof(Core::CPUState, gregs[0]);
-    const auto end = offsetof(Core::CPUState, gregs[16]);
-
-    if (Offset >= begin && Offset < end) {
-      const auto reg = (Offset - begin) / Core::CPUState::GPR_REG_SIZE;
-      LOGMAN_THROW_AA_FMT(Class.Val == IR::GPRClass.Val, "unexpected Class {}", Class);
-
-      // 0..15 -> 16 in total
-      return reg < Core::CPUState::NUM_GPRS;
-    }
-
-    return false;
-  }
-
-  bool IsStaticAllocFpr(uint32_t Offset, RegisterClassType Class, bool AllowGpr) const {
-    const auto [begin, end] = [this]() -> std::pair<ptrdiff_t, ptrdiff_t> {
-      if (SupportsAVX) {
-        return {
-          offsetof(Core::CPUState, xmm.avx.data[0][0]),
-          offsetof(Core::CPUState, xmm.avx.data[16][0]),
-        };
-      } else {
-        return {
-          offsetof(Core::CPUState, xmm.sse.data[0][0]),
-          offsetof(Core::CPUState, xmm.sse.data[16][0]),
-        };
-      }
-    }();
-
-    if (Offset >= begin && Offset < end) {
-      const auto size = SupportsAVX ? Core::CPUState::XMM_AVX_REG_SIZE
-                                    : Core::CPUState::XMM_SSE_REG_SIZE;
-      const auto reg = (Offset - begin) / size;
-      LOGMAN_THROW_AA_FMT(Class.Val == IR::FPRClass.Val || (AllowGpr && Class.Val == IR::GPRClass.Val), "unexpected Class {}, AllowGpr {}", Class, AllowGpr);
-
-      // 0..15 -> 16 in total
-      return reg < Core::CPUState::NUM_XMMS;
-    }
-
-    return false;
-  }
-};
-
-/**
- * @brief This pass replaces Load/Store Context with Load/Store Register for Statically Mapped registers. It also does some validation.
- *
- */
-bool StaticRegisterAllocationPass::Run(IREmitter *IREmit) {
-  auto CurrentIR = IREmit->ViewIR();
-
-  for (auto [BlockNode, BlockIROp] : CurrentIR.GetBlocks()) {
-      for (auto [CodeNode, IROp] : CurrentIR.GetCode(BlockNode)) {
-        IREmit->SetWriteCursor(CodeNode);
-
-        if (IROp->Op == OP_LOADCONTEXT) {
-            auto Op = IROp->CW<IR::IROp_LoadContext>();
-
-            if (IsStaticAllocGpr(Op->Offset, Op->Class) || IsStaticAllocFpr(Op->Offset, Op->Class, true)) {
-              auto GeneralClass = Op->Class;
-              if (IsStaticAllocFpr(Op->Offset, GeneralClass, true) && GeneralClass == GPRClass) {
-                GeneralClass = FPRClass;
-              }
-              auto StaticClass = GeneralClass == GPRClass ? GPRFixedClass : FPRFixedClass;
-              OrderedNode *sraReg = IREmit->_LoadRegister(false, Op->Offset, GeneralClass, StaticClass, Op->Header.Size);
-              if (GeneralClass != Op->Class) {
-                sraReg = IREmit->_VExtractToGPR(Op->Header.Size, Op->Header.Size, sraReg, 0);
-              }
-              IREmit->ReplaceAllUsesWith(CodeNode, sraReg);
-            }
-        } if (IROp->Op == OP_STORECONTEXT) {
-            auto Op = IROp->CW<IR::IROp_StoreContext>();
-
-            if (IsStaticAllocGpr(Op->Offset, Op->Class) || IsStaticAllocFpr(Op->Offset, Op->Class, true)) {
-              auto val = IREmit->UnwrapNode(Op->Value);
-
-              auto GeneralClass = Op->Class;
-              if (IsStaticAllocFpr(Op->Offset, GeneralClass, true) && GeneralClass == GPRClass) {
-                val = IREmit->_VCastFromGPR(Op->Header.Size, Op->Header.Size, val);
-                GeneralClass = FPRClass;
-              }
-
-              auto StaticClass = GeneralClass == GPRClass ? GPRFixedClass : FPRFixedClass;
-              IREmit->_StoreRegister(val, false, Op->Offset, GeneralClass, StaticClass, Op->Header.Size);
-
-              IREmit->Remove(CodeNode);
-            }
-        }
-    }
-  }
-
-  return true;
-}
-
-std::unique_ptr<FEXCore::IR::Pass> CreateStaticRegisterAllocationPass(bool SupportsAVX) {
-  return std::make_unique<StaticRegisterAllocationPass>(SupportsAVX);
-}
-
-}

External/FEXCore/Source/Interface/IR/Passes/SyscallOptimization.cpp

@@ -11,6 +11,7 @@ $end_info$
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/HLE/SyscallHandler.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <memory>
 #include <stdint.h>
@@ -23,6 +24,8 @@ public:
 };
 
 bool SyscallOptimization::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::SyscallOpt");
+
   bool Changed = false;
   auto CurrentIR = IREmit->ViewIR();
 

External/FEXCore/Source/Interface/IR/Passes/ValueDominanceValidation.cpp

@@ -11,6 +11,7 @@ $end_info$
 #include <FEXCore/IR/IREmitter.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <functional>
 #include <memory>
@@ -36,6 +37,8 @@ public:
 };
 
 bool ValueDominanceValidation::Run(IREmitter *IREmit) {
+  FEXCORE_PROFILE_SCOPED("PassManager::ValueDominanceValidation");
+
   bool HadError = false;
   auto CurrentIR = IREmit->ViewIR();
 

External/FEXCore/Source/Utils/Allocator.cpp

@@ -145,8 +145,10 @@ namespace FEXCore::Allocator {
   #define STEAL_LOG(...) // fprintf(stderr, __VA_ARGS__)
 
   std::vector<MemoryRegion> StealMemoryRegion(uintptr_t Begin, uintptr_t End) {
+    void * const StackLocation = alloca(0);
+    const uintptr_t StackLocation_u64 = reinterpret_cast<uintptr_t>(StackLocation);
     std::vector<MemoryRegion> Regions;
-    
+
     int MapsFD = open("/proc/self/maps", O_RDONLY);
     LogMan::Throw::AFmt(MapsFD != -1, "Failed to open /proc/self/maps");
 
@@ -155,6 +157,8 @@ namespace FEXCore::Allocator {
     uintptr_t RegionBegin = 0;
     uintptr_t RegionEnd = 0;
 
+    uintptr_t PreviousMapEnd = 0;
+
     char Buffer[2048];
     const char *Cursor;
     ssize_t Remaining = 0;
@@ -162,7 +166,7 @@ namespace FEXCore::Allocator {
     for(;;) {
 
       if (Remaining == 0) {
-        do { 
+        do {
           Remaining = read(MapsFD, Buffer, sizeof(Buffer));
         } while ( Remaining == -1 && errno == EAGAIN);
 
@@ -172,8 +176,8 @@ namespace FEXCore::Allocator {
       if (Remaining == 0 && State == ParseBegin) {
         STEAL_LOG("[%d] EndOfFile; RegionBegin: %016lX RegionEnd: %016lX\n", __LINE__, RegionBegin, RegionEnd);
 
-        auto MapBegin = std::max(RegionEnd, Begin);
-        auto MapEnd = End;
+        const auto MapBegin = std::max(RegionEnd, Begin);
+        const auto MapEnd = End;
 
         STEAL_LOG("     MapBegin: %016lX MapEnd: %016lX\n", MapBegin, MapEnd);
 
@@ -209,9 +213,12 @@ namespace FEXCore::Allocator {
         if (c == '-') {
           STEAL_LOG("[%d] ParseBegin; RegionBegin: %016lX RegionEnd: %016lX\n", __LINE__, RegionBegin, RegionEnd);
 
-          auto MapBegin = std::max(RegionEnd, Begin);
-          auto MapEnd = std::min(RegionBegin, End);
-          
+          const auto MapBegin = std::max(RegionEnd, Begin);
+          const auto MapEnd = std::min(RegionBegin, End);
+
+          // Store the location we are going to map.
+          PreviousMapEnd = MapEnd;
+
           STEAL_LOG("     MapBegin: %016lX MapEnd: %016lX\n", MapBegin, MapEnd);
 
           if (MapEnd > MapBegin) {
@@ -225,6 +232,7 @@ namespace FEXCore::Allocator {
 
             Regions.push_back({(void*)MapBegin, MapSize});
           }
+
           RegionBegin = 0;
           RegionEnd = 0;
           State = ParseEnd;
@@ -240,6 +248,22 @@ namespace FEXCore::Allocator {
           STEAL_LOG("[%d] ParseEnd; RegionBegin: %016lX RegionEnd: %016lX\n", __LINE__, RegionBegin, RegionEnd);
 
           State = ScanEnd;
+
+          // If the previous map's ending and the region we just parsed overlap the stack then we need to save the stack mapping.
+          // Otherwise we will have severely limited stack size which crashes quickly.
+          if (PreviousMapEnd <= StackLocation_u64 && RegionEnd > StackLocation_u64) {
+            auto BelowStackRegion = Regions.back();
+            LOGMAN_THROW_AA_FMT(reinterpret_cast<uint64_t>(BelowStackRegion.Ptr) + BelowStackRegion.Size == PreviousMapEnd,
+              "This needs to match");
+
+            // Allocate the region under the stack as READ | WRITE so the stack can still grow
+            auto Alloc = mmap(BelowStackRegion.Ptr, BelowStackRegion.Size, PROT_READ | PROT_WRITE, MAP_ANONYMOUS | MAP_NORESERVE | MAP_PRIVATE | MAP_FIXED, -1, 0);
+
+            LogMan::Throw::AFmt(Alloc != MAP_FAILED, "mmap({:x},{:x}) failed", BelowStackRegion.Ptr, BelowStackRegion.Size);
+            LogMan::Throw::AFmt(Alloc == BelowStackRegion.Ptr, "mmap({},{:x}) returned {} instead of {:x}", Alloc, BelowStackRegion.Ptr);
+
+            Regions.pop_back();
+          }
           continue;
         } else {
           LogMan::Throw::AFmt(std::isalpha(c) || std::isdigit(c), "Unexpected char '{}' in ParseEnd", c);

External/FEXCore/Source/Utils/Allocator/64BitAllocator.cpp

@@ -69,11 +69,14 @@ namespace Alloc::OSAllocator {
       struct LiveVMARegion {
         ReservedVMARegion *SlabInfo;
         uint64_t FreeSpace{};
+        uint64_t NumManagedPages{};
         uint32_t LastPageAllocation{};
+        bool HadMunmap{};
 
         // Align UsedPages so it pads to the next page.
         // Necessary to take advantage of madvise zero page pooling.
-        alignas(4096) FEXCore::FlexBitSet<uint64_t> UsedPages;
+        using FlexBitElementType = uint64_t;
+        alignas(4096) FEXCore::FlexBitSet<FlexBitElementType> UsedPages;
 
         // This returns the size of the LiveVMARegion in addition to the flex set that tracks the used data
         // The LiveVMARegion lives at the start of the VMA region which means on initialization we need to set that
@@ -85,8 +88,8 @@ namespace Alloc::OSAllocator {
           // 0x100'0000 Pages
           // 1 bit per page for tracking means 0x20'0000 (Pages / 8) bytes of flex space
           // Which is 2MB of tracking
-          uint64_t NumElements = (Size >> FHU::FEX_PAGE_SHIFT) * sizeof(uint64_t);
-          return sizeof(LiveVMARegion) + FEXCore::FlexBitSet<uint64_t>::Size(NumElements);
+          uint64_t NumElements = (Size >> FHU::FEX_PAGE_SHIFT) * sizeof(FlexBitElementType);
+          return sizeof(LiveVMARegion) + FEXCore::FlexBitSet<FlexBitElementType>::Size(NumElements);
         }
 
         static void InitializeVMARegionUsed(LiveVMARegion *Region, size_t AdditionalSize) {
@@ -95,19 +98,21 @@ namespace Alloc::OSAllocator {
 
           Region->FreeSpace = Region->SlabInfo->RegionSize - SizePlusManagedData;
 
-          size_t NumPages = SizePlusManagedData >> FHU::FEX_PAGE_SHIFT;
+          size_t NumManagedPages = SizePlusManagedData >> FHU::FEX_PAGE_SHIFT;
+          size_t ManagedSize  = NumManagedPages << FHU::FEX_PAGE_SHIFT;
 
           // Use madvise to set the full tracking region to zero.
           // This ensures unused pages are zero, while not having the backing pages consuming memory.
-          ::madvise(Region->UsedPages.Memory + (NumPages * 4096), (Region->SlabInfo->RegionSize >> FHU::FEX_PAGE_SHIFT) - (NumPages * 4096), MADV_DONTNEED);
+          ::madvise(Region->UsedPages.Memory + ManagedSize, (Region->SlabInfo->RegionSize >> FHU::FEX_PAGE_SHIFT) - ManagedSize, MADV_DONTNEED);
 
           // Use madvise to claim WILLNEED on the beginning pages for initial state tracking.
           // Improves performance of the following MemClear by not doing a page level fault dance for data necessary to track >170TB of used pages.
-          ::madvise(Region->UsedPages.Memory, NumPages * 4096, MADV_WILLNEED);
+          ::madvise(Region->UsedPages.Memory, ManagedSize, MADV_WILLNEED);
 
           // Set our reserved pages
-          Region->UsedPages.MemSet(NumPages);
-          Region->LastPageAllocation = NumPages;
+          Region->UsedPages.MemSet(NumManagedPages);
+          Region->LastPageAllocation = NumManagedPages;
+          Region->NumManagedPages = NumManagedPages;
         }
       };
 
@@ -129,6 +134,7 @@ namespace Alloc::OSAllocator {
         ReservedVMARegion *ReservedRegion = *ReservedIterator;
 
         ReservedRegions->erase(ReservedIterator);
+
         // mprotect the new region we've allocated
         size_t SizeOfLiveRegion = FEXCore::AlignUp(LiveVMARegion::GetSizeWithFlexSet(ReservedRegion->RegionSize), FHU::FEX_PAGE_SIZE);
         size_t SizePlusManagedData = UsedSize + SizeOfLiveRegion;
@@ -152,6 +158,9 @@ namespace Alloc::OSAllocator {
 
       // 32-bit old kernel workarounds
       std::vector<FEXCore::Allocator::MemoryRegion> Steal32BitIfOldKernel();
+
+      void AllocateMemoryRegions(std::vector<FEXCore::Allocator::MemoryRegion> const &Ranges);
+      LiveVMARegion *FindLiveRegionForAddress(uintptr_t Addr, uintptr_t AddrEnd);
   };
 
 void OSAllocator_64Bit::DetermineVASize() {
@@ -167,6 +176,42 @@ void OSAllocator_64Bit::DetermineVASize() {
   UPPER_BOUND_PAGE = UPPER_BOUND / FHU::FEX_PAGE_SIZE;
 }
 
+OSAllocator_64Bit::LiveVMARegion *OSAllocator_64Bit::FindLiveRegionForAddress(uintptr_t Addr, uintptr_t AddrEnd) {
+  LiveVMARegion *LiveRegion{};
+
+  // Check active slabs to see if we can fit this
+  for (auto it = LiveRegions->begin(); it != LiveRegions->end(); ++it) {
+    uintptr_t RegionBegin = (*it)->SlabInfo->Base;
+    uintptr_t RegionEnd = RegionBegin + (*it)->SlabInfo->RegionSize;
+
+    if (Addr >= RegionBegin &&
+        Addr < RegionEnd) {
+      LiveRegion = *it;
+      // Leave our loop
+      break;
+    }
+  }
+
+  // Couldn't find an active region that fit
+  // Check reserved regions
+  if (!LiveRegion) {
+    // Didn't have a slab that fit this range
+    // Check our reserved regions to see if we have one that fits
+    for (auto it = ReservedRegions->begin(); it != ReservedRegions->end(); ++it) {
+      ReservedVMARegion *ReservedRegion = *it;
+      uintptr_t RegionEnd = ReservedRegion->Base + ReservedRegion->RegionSize;
+      if (Addr >= ReservedRegion->Base &&
+          AddrEnd < RegionEnd) {
+        // Found one, let's make it active
+        LiveRegion = MakeRegionActive(it, 0);
+        break;
+      }
+    }
+  }
+
+  return LiveRegion;
+}
+
 void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset) {
   if (addr != 0 &&
       addr < reinterpret_cast<void*>(LOWER_BOUND)) {
@@ -205,41 +250,13 @@ void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, in
   LiveVMARegion *LiveRegion{};
 
   if (Fixed || Addr != 0) {
-    // Check active slabs to see if we can fit this
-    for (auto it = LiveRegions->begin(); it != LiveRegions->end(); ++it) {
-      uintptr_t RegionBegin = (*it)->SlabInfo->Base;
-      uintptr_t RegionEnd = RegionBegin + (*it)->SlabInfo->RegionSize;
-
-      if (Addr >= RegionBegin &&
-          Addr < RegionEnd) {
-        LiveRegion = *it;
-        // Leave our loop
-        break;
-      }
-    }
-
-    // Couldn't find an active region that fit
-    // Check reserved regions
-    if (!LiveRegion) {
-      // Didn't have a slab that fit this range
-      // Check our reserved regions to see if we have one that fits
-      for (auto it = ReservedRegions->begin(); it != ReservedRegions->end(); ++it) {
-        ReservedVMARegion *ReservedRegion = *it;
-        uintptr_t RegionEnd = ReservedRegion->Base + ReservedRegion->RegionSize;
-        if (Addr >= ReservedRegion->Base &&
-            AddrEnd < RegionEnd) {
-          // Found one, let's make it active
-          LiveRegion = MakeRegionActive(it, 0);
-          break;
-        }
-      }
-    }
+    LiveRegion = FindLiveRegionForAddress(Addr, AddrEnd);
   }
 
   again:
 
   auto CheckIfRangeFits = [&AllocatedOffset](LiveVMARegion *Region, uint64_t length, int prot, int flags, int fd, off_t offset, uint64_t StartingPosition = 0) -> std::pair<LiveVMARegion*, void*> {
-    uint64_t AllocatedPage{};
+    uint64_t AllocatedPage{~0ULL};
     uint64_t NumberOfPages = length >> FHU::FEX_PAGE_SHIFT;
 
     if (Region->FreeSpace >= length) {
@@ -249,72 +266,29 @@ void *OSAllocator_64Bit::Mmap(void *addr, size_t length, int prot, int flags, in
         : Region->LastPageAllocation;
       size_t RegionNumberOfPages = Region->SlabInfo->RegionSize >> FHU::FEX_PAGE_SHIFT;
 
-      // Backward scan
-      // We need to do a backward scan first to fill any holes
-      // Otherwise we will very quickly run out of VMA regions (65k maximum)
-      for (size_t CurrentPage = LastAllocation;
-           CurrentPage >= NumberOfPages;) {
-        size_t Remaining = NumberOfPages;
-        assert(Remaining <= CurrentPage);
 
-        while (Remaining) {
-          if (Region->UsedPages[CurrentPage - Remaining]) {
-            // Has an intersecting range
-            break;
-          }
-          --Remaining;
-        }
+      if (Region->HadMunmap) {
+        // Backward scan
+        // We need to do a backward scan first to fill any holes
+        // Otherwise we will very quickly run out of VMA regions (65k maximum)
+        auto SearchResult = Region->UsedPages.BackwardScanForRange<true>(LastAllocation, NumberOfPages, Region->NumManagedPages);
 
-        if (Remaining) {
-          // Didn't find a slab range
-          CurrentPage -= Remaining;
-        }
-        else {
-          // We have a slab range
-          CurrentPage -= NumberOfPages;
+        AllocatedPage = SearchResult.FoundElement;
 
-          // Keep scanning backwards to not introduce ANOTHER gap
-          while (CurrentPage >= 1) {
-            if (Region->UsedPages[CurrentPage - 1]) {
-              // Found a used page, we can leave now
-              break;
-            }
-            --CurrentPage;
-          }
-          AllocatedPage = CurrentPage;
-          break;
+        // If we didn't even have a one page free in the backward search, then unclaim HadMunmap.
+        // Switching over to default forward search.
+        if (SearchResult.FoundElement == ~0ULL && !SearchResult.FoundHole) {
+          Region->HadMunmap = false;
         }
       }
 
       // Foward Scan
-      if (AllocatedPage == 0) {
-        for (size_t CurrentPage = LastAllocation;
-             CurrentPage < (RegionNumberOfPages - NumberOfPages);) {
-          // If we have enough free space, check if we have enough free pages that are contiguous
-          size_t Remaining = NumberOfPages;
-
-          assert((CurrentPage + Remaining - 1) < RegionNumberOfPages);
-          while (Remaining) {
-            if (Region->UsedPages[CurrentPage + Remaining - 1]) {
-              // Has an intersecting range
-              break;
-            }
-            --Remaining;
-          }
-
-          if (Remaining) {
-            // Didn't find a slab range
-            CurrentPage += Remaining;
-          }
-          else {
-            // We have a slab range
-            AllocatedPage = CurrentPage;
-            break;
-          }
-        }
+      if (AllocatedPage == ~0ULL) {
+        auto SearchResult = Region->UsedPages.ForwardScanForRange<true>(LastAllocation, NumberOfPages, RegionNumberOfPages);
+        AllocatedPage = SearchResult.FoundElement;
       }
 
-      if (AllocatedPage) {
+      if (AllocatedPage != ~0ULL) {
         AllocatedOffset = Region->SlabInfo->Base + AllocatedPage * FHU::FEX_PAGE_SIZE;
 
         // We need to setup protections for this
@@ -497,6 +471,8 @@ int OSAllocator_64Bit::Munmap(void *addr, size_t length) {
       // This will let us more quickly fill holes
       (*it)->LastPageAllocation = std::min((*it)->LastPageAllocation, SlabPageBegin);
 
+      (*it)->HadMunmap = true;
+
       // XXX: Move region back to reserved list
       return 0;
     }
@@ -537,12 +513,7 @@ std::vector<FEXCore::Allocator::MemoryRegion> OSAllocator_64Bit::Steal32BitIfOld
   return FEXCore::Allocator::StealMemoryRegion(LOWER_BOUND_32, UPPER_BOUND_32);
 }
 
-OSAllocator_64Bit::OSAllocator_64Bit() {
-  DetermineVASize();
-  auto LowMem = Steal32BitIfOldKernel();
-
-  auto Ranges = FEXCore::Allocator::StealMemoryRegion(LOWER_BOUND, UPPER_BOUND);
-
+void OSAllocator_64Bit::AllocateMemoryRegions(std::vector<FEXCore::Allocator::MemoryRegion> const &Ranges) {
   for (auto [Ptr, AllocationSize]: Ranges) {
     if (!ObjectAlloc) {
       auto MaxSize = std::min(size_t(64) * 1024 * 1024, AllocationSize);
@@ -564,12 +535,22 @@ OSAllocator_64Bit::OSAllocator_64Bit() {
         continue;
       }
     }
-    
+
     ReservedVMARegion *Region = ObjectAlloc->new_construct<ReservedVMARegion>();
     Region->Base = reinterpret_cast<uint64_t>(Ptr);
     Region->RegionSize = AllocationSize;
     ReservedRegions->emplace_back(Region);
   }
+}
+
+
+OSAllocator_64Bit::OSAllocator_64Bit() {
+  DetermineVASize();
+  auto LowMem = Steal32BitIfOldKernel();
+
+  auto Ranges = FEXCore::Allocator::StealMemoryRegion(LOWER_BOUND, UPPER_BOUND);
+
+  AllocateMemoryRegions(Ranges);
 
   FEXCore::Allocator::ReclaimMemoryRegion(LowMem);
 }

External/FEXCore/Source/Utils/Allocator/FlexBitSet.h

@@ -1,6 +1,7 @@
 #pragma once
 
 #include <FEXCore/Utils/MathUtils.h>
+#include <FEXCore/Utils/LogManager.h>
 
 #include <cstddef>
 #include <cstdint>
@@ -38,6 +39,109 @@ struct FlexBitSet final {
     memset(Memory, 0xFF, FEXCore::AlignUp(Elements / MinimumSizeBits, MinimumSizeBits));
   }
 
+  // Range scanning results
+  struct BitsetScanResults {
+    // Which element was found. ~0ULL if not found.
+    size_t FoundElement;
+    // During the scan, found a hole in the allocations that didn't fit.
+    bool FoundHole;
+  };
+
+  // TODO: Make {Forward,Backward}ScanForRange faster
+  // Currently these functions test a single bit at a time, which is fairly costly.
+  // The compiler emits a full element load per iteration, wasting a bunch of time on loads.
+  // If we change these functions to have a pre-amble and post-amble to align the primary loop to the element size then this can go significantly
+  // faster.
+  //
+  // Once the element scanning is aligned to the element size, we can then use native count leading zero(CLZ) and count trailing zero(CTZ)
+  // instructions on a full element to scan uint64_t elements per loop iteration.
+
+  // Implementation details:
+  // Template argument WantUnset
+  // Used to determine if the desired range is for set or unset ranges.
+  // Typically `WantUnset` should be true. Used for finding a unset range inside of a range will set elements.
+  //
+  // @param BeginningElement - The first element in the set to start scanning from.
+  // @param ElementCount - How many elements to find a range for fitting.
+  // @param MinimumElement - Minimum element in the set to search to
+  //
+  // @return The scan results
+  template<bool WantUnset>
+  BitsetScanResults BackwardScanForRange(size_t BeginningElement, size_t ElementCount, size_t MinimumElement) {
+    bool FoundHole {};
+    for (size_t CurrentPage = BeginningElement;
+         CurrentPage >= (MinimumElement + ElementCount);) {
+      size_t Remaining = ElementCount;
+      LOGMAN_THROW_AA_FMT(Remaining <= CurrentPage, "Scanning less than available range");
+
+      while (Remaining) {
+        if (this->Get(CurrentPage - Remaining) == WantUnset) {
+          // Has an intersecting range
+          break;
+        }
+        --Remaining;
+      }
+
+      if (Remaining) {
+        // If we found at least one Element hole then track that
+        if (Remaining != ElementCount) {
+          FoundHole = true;
+        }
+
+        // Didn't find a slab range
+        CurrentPage -= Remaining;
+      }
+      else {
+        // We have a slab range
+        return BitsetScanResults{CurrentPage - ElementCount, FoundHole};
+      }
+    }
+
+    return BitsetScanResults {~0ULL, FoundHole};
+  }
+
+  // @param BeginningElement - The first element in the set to start scanning from.
+  // @param ElementCount - How many elements to find a range for fitting.
+  // @param ElementsInSet - How many elements are in the full set.
+  //
+  // @return The scan results
+  template<bool WantUnset>
+  BitsetScanResults ForwardScanForRange(size_t BeginningElement, size_t ElementCount, size_t ElementsInSet) {
+    bool FoundHole {};
+
+    for (size_t CurrentElement = BeginningElement;
+         CurrentElement < (ElementsInSet - ElementCount);) {
+      // If we have enough free space, check if we have enough free pages that are contiguous
+      size_t Remaining = ElementCount;
+
+      LOGMAN_THROW_AA_FMT((CurrentElement + Remaining - 1) < ElementsInSet, "Scanning less than available range");
+
+      while (Remaining) {
+        if (this->Get(CurrentElement + Remaining - 1) == WantUnset) {
+          // Has an intersecting range
+          break;
+        }
+        --Remaining;
+      }
+
+      if (Remaining) {
+        // If we found at least one Element hole then track that
+        if (Remaining != ElementCount) {
+          FoundHole = true;
+        }
+
+        // Didn't find a slab range
+        CurrentElement += Remaining;
+      }
+      else {
+        // We have a slab range
+        return BitsetScanResults {CurrentElement, FoundHole};
+      }
+    }
+
+    return BitsetScanResults {~0ULL, FoundHole};
+  }
+
   // This very explicitly doesn't let you take an address
   // Is only a getter
   bool operator[](size_t Element) const {

External/FEXCore/Source/Utils/Profiler.cpp

@@ -0,0 +1,116 @@
+#include <array>
+#include <cstdint>
+#include <fcntl.h>
+#include <limits.h>
+#include <linux/magic.h>
+#include <string>
+#include <sys/stat.h>
+#include <sys/vfs.h>
+
+#include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
+
+#define BACKEND_OFF 0
+#define BACKEND_GPUVIS 1
+
+#ifdef ENABLE_FEXCORE_PROFILER
+#if FEXCORE_PROFILER_BACKEND == BACKEND_GPUVIS
+namespace FEXCore::Profiler {
+  ProfilerBlock::ProfilerBlock(std::string_view const Format)
+    : DurationBegin {GetTime()}
+    , Format {Format} {
+    }
+
+  ProfilerBlock::~ProfilerBlock() {
+    auto Duration = GetTime() - DurationBegin;
+    TraceObject(Format, Duration);
+  }
+}
+
+namespace GPUVis {
+  // ftrace FD for writing trace data.
+  // Needs to be a raw FD since we hold this open for the entire application execution.
+  static int TraceFD {-1};
+
+  // Need to search the paths to find the real trace path
+  static std::array<char const*, 2> TraceFSDirectories {
+    "/sys/kernel/tracing",
+    "/sys/kernel/debug/tracing",
+  };
+
+  static bool IsTraceFS(char const* Path) {
+    struct statfs stat;
+    if (statfs(Path, &stat)) {
+      return false;
+    }
+    return stat.f_type == TRACEFS_MAGIC;
+  }
+
+  void Init() {
+    for (auto Path : TraceFSDirectories) {
+      if (IsTraceFS(Path)) {
+        std::string FilePath = fmt::format("{}/trace_marker", Path);
+        TraceFD = open(FilePath.c_str(), O_WRONLY | O_CLOEXEC);
+        if (TraceFD != -1) {
+          // Opened TraceFD, early exit
+          break;
+        }
+      }
+    }
+  }
+
+  void Shutdown() {
+    if (TraceFD != -1) {
+      close(TraceFD);
+      TraceFD = -1;
+    }
+  }
+
+  void TraceObject(std::string_view const Format, uint64_t Duration) {
+    if (TraceFD != -1) {
+      // Print the duration as something that began negative duration ago
+      std::string Event = fmt::format("{} (lduration=-{})\n", Format, Duration);
+      write(TraceFD, Event.c_str(), Event.size());
+    }
+  }
+
+  void TraceObject(std::string_view const Format) {
+    if (TraceFD != -1) {
+      std::string Event = fmt::format("{}\n", Format);
+      write(TraceFD, Format.data(), Format.size());
+    }
+  }
+}
+#else
+#error Unknown profiler backend
+#endif
+#endif
+
+namespace FEXCore::Profiler {
+#ifdef ENABLE_FEXCORE_PROFILER
+  void Init() {
+#if FEXCORE_PROFILER_BACKEND == BACKEND_GPUVIS
+    GPUVis::Init();
+#endif
+  }
+
+  void Shutdown() {
+#if FEXCORE_PROFILER_BACKEND == BACKEND_GPUVIS
+    GPUVis::Shutdown();
+#endif
+  }
+
+  void TraceObject(std::string_view const Format, uint64_t Duration) {
+#if FEXCORE_PROFILER_BACKEND == BACKEND_GPUVIS
+    GPUVis::TraceObject(Format, Duration);
+#endif
+  }
+
+  void TraceObject(std::string_view const Format) {
+#if FEXCORE_PROFILER_BACKEND == BACKEND_GPUVIS
+    GPUVis::TraceObject(Format);
+#endif
+
+  }
+#endif
+}

External/FEXCore/include/FEXCore/Config/Config.h

@@ -74,7 +74,9 @@ namespace Handler {
     LAYER_GLOBAL_MAIN, ///< /usr/share/fex-emu/Config.json by default
     LAYER_MAIN,
     LAYER_ARGUMENTS,
+    LAYER_GLOBAL_STEAM_APP,
     LAYER_GLOBAL_APP,
+    LAYER_LOCAL_STEAM_APP,
     LAYER_LOCAL_APP,
     LAYER_ENVIRONMENT,
     LAYER_TOP,
@@ -272,7 +274,7 @@ namespace Type {
    *
    * @return unique_ptr for that layer
    */
-  FEX_DEFAULT_VISIBILITY std::unique_ptr<FEXCore::Config::Layer> CreateAppLayer(const std::string& Filename, bool Global);
+  FEX_DEFAULT_VISIBILITY std::unique_ptr<FEXCore::Config::Layer> CreateAppLayer(const std::string& Filename, FEXCore::Config::LayerType Type);
 
   /**
    * @brief iCreate an environment configuration loader

External/FEXCore/include/FEXCore/Core/CoreState.h

@@ -28,9 +28,16 @@ namespace FEXCore::Core {
 
     uint64_t rip; ///< Current core's RIP. May not be entirely accurate while JIT is active
     uint64_t gregs[16];
-    uint16_t es, cs, ss, ds;
-    uint64_t gs;
-    uint64_t fs;
+    // Raw segment register indexes
+    uint16_t es_idx, cs_idx, ss_idx, ds_idx;
+    uint16_t gs_idx, fs_idx;
+    uint16_t _pad[2];
+
+    // Segment registers holding base addresses
+    uint32_t es_cached, cs_cached, ss_cached, ds_cached;
+    uint64_t gs_cached;
+    uint64_t fs_cached;
+    uint64_t _pad2[1];
     XMMRegs xmm;
     uint8_t flags[48];
     uint64_t mm[8][2];
@@ -211,12 +218,13 @@ namespace FEXCore::Core {
 
     uint32_t SignalHandlerRefCounter{};
 
-    struct SynchronousFaultDataStruct {
+    struct alignas(8) SynchronousFaultDataStruct {
       bool FaultToTopAndGeneratedException{};
       uint8_t Signal;
-      uint32_t TrapNo;
-      uint32_t err_code;
-      uint32_t si_code;
+      uint8_t TrapNo;
+      uint8_t si_code;
+      uint16_t err_code;
+      uint32_t _pad : 16;
     } SynchronousFaultData;
 
     InternalThreadState* Thread;
@@ -230,6 +238,9 @@ namespace FEXCore::Core {
   static_assert(offsetof(CpuStateFrame, Pointers) + sizeof(CpuStateFrame::Pointers) <= 32760, "JITPointers maximum pointer needs to be less than architecture maximum 32768");
 
   static_assert(std::is_standard_layout<CpuStateFrame>::value, "This needs to be standard layout");
+  static_assert(sizeof(CpuStateFrame::SynchronousFaultData) == 8, "This needs to be 8 bytes");
+  static_assert(std::alignment_of_v<CpuStateFrame::SynchronousFaultDataStruct> == 8, "This needs to be 8 bytes");
+  static_assert(offsetof(CpuStateFrame, SynchronousFaultData) % 8 == 0, "This needs to be aligned");
 
   FEX_DEFAULT_VISIBILITY std::string_view const& GetFlagName(unsigned Flag);
   FEX_DEFAULT_VISIBILITY std::string_view const& GetGRegName(unsigned Reg);

External/FEXCore/include/FEXCore/Core/HostFeatures.h

@@ -27,6 +27,7 @@ class HostFeatures final {
     bool SupportsSHA{};
     bool SupportsBMI1{};
     bool SupportsBMI2{};
+    bool SupportsPMULL_128Bit{};
 
     // Float exception behaviour
     bool SupportsFlushInputsToZero{};

External/FEXCore/include/FEXCore/Debug/InternalThreadState.h

@@ -8,8 +8,8 @@
 #include <FEXCore/Utils/InterruptableConditionVariable.h>
 #include <FEXCore/Utils/Threads.h>
 
-#include <map>
-#include <unordered_map>
+#include <tsl/robin_map.h>
+
 #include <shared_mutex>
 
 namespace FEXCore {
@@ -101,7 +101,7 @@ namespace FEXCore::Core {
     std::unique_ptr<FEXCore::CPU::CPUBackend> CPUBackend;
     std::unique_ptr<FEXCore::LookupCache> LookupCache;
 
-    std::unordered_map<uint64_t, LocalIREntry> DebugStore;
+    tsl::robin_map<uint64_t, LocalIREntry> DebugStore;
 
     std::unique_ptr<FEXCore::Frontend::Decoder> FrontendDecoder;
     std::unique_ptr<FEXCore::IR::PassManager> PassManager;
@@ -115,7 +115,7 @@ namespace FEXCore::Core {
 
     std::shared_mutex ObjectCacheRefCounter{};
     bool DestroyedByParent{false};  // Should the parent destroy this thread, or it destory itself
-    
+
     alignas(16) FEXCore::Core::CpuStateFrame BaseFrameState{};
 
   };

External/FEXCore/include/FEXCore/IR/IREmitter.h

@@ -1,4 +1,5 @@
 #pragma once
+#include <FEXCore/Core/CoreState.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/IR/IR.h>
 

External/FEXCore/include/FEXCore/IR/IntrusiveIRList.h

@@ -280,6 +280,11 @@ public:
     return Wrapper.GetNode(GetListData());
   }
 
+  ///< Gets an OrderedNode from the IRListView as an OrderedNodeWrapper.
+  [[nodiscard]] OrderedNodeWrapper WrapNode(OrderedNode *Node) const {
+    return Node->Wrapped(GetListData());
+  }
+
 private:
   struct BlockRange {
     using iterator = NodeIterator;