Docs: Update for release FEX-2301

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

.github/workflows/ccpp.yml

@@ -29,6 +29,19 @@ jobs:
     - name: Set runner label
       run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
 
+    - name: Set rootfs paths
+      run: |
+        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+
+    - name: Update RootFS cache
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      run: $GITHUB_WORKSPACE/Scripts/CI_FetchRootFS.py
+
     - name : submodule checkout
       # Need to update submodules
       run: |
@@ -51,7 +64,7 @@ jobs:
       # Note the current convention is to use the -S and -B options here to specify source
       # and build directories, but this is only available with CMake 3.13 and higher.
       # The CMake binaries on the Github Actions machines are (as of this writing) 3.12
-      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DENABLE_INTERPRETER=True
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True -DENABLE_INTERPRETER=True -DBUILD_FEX_LINUX_TESTS=True -DBUILD_THUNKS=True -DCMAKE_INSTALL_PREFIX=${{runner.workspace}}/build/install
 
     - name: Build
       working-directory: ${{runner.workspace}}/build
@@ -153,6 +166,62 @@ jobs:
       working-directory: ${{runner.workspace}}/build
       run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_APITests.log || true
 
+    - name: FEXLinuxTests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target fex_linux_tests_all
+
+    - name: FEXLinuxTests Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_FEXLinuxTests.log || true
+
+    - name: Thunkgen tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target thunkgen_tests
+
+    - name: Thunkgen Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ThunkgenTests.log || true
+
+    - name: Install
+      if: matrix.arch[1] == 'x64'
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target install
+
+    - name: Test GL No-Thunks
+      if: matrix.arch[1] == 'x64'
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      env:
+        DISPLAY: ":0"
+      run: cmake --build . --config $BUILD_TYPE --target thunk_functional_tests_nothunks
+
+    - name: No thunks Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_NoThunkResults.log || true
+
+    - name: Test GL Thunks
+      if: matrix.arch[1] == 'x64'
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      env:
+        DISPLAY: ":0"
+      run: cmake --build . --config $BUILD_TYPE --target thunk_functional_tests_thunks
+
+    - name: Thunks Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ThunkResults.log || true
+
     - name: Truncate test results
       if: ${{ always() }}
       shell: bash
@@ -172,4 +241,3 @@ jobs:
         name: Results-${{ env.runner_name }}
         path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
         retention-days: 3
-

.github/workflows/vixl_simulator.yml

@@ -0,0 +1,118 @@
+name: Vixl Simulator run
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+env:
+  # Customize the CMake build type here (Release, Debug, RelWithDebInfo, etc.)
+  BUILD_TYPE: Release
+  CC: clang
+  CXX: clang++
+
+jobs:
+  build:
+    runs-on: ${{ matrix.arch }}
+    strategy:
+      matrix:
+        # Only the x86-64 runner is fast enough to run this
+        arch: [[self-hosted, x64], [self-hosted, ARMv8.4]]
+      fail-fast: false
+
+    steps:
+    - uses: actions/checkout@v2
+
+    - name: Set runner label
+      run: echo "runner_label=${{ matrix.arch[1] }}" >> $GITHUB_ENV
+
+    - name: Set rootfs paths
+      run: |
+        echo "FEX_ROOTFS_MOUNT=/mnt/AutoNFS/rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS_PATH=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "FEX_ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+        echo "ROOTFS=$HOME/Rootfs/" >> $GITHUB_ENV
+
+    - name: Update RootFS cache
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      run: $GITHUB_WORKSPACE/Scripts/CI_FetchRootFS.py
+
+    - name : submodule checkout
+      # Need to update submodules
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1
+
+    - name: Clean Build Environment
+      run: rm -Rf ${{runner.workspace}}/build
+
+    - name: Create Build Environment
+      # Some projects don't allow in-source building, so create a separate build directory
+      # We'll use this as our working directory for all subsequent commands
+      run: cmake -E make_directory ${{runner.workspace}}/build
+
+    - name: Configure CMake
+      # Use a bash shell so we can use the same syntax for environment variable
+      # access regardless of the host operating system
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      # Note the current convention is to use the -S and -B options here to specify source
+      # and build directories, but this is only available with CMake 3.13 and higher.
+      # The CMake binaries on the Github Actions machines are (as of this writing) 3.12
+      run: cmake $GITHUB_WORKSPACE -DCMAKE_BUILD_TYPE=$BUILD_TYPE -G Ninja -DENABLE_VIXL_SIMULATOR=True -DENABLE_LTO=False -DENABLE_ASSERTIONS=True -DENABLE_X86_HOST_DEBUG=True
+
+    - name: Build
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the build.  You can specify a specific target with "--target <NAME>"
+      run: cmake --build . --config $BUILD_TYPE
+
+    - name: ASM Tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the unit tests
+      run: cmake --build . --config $BUILD_TYPE --target asm_tests
+
+    - name: ASM Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_ASM.log || true
+
+    - name: IR Tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      # Execute the unit tests
+      run: cmake --build . --config $BUILD_TYPE --target ir_tests
+
+    - name: IR Test Results move
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_IR.log || true
+
+    - name: Truncate test results
+      if: ${{ always() }}
+      shell: bash
+      working-directory: ${{runner.workspace}}/build
+      # Cap out the log files at 20M in case something crash spins and dumps fault text
+      # ASM tests get quite close to 10MB
+      run: truncate --size=<20M ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log || true
+
+    - name: Set runner name
+      if: ${{ always() }}
+      run: echo "runner_name=$(hostname)" >> $GITHUB_ENV
+
+    - name: Upload results
+      if: ${{ always() }}
+      uses: 'actions/upload-artifact@v2'
+      with:
+        name: Results-${{ env.runner_name }}
+        path: ${{runner.workspace}}/build/Testing/Temporary/LastTest_*.log
+        retention-days: 3
+

.gitignore

@@ -10,3 +10,4 @@ out/
 .vscode/
 .vs/
 *.pyc
+.cache

.gitmodules

@@ -49,3 +49,7 @@
 	shallow = true
 	path = External/robin-map
 	url = https://github.com/Tessil/robin-map.git
+[submodule "External/Vulkan-Headers"]
+	shallow = true
+	path = External/Vulkan-Headers
+	url = https://github.com/KhronosGroup/Vulkan-Headers.git

CI/GLThunks.json

@@ -0,0 +1,5 @@
+{
+  "ThunksDB": {
+    "GL": 1
+  }
+}

CI/VulkanThunks.json

@@ -0,0 +1,5 @@
+{
+  "ThunksDB": {
+    "Vulkan": 1
+  }
+}

CMakeFiles/cmake_uninstall.cmake.in

@@ -0,0 +1,21 @@
+if(NOT EXISTS "@CMAKE_BINARY_DIR@/install_manifest.txt")
+  message(FATAL_ERROR "Cannot find install manifest: @CMAKE_BINARY_DIR@/install_manifest.txt")
+endif()
+
+file(READ "@CMAKE_BINARY_DIR@/install_manifest.txt" files)
+string(REGEX REPLACE "\n" ";" files "${files}")
+foreach(file ${files})
+  message(STATUS "Uninstalling $ENV{DESTDIR}${file}")
+  if(IS_SYMLINK "$ENV{DESTDIR}${file}" OR EXISTS "$ENV{DESTDIR}${file}")
+    exec_program(
+      "@CMAKE_COMMAND@" ARGS "-E remove \"$ENV{DESTDIR}${file}\""
+      OUTPUT_VARIABLE rm_out
+      RETURN_VALUE rm_retval
+      )
+    if(NOT "${rm_retval}" STREQUAL 0)
+      message(FATAL_ERROR "Problem when removing $ENV{DESTDIR}${file}")
+    endif()
+  else(IS_SYMLINK "$ENV{DESTDIR}${file}" OR EXISTS "$ENV{DESTDIR}${file}")
+    message(STATUS "File $ENV{DESTDIR}${file} does not exist.")
+  endif()
+endforeach()

CMakeLists.txt

@@ -1,8 +1,13 @@
 cmake_minimum_required(VERSION 3.14)
 project(FEX)
 
+INCLUDE (CheckIncludeFiles)
+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(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)
@@ -12,10 +17,10 @@ option(ENABLE_MOLD "Enable linking with mold" FALSE)
 option(ENABLE_ASAN "Enables Clang ASAN" FALSE)
 option(ENABLE_TSAN "Enables Clang TSAN" FALSE)
 option(ENABLE_ASSERTIONS "Enables assertions in build" FALSE)
+option(ENABLE_GDB_SYMBOLS "Enables GDBSymbols integration support" ${HAVE_GDB_JIT_READER_H})
 option(ENABLE_VISUAL_DEBUGGER "Enables the visual debugger for compiling" FALSE)
 option(ENABLE_STRICT_WERROR "Enables stricter -Werror for CI" FALSE)
 option(ENABLE_WERROR "Enables -Werror" FALSE)
-option(ENABLE_STATIC_PIE "Enables static-pie build" FALSE)
 option(ENABLE_JEMALLOC "Enables jemalloc allocator" TRUE)
 option(ENABLE_OFFLINE_TELEMETRY "Enables FEX offline telemetry" TRUE)
 option(ENABLE_COMPILE_TIME_TRACE "Enables time trace compile option" FALSE)
@@ -23,11 +28,37 @@ option(ENABLE_LIBCXX "Enables LLVM libc++" FALSE)
 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_C_COMPILER "x86_64-linux-gnu-gcc" CACHE STRING "c compiler for compiling x86 guest libs")
-set (X86_CXX_COMPILER "x86_64-linux-gnu-g++" CACHE STRING "c++ compiler for compiling x86 guest libs")
+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 ${FEXCORE_PROFILER_BACKEND}")
+  endif()
+endif()
+
+# uninstall target
+if(NOT TARGET uninstall)
+  configure_file(
+    "${CMAKE_CURRENT_SOURCE_DIR}/CMakeFiles/cmake_uninstall.cmake.in"
+    "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/cmake_uninstall.cmake"
+    IMMEDIATE @ONLY)
+
+  add_custom_target(uninstall
+    COMMAND ${CMAKE_COMMAND} -P ${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/cmake_uninstall.cmake)
+endif()
+
 # These options are meant for package management
 set (TUNE_CPU "native" CACHE STRING "Override the CPU the build is tuned for")
 set (TUNE_ARCH "generic" CACHE STRING "Override the Arch the build is tuned for")
@@ -43,6 +74,12 @@ if (ENABLE_ASSERTIONS)
   add_definitions(-DASSERTIONS_ENABLED=1)
 endif()
 
+if (ENABLE_GDB_SYMBOLS)
+  message(STATUS "GDBSymbols support enabled")
+  add_definitions(-DGDB_SYMBOLS_ENABLED=1)
+endif()
+
+
 if (ENABLE_INTERPRETER)
   message(STATUS "Interpreter enabled")
   add_definitions(-DINTERPRETER_ENABLED=1)
@@ -77,7 +114,7 @@ if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
   set (CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcx16")
 endif()
 
-if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
+if (CMAKE_SYSTEM_PROCESSOR MATCHES "^aarch64|^arm64|^armv8\.*")
   set(_M_ARM_64 1)
   add_definitions(-D_M_ARM_64=1)
 endif()
@@ -129,130 +166,6 @@ if(DEFINED ENV{TERMUX_VERSION} OR ENABLE_TERMUX_BUILD)
   set(ENABLE_JEMALLOC FALSE)
 endif()
 
-if (ENABLE_STATIC_PIE)
-  if (_M_ARM_64 AND ENABLE_LLD)
-    message (FATAL_ERROR "Static linking does not currently work with AArch64+lld. Use GNU ld for now.")
-  endif()
-
-  file(WRITE ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/Determine_iplt.c
-    "int main(int argc, char* argv[])
-      {
-        return 0;
-      }")
-
-  # Compile the test application with our LD_OVERRIDE and static-pie options
-  try_compile(
-    COMPILE_RESULT
-    ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp
-    ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/Determine_iplt.c
-    COMPILE_DEFINITIONS "-fPIE ${LD_OVERRIDE}"
-    LINK_LIBRARIES "-static-pie ${LD_OVERRIDE}"
-    COPY_FILE ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/Determine_iplt
-  )
-
-  if (${COMPILE_RESULT})
-    # Read the symbols from the elf
-    execute_process(COMMAND
-      readelf -s ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/Determine_iplt
-      OUTPUT_FILE ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/plt_out.txt
-      OUTPUT_VARIABLE PLT_SYMBOLS)
-
-    # Pull out the __rela_iplt_{start,end} symbols if they exist
-    execute_process(COMMAND
-      "grep" "__rela_iplt" ${PROJECT_BINARY_DIR}/CMakeFiles/CMakeTmp/plt_out.txt
-      OUTPUT_VARIABLE PLT_SYMBOLS)
-
-    set (SYMBOLS_FINE TRUE)
-    set (HAS_IPLT -1)
-    # Check if we have any symbols in our grep output
-    # The symbols must either not exist at all OR the symbols are zero
-    if (PLT_SYMBOLS)
-      string(FIND ${PLT_SYMBOLS} "__rela_iplt_start" HAS_IPLT)
-    endif()
-
-    if (NOT HAS_IPLT EQUAL -1)
-      # We have some symbols from readelf. Let's parse the results to check if they are zero
-      # Format: '35: 0000000000000000     0 NOTYPE  LOCAL  HIDDEN   UND __rela_iplt_start'
-      string(REPLACE "\n" ";" SYMBOL_LIST ${PLT_SYMBOLS})
-      foreach (SYMBOL ${SYMBOL_LIST})
-        # strip any leading and trailing whitespace
-        string (STRIP ${SYMBOL} SYMBOL)
-        # Convert string to a list
-        string(REPLACE " " ";" SYMBOL_VALUES ${SYMBOL}})
-        # Pull out the address argument
-        list(GET SYMBOL_VALUES 1 OFFSET)
-
-        # Check against integer zero
-        if (NOT ${OFFSET} EQUAL 0)
-          # Symbol wasn't zero, this now fails
-          set (SYMBOLS_FINE FALSE)
-        endif()
-      endforeach()
-    endif()
-
-    if (SYMBOLS_FINE)
-      # We can now exnable static-pie
-      set (STATIC_PIE_OPTIONS "-static-pie")
-      # Pthreads has an issue with exposing symbols
-      # We need to make some concessions to the pthread gods
-      if (ENABLE_LLD)
-        set (PTHREAD_LIB
-          -Wl,--undefined-glob=pthread_*
-          -Wl,--undefined=__cxa_finalize
-          -Wl,--undefined=_pthread_cleanup_push_defer
-          -Wl,--undefined=_pthread_cleanup_pop_restore
-          -Wl,--undefined=__pthread_cleanup_upto
-          pthread)
-      else()
-        set (PTHREAD_LIB
-          -Wl,--undefined=pthread_join
-          -Wl,--undefined=pthread_attr_getdetachstate
-          -Wl,--undefined=pthread_sigmask
-          -Wl,--undefined=pthread_mutex_lock
-          -Wl,--undefined=pthread_cond_init
-          -Wl,--undefined=pthread_attr_init
-          -Wl,--undefined=pthread_mutex_unlock
-          -Wl,--undefined=pthread_mutexattr_destroy
-          -Wl,--undefined=pthread_detach
-          -Wl,--undefined=pthread_mutex_init
-          -Wl,--undefined=pthread_getattr_np
-          -Wl,--undefined=pthread_cond_timedwait
-          -Wl,--undefined=pthread_attr_destroy
-          -Wl,--undefined=pthread_mutexattr_settype
-          -Wl,--undefined=pthread_rwlock_unlock
-          -Wl,--undefined=pthread_rwlock_wrlock
-          -Wl,--undefined=pthread_setspecific
-          -Wl,--undefined=pthread_create
-          -Wl,--undefined=pthread_cond_clockwait
-          -Wl,--undefined=pthread_key_create
-          -Wl,--undefined=pthread_rwlock_rdlock
-          -Wl,--undefined=pthread_setname_np
-          -Wl,--undefined=pthread_cond_signal
-          -Wl,--undefined=pthread_mutexattr_init
-          -Wl,--undefined=pthread_attr_setstack
-          -Wl,--undefined=pthread_self
-          -Wl,--undefined=pthread_getaffinity_np
-          -Wl,--undefined=pthread_cond_wait
-          -Wl,--undefined=pthread_mutex_trylock
-          -Wl,--undefined=pthread_cond_broadcast
-          -Wl,--undefined=pthread_cond_destroy
-          -Wl,--undefined=pthread_getspecific
-          -Wl,--undefined=pthread_key_delete
-          -Wl,--undefined=pthread_once
-          -Wl,--undefined=__cxa_finalize
-          -Wl,--undefined=_pthread_cleanup_push_defer
-          -Wl,--undefined=_pthread_cleanup_pop_restore
-          -Wl,--undefined=__pthread_cleanup_upto
-          pthread)
-      endif()
-    else()
-      message (FATAL_ERROR "Application has __rela_iplt_{start,end} symbols. Which means static-pie can't be enabled")
-    endif()
-  else()
-    message (FATAL_ERROR "Couldn't compile static-pie test. Static-pie can't be enabled! Is your glibc compiled without static-pie?")
-  endif()
-endif()
-
 if (ENABLE_ASAN)
   add_definitions(-DENABLE_ASAN=1)
   add_compile_options(-fno-omit-frame-pointer -fsanitize=address -fsanitize-address-use-after-scope)
@@ -477,8 +390,6 @@ if (BUILD_TESTS)
 endif()
 
 add_subdirectory(FEXHeaderUtils/)
-include_directories(FEXHeaderUtils/)
-
 add_subdirectory(External/FEXCore)
 
 # Binfmt_misc files must be installed prior to Source/ installs
@@ -488,15 +399,20 @@ add_subdirectory(Source/)
 add_subdirectory(Data/AppConfig/)
 
 # Install the ThunksDB file
-install(
-  FILES ${CMAKE_CURRENT_SOURCE_DIR}/Data/ThunksDB.json
-  DESTINATION ${DATA_DIRECTORY}/)
+file(GLOB CONFIG_SOURCES CONFIGURE_DEPENDS ${CMAKE_CURRENT_SOURCE_DIR}/Data/*.json)
+
+# Any application configuration json file gets installed
+foreach(CONFIG_SRC ${CONFIG_SOURCES})
+  install(FILES ${CONFIG_SRC}
+    DESTINATION ${DATA_DIRECTORY}/)
+endforeach()
 
 if (BUILD_TESTS)
   add_subdirectory(unittests/)
 endif()
 
 if (BUILD_THUNKS)
+  set (FEX_PROJECT_SOURCE_DIR ${PROJECT_SOURCE_DIR})
   add_subdirectory(ThunkLibs/Generator)
 
   # Thunk targets for both host libraries and IDE integration
@@ -512,11 +428,31 @@ if (BUILD_THUNKS)
     SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/ThunkLibs/GuestLibs"
     BINARY_DIR "Guest"
     CMAKE_ARGS
+      "-DBITNESS=64"
       "-DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}"
-      "-DX86_C_COMPILER:STRING=${X86_C_COMPILER}"
-      "-DX86_CXX_COMPILER:STRING=${X86_CXX_COMPILER}"
+      "-DENABLE_CLANG_THUNKS=${ENABLE_CLANG_THUNKS}"
+      "-DCMAKE_TOOLCHAIN_FILE:FILEPATH=${X86_64_TOOLCHAIN_FILE}"
       "-DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}"
       "-DSTRUCT_VERIFIER=${CMAKE_SOURCE_DIR}/Scripts/StructPackVerifier.py"
+      "-DFEX_PROJECT_SOURCE_DIR=${FEX_PROJECT_SOURCE_DIR}"
+      "-DGENERATOR_EXE=$<TARGET_FILE:thunkgen>"
+    INSTALL_COMMAND ""
+    BUILD_ALWAYS ON
+    DEPENDS thunkgen
+  )
+
+  ExternalProject_Add(guest-libs-32
+    PREFIX guest-libs-32
+    SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/ThunkLibs/GuestLibs"
+    BINARY_DIR "Guest_32"
+    CMAKE_ARGS
+      "-DBITNESS=32"
+      "-DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}"
+      "-DENABLE_CLANG_THUNKS=${ENABLE_CLANG_THUNKS}"
+      "-DCMAKE_TOOLCHAIN_FILE:FILEPATH=${X86_32_TOOLCHAIN_FILE}"
+      "-DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}"
+      "-DSTRUCT_VERIFIER=${CMAKE_SOURCE_DIR}/Scripts/StructPackVerifier.py"
+      "-DFEX_PROJECT_SOURCE_DIR=${FEX_PROJECT_SOURCE_DIR}"
       "-DGENERATOR_EXE=$<TARGET_FILE:thunkgen>"
     INSTALL_COMMAND ""
     BUILD_ALWAYS ON
@@ -531,6 +467,28 @@ if (BUILD_THUNKS)
     )"
     DEPENDS guest-libs
   )
+
+  install(
+    CODE "MESSAGE(\"-- Installing: guest-libs-32\")"
+    CODE "
+    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target install
+    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest_32
+    )"
+    DEPENDS guest-libs-32
+  )
+
+  add_custom_target(uninstall_guest-libs
+    COMMAND ${CMAKE_COMMAND} "--build" "." "--target" "uninstall"
+    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest
+  )
+
+  add_custom_target(uninstall_guest-libs-32
+    COMMAND ${CMAKE_COMMAND} "--build" "." "--target" "uninstall"
+    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest_32
+  )
+
+  add_dependencies(uninstall uninstall_guest-libs)
+  add_dependencies(uninstall uninstall_guest-libs-32)
 endif()
 
 set(FEX_VERSION_MAJOR "0")
@@ -583,15 +541,9 @@ endif()
 
 # Package creation
 set (CPACK_GENERATOR "DEB")
-if (ENABLE_STATIC_PIE)
-  set (CPACK_PACKAGE_NAME fex-emu-static)
-  set (CPACK_DEBIAN_PACKAGE_CONFLICTS "fex-emu")
-else()
-  set (CPACK_PACKAGE_NAME fex-emu)
-  set (CPACK_DEBIAN_PACKAGE_CONFLICTS "fex-emu-static")
-endif()
+set (CPACK_PACKAGE_NAME fex-emu)
 set (CPACK_PACKAGE_FILE_NAME "${CPACK_PACKAGE_NAME}-${GIT_DESCRIBE_STRING}_${CMAKE_SYSTEM_PROCESSOR}")
-set (CPACK_PACKAGE_CONTACT "FEX-Emu Maintainers <team@fex-emu.org>")
+set (CPACK_PACKAGE_CONTACT "FEX-Emu Maintainers <team@fex-emu.com>")
 set (CPACK_PACKAGE_VERSION_MAJOR "${FEX_VERSION_MAJOR}")
 set (CPACK_PACKAGE_VERSION_MINOR "${FEX_VERSION_MINOR}")
 set (CPACK_PACKAGE_VERSION_PATCH "${FEX_VERSION_PATCH}")

CODE_OF_CONDUCT.md

@@ -55,7 +55,7 @@ further defined and clarified by project maintainers.
 ## Enforcement
 
 Instances of abusive, harassing, or otherwise unacceptable behavior may be
-reported by contacting the project team at team@fex-emu.org. All
+reported by contacting the project team at team@fex-emu.com. All
 complaints will be reviewed and investigated and will result in a response that
 is deemed necessary and appropriate to the circumstances. The project team is
 obligated to maintain confidentiality with regard to the reporter of an incident.

Data/AppConfig/CMakeLists.txt

@@ -11,15 +11,15 @@ endforeach()
 # First generate then install it
 foreach(GEN_CONFIG_SRC ${GEN_CONFIG_SOURCES})
   # Get the filename only component
-  get_filename_component(CONFIG_NAME ${GEN_CONFIG_SRC} NAME_WE)
+  get_filename_component(CONFIG_NAME ${GEN_CONFIG_SRC} NAME_WLE)
 
   # Configure it
   configure_file(
     ${GEN_CONFIG_SRC}
-    ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME}.json)
+    ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME})
 
   # Then install the configured json
   install(
-    FILES ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME}.json
+    FILES ${CMAKE_BINARY_DIR}/Data/AppConfig/${CONFIG_NAME}
     DESTINATION ${DATA_DIRECTORY}/AppConfig/)
 endforeach()

Data/AppConfig/VC_redist.x64.exe.json

@@ -0,0 +1,5 @@
+{
+  "Config": {
+    "x86dec_SynchronizeRIPOnAllBlocks": "1"
+  }
+}

Data/AppConfig/VC_redist.x86.exe.json

@@ -0,0 +1,5 @@
+{
+  "Config": {
+    "x86dec_SynchronizeRIPOnAllBlocks": "1"
+  }
+}

Data/AppConfig/vcredist_x64.exe.json

@@ -0,0 +1,5 @@
+{
+  "Config": {
+    "x86dec_SynchronizeRIPOnAllBlocks": "1"
+  }
+}

Data/AppConfig/vcredist_x86.exe.json

@@ -0,0 +1,5 @@
+{
+  "Config": {
+    "x86dec_SynchronizeRIPOnAllBlocks": "1"
+  }
+}

Data/ThunksDB.json

@@ -6,18 +6,10 @@
         "X11"
       ],
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libGL.so",
-        "/usr/lib/x86_64-linux-gnu/libGL.so.1",
-        "/usr/lib/x86_64-linux-gnu/libGL.so.1.2.0",
-        "/usr/lib/x86_64-linux-gnu/libGL.so.1.7.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so.1.2.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGL.so.1.7.0",
-        "/lib/x86_64-linux-gnu/libGL.so",
-        "/lib/x86_64-linux-gnu/libGL.so.1",
-        "/lib/x86_64-linux-gnu/libGL.so.1.2.0",
-        "/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": {
@@ -26,322 +18,159 @@
         "X11"
       ],
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libGLESv2.so",
-        "/usr/lib/x86_64-linux-gnu/libGLESv2.so.2",
-        "/usr/lib/x86_64-linux-gnu/libGLESv2.so.2.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGLESv2.so",
-        "/usr/local/lib/x86_64-linux-gnu/libGLESv2.so.2",
-        "/usr/local/lib/x86_64-linux-gnu/libGLESv2.so.2.0.0",
-        "/lib/x86_64-linux-gnu/libGLESv2.so",
-        "/lib/x86_64-linux-gnu/libGLESv2.so.2",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libX11.so",
-        "/usr/lib/x86_64-linux-gnu/libX11.so.6",
-        "/usr/lib/x86_64-linux-gnu/libX11.so.6.4.0",
-        "/usr/local/lib/x86_64-linux-gnu/libX11.so",
-        "/usr/local/lib/x86_64-linux-gnu/libX11.so.6",
-        "/usr/local/lib/x86_64-linux-gnu/libX11.so.6.4.0",
-        "/lib/x86_64-linux-gnu/libX11.so",
-        "/lib/x86_64-linux-gnu/libX11.so.6",
-        "/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-radeon": {
-      "Library": "libvulkan_radeon-guest.so",
+    "Vulkan": {
+      "Library": "libvulkan-guest.so",
       "Depends": [
         "xcb"
       ],
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_radeon.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_radeon.so",
-        "/lib/x86_64-linux-gnu/libvulkan_radeon.so"
+        "@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": [
         "Vulkan library relies on xcb, otherwise it crashes with jemalloc"
       ]
     },
-    "Vulkan-lavapipe": {
-      "Library": "libvulkan_lvp-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_lvp.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_lvp.so",
-        "/lib/x86_64-linux-gnu/libvulkan_lvp.so"
-      ]
-    },
-    "Vulkan-freedreno": {
-      "Library": "libvulkan_freedreno-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_freedreno.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_freedreno.so",
-        "/lib/x86_64-linux-gnu/libvulkan_freedreno.so"
-      ]
-    },
-    "Vulkan-intel": {
-      "Library": "libvulkan_intel-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_intel.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_intel.so",
-        "/lib/x86_64-linux-gnu/libvulkan_intel.so"
-      ]
-    },
-    "Vulkan-panfrost": {
-      "Library": "libvulkan_panfrost-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_panfrost.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_panfrost.so",
-        "/lib/x86_64-linux-gnu/libvulkan_panfrost.so"
-      ]
-    },
-    "Vulkan-nvidia": {
-      "Library": "libvulkan_nvidia-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libGLX_nvidia.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libGLX_nvidia.so.0",
-        "/lib/x86_64-linux-gnu/libGLX_nvidia.so.0"
-      ],
-      "Comment": [
-        "Not currently wired up"
-      ]
-    },
-    "Vulkan-virtio": {
-      "Library": "libvulkan_virtio-guest.so",
-      "Depends": [
-        "xcb"
-      ],
-      "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libvulkan_virtio.so",
-        "/usr/local/lib/x86_64-linux-gnu/libvulkan_virtio.so",
-        "/lib/x86_64-linux-gnu/libvulkan_virtio.so"
-      ]
-    },
     "xcb": {
       "Library": "libxcb-guest.so",
       "Overlay": [
-        "/usr/lib/x86_64-linux-gnu/libxcb.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb.so.1",
-        "/usr/lib/x86_64-linux-gnu/libxcb.so.1.1.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb.so.1.1.0",
-        "/lib/x86_64-linux-gnu/libxcb.so",
-        "/lib/x86_64-linux-gnu/libxcb.so.1",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri2.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri2.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri2.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri2.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri2.so",
-        "/lib/x86_64-linux-gnu/libxcb-dri2.so.0",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri3.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-dri3.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri3.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-dri3.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-dri3.so",
-        "/lib/x86_64-linux-gnu/libxcb-dri3.so.0",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-xfixes.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-xfixes.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-xfixes.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-xfixes.so",
-        "/lib/x86_64-linux-gnu/libxcb-xfixes.so.0",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-shm.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-shm.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-shm.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-shm.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-shm.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-shm.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-shm.so",
-        "/lib/x86_64-linux-gnu/libxcb-shm.so.0",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-sync.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-sync.so.1",
-        "/usr/lib/x86_64-linux-gnu/libxcb-sync.so.1.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-sync.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-sync.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-sync.so.1.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-sync.so",
-        "/lib/x86_64-linux-gnu/libxcb-sync.so.1",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-randr.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-randr.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-randr.so.0.1.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-randr.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-randr.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-randr.so.0.1.0",
-        "/lib/x86_64-linux-gnu/libxcb-randr.so",
-        "/lib/x86_64-linux-gnu/libxcb-randr.so.0",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-present.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-present.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-present.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-present.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-present.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-present.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-present.so",
-        "/lib/x86_64-linux-gnu/libxcb-present.so.0",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libxcb-glx.so",
-        "/usr/lib/x86_64-linux-gnu/libxcb-glx.so.0",
-        "/usr/lib/x86_64-linux-gnu/libxcb-glx.so.0.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-glx.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-glx.so.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxcb-glx.so.0.0.0",
-        "/lib/x86_64-linux-gnu/libxcb-glx.so",
-        "/lib/x86_64-linux-gnu/libxcb-glx.so.0",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libxshmfence.so",
-        "/usr/lib/x86_64-linux-gnu/libxshmfence.so.1",
-        "/usr/lib/x86_64-linux-gnu/libxshmfence.so.1.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libxshmfence.so",
-        "/usr/local/lib/x86_64-linux-gnu/libxshmfence.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libxshmfence.so.1.0.0",
-        "/lib/x86_64-linux-gnu/libxshmfence.so",
-        "/lib/x86_64-linux-gnu/libxshmfence.so.1",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libdrm.so",
-        "/usr/lib/x86_64-linux-gnu/libdrm.so.2",
-        "/usr/lib/x86_64-linux-gnu/libdrm.so.2.4.0",
-        "/usr/local/lib/x86_64-linux-gnu/libdrm.so",
-        "/usr/local/lib/x86_64-linux-gnu/libdrm.so.2",
-        "/usr/local/lib/x86_64-linux-gnu/libdrm.so.2.4.0",
-        "/lib/x86_64-linux-gnu/libdrm.so",
-        "/lib/x86_64-linux-gnu/libdrm.so.2",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libasound.so",
-        "/usr/lib/x86_64-linux-gnu/libasound.so.2",
-        "/usr/lib/x86_64-linux-gnu/libasound.so.2.0.0",
-        "/usr/local/lib/x86_64-linux-gnu/libasound.so",
-        "/usr/local/lib/x86_64-linux-gnu/libasound.so.2",
-        "/usr/local/lib/x86_64-linux-gnu/libasound.so.2.0.0",
-        "/lib/x86_64-linux-gnu/libasound.so",
-        "/lib/x86_64-linux-gnu/libasound.so.2",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libXrender.so",
-        "/usr/lib/x86_64-linux-gnu/libXrender.so.1",
-        "/usr/lib/x86_64-linux-gnu/libXrender.so.1.3.0",
-        "/usr/local/lib/x86_64-linux-gnu/libXrender.so",
-        "/usr/local/lib/x86_64-linux-gnu/libXrender.so.1",
-        "/usr/local/lib/x86_64-linux-gnu/libXrender.so.1.3.0",
-        "/lib/x86_64-linux-gnu/libXrender.so",
-        "/lib/x86_64-linux-gnu/libXrender.so.1",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libXext.so",
-        "/usr/lib/x86_64-linux-gnu/libXext.so.6",
-        "/usr/lib/x86_64-linux-gnu/libXext.so.6.4.0",
-        "/usr/local/lib/x86_64-linux-gnu/libXext.so",
-        "/usr/local/lib/x86_64-linux-gnu/libXext.so.6",
-        "/usr/local/lib/x86_64-linux-gnu/libXext.so.6.4.0",
-        "/lib/x86_64-linux-gnu/libXext.so",
-        "/lib/x86_64-linux-gnu/libXext.so.6",
-        "/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": [
-        "/usr/lib/x86_64-linux-gnu/libXfixes.so",
-        "/usr/lib/x86_64-linux-gnu/libXfixes.so.3",
-        "/usr/lib/x86_64-linux-gnu/libXfixes.so.3.1.0",
-        "/usr/local/lib/x86_64-linux-gnu/libXfixes.so",
-        "/usr/local/lib/x86_64-linux-gnu/libXfixes.so.3",
-        "/usr/local/lib/x86_64-linux-gnu/libXfixes.so.3.1.0",
-        "/lib/x86_64-linux-gnu/libXfixes.so",
-        "/lib/x86_64-linux-gnu/libXfixes.so.3",
-        "/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@/@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

@@ -9,12 +9,19 @@ if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcx16")
 endif()
 
-if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
+if (CMAKE_SYSTEM_PROCESSOR MATCHES "^aarch64|^arm64|^armv8\.*")
   set(_M_ARM_64 1)
 endif()
 
-set(ENABLE_JIT_X86_64 ${_M_X86_64} CACHE BOOL "Enable the x86_64 JIT")
-set(ENABLE_JIT_ARM64 ${_M_ARM_64} CACHE BOOL "Enable the ARM64 JIT")
+if (ENABLE_VIXL_SIMULATOR)
+  # If the vixl simulator is enabled then we are using the ARM64 JIT
+  option(ENABLE_JIT_X86_64 "Enable the x86_64 JIT" FALSE)
+  option(ENABLE_JIT_ARM64 "Enable the ARM64 JIT" TRUE)
+else()
+  option(ENABLE_JIT_X86_64 "Enable the x86_64 JIT" ${_M_X86_64})
+  option(ENABLE_JIT_ARM64 "Enable the ARM64 JIT" ${_M_ARM_64})
+endif()
+
 option(ENABLE_CLANG_FORMAT "Run clang format over the source" FALSE)
 
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
@@ -27,7 +34,6 @@ set(CMAKE_INCLUDE_CURRENT_DIR ON)
 include(CheckCXXCompilerFlag)
 include(CheckIncludeFileCXX)
 
-
 if (EXISTS ${CMAKE_CURRENT_DIR}/External/vixl/)
     # Useful to have for freestanding libFEXCore
     add_subdirectory(External/vixl/)
@@ -46,14 +52,14 @@ if (OVERRIDE_VERSION STREQUAL "detect")
 
   if (GIT_FOUND)
     execute_process(
-      COMMAND ${GIT_EXECUTABLE} rev-parse --short HEAD
+      COMMAND ${GIT_EXECUTABLE} rev-parse --short=7 HEAD
       WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
       OUTPUT_VARIABLE GIT_SHORT_HASH
       ERROR_QUIET
       OUTPUT_STRIP_TRAILING_WHITESPACE
     )
     execute_process(
-      COMMAND ${GIT_EXECUTABLE} describe
+      COMMAND ${GIT_EXECUTABLE} describe --abbrev=7
       WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
       OUTPUT_VARIABLE GIT_DESCRIBE_STRING
       ERROR_QUIET
@@ -76,3 +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/Scripts/json_ir_generator.py

@@ -321,8 +321,6 @@ def print_ir_structs(defines):
 
 
         if op.SSAArgNum > 0:
-            # Add helpers for accessing SSA arguments, given how frequently they're accessed
-
             output_file.write("\t// Get index of argument by name\n")
             SSAArg = 0
             for arg in op.Arguments:
@@ -330,14 +328,6 @@ def print_ir_structs(defines):
                     output_file.write("\tstatic constexpr size_t {}_Index = {};\n".format(arg.Name, SSAArg))
                     SSAArg = SSAArg + 1
 
-            output_file.write("\n")
-            output_file.write("\t[[nodiscard]] OrderedNodeWrapper& Args(size_t Index) {\n")
-            output_file.write("\t\treturn Header.Args[Index];\n")
-            output_file.write("\t}\n")
-            output_file.write("\t[[nodiscard]] const OrderedNodeWrapper& Args(size_t Index) const {\n")
-            output_file.write("\t\treturn Header.Args[Index];\n")
-            output_file.write("\t}\n")
-
 
         output_file.write("};\n")
 

External/FEXCore/Source/CMakeLists.txt

@@ -81,6 +81,7 @@ set (SRCS
   Interface/Core/LookupCache.cpp
   Interface/Core/BlockSamplingData.cpp
   Interface/Core/Core.cpp
+  Interface/Core/CPUBackend.cpp
   Interface/Core/CPUID.cpp
   Interface/Core/Frontend.cpp
   Interface/Core/GdbServer.cpp
@@ -117,6 +118,7 @@ set (SRCS
   Interface/Core/X86Tables/X87Tables.cpp
   Interface/Core/X86Tables/XOPTables.cpp
   Interface/HLE/Thunks/Thunks.cpp
+  Interface/GDBJIT/GDBJIT.cpp
   Interface/IR/AOTIR.cpp
   Interface/IR/IRDumper.cpp
   Interface/IR/IRParser.cpp
@@ -133,7 +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
@@ -141,6 +142,7 @@ set (SRCS
   Utils/NetStream.cpp
   Utils/Telemetry.cpp
   Utils/Threads.cpp
+  Utils/Profiler.cpp
   )
 
 if (ENABLE_INTERPRETER)
@@ -175,6 +177,15 @@ if (_M_ARM_64)
   list(APPEND DEFINES -D_M_ARM_64=1)
 endif()
 
+if (ENABLE_VIXL_SIMULATOR)
+  # We can run the simulator on both x86-64 or AArch64 hosts
+  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
@@ -206,10 +217,12 @@ if (ENABLE_JIT_ARM64)
     Interface/Core/JIT/Arm64/MemoryOps.cpp
     Interface/Core/JIT/Arm64/MiscOps.cpp
     Interface/Core/JIT/Arm64/MoveOps.cpp
-    Interface/Core/JIT/Arm64/VectorOps.cpp)
+    Interface/Core/JIT/Arm64/VectorOps.cpp
+    Interface/Core/JIT/Arm64/Arm64Relocations.cpp
+    )
 endif()
 
-set (LIBS vixl dl xxhash tiny-json)
+set (LIBS fmt::fmt vixl dl xxhash tiny-json FEXHeaderUtils)
 if (ENABLE_JEMALLOC)
   list (APPEND LIBS FEX_jemalloc)
 endif()
@@ -224,13 +237,11 @@ set(OUTPUT_IR_FOLDER "${CMAKE_BINARY_DIR}/include/FEXCore/IR")
 set(OUTPUT_NAME "${OUTPUT_IR_FOLDER}/IRDefines.inc")
 set(INPUT_NAME "${CMAKE_CURRENT_SOURCE_DIR}/Interface/IR/IR.json")
 
-add_custom_target(CREATE_IR_FOLDER ALL
-  COMMAND ${CMAKE_COMMAND} -E make_directory "${OUTPUT_IR_FOLDER}")
+file(MAKE_DIRECTORY "${OUTPUT_IR_FOLDER}")
 
 add_custom_command(
   OUTPUT "${OUTPUT_NAME}"
   DEPENDS "${INPUT_NAME}"
-  DEPENDS CREATE_IR_FOLDER
   DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py"
   COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_generator.py" "${INPUT_NAME}" "${OUTPUT_NAME}"
   )
@@ -244,7 +255,6 @@ set(OUTPUT_IR_DOC "${CMAKE_BINARY_DIR}/IR.md")
 add_custom_command(
   OUTPUT "${OUTPUT_IR_DOC}"
   DEPENDS "${INPUT_NAME}"
-  DEPENDS CREATE_IR_FOLDER
   DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py"
   COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/json_ir_doc_generator.py" "${INPUT_NAME}" "${OUTPUT_IR_DOC}"
   )
@@ -265,15 +275,13 @@ set(INPUT_CONFIG_NAME "${CMAKE_BINARY_DIR}/generated/Config/Config.json")
 set(OUTPUT_MAN_NAME "${CMAKE_BINARY_DIR}/generated/FEX.1")
 set(OUTPUT_MAN_NAME_COMPRESS "${CMAKE_BINARY_DIR}/generated/FEX.1.gz")
 
-add_custom_target(CREATE_CONFIG_FOLDER ALL
-  COMMAND ${CMAKE_COMMAND} -E make_directory "${OUTPUT_CONFIG_FOLDER}")
+file(MAKE_DIRECTORY "${OUTPUT_CONFIG_FOLDER}")
 
 add_custom_command(
   OUTPUT "${OUTPUT_CONFIG_NAME}"
   OUTPUT "${OUTPUT_CONFIG_OPTION_NAME}"
   OUTPUT "${OUTPUT_MAN_NAME}"
   DEPENDS "${INPUT_CONFIG_NAME}"
-  DEPENDS CREATE_CONFIG_FOLDER
   DEPENDS "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/config_generator.py"
   COMMAND "python3" "${CMAKE_CURRENT_SOURCE_DIR}/../Scripts/config_generator.py" "${INPUT_CONFIG_NAME}" "${OUTPUT_CONFIG_NAME}" "${OUTPUT_MAN_NAME}"
   "${OUTPUT_CONFIG_OPTION_NAME}"
@@ -360,14 +368,14 @@ endfunction()
 
 # Build FEXCore_Config static library
 add_library(FEXCore_Base STATIC ${FEXCORE_BASE_SRCS})
-target_link_libraries(FEXCore_Base fmt::fmt tiny-json)
+target_link_libraries(FEXCore_Base ${LIBS})
 AddDefaultOptionsToTarget(FEXCore_Base)
 
 function(AddObject Name Type)
   add_library(${Name} ${Type} ${SRCS})
   add_dependencies(${Name} IR_INC)
 
-  target_link_libraries(${Name} FEXCore_Base ${LIBS})
+  target_link_libraries(${Name} FEXCore_Base)
   AddDefaultOptionsToTarget(${Name})
 
   set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
@@ -375,7 +383,7 @@ endfunction()
 
 function(AddLibrary Name Type)
   add_library(${Name} ${Type} $<TARGET_OBJECTS:${PROJECT_NAME}_object>)
-  target_link_libraries(${Name} FEXCore_Base ${LIBS})
+  target_link_libraries(${Name} FEXCore_Base)
   set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
 
   AddDefaultOptionsToTarget(${Name})

External/FEXCore/Source/Common/BitSet.h

@@ -20,12 +20,12 @@ struct BitSet final {
   ElementType *Memory;
   void Allocate(size_t Elements) {
     size_t AllocateSize = AlignUp(Elements, MinimumSizeBits) / MinimumSize;
-    LOGMAN_THROW_A_FMT((AllocateSize * MinimumSize) >= Elements, "Fail");
+    LOGMAN_THROW_AA_FMT((AllocateSize * MinimumSize) >= Elements, "Fail");
     Memory = static_cast<ElementType*>(FEXCore::Allocator::malloc(AllocateSize));
   }
   void Realloc(size_t Elements) {
     size_t AllocateSize = AlignUp(Elements, MinimumSizeBits) / MinimumSize;
-    LOGMAN_THROW_A_FMT((AllocateSize * MinimumSize) >= Elements, "Fail");
+    LOGMAN_THROW_AA_FMT((AllocateSize * MinimumSize) >= Elements, "Fail");
     Memory = static_cast<ElementType*>(FEXCore::Allocator::realloc(Memory, AllocateSize));
   }
   void Free() {
@@ -64,7 +64,7 @@ struct BitSetView final {
   ElementType *Memory;
 
   void GetView(BitSet<T> &Set, uint64_t ElementOffset) {
-    LOGMAN_THROW_A_FMT((ElementOffset % MinimumSize) == 0,
+    LOGMAN_THROW_AA_FMT((ElementOffset % MinimumSize) == 0,
                        "Bitset view offset needs to be aligned to size of backing element");
     Memory = &Set.Memory[ElementOffset / MinimumSizeBits];
   }

External/FEXCore/Source/Common/JitSymbols.cpp

@@ -7,6 +7,11 @@
 
 namespace FEXCore {
   JITSymbols::JITSymbols() : fp{nullptr, std::fclose} {
+  }
+
+  JITSymbols::~JITSymbols() = default;
+
+  void JITSymbols::InitFile() {
     const auto PerfMap = fmt::format("/tmp/perf-{}.map", getpid());
 
     fp.reset(fopen(PerfMap.c_str(), "wb"));
@@ -16,8 +21,6 @@ namespace FEXCore {
     }
   }
 
-  JITSymbols::~JITSymbols() = default;
-
   void JITSymbols::Register(const void *HostAddr, uint64_t GuestAddr, uint32_t CodeSize) {
     if (!fp) return;
 

External/FEXCore/Source/Common/JitSymbols.h

@@ -11,6 +11,7 @@ public:
   JITSymbols();
   ~JITSymbols();
 
+  void InitFile();
   void Register(const void *HostAddr, uint64_t GuestAddr, uint32_t CodeSize);
   void Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name);
   void Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name, uintptr_t Offset);

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

@@ -82,7 +82,7 @@ namespace JSON {
     json_t const* ConfigList = json_getProperty(json, "Config");
 
     if (!ConfigList) {
-      LogMan::Msg::EFmt("Couldn't get config list");
+      // This is a non-error if the configuration file exists but no Config section
       return;
     }
 
@@ -154,15 +154,20 @@ namespace JSON {
     return ConfigDir;
   }
 
-  std::string GetConfigFileLocation() {
+  std::string GetConfigFileLocation(bool Global) {
     std::string ConfigFile{};
-    const char *AppConfig = getenv("FEX_APP_CONFIG");
-    if (AppConfig) {
-      // App config environment variable overwrites only the config file
-      ConfigFile = AppConfig;
+    if (Global) {
+      ConfigFile = GetConfigDirectory(true) + "Config.json";
     }
     else {
-      ConfigFile = GetConfigDirectory(false) + "Config.json";
+      const char *AppConfig = getenv("FEX_APP_CONFIG");
+      if (AppConfig) {
+        // App config environment variable overwrites only the config file
+        ConfigFile = AppConfig;
+      }
+      else {
+        ConfigFile = GetConfigDirectory(false) + "Config.json";
+      }
     }
     return ConfigFile;
   }
@@ -206,9 +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, 6> 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,
@@ -371,6 +379,22 @@ namespace JSON {
     return {};
   }
 
+
+  std::string FindContainer() {
+    // We only support pressure-vessel at the moment
+    const static std::string ContainerManager = "/run/host/container-manager";
+    if (std::filesystem::exists(ContainerManager)) {
+      std::vector<char> Manager{};
+      if (FEXCore::FileLoading::LoadFile(Manager, ContainerManager)) {
+        // Trim the whitespace, may contain a newline
+        std::string ManagerStr = Manager.data();
+        ManagerStr = FEXCore::StringUtils::Trim(ManagerStr);
+        return ManagerStr;
+      }
+    }
+    return {};
+  }
+
   std::string FindContainerPrefix() {
     // We only support pressure-vessel at the moment
     const static std::string ContainerManager = "/run/host/container-manager";
@@ -597,7 +621,7 @@ namespace JSON {
   // Application loaders
   class MainLoader final : public FEXCore::Config::OptionMapper {
   public:
-    explicit MainLoader();
+    explicit MainLoader(FEXCore::Config::LayerType Type);
     explicit MainLoader(std::string ConfigFile);
     void Load() override;
 
@@ -607,7 +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:
@@ -643,9 +667,9 @@ namespace JSON {
     }
   }
 
-  MainLoader::MainLoader()
-    : FEXCore::Config::OptionMapper(FEXCore::Config::LayerType::LAYER_MAIN)
-    , Config{FEXCore::Config::GetConfigFileLocation()} {
+  MainLoader::MainLoader(FEXCore::Config::LayerType Type)
+    : FEXCore::Config::OptionMapper(Type)
+    , Config{FEXCore::Config::GetConfigFileLocation(Type == FEXCore::Config::LayerType::LAYER_GLOBAL_MAIN)} {
   }
 
   MainLoader::MainLoader(std::string ConfigFile)
@@ -659,8 +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
@@ -719,17 +745,21 @@ namespace JSON {
     }
   }
 
+  std::unique_ptr<FEXCore::Config::Layer> CreateGlobalMainLayer() {
+    return std::make_unique<FEXCore::Config::MainLoader>(FEXCore::Config::LayerType::LAYER_GLOBAL_MAIN);
+  }
+
   std::unique_ptr<FEXCore::Config::Layer> CreateMainLayer(std::string const *File) {
     if (File) {
       return std::make_unique<FEXCore::Config::MainLoader>(*File);
     }
     else {
-      return std::make_unique<FEXCore::Config::MainLoader>();
+      return std::make_unique<FEXCore::Config::MainLoader>(FEXCore::Config::LayerType::LAYER_MAIN);
     }
   }
 
-  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": {
@@ -49,6 +50,13 @@
           "Cache JIT object code to drive.",
           "Allows JIT code to be shared between applications"
         ]
+      },
+      "EnableAVX": {
+        "Type": "bool",
+        "Default": "true",
+        "Desc": [
+          "Determines whether or not we use the expanded register file for AVX or not"
+        ]
       }
     },
     "Emulation": {
@@ -83,6 +91,20 @@
           "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/",
+        "Desc": [
+          "Folder to find the 32-bit guest-side thunking libraries."
+        ]
+      },
       "ThunkConfig": {
         "Type": "str",
         "Default": "",
@@ -208,6 +230,16 @@
           "Useful for determining hot blocks of code",
           "Has some file writing overhead per JIT block"
         ]
+      },
+      "GDBSymbols": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Integrates with GDB using the JIT interface.",
+          "Needs the fex jit loader in GDB, which can be loaded via `jit-reader-load libFEXGDBReader.so.`",
+          "Also needs x86_64-linux-gnu-objdump in PATH.",
+          "Can be very slow."
+        ]
       }
     },
     "Logging": {
@@ -219,22 +251,13 @@
           "Disables logging"
         ]
       },
-      "OutputSocket": {
-        "Type": "str",
-        "Default": "",
-        "Desc": [
-          "Socket to connect to",
-          "eg: localhost:8087",
-          "If set will override the OutputLog location"
-        ]
-      },
       "OutputLog": {
         "Type": "str",
-        "Default": "stderr",
+        "Default": "server",
         "ShortArg": "o",
         "Desc": [
           "File to write FEX output to.",
-          "[stdout, stderr, <Filename>]"
+          "[stdout, stderr, server, <Filename>]"
         ]
       }
     },
@@ -260,6 +283,14 @@
           "Highly likely to break any multithreaded application if disabled."
         ]
       },
+      "TSOAutoMigration": {
+        "Type": "bool",
+        "Default": "true",
+        "Desc": [
+          "Automatically enables TSO when shared memory is used.",
+          "Should work without issues in most cases."
+        ]
+      },
       "X87ReducedPrecision": {
         "Type": "bool",
         "Default": "false",
@@ -300,6 +331,16 @@
           "Forces a process to stall out on initialization",
           "Useful for a process that keeps restarting and doesn't work"
         ]
+      },
+      "x86dec_SynchronizeRIPOnAllBlocks": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "An application that uses try-catch or longjump extensively needs the ability to do context aware state flushing",
+          "In the case of FEX's block-linking, it won't always ensure that RIP is synchronized.",
+          "If an exception occurs and RIP isn't synchronized, then FEX's exception stack restore may not long jump as expected",
+          "Can be useful for Wine applications that rely on stack unwinding"
+        ]
       }
     },
     "Misc": {
@@ -325,6 +366,13 @@
         "Desc": [
           "Loads an AOT IR cache for the loaded executable."
         ]
+      },
+      "ServerSocketPath": {
+        "Type": "str",
+        "Default": "",
+        "Desc": [
+          "Override for a FEXServer socket path. Only useful for chroots."
+        ]
       }
     }
   },
@@ -342,6 +390,17 @@
         "Type": "str",
         "Default": ""
       },
+      "APP_CONFIG_NAME": {
+        "Type": "str",
+        "Default": "",
+        "Desc": [
+          "This is the application config name that has been loaded.",
+          "This differs from APP_FILENAME in two ways",
+          "Where APP_FILENAME always points to the executable path that FEX-Emu is executing.",
+          "This matches what is used to load the AppLayer configuration name.",
+          "When running through a compatibility layer like wine, this will only be the exe name, instead of wine full path."
+        ]
+      },
       "IS64BIT_MODE": {
         "Type": "bool",
         "Default": "false"

External/FEXCore/Source/Interface/Context/Context.cpp

@@ -43,8 +43,8 @@ namespace FEXCore::Context {
     delete CTX;
   }
 
-  FEXCore::Core::InternalThreadState* InitCore(FEXCore::Context::Context *CTX, FEXCore::CodeLoader *Loader) {
-    return CTX->InitCore(Loader);
+  FEXCore::Core::InternalThreadState* InitCore(FEXCore::Context::Context *CTX, uint64_t InitialRIP, uint64_t StackPointer) {
+    return CTX->InitCore(InitialRIP, StackPointer);
   }
 
   void SetExitHandler(FEXCore::Context::Context *CTX, ExitHandler handler) {
@@ -110,6 +110,10 @@ namespace FEXCore::Context {
   void RegisterExternalSyscallVisitor(FEXCore::Context::Context *CTX, [[maybe_unused]] uint64_t Syscall, [[maybe_unused]] FEXCore::HLE::SyscallVisitor *Visitor) {
   }
 
+  HostFeatures GetHostFeatures(const FEXCore::Context::Context *CTX) {
+    return CTX->HostFeatures;
+  }
+
   void HandleCallback(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread, uint64_t RIP) {
     CTX->HandleCallback(Thread, RIP);
   }
@@ -156,6 +160,7 @@ namespace FEXCore::Context {
 
   void SetSyscallHandler(FEXCore::Context::Context *CTX, FEXCore::HLE::SyscallHandler *Handler) {
     CTX->SyscallHandler = Handler;
+    CTX->SourcecodeResolver = Handler->GetSourcecodeResolver();
   }
 
   FEXCore::CPUID::FunctionResults RunCPUIDFunction(FEXCore::Context::Context *CTX, uint32_t Function, uint32_t Leaf) {
@@ -193,6 +198,14 @@ namespace FEXCore::Context {
     return CTX->UnloadAOTIRCacheEntry(Entry);
   }
 
+  CustomIRResult AddCustomIREntrypoint(FEXCore::Context::Context *CTX, uintptr_t Entrypoint, std::function<void(uintptr_t Entrypoint, FEXCore::IR::IREmitter *)> Handler, void *Creator, void *Data) {
+    return CTX->AddCustomIREntrypoint(Entrypoint, Handler, Creator, Data);
+  }
+
+  void AppendThunkDefinitions(FEXCore::Context::Context *CTX, std::vector<FEXCore::IR::ThunkDefinition> const& Definitions) {
+    CTX->AppendThunkDefinitions(Definitions);
+  }
+
 namespace Debug {
   void CompileRIP(FEXCore::Context::Context *CTX, uint64_t RIP) {
     CTX->CompileRIP(CTX->ParentThread, RIP);

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

@@ -1,18 +1,21 @@
 #pragma once
 
 #include "Common/JitSymbols.h"
+#include "FEXHeaderUtils/ScopedSignalMask.h"
 #include "Interface/Core/CPUID.h"
-#include "Interface/Core/HostFeatures.h"
 #include "Interface/Core/X86HelperGen.h"
 #include "Interface/Core/ObjectCache/ObjectCacheService.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/IR/AOTIR.h"
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/Context.h>
 #include <FEXCore/Core/CoreState.h>
+#include <FEXCore/Core/HostFeatures.h>
 #include <FEXCore/Core/SignalDelegator.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/CompilerDefs.h>
 #include <FEXCore/Utils/Event.h>
+#include <FEXHeaderUtils/Syscalls.h>
 #include <stdint.h>
 
 
@@ -42,10 +45,14 @@ namespace CodeSerialize {
 namespace CPU {
   class Arm64JITCore;
   class X86JITCore;
+  class InterpreterCore;
+  class Dispatcher;
 }
 namespace HLE {
 struct SyscallArguments;
 class SyscallHandler;
+class SourcecodeResolver;
+struct SourcecodeMap;
 }
 }
 
@@ -72,6 +79,7 @@ namespace FEXCore::Context {
     friend class FEXCore::CPU::X86JITCore;
   #endif
 
+    friend class FEXCore::CPU::InterpreterCore;
     friend class FEXCore::IR::Validation::IRValidation;
 
     struct {
@@ -86,6 +94,7 @@ namespace FEXCore::Context {
       FEX_CONFIG_OPT(GdbServer, GDBSERVER);
       FEX_CONFIG_OPT(Is64BitMode, IS64BIT_MODE);
       FEX_CONFIG_OPT(TSOEnabled, TSOENABLED);
+      FEX_CONFIG_OPT(TSOAutoMigration, TSOAUTOMIGRATION);
       FEX_CONFIG_OPT(ABILocalFlags, ABILOCALFLAGS);
       FEX_CONFIG_OPT(ABINoPF, ABINOPF);
       FEX_CONFIG_OPT(AOTIRCapture, AOTIRCAPTURE);
@@ -96,27 +105,28 @@ 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);
       FEX_CONFIG_OPT(GlobalJITNaming, GLOBALJITNAMING);
       FEX_CONFIG_OPT(LibraryJITNaming, LIBRARYJITNAMING);
       FEX_CONFIG_OPT(BlockJITNaming, BLOCKJITNAMING);
+      FEX_CONFIG_OPT(GDBSymbols, GDBSYMBOLS);
       FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
       FEX_CONFIG_OPT(CacheObjectCodeCompilation, CACHEOBJECTCODECOMPILATION);
       FEX_CONFIG_OPT(x87ReducedPrecision, X87REDUCEDPRECISION);
+      FEX_CONFIG_OPT(x86dec_SynchronizeRIPOnAllBlocks, X86DEC_SYNCHRONIZERIPONALLBLOCKS);
+      FEX_CONFIG_OPT(EnableAVX, ENABLEAVX);
     } Config;
 
-    using IntCallbackReturn =  FEX_NAKED void(*)(FEXCore::Core::InternalThreadState *Thread, volatile void *Host_RSP);
-    IntCallbackReturn InterpreterCallbackReturn;
-
     FEXCore::HostFeatures HostFeatures;
 
     std::mutex ThreadCreationMutex;
-    uint64_t ThreadID{};
     FEXCore::Core::InternalThreadState* ParentThread;
     std::vector<FEXCore::Core::InternalThreadState*> Threads;
     std::atomic_bool CoreShuttingDown{false};
+    bool NeedToCheckXID{true};
 
     std::mutex IdleWaitMutex;
     std::condition_variable IdleWaitCV;
@@ -125,9 +135,13 @@ namespace FEXCore::Context {
     Event PauseWait;
     bool Running{};
 
+    std::shared_mutex CodeInvalidationMutex;
+
     FEXCore::CPUIDEmu CPUID;
     FEXCore::HLE::SyscallHandler *SyscallHandler{};
+    FEXCore::HLE::SourcecodeResolver *SourcecodeResolver{};
     std::unique_ptr<FEXCore::ThunkHandler> ThunkHandler;
+    std::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
 
     CustomCPUFactoryType CustomCPUFactory;
     FEXCore::Context::ExitHandler CustomExitHandler;
@@ -142,7 +156,7 @@ namespace FEXCore::Context {
     Context();
     ~Context();
 
-    FEXCore::Core::InternalThreadState* InitCore(FEXCore::CodeLoader *Loader);
+    FEXCore::Core::InternalThreadState* InitCore(uint64_t InitialRIP, uint64_t StackPointer);
     FEXCore::Context::ExitReason RunUntilExit();
     int GetProgramStatus() const;
     bool IsPaused() const { return !Running; }
@@ -162,15 +176,33 @@ namespace FEXCore::Context {
     void RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
     void RegisterFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
 
-    // Must be called from owning thread
-    static void RemoveThreadCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
+    static void ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
+    static void ThreadAddBlockLink(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestDestination, uintptr_t HostLink, const std::function<void()> &delinker);
 
-    // Wrapper which takes CpuStateFrame instead of InternalThreadState
-    // Must be called from owning thread
-    static void RemoveThreadCodeEntryFromJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
-      RemoveThreadCodeEntry(Frame->Thread, GuestRIP);
+    template<auto Fn>
+    static uint64_t ThreadExitFunctionLink(FEXCore::Core::CpuStateFrame *Frame, uint64_t *record) {
+      FHU::ScopedSignalMaskWithSharedLock lk(Frame->Thread->CTX->CodeInvalidationMutex);
+
+      return Fn(Frame, record);
     }
 
+    // Wrapper which takes CpuStateFrame instead of InternalThreadState and unique_locks CodeInvalidationMutex
+    // Must be called from owning thread
+    static void ThreadRemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
+      auto Thread = Frame->Thread;
+      
+      LogMan::Throw::AFmt(Thread->ThreadManager.GetTID() == FHU::Syscalls::gettid(), "Must be called from owning thread {}, not {}", Thread->ThreadManager.GetTID(), FHU::Syscalls::gettid());
+
+      FHU::ScopedSignalMaskWithUniqueLock lk(Thread->CTX->CodeInvalidationMutex);
+
+      ThreadRemoveCodeEntry(Thread, GuestRIP);
+    }
+
+    // returns false if a handler was already registered
+    CustomIRResult AddCustomIREntrypoint(uintptr_t Entrypoint, std::function<void(uintptr_t Entrypoint, FEXCore::IR::IREmitter *)> Handler, void *Creator, void *Data);
+
+    void RemoveCustomIREntrypoint(uintptr_t Entrypoint);
+
     // Debugger interface
     void CompileRIP(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP);
     uint64_t GetThreadCount() const;
@@ -180,21 +212,19 @@ namespace FEXCore::Context {
 
     struct GenerateIRResult {
       FEXCore::IR::IRListView* IRList;
-      // User's responsibility to deallocate this.
-      FEXCore::IR::RegisterAllocationData* RAData;
+      FEXCore::IR::RegisterAllocationData::UniquePtr RAData;
       uint64_t TotalInstructions;
       uint64_t TotalInstructionsLength;
       uint64_t StartAddr;
       uint64_t Length;
     };
-    [[nodiscard]] GenerateIRResult GenerateIR(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
+    [[nodiscard]] GenerateIRResult GenerateIR(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP, bool ExtendedDebugInfo);
 
     struct CompileCodeResult {
       void* CompiledCode;
       FEXCore::IR::IRListView* IRData;
       FEXCore::Core::DebugData* DebugData;
-      // User's responsibility to deallocate this.
-      FEXCore::IR::RegisterAllocationData* RAData;
+      FEXCore::IR::RegisterAllocationData::UniquePtr RAData;
       bool GeneratedIR;
       uint64_t StartAddr;
       uint64_t Length;
@@ -207,7 +237,7 @@ namespace FEXCore::Context {
 
     // Used for thread creation from syscalls
     /**
-     * @brief Used to create FEX thread objects in preparation for creating a true OS thread
+     * @brief Used to create FEX thread objects in preparation for creating a true OS thread. Does set a TID or PID.
      *
      * @param NewThreadState The initial thread state to setup for our state
      * @param ParentTID The PID that was the parent thread that created this
@@ -230,7 +260,7 @@ namespace FEXCore::Context {
     FEXCore::Core::InternalThreadState* CreateThread(FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID);
 
     /**
-     * @brief Initializes the TLS data for a thread
+     * @brief Initializes TID, PID and TLS data for a thread
      *
      * @param Thread The internal FEX thread state object
      */
@@ -294,11 +324,17 @@ namespace FEXCore::Context {
       IRCaptureCache.SetAOTIRRenamer(CacheRenamer);
     }
 
+    void AppendThunkDefinitions(std::vector<FEXCore::IR::ThunkDefinition> const& Definitions);
+
     FEXCore::Utils::PooledAllocatorMMap OpDispatcherAllocator;
     FEXCore::Utils::PooledAllocatorMMap FrontendAllocator;
 
+    void MarkMemoryShared();
+
+    bool IsTSOEnabled() { return (IsMemoryShared || !Config.TSOAutoMigration) && Config.TSOEnabled; }
+
   protected:
-    void ClearCodeCache(FEXCore::Core::InternalThreadState *Thread, bool AlsoClearIRCache);
+    void ClearCodeCache(FEXCore::Core::InternalThreadState *Thread);
 
   private:
     /**
@@ -317,17 +353,15 @@ namespace FEXCore::Context {
      *
      * InitializeCompiler is called inside of CreateThread, so you likely don't need this
      */
-    void InitializeCompiler(FEXCore::Core::InternalThreadState* State);
+    void InitializeCompiler(FEXCore::Core::InternalThreadState* Thread);
 
     void WaitForIdleWithTimeout();
 
     void NotifyPause();
 
-    void AddBlockMapping(FEXCore::Core::InternalThreadState *Thread, uint64_t Address, void *Ptr, uint64_t Start, uint64_t Length);
-    FEXCore::CodeLoader *LocalLoader{};
+    void AddBlockMapping(FEXCore::Core::InternalThreadState *Thread, uint64_t Address, void *Ptr);
 
     // Entry Cache
-    uint64_t StartingRIP;
     std::mutex ExitMutex;
     std::unique_ptr<GdbServer> DebugServer;
 
@@ -335,7 +369,13 @@ namespace FEXCore::Context {
     std::unique_ptr<FEXCore::CodeSerialize::CodeObjectSerializeService> CodeObjectCacheService;
 
     bool StartPaused = false;
+    bool IsMemoryShared = false;
     FEX_CONFIG_OPT(AppFilename, APP_FILENAME);
+
+    std::shared_mutex CustomIRMutex;
+    std::unordered_map<uint64_t, std::tuple<std::function<void(uintptr_t Entrypoint, FEXCore::IR::IREmitter *)>, void *, void *>> CustomIRHandlers;
+    FEXCore::CPU::CPUBackendFeatures BackendFeatures;
+    FEXCore::CPU::DispatcherConfig DispatcherConfig;
   };
 
   uint64_t HandleSyscall(FEXCore::HLE::SyscallHandler *Handler, FEXCore::Core::CpuStateFrame *Frame, FEXCore::HLE::SyscallArguments *Args);

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

@@ -1,14 +1,14 @@
 #include "Interface/Core/ArchHelpers/Arm64.h"
 #include "Interface/Core/ArchHelpers/MContext.h"
+#include "Interface/Core/ArchHelpers/CodeEmitter/Buffer.h"
 
+#include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/Telemetry.h>
 
 #include <atomic>
-#include <stdint.h>
-
-#include <signal.h>
-#include "aarch64/cpu-aarch64.h"
+#include <csignal>
+#include <cstdint>
 
 namespace FEXCore::ArchHelpers::Arm64 {
 FEXCORE_TELEMETRY_STATIC_INIT(SplitLock, TYPE_HAS_SPLIT_LOCKS);
@@ -571,7 +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;
     }
   }
@@ -1986,7 +1986,8 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
         DesiredFunction = NEGDesired;
         break;
       default:
-        LogMan::Msg::EFmt("Unhandled JIT SIGBUS Atomic mem op 0x{:02x}", AtomicOp);
+        LogMan::Msg::EFmt("Unhandled JIT SIGBUS Atomic mem op 0x{:02x}",
+                          ToUnderlying(AtomicOp));
         return false;
     }
 
@@ -2039,7 +2040,8 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
         DesiredFunction = NEGDesired;
         break;
       default:
-        LogMan::Msg::EFmt("Unhandled JIT SIGBUS Atomic mem op 0x{:02x}", AtomicOp);
+        LogMan::Msg::EFmt("Unhandled JIT SIGBUS Atomic mem op 0x{:02x}",
+                          ToUnderlying(AtomicOp));
         return false;
     }
 
@@ -2092,7 +2094,8 @@ uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info) {
         DesiredFunction = NEGDesired;
         break;
       default:
-        LogMan::Msg::EFmt("Unhandled JIT SIGBUS Atomic mem op 0x{:02x}", AtomicOp);
+        LogMan::Msg::EFmt("Unhandled JIT SIGBUS Atomic mem op 0x{:02x}",
+                          ToUnderlying(AtomicOp));
         return false;
     }
 
@@ -2307,7 +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"
@@ -7,37 +8,37 @@
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/MathUtils.h>
 
-#include "aarch64/cpu-aarch64.h"
-#include "cpu-features.h"
-#include "aarch64/instructions-aarch64.h"
-#include "utils-vixl.h"
+#include <aarch64/cpu-aarch64.h>
+#include <aarch64/instructions-aarch64.h>
+#include <cpu-features.h>
+#include <utils-vixl.h>
 
+#include <array>
 #include <tuple>
+#include <utility>
 
 namespace FEXCore::CPU {
-#define STATE x28
 
 // 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();
-
-  auto Features = vixl::CPUFeatures::InferFromOS();
-  if (ctx->HostFeatures.SupportsAtomics) {
-    // Hypervisor can hide this on the c630?
-    Features.Combine(vixl::CPUFeatures::Feature::kLORegions);
-  }
-
-  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();
@@ -84,17 +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;
       }
     }
@@ -110,116 +111,144 @@ 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) {
-      for (size_t i = 0; i < SRAFPR.size(); i+=2) {
-        auto Reg1 = SRAFPR[i];
-        auto Reg2 = SRAFPR[i+1];
+  if (FPRs) {
+    if (EmitterCTX->HostFeatures.SupportsAVX) {
+      for (size_t i = 0; i < SRAFPR.size(); i++) {
+        const auto Reg = SRAFPR[i];
 
-        if (((1U << Reg1.GetCode()) & FPRSpillMask) &&
-            ((1U << Reg2.GetCode()) & FPRSpillMask)) {
-          stp(Reg1.Q(), Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+        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 if (((1U << Reg1.GetCode()) & FPRSpillMask)) {
-          str(Reg1.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+      }
+    } 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 if (((1U << Reg2.GetCode()) & FPRSpillMask)) {
-          str(Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i+1][0])));
+      }
+      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.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.Idx()) & FPRSpillMask)) {
+            str(Reg1.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
+          }
+          else if (((1U << Reg2.Idx()) & FPRSpillMask)) {
+            str(Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i+1][0]));
+          }
         }
       }
     }
@@ -227,99 +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])));
+  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<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.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 if (((1U << Reg1.GetCode()) & GPRFillMask)) {
-        ldr(Reg1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
+    } 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 if (((1U << Reg2.GetCode()) & GPRFillMask)) {
-        ldr(Reg2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i+1])));
+      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.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.Idx()) & FPRFillMask)) {
+            ldr(Reg1.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i][0]));
+          }
+          else if (((1U << Reg2.Idx()) & FPRFillMask)) {
+            ldr(Reg2.Q(), STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[i+1][0]));
+          }
+        }
       }
     }
+  }
 
-    if (FPRs) {
-      for (size_t i = 0; i < SRAFPR.size(); i+=2) {
-        auto Reg1 = SRAFPR[i];
-        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[i][0])));
-        }
-        else if (((1U << Reg1.GetCode()) & FPRFillMask)) {
-          ldr(Reg1.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
-        }
-        else if (((1U << Reg2.GetCode()) & FPRFillMask)) {
-          ldr(Reg2.Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[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() {
-  uint64_t SPOffset = AlignUp((RA64.size() + 1) * 8 + RAFPR.size() * 16, 16);
+void Arm64Emitter::PushDynamicRegsAndLR(FEXCore::ARMEmitter::Register TmpReg) {
+  const auto CanUseSVE = EmitterCTX->HostFeatures.SupportsAVX;
+  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);
 
-  for (auto RA : RAFPR)
-  {
-    str(RA.Q(), MemOperand(sp, i * 8));
-    i+=2;
+  // rsp capable move
+  add(ARMEmitter::Size::i64Bit, TmpReg, ARMEmitter::Reg::rsp, 0);
+
+  if (CanUseSVE) {
+    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 {
+    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 (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() {
-  uint64_t SPOffset = AlignUp((RA64.size() + 1) * 8 + RAFPR.size() * 16, 16);
-  int i = 0;
+  const auto CanUseSVE = EmitterCTX->HostFeatures.SupportsAVX;
 
-  for (auto RA : RAFPR)
-  {
-    ldr(RA.Q(), MemOperand(sp, i * 8));
-    i+=2;
-  }
-
-#if 0 // All GPRs should be caller saved
-  for (auto RA : RA64)
-  {
-    ldr(RA, MemOperand(sp, i * 8));
-    i++;
-  }
-#endif
-
-  ldr(lr, MemOperand(sp, i * 8));
-
-  add(sp, sp, SPOffset);
-}
-
-void Arm64Emitter::ResetStack() {
-  if (SpillSlots == 0)
-    return;
-
-  if (IsImmAddSub(SpillSlots * 16)) {
-    add(sp, sp, SpillSlots * 16);
+  if (CanUseSVE) {
+    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 {
-   // Too big to fit in a 12bit immediate
-   LoadConstant(x0, SpillSlots * 16);
-   add(sp, sp, x0);
+    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);
+    }
   }
+
+  ldr<ARMEmitter::IndexType::POST>(ARMEmitter::XReg::lr, ARMEmitter::Reg::rsp, 16);
 }
 
 void Arm64Emitter::Align16B() {
@@ -329,122 +396,4 @@ void Arm64Emitter::Align16B() {
   }
 }
 
-uint64_t Arm64Emitter::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
-  switch (Op) {
-    case FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER:
-      return Dispatcher->ExitFunctionLinkerAddress;
-    break;
-    default:
-      ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op));
-    break;
-  }
-  return ~0ULL;
-}
-
-void Arm64Emitter::InsertNamedThunkRelocation(vixl::aarch64::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<uint64_t>();
-  MoveABI.NamedThunkMove.Offset = CurrentCursor - GuestEntry;
-  MoveABI.NamedThunkMove.Symbol = Sum;
-  MoveABI.NamedThunkMove.RegisterIndex = Reg.GetCode();
-
-  uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Sum));
-
-  LoadConstant(Reg, Pointer, EmitterCTX->Config.CacheObjectCodeCompilation());
-  Relocations.emplace_back(MoveABI);
-}
-
-Arm64Emitter::NamedSymbolLiteralPair Arm64Emitter::InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
-  uint64_t Pointer = GetNamedSymbolLiteral(Op);
-
-  Arm64Emitter::NamedSymbolLiteralPair Lit {
-    .Lit = Literal(Pointer),
-    .MoveABI = {
-      .NamedSymbolLiteral = {
-        .Header = {
-          .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL,
-        },
-        .Symbol = Op,
-        .Offset = 0,
-      },
-    },
-  };
-  return Lit;
-}
-
-void Arm64Emitter::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit) {
-  // Offset is the offset from the entrypoint of the block
-  auto CurrentCursor = GetCursorAddress<uint64_t>();
-  Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - GuestEntry;
-
-  place(&Lit.Lit);
-  Relocations.emplace_back(Lit.MoveABI);
-}
-
-void Arm64Emitter::InsertGuestRIPMove(vixl::aarch64::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<uint64_t>();
-  MoveABI.GuestRIPMove.Offset = CurrentCursor - GuestEntry;
-  MoveABI.GuestRIPMove.GuestRIP = Constant;
-  MoveABI.GuestRIPMove.RegisterIndex = Reg.GetCode();
-
-  LoadConstant(Reg, Constant, EmitterCTX->Config.CacheObjectCodeCompilation());
-  Relocations.emplace_back(MoveABI);
-}
-
-bool Arm64Emitter::ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations) {
-  size_t DataIndex{};
-  for (size_t j = 0; j < NumRelocations; ++j) {
-    const FEXCore::CPU::Relocation *Reloc = reinterpret_cast<const FEXCore::CPU::Relocation *>(&EntryRelocations[DataIndex]);
-    LOGMAN_THROW_A_FMT((DataIndex % alignof(Relocation)) == 0, "Alignment of relocation wasn't adhered to");
-
-    switch (Reloc->Header.Type) {
-      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);
-
-        // Generate a literal so we can place it
-        Literal<uint64_t> Lit(Pointer);
-        place(&Lit);
-
-        DataIndex += sizeof(Reloc->NamedSymbolLiteral);
-        break;
-      }
-      case FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE: {
-        uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Reloc->NamedThunkMove.Symbol));
-        if (Pointer == ~0ULL) {
-          return false;
-        }
-
-        // 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);
-        DataIndex += sizeof(Reloc->NamedThunkMove);
-        break;
-      }
-      case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE: {
-        // XXX: Reenable once the JIT Object Cache is upstream
-        // XXX: Should spin the relocation list, create a list of guest RIP moves, and ask for them all once, reduces lock contention.
-        uint64_t Pointer = ~0ULL; // EmitterCTX->JITObjectCache->FindRelocatedRIP(Reloc->GuestRIPMove.GuestRIP);
-        if (Pointer == ~0ULL) {
-          return false;
-        }
-
-        // 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);
-        DataIndex += sizeof(Reloc->GuestRIPMove);
-        break;
-      }
-    }
-  }
-
-  return true;
-}
-
 }

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

@@ -1,73 +1,101 @@
 #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"
 
-#include "aarch64/assembler-aarch64.h"
-#include "aarch64/constants-aarch64.h"
-#include "aarch64/cpu-aarch64.h"
-#include "aarch64/operands-aarch64.h"
-#include "platform-vixl.h"
-#include "FEXCore/Config/Config.h"
+#include <aarch64/assembler-aarch64.h>
+#include <aarch64/constants-aarch64.h>
+#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>
+#endif
+
+#include <FEXCore/Config/Config.h>
 
 #include <array>
-#include <stddef.h>
-#include <stdint.h>
+#include <cstddef>
+#include <cstdint>
 #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
+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.
+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
+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.
+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);
-
-  std::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
+  ~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
+  //       TMP4 is left alone.
   void SpillStaticRegs(bool FPRs = true, uint32_t GPRSpillMask = ~0U, uint32_t FPRSpillMask = ~0U);
   void FillStaticRegs(bool FPRs = true, uint32_t GPRFillMask = ~0U, uint32_t FPRFillMask = ~0U);
 
@@ -77,75 +105,83 @@ 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 ResetStack();
   void Align16B();
-  /**
-   * @name Relocations
-   * @{ */
 
-    uint64_t GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+#ifdef VIXL_SIMULATOR
+  // Generates a vixl simulator runtime call.
+  //
+  // This matches behaviour of vixl's macro assembler, but we need to reimplement it since we aren't using the macro assembler.
+  // This isn't too complex with how vixl emits this.
+  //
+  // Emit:
+  // 1) hlt(kRuntimeCallOpcode)
+  // 2) Simulator wrapper handler
+  // 3) Function to call
+  // 4) Style of the function call (Call versus tail-call)
 
-    /**
-     * @brief A literal pair relocation object for named symbol literals
-     */
-    struct NamedSymbolLiteralPair {
-      Literal<uint64_t> Lit;
-      Relocation MoveABI{};
-    };
+  template<typename R, typename... P>
+  void GenerateRuntimeCall(R (*Function)(P...)) {
+    uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<R, P...>::Wrapper));
 
-    /**
-     * @brief Inserts a thunk relocation
-     *
-     * @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);
+    uintptr_t FunctionAddress = reinterpret_cast<uintptr_t>(Function);
 
-    /**
-     * @brief Inserts a guest GPR move relocation
-     *
-     * @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);
+    hlt(vixl::aarch64::kRuntimeCallOpcode);
 
-    /**
-     * @brief Inserts a named symbol as a literal in memory
-     *
-     * Need to use `PlaceNamedSymbolLiteral` with the return value to place the literal in the desired location
-     *
-     * @param Op The named symbol to place
-     *
-     * @return A temporary `NamedSymbolLiteralPair`
-     */
-    NamedSymbolLiteralPair InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+    // Simulator wrapper address pointer.
+    dc64(SimulatorWrapperAddress);
 
-    /**
-     * @brief Place the named symbol literal relocation in memory
-     *
-     * @param Lit - Which literal to place
-     */
-    void PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit);
+    // Runtime function address to call
+    dc64(FunctionAddress);
 
-    std::vector<FEXCore::CPU::Relocation> Relocations;
+    // Call type
+    dc32(vixl::aarch64::kCallRuntime);
+  }
 
-    ///< Relocation code loading
-    bool ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations);
+  template<typename R, typename... P>
+  void GenerateIndirectRuntimeCall(ARMEmitter::Register Reg) {
+    uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<R, P...>::Wrapper));
 
-  /**  @} */
+    hlt(vixl::aarch64::kIndirectRuntimeCallOpcode);
 
-  uint32_t SpillSlots{};
-  /**
-  * @brief Current guest RIP entrypoint
-  */
-  uint64_t GuestEntry{};
+    // Simulator wrapper address pointer.
+    dc64(SimulatorWrapperAddress);
 
+    // Register that contains the function to call
+    dc32(Reg.Idx());
+
+    // Call type
+    dc32(vixl::aarch64::kCallRuntime);
+  }
+
+  template<>
+  void GenerateIndirectRuntimeCall<float, __uint128_t>(ARMEmitter::Register Reg) {
+    uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<float, __uint128_t>::Wrapper));
+
+    hlt(vixl::aarch64::kIndirectRuntimeCallOpcode);
+
+    // Simulator wrapper address pointer.
+    dc64(SimulatorWrapperAddress);
+
+    // Register that contains the function to call
+    dc32(Reg.Idx());
+
+    // Call type
+    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;
 }
@@ -124,11 +130,66 @@ static inline void SetArmReg(void* ucontext, uint32_t id, uint64_t val) {
 static inline __uint128_t GetArmFPR(void* ucontext, uint32_t id) {
   auto MContext = GetMContext(ucontext);
   HostFPRState *HostState = reinterpret_cast<HostFPRState*>(&MContext->__reserved[0]);
-  LOGMAN_THROW_A_FMT(HostState->Head.Magic == FPR_MAGIC, "Wrong FPR Magic: 0x{:08x}", HostState->Head.Magic);
+  LOGMAN_THROW_AA_FMT(HostState->Head.Magic == FPR_MAGIC, "Wrong FPR Magic: 0x{:08x}", HostState->Head.Magic);
 
   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) {
@@ -143,7 +204,7 @@ static inline void BackupContext(void* ucontext, T *Backup) {
 
     // Host FPR state starts at _mcontext->reserved[0];
     HostFPRState *HostState = reinterpret_cast<HostFPRState*>(&_mcontext->__reserved[0]);
-    LOGMAN_THROW_A_FMT(HostState->Head.Magic == FPR_MAGIC, "Wrong FPR Magic: 0x{:08x}", HostState->Head.Magic);
+    LOGMAN_THROW_AA_FMT(HostState->Head.Magic == FPR_MAGIC, "Wrong FPR Magic: 0x{:08x}", HostState->Head.Magic);
     Backup->FPSR = HostState->FPSR;
     Backup->FPCR = HostState->FPCR;
     memcpy(&Backup->FPRs[0], &HostState->FPRs[0], 32 * sizeof(__uint128_t));
@@ -163,7 +224,7 @@ static inline void RestoreContext(void* ucontext, T *Backup) {
    auto _mcontext = GetMContext(ucontext);
 
     HostFPRState *HostState = reinterpret_cast<HostFPRState*>(&_mcontext->__reserved[0]);
-    LOGMAN_THROW_A_FMT(HostState->Head.Magic == FPR_MAGIC, "Wrong FPR Magic: 0x{:08x}", HostState->Head.Magic);
+    LOGMAN_THROW_AA_FMT(HostState->Head.Magic == FPR_MAGIC, "Wrong FPR Magic: 0x{:08x}", HostState->Head.Magic);
     memcpy(&HostState->FPRs[0], &Backup->FPRs[0], 32 * sizeof(__uint128_t));
     HostState->FPCR = Backup->FPCR;
     HostState->FPSR = Backup->FPSR;
@@ -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/CPUBackend.cpp

@@ -0,0 +1,87 @@
+#include "Interface/Context/Context.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
+#include <FEXCore/Core/CPUBackend.h>
+
+namespace FEXCore {
+namespace CPU {
+
+CPUBackend::CPUBackend(FEXCore::Core::InternalThreadState *ThreadState, size_t InitialCodeSize, size_t MaxCodeSize)
+    : ThreadState(ThreadState), InitialCodeSize(InitialCodeSize), MaxCodeSize(MaxCodeSize) {}
+
+CPUBackend::~CPUBackend() {
+  for (auto CodeBuffer : CodeBuffers) {
+    FreeCodeBuffer(CodeBuffer);
+  }
+  CodeBuffers.clear();
+}
+
+auto CPUBackend::GetEmptyCodeBuffer() -> CodeBuffer * {
+  if (ThreadState->CurrentFrame->SignalHandlerRefCounter == 0) {
+    if (CodeBuffers.empty()) {
+      auto NewCodeBuffer = AllocateNewCodeBuffer(InitialCodeSize);
+      EmplaceNewCodeBuffer(NewCodeBuffer);
+    } else {
+      if (CodeBuffers.size() > 1) {
+        // If we have more than one code buffer we are tracking then walk them and delete
+        // This is a cleanup step
+        for (size_t i = 1; i < CodeBuffers.size(); i++) {
+          FreeCodeBuffer(CodeBuffers[i]);
+        }
+        CodeBuffers.resize(1);
+      }
+      // Set the current code buffer to the initial
+      CurrentCodeBuffer = &CodeBuffers[0];
+
+      if (CurrentCodeBuffer->Size != MaxCodeSize) {
+        FreeCodeBuffer(*CurrentCodeBuffer);
+
+        // Resize the code buffer and reallocate our code size
+        CurrentCodeBuffer->Size *= 1.5;
+        CurrentCodeBuffer->Size = std::min(CurrentCodeBuffer->Size, MaxCodeSize);
+
+        *CurrentCodeBuffer = AllocateNewCodeBuffer(CurrentCodeBuffer->Size);
+      }
+    }
+  } else {
+    // We have signal handlers that have generated code
+    // This means that we can not safely clear the code at this point in time
+    // Allocate some new code buffers that we can switch over to instead
+    auto NewCodeBuffer = AllocateNewCodeBuffer(InitialCodeSize);
+    EmplaceNewCodeBuffer(NewCodeBuffer);
+  }
+
+  return CurrentCodeBuffer;
+}
+
+auto CPUBackend::AllocateNewCodeBuffer(size_t Size) -> CodeBuffer {
+  CodeBuffer Buffer;
+  Buffer.Size = Size;
+  Buffer.Ptr = static_cast<uint8_t *>(
+      FEXCore::Allocator::mmap(nullptr, Buffer.Size, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
+  LOGMAN_THROW_AA_FMT(!!Buffer.Ptr, "Couldn't allocate code buffer");
+
+  if (ThreadState->CTX->Config.GlobalJITNaming()) {
+    ThreadState->CTX->Symbols.RegisterJITSpace(Buffer.Ptr, Buffer.Size);
+  }
+  return Buffer;
+}
+
+void CPUBackend::FreeCodeBuffer(CodeBuffer Buffer) {
+  FEXCore::Allocator::munmap(Buffer.Ptr, Buffer.Size);
+}
+
+bool CPUBackend::IsAddressInCodeBuffer(uintptr_t Address) const {
+  for (auto &Buffer: CodeBuffers) {
+    auto start = (uintptr_t)Buffer.Ptr;
+    auto end = start + Buffer.Size;
+
+    if (Address >= start && Address < end) {
+      return true;
+    }
+  }
+
+  return false;
+}
+
+}
+}

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

@@ -8,11 +8,11 @@ $end_info$
 #include "Common/StringConv.h"
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUID.h"
-#include "Interface/Core/HostFeatures.h"
 #include "Utils/FileLoading.h"
 
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Core/CPUID.h>
+#include <FEXCore/Core/HostFeatures.h>
 #include <FEXHeaderUtils/Syscalls.h>
 
 #include "git_version.h"
@@ -39,6 +39,7 @@ namespace ProductNames {
   static const char ARM_A78C[] = "Cortex-A78C";
   static const char ARM_A710[] = "Cortex-A710";
   static const char ARM_X1[] = "Cortex-X1";
+  static const char ARM_X1C[] = "Cortex-X1C";
   static const char ARM_X2[] = "Cortex-X2";
   static const char ARM_N1[] = "Neoverse N1";
   static const char ARM_N2[] = "Neoverse N2";
@@ -83,7 +84,10 @@ static uint32_t CalculateNumberOfCPUs() {
   return CPUs;
 }
 
+// TODO: Replace usages with CTX->HostFeatures.EnableAVX
+//       when AVX implementations are further along.
 constexpr uint32_t SUPPORTS_AVX = 0;
+
 // #define CPUID_AMD
 #ifdef CPUID_AMD
 constexpr uint32_t FAMILY_IDENTIFIER =
@@ -151,7 +155,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
   // CPU priority order
   // This is mostly arbitrary but will sort by some sort of CPU priority by performance
   // Relative list so things they will commonly end up in big.little configurations sort of relate
-  static constexpr std::array<CPUMIDR, 35> CPUMIDRs = {{
+  static constexpr std::array<CPUMIDR, 36> CPUMIDRs = {{
     // Typically big CPU cores
     {0x61, 0x023, 1, ProductNames::ARM_Firestorm}, // Apple M1 Firestorm
 
@@ -161,6 +165,7 @@ void CPUIDEmu::SetupHostHybridFlag() {
     {0x41, 0xd49, 1, ProductNames::ARM_N2}, // N2
     {0x41, 0xd48, 1, ProductNames::ARM_X2}, // X2
     {0x41, 0xd47, 1, ProductNames::ARM_A710}, // A710
+    {0x41, 0xd4C, 1, ProductNames::ARM_X1C}, // X1C
     {0x41, 0xd44, 1, ProductNames::ARM_X1}, // X1
     {0x41, 0xd42, 1, ProductNames::ARM_A78AE}, // A78AE
     {0x41, 0xd41, 1, ProductNames::ARM_A78}, // A78
@@ -416,7 +421,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
 
   Res.ecx =
     (1 <<  0) | // SSE3
-    (0 <<  1) | // PCLMULQDQ
+    (CTX->HostFeatures.SupportsPMULL_128Bit <<  1) | // PCLMULQDQ
     (1 <<  2) | // DS area supports 64bit layout
     (1 <<  3) | // MWait
     (0 <<  4) | // DS-CPL
@@ -446,7 +451,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
     (SUPPORTS_AVX << 28) | // AVX
     (0 << 29) | // F16C
     (CTX->HostFeatures.SupportsRAND << 30) | // RDRAND
-    (0 << 31);  // Hypervisor always returns zero
+    (1 << 31);  // Hypervisor always returns one
 
   Res.edx =
     (1 <<  0) | // FPU

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

@@ -7,6 +7,7 @@ desc: Glues Frontend, OpDispatcher and IR Opts & Compilation, LookupCache, Dispa
 $end_info$
 */
 
+#include <cstdint>
 #include "Interface/Context/Context.h"
 #include "Interface/Core/LookupCache.h"
 #include "Interface/Core/Core.h"
@@ -17,6 +18,7 @@ $end_info$
 #include "Interface/Core/OpcodeDispatcher.h"
 #include "Interface/Core/Interpreter/InterpreterCore.h"
 #include "Interface/Core/JIT/JITCore.h"
+#include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/HLE/Thunks/Thunks.h"
 #include "Interface/IR/Passes/RegisterAllocationPass.h"
 #include "Interface/IR/Passes.h"
@@ -32,6 +34,7 @@ $end_info$
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Debug/X86Tables.h>
 #include <FEXCore/HLE/SyscallHandler.h>
+#include <FEXCore/HLE/SourcecodeResolver.h>
 #include <FEXCore/HLE/Linux/ThreadManagement.h>
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IREmitter.h>
@@ -41,6 +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>
 
@@ -49,7 +53,6 @@ $end_info$
 #include <atomic>
 #include <chrono>
 #include <condition_variable>
-#include <cstdint>
 #include <filesystem>
 #include <functional>
 #include <fstream>
@@ -74,6 +77,7 @@ $end_info$
 #include <vector>
 #include <xxhash.h>
 
+
 namespace FEXCore::CPU {
   bool CreateCPUCore(FEXCore::Context::Context *CTX) {
     // This should be used for generating things that are shared between threads
@@ -151,6 +155,16 @@ namespace FEXCore::Context {
     if (Config.CacheObjectCodeCompilation() != FEXCore::Config::ConfigObjectCodeHandler::CONFIG_NONE) {
       CodeObjectCacheService = std::make_unique<FEXCore::CodeSerialize::CodeObjectSerializeService>(this);
     }
+    if (!Config.EnableAVX) {
+      HostFeatures.SupportsAVX = false;
+    }
+
+    if (Config.BlockJITNaming() ||
+        Config.GlobalJITNaming() ||
+        Config.LibraryJITNaming()) {
+      // Only initialize symbols file if enabled. Ensures we don't pollute /tmp with empty files.
+      Symbols.InitFile();
+    }
   }
 
   Context::~Context() {
@@ -177,15 +191,17 @@ namespace FEXCore::Context {
 
     // Initialize default CPU state
     NewThreadState.rip = ~0ULL;
-    for (int i = 0; i < 16; ++i) {
-      NewThreadState.gregs[i] = 0;
+    for (auto& greg : NewThreadState.gregs) {
+      greg = 0;
     }
 
-    for (int i = 0; i < 16; ++i) {
-      NewThreadState.xmm[i][0] = 0xDEADBEEFULL;
-      NewThreadState.xmm[i][1] = 0xBAD0DAD1ULL;
+    for (auto& xmm : NewThreadState.xmm.avx.data) {
+      xmm[0] = 0xDEADBEEFULL;
+      xmm[1] = 0xBAD0DAD1ULL;
+      xmm[2] = 0xDEADCAFEULL;
+      xmm[3] = 0xBAD2CAD3ULL;
     }
-    memset(NewThreadState.flags, 0, 32);
+    memset(NewThreadState.flags, 0, Core::CPUState::NUM_EFLAG_BITS);
     NewThreadState.flags[1] = 1;
     NewThreadState.flags[9] = 1;
     NewThreadState.FCW = 0x37F;
@@ -193,19 +209,22 @@ namespace FEXCore::Context {
     return NewThreadState;
   }
 
-  FEXCore::Core::InternalThreadState* Context::InitCore(FEXCore::CodeLoader *Loader) {
-    // Initialize the CPU core signal handlers
+  FEXCore::Core::InternalThreadState* Context::InitCore(uint64_t InitialRIP, uint64_t StackPointer) {
+    // Initialize the CPU core signal handlers & DispatcherConfig
     switch (Config.Core) {
 #ifdef INTERPRETER_ENABLED
     case FEXCore::Config::CONFIG_INTERPRETER:
       FEXCore::CPU::InitializeInterpreterSignalHandlers(this);
+      BackendFeatures = FEXCore::CPU::GetInterpreterBackendFeatures();
       break;
 #endif
     case FEXCore::Config::CONFIG_IRJIT:
 #if (_M_X86_64 && JIT_X86_64)
       FEXCore::CPU::InitializeX86JITSignalHandlers(this);
-#elif (_M_ARM_64 && JIT_ARM64)
+      BackendFeatures = FEXCore::CPU::GetX86JITBackendFeatures();
+#elif (_M_ARM_64 && JIT_ARM64) || defined(VIXL_SIMULATOR)
       FEXCore::CPU::InitializeArm64JITSignalHandlers(this);
+      BackendFeatures = FEXCore::CPU::GetArm64JITBackendFeatures();
 #else
       ERROR_AND_DIE_FMT("FEXCore has been compiled without a viable JIT core");
 #endif
@@ -218,6 +237,32 @@ namespace FEXCore::Context {
       break;
     }
 
+    DispatcherConfig.StaticRegisterAllocation = Config.StaticRegisterAllocation && BackendFeatures.SupportsStaticRegisterAllocation;
+
+#if JIT_ARM64
+    Dispatcher = FEXCore::CPU::Dispatcher::CreateArm64(this, DispatcherConfig);
+#elif JIT_X86_64
+    Dispatcher = FEXCore::CPU::Dispatcher::CreateX86(this, DispatcherConfig);
+#else
+    ERROR_AND_DIE_FMT("FEXCore has been compiled with an unknown target");
+#endif
+
+    // Initialize common signal handlers
+
+    auto PauseHandler = [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
+      return Thread->CTX->Dispatcher->HandleSignalPause(Thread, Signal, info, ucontext);
+    };
+
+    SignalDelegation->RegisterHostSignalHandler(SignalDelegator::SIGNAL_FOR_PAUSE, PauseHandler, true);
+
+    auto GuestSignalHandler = [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) -> bool {
+      return Thread->CTX->Dispatcher->HandleGuestSignal(Thread, Signal, info, ucontext, GuestAction, GuestStack);
+    };
+
+    for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
+      SignalDelegation->RegisterHostSignalHandlerForGuest(Signal, GuestSignalHandler);
+    }
+
     // Initialize GDBServer after the signal handlers are installed
     // It may install its own handlers that need to be executed AFTER the CPU cores
     if (Config.GdbServer) {
@@ -229,7 +274,6 @@ namespace FEXCore::Context {
 
     ThunkHandler.reset(FEXCore::ThunkHandler::Create());
 
-    LocalLoader = Loader;
     using namespace FEXCore::Core;
 
     FEXCore::Core::CPUState NewThreadState = CreateDefaultCPUState();
@@ -238,9 +282,9 @@ namespace FEXCore::Context {
     // We are the parent thread
     ParentThread = Thread;
 
-    Thread->CurrentFrame->State.gregs[X86State::REG_RSP] = Loader->GetStackPointer();
+    Thread->CurrentFrame->State.gregs[X86State::REG_RSP] = StackPointer;
 
-    Thread->CurrentFrame->State.rip = StartingRIP = Loader->DefaultRIP();
+    Thread->CurrentFrame->State.rip = InitialRIP;
 
     InitializeThreadData(Thread);
     return Thread;
@@ -258,7 +302,7 @@ namespace FEXCore::Context {
   }
 
   void Context::HandleCallback(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP) {
-    Thread->CPUBackend->CallbackPtr(Thread->CurrentFrame, RIP);
+    Thread->CTX->Dispatcher->ExecuteJITCallback(Thread->CurrentFrame, RIP);
   }
 
   void Context::RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required) {
@@ -352,10 +396,7 @@ namespace FEXCore::Context {
       // Walk the threads and tell them to clear their caches
       // Useful when our block size is set to a large number and we need to step a single instruction
       for (auto &Thread : Threads) {
-        // Wait for thread to be fully constructed
-        // XXX: Look into thread partial construction issues
-        while(Thread->RunningEvents.WaitingToStart.load()) ;
-        ClearCodeCache(Thread, true);
+        ClearCodeCache(Thread);
       }
     }
     CoreRunningMode PreviousRunningMode = this->Config.RunningMode;
@@ -448,22 +489,6 @@ namespace FEXCore::Context {
 
   void Context::InitializeThreadData(FEXCore::Core::InternalThreadState *Thread) {
     Thread->CPUBackend->Initialize();
-
-    auto IRHandler = [Thread](uint64_t Addr, IR::IREmitter *IR) -> void {
-      // Run the passmanager over the IR from the dispatcher
-      Thread->PassManager->Run(IR);
-      Core::LocalIREntry Entry = {Addr, 0ULL,
-        decltype(Entry.IR)(IR->CreateIRCopy()),
-        decltype(Entry.RAData)(Thread->PassManager->HasPass("RA")
-          ? Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA")->PullAllocationData()
-          : nullptr),
-        decltype(Entry.DebugData)(new Core::DebugData())
-      };
-      std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
-      Thread->LocalIRCache.insert({Addr, std::move(Entry)});
-    };
-
-    LocalLoader->AddIR(IRHandler);
   }
 
   struct ExecutionThreadHandler {
@@ -483,10 +508,21 @@ namespace FEXCore::Context {
     ExecutionThreadHandler *Arg = reinterpret_cast<ExecutionThreadHandler*>(FEXCore::Allocator::malloc(sizeof(ExecutionThreadHandler)));
     Arg->This = this;
     Arg->Thread = Thread;
+    Thread->StartPaused = NeedToCheckXID;
     Thread->ExecutionThread = FEXCore::Threads::Thread::Create(ThreadHandler, Arg);
 
     // Wait for the thread to have started
     Thread->ThreadWaiting.Wait();
+
+    if (NeedToCheckXID) {
+      // The first time an application creates a thread, GLIBC installs their SETXID signal handler.
+      // FEX needs to capture all signals and defer them to the guest.
+      // Once FEX creates its first guest thread, overwrite the GLIBC SETXID handler *again* to ensure
+      // FEX maintains control of the signal handler on this signal.
+      NeedToCheckXID = false;
+      SignalDelegation->CheckXIDHandler();
+      Thread->StartRunning.NotifyAll();
+    }
   }
 
   void Context::InitializeThreadTLSData(FEXCore::Core::InternalThreadState *Thread) {
@@ -502,50 +538,51 @@ namespace FEXCore::Context {
     Thread->StartRunning.NotifyAll();
   }
 
-  void Context::InitializeCompiler(FEXCore::Core::InternalThreadState* State) {
-    State->OpDispatcher = std::make_unique<FEXCore::IR::OpDispatchBuilder>(this);
-    State->OpDispatcher->SetMultiblock(Config.Multiblock);
-    State->LookupCache = std::make_unique<FEXCore::LookupCache>(this);
-    State->FrontendDecoder = std::make_unique<FEXCore::Frontend::Decoder>(this);
-    State->PassManager = std::make_unique<FEXCore::IR::PassManager>();
-    State->PassManager->RegisterExitHandler([this]() {
+  void Context::InitializeCompiler(FEXCore::Core::InternalThreadState* Thread) {
+    Thread->OpDispatcher = std::make_unique<FEXCore::IR::OpDispatchBuilder>(this);
+    Thread->OpDispatcher->SetMultiblock(Config.Multiblock);
+    Thread->LookupCache = std::make_unique<FEXCore::LookupCache>(this);
+    Thread->FrontendDecoder = std::make_unique<FEXCore::Frontend::Decoder>(this);
+    Thread->PassManager = std::make_unique<FEXCore::IR::PassManager>();
+    Thread->PassManager->RegisterExitHandler([this]() {
         Stop(false /* Ignore current thread */);
     });
 
-    State->CTX = this;
+    Thread->CurrentFrame->Pointers.Common.L1Pointer = Thread->LookupCache->GetL1Pointer();
+    Thread->CurrentFrame->Pointers.Common.L2Pointer = Thread->LookupCache->GetPagePointer();
 
-    #if _M_ARM_64
-    bool DoSRA = State->CTX->Config.StaticRegisterAllocation;
-    #else
-    bool DoSRA = false;
-    #endif
+    Dispatcher->InitThreadPointers(Thread);
 
-    State->PassManager->AddDefaultPasses(this, Config.Core == FEXCore::Config::CONFIG_IRJIT, DoSRA);
-    State->PassManager->AddDefaultValidationPasses();
+    Thread->CTX = this;
 
-    State->PassManager->RegisterSyscallHandler(SyscallHandler);
+    bool DoSRA = DispatcherConfig.StaticRegisterAllocation;
+
+    Thread->PassManager->AddDefaultPasses(this, Config.Core == FEXCore::Config::CONFIG_IRJIT, DoSRA);
+    Thread->PassManager->AddDefaultValidationPasses();
+
+    Thread->PassManager->RegisterSyscallHandler(SyscallHandler);
 
     // Create CPU backend
     switch (Config.Core) {
 #ifdef INTERPRETER_ENABLED
     case FEXCore::Config::CONFIG_INTERPRETER:
-      State->CPUBackend = FEXCore::CPU::CreateInterpreterCore(this, State);
+      Thread->CPUBackend = FEXCore::CPU::CreateInterpreterCore(this, Thread);
       break;
 #endif
     case FEXCore::Config::CONFIG_IRJIT:
-      State->PassManager->InsertRegisterAllocationPass(DoSRA);
+      Thread->PassManager->InsertRegisterAllocationPass(DoSRA, HostFeatures.SupportsAVX);
 
 #if (_M_X86_64 && JIT_X86_64)
-      State->CPUBackend = FEXCore::CPU::CreateX86JITCore(this, State);
-#elif (_M_ARM_64 && JIT_ARM64)
-      State->CPUBackend = FEXCore::CPU::CreateArm64JITCore(this, State);
+      Thread->CPUBackend = FEXCore::CPU::CreateX86JITCore(this, Thread);
+#elif (_M_ARM_64 && JIT_ARM64) || defined(VIXL_SIMULATOR)
+      Thread->CPUBackend = FEXCore::CPU::CreateArm64JITCore(this, Thread);
 #else
       ERROR_AND_DIE_FMT("FEXCore has been compiled without a viable JIT core");
 #endif
 
       break;
     case FEXCore::Config::CONFIG_CUSTOM:
-      State->CPUBackend = CustomCPUFactory(this, State);
+      Thread->CPUBackend = CustomCPUFactory(this, Thread);
       break;
     default:
       ERROR_AND_DIE_FMT("Unknown core configuration");
@@ -554,14 +591,7 @@ namespace FEXCore::Context {
   }
 
   FEXCore::Core::InternalThreadState* Context::CreateThread(FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID) {
-    FEXCore::Core::InternalThreadState *Thread{};
-
-    // Grab the new thread object
-    {
-      std::lock_guard<std::mutex> lk(ThreadCreationMutex);
-      Thread = Threads.emplace_back(new FEXCore::Core::InternalThreadState{});
-      Thread->ThreadManager.TID = ++ThreadID;
-    }
+    FEXCore::Core::InternalThreadState *Thread = new FEXCore::Core::InternalThreadState{};
 
     // Copy over the new thread state to the new object
     memcpy(Thread->CurrentFrame, NewThreadState, sizeof(FEXCore::Core::CPUState));
@@ -573,13 +603,19 @@ namespace FEXCore::Context {
     InitializeCompiler(Thread);
     InitializeThreadData(Thread);
 
+    // Insert after the Thread object has been fully initialized
+    {
+      std::lock_guard lk(ThreadCreationMutex);
+      Threads.push_back(Thread);
+    }
+
     return Thread;
   }
 
   void Context::DestroyThread(FEXCore::Core::InternalThreadState *Thread) {
     // remove new thread object
     {
-      std::lock_guard<std::mutex> lk(ThreadCreationMutex);
+      std::lock_guard lk(ThreadCreationMutex);
 
       auto It = std::find(Threads.begin(), Threads.end(), Thread);
       LOGMAN_THROW_A_FMT(It != Threads.end(), "Thread wasn't in Threads");
@@ -634,14 +670,13 @@ namespace FEXCore::Context {
     FEXCore::Threads::Thread::CleanupAfterFork();
   }
 
-  void Context::AddBlockMapping(FEXCore::Core::InternalThreadState *Thread, uint64_t Address, void *Ptr, uint64_t Start, uint64_t Length) {
-    // Only call MarkGuestExecutableRange if new pages are marked as containing code
-    if (Thread->LookupCache->AddBlockMapping(Address, Ptr, Start, Length)) {
-      Thread->CTX->SyscallHandler->MarkGuestExecutableRange(Start, Length);
-    }
+  void Context::AddBlockMapping(FEXCore::Core::InternalThreadState *Thread, uint64_t Address, void *Ptr) {
+    Thread->LookupCache->AddBlockMapping(Address, Ptr);
   }
 
-  void Context::ClearCodeCache(FEXCore::Core::InternalThreadState *Thread, bool AlsoClearIRCache) {
+  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
@@ -651,18 +686,16 @@ namespace FEXCore::Context {
 
     Thread->LookupCache->ClearCache();
     Thread->CPUBackend->ClearCache();
-
-    if (AlsoClearIRCache) {
-      Thread->LocalIRCache.clear();
-    }
+    Thread->DebugStore.clear();
   }
 
-  static void IRDumper(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP, IR::RegisterAllocationData* RA) {
+  static void IRDumper(FEXCore::Core::InternalThreadState *Thread, IR::IREmitter *IREmitter, uint64_t GuestRIP, IR::RegisterAllocationData* RA) {
     FILE* f = nullptr;
     bool CloseAfter = false;
     const auto DumpIRStr = Thread->CTX->Config.DumpIR();
 
-    if (DumpIRStr =="stderr") {
+    // DumpIRStr might be no if not dumping but ShouldDump is set in OpDisp
+    if (DumpIRStr =="stderr" || DumpIRStr =="no") {
       f = stderr;
     }
     else if (DumpIRStr =="stdout") {
@@ -676,7 +709,7 @@ namespace FEXCore::Context {
 
     if (f) {
       std::stringstream out;
-      auto NewIR = Thread->OpDispatcher->ViewIR();
+      auto NewIR = IREmitter->ViewIR();
       FEXCore::IR::Dump(&out, &NewIR, RA);
       fmt::print(f,"IR-{} 0x{:x}:\n{}\n@@@@@\n", RA ? "post" : "pre", GuestRIP, out.str());
 
@@ -686,12 +719,12 @@ namespace FEXCore::Context {
     }
   };
 
-  static void ValidateIR(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
+  static void ValidateIR(FEXCore::Context::Context *ctx, IR::IREmitter *IREmitter) {
     // Convert to text, Parse, Convert to text again and make sure the texts match
     std::stringstream out;
     static auto compaction = IR::CreateIRCompaction(ctx->OpDispatcherAllocator);
-    compaction->Run(Thread->OpDispatcher.get());
-    auto NewIR = Thread->OpDispatcher->ViewIR();
+    compaction->Run(IREmitter);
+    auto NewIR = IREmitter->ViewIR();
     Dump(&out, &NewIR, nullptr);
     out.seekg(0);
     FEXCore::Utils::PooledAllocatorMalloc Allocator;
@@ -710,151 +743,181 @@ namespace FEXCore::Context {
     }
   }
 
-  Context::GenerateIRResult Context::GenerateIR(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) {
-    uint8_t const *GuestCode{};
-    GuestCode = reinterpret_cast<uint8_t const*>(GuestRIP);
+  Context::GenerateIRResult Context::GenerateIR(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP, bool ExtendedDebugInfo) {
+    FEXCORE_PROFILE_SCOPED("GenerateIR");
 
-    bool HadDispatchError {false};
+    Thread->OpDispatcher->ReownOrClaimBuffer();
+    Thread->OpDispatcher->ResetWorkingList();
 
     uint64_t TotalInstructions {0};
     uint64_t TotalInstructionsLength {0};
 
-    Thread->FrontendDecoder->DecodeInstructionsAtEntry(GuestCode, GuestRIP);
 
-    auto CodeBlocks = Thread->FrontendDecoder->GetDecodedBlocks();
+    std::shared_lock lk(CustomIRMutex);
 
-    Thread->OpDispatcher->ReownOrClaimBuffer();
-    Thread->OpDispatcher->ResetWorkingList();
-    Thread->OpDispatcher->BeginFunction(GuestRIP, CodeBlocks);
+    auto Handler = CustomIRHandlers.find(GuestRIP);
+    if (Handler != CustomIRHandlers.end()) {
+      TotalInstructions = 1;
+      TotalInstructionsLength = 1;
+      std::get<0>(Handler->second)(GuestRIP, Thread->OpDispatcher.get());
+      lk.unlock();
+    } else {
+      lk.unlock();
+      uint8_t const *GuestCode{};
+      GuestCode = reinterpret_cast<uint8_t const*>(GuestRIP);
 
-    const uint8_t GPRSize = GetGPRSize();
+      bool HadDispatchError {false};
 
-    for (size_t j = 0; j < CodeBlocks->size(); ++j) {
-      FEXCore::Frontend::Decoder::DecodedBlocks const &Block = CodeBlocks->at(j);
-      // Set the block entry point
-      Thread->OpDispatcher->SetNewBlockIfChanged(Block.Entry);
+      Thread->FrontendDecoder->DecodeInstructionsAtEntry(GuestCode, GuestRIP, [Thread](uint64_t BlockEntry, uint64_t Start, uint64_t Length) {
+        if (Thread->LookupCache->AddBlockExecutableRange(BlockEntry, Start, Length)) {
+          Thread->CTX->SyscallHandler->MarkGuestExecutableRange(Start, Length);
+        }
+      });
 
-      uint64_t BlockInstructionsLength {};
+      auto CodeBlocks = Thread->FrontendDecoder->GetDecodedBlocks();
 
-      // Reset any block-specific state
-      Thread->OpDispatcher->StartNewBlock();
+      Thread->OpDispatcher->BeginFunction(GuestRIP, CodeBlocks);
 
-      uint64_t InstsInBlock = Block.NumInstructions;
+      const uint8_t GPRSize = GetGPRSize();
 
-      for (size_t i = 0; i < InstsInBlock; ++i) {
-        FEXCore::X86Tables::X86InstInfo const* TableInfo {nullptr};
-        FEXCore::X86Tables::DecodedInst const* DecodedInfo {nullptr};
+      for (size_t j = 0; j < CodeBlocks->size(); ++j) {
+        FEXCore::Frontend::Decoder::DecodedBlocks const &Block = CodeBlocks->at(j);
+        // Set the block entry point
+        Thread->OpDispatcher->SetNewBlockIfChanged(Block.Entry);
 
-        TableInfo = Block.DecodedInstructions[i].TableInfo;
-        DecodedInfo = &Block.DecodedInstructions[i];
-        bool IsLocked = DecodedInfo->Flags & FEXCore::X86Tables::DecodeFlags::FLAG_LOCK;
+        uint64_t BlockInstructionsLength {};
 
-        if (Config.SMCChecks == FEXCore::Config::CONFIG_SMC_FULL) {
-          auto ExistingCodePtr = reinterpret_cast<uint64_t*>(Block.Entry + BlockInstructionsLength);
+        // Reset any block-specific state
+        Thread->OpDispatcher->StartNewBlock();
 
-          auto CodeChanged = Thread->OpDispatcher->_ValidateCode(ExistingCodePtr[0], ExistingCodePtr[1], (uintptr_t)ExistingCodePtr - GuestRIP, DecodedInfo->InstSize);
-
-          auto InvalidateCodeCond = Thread->OpDispatcher->_CondJump(CodeChanged);
-
-          auto CurrentBlock = Thread->OpDispatcher->GetCurrentBlock();
-          auto CodeWasChangedBlock = Thread->OpDispatcher->CreateNewCodeBlockAtEnd();
-          Thread->OpDispatcher->SetTrueJumpTarget(InvalidateCodeCond, CodeWasChangedBlock);
-
-          Thread->OpDispatcher->SetCurrentCodeBlock(CodeWasChangedBlock);
-          Thread->OpDispatcher->_RemoveThreadCodeEntry();
-          Thread->OpDispatcher->_ExitFunction(Thread->OpDispatcher->_EntrypointOffset(Block.Entry + BlockInstructionsLength - GuestRIP, GPRSize));
-
-          auto NextOpBlock = Thread->OpDispatcher->CreateNewCodeBlockAfter(CurrentBlock);
-
-          Thread->OpDispatcher->SetFalseJumpTarget(InvalidateCodeCond, NextOpBlock);
-          Thread->OpDispatcher->SetCurrentCodeBlock(NextOpBlock);
+        if (Config.x86dec_SynchronizeRIPOnAllBlocks) {
+          // Ensure the RIP is synchronized to the context on block entry.
+          // In the case of block linking, the RIP may not have synchronized.
+          auto NewRIP = Thread->OpDispatcher->_EntrypointOffset(Block.Entry - GuestRIP, GPRSize);
+          Thread->OpDispatcher->_StoreContext(GPRSize, IR::GPRClass, NewRIP, offsetof(FEXCore::Core::CPUState, rip));
         }
 
-        if (TableInfo && TableInfo->OpcodeDispatcher) {
-          auto Fn = TableInfo->OpcodeDispatcher;
-          Thread->OpDispatcher->HandledLock = false;
-          Thread->OpDispatcher->ResetDecodeFailure();
-          std::invoke(Fn, Thread->OpDispatcher, DecodedInfo);
-          if (Thread->OpDispatcher->HadDecodeFailure()) {
-            HadDispatchError = true;
+        uint64_t InstsInBlock = Block.NumInstructions;
+
+        for (size_t i = 0; i < InstsInBlock; ++i) {
+          FEXCore::X86Tables::X86InstInfo const* TableInfo {nullptr};
+          FEXCore::X86Tables::DecodedInst const* DecodedInfo {nullptr};
+
+          TableInfo = Block.DecodedInstructions[i].TableInfo;
+          DecodedInfo = &Block.DecodedInstructions[i];
+          bool IsLocked = DecodedInfo->Flags & FEXCore::X86Tables::DecodeFlags::FLAG_LOCK;
+
+          if (ExtendedDebugInfo) {
+            Thread->OpDispatcher->_GuestOpcode(Block.Entry + BlockInstructionsLength - GuestRIP);
+          }
+
+          if (Config.SMCChecks == FEXCore::Config::CONFIG_SMC_FULL) {
+            auto ExistingCodePtr = reinterpret_cast<uint64_t*>(Block.Entry + BlockInstructionsLength);
+
+            auto CodeChanged = Thread->OpDispatcher->_ValidateCode(ExistingCodePtr[0], ExistingCodePtr[1], (uintptr_t)ExistingCodePtr - GuestRIP, DecodedInfo->InstSize);
+
+            auto InvalidateCodeCond = Thread->OpDispatcher->_CondJump(CodeChanged);
+
+            auto CurrentBlock = Thread->OpDispatcher->GetCurrentBlock();
+            auto CodeWasChangedBlock = Thread->OpDispatcher->CreateNewCodeBlockAtEnd();
+            Thread->OpDispatcher->SetTrueJumpTarget(InvalidateCodeCond, CodeWasChangedBlock);
+
+            Thread->OpDispatcher->SetCurrentCodeBlock(CodeWasChangedBlock);
+            Thread->OpDispatcher->_ThreadRemoveCodeEntry();
+            Thread->OpDispatcher->_ExitFunction(Thread->OpDispatcher->_EntrypointOffset(Block.Entry + BlockInstructionsLength - GuestRIP, GPRSize));
+
+            auto NextOpBlock = Thread->OpDispatcher->CreateNewCodeBlockAfter(CurrentBlock);
+
+            Thread->OpDispatcher->SetFalseJumpTarget(InvalidateCodeCond, NextOpBlock);
+            Thread->OpDispatcher->SetCurrentCodeBlock(NextOpBlock);
+          }
+
+          if (TableInfo && TableInfo->OpcodeDispatcher) {
+            auto Fn = TableInfo->OpcodeDispatcher;
+            Thread->OpDispatcher->HandledLock = false;
+            Thread->OpDispatcher->ResetDecodeFailure();
+            std::invoke(Fn, Thread->OpDispatcher, DecodedInfo);
+            if (Thread->OpDispatcher->HadDecodeFailure()) {
+              HadDispatchError = true;
+            }
+            else {
+              if (Thread->OpDispatcher->HandledLock != IsLocked) {
+                HadDispatchError = true;
+                LogMan::Msg::EFmt("Missing LOCK HANDLER at 0x{:x}{{'{}'}}", Block.Entry + BlockInstructionsLength, TableInfo->Name ?: "UND");
+              }
+              BlockInstructionsLength += DecodedInfo->InstSize;
+              TotalInstructionsLength += DecodedInfo->InstSize;
+              ++TotalInstructions;
+            }
           }
           else {
-            if (Thread->OpDispatcher->HandledLock != IsLocked) {
-              HadDispatchError = true;
-              LogMan::Msg::EFmt("Missing LOCK HANDLER at 0x{:x}{{'{}'}}", Block.Entry + BlockInstructionsLength, TableInfo->Name ?: "UND");
-            }
-            BlockInstructionsLength += DecodedInfo->InstSize;
-            TotalInstructionsLength += DecodedInfo->InstSize;
-            ++TotalInstructions;
+            // Invalid instruction
+            Thread->OpDispatcher->InvalidOp(DecodedInfo);
+            Thread->OpDispatcher->_ExitFunction(Thread->OpDispatcher->_EntrypointOffset(Block.Entry - GuestRIP, GPRSize));
           }
-        }
-        else {
-          // Invalid instruction
-          Thread->OpDispatcher->InvalidOp(DecodedInfo);
-          Thread->OpDispatcher->_ExitFunction(Thread->OpDispatcher->_EntrypointOffset(Block.Entry - GuestRIP, GPRSize));
-        }
 
-        // If we had a dispatch error then leave early
-        if (HadDispatchError) {
-          if (TotalInstructions == 0) {
+          const bool NeedsBlockEnd = (HadDispatchError && TotalInstructions > 0) ||
+            (Thread->OpDispatcher->NeedsBlockEnder() && i + 1 == InstsInBlock);
+
+          // 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 };
           }
-          else {
+
+          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;
+
+          if (Thread->OpDispatcher->FinishOp(DecodedInfo->PC + DecodedInfo->InstSize, i + 1 == InstsInBlock)) {
+            break;
+          }
         }
       }
+
+      Thread->OpDispatcher->Finalize();
+
+      Thread->FrontendDecoder->DelayedDisownBuffer();
     }
 
-    Thread->OpDispatcher->Finalize();
+    IR::IREmitter *IREmitter = Thread->OpDispatcher.get();
 
+    auto ShouldDump = Thread->CTX->Config.DumpIR() != "no" || Thread->OpDispatcher->ShouldDump;
     // Debug
     {
-      if (Thread->CTX->Config.DumpIR() != "no") {
-        IRDumper(Thread, GuestRIP, nullptr);
+      if (ShouldDump) {
+        IRDumper(Thread, IREmitter, GuestRIP, nullptr);
       }
 
       if (Thread->CTX->Config.ValidateIRarser) {
-        ValidateIR(this, Thread);
+        ValidateIR(this, IREmitter);
       }
     }
 
     // Run the passmanager over the IR from the dispatcher
-    Thread->PassManager->Run(Thread->OpDispatcher.get());
+    Thread->PassManager->Run(IREmitter);
 
     // Debug
     {
-      if (Thread->CTX->Config.DumpIR() != "no") {
-        IRDumper(Thread, GuestRIP, Thread->PassManager->HasPass("RA") ? Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA")->GetAllocationData() : nullptr);
-      }
-
-      if (Thread->OpDispatcher->ShouldDump) {
-        std::stringstream out;
-        auto NewIR = Thread->OpDispatcher->ViewIR();
-        FEXCore::IR::Dump(&out, &NewIR, Thread->PassManager->HasPass("RA") ? Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA")->GetAllocationData() : nullptr);
-        LogMan::Msg::IFmt("IR 0x{:x}:\n{}\n@@@@@\n", GuestRIP, out.str());
+      if (ShouldDump) {
+        IRDumper(Thread, IREmitter, GuestRIP, Thread->PassManager->HasPass("RA") ? Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA")->GetAllocationData() : nullptr);
       }
     }
 
     auto RAData = Thread->PassManager->HasPass("RA") ? Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA")->PullAllocationData() : nullptr;
-    auto IRList = Thread->OpDispatcher->CreateIRCopy();
+    auto IRList = IREmitter->CreateIRCopy();
 
-    Thread->OpDispatcher->DelayedDisownBuffer();
-    Thread->FrontendDecoder->DelayedDisownBuffer();
+    IREmitter->DelayedDisownBuffer();
 
     return {
       .IRList = IRList,
-      .RAData = RAData.release(),
+      .RAData = std::move(RAData),
       .TotalInstructions = TotalInstructions,
       .TotalInstructionsLength = TotalInstructionsLength,
       .StartAddr = Thread->FrontendDecoder->DecodedMinAddress,
@@ -865,27 +928,11 @@ namespace FEXCore::Context {
   Context::CompileCodeResult Context::CompileCode(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) {
     FEXCore::IR::IRListView *IRList {};
     FEXCore::Core::DebugData *DebugData {};
-    FEXCore::IR::RegisterAllocationData *RAData {};
+    FEXCore::IR::RegisterAllocationData::UniquePtr RAData {};
     bool GeneratedIR {};
     uint64_t StartAddr {};
     uint64_t Length {};
 
-    std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
-    // Do we already have this in the IR cache?
-    auto LocalEntry = Thread->LocalIRCache.find(GuestRIP);
-
-    if (LocalEntry != Thread->LocalIRCache.end()) {
-      // Entry already exists
-      // pull in the data
-      IRList = LocalEntry->second.IR.get();
-      DebugData = LocalEntry->second.DebugData.get();
-      RAData = LocalEntry->second.RAData.get();
-      StartAddr = LocalEntry->second.StartAddr;
-      Length = LocalEntry->second.Length;
-
-      GeneratedIR = false;
-    }
-
     // JIT Code object cache lookup
     if (CodeObjectCacheService) {
       auto CodeCacheEntry = CodeObjectCacheService->FetchCodeObjectFromCache(GuestRIP);
@@ -893,25 +940,33 @@ namespace FEXCore::Context {
         auto CompiledCode = Thread->CPUBackend->RelocateJITObjectCode(GuestRIP, CodeCacheEntry);
         if (CompiledCode) {
           return {
-            .CompiledCode = CompiledCode,
-            .IRData      = nullptr, // No IR data generated
-            .DebugData   = nullptr, // nullptr here ensures that code serialization doesn't occur on from cache read
-            .RAData      = nullptr, // No RA data generated
-            .GeneratedIR = false, // nullptr here ensures IR cache mechanisms won't run
-            .StartAddr   = 0, // Unused
-            .Length      = 0, // Unused
+              .CompiledCode = CompiledCode,
+              .IRData = nullptr,    // No IR data generated
+              .DebugData = nullptr, // nullptr here ensures that code serialization doesn't occur on from cache read
+              .RAData = nullptr,    // No RA data generated
+              .GeneratedIR = false, // nullptr here ensures IR cache mechanisms won't run
+              .StartAddr = 0,       // Unused
+              .Length = 0,          // Unused
           };
         }
       }
     }
 
+    if (SourcecodeResolver && Config.GDBSymbols()) {
+      auto AOTIRCacheEntry = SyscallHandler->LookupAOTIRCacheEntry(GuestRIP);
+      if (AOTIRCacheEntry.Entry && !AOTIRCacheEntry.Entry->ContainsCode) {
+        AOTIRCacheEntry.Entry->SourcecodeMap =
+            SourcecodeResolver->GenerateMap(AOTIRCacheEntry.Entry->Filename, AOTIRCacheEntry.Entry->FileId);
+      }
+    }
+
     // AOT IR bookkeeping and cache
     {
       auto [IRCopy, RACopy, DebugDataCopy, _StartAddr, _Length, _GeneratedIR] = IRCaptureCache.PreGenerateIRFetch(GuestRIP, IRList);
       if (_GeneratedIR) {
         // Setup pointers to internal structures
         IRList = IRCopy;
-        RAData = RACopy;
+        RAData = std::move(RACopy);
         DebugData = DebugDataCopy;
         StartAddr = _StartAddr;
         Length = _Length;
@@ -921,11 +976,11 @@ namespace FEXCore::Context {
 
     if (IRList == nullptr) {
       // Generate IR + Meta Info
-      auto [IRCopy, RACopy, TotalInstructions, TotalInstructionsLength, _StartAddr, _Length] = GenerateIR(Thread, GuestRIP);
+      auto [IRCopy, RACopy, TotalInstructions, TotalInstructionsLength, _StartAddr, _Length] = GenerateIR(Thread, GuestRIP, Config.GDBSymbols());
 
       // Setup pointers to internal structures
       IRList = IRCopy;
-      RAData = RACopy;
+      RAData = std::move(RACopy);
       DebugData = new FEXCore::Core::DebugData();
       StartAddr = _StartAddr;
       Length = _Length;
@@ -942,10 +997,10 @@ namespace FEXCore::Context {
     }
     // Attempt to get the CPU backend to compile this code
     return {
-      .CompiledCode = Thread->CPUBackend->CompileCode(GuestRIP, IRList, DebugData, RAData),
+      .CompiledCode = Thread->CPUBackend->CompileCode(GuestRIP, IRList, DebugData, RAData.get(), GetGdbServerStatus()),
       .IRData = IRList,
       .DebugData = DebugData,
-      .RAData = RAData,
+      .RAData = std::move(RAData),
       .GeneratedIR = GeneratedIR,
       .StartAddr = StartAddr,
       .Length = Length,
@@ -964,10 +1019,11 @@ namespace FEXCore::Context {
   }
 
   uintptr_t Context::CompileBlock(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
+    FEXCORE_PROFILE_SCOPED("CompileBlock");
     auto Thread = Frame->Thread;
 
-    // Needs to be held for SMC interactions around concurrent compile and invalidation hazards
-    auto InvalidationLk = Thread->CTX->SyscallHandler->CompileCodeLock(GuestRIP);
+    // Invalidate might take a unique lock on this, to guarantee that during invalidation no code gets compiled
+    std::shared_lock lk(CodeInvalidationMutex);
 
     // Is the code in the cache?
     // The backends only check L1 and L2, not L3
@@ -978,16 +1034,14 @@ namespace FEXCore::Context {
     void *CodePtr {};
     FEXCore::IR::IRListView *IRList {};
     FEXCore::Core::DebugData *DebugData {};
-    FEXCore::IR::RegisterAllocationData *RAData {};
 
     bool GeneratedIR {};
     uint64_t StartAddr {}, Length {};
 
-    auto [Code, IR, Data, RA, Generated, _StartAddr, _Length] = CompileCode(Thread, GuestRIP);
+    auto [Code, IR, Data, RAData, Generated, _StartAddr, _Length] = CompileCode(Thread, GuestRIP);
     CodePtr = Code;
     IRList = IR;
     DebugData = Data;
-    RAData = RA;
     GeneratedIR = Generated;
     StartAddr = _StartAddr;
     Length = _Length;
@@ -998,22 +1052,25 @@ namespace FEXCore::Context {
 
     // The core managed to compile the code.
     if (Config.BlockJITNaming()) {
+      auto FragmentBasePtr = reinterpret_cast<uint8_t *>(CodePtr);
+
       if (DebugData) {
-        auto GuestRIPLookup = this->SyscallHandler->LookupAOTIRCacheEntry(GuestRIP);
+        auto GuestRIPLookup = SyscallHandler->LookupAOTIRCacheEntry(GuestRIP);
 
         if (DebugData->Subblocks.size()) {
           for (auto& Subblock: DebugData->Subblocks) {
+            auto BlockBasePtr = FragmentBasePtr + Subblock.HostCodeOffset;
             if (GuestRIPLookup.Entry) {
-              Symbols.Register(CodePtr, DebugData->HostCodeSize, GuestRIPLookup.Entry->Filename, GuestRIP - GuestRIPLookup.Offset);
+              Symbols.Register(BlockBasePtr, DebugData->HostCodeSize, GuestRIPLookup.Entry->Filename, GuestRIP - GuestRIPLookup.VAFileStart);
             } else {
-            Symbols.Register((void*)Subblock.HostCodeStart, GuestRIP, Subblock.HostCodeSize);
-          }
+              Symbols.Register(BlockBasePtr, GuestRIP, Subblock.HostCodeSize);
+            }
           }
         } else {
           if (GuestRIPLookup.Entry) {
-            Symbols.Register(CodePtr, DebugData->HostCodeSize, GuestRIPLookup.Entry->Filename, GuestRIP - GuestRIPLookup.Offset);
+            Symbols.Register(FragmentBasePtr, DebugData->HostCodeSize, GuestRIPLookup.Entry->Filename, GuestRIP - GuestRIPLookup.VAFileStart);
         } else {
-          Symbols.Register(CodePtr, GuestRIP, DebugData->HostCodeSize);
+          Symbols.Register(FragmentBasePtr, GuestRIP, DebugData->HostCodeSize);
           }
         }
       }
@@ -1046,7 +1103,7 @@ namespace FEXCore::Context {
         GuestRIP,
         StartAddr,
         Length,
-        RAData,
+        std::move(RAData),
         IRList,
         DebugData,
         GeneratedIR)) {
@@ -1055,7 +1112,8 @@ namespace FEXCore::Context {
     }
 
     // Insert to lookup cache
-    AddBlockMapping(Thread, GuestRIP, CodePtr, StartAddr, Length);
+    // Pages containing this block are added via AddBlockExecutableRange before each page gets accessed in the frontend
+    AddBlockMapping(Thread, GuestRIP, CodePtr);
 
     return (uintptr_t)CodePtr;
   }
@@ -1071,7 +1129,7 @@ namespace FEXCore::Context {
     // Now notify the thread that we are initialized
     Thread->ThreadWaiting.NotifyAll();
 
-    if (Thread != Thread->CTX->ParentThread || StartPaused) {
+    if (Thread != Thread->CTX->ParentThread || StartPaused || Thread->StartPaused) {
       // Parent thread doesn't need to wait to run
       Thread->StartRunning.Wait();
     }
@@ -1083,7 +1141,7 @@ namespace FEXCore::Context {
 
       Thread->RunningEvents.Running = true;
 
-      Thread->CPUBackend->ExecuteDispatch(Thread->CurrentFrame);
+      Thread->CTX->Dispatcher->ExecuteDispatch(Thread->CurrentFrame);
 
       Thread->RunningEvents.Running = false;
     }
@@ -1125,35 +1183,105 @@ namespace FEXCore::Context {
 
     for (auto it = lower; it != upper; it++) {
       for (auto Address: it->second) {
-        Context::RemoveThreadCodeEntry(Thread, Address);
+        Context::ThreadRemoveCodeEntry(Thread, Address);
       }
       it->second.clear();
     }
   }
 
-  void InvalidateGuestCodeRange(FEXCore::Context::Context *CTX, uint64_t Start, uint64_t Length) {
-    std::lock_guard<std::mutex> lk(CTX->ThreadCreationMutex);
+  static void InvalidateGuestCodeRangeInternal(FEXCore::Context::Context *CTX, uint64_t Start, uint64_t Length) {
+    std::lock_guard lk(CTX->ThreadCreationMutex);
+
     for (auto &Thread : CTX->Threads) {
-      if (Thread->RunningEvents.Running.load()) {
-        InvalidateGuestThreadCodeRange(Thread, Start, Length);
+      InvalidateGuestThreadCodeRange(Thread, Start, Length);
+    }
+  }
+
+  void InvalidateGuestCodeRange(FEXCore::Context::Context *CTX, uint64_t Start, uint64_t Length) {
+    FHU::ScopedSignalMaskWithUniqueLock CodeInvalidationLock(CTX->CodeInvalidationMutex);
+
+    InvalidateGuestCodeRangeInternal(CTX, Start, Length);
+  }
+
+  void InvalidateGuestCodeRange(FEXCore::Context::Context *CTX, uint64_t Start, uint64_t Length, std::function<void(uint64_t start, uint64_t Length)> CallAfter) {
+    FHU::ScopedSignalMaskWithUniqueLock CodeInvalidationLock(CTX->CodeInvalidationMutex);
+
+    InvalidateGuestCodeRangeInternal(CTX, Start, Length);
+    CallAfter(Start, Length);
+  }
+
+  void Context::MarkMemoryShared() {
+    if (!IsMemoryShared) {
+      IsMemoryShared = true;
+
+      if (Config.TSOAutoMigration) {
+        std::lock_guard<std::mutex> lkThreads(ThreadCreationMutex);
+        LogMan::Throw::AFmt(Threads.size() == 1, "First MarkMemoryShared called must be before creating any threads");
+
+        auto Thread = Threads[0];
+
+        // Only the lookup cache is cleared here, so that old code can keep running until next compilation
+        std::lock_guard<std::recursive_mutex> lkLookupCache(Thread->LookupCache->WriteLock);
+        Thread->LookupCache->ClearCache();
+
+        // DebugStore also needs to be cleared
+        Thread->DebugStore.clear();
       }
     }
   }
 
-  void Context::RemoveThreadCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) {
+  void MarkMemoryShared(FEXCore::Context::Context *CTX) {
+    CTX->MarkMemoryShared();
+  }
+
+  void Context::ThreadAddBlockLink(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestDestination, uintptr_t HostLink, const std::function<void()> &delinker) {
+    std::shared_lock lk(Thread->CTX->CodeInvalidationMutex);
+
+    Thread->LookupCache->AddBlockLink(GuestDestination, HostLink, delinker);
+  }
+
+  void Context::ThreadRemoveCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) {
+    LogMan::Throw::AFmt(Thread->CTX->CodeInvalidationMutex.try_lock() == false, "CodeInvalidationMutex needs to be unique_locked here");
+
     std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
 
-    Thread->LocalIRCache.erase(GuestRIP);
+    Thread->DebugStore.erase(GuestRIP);
     Thread->LookupCache->Erase(GuestRIP);
   }
 
+  CustomIRResult Context::AddCustomIREntrypoint(uintptr_t Entrypoint, std::function<void(uintptr_t Entrypoint, FEXCore::IR::IREmitter *)> Handler, void *Creator, void *Data) {
+    LOGMAN_THROW_A_FMT(Config.Is64BitMode || !(Entrypoint >> 32), "64-bit Entrypoint in 32-bit mode {:x}", Entrypoint);
+
+    std::unique_lock lk(CustomIRMutex);
+
+    auto InsertedIterator = CustomIRHandlers.emplace(Entrypoint, std::tuple(Handler, Creator, Data));
+
+    if (!InsertedIterator.second) {
+      const auto &[fn, Creator, Data] = InsertedIterator.first->second;
+      return CustomIRResult(std::move(lk), Creator, Data);
+    } else {
+      lk.unlock();
+      return CustomIRResult(std::move(lk), 0, 0);
+    }
+  }
+
+  void Context::RemoveCustomIREntrypoint(uintptr_t Entrypoint) {
+    LOGMAN_THROW_A_FMT(Config.Is64BitMode || !(Entrypoint >> 32), "64-bit Entrypoint in 32-bit mode {:x}", Entrypoint);
+
+    std::scoped_lock lk(CustomIRMutex);
+
+    InvalidateGuestCodeRange(this, Entrypoint, 1, [this](uint64_t Entrypoint, uint64_t) {
+      CustomIRHandlers.erase(Entrypoint);
+    });
+  }
+
   // Debug interface
   void Context::CompileRIP(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP) {
     uint64_t RIPBackup = Thread->CurrentFrame->State.rip;
     Thread->CurrentFrame->State.rip = RIP;
 
     // Erase the RIP from all the storage backings if it exists
-    RemoveThreadCodeEntry(Thread, RIP);
+    ThreadRemoveCodeEntry(Thread, RIP);
 
     // We don't care if compilation passes or not
     CompileBlock(Thread->CurrentFrame, RIP);
@@ -1171,8 +1299,8 @@ namespace FEXCore::Context {
 
   bool Context::GetDebugDataForRIP(uint64_t RIP, FEXCore::Core::DebugData *Data) {
     std::lock_guard<std::recursive_mutex> lk(ParentThread->LookupCache->WriteLock);
-    auto it = ParentThread->LocalIRCache.find(RIP);
-    if (it == ParentThread->LocalIRCache.end()) {
+    auto it = ParentThread->DebugStore.find(RIP);
+    if (it == ParentThread->DebugStore.end()) {
       return false;
     }
 
@@ -1211,6 +1339,11 @@ namespace FEXCore::Context {
     }
   }
 
+
+  void Context::AppendThunkDefinitions(std::vector<FEXCore::IR::ThunkDefinition> const& Definitions) {
+    ThunkHandler->AppendThunkDefinitions(Definitions);
+  }
+
   void ConfigureAOTGen(FEXCore::Core::InternalThreadState *Thread, std::set<uint64_t> *ExternalBranches, uint64_t SectionMaxAddress) {
     Thread->FrontendDecoder->SetExternalBranches(ExternalBranches);
     Thread->FrontendDecoder->SetSectionMaxAddress(SectionMaxAddress);

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"
@@ -16,34 +17,44 @@
 #include <array>
 #include <bit>
 #include <cmath>
+#include <cstddef>
 #include <cstdint>
 #include <memory>
-#include <stddef.h>
 
-#include "aarch64/assembler-aarch64.h"
-#include "aarch64/constants-aarch64.h"
-#include "aarch64/operands-aarch64.h"
-#include "aarch64/cpu-aarch64.h"
-#include "code-buffer-vixl.h"
-#include "platform-vixl.h"
+#include <aarch64/assembler-aarch64.h>
+#include <aarch64/constants-aarch64.h>
+#include <aarch64/cpu-aarch64.h>
+#include <aarch64/operands-aarch64.h>
+#include <code-buffer-vixl.h>
+#include <platform-vixl.h>
 
 #include <sys/syscall.h>
 #include <unistd.h>
 
 namespace FEXCore::CPU {
 
-using namespace vixl;
-using namespace vixl::aarch64;
+constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096;
 
-static constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096;
-#define STATE x28
+Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &config)
+  : FEXCore::CPU::Dispatcher(ctx, config), Arm64Emitter(ctx, MAX_DISPATCHER_CODE_SIZE)
+#ifdef VIXL_SIMULATOR
+  , Simulator {&Decoder}
+#endif
+{
+#ifdef VIXL_SIMULATOR
+  // Hardcode a 256-bit vector width if we are running in the simulator.
+  Simulator.SetVectorLengthInBits(256);
+#endif
 
-Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config)
-  : FEXCore::CPU::Dispatcher(ctx, Thread), Arm64Emitter(ctx, MAX_DISPATCHER_CODE_SIZE) {
-  SRAEnabled = config.StaticRegisterAssignment;
-  SetAllowAssembler(true);
+  EmitDispatcher();
+}
 
-  DispatchPtr = GetCursorAddress<CPUBackend::AsmDispatch>();
+void Arm64Dispatcher::EmitDispatcher() {
+#ifdef VIXL_DISASSEMBLER
+  const auto DisasmBegin = GetCursorAddress<const vixl::aarch64::Instruction*>();
+#endif
+
+  DispatchPtr = GetCursorAddress<AsmDispatch>();
 
   // while (true) {
   //    Ptr = FindBlock(RIP)
@@ -53,12 +64,9 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   //    Ptr();
   // }
 
-  Literal l_PagePtr {Thread->LookupCache->GetPagePointer()};
-  Literal l_CTX {reinterpret_cast<uintptr_t>(CTX)};
-  Literal l_Sleep {reinterpret_cast<uint64_t>(SleepThread)};
-  Literal l_CompileBlock {GetCompileBlockPtr()};
-  Literal l_ExitFunctionLink {config.ExitFunctionLink};
-  Literal l_ExitFunctionLinkThis {config.ExitFunctionLinkThis};
+  ARMEmitter::ForwardLabel l_CTX;
+  ARMEmitter::ForwardLabel l_Sleep;
+  ARMEmitter::ForwardLabel l_CompileBlock;
 
   // Push all the register we need to save
   PushCalleeSavedRegisters();
@@ -66,135 +74,114 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   // 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, MemOperand(STATE, offsetof(FEXCore::Core::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>();
 
-  if (SRAEnabled) {
+  if (config.StaticRegisterAllocation) {
     FillStaticRegs();
   }
 
   // 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, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
-  auto RipReg = x2;
+
+  auto RipReg = ARMEmitter::XReg::x2;
+  ldr(RipReg, STATE_PTR(CpuStateFrame, State.rip));
 
   // L1 Cache
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.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);
 
-  if (!config.ExecuteBlocksWithCall) {
-    br(x3);
-  } else {
-    b(&CallBlock);
-  }
+  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, &l_PagePtr);
+  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 = Thread->LookupCache->GetVirtualMemorySize();
+  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, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.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
-      if (!config.ExecuteBlocksWithCall) {
-        br(x3);
-      } else {
-        bind(&CallBlock);
-        mov(x0, STATE);
-        blr(x3);
-
-        if (CTX->GetGdbServerStatus()) {
-          // 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(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
-          ldr(x0, &l_CTX);
-          ldr(w0, MemOperand(x0, offsetof(FEXCore::Context::Context, Config.RunningMode)));
-          // If the value == 0 then branch to the top
-          cbz(x0, &LoopTop);
-          // Else we need to pause now
-          b(&ThreadPauseHandler);
-        } else {
-          // Unconditionally loop to the top
-          // We will only stop on error when compiling a block or signal
-          b(&LoopTop);
-        }
-      }
+      br(ARMEmitter::Reg::r3);
     }
   }
 
   {
-    bind(&ExitSpillSRA);
     ThreadStopHandlerAddressSpillSRA = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
     ThreadStopHandlerAddress = GetCursorAddress<uint64_t>();
@@ -206,10 +193,9 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     ret();
   }
 
-  constexpr bool SignalSafeCompile = true;
   {
     ExitFunctionLinkerAddress = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
     if (SignalSafeCompile) {
@@ -221,51 +207,54 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
       // 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);
     }
 
-    ldr(x0, &l_ExitFunctionLinkThis);
-    mov(x1, STATE);
-    mov(x2, lr);
+    mov(ARMEmitter::XReg::x0, STATE);
+    mov(ARMEmitter::XReg::x1, ARMEmitter::XReg::lr);
 
-    ldr(x3, &l_ExitFunctionLink);
-    blr(x3);
+    ldr(ARMEmitter::XReg::x2, STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uintptr_t, void *, void *>(ARMEmitter::Reg::r2);
+#else
+    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 (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       FillStaticRegs();
-    br(x0);
+
+    br(ARMEmitter::Reg::r0);
   }
 
   // Need to create the block
   {
-    bind(&NoBlock);
+    Bind(&NoBlock);
 
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
     if (SignalSafeCompile) {
@@ -277,42 +266,45 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
       // 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
-    blr(x3); // { CTX, Frame, RIP}
-
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<void, void *, uint64_t, void *>(ARMEmitter::Reg::r3);
+#else
+    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 (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       FillStaticRegs();
 
     b(&LoopTop);
@@ -329,54 +321,59 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   {
     // Guest SIGILL handler
     // Needs to be distinct from the SignalHandlerReturnAddress
-    UnimplementedInstructionAddress = GetCursorAddress<uint64_t>();
+    GuestSignal_SIGILL = GetCursorAddress<uint64_t>();
 
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
     hlt(0);
   }
 
   {
-    // Guest Overflow handler
+    // Guest SIGTRAP handler
     // Needs to be distinct from the SignalHandlerReturnAddress
-    OverflowExceptionInstructionAddress = GetCursorAddress<uint64_t>();
+    GuestSignal_SIGTRAP  = GetCursorAddress<uint64_t>();
 
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
-    LoadConstant(x0, reinterpret_cast<uint64_t>(&SynchronousFaultData));
-    LoadConstant(w1, 1);
-    strb(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, FaultToTopAndGeneratedException)));
-    LoadConstant(w1, X86State::X86_TRAPNO_OF);
-    str(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, TrapNo)));
-    LoadConstant(w1, 0x80);
-    str(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, si_code)));
-    LoadConstant(x1, 0);
-    str(w1, MemOperand(x0, offsetof(Dispatcher::SynchronousFaultDataStruct, err_code)));
+    brk(0);
+  }
+
+  {
+    // Guest Overflow handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGSEGV = GetCursorAddress<uint64_t>();
+
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs();
 
     // hlt/udf = SIGILL
     // brk = SIGTRAP
     // ??? = SIGSEGV
     // Force a SIGSEGV by loading zero
-    ldr(x1, MemOperand(x1));
+    LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, 0);
+    ldr(ARMEmitter::XReg::x1, ARMEmitter::Reg::r1);
   }
 
   {
     ThreadPauseHandlerAddressSpillSRA = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
+    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);
-    blr(x2);
+    ldr(ARMEmitter::XReg::x0, &l_CTX);
+    mov(ARMEmitter::XReg::x1, STATE);
+    ldr(ARMEmitter::XReg::x2, &l_Sleep);
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<void, void *, void *>(ARMEmitter::Reg::r2);
+#else
+    blr(ARMEmitter::Reg::r2);
+#endif
 
     PauseReturnInstruction = GetCursorAddress<uint64_t>();
     // Fault to start running again
@@ -399,58 +396,54 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     // On return to the thunk, the thunk can get whatever its return value is from the thread context depending on ABI handling on its end
     // When the thunk itself returns, it'll do its regular return logic there
     // void ReentrantCallback(FEXCore::Core::InternalThreadState *Thread, uint64_t RIP);
-    CallbackPtr = GetCursorAddress<CPUBackend::JITCallback>();
+    CallbackPtr = GetCursorAddress<JITCallback>();
 
     // We expect the thunk to have previously pushed the registers it was using
     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
-    LoadConstant(x0, reinterpret_cast<uint64_t>(&SignalHandlerRefCounter));
-    ldr(w2, MemOperand(x0));
-    add(w2, w2, 1);
-    str(w2, MemOperand(x0));
+    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, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])));
-    sub(x2, x2, 16);
-    str(x2, MemOperand(STATE, offsetof(FEXCore::Core::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, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.rip)));
+    str(ARMEmitter::XReg::x1, STATE_PTR(CpuStateFrame, State.rip));
 
     // load static regs
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       FillStaticRegs();
 
     // Now go back to the regular dispatcher loop
     b(&LoopTop);
   }
 
-  // Long division helpers
-  uint64_t LUDIVHandler{};
-  uint64_t LDIVHandler{};
-  uint64_t LUREMHandler{};
-  uint64_t LREMHandler{};
-
   {
-    LUDIVHandler = GetCursorAddress<uint64_t>();
-
-    PushDynamicRegsAndLR();
-
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LUDIV)));
+    LUDIVHandlerAddress = GetCursorAddress<uint64_t>();
 
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
     SpillStaticRegs();
-    blr(x3);
+
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUDIV));
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+    blr(ARMEmitter::Reg::r3);
+#endif
     FillStaticRegs();
 
     // Result is now in x0
@@ -462,14 +455,17 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   }
 
   {
-    LDIVHandler = GetCursorAddress<uint64_t>();
-
-    PushDynamicRegsAndLR();
-
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LDIV)));
+    LDIVHandlerAddress = GetCursorAddress<uint64_t>();
 
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
     SpillStaticRegs();
-    blr(x3);
+
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LDIV));
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+    blr(ARMEmitter::Reg::r3);
+#endif
     FillStaticRegs();
 
     // Result is now in x0
@@ -481,14 +477,17 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   }
 
   {
-    LUREMHandler = GetCursorAddress<uint64_t>();
-
-    PushDynamicRegsAndLR();
-
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LUREM)));
+    LUREMHandlerAddress = GetCursorAddress<uint64_t>();
 
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
     SpillStaticRegs();
-    blr(x3);
+
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LUREM));
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+    blr(ARMEmitter::Reg::r3);
+#endif
     FillStaticRegs();
 
     // Result is now in x0
@@ -500,14 +499,18 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   }
 
   {
-    LREMHandler = GetCursorAddress<uint64_t>();
-
-    PushDynamicRegsAndLR();
-
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.LREM)));
+    LREMHandlerAddress = GetCursorAddress<uint64_t>();
 
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
     SpillStaticRegs();
-    blr(x3);
+
+    ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.AArch64.LREM));
+
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<uint64_t, uint64_t, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+    blr(ARMEmitter::Reg::r3);
+#endif
     FillStaticRegs();
 
     // Result is now in x0
@@ -518,19 +521,16 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     ret();
   }
 
-  place(&l_PagePtr);
-  place(&l_CTX);
-  place(&l_Sleep);
-  place(&l_CompileBlock);
-  place(&l_ExitFunctionLink);
-  place(&l_ExitFunctionLinkThis);
+  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());
@@ -540,54 +540,129 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
     CTX->Symbols.RegisterJITSpace(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr));
   }
 
-  // Setup dispatcher specific pointers that need to be accessed from JIT code
-  {
-    auto &Pointers = ThreadState->CurrentFrame->Pointers.AArch64;
-
-    Pointers.DispatcherLoopTop = AbsoluteLoopTopAddress;
-    Pointers.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
-    Pointers.ThreadStopHandlerSpillSRA = ThreadStopHandlerAddressSpillSRA;
-    Pointers.ThreadPauseHandlerSpillSRA = ThreadPauseHandlerAddressSpillSRA;
-    Pointers.UnimplementedInstructionHandler = UnimplementedInstructionAddress;
-    Pointers.OverflowExceptionHandler = OverflowExceptionInstructionAddress;
-    Pointers.SignalReturnHandler = SignalHandlerReturnAddress;
-    Pointers.L1Pointer = Thread->LookupCache->GetL1Pointer();
-    Pointers.LUDIVHandler = LUDIVHandler;
-    Pointers.LDIVHandler = LDIVHandler;
-    Pointers.LUREMHandler = LUREMHandler;
-    Pointers.LREMHandler = LREMHandler;
-  }
+#ifdef VIXL_DISASSEMBLER
+  const auto DisasmEnd = GetCursorAddress<const vixl::aarch64::Instruction*>();
+  Disasm.DisassembleBuffer(DisasmBegin, DisasmEnd);
+#endif
 }
 
-void Arm64Dispatcher::SpillSRA(void *ucontext, uint32_t IgnoreMask) {
-  for(int i = 0; i < SRA64.size(); i++) {
-    if (IgnoreMask & (1U << SRA64[i].GetCode())) {
-      // Skip this one, it's already spilled
-      continue;
-    }
-    ThreadState->CurrentFrame->State.gregs[i] = ArchHelpers::Context::GetArmReg(ucontext, SRA64[i].GetCode());
-  }
-
-  for(int i = 0; i < SRAFPR.size(); i++) {
-    auto FPR = ArchHelpers::Context::GetArmFPR(ucontext, SRAFPR[i].GetCode());
-    memcpy(&ThreadState->CurrentFrame->State.xmm[i][0], &FPR, sizeof(__uint128_t));
-  }
+#ifdef VIXL_SIMULATOR
+void Arm64Dispatcher::ExecuteDispatch(FEXCore::Core::CpuStateFrame *Frame) {
+  Simulator.WriteXRegister(0, reinterpret_cast<int64_t>(Frame));
+  Simulator.RunFrom(reinterpret_cast<vixl::aarch64::Instruction const*>(DispatchPtr));
 }
 
-#ifdef _M_ARM_64
-
-void InterpreterCore::CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
-  DispatcherConfig config;
-  config.ExecuteBlocksWithCall = true;
-
-  Dispatcher = std::make_unique<Arm64Dispatcher>(ctx, Thread, config);
-  DispatchPtr = Dispatcher->DispatchPtr;
-  CallbackPtr = Dispatcher->CallbackPtr;
-
-  // TODO: It feels wrong to initialize this way
-  ctx->InterpreterCallbackReturn = Dispatcher->ReturnPtr;
+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<vixl::aarch64::Instruction const*>(CallbackPtr));
 }
 
 #endif
 
+size_t Arm64Dispatcher::GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) {
+  FEXCore::ARMEmitter::Emitter emit{CodeBuffer, MaxGDBPauseCheckSize};
+
+  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(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(ARMEmitter::Size::i32Bit, ARMEmitter::Reg::r0, &RunBlock);
+  {
+    ARMEmitter::ForwardLabel l_GuestRIP;
+    // Make sure RIP is syncronized to the context
+    emit.ldr(ARMEmitter::XReg::x0, &l_GuestRIP);
+    emit.str(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, State.rip));
+
+    // Stop the thread
+    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);
+
+  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");
+
+  FEXCore::ARMEmitter::Emitter emit{CodeBuffer, MaxInterpreterTrampolineSize};
+  ARMEmitter::ForwardLabel InlineIRData;
+
+  emit.mov(ARMEmitter::XReg::x0, STATE);
+  emit.adr(ARMEmitter::Reg::r1, &InlineIRData);
+
+  emit.ldr(ARMEmitter::XReg::x3, STATE_PTR(CpuStateFrame, Pointers.Interpreter.FragmentExecuter));
+  emit.blr(ARMEmitter::Reg::r3);
+
+  emit.ldr(ARMEmitter::XReg::x0, STATE_PTR(CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+  emit.br(ARMEmitter::Reg::r0);
+
+  emit.Bind(&InlineIRData);
+
+  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].Idx())) {
+      // Skip this one, it's already spilled
+      continue;
+    }
+    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].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].Idx());
+      memcpy(&Thread->CurrentFrame->State.xmm.sse.data[i][0], &FPR, sizeof(__uint128_t));
+    }
+  }
+}
+
+void Arm64Dispatcher::InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) {
+  // Setup dispatcher specific pointers that need to be accessed from JIT code
+  {
+    auto &Common = Thread->CurrentFrame->Pointers.Common;
+
+    Common.DispatcherLoopTop = AbsoluteLoopTopAddress;
+    Common.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
+    Common.ExitFunctionLinker = ExitFunctionLinkerAddress;
+    Common.ThreadStopHandlerSpillSRA = ThreadStopHandlerAddressSpillSRA;
+    Common.ThreadPauseHandlerSpillSRA = ThreadPauseHandlerAddressSpillSRA;
+    Common.GuestSignal_SIGILL = GuestSignal_SIGILL;
+    Common.GuestSignal_SIGTRAP = GuestSignal_SIGTRAP;
+    Common.GuestSignal_SIGSEGV = GuestSignal_SIGSEGV;
+    Common.SignalReturnHandler = SignalHandlerReturnAddress;
+
+    auto &AArch64 = Thread->CurrentFrame->Pointers.AArch64;
+    AArch64.LUDIVHandler = LUDIVHandlerAddress;
+    AArch64.LDIVHandler = LDIVHandlerAddress;
+    AArch64.LUREMHandler = LUREMHandlerAddress;
+    AArch64.LREMHandler = LREMHandlerAddress;
+  }
+}
+
+std::unique_ptr<Dispatcher> Dispatcher::CreateArm64(FEXCore::Context::Context *CTX, const DispatcherConfig &Config) {
+  return std::make_unique<Arm64Dispatcher>(CTX, Config);
+}
+
 }

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

@@ -3,6 +3,10 @@
 #include "Interface/Core/ArchHelpers/Arm64Emitter.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 
+#ifdef VIXL_SIMULATOR
+#include <aarch64/simulator-aarch64.h>
+#endif
+
 namespace FEXCore::Context {
 struct Context;
 }
@@ -11,14 +15,39 @@ 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 {
   public:
-    Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config);
+    Arm64Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &config);
+    void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) override;
+    size_t GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) override;
+    size_t GenerateInterpreterTrampoline(uint8_t *CodeBuffer) override;
+
+#ifdef VIXL_SIMULATOR
+  void ExecuteDispatch(FEXCore::Core::CpuStateFrame *Frame) override;
+  void ExecuteJITCallback(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP) override;
+#endif
+
+  void EmitDispatcher();
 
   protected:
-    void SpillSRA(void *ucontext, uint32_t IgnoreMask) override;
+    void SpillSRA(FEXCore::Core::InternalThreadState *Thread, void *ucontext, uint32_t IgnoreMask) override;
+
+  private:
+    // Long division helpers
+    uint64_t LUDIVHandlerAddress{};
+    uint64_t LDIVHandlerAddress{};
+    uint64_t LUREMHandlerAddress{};
+    uint64_t LREMHandlerAddress{};
+
+#ifdef VIXL_SIMULATOR
+    vixl::aarch64::Decoder Decoder;
+    vixl::aarch64::Simulator Simulator;
+#endif
 };
 
 }

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

@@ -1,4 +1,5 @@
 
+#include "Interface/Context/Context.h"
 #include "Interface/Core/ArchHelpers/MContext.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/X86HelperGen.h"
@@ -39,7 +40,7 @@ void Dispatcher::SleepThread(FEXCore::Context::Context *ctx, FEXCore::Core::CpuS
   ctx->IdleWaitCV.notify_all();
 }
 
-ArchHelpers::Context::ContextBackup* Dispatcher::StoreThreadState(int Signal, void *ucontext) {
+ArchHelpers::Context::ContextBackup* Dispatcher::StoreThreadState(FEXCore::Core::InternalThreadState *Thread, int Signal, void *ucontext) {
   // We can end up getting a signal at any point in our host state
   // Jump to a handler that saves all state so we can safely return
   uint64_t OldSP = ArchHelpers::Context::GetSp(ucontext);
@@ -64,7 +65,7 @@ ArchHelpers::Context::ContextBackup* Dispatcher::StoreThreadState(int Signal, vo
   // Save guest state
   // We can't guarantee if registers are in context or host GPRs
   // So we need to save everything
-  memcpy(&Context->GuestState, ThreadState->CurrentFrame, sizeof(FEXCore::Core::CPUState));
+  memcpy(&Context->GuestState, Thread->CurrentFrame, sizeof(FEXCore::Core::CPUState));
 
   // Set the new SP
   ArchHelpers::Context::SetSp(ucontext, NewSP);
@@ -81,13 +82,13 @@ ArchHelpers::Context::ContextBackup* Dispatcher::StoreThreadState(int Signal, vo
   Context->SigInfoLocation = 0;
 
   // Store fault to top status and then reset it
-  Context->FaultToTopAndGeneratedException = SynchronousFaultData.FaultToTopAndGeneratedException;
-  SynchronousFaultData.FaultToTopAndGeneratedException = false;
+  Context->FaultToTopAndGeneratedException = Thread->CurrentFrame->SynchronousFaultData.FaultToTopAndGeneratedException;
+  Thread->CurrentFrame->SynchronousFaultData.FaultToTopAndGeneratedException = false;
 
   return Context;
 }
 
-void Dispatcher::RestoreThreadState(void *ucontext) {
+void Dispatcher::RestoreThreadState(FEXCore::Core::InternalThreadState *Thread, void *ucontext) {
   uint64_t OldSP{};
   if (CTX->Config.Core() == FEXCore::Config::CONFIG_IRJIT) {
     OldSP = ArchHelpers::Context::GetSp(ucontext);
@@ -98,17 +99,18 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
     SignalFrames.pop();
   }
 
+  const bool IsAVXEnabled = CTX->Config.EnableAVX;
   uintptr_t NewSP = OldSP;
   auto Context = reinterpret_cast<ArchHelpers::Context::ContextBackup*>(NewSP);
 
   // First thing, reset the guest state
-  memcpy(ThreadState->CurrentFrame, &Context->GuestState, sizeof(FEXCore::Core::CPUState));
+  memcpy(Thread->CurrentFrame, &Context->GuestState, sizeof(FEXCore::Core::CPUState));
 
   // Now restore host state
   ArchHelpers::Context::RestoreContext(ucontext, Context);
 
   if (Context->UContextLocation) {
-    auto Frame = ThreadState->CurrentFrame;
+    auto Frame = Thread->CurrentFrame;
 
     if (Context->Flags &ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_INJIT) {
       // XXX: Unsupported since it needs state reconstruction
@@ -132,7 +134,7 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
         Frame->State.rip = guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_RIP];
         // XXX: Full context setting
         uint32_t eflags = guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_EFL];
-        for (size_t i = 0; i < 32; ++i) {
+        for (size_t i = 0; i < Core::CPUState::NUM_EFLAG_BITS; ++i) {
           Frame->State.flags[i] = (eflags & (1U << i)) ? 1 : 0;
         }
 
@@ -158,10 +160,22 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
             COPY_REG(RCX);
             COPY_REG(RSP);
 #undef COPY_REG
-        FEXCore::x86_64::_libc_fpstate *fpstate = reinterpret_cast<FEXCore::x86_64::_libc_fpstate*>(guest_uctx->uc_mcontext.fpregs);
+        auto *xstate = reinterpret_cast<x86_64::xstate*>(guest_uctx->uc_mcontext.fpregs);
+        auto *fpstate = &xstate->fpstate;
+
         // Copy float registers
         memcpy(Frame->State.mm, fpstate->_st, sizeof(Frame->State.mm));
-        memcpy(Frame->State.xmm, fpstate->_xmm, sizeof(Frame->State.xmm));
+
+        if (IsAVXEnabled) {
+          for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
+            memcpy(&Frame->State.xmm.avx.data[i][0], &fpstate->_xmm[i], sizeof(__uint128_t));
+          }
+          for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
+            memcpy(&Frame->State.xmm.avx.data[i][2], &xstate->ymmh.ymmh_space[i], sizeof(__uint128_t));
+          }
+        } else {
+          memcpy(Frame->State.xmm.sse.data, fpstate->_xmm, sizeof(Frame->State.xmm.sse.data));
+        }
 
         // FCW store default
         Frame->State.FCW = fpstate->fcw;
@@ -190,7 +204,7 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
         // XXX: Full context setting
         // First 32-bytes of flags is EFLAGS broken out
         uint32_t eflags = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EFL];
-        for (size_t i = 0; i < 32; ++i) {
+        for (size_t i = 0; i < Core::CPUState::NUM_EFLAG_BITS; ++i) {
           Frame->State.flags[i] = (eflags & (1U << i)) ? 1 : 0;
         }
 
@@ -198,12 +212,20 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
         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);
@@ -215,16 +237,26 @@ void Dispatcher::RestoreThreadState(void *ucontext) {
         COPY_REG(RCX);
         COPY_REG(RSP);
 #undef COPY_REG
-        FEXCore::x86::_libc_fpstate *fpstate = reinterpret_cast<FEXCore::x86::_libc_fpstate*>(guest_uctx->uc_mcontext.fpregs);
+        auto *xstate = reinterpret_cast<x86::xstate*>(guest_uctx->uc_mcontext.fpregs);
+        auto *fpstate = &xstate->fpstate;
 
         // Copy float registers
-        for (size_t i = 0; i < 8; ++i) {
+        for (size_t i = 0; i < Core::CPUState::NUM_MMS; ++i) {
           // 32-bit st register size is only 10 bytes. Not padded to 16byte like x86-64
           memcpy(&Frame->State.mm[i], &fpstate->_st[i], 10);
         }
 
         // Extended XMM state
-        memcpy(fpstate->_xmm, Frame->State.xmm, sizeof(Frame->State.xmm));
+        if (IsAVXEnabled) {
+          for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
+            memcpy(&fpstate->_xmm[i], &Frame->State.xmm.avx.data[i][0], sizeof(__uint128_t));
+          }
+          for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
+            memcpy(&xstate->ymmh.ymmh_space[i], &Frame->State.xmm.avx.data[i][2], sizeof(__uint128_t));
+          }
+        } else {
+          memcpy(Frame->State.xmm.sse.data, fpstate->_xmm, sizeof(Frame->State.xmm.sse.data));
+        }
 
         // FCW store default
         Frame->State.FCW = fpstate->fcw;
@@ -256,15 +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;
   }
@@ -273,14 +304,34 @@ static uint32_t ConvertSignalToError(int Signal, siginfo_t *HostSigInfo) {
   return 0;
 }
 
-bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) {
-  auto ContextBackup = StoreThreadState(Signal, ucontext);
+template <typename T>
+static void SetXStateInfo(T* xstate, bool is_avx_enabled) {
+  auto* fpstate = &xstate->fpstate;
 
-  auto Frame = ThreadState->CurrentFrame;
+  fpstate->sw_reserved.magic1 = x86_64::fpx_sw_bytes::FP_XSTATE_MAGIC;
+  fpstate->sw_reserved.extended_size = is_avx_enabled ? sizeof(T) : 0;
+
+  fpstate->sw_reserved.xfeatures |= x86_64::fpx_sw_bytes::FEATURE_FP |
+                                    x86_64::fpx_sw_bytes::FEATURE_SSE;
+  if (is_avx_enabled) {
+    fpstate->sw_reserved.xfeatures |= x86_64::fpx_sw_bytes::FEATURE_YMM;
+  }
+
+  fpstate->sw_reserved.xstate_size = fpstate->sw_reserved.extended_size;
+
+  if (is_avx_enabled) {
+    xstate->xstate_hdr.xfeatures = 0;
+  }
+}
+
+bool Dispatcher::HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) {
+  auto ContextBackup = StoreThreadState(Thread, Signal, ucontext);
+
+  auto Frame = Thread->CurrentFrame;
 
   // Ref count our faults
   // We use this to track if it is safe to clear cache
-  ++SignalHandlerRefCounter;
+  ++Thread->CurrentFrame->SignalHandlerRefCounter;
 
   uint64_t OldPC = ArchHelpers::Context::GetPc(ucontext);
   // Set the new PC
@@ -292,14 +343,15 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
   uint64_t NewGuestSP = OldGuestSP;
 
   // Pulling from context here
-  bool Is64BitMode = CTX->Config.Is64BitMode;
-  uint64_t SignalReturn = CTX->X86CodeGen.SignalReturn;
+  const bool Is64BitMode = CTX->Config.Is64BitMode;
+  const bool IsAVXEnabled = CTX->Config.EnableAVX;
+  const uint64_t SignalReturn = CTX->X86CodeGen.SignalReturn;
 
   // Spill the SRA regardless of signal handler type
   // We are going to be returning to the top of the dispatcher which will fill again
   // Otherwise we might load garbage
-  if (SRAEnabled) {
-    if (IsAddressInJITCode(OldPC, false)) {
+  if (config.StaticRegisterAllocation) {
+    if (Thread->CPUBackend->IsAddressInCodeBuffer(OldPC)) {
       uint32_t IgnoreMask{};
 #ifdef _M_ARM_64
       if (Frame->InSyscallInfo != 0) {
@@ -323,11 +375,11 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
 #endif
 
       // We are in jit, SRA must be spilled
-      SpillSRA(ucontext, IgnoreMask);
+      SpillSRA(Thread, ucontext, IgnoreMask);
 
       ContextBackup->Flags |= ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_INJIT;
     } else {
-      if (!IsAddressInJITCode(OldPC, true)) {
+      if (!IsAddressInDispatcher(OldPC)) {
         // This is likely to cause issues but in some cases it isn't fatal
         // This can also happen if we have put a signal on hold, then we just reenabled the signal
         // So we are in the syscall handler
@@ -373,8 +425,13 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
   if (GuestAction->sa_flags & SA_SIGINFO) {
     // Setup ucontext a bit
     if (Is64BitMode) {
-      NewGuestSP -= sizeof(FEXCore::x86_64::_libc_fpstate);
-      NewGuestSP = AlignDown(NewGuestSP, alignof(FEXCore::x86_64::_libc_fpstate));
+      if (IsAVXEnabled) {
+        NewGuestSP -= sizeof(x86_64::xstate);
+        NewGuestSP = AlignDown(NewGuestSP, alignof(x86_64::xstate));
+      } else {
+        NewGuestSP -= sizeof(x86_64::_libc_fpstate);
+        NewGuestSP = AlignDown(NewGuestSP, alignof(x86_64::_libc_fpstate));
+      }
       uint64_t FPStateLocation = NewGuestSP;
 
       NewGuestSP -= sizeof(FEXCore::x86_64::ucontext_t);
@@ -396,8 +453,9 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       guest_uctx->uc_flags = FEXCore::x86_64::UC_FP_XSTATE;
 
       // Pointer to where the fpreg memory is
-      guest_uctx->uc_mcontext.fpregs = reinterpret_cast<FEXCore::x86_64::_libc_fpstate*>(FPStateLocation);
-      FEXCore::x86_64::_libc_fpstate *fpstate = reinterpret_cast<FEXCore::x86_64::_libc_fpstate*>(FPStateLocation);
+      guest_uctx->uc_mcontext.fpregs = reinterpret_cast<x86_64::_libc_fpstate*>(FPStateLocation);
+      auto *xstate = reinterpret_cast<x86_64::xstate*>(FPStateLocation);
+      SetXStateInfo(xstate, IsAVXEnabled);
 
       guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_RIP] = Frame->State.rip;
       guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_EFL] = 0;
@@ -409,15 +467,16 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       *guest_siginfo = *HostSigInfo;
 
       if (ContextBackup->FaultToTopAndGeneratedException) {
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_TRAPNO] = SynchronousFaultData.TrapNo;
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_ERR] = SynchronousFaultData.err_code;
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_TRAPNO] = Frame->SynchronousFaultData.TrapNo;
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_ERR] = Frame->SynchronousFaultData.err_code;
 
         // Overwrite si_code
-        guest_siginfo->si_code = SynchronousFaultData.si_code;
+        guest_siginfo->si_code = Thread->CurrentFrame->SynchronousFaultData.si_code;
+        Signal = Frame->SynchronousFaultData.Signal;
       }
       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;
@@ -442,9 +501,21 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       COPY_REG(RSP);
 #undef COPY_REG
 
+      auto* fpstate = &xstate->fpstate;
+
       // Copy float registers
       memcpy(fpstate->_st, Frame->State.mm, sizeof(Frame->State.mm));
-      memcpy(fpstate->_xmm, Frame->State.xmm, sizeof(Frame->State.xmm));
+
+      if (IsAVXEnabled) {
+        for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
+          memcpy(&fpstate->_xmm[i], &Frame->State.xmm.avx.data[i][0], sizeof(__uint128_t));
+        }
+        for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
+          memcpy(&xstate->ymmh.ymmh_space[i], &Frame->State.xmm.avx.data[i][2], sizeof(__uint128_t));
+        }
+      } else {
+        memcpy(fpstate->_xmm, Frame->State.xmm.sse.data, sizeof(Frame->State.xmm.sse.data));
+      }
 
       // FCW store default
       fpstate->fcw = Frame->State.FCW;
@@ -469,8 +540,13 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
     else {
       ContextBackup->Flags |= ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_32BIT;
 
-      NewGuestSP -= sizeof(FEXCore::x86::_libc_fpstate);
-      NewGuestSP = AlignDown(NewGuestSP, alignof(FEXCore::x86::_libc_fpstate));
+      if (IsAVXEnabled) {
+        NewGuestSP -= sizeof(x86::xstate);
+        NewGuestSP = AlignDown(NewGuestSP, alignof(x86::xstate));
+      } else {
+        NewGuestSP -= sizeof(x86::_libc_fpstate);
+        NewGuestSP = AlignDown(NewGuestSP, alignof(x86::_libc_fpstate));
+      }
       uint64_t FPStateLocation = NewGuestSP;
 
       NewGuestSP -= sizeof(FEXCore::x86::ucontext_t);
@@ -493,27 +569,30 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
 
       // Pointer to where the fpreg memory is
       guest_uctx->uc_mcontext.fpregs = static_cast<uint32_t>(FPStateLocation);
-      FEXCore::x86::_libc_fpstate *fpstate = reinterpret_cast<FEXCore::x86::_libc_fpstate*>(FPStateLocation);
+      auto *xstate = reinterpret_cast<x86::xstate*>(FPStateLocation);
+      SetXStateInfo(xstate, IsAVXEnabled);
+
+      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;
 
-      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;
       if (ContextBackup->FaultToTopAndGeneratedException) {
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = SynchronousFaultData.TrapNo;
-        guest_siginfo->si_code = SynchronousFaultData.si_code;
-        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ERR] = SynchronousFaultData.err_code;
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = Frame->SynchronousFaultData.TrapNo;
+        guest_siginfo->si_code = Frame->SynchronousFaultData.si_code;
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ERR] = Frame->SynchronousFaultData.err_code;
+        Signal = Frame->SynchronousFaultData.Signal;
       }
       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];
@@ -527,15 +606,26 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       COPY_REG(RSP);
 #undef COPY_REG
 
+      auto *fpstate = &xstate->fpstate;
+
       // Copy float registers
-      for (size_t i = 0; i < 8; ++i) {
+      for (size_t i = 0; i < Core::CPUState::NUM_MMS; ++i) {
         // 32-bit st register size is only 10 bytes. Not padded to 16byte like x86-64
         memcpy(&fpstate->_st[i], &Frame->State.mm[i], 10);
       }
 
       // Extended XMM state
       fpstate->status = FEXCore::x86::fpstate_magic::MAGIC_XFPSTATE;
-      memcpy(fpstate->_xmm, Frame->State.xmm, sizeof(Frame->State.xmm));
+      if (IsAVXEnabled) {
+        for (size_t i = 0; i < std::size(Frame->State.xmm.avx.data); i++) {
+          memcpy(&fpstate->_xmm[i], &Frame->State.xmm.avx.data[i][0], sizeof(__uint128_t));
+        }
+        for (size_t i = 0; i < std::size(Frame->State.xmm.avx.data); i++) {
+          memcpy(&xstate->ymmh.ymmh_space[i], &Frame->State.xmm.avx.data[i][2], sizeof(__uint128_t));
+        }
+      } else {
+        memcpy(fpstate->_xmm, Frame->State.xmm.sse.data, sizeof(Frame->State.xmm.sse.data));
+      }
 
       // FCW store default
       fpstate->fcw = Frame->State.FCW;
@@ -618,14 +708,14 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
   else {
     NewGuestSP -= 4;
     *(uint32_t*)NewGuestSP = SignalReturn;
-    LOGMAN_THROW_A_FMT(SignalReturn < 0x1'0000'0000ULL, "This needs to be below 4GB");
+    LOGMAN_THROW_AA_FMT(SignalReturn < 0x1'0000'0000ULL, "This needs to be below 4GB");
     Frame->State.gregs[FEXCore::X86State::REG_RSP] = NewGuestSP;
   }
 
   // The guest starts its signal frame with a zero initialized FPU
   // Set that up now. Little bit costly but it's a requirement
   // This state will be restored on rt_sigreturn
-  memset(Frame->State.xmm, 0, sizeof(Frame->State.xmm));
+  memset(Frame->State.xmm.avx.data, 0, sizeof(Frame->State.xmm));
   memset(Frame->State.mm, 0, sizeof(Frame->State.mm));
   Frame->State.FCW = 0x37F;
   Frame->State.FTW = 0xFFFF;
@@ -633,44 +723,44 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
   return true;
 }
 
-bool Dispatcher::HandleSIGILL(int Signal, void *info, void *ucontext) {
+bool Dispatcher::HandleSIGILL(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) {
 
   if (ArchHelpers::Context::GetPc(ucontext) == SignalHandlerReturnAddress) {
-    RestoreThreadState(ucontext);
+    RestoreThreadState(Thread, ucontext);
 
     // Ref count our faults
     // We use this to track if it is safe to clear cache
-    --SignalHandlerRefCounter;
+    --Thread->CurrentFrame->SignalHandlerRefCounter;
     return true;
   }
 
   if (ArchHelpers::Context::GetPc(ucontext) == PauseReturnInstruction) {
-    RestoreThreadState(ucontext);
+    RestoreThreadState(Thread, ucontext);
 
     // Ref count our faults
     // We use this to track if it is safe to clear cache
-    --SignalHandlerRefCounter;
+    --Thread->CurrentFrame->SignalHandlerRefCounter;
     return true;
   }
 
   return false;
 }
 
-bool Dispatcher::HandleSignalPause(int Signal, void *info, void *ucontext) {
-  FEXCore::Core::SignalEvent SignalReason = ThreadState->SignalReason.load();
-  auto Frame = ThreadState->CurrentFrame;
+bool Dispatcher::HandleSignalPause(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) {
+  FEXCore::Core::SignalEvent SignalReason = Thread->SignalReason.load();
+  auto Frame = Thread->CurrentFrame;
 
   if (SignalReason == FEXCore::Core::SignalEvent::Pause) {
     // Store our thread state so we can come back to this
-    StoreThreadState(Signal, ucontext);
+    StoreThreadState(Thread, Signal, ucontext);
 
-    if (SRAEnabled && IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), false)) {
+    if (config.StaticRegisterAllocation && Thread->CPUBackend->IsAddressInCodeBuffer(ArchHelpers::Context::GetPc(ucontext))) {
       // We are in jit, SRA must be spilled
       ArchHelpers::Context::SetPc(ucontext, ThreadPauseHandlerAddressSpillSRA);
     } else {
-      if (SRAEnabled) {
+      if (config.StaticRegisterAllocation) {
         // We are in non-jit, SRA is already spilled
-        LOGMAN_THROW_A_FMT(!IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), true),
+        LOGMAN_THROW_A_FMT(!IsAddressInDispatcher(ArchHelpers::Context::GetPc(ucontext)),
                            "Signals in dispatcher have unsynchronized context");
       }
       ArchHelpers::Context::SetPc(ucontext, ThreadPauseHandlerAddress);
@@ -681,9 +771,9 @@ bool Dispatcher::HandleSignalPause(int Signal, void *info, void *ucontext) {
 
     // Ref count our faults
     // We use this to track if it is safe to clear cache
-    ++SignalHandlerRefCounter;
+    ++Thread->CurrentFrame->SignalHandlerRefCounter;
 
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
+    Thread->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
     return true;
   }
 
@@ -694,16 +784,16 @@ bool Dispatcher::HandleSignalPause(int Signal, void *info, void *ucontext) {
     ArchHelpers::Context::SetSp(ucontext, Frame->ReturningStackLocation);
 
     // Our ref counting doesn't matter anymore
-    SignalHandlerRefCounter = 0;
+    Thread->CurrentFrame->SignalHandlerRefCounter = 0;
 
     // Set the new PC
-    if (SRAEnabled && IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), false)) {
+    if (config.StaticRegisterAllocation && Thread->CPUBackend->IsAddressInCodeBuffer(ArchHelpers::Context::GetPc(ucontext))) {
       // We are in jit, SRA must be spilled
       ArchHelpers::Context::SetPc(ucontext, ThreadStopHandlerAddressSpillSRA);
     } else {
-      if (SRAEnabled) {
+      if (config.StaticRegisterAllocation) {
         // We are in non-jit, SRA is already spilled
-        LOGMAN_THROW_A_FMT(!IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), true),
+        LOGMAN_THROW_A_FMT(!IsAddressInDispatcher(ArchHelpers::Context::GetPc(ucontext)),
                            "Signals in dispatcher have unsynchronized context");
       }
       ArchHelpers::Context::SetPc(ucontext, ThreadStopHandlerAddress);
@@ -712,24 +802,24 @@ bool Dispatcher::HandleSignalPause(int Signal, void *info, void *ucontext) {
     // We need to be a little bit careful here
     // If we were already paused (due to GDB) and we are immediately stopping (due to gdb kill)
     // Then we need to ensure we don't double decrement our idle thread counter
-    if (ThreadState->RunningEvents.ThreadSleeping) {
+    if (Thread->RunningEvents.ThreadSleeping) {
       // If the thread was sleeping then its idle counter was decremented
       // Reincrement it here to not break logic
-      ++ThreadState->CTX->IdleWaitRefCount;
+      ++Thread->CTX->IdleWaitRefCount;
     }
 
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
+    Thread->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
     return true;
   }
 
   if (SignalReason == FEXCore::Core::SignalEvent::Return) {
-    RestoreThreadState(ucontext);
+    RestoreThreadState(Thread, ucontext);
 
     // Ref count our faults
     // We use this to track if it is safe to clear cache
-    --SignalHandlerRefCounter;
+    --Thread->CurrentFrame->SignalHandlerRefCounter;
 
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
+    Thread->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
     return true;
   }
 
@@ -748,28 +838,4 @@ uint64_t Dispatcher::GetCompileBlockPtr() {
   return CompileBlockPtr.Data;
 }
 
-void Dispatcher::RemoveCodeBuffer(uint8_t* start_to_remove) {
-  for (auto iter = CodeBuffers.begin(); iter != CodeBuffers.end(); ++iter) {
-    auto [start, end] = *iter;
-    if (start == reinterpret_cast<uint64_t>(start_to_remove)) {
-      CodeBuffers.erase(iter);
-      return;
-    }
-  }
-}
-
-bool Dispatcher::IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher) const {
-  for (auto [start, end] : CodeBuffers) {
-    if (Address >= start && Address < end) {
-      return true;
-    }
-  }
-
-  if (IncludeDispatcher && IsAddressInDispatcher(Address)) {
-    return true;
-  }
-
-  return false;
-}
-
 }

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

@@ -1,8 +1,6 @@
 #pragma once
 
 #include <FEXCore/Core/CPUBackend.h>
-
-#include "Interface/Context/Context.h"
 #include "Interface/Core/ArchHelpers/MContext.h"
 
 #include <cstdint>
@@ -21,21 +19,19 @@ struct CpuStateFrame;
 struct InternalThreadState;
 }
 
+namespace FEXCore::Context {
+struct Context;
+}
+
 namespace FEXCore::CPU {
 
 struct DispatcherConfig {
-  bool ExecuteBlocksWithCall = false;
-  uintptr_t ExitFunctionLink = 0;
-  uintptr_t ExitFunctionLinkThis = 0;
-  bool StaticRegisterAssignment = false;
+  bool StaticRegisterAllocation = false;
 };
 
 class Dispatcher {
 public:
   virtual ~Dispatcher() = default;
-  CPUBackend::AsmDispatch DispatchPtr;
-  CPUBackend::JITCallback CallbackPtr;
-  FEXCore::Context::Context::IntCallbackReturn ReturnPtr;
 
   /**
    * @name Dispatch Helper functions
@@ -48,61 +44,70 @@ public:
   uint64_t ThreadPauseHandlerAddressSpillSRA{};
   uint64_t ExitFunctionLinkerAddress{};
   uint64_t SignalHandlerReturnAddress{};
-  uint64_t UnimplementedInstructionAddress{};
-  uint64_t OverflowExceptionInstructionAddress{};
+  uint64_t GuestSignal_SIGILL{};
+  uint64_t GuestSignal_SIGTRAP{};
+  uint64_t GuestSignal_SIGSEGV{};
+  uint64_t IntCallbackReturnAddress{};
 
   uint64_t PauseReturnInstruction{};
 
   /**  @} */
 
-  uint32_t SignalHandlerRefCounter{};
-  struct SynchronousFaultDataStruct {
-    bool FaultToTopAndGeneratedException{};
-    uint32_t TrapNo;
-    uint32_t err_code;
-    uint32_t si_code;
-  } SynchronousFaultData;
-
   uint64_t Start{};
   uint64_t End{};
 
-  bool HandleGuestSignal(int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack);
-  bool HandleSIGILL(int Signal, void *info, void *ucontext);
-  bool HandleSignalPause(int Signal, void *info, void *ucontext);
+  bool HandleGuestSignal(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack);
+  bool HandleSIGILL(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext);
+  bool HandleSignalPause(FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext);
 
-  void RegisterCodeBuffer(uint8_t* start, size_t size) {
-    CodeBuffers.emplace_back(reinterpret_cast<uint64_t>(start),
-        reinterpret_cast<uint64_t>(start + size));
-  }
-
-  void RemoveCodeBuffer(uint8_t* start);
-
-  bool IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher = true) const;
   bool IsAddressInDispatcher(uint64_t Address) const {
     return Address >= Start && Address < End;
   }
 
-protected:
-  Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
-    : CTX {ctx}
-    , ThreadState {Thread} {}
+  virtual void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) = 0;
 
-  ArchHelpers::Context::ContextBackup* StoreThreadState(int Signal, void *ucontext);
-  void RestoreThreadState(void *ucontext);
+  // These are across all arches for now
+  static constexpr size_t MaxGDBPauseCheckSize = 128;
+  static constexpr size_t MaxInterpreterTrampolineSize = 128;
+
+  virtual size_t GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) = 0;
+  virtual size_t GenerateInterpreterTrampoline(uint8_t *CodeBuffer) = 0;
+
+  static std::unique_ptr<Dispatcher> CreateX86(FEXCore::Context::Context *CTX, const DispatcherConfig &Config);
+  static std::unique_ptr<Dispatcher> CreateArm64(FEXCore::Context::Context *CTX, const DispatcherConfig &Config);
+
+  virtual void ExecuteDispatch(FEXCore::Core::CpuStateFrame *Frame) {
+    DispatchPtr(Frame);
+  }
+
+  virtual void ExecuteJITCallback(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP) {
+    CallbackPtr(Frame, RIP);
+  }
+
+protected:
+  Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &Config)
+    : CTX {ctx}
+    , config {Config}
+    {}
+
+  ArchHelpers::Context::ContextBackup* StoreThreadState(FEXCore::Core::InternalThreadState *Thread, int Signal, void *ucontext);
+  void RestoreThreadState(FEXCore::Core::InternalThreadState *Thread, void *ucontext);
   std::stack<uint64_t, std::vector<uint64_t>> SignalFrames;
 
-  bool SRAEnabled = false;
-  virtual void SpillSRA(void *ucontext, uint32_t IgnoreMask) {}
+  virtual void SpillSRA(FEXCore::Core::InternalThreadState *Thread, void *ucontext, uint32_t IgnoreMask) {}
 
   FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *ThreadState;
+  DispatcherConfig config;
 
   static void SleepThread(FEXCore::Context::Context *ctx, FEXCore::Core::CpuStateFrame *Frame);
 
   static uint64_t GetCompileBlockPtr();
 
-private:
-  std::vector<std::tuple<uint64_t, uint64_t>> CodeBuffers; // Start, End
+  using AsmDispatch = void(*)(FEXCore::Core::CpuStateFrame *Frame);
+  using JITCallback = void(*)(FEXCore::Core::CpuStateFrame *Frame, uint64_t RIP);
+
+  AsmDispatch DispatchPtr;
+  JITCallback CallbackPtr;
 };
 
 }

External/FEXCore/Source/Interface/Core/Dispatcher/X86Dispatcher.cpp

@@ -18,21 +18,26 @@
 #include <stddef.h>
 #include <stdint.h>
 #include <sys/mman.h>
-#include "xbyak/xbyak.h"
+#include <xbyak/xbyak.h>
+
+#define STATE_PTR(STATE_TYPE, FIELD) \
+  [STATE + offsetof(FEXCore::Core::STATE_TYPE, FIELD)]
 
 namespace FEXCore::CPU {
 static constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096;
 #define STATE r14
 
-X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config)
-  : Dispatcher(ctx, Thread)
+X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &config)
+  : Dispatcher(ctx, config)
   , Xbyak::CodeGenerator(MAX_DISPATCHER_CODE_SIZE,
       FEXCore::Allocator::mmap(nullptr, MAX_DISPATCHER_CODE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0),
       nullptr) {
 
+  LOGMAN_THROW_AA_FMT(!config.StaticRegisterAllocation, "X86 dispatcher does not support SRA");
+
   using namespace Xbyak;
   using namespace Xbyak::util;
-  DispatchPtr = getCurr<CPUBackend::AsmDispatch>();
+  DispatchPtr = getCurr<AsmDispatch>();
 
   // Temp registers
   // rax, rcx, rdx, rsi, r8, r9,
@@ -78,11 +83,10 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
 
   // Save this stack pointer so we can cleanly shutdown the emulation with a long jump
   // regardless of where we were in the stack
-  mov(qword [rdi + offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)], rsp);
+  mov(qword STATE_PTR(CpuStateFrame, ReturningStackLocation), rsp);
 
   Label LoopTop;
   Label FullLookup;
-  Label CallBlock;
   Label NoBlock;
   Label ExitBlock;
   Label ThreadPauseHandler;
@@ -92,29 +96,24 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
 
   {
     // Load our RIP
-    mov(rdx, qword [STATE + offsetof(FEXCore::Core::CPUState, rip)]);
+    mov(rdx, qword STATE_PTR(CPUState, rip));
 
     // L1 Cache
-    mov(r13, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.L1Pointer)]);
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
     mov(rax, rdx);
 
     and_(rax, LookupCache::L1_ENTRIES_MASK);
     shl(rax, 4);
-    cmp(qword[r13 + rax + 8], rdx);
+    cmp(qword[r13 + rax + offsetof(FEXCore::LookupCache::LookupCacheEntry, GuestCode)], rdx);
     jne(FullLookup);
 
-    if (!config.ExecuteBlocksWithCall) {
-      jmp(qword[r13 + rax + 0]);
-    } else {
-      mov(rax, qword[r13 + rax + 0]);
-      jmp(CallBlock);
-    }
+    jmp(qword[r13 + rax + offsetof(FEXCore::LookupCache::LookupCacheEntry, HostCode)]);
 
     L(FullLookup);
-    mov(r13, Thread->LookupCache->GetPagePointer());
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L2Pointer));
 
     // Full lookup
-    uint64_t VirtualMemorySize = Thread->LookupCache->GetVirtualMemorySize();
+    uint64_t VirtualMemorySize = CTX->Config.VirtualMemSize;
     mov(rax, rdx);
     mov(rbx, VirtualMemorySize - 1);
     and_(rax, rbx);
@@ -143,7 +142,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     je(NoBlock);
 
     // Update L1
-    mov(r13, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.L1Pointer)]);
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
     mov(rcx, rdx);
     and_(rcx, LookupCache::L1_ENTRIES_MASK);
     shl(rcx, 1);
@@ -151,30 +150,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     mov(qword[r13 + rcx*8 + 0], rax);
 
     // Real block if we made it here
-    if (!config.ExecuteBlocksWithCall) {
-      jmp(rax);
-    } else {
-      L(CallBlock);
-      mov(rdi, STATE);
-      call(rax);
-
-      if (CTX->GetGdbServerStatus()) {
-        // 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(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
-        mov(rax, qword [STATE + (offsetof(FEXCore::Core::InternalThreadState, CTX))]);
-
-        // If the value == 0 then branch to the top
-        cmp(dword [rax + (offsetof(FEXCore::Context::Context, Config.RunningMode))], 0);
-        je(LoopTop);
-        // Else we need to pause now
-        jmp(ThreadPauseHandler);
-        ud2();
-      }
-      else {
-        jmp(LoopTop);
-      }
-    }
+    jmp(rax);
   }
 
   {
@@ -282,13 +258,11 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
       mov(rax, r9);
     }
 
-    // {rdi, rsi, rdx}
-    mov(rdi, config.ExitFunctionLinkThis);
-    mov(rsi, STATE);
-    mov(rdx, rax); // rax is set at the block end
+    // {rdi, rsi}
+    mov(rdi, STATE);
+    mov(rsi, rax); // rax is set at the block end
 
-    mov(rax, config.ExitFunctionLink);
-    call(rax);
+    call(qword STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
 
     if (SignalSafeCompile) {
       // Now restore the signal mask
@@ -330,7 +304,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
   }
 
   {
-    CallbackPtr = getCurr<CPUBackend::JITCallback>();
+    CallbackPtr = getCurr<JITCallback>();
 
     push(rbx);
     push(rbp);
@@ -345,7 +319,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     // XXX: XMM?
 
     // Make sure to adjust the refcounter so we don't clear the cache now
-    add(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SignalHandlerRefCountPointer)], 1);
+    add(qword STATE_PTR(CpuStateFrame, SignalHandlerRefCounter), 1);
 
     // 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
@@ -353,12 +327,12 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
 
     // Store the trampoline to the guest stack
     // Guest stack is now correctly misaligned after a regular call instruction
-    sub(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])], 16);
-    mov(rbx, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])]);
+    sub(qword STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]), 16);
+    mov(rbx, qword STATE_PTR(CpuStateFrame, State.gregs[X86State::REG_RSP]));
     mov(qword [rbx], rax);
 
     // Store RIP to the context state
-    mov(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.rip)], rsi);
+    mov(qword STATE_PTR(CpuStateFrame, State.rip), rsi);
 
     // Back to the loop top now
     jmp(LoopTop);
@@ -373,30 +347,34 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
   {
     // Guest SIGILL handler
     // Needs to be distinct from the SignalHandlerReturnAddress
-    UnimplementedInstructionAddress = getCurr<uint64_t>();
+    GuestSignal_SIGILL = getCurr<uint64_t>();
     ud2();
   }
 
   {
-    // Guest Overflow handler
+    // Guest SIGTRAP handler
     // Needs to be distinct from the SignalHandlerReturnAddress
-    OverflowExceptionInstructionAddress = getCurr<uint64_t>();
+    GuestSignal_SIGTRAP = getCurr<uint64_t>();
 
     // ud2 = SIGILL
     // int3 = SIGTRAP
     // hlt = SIGSEGV
+    int3();
+  }
 
-    mov(rax, reinterpret_cast<uint64_t>(&SynchronousFaultData));
-    add(byte [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, FaultToTopAndGeneratedException)], 1);
-    mov(dword [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, TrapNo)], X86State::X86_TRAPNO_OF);
-    mov(dword [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, err_code)], 0);
-    mov(dword [rax + offsetof(Dispatcher::SynchronousFaultDataStruct, si_code)], 0x80);
+  {
+    // Guest SIGSEGV handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGSEGV = getCurr<uint64_t>();
 
+    // ud2 = SIGILL
+    // int3 = SIGTRAP
+    // hlt = SIGSEGV
     hlt();
   }
 
   {
-    ReturnPtr = getCurr<FEXCore::Context::Context::IntCallbackReturn>();
+    IntCallbackReturnAddress = getCurr<uint64_t>();
 //  using CallbackReturn =  FEX_NAKED void(*)(FEXCore::Core::InternalThreadState *Thread, volatile void *Host_RSP);
 
     // rdi = thread
@@ -430,39 +408,92 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     CTX->Symbols.RegisterJITSpace(reinterpret_cast<void*>(Start), End-Start);
   }
 
-  // Setup dispatcher specific pointers that need to be accessed from JIT code
-  {
-    auto &Pointers = ThreadState->CurrentFrame->Pointers.X86;
+}
 
-    Pointers.DispatcherLoopTop = AbsoluteLoopTopAddress;
-    Pointers.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
-    Pointers.ThreadStopHandler = ThreadStopHandlerAddress;
-    Pointers.ThreadPauseHandler = ThreadPauseHandlerAddress;
-    Pointers.UnimplementedInstructionHandler = UnimplementedInstructionAddress;
-    Pointers.OverflowExceptionHandler = OverflowExceptionInstructionAddress;
-    Pointers.SignalReturnHandler = SignalHandlerReturnAddress;
-    Pointers.L1Pointer = Thread->LookupCache->GetL1Pointer();
+// Used by GenerateGDBPauseCheck, GenerateInterpreterTrampoline
+static thread_local Xbyak::CodeGenerator emit(1, &emit); // actual emit target set with setNewBuffer
+
+size_t X86Dispatcher::GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) {
+  using namespace Xbyak;
+  using namespace Xbyak::util;
+
+  emit.setNewBuffer(CodeBuffer, MaxGDBPauseCheckSize);
+  
+  Label 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(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
+  emit.mov(rax, reinterpret_cast<uint64_t>(CTX));
+
+  // If the value == 0 then we don't need to stop
+  emit.cmp(dword [rax + (offsetof(FEXCore::Context::Context, Config.RunningMode))], 0);
+  emit.je(RunBlock);
+  {
+    // Make sure RIP is syncronized to the context
+    emit.mov(rax, GuestRIP);
+    emit.mov(qword STATE_PTR(CpuStateFrame, State.rip), rax);
+
+    // Stop the thread
+    emit.mov(rax, qword STATE_PTR(CpuStateFrame, Pointers.Common.ThreadPauseHandlerSpillSRA));
+    emit.jmp(rax);
   }
+
+  emit.L(RunBlock);
+
+  emit.ready();
+
+  return emit.getSize();
+}
+
+
+size_t X86Dispatcher::GenerateInterpreterTrampoline(uint8_t *CodeBuffer) {
+  using namespace Xbyak;
+  using namespace Xbyak::util;
+
+  emit.setNewBuffer(CodeBuffer, MaxInterpreterTrampolineSize);
+
+  Label InlineIRData;
+  
+  emit.mov(rdi, STATE);
+  emit.lea(rsi, ptr[rip + InlineIRData]);
+  emit.call(qword STATE_PTR(CpuStateFrame, Pointers.Interpreter.FragmentExecuter));
+
+  emit.jmp(qword STATE_PTR(CpuStateFrame, Pointers.Common.DispatcherLoopTop));
+
+  emit.L(InlineIRData);
+
+  emit.ready();
+
+  return emit.getSize();
 }
 
 X86Dispatcher::~X86Dispatcher() {
   FEXCore::Allocator::munmap(top_, MAX_DISPATCHER_CODE_SIZE);
 }
 
-#ifdef _M_X86_64
+void X86Dispatcher::InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) {
+  // Setup dispatcher specific pointers that need to be accessed from JIT code
+  {
+    auto &Common = Thread->CurrentFrame->Pointers.Common;
 
-void InterpreterCore::CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
-  DispatcherConfig config;
-  config.ExecuteBlocksWithCall = true;
+    Common.DispatcherLoopTop = AbsoluteLoopTopAddress;
+    Common.DispatcherLoopTopFillSRA = AbsoluteLoopTopAddressFillSRA;
+    Common.ExitFunctionLinker = ExitFunctionLinkerAddress;
+    Common.ThreadStopHandlerSpillSRA = ThreadStopHandlerAddress;
+    Common.ThreadPauseHandlerSpillSRA = ThreadPauseHandlerAddress;
+    Common.GuestSignal_SIGILL = GuestSignal_SIGILL;
+    Common.GuestSignal_SIGTRAP = GuestSignal_SIGTRAP;
+    Common.GuestSignal_SIGSEGV = GuestSignal_SIGSEGV;
+    Common.SignalReturnHandler = SignalHandlerReturnAddress;
 
-  Dispatcher = std::make_unique<X86Dispatcher>(ctx, Thread, config);
-  DispatchPtr = Dispatcher->DispatchPtr;
-  CallbackPtr = Dispatcher->CallbackPtr;
-
-  // TODO: It feels wrong to initialize this way
-  ctx->InterpreterCallbackReturn = Dispatcher->ReturnPtr;
+    auto &Interpreter = Thread->CurrentFrame->Pointers.Interpreter;
+    (uintptr_t&)Interpreter.CallbackReturn = IntCallbackReturnAddress;
+  }
 }
 
-#endif
+std::unique_ptr<Dispatcher> Dispatcher::CreateX86(FEXCore::Context::Context *CTX, const DispatcherConfig &Config) {
+  return std::make_unique<X86Dispatcher>(CTX, Config);
+}
 
 }

External/FEXCore/Source/Interface/Core/Dispatcher/X86Dispatcher.h

@@ -17,7 +17,10 @@ namespace FEXCore::CPU {
 
 class X86Dispatcher final : public Dispatcher, public Xbyak::CodeGenerator {
   public:
-    X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, DispatcherConfig &config);
+    X86Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &config);
+    void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) override;
+    size_t GenerateGDBPauseCheck(uint8_t *CodeBuffer, uint64_t GuestRIP) override;
+    size_t GenerateInterpreterTrampoline(uint8_t *CodeBuffer) override;
 
     virtual ~X86Dispatcher() override;
 };

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

@@ -18,7 +18,9 @@ $end_info$
 #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>
 #include <sys/mman.h>
 
@@ -187,7 +189,7 @@ Decoder::~Decoder() {
 
 uint8_t Decoder::ReadByte() {
   uint8_t Byte = InstStream[InstructionSize];
-  LOGMAN_THROW_A_FMT(InstructionSize < MAX_INST_SIZE, "Max instruction size exceeded!");
+  LOGMAN_THROW_AA_FMT(InstructionSize < MAX_INST_SIZE, "Max instruction size exceeded!");
   Instruction[InstructionSize] = Byte;
   InstructionSize++;
   return Byte;
@@ -199,14 +201,7 @@ uint8_t Decoder::PeekByte(uint8_t Offset) const {
 }
 
 uint64_t Decoder::ReadData(uint8_t Size) {
-  if (Size == 0) {
-    return 0;
-  }
-
-  if (Size > sizeof(uint64_t)) {
-    LOGMAN_MSG_A_FMT("Unknown data size to read");
-    return 0;
-  }
+  LOGMAN_THROW_AA_FMT(Size != 0 && Size <= sizeof(uint64_t), "Unknown data size to read");
 
   uint64_t Res = 0;
   std::memcpy(&Res, &InstStream[InstructionSize], Size);
@@ -346,13 +341,15 @@ void Decoder::DecodeModRM_64(X86Tables::DecodedOperand *Operand, X86Tables::ModR
     Operand->Data.SIB.Index = MapModRMToReg(DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_X ? 1 : 0, SIB.index, false, false, false, false, 0b100);
     Operand->Data.SIB.Base  = MapModRMToReg(DecodeInst->Flags & DecodeFlags::FLAG_REX_XGPR_B ? 1 : 0, SIB.base, false, false, false, false, ModRM.mod == 0 ? 0b101 : 16);
 
-    LOGMAN_THROW_A_FMT(Displacement <= 4, "Number of bytes should be <= 4 for literal src");
+    LOGMAN_THROW_AA_FMT(Displacement <= 4, "Number of bytes should be <= 4 for literal src");
 
-    uint64_t Literal = ReadData(Displacement);
-    if (Displacement == 1) {
-      Literal = static_cast<int8_t>(Literal);
+    if (Displacement) {
+      uint64_t Literal = ReadData(Displacement);
+      if (Displacement == 1) {
+        Literal = static_cast<int8_t>(Literal);
+      }
+      Operand->Data.SIB.Offset = Literal;
     }
-    Operand->Data.SIB.Offset = Literal;
   }
   else if (ModRM.mod == 0) {
     // Explained in Table 1-14. "Operand Addressing Using ModRM and SIB Bytes"
@@ -403,7 +400,7 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
     return false;
   }
 
-  LOGMAN_THROW_A_FMT(!(Info->Type >= FEXCore::X86Tables::TYPE_GROUP_1 && Info->Type <= FEXCore::X86Tables::TYPE_GROUP_P),
+  LOGMAN_THROW_AA_FMT(!(Info->Type >= FEXCore::X86Tables::TYPE_GROUP_1 && Info->Type <= FEXCore::X86Tables::TYPE_GROUP_P),
                      "Group Ops should have been decoded before this!");
 
   uint8_t DestSize{};
@@ -454,8 +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 ||
@@ -486,7 +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 ||
@@ -527,7 +540,7 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
   }
 
   if (HAS_NON_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_REX_IN_BYTE)) {
-    LOGMAN_THROW_A_FMT(!HasMODRM, "This instruction shouldn't have ModRM!");
+    LOGMAN_THROW_AA_FMT(!HasMODRM, "This instruction shouldn't have ModRM!");
 
     // If the REX is in the byte that means the lower nibble of the OP contains the destination GPR
     // This also means that the destination is always a GPR on these ones
@@ -636,7 +649,7 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
   }
 
   if (Bytes != 0) {
-    LOGMAN_THROW_A_FMT(Bytes <= 8, "Number of bytes should be <= 8 for literal src");
+    LOGMAN_THROW_AA_FMT(Bytes <= 8, "Number of bytes should be <= 8 for literal src");
 
     DecodeInst->Src[CurrentSrc].Data.Literal.Size = Bytes;
 
@@ -661,7 +674,7 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op,
     DecodeInst->Src[CurrentSrc].Data.Literal.Value = Literal;
   }
 
-  LOGMAN_THROW_A_FMT(Bytes == 0, "Inst at 0x{:x}: 0x{:04x} '{}' Had an instruction of size {} with {} remaining",
+  LOGMAN_THROW_AA_FMT(Bytes == 0, "Inst at 0x{:x}: 0x{:04x} '{}' Had an instruction of size {} with {} remaining",
                      DecodeInst->PC, DecodeInst->OP, DecodeInst->TableInfo->Name ?: "UND", InstructionSize, Bytes);
   DecodeInst->InstSize = InstructionSize;
   return true;
@@ -687,7 +700,7 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
     return false;
   }
 
-  LOGMAN_THROW_A_FMT(Info->Type != FEXCore::X86Tables::TYPE_REX_PREFIX,
+  LOGMAN_THROW_AA_FMT(Info->Type != FEXCore::X86Tables::TYPE_REX_PREFIX,
                      "REX PREFIX should have been decoded before this!");
 
   if (Info->Type >= FEXCore::X86Tables::TYPE_GROUP_1 &&
@@ -744,7 +757,7 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
         3,
       };
       uint8_t Field = RegToField[ModRM.reg];
-      LOGMAN_THROW_A_FMT(Field != 255, "Invalid field selected!");
+      LOGMAN_THROW_AA_FMT(Field != 255, "Invalid field selected!");
 
       LocalOp = (Field << 3) | ModRM.rm;
       return NormalOp(&SecondModRMTableOps[LocalOp], LocalOp);
@@ -780,6 +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();
@@ -787,6 +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;
@@ -1116,6 +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;
@@ -1135,7 +1179,8 @@ const uint8_t *Decoder::AdjustAddrForSpecialRegion(uint8_t const* _InstStream, u
   return _InstStream - EntryPoint + RIP;
 }
 
-void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC) {
+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();
@@ -1165,6 +1210,12 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
   // Entry is a jump target
   BlocksToDecode.emplace(PC);
 
+  uint64_t CurrentCodePage = PC & FHU::FEX_PAGE_MASK;
+
+  std::set<uint64_t> CodePages = { CurrentCodePage };
+
+  AddContainedCodePage(PC, CurrentCodePage, FHU::FEX_PAGE_SIZE);
+
   while (!BlocksToDecode.empty()) {
     auto BlockDecodeIt = BlocksToDecode.begin();
     uint64_t RIPToDecode = *BlockDecodeIt;
@@ -1181,10 +1232,33 @@ void Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
     InstStream = AdjustAddrForSpecialRegion(_InstStream, EntryPoint, RIPToDecode);
 
     while (1) {
+
+      // MAX_INST_SIZE assumes worst case
+      auto OpMinAddress = RIPToDecode + PCOffset;
+      auto OpMaxAddress = OpMinAddress + MAX_INST_SIZE;
+
+      auto OpMinPage = OpMinAddress & FHU::FEX_PAGE_MASK;
+      auto OpMaxPage = OpMaxAddress & FHU::FEX_PAGE_MASK;
+
+
+      if (OpMinPage != CurrentCodePage) {
+        CurrentCodePage = OpMinPage;
+        if (CodePages.insert(CurrentCodePage).second) {
+          AddContainedCodePage(PC, CurrentCodePage, FHU::FEX_PAGE_SIZE);
+        }
+      }
+
+      if (OpMaxPage != CurrentCodePage) {
+        CurrentCodePage = OpMaxPage;
+        if (CodePages.insert(CurrentCodePage).second) {
+          AddContainedCodePage(PC, CurrentCodePage, FHU::FEX_PAGE_SIZE);
+        }
+      }
+
       bool ErrorDuringDecoding = !DecodeInstruction(RIPToDecode + PCOffset);
 
       if (ErrorDuringDecoding) {
-        LogMan::Msg::DFmt("Couldn't Decode something at 0x{:x}, Started at 0x{:x}", PC + PCOffset, PC);
+        LogMan::Msg::DFmt("Couldn't Decode something at 0x{:x}, Started at 0x{:x}", RIPToDecode + PCOffset, PC);
         // Put an invalid instruction in the stream so the core can raise SIGILL if hit
         CurrentBlockDecoding.HasInvalidInstruction = true;
         // Error while decoding instruction. We don't know the table or instruction size
@@ -1210,23 +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

@@ -27,7 +27,7 @@ public:
 
   Decoder(FEXCore::Context::Context *ctx);
   ~Decoder();
-  void DecodeInstructionsAtEntry(uint8_t const* InstStream, uint64_t PC);
+  void DecodeInstructionsAtEntry(uint8_t const* InstStream, uint64_t PC, std::function<void(uint64_t BlockEntry, uint64_t Start, uint64_t Length)> AddContainedCodePage);
 
   std::vector<DecodedBlocks> const *GetDecodedBlocks() const {
     return &Blocks;
@@ -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/GdbServer.cpp

@@ -125,7 +125,7 @@ static std::string hexstring(std::istringstream &ss, int delm) {
   return ret;
 }
 
-static std::string encodeHex(unsigned char *data, size_t length) {
+static std::string encodeHex(const unsigned char *data, size_t length) {
   std::ostringstream ss;
 
   for (size_t i=0; i < length; i++) {
@@ -251,15 +251,15 @@ void GdbServer::SendACK(std::ostream &stream, bool NACK) {
 }
 
 struct FEX_PACKED GDBContextDefinition {
-  uint64_t gregs[16];
+  uint64_t gregs[Core::CPUState::NUM_GPRS];
   uint64_t rip;
   uint32_t eflags;
   uint32_t cs, ss, ds, es, fs, gs;
-  X80SoftFloat mm[8];
+  X80SoftFloat mm[Core::CPUState::NUM_MMS];
   uint32_t fctrl;
   uint32_t fstat;
   uint32_t dummies[6];
-  uint64_t xmm[16][2];
+  uint64_t xmm[Core::CPUState::NUM_XMMS][4];
   uint32_t mxcsr;
 };
 
@@ -288,12 +288,12 @@ std::string GdbServer::readRegs() {
   memcpy(&GDB.gregs[0], &state.gregs[0], sizeof(GDB.gregs));
   memcpy(&GDB.rip, &state.rip, sizeof(GDB.rip));
 
-  for (size_t i = 0; i < 32; ++i) {
+  for (size_t i = 0; i < Core::CPUState::NUM_EFLAG_BITS; ++i) {
     uint64_t Flag = state.flags[i];
     GDB.eflags |= (Flag << i);
   }
 
-  for (size_t i = 0; i < 8; ++i) {
+  for (size_t i = 0; i < Core::CPUState::NUM_MMS; ++i) {
     memcpy(&GDB.mm[i], &state.mm[i], sizeof(GDB.mm));
   }
 
@@ -306,7 +306,7 @@ std::string GdbServer::readRegs() {
   GDB.fstat |= static_cast<uint32_t>(state.flags[FEXCore::X86State::X87FLAG_C2_LOC]) << 10;
   GDB.fstat |= static_cast<uint32_t>(state.flags[FEXCore::X86State::X87FLAG_C3_LOC]) << 14;
 
-  memcpy(&GDB.xmm[0], &state.xmm[0], sizeof(GDB.xmm));
+  memcpy(&GDB.xmm[0], &state.xmm.avx.data[0], sizeof(GDB.xmm));
 
   return encodeHex((unsigned char *)&GDB, sizeof(GDBContextDefinition));
 }
@@ -346,7 +346,7 @@ GdbServer::HandledPacketType GdbServer::readReg(const std::string& packet) {
   }
   else if (addr == offsetof(GDBContextDefinition, eflags)) {
     uint32_t eflags{};
-    for (size_t i = 0; i < 32; ++i) {
+    for (size_t i = 0; i < Core::CPUState::NUM_EFLAG_BITS; ++i) {
       uint64_t Flag = state.flags[i];
       eflags |= (Flag << i);
     }
@@ -382,7 +382,9 @@ GdbServer::HandledPacketType GdbServer::readReg(const std::string& packet) {
   }
   else if (addr >= offsetof(GDBContextDefinition, xmm[0][0]) &&
            addr < offsetof(GDBContextDefinition, xmm[16][0])) {
-    return {encodeHex((unsigned char *)(&state.xmm[(addr - offsetof(GDBContextDefinition, xmm[0][0])) / 16][0]), 16), HandledPacketType::TYPE_ACK};
+    const auto XmmIndex = (addr - offsetof(GDBContextDefinition, xmm[0][0])) / Core::CPUState::XMM_AVX_REG_SIZE;
+    const auto *Data = (unsigned char *)&state.xmm.avx.data[XmmIndex][0];
+    return {encodeHex(Data, Core::CPUState::XMM_AVX_REG_SIZE), HandledPacketType::TYPE_ACK};
   }
   else if (addr == offsetof(GDBContextDefinition, mxcsr)) {
     uint32_t Empty{};
@@ -424,7 +426,7 @@ std::string buildTargetXML() {
       // We want to just memcpy our x86 state to gdb, so we tell it the ordering.
 
       // GPRs
-      for (int i=0; i < 16; i++) {
+      for (uint32_t i = 0; i < Core::CPUState::NUM_GPRS; i++) {
         reg(FEXCore::Core::GetGRegName(i), "int64", 64);
       }
 
@@ -481,13 +483,37 @@ std::string buildTargetXML() {
         )";
 
       // SSE regs
-      for (int i=0; i < 16; i++) {
+      for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
           reg("xmm" + std::to_string(i), "vec128", 128);
       }
 
       reg("mxcsr", "int", 32);
 
     xml << "</feature>\n";
+
+    xml << "<feature name='org.gnu.gdb.i386.avx'>";
+    xml <<
+        R"(<vector id="v4f" type="ieee_single" count="4"/>
+        <vector id="v2d" type="ieee_double" count="2"/>
+        <vector id="v16i8" type="int8" count="16"/>
+        <vector id="v8i16" type="int16" count="8"/>
+        <vector id="v4i32" type="int32" count="4"/>
+        <vector id="v2i64" type="int64" count="2"/>
+        <union id="vec128">
+          <field name="v4_float" type="v4f"/>
+          <field name="v2_double" type="v2d"/>
+          <field name="v16_int8" type="v16i8"/>
+          <field name="v8_int16" type="v8i16"/>
+          <field name="v4_int32" type="v4i32"/>
+          <field name="v2_int64" type="v2i64"/>
+          <field name="uint128" type="uint128"/>
+        </union>
+        )";
+    for (size_t i = 0; i < Core::CPUState::NUM_XMMS; i++) {
+        reg(fmt::format("ymm{}h", i), "vec128", 128);
+    }
+    xml << "</feature>\n";
+
   xml << "</target>";
   xml << std::flush;
 

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

@@ -1,7 +1,7 @@
 #include "Interface/Core/CPUID.h"
-#include "Interface/Core/HostFeatures.h"
+#include <FEXCore/Core/HostFeatures.h>
 
-#ifdef _M_ARM_64
+#if defined(_M_ARM_64) || defined(VIXL_SIMULATOR)
 #include "aarch64/assembler-aarch64.h"
 #include "aarch64/cpu-aarch64.h"
 #include "aarch64/disasm-aarch64.h"
@@ -50,8 +50,12 @@ static uint32_t GetDCZID() {
 
 
 HostFeatures::HostFeatures() {
-#ifdef _M_ARM_64
+#if defined(_M_ARM_64) || defined(VIXL_SIMULATOR)
+#ifdef VIXL_SIMULATOR
+  auto Features = vixl::CPUFeatures::All();
+#else
   auto Features = vixl::CPUFeatures::InferFromOS();
+#endif
   SupportsAES = Features.Has(vixl::CPUFeatures::Feature::kAES);
   SupportsCRC = Features.Has(vixl::CPUFeatures::Feature::kCRC32);
   SupportsAtomics = Features.Has(vixl::CPUFeatures::Feature::kAtomics);
@@ -61,7 +65,26 @@ 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;
+#ifdef VIXL_SIMULATOR
+  // Hardcode enable SVE with 256-bit wide registers.
+  SupportsAVX = true;
+#else
+  SupportsAVX = Features.Has(vixl::CPUFeatures::Feature::kSVE2) &&
+                vixl::aarch64::CPU::ReadSVEVectorLengthInBits() >= 256;
+#endif
+  SupportsSHA = true;
+  SupportsBMI1 = true;
+  SupportsBMI2 = true;
+
+  if (!SupportsAtomics) {
+    WARN_ONCE_FMT("Host CPU doesn't support atomics. Expect bad performance");
+  }
+
+#ifdef _M_ARM_64
   // We need to get the CPU's cache line size
   // We expect sane targets that have correct cacheline sizes across clusters
   uint64_t CTR;
@@ -71,31 +94,6 @@ HostFeatures::HostFeatures() {
   DCacheLineSize = 4 << ((CTR >> 16) & 0xF);
   ICacheLineSize = 4 << (CTR & 0xF);
 
-  if (!SupportsAtomics) {
-    WARN_ONCE_FMT("Host CPU doesn't support atomics. Expect bad performance");
-  }
-#endif
-#ifdef _M_X86_64
-  Xbyak::util::Cpu Features{};
-  SupportsAES = Features.has(Xbyak::util::Cpu::tAESNI);
-  SupportsCRC = Features.has(Xbyak::util::Cpu::tSSE42);
-  SupportsRAND = Features.has(Xbyak::util::Cpu::tRDRAND) && Features.has(Xbyak::util::Cpu::tRDSEED);
-  SupportsRCPC = true;
-  SupportsTSOImm9 = true;
-
-  // xbyak doesn't know how to check for CLZero
-  uint32_t eax, ebx, ecx, edx;
-  // First ensure we support a new enough extended CPUID function range
-  __cpuid(0x8000'0000, eax, ebx, ecx, edx);
-  if (eax >= 0x8000'0008U) {
-    // CLZero defined in 8000_00008_EBX[bit 0]
-    __cpuid(0x8000'0008, eax, ebx, ecx, edx);
-    SupportsCLZERO = ebx & 1;
-  }
-
-  SupportsFlushInputsToZero = true;
-  SupportsFloatExceptions = true;
-#else
   // Test if this CPU supports float exception trapping by attempting to enable
   // On unsupported these bits are architecturally defined as RAZ/WI
   constexpr uint32_t ExceptionEnableTraps =
@@ -116,6 +114,40 @@ HostFeatures::HostFeatures() {
   SetFPCR(OriginalFPCR);
 #endif
 
+#endif
+#if defined(_M_X86_64) && !defined(VIXL_SIMULATOR)
+  Xbyak::util::Cpu Features{};
+  SupportsAES = Features.has(Xbyak::util::Cpu::tAESNI);
+  SupportsCRC = Features.has(Xbyak::util::Cpu::tSSE42);
+  SupportsRAND = Features.has(Xbyak::util::Cpu::tRDRAND) && Features.has(Xbyak::util::Cpu::tRDSEED);
+  SupportsRCPC = true;
+  SupportsTSOImm9 = true;
+  Supports3DNow = Features.has(Xbyak::util::Cpu::t3DN) && Features.has(Xbyak::util::Cpu::tE3DN);
+  SupportsSSE4A = Features.has(Xbyak::util::Cpu::tSSE4a);
+  SupportsAVX = true;
+  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;
+  // First ensure we support a new enough extended CPUID function range
+  __cpuid(0x8000'0000, eax, ebx, ecx, edx);
+  if (eax >= 0x8000'0008U) {
+    // CLZero defined in 8000_00008_EBX[bit 0]
+    __cpuid(0x8000'0008, eax, ebx, ecx, edx);
+    SupportsCLZERO = ebx & 1;
+  }
+
+  SupportsFlushInputsToZero = true;
+  SupportsFloatExceptions = true;
+#endif
+
+#ifdef VIXL_SIMULATOR
+  // simulator doesn't support dc(ZVA)
+  SupportsCLZERO = false;
+#else
   // Check if we can support cacheline clears
   uint32_t DCZID = GetDCZID();
   if ((DCZID & DCZID_DZP_MASK) == 0) {
@@ -125,5 +157,6 @@ HostFeatures::HostFeatures() {
     // This means we can use the instruction
     SupportsCLZERO = DCZID_Bytes == CPUIDEmu::CACHELINE_SIZE;
   }
+#endif
 }
 }

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

@@ -20,7 +20,7 @@ DEF_OP(TruncElementPair) {
 
   switch (IROp->Size) {
     case 4: {
-      uint64_t *Src = GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+      uint64_t *Src = GetSrc<uint64_t*>(Data->SSAData, Op->Pair);
       uint64_t Result{};
       Result = Src[0] & ~0U;
       Result |= Src[1] << 32;
@@ -69,11 +69,11 @@ DEF_OP(CycleCounter) {
 
 DEF_OP(Add) {
   auto Op = IROp->C<IR::IROp_Add>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a + b; };
+  auto *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  auto *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a + b; };
 
   switch (OpSize) {
     DO_OP(4, uint32_t, Func)
@@ -84,11 +84,11 @@ DEF_OP(Add) {
 
 DEF_OP(Sub) {
   auto Op = IROp->C<IR::IROp_Sub>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a - b; };
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a - b; };
 
   switch (OpSize) {
     DO_OP(4, uint32_t, Func)
@@ -99,9 +99,9 @@ DEF_OP(Sub) {
 
 DEF_OP(Neg) {
   auto Op = IROp->C<IR::IROp_Neg>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Src);
   switch (OpSize) {
     case 4:
       GD = -static_cast<int32_t>(Src);
@@ -115,10 +115,10 @@ DEF_OP(Neg) {
 
 DEF_OP(Mul) {
   auto Op = IROp->C<IR::IROp_Mul>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
 
   switch (OpSize) {
     case 4:
@@ -138,10 +138,10 @@ DEF_OP(Mul) {
 
 DEF_OP(UMul) {
   auto Op = IROp->C<IR::IROp_UMul>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
 
   switch (OpSize) {
     case 4:
@@ -161,9 +161,9 @@ DEF_OP(UMul) {
 
 DEF_OP(Div) {
   auto Op = IROp->C<IR::IROp_Div>();
-  uint8_t OpSize = IROp->Size;
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint8_t OpSize = IROp->Size;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
 
   switch (OpSize) {
     case 1:
@@ -179,7 +179,7 @@ DEF_OP(Div) {
       GD = static_cast<int64_t>(Src1) / static_cast<int64_t>(Src2);
       break;
     case 16: {
-      __int128_t Tmp = *GetSrc<__int128_t*>(Data->SSAData, Op->Header.Args[0]) / *GetSrc<__int128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __int128_t Tmp = *GetSrc<__int128_t*>(Data->SSAData, Op->Src1) / *GetSrc<__int128_t*>(Data->SSAData, Op->Src2);
       memcpy(GDP, &Tmp, 16);
       break;
     }
@@ -189,10 +189,10 @@ DEF_OP(Div) {
 
 DEF_OP(UDiv) {
   auto Op = IROp->C<IR::IROp_UDiv>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
 
   switch (OpSize) {
     case 1:
@@ -208,7 +208,7 @@ DEF_OP(UDiv) {
       GD = static_cast<uint64_t>(Src1) / static_cast<uint64_t>(Src2);
       break;
     case 16: {
-      __uint128_t Tmp = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]) / *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __uint128_t Tmp = *GetSrc<__uint128_t*>(Data->SSAData, Op->Src1) / *GetSrc<__uint128_t*>(Data->SSAData, Op->Src2);
       memcpy(GDP, &Tmp, 16);
       break;
     }
@@ -218,10 +218,10 @@ DEF_OP(UDiv) {
 
 DEF_OP(Rem) {
   auto Op = IROp->C<IR::IROp_Rem>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
 
   switch (OpSize) {
     case 1:
@@ -237,7 +237,7 @@ DEF_OP(Rem) {
       GD = static_cast<int64_t>(Src1) % static_cast<int64_t>(Src2);
       break;
     case 16: {
-      __int128_t Tmp = *GetSrc<__int128_t*>(Data->SSAData, Op->Header.Args[0]) % *GetSrc<__int128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __int128_t Tmp = *GetSrc<__int128_t*>(Data->SSAData, Op->Src1) % *GetSrc<__int128_t*>(Data->SSAData, Op->Src2);
       memcpy(GDP, &Tmp, 16);
       break;
     }
@@ -247,10 +247,10 @@ DEF_OP(Rem) {
 
 DEF_OP(URem) {
   auto Op = IROp->C<IR::IROp_URem>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
 
   switch (OpSize) {
     case 1:
@@ -266,7 +266,7 @@ DEF_OP(URem) {
       GD = static_cast<uint64_t>(Src1) % static_cast<uint64_t>(Src2);
       break;
     case 16: {
-      __uint128_t Tmp = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]) % *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+      __uint128_t Tmp = *GetSrc<__uint128_t*>(Data->SSAData, Op->Src1) % *GetSrc<__uint128_t*>(Data->SSAData, Op->Src2);
       memcpy(GDP, &Tmp, 16);
       break;
     }
@@ -276,10 +276,10 @@ DEF_OP(URem) {
 
 DEF_OP(MulH) {
   auto Op = IROp->C<IR::IROp_MulH>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
 
   switch (OpSize) {
     case 4: {
@@ -298,10 +298,10 @@ DEF_OP(MulH) {
 
 DEF_OP(UMulH) {
   auto Op = IROp->C<IR::IROp_UMulH>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
   switch (OpSize) {
     case 4:
       GD = static_cast<uint64_t>(Src1) * static_cast<uint64_t>(Src2);
@@ -324,11 +324,11 @@ DEF_OP(UMulH) {
 
 DEF_OP(Or) {
   auto Op = IROp->C<IR::IROp_Or>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a | b; };
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a | b; };
 
   switch (OpSize) {
     DO_OP(1, uint8_t,  Func)
@@ -342,11 +342,11 @@ DEF_OP(Or) {
 
 DEF_OP(And) {
   auto Op = IROp->C<IR::IROp_And>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a & b; };
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a & b; };
 
   switch (OpSize) {
     DO_OP(1, uint8_t,  Func)
@@ -361,8 +361,8 @@ DEF_OP(Andn) {
   auto Op = IROp->C<IR::IROp_Andn>();
   const uint8_t OpSize = IROp->Size;
 
-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
   constexpr auto Func = [](auto a, auto b) {
     using Type = decltype(a);
     return static_cast<Type>(a & static_cast<Type>(~b));
@@ -379,11 +379,11 @@ DEF_OP(Andn) {
 
 DEF_OP(Xor) {
   auto Op = IROp->C<IR::IROp_Xor>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
-  auto Func = [](auto a, auto b) { return a ^ b; };
+  void *Src1 = GetSrc<void*>(Data->SSAData, Op->Src1);
+  void *Src2 = GetSrc<void*>(Data->SSAData, Op->Src2);
+  const auto Func = [](auto a, auto b) { return a ^ b; };
 
   switch (OpSize) {
     DO_OP(1, uint8_t,  Func)
@@ -396,11 +396,11 @@ DEF_OP(Xor) {
 
 DEF_OP(Lshl) {
   auto Op = IROp->C<IR::IROp_Lshl>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  uint8_t Mask = OpSize * 8 - 1;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+  const uint8_t Mask = OpSize * 8 - 1;
   switch (OpSize) {
     case 4:
       GD = static_cast<uint32_t>(Src1) << (Src2 & Mask);
@@ -414,11 +414,11 @@ DEF_OP(Lshl) {
 
 DEF_OP(Lshr) {
   auto Op = IROp->C<IR::IROp_Lshr>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  uint8_t Mask = OpSize * 8 - 1;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+  const uint8_t Mask = OpSize * 8 - 1;
   switch (OpSize) {
     case 4:
       GD = static_cast<uint32_t>(Src1) >> (Src2 & Mask);
@@ -432,11 +432,11 @@ DEF_OP(Lshr) {
 
 DEF_OP(Ashr) {
   auto Op = IROp->C<IR::IROp_Ashr>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  uint8_t Mask = OpSize * 8 - 1;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+  const uint8_t Mask = OpSize * 8 - 1;
   switch (OpSize) {
     case 4:
       GD = (uint32_t)(static_cast<int32_t>(Src1) >> (Src2 & Mask));
@@ -450,12 +450,12 @@ DEF_OP(Ashr) {
 
 DEF_OP(Ror) {
   auto Op = IROp->C<IR::IROp_Ror>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  auto Ror = [] (auto In, auto R) {
-    auto RotateMask = sizeof(In) * 8 - 1;
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+  const auto Ror = [] (auto In, auto R) {
+    const auto RotateMask = sizeof(In) * 8 - 1;
     R &= RotateMask;
     return (In >> R) | (In << (sizeof(In) * 8 - R));
   };
@@ -474,11 +474,11 @@ DEF_OP(Ror) {
 
 DEF_OP(Extr) {
   auto Op = IROp->C<IR::IROp_Extr>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  auto Extr = [] (auto Src1, auto Src2, uint8_t lsb) -> decltype(Src1) {
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+  const auto Extr = [] (auto Src1, auto Src2, uint8_t lsb) -> decltype(Src1) {
     __uint128_t Result{};
     Result = Src1;
     Result <<= sizeof(Src1) * 8;
@@ -500,7 +500,7 @@ DEF_OP(Extr) {
 }
 
 DEF_OP(PDep) {
-  const auto Op = IROp->C<IR::IROp_PExt>();
+  const auto Op = IROp->C<IR::IROp_PDep>();
   const auto OpSize = IROp->Size;
 
   if (OpSize != 4 && OpSize != 8) {
@@ -508,10 +508,10 @@ DEF_OP(PDep) {
     return;
   }
 
-  const uint64_t Input = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Args(0))
-                                     : *GetSrc<uint64_t*>(Data->SSAData, Op->Args(0));
-  uint64_t Mask = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Args(1))
-                              : *GetSrc<uint64_t*>(Data->SSAData, Op->Args(1));
+  const uint64_t Input = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Input)
+                                     : *GetSrc<uint64_t*>(Data->SSAData, Op->Input);
+  uint64_t Mask = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Mask)
+                              : *GetSrc<uint64_t*>(Data->SSAData, Op->Mask);
 
   uint64_t Result = 0;
   for (uint64_t Index = 0; Mask > 0; Index++) {
@@ -532,10 +532,10 @@ DEF_OP(PExt) {
     return;
   }
 
-  const uint64_t Input = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Args(0))
-                                     : *GetSrc<uint64_t*>(Data->SSAData, Op->Args(0));
-  uint64_t Mask = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Args(1))
-                              : *GetSrc<uint64_t*>(Data->SSAData, Op->Args(1));
+  const uint64_t Input = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Input)
+                                     : *GetSrc<uint64_t*>(Data->SSAData, Op->Input);
+  uint64_t Mask = OpSize == 4 ? *GetSrc<uint32_t*>(Data->SSAData, Op->Mask)
+                              : *GetSrc<uint64_t*>(Data->SSAData, Op->Mask);
 
   uint64_t Result = 0;
   for (uint64_t Offset = 0; Mask > 0; Offset++) {
@@ -549,39 +549,39 @@ DEF_OP(PExt) {
 
 DEF_OP(LDiv) {
   auto Op = IROp->C<IR::IROp_LDiv>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
   // Each source is OpSize in size
   // So you can have up to a 128bit divide from x86-64
   switch (OpSize) {
     case 2: {
-      uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
-      int16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[2]);
-      int32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
-      int32_t Res = Source / Divisor;
+      const uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Lower);
+      const uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Upper);
+      const int16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Divisor);
+      const int32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
+      const int32_t Res = Source / Divisor;
 
       // We only store the lower bits of the result
       GD = static_cast<int16_t>(Res);
       break;
     }
     case 4: {
-      uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
-      int32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-      int64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
-      int64_t Res = Source / Divisor;
+      const uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Lower);
+      const uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Upper);
+      const int32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Divisor);
+      const int64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
+      const int64_t Res = Source / Divisor;
 
       // We only store the lower bits of the result
       GD = static_cast<int32_t>(Res);
       break;
     }
     case 8: {
-      uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-      int64_t Divisor = *GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[2]);
-      __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
-      __int128_t Res = Source / Divisor;
+      const uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+      const uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+      const int64_t Divisor = *GetSrc<int64_t*>(Data->SSAData, Op->Divisor);
+      const __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
+      const __int128_t Res = Source / Divisor;
 
       // We only store the lower bits of the result
       memcpy(GDP, &Res, OpSize);
@@ -593,39 +593,39 @@ DEF_OP(LDiv) {
 
 DEF_OP(LUDiv) {
   auto Op = IROp->C<IR::IROp_LUDiv>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
   // Each source is OpSize in size
   // So you can have up to a 128bit divide from x86-64
   switch (OpSize) {
     case 2: {
-      uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[2]);
-      uint32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
-      uint32_t Res = Source / Divisor;
+      const uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Lower);
+      const uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Upper);
+      const uint16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Divisor);
+      const uint32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
+      const uint32_t Res = Source / Divisor;
 
       // We only store the lower bits of the result
       GD = static_cast<uint16_t>(Res);
       break;
     }
     case 4: {
-      uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-      uint64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
-      uint64_t Res = Source / Divisor;
+      const uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Lower);
+      const uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Upper);
+      const uint32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Divisor);
+      const uint64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
+      const uint64_t Res = Source / Divisor;
 
       // We only store the lower bits of the result
       GD = static_cast<uint32_t>(Res);
       break;
     }
     case 8: {
-      uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint64_t Divisor = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[2]);
-      __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
-      __uint128_t Res = Source / Divisor;
+      const uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+      const uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+      const uint64_t Divisor = *GetSrc<uint64_t*>(Data->SSAData, Op->Divisor);
+      const __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
+      const __uint128_t Res = Source / Divisor;
 
       // We only store the lower bits of the result
       memcpy(GDP, &Res, OpSize);
@@ -637,39 +637,39 @@ DEF_OP(LUDiv) {
 
 DEF_OP(LRem) {
   auto Op = IROp->C<IR::IROp_LRem>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
   // Each source is OpSize in size
   // So you can have up to a 128bit Remainder from x86-64
   switch (OpSize) {
     case 2: {
-      uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
-      int16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[2]);
-      int32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
-      int32_t Res = Source % Divisor;
+      const uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Lower);
+      const uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Upper);
+      const int16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Divisor);
+      const int32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
+      const int32_t Res = Source % Divisor;
 
       // We only store the lower bits of the result
       GD = static_cast<int16_t>(Res);
       break;
     }
     case 4: {
-      uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
-      int32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-      int64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
-      int64_t Res = Source % Divisor;
+      const uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Lower);
+      const uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Upper);
+      const int32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Divisor);
+      const int64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
+      const int64_t Res = Source % Divisor;
 
       // We only store the lower bits of the result
       GD = static_cast<int32_t>(Res);
       break;
     }
     case 8: {
-      uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-      int64_t Divisor = *GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[2]);
-      __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
-      __int128_t Res = Source % Divisor;
+      const uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+      const uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+      const int64_t Divisor = *GetSrc<int64_t*>(Data->SSAData, Op->Divisor);
+      const __int128_t Source = (static_cast<__int128_t>(SrcHigh) << 64) | SrcLow;
+      const __int128_t Res = Source % Divisor;
       // We only store the lower bits of the result
       memcpy(GDP, &Res, OpSize);
       break;
@@ -680,39 +680,39 @@ DEF_OP(LRem) {
 
 DEF_OP(LURem) {
   auto Op = IROp->C<IR::IROp_LURem>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
   // Each source is OpSize in size
   // So you can have up to a 128bit Remainder from x86-64
   switch (OpSize) {
     case 2: {
-      uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[2]);
-      uint32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
-      uint32_t Res = Source % Divisor;
+      const uint16_t SrcLow = *GetSrc<uint16_t*>(Data->SSAData, Op->Lower);
+      const uint16_t SrcHigh = *GetSrc<uint16_t*>(Data->SSAData, Op->Upper);
+      const uint16_t Divisor = *GetSrc<uint16_t*>(Data->SSAData, Op->Divisor);
+      const uint32_t Source = (static_cast<uint32_t>(SrcHigh) << 16) | SrcLow;
+      const uint32_t Res = Source % Divisor;
 
       // We only store the lower bits of the result
       GD = static_cast<uint16_t>(Res);
       break;
     }
     case 4: {
-      uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-      uint64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
-      uint64_t Res = Source % Divisor;
+      const uint32_t SrcLow = *GetSrc<uint32_t*>(Data->SSAData, Op->Lower);
+      const uint32_t SrcHigh = *GetSrc<uint32_t*>(Data->SSAData, Op->Upper);
+      const uint32_t Divisor = *GetSrc<uint32_t*>(Data->SSAData, Op->Divisor);
+      const uint64_t Source = (static_cast<uint64_t>(SrcHigh) << 32) | SrcLow;
+      const uint64_t Res = Source % Divisor;
 
       // We only store the lower bits of the result
       GD = static_cast<uint32_t>(Res);
       break;
     }
     case 8: {
-      uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-      uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-      uint64_t Divisor = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[2]);
-      __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
-      __uint128_t Res = Source % Divisor;
+      const uint64_t SrcLow = *GetSrc<uint64_t*>(Data->SSAData, Op->Lower);
+      const uint64_t SrcHigh = *GetSrc<uint64_t*>(Data->SSAData, Op->Upper);
+      const uint64_t Divisor = *GetSrc<uint64_t*>(Data->SSAData, Op->Divisor);
+      const __uint128_t Source = (static_cast<__uint128_t>(SrcHigh) << 64) | SrcLow;
+      const __uint128_t Res = Source % Divisor;
       // We only store the lower bits of the result
       memcpy(GDP, &Res, OpSize);
       break;
@@ -723,62 +723,62 @@ DEF_OP(LURem) {
 
 DEF_OP(Not) {
   auto Op = IROp->C<IR::IROp_Not>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
   const uint64_t mask[9]= { 0, 0xFF, 0xFFFF, 0, 0xFFFFFFFF, 0, 0, 0, 0xFFFFFFFFFFFFFFFFULL };
-  uint64_t Mask = mask[OpSize];
+  const uint64_t Mask = mask[OpSize];
   GD = (~Src) & Mask;
 }
 
 DEF_OP(Popcount) {
   auto Op = IROp->C<IR::IROp_Popcount>();
-  uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
   GD = std::popcount(Src);
 }
 
 DEF_OP(FindLSB) {
   auto Op = IROp->C<IR::IROp_FindLSB>();
-  uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Result = FindFirstSetBit(Src);
+  const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
+  const uint64_t Result = FindFirstSetBit(Src);
   GD = Result - 1;
 }
 
 DEF_OP(FindMSB) {
   auto Op = IROp->C<IR::IROp_FindMSB>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
   switch (OpSize) {
-    case 1: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]))) - 1; break;
-    case 2: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]))) - 1; break;
-    case 4: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]))) - 1; break;
-    case 8: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]))) - 1; break;
+    case 1: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint8_t*>(Data->SSAData, Op->Src))) - 1; break;
+    case 2: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint16_t*>(Data->SSAData, Op->Src))) - 1; break;
+    case 4: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint32_t*>(Data->SSAData, Op->Src))) - 1; break;
+    case 8: GD = (OpSize * 8 - std::countl_zero(*GetSrc<uint64_t*>(Data->SSAData, Op->Src))) - 1; break;
     default: LOGMAN_MSG_A_FMT("Unknown FindMSB size: {}", OpSize); break;
   }
 }
 
 DEF_OP(FindTrailingZeros) {
   auto Op = IROp->C<IR::IROp_FindTrailingZeros>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
   switch (OpSize) {
     case 1: {
-      auto Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Src);
       GD = std::countr_zero(Src);
       break;
     }
     case 2: {
-      auto Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Src);
       GD = std::countr_zero(Src);
       break;
     }
     case 4: {
-      auto Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Src);
       GD = std::countr_zero(Src);
       break;
     }
     case 8: {
-      auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
       GD = std::countr_zero(Src);
       break;
     }
@@ -788,26 +788,26 @@ DEF_OP(FindTrailingZeros) {
 
 DEF_OP(CountLeadingZeroes) {
   auto Op = IROp->C<IR::IROp_CountLeadingZeroes>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
   switch (OpSize) {
     case 1: {
-      auto Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Src);
       GD = std::countl_zero(Src);
       break;
     }
     case 2: {
-      auto Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Src);
       GD = std::countl_zero(Src);
       break;
     }
     case 4: {
-      auto Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Src);
       GD = std::countl_zero(Src);
       break;
     }
     case 8: {
-      auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+      const auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
       GD = std::countl_zero(Src);
       break;
     }
@@ -817,12 +817,12 @@ DEF_OP(CountLeadingZeroes) {
 
 DEF_OP(Rev) {
   auto Op = IROp->C<IR::IROp_Rev>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
   switch (OpSize) {
-    case 2: GD = BSwap16(*GetSrc<uint16_t*>(Data->SSAData, Op->Header.Args[0])); break;
-    case 4: GD = BSwap32(*GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[0])); break;
-    case 8: GD = BSwap64(*GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0])); break;
+    case 2: GD = BSwap16(*GetSrc<uint16_t*>(Data->SSAData, Op->Src)); break;
+    case 4: GD = BSwap32(*GetSrc<uint32_t*>(Data->SSAData, Op->Src)); break;
+    case 8: GD = BSwap64(*GetSrc<uint64_t*>(Data->SSAData, Op->Src)); break;
     default: LOGMAN_MSG_A_FMT("Unknown REV size: {}", OpSize); break;
   }
 }
@@ -830,34 +830,36 @@ DEF_OP(Rev) {
 DEF_OP(Bfi) {
   auto Op = IROp->C<IR::IROp_Bfi>();
   uint64_t SourceMask = (1ULL << Op->Width) - 1;
-  if (Op->Width == 64)
+  if (Op->Width == 64) {
     SourceMask = ~0ULL;
-  uint64_t DestMask = ~(SourceMask << Op->lsb);
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
-  uint64_t Res = (Src1 & DestMask) | ((Src2 & SourceMask) << Op->lsb);
+  }
+  const uint64_t DestMask = ~(SourceMask << Op->lsb);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Dest);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
+  const uint64_t Res = (Src1 & DestMask) | ((Src2 & SourceMask) << Op->lsb);
   GD = Res;
 }
 
 DEF_OP(Bfe) {
   auto Op = IROp->C<IR::IROp_Bfe>();
 
-  LOGMAN_THROW_A_FMT(IROp->Size <= 8, "OpSize is too large for BFE: {}", IROp->Size);
+  LOGMAN_THROW_AA_FMT(IROp->Size <= 8, "OpSize is too large for BFE: {}", IROp->Size);
   uint64_t SourceMask = (1ULL << Op->Width) - 1;
-  if (Op->Width == 64)
+  if (Op->Width == 64) {
     SourceMask = ~0ULL;
+  }
   SourceMask <<= Op->lsb;
-  uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Src);
   GD = (Src & SourceMask) >> Op->lsb;
 }
 
 DEF_OP(Sbfe) {
   auto Op = IROp->C<IR::IROp_Sbfe>();
 
-  LOGMAN_THROW_A_FMT(IROp->Size <= 8, "OpSize is too large for SBFE: {}", IROp->Size);
-  int64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t ShiftLeftAmount = (64 - (Op->Width + Op->lsb));
-  uint64_t ShiftRightAmount = ShiftLeftAmount + Op->lsb;
+  LOGMAN_THROW_AA_FMT(IROp->Size <= 8, "OpSize is too large for SBFE: {}", IROp->Size);
+  int64_t Src = *GetSrc<int64_t*>(Data->SSAData, Op->Src);
+  const uint64_t ShiftLeftAmount = (64 - (Op->Width + Op->lsb));
+  const uint64_t ShiftRightAmount = ShiftLeftAmount + Op->lsb;
   Src <<= ShiftLeftAmount;
   Src >>= ShiftRightAmount;
   GD = Src;
@@ -865,20 +867,20 @@ DEF_OP(Sbfe) {
 
 DEF_OP(Select) {
   auto Op = IROp->C<IR::IROp_Select>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp2);
 
   uint64_t ArgTrue;
   uint64_t ArgFalse;
 
   if (OpSize == 4) {
-    ArgTrue = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[2]);
-    ArgFalse = *GetSrc<uint32_t*>(Data->SSAData, Op->Header.Args[3]);
+    ArgTrue = *GetSrc<uint32_t*>(Data->SSAData, Op->TrueVal);
+    ArgFalse = *GetSrc<uint32_t*>(Data->SSAData, Op->FalseVal);
   } else {
-    ArgTrue = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[2]);
-    ArgFalse = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[3]);
+    ArgTrue = *GetSrc<uint64_t*>(Data->SSAData, Op->TrueVal);
+    ArgFalse = *GetSrc<uint64_t*>(Data->SSAData, Op->FalseVal);
   }
 
   bool CompResult;
@@ -892,57 +894,75 @@ 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;
 
-  uint32_t SourceSize = GetOpSize(Data->CurrentIR, Op->Header.Args[0]);
+  constexpr auto AVXRegSize = Core::CPUState::XMM_AVX_REG_SIZE;
+  constexpr auto SSERegSize = Core::CPUState::XMM_SSE_REG_SIZE;
+  constexpr auto SSEBitSize = SSERegSize * 8;
 
-  LOGMAN_THROW_A_FMT(IROp->Size <= 16, "OpSize is too large for VExtractToGPR: {}", IROp->Size);
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto ElementSizeBits = ElementSize * 8;
+  const auto Shift = ElementSizeBits * Op->Index;
 
-  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)
+  const uint32_t SourceSize = GetOpSize(Data->CurrentIR, Op->Vector);
+
+  LOGMAN_THROW_AA_FMT(OpSize <= AVXRegSize,
+                      "OpSize is too large for VExtractToGPR: {}", OpSize);
+
+  if (SourceSize >= SSERegSize) {
+    __uint128_t SourceMask = (1ULL << ElementSizeBits) - 1;
+    if (ElementSize == 8) {
       SourceMask = ~0ULL;
+    }
 
-    __uint128_t Src = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-    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->Header.Args[0]);
-    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;
   }
 }
 
 DEF_OP(Float_ToGPR_ZS) {
   auto Op = IROp->C<IR::IROp_Float_ToGPR_ZS>();
-  uint16_t Conv = (IROp->Size << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (IROp->Size << 8) | Op->SrcElementSize;
   switch (Conv) {
     case 0x0804: { // int64_t <- float
-      int64_t Dst = (int64_t)std::trunc(*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]));
+      const int64_t Dst = (int64_t)std::trunc(*GetSrc<float*>(Data->SSAData, Op->Scalar));
       memcpy(GDP, &Dst, IROp->Size);
       break;
     }
     case 0x0808: { // int64_t <- double
-      int64_t Dst = (int64_t)std::trunc(*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]));
+      const int64_t Dst = (int64_t)std::trunc(*GetSrc<double*>(Data->SSAData, Op->Scalar));
       memcpy(GDP, &Dst, IROp->Size);
       break;
     }
     case 0x0404: { // int32_t <- float
-      int32_t Dst = (int32_t)std::trunc(*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]));
+      const int32_t Dst = (int32_t)std::trunc(*GetSrc<float*>(Data->SSAData, Op->Scalar));
       memcpy(GDP, &Dst, IROp->Size);
       break;
     }
     case 0x0408: { // int32_t <- double
-      int32_t Dst = (int32_t)std::trunc(*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]));
+      const int32_t Dst = (int32_t)std::trunc(*GetSrc<double*>(Data->SSAData, Op->Scalar));
       memcpy(GDP, &Dst, IROp->Size);
       break;
     }
@@ -951,25 +971,25 @@ DEF_OP(Float_ToGPR_ZS) {
 
 DEF_OP(Float_ToGPR_S) {
   auto Op = IROp->C<IR::IROp_Float_ToGPR_S>();
-  uint16_t Conv = (IROp->Size << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (IROp->Size << 8) | Op->SrcElementSize;
   switch (Conv) {
     case 0x0804: { // int64_t <- float
-      int64_t Dst = (int64_t)std::nearbyint(*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]));
+      const int64_t Dst = (int64_t)std::nearbyint(*GetSrc<float*>(Data->SSAData, Op->Scalar));
       memcpy(GDP, &Dst, IROp->Size);
       break;
     }
     case 0x0808: { // int64_t <- double
-      int64_t Dst = (int64_t)std::nearbyint(*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]));
+      const int64_t Dst = (int64_t)std::nearbyint(*GetSrc<double*>(Data->SSAData, Op->Scalar));
       memcpy(GDP, &Dst, IROp->Size);
       break;
     }
     case 0x0404: { // int32_t <- float
-      int32_t Dst = (int32_t)std::nearbyint(*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]));
+      const int32_t Dst = (int32_t)std::nearbyint(*GetSrc<float*>(Data->SSAData, Op->Scalar));
       memcpy(GDP, &Dst, IROp->Size);
       break;
     }
     case 0x0408: { // int32_t <- double
-      int32_t Dst = (int32_t)std::nearbyint(*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]));
+      const int32_t Dst = (int32_t)std::nearbyint(*GetSrc<double*>(Data->SSAData, Op->Scalar));
       memcpy(GDP, &Dst, IROp->Size);
       break;
     }
@@ -980,9 +1000,9 @@ DEF_OP(FCmp) {
   auto Op = IROp->C<IR::IROp_FCmp>();
   uint32_t ResultFlags{};
   if (Op->ElementSize == 4) {
-    float Src1 = *GetSrc<float*>(Data->SSAData, Op->Header.Args[0]);
-    float Src2 = *GetSrc<float*>(Data->SSAData, Op->Header.Args[1]);
-    bool Unordered = std::isnan(Src1) || std::isnan(Src2);
+    const float Src1 = *GetSrc<float*>(Data->SSAData, Op->Scalar1);
+    const float Src2 = *GetSrc<float*>(Data->SSAData, Op->Scalar2);
+    const bool Unordered = std::isnan(Src1) || std::isnan(Src2);
     if (Op->Flags & (1 << IR::FCMP_FLAG_LT)) {
       if (Unordered || (Src1 < Src2)) {
         ResultFlags |= (1 << IR::FCMP_FLAG_LT);
@@ -1000,9 +1020,9 @@ DEF_OP(FCmp) {
     }
   }
   else {
-    double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-    double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
-    bool Unordered = std::isnan(Src1) || std::isnan(Src2);
+    const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Scalar1);
+    const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Scalar2);
+    const bool Unordered = std::isnan(Src1) || std::isnan(Src2);
     if (Op->Flags & (1 << IR::FCMP_FLAG_LT)) {
       if (Unordered || (Src1 < Src2)) {
         ResultFlags |= (1 << IR::FCMP_FLAG_LT);

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

@@ -4,6 +4,7 @@ tags: backend|interpreter
 $end_info$
 */
 
+#include "Interface/Context/Context.h"
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"
@@ -25,20 +26,13 @@ static void SignalReturn(FEXCore::Core::InternalThreadState *Thread) {
 }
 
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
-DEF_OP(GuestCallDirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestCallIndirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
 
 DEF_OP(SignalReturn) {
   SignalReturn(Data->State);
 }
 
 DEF_OP(CallbackReturn) {
-  Data->State->CTX->InterpreterCallbackReturn(Data->State, Data->StackEntry);
+  Data->State->CurrentFrame->Pointers.Interpreter.CallbackReturn(Data->State, Data->StackEntry);
 }
 
 DEF_OP(ExitFunction) {
@@ -48,7 +42,7 @@ DEF_OP(ExitFunction) {
   uintptr_t* ContextPtr = reinterpret_cast<uintptr_t*>(Data->State->CurrentFrame);
 
   void *ContextData = reinterpret_cast<void*>(ContextPtr);
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
+  void *Src = GetSrc<void*>(Data->SSAData, Op->NewRIP);
 
   memcpy(ContextData, Src, OpSize);
 
@@ -57,22 +51,22 @@ DEF_OP(ExitFunction) {
 
 DEF_OP(Jump) {
   auto Op = IROp->C<IR::IROp_Jump>();
-  uintptr_t ListBegin = Data->CurrentIR->GetListData();
-  uintptr_t DataBegin = Data->CurrentIR->GetData();
+  const uintptr_t ListBegin = Data->CurrentIR->GetListData();
+  const uintptr_t DataBegin = Data->CurrentIR->GetData();
 
-  Data->BlockIterator = IR::NodeIterator(ListBegin, DataBegin, Op->Header.Args[0]);
+  Data->BlockIterator = IR::NodeIterator(ListBegin, DataBegin, Op->TargetBlock);
   Data->BlockResults.Redo = true;
 }
 
 DEF_OP(CondJump) {
   auto Op = IROp->C<IR::IROp_CondJump>();
-  uintptr_t ListBegin = Data->CurrentIR->GetListData();
-  uintptr_t DataBegin = Data->CurrentIR->GetData();
+  const uintptr_t ListBegin = Data->CurrentIR->GetListData();
+  const uintptr_t DataBegin = Data->CurrentIR->GetData();
 
   bool CompResult;
 
-  uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp1);
-  uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp2);
+  const uint64_t Src1 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp1);
+  const uint64_t Src2 = *GetSrc<uint64_t*>(Data->SSAData, Op->Cmp2);
 
   if (Op->CompareSize == 4)
     CompResult = IsConditionTrue<uint32_t, int32_t, float>(Op->Cond.Val, Src1, Src2);
@@ -133,7 +127,7 @@ DEF_OP(Thunk) {
   auto Op = IROp->C<IR::IROp_Thunk>();
 
   auto thunkFn = Data->State->CTX->ThunkHandler->LookupThunk(Op->ThunkNameHash);
-  thunkFn(*GetSrc<void**>(Data->SSAData, Op->Header.Args[0]));
+  thunkFn(*GetSrc<void**>(Data->SSAData, Op->ArgPtr));
 }
 
 DEF_OP(ValidateCode) {
@@ -147,15 +141,15 @@ DEF_OP(ValidateCode) {
   }
 }
 
-DEF_OP(RemoveThreadCodeEntry) {
-  Data->State->CTX->RemoveThreadCodeEntry(Data->State, Data->CurrentEntry);
+DEF_OP(ThreadRemoveCodeEntry) {
+  Data->State->CTX->ThreadRemoveCodeEntryFromJit(Data->State->CurrentFrame, Data->CurrentEntry);
 }
 
 DEF_OP(CPUID) {
   auto Op = IROp->C<IR::IROp_CPUID>();
   uint64_t *DstPtr = GetDest<uint64_t*>(Data->SSAData, Node);
-  uint64_t Arg = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
-  uint64_t Leaf = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[1]);
+  const uint64_t Arg = *GetSrc<uint64_t*>(Data->SSAData, Op->Function);
+  const uint64_t Leaf = *GetSrc<uint64_t*>(Data->SSAData, Op->Leaf);
 
   auto Results = Data->State->CTX->CPUID.RunFunction(Arg, Leaf);
   memcpy(DstPtr, &Results, sizeof(uint32_t) * 4);

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

@@ -13,53 +13,77 @@ $end_info$
 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>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_VInsGPR>();
+  const auto OpSize = IROp->Size;
 
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  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);
 }
 
 DEF_OP(VCastFromGPR) {
   auto Op = IROp->C<IR::IROp_VCastFromGPR>();
-  memcpy(GDP, GetSrc<void*>(Data->SSAData, Op->Header.Args[0]), Op->Header.ElementSize);
+  memcpy(GDP, GetSrc<void*>(Data->SSAData, Op->Src), Op->Header.ElementSize);
 }
 
 DEF_OP(Float_FromGPR_S) {
   auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
 
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
   switch (Conv) {
     case 0x0404: { // Float <- int32_t
-      float Dst = (float)*GetSrc<int32_t*>(Data->SSAData, Op->Header.Args[0]);
+      const float Dst = (float)*GetSrc<int32_t*>(Data->SSAData, Op->Src);
       memcpy(GDP, &Dst, Op->Header.ElementSize);
       break;
     }
     case 0x0408: { // Float <- int64_t
-      float Dst = (float)*GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[0]);
+      const float Dst = (float)*GetSrc<int64_t*>(Data->SSAData, Op->Src);
       memcpy(GDP, &Dst, Op->Header.ElementSize);
       break;
     }
     case 0x0804: { // Double <- int32_t
-      double Dst = (double)*GetSrc<int32_t*>(Data->SSAData, Op->Header.Args[0]);
+      const double Dst = (double)*GetSrc<int32_t*>(Data->SSAData, Op->Src);
       memcpy(GDP, &Dst, Op->Header.ElementSize);
       break;
     }
     case 0x0808: { // Double <- int64_t
-      double Dst = (double)*GetSrc<int64_t*>(Data->SSAData, Op->Header.Args[0]);
+      const double Dst = (double)*GetSrc<int64_t*>(Data->SSAData, Op->Src);
       memcpy(GDP, &Dst, Op->Header.ElementSize);
       break;
     }
@@ -68,15 +92,15 @@ DEF_OP(Float_FromGPR_S) {
 
 DEF_OP(Float_FToF) {
   auto Op = IROp->C<IR::IROp_Float_FToF>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
   switch (Conv) {
     case 0x0804: { // Double <- Float
-      double Dst = (double)*GetSrc<float*>(Data->SSAData, Op->Header.Args[0]);
+      const double Dst = (double)*GetSrc<float*>(Data->SSAData, Op->Scalar);
       memcpy(GDP, &Dst, 8);
       break;
     }
     case 0x0408: { // Float <- Double
-      float Dst = (float)*GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
+      const float Dst = (float)*GetSrc<double*>(Data->SSAData, Op->Scalar);
       memcpy(GDP, &Dst, 4);
       break;
     }
@@ -86,68 +110,78 @@ DEF_OP(Float_FToF) {
 
 DEF_OP(Vector_SToF) {
   auto Op = IROp->C<IR::IROp_Vector_SToF>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
 
-  auto Func = [](auto a, auto min, auto max) { return a; };
-  switch (Op->Header.ElementSize) {
+  const auto Func = [](auto a, auto min, auto max) { return a; };
+  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>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_Vector_FToZS>();
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
 
-  auto Func = [](auto a, auto min, auto max) { return std::trunc(a); };
-  switch (Op->Header.ElementSize) {
+  const auto Func = [](auto a, auto min, auto max) { return std::trunc(a); };
+  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>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_Vector_FToS>();
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  uint8_t Elements = OpSize / Op->Header.ElementSize;
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
 
-  auto Func = [](auto a, auto min, auto max) { return std::nearbyint(a); };
-  switch (Op->Header.ElementSize) {
+  const auto Func = [](auto a, auto min, auto max) { return std::nearbyint(a); };
+  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>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_Vector_FToF>();
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t ElementSize = Op->Header.ElementSize;
+  const uint16_t Conv = (ElementSize << 8) | Op->SrcElementSize;
 
-  auto Func = [](auto a, auto min, auto max) { return a; };
+  const auto Func = [](auto a, auto min, auto max) { return a; };
   switch (Conv) {
     case 0x0804: { // Double <- float
       // Only the lower elements from the source
@@ -161,56 +195,59 @@ 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>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_Vector_FToI>();
+  const uint8_t OpSize = IROp->Size;
 
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  uint8_t Tmp[16]{};
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
 
-  uint8_t Elements = OpSize / Op->Header.ElementSize;
-  auto Func_Nearest = [](auto a) { return std::rint(a); };
-  auto Func_Neg = [](auto a) { return std::floor(a); };
-  auto Func_Pos = [](auto a) { return std::ceil(a); };
-  auto Func_Trunc = [](auto a) { return std::trunc(a); };
-  auto Func_Host = [](auto a) { return std::rint(a); };
+  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); };
+  const auto Func_Trunc = [](auto a) { return std::trunc(a); };
+  const auto Func_Host = [](auto a) { return std::rint(a); };
 
   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/EncryptionOps.cpp

@@ -360,7 +360,7 @@ namespace FEXCore::CPU {
 
 DEF_OP(AESImc) {
   auto Op = IROp->C<IR::IROp_VAESImc>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Vector);
 
   // Pseudo-code
   // Dst = InvMixColumns(STATE)
@@ -371,8 +371,8 @@ DEF_OP(AESImc) {
 
 DEF_OP(AESEnc) {
   auto Op = IROp->C<IR::IROp_VAESEnc>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);
 
   // Pseudo-code
   // STATE = Src1
@@ -391,8 +391,8 @@ DEF_OP(AESEnc) {
 
 DEF_OP(AESEncLast) {
   auto Op = IROp->C<IR::IROp_VAESEncLast>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);
 
   // Pseudo-code
   // STATE = Src1
@@ -409,8 +409,8 @@ DEF_OP(AESEncLast) {
 
 DEF_OP(AESDec) {
   auto Op = IROp->C<IR::IROp_VAESDec>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);
 
   // Pseudo-code
   // STATE = Src1
@@ -429,8 +429,8 @@ DEF_OP(AESDec) {
 
 DEF_OP(AESDecLast) {
   auto Op = IROp->C<IR::IROp_VAESDecLast>();
-  __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-  __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[1]);
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);
 
   // Pseudo-code
   // STATE = Src1
@@ -447,7 +447,7 @@ DEF_OP(AESDecLast) {
 
 DEF_OP(AESKeyGenAssist) {
   auto Op = IROp->C<IR::IROp_VAESKeyGenAssist>();
-  uint8_t *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint8_t *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->Src);
 
   // Pseudo-code
   // X3 = Src1[127:96]
@@ -513,6 +513,44 @@ DEF_OP(CRC32) {
   memcpy(GDP, &Tmp, sizeof(Tmp));
 }
 
+DEF_OP(PCLMUL) {
+  auto Op = IROp->C<IR::IROp_PCLMUL>();
+
+  const auto Selector = Op->Selector;
+  auto* Dst  = GetDest<uint64_t*>(Data->SSAData, Node);
+  auto* Src1 = GetSrc<uint64_t*>(Data->SSAData, Op->Src1);
+  auto* Src2 = GetSrc<uint64_t*>(Data->SSAData, Op->Src2);
+
+  const uint64_t TMP1 = (Selector & 0x01) == 0 ? Src1[0] : Src1[1];
+  const uint64_t TMP2 = (Selector & 0x10) == 0 ? Src2[0] : Src2[1];
+  
+  const auto make_lo = [](uint64_t lhs, uint64_t rhs) {
+    uint64_t result = 0;
+
+    for (size_t i = 0; i < 64; i++) {
+      if ((lhs & (1ULL << i)) != 0) {
+        result ^= rhs << i;
+      }
+    }
+
+    return result;
+  };
+  const auto make_hi = [](uint64_t lhs, uint64_t rhs) {
+    uint64_t result = 0;
+
+    for (size_t i = 1; i < 64; i++) {
+      if ((lhs & (1ULL << i)) != 0) {
+        result ^= rhs >> (64 - i);
+      }
+    }
+
+    return result;
+  };
+
+  Dst[0] = make_lo(TMP1, TMP2);
+  Dst[1] = make_hi(TMP1, TMP2);
+}
+
 #undef DEF_OP
 
 } // namespace FEXCore::CPU

External/FEXCore/Source/Interface/Core/Interpreter/F80Ops.cpp

@@ -20,99 +20,90 @@ DEF_OP(F80LOADFCW) {
 
 DEF_OP(F80ADD) {
   auto Op = IROp->C<IR::IROp_F80Add>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FADD(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FADD(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80SUB) {
   auto Op = IROp->C<IR::IROp_F80Sub>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FSUB(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FSUB(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80MUL) {
   auto Op = IROp->C<IR::IROp_F80Mul>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FMUL(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FMUL(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80DIV) {
   auto Op = IROp->C<IR::IROp_F80Div>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FDIV(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FDIV(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80FYL2X) {
   auto Op = IROp->C<IR::IROp_F80FYL2X>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FYL2X(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FYL2X(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80ATAN) {
   auto Op = IROp->C<IR::IROp_F80ATAN>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FATAN(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FATAN(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80FPREM1) {
   auto Op = IROp->C<IR::IROp_F80FPREM1>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FREM1(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FREM1(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80FPREM) {
   auto Op = IROp->C<IR::IROp_F80FPREM>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FREM(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FREM(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80SCALE) {
   auto Op = IROp->C<IR::IROp_F80SCALE>();
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FSCALE(Src1, Src2);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
+  const auto Tmp = X80SoftFloat::FSCALE(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80CVT) {
   auto Op = IROp->C<IR::IROp_F80CVT>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
 
   switch (OpSize) {
     case 4: {
@@ -131,9 +122,9 @@ DEF_OP(F80CVT) {
 
 DEF_OP(F80CVTINT) {
   auto Op = IROp->C<IR::IROp_F80CVTInt>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
 
   switch (OpSize) {
     case 2: {
@@ -160,13 +151,13 @@ DEF_OP(F80CVTTO) {
 
   switch (Op->SrcSize) {
     case 4: {
-      float Src = *GetSrc<float *>(Data->SSAData, Op->Header.Args[0]);
+      float Src = *GetSrc<float *>(Data->SSAData, Op->X80Src);
       X80SoftFloat Tmp = Src;
       memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
       break;
     }
     case 8: {
-      double Src = *GetSrc<double *>(Data->SSAData, Op->Header.Args[0]);
+      double Src = *GetSrc<double *>(Data->SSAData, Op->X80Src);
       X80SoftFloat Tmp = Src;
       memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
       break;
@@ -180,13 +171,13 @@ DEF_OP(F80CVTTOINT) {
 
   switch (Op->SrcSize) {
     case 2: {
-      int16_t Src = *GetSrc<int16_t*>(Data->SSAData, Op->Header.Args[0]);
+      int16_t Src = *GetSrc<int16_t*>(Data->SSAData, Op->Src);
       X80SoftFloat Tmp = Src;
       memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
       break;
     }
     case 4: {
-      int32_t Src = *GetSrc<int32_t*>(Data->SSAData, Op->Header.Args[0]);
+      int32_t Src = *GetSrc<int32_t*>(Data->SSAData, Op->Src);
       X80SoftFloat Tmp = Src;
       memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
       break;
@@ -197,77 +188,73 @@ DEF_OP(F80CVTTOINT) {
 
 DEF_OP(F80ROUND) {
   auto Op = IROp->C<IR::IROp_F80Round>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FRNDINT(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FRNDINT(Src);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80F2XM1) {
   auto Op = IROp->C<IR::IROp_F80F2XM1>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::F2XM1(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::F2XM1(Src);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80TAN) {
   auto Op = IROp->C<IR::IROp_F80TAN>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FTAN(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FTAN(Src);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80SQRT) {
   auto Op = IROp->C<IR::IROp_F80SQRT>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FSQRT(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FSQRT(Src);
 
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80SIN) {
   auto Op = IROp->C<IR::IROp_F80SIN>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FSIN(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FSIN(Src);
+
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80COS) {
   auto Op = IROp->C<IR::IROp_F80COS>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FCOS(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FCOS(Src);
+
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80XTRACT_EXP) {
   auto Op = IROp->C<IR::IROp_F80XTRACT_EXP>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FXTRACT_EXP(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FXTRACT_EXP(Src);
+
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80XTRACT_SIG) {
   auto Op = IROp->C<IR::IROp_F80XTRACT_SIG>();
-  X80SoftFloat Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Tmp;
-  Tmp = X80SoftFloat::FXTRACT_SIG(Src);
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+  const auto Tmp = X80SoftFloat::FXTRACT_SIG(Src);
+
   memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
 }
 
 DEF_OP(F80CMP) {
   auto Op = IROp->C<IR::IROp_F80Cmp>();
   uint32_t ResultFlags{};
-  X80SoftFloat Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]);
-  X80SoftFloat Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[1]);
+  const auto Src1 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src1);
+  const auto Src2 = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src2);
   bool eq, lt, nan;
   X80SoftFloat::FCMP(Src1, Src2, &eq, &lt, &nan);
   if (Op->Flags & (1 << IR::FCMP_FLAG_LT) &&
@@ -288,7 +275,7 @@ DEF_OP(F80CMP) {
 
 DEF_OP(F80BCDLOAD) {
   auto Op = IROp->C<IR::IROp_F80BCDLoad>();
-  uint8_t *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+  const uint8_t *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->X80Src);
   uint64_t BCD{};
   // We walk through each uint8_t and pull out the BCD encoding
   // Each 4bit split is a digit
@@ -323,7 +310,7 @@ DEF_OP(F80BCDLOAD) {
 
 DEF_OP(F80BCDSTORE) {
   auto Op = IROp->C<IR::IROp_F80BCDStore>();
-  X80SoftFloat Src1 = X80SoftFloat::FRNDINT(*GetSrc<X80SoftFloat*>(Data->SSAData, Op->Header.Args[0]));
+  X80SoftFloat Src1 = X80SoftFloat::FRNDINT(*GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src));
   bool Negative = Src1.Sign;
 
   // Clear the Sign bit
@@ -358,74 +345,74 @@ DEF_OP(F80BCDSTORE) {
 
 DEF_OP(F64SIN) {
   auto Op = IROp->C<IR::IROp_F64SIN>();
-  double Src = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-  double Tmp = sin(Src);
+  const double Src = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Tmp = sin(Src);
   memcpy(GDP, &Tmp, sizeof(double));
 }
 
 DEF_OP(F64COS) {
   auto Op = IROp->C<IR::IROp_F64COS>();
-  double Src = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-  double Tmp = cos(Src);
+  const double Src = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Tmp = cos(Src);
   memcpy(GDP, &Tmp, sizeof(double));
 }
 
 DEF_OP(F64TAN) {
   auto Op = IROp->C<IR::IROp_F64TAN>();
-  double Src = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-  double Tmp = tan(Src);
+  const double Src = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Tmp = tan(Src);
   memcpy(GDP, &Tmp, sizeof(double));
 }
 
 DEF_OP(F64F2XM1) {
   auto Op = IROp->C<IR::IROp_F64F2XM1>();
-  double Src = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-  double Tmp = exp2(Src) - 1.0;
+  const double Src = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Tmp = exp2(Src) - 1.0;
   memcpy(GDP, &Tmp, sizeof(double));
 }
 
 DEF_OP(F64ATAN) {
   auto Op = IROp->C<IR::IROp_F64ATAN>();
-  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
-  double Tmp = atan2(Src1, Src2);
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src1);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double Tmp = atan2(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(double));
 }
 
 DEF_OP(F64FPREM) {
   auto Op = IROp->C<IR::IROp_F64FPREM>();
-  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
-  double Tmp = fmod(Src1, Src2);
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src1);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double Tmp = fmod(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(double));
 }
 
 DEF_OP(F64FPREM1) {
   auto Op = IROp->C<IR::IROp_F64FPREM1>();
-  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
-  double Tmp = remainder(Src1, Src2);
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src1);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double Tmp = remainder(Src1, Src2);
 
   memcpy(GDP, &Tmp, sizeof(double));
 }
 
 DEF_OP(F64FYL2X) {
   auto Op = IROp->C<IR::IROp_F64FYL2X>();
-  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
-  double Tmp = Src2 * log2(Src1);
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double Tmp = Src2 * log2(Src1);
 
   memcpy(GDP, &Tmp, sizeof(double));
 }
 
 DEF_OP(F64SCALE) {
   auto Op = IROp->C<IR::IROp_F64SCALE>();
-  double Src1 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[0]);
-  double Src2 = *GetSrc<double*>(Data->SSAData, Op->Header.Args[1]);
-  double trunc = (double)(int64_t)(Src2); //truncate
-  double Tmp = Src1 * exp2(trunc);
+  const double Src1 = *GetSrc<double*>(Data->SSAData, Op->Src1);
+  const double Src2 = *GetSrc<double*>(Data->SSAData, Op->Src2);
+  const double trunc = (double)(int64_t)(Src2); //truncate
+  const double Tmp = Src1 * exp2(trunc);
 
   memcpy(GDP, &Tmp, sizeof(double));
 }

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

@@ -14,7 +14,7 @@ namespace FEXCore::CPU {
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(GetHostFlag) {
   auto Op = IROp->C<IR::IROp_GetHostFlag>();
-  GD = (*GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]) >> Op->Flag) & 1;
+  GD = (*GetSrc<uint64_t*>(Data->SSAData, Op->Value) >> Op->Flag) & 1;
 }
 #undef DEF_OP
 

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

@@ -8,6 +8,7 @@
 #include <FEXCore/IR/IntrusiveIRList.h>
 
 namespace FEXCore::CPU {
+class Dispatcher;
 class X86DispatchGenerator;
 class Arm64DispatchGenerator;
 
@@ -20,7 +21,7 @@ using DestMapType = std::vector<uint32_t>;
 
 class InterpreterCore final : public CPUBackend {
 public:
-  explicit InterpreterCore(FEXCore::Context::Context *ctx,
+  explicit InterpreterCore(Dispatcher *Dispatch,
                            FEXCore::Core::InternalThreadState *Thread);
 
   [[nodiscard]] std::string GetName() override { return "Interpreter"; }
@@ -28,23 +29,19 @@ public:
   [[nodiscard]] void *CompileCode(uint64_t Entry,
                                   FEXCore::IR::IRListView const *IR,
                                   FEXCore::Core::DebugData *DebugData,
-                                  FEXCore::IR::RegisterAllocationData *RAData) override;
+                                  FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) override;
 
   [[nodiscard]] void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }
 
   [[nodiscard]] bool NeedsOpDispatch() override { return true; }
 
-  void CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread);
-
   static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
-
-  bool NeedsRetainedIRCopy() const override { return true; }
+  
+  void ClearCache() override;
 
 private:
-  FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *State;
-
-  std::unique_ptr<Dispatcher> Dispatcher{};
+  size_t BufferUsed;
+  Dispatcher *Dispatch;
 };
 
 template<typename T>

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

@@ -9,70 +9,101 @@
 #include <FEXCore/Core/SignalDelegator.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/MathUtils.h>
 
 #include <memory>
 #include <signal.h>
 #include <stdint.h>
-#include <unordered_map>
 #include <utility>
 
 #include "InterpreterOps.h"
 
+#if defined(_M_X86_64)
+  #include "Interface/Core/Dispatcher/X86Dispatcher.h"
+#elif defined(_M_ARM_64)
+  #include "Interface/Core/Dispatcher/Arm64Dispatcher.h"
+#else
+  #error missing arch
+#endif
+
+static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
+static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 128;
+
 namespace FEXCore::IR {
   class IRListView;
   class RegisterAllocationData;
 }
 
+
 namespace FEXCore::CPU {
-class CPUBackend;
 
-static void InterpreterExecution(FEXCore::Core::CpuStateFrame *Frame) {
-  auto Thread = Frame->Thread;
+InterpreterCore::InterpreterCore(Dispatcher *Dispatcher, FEXCore::Core::InternalThreadState *Thread)
+  : CPUBackend(Thread, INITIAL_CODE_SIZE, MAX_CODE_SIZE)
+  , Dispatch(Dispatcher)
+  {
 
-  auto LocalEntry = Thread->LocalIRCache.find(Thread->CurrentFrame->State.rip);
+  auto &Interpreter = Thread->CurrentFrame->Pointers.Interpreter;
 
-  InterpreterOps::InterpretIR(Thread, Thread->CurrentFrame->State.rip, LocalEntry->second.IR.get(), LocalEntry->second.DebugData.get());
-}
+  Interpreter.FragmentExecuter = reinterpret_cast<uint64_t>(&InterpreterOps::InterpretIR);
 
-InterpreterCore::InterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
-  : CTX {ctx}
-  , State {Thread} {
-
-  CreateAsmDispatch(ctx, Thread);
+  ClearCache();
 }
 
 void InterpreterCore::InitializeSignalHandlers(FEXCore::Context::Context *CTX) {
-  CTX->SignalDelegation->RegisterHostSignalHandler(SignalDelegator::SIGNAL_FOR_PAUSE, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-    InterpreterCore *Core = reinterpret_cast<InterpreterCore*>(Thread->CPUBackend.get());
-    return Core->Dispatcher->HandleSignalPause(Signal, info, ucontext);
-  }, true);
 
 #ifdef _M_ARM_64
   CTX->SignalDelegation->RegisterHostSignalHandler(SIGBUS, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
     return FEXCore::ArchHelpers::Arm64::HandleSIGBUS(true, Signal, info, ucontext);
   }, true);
 #endif
-
-  auto GuestSignalHandler = [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) -> bool {
-    InterpreterCore *Core = reinterpret_cast<InterpreterCore*>(Thread->CPUBackend.get());
-    return Core->Dispatcher->HandleGuestSignal(Signal, info, ucontext, GuestAction, GuestStack);
-  };
-
-  for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
-    CTX->SignalDelegation->RegisterHostSignalHandlerForGuest(Signal, GuestSignalHandler);
-  }
 }
 
-void *InterpreterCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData) {
-  return reinterpret_cast<void*>(InterpreterExecution);
+void *InterpreterCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) {
+
+  const auto IRSize = AlignUp(IR->GetInlineSize(), 16);
+  const auto MaxSize = IRSize + Dispatcher::MaxInterpreterTrampolineSize + GDBEnabled * Dispatcher::MaxGDBPauseCheckSize;
+
+  if ((BufferUsed + MaxSize) > CurrentCodeBuffer->Size) {
+    ThreadState->CTX->ClearCodeCache(ThreadState);
+  }
+
+  const auto BufferStart = CurrentCodeBuffer->Ptr + BufferUsed;
+
+  auto DestBuffer = BufferStart;
+
+  if (GDBEnabled) {
+    const auto GDBSize = Dispatch->GenerateGDBPauseCheck(DestBuffer, Entry);
+    DestBuffer += GDBSize;
+    BufferUsed += GDBSize;
+  }
+
+  const auto TrampolineSize = Dispatch->GenerateInterpreterTrampoline(DestBuffer);
+  DestBuffer += TrampolineSize;
+  BufferUsed += TrampolineSize;
+
+
+  IR->Serialize(DestBuffer);
+  DestBuffer += IRSize;
+  BufferUsed += IRSize;
+
+  return BufferStart;
+}
+
+void InterpreterCore::ClearCache() {
+  // Calling this one is needed to setup the initial CurrentCodeBuffer
+  [[maybe_unused]] auto CodeBuffer = GetEmptyCodeBuffer();
+  BufferUsed = 0;
 }
 
 std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
-  return std::make_unique<InterpreterCore>(ctx, Thread);
+  return std::make_unique<InterpreterCore>(ctx->Dispatcher.get(), Thread);
 }
 
 void InitializeInterpreterSignalHandlers(FEXCore::Context::Context *CTX) {
   InterpreterCore::InitializeSignalHandlers(CTX);
 }
 
+CPUBackendFeatures GetInterpreterBackendFeatures() {
+  return CPUBackendFeatures { };
+}
 }

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

@@ -12,10 +12,11 @@ namespace FEXCore::Core {
 
 namespace FEXCore::CPU {
 class CPUBackend;
+struct DispatcherConfig;
 
 [[nodiscard]] std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx,
                                                                 FEXCore::Core::InternalThreadState *Thread);
-
 void InitializeInterpreterSignalHandlers(FEXCore::Context::Context *CTX);
+CPUBackendFeatures GetInterpreterBackendFeatures();
 
 } // namespace FEXCore::CPU

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

@@ -159,21 +159,26 @@
   break;                                            \
   }
 
+struct InterpVector256 {
+  __uint128_t Lower;
+  __uint128_t Upper;
+};
+
 template<typename Res>
 Res GetDest(void* SSAData, FEXCore::IR::OrderedNodeWrapper Op) {
-  auto DstPtr = &reinterpret_cast<__uint128_t*>(SSAData)[Op.ID().Value];
+  auto DstPtr = &reinterpret_cast<InterpVector256*>(SSAData)[Op.ID().Value];
   return reinterpret_cast<Res>(DstPtr);
 }
 
 template<typename Res>
 Res GetDest(void* SSAData, FEXCore::IR::NodeID Op) {
-  auto DstPtr = &reinterpret_cast<__uint128_t*>(SSAData)[Op.Value];
+  auto DstPtr = &reinterpret_cast<InterpVector256*>(SSAData)[Op.Value];
   return reinterpret_cast<Res>(DstPtr);
 }
 
 
 template<typename Res>
 Res GetSrc(void* SSAData, FEXCore::IR::OrderedNodeWrapper Src) {
-  auto DstPtr = &reinterpret_cast<__uint128_t*>(SSAData)[Src.ID().Value];
+  auto DstPtr = &reinterpret_cast<InterpVector256*>(SSAData)[Src.ID().Value];
   return reinterpret_cast<Res>(DstPtr);
 }

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

@@ -1,5 +1,6 @@
 #include "Interface/Context/Context.h"
 #include "Interface/Core/CPUID.h"
+#include "InterpreterDefines.h"
 #include "InterpreterOps.h"
 #include "F80Ops.h"
 
@@ -112,8 +113,6 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(ATOMICFETCHNEG,         AtomicFetchNeg);
 
   // Branch ops
-  REGISTER_OP(GUESTCALLDIRECT,        GuestCallDirect);
-  REGISTER_OP(GUESTCALLINDIRECT,      GuestCallIndirect);
   REGISTER_OP(SIGNALRETURN,           SignalReturn);
   REGISTER_OP(CALLBACKRETURN,         CallbackReturn);
   REGISTER_OP(EXITFUNCTION,           ExitFunction);
@@ -123,7 +122,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(INLINESYSCALL,          InlineSyscall);
   REGISTER_OP(THUNK,                  Thunk);
   REGISTER_OP(VALIDATECODE,           ValidateCode);
-  REGISTER_OP(REMOVETHREADCODEENTRY,        RemoveThreadCodeEntry);
+  REGISTER_OP(THREADREMOVECODEENTRY,        ThreadRemoveCodeEntry);
   REGISTER_OP(CPUID,                  CPUID);
 
   // Conversion ops
@@ -155,8 +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);
 
@@ -166,6 +163,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(CODEBLOCK,              NoOp);
   REGISTER_OP(BEGINBLOCK,             NoOp);
   REGISTER_OP(ENDBLOCK,               NoOp);
+  REGISTER_OP(GUESTOPCODE,            NoOp);
   REGISTER_OP(FENCE,                  Fence);
   REGISTER_OP(BREAK,                  Break);
   REGISTER_OP(PHI,                    NoOp);
@@ -181,13 +179,10 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   // Move ops
   REGISTER_OP(EXTRACTELEMENTPAIR,     ExtractElementPair);
   REGISTER_OP(CREATEELEMENTPAIR,      CreateElementPair);
-  REGISTER_OP(MOV,                    Mov);
 
   // Vector ops
   REGISTER_OP(VECTORZERO,             VectorZero);
   REGISTER_OP(VECTORIMM,              VectorImm);
-  REGISTER_OP(SPLATVECTOR2,           SplatVector);
-  REGISTER_OP(SPLATVECTOR4,           SplatVector);
   REGISTER_OP(VMOV,                   VMov);
   REGISTER_OP(VAND,                   VAnd);
   REGISTER_OP(VBIC,                   VBic);
@@ -246,18 +241,13 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(VUSHRS,                 VUShrS);
   REGISTER_OP(VSSHRS,                 VSShrS);
   REGISTER_OP(VINSELEMENT,            VInsElement);
-  REGISTER_OP(VINSSCALARELEMENT,      VInsScalarElement);
-  REGISTER_OP(VEXTRACTELEMENT,        VExtractElement);
   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);
   REGISTER_OP(VUSHRNI,                VUShrNI);
   REGISTER_OP(VUSHRNI2,               VUShrNI2);
-  REGISTER_OP(VBITCAST,               VBitcast);
   REGISTER_OP(VSXTL,                  VSXTL);
   REGISTER_OP(VSXTL2,                 VSXTL2);
   REGISTER_OP(VUXTL,                  VUXTL);
@@ -284,6 +274,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(VAESDECLAST,            AESDecLast);
   REGISTER_OP(VAESKEYGENASSIST,       AESKeyGenAssist);
   REGISTER_OP(CRC32,                  CRC32);
+  REGISTER_OP(PCLMUL,                 PCLMUL);
 
   // F80 ops
   REGISTER_OP(F80LOADFCW,             F80LOADFCW);
@@ -333,32 +324,37 @@ void InterpreterOps::Op_Unhandled(FEXCore::IR::IROp_Header *IROp, IROpData *Data
 void InterpreterOps::Op_NoOp(FEXCore::IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node) {
 }
 
-void InterpreterOps::InterpretIR(FEXCore::Core::InternalThreadState *Thread, uint64_t Entry, FEXCore::IR::IRListView *CurrentIR, FEXCore::Core::DebugData *DebugData) {
+void InterpreterOps::InterpretIR(FEXCore::Core::CpuStateFrame *Frame, FEXCore::IR::IRListView const *CurrentIR) {
   volatile void *StackEntry = alloca(0);
 
-  uintptr_t ListSize = CurrentIR->GetSSACount();
+  const uintptr_t ListSize = CurrentIR->GetSSACount();
 
   static_assert(sizeof(FEXCore::IR::IROp_Header) == 4);
   static_assert(sizeof(FEXCore::IR::OrderedNode) == 16);
 
   auto BlockEnd = CurrentIR->GetBlocks().end();
 
-  InterpreterOps::IROpData OpData{};
-  OpData.State = Thread;
-  OpData.SSAData = alloca(ListSize * 16);
-  OpData.CurrentEntry = Entry;
-  OpData.CurrentIR = CurrentIR;
-  OpData.StackEntry = StackEntry;
-  OpData.BlockIterator = CurrentIR->GetBlocks().begin();
+  constexpr size_t ListEntrySizeInBytes = sizeof(InterpVector256);
+  const size_t SSADataSize = ListSize * ListEntrySizeInBytes;
 
-  // Clear them all to zero. Required for Zero-extend semantics
-  memset(OpData.SSAData, 0, ListSize * 16);
+  InterpreterOps::IROpData OpData{
+    .State = Frame->Thread,
+    .CurrentEntry = Frame->State.rip,
+    .CurrentIR = CurrentIR,
+    .StackEntry = StackEntry,
+    .SSAData = alloca(SSADataSize),
+    .BlockResults = {},
+    .BlockIterator = CurrentIR->GetBlocks().begin(),
+  };
+
+  // Clear all SSAData entries to zero. Required for Zero-extend semantics
+  memset(OpData.SSAData, 0, SSADataSize);
 
   while (1) {
     using namespace FEXCore::IR;
     auto [BlockNode, BlockHeader] = OpData.BlockIterator();
     auto BlockIROp = BlockHeader->CW<IROp_CodeBlock>();
-    LOGMAN_THROW_A_FMT(BlockIROp->Header.Op == IR::OP_CODEBLOCK, "IR type failed to be a code block");
+    LOGMAN_THROW_AA_FMT(BlockIROp->Header.Op == IR::OP_CODEBLOCK, "IR type failed to be a code block");
 
     // Reset the block results per block
     memset(&OpData.BlockResults, 0, sizeof(OpData.BlockResults));

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

@@ -47,14 +47,14 @@ namespace FEXCore::CPU {
   class InterpreterOps {
 
     public:
-      static void InterpretIR(FEXCore::Core::InternalThreadState *Thread, uint64_t Entry, FEXCore::IR::IRListView *CurrentIR, FEXCore::Core::DebugData *DebugData);
+      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{};
         uint64_t CurrentEntry{};
-        FEXCore::IR::IRListView *CurrentIR{};
+        FEXCore::IR::IRListView const *CurrentIR{};
         volatile void *StackEntry{};
         void *SSAData{};
         struct {
@@ -142,8 +142,6 @@ namespace FEXCore::CPU {
   DEF_OP(AtomicFetchNeg);
 
   ///< Branch ops
-  DEF_OP(GuestCallDirect);
-  DEF_OP(GuestCallIndirect);
   DEF_OP(SignalReturn);
   DEF_OP(CallbackReturn);
   DEF_OP(ExitFunction);
@@ -153,7 +151,7 @@ namespace FEXCore::CPU {
   DEF_OP(InlineSyscall);
   DEF_OP(Thunk);
   DEF_OP(ValidateCode);
-  DEF_OP(RemoveThreadCodeEntry);
+  DEF_OP(ThreadRemoveCodeEntry);
   DEF_OP(CPUID);
 
   ///< Conversion ops
@@ -183,8 +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);
 
@@ -209,7 +205,6 @@ namespace FEXCore::CPU {
   ///< Vector ops
   DEF_OP(VectorZero);
   DEF_OP(VectorImm);
-  DEF_OP(SplatVector);
   DEF_OP(VMov);
   DEF_OP(VAnd);
   DEF_OP(VBic);
@@ -266,18 +261,13 @@ namespace FEXCore::CPU {
   DEF_OP(VUShrS);
   DEF_OP(VSShrS);
   DEF_OP(VInsElement);
-  DEF_OP(VInsScalarElement);
-  DEF_OP(VExtractElement);
   DEF_OP(VDupElement);
   DEF_OP(VExtr);
-  DEF_OP(VSLI);
-  DEF_OP(VSRI);
   DEF_OP(VUShrI);
   DEF_OP(VSShrI);
   DEF_OP(VShlI);
   DEF_OP(VUShrNI);
   DEF_OP(VUShrNI2);
-  DEF_OP(VBitcast);
   DEF_OP(VSXTL);
   DEF_OP(VSXTL2);
   DEF_OP(VUXTL);
@@ -304,6 +294,7 @@ namespace FEXCore::CPU {
   DEF_OP(AESDecLast);
   DEF_OP(AESKeyGenAssist);
   DEF_OP(CRC32);
+  DEF_OP(PCLMUL);
 
   ///< F80 ops
   DEF_OP(F80LOADFCW);
@@ -408,7 +399,7 @@ namespace FEXCore::CPU {
     return CompResult;
   }
 
-  static uint8_t GetOpSize(FEXCore::IR::IRListView *CurrentIR, IR::OrderedNodeWrapper Node) {
+  static uint8_t GetOpSize(FEXCore::IR::IRListView const *CurrentIR, IR::OrderedNodeWrapper Node) {
     auto IROp = CurrentIR->GetOp<FEXCore::IR::IROp_Header>(Node);
     return IROp->Size;
   }

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

@@ -4,6 +4,7 @@ tags: backend|interpreter
 $end_info$
 */
 
+#include "Interface/Core/CPUID.h"
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"
@@ -24,93 +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 *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
+  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->Header.Args[0]);
-
-  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->Header.Args[1]);
+  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);
-  void *Src = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  memcpy(MemData, Src, IROp->Size);
+  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, OpSize);
 }
 
 DEF_OP(SpillRegister) {
@@ -133,7 +180,7 @@ DEF_OP(LoadFlag) {
 
 DEF_OP(StoreFlag) {
   auto Op = IROp->C<IR::IROp_StoreFlag>();
-  uint8_t Arg = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+  uint8_t Arg = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
 
   uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
   ContextPtr += offsetof(FEXCore::Core::CPUState, flags[0]);
@@ -143,8 +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);
 
@@ -157,7 +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);
@@ -179,16 +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);
 
@@ -220,41 +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->Header.Args[0]);
-
-  memcpy(GDP, GetSrc<void*>(Data->SSAData, Op->Header.Args[1]), 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->Header.Args[0]); \
-      memcpy(MemData, &GetSrc<y*>(Data->SSAData, Op->Header.Args[1])[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

@@ -4,6 +4,8 @@ tags: backend|interpreter
 $end_info$
 */
 
+#include "Interface/Context/Context.h"
+
 #include "Interface/Core/Interpreter/InterpreterClass.h"
 #include "Interface/Core/Interpreter/InterpreterOps.h"
 #include "Interface/Core/Interpreter/InterpreterDefines.h"
@@ -17,13 +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) {
@@ -44,14 +39,26 @@ DEF_OP(Fence) {
 
 DEF_OP(Break) {
   auto Op = IROp->C<IR::IROp_Break>();
-  switch (Op->Reason) {
-    case FEXCore::IR::Break_Halt: // HLT
-      StopThread(Data->State);
+
+  Data->State->CurrentFrame->SynchronousFaultData.FaultToTopAndGeneratedException = 1;
+  Data->State->CurrentFrame->SynchronousFaultData.Signal = Op->Reason.Signal;
+  Data->State->CurrentFrame->SynchronousFaultData.TrapNo = Op->Reason.TrapNumber;
+  Data->State->CurrentFrame->SynchronousFaultData.err_code = Op->Reason.ErrorRegister;
+  Data->State->CurrentFrame->SynchronousFaultData.si_code = Op->Reason.si_code;
+
+  switch (Op->Reason.Signal) {
+  case SIGILL:
+    FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGILL);
     break;
-    case FEXCore::IR::Break_InvalidInstruction:
-      FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGILL);
+  case SIGTRAP:
+    FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGTRAP);
+    break;
+  case SIGSEGV:
+    FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGSEGV);
+    break;
+  default:
+    FHU::Syscalls::tgkill(Data->State->ThreadManager.PID, Data->State->ThreadManager.TID, SIGTRAP);
     break;
-  default: LOGMAN_MSG_A_FMT("Unknown Break Reason: {}", Op->Reason); break;
   }
 }
 
@@ -86,7 +93,7 @@ DEF_OP(GetRoundingMode) {
 
 DEF_OP(SetRoundingMode) {
   auto Op = IROp->C<IR::IROp_SetRoundingMode>();
-  uint8_t GuestRounding = *GetSrc<uint8_t*>(Data->SSAData, Op->Header.Args[0]);
+  const auto GuestRounding = *GetSrc<uint8_t*>(Data->SSAData, Op->RoundMode);
 #ifdef _M_ARM_64
   uint64_t HostRounding{};
   __asm volatile(R"(
@@ -126,16 +133,16 @@ DEF_OP(SetRoundingMode) {
 
 DEF_OP(Print) {
   auto Op = IROp->C<IR::IROp_Print>();
-  uint8_t OpSize = IROp->Size;
+  const uint8_t OpSize = IROp->Size;
 
   if (OpSize <= 8) {
-    uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[0]);
+    const auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
     LogMan::Msg::IFmt(">>>> Value in Arg: 0x{:x}, {}", Src, Src);
   }
   else if (OpSize == 16) {
-    __uint128_t Src = *GetSrc<__uint128_t*>(Data->SSAData, Op->Header.Args[0]);
-    uint64_t Src0 = Src;
-    uint64_t Src1 = Src >> 64;
+    const auto Src = *GetSrc<__uint128_t*>(Data->SSAData, Op->Value);
+    const uint64_t Src0 = Src;
+    const uint64_t Src1 = Src >> 64;
     LogMan::Msg::IFmt(">>>> Value[0] in Arg: 0x{:x}, {}", Src0, Src0);
     LogMan::Msg::IFmt("     Value[1] in Arg: 0x{:x}, {}", Src1, Src1);
   }

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

@@ -14,15 +14,15 @@ namespace FEXCore::CPU {
 #define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
 DEF_OP(ExtractElementPair) {
   auto Op = IROp->C<IR::IROp_ExtractElementPair>();
-  uintptr_t Src = GetSrc<uintptr_t>(Data->SSAData, Op->Header.Args[0]);
+  const auto Src = GetSrc<uintptr_t>(Data->SSAData, Op->Pair);
   memcpy(GDP,
     reinterpret_cast<void*>(Src + Op->Header.Size * Op->Element), Op->Header.Size);
 }
 
 DEF_OP(CreateElementPair) {
   auto Op = IROp->C<IR::IROp_CreateElementPair>();
-  void *Src_Lower = GetSrc<void*>(Data->SSAData, Op->Header.Args[0]);
-  void *Src_Upper = GetSrc<void*>(Data->SSAData, Op->Header.Args[1]);
+  const void *Src_Lower = GetSrc<void*>(Data->SSAData, Op->Lower);
+  const void *Src_Upper = GetSrc<void*>(Data->SSAData, Op->Upper);
 
   uint8_t *Dst = GetDest<uint8_t*>(Data->SSAData, Node);
 
@@ -30,13 +30,6 @@ DEF_OP(CreateElementPair) {
   memcpy(Dst + IROp->ElementSize, Src_Upper, IROp->ElementSize);
 }
 
-DEF_OP(Mov) {
-  auto Op = IROp->C<IR::IROp_Mov>();
-  uint8_t OpSize = IROp->Size;
-
-  memcpy(GDP, GetSrc<void*>(Data->SSAData, Op->Header.Args[0]), OpSize);
-}
-
 #undef DEF_OP
 
 } // namespace FEXCore::CPU

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

@@ -0,0 +1,131 @@
+/*
+$info$
+tags: backend|arm64
+desc: relocation logic of the arm64 splatter backend
+$end_info$
+*/
+#include "Interface/Context/Context.h"
+#include "Interface/Core/JIT/Arm64/JITClass.h"
+#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:
+      return ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker;
+    break;
+    default:
+      ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op));
+    break;
+  }
+  return ~0ULL;
+}
+
+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.Idx();
+
+  uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Sum));
+
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Pointer, EmitterCTX->Config.CacheObjectCodeCompilation());
+  Relocations.emplace_back(MoveABI);
+}
+
+Arm64JITCore::NamedSymbolLiteralPair Arm64JITCore::InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
+  uint64_t Pointer = GetNamedSymbolLiteral(Op);
+
+  Arm64JITCore::NamedSymbolLiteralPair Lit {
+    .Lit = Pointer,
+    .MoveABI = {
+      .NamedSymbolLiteral = {
+        .Header = {
+          .Type = FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL,
+        },
+        .Symbol = Op,
+        .Offset = 0,
+      },
+    },
+  };
+  return Lit;
+}
+
+void Arm64JITCore::PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit) {
+  // Offset is the offset from the entrypoint of the block
+  auto CurrentCursor = GetCursorAddress<uint8_t *>();
+  Lit.MoveABI.NamedSymbolLiteral.Offset = CurrentCursor - GuestEntry;
+
+  Bind(&Lit.Loc);
+  dc64(Lit.Lit);
+  Relocations.emplace_back(Lit.MoveABI);
+}
+
+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.Idx();
+
+  LoadConstant(ARMEmitter::Size::i64Bit, Reg, Constant, EmitterCTX->Config.CacheObjectCodeCompilation());
+  Relocations.emplace_back(MoveABI);
+}
+
+bool Arm64JITCore::ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations) {
+  size_t DataIndex{};
+  for (size_t j = 0; j < NumRelocations; ++j) {
+    const FEXCore::CPU::Relocation *Reloc = reinterpret_cast<const FEXCore::CPU::Relocation *>(&EntryRelocations[DataIndex]);
+    LOGMAN_THROW_AA_FMT((DataIndex % alignof(Relocation)) == 0, "Alignment of relocation wasn't adhered to");
+
+    switch (Reloc->Header.Type) {
+      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
+        SetCursorOffset(CursorEntry + Reloc->NamedSymbolLiteral.Offset);
+
+        // Generate a literal so we can place it
+        dc64(Pointer);
+
+        DataIndex += sizeof(Reloc->NamedSymbolLiteral);
+        break;
+      }
+      case FEXCore::CPU::RelocationTypes::RELOC_NAMED_THUNK_MOVE: {
+        uint64_t Pointer = reinterpret_cast<uint64_t>(EmitterCTX->ThunkHandler->LookupThunk(Reloc->NamedThunkMove.Symbol));
+        if (Pointer == ~0ULL) {
+          return false;
+        }
+
+        // Relocation occurs at the cursorEntry + offset relative to that cursor.
+        SetCursorOffset(CursorEntry + Reloc->NamedThunkMove.Offset);
+        LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc->NamedThunkMove.RegisterIndex), Pointer, true);
+        DataIndex += sizeof(Reloc->NamedThunkMove);
+        break;
+      }
+      case FEXCore::CPU::RelocationTypes::RELOC_GUEST_RIP_MOVE: {
+        // XXX: Reenable once the JIT Object Cache is upstream
+        // XXX: Should spin the relocation list, create a list of guest RIP moves, and ask for them all once, reduces lock contention.
+        uint64_t Pointer = ~0ULL; // EmitterCTX->JITObjectCache->FindRelocatedRIP(Reloc->GuestRIPMove.GuestRIP);
+        if (Pointer == ~0ULL) {
+          return false;
+        }
+
+        // Relocation occurs at the cursorEntry + offset relative to that cursor.
+        SetCursorOffset(CursorEntry + Reloc->GuestRIPMove.Offset);
+        LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Register(Reloc->GuestRIPMove.RegisterIndex), Pointer, true);
+        DataIndex += sizeof(Reloc->GuestRIPMove);
+        break;
+      }
+    }
+  }
+
+  return true;
+}
+}
+

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

@@ -4,7 +4,9 @@ tags: backend|arm64
 $end_info$
 */
 
+#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"
@@ -16,16 +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)
-DEF_OP(GuestCallDirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestCallIndirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
+#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
@@ -33,12 +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.AArch64.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();
 
@@ -46,15 +38,15 @@ DEF_OP(CallbackReturn) {
   ResetStack();
 
   // We can now lower the ref counter again
-  ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SignalHandlerRefCountPointer)));
-  ldr(w2, MemOperand(x0));
-  sub(w2, w2, 1);
-  str(w2, MemOperand(x0));
+
+  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();
 
@@ -65,110 +57,111 @@ 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{Dispatcher->ExitFunctionLinkerAddress};
-    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->Header.Args[0].ID());
+
+    ARMEmitter::ForwardLabel FullLookup;
+    auto RipReg = GetReg(Op->NewRIP.ID());
 
     // L1 Cache
-    ldr(x0, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.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.AArch64.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);
   }
 }
 
 DEF_OP(Jump) {
   const auto Op = IROp->C<IR::IROp_Jump>();
-  const auto ArgID = Op->Args(0).ID();
+  const auto Target = Op->TargetBlock.ID();
 
-  PendingTargetLabel = &JumpTargets.try_emplace(ArgID).first->second;
+  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;
@@ -182,7 +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();
@@ -193,19 +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.AArch64.SyscallHandlerObj)));
-  ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.SyscallHandlerFunc)));
-  mov(x1, STATE);
-  mov(x2, sp);
-  blr(x3);
+  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());
 
-  add(sp, sp, SPOffset);
+  // SP supporting move
+  add(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, ARMEmitter::Reg::rsp, 0);
+#ifdef VIXL_SIMULATOR
+  GenerateIndirectRuntimeCall<uint64_t, void*, void*, void*>(ARMEmitter::Reg::r3);
+#else
+  blr(ARMEmitter::Reg::r3);
+#endif
+
+  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) {
@@ -221,7 +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);
   }
 }
 
@@ -238,30 +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
@@ -273,66 +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);
       }
     }
   }
@@ -340,16 +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
 
@@ -360,16 +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);
   }
 }
 
@@ -381,13 +330,17 @@ DEF_OP(Thunk) {
 
   SpillStaticRegs(); // spill to ctx before ra64 spill
 
-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
 
-  mov(x0, GetReg<RA_64>(Op->Header.Args[0].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);
-  blr(x2);
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, (uintptr_t)thunkFn);
+#ifdef VIXL_SIMULATOR
+  GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
+#else
+  blr(ARMEmitter::Reg::r2);
+#endif
 
   PopDynamicRegsAndLR();
 
@@ -400,61 +353,67 @@ 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;
   }
 }
 
-DEF_OP(RemoveThreadCodeEntry) {
+DEF_OP(ThreadRemoveCodeEntry) {
   // Arguments are passed as follows:
   // 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.AArch64.RemoveThreadCodeEntryFromJIT)));
+  ldr(ARMEmitter::XReg::x2, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadRemoveCodeEntryFromJIT));
   SpillStaticRegs();
-  blr(x2);
+#ifdef VIXL_SIMULATOR
+  GenerateIndirectRuntimeCall<void, void*, void*>(ARMEmitter::Reg::r2);
+#else
+  blr(ARMEmitter::Reg::r2);
+#endif
   FillStaticRegs();
 
   // Fix the stack and any values that were stepped on
@@ -464,33 +423,36 @@ DEF_OP(RemoveThreadCodeEntry) {
 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.AArch64.CPUIDObj)));
-  ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.CPUIDFunction)));
-  mov(x1, GetReg<RA_64>(Op->Header.Args[0].ID()));
-  mov(x2, GetReg<RA_64>(Op->Header.Args[1].ID()));
-  SpillStaticRegs();
-  blr(x3);
+  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>(ARMEmitter::Reg::r3);
+#else
+  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
 void Arm64JITCore::RegisterBranchHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(GUESTCALLDIRECT,   GuestCallDirect);
-  REGISTER_OP(GUESTCALLINDIRECT, GuestCallIndirect);
   REGISTER_OP(SIGNALRETURN,      SignalReturn);
   REGISTER_OP(CALLBACKRETURN,    CallbackReturn);
   REGISTER_OP(EXITFUNCTION,      ExitFunction);
@@ -500,7 +462,7 @@ void Arm64JITCore::RegisterBranchHandlers() {
   REGISTER_OP(INLINESYSCALL,     InlineSyscall);
   REGISTER_OP(THUNK,             Thunk);
   REGISTER_OP(VALIDATECODE,      ValidateCode);
-  REGISTER_OP(REMOVETHREADCODEENTRY,   RemoveThreadCodeEntry);
+  REGISTER_OP(THREADREMOVECODEENTRY,   ThreadRemoveCodeEntry);
   REGISTER_OP(CPUID,             CPUID);
 #undef REGISTER_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->Header.Args[0].ID()));
-  switch (Op->Header.ElementSize) {
-    case 1: {
-      ins(GetDst(Node).V16B(), Op->DestIdx, GetReg<RA_32>(Op->Header.Args[1].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->Header.Args[1].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->Header.Args[1].ID()));
-    break;
-    }
-    case 8: {
-      ins(GetDst(Node).V2D(), Op->DestIdx, GetReg<RA_64>(Op->Header.Args[1].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->Header.Args[0].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->Header.Args[0].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->Header.Args[0].ID()).W());
+      fmov(ARMEmitter::Size::i32Bit, Dst.S(), Src);
       break;
     case 8:
-      fmov(GetDst(Node).D(), GetReg<RA_64>(Op->Header.Args[0].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>();
-  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->Header.Args[0].ID()));
+      scvtf(ARMEmitter::Size::i32Bit, Dst.S(), Src);
       break;
     }
     case 0x0408: { // Float <- int64_t
-      scvtf(GetDst(Node).S(), GetReg<RA_64>(Op->Header.Args[0].ID()));
+      scvtf(ARMEmitter::Size::i64Bit, Dst.S(), Src);
       break;
     }
     case 0x0804: { // Double <- int32_t
-      scvtf(GetDst(Node).D(), GetReg<RA_32>(Op->Header.Args[0].ID()));
+      scvtf(ARMEmitter::Size::i32Bit, Dst.D(), Src);
       break;
     }
     case 0x0808: { // Double <- int64_t
-      scvtf(GetDst(Node).D(), GetReg<RA_64>(Op->Header.Args[0].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>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  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->Header.Args[0].ID()).S());
+      fcvt(Dst.D(), Src.S());
       break;
     }
     case 0x0408: { // Float <- Double
-      fcvt(GetDst(Node).S(), GetSrc(Op->Header.Args[0].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->Header.Args[0].ID()).V4S());
-    break;
-    case 8:
-      scvtf(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].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->Header.Args[0].ID()).V4S());
-    break;
-    case 8:
-      fcvtzs(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].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->Header.Args[0].ID()).V4S());
-      fcvtzs(GetDst(Node).V4S(), GetDst(Node).V4S());
-    break;
-    case 8:
-      frinti(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].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->Header.Args[0].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->Header.Args[0].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->Header.Args[0].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->Header.Args[0].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->Header.Args[0].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->Header.Args[0].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->Header.Args[0].ID()).V4S());
+      case FEXCore::IR::Round_Host.Val:
+        frinti(SubEmitSize, Dst.Z(), Mask, Vector.Z());
         break;
-        case 8:
-          frintp(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].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->Header.Args[0].ID()).V4S());
+      case FEXCore::IR::Round_Negative_Infinity.Val:
+        frintm(SubEmitSize, Dst.Q(), Vector.Q());
         break;
-        case 8:
-          frintz(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].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->Header.Args[0].ID()).V4S());
+      case FEXCore::IR::Round_Towards_Zero.Val:
+        frintz(SubEmitSize, Dst.Q(), Vector.Q());
         break;
-        case 8:
-          frinti(GetDst(Node).V2D(), GetSrc(Op->Header.Args[0].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

@@ -4,114 +4,148 @@ 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->Header.Args[0].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->Header.Args[0].ID()).V16B());
-  aese(VTMP1.V16B(), VTMP2.V16B());
-  aesmc(VTMP1.V16B(), VTMP1.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].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->Header.Args[0].ID()).V16B());
-  aese(VTMP1.V16B(), VTMP2.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].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->Header.Args[0].ID()).V16B());
-  aesd(VTMP1.V16B(), VTMP2.V16B());
-  aesimc(VTMP1.V16B(), VTMP1.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].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->Header.Args[0].ID()).V16B());
-  aesd(VTMP1.V16B(), VTMP2.V16B());
-  eor(GetDst(Node).V16B(), VTMP1.V16B(), GetSrc(Op->Header.Args[1].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->Header.Args[0].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);
   }
 }
 
+DEF_OP(PCLMUL) {
+  auto Op = IROp->C<IR::IROp_PCLMUL>();
+
+  auto Dst  = GetVReg(Node);
+  auto Src1 = GetVReg(Op->Src1.ID());
+  auto Src2 = GetVReg(Op->Src2.ID());
+
+  switch (Op->Selector) {
+  case 0b00000000:
+    pmull(ARMEmitter::SubRegSize::i128Bit, Dst.D(), Src1.D(), Src2.D());
+    break;
+  case 0b00000001:
+    dup(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), Src1.Q(), 1);
+    pmull(ARMEmitter::SubRegSize::i128Bit, Dst.D(), VTMP1.D(), Src2.D());
+    break;
+  case 0b00010000:
+    dup(ARMEmitter::SubRegSize::i64Bit, VTMP1.Q(), Src2.Q(), 1);
+    pmull(ARMEmitter::SubRegSize::i128Bit, Dst.D(), VTMP1.D(), Src1.D());
+    break;
+  case 0b00010001:
+    pmull2(ARMEmitter::SubRegSize::i128Bit, Dst.Q(), Src1.Q(), Src2.Q());
+    break;
+  default:
+    LOGMAN_MSG_A_FMT("Unknown PCLMUL selector: {}", Op->Selector);
+    break;
+  }
+}
+
 #undef DEF_OP
 void Arm64JITCore::RegisterEncryptionHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
-  REGISTER_OP(VAESIMC,     AESImc);
-  REGISTER_OP(VAESENC,     AESEnc);
-  REGISTER_OP(VAESENCLAST, AESEncLast);
-  REGISTER_OP(VAESDEC,     AESDec);
-  REGISTER_OP(VAESDECLAST, AESDecLast);
-  REGISTER_OP(VAESKEYGENASSIST, AESKeyGenAssist);
+  REGISTER_OP(VAESIMC,           AESImc);
+  REGISTER_OP(VAESENC,           AESEnc);
+  REGISTER_OP(VAESENCLAST,       AESEncLast);
+  REGISTER_OP(VAESDEC,           AESDec);
+  REGISTER_OP(VAESDECLAST,       AESDecLast);
+  REGISTER_OP(VAESKEYGENASSIST,  AESKeyGenAssist);
   REGISTER_OP(CRC32,             CRC32);
+  REGISTER_OP(PCLMUL,            PCLMUL);
 #undef REGISTER_OP
 }
 }

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->Header.Args[0].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/JITClass.h

@@ -6,42 +6,31 @@ $end_info$
 
 #pragma once
 
+#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"
-#include "aarch64/disasm-aarch64.h"
-#include "aarch64/assembler-aarch64.h"
+#include <aarch64/assembler-aarch64.h>
+#include <aarch64/disasm-aarch64.h>
 
 #include <FEXCore/Core/CPUBackend.h>
 #include <FEXCore/IR/IR.h>
 #include <FEXCore/IR/IntrusiveIRList.h>
 
-#define STATE x28
-#define TMP1 x0
-#define TMP2 x1
-#define TMP3 x2
-#define TMP4 x3
-
-#define VTMP1 v1
-#define VTMP2 v2
-#define VTMP3 v3
+#include <array>
+#include <cstdint>
+#include <map>
+#include <utility>
+#include <vector>
 
 namespace FEXCore::Core {
   struct InternalThreadState;
 }
 
 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-
 class Arm64JITCore final : public CPUBackend, public Arm64Emitter  {
 public:
-  struct CodeBuffer {
-    uint8_t *Ptr;
-    size_t Size;
-  };
-
   explicit Arm64JITCore(FEXCore::Context::Context *ctx,
                         FEXCore::Core::InternalThreadState *Thread);
   ~Arm64JITCore() override;
@@ -51,7 +40,7 @@ public:
   [[nodiscard]] void *CompileCode(uint64_t Entry,
                                   FEXCore::IR::IRListView const *IR,
                                   FEXCore::Core::DebugData *DebugData,
-                                  FEXCore::IR::RegisterAllocationData *RAData) override;
+                                  FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) override;
 
   [[nodiscard]] void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }
 
@@ -59,28 +48,20 @@ public:
 
   void ClearCache() override;
 
-  static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
-  [[nodiscard]] CodeBuffer AllocateNewCodeBuffer(size_t Size);
-
-  bool IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher = true) const override {
-    return Dispatcher->IsAddressInJITCode(Address, IncludeDispatcher);
-  }
-
   static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
 
   void ClearRelocations() override { Relocations.clear(); }
 
 private:
   FEX_CONFIG_OPT(ParanoidTSO, PARANOIDTSO);
+  const bool HostSupportsSVE{};
 
-  std::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
-  Label *PendingTargetLabel;
+  ARMEmitter::BiDirectionalLabel *PendingTargetLabel;
   FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *ThreadState;
   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
@@ -103,38 +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;
 
@@ -147,40 +162,80 @@ private:
   vixl::aarch64::Decoder Decoder;
 #endif
 
-  void EmplaceNewCodeBuffer(CodeBuffer Buffer) {
-    CurrentCodeBuffer = &CodeBuffers.emplace_back(Buffer);
-  }
-
-  void FreeCodeBuffer(CodeBuffer Buffer);
-
-  // This is the initial code buffer that we will fall back to
-  // In a program without signals and code clearing, we will typically
-  // only have this code buffer
-  CodeBuffer InitialCodeBuffer{};
-  // This is the array of /additional/ code buffers that we may need to allocate
-  // Allocation only occurs when we've hit signals and need to clear code cache
-  // For code safety we can't delete code buffers until outside of all signals
-  std::vector<CodeBuffer> CodeBuffers{};
-
-  // This is the current code buffer that we are tracking
-  CodeBuffer *CurrentCodeBuffer{};
-
-  // We don't want to mvoe above 128MB atm because that means we will have to encode longer jumps
-  static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 128;
-  static constexpr size_t MAX_DISPATCHER_CODE_SIZE = 4096 * 2;
-
 #if DEBUG
   vixl::aarch64::Disassembler Disasm;
 #endif
 
-  static uint64_t ExitFunctionLink(Arm64JITCore *core, FEXCore::Core::CpuStateFrame *Frame, uint64_t *record);
-
   // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
   void EmitDetectionString();
   IR::RegisterAllocationPass *RAPass;
   IR::RegisterAllocationData *RAData;
+  FEXCore::Core::DebugData *DebugData;
 
-  using OpHandler = void (Arm64JITCore::*)(IR::IROp_Header *IROp, IR::NodeID Node);
+  void ResetStack();
+  /**
+   * @name Relocations
+   * @{ */
+
+    uint64_t GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+
+    /**
+     * @brief A literal pair relocation object for named symbol literals
+     */
+    struct NamedSymbolLiteralPair {
+      ARMEmitter::ForwardLabel Loc;
+      uint64_t Lit;
+      Relocation MoveABI{};
+    };
+
+    /**
+     * @brief Inserts a thunk relocation
+     *
+     * @param Reg - The GPR to move the thunk handler in to
+     * @param Sum - The hash of the thunk
+     */
+    void InsertNamedThunkRelocation(ARMEmitter::Register Reg, const IR::SHA256Sum &Sum);
+
+    /**
+     * @brief Inserts a guest GPR move relocation
+     *
+     * @param Reg - The GPR to move the guest RIP in to
+     * @param Constant - The guest RIP that will be relocated
+     */
+    void InsertGuestRIPMove(ARMEmitter::Register Reg, uint64_t Constant);
+
+    /**
+     * @brief Inserts a named symbol as a literal in memory
+     *
+     * Need to use `PlaceNamedSymbolLiteral` with the return value to place the literal in the desired location
+     *
+     * @param Op The named symbol to place
+     *
+     * @return A temporary `NamedSymbolLiteralPair`
+     */
+    NamedSymbolLiteralPair InsertNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op);
+
+    /**
+     * @brief Place the named symbol literal relocation in memory
+     *
+     * @param Lit - Which literal to place
+     */
+    void PlaceNamedSymbolLiteral(NamedSymbolLiteralPair &Lit);
+
+    std::vector<FEXCore::CPU::Relocation> Relocations;
+
+    ///< Relocation code loading
+    bool ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint64_t CursorEntry, size_t NumRelocations, const char* EntryRelocations);
+
+  /**  @} */
+
+  uint32_t SpillSlots{};
+  /**
+  * @brief Current guest RIP entrypoint
+  */
+  uint8_t *GuestEntry{};
+
+  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();
@@ -192,7 +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);
@@ -269,8 +324,6 @@ private:
   DEF_OP(AtomicFetchNeg);
 
   ///< Branch ops
-  DEF_OP(GuestCallDirect);
-  DEF_OP(GuestCallIndirect);
   DEF_OP(SignalReturn);
   DEF_OP(CallbackReturn);
   DEF_OP(ExitFunction);
@@ -280,7 +333,7 @@ private:
   DEF_OP(InlineSyscall);
   DEF_OP(Thunk);
   DEF_OP(ValidateCode);
-  DEF_OP(RemoveThreadCodeEntry);
+  DEF_OP(ThreadRemoveCodeEntry);
   DEF_OP(CPUID);
 
   ///< Conversion ops
@@ -314,13 +367,11 @@ private:
   DEF_OP(StoreMemTSO);
   DEF_OP(ParanoidLoadMemTSO);
   DEF_OP(ParanoidStoreMemTSO);
-  DEF_OP(VLoadMemElement);
-  DEF_OP(VStoreMemElement);
   DEF_OP(CacheLineClear);
   DEF_OP(CacheLineZero);
 
   ///< Misc ops
-  DEF_OP(EndBlock);
+  DEF_OP(GuestOpcode);
   DEF_OP(Fence);
   DEF_OP(Break);
   DEF_OP(Phi);
@@ -335,13 +386,10 @@ private:
   ///< Move ops
   DEF_OP(ExtractElementPair);
   DEF_OP(CreateElementPair);
-  DEF_OP(Mov);
 
   ///< Vector ops
   DEF_OP(VectorZero);
   DEF_OP(VectorImm);
-  DEF_OP(SplatVector2);
-  DEF_OP(SplatVector4);
   DEF_OP(VMov);
   DEF_OP(VAnd);
   DEF_OP(VBic);
@@ -400,18 +448,13 @@ private:
   DEF_OP(VUShrS);
   DEF_OP(VSShrS);
   DEF_OP(VInsElement);
-  DEF_OP(VInsScalarElement);
-  DEF_OP(VExtractElement);
   DEF_OP(VDupElement);
   DEF_OP(VExtr);
-  DEF_OP(VSLI);
-  DEF_OP(VSRI);
   DEF_OP(VUShrI);
   DEF_OP(VSShrI);
   DEF_OP(VShlI);
   DEF_OP(VUShrNI);
   DEF_OP(VUShrNI2);
-  DEF_OP(VBitcast);
   DEF_OP(VSXTL);
   DEF_OP(VSXTL2);
   DEF_OP(VUXTL);
@@ -437,9 +480,8 @@ private:
   DEF_OP(AESDecLast);
   DEF_OP(AESKeyGenAssist);
   DEF_OP(CRC32);
+  DEF_OP(PCLMUL);
 #undef DEF_OP
 };
 
-
-}
-
+} // namespace FEXCore::CPU

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

@@ -4,24 +4,31 @@ tags: backend|arm64
 $end_info$
 */
 
+#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>();
+  // metadata
+  DebugData->GuestOpcodes.push_back({Op->GuestEntryOffset, GetCursorAddress<uint8_t*>() - GuestEntry});
+}
 
 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;
   }
@@ -29,123 +36,124 @@ DEF_OP(Fence) {
 
 DEF_OP(Break) {
   auto Op = IROp->C<IR::IROp_Break>();
-  switch (Op->Reason) {
-    case FEXCore::IR::Break_Unimplemented: // Hard fault
-    case FEXCore::IR::Break_Interrupt: // Guest ud2
-      hlt(4);
-      break;
-    case FEXCore::IR::Break_Overflow: // overflow
-      ResetStack();
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.OverflowExceptionHandler)));
-      br(TMP1);
-      break;
-    case FEXCore::IR::Break_Halt: { // HLT
-      // Time to quit
-      // Set our stack to the starting stack location
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)));
-      add(sp, TMP1, 0);
 
-      // Now we need to jump to the thread stop handler
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.ThreadStopHandlerSpillSRA)));
-      br(TMP1);
-      break;
-    }
-    case FEXCore::IR::Break_Interrupt3: { // INT3
-      ResetStack();
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.ThreadPauseHandlerSpillSRA)));
-      br(TMP1);
-      break;
-    }
-    case FEXCore::IR::Break_InvalidInstruction:
-    {
-      ResetStack();
+  // First we must reset the stack
+  ResetStack();
 
-      ldr(TMP1, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.UnimplementedInstructionHandler)));
-      br(TMP1);
+  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,
+  };
 
-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Break reason: {}", Op->Reason);
+  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, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGILL));
+    br(TMP1);
+    break;
+  case SIGTRAP:
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP));
+    br(TMP1);
+    break;
+  case SIGSEGV:
+    ldr(TMP1, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGSEGV));
+    br(TMP1);
+    break;
+  default:
+    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->Header.Args[0].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->Header.Args[0].ID())) {
-    mov(x0, GetReg<RA_64>(Op->Header.Args[0].ID()));
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.PrintValue)));
+  if (IsGPR(Op->Value.ID())) {
+    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->Header.Args[0].ID()).V1D());
-    // Bug in vixl that source vector needs to b V1D rather than V2D?
-    fmov(x1, GetSrc(Op->Header.Args[0].ID()).V1D(), 1);
-    ldr(x3, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.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));
   }
-  SpillStaticRegs();
-  blr(x3);
-  FillStaticRegs();
 
+  blr(ARMEmitter::Reg::r3);
+
+  FillStaticRegs();
   PopDynamicRegsAndLR();
 }
 
@@ -163,27 +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
@@ -191,14 +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) {
@@ -206,21 +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
@@ -231,6 +238,7 @@ void Arm64JITCore::RegisterMiscHandlers() {
   REGISTER_OP(CODEBLOCK,  NoOp);
   REGISTER_OP(BEGINBLOCK, NoOp);
   REGISTER_OP(ENDBLOCK,   NoOp);
+  REGISTER_OP(GUESTOPCODE, GuestOpcode);
   REGISTER_OP(FENCE,      Fence);
   REGISTER_OP(BREAK,      Break);
   REGISTER_OP(PHI,        NoOp);

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

@@ -7,78 +7,45 @@ $end_info$
 #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->Header.Args[0].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->Header.Args[0].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->Header.Args[0].ID());
-      RegSecond = GetReg<RA_32>(Op->Header.Args[1].ID());
-      RegTmp = w0;
-      break;
-    }
-    case 8: {
-      Dst = GetSrcPair<RA_64>(Node);
-      RegFirst = GetReg<RA_64>(Op->Header.Args[0].ID());
-      RegSecond = GetReg<RA_64>(Op->Header.Args[1].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);
   }
 }
 
-DEF_OP(Mov) {
-  auto Op = IROp->C<IR::IROp_Mov>();
-  mov(GetReg<RA_64>(Node), GetReg<RA_64>(Op->Header.Args[0].ID()));
-}
-
 #undef DEF_OP
 void Arm64JITCore::RegisterMoveHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &Arm64JITCore::Op_##x
   REGISTER_OP(EXTRACTELEMENTPAIR, ExtractElementPair);
   REGISTER_OP(CREATEELEMENTPAIR,  CreateElementPair);
-  REGISTER_OP(MOV,                Mov);
 #undef REGISTER_OP
 }
 }

External/FEXCore/Source/Interface/Core/JIT/JITCore.h

@@ -16,9 +16,11 @@ class CPUBackend;
 [[nodiscard]] std::unique_ptr<CPUBackend> CreateX86JITCore(FEXCore::Context::Context *ctx,
                                                            FEXCore::Core::InternalThreadState *Thread);
 void InitializeX86JITSignalHandlers(FEXCore::Context::Context *CTX);
+CPUBackendFeatures GetX86JITBackendFeatures();
 
 [[nodiscard]] std::unique_ptr<CPUBackend> CreateArm64JITCore(FEXCore::Context::Context *ctx,
                                                              FEXCore::Core::InternalThreadState *Thread);
 void InitializeArm64JITSignalHandlers(FEXCore::Context::Context *CTX);
+CPUBackendFeatures GetArm64JITBackendFeatures();
 
 } // namespace FEXCore::CPU

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

@@ -29,31 +29,24 @@ $end_info$
 
 namespace FEXCore::CPU {
 #define DEF_OP(x) void X86JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
-DEF_OP(GuestCallDirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestCallIndirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
 
 DEF_OP(SignalReturn) {
   // Adjust the stack first for a regular return
   if (SpillSlots) {
-    add(rsp, SpillSlots * 16); // + 8 to consume return address
+    add(rsp, SpillSlots * MaxSpillSlotSize); // + 8 to consume return address
   }
 
-  jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SignalReturnHandler)]);
+  jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SignalReturnHandler)]);
 }
 
 DEF_OP(CallbackReturn) {
   // Adjust the stack first for a regular return
   if (SpillSlots) {
-    add(rsp, SpillSlots * 16); // + 8 to consume return address
+    add(rsp, SpillSlots * MaxSpillSlotSize); // + 8 to consume return address
   }
 
   // Make sure to adjust the refcounter so we don't clear the cache now
-  sub(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SignalHandlerRefCountPointer)], 1);
+  sub(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, SignalHandlerRefCounter)], 1);
 
   // We need to adjust an additional 8 bytes to get back to the original "misaligned" RSP state
   add(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.gregs[X86State::REG_RSP])], 8);
@@ -78,7 +71,7 @@ DEF_OP(ExitFunction) {
 
 
   if (SpillSlots) {
-    add(rsp, SpillSlots * 16);
+    add(rsp, SpillSlots * MaxSpillSlotSize);
   }
 
   uint64_t NewRIP;
@@ -91,14 +84,15 @@ DEF_OP(ExitFunction) {
     jmp(qword[rax]);
 
     L(l_BranchHost);
-    dq(Dispatcher->ExitFunctionLinkerAddress);
+    //FEX_TODO(this is not per thread)
+    dq(ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker);
     L(l_BranchGuest);
     dq(NewRIP);
   } else {
     Xbyak::Reg RipReg = GetSrc<RA_64>(Op->NewRIP.ID());
 
     // L1 Cache
-    mov(rcx, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.L1Pointer)]);
+    mov(rcx, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.L1Pointer)]);
 
     mov(rax, RipReg);
 
@@ -113,7 +107,7 @@ DEF_OP(ExitFunction) {
 
     L(FullLookup);
     mov(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, State.rip)], RipReg);
-    jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.DispatcherLoopTop)]);
+    jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.DispatcherLoopTop)]);
   }
 
 #ifdef BLOCKSTATS
@@ -123,9 +117,9 @@ DEF_OP(ExitFunction) {
 
 DEF_OP(Jump) {
   const auto Op = IROp->C<IR::IROp_Jump>();
-  const auto ArgID = Op->Args(0).ID();
+  const auto Target = Op->TargetBlock.ID();
 
-  PendingTargetLabel = &JumpTargets.try_emplace(ArgID).first->second;
+  PendingTargetLabel = &JumpTargets.try_emplace(Target).first->second;
 }
 
 #define GRCMP(Node) (Op->CompareSize == 4 ? GetSrc<RA_32>(Node) : GetSrc<RA_64>(Node))
@@ -181,13 +175,13 @@ DEF_OP(Syscall) {
   }
 
   mov(rsi, STATE); // Move thread in to rsi
-  mov(rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SyscallHandlerObj)]);
+  mov(rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerObj)]);
   mov(rdx, rsp);
 
   if (NumPush & 1)
     sub(rsp, 8); // Align
   // {rdi, rsi, rdx}
-  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.SyscallHandlerFunc)]);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.SyscallHandlerFunc)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -215,7 +209,7 @@ DEF_OP(Thunk) {
   if (NumPush & 1)
     sub(rsp, 8); // Align
 
-  mov(rdi, GetSrc<RA_64>(Op->Header.Args[0].ID()));
+  mov(rdi, GetSrc<RA_64>(Op->ArgPtr.ID()));
 
   auto thunkFn = ThreadState->CTX->ThunkHandler->LookupThunk(Op->ThunkNameHash);
 
@@ -259,7 +253,7 @@ DEF_OP(ValidateCode) {
   }
 }
 
-DEF_OP(RemoveThreadCodeEntry) {
+DEF_OP(ThreadRemoveCodeEntry) {
   auto NumPush = RA64.size();
 
   for (auto &Reg : RA64)
@@ -272,7 +266,7 @@ DEF_OP(RemoveThreadCodeEntry) {
   mov(rax, Entry); // imm64 move
   mov(rsi, rax);
 
-  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.RemoveThreadCodeEntryFromJIT)]);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.ThreadRemoveCodeEntryFromJIT)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -294,9 +288,9 @@ DEF_OP(CPUID) {
   // Result: RAX, RDX. 4xi32
 
   // rsi can be in the source registers, so copy argument to edx first
-  mov (edx, GetSrc<RA_32>(Op->Header.Args[1].ID()));
-  mov (esi, GetSrc<RA_32>(Op->Header.Args[0].ID()));
-  mov (rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.CPUIDObj)]);
+  mov (edx, GetSrc<RA_32>(Op->Leaf.ID()));
+  mov (esi, GetSrc<RA_32>(Op->Function.ID()));
+  mov (rdi, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj)]);
 
   auto NumPush = RA64.size();
 
@@ -304,7 +298,7 @@ DEF_OP(CPUID) {
     sub(rsp, 8); // Align
 
   // {rdi, rsi, rdx}
-  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.CPUIDFunction)]);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDFunction)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -320,8 +314,6 @@ DEF_OP(CPUID) {
 #undef DEF_OP
 void X86JITCore::RegisterBranchHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
-  REGISTER_OP(GUESTCALLDIRECT,   GuestCallDirect);
-  REGISTER_OP(GUESTCALLINDIRECT, GuestCallIndirect);
   REGISTER_OP(SIGNALRETURN,      SignalReturn);
   REGISTER_OP(CALLBACKRETURN,    CallbackReturn);
   REGISTER_OP(EXITFUNCTION,      ExitFunction);
@@ -330,7 +322,7 @@ void X86JITCore::RegisterBranchHandlers() {
   REGISTER_OP(SYSCALL,           Syscall);
   REGISTER_OP(THUNK,             Thunk);
   REGISTER_OP(VALIDATECODE,      ValidateCode);
-  REGISTER_OP(REMOVETHREADCODEENTRY,   RemoveThreadCodeEntry);
+  REGISTER_OP(THREADREMOVECODEENTRY,   ThreadRemoveCodeEntry);
   REGISTER_OP(CPUID,             CPUID);
 #undef REGISTER_OP
 }

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);
   }
 }
 
@@ -45,56 +94,64 @@ DEF_OP(VCastFromGPR) {
   auto Op = IROp->C<IR::IROp_VCastFromGPR>();
   switch (Op->Header.ElementSize) {
     case 1:
-      movzx(rax, GetSrc<RA_8>(Op->Header.Args[0].ID()));
+      movzx(rax, GetSrc<RA_8>(Op->Src.ID()));
       vmovq(GetDst(Node), rax);
       break;
     case 2:
-      movzx(rax, GetSrc<RA_16>(Op->Header.Args[0].ID()));
+      movzx(rax, GetSrc<RA_16>(Op->Src.ID()));
       vmovq(GetDst(Node), rax);
       break;
     case 4:
-      vmovd(GetDst(Node), GetSrc<RA_32>(Op->Header.Args[0].ID()).cvt32());
+      vmovd(GetDst(Node), GetSrc<RA_32>(Op->Src.ID()).cvt32());
       break;
     case 8:
-      vmovq(GetDst(Node), GetSrc<RA_64>(Op->Header.Args[0].ID()).cvt64());
+      vmovq(GetDst(Node), GetSrc<RA_64>(Op->Src.ID()).cvt64());
       break;
     default: LOGMAN_MSG_A_FMT("Unknown VCastFromGPR element size: {}", Op->Header.ElementSize);
   }
 }
 
 DEF_OP(Float_FromGPR_S) {
-  auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
-  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
-      cvtsi2ss(GetDst(Node), GetSrc<RA_32>(Op->Header.Args[0].ID()));
+      cvtsi2ss(GetDst(Node), GetSrc<RA_32>(Op->Src.ID()));
       break;
     }
     case 0x0408: { // Float <- int64_t
-      cvtsi2ss(GetDst(Node), GetSrc<RA_64>(Op->Header.Args[0].ID()));
+      cvtsi2ss(GetDst(Node), GetSrc<RA_64>(Op->Src.ID()));
       break;
     }
     case 0x0804: { // Double <- int32_t
-      cvtsi2sd(GetDst(Node), GetSrc<RA_32>(Op->Header.Args[0].ID()));
+      cvtsi2sd(GetDst(Node), GetSrc<RA_32>(Op->Src.ID()));
       break;
     }
     case 0x0808: { // Double <- int64_t
-      cvtsi2sd(GetDst(Node), GetSrc<RA_64>(Op->Header.Args[0].ID()));
+      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;
   }
 }
 
 DEF_OP(Float_FToF) {
   auto Op = IROp->C<IR::IROp_Float_FToF>();
-  uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+
   switch (Conv) {
     case 0x0804: { // Double <- Float
-      cvtss2sd(GetDst(Node), GetSrc(Op->Header.Args[0].ID()));
+      cvtss2sd(GetDst(Node), GetSrc(Op->Scalar.ID()));
       break;
     }
     case 0x0408: { // Float <- Double
-      cvtsd2ss(GetDst(Node), GetSrc(Op->Header.Args[0].ID()));
+      cvtsd2ss(GetDst(Node), GetSrc(Op->Scalar.ID()));
       break;
     }
     default: LOGMAN_MSG_A_FMT("Unknown Float_FToF sizes: 0x{:x}", Conv);
@@ -102,99 +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->Header.Args[0].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->Header.Args[0].ID()), 1);
-      pextrq(rcx, GetSrc(Op->Header.Args[0].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->Header.Args[0].ID()));
-    break;
+      if (Is256Bit) {
+        vcvttps2dq(ToYMM(Dst), ToYMM(Vector));
+      } else {
+        vcvttps2dq(Dst, Vector);
+      }
+      break;
     case 8:
-      cvttpd2dq(GetDst(Node), GetSrc(Op->Header.Args[0].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->Header.Args[0].ID()));
-    break;
+      if (Is256Bit) {
+        vcvtps2dq(ToYMM(Dst), ToYMM(Vector));
+      } else {
+        vcvtps2dq(Dst, Vector);
+      }
+      break;
     case 8:
-      cvtpd2dq(GetDst(Node), GetSrc(Op->Header.Args[0].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>();
-  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->Header.Args[0].ID()));
+      if (Is256Bit) {
+        vcvtps2pd(ToYMM(Dst), Vector);
+      } else {
+        vcvtps2pd(Dst, Vector);
+      }
       break;
     }
     case 0x0408: { // Float <- Double
-      cvtpd2ps(GetDst(Node), GetSrc(Op->Header.Args[0].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->Header.Args[0].ID()), RoundMode);
-    break;
+      if (Is256Bit) {
+        vroundps(ToYMM(Dst), ToYMM(Vector), RoundMode);
+      } else {
+        vroundps(Dst, Vector, RoundMode);
+      }
+      break;
     case 8:
-      roundpd(GetDst(Node), GetSrc(Op->Header.Args[0].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/EncryptionOps.cpp

@@ -17,44 +17,44 @@ namespace FEXCore::CPU {
 
 DEF_OP(AESImc) {
   auto Op = IROp->C<IR::IROp_VAESImc>();
-  vaesimc(GetDst(Node), GetSrc(Op->Header.Args[0].ID()));
+  vaesimc(GetDst(Node), GetSrc(Op->Vector.ID()));
 }
 
 DEF_OP(AESEnc) {
   auto Op = IROp->C<IR::IROp_VAESEnc>();
-  vaesenc(GetDst(Node), GetSrc(Op->Header.Args[0].ID()), GetSrc(Op->Header.Args[1].ID()));
+  vaesenc(GetDst(Node), GetSrc(Op->State.ID()), GetSrc(Op->Key.ID()));
 }
 
 DEF_OP(AESEncLast) {
   auto Op = IROp->C<IR::IROp_VAESEncLast>();
-  vaesenclast(GetDst(Node), GetSrc(Op->Header.Args[0].ID()), GetSrc(Op->Header.Args[1].ID()));
+  vaesenclast(GetDst(Node), GetSrc(Op->State.ID()), GetSrc(Op->Key.ID()));
 }
 
 DEF_OP(AESDec) {
   auto Op = IROp->C<IR::IROp_VAESDec>();
-  vaesdec(GetDst(Node), GetSrc(Op->Header.Args[0].ID()), GetSrc(Op->Header.Args[1].ID()));
+  vaesdec(GetDst(Node), GetSrc(Op->State.ID()), GetSrc(Op->Key.ID()));
 }
 
 DEF_OP(AESDecLast) {
   auto Op = IROp->C<IR::IROp_VAESDecLast>();
-  vaesdeclast(GetDst(Node), GetSrc(Op->Header.Args[0].ID()), GetSrc(Op->Header.Args[1].ID()));
+  vaesdeclast(GetDst(Node), GetSrc(Op->State.ID()), GetSrc(Op->Key.ID()));
 }
 
 DEF_OP(AESKeyGenAssist) {
   auto Op = IROp->C<IR::IROp_VAESKeyGenAssist>();
-  vaeskeygenassist(GetDst(Node), GetSrc(Op->Header.Args[0].ID()), Op->RCON);
+  vaeskeygenassist(GetDst(Node), GetSrc(Op->Src.ID()), Op->RCON);
 }
 
 DEF_OP(CRC32) {
   auto Op = IROp->C<IR::IROp_CRC32>();
   switch (IROp->Size) {
   case 4:
-    mov(TMP1, GetSrc<RA_32>(Op->Header.Args[1].ID()));
-    mov(GetDst<RA_32>(Node), GetSrc<RA_32>(Op->Header.Args[0].ID()));
+    mov(TMP1, GetSrc<RA_32>(Op->Src2.ID()));
+    mov(GetDst<RA_32>(Node), GetSrc<RA_32>(Op->Src1.ID()));
   break;
   case 8:
-    mov(TMP1, GetSrc<RA_64>(Op->Header.Args[1].ID()));
-    mov(GetDst<RA_64>(Node), GetSrc<RA_64>(Op->Header.Args[0].ID()));
+    mov(TMP1, GetSrc<RA_64>(Op->Src2.ID()));
+    mov(GetDst<RA_64>(Node), GetSrc<RA_64>(Op->Src1.ID()));
   break;
   default: LOGMAN_MSG_A_FMT("Unknown CRC32 size: {}", IROp->Size);
   }
@@ -75,16 +75,37 @@ DEF_OP(CRC32) {
   }
 }
 
+DEF_OP(PCLMUL) {
+  auto Op = IROp->C<IR::IROp_PCLMUL>();
+
+  auto Dst = GetDst(Node);
+  auto Src1 = GetSrc(Op->Src1.ID());
+  auto Src2 = GetSrc(Op->Src2.ID());
+
+  switch (Op->Selector) {
+  case 0b00000000:
+  case 0b00000001:
+  case 0b00010000:
+  case 0b00010001:
+    vpclmulqdq(Dst, Src1, Src2, Op->Selector);
+    break;
+  default:
+    LOGMAN_MSG_A_FMT("Unknown PCLMUL selector: {}", Op->Selector);
+    break;
+  }
+}
+
 #undef DEF_OP
 void X86JITCore::RegisterEncryptionHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
-  REGISTER_OP(VAESIMC,     AESImc);
-  REGISTER_OP(VAESENC,     AESEnc);
-  REGISTER_OP(VAESENCLAST, AESEncLast);
-  REGISTER_OP(VAESDEC,     AESDec);
-  REGISTER_OP(VAESDECLAST, AESDecLast);
-  REGISTER_OP(VAESKEYGENASSIST, AESKeyGenAssist);
+  REGISTER_OP(VAESIMC,           AESImc);
+  REGISTER_OP(VAESENC,           AESEnc);
+  REGISTER_OP(VAESENCLAST,       AESEncLast);
+  REGISTER_OP(VAESDEC,           AESDec);
+  REGISTER_OP(VAESDECLAST,       AESDecLast);
+  REGISTER_OP(VAESKEYGENASSIST,  AESKeyGenAssist);
   REGISTER_OP(CRC32,             CRC32);
+  REGISTER_OP(PCLMUL,            PCLMUL);
 #undef REGISTER_OP
 }
 }

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

@@ -18,7 +18,7 @@ namespace FEXCore::CPU {
 DEF_OP(GetHostFlag) {
   auto Op = IROp->C<IR::IROp_GetHostFlag>();
 
-  mov(rax, GetSrc<RA_64>(Op->Header.Args[0].ID()));
+  mov(rax, GetSrc<RA_64>(Op->Value.ID()));
   shr(rax, Op->Flag);
   and_(rax, 1);
   mov(GetDst<RA_64>(Node), rax);

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

@@ -25,7 +25,9 @@ $end_info$
 #include <FEXCore/IR/IntrusiveIRList.h>
 #include <FEXCore/IR/RegisterAllocationData.h>
 #include <FEXCore/Utils/Allocator.h>
+#include <FEXCore/Utils/EnumUtils.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/Profiler.h>
 
 #include <algorithm>
 #include <array>
@@ -43,6 +45,9 @@ $end_info$
 // #define DEBUG_RA 1
 // #define DEBUG_CYCLES
 
+static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
+static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 256;
+
 namespace {
 static void PrintValue(uint64_t Value) {
   LogMan::Msg::DFmt("Value: 0x{:x}", Value);
@@ -55,53 +60,43 @@ static void PrintVectorValue(uint64_t Value, uint64_t ValueUpper) {
 
 namespace FEXCore::CPU {
 
-CodeBuffer AllocateNewCodeBuffer(FEXCore::Context::Context *CTX, size_t Size) {
-  CodeBuffer Buffer;
-  Buffer.Size = Size;
-  Buffer.Ptr = static_cast<uint8_t*>(
-               FEXCore::Allocator::mmap(nullptr,
-                    Buffer.Size,
-                    PROT_READ | PROT_WRITE | PROT_EXEC,
-                    MAP_PRIVATE | MAP_ANONYMOUS,
-                    -1, 0));
-  LOGMAN_THROW_A_FMT(Buffer.Ptr != reinterpret_cast<uint8_t*>(~0ULL), "Couldn't allocate code buffer");
-  if (CTX->Config.GlobalJITNaming()) {
-    CTX->Symbols.RegisterJITSpace(Buffer.Ptr, Buffer.Size);
-  }
-  return Buffer;
-}
-
-void FreeCodeBuffer(CodeBuffer Buffer) {
-  FEXCore::Allocator::munmap(Buffer.Ptr, Buffer.Size);
-}
-
 void X86JITCore::PushRegs() {
-  sub(rsp, 16 * RAXMM_x.size());
+  const auto AVXRegSize = Core::CPUState::XMM_AVX_REG_SIZE;
+
+  sub(rsp, AVXRegSize * RAXMM_x.size());
   for (size_t i = 0; i < RAXMM_x.size(); ++i) {
-    movaps(ptr[rsp + i * 16], RAXMM_x[i]);
+    vmovups(ptr[rsp + i * AVXRegSize], ToYMM(RAXMM_x[i]));
   }
 
-  for (auto &Reg : RA64)
+  for (const auto &Reg : RA64) {
     push(Reg);
+  }
 
-  auto NumPush = RA64.size();
-  if (NumPush & 1)
-    sub(rsp, 8); // Align
+  const auto NumPush = RA64.size();
+  if ((NumPush & 1) != 0) {
+    // Align
+    sub(rsp, 8);
+  }
 }
 
 void X86JITCore::PopRegs() {
-  auto NumPush = RA64.size();
+  const auto AVXRegSize = Core::CPUState::XMM_AVX_REG_SIZE;
+  const auto NumPush = RA64.size();
 
-  if (NumPush & 1)
-    add(rsp, 8); // Align
-  for (uint32_t i = RA64.size(); i > 0; --i)
-    pop(RA64[i - 1]);
-
-  for (size_t i = 0; i < RAXMM_x.size(); ++i) {
-    movaps(RAXMM_x[i], ptr[rsp + i * 16]);
+  if ((NumPush & 1) != 0) {
+    // Align
+    add(rsp, 8);
   }
 
-  add(rsp, 16 * RAXMM_x.size());
+  for (uint32_t i = RA64.size(); i > 0; --i) {
+    pop(RA64[i - 1]);
+  }
+
+  for (size_t i = 0; i < RAXMM_x.size(); ++i) {
+    vmovups(ToYMM(RAXMM_x[i]), ptr[rsp + i * AVXRegSize]);
+  }
+
+  add(rsp, AVXRegSize * RAXMM_x.size());
 }
 
 void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
@@ -115,7 +110,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
       case FABI_VOID_U16: {
         PushRegs();
         mov(edi, GetSrc<RA_32>(IROp->Args[0].ID()));
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
         break;
@@ -124,7 +119,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushRegs();
 
         movss(xmm0, GetSrc(IROp->Args[0].ID()));
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -138,7 +133,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushRegs();
 
         movsd(xmm0, GetSrc(IROp->Args[0].ID()));
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -153,7 +148,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         PushRegs();
 
         mov(edi, GetSrc<RA_32>(IROp->Args[0].ID()));
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -169,7 +164,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -183,7 +178,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -196,7 +191,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
 
         movsd(xmm0, GetSrc(IROp->Args[0].ID()));
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -210,7 +205,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movsd(xmm0, GetSrc(IROp->Args[0].ID()));
         movsd(xmm1, GetSrc(IROp->Args[1].ID()));
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -224,7 +219,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -237,7 +232,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -250,7 +245,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -266,7 +261,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdx, GetSrc(IROp->Args[1].ID()));
         pextrq(rcx, GetSrc(IROp->Args[1].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -279,7 +274,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdi, GetSrc(IROp->Args[0].ID()));
         pextrq(rsi, GetSrc(IROp->Args[0].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -297,7 +292,7 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
         movq(rdx, GetSrc(IROp->Args[1].ID()));
         pextrq(rcx, GetSrc(IROp->Args[1].ID()), 1);
 
-        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.FallbackHandlerPointers[Info.HandlerIndex])]);
+        call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.FallbackHandlerPointers[Info.HandlerIndex])]);
 
         PopRegs();
 
@@ -310,24 +305,42 @@ void X86JITCore::Op_Unhandled(IR::IROp_Header *IROp, IR::NodeID Node) {
       case FABI_UNKNOWN:
       default:
 #if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
-        LOGMAN_MSG_A_FMT("Unhandled IR Fallback ABI: {} {}", FEXCore::IR::GetName(IROp->Op), Info.ABI);
+        LOGMAN_MSG_A_FMT("Unhandled IR Fallback ABI: {} {}",
+                         IR::GetName(IROp->Op), ToUnderlying(Info.ABI));
 #endif
         break;
     }
   }
 }
 
+static uint64_t X86JITCore_ExitFunctionLink(FEXCore::Core::CpuStateFrame *Frame, uint64_t *record) {
+  auto Thread = Frame->Thread;
+  auto GuestRip = record[1];
+
+  auto HostCode = Thread->LookupCache->FindBlock(GuestRip);
+
+  if (!HostCode) {
+    Thread->CurrentFrame->State.rip = GuestRip;
+    return Frame->Pointers.Common.DispatcherLoopTop;
+  }
+
+  auto LinkerAddress = Frame->Pointers.Common.ExitFunctionLinker;
+  Context::Context::ThreadAddBlockLink(Thread, GuestRip, (uintptr_t)record, [record, LinkerAddress]{
+    // undo the link
+    record[0] = LinkerAddress;
+  });
+
+  record[0] = HostCode;
+  return HostCode;
+}
+
 void X86JITCore::Op_NoOp(IR::IROp_Header *IROp, IR::NodeID Node) {
 }
 
-X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, CodeBuffer Buffer)
-  : CodeGenerator(Buffer.Size, Buffer.Ptr, nullptr)
-  , CTX {ctx}
-  , ThreadState {Thread}
-  , InitialCodeBuffer {Buffer}
-{
-  CurrentCodeBuffer = &InitialCodeBuffer;
-  EmitDetectionString();
+X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
+  : CPUBackend(Thread, INITIAL_CODE_SIZE, MAX_CODE_SIZE)
+  , CodeGenerator(0, this, nullptr) // this is not used here
+  , CTX {ctx} {
 
   RAPass = Thread->PassManager->GetPass<IR::RegisterAllocationPass>("RA");
 
@@ -356,69 +369,39 @@ X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalTh
   RegisterVectorHandlers();
   RegisterEncryptionHandlers();
 
-  DispatcherConfig config;
-  config.ExitFunctionLink = reinterpret_cast<uintptr_t>(&ExitFunctionLink);
-  config.ExitFunctionLinkThis = reinterpret_cast<uintptr_t>(this);
-  config.StaticRegisterAssignment = ctx->Config.StaticRegisterAllocation;
-
-  Dispatcher = std::make_unique<X86Dispatcher>(CTX, ThreadState, config);
-  DispatchPtr = Dispatcher->DispatchPtr;
-  CallbackPtr = Dispatcher->CallbackPtr;
-
   {
-    // Set up pointers that the JIT needs to load
-    auto &Pointers = ThreadState->CurrentFrame->Pointers.X86;
-    // Process specific
-    Pointers.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
-    Pointers.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
-    Pointers.RemoveThreadCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::RemoveThreadCodeEntryFromJit);
-    Pointers.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);
+    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);
+    Common.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);
 
     {
       FEXCore::Utils::MemberFunctionToPointerCast PMF(&FEXCore::CPUIDEmu::RunFunction);
-      Pointers.CPUIDFunction = PMF.GetConvertedPointer();
+      Common.CPUIDFunction = PMF.GetConvertedPointer();
     }
 
-    Pointers.SyscallHandlerObj = reinterpret_cast<uint64_t>(CTX->SyscallHandler);
-    Pointers.SyscallHandlerFunc = reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall);
+    Common.SyscallHandlerObj = reinterpret_cast<uint64_t>(CTX->SyscallHandler);
+    Common.SyscallHandlerFunc = reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall);
+    Common.ExitFunctionLink = reinterpret_cast<uintptr_t>(&Context::Context::ThreadExitFunctionLink<X86JITCore_ExitFunctionLink>);
 
     // Fill in the fallback handlers
-    InterpreterOps::FillFallbackIndexPointers(Pointers.FallbackHandlerPointers);
-
-    // Thread Specific
-    Pointers.SignalHandlerRefCountPointer = reinterpret_cast<uint64_t>(&Dispatcher->SignalHandlerRefCounter);
+    InterpreterOps::FillFallbackIndexPointers(Common.FallbackHandlerPointers);
   }
+
+  // Must be done after Dispatcher init
+  ClearCache();
 }
 
 void X86JITCore::InitializeSignalHandlers(FEXCore::Context::Context *CTX) {
   CTX->SignalDelegation->RegisterHostSignalHandler(SIGILL, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-    X86JITCore *Core = reinterpret_cast<X86JITCore*>(Thread->CPUBackend.get());
-    return Core->Dispatcher->HandleSIGILL(Signal, info, ucontext);
+    return Thread->CTX->Dispatcher->HandleSIGILL(Thread, Signal, info, ucontext);
   }, true);
-
-  CTX->SignalDelegation->RegisterHostSignalHandler(SignalDelegator::SIGNAL_FOR_PAUSE, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-    X86JITCore *Core = reinterpret_cast<X86JITCore*>(Thread->CPUBackend.get());
-    return Core->Dispatcher->HandleSignalPause(Signal, info, ucontext);
-  }, true);
-
-  auto GuestSignalHandler = [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext, GuestSigAction *GuestAction, stack_t *GuestStack) -> bool {
-    X86JITCore *Core = reinterpret_cast<X86JITCore*>(Thread->CPUBackend.get());
-    return Core->Dispatcher->HandleGuestSignal(Signal, info, ucontext, GuestAction, GuestStack);
-  };
-
-  for (uint32_t Signal = 0; Signal <= SignalDelegator::MAX_SIGNALS; ++Signal) {
-    CTX->SignalDelegation->RegisterHostSignalHandlerForGuest(Signal, GuestSignalHandler);
-  }
 }
 
 X86JITCore::~X86JITCore() {
-  for (auto CodeBuffer : CodeBuffers) {
-    FreeCodeBuffer(CodeBuffer);
-  }
-  CodeBuffers.clear();
 
-
-  FreeCodeBuffer(InitialCodeBuffer);
 }
 
 void X86JITCore::EmitDetectionString() {
@@ -429,51 +412,15 @@ void X86JITCore::EmitDetectionString() {
 }
 
 void X86JITCore::ClearCache() {
-  if (Dispatcher->SignalHandlerRefCounter == 0) {
-    if (!CodeBuffers.empty()) {
-      // If we have more than one code buffer we are tracking then walk them and delete
-      // This is a cleanup step
-      for (auto CodeBuffer : CodeBuffers) {
-        FreeCodeBuffer(CodeBuffer);
-      }
-      CodeBuffers.clear();
-
-      // Set the current code buffer to the initial
-      setNewBuffer(InitialCodeBuffer.Ptr, InitialCodeBuffer.Size);
-      CurrentCodeBuffer = &InitialCodeBuffer;
-    }
-
-    if (CurrentCodeBuffer->Size == MAX_CODE_SIZE) {
-      // Rewind to the start of the code cache start
-      reset();
-    }
-    else {
-      FreeCodeBuffer(InitialCodeBuffer);
-
-      // Resize the code buffer and reallocate our code size
-      CurrentCodeBuffer->Size *= 1.5;
-      CurrentCodeBuffer->Size = std::min(CurrentCodeBuffer->Size, MAX_CODE_SIZE);
-
-      InitialCodeBuffer = AllocateNewCodeBuffer(CTX, CurrentCodeBuffer->Size);
-      setNewBuffer(InitialCodeBuffer.Ptr, InitialCodeBuffer.Size);
-    }
-  }
-  else {
-    // We have signal handlers that have generated code
-    // This means that we can not safely clear the code at this point in time
-    // Allocate some new code buffers that we can switch over to instead
-    auto NewCodeBuffer = AllocateNewCodeBuffer(CTX, X86JITCore::INITIAL_CODE_SIZE);
-    EmplaceNewCodeBuffer(NewCodeBuffer);
-    setNewBuffer(NewCodeBuffer.Ptr, NewCodeBuffer.Size);
-  }
-
+  auto CodeBuffer = GetEmptyCodeBuffer();
+  setNewBuffer(CodeBuffer->Ptr, CodeBuffer->Size);
   EmitDetectionString();
 }
 
 IR::PhysicalRegister X86JITCore::GetPhys(IR::NodeID Node) const {
   auto PhyReg = RAData->GetNodeRegister(Node);
 
-  LOGMAN_THROW_A_FMT(PhyReg.Raw != 255, "Couldn't Allocate register for node: ssa{}. Class: {}", Node, PhyReg.Class);
+  LOGMAN_THROW_AA_FMT(PhyReg.Raw != 255, "Couldn't Allocate register for node: ssa{}. Class: {}", Node, PhyReg.Class);
 
   return PhyReg;
 }
@@ -635,48 +582,37 @@ std::tuple<X86JITCore::SetCC, X86JITCore::CMovCC, X86JITCore::JCC> X86JITCore::G
   return { &CodeGenerator::sete , &CodeGenerator::cmove , &CodeGenerator::je  };
 }
 
-void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRListView const *IR, [[maybe_unused]] FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData) {
+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();
 
   this->Entry = Entry;
   this->RAData = RAData;
+  this->DebugData = DebugData;
 
   // Fairly excessive buffer range to make sure we don't overflow
-  uint32_t BufferRange = SSACount * 16;
+  uint32_t BufferRange = SSACount * 16 + GDBEnabled * Dispatcher::MaxGDBPauseCheckSize;
   if ((getSize() + BufferRange) > CurrentCodeBuffer->Size) {
-    ThreadState->CTX->ClearCodeCache(ThreadState, false);
+    CTX->ClearCodeCache(ThreadState);
   }
 
-	void *GuestEntry = getCurr<void*>();
+	GuestEntry = getCurr<uint8_t*>();
   CursorEntry = getSize();
   this->IR = IR;
 
-  if (CTX->GetGdbServerStatus()) {
-    Label 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(CTX->Config.RunningMode) == 4, "This is expected to be size of 4");
-    mov(rax, reinterpret_cast<uint64_t>(CTX));
-
-    // If the value == 0 then branch to the top
-    cmp(dword [rax + (offsetof(FEXCore::Context::Context, Config.RunningMode))], 0);
-    je(RunBlock);
-    // Else we need to pause now
-    mov(rax, Dispatcher->ThreadPauseHandlerAddress);
-    jmp(rax);
-    ud2();
-
-    L(RunBlock);
+  if (GDBEnabled) {
+    auto GDBSize = CTX->Dispatcher->GenerateGDBPauseCheck(GuestEntry, Entry);
+    setSize(getSize() + GDBSize);
   }
 
-  LOGMAN_THROW_A_FMT(RAData != nullptr, "Needs RA");
+  LOGMAN_THROW_AA_FMT(RAData != nullptr, "Needs RA");
 
   SpillSlots = RAData->SpillSlots();
 
   if (SpillSlots) {
-    sub(rsp, SpillSlots * 16);
+    sub(rsp, SpillSlots * MaxSpillSlotSize);
   }
 
 #ifdef BLOCKSTATS
@@ -726,12 +662,13 @@ void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRLi
 
   for (auto [BlockNode, BlockHeader] : IR->GetBlocks()) {
     using namespace FEXCore::IR;
-    {
 #if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
-      auto BlockIROp = BlockHeader->CW<IROp_CodeBlock>();
-      LOGMAN_THROW_A_FMT(BlockIROp->Header.Op == IR::OP_CODEBLOCK, "IR type failed to be a code block");
+    auto BlockIROp = BlockHeader->CW<IROp_CodeBlock>();
+    LOGMAN_THROW_AA_FMT(BlockIROp->Header.Op == IR::OP_CODEBLOCK, "IR type failed to be a code block");
 #endif
 
+    auto BlockStartHostCode = getCurr<uint8_t *>();
+    {
       const auto Node = IR->GetID(BlockNode);
       const auto IsTarget = JumpTargets.try_emplace(Node).first;
 
@@ -786,6 +723,13 @@ void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRLi
       OpHandler Handler = OpHandlers[IROp->Op];
       (this->*Handler)(IROp, ID);
     }
+
+    if (DebugData) {
+      DebugData->Subblocks.push_back({
+        static_cast<uint32_t>(BlockStartHostCode - GuestEntry),
+        static_cast<uint32_t>(getCurr<uint8_t *>() - BlockStartHostCode)
+      });
+    }
   }
 
   // Make sure last branch is generated. It certainly can't be eliminated here.
@@ -808,29 +752,12 @@ void *X86JITCore::CompileCode(uint64_t Entry, [[maybe_unused]] FEXCore::IR::IRLi
   return GuestEntry;
 }
 
-uint64_t X86JITCore::ExitFunctionLink(X86JITCore *core, FEXCore::Core::CpuStateFrame *Frame, uint64_t *record) {
-  auto Thread = Frame->Thread;
-  auto GuestRip = record[1];
-
-  auto HostCode = Thread->LookupCache->FindBlock(GuestRip);
-
-  if (!HostCode) {
-    Thread->CurrentFrame->State.rip = GuestRip;
-    return core->Dispatcher->AbsoluteLoopTopAddress;
-  }
-
-  auto LinkerAddress = core->Dispatcher->ExitFunctionLinkerAddress;
-  Thread->LookupCache->AddBlockLink(GuestRip, (uintptr_t)record, [record, LinkerAddress]{
-    // undo the link
-    record[0] = LinkerAddress;
-  });
-
-  record[0] = HostCode;
-  return HostCode;
+std::unique_ptr<CPUBackend> CreateX86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
+  return std::make_unique<X86JITCore>(ctx, Thread);
 }
 
-std::unique_ptr<CPUBackend> CreateX86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread) {
-  return std::make_unique<X86JITCore>(ctx, Thread, AllocateNewCodeBuffer(ctx, X86JITCore::INITIAL_CODE_SIZE));
+CPUBackendFeatures GetX86JITBackendFeatures() {
+  return CPUBackendFeatures { };
 }
 
 void InitializeX86JITSignalHandlers(FEXCore::Context::Context *CTX) {

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

@@ -6,6 +6,7 @@ $end_info$
 
 #pragma once
 
+#include <FEXCore/IR/RegisterAllocationData.h>
 #include "Interface/Core/BlockSamplingData.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/ObjectCache/Relocations.h"
@@ -25,13 +26,6 @@ using namespace Xbyak;
 #include <tuple>
 
 namespace FEXCore::CPU {
-struct CodeBuffer {
-  uint8_t *Ptr;
-  size_t Size;
-};
-
-[[nodiscard]] CodeBuffer AllocateNewCodeBuffer(size_t Size);
-void FreeCodeBuffer(CodeBuffer Buffer);
 
 // Temp registers
 // rax, rcx, rdx, rsi, r8, r9,
@@ -58,8 +52,7 @@ const std::array<Xbyak::Xmm, 11> RAXMM_x = {  xmm1, xmm2, xmm3, xmm4, xmm5, xmm6
 class X86JITCore final : public CPUBackend, public Xbyak::CodeGenerator {
 public:
   explicit X86JITCore(FEXCore::Context::Context *ctx,
-                      FEXCore::Core::InternalThreadState *Thread,
-                      CodeBuffer Buffer);
+                      FEXCore::Core::InternalThreadState *Thread);
   ~X86JITCore() override;
 
   [[nodiscard]] std::string GetName() override { return "JIT"; }
@@ -67,7 +60,7 @@ public:
   [[nodiscard]] void *CompileCode(uint64_t Entry,
                                   FEXCore::IR::IRListView const *IR,
                                   FEXCore::Core::DebugData *DebugData,
-                                  FEXCore::IR::RegisterAllocationData *RAData) override;
+                                  FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) override;
 
   [[nodiscard]] void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }
 
@@ -75,13 +68,6 @@ public:
 
   void ClearCache() override;
 
-  static constexpr size_t INITIAL_CODE_SIZE = 1024 * 1024 * 16;
-  static constexpr size_t MAX_CODE_SIZE = 1024 * 1024 * 256;
-
-  bool IsAddressInJITCode(uint64_t Address, bool IncludeDispatcher = true) const override {
-    return Dispatcher->IsAddressInJITCode(Address, IncludeDispatcher);
-  }
-
   static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
 
   void ClearRelocations() override { Relocations.clear(); }
@@ -150,9 +136,7 @@ private:
 
   Label* PendingTargetLabel{};
   FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *ThreadState;
   FEXCore::IR::IRListView const *IR;
-  std::unique_ptr<FEXCore::CPU::Dispatcher> Dispatcher;
   uint64_t Entry;
 
   std::unordered_map<IR::NodeID, Label> JumpTargets;
@@ -197,6 +181,10 @@ private:
   [[nodiscard]] Xbyak::Xmm GetSrc(IR::NodeID Node) const;
   [[nodiscard]] Xbyak::Xmm GetDst(IR::NodeID Node) const;
 
+  [[nodiscard]] static Xbyak::Ymm ToYMM(const Xbyak::Xmm& xmm) {
+    return Xbyak::Ymm{xmm.getIdx()};
+  }
+
   [[nodiscard]] Xbyak::RegExp GenerateModRM(Xbyak::Reg Base, IR::OrderedNodeWrapper Offset,
                                             IR::MemOffsetType OffsetType, uint8_t OffsetScale) const;
 
@@ -205,33 +193,23 @@ private:
 
   IR::RegisterAllocationPass *RAPass;
   FEXCore::IR::RegisterAllocationData *RAData;
+  FEXCore::Core::DebugData *DebugData;
 
 #ifdef BLOCKSTATS
   bool GetSamplingData {true};
 #endif
 
-  void EmplaceNewCodeBuffer(CodeBuffer Buffer) {
-    CurrentCodeBuffer = &CodeBuffers.emplace_back(Buffer);
-  }
-
-  static uint64_t ExitFunctionLink(X86JITCore* code, FEXCore::Core::CpuStateFrame *Frame, uint64_t *record);
+  static uint64_t ExitFunctionLink(FEXCore::Core::CpuStateFrame *Frame, uint64_t *record);
 
   // This is purely a debugging aid for developers to see if they are in JIT code space when inspecting raw memory
   void EmitDetectionString();
 
-  // This is the initial code buffer that we will fall back to
-  // In a program without signals and code clearing, we will typically
-  // only have this code buffer
-  CodeBuffer InitialCodeBuffer{};
-  // This is the array of /additional/ code buffers that we may need to allocate
-  // Allocation only occurs when we've hit signals and need to clear code cache
-  // For code safety we can't delete code buffers until outside of all signals
-  std::vector<CodeBuffer> CodeBuffers{};
-
-  // This is the current code buffer that we are tracking
-  CodeBuffer *CurrentCodeBuffer{};
-
   uint32_t SpillSlots{};
+  /**
+  * @brief Current guest RIP entrypoint
+  */
+  uint8_t *GuestEntry{};
+
   using SetCC = void (X86JITCore::*)(const Operand& op);
   using CMovCC = void (X86JITCore::*)(const Reg& reg, const Operand& op);
   using JCC = void (X86JITCore::*)(const Label& label, LabelType type);
@@ -330,8 +308,6 @@ private:
   DEF_OP(AtomicFetchNeg);
 
   ///< Branch ops
-  DEF_OP(GuestCallDirect);
-  DEF_OP(GuestCallIndirect);
   DEF_OP(SignalReturn);
   DEF_OP(CallbackReturn);
   DEF_OP(ExitFunction);
@@ -340,7 +316,7 @@ private:
   DEF_OP(Syscall);
   DEF_OP(Thunk);
   DEF_OP(ValidateCode);
-  DEF_OP(RemoveThreadCodeEntry);
+  DEF_OP(ThreadRemoveCodeEntry);
   DEF_OP(CPUID);
 
   ///< Conversion ops
@@ -361,6 +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);
@@ -369,13 +347,11 @@ private:
   DEF_OP(StoreFlag);
   DEF_OP(LoadMem);
   DEF_OP(StoreMem);
-  DEF_OP(VLoadMemElement);
-  DEF_OP(VStoreMemElement);
   DEF_OP(CacheLineClear);
   DEF_OP(CacheLineZero);
 
   ///< Misc ops
-  DEF_OP(EndBlock);
+  DEF_OP(GuestOpcode);
   DEF_OP(Fence);
   DEF_OP(Break);
   DEF_OP(Phi);
@@ -390,12 +366,10 @@ private:
   ///< Move ops
   DEF_OP(ExtractElementPair);
   DEF_OP(CreateElementPair);
-  DEF_OP(Mov);
 
   ///< Vector ops
   DEF_OP(VectorZero);
   DEF_OP(VectorImm);
-  DEF_OP(SplatVector);
   DEF_OP(VMov);
   DEF_OP(VAnd);
   DEF_OP(VBic);
@@ -454,18 +428,13 @@ private:
   DEF_OP(VUShrS);
   DEF_OP(VSShrS);
   DEF_OP(VInsElement);
-  DEF_OP(VInsScalarElement);
-  DEF_OP(VExtractElement);
   DEF_OP(VDupElement);
   DEF_OP(VExtr);
-  DEF_OP(VSLI);
-  DEF_OP(VSRI);
   DEF_OP(VUShrI);
   DEF_OP(VSShrI);
   DEF_OP(VShlI);
   DEF_OP(VUShrNI);
   DEF_OP(VUShrNI2);
-  DEF_OP(VBitcast);
   DEF_OP(VSXTL);
   DEF_OP(VSXTL2);
   DEF_OP(VUXTL);
@@ -491,6 +460,7 @@ private:
   DEF_OP(AESDecLast);
   DEF_OP(AESKeyGenAssist);
   DEF_OP(CRC32);
+  DEF_OP(PCLMUL);
 #undef DEF_OP
 };
 

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

@@ -4,6 +4,7 @@ tags: backend|x86-64
 $end_info$
 */
 
+#include "Interface/Core/CPUID.h"
 #include "Interface/Core/JIT/x86_64/JITClass.h"
 
 #include <FEXCore/Core/CoreState.h>
@@ -20,130 +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) {
@@ -152,21 +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;
@@ -185,53 +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;
@@ -244,12 +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) {
@@ -257,10 +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:
@@ -269,57 +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;
@@ -332,10 +487,10 @@ DEF_OP(StoreContextIndexed) {
 }
 
 DEF_OP(SpillRegister) {
-  auto Op = IROp->C<IR::IROp_SpillRegister>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_SpillRegister>();
+  const uint8_t OpSize = IROp->Size;
+  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;
 
-  uint32_t SlotOffset = Op->Slot * 16;
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (OpSize) {
       case 1: {
@@ -354,36 +509,44 @@ DEF_OP(SpillRegister) {
         mov(qword [rsp + SlotOffset], GetSrc<RA_64>(Op->Value.ID()));
         break;
       }
-      default:  LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize);
+      default:
+        LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize);
+        break;
     }
   } else if (Op->Class == FEXCore::IR::FPRClass) {
+    const auto Src = GetSrc(Op->Value.ID());
+
     switch (OpSize) {
       case 4: {
-        movss(dword [rsp + SlotOffset], GetSrc(Op->Value.ID()));
+        movss(dword [rsp + SlotOffset], Src);
         break;
       }
       case 8: {
-        movsd(qword [rsp + SlotOffset], GetSrc(Op->Value.ID()));
+        movsd(qword [rsp + SlotOffset], Src);
         break;
       }
       case 16: {
-        movaps(xword [rsp + SlotOffset], GetSrc(Op->Value.ID()));
+        movaps(xword [rsp + SlotOffset], Src);
         break;
       }
-      default:  LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize);
+      case 32: {
+        vmovups(yword [rsp + SlotOffset], ToYMM(Src));
+        break;
+      }
+      default:
+        LOGMAN_MSG_A_FMT("Unhandled SpillRegister size: {}", OpSize);
+        break;
     }
   } else {
     LOGMAN_MSG_A_FMT("Unhandled SpillRegister class: {}", Op->Class.Val);
   }
-
-
 }
 
 DEF_OP(FillRegister) {
-  auto Op = IROp->C<IR::IROp_FillRegister>();
-  uint8_t OpSize = IROp->Size;
+  const auto Op = IROp->C<IR::IROp_FillRegister>();
+  const uint8_t OpSize = IROp->Size;
+  const uint32_t SlotOffset = Op->Slot * MaxSpillSlotSize;
 
-  uint32_t SlotOffset = Op->Slot * 16;
   if (Op->Class == FEXCore::IR::GPRClass) {
     switch (OpSize) {
       case 1: {
@@ -402,23 +565,33 @@ DEF_OP(FillRegister) {
         mov(GetDst<RA_64>(Node), qword [rsp + SlotOffset]);
         break;
       }
-      default:  LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize);
+      default:
+        LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize);
+        break;
     }
   } else if (Op->Class == FEXCore::IR::FPRClass) {
+    const auto Dst = GetDst(Node);
+
     switch (OpSize) {
       case 4: {
-        movss(GetDst(Node), dword [rsp + SlotOffset]);
+        vmovss(Dst, dword [rsp + SlotOffset]);
         break;
       }
       case 8: {
-        movsd(GetDst(Node), qword [rsp + SlotOffset]);
+        vmovsd(Dst, qword [rsp + SlotOffset]);
         break;
       }
       case 16: {
-        movaps(GetDst(Node), xword [rsp + SlotOffset]);
+        vmovaps(Dst, xword [rsp + SlotOffset]);
         break;
       }
-      default:  LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize);
+      case 32: {
+        vmovups(ToYMM(Dst), yword [rsp + SlotOffset]);
+        break;
+      }
+      default:
+        LOGMAN_MSG_A_FMT("Unhandled FillRegister size: {}", OpSize);
+        break;
     }
   } else {
     LOGMAN_MSG_A_FMT("Unhandled FillRegister class: {}", Op->Class.Val);
@@ -463,130 +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>();
 
@@ -617,8 +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);
@@ -629,8 +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

@@ -7,6 +7,7 @@ $end_info$
 #include "Interface/Context/Context.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/JIT/x86_64/JITClass.h"
+#include "FEXCore/Debug/InternalThreadState.h"
 
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Utils/LogManager.h>
@@ -20,6 +21,12 @@ $end_info$
 namespace FEXCore::CPU {
 #define DEF_OP(x) void X86JITCore::Op_##x(IR::IROp_Header *IROp, IR::NodeID Node)
 
+DEF_OP(GuestOpcode) {
+  auto Op = IROp->C<IR::IROp_GuestOpcode>();
+  // metadata
+  DebugData->GuestOpcodes.push_back({Op->GuestEntryOffset, getCurr<uint8_t*>() - GuestEntry});
+}
+
 DEF_OP(Fence) {
   auto Op = IROp->C<IR::IROp_Fence>();
   switch (Op->Fence) {
@@ -38,56 +45,38 @@ DEF_OP(Fence) {
 
 DEF_OP(Break) {
   auto Op = IROp->C<IR::IROp_Break>();
-  switch (Op->Reason) {
-    case FEXCore::IR::Break_Unimplemented: // Hard fault
-    case FEXCore::IR::Break_Interrupt: // Guest ud2
-      ud2();
-      break;
-    case FEXCore::IR::Break_Overflow: // overflow
-      // Need to be outside of JIT cache space to ensure cache clearing correctness
-      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.OverflowExceptionHandler)]);
-      break;
-    case FEXCore::IR::Break_Halt: { // HLT
-      // Time to quit
-      // Set our stack to the starting stack location
-      mov(rsp, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)]);
 
-      // Now we need to jump to the thread stop handler
-      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.ThreadStopHandler)]);
-      break;
-    }
-    case FEXCore::IR::Break_Interrupt3: // INT3
-    {
-      if (CTX->GetGdbServerStatus()) {
-        // Adjust the stack first for a regular return
-        if (SpillSlots) {
-          add(rsp, SpillSlots * 16);
-        }
+  if (SpillSlots) {
+    add(rsp, SpillSlots * MaxSpillSlotSize);
+  }
 
-        // This jump target needs to be a constant offset here
-        jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.ThreadPauseHandler)]);
-      }
-      else {
-        // If we don't have a gdb server attached then....crash?
-        // Treat this case like HLT
-        mov(rsp, qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, ReturningStackLocation)]);
+  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,
+  };
 
-        // Now we need to jump to the thread stop handler
-        jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.ThreadStopHandler)]);
-      }
+  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:
+    jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGILL)]);
+    break;
+  case SIGTRAP:
+    jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP)]);
+    break;
+  case SIGSEGV:
+    jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGSEGV)]);
+    break;
+  default:
+    jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.GuestSignal_SIGTRAP)]);
     break;
-    }
-    case FEXCore::IR::Break_InvalidInstruction:
-    {
-      if (SpillSlots) {
-        add(rsp, SpillSlots * 16);
-      }
-
-      // Need to be outside of JIT cache space to ensure cache clearing correctness
-      jmp(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.UnimplementedInstructionHandler)]);
-      break;
-    }
-    default: LOGMAN_MSG_A_FMT("Unknown Break reason: {}", Op->Reason);
   }
 }
 
@@ -103,7 +92,7 @@ DEF_OP(GetRoundingMode) {
 
 DEF_OP(SetRoundingMode) {
   auto Op = IROp->C<IR::IROp_SetRoundingMode>();
-  auto Src = GetSrc<RA_32>(Op->Header.Args[0].ID());
+  auto Src = GetSrc<RA_32>(Op->RoundMode.ID());
 
   // Load old mxcsr
   // Only stores to memory
@@ -128,15 +117,15 @@ DEF_OP(Print) {
   auto Op = IROp->C<IR::IROp_Print>();
 
   PushRegs();
-  if (IsGPR(Op->Header.Args[0].ID())) {
-    mov (rdi, GetSrc<RA_64>(Op->Header.Args[0].ID()));
-    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.PrintValue)]);
+  if (IsGPR(Op->Value.ID())) {
+    mov (rdi, GetSrc<RA_64>(Op->Value.ID()));
+    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintValue)]);
   }
   else {
-    pextrq(rdi, GetSrc(Op->Header.Args[0].ID()), 0);
-    pextrq(rsi, GetSrc(Op->Header.Args[0].ID()), 1);
+    pextrq(rdi, GetSrc(Op->Value.ID()), 0);
+    pextrq(rsi, GetSrc(Op->Value.ID()), 1);
 
-    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.PrintVectorValue)]);
+    call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.PrintVectorValue)]);
   }
 
   PopRegs();
@@ -178,6 +167,7 @@ void X86JITCore::RegisterMiscHandlers() {
   REGISTER_OP(CODEBLOCK,  NoOp);
   REGISTER_OP(BEGINBLOCK, NoOp);
   REGISTER_OP(ENDBLOCK,   NoOp);
+  REGISTER_OP(GUESTOPCODE, GuestOpcode);
   REGISTER_OP(FENCE,      Fence);
   REGISTER_OP(BREAK,      Break);
   REGISTER_OP(PHI,        NoOp);

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

@@ -20,13 +20,13 @@ DEF_OP(ExtractElementPair) {
   auto Op = IROp->C<IR::IROp_ExtractElementPair>();
   switch (Op->Header.Size) {
     case 4: {
-      auto Src = GetSrcPair<RA_32>(Op->Header.Args[0].ID());
+      auto Src = GetSrcPair<RA_32>(Op->Pair.ID());
       std::array<Xbyak::Reg, 2> Regs = {Src.first, Src.second};
       mov (GetDst<RA_32>(Node), Regs[Op->Element]);
       break;
     }
     case 8: {
-      auto Src = GetSrcPair<RA_64>(Op->Header.Args[0].ID());
+      auto Src = GetSrcPair<RA_64>(Op->Pair.ID());
       std::array<Xbyak::Reg, 2> Regs = {Src.first, Src.second};
       mov (GetDst<RA_64>(Node), Regs[Op->Element]);
       break;
@@ -45,15 +45,15 @@ DEF_OP(CreateElementPair) {
   switch (IROp->ElementSize) {
     case 4: {
       Dst = GetSrcPair<RA_32>(Node);
-      RegFirst = GetSrc<RA_32>(Op->Header.Args[0].ID());
-      RegSecond = GetSrc<RA_32>(Op->Header.Args[1].ID());
+      RegFirst = GetSrc<RA_32>(Op->Lower.ID());
+      RegSecond = GetSrc<RA_32>(Op->Upper.ID());
       RegTmp = eax;
       break;
     }
     case 8: {
       Dst = GetSrcPair<RA_64>(Node);
-      RegFirst = GetSrc<RA_64>(Op->Header.Args[0].ID());
-      RegSecond = GetSrc<RA_64>(Op->Header.Args[1].ID());
+      RegFirst = GetSrc<RA_64>(Op->Lower.ID());
+      RegSecond = GetSrc<RA_64>(Op->Upper.ID());
       RegTmp = rax;
       break;
     }
@@ -73,17 +73,11 @@ DEF_OP(CreateElementPair) {
   }
 }
 
-DEF_OP(Mov) {
-  auto Op = IROp->C<IR::IROp_Mov>();
-  mov (GetDst<RA_64>(Node), GetSrc<RA_64>(Op->Header.Args[0].ID()));
-}
-
 #undef DEF_OP
 void X86JITCore::RegisterMoveHandlers() {
 #define REGISTER_OP(op, x) OpHandlers[FEXCore::IR::IROps::OP_##op] = &X86JITCore::Op_##x
   REGISTER_OP(EXTRACTELEMENTPAIR, ExtractElementPair);
   REGISTER_OP(CREATEELEMENTPAIR,  CreateElementPair);
-  REGISTER_OP(MOV,                Mov);
 #undef REGISTER_OP
 }
 }

External/FEXCore/Source/Interface/Core/JIT/x86_64/x64Relocations.cpp

@@ -4,6 +4,7 @@ tags: backend|x86-64
 desc: relocation logic of the x86-64 splatter backend
 $end_info$
 */
+#include "Interface/Context/Context.h"
 #include "Interface/Core/JIT/x86_64/JITClass.h"
 #include "Interface/HLE/Thunks/Thunks.h"
 
@@ -11,7 +12,7 @@ namespace FEXCore::CPU {
 uint64_t X86JITCore::GetNamedSymbolLiteral(FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol Op) {
   switch (Op) {
     case FEXCore::CPU::RelocNamedSymbolLiteral::NamedSymbol::SYMBOL_LITERAL_EXITFUNCTION_LINKER:
-      return Dispatcher->ExitFunctionLinkerAddress;
+      return ThreadState->CurrentFrame->Pointers.Common.ExitFunctionLinker;
     break;
     default:
       ERROR_AND_DIE_FMT("Unknown named symbol literal: {}", static_cast<uint32_t>(Op));
@@ -90,7 +91,7 @@ bool X86JITCore::ApplyRelocations(uint64_t GuestEntry, uint64_t CodeEntry, uint6
   size_t DataIndex{};
   for (size_t j = 0; j < NumRelocations; ++j) {
     const FEXCore::CPU::Relocation *Reloc = reinterpret_cast<const FEXCore::CPU::Relocation *>(&EntryRelocations[DataIndex]);
-    LOGMAN_THROW_A_FMT((DataIndex % alignof(Relocation)) == 0, "Alignment of relocation wasn't adhered to");
+    LOGMAN_THROW_AA_FMT((DataIndex % alignof(Relocation)) == 0, "Alignment of relocation wasn't adhered to");
 
     switch (Reloc->Header.Type) {
       case FEXCore::CPU::RelocationTypes::RELOC_NAMED_SYMBOL_LITERAL: {

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]
   //       |
@@ -29,54 +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));
-  LOGMAN_THROW_A_FMT(PageMemory != -1ULL, "Failed to allocate page memory");
+  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));
-  LOGMAN_THROW_A_FMT(L1Pointer != -1ULL, "Failed to allocate L1Pointer");
+  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);
 
-void LookupCache::HintUsedRange(uint64_t Address, uint64_t Size) {
-  // Tell the kernel we will definitely need [Address, Address+Size) mapped for the page pointer
-  // Page Pointer is allocated per page, so shift by page size
-  Address >>= 12;
-  Size >>= 12;
-  madvise(reinterpret_cast<void*>(PagePointer + Address), Size, MADV_WILLNEED);
+  // 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;
   };
@@ -25,9 +27,6 @@ public:
   LookupCache(FEXCore::Context::Context *CTX);
   ~LookupCache();
 
-  using LookupCacheIter = uintptr_t;
-  uintptr_t End() { return 0; }
-
   uintptr_t FindBlock(uint64_t Address) {
     // Try L1, no lock needed
     auto &L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
@@ -57,7 +56,7 @@ public:
         return L1Entry.HostCode;
       }
     }
-    
+
     // Try L3
     auto HostCode = BlockList.find(Address);
 
@@ -65,37 +64,41 @@ public:
       CacheBlockMapping(Address, HostCode->second);
       return HostCode->second;
     }
-      
+
     // Failed to find
     return 0;
   }
 
   std::map<uint64_t, std::vector<uint64_t>> CodePages;
 
+  // Appends Block {Address} to CodePages [Start, Start + Length)
   // Returns true if new pages are marked as containing code
-  bool AddBlockMapping(uint64_t Address, void *HostCode, uint64_t Start, uint64_t Length) {
+  bool AddBlockExecutableRange(uint64_t Address, uint64_t Start, uint64_t Length) {
     std::lock_guard<std::recursive_mutex> lk(WriteLock);
-    
-#if defined(ASSERTIONS_ENABLED) && ASSERTIONS_ENABLED
-    auto InsertPoint =
-#endif
-    BlockList.emplace(Address, (uintptr_t)HostCode);
-    LOGMAN_THROW_A_FMT(InsertPoint.second == true, "Dupplicate block mapping added");
 
     bool rv = false;
 
-    for (auto CurrentPage = Start >> 12, EndPage = (Start + Length) >> 12; CurrentPage <= EndPage; CurrentPage++) {
-      rv |= CodePages[CurrentPage].size() == 0;
-      CodePages[CurrentPage].push_back(Address);
+    for (auto CurrentPage = Start >> 12, EndPage = (Start + Length -1) >> 12; CurrentPage <= EndPage; CurrentPage++) {
+      auto &CodePage = CodePages[CurrentPage];
+      rv |= CodePage.size() == 0;
+      CodePage.push_back(Address);
     }
 
+    return rv;
+  }
+
+  // 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");
+
     // There is no need to update L1 or L2, they will get updated on first lookup
     // However, adding to L1 here increases performance
     auto &L1Entry = reinterpret_cast<LookupCacheEntry*>(L1Pointer)[Address & L1_ENTRIES_MASK];
     L1Entry.GuestCode = Address;
     L1Entry.HostCode = (uintptr_t)HostCode;
-
-    return rv;
   }
 
   void Erase(uint64_t Address) {
@@ -103,9 +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();
     }
 
@@ -143,14 +146,12 @@ 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();
   void ClearL2Cache();
 
-  void HintUsedRange(uint64_t Address, uint64_t Size);
-
   uintptr_t GetL1Pointer() const { return L1Pointer; }
   uintptr_t GetPagePointer() const { return PagePointer; }
   uintptr_t GetVirtualMemorySize() const { return VirtualMemSize; }
@@ -158,8 +159,8 @@ public:
   constexpr static size_t L1_ENTRIES = 1 * 1024 * 1024; // Must be a power of 2
   constexpr static size_t L1_ENTRIES_MASK = L1_ENTRIES - 1;
 
-  // This needs to be taken before reads or writes to L2, L3, CodePages, Thread::LocalIRCache,
-  // and before writes to L1. Concurrent access from a thread that this LookupCache doesn't belong to 
+  // 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
   // 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.
@@ -168,7 +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
@@ -238,9 +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

@@ -76,6 +76,8 @@ public:
   OrderedNode* flagsOpSrcSigned{};
 
   FEXCore::Context::Context *CTX{};
+
+  // Used during new op bringup
   bool ShouldDump {false};
 
   struct JumpTargetInfo {
@@ -151,8 +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();
@@ -276,6 +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);
@@ -290,6 +299,8 @@ public:
   void WriteSegmentReg(OpcodeArgs);
   void EnterOp(OpcodeArgs);
 
+  void SGDTOp(OpcodeArgs);
+
   // SSE
   void MOVAPSOp(OpcodeArgs);
   void MOVUPSOp(OpcodeArgs);
@@ -312,10 +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>
@@ -340,8 +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);
@@ -368,15 +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);
@@ -396,6 +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);
@@ -513,7 +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);
 
@@ -538,12 +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);
 
@@ -568,12 +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);
 
@@ -621,6 +727,8 @@ public:
   void DPPOp(OpcodeArgs);
 
   void MPSADBWOp(OpcodeArgs);
+  void PCLMULQDQOp(OpcodeArgs);
+  void VPCLMULQDQOp(OpcodeArgs);
 
   void CRC32(OpcodeArgs);
 
@@ -628,8 +736,6 @@ public:
 
   void InvalidOp(OpcodeArgs);
 
-#undef OpcodeArgs
-
   void SetPackedRFLAG(bool Lower8, OrderedNode *Src);
   OrderedNode *GetPackedRFLAG(bool Lower8);
 
@@ -638,10 +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
@@ -653,6 +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);
@@ -661,7 +849,7 @@ private:
   void StoreResult(FEXCore::IR::RegisterClassType Class, FEXCore::X86Tables::DecodedOp Op, OrderedNode *const Src, int8_t Align, MemoryAccessType AccessType = MemoryAccessType::ACCESS_DEFAULT);
 
   [[nodiscard]] static uint32_t GPROffset(X86State::X86Reg reg) {
-    LOGMAN_THROW_A_FMT(reg <= X86State::X86Reg::REG_R15, "Invalid reg used");
+    LOGMAN_THROW_AA_FMT(reg <= X86State::X86Reg::REG_R15, "Invalid reg used");
     return static_cast<uint32_t>(offsetof(Core::CPUState, gregs[static_cast<size_t>(reg)]));
   }
 
@@ -1236,14 +1424,14 @@ private:
   uint64_t Entry;
 
   OrderedNode* _StoreMemAutoTSO(FEXCore::IR::RegisterClassType Class, uint8_t Size, OrderedNode *Addr, OrderedNode *Value, uint8_t Align = 1) {
-    if (CTX->Config.TSOEnabled)
+    if (CTX->IsTSOEnabled())
       return _StoreMemTSO(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
     else
       return _StoreMem(Class, Size, Value, Addr, Invalid(), Align, MEM_OFFSET_SXTX, 1);
   }
 
   OrderedNode* _LoadMemAutoTSO(FEXCore::IR::RegisterClassType Class, uint8_t Size, OrderedNode *ssa0, uint8_t Align = 1) {
-    if (CTX->Config.TSOEnabled)
+    if (CTX->IsTSOEnabled())
       return _LoadMemTSO(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);
     else
       return _LoadMem(Class, Size, ssa0, Invalid(), Align, MEM_OFFSET_SXTX, 1);

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[0]));
+  // Hardcoded to XMM0
+  auto XMM0 = LoadXMMRegister(0);
 
   auto A0 = _VExtractToGPR(16, 4, Src, 3);
   auto B0 = _VExtractToGPR(16, 4, Src, 2);
@@ -263,46 +261,163 @@ 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 *Result = AESKeyGenAssistImpl(Op);
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::VAESKeyGenAssistOp(OpcodeArgs) {
+  OrderedNode *Assist = AESKeyGenAssistImpl(Op);
+  OrderedNode *Result = _VMov(16, Assist);
+  StoreResult(FPRClass, Op, Result, -1);
+}
+
+void OpDispatchBuilder::PCLMULQDQOp(OpcodeArgs) {
+  LOGMAN_THROW_A_FMT(Op->Src[1].IsLiteral(), "Selector needs to be literal here");
+
+  OrderedNode *Dest = LoadSource(FPRClass, Op, Op->Dest, Op->Flags, -1);
   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;
+  const auto Selector = static_cast<uint8_t>(Op->Src[1].Data.Literal.Value);
+
+  auto Res = _PCLMUL(Dest, Src, Selector);
+  StoreResult(FPRClass, Op, Res, -1);
+}
+
+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);
+
+  OrderedNode *Res = _PCLMUL(Src1, Src2, Selector);
+  if (Is128Bit) {
+    Res = _VMov(16, Res);
+  }
 
-  auto Res = _VAESKeyGenAssist(Src, RCON);
   StoreResult(FPRClass, Op, Res, -1);
 }