Emitter: Remove unused header

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

.github/ISSUE_TEMPLATE/potential-game-bug.md

@@ -0,0 +1,45 @@
+---
+name: Potential Game Bug
+about: A bug in FEX-Emu that causes a problem in a game
+title: "[Game]: [Short Problem Description]"
+labels: Game related
+assignees: ''
+
+---
+
+**What Game**
+The game name.
+A link to the storefront where to get the game. GOG, Steam, Itch.io, etc
+
+**Describe the bug**
+A clear and concise description of what the bug is.
+
+**To Reproduce**
+Steps to reproduce the behavior:
+1. Go to '...'
+2. Click on '....'
+3. Scroll down to '....'
+4. See error
+
+**Expected behavior**
+A clear and concise description of what you expected to happen.
+
+**Screenshots and Video**
+If applicable, add screenshots and video to help explain your problem.
+
+**System information:**
+ - OS: [eg: Ubuntu 21.10]
+ - CPU/SoC: [eg: Snapdragon 888, Intel Core i8-12900k]
+ - Video driver version: [eg: OpenGL ES 3.2 Mesa 22.0.0-devel (git-9ff086052a)]
+ - RootFS used: [eg: Ubuntu 21.10 Official Rootfs]
+ - FEX version: (FEXGetConfig --version) [eg: FEX-2112-155-gc691d709]
+ - Thunks Enabled: [Yes/No]
+
+**Additional context**
+ - Is this an x86 or x86-64 game: [x86/x86-64/Both]
+ - Does this reproduce on x86-64 host with FEX: [Yes/No/Untested]
+ - Does this reproduce on AArch64 with Radeon/Intel/Nvidia: [Yes/No/Untested]
+ - Is this a Vulkan game: [Yes/No/Unknown]
+   - If Yes, What is your Vulkan driver:
+
+Add any other context about the problem here.

.github/workflows/ccpp.yml

@@ -29,9 +29,24 @@ 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: git submodule update --init --depth 1
+      run: |
+        git submodule sync --recursive
+        git submodule update --init --depth 1
 
     - name: Clean Build Environment
       run: rm -Rf ${{runner.workspace}}/build
@@ -49,7 +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
+      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
@@ -140,6 +155,73 @@ jobs:
       working-directory: ${{runner.workspace}}/build
       run: mv ${{runner.workspace}}/build/Testing/Temporary/LastTest.log ${{runner.workspace}}/build/Testing/Temporary/LastTest_StructVerifier.log || true
 
+    - name: APITest tests
+      working-directory: ${{runner.workspace}}/build
+      shell: bash
+      run: cmake --build . --config $BUILD_TYPE --target api_tests
+
+    - name: APITest 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_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: ${{ always() }}
       shell: bash
@@ -159,4 +241,3 @@ jobs:
         name: Results-${{ 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

@@ -1,7 +1,7 @@
 [submodule "External/vixl"]
 	shallow = true
 	path = External/vixl
-	url = https://github.com/Sonicadvance1/vixl.git
+	url = https://github.com/FEX-Emu/vixl.git
 [submodule "External/cpp-optparse"]
 	path = External/cpp-optparse
 	url = https://github.com/Sonicadvance1/cpp-optparse
@@ -42,7 +42,14 @@
 [submodule "External/xxhash"]
 	path = External/xxhash
 	url = https://github.com/FEX-Emu/xxHash.git
-[submodule "External/Vulkan-Docs"]
-  shallow = true
-	path = External/Vulkan-Docs
-	url = https://github.com/KhronosGroup/Vulkan-Docs.git
+[submodule "External/Catch2"]
+	path = External/Catch2
+	url = https://github.com/catchorg/Catch2.git
+[submodule "External/robin-map"]
+	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

@@ -1,27 +1,69 @@
 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)
 option(ENABLE_XRAY "Enable building with LLVM X-Ray" FALSE)
-option(ENABLE_LLD "Enable linking with LLD" FALSE)
+option(ENABLE_LLD "Enable linking with lld" FALSE)
+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)
+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 ${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/cmake_uninstall.cmake)
+endif()
+
+# These options are meant for package management
+set (TUNE_CPU "native" CACHE STRING "Override the CPU the build is tuned for")
+set (TUNE_ARCH "generic" CACHE STRING "Override the Arch the build is tuned for")
+set (OVERRIDE_VERSION "detect" CACHE STRING "Override the FEX version in the format of <MMYY>{.<REV>}")
+
 string(TOUPPER "${CMAKE_BUILD_TYPE}" CMAKE_BUILD_TYPE)
 if (CMAKE_BUILD_TYPE MATCHES "DEBUG")
   set(ENABLE_ASSERTIONS TRUE)
@@ -32,6 +74,17 @@ 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)
+endif()
+
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/Bin)
@@ -61,15 +114,17 @@ if (CMAKE_SYSTEM_PROCESSOR MATCHES "x86_64")
   set (CMAKE_CXX_FLAGS "${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()
 
-find_program(CCACHE_PROGRAM ccache)
-if(CCACHE_PROGRAM)
-  message(STATUS "CCache enabled")
-  set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "${CCACHE_PROGRAM}")
+if (ENABLE_CCACHE)
+  find_program(CCACHE_PROGRAM ccache)
+  if(CCACHE_PROGRAM)
+    message(STATUS "CCache enabled")
+    set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "${CCACHE_PROGRAM}")
+  endif()
 endif()
 
 if (ENABLE_XRAY)
@@ -77,10 +132,26 @@ if (ENABLE_XRAY)
   link_libraries(-fxray-instrument)
 endif()
 
+if (ENABLE_COMPILE_TIME_TRACE)
+  add_compile_options(-ftime-trace)
+  link_libraries(-ftime-trace)
+endif()
+
 set (PTHREAD_LIB pthread)
-if (ENABLE_LLD)
+
+if (ENABLE_LLD AND ENABLE_MOLD)
+  message (FATAL_ERROR "Cannot enable both lld and mold")
+elseif (ENABLE_LLD)
   set (LD_OVERRIDE "-fuse-ld=lld")
-  link_libraries(${LD_OVERRIDE})
+  add_link_options(${LD_OVERRIDE})
+elseif (ENABLE_MOLD)
+  add_link_options("-fuse-ld=mold")
+endif()
+
+if (ENABLE_LIBCXX)
+  message(WARNING "This is an unsupported configuration and should only be used for testing")
+  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -stdlib=libc++")
+  set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -stdlib=libc++ -lc++abi")
 endif()
 
 if (NOT ENABLE_OFFLINE_TELEMETRY)
@@ -88,128 +159,11 @@ if (NOT ENABLE_OFFLINE_TELEMETRY)
   add_definitions(-DFEX_DISABLE_TELEMETRY=1)
 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!")
-  endif()
+if(DEFINED ENV{TERMUX_VERSION} OR ENABLE_TERMUX_BUILD)
+  add_definitions(-DTERMUX_BUILD=1)
+  set(TERMUX_BUILD 1)
+  # Termux doesn't support Jemalloc due to bad interactions between emutls, jemalloc, and scudo
+  set(ENABLE_JEMALLOC FALSE)
 endif()
 
 if (ENABLE_ASAN)
@@ -241,6 +195,8 @@ set (CMAKE_LINKER_FLAGS_RELWITHDEBINFO "${CMAKE_LINKER_FLAGS_RELWITHDEBINFO} -fn
 set (CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} -fomit-frame-pointer")
 set (CMAKE_LINKER_FLAGS_RELEASE "${CMAKE_LINKER_FLAGS_RELEASE} -fomit-frame-pointer")
 
+include_directories(External/robin-map/include/)
+
 add_subdirectory(External/vixl/)
 include_directories(External/vixl/src/)
 
@@ -251,17 +207,23 @@ endif()
 
 find_package(PkgConfig REQUIRED)
 find_package(Python 3.0 REQUIRED COMPONENTS Interpreter)
-pkg_check_modules(XXHASH libxxhash>=0.8.0 QUIET)
 
-if (NOT XXHASH_FOUND)
-  message(STATUS "xxHash not found. Using Externals")
-  add_subdirectory(External/xxhash/)
-  include_directories(External/xxhash/)
-endif()
+add_subdirectory(External/xxhash/)
+include_directories(External/xxhash/)
 
 add_definitions(-Wno-trigraphs)
 add_definitions(-DGLOBAL_DATA_DIRECTORY="${DATA_DIRECTORY}/")
 
+if (BUILD_TESTS)
+  option(CATCH_BUILD_STATIC_LIBRARY "" ON)
+  set(CATCH_BUILD_STATIC_LIBRARY ON)
+  add_subdirectory(External/Catch2/)
+
+  # Pull in catch_discover_tests definition
+  list(APPEND CMAKE_MODULE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/External/Catch2/contrib/")
+  include(Catch)
+endif()
+
 add_subdirectory(External/cpp-optparse/)
 include_directories(External/cpp-optparse/)
 
@@ -300,11 +262,6 @@ if(ENUM_ENUM_WARNING)
   add_compile_options(-Wno-deprecated-enum-enum-conversion)
 endif()
 
-check_cxx_compiler_flag("-march=native" COMPILER_SUPPORTS_MARCH_NATIVE)
-if(COMPILER_SUPPORTS_MARCH_NATIVE)
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native")
-endif()
-
 if(ENABLE_WERROR OR ENABLE_STRICT_WERROR)
   add_compile_options(-Werror)
   if (NOT ENABLE_STRICT_WERROR)
@@ -313,29 +270,52 @@ if(ENABLE_WERROR OR ENABLE_STRICT_WERROR)
   endif()
 endif()
 
-if(_M_ARM_64)
-  if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 999999.0)
-    # Clang 12.0 fixed the -mcpu=native bug with mixed big.little implementers
-    # Clang can not currently check for native Apple M1 type in hypervisor. Currently disabled
-    check_cxx_compiler_flag("-mcpu=native" COMPILER_SUPPORTS_CPU_TYPE)
-    if(COMPILER_SUPPORTS_CPU_TYPE)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcpu=native")
+if (NOT TUNE_ARCH STREQUAL "generic")
+  check_cxx_compiler_flag("-march=${TUNE_ARCH}" COMPILER_SUPPORTS_ARCH_TYPE)
+  if(COMPILER_SUPPORTS_ARCH_TYPE)
+    add_compile_options("-march=${TUNE_ARCH}")
+  else()
+    message(FATAL_ERROR "Trying to compile arch type '${TUNE_ARCH}' but the compiler doesn't support this")
+  endif()
+endif()
+
+if (TUNE_CPU STREQUAL "native")
+  if(_M_ARM_64)
+    if (CMAKE_CXX_COMPILER_VERSION VERSION_GREATER_EQUAL 999999.0)
+      # Clang 12.0 fixed the -mcpu=native bug with mixed big.little implementers
+      # Clang can not currently check for native Apple M1 type in hypervisor. Currently disabled
+      check_cxx_compiler_flag("-mcpu=native" COMPILER_SUPPORTS_CPU_TYPE)
+      if(COMPILER_SUPPORTS_CPU_TYPE)
+        add_compile_options("-mcpu=native")
+      endif()
+    else()
+      # Due to an oversight in llvm, it declares any reasonably new Kryo CPU to only be ARMv8.0
+      # Manually detect newer CPU revisions until clang and llvm fixes their bug
+      # This script will either provide a supported CPU or 'native'
+      # Additionally -march doesn't work under AArch64+Clang, so you have to use -mcpu or -mtune
+      execute_process(COMMAND python3 "${PROJECT_SOURCE_DIR}/Scripts/aarch64_fit_native.py" "/proc/cpuinfo" "${CMAKE_CXX_COMPILER_VERSION}"
+        OUTPUT_VARIABLE AARCH64_CPU)
+
+      string(STRIP ${AARCH64_CPU} AARCH64_CPU)
+
+      check_cxx_compiler_flag("-mcpu=${AARCH64_CPU}" COMPILER_SUPPORTS_CPU_TYPE)
+      if(COMPILER_SUPPORTS_CPU_TYPE)
+        add_compile_options("-mcpu=${AARCH64_CPU}")
+      endif()
     endif()
   else()
-    # Due to an oversight in llvm, it declares any reasonably new Kryo CPU to only be ARMv8.0
-    # Manually detect newer CPU revisions until clang and llvm fixes their bug
-    # This script will either provide a supported CPU or 'native'
-    # Additionally -march doesn't work under AArch64+Clang, so you have to use -mcpu or -mtune
-    execute_process(COMMAND python3 "${PROJECT_SOURCE_DIR}/Scripts/aarch64_fit_native.py" "/proc/cpuinfo" "${CMAKE_CXX_COMPILER_VERSION}"
-      OUTPUT_VARIABLE AARCH64_CPU)
-
-    string(STRIP ${AARCH64_CPU} AARCH64_CPU)
-
-    check_cxx_compiler_flag("-mcpu=${AARCH64_CPU}" COMPILER_SUPPORTS_CPU_TYPE)
-    if(COMPILER_SUPPORTS_CPU_TYPE)
-      set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcpu=${AARCH64_CPU}")
+    check_cxx_compiler_flag("-march=native" COMPILER_SUPPORTS_MARCH_NATIVE)
+    if(COMPILER_SUPPORTS_MARCH_NATIVE)
+      add_compile_options("-march=native")
     endif()
   endif()
+else()
+  check_cxx_compiler_flag("-mcpu=${TUNE_CPU}" COMPILER_SUPPORTS_CPU_TYPE)
+  if(COMPILER_SUPPORTS_CPU_TYPE)
+    add_compile_options("-mcpu=${TUNE_CPU}")
+  else()
+    message(FATAL_ERROR "Trying to compile cpu type '${TUNE_CPU}' but the compiler doesn't support this")
+  endif()
 endif()
 
 if (ENABLE_IWYU)
@@ -408,6 +388,8 @@ if (BUILD_TESTS)
   enable_testing()
   message(STATUS "Unit tests are enabled")
 endif()
+
+add_subdirectory(FEXHeaderUtils/)
 add_subdirectory(External/FEXCore)
 
 # Binfmt_misc files must be installed prior to Source/ installs
@@ -417,120 +399,163 @@ 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 ${PROJECT_SOURCE_DIR})
+  add_subdirectory(ThunkLibs/Generator)
+
+  # Thunk targets for both host libraries and IDE integration
+  add_subdirectory(ThunkLibs/HostLibs)
+
+  # Thunk targets for IDE integration of guest code, only
+  add_subdirectory(ThunkLibs/GuestLibs)
+
+  # Thunk targets for guest libraries
   include(ExternalProject)
-
-  ExternalProject_Add(host-libs
-    PREFIX host-libs
-    SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}/ThunkLibs/HostLibs"
-    BINARY_DIR "Host"
-    CMAKE_ARGS
-      "-DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE}"
-      "-DCMAKE_INSTALL_PREFIX=${CMAKE_INSTALL_PREFIX}"
-      "-DVULKAN_XML=${CMAKE_SOURCE_DIR}/External/Vulkan-Docs/xml/vk.xml"
-    INSTALL_COMMAND ""
-    BUILD_ALWAYS ON
-  )
-
-  install(
-    CODE "MESSAGE(\"-- Installing: host-libs\")"
-    CODE "
-    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target ThunkHostsInstall
-    WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Host
-    )"
-    DEPENDS host-libs
-  )
-
   ExternalProject_Add(guest-libs
     PREFIX guest-libs
     SOURCE_DIR "${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=${CMAKE_SOURCE_DIR}/Scripts/StructPackVerifier.py"
-      "-DVULKAN_XML=${CMAKE_SOURCE_DIR}/External/Vulkan-Docs/xml/vk.xml"
+      "-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
+    DEPENDS thunkgen
   )
 
   install(
     CODE "MESSAGE(\"-- Installing: guest-libs\")"
     CODE "
-    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target ThunkGuestsInstall
+    EXECUTE_PROCESS(COMMAND ${CMAKE_COMMAND} --build . --target install
     WORKING_DIRECTORY ${CMAKE_BINARY_DIR}/Guest
     )"
     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")
 set(FEX_VERSION_MINOR "0")
 set(FEX_VERSION_PATCH "0")
 
-find_package(Git)
-if (GIT_FOUND)
-  execute_process(
-    COMMAND ${GIT_EXECUTABLE} describe --abbrev=0
-    WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
-    OUTPUT_VARIABLE GIT_DESCRIBE_STRING
-    RESULT_VARIABLE GIT_ERROR
-    ERROR_QUIET
-    OUTPUT_STRIP_TRAILING_WHITESPACE
-  )
+if (OVERRIDE_VERSION STREQUAL "detect")
+  find_package(Git)
+  if (GIT_FOUND)
+    execute_process(
+      COMMAND ${GIT_EXECUTABLE} describe --abbrev=0
+      WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
+      OUTPUT_VARIABLE GIT_DESCRIBE_STRING
+      RESULT_VARIABLE GIT_ERROR
+      ERROR_QUIET
+      OUTPUT_STRIP_TRAILING_WHITESPACE
+    )
 
-  if (NOT ${GIT_ERROR} EQUAL 0)
-    # Likely built in a way that doesn't have tags
-    # Setup a version tag that is unknown
-    set(GIT_DESCRIBE_STRING "FEX-0000")
+    if (NOT ${GIT_ERROR} EQUAL 0)
+      # Likely built in a way that doesn't have tags
+      # Setup a version tag that is unknown
+      set(GIT_DESCRIBE_STRING "FEX-0000")
+    endif()
   endif()
+else()
+  set(GIT_DESCRIBE_STRING "FEX-${OVERRIDE_VERSION}")
+endif()
 
-  # Change something like `FEX-2106.1-76-<hash>` in to a list
-  string(REPLACE "-" ";" DESCRIBE_LIST ${GIT_DESCRIBE_STRING})
+# Parse the version here
+# Change something like `FEX-2106.1-76-<hash>` in to a list
+string(REPLACE "-" ";" DESCRIBE_LIST ${GIT_DESCRIBE_STRING})
 
-  # Extract the `2106.1` element
-  list(GET DESCRIBE_LIST 1 DESCRIBE_LIST)
+# Extract the `2106.1` element
+list(GET DESCRIBE_LIST 1 DESCRIBE_LIST)
 
-  # Change `2106.1` in to a list
-  string(REPLACE "." ";" DESCRIBE_LIST ${DESCRIBE_LIST})
+# Change `2106.1` in to a list
+string(REPLACE "." ";" DESCRIBE_LIST ${DESCRIBE_LIST})
 
-  # Calculate list size
-  list(LENGTH DESCRIBE_LIST LIST_SIZE)
+# Calculate list size
+list(LENGTH DESCRIBE_LIST LIST_SIZE)
 
-  # Pull out the major version
-  list(GET DESCRIBE_LIST 0 FEX_VERSION_MAJOR)
+# Pull out the major version
+list(GET DESCRIBE_LIST 0 FEX_VERSION_MAJOR)
 
-  # Minor version only exists if there is a .1 at the end
-  # eg: 2106 versus 2106.1
-  if (LIST_SIZE GREATER 1)
-    list(GET DESCRIBE_LIST 1 FEX_VERSION_MINOR)
-  endif()
+# Minor version only exists if there is a .1 at the end
+# eg: 2106 versus 2106.1
+if (LIST_SIZE GREATER 1)
+  list(GET DESCRIBE_LIST 1 FEX_VERSION_MINOR)
 endif()
 
 # Package creation
 set (CPACK_GENERATOR "DEB")
 set (CPACK_PACKAGE_NAME fex-emu)
-set (CPACK_PACKAGE_CONTACT "team@fex-emu.org")
+set (CPACK_PACKAGE_FILE_NAME "${CPACK_PACKAGE_NAME}-${GIT_DESCRIBE_STRING}_${CMAKE_SYSTEM_PROCESSOR}")
+set (CPACK_PACKAGE_CONTACT "FEX-Emu Maintainers <team@fex-emu.com>")
 set (CPACK_PACKAGE_VERSION_MAJOR "${FEX_VERSION_MAJOR}")
 set (CPACK_PACKAGE_VERSION_MINOR "${FEX_VERSION_MINOR}")
 set (CPACK_PACKAGE_VERSION_PATCH "${FEX_VERSION_PATCH}")
+set (CPACK_PACKAGE_DESCRIPTION_FILE "${CMAKE_CURRENT_SOURCE_DIR}/CPack/Description.txt")
 
 # Debian defines
-set (CPACK_DEBIAN_PACKAGE_DEPENDS "libstdc++6")
-set (CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA "${CMAKE_CURRENT_SOURCE_DIR}/CPack/postinst;${CMAKE_CURRENT_SOURCE_DIR}/CPack/prerm")
+set (CPACK_DEBIAN_PACKAGE_DEPENDS "libc6, libstdc++6, libepoxy0, libsdl2-2.0-0, libegl1, libx11-6, squashfuse")
+set (CPACK_DEBIAN_PACKAGE_CONTROL_EXTRA
+  "${CMAKE_CURRENT_SOURCE_DIR}/CPack/postinst;${CMAKE_CURRENT_SOURCE_DIR}/CPack/prerm;${CMAKE_CURRENT_SOURCE_DIR}/CPack/triggers")
 if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64")
   # binfmt_misc conflicts with qemu-user-static
   # We also only install binfmt_misc on aarch64 hosts
-  set (CPACK_DEBIAN_PACKAGE_CONFLICTS "qemu-user-static")
+  set (CPACK_DEBIAN_PACKAGE_CONFLICTS "${CPACK_DEBIAN_PACKAGE_CONFLICTS}, qemu-user-static")
 endif()
 include (CPack)

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.

CPack/Description.txt

@@ -0,0 +1,3 @@
+x86 and x86-64 Linux emulator
+
+FEX is very much work in progress, so expect things to change.

CPack/triggers

@@ -0,0 +1 @@
+activate-noawait ldconfig

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/steamwebhelper.json

@@ -1,5 +1,5 @@
 {
   "Config": {
-    "StallProcess": "1"
+    "AdditionalArguments": "--no-sandbox"
   }
 }

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

@@ -6,14 +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",
-        "/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": {
@@ -22,264 +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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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",
-        "/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

@@ -9,14 +9,20 @@ 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)
-option(ENABLE_JITSYMBOLS "Enable visibility of JITSymbols in profiling tools" FALSE)
 
 set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 cmake_policy(SET CMP0083 NEW) # Follow new PIE policy
@@ -28,7 +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/)
@@ -38,27 +43,32 @@ endif()
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
-# Find our git hash
-find_package(Git)
-
 set(GIT_SHORT_HASH "Unknown")
 set(GIT_DESCRIBE_STRING "FEX-Unknown")
 
-if (GIT_FOUND)
-  execute_process(
-    COMMAND ${GIT_EXECUTABLE} rev-parse --short HEAD
-    WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
-    OUTPUT_VARIABLE GIT_SHORT_HASH
-    ERROR_QUIET
-    OUTPUT_STRIP_TRAILING_WHITESPACE
-  )
-  execute_process(
-    COMMAND ${GIT_EXECUTABLE} describe
-    WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
-    OUTPUT_VARIABLE GIT_DESCRIBE_STRING
-    ERROR_QUIET
-    OUTPUT_STRIP_TRAILING_WHITESPACE
-  )
+if (OVERRIDE_VERSION STREQUAL "detect")
+# Find our git hash
+  find_package(Git)
+
+  if (GIT_FOUND)
+    execute_process(
+      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 ${GIT_EXECUTABLE} describe --abbrev=7
+      WORKING_DIRECTORY "${CMAKE_SOURCE_DIR}"
+      OUTPUT_VARIABLE GIT_DESCRIBE_STRING
+      ERROR_QUIET
+      OUTPUT_STRIP_TRAILING_WHITESPACE
+    )
+  endif()
+else()
+  set(GIT_SHORT_HASH "${OVERRIDE_VERSION}")
+  set(GIT_DESCRIBE_STRING "FEX-${OVERRIDE_VERSION}")
 endif()
 
 configure_file(
@@ -72,3 +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/Readme.md

@@ -18,14 +18,8 @@ This project aims to provide a fast and functional x86-64 emulation library that
 * Portable library implementation in order to support easy integration in to applications
 ### Target Host Architecture
 The target host architecture for this library is AArch64. Specifically the ARMv8.1 version or newer.
-The CPU IR is designed with AArch64 in mind but there is a desire to run the recompiled code on other architectures as well.
-Multiple architecture support is desired for easier bringup and debugging, performance isn't as much of a priority there (ex. x86-64(guest) translated to x86-64(host))
-### Not currently goals but will be in the future
-* 32bit x86 support
-  * This will be a desire in the future, but to lower the amount of work required, decided to push this off for now.
-* Integration in to WINE
-* Later generation of x86-64 instruction sets
-  * Including AVX, F16C, XOP, FMA, AVX2, etc
+The CPU IR is designed with AArch64 in mind but should allow for other architectures as well.
+x86-64 host support is available for ease of development, but is not a priority.
 ### Not desired
 * Kernel space emulation
 * CPL0-2 emulation
@@ -33,7 +27,3 @@ Multiple architecture support is desired for easier bringup and debugging, perfo
 * IRQs
 * SVM
 * "Cycle Accurate" emulation
-### Dependencies
- * clang-tidy if you want to ensure the code stays tidy
- * cmake
- * A C++17 compliant compiler (There are assumptions made about using Clang and LTO)

External/FEXCore/Scripts/config_generator.py

@@ -374,7 +374,7 @@ def print_parse_argloader_options(options):
             conversion_func = "std::to_string"
             if ("ArgumentHandler" in op_vals):
                 NeedsString = True
-                conversion_func = "FEX::Handler::{0}".format(op_vals["ArgumentHandler"])
+                conversion_func = "FEXCore::Config::Handler::{0}".format(op_vals["ArgumentHandler"])
             if (value_type == "str"):
                 NeedsString = True
                 conversion_func = ""
@@ -396,6 +396,21 @@ def print_parse_argloader_options(options):
 
     output_argloader.write("#endif\n")
 
+
+def print_parse_envloader_options(options):
+    output_argloader.write("#ifdef ENVLOADER\n")
+    output_argloader.write("#undef ENVLOADER\n")
+    output_argloader.write("if (false) {}\n")
+
+    for op_group, group_vals in options.items():
+        for op_key, op_vals in group_vals.items():
+            if ("ArgumentHandler" in op_vals):
+                conversion_func = "FEXCore::Config::Handler::{0}".format(op_vals["ArgumentHandler"])
+                output_argloader.write("else if (Key == \"FEX_{0}\") {{\n".format(op_key.upper()))
+                output_argloader.write("Value = {0}(Value);\n".format(conversion_func))
+                output_argloader.write("}\n")
+    output_argloader.write("#endif\n")
+
 def check_for_duplicate_options(options):
     short_map = []
     long_map = []
@@ -470,4 +485,8 @@ output_man.close()
 output_argloader = open(output_argumentloader_filename, "w")
 print_argloader_options(options);
 print_parse_argloader_options(options);
+
+# Generate environment loader code
+print_parse_envloader_options(options);
+
 output_argloader.close()

External/FEXCore/Scripts/json_ir_doc_generator.py

@@ -7,17 +7,12 @@ OpClasses = collections.OrderedDict()
 def get_ir_classes(ops, defines):
     global OpClasses
 
-    for op_key, op_vals in ops.items():
-        if not ("Last" in op_vals):
-            OpClass = "#Unknown"
+    for op_class, opslist in ops.items():
+        if not (op_class in OpClasses):
+            OpClasses[op_class] = []
 
-            if ("OpClass" in op_vals):
-                OpClass = op_vals["OpClass"]
-
-            if not (OpClass in OpClasses):
-                OpClasses[OpClass] = []
-
-            OpClasses[OpClass].append([op_key, op_vals])
+        for op, op_val in opslist.items():
+            OpClasses[op_class].append([op, op_val])
 
     # Sort the dictionary after we are done parsing it
     OpClasses = collections.OrderedDict(sorted(OpClasses.items()))
@@ -38,41 +33,9 @@ def print_ir_ops():
             op_key = op[0]
             op_vals = op[1]
             output_file.write("## %s\n" % (op_key))
-            HasDest = ("HasDest" in op_vals and op_vals["HasDest"] == True)
-            HasSSAArgs = ("SSAArgs" in op_vals and len(op_vals["SSAArgs"]) > 0)
-            HasSSAArgNames = "SSANames" in op_vals
-            HasArgs = "Args" in op_vals
-            SSAArgsCount = 0
-            ArgCount = 0
-            if (HasSSAArgs):
-                SSAArgsCount = int(op_vals["SSAArgs"])
-            if (HasArgs):
-                ArgCount = len(op_vals["Args"])
-
-            TotalArgsCount = SSAArgsCount + (ArgCount / 2)
 
             output_file.write(">")
-            if (HasDest):
-                output_file.write("%dest = ")
-
-            output_file.write("%s " % op_key)
-
-            ArgComma = (", ", "")
-            if (HasSSAArgs):
-                for i in range(0, SSAArgsCount):
-                    FinalArg = (i + 1) == TotalArgsCount
-                    if (HasSSAArgNames):
-                        output_file.write("%%%s%s" % (op_vals["SSANames"][i], ArgComma[FinalArg]))
-                    else:
-                        output_file.write("%%ssa%d%s" % (i, ArgComma[FinalArg]))
-
-            if (HasArgs):
-                Args = op_vals["Args"]
-                for i in range(0, ArgCount, 2):
-                    FinalArg = ((i / 2) + SSAArgsCount + 1) == TotalArgsCount
-                    data_type = Args[i]
-                    data_name = Args[i + 1]
-                    output_file.write("\<%s %s\>%s" % (data_type, data_name, ArgComma[FinalArg]))
+            output_file.write(op_key)
 
             output_file.write("\n\n")
 

External/FEXCore/Source/CMakeLists.txt

@@ -1,9 +1,15 @@
 set (MAN_DIR ${CMAKE_INSTALL_PREFIX}/share/man CACHE PATH "MAN_DIR")
 
-set (SRCS
+set (FEXCORE_BASE_SRCS
   Common/Paths.cpp
+  Interface/Config/Config.cpp
+  Utils/FileLoading.cpp
+  Utils/ForcedAssert.cpp
+  Utils/LogManager.cpp
+  )
+
+set (SRCS
   Common/JitSymbols.cpp
-  Common/NetStream.cpp
   Common/SoftFloat-3e/extF80_add.c
   Common/SoftFloat-3e/extF80_div.c
   Common/SoftFloat-3e/extF80_sub.c
@@ -71,20 +77,23 @@ set (SRCS
   Common/SoftFloat-3e/f32_to_extF80.c
   Common/SoftFloat-3e/s_normSubnormalF32Sig.c
   Common/SoftFloat-3e/s_f32UIToCommonNaN.c
-  Interface/Config/Config.cpp
   Interface/Context/Context.cpp
   Interface/Core/LookupCache.cpp
   Interface/Core/BlockSamplingData.cpp
-  Interface/Core/CompileService.cpp
   Interface/Core/Core.cpp
+  Interface/Core/CPUBackend.cpp
   Interface/Core/CPUID.cpp
   Interface/Core/Frontend.cpp
   Interface/Core/GdbServer.cpp
   Interface/Core/HostFeatures.cpp
+  Interface/Core/ObjectCache/JobHandling.cpp
+  Interface/Core/ObjectCache/NamedRegionObjectHandler.cpp
+  Interface/Core/ObjectCache/ObjectCacheService.cpp
   Interface/Core/OpcodeDispatcher/Crypto.cpp
   Interface/Core/OpcodeDispatcher/Flags.cpp
   Interface/Core/OpcodeDispatcher/Vector.cpp
   Interface/Core/OpcodeDispatcher/X87.cpp
+  Interface/Core/OpcodeDispatcher/X87F64.cpp
   Interface/Core/OpcodeDispatcher.cpp
   Interface/Core/SignalDelegator.cpp
   Interface/Core/X86Tables.cpp
@@ -95,8 +104,7 @@ set (SRCS
   Interface/Core/Dispatcher/Dispatcher.cpp
   Interface/Core/Dispatcher/X86Dispatcher.cpp
   Interface/Core/Dispatcher/Arm64Dispatcher.cpp
-  Interface/Core/Interpreter/InterpreterCore.cpp
-  Interface/Core/Interpreter/InterpreterOps.cpp
+  Interface/Core/Interpreter/InterpreterFallbacks.cpp
   Interface/Core/X86Tables/BaseTables.cpp
   Interface/Core/X86Tables/DDDTables.cpp
   Interface/Core/X86Tables/EVEXTables.cpp
@@ -110,6 +118,8 @@ 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
   Interface/IR/IREmitter.cpp
@@ -125,16 +135,33 @@ 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
   Utils/Allocator/64BitAllocator.cpp
-  Utils/LogManager.cpp
+  Utils/NetStream.cpp
   Utils/Telemetry.cpp
   Utils/Threads.cpp
+  Utils/Profiler.cpp
   )
 
+if (ENABLE_INTERPRETER)
+  list(APPEND SRCS
+    Interface/Core/Interpreter/InterpreterCore.cpp
+    Interface/Core/Interpreter/InterpreterOps.cpp
+    Interface/Core/Interpreter/ALUOps.cpp
+    Interface/Core/Interpreter/AtomicOps.cpp
+    Interface/Core/Interpreter/BranchOps.cpp
+    Interface/Core/Interpreter/ConversionOps.cpp
+    Interface/Core/Interpreter/EncryptionOps.cpp
+    Interface/Core/Interpreter/F80Ops.cpp
+    Interface/Core/Interpreter/FlagOps.cpp
+    Interface/Core/Interpreter/MemoryOps.cpp
+    Interface/Core/Interpreter/MiscOps.cpp
+    Interface/Core/Interpreter/MoveOps.cpp
+    Interface/Core/Interpreter/VectorOps.cpp)
+endif()
+
 if(_M_ARM_64)
   list(APPEND SRCS
     Interface/Core/ArchHelpers/Arm64.cpp)
@@ -150,6 +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
@@ -162,7 +198,9 @@ if (ENABLE_JIT_X86_64)
     Interface/Core/JIT/x86_64/MemoryOps.cpp
     Interface/Core/JIT/x86_64/MiscOps.cpp
     Interface/Core/JIT/x86_64/MoveOps.cpp
-    Interface/Core/JIT/x86_64/VectorOps.cpp)
+    Interface/Core/JIT/x86_64/VectorOps.cpp
+    Interface/Core/JIT/x86_64/x64Relocations.cpp
+    )
   list(APPEND DEFINES -DJIT_X86_64)
 endif()
 
@@ -179,14 +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()
 
-if (ENABLE_JITSYMBOLS)
-  list(APPEND DEFINES -DENABLE_JITSYMBOLS=1)
-endif()
-
-set (LIBS vixl dl fmt::fmt xxhash tiny-json)
+set (LIBS fmt::fmt vixl dl xxhash tiny-json FEXHeaderUtils)
 if (ENABLE_JEMALLOC)
   list (APPEND LIBS FEX_jemalloc)
 endif()
@@ -201,13 +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}"
   )
@@ -221,7 +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}" "${OUTPUT_IR_DOC}"
   )
@@ -240,21 +273,26 @@ set(OUTPUT_CONFIG_NAME "${OUTPUT_CONFIG_FOLDER}/ConfigValues.inl")
 set(OUTPUT_CONFIG_OPTION_NAME "${OUTPUT_CONFIG_FOLDER}/ConfigOptions.inl")
 set(INPUT_CONFIG_NAME "${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}"
   )
 
+add_custom_command(
+  OUTPUT "${OUTPUT_MAN_NAME_COMPRESS}"
+  DEPENDS "${OUTPUT_MAN_NAME}"
+  COMMAND "gzip" "-kf9n" "${OUTPUT_MAN_NAME}"
+  )
+
 set_source_files_properties(${OUTPUT_CONFIG_NAME} PROPERTIES
   GENERATED TRUE)
 set_source_files_properties(${OUTPUT_CONFIG_OPTION_NAME} PROPERTIES
@@ -262,32 +300,28 @@ set_source_files_properties(${OUTPUT_CONFIG_OPTION_NAME} PROPERTIES
 
 set_source_files_properties(${OUTPUT_MAN_NAME} PROPERTIES
   GENERATED TRUE)
+set_source_files_properties(${OUTPUT_MAN_NAME_COMPRESS} PROPERTIES
+  GENERATED TRUE)
 
 # Create the target
 add_custom_target(CONFIG_INC
   DEPENDS "${OUTPUT_CONFIG_NAME}"
   DEPENDS "${OUTPUT_CONFIG_OPTION_NAME}"
-  DEPENDS "${OUTPUT_MAN_NAME}")
+  DEPENDS "${OUTPUT_MAN_NAME}"
+  DEPENDS "${OUTPUT_MAN_NAME_COMPRESS}")
 
-# Install the man page
-install(FILES ${OUTPUT_MAN_NAME} DESTINATION ${MAN_DIR}/man1)
+# Install the compressed man page
+install(FILES ${OUTPUT_MAN_NAME_COMPRESS} DESTINATION ${MAN_DIR}/man1)
 
 # Add in diagnostic colours if the option is available.
 # Ninja code generator will kill colours if this isn't here
 check_cxx_compiler_flag(-fdiagnostics-color=always GCC_COLOR)
 check_cxx_compiler_flag(-fcolor-diagnostics CLANG_COLOR)
 
-function(AddObject Name Type)
-  add_library(${Name} ${Type} ${SRCS})
-  add_dependencies(${Name} IR_INC)
-  add_dependencies(${Name} CONFIG_INC)
-
-  target_link_libraries(${Name} ${LIBS})
-  set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
+function(AddDefaultOptionsToTarget Name)
   set_target_properties(${Name} PROPERTIES C_VISIBILITY_PRESET hidden)
   set_target_properties(${Name} PROPERTIES CXX_VISIBILITY_PRESET hidden)
   set_target_properties(${Name} PROPERTIES VISIBILITY_INLINES_HIDDEN TRUE)
-
   target_include_directories(${Name} PUBLIC "${CMAKE_CURRENT_BINARY_DIR}")
 
   target_include_directories(${Name} PRIVATE IncludePrivate/)
@@ -297,6 +331,7 @@ function(AddObject Name Type)
   target_include_directories(${Name} PUBLIC "${CMAKE_BINARY_DIR}/include/")
 
   target_compile_definitions(${Name} PRIVATE ${DEFINES})
+  add_dependencies(${Name} CONFIG_INC)
 
   target_compile_options(${Name}
     PRIVATE
@@ -307,6 +342,7 @@ function(AddObject Name Type)
 
     -Wno-trigraphs
     -ffunction-sections
+    -fwrapv
   )
 
   if (GCC_COLOR)
@@ -319,19 +355,6 @@ function(AddObject Name Type)
       PRIVATE
       "-fcolor-diagnostics")
   endif()
-endfunction()
-
-function(AddLibrary Name Type)
-  add_library(${Name} ${Type} $<TARGET_OBJECTS:${PROJECT_NAME}_object>)
-  target_link_libraries(${Name} ${LIBS})
-  set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
-  set_target_properties(${Name} PROPERTIES C_VISIBILITY_PRESET hidden)
-  set_target_properties(${Name} PROPERTIES CXX_VISIBILITY_PRESET hidden)
-  set_target_properties(${Name} PROPERTIES VISIBILITY_INLINES_HIDDEN TRUE)
-
-  target_include_directories(${Name} PUBLIC "${CMAKE_CURRENT_BINARY_DIR}")
-  target_include_directories(${Name} PUBLIC "${PROJECT_SOURCE_DIR}/include/")
-  target_include_directories(${Name} PUBLIC "${CMAKE_BINARY_DIR}/include/")
 
   if (CMAKE_BUILD_TYPE MATCHES "RELEASE")
     target_link_options(${Name}
@@ -343,6 +366,29 @@ function(AddLibrary Name Type)
   endif()
 endfunction()
 
+# Build FEXCore_Config static library
+add_library(FEXCore_Base STATIC ${FEXCORE_BASE_SRCS})
+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(${Name} FEXCore_Base)
+  AddDefaultOptionsToTarget(${Name})
+
+  set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
+endfunction()
+
+function(AddLibrary Name Type)
+  add_library(${Name} ${Type} $<TARGET_OBJECTS:${PROJECT_NAME}_object>)
+  target_link_libraries(${Name} FEXCore_Base)
+  set_target_properties(${Name} PROPERTIES OUTPUT_NAME FEXCore)
+
+  AddDefaultOptionsToTarget(${Name})
+endfunction()
+
 AddObject(${PROJECT_NAME}_object OBJECT)
 AddLibrary(${PROJECT_NAME} STATIC)
 AddLibrary(${PROJECT_NAME}_shared SHARED)

External/FEXCore/Source/Common/BitSet.h

@@ -1,13 +1,16 @@
 #pragma once
-#include "Common/MathUtils.h"
+
 #include <FEXCore/Utils/Allocator.h>
+#include <FEXCore/Utils/MathUtils.h>
 #include <FEXCore/Utils/LogManager.h>
 
+#include <cstdint>
+#include <cstdlib>
 #include <cstring>
-#include <stdint.h>
-#include <stdlib.h>
 #include <type_traits>
 
+namespace FEXCore {
+
 template<typename T>
 struct BitSet final {
   using ElementType = T;
@@ -17,12 +20,12 @@ struct BitSet final {
   ElementType *Memory;
   void Allocate(size_t Elements) {
     size_t AllocateSize = AlignUp(Elements, MinimumSizeBits) / MinimumSize;
-    LOGMAN_THROW_A((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((AllocateSize * MinimumSize) >= Elements, "Fail");
+    LOGMAN_THROW_AA_FMT((AllocateSize * MinimumSize) >= Elements, "Fail");
     Memory = static_cast<ElementType*>(FEXCore::Allocator::realloc(Memory, AllocateSize));
   }
   void Free() {
@@ -61,8 +64,8 @@ struct BitSetView final {
   ElementType *Memory;
 
   void GetView(BitSet<T> &Set, uint64_t ElementOffset) {
-    LOGMAN_THROW_A((ElementOffset % MinimumSize) == 0,
-        "Bitset view offset needs to be aligned to size of backing element");
+    LOGMAN_THROW_AA_FMT((ElementOffset % MinimumSize) == 0,
+                       "Bitset view offset needs to be aligned to size of backing element");
     Memory = &Set.Memory[ElementOffset / MinimumSizeBits];
   }
 
@@ -87,11 +90,12 @@ struct BitSetView final {
   bool operator[](T Element) {
     return Get(Element);
   }
-
 };
 
 static_assert(sizeof(BitSet<uint32_t>) == sizeof(uintptr_t), "Needs to just be a pointer");
-static_assert(std::is_trivially_copyable<BitSet<uint32_t>>::value, "Needs to trivially copyable");
+static_assert(std::is_trivially_copyable_v<BitSet<uint32_t>>, "Needs to trivially copyable");
 
 static_assert(sizeof(BitSetView<uint32_t>) == sizeof(uintptr_t), "Needs to just be a pointer");
-static_assert(std::is_trivially_copyable<BitSetView<uint32_t>>::value, "Needs to trivially copyable");
+static_assert(std::is_trivially_copyable_v<BitSetView<uint32_t>>, "Needs to trivially copyable");
+
+} // namespace FEXCore

External/FEXCore/Source/Common/JitSymbols.cpp

@@ -1,45 +1,64 @@
 #include "Common/JitSymbols.h"
 
 #include <string>
-#include <sstream>
 #include <unistd.h>
 
-namespace FEXCore {
-  JITSymbols::JITSymbols() {
-    std::stringstream PerfMap;
-    PerfMap << "/tmp/perf-" << getpid() << ".map";
+#include <fmt/format.h>
 
-    fp = fopen(PerfMap.str().c_str(), "wb");
+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"));
     if (fp) {
       // Disable buffering on this file
-      setvbuf(fp, nullptr, _IONBF, 0);
+      setvbuf(fp.get(), nullptr, _IONBF, 0);
     }
   }
 
-  JITSymbols::~JITSymbols() {
-    if (fp) {
-      fclose(fp);
-    }
-  }
-
-  void JITSymbols::Register(void *HostAddr, uint64_t GuestAddr, uint32_t CodeSize) {
+  void JITSymbols::Register(const void *HostAddr, uint64_t GuestAddr, uint32_t CodeSize) {
     if (!fp) return;
 
     // Linux perf format is very straightforward
     // `<HostPtr> <Size> <Name>\n`
-    std::stringstream String;
-    String << std::hex << HostAddr << " " << CodeSize << " " << "JIT_0x" << GuestAddr << "_" << HostAddr << std::endl;
-    fwrite(String.str().c_str(), 1, String.str().size(), fp);
+    fmt::print(fp.get(), "{} {:x} JIT_0x{:x}_{}\n", HostAddr, CodeSize, GuestAddr, HostAddr);
   }
 
-  void JITSymbols::Register(void *HostAddr, uint32_t CodeSize, std::string const &Name) {
+  void JITSymbols::Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name) {
     if (!fp) return;
 
     // Linux perf format is very straightforward
     // `<HostPtr> <Size> <Name>\n`
-    std::stringstream String;
-    String << std::hex << HostAddr << " " << CodeSize << " " << Name << "_" << HostAddr << std::endl;
-    fwrite(String.str().c_str(), 1, String.str().size(), fp);
+    fmt::print(fp.get(), "{} {:x} {}_{}\n", HostAddr, CodeSize, Name, HostAddr);
   }
-}
 
+  void JITSymbols::Register(const void *HostAddr, uint32_t CodeSize, std::string_view Name, uintptr_t Offset) {
+    if (!fp) return;
+
+    // Linux perf format is very straightforward
+    // `<HostPtr> <Size> <Name>\n`
+    fmt::print(fp.get(), "{} {:x} {}+0x{:x} ({})\n", HostAddr, CodeSize, Name, Offset, HostAddr);
+  }
+
+  void JITSymbols::RegisterNamedRegion(const void *HostAddr, uint32_t CodeSize, std::string_view Name) {
+    if (!fp) return;
+
+    // Linux perf format is very straightforward
+    // `<HostPtr> <Size> <Name>\n`
+    fmt::print(fp.get(), "{} {:x} {}\n", HostAddr, CodeSize, Name);
+  }
+
+  void JITSymbols::RegisterJITSpace(const void *HostAddr, uint32_t CodeSize) {
+    if (!fp) return;
+
+    // Linux perf format is very straightforward
+    // `<HostPtr> <Size> <Name>\n`
+    fmt::print(fp.get(), "{} {:x} FEXJIT\n", HostAddr, CodeSize);
+  }
+
+} // namespace FEXCore

External/FEXCore/Source/Common/JitSymbols.h

@@ -1,17 +1,26 @@
 #pragma once
+
 #include <cstdint>
 #include <cstdio>
-#include <string>
+#include <memory>
+#include <string_view>
 
 namespace FEXCore {
 class JITSymbols final {
 public:
   JITSymbols();
   ~JITSymbols();
-  void Register(void *HostAddr, uint64_t GuestAddr, uint32_t CodeSize);
-  void Register(void *HostAddr, uint32_t CodeSize, std::string const &Name);
+
+  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);
+  void RegisterNamedRegion(const void *HostAddr, uint32_t CodeSize, std::string_view Name);
+  void RegisterJITSpace(const void *HostAddr, uint32_t CodeSize);
 
 private:
-  FILE* fp{};
+  using FILEPtr = std::unique_ptr<FILE, decltype(&std::fclose)>;
+
+  FILEPtr fp;
 };
 }

External/FEXCore/Source/Common/MathUtils.h

@@ -1,13 +0,0 @@
-#pragma once
-
-#include <stdint.h>
-
-static inline uint64_t AlignUp(uint64_t value, uint64_t size) {
-  return value + (size - value % size) % size;
-};
-
-static inline uint64_t AlignDown(uint64_t value, uint64_t size) {
-  return value - value % size;
-};
-
-

External/FEXCore/Source/Common/NetStream.h

@@ -1,42 +0,0 @@
-#pragma once
-
-#include <array>
-#include <iostream>
-#include <iterator>
-#include <string.h>
-
-class NetStream : public std::iostream {
-public:
-    NetStream(int socketfd) : std::iostream(new NetBuf(socketfd)) {}
-    virtual ~NetStream();
-
-private:
-    class NetBuf : public std::streambuf {
-
-    public:
-        NetBuf(int socketfd) {
-            socket = socketfd;
-            reset_output_buffer();
-        }
-        virtual ~NetBuf();
-
-    protected:
-        virtual std::streamsize xsputn(const char* buffer, std::streamsize size);
-
-        virtual std::streambuf::int_type underflow();
-        virtual std::streambuf::int_type overflow(std::streambuf::int_type ch);
-        virtual int sync();
-
-    private:
-        void reset_output_buffer() {
-            // we always leave room for one extra char
-            setp(std::begin(output_buffer), std::end(output_buffer) -1);
-        }
-
-        int flushBuffer(const char *buffer, size_t size);
-
-        int socket;
-        std::array<char, 1400> output_buffer;
-        std::array<char, 1500> input_buffer; // enough for a typical packet
-    };
-};

External/FEXCore/Source/Common/Paths.cpp

@@ -72,7 +72,7 @@ namespace FEXCore::Paths {
     // Ensure the folder structure is created for our Data
     if (!std::filesystem::exists(*EntryCache, ec) &&
         !std::filesystem::create_directories(*EntryCache, ec)) {
-      LogMan::Msg::D("Couldn't create EntryCache directory: '%s'", EntryCache->c_str());
+      LogMan::Msg::DFmt("Couldn't create EntryCache directory: '{}'", *EntryCache);
     }
   }
 

External/FEXCore/Source/Common/SoftFloat.h

@@ -57,51 +57,183 @@ struct X80SoftFloat {
 
   // Ops
   static X80SoftFloat FADD(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    faddp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_add(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FSUB(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fsubp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_sub(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FMUL(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fmulp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_mul(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FDIV(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fdivp;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_div(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FREM(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
-    X80SoftFloat Rem = extF80_rem(lhs, rhs);
-    if (SignBit(Rem)) {
-      Rem = extF80_add(Rem, rhs);
-    }
-    else {
-      Rem.Sign = SignBit(lhs);
-    }
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fprem;
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
 
-    return Rem;
+    return Result;
+#else
+    return extF80_rem(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FREM1(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fprem1;
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     return extF80_rem(lhs, rhs);
+#endif
   }
 
   static X80SoftFloat FRNDINT(X80SoftFloat const &lhs) {
     return extF80_roundToInt(lhs, softfloat_roundingMode, false);
   }
 
+  static X80SoftFloat FRNDINT(X80SoftFloat const &lhs, uint_fast8_t RoundMode) {
+    return extF80_roundToInt(lhs, RoundMode, false);
+  }
+
   static X80SoftFloat FXTRACT_SIG(X80SoftFloat const &lhs) {
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fxtract;
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     X80SoftFloat Tmp = lhs;
     Tmp.Exponent = 0x3FFF;
     Tmp.Sign = lhs.Sign;
     return Tmp;
+#endif
   }
 
   static X80SoftFloat FXTRACT_EXP(X80SoftFloat const &lhs) {
+#if defined(DEBUG_X86_FLOAT)
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fxtract;
+    ffreep %%st(0);
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     int32_t TrueExp = lhs.Exponent - ExponentBias;
     return i32_to_extF80(TrueExp);
+#endif
   }
 
   static void FCMP(X80SoftFloat const &lhs, X80SoftFloat const &rhs, bool *eq, bool *lt, bool *nan) {
@@ -111,62 +243,190 @@ struct X80SoftFloat {
   }
 
   static X80SoftFloat FSCALE(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
-    WARN_ONCE("x87: Application used FSCALE which may have accuracy problems");
-    X80SoftFloat Int = FRNDINT(rhs);
+    WARN_ONCE_FMT("x87: Application used FSCALE which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st1
+    fldt %[lhs]; # st0
+    fscale; # st0 = st0 * 2^(rdint(st1))
+    fstpt %[result];
+    ffreep %%st(0);
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
+    X80SoftFloat Int = FRNDINT(rhs, softfloat_round_minMag);
     BIGFLOAT Src2_d = Int;
     Src2_d = exp2l(Src2_d);
     X80SoftFloat Src2_X80 = Src2_d;
     X80SoftFloat Result = extF80_mul(lhs, Src2_X80);
     return Result;
+#endif
   }
 
   static X80SoftFloat F2XM1(X80SoftFloat const &lhs) {
-    WARN_ONCE("x87: Application used F2XM1 which may have accuracy problems");
+    WARN_ONCE_FMT("x87: Application used F2XM1 which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    f2xm1; # st0 = 2^st(0) - 1
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     BIGFLOAT Src1_d = lhs;
     BIGFLOAT Result = exp2l(Src1_d);
     Result -= 1.0;
     return Result;
+#endif
   }
 
   static X80SoftFloat FYL2X(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
-    WARN_ONCE("x87: Application used FYL2X which may have accuracy problems");
+    WARN_ONCE_FMT("x87: Application used FYL2X which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[rhs]; # st(1)
+    fldt %[lhs]; # st(0)
+    fyl2x; # st(1) * log2l(st(0))
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     BIGFLOAT Src1_d = lhs;
     BIGFLOAT Src2_d = rhs;
     BIGFLOAT Tmp = Src2_d * log2l(Src1_d);
     return Tmp;
+#endif
   }
 
   static X80SoftFloat FATAN(X80SoftFloat const &lhs, X80SoftFloat const &rhs) {
-    WARN_ONCE("x87: Application used FATAN which may have accuracy problems");
+    WARN_ONCE_FMT("x87: Application used FATAN which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs];
+    fldt %[rhs];
+    fpatan;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    , [rhs] "m" (rhs)
+    : "st", "st(1)");
+
+    return Result;
+#else
     BIGFLOAT Src1_d = lhs;
     BIGFLOAT Src2_d = rhs;
     BIGFLOAT Tmp = atan2l(Src1_d, Src2_d);
     return Tmp;
+#endif
   }
 
   static X80SoftFloat FTAN(X80SoftFloat const &lhs) {
-    WARN_ONCE("x87: Application used FTAN which may have accuracy problems");
+    WARN_ONCE_FMT("x87: Application used FTAN which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fptan;
+    ffreep %%st(0);
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     BIGFLOAT Src_d = lhs;
     Src_d = tanl(Src_d);
     return Src_d;
+#endif
   }
 
   static X80SoftFloat FSIN(X80SoftFloat const &lhs) {
-    WARN_ONCE("x87: Application used FSIN which may have accuracy problems");
+    WARN_ONCE_FMT("x87: Application used FSIN which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fsin;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     BIGFLOAT Src_d = lhs;
     Src_d = sinl(Src_d);
     return Src_d;
+#endif
   }
 
   static X80SoftFloat FCOS(X80SoftFloat const &lhs) {
-    WARN_ONCE("x87: Application used FCOS which may have accuracy problems");
+    WARN_ONCE_FMT("x87: Application used FCOS which may have accuracy problems");
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fcos;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     BIGFLOAT Src_d = lhs;
     Src_d = cosl(Src_d);
     return Src_d;
+#endif
   }
 
   static X80SoftFloat FSQRT(X80SoftFloat const &lhs) {
+#ifdef DEBUG_X86_FLOAT
+    BIGFLOAT Result;
+    asm (R"(
+    fninit;
+    fldt %[lhs]; # st0
+    fsqrt;
+    fstpt %[result];
+    )"
+    : [result] "=m" (Result)
+    : [lhs] "m" (lhs)
+    : "st");
+
+    return Result;
+#else
     return extF80_sqrt(lhs);
+#endif
   }
 
   operator float() const {

External/FEXCore/Source/Common/StringUtils.h

@@ -0,0 +1,29 @@
+#pragma once
+#include <string>
+
+namespace FEXCore::StringUtils {
+  // Trim the left side of the string of whitespace and new lines
+  [[maybe_unused]] static std::string LeftTrim(std::string String, std::string TrimTokens = " \t\n\r") {
+    size_t pos = std::string::npos;
+    if ((pos = String.find_first_not_of(TrimTokens)) != std::string::npos) {
+      String.erase(0, pos);
+    }
+
+    return String;
+  }
+
+  // Trim the right side of the string of whitespace and new lines
+  [[maybe_unused]] static std::string RightTrim(std::string String, std::string TrimTokens = " \t\n\r") {
+    size_t pos = std::string::npos;
+    if ((pos = String.find_last_not_of(TrimTokens)) != std::string::npos) {
+      String.erase(String.begin() + pos + 1, String.end());
+    }
+
+    return String;
+  }
+
+  // Trim both the left and right of the string of whitespace and new lines
+  [[maybe_unused]] static std::string Trim(std::string String, std::string TrimTokens = " \t\n\r") {
+    return RightTrim(LeftTrim(String, TrimTokens), TrimTokens);
+  }
+}

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

@@ -1,5 +1,7 @@
 #include "Common/StringConv.h"
+#include "Common/StringUtils.h"
 #include "Common/Paths.h"
+#include "Utils/FileLoading.h"
 
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Utils/LogManager.h>
@@ -40,45 +42,6 @@ namespace DefaultValues {
 #include <FEXCore/Config/ConfigValues.inl>
 }
 
-static bool LoadConfigFile(std::vector<char> &Data, const std::string &Config) {
-  std::fstream ConfigFile;
-  ConfigFile.open(Config, std::ios::in);
-
-  if (!ConfigFile.is_open()) {
-    return false;
-  }
-
-  if (!ConfigFile.seekg(0, std::fstream::end)) {
-    LogMan::Msg::D("Couldn't load configuration file: Seek end");
-    return false;
-  }
-
-  auto FileSize = ConfigFile.tellg();
-  if (ConfigFile.fail()) {
-    LogMan::Msg::D("Couldn't load configuration file: tellg");
-    return false;
-  }
-
-  if (!ConfigFile.seekg(0, std::fstream::beg)) {
-    LogMan::Msg::D("Couldn't load configuration file: Seek beginning");
-    return false;
-  }
-
-  if (FileSize > 0) {
-    Data.resize(FileSize);
-    if (!ConfigFile.read(&Data.at(0), FileSize)) {
-      // Probably means permissions aren't set. Just early exit
-      return false;
-    }
-    ConfigFile.close();
-  }
-  else {
-    return false;
-  }
-
-  return true;
-}
-
 namespace JSON {
   struct JsonAllocator {
     jsonPool_t PoolObject;
@@ -99,7 +62,7 @@ namespace JSON {
 
   static void LoadJSonConfig(const std::string &Config, std::function<void(const char *Name, const char *ConfigSring)> Func) {
     std::vector<char> Data;
-    if (!LoadConfigFile(Data, Config)) {
+    if (!FEXCore::FileLoading::LoadFile(Data, Config)) {
       return;
     }
 
@@ -112,14 +75,14 @@ namespace JSON {
 
     json_t const *json = json_createWithPool(&Data.at(0), &Pool.PoolObject);
     if (!json) {
-      LogMan::Msg::E("Couldn't create json");
+      LogMan::Msg::EFmt("Couldn't create json");
       return;
     }
 
     json_t const* ConfigList = json_getProperty(json, "Config");
 
     if (!ConfigList) {
-      LogMan::Msg::E("Couldn't get config list");
+      // This is a non-error if the configuration file exists but no Config section
       return;
     }
 
@@ -130,12 +93,12 @@ namespace JSON {
       const char* ConfigString = json_getValue(ConfigItem);
 
       if (!ConfigName) {
-        LogMan::Msg::E("Couldn't get config name");
+        LogMan::Msg::EFmt("Couldn't get config name");
         return;
       }
 
       if (!ConfigString) {
-        LogMan::Msg::E("Couldn't get ConfigString for '%s'", ConfigName);
+        LogMan::Msg::EFmt("Couldn't get ConfigString for '{}'", ConfigName);
         return;
       }
 
@@ -191,15 +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;
   }
@@ -211,7 +179,7 @@ namespace JSON {
     if (!Global &&
         !std::filesystem::exists(ConfigFile, ec) &&
         !std::filesystem::create_directories(ConfigFile, ec)) {
-      LogMan::Msg::D("Couldn't create config directory: '%s'", ConfigFile.c_str());
+      LogMan::Msg::DFmt("Couldn't create config directory: '{}'", ConfigFile);
       // Let's go local in this case
       return "./" + Filename + ".json";
     }
@@ -243,9 +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,
@@ -408,38 +379,31 @@ namespace JSON {
     return {};
   }
 
-  std::string ltrim(std::string String) {
-    size_t pos = std::string::npos;
-    if ((pos = String.find_first_not_of(" \t\n\r")) != std::string::npos) {
-      String.erase(0, pos);
+
+  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 String;
+    return {};
   }
 
-  std::string rtrim(std::string String) {
-    size_t pos = std::string::npos;
-    if ((pos = String.find_last_not_of(" \t\n\r")) != std::string::npos) {
-      String.erase(String.begin() + pos + 1, String.end());
-    }
-
-    return String;
-  }
-
-  std::string trim(std::string String) {
-    return rtrim(ltrim(String));
-  }
-
-
   std::string FindContainerPrefix() {
     // 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 (LoadConfigFile(Manager, ContainerManager)) {
+      if (FEXCore::FileLoading::LoadFile(Manager, ContainerManager)) {
         // Trim the whitespace, may contain a newline
         std::string ManagerStr = Manager.data();
-        ManagerStr = trim(ManagerStr);
+        ManagerStr = FEXCore::StringUtils::Trim(ManagerStr);
         if (strncmp(ManagerStr.data(), "pressure-vessel", Manager.size()) == 0) {
           // We are running inside of pressure vessel
           // Our $CMAKE_INSTALL_PREFIX paths are now inside of /run/host/$CMAKE_INSTALL_PREFIX
@@ -454,7 +418,9 @@ namespace JSON {
     Meta->Load();
 
     // Do configuration option fix ups after everything is reloaded
-    if (FEXCore::Config::Exists(FEXCore::Config::CONFIG_THREADS)) {
+    {
+      // Always fix up the number of threads and create the configuration
+      // Otherwise the application could receive zero as the number of threads
       FEX_CONFIG_OPT(Cores, THREADS);
       if (Cores == 0) {
         // When the number of emulated CPU cores is zero then auto detect
@@ -462,6 +428,35 @@ namespace JSON {
       }
     }
 
+    if (FEXCore::Config::Exists(FEXCore::Config::CONFIG_CORE)) {
+      // Sanitize Core option
+      FEX_CONFIG_OPT(Core, CORE);
+#if (_M_X86_64)
+      constexpr uint32_t MaxCoreNumber = 2;
+#else
+      constexpr uint32_t MaxCoreNumber = 1;
+#endif
+#ifdef INTERPRETER_ENABLED
+      constexpr uint32_t MinCoreNumber = 0;
+#else
+      constexpr uint32_t MinCoreNumber = 1;
+#endif
+      if (Core > MaxCoreNumber || Core < MinCoreNumber) {
+        // Sanitize the core option by setting the core to the JIT if invalid
+        FEXCore::Config::EraseSet(FEXCore::Config::CONFIG_CORE, std::to_string(FEXCore::Config::CONFIG_IRJIT));
+      }
+    }
+
+    if (FEXCore::Config::Exists(FEXCore::Config::CONFIG_CACHEOBJECTCODECOMPILATION)) {
+      FEX_CONFIG_OPT(CacheObjectCodeCompilation, CACHEOBJECTCODECOMPILATION);
+      FEX_CONFIG_OPT(Core, CORE);
+
+      if (CacheObjectCodeCompilation() && Core() == FEXCore::Config::CONFIG_INTERPRETER) {
+        // If running the interpreter then disable cache code compilation
+        FEXCore::Config::Erase(FEXCore::Config::CONFIG_CACHEOBJECTCODECOMPILATION);
+      }
+    }
+
     std::string ContainerPrefix { FindContainerPrefix() };
     auto ExpandPathIfExists = [&ContainerPrefix](FEXCore::Config::ConfigOption Config, std::string PathName) {
       auto NewPath = ExpandPath(ContainerPrefix, PathName);
@@ -539,7 +534,7 @@ namespace JSON {
     return Meta->Get(Option);
   }
 
-  void Set(ConfigOption Option, std::string Data) {
+  void Set(ConfigOption Option, std::string_view Data) {
     Meta->Set(Option, Data);
   }
 
@@ -547,7 +542,7 @@ namespace JSON {
     Meta->Erase(Option);
   }
 
-  void EraseSet(ConfigOption Option, std::string Data) {
+  void EraseSet(ConfigOption Option, std::string_view Data) {
     Meta->EraseSet(Option, Data);
   }
 
@@ -626,7 +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;
 
@@ -636,7 +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:
@@ -672,9 +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)
@@ -688,8 +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
@@ -716,9 +713,13 @@ namespace JSON {
       if (std::string::npos == pos)
         continue;
 
-      std::string_view Ident = Var.substr(0,pos);
-      std::string_view Value = Var.substr(pos+1);
-      EnvMap[Ident]=Value;
+      std::string_view Key = Var.substr(0,pos);
+      std::string_view Value {Var.substr(pos+1)};
+
+#define ENVLOADER
+#include <FEXCore/Config/ConfigOptions.inl>
+
+      EnvMap[Key]=Value;
     }
 
     std::function GetVar = [=](const std::string_view id)  -> std::optional<std::string_view> {
@@ -744,17 +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": {
@@ -32,12 +33,30 @@
       },
       "Threads": {
         "Type": "uint32",
-        "Default": "1",
+        "Default": "0",
         "ShortArg": "T",
         "Desc": [
           "Number of physical hardware threads to tell the process we have.",
           "0 will auto detect."
         ]
+      },
+      "CacheObjectCodeCompilation": {
+        "Type": "uint32",
+        "Default": "FEXCore::Config::ConfigObjectCodeHandler::CONFIG_NONE",
+        "TextDefault": "none",
+        "Choices": [ "none", "read", "readwrite" ],
+        "ArgumentHandler": "CacheObjectCodeHandler",
+        "Desc": [
+          "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": {
@@ -72,6 +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": "",
@@ -104,6 +137,13 @@
           "This can be useful for setting environment variables that thunks can pick up.",
           "Typically isn't necessary since the guest libc isn't thunked. But is possible."
         ]
+      },
+      "AdditionalArguments": {
+        "Type": "strarray",
+        "Default": "",
+        "Desc": [
+          "Allows the user to pass additional arguments to the application"
+        ]
       }
     },
     "Debug": {
@@ -163,8 +203,44 @@
           "Potentially useful for debugging memory problems",
           "32-bit allocator is always used if your host kernel is older than 4.17"
         ]
+      },
+      "GlobalJITNaming": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Uses JITSymbols to name all JIT state as one symbol",
+          "Useful for querying how much time is spent inside of the JIT",
+          "Profiling tools will show JIT time as FEXJIT"
+        ]
+      },
+      "LibraryJITNaming": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Uses JITSymbols to name JIT symbols grouped by library",
+          "Useful for querying how much time is spent in each guest library",
+          "Can be used to help guide thunk generation"
+        ]
+      },
+      "BlockJITNaming": {
+        "Type": "bool",
+        "Default": "false",
+        "Desc": [
+          "Uses JITSymbols to name JIT symbols",
+          "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": {
       "SilentLog": {
@@ -177,25 +253,26 @@
       },
       "OutputLog": {
         "Type": "str",
-        "Default": "stderr",
+        "Default": "server",
         "ShortArg": "o",
         "Desc": [
           "File to write FEX output to.",
-          "[stdout, stderr, <Filename>]"
+          "[stdout, stderr, server, <Filename>]"
         ]
       }
     },
     "Hacks": {
       "SMCChecks": {
         "Type": "uint8",
-        "Default": "FEXCore::Config::CONFIG_SMC_MMAN",
-        "TextDefault": "mman",
+        "Default": "FEXCore::Config::CONFIG_SMC_MTRACK",
+        "TextDefault": "mtrack",
         "ArgumentHandler": "SMCCheckHandler",
         "Desc": [
           "Checks code for modification before execution.",
           "\tnone: No checks",
-          "\tmman: Invalidate on mmap, mprotect, munmap",
-          "\tfull: Validate code before every run (slow)"
+          "\tmtrack: Page tracking based invalidation",
+          "\tfull: Validate code before every run (slow)",
+          "\tmman: Invalidate on mmap, mprotect, munmap (deprecated, use mtrack)"
         ]
       },
       "TSOEnabled": {
@@ -206,6 +283,21 @@
           "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",
+        "Desc": [
+          "Emulates X87 floating point using 64-bit precision. This reduces emulation accuracy and may result in rendering bugs."
+        ]
+      },
       "ABILocalFlags": {
         "Type": "bool",
         "Default": "false",
@@ -239,6 +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": {
@@ -264,6 +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."
+        ]
       }
     }
   },
@@ -281,6 +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

@@ -43,15 +43,15 @@ 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) {
     CTX->CustomExitHandler = std::move(handler);
   }
 
-  ExitHandler GetExitHandler(FEXCore::Context::Context *CTX) {
+  ExitHandler GetExitHandler(const FEXCore::Context::Context *CTX) {
     return CTX->CustomExitHandler;
   }
 
@@ -71,23 +71,23 @@ namespace FEXCore::Context {
     return CTX->RunUntilExit();
   }
 
-  int GetProgramStatus(FEXCore::Context::Context *CTX) {
+  int GetProgramStatus(const FEXCore::Context::Context *CTX) {
     return CTX->GetProgramStatus();
   }
 
-  FEXCore::Context::ExitReason GetExitReason(FEXCore::Context::Context *CTX) {
+  FEXCore::Context::ExitReason GetExitReason(const FEXCore::Context::Context *CTX) {
     return CTX->ParentThread->ExitReason;
   }
 
-  bool IsDone(FEXCore::Context::Context *CTX) {
+  bool IsDone(const FEXCore::Context::Context *CTX) {
     return CTX->IsPaused();
   }
 
-  void GetCPUState(FEXCore::Context::Context *CTX, FEXCore::Core::CPUState *State) {
+  void GetCPUState(const FEXCore::Context::Context *CTX, FEXCore::Core::CPUState *State) {
     memcpy(State, CTX->ParentThread->CurrentFrame, sizeof(FEXCore::Core::CPUState));
   }
 
-  void SetCPUState(FEXCore::Context::Context *CTX, FEXCore::Core::CPUState *State) {
+  void SetCPUState(FEXCore::Context::Context *CTX, const FEXCore::Core::CPUState *State) {
     memcpy(CTX->ParentThread->CurrentFrame, State, sizeof(FEXCore::Core::CPUState));
   }
 
@@ -110,16 +110,20 @@ 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);
   }
 
   void RegisterHostSignalHandler(FEXCore::Context::Context *CTX, int Signal, HostSignalDelegatorFunction Func, bool Required) {
-      CTX->RegisterHostSignalHandler(Signal, Func, Required);
+      CTX->RegisterHostSignalHandler(Signal, std::move(Func), Required);
   }
 
   void RegisterFrontendHostSignalHandler(FEXCore::Context::Context *CTX, int Signal, HostSignalDelegatorFunction Func, bool Required) {
-    CTX->RegisterFrontendHostSignalHandler(Signal, Func, Required);
+    CTX->RegisterFrontendHostSignalHandler(Signal, std::move(Func), Required);
   }
 
   FEXCore::Core::InternalThreadState* CreateThread(FEXCore::Context::Context *CTX, FEXCore::Core::CPUState *NewThreadState, uint64_t ParentTID) {
@@ -149,25 +153,34 @@ namespace FEXCore::Context {
   void CleanupAfterFork(FEXCore::Context::Context *CTX, FEXCore::Core::InternalThreadState *Thread) {
     CTX->CleanupAfterFork(Thread);
   }
-
+  
   void SetSignalDelegator(FEXCore::Context::Context *CTX, FEXCore::SignalDelegator *SignalDelegation) {
     CTX->SignalDelegation = SignalDelegation;
   }
 
   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) {
     return CTX->CPUID.RunFunction(Function, Leaf);
   }
 
-  void SetAOTIRLoader(FEXCore::Context::Context *CTX, std::function<int(const std::string&)> CacheReader) {
-    CTX->AOTIRLoader = CacheReader;
+  FEX_DEFAULT_VISIBILITY FEXCore::CPUID::FunctionResults RunCPUIDFunctionName(FEXCore::Context::Context *CTX, uint32_t Function, uint32_t Leaf, uint32_t CPU) {
+    return CTX->CPUID.RunFunctionName(Function, Leaf, CPU);
   }
 
-  void SetAOTIRWriter(FEXCore::Context::Context *CTX, std::function<std::unique_ptr<std::ostream>(const std::string&)> CacheWriter) {
-    CTX->AOTIRWriter = CacheWriter;
+  void SetAOTIRLoader(FEXCore::Context::Context *CTX, std::function<int(const std::string&)> CacheReader) {
+    CTX->SetAOTIRLoader(CacheReader);
+  }
+
+  void SetAOTIRWriter(FEXCore::Context::Context *CTX, std::function<std::unique_ptr<std::ofstream>(const std::string&)> CacheWriter) {
+    CTX->SetAOTIRWriter(CacheWriter);
+  }
+
+  void SetAOTIRRenamer(FEXCore::Context::Context *CTX, std::function<void(const std::string&)> CacheRenamer) {
+    CTX->SetAOTIRRenamer(CacheRenamer);
   }
 
   void FinalizeAOTIRCache(FEXCore::Context::Context *CTX) {
@@ -178,11 +191,19 @@ namespace FEXCore::Context {
     CTX->WriteFilesWithCode(Writer);
   }
 
-  void AddNamedRegion(FEXCore::Context::Context *CTX, uintptr_t Base, uintptr_t Length, uintptr_t Offset, const std::string& Name) {
-    return CTX->AddNamedRegion(Base, Length, Offset, Name);
+  IR::AOTIRCacheEntry *LoadAOTIRCacheEntry(FEXCore::Context::Context *CTX, const std::string &Name) {
+    return CTX->LoadAOTIRCacheEntry(Name);
   }
-  void RemoveNamedRegion(FEXCore::Context::Context *CTX, uintptr_t Base, uintptr_t Length) {
-    return CTX->RemoveNamedRegion(Base, Length);
+  void UnloadAOTIRCacheEntry(FEXCore::Context::Context *CTX, IR::AOTIRCacheEntry *Entry) {
+    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 {

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

@@ -1,20 +1,23 @@
 #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>
 
-#ifdef ENABLE_JITSYMBOLS
-#include <Common/JITSymbols.h>
-#endif
 
 #include <atomic>
 #include <condition_variable>
@@ -35,13 +38,21 @@ class CodeLoader;
 class ThunkHandler;
 class GdbServer;
 
+namespace CodeSerialize {
+  class CodeObjectSerializeService;
+}
+
 namespace CPU {
   class Arm64JITCore;
   class X86JITCore;
+  class InterpreterCore;
+  class Dispatcher;
 }
 namespace HLE {
 struct SyscallArguments;
 class SyscallHandler;
+class SourcecodeResolver;
+struct SourcecodeMap;
 }
 }
 
@@ -59,38 +70,6 @@ namespace FEXCore::Context {
     MODE_SINGLESTEP = 1,
   };
 
-  struct AOTIRInlineEntry {
-    uint64_t GuestHash;
-    uint64_t GuestLength;
-
-    /* RAData followed by IRData */
-    uint8_t InlineData[0];
-
-    IR::RegisterAllocationData *GetRAData();
-    IR::IRListView *GetIRData();
-  };
-
-  struct AOTIRInlineIndexEntry {
-    uint64_t GuestStart;
-    uint64_t DataOffset;
-  };
-
-  struct AOTIRInlineIndex {
-    uint64_t Count;
-    uint64_t DataBase;
-    AOTIRInlineIndexEntry Entries[0];
-
-    AOTIRInlineEntry *Find(uint64_t GuestStart);
-    AOTIRInlineEntry *GetInlineEntry(uint64_t DataOffset);
-  };
-
-  struct AOTIRCaptureCacheEntry {
-    std::unique_ptr<std::ostream> Stream;
-    std::map<uint64_t, uint64_t> Index;
-
-    void AppendAOTIRCaptureCache(uint64_t GuestRIP, uint64_t Start, uint64_t Length, uint64_t Hash, FEXCore::IR::IRListView *IRList, FEXCore::IR::RegisterAllocationData *RAData);
-  };
-
   struct Context {
     friend class FEXCore::HLE::SyscallHandler;
   #ifdef JIT_ARM64
@@ -100,6 +79,7 @@ namespace FEXCore::Context {
     friend class FEXCore::CPU::X86JITCore;
   #endif
 
+    friend class FEXCore::CPU::InterpreterCore;
     friend class FEXCore::IR::Validation::IRValidation;
 
     struct {
@@ -114,6 +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);
@@ -124,21 +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;
@@ -147,37 +135,17 @@ 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;
 
-    struct AOTIRCacheEntry {
-      AOTIRInlineIndex *Array;
-      void *mapping;
-      size_t size;
-    };
-
-    std::unordered_map<std::string, AOTIRCacheEntry> AOTIRCache;
-    std::function<int(const std::string&)> AOTIRLoader;
-    std::function<std::unique_ptr<std::ostream>(const std::string&)> AOTIRWriter;
-    std::unordered_map<std::string, AOTIRCaptureCacheEntry> AOTIRCaptureCache;
-
-    struct AddrToFileEntry {
-      uint64_t Start;
-      uint64_t Len;
-      uint64_t Offset;
-      std::string fileid;
-      std::string filename;
-      void *CachedFileEntry;
-      bool ContainsCode;
-    };
-
-    std::map<uint64_t, AddrToFileEntry> AddrToFile;
-    std::map<std::string, std::string> FilesWithCode;
-
 #ifdef BLOCKSTATS
     std::unique_ptr<FEXCore::BlockSamplingData> BlockData;
 #endif
@@ -188,7 +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; }
@@ -208,13 +176,33 @@ namespace FEXCore::Context {
     void RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
     void RegisterFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
 
-    static void RemoveCodeEntry(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
-    static void RemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
-      RemoveCodeEntry(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;
@@ -224,21 +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;
@@ -249,24 +235,9 @@ namespace FEXCore::Context {
     // same as CompileBlock, but aborts on failure
     void CompileBlockJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP);
 
-    bool LoadAOTIRCache(int streamfd);
-    void FinalizeAOTIRCache();
-    void WriteFilesWithCode(std::function<void(const std::string& fileid, const std::string& filename)> Writer);
-    /**
-     * @brief Initializes the JIT compilers for the thread
-     *
-     * @param State The internal FEX thread state object
-     * @param CompileThread Is this for the compile service or not?
-     *
-     * InitializeCompiler is called inside of CreateThread, so you likely don't need this
-     * This is exposed because the CompileService needs to initialize compilers while copying data from
-     * the paired InternalThreadState that it is compiling code for
-     */
-    void InitializeCompiler(FEXCore::Core::InternalThreadState* State, bool CompileThread);
-
     // 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
@@ -289,7 +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
      */
@@ -325,18 +296,45 @@ namespace FEXCore::Context {
 
     uint8_t GetGPRSize() const { return Config.Is64BitMode ? 8 : 4; }
 
-    void AddNamedRegion(uintptr_t Base, uintptr_t Size, uintptr_t Offset, const std::string &filename);
-    void RemoveNamedRegion(uintptr_t Base, uintptr_t Size);
+    IR::AOTIRCacheEntry *LoadAOTIRCacheEntry(const std::string &filename);
+    void UnloadAOTIRCacheEntry(IR::AOTIRCacheEntry *Entry);
 
-#if ENABLE_JITSYMBOLS
     FEXCore::JITSymbols Symbols;
-#endif
 
     // Public for threading
     void ExecutionThread(FEXCore::Core::InternalThreadState *Thread);
 
+    void FinalizeAOTIRCache() {
+      IRCaptureCache.FinalizeAOTIRCache();
+    }
+
+    void WriteFilesWithCode(std::function<void(const std::string& fileid, const std::string& filename)> Writer) {
+      IRCaptureCache.WriteFilesWithCode(Writer);
+    }
+
+    void SetAOTIRLoader(std::function<int(const std::string&)> CacheReader) {
+      IRCaptureCache.SetAOTIRLoader(CacheReader);
+    }
+
+    void SetAOTIRWriter(std::function<std::unique_ptr<std::ofstream>(const std::string&)> CacheWriter) {
+      IRCaptureCache.SetAOTIRWriter(CacheWriter);
+    }
+
+    void SetAOTIRRenamer(std::function<void(const std::string&)> CacheRenamer) {
+      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:
     /**
@@ -348,28 +346,36 @@ namespace FEXCore::Context {
      */
     void InitializeThreadData(FEXCore::Core::InternalThreadState *Thread);
 
+    /**
+     * @brief Initializes the JIT compilers for the thread
+     *
+     * @param State The internal FEX thread state object
+     *
+     * InitializeCompiler is called inside of CreateThread, so you likely don't need this
+     */
+    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;
 
-    std::shared_mutex AOTIRCacheLock;
-    std::shared_mutex AOTIRCaptureCacheWriteoutLock;
-    std::atomic<bool> AOTIRCaptureCacheWriteoutFlusing;
-
-    std::queue<std::function<void()>> AOTIRCaptureCacheWriteoutQueue;
-    void AOTIRCaptureCacheWriteoutQueue_Flush();
-    void AOTIRCaptureCacheWriteoutQueue_Append(const std::function<void()> &fn);
+    IR::AOTIRCaptureCache IRCaptureCache;
+    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.h

@@ -12,6 +12,10 @@ namespace FEXCore::ArchHelpers::Arm64 {
   constexpr uint32_t ATOMIC_MEM_MASK = 0x3B200C00;
   constexpr uint32_t ATOMIC_MEM_INST = 0x38200000;
 
+  constexpr uint32_t RCPC2_MASK  = 0x3F'E0'0C'00;
+  constexpr uint32_t LDAPUR_INST = 0x19'40'00'00;
+  constexpr uint32_t STLUR_INST  = 0x19'00'00'00;
+
   constexpr uint32_t LDAXP_MASK = 0xBF'FF'80'00;
   constexpr uint32_t LDAXP_INST = 0x88'7F'80'00;
 
@@ -27,24 +31,36 @@ namespace FEXCore::ArchHelpers::Arm64 {
   constexpr uint32_t CBNZ_MASK = 0x7F'00'00'00;
   constexpr uint32_t CBNZ_INST = 0x35'00'00'00;
 
-  constexpr uint32_t ALU_OP_MASK = 0x7F'00'00'00;
-  constexpr uint32_t ADD_INST    = 0x0B'00'00'00;
-  constexpr uint32_t SUB_INST    = 0x4B'00'00'00;
-  constexpr uint32_t CMP_INST    = 0x6B'00'00'00;
-  constexpr uint32_t AND_INST    = 0x0A'00'00'00;
-  constexpr uint32_t OR_INST     = 0x2A'00'00'00;
-  constexpr uint32_t EOR_INST    = 0x4A'00'00'00;
-  
+  constexpr uint32_t ALU_OP_MASK    = 0x7F'20'00'00;
+  constexpr uint32_t ADD_INST       = 0x0B'00'00'00;
+  constexpr uint32_t SUB_INST       = 0x4B'00'00'00;
+  constexpr uint32_t ADD_SHIFT_INST = 0x0B'20'00'00;
+  constexpr uint32_t SUB_SHIFT_INST = 0x4B'20'00'00;
+  constexpr uint32_t CMP_INST       = 0x6B'00'00'00;
+  constexpr uint32_t CMP_SHIFT_INST = 0x6B'20'00'00;
+  constexpr uint32_t AND_INST       = 0x0A'00'00'00;
+  constexpr uint32_t BIC_INST       = 0x0A'20'00'00;
+  constexpr uint32_t OR_INST        = 0x2A'00'00'00;
+  constexpr uint32_t ORN_INST       = 0x2A'20'00'00;
+  constexpr uint32_t EOR_INST       = 0x4A'00'00'00;
+  constexpr uint32_t EON_INST       = 0x4A'20'00'00;
+
   constexpr uint32_t CCMP_MASK   = 0x7F'E0'0C'10;
   constexpr uint32_t CCMP_INST   = 0x7A'40'00'00;
-  
+
+  constexpr uint32_t CLREX_MASK = 0xFF'FF'F0'FF;
+  constexpr uint32_t CLREX_INST = 0xD5'03'30'5F;
+
   enum ExclusiveAtomicPairType {
     TYPE_SWAP,
     TYPE_ADD,
     TYPE_SUB,
     TYPE_AND,
+    TYPE_BIC,
     TYPE_OR,
+    TYPE_ORN,
     TYPE_EOR,
+    TYPE_EON,
     TYPE_NEG, // This is just a sub with zero. Need to know the differences
   };
 
@@ -65,6 +81,9 @@ namespace FEXCore::ArchHelpers::Arm64 {
   constexpr uint32_t RN_OFFSET = 5;
   constexpr uint32_t RM_OFFSET = 16;
 
+  constexpr uint32_t DMB = 0b1101'0101'0000'0011'0011'0000'1011'1111 |
+    0b1011'0000'0000; // Inner shareable all
+
   inline uint32_t GetRdReg(uint32_t Instr) {
     return (Instr >> RD_OFFSET) & REGISTER_MASK;
   }
@@ -77,12 +96,14 @@ namespace FEXCore::ArchHelpers::Arm64 {
     return (Instr >> RM_OFFSET) & REGISTER_MASK;
   }
 
-  bool HandleAtomicLoad(void *_ucontext, void *_info, uint32_t Instr);
-  bool HandleAtomicStore(void *_ucontext, void *_info, uint32_t Instr);
+  bool HandleAtomicLoad(void *_ucontext, void *_info, uint32_t Instr, int64_t Offset);
+  bool HandleAtomicStore(void *_ucontext, void *_info, uint32_t Instr, int64_t Offset);
   bool HandleAtomicLoad128(void *_ucontext, void *_info, uint32_t Instr);
   uint64_t HandleAtomicLoadstoreExclusive(void *_ucontext, void *_info);
   bool HandleCASPAL(void *_ucontext, void *_info, uint32_t Instr);
   uint64_t HandleCASPAL_ARMv8(void *_ucontext, void *_info, uint32_t Instr);
+  bool HandleAtomicVectorStore(void *_ucontext, void *_info, uint32_t Instr);
   bool HandleCASAL(void *_ucontext, void *_info, uint32_t Instr);
   bool HandleAtomicMemOp(void *_ucontext, void *_info, uint32_t Instr);
+  [[nodiscard]] bool HandleSIGBUS(bool ParanoidTSO, int Signal, void *info, void *ucontext);
 }

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

@@ -1,66 +1,109 @@
 #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"
 
-#include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Core/CoreState.h>
+#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(size_t size) : vixl::aarch64::Assembler(size, vixl::aarch64::PositionDependentCode) {
+Arm64Emitter::Arm64Emitter(FEXCore::Context::Context *ctx, size_t size)
+  : 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();
-  SupportsAtomics = Features.Has(vixl::CPUFeatures::Feature::kAtomics);
-  // RCPC is bugged on Snapdragon 865
-  // Causes glibc cond16 test to immediately throw assert
-  // __pthread_mutex_cond_lock: Assertion `mutex->__data.__owner == 0'
-  SupportsRCPC = false; //Features.Has(vixl::CPUFeatures::Feature::kRCpc);
-
-  if (SupportsAtomics) {
-    // Hypervisor can hide this on the c630?
-    Features.Combine(vixl::CPUFeatures::Feature::kLORegions);
-  }
-
-  SetCPUFeatures(Features);
-
-  if (!SupportsAtomics) {
-    WARN_ONCE("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;
-  __asm volatile ("mrs %[ctr], ctr_el0"
-    : [ctr] "=r"(CTR));
-
-  DCacheLineSize = 4 << ((CTR >> 16) & 0xF);
-  ICacheLineSize = 4 << (CTR & 0xF);
-#endif
 }
 
-void Arm64Emitter::LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant) {
-  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();
+    }
     return;
   }
 
-  movz(Reg, (Constant) & 0xFFFF, 0);
-  for (int i = 1; i < Segments; ++i) {
+  int NumMoves = 1;
+  int RequiredMoveSegments{};
+
+  // Count the number of move segments
+  // We only want to use ADRP+ADD if we have more than 1 segment
+  for (size_t i = 0; i < Segments; ++i) {
     uint16_t Part = (Constant >> (i * 16)) & 0xFFFF;
-    if (Part) {
-      movk(Reg, Part, i * 16);
+    if (Part != 0) {
+      ++RequiredMoveSegments;
+    }
+  }
+
+  // ADRP+ADD is specifically optimized in hardware
+  // Check if we can use this
+  auto PC = GetCursorAddress<uint64_t>();
+
+  // PC aligned to page
+  uint64_t AlignedPC = PC & ~0xFFFULL;
+
+  // Offset from aligned PC
+  int64_t AlignedOffset = static_cast<int64_t>(Constant) - static_cast<int64_t>(AlignedPC);
+
+  // If the aligned offset is within the 4GB window then we can use ADRP+ADD
+  // and the number of move segments more than 1
+  if (RequiredMoveSegments > 1 && vixl::IsInt32(AlignedOffset)) {
+    // If this is 4k page aligned then we only need ADRP
+    if ((AlignedOffset & 0xFFF) == 0) {
+      adrp(Reg, AlignedOffset >> 12);
+    }
+    else {
+      // If the constant is within 1MB of PC then we can still use ADR to load in a single instruction
+      // 21-bit signed integer here
+      int64_t SmallOffset = static_cast<int64_t>(Constant) - static_cast<int64_t>(PC);
+      if (vixl::IsInt21(SmallOffset)) {
+        adr(Reg, SmallOffset);
+      }
+      else {
+        // Need to use ADRP + ADD
+        adrp(Reg, AlignedOffset >> 12);
+        add(s, Reg, Reg, Constant & 0xFFF);
+        NumMoves = 2;
+      }
+    }
+  }
+  else {
+    movz(s, Reg, (Constant) & 0xFFFF, 0);
+    for (int i = 1; i < Segments; ++i) {
+      uint16_t Part = (Constant >> (i * 16)) & 0xFFFF;
+      if (Part) {
+        movk(s, Reg, Part, i * 16);
+        ++NumMoves;
+      }
+    }
+  }
+
+  if (NOPPad) {
+    for (int i = NumMoves; i < Segments; ++i) {
+      nop();
     }
   }
 }
@@ -68,172 +111,288 @@ 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::XRegister, vixl::aarch64::XRegister>, 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::XRegister, vixl::aarch64::XRegister>, 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()) {
+    return;
+  }
 
-void Arm64Emitter::SpillStaticRegs() {
-  if (StaticRegisterAllocation()) {
-    for (size_t i = 0; i < SRA64.size(); i+=2) {
-        stp(SRA64[i], SRA64[i+1], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
+  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]));
+    }
+  }
 
-    for (size_t i = 0; i < SRAFPR.size(); i+=2) {
-      stp(SRAFPR[i].Q(), SRAFPR[i+1].Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+  if (FPRs) {
+    if (EmitterCTX->HostFeatures.SupportsAVX) {
+      for (size_t i = 0; i < SRAFPR.size(); i++) {
+        const auto Reg = SRAFPR[i];
+
+        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 (GPRSpillMask && FPRSpillMask == ~0U) {
+        // Optimize the common case where we can spill four registers per instruction
+        auto TmpReg = SRA64[__builtin_ffs(GPRSpillMask)];
+
+        // Load the sse offset in to the temporary register
+        add(ARMEmitter::Size::i64Bit, TmpReg, STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[0][0]));
+        for (size_t i = 0; i < SRAFPR.size(); i += 4) {
+          const auto Reg1 = SRAFPR[i];
+          const auto Reg2 = SRAFPR[i + 1];
+          const auto Reg3 = SRAFPR[i + 2];
+          const auto Reg4 = SRAFPR[i + 3];
+          st1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), TmpReg, 64);
+        }
+      }
+      else {
+        for (size_t i = 0; i < SRAFPR.size(); i += 2) {
+          const auto Reg1 = SRAFPR[i];
+          const auto Reg2 = SRAFPR[i + 1];
+
+          if (((1U << Reg1.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]));
+          }
+        }
+      }
     }
   }
 }
 
-void Arm64Emitter::FillStaticRegs() {
-  if (StaticRegisterAllocation()) {
-    for (size_t i = 0; i < SRA64.size(); i+=2) {
-      ldp(SRA64[i], SRA64[i+1], MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.gregs[i])));
-    }
+void Arm64Emitter::FillStaticRegs(bool FPRs, uint32_t GPRFillMask, uint32_t FPRFillMask) {
+  if (!StaticRegisterAllocation()) {
+    return;
+  }
 
-    for (size_t i = 0; i < SRAFPR.size(); i+=2) {
-      ldp(SRAFPR[i].Q(), SRAFPR[i+1].Q(), MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, State.xmm[i][0])));
+  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 (GPRFillMask && FPRFillMask == ~0U) {
+        // Optimize the common case where we can fill four registers per instruction.
+        // Use one of the filling static registers before we fill it.
+        auto TmpReg = SRA64[__builtin_ffs(GPRFillMask)];
+
+        // Load the sse offset in to the temporary register
+        add(ARMEmitter::Size::i64Bit, TmpReg, STATE.R(), offsetof(FEXCore::Core::CpuStateFrame, State.xmm.sse.data[0][0]));
+        for (size_t i = 0; i < SRAFPR.size(); i += 4) {
+          const auto Reg1 = SRAFPR[i];
+          const auto Reg2 = SRAFPR[i + 1];
+          const auto Reg3 = SRAFPR[i + 2];
+          const auto Reg4 = SRAFPR[i + 3];
+          ld1<ARMEmitter::SubRegSize::i64Bit>(Reg1.Q(), Reg2.Q(), Reg3.Q(), Reg4.Q(), TmpReg, 64);
+        }
+      }
+      else {
+        for (size_t i = 0; i < SRAFPR.size(); i += 2) {
+          const auto Reg1 = SRAFPR[i];
+          const auto Reg2 = SRAFPR[i + 1];
+
+          if (((1U << Reg1.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]));
+          }
+        }
+      }
+    }
+  }
+
+  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() {
   uint64_t CurrentOffset = GetCursorAddress<uint64_t>();
   for (uint64_t i = (16 - (CurrentOffset & 0xF)); i != 0; i -= 4) {
-      nop();
+    nop();
   }
 }
 

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

@@ -1,87 +1,187 @@
 #pragma once
 
-#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 "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>
+#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(size_t size);
+  Arm64Emitter(FEXCore::Context::Context *ctx, size_t size);
+  ~Arm64Emitter();
 
+  FEXCore::Context::Context *EmitterCTX;
   vixl::aarch64::CPU CPU;
-  bool SupportsAtomics{};
-  bool SupportsRCPC{};
+  void LoadConstant(FEXCore::ARMEmitter::Size s, FEXCore::ARMEmitter::Register Reg, uint64_t Constant, bool NOPPad = false);
 
-  void LoadConstant(vixl::aarch64::Register Reg, uint64_t Constant);
-  void SpillStaticRegs();
-  void FillStaticRegs();
 
-  void PushDynamicRegsAndLR();
+  // 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);
+
+  static constexpr uint32_t CALLER_GPR_MASK = 0b0011'1111'1111'1111'1111;
+
+  // This isn't technically true because the lower 64-bits of v8..v15 are callee saved
+  // We can't guarantee only the lower 64bits are used so flush everything
+  static constexpr uint32_t CALLER_FPR_MASK = ~0U;
+
+  void PushDynamicRegsAndLR(FEXCore::ARMEmitter::Register TmpReg);
   void PopDynamicRegsAndLR();
 
   void PushCalleeSavedRegisters();
   void PopCalleeSavedRegisters();
 
-  void ResetStack();
   void Align16B();
 
-  uint32_t SpillSlots{};
+#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)
 
-  uint32_t DCacheLineSize{};
-  uint32_t ICacheLineSize{};
+  template<typename R, typename... P>
+  void GenerateRuntimeCall(R (*Function)(P...)) {
+    uintptr_t SimulatorWrapperAddress = reinterpret_cast<uintptr_t>(
+      &(vixl::aarch64::Simulator::RuntimeCallStructHelper<R, P...>::Wrapper));
 
+    uintptr_t FunctionAddress = reinterpret_cast<uintptr_t>(Function);
+
+    hlt(vixl::aarch64::kRuntimeCallOpcode);
+
+    // Simulator wrapper address pointer.
+    dc64(SimulatorWrapperAddress);
+
+    // Runtime function address to call
+    dc64(FunctionAddress);
+
+    // Call type
+    dc32(vixl::aarch64::kCallRuntime);
+  }
+
+  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);
+
+    // 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/Arm64_stubs.cpp

@@ -12,15 +12,15 @@ namespace FEXCore::ArchHelpers::Arm64 {
 // Obvously such a configuration can't do the actual arm64-specific stuff
 
 bool HandleCASPAL(void *_ucontext, void *_info, uint32_t Instr) {
-    ERROR_AND_DIE("HandleCASPAL Not Implemented");
+    ERROR_AND_DIE_FMT("HandleCASPAL Not Implemented");
 }
 
 bool HandleCASAL(void *_ucontext, void *_info, uint32_t Instr) {
-    ERROR_AND_DIE("HandleCASAL Not Implemented");
+    ERROR_AND_DIE_FMT("HandleCASAL Not Implemented");
 }
 
 bool HandleAtomicMemOp(void *_ucontext, void *_info, uint32_t Instr) {
-    ERROR_AND_DIE("HandleAtomicMemOp Not Implemented");
+    ERROR_AND_DIE_FMT("HandleAtomicMemOp Not Implemented");
 }
 #endif
 

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>
@@ -10,20 +11,29 @@
 #include <stdint.h>
 #include <type_traits>
 
-
 namespace FEXCore::ArchHelpers::Context {
 
+enum ContextFlags : uint32_t {
+  CONTEXT_FLAG_INJIT = (1U << 0),
+  CONTEXT_FLAG_32BIT = (1U << 1),
+};
+
 struct X86ContextBackup {
   // Host State
   // RIP and RSP is stored in GPRs here
   uint64_t GPRs[23];
   FEXCore::x86_64::_libc_fpstate FPRState;
   uint64_t sa_mask;
+  bool FaultToTopAndGeneratedException;
 
   // Guest state
   int Signal;
+  uint32_t Flags;
+  uint64_t OriginalRIP;
+  uint64_t FPStateLocation;
+  uint64_t UContextLocation;
+  uint64_t SigInfoLocation;
   FEXCore::Core::CPUState GuestState;
-
   static constexpr int RedZoneSize = 128;
 };
 
@@ -37,9 +47,15 @@ struct ArmContextBackup {
   uint32_t FPCR;
   __uint128_t FPRs[32];
   uint64_t sa_mask;
+  bool FaultToTopAndGeneratedException;
 
   // Guest state
   int Signal;
+  uint32_t Flags;
+  uint64_t OriginalRIP;
+  uint64_t FPStateLocation;
+  uint64_t UContextLocation;
+  uint64_t SigInfoLocation;
   FEXCore::Core::CPUState GuestState;
 
   // Arm64 doesn't have a red zone
@@ -59,6 +75,26 @@ 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;
+  uint32_t Size;
+};
+
+struct HostFPRState {
+  HostCTXHeader Head;
+  uint32_t FPSR;
+  uint32_t FPCR;
+  __uint128_t FPRs[32];
+};
+
+struct HostESRState {
+  HostCTXHeader Head;
+  uint64_t ESR;
+};
+
 static inline uint64_t GetSp(void* ucontext) {
   return GetMContext(ucontext)->sp;
 }
@@ -91,19 +127,68 @@ static inline void SetArmReg(void* ucontext, uint32_t id, uint64_t val) {
   GetMContext(ucontext)->regs[id] = val;
 }
 
-constexpr uint32_t FPR_MAGIC = 0x46508001U;
+static inline __uint128_t GetArmFPR(void* ucontext, uint32_t id) {
+  auto MContext = GetMContext(ucontext);
+  HostFPRState *HostState = reinterpret_cast<HostFPRState*>(&MContext->__reserved[0]);
+  LOGMAN_THROW_AA_FMT(HostState->Head.Magic == FPR_MAGIC, "Wrong FPR Magic: 0x{:08x}", HostState->Head.Magic);
 
-struct HostCTXHeader {
-  uint32_t Magic;
-  uint32_t Size;
-};
+  return HostState->FPRs[id];
+}
 
-struct HostFPRState {
-  HostCTXHeader Head;
-  uint32_t FPSR;
-  uint32_t FPCR;
-  __uint128_t FPRs[32];
-};
+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>
@@ -119,7 +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(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));
@@ -127,7 +212,8 @@ static inline void BackupContext(void* ucontext, T *Backup) {
     // Save the signal mask so we can restore it
     memcpy(&Backup->sa_mask, &_ucontext->uc_sigmask, sizeof(uint64_t));
   } else {
-      ERROR_AND_DIE("Wrong context type"); // This must be a runtime error
+    // This must be a runtime error
+    ERROR_AND_DIE_FMT("Wrong context type");
   }
 }
 
@@ -138,7 +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(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;
@@ -152,7 +238,8 @@ static inline void RestoreContext(void* ucontext, T *Backup) {
     // Restore the signal mask now
     memcpy(&_ucontext->uc_sigmask, &Backup->sa_mask, sizeof(uint64_t));
   } else {
-      ERROR_AND_DIE("Wrong context type"); // This must be a runtime error
+    // This must be a runtime error
+    ERROR_AND_DIE_FMT("Wrong context type");
   }
 }
 
@@ -185,11 +272,19 @@ static inline void SetState(void* ucontext, uint64_t val) {
 }
 
 static inline uint64_t GetArmReg(void* ucontext, uint32_t id) {
-  ERROR_AND_DIE("Not impelented for x86 host");
+  ERROR_AND_DIE_FMT("Not impelented for x86 host");
 }
 
 static inline void SetArmReg(void* ucontext, uint32_t id, uint64_t val) {
-  ERROR_AND_DIE("Not impelented for x86 host");
+  ERROR_AND_DIE_FMT("Not impelented for x86 host");
+}
+
+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;
@@ -208,7 +303,8 @@ static inline void BackupContext(void* ucontext, T *Backup) {
     // Save the signal mask so we can restore it
     memcpy(&Backup->sa_mask, &_ucontext->uc_sigmask, sizeof(uint64_t));
   } else {
-    ERROR_AND_DIE("Wrong context type"); // This must be a runtime error
+    // This must be a runtime error
+    ERROR_AND_DIE_FMT("Wrong context type");
   }
 }
 
@@ -226,7 +322,8 @@ static inline void RestoreContext(void* ucontext, T *Backup) {
     // Restore the signal mask now
     memcpy(&_ucontext->uc_sigmask, &Backup->sa_mask, sizeof(uint64_t));
   } else {
-    ERROR_AND_DIE("Wrong context type"); // This must be a runtime error
+    // This must be a runtime error
+    ERROR_AND_DIE_FMT("Wrong context type");
   }
 }
 

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

@@ -27,7 +27,7 @@ namespace FEXCore {
              << std::endl;
     }
     Output.close();
-    LogMan::Msg::D("Dumped %d blocks of sampling data", SamplingMap.size());
+    LogMan::Msg::DFmt("Dumped {} blocks of sampling data", SamplingMap.size());
   }
 
   BlockSamplingData::BlockData *BlockSamplingData::GetBlockData(uint64_t RIP) {

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

@@ -5,11 +5,16 @@ desc: Handles presented capability bits for guest cpu
 $end_info$
 */
 
-#include <FEXCore/Config/Config.h>
-#include <FEXCore/Core/CPUID.h>
+#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"
 
 #include <cstring>
@@ -18,7 +23,71 @@ $end_info$
 #endif
 
 namespace FEXCore {
+namespace ProductNames {
+#ifdef _M_ARM_64
+  static const char ARM_UNKNOWN[] = "Unknown ARM CPU";
+  static const char ARM_A57[] = "Cortex-A57";
+  static const char ARM_A72[] = "Cortex-A72";
+  static const char ARM_A73[] = "Cortex-A73";
+  static const char ARM_A75[] = "Cortex-A75";
+  static const char ARM_A76[] = "Cortex-A76";
+  static const char ARM_A76AE[] = "Cortex-A76AE";
+  static const char ARM_V1[] = "Neoverse V1";
+  static const char ARM_A77[] = "Cortex-A77";
+  static const char ARM_A78[] = "Cortex-A78";
+  static const char ARM_A78AE[] = "Cortex-A78AE";
+  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";
+  static const char ARM_E1[] = "Neoverse E1";
+  static const char ARM_A35[] = "Cortex-A35";
+  static const char ARM_A53[] = "Cortex-A53";
+  static const char ARM_A55[] = "Cortex-A55";
+  static const char ARM_A65[] = "Cortex-A65";
+  static const char ARM_A510[] = "Cortex-A510";
+
+  static const char ARM_Kryo200[] = "Kryo 2xx";
+  static const char ARM_Kryo300[] = "Kryo 3xx";
+  static const char ARM_Kryo400[] = "Kryo 4xx/5xx";
+
+  static const char ARM_Kryo200S[] = "Kryo 2xx S";
+  static const char ARM_Kryo300S[] = "Kryo 3xx S";
+  static const char ARM_Kryo400S[] = "Kryo 4xx/5xx S";
+
+  static const char ARM_Denver[] = "Nvidia Denver";
+  static const char ARM_Carmel[] = "Nvidia Carmel";
+
+  static const char ARM_Firestorm[] = "Apple Firestorm";
+  static const char ARM_Icestorm[] = "Apple Icestorm";
+#else
+  static const char UNKNOWN[] = "Unknown CPU";
+#endif
+}
+
+static uint32_t GetCPUID() {
+  uint32_t CPU{};
+  FHU::Syscalls::getcpu(&CPU, nullptr);
+  return CPU;
+}
+
+static uint32_t CalculateNumberOfCPUs() {
+  size_t CPUs = 1;
+
+  while(std::filesystem::exists("/sys/devices/system/cpu/cpu" + std::to_string(CPUs))) {
+    CPUs++;
+  }
+
+  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 =
@@ -45,6 +114,246 @@ static uint32_t GetCycleCounterFrequency() {
       : [Res] "=r" (Result));
   return Result;
 }
+
+void CPUIDEmu::SetupHostHybridFlag() {
+  size_t CPUs = CalculateNumberOfCPUs();
+  PerCPUData.resize(CPUs);
+
+  uint64_t MIDR{};
+  for (size_t i = 0; i < CPUs; ++i) {
+    std::error_code ec{};
+    std::string MIDRPath = "/sys/devices/system/cpu/cpu" + std::to_string(i) + "/regs/identification/midr_el1";
+    if (std::filesystem::exists(MIDRPath, ec)) {
+      std::vector<char> Data{};
+      // Needs to be a fixed size since depending on kernel it will try to read a full page of data and fail
+      // Only read 18 bytes for a 64bit value prefixed with 0x
+      if (FEXCore::FileLoading::LoadFile(Data, MIDRPath, 18)) {
+        uint64_t NewMIDR{};
+        std::string_view MIDRView(&Data.at(0), 18);
+        if (FEXCore::StrConv::Conv(MIDRView, &NewMIDR)) {
+          if (MIDR != 0 && MIDR != NewMIDR) {
+            // CPU mismatch, claim hybrid
+            Hybrid = true;
+          }
+
+          // Truncate to 32-bits, top 32-bits are all reserved in MIDR
+          PerCPUData[i].ProductName = ProductNames::ARM_UNKNOWN;
+          PerCPUData[i].MIDR = NewMIDR;
+          MIDR = NewMIDR;
+        }
+      }
+    }
+  }
+
+  struct CPUMIDR {
+    uint8_t Implementer;
+    uint16_t Part;
+    bool DefaultBig; // Defaults to a big core
+    const char *ProductName{};
+  };
+
+  // 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, 36> CPUMIDRs = {{
+    // Typically big CPU cores
+    {0x61, 0x023, 1, ProductNames::ARM_Firestorm}, // Apple M1 Firestorm
+
+    {0x41, 0xd49, 1, ProductNames::ARM_N2}, // N2
+    {0x41, 0xd4b, 1, ProductNames::ARM_A78C}, // A78C
+    {0x41, 0xd4a, 1, ProductNames::ARM_E1}, // E1
+    {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
+    {0x41, 0xd40, 1, ProductNames::ARM_V1}, // V1
+    {0x41, 0xd0e, 1, ProductNames::ARM_A76AE}, // A76AE
+    {0x41, 0xd0d, 1, ProductNames::ARM_A77}, // A77
+    {0x41, 0xd0c, 1, ProductNames::ARM_N1}, // N1
+    {0x41, 0xd0b, 1, ProductNames::ARM_A76}, // A76
+    {0x51, 0x804, 1, ProductNames::ARM_Kryo400}, // Kryo 4xx Gold (A76 based)
+    {0x41, 0xd0a, 1, ProductNames::ARM_A75}, // A75
+    {0x51, 0x802, 1, ProductNames::ARM_Kryo300}, // Kryo 3xx Gold (A75 based)
+    {0x41, 0xd09, 1, ProductNames::ARM_A73}, // A73
+    {0x51, 0x800, 1, ProductNames::ARM_Kryo200}, // Kryo 2xx Gold (A73 based)
+    {0x41, 0xd08, 1, ProductNames::ARM_A72}, // A72
+
+    {0x4e, 0x004, 1, ProductNames::ARM_Carmel}, // Carmel
+
+    // Denver rated above A57 to match TX2 weirdness
+    {0x4e, 0x003, 1, ProductNames::ARM_Denver}, // Denver
+
+    {0x41, 0xd07, 1, ProductNames::ARM_A57}, // A57
+
+    // Typically Little CPU cores
+    {0x61, 0x022, 0, ProductNames::ARM_Icestorm}, // Apple M1 Icestorm
+    {0x41, 0xd46, 0, ProductNames::ARM_A510}, // A510
+    {0x41, 0xd06, 0, ProductNames::ARM_A65}, // A65
+    {0x41, 0xd05, 0, ProductNames::ARM_A55}, // A55
+    {0x51, 0x805, 0, ProductNames::ARM_Kryo400S}, // Kryo 4xx/5xx Silver (A55 based)
+    {0x51, 0x803, 0, ProductNames::ARM_Kryo300S}, // Kryo 3xx Silver (A55 based)
+    {0x41, 0xd03, 0, ProductNames::ARM_A53}, // A53
+    {0x51, 0x801, 0, ProductNames::ARM_Kryo200S}, // Kryo 2xx Silver (A53 based)
+    {0x41, 0xd04, 0, ProductNames::ARM_A35}, // A35
+
+    {0x41, 0, 0, ProductNames::ARM_UNKNOWN}, // Invalid CPU or Apple CPU inside Parallels VM
+    {0x0, 0, 0, ProductNames::ARM_UNKNOWN}, // Invalid starting point is lowest ranked
+  }};
+
+  auto FindDefinedMIDR = [](uint32_t MIDR) -> const CPUMIDR* {
+    uint8_t Implementer = MIDR >> 24;
+    uint16_t Part = (MIDR >> 4) & 0xFFF;
+
+    for (auto &MIDROption : CPUMIDRs) {
+      if (MIDROption.Implementer == Implementer &&
+          MIDROption.Part == Part) {
+        return &MIDROption;
+      }
+    }
+
+    return nullptr;
+  };
+
+  if (Hybrid) {
+    // Walk the MIDRs and calculate big little designs
+    std::vector<const CPUMIDR*> BigCores;
+    std::vector<const CPUMIDR*> LittleCores;
+
+    // Separate CPU cores out to big or little selected
+    for (size_t i = 0; i < CPUs; ++i) {
+      uint32_t MIDR = PerCPUData[i].MIDR;
+      auto MIDROption = FindDefinedMIDR(MIDR);
+      if (MIDROption) {
+        // Found one
+        if (MIDROption->DefaultBig) {
+          BigCores.emplace_back(MIDROption);
+        }
+        else {
+          LittleCores.emplace_back(MIDROption);
+        }
+      }
+      else {
+        // If we didn't insert this MIDR then claim it is a little core.
+        LittleCores.emplace_back(&CPUMIDRs.back());
+      }
+    }
+
+    if (LittleCores.empty()) {
+      // If we only ended up with big cores then we need to move some to be little cores
+      uint32_t LowestMIDR = ~0U;
+      uint32_t LowestMIDRIdx = 0;
+      // Walk all the big cores
+      for (size_t i = 0; i < BigCores.size(); ++i) {
+        uint8_t Implementer = BigCores[i]->Implementer;
+        uint16_t Part = BigCores[i]->Part;
+
+        // Walk our list of CPUMIDRs to find the most little core
+        for (size_t j = LowestMIDRIdx; j < CPUMIDRs.size(); ++j) {
+          auto &MIDROption = CPUMIDRs[i];
+          if ((MIDROption.Implementer == Implementer &&
+              MIDROption.Part == Part) ||
+              (MIDROption.Implementer == 0 &&
+               MIDROption.Part == 0)) {
+
+            LowestMIDRIdx = j;
+            LowestMIDR = MIDR;
+            break;
+          }
+        }
+      }
+
+      // Now we WILL have found a big core to demote to little status
+      // Demote them
+      std::erase_if(BigCores, [&LittleCores, LowestMIDR](auto *Entry) {
+        // Demote by erase copy to little array
+        uint8_t Implementer = LowestMIDR >> 24;
+        uint16_t Part = (LowestMIDR >> 4) & 0xFFF;
+
+        if (Entry->Implementer == Implementer &&
+            Entry->Part == Part) {
+          // Add it to the BigCore list
+          LittleCores.emplace_back(Entry);
+          return true;
+        }
+        return false;
+      });
+    }
+
+    if (BigCores.empty()) {
+      // We never found a CPU core we understand
+      // Grab the first core, consider it as little, move everything else to Big
+      uint32_t LittleMIDR = PerCPUData[0].MIDR;
+      // Now walk the little cores and move them to Big if they don't match
+      std::erase_if(LittleCores, [&BigCores, LittleMIDR](auto *Entry) {
+        // You're promoted now
+        uint8_t Implementer = LittleMIDR >> 24;
+        uint16_t Part = (LittleMIDR >> 4) & 0xFFF;
+
+        if (Entry->Implementer != Implementer ||
+            Entry->Part != Part) {
+          // Add it to the BigCore list
+          BigCores.emplace_back(Entry);
+          return true;
+        }
+        return false;
+      });
+    }
+
+    // Now walk the per CPU data one more time and set if it is big or little
+    for (auto &Data : PerCPUData) {
+      uint8_t Implementer = Data.MIDR >> 24;
+      uint16_t Part = (Data.MIDR >> 4) & 0xFFF;
+
+      bool FoundBig{};
+      const CPUMIDR *MIDR{};
+      for (auto Big : BigCores) {
+        if (Big->Implementer == Implementer &&
+            Big->Part == Part) {
+          FoundBig = true;
+          MIDR = Big;
+          break;
+        }
+      }
+
+      if (!FoundBig) {
+        for (auto Little : LittleCores) {
+          if (Little->Implementer == Implementer &&
+              Little->Part == Part) {
+            MIDR = Little;
+            break;
+          }
+        }
+      }
+
+      Data.IsBig = FoundBig;
+      if (MIDR) {
+        Data.ProductName = MIDR->ProductName ?: ProductNames::ARM_UNKNOWN;
+      }
+      else {
+        Data.ProductName = ProductNames::ARM_UNKNOWN;
+      }
+    }
+  }
+  else {
+    // If we aren't hybrid then just claim everything is big
+    for (size_t i = 0; i < CPUs; ++i) {
+      uint32_t MIDR = PerCPUData[i].MIDR;
+      auto MIDROption = FindDefinedMIDR(MIDR);
+
+      PerCPUData[i].IsBig = true;
+      if (MIDROption) {
+        PerCPUData[i].ProductName = MIDROption->ProductName ?: ProductNames::ARM_UNKNOWN;
+      }
+      else {
+        PerCPUData[i].ProductName = ProductNames::ARM_UNKNOWN;
+      }
+    }
+  }
+}
+
 #else
 static uint32_t GetCycleCounterFrequency() {
   uint32_t eax, ebx, ecx, edx;
@@ -58,6 +367,24 @@ static uint32_t GetCycleCounterFrequency() {
   }
   return 0;
 }
+
+void CPUIDEmu::SetupHostHybridFlag() {
+  uint32_t eax, ebx, ecx, edx;
+  __cpuid(0, eax, ebx, ecx, edx);
+  if (eax >= 0x7) {
+    __cpuid(0x7, eax, ebx, ecx, edx);
+    // Bit 15 of edx claims hybrid CPU
+    Hybrid = (edx & (1U << 15)) != 0;
+  }
+
+  size_t CPUs = CalculateNumberOfCPUs();
+  PerCPUData.resize(CPUs);
+  for (size_t i = 0; i < CPUs; ++i) {
+    PerCPUData[i].IsBig = true;
+    PerCPUData[i].ProductName = ProductNames::UNKNOWN;
+  }
+}
+
 #endif
 
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_0h(uint32_t Leaf) {
@@ -82,6 +409,8 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_0h(uint32_t Leaf) {
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
   FEXCore::CPUID::FunctionResults Res{};
   uint32_t CoreCount = Cores();
+  // XXX: Enable once the rest of the SSE4.2 instructions are emulated
+  uint32_t SupportsSSE42 = CTX->HostFeatures.SupportsCRC && false ? 1 : 0;
 
   Res.eax = FAMILY_IDENTIFIER;
 
@@ -92,7 +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
@@ -111,7 +440,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
     (0 << 17) | // Process-context identifiers
     (0 << 18) | // Prefetching from memory mapped device
     (1 << 19) | // SSE4.1
-    (0 << 20) | // SSE4.2
+    (SupportsSSE42 << 20) | // SSE4.2
     (0 << 21) | // X2APIC
     (1 << 22) | // MOVBE
     (1 << 23) | // POPCNT
@@ -121,8 +450,8 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_01h(uint32_t Leaf) {
     (0 << 27) | // OSXSAVE
     (SUPPORTS_AVX << 28) | // AVX
     (0 << 29) | // F16C
-    (0 << 30) | // RDRAND
-    (0 << 31);  // Hypervisor always returns zero
+    (CTX->HostFeatures.SupportsRAND << 30) | // RDRAND
+    (1 << 31);  // Hypervisor always returns one
 
   Res.edx =
     (1 <<  0) | // FPU
@@ -308,12 +637,12 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_07h(uint32_t Leaf) {
       (1 <<  0) | // FS/GS support
       (0 <<  1) | // TSC adjust MSR
       (0 <<  2) | // SGX
-      (0 <<  3) | // BMI1
+      (1 <<  3) | // BMI1
       (0 <<  4) | // Intel Hardware Lock Elison
       (0 <<  5) | // AVX2 support
       (1 <<  6) | // FPU data pointer updated only on exception
       (1 <<  7) | // SMEP support
-      (0 <<  8) | // BMI2
+      (1 <<  8) | // BMI2
       (0 <<  9) | // Enhanced REP MOVSB/STOSB
       (1 << 10) | // INVPCID for system software control of process-context
       (0 << 11) | // Restricted transactional memory
@@ -323,8 +652,8 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_07h(uint32_t Leaf) {
       (0 << 15) | // Intel Resource Directory Technology Allocation
       (0 << 16) | // Reserved
       (0 << 17) | // Reserved
-      (0 << 18) | // RDSEED
-      (0 << 19) | // ADCX and ADOX instructions
+      (CTX->HostFeatures.SupportsRAND << 18) | // RDSEED
+      (1 << 19) | // ADCX and ADOX instructions
       (0 << 20) | // SMAP Supervisor mode access prevention and CLAC/STAC instructions
       (0 << 21) | // Reserved
       (0 << 22) | // Reserved
@@ -334,7 +663,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_07h(uint32_t Leaf) {
       (0 << 26) | // Reserved
       (0 << 27) | // Reserved
       (0 << 28) | // Reserved
-      (0 << 29) | // SHA instructions
+      (1 << 29) | // SHA instructions
       (0 << 30) | // Reserved
       (0 << 31);  // Reserved
 
@@ -382,29 +711,29 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_07h(uint32_t Leaf) {
       (0 <<  6) | // Reserved
       (0 <<  7) | // Reserved
       (0 <<  8) | // AVX512_VP2INTERSECT
-      (0 <<  9) | // Reserved
+      (0 <<  9) | // SRBDS_CTRL (Special Register Buffer Data Sampling Mitigations)
       (0 << 10) | // VERW clears CPU buffers
       (0 << 11) | // Reserved
       (0 << 12) | // Reserved
-      (0 << 13) | // Reserved
+      (0 << 13) | // TSX Force Abort (TSX will force abort if attempted)
       (0 << 14) | // SERIALIZE instruction
-      (0 << 15) | // Reserved
-      (0 << 16) | // Reserved
+      ((Hybrid ? 1U : 0U) << 15) | // Hybrid
+      (0 << 16) | // TSXLDTRK (TSX Suspend load address tracking) - Allows untracked memory loads inside TSX region
       (0 << 17) | // Reserved
       (0 << 18) | // Intel PCONFIG
       (0 << 19) | // Intel Architectural LBR
       (0 << 20) | // Intel CET
       (0 << 21) | // Reserved
-      (0 << 22) | // Reserved
-      (0 << 23) | // Reserved
-      (0 << 24) | // Reserved
-      (0 << 25) | // Reserved
-      (0 << 26) | // Reserved
-      (0 << 27) | // Reserved
+      (0 << 22) | // AMX-BF16 - Tile computation on bfloat16
+      (0 << 23) | // AVX512_FP16 - FP16 AVX512 instructions
+      (0 << 24) | // AMX-tile - If AMX is implemented
+      (0 << 25) | // AMX-int8 - AMX on 8-bit integers
+      (0 << 26) | // IBRS_IBPB - Speculation control
+      (0 << 27) | // STIBP - Single Thread Indirect Branch Predictor, Part of IBC
       (0 << 28) | // L1D Flush
-      (0 << 29) | // Arch capabilities
-      (0 << 30) | // Reserved
-      (0 << 31);  // Reserved
+      (0 << 29) | // Arch capabilities - Speculative side channel mitigations
+      (0 << 30) | // Arch capabilities - MSR module specific
+      (0 << 31);  // SSBD - Speculative Store Bypass Disable
   }
 
   return Res;
@@ -476,6 +805,59 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_15h(uint32_t Leaf) {
   return Res;
 }
 
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_1Ah(uint32_t Leaf) {
+  FEXCore::CPUID::FunctionResults Res{};
+  if (Hybrid) {
+    uint32_t CPU = GetCPUID();
+    auto &Data = PerCPUData[CPU];
+    // 0x40 is a big CPU
+    // 0x20 is a little CPU
+    Res.eax |= (Data.IsBig ? 0x40 : 0x20) << 24;
+  }
+  return Res;
+}
+
+// Hypervisor CPUID information leaf
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0000h(uint32_t Leaf) {
+  FEXCore::CPUID::FunctionResults Res{};
+  // Maximum supported hypervisor leafs
+  // We only expose the information leaf
+  //
+  // Common courtesy to follow VMWare's "Hypervisor CPUID Interface proposal"
+  // 4000_0000h - Information leaf. Advertising to the software which hypervisor this is
+  // 4000_0001h - 4000_000Fh - Hypervisor specific leafs. FEX can use these for anything
+  // 4000_0010h - 4000_00FFh - "Generic Leafs" - Try not to overwrite, other hypervisors might expect information in these
+  //
+  // CPUID documentation information:
+  // 4000_0000h - 4FFF_FFFFh - No existing or future CPU will return information in this range
+  // Reserved entirely for VMs to do whatever they want.
+  Res.eax = 0x40000001;
+
+  // EBX, EDX, ECX become the hypervisor ID signature
+  constexpr static char HypervisorID[12] = "FEXIFEXIEMU";
+  memcpy(&Res.ebx, HypervisorID, sizeof(HypervisorID));
+  return Res;
+}
+
+// Hypervisor CPUID information leaf
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_4000_0001h(uint32_t Leaf) {
+  FEXCore::CPUID::FunctionResults Res{};
+  if (Leaf == 0) {
+    // EAX[3:0] Is the host architecture that FEX is running under
+#ifdef _M_X86_64
+    // EAX[3:0] = 1 = x86_64 host architecture
+    Res.eax |= 0b0001;
+#elif defined(_M_ARM_64)
+    // EAX[3:0] = 2 = AArch64 host architecture
+    Res.eax |= 0b0010;
+#else
+    // EAX[3:0] = 0 = Unknown architecture
+#endif
+  }
+
+  return Res;
+}
+
 // Highest extended function implemented
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0000h(uint32_t Leaf) {
   FEXCore::CPUID::FunctionResults Res{};
@@ -563,35 +945,52 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0001h(uint32_t Leaf) {
     (1 << 24) | // FXSAVE/FXRSTOR
     (1 << 25) | // FXSAVE/FXRSTOR Optimizations
     (0 << 26) | // 1 gigabit pages
-    (0 << 27) | // RDTSCP
+    (1 << 27) | // RDTSCP
     (0 << 28) | // Reserved
     (1 << 29) | // Long Mode
-    (0 << 30) | // 3DNow! Extensions
-    (0 << 31);  // 3DNow!
+    (1 << 30) | // 3DNow! Extensions
+    (1 << 31);  // 3DNow!
   return Res;
 }
 
-constexpr char ProcessorBrand[48] = {
+constexpr char ProcessorBrand[32] = {
   GIT_DESCRIBE_STRING
-  "\0"
 };
 
+constexpr ssize_t DESCRIBE_STR_SIZE = std::char_traits<char>::length(GIT_DESCRIBE_STRING);
+static_assert(DESCRIBE_STR_SIZE < 32);
+
 //Processor brand string
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0002h(uint32_t Leaf) {
-  FEXCore::CPUID::FunctionResults Res{};
-  memcpy(&Res, &ProcessorBrand[0], sizeof(FEXCore::CPUID::FunctionResults));
-  return Res;
+  return Function_8000_0002h(Leaf, GetCPUID());
 }
 
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0003h(uint32_t Leaf) {
-  FEXCore::CPUID::FunctionResults Res{};
-  memcpy(&Res, &ProcessorBrand[16], sizeof(FEXCore::CPUID::FunctionResults));
-  return Res;
+  return Function_8000_0003h(Leaf, GetCPUID());
 }
 
 FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0004h(uint32_t Leaf) {
+  return Function_8000_0004h(Leaf, GetCPUID());
+}
+
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0002h(uint32_t Leaf, uint32_t CPU) {
   FEXCore::CPUID::FunctionResults Res{};
-  memcpy(&Res, &ProcessorBrand[32], sizeof(FEXCore::CPUID::FunctionResults));
+  memset(&Res, ' ', sizeof(FEXCore::CPUID::FunctionResults));
+  memcpy(&Res, &ProcessorBrand[0], std::min(16L, DESCRIBE_STR_SIZE));
+  return Res;
+}
+
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0003h(uint32_t Leaf, uint32_t CPU) {
+  FEXCore::CPUID::FunctionResults Res{};
+  memset(&Res, ' ', sizeof(FEXCore::CPUID::FunctionResults));
+  memcpy(&Res, &ProcessorBrand[16], std::max(0L, DESCRIBE_STR_SIZE - 16));
+  return Res;
+}
+
+FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0004h(uint32_t Leaf, uint32_t CPU) {
+  FEXCore::CPUID::FunctionResults Res{};
+  auto &Data = PerCPUData[CPU];
+  memcpy(&Res, Data.ProductName, std::min(strlen(Data.ProductName), sizeof(FEXCore::CPUID::FunctionResults)));
   return Res;
 }
 
@@ -684,7 +1083,7 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_8000_0008h(uint32_t Leaf) {
   Res.ebx =
     (0 << 2) | // XSaveErPtr: Saving and restoring error pointers
     (0 << 1) | // IRPerf: Instructions retired count support
-    (0 << 0);  // CLZERO support
+    (CTX->HostFeatures.SupportsCLZERO << 0);  // CLZERO support
 
   uint32_t CoreCount = Cores() - 1;
   Res.ecx =
@@ -815,25 +1214,25 @@ FEXCore::CPUID::FunctionResults CPUIDEmu::Function_Reserved(uint32_t Leaf) {
 
 void CPUIDEmu::Init(FEXCore::Context::Context *ctx) {
   CTX = ctx;
-  using namespace std::placeholders;
-  RegisterFunction(0, std::bind(&CPUIDEmu::Function_0h, this, _1));
-  RegisterFunction(1, std::bind(&CPUIDEmu::Function_01h, this, _1));
-  RegisterFunction(2, std::bind(&CPUIDEmu::Function_02h, this, _1));
+
+  RegisterFunction(0, &CPUIDEmu::Function_0h);
+  RegisterFunction(1, &CPUIDEmu::Function_01h);
+  RegisterFunction(2, &CPUIDEmu::Function_02h);
   // 3: Serial Number(previously), now reserved
 #ifndef CPUID_AMD
   // Deterministic cache parameters for each level
-  RegisterFunction(0x4, std::bind(&CPUIDEmu::Function_04h, this, _1));
+  RegisterFunction(0x4, &CPUIDEmu::Function_04h);
 #endif
   // 5: Monitor/mwait
   // Thermal and power management
-  RegisterFunction(6, std::bind(&CPUIDEmu::Function_06h, this, _1));
+  RegisterFunction(6, &CPUIDEmu::Function_06h);
   // Extended feature flags
-  RegisterFunction(7, std::bind(&CPUIDEmu::Function_07h, this, _1));
+  RegisterFunction(7, &CPUIDEmu::Function_07h);
   // 9: Direct Cache Access information
   // 0x0A: Architectural performance monitoring
   // 0x0B: Extended topology enumeration
   // 0x0D: Processor extended state enumeration
-  RegisterFunction(0x0D, std::bind(&CPUIDEmu::Function_0Dh, this, _1));
+  RegisterFunction(0x0D, &CPUIDEmu::Function_0Dh);
   // 0x0F: Intel RDT monitoring
   // 0x10: Intel RDT allocation enumeration
   // 0x12: Intel SGX capability enumeration
@@ -842,38 +1241,46 @@ void CPUIDEmu::Init(FEXCore::Context::Context *ctx) {
 #ifndef CPUID_AMD
   // Timestamp counter information
   // Doesn't exist on AMD hardware
-  RegisterFunction(0x15, std::bind(&CPUIDEmu::Function_15h, this, _1));
+  RegisterFunction(0x15, &CPUIDEmu::Function_15h);
 #endif
   // 0x16: Processor frequency information
   // 0x17: SoC vendor attribute enumeration
 
+  // 0x1A: Hybrid Information Sub-leaf
+#ifndef CPUID_AMD
+  RegisterFunction(0x1A, &CPUIDEmu::Function_1Ah);
+#endif
+  // Hypervisor CPUID information leaf
+  RegisterFunction(0x4000'0000, &CPUIDEmu::Function_4000_0000h);
+  RegisterFunction(0x4000'0001, &CPUIDEmu::Function_4000_0001h);
+
   // Largest extended function number
-  RegisterFunction(0x8000'0000, std::bind(&CPUIDEmu::Function_8000_0000h, this, _1));
+  RegisterFunction(0x8000'0000, &CPUIDEmu::Function_8000_0000h);
   // Processor vendor
-  RegisterFunction(0x8000'0001, std::bind(&CPUIDEmu::Function_8000_0001h, this, _1));
+  RegisterFunction(0x8000'0001, &CPUIDEmu::Function_8000_0001h);
   // Processor brand string
-  RegisterFunction(0x8000'0002, std::bind(&CPUIDEmu::Function_8000_0002h, this, _1));
+  RegisterFunction(0x8000'0002, &CPUIDEmu::Function_8000_0002h);
   // Processor brand string continued
-  RegisterFunction(0x8000'0003, std::bind(&CPUIDEmu::Function_8000_0003h, this, _1));
+  RegisterFunction(0x8000'0003, &CPUIDEmu::Function_8000_0003h);
   // Processor brand string continued
-  RegisterFunction(0x8000'0004, std::bind(&CPUIDEmu::Function_8000_0004h, this, _1));
+  RegisterFunction(0x8000'0004, &CPUIDEmu::Function_8000_0004h);
   // 0x8000'0005: L1 Cache and TLB identifiers
 #ifdef CPUID_AMD
-  RegisterFunction(0x8000'0005, std::bind(&CPUIDEmu::Function_8000_0005h, this, _1));
+  RegisterFunction(0x8000'0005, &CPUIDEmu::Function_8000_0005h);
 #else
   // This is full reserved on Intel platforms
-  RegisterFunction(0x8000'0005, std::bind(&CPUIDEmu::Function_Reserved, this, _1));
+  RegisterFunction(0x8000'0005, &CPUIDEmu::Function_Reserved);
 #endif
   // 0x8000'0006: L2 Cache identifiers
-  RegisterFunction(0x8000'0006, std::bind(&CPUIDEmu::Function_8000_0006h, this, _1));
+  RegisterFunction(0x8000'0006, &CPUIDEmu::Function_8000_0006h);
   // Advanced power management information
-  RegisterFunction(0x8000'0007, std::bind(&CPUIDEmu::Function_8000_0007h, this, _1));
+  RegisterFunction(0x8000'0007, &CPUIDEmu::Function_8000_0007h);
   // Virtual and physical address sizes
-  RegisterFunction(0x8000'0008, std::bind(&CPUIDEmu::Function_8000_0008h, this, _1));
+  RegisterFunction(0x8000'0008, &CPUIDEmu::Function_8000_0008h);
 
   // 0x8000'000A: SVM Revision
   // TLB 1GB page identifiers
-  RegisterFunction(0x8000'0019, std::bind(&CPUIDEmu::Function_8000_0019h, this, _1));
+  RegisterFunction(0x8000'0019, &CPUIDEmu::Function_8000_0019h);
 
   // 0x8000'001A: Performance optimization identifiers
   // 0x8000'001B: Instruction based sampling identifiers
@@ -881,10 +1288,13 @@ void CPUIDEmu::Init(FEXCore::Context::Context *ctx) {
   // 0x8000'001D: Cache properties
 #ifdef CPUID_AMD
   // Deterministic cache parameters for each level
-  RegisterFunction(0x8000'001D, std::bind(&CPUIDEmu::Function_8000_001Dh, this, _1));
+  RegisterFunction(0x8000'001D, &CPUIDEmu::Function_8000_001Dh);
 #endif
   // 0x8000'001E: Extended APIC ID
   // 0x8000'001F: AMD Secure Encryption
+
+  // Setup some state tracking
+  SetupHostHybridFlag();
 }
 }
 

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

@@ -1,13 +1,13 @@
 #pragma once
-#include <functional>
+
+#include <cstdint>
 #include <unordered_map>
+#include <utility>
+#include <vector>
 
 #include <FEXCore/Core/CPUID.h>
 #include <FEXCore/Config/Config.h>
 
-#include <cstdint>
-#include <utility>
-
 namespace FEXCore {
 namespace Context {
   struct Context;
@@ -24,27 +24,50 @@ private:
   constexpr static uint32_t CPUID_VENDOR_AMD3 = 0x444D4163; // "cAMD"
 
 public:
+  // X86 cacheline size effectively has to be hardcoded to 64
+  // if we report anything differently then applications are likely to break
+  constexpr static uint64_t CACHELINE_SIZE = 64;
+
   void Init(FEXCore::Context::Context *ctx);
 
   FEXCore::CPUID::FunctionResults RunFunction(uint32_t Function, uint32_t Leaf) {
-    auto Handler = FunctionHandlers.find(Function);
+    const auto Handler = FunctionHandlers.find(Function);
 
     if (Handler == FunctionHandlers.end()) {
       return Function_Reserved(Leaf);
     }
 
-    return Handler->second(Leaf);
+    return (this->*Handler->second)(Leaf);
   }
+
+  FEXCore::CPUID::FunctionResults RunFunctionName(uint32_t Function, uint32_t Leaf, uint32_t CPU) {
+    if (Function == 0x8000'0002U)
+      return Function_8000_0002h(Leaf, CPU % PerCPUData.size());
+    else if (Function == 0x8000'0003U)
+      return Function_8000_0003h(Leaf, CPU % PerCPUData.size());
+    else
+      return Function_8000_0004h(Leaf, CPU % PerCPUData.size());
+  }
+
 private:
   FEXCore::Context::Context *CTX;
+  bool Hybrid{};
   FEX_CONFIG_OPT(Cores, THREADS);
 
-  using FunctionHandler = std::function<FEXCore::CPUID::FunctionResults(uint32_t Leaf)>;
+  using FunctionHandler = FEXCore::CPUID::FunctionResults (CPUIDEmu::*)(uint32_t Leaf);
   void RegisterFunction(uint32_t Function, FunctionHandler Handler) {
-    FunctionHandlers[Function] = Handler;
+    FunctionHandlers.insert_or_assign(Function, Handler);
   }
 
   std::unordered_map<uint32_t, FunctionHandler> FunctionHandlers;
+  struct CPUData {
+    const char *ProductName{};
+#ifdef _M_ARM_64
+    uint32_t MIDR{};
+#endif
+    bool IsBig{};
+  };
+  std::vector<CPUData> PerCPUData{};
 
   // Functions
   FEXCore::CPUID::FunctionResults Function_0h(uint32_t Leaf);
@@ -55,11 +78,19 @@ private:
   FEXCore::CPUID::FunctionResults Function_07h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_0Dh(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_15h(uint32_t Leaf);
+  FEXCore::CPUID::FunctionResults Function_1Ah(uint32_t Leaf);
+  FEXCore::CPUID::FunctionResults Function_4000_0000h(uint32_t Leaf);
+  FEXCore::CPUID::FunctionResults Function_4000_0001h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0000h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0001h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0002h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0003h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0004h(uint32_t Leaf);
+
+  FEXCore::CPUID::FunctionResults Function_8000_0002h(uint32_t Leaf, uint32_t CPU);
+  FEXCore::CPUID::FunctionResults Function_8000_0003h(uint32_t Leaf, uint32_t CPU);
+  FEXCore::CPUID::FunctionResults Function_8000_0004h(uint32_t Leaf, uint32_t CPU);
+
   FEXCore::CPUID::FunctionResults Function_8000_0005h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0006h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_0007h(uint32_t Leaf);
@@ -68,5 +99,8 @@ private:
   FEXCore::CPUID::FunctionResults Function_8000_0019h(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_8000_001Dh(uint32_t Leaf);
   FEXCore::CPUID::FunctionResults Function_Reserved(uint32_t Leaf);
+
+  void SetupHostHybridFlag();
+
 };
 }

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

@@ -1,181 +0,0 @@
-#include "Interface/Context/Context.h"
-#include "Interface/Core/LookupCache.h"
-#include "Interface/Core/CompileService.h"
-#include "Interface/Core/OpcodeDispatcher.h"
-#include "FEXCore/Debug/InternalThreadState.h"
-#include "FEXCore/HLE/Linux/ThreadManagement.h"
-#include "Interface/IR/PassManager.h"
-
-#include <FEXCore/Core/CPUBackend.h>
-#include <FEXCore/Core/CoreState.h>
-#include <FEXCore/Core/SignalDelegator.h>
-#include <FEXCore/Utils/Event.h>
-#include <FEXCore/Utils/LogManager.h>
-#include <FEXCore/Utils/Threads.h>
-
-#include <memory>
-#include <pthread.h>
-#include <stdio.h>
-
-namespace FEXCore {
-  static void* ThreadHandler(void *Arg) {
-    FEXCore::CompileService *This = reinterpret_cast<FEXCore::CompileService*>(Arg);
-    This->ExecutionThread();
-    return nullptr;
-  }
-
-  CompileService::CompileService(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread)
-    : CTX {ctx}
-    , ParentThread {Thread} {
-
-    CompileThreadData = std::make_unique<FEXCore::Core::InternalThreadState>();
-    CompileThreadData->IsCompileService = true;
-
-    // We need a compiler for this work thread
-    CTX->InitializeCompiler(CompileThreadData.get(), true);
-    CompileThreadData->CPUBackend->CopyNecessaryDataForCompileThread(ParentThread->CPUBackend.get());
-
-    WorkerThread = FEXCore::Threads::Thread::Create(ThreadHandler, this);
-  }
-
-  void CompileService::Initialize() {
-    // Share CompileService which = this
-    CompileThreadData->CompileService = ParentThread->CompileService;
-  }
-
-  void CompileService::Shutdown() {
-    ShuttingDown = true;
-    // Kick the working thread
-    StartWork.NotifyAll();
-    WorkerThread->join(nullptr);
-  }
-
-  void CompileService::ClearCache(FEXCore::Core::InternalThreadState *Thread) {
-    // On cache clear we need to spin down the execution thread to ensure it isn't trying to give us more work items
-    if (CompileMutex.try_lock()) {
-      // We can only clear these things if we pulled the compile mutex
-
-      // Grab the work queue and clear it
-      // We don't need to grab the queue mutex since this thread will no longer receive any work events
-      // Threads are bounded 1:1
-      while (WorkQueue.size()) {
-        WorkItem *Item = WorkQueue.front();
-        WorkQueue.pop();
-        delete Item;
-      }
-
-      // Go through the garbage collection array and clear it
-      // It's safe to clear things that aren't marked safe since we are clearing cache
-      if (GCArray.size()) {
-        // Clean up our GC array
-        for (auto it = GCArray.begin(); it != GCArray.end();) {
-          delete *it;
-          it = GCArray.erase(it);
-        }
-      }
-
-      LOGMAN_THROW_A(CompileThreadData->LocalIRCache.size() == 0, "Compile service must never have LocalIRCache");
-
-      CompileMutex.unlock();
-    }
-
-    // Clear the inverse cache of what is calling us from the Context ClearCache routine
-    auto SelectedThread = Thread->IsCompileService ? ParentThread : Thread;
-    SelectedThread->LookupCache->ClearCache();
-    SelectedThread->CPUBackend->ClearCache();
-  }
-
-
-  CompileService::WorkItem *CompileService::CompileCode(uint64_t RIP) {
-    // Tell the worker thread to compile code for us
-    WorkItem *Item = new WorkItem{};
-    Item->RIP = RIP;
-
-    {
-      // Fill the threads work queue
-      std::scoped_lock<std::mutex> lk(QueueMutex);
-      WorkQueue.emplace(Item);
-    }
-
-    // Notify the thread that it has more work
-    StartWork.NotifyAll();
-
-    return Item;
-  }
-
-  void CompileService::ExecutionThread() {
-    // Ignore signals coming from the guest
-    CTX->SignalDelegation->MaskThreadSignals();
-
-    // Set our thread name so we can see its relation
-    char ThreadName[16]{};
-    snprintf(ThreadName, 16, "%ld-CS", ParentThread->ThreadManager.TID.load());
-    pthread_setname_np(pthread_self(), ThreadName);
-
-    while (true) {
-      // Wait for work
-      StartWork.Wait();
-      if (ShuttingDown.load()) {
-        break;
-      }
-      std::scoped_lock<std::mutex> lk(CompileMutex);
-
-      size_t WorkItems{};
-
-      do {
-        // Grab a work item
-        WorkItem *Item{};
-        {
-          std::scoped_lock<std::mutex> lk(QueueMutex);
-          WorkItems = WorkQueue.size();
-          if (WorkItems) {
-            Item = WorkQueue.front();
-            WorkQueue.pop();
-          }
-        }
-
-        // If we had a work item then work on it
-        if (Item) {
-          // Make sure it's not in lookup cache by accident
-          LOGMAN_THROW_A(CompileThreadData->LookupCache->FindBlock(Item->RIP) == 0, "Compile Service must never have entries in the LookupCache");
-
-          // Code isn't in cache, compile now
-          // Set our thread state's RIP
-          CompileThreadData->CurrentFrame->State.rip = Item->RIP;
-
-          auto [CodePtr, IRList, DebugData, RAData, Generated, StartAddr, Length] = CTX->CompileCode(CompileThreadData.get(), Item->RIP);
-
-          LOGMAN_THROW_A(Generated == true, "Compile Service doesn't have IR Cache");
-
-          if (!CodePtr) {
-            // XXX: We currently have the expectation that compile service code will be significantly smaller than regular thread's code
-            ERROR_AND_DIE("Couldn't compile code for thread at RIP: 0x%lx", Item->RIP);
-          }
-
-          Item->CodePtr = CodePtr;
-          Item->IRList = IRList;
-          Item->DebugData = DebugData;
-          Item->RAData = RAData;
-          Item->StartAddr = StartAddr;
-          Item->Length = Length;
-
-          GCArray.emplace_back(Item);
-          Item->ServiceWorkDone.NotifyAll();
-        }
-      } while (WorkItems != 0);
-
-      if (GCArray.size()) {
-        // Clean up our GC array
-        for (auto it = GCArray.begin(); it != GCArray.end();) {
-          if ((*it)->SafeToClear) {
-            delete *it;
-            it = GCArray.erase(it);
-          }
-          else {
-            ++it;
-          }
-        }
-      }
-    }
-  }
-}

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

@@ -1,65 +0,0 @@
-#pragma once
-
-#include <FEXCore/Debug/InternalThreadState.h>
-#include <FEXCore/Utils/Event.h>
-#include <FEXCore/Utils/Threads.h>
-
-#include <atomic>
-#include <memory>
-#include <mutex>
-#include <queue>
-#include <stdint.h>
-#include <vector>
-
-namespace FEXCore {
-namespace Context {
-  struct Context;
-}
-namespace IR {
-  class IRListView;
-  class RegisterAllocationData;
-};
-class CompileService final {
-  public:
-    CompileService(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread);
-    void Initialize();
-    void Shutdown();
-
-    struct WorkItem {
-      // Incoming
-      uint64_t RIP{};
-
-      // Outgoing
-      void *CodePtr{};
-      FEXCore::IR::IRListView *IRList{};
-      FEXCore::IR::RegisterAllocationData *RAData{};
-      FEXCore::Core::DebugData *DebugData{};
-      uint64_t StartAddr;
-      uint64_t Length;
-
-      // Communication
-      Event ServiceWorkDone{};
-      std::atomic_bool SafeToClear{};
-    };
-
-    WorkItem *CompileCode(uint64_t RIP);
-    void ClearCache(FEXCore::Core::InternalThreadState *Thread);
-
-    // Public for threading
-    void ExecutionThread();
-
-  private:
-    FEXCore::Context::Context *CTX;
-    FEXCore::Core::InternalThreadState *ParentThread;
-
-    std::unique_ptr<FEXCore::Threads::Thread> WorkerThread;
-    std::unique_ptr<FEXCore::Core::InternalThreadState> CompileThreadData;
-
-    std::mutex QueueMutex{};
-    std::mutex CompileMutex{};
-    std::queue<WorkItem*> WorkQueue{};
-    std::vector<WorkItem*> GCArray{};
-    Event StartWork{};
-    std::atomic_bool ShuttingDown{false};
-};
-}

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"
@@ -11,38 +12,49 @@
 #include <FEXCore/Core/CoreState.h>
 #include <FEXCore/Core/X86Enums.h>
 #include <FEXCore/Debug/InternalThreadState.h>
+#include <FEXHeaderUtils/Syscalls.h>
+
 #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>
 
-#ifdef ENABLE_JITSYMBOLS
+#include <sys/syscall.h>
 #include <unistd.h>
-#endif
 
 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)
-  : Dispatcher(ctx, Thread), Arm64Emitter(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)
@@ -52,15 +64,9 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   //    Ptr();
   // }
 
-  uint64_t VirtualMemorySize = Thread->LookupCache->GetVirtualMemorySize();
-  Literal l_VirtualMemory {VirtualMemorySize};
-  Literal l_PagePtr {Thread->LookupCache->GetPagePointer()};
-  Literal l_L1Ptr {Thread->LookupCache->GetL1Pointer()};
-  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();
@@ -68,134 +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, &l_L1Ptr);
+  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 = CTX->Config.VirtualMemSize;
   if (std::popcount(VirtualMemorySize) == 1) {
-    and_(x3, RipReg, Thread->LookupCache->GetVirtualMemorySize() - 1);
+    and_(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r3, RipReg.R(), VirtualMemorySize - 1);
   }
   else {
-    ldr(x3, &l_VirtualMemory);
-    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, &l_L1Ptr);
+      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>();
@@ -209,36 +195,116 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
 
   {
     ExitFunctionLinkerAddress = GetCursorAddress<uint64_t>();
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
-    ldr(x0, &l_ExitFunctionLinkThis);
-    mov(x1, STATE);
-    mov(x2, lr);
+    if (SignalSafeCompile) {
+      // When compiling code, mask all signals to reduce the chance of reentrant allocations
+      // Args:
+      // X0: SETMASK
+      // X1: Pointer to mask value (uint64_t)
+      // X2: Pointer to old mask value (uint64_t)
+      // X3: Size of mask, sizeof(uint64_t)
+      // X8: Syscall
 
-    ldr(x3, &l_ExitFunctionLink);
-    blr(x3);
+      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);
+    }
 
-    if (SRAEnabled)
+    mov(ARMEmitter::XReg::x0, STATE);
+    mov(ARMEmitter::XReg::x1, ARMEmitter::XReg::lr);
+
+    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(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(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);
+      mov(ARMEmitter::XReg::x0, ARMEmitter::XReg::x4);
+    }
+
+    if (config.StaticRegisterAllocation)
       FillStaticRegs();
-    br(x0);
+
+    br(ARMEmitter::Reg::r0);
   }
 
   // Need to create the block
   {
-    bind(&NoBlock);
+    Bind(&NoBlock);
 
-    ldr(x0, &l_CTX);
-    mov(x1, STATE);
-    ldr(x3, &l_CompileBlock);
-
-    if (SRAEnabled)
+    if (config.StaticRegisterAllocation)
       SpillStaticRegs();
 
-    // X2 contains our guest RIP
-    blr(x3); // { CTX, Frame, RIP}
+    if (SignalSafeCompile) {
+      // When compiling code, mask all signals to reduce the chance of reentrant allocations
+      // Args:
+      // X0: SETMASK
+      // X1: Pointer to mask value (uint64_t)
+      // X2: Pointer to old mask value (uint64_t)
+      // X3: Size of mask, sizeof(uint64_t)
+      // X8: Syscall
 
-    if (SRAEnabled)
+      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(ARMEmitter::XReg::x2, ARMEmitter::Reg::rsp, 8);
+    }
+
+    ldr(ARMEmitter::XReg::x0, &l_CTX);
+    mov(ARMEmitter::XReg::x1, STATE);
+    ldr(ARMEmitter::XReg::x3, &l_CompileBlock);
+
+    // X2 contains our guest RIP
+#ifdef VIXL_SIMULATOR
+    GenerateIndirectRuntimeCall<void, void *, uint64_t, void *>(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(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(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::rsp, ARMEmitter::Reg::rsp, 16);
+    }
+
+    if (config.StaticRegisterAllocation)
       FillStaticRegs();
 
     b(&LoopTop);
@@ -253,20 +319,61 @@ Arm64Dispatcher::Arm64Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::
   }
 
   {
+    // Guest SIGILL handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGILL = GetCursorAddress<uint64_t>();
+
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs();
+
+    hlt(0);
+  }
+
+  {
+    // Guest SIGTRAP handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGTRAP  = GetCursorAddress<uint64_t>();
+
+    if (config.StaticRegisterAllocation)
+      SpillStaticRegs();
+
+    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
+    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
@@ -289,93 +396,273 @@ 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);
   }
 
-  place(&l_VirtualMemory);
-  place(&l_PagePtr);
-  place(&l_L1Ptr);
-  place(&l_CTX);
-  place(&l_Sleep);
-  place(&l_CompileBlock);
-  place(&l_ExitFunctionLink);
-  place(&l_ExitFunctionLinkThis);
+  {
+    LUDIVHandlerAddress = GetCursorAddress<uint64_t>();
 
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
+    SpillStaticRegs();
+
+    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
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  {
+    LDIVHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
+    SpillStaticRegs();
+
+    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
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  {
+    LUREMHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
+    SpillStaticRegs();
+
+    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
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  {
+    LREMHandlerAddress = GetCursorAddress<uint64_t>();
+
+    PushDynamicRegsAndLR(ARMEmitter::Reg::r3);
+    SpillStaticRegs();
+
+    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
+    // Fix the stack and any values that were stepped on
+    PopDynamicRegsAndLR();
+
+    // Go back to our code block
+    ret();
+  }
+
+  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 ENABLE_JITSYMBOLS
-  std::string Name = "Dispatch_" + std::to_string(::gettid());
-  CTX->Symbols.Register(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr), Name);
+  if (CTX->Config.BlockJITNaming()) {
+    std::string Name = "Dispatch_" + std::to_string(FHU::Syscalls::gettid());
+    CTX->Symbols.Register(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr), Name);
+  }
+  if (CTX->Config.GlobalJITNaming()) {
+    CTX->Symbols.RegisterJITSpace(reinterpret_cast<void*>(DispatchPtr), End - reinterpret_cast<uint64_t>(DispatchPtr));
+  }
+
+#ifdef VIXL_DISASSEMBLER
+  const auto DisasmEnd = GetCursorAddress<const vixl::aarch64::Instruction*>();
+  Disasm.DisassembleBuffer(DisasmBegin, DisasmEnd);
 #endif
 }
 
-void Arm64Dispatcher::SpillSRA(void *ucontext) {
-  for(int i = 0; i < SRA64.size(); i++) {
-    ThreadState->CurrentFrame->State.gregs[i] = ArchHelpers::Context::GetArmReg(ucontext, SRA64[i].GetCode());
-  }
-  // TODO: Also recover FPRs, not sure where the neon context is
-  // This is usually not needed
-  /*
-  for(int i = 0; i < SRAFPR.size(); i++) {
-      State->State.State.xmm[i][0] = _mcontext.neon[SRAFPR[i].GetCode()];
-      State->State.State.xmm[i][0] = _mcontext.neon[SRAFPR[i].GetCode()];
-  }
-  */
+#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) 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

@@ -1,5 +1,5 @@
 
-#include "Common/MathUtils.h"
+#include "Interface/Context/Context.h"
 #include "Interface/Core/ArchHelpers/MContext.h"
 #include "Interface/Core/Dispatcher/Dispatcher.h"
 #include "Interface/Core/X86HelperGen.h"
@@ -12,12 +12,13 @@
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/Event.h>
 #include <FEXCore/Utils/LogManager.h>
+#include <FEXCore/Utils/MathUtils.h>
 
 #include <atomic>
 #include <condition_variable>
-#include <bits/types/siginfo_t.h>
+#include <csignal>
+#include <cstring>
 #include <signal.h>
-#include <string.h>
 
 namespace FEXCore::CPU {
 
@@ -27,15 +28,19 @@ void Dispatcher::SleepThread(FEXCore::Context::Context *ctx, FEXCore::Core::CpuS
   --ctx->IdleWaitRefCount;
   ctx->IdleWaitCV.notify_all();
 
+  Thread->RunningEvents.ThreadSleeping = true;
+
   // Go to sleep
   Thread->StartRunning.Wait();
 
   Thread->RunningEvents.Running = true;
   ++ctx->IdleWaitRefCount;
+  Thread->RunningEvents.ThreadSleeping = false;
+
   ctx->IdleWaitCV.notify_all();
 }
 
-void 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);
@@ -60,26 +65,212 @@ void Dispatcher::StoreThreadState(int Signal, void *ucontext) {
   // 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);
 
-  SignalFrames.push(NewSP);
+  // Signal frames are only used on the interpreter
+  // The JITS require the stack to be setup correctly on rt_sigreturn
+  if (CTX->Config.Core() == FEXCore::Config::CONFIG_INTERPRETER) {
+    SignalFrames.push(NewSP);
+  }
+
+  Context->Flags = 0;
+  Context->FPStateLocation = 0;
+  Context->UContextLocation = 0;
+  Context->SigInfoLocation = 0;
+
+  // Store fault to top status and then reset it
+  Context->FaultToTopAndGeneratedException = Thread->CurrentFrame->SynchronousFaultData.FaultToTopAndGeneratedException;
+  Thread->CurrentFrame->SynchronousFaultData.FaultToTopAndGeneratedException = false;
+
+  return Context;
 }
 
-void Dispatcher::RestoreThreadState(void *ucontext) {
-  LOGMAN_THROW_A(!SignalFrames.empty(), "Trying to restore a signal frame when we don't have any");
-  uint64_t OldSP = SignalFrames.top();
-  SignalFrames.pop();
+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);
+  }
+  else {
+    LOGMAN_THROW_A_FMT(!SignalFrames.empty(), "Trying to restore a signal frame when we don't have any");
+    OldSP = SignalFrames.top();
+    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 = Thread->CurrentFrame;
+
+    if (Context->Flags &ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_INJIT) {
+      // XXX: Unsupported since it needs state reconstruction
+      // If we are in the JIT then SRA might need to be restored to values from the context
+      // We can't currently support this since it might result in tearing without real state reconstruction
+    }
+
+    if (!(Context->Flags & ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_32BIT)) {
+      auto *guest_uctx = reinterpret_cast<FEXCore::x86_64::ucontext_t*>(Context->UContextLocation);
+      [[maybe_unused]] auto *guest_siginfo = reinterpret_cast<siginfo_t*>(Context->SigInfoLocation);
+
+      // If the guest modified the RIP then we need to take special precautions here
+      if (Context->OriginalRIP != guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_RIP] ||
+          Context->FaultToTopAndGeneratedException) {
+        // Hack! Go back to the top of the dispatcher top
+        // This is only safe inside the JIT rather than anything outside of it
+        ArchHelpers::Context::SetPc(ucontext, AbsoluteLoopTopAddressFillSRA);
+        // Set our state register to point to our guest thread data
+        ArchHelpers::Context::SetState(ucontext, reinterpret_cast<uint64_t>(Frame));
+
+        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 < Core::CPUState::NUM_EFLAG_BITS; ++i) {
+          Frame->State.flags[i] = (eflags & (1U << i)) ? 1 : 0;
+        }
+
+        Frame->State.flags[1] = 1;
+        Frame->State.flags[9] = 1;
+
+#define COPY_REG(x) \
+            Frame->State.gregs[X86State::REG_##x] = guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_##x];
+            COPY_REG(R8);
+            COPY_REG(R9);
+            COPY_REG(R10);
+            COPY_REG(R11);
+            COPY_REG(R12);
+            COPY_REG(R13);
+            COPY_REG(R14);
+            COPY_REG(R15);
+            COPY_REG(RDI);
+            COPY_REG(RSI);
+            COPY_REG(RBP);
+            COPY_REG(RBX);
+            COPY_REG(RDX);
+            COPY_REG(RAX);
+            COPY_REG(RCX);
+            COPY_REG(RSP);
+#undef COPY_REG
+        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));
+
+        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;
+        Frame->State.FTW = fpstate->ftw;
+
+        // Deconstruct FSW
+        Frame->State.flags[FEXCore::X86State::X87FLAG_C0_LOC] = (fpstate->fsw >> 8) & 1;
+        Frame->State.flags[FEXCore::X86State::X87FLAG_C1_LOC] = (fpstate->fsw >> 9) & 1;
+        Frame->State.flags[FEXCore::X86State::X87FLAG_C2_LOC] = (fpstate->fsw >> 10) & 1;
+        Frame->State.flags[FEXCore::X86State::X87FLAG_C3_LOC] = (fpstate->fsw >> 14) & 1;
+        Frame->State.flags[FEXCore::X86State::X87FLAG_TOP_LOC] = (fpstate->fsw >> 11) & 0b111;
+      }
+    }
+    else {
+      auto *guest_uctx = reinterpret_cast<FEXCore::x86::ucontext_t*>(Context->UContextLocation);
+      [[maybe_unused]] auto *guest_siginfo = reinterpret_cast<FEXCore::x86::siginfo_t*>(Context->SigInfoLocation);
+      // If the guest modified the RIP then we need to take special precautions here
+      if (Context->OriginalRIP != guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EIP] ||
+          Context->FaultToTopAndGeneratedException) {
+        // Hack! Go back to the top of the dispatcher top
+        // This is only safe inside the JIT rather than anything outside of it
+        ArchHelpers::Context::SetPc(ucontext, AbsoluteLoopTopAddressFillSRA);
+        // Set our state register to point to our guest thread data
+        ArchHelpers::Context::SetState(ucontext, reinterpret_cast<uint64_t>(Frame));
+
+        // 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 < Core::CPUState::NUM_EFLAG_BITS; ++i) {
+          Frame->State.flags[i] = (eflags & (1U << i)) ? 1 : 0;
+        }
+
+        Frame->State.flags[1] = 1;
+        Frame->State.flags[9] = 1;
+
+        Frame->State.rip = guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_EIP];
+        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);
+        COPY_REG(RSI);
+        COPY_REG(RBP);
+        COPY_REG(RBX);
+        COPY_REG(RDX);
+        COPY_REG(RAX);
+        COPY_REG(RCX);
+        COPY_REG(RSP);
+#undef COPY_REG
+        auto *xstate = reinterpret_cast<x86::xstate*>(guest_uctx->uc_mcontext.fpregs);
+        auto *fpstate = &xstate->fpstate;
+
+        // Copy float registers
+        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
+        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;
+        Frame->State.FTW = fpstate->ftw;
+
+        // Deconstruct FSW
+        Frame->State.flags[FEXCore::X86State::X87FLAG_C0_LOC] = (fpstate->fsw >> 8) & 1;
+        Frame->State.flags[FEXCore::X86State::X87FLAG_C1_LOC] = (fpstate->fsw >> 9) & 1;
+        Frame->State.flags[FEXCore::X86State::X87FLAG_C2_LOC] = (fpstate->fsw >> 10) & 1;
+        Frame->State.flags[FEXCore::X86State::X87FLAG_C3_LOC] = (fpstate->fsw >> 14) & 1;
+        Frame->State.flags[FEXCore::X86State::X87FLAG_TOP_LOC] = (fpstate->fsw >> 11) & 0b111;
+      }
+    }
+  }
 }
 
 static uint32_t ConvertSignalToTrapNo(int Signal, siginfo_t *HostSigInfo) {
@@ -97,15 +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;
   }
@@ -114,13 +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) {
-  StoreThreadState(Signal, ucontext);
-  auto Frame = ThreadState->CurrentFrame;
+template <typename T>
+static void SetXStateInfo(T* xstate, bool is_avx_enabled) {
+  auto* fpstate = &xstate->fpstate;
+
+  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
@@ -132,23 +343,51 @@ 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) {
+        // We are in a syscall, this means we are in a weird register state
+        // We need to spill SRA but only some of it, since some values have already been spilled
+        // Lower 16 bits tells us which registers are already spilled to the context
+        // So we ignore spilling those ones
+        uint16_t NumRegisters = std::popcount(Frame->InSyscallInfo & 0xFFFF);
+        if (NumRegisters >= 4) {
+          // Unhandled case
+          IgnoreMask = 0;
+        }
+        else {
+          IgnoreMask = Frame->InSyscallInfo & 0xFFFF;
+        }
+      }
+      else {
+        // We must spill everything
+        IgnoreMask = 0;
+      }
+#endif
+
       // We are in jit, SRA must be spilled
-      SpillSRA(ucontext);
+      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
         // Only throw a log message in this case
-        LogMan::Msg::E("Signals in dispatcher have unsynchronized context");
+        if constexpr (false) {
+          // XXX: Messages in the signal handler can cause us to crash
+          LogMan::Msg::EFmt("Signals in dispatcher have unsynchronized context");
+        }
       }
     }
   }
@@ -180,12 +419,19 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
   // siginfo_t
   siginfo_t *HostSigInfo = reinterpret_cast<siginfo_t*>(info);
 
-  if (GuestAction->sa_flags & SA_SIGINFO) {
+  // Backup where we think the RIP currently is
+  ContextBackup->OriginalRIP = Frame->State.rip;
 
+  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);
@@ -196,6 +442,10 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       NewGuestSP = AlignDown(NewGuestSP, alignof(siginfo_t));
       uint64_t SigInfoLocation = NewGuestSP;
 
+      ContextBackup->FPStateLocation = FPStateLocation;
+      ContextBackup->UContextLocation = UContextLocation;
+      ContextBackup->SigInfoLocation = SigInfoLocation;
+
       FEXCore::x86_64::ucontext_t *guest_uctx = reinterpret_cast<FEXCore::x86_64::ucontext_t*>(UContextLocation);
       siginfo_t *guest_siginfo = reinterpret_cast<siginfo_t*>(SigInfoLocation);
 
@@ -203,14 +453,31 @@ 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;
       guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_CSGSFS] = 0;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_ERR] = ConvertSignalToError(Signal, HostSigInfo);
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86_64::FEX_REG_TRAPNO] = ConvertSignalToTrapNo(Signal, HostSigInfo);
+
+      // aarch64 and x86_64 siginfo_t matches. We can just copy this over
+      // SI_USER could also potentially have random data in it, needs to be bit perfect
+      // For guest faults we don't have a real way to reconstruct state to a real guest RIP
+      *guest_siginfo = *HostSigInfo;
+
+      if (ContextBackup->FaultToTopAndGeneratedException) {
+        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 = 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(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;
 
@@ -234,9 +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;
@@ -255,17 +534,19 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       guest_uctx->uc_stack.ss_sp = GuestStack->ss_sp;
       guest_uctx->uc_stack.ss_size = GuestStack->ss_size;
 
-      // aarch64 and x86_64 siginfo_t matches. We can just copy this over
-      // SI_USER could also potentially have random data in it, needs to be bit perfect
-      // For guest faults we don't have a real way to reconstruct state to a real guest RIP
-      *guest_siginfo = *HostSigInfo;
-
       Frame->State.gregs[X86State::REG_RSI] = SigInfoLocation;
       Frame->State.gregs[X86State::REG_RDX] = UContextLocation;
     }
     else {
-      NewGuestSP -= sizeof(FEXCore::x86::_libc_fpstate);
-      NewGuestSP = AlignDown(NewGuestSP, alignof(FEXCore::x86::_libc_fpstate));
+      ContextBackup->Flags |= ArchHelpers::Context::ContextFlags::CONTEXT_FLAG_32BIT;
+
+      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);
@@ -276,6 +557,10 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       NewGuestSP = AlignDown(NewGuestSP, alignof(FEXCore::x86::siginfo_t));
       uint64_t SigInfoLocation = NewGuestSP;
 
+      ContextBackup->FPStateLocation = FPStateLocation;
+      ContextBackup->UContextLocation = UContextLocation;
+      ContextBackup->SigInfoLocation = SigInfoLocation;
+
       FEXCore::x86::ucontext_t *guest_uctx = reinterpret_cast<FEXCore::x86::ucontext_t*>(UContextLocation);
       FEXCore::x86::siginfo_t *guest_siginfo = reinterpret_cast<FEXCore::x86::siginfo_t*>(SigInfoLocation);
 
@@ -284,19 +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_GS] = Frame->State.gs;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_FS] = Frame->State.fs;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ES] = Frame->State.es;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS] = Frame->State.ds;
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = ConvertSignalToTrapNo(Signal, HostSigInfo);
-      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ERR] = ConvertSignalToError(Signal, HostSigInfo);
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_CS] = Frame->State.cs_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_DS] = Frame->State.ds_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_ES] = Frame->State.es_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_FS] = Frame->State.fs_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_GS] = Frame->State.gs_idx;
+      guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_SS] = Frame->State.ss_idx;
+
+      if (ContextBackup->FaultToTopAndGeneratedException) {
+        guest_uctx->uc_mcontext.gregs[FEXCore::x86::FEX_REG_TRAPNO] = Frame->SynchronousFaultData.TrapNo;
+        guest_siginfo->si_code = Frame->SynchronousFaultData.si_code;
+        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(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];
@@ -310,15 +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;
@@ -339,7 +646,6 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
       // These three elements are in every siginfo
       guest_siginfo->si_signo = HostSigInfo->si_signo;
       guest_siginfo->si_errno = HostSigInfo->si_errno;
-      guest_siginfo->si_code = HostSigInfo->si_code;
 
       switch (Signal) {
         case SIGSEGV:
@@ -368,7 +674,7 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
           guest_siginfo->_sifields._timer.sigval.sival_int = HostSigInfo->si_int;
           break;
         default:
-          LogMan::Msg::E("Unhandled siginfo_t for signal: %d\n", Signal);
+          LogMan::Msg::EFmt("Unhandled siginfo_t for signal: {}\n", Signal);
           break;
       }
 
@@ -402,14 +708,14 @@ bool Dispatcher::HandleGuestSignal(int Signal, void *info, void *ucontext, Guest
   else {
     NewGuestSP -= 4;
     *(uint32_t*)NewGuestSP = SignalReturn;
-    LOGMAN_THROW_A(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;
@@ -417,44 +723,45 @@ 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(!IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), true), "Signals in dispatcher have unsynchronized context");
+        LOGMAN_THROW_A_FMT(!IsAddressInDispatcher(ArchHelpers::Context::GetPc(ucontext)),
+                           "Signals in dispatcher have unsynchronized context");
       }
       ArchHelpers::Context::SetPc(ucontext, ThreadPauseHandlerAddress);
     }
@@ -464,9 +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;
   }
 
@@ -477,32 +784,42 @@ 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(!IsAddressInJITCode(ArchHelpers::Context::GetPc(ucontext), true), "Signals in dispatcher have unsynchronized context");
+        LOGMAN_THROW_A_FMT(!IsAddressInDispatcher(ArchHelpers::Context::GetPc(ucontext)),
+                           "Signals in dispatcher have unsynchronized context");
       }
       ArchHelpers::Context::SetPc(ucontext, ThreadStopHandlerAddress);
     }
 
-    ThreadState->SignalReason.store(FEXCore::Core::SignalEvent::Nothing);
+    // 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 (Thread->RunningEvents.ThreadSleeping) {
+      // If the thread was sleeping then its idle counter was decremented
+      // Reincrement it here to not break logic
+      ++Thread->CTX->IdleWaitRefCount;
+    }
+
+    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;
   }
 
@@ -521,27 +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) {
-    return IsAddressInDispatcher(Address);
-  }
-  return false;
-}
-
 }

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

@@ -1,11 +1,10 @@
 #pragma once
 
 #include <FEXCore/Core/CPUBackend.h>
+#include "Interface/Core/ArchHelpers/MContext.h"
 
-#include "Interface/Context/Context.h"
-
-#include <bits/types/stack_t.h>
 #include <cstdint>
+#include <signal.h>
 #include <stddef.h>
 #include <stack>
 #include <tuple>
@@ -20,21 +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
@@ -47,52 +44,70 @@ public:
   uint64_t ThreadPauseHandlerAddressSpillSRA{};
   uint64_t ExitFunctionLinkerAddress{};
   uint64_t SignalHandlerReturnAddress{};
+  uint64_t GuestSignal_SIGILL{};
+  uint64_t GuestSignal_SIGTRAP{};
+  uint64_t GuestSignal_SIGSEGV{};
+  uint64_t IntCallbackReturnAddress{};
+
   uint64_t PauseReturnInstruction{};
 
   /**  @} */
 
-  uint32_t SignalHandlerRefCounter{};
-
   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;
   }
 
+  virtual void InitThreadPointers(FEXCore::Core::InternalThreadState *Thread) = 0;
+
+  // 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, FEXCore::Core::InternalThreadState *Thread)
+  Dispatcher(FEXCore::Context::Context *ctx, const DispatcherConfig &Config)
     : CTX {ctx}
-    , ThreadState {Thread} {}
+    , config {Config}
+    {}
 
-  void StoreThreadState(int Signal, void *ucontext);
-  void RestoreThreadState(void *ucontext);
-  std::stack<uint64_t> SignalFrames;
+  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) {}
+  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

@@ -11,27 +11,33 @@
 #include <FEXCore/Core/CPUBackend.h>
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/Allocator.h>
+#include <FEXHeaderUtils/Syscalls.h>
 
 #include <cmath>
 #include <memory>
 #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,
@@ -77,11 +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;
@@ -91,30 +96,26 @@ 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, Thread->LookupCache->GetL1Pointer());
+    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 = CTX->Config.VirtualMemSize;
     mov(rax, rdx);
-    mov(rbx, Thread->LookupCache->GetVirtualMemorySize() - 1);
+    mov(rbx, VirtualMemorySize - 1);
     and_(rax, rbx);
     shr(rax, 12);
 
@@ -141,8 +142,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     je(NoBlock);
 
     // Update L1
-    
-    mov(r13, Thread->LookupCache->GetL1Pointer());
+    mov(r13, qword STATE_PTR(CpuStateFrame, Pointers.Common.L1Pointer));
     mov(rcx, rdx);
     and_(rcx, LookupCache::L1_ENTRIES_MASK);
     shl(rcx, 1);
@@ -150,30 +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);
   }
 
   {
@@ -192,10 +169,37 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
     ret();
   }
 
+  constexpr bool SignalSafeCompile = true;
   // Block creation
   {
     L(NoBlock);
 
+    if (SignalSafeCompile) {
+      // When compiling code, mask all signals to reduce the chance of reentrant allocations
+      // RDI: SETMASK
+      // RSI: Pointer to mask value (uint64_t)
+      // RDX: Pointer to old mask value (uint64_t)
+      // R10: Size of mask, sizeof(uint64_t)
+      // RAX: Syscall
+
+      // Backup rdx
+      mov(r9, rdx);
+
+      mov(rdi, ~0ULL);
+      sub(rsp, 16);
+      mov(qword [rsp], rdi);
+      mov(qword [rsp + 8], rdi);
+
+      mov(rdi, SIG_SETMASK);
+      mov(rsi, rsp);
+      mov(rdx, rsp);
+      mov(r10, 8);
+      mov(rax, SYS_rt_sigprocmask);
+      syscall();
+
+      mov(rdx, r9);
+    }
+
     // {rdi, rsi, rdx}
     mov(rdi, reinterpret_cast<uint64_t>(CTX));
     mov(rsi, STATE);
@@ -203,20 +207,84 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
 
     call(rax);
 
+    if (SignalSafeCompile) {
+      // Now restore the signal mask
+      // Living in the same location
+      // Backup rdx
+      mov(r9, rdx);
+
+      mov(rdi, SIG_SETMASK);
+      mov(rsi, rsp);
+      mov(rdx, 0); // Don't care about result
+      mov(r10, 8);
+      mov(rax, SYS_rt_sigprocmask);
+      syscall();
+
+      // Bring stack back
+      add(rsp, 16);
+
+      mov(rdx, r9);
+    }
+
     // rdx already contains RIP here
     jmp(LoopTop);
   }
 
   {
     ExitFunctionLinkerAddress = getCurr<uint64_t>();
-    // {rdi, rsi, rdx}
-    mov(rdi, config.ExitFunctionLinkThis);
-    mov(rsi, STATE);
-    mov(rdx, rax); // rax is set at the block end
+    if (SignalSafeCompile) {
+      // When compiling code, mask all signals to reduce the chance of reentrant allocations
+      // RDI: SETMASK
+      // RSI: Pointer to mask value (uint64_t)
+      // RDX: Pointer to old mask value (uint64_t)
+      // R10: Size of mask, sizeof(uint64_t)
+      // RAX: Syscall
 
-    mov(rax, config.ExitFunctionLink);
-    call(rax);
-    jmp(rax);
+      // Backup rax
+      mov(r9, rax);
+
+      mov(rdi, ~0ULL);
+      sub(rsp, 16);
+      mov(qword [rsp], rdi);
+      mov(qword [rsp + 8], rdi);
+
+      mov(rdi, SIG_SETMASK);
+      mov(rsi, rsp);
+      mov(rdx, rsp);
+      mov(r10, 8);
+      mov(rax, SYS_rt_sigprocmask);
+      syscall();
+
+      mov(rax, r9);
+    }
+
+    // {rdi, rsi}
+    mov(rdi, STATE);
+    mov(rsi, rax); // rax is set at the block end
+
+    call(qword STATE_PTR(CpuStateFrame, Pointers.Common.ExitFunctionLink));
+
+    if (SignalSafeCompile) {
+      // Now restore the signal mask
+      // Living in the same location
+      // Backup rax
+      mov(r9, rax);
+
+      mov(rdi, SIG_SETMASK);
+      mov(rsi, rsp);
+      mov(rdx, 0); // Don't care about result
+      mov(r10, 8);
+      mov(rax, SYS_rt_sigprocmask);
+      syscall();
+
+      // Bring stack back
+      add(rsp, 16);
+
+      jmp(r9);
+    }
+    else {
+      jmp(rax);
+    }
   }
 
   {
@@ -236,7 +304,7 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
   }
 
   {
-    CallbackPtr = getCurr<CPUBackend::JITCallback>();
+    CallbackPtr = getCurr<JITCallback>();
 
     push(rbx);
     push(rbp);
@@ -251,8 +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
-    mov(rax, reinterpret_cast<uint64_t>(&SignalHandlerRefCounter));
-    add(dword [rax], 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
@@ -260,12 +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);
@@ -274,13 +341,40 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
   {
     // Signal return handler
     SignalHandlerReturnAddress = getCurr<uint64_t>();
-
     ud2();
   }
 
+  {
+    // Guest SIGILL handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGILL = getCurr<uint64_t>();
+    ud2();
+  }
 
   {
-    ReturnPtr = getCurr<FEXCore::Context::Context::IntCallbackReturn>();
+    // Guest SIGTRAP handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGTRAP = getCurr<uint64_t>();
+
+    // ud2 = SIGILL
+    // int3 = SIGTRAP
+    // hlt = SIGSEGV
+    int3();
+  }
+
+  {
+    // Guest SIGSEGV handler
+    // Needs to be distinct from the SignalHandlerReturnAddress
+    GuestSignal_SIGSEGV = getCurr<uint64_t>();
+
+    // ud2 = SIGILL
+    // int3 = SIGTRAP
+    // hlt = SIGSEGV
+    hlt();
+  }
+
+  {
+    IntCallbackReturnAddress = getCurr<uint64_t>();
 //  using CallbackReturn =  FEX_NAKED void(*)(FEXCore::Core::InternalThreadState *Thread, volatile void *Host_RSP);
 
     // rdi = thread
@@ -306,30 +400,100 @@ X86Dispatcher::X86Dispatcher(FEXCore::Context::Context *ctx, FEXCore::Core::Inte
   Start = reinterpret_cast<uint64_t>(getCode());
   End = Start + getSize();
 
-  #if ENABLE_JITSYMBOLS
-    std::string Name = "Dispatch_" + std::to_string(::gettid());
+  if (CTX->Config.BlockJITNaming()) {
+    std::string Name = "Dispatch_" + std::to_string(FHU::Syscalls::gettid());
     CTX->Symbols.Register(reinterpret_cast<void*>(Start), End-Start, Name);
-  #endif
+  }
+  if (CTX->Config.GlobalJITNaming()) {
+    CTX->Symbols.RegisterJITSpace(reinterpret_cast<void*>(Start), End-Start);
+  }
+
+}
+
+// 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>
 
@@ -127,24 +129,67 @@ static uint32_t MapModRMToReg(uint8_t REX, uint8_t bits, bool HighBits, bool Has
   return (*GPRs)[(REX << 3) | bits];
 }
 
+static uint32_t MapVEXToReg(uint8_t vvvv, bool HasXMM) {
+  using GPRArray = std::array<uint32_t, 16>;
+
+  static constexpr GPRArray GPRIndexes = {
+    FEXCore::X86State::REG_RAX,
+    FEXCore::X86State::REG_RCX,
+    FEXCore::X86State::REG_RDX,
+    FEXCore::X86State::REG_RBX,
+    FEXCore::X86State::REG_RSP,
+    FEXCore::X86State::REG_RBP,
+    FEXCore::X86State::REG_RSI,
+    FEXCore::X86State::REG_RDI,
+    FEXCore::X86State::REG_R8,
+    FEXCore::X86State::REG_R9,
+    FEXCore::X86State::REG_R10,
+    FEXCore::X86State::REG_R11,
+    FEXCore::X86State::REG_R12,
+    FEXCore::X86State::REG_R13,
+    FEXCore::X86State::REG_R14,
+    FEXCore::X86State::REG_R15,
+  };
+
+  static constexpr GPRArray XMMIndexes = {
+    FEXCore::X86State::REG_XMM_0,
+    FEXCore::X86State::REG_XMM_1,
+    FEXCore::X86State::REG_XMM_2,
+    FEXCore::X86State::REG_XMM_3,
+    FEXCore::X86State::REG_XMM_4,
+    FEXCore::X86State::REG_XMM_5,
+    FEXCore::X86State::REG_XMM_6,
+    FEXCore::X86State::REG_XMM_7,
+    FEXCore::X86State::REG_XMM_8,
+    FEXCore::X86State::REG_XMM_9,
+    FEXCore::X86State::REG_XMM_10,
+    FEXCore::X86State::REG_XMM_11,
+    FEXCore::X86State::REG_XMM_12,
+    FEXCore::X86State::REG_XMM_13,
+    FEXCore::X86State::REG_XMM_14,
+    FEXCore::X86State::REG_XMM_15,
+  };
+
+  if (HasXMM) {
+    return XMMIndexes[vvvv];
+  } else {
+    return GPRIndexes[vvvv];
+  }
+}
+
 Decoder::Decoder(FEXCore::Context::Context *ctx)
   : CTX {ctx}
-  , OSABI { ctx->SyscallHandler ? ctx->SyscallHandler->GetOSABI() : FEXCore::HLE::SyscallOSABI::OS_UNKNOWN } {
-  // Using mmap is a start-up time optimization
-  // Take advantage of page faulting to reduce startup time for minimal runtime cost
-  DecodedBuffer =
-    reinterpret_cast<FEXCore::X86Tables::DecodedInst *>(
-      FEXCore::Allocator::mmap(0, sizeof(FEXCore::X86Tables::DecodedInst) * DefaultDecodedBufferSize,
-      PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0));
+  , OSABI { ctx->SyscallHandler ? ctx->SyscallHandler->GetOSABI() : FEXCore::HLE::SyscallOSABI::OS_UNKNOWN }
+  , PoolObject {ctx->FrontendAllocator, sizeof(FEXCore::X86Tables::DecodedInst) * DefaultDecodedBufferSize} {
 }
 
 Decoder::~Decoder() {
-  FEXCore::Allocator::munmap(DecodedBuffer, sizeof(FEXCore::X86Tables::DecodedInst) * DefaultDecodedBufferSize);
+  PoolObject.UnclaimBuffer();
 }
 
 uint8_t Decoder::ReadByte() {
   uint8_t Byte = InstStream[InstructionSize];
-  LOGMAN_THROW_A(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;
@@ -156,14 +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("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);
@@ -303,21 +341,21 @@ 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);
 
-    uint64_t Literal {0};
-    LOGMAN_THROW_A(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");
 
-    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"
     if (ModRM.rm == 0b101) {
       // 32bit Displacement
-      uint32_t Literal;
-      Literal = ReadData(4);
+      const uint32_t Literal = ReadData(4);
 
       Operand->Type = DecodedOperand::OpType::RIPRelative;
       Operand->Data.RIPLiteral.Value.u = Literal;
@@ -330,8 +368,7 @@ void Decoder::DecodeModRM_64(X86Tables::DecodedOperand *Operand, X86Tables::ModR
   }
   else {
     uint8_t DisplacementSize = ModRM.mod == 1 ? 1 : 4;
-    uint32_t Literal{};
-    Literal = ReadData(DisplacementSize);
+    uint32_t Literal = ReadData(DisplacementSize);
     if (DisplacementSize == 1) {
       Literal = static_cast<int8_t>(Literal);
     }
@@ -343,32 +380,33 @@ void Decoder::DecodeModRM_64(X86Tables::DecodedOperand *Operand, X86Tables::ModR
   }
 }
 
-bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op) {
+bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op, DecodedHeader Options) {
   DecodeInst->OP = Op;
   DecodeInst->TableInfo = Info;
 
   // XXX: Once we support 32bit x86 then this will be necessary to support
   if (Info->Type == FEXCore::X86Tables::TYPE_LEGACY_PREFIX) {
-    LogMan::Msg::D("Legacy Prefix");
+    LogMan::Msg::DFmt("Legacy Prefix");
     return false;
   }
 
   if (Info->Type == FEXCore::X86Tables::TYPE_UNKNOWN) {
-    LogMan::Msg::D("Unknown instruction: %s 0x%04x 0x%lx", Info->Name, Op, DecodeInst->PC);
+    LogMan::Msg::DFmt("Unknown instruction: {} 0x{:04x} 0x{:x}", Info->Name ?: "UND", Op, DecodeInst->PC);
     return false;
   }
 
   if (Info->Type == FEXCore::X86Tables::TYPE_INVALID) {
-    LogMan::Msg::D("Invalid or Unknown instruction: %s 0x%04x 0x%lx", Info->Name, Op, DecodeInst->PC);
+    LogMan::Msg::DFmt("Invalid or Unknown instruction: {} 0x{:04x} 0x{:x}", Info->Name ?: "UND", Op, DecodeInst->PC);
     return false;
   }
 
-  LOGMAN_THROW_A(!(Info->Type >= FEXCore::X86Tables::TYPE_GROUP_1 && Info->Type <= FEXCore::X86Tables::TYPE_GROUP_P),
-    "Group Ops should have been decoded before this!");
+  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{};
-  bool HasWideningDisplacement = FEXCore::X86Tables::DecodeFlags::GetOpAddr(DecodeInst->Flags, 0) & FEXCore::X86Tables::DecodeFlags::FLAG_WIDENING_SIZE_LAST;
-  bool HasNarrowingDisplacement = FEXCore::X86Tables::DecodeFlags::GetOpAddr(DecodeInst->Flags, 0) & FEXCore::X86Tables::DecodeFlags::FLAG_OPERAND_SIZE_LAST;
+  const bool HasWideningDisplacement = (FEXCore::X86Tables::DecodeFlags::GetOpAddr(DecodeInst->Flags, 0) & FEXCore::X86Tables::DecodeFlags::FLAG_WIDENING_SIZE_LAST) != 0 ||
+                                       (Options.w && CTX->Config.Is64BitMode);
+  const bool HasNarrowingDisplacement = (FEXCore::X86Tables::DecodeFlags::GetOpAddr(DecodeInst->Flags, 0) & FEXCore::X86Tables::DecodeFlags::FLAG_OPERAND_SIZE_LAST) != 0;
 
   bool HasXMMSrc = !!(Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_XMM_FLAGS) &&
     !HAS_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_SRC_GPR) &&
@@ -401,8 +439,8 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op)
   // New instruction size decoding
   {
     // Decode destinations first
-    uint32_t DstSizeFlag = FEXCore::X86Tables::InstFlags::GetSizeDstFlags(Info->Flags);
-    uint32_t SrcSizeFlag = FEXCore::X86Tables::InstFlags::GetSizeSrcFlags(Info->Flags);
+    const auto DstSizeFlag = FEXCore::X86Tables::InstFlags::GetSizeDstFlags(Info->Flags);
+    const auto SrcSizeFlag = FEXCore::X86Tables::InstFlags::GetSizeSrcFlags(Info->Flags);
 
     if (DstSizeFlag == FEXCore::X86Tables::InstFlags::SIZE_8BIT) {
       DecodeInst->Flags |= DecodeFlags::GenSizeDstSize(DecodeFlags::SIZE_8BIT);
@@ -413,8 +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 ||
@@ -445,7 +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 ||
@@ -486,7 +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(!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
@@ -537,8 +591,11 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op)
         return false;
     }
     else {
-      auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
-      (this->*Disp)(&NonGPR, ModRM);
+      // Only decode if we haven't pre-decoded
+      if (NonGPR.IsNone()) {
+        auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
+        (this->*Disp)(&NonGPR, ModRM);
+      }
     }
 
     return true;
@@ -546,6 +603,13 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op)
 
   size_t CurrentSrc = 0;
 
+  if ((Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_VEX_1ST_SRC) != 0) {
+    DecodeInst->Src[CurrentSrc].Type = DecodedOperand::OpType::GPR;
+    DecodeInst->Src[CurrentSrc].Data.GPR.HighBits = false;
+    DecodeInst->Src[CurrentSrc].Data.GPR.GPR = MapVEXToReg(Options.vvvv, HasXMMSrc);
+    ++CurrentSrc;
+  }
+
   if (Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_MODRM) {
     if (Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_SF_MOD_DST) {
       if (!ModRMOperand(DecodeInst->Src[CurrentSrc], DecodeInst->Dest, HasXMMSrc, HasXMMDst, HasMMSrc, HasMMDst, Is8BitSrc, Is8BitDest))
@@ -558,6 +622,13 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op)
     ++CurrentSrc;
   }
 
+  if ((Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_VEX_2ND_SRC) != 0) {
+    DecodeInst->Src[CurrentSrc].Type = DecodedOperand::OpType::GPR;
+    DecodeInst->Src[CurrentSrc].Data.GPR.HighBits = false;
+    DecodeInst->Src[CurrentSrc].Data.GPR.GPR = MapVEXToReg(Options.vvvv, HasXMMSrc);
+    ++CurrentSrc;
+  }
+
   if (HAS_NON_XMM_SUBFLAG(Info->Flags, FEXCore::X86Tables::InstFlags::FLAGS_SF_SRC_RAX)) {
     DecodeInst->Src[CurrentSrc].Type = DecodedOperand::OpType::GPR;
     DecodeInst->Src[CurrentSrc].Data.GPR.HighBits = false;
@@ -571,8 +642,14 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op)
     ++CurrentSrc;
   }
 
+  if ((Info->Flags & FEXCore::X86Tables::InstFlags::FLAGS_VEX_DST) != 0) {
+    CurrentDest->Type = DecodedOperand::OpType::GPR;
+    CurrentDest->Data.GPR.HighBits = false;
+    CurrentDest->Data.GPR.GPR = MapVEXToReg(Options.vvvv, HasXMMDst);
+  }
+
   if (Bytes != 0) {
-    LOGMAN_THROW_A(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;
 
@@ -597,7 +674,8 @@ bool Decoder::NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op)
     DecodeInst->Src[CurrentSrc].Data.Literal.Value = Literal;
   }
 
-  LOGMAN_THROW_A(Bytes == 0, "Inst at 0x%lx: 0x%04x '%s' Had an instruction of size %d with %d remaining", DecodeInst->PC, DecodeInst->OP, DecodeInst->TableInfo->Name, InstructionSize, Bytes);
+  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;
 }
@@ -608,21 +686,22 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
 
   // XXX: Once we support 32bit x86 then this will be necessary to support
   if (Info->Type == FEXCore::X86Tables::TYPE_LEGACY_PREFIX) {
-    LogMan::Msg::D("Legacy Prefix");
+    LogMan::Msg::DFmt("Legacy Prefix");
     return false;
   }
 
   if (Info->Type == FEXCore::X86Tables::TYPE_UNKNOWN) {
-    LogMan::Msg::D("Unknown instruction: %s 0x%04x 0x%lx", Info->Name, Op, DecodeInst->PC);
+    LogMan::Msg::DFmt("Unknown instruction: {} 0x{:04x} 0x{:x}", Info->Name ?: "UND", Op, DecodeInst->PC);
     return false;
   }
 
   if (Info->Type == FEXCore::X86Tables::TYPE_INVALID) {
-    LogMan::Msg::D("Invalid or Unknown instruction: %s 0x%04x 0x%lx", Info->Name, Op, DecodeInst->PC);
+    LogMan::Msg::DFmt("Invalid or Unknown instruction: {} 0x{:04x} 0x{:x}", Info->Name ?: "UND", Op, DecodeInst->PC);
     return false;
   }
 
-  LOGMAN_THROW_A(Info->Type != FEXCore::X86Tables::TYPE_REX_PREFIX, "REX PREFIX should have been decoded before this!");
+  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 &&
       Info->Type <= FEXCore::X86Tables::TYPE_GROUP_11) {
@@ -678,7 +757,7 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
         3,
       };
       uint8_t Field = RegToField[ModRM.reg];
-      LOGMAN_THROW_A(Field != 255, "Invalid field selected!");
+      LOGMAN_THROW_AA_FMT(Field != 255, "Invalid field selected!");
 
       LocalOp = (Field << 3) | ModRM.rm;
       return NormalOp(&SecondModRMTableOps[LocalOp], LocalOp);
@@ -703,18 +782,35 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
     FEXCORE_TELEMETRY_SET(VEXOpTelem, 1);
     uint16_t map_select = 1;
     uint16_t pp = 0;
+    const uint8_t Byte1 = ReadByte();
+    DecodedHeader options{};
 
-    uint8_t Byte1 = ReadByte();
+    if ((Byte1 & 0b10000000) == 0) {
+      LOGMAN_THROW_A_FMT(CTX->Config.Is64BitMode, "VEX.R shouldn't be 0 in 32-bit mode!");
+      DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_R;
+    }
 
     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
-      uint8_t Byte2 = ReadByte();
+      const uint8_t Byte2 = ReadByte();
       pp = Byte2 & 0b11;
       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;
+      }
+      if (CTX->Config.Is64BitMode && (Byte1 & 0b00100000) == 0) {
+        DecodeInst->Flags |= DecodeFlags::FLAG_REX_XGPR_B;
+      }
       if (!(map_select >= 1 && map_select <= 3)) {
-        LogMan::Msg::E("We don't understand a map_select of: %d", map_select);
+        LogMan::Msg::EFmt("We don't understand a map_select of: {}", map_select);
         return false;
       }
     }
@@ -740,10 +836,10 @@ bool Decoder::NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16
 #define OPD(group, pp, opcode) (((group - TYPE_VEX_GROUP_12) << 4) | (pp << 3) | (opcode))
       Op = OPD(LocalInfo->Type, pp, ModRM.reg);
 #undef OPD
-      return NormalOp(&VEXTableGroupOps[Op], Op);
+      return NormalOp(&VEXTableGroupOps[Op], Op, options);
+    } else {
+      return NormalOp(LocalInfo, Op, options);
     }
-    else
-      return NormalOp(LocalInfo, Op);
   }
   else if (Info->Type == FEXCore::X86Tables::TYPE_GROUP_EVEX) {
     FEXCORE_TELEMETRY_SET(EVEXOpTelem, 1);
@@ -774,7 +870,7 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
     case 0x0F: {// Escape Op
       uint8_t EscapeOp = ReadByte();
       switch (EscapeOp) {
-      case 0x0F: { // 3DNow!
+      case 0x0F: [[unlikely]] { // 3DNow!
         // 3DNow! Instruction Encoding: 0F 0F [ModRM] [SIB] [Displacement] [Opcode]
         // Decode ModRM
         uint8_t ModRMByte = ReadByte();
@@ -787,8 +883,12 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
         const bool Has16BitAddressing = !CTX->Config.Is64BitMode &&
           DecodeInst->Flags & DecodeFlags::FLAG_ADDRESS_SIZE;
 
-        auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
-        (this->*Disp)(&DecodeInst->Src[0], ModRM);
+        // All 3DNow! instructions have the second argument as the rm handler
+        // We need to decode it upfront to get the displacement out of the way
+        if (ModRM.mod != 0b11) {
+          auto Disp = DecodeModRMs_Disp[Has16BitAddressing];
+          (this->*Disp)(&DecodeInst->Src[0], ModRM);
+        }
 
         // Take a peek at the op just past the displacement
         uint8_t LocalOp = ReadByte();
@@ -797,14 +897,20 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
       }
       case 0x38: { // F38 Table!
         constexpr uint16_t PF_38_NONE = 0;
-        constexpr uint16_t PF_38_66 = 1;
-        constexpr uint16_t PF_38_F2 = 2;
+        constexpr uint16_t PF_38_66   = (1U << 0);
+        constexpr uint16_t PF_38_F2   = (1U << 1);
+        constexpr uint16_t PF_38_F3   = (1U << 2);
 
         uint16_t Prefix = PF_38_NONE;
-        if (DecodeInst->LastEscapePrefix == 0xF2) // REPNE
-          Prefix = PF_38_F2;
-        else if (DecodeInst->LastEscapePrefix == 0x66) // Operand Size
-          Prefix = PF_38_66;
+        if (DecodeInst->Flags & DecodeFlags::FLAG_OPERAND_SIZE) {
+          Prefix |= PF_38_66;
+        }
+        if (DecodeInst->Flags & DecodeFlags::FLAG_REPNE_PREFIX) {
+          Prefix |= PF_38_F2;
+        }
+        if (DecodeInst->Flags & DecodeFlags::FLAG_REP_PREFIX) {
+          Prefix |= PF_38_F3;
+        }
 
         uint16_t LocalOp = (Prefix << 8) | ReadByte();
         return NormalOpHeader(&FEXCore::X86Tables::H0F38TableOps[LocalOp], LocalOp);
@@ -919,7 +1025,7 @@ bool Decoder::DecodeInstruction(uint64_t PC) {
       auto Info = &FEXCore::X86Tables::BaseOps[Op];
 
       if (Info->Type == FEXCore::X86Tables::TYPE_REX_PREFIX) {
-        LOGMAN_THROW_A(CTX->Config.Is64BitMode, "Got REX prefix in 32bit mode");
+        LOGMAN_THROW_A_FMT(CTX->Config.Is64BitMode, "Got REX prefix in 32bit mode");
         DecodeInst->Flags |= DecodeFlags::FLAG_REX_PREFIX;
 
         // Widening displacement
@@ -972,13 +1078,13 @@ void Decoder::BranchTargetInMultiblockRange() {
       // auto RIPOffset = LoadSource(Op, Op->Src[0], Op->Flags);
       // auto RIPTargetConst = _Constant(Op->PC + Op->InstSize);
       // Target offset is PC + InstSize + Literal
-      LOGMAN_THROW_A(DecodeInst->Src[0].IsLiteral(), "Had wrong operand type");
+      LOGMAN_THROW_A_FMT(DecodeInst->Src[0].IsLiteral(), "Had wrong operand type");
       TargetRIP = DecodeInst->PC + DecodeInst->InstSize + DecodeInst->Src[0].Data.Literal.Value;
     break;
     }
     case 0xE9:
     case 0xEB: // Both are unconditional JMP instructions
-      LOGMAN_THROW_A(DecodeInst->Src[0].IsLiteral(), "Had wrong operand type");
+      LOGMAN_THROW_A_FMT(DecodeInst->Src[0].IsLiteral(), "Had wrong operand type");
       TargetRIP = DecodeInst->PC + DecodeInst->InstSize + DecodeInst->Src[0].Data.Literal.Value;
       Conditional = false;
     break;
@@ -1025,6 +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;
@@ -1044,7 +1179,8 @@ const uint8_t *Decoder::AdjustAddrForSpecialRegion(uint8_t const* _InstStream, u
   return _InstStream - EntryPoint + RIP;
 }
 
-bool 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();
@@ -1052,13 +1188,13 @@ bool Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
   DecodedSize = 0;
   MaxCondBranchForward = 0;
   MaxCondBranchBackwards = ~0ULL;
+  DecodedBuffer = PoolObject.ReownOrClaimBuffer();
 
   // XXX: Load symbol data
   SymbolAvailable = false;
   EntryPoint = PC;
   InstStream = _InstStream;
 
-  bool ErrorDuringDecoding = false;
   uint64_t TotalInstructions{};
 
   // If we don't have symbols available then we become a bit optimistic about multiblock ranges
@@ -1074,6 +1210,12 @@ bool 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;
@@ -1090,20 +1232,38 @@ bool Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
     InstStream = AdjustAddrForSpecialRegion(_InstStream, EntryPoint, RIPToDecode);
 
     while (1) {
-      ErrorDuringDecoding = !DecodeInstruction(RIPToDecode + PCOffset);
+
+      // 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::D("Couldn't Decode something at 0x%lx, Started at 0x%lx", PC + PCOffset, PC);
-        if (Blocks.size() == 1) {
-          return false;
-        }
-        LOGMAN_THROW_A(Blocks.size() != 1, "Decode Error in entry block");
-
+        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;
-        if (ErrorDuringDecoding && Blocks.size() != 1) {
-          ErrorDuringDecoding = false;
-        }
-        break;
+        // Error while decoding instruction. We don't know the table or instruction size
+        DecodeInst->TableInfo = nullptr;
+        DecodeInst->InstSize = 0;
       }
 
       DecodedMinAddress = std::min(DecodedMinAddress, RIPToDecode + PCOffset);
@@ -1112,6 +1272,11 @@ bool Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
       ++BlockNumberOfInstructions;
       ++DecodedSize;
 
+      // Can not continue this block at all on invalid instruction
+      if (CurrentBlockDecoding.HasInvalidInstruction) {
+        break;
+      }
+
       bool CanContinue = false;
       if (!(DecodeInst->TableInfo->Flags &
           (FEXCore::X86Tables::InstFlags::FLAGS_BLOCK_END | FEXCore::X86Tables::InstFlags::FLAGS_SETS_RIP))) {
@@ -1119,23 +1284,21 @@ bool 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;
       }
 
@@ -1156,7 +1319,6 @@ bool Decoder::DecodeInstructionsAtEntry(uint8_t const* _InstStream, uint64_t PC)
   std::sort(Blocks.begin(), Blocks.end(), [](const FEXCore::Frontend::Decoder::DecodedBlocks& a, const FEXCore::Frontend::Decoder::DecodedBlocks& b) {
     return a.Entry < b.Entry;
   });
-  return !ErrorDuringDecoding;
 }
 
 }

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

@@ -27,7 +27,7 @@ public:
 
   Decoder(FEXCore::Context::Context *ctx);
   ~Decoder();
-  bool 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;
@@ -35,26 +35,42 @@ public:
 
   uint64_t DecodedMinAddress {};
   uint64_t DecodedMaxAddress {~0ULL};
-  
+
   void SetSectionMaxAddress(uint64_t v) { SectionMaxAddress = v; }
   void SetExternalBranches(std::set<uint64_t> *v) { ExternalBranches = v; }
+
+  void DelayedDisownBuffer() {
+    PoolObject.DelayedDisownBuffer();
+  }
+
 private:
+  // To pass any information from instruction prefixes
+  // down into the actual instruction handling machinery.
+  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;
   const FEXCore::HLE::SyscallOSABI OSABI{};
 
   bool DecodeInstruction(uint64_t PC);
 
   void BranchTargetInMultiblockRange();
+  bool BranchTargetCanContinue(bool FinalInstruction) const;
 
   uint8_t ReadByte();
   uint8_t PeekByte(uint8_t Offset) const;
   uint64_t ReadData(uint8_t Size);
   void SkipBytes(uint8_t Size) { InstructionSize += Size; }
-  bool NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op);
+
+  bool NormalOp(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op, DecodedHeader Options = {});
   bool NormalOpHeader(FEXCore::X86Tables::X86InstInfo const *Info, uint16_t Op);
 
   static constexpr size_t DefaultDecodedBufferSize = 0x10000;
   FEXCore::X86Tables::DecodedInst *DecodedBuffer{};
+  Utils::FixedSizePooledAllocation<FEXCore::X86Tables::DecodedInst*, 5000, 500> PoolObject;
   size_t DecodedSize {};
 
   uint8_t const *InstStream;
@@ -83,7 +99,7 @@ private:
   void DecodeModRM_16(X86Tables::DecodedOperand *Operand, X86Tables::ModRMDecoded ModRM);
   void DecodeModRM_64(X86Tables::DecodedOperand *Operand, X86Tables::ModRMDecoded ModRM);
 
-  const std::array<DecodeModRMPtr, 2> DecodeModRMs_Disp {
+  static constexpr std::array<DecodeModRMPtr, 2> DecodeModRMs_Disp{
     &FEXCore::Frontend::Decoder::DecodeModRM_64,
     &FEXCore::Frontend::Decoder::DecodeModRM_16,
   };

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

@@ -6,8 +6,10 @@ $end_info$
 #pragma once
 
 #include <FEXCore/Config/Config.h>
+#include <FEXCore/Utils/Event.h>
 #include <FEXCore/Utils/Threads.h>
 
+#include <atomic>
 #include <istream>
 #include <memory>
 #include <mutex>
@@ -27,9 +29,14 @@ public:
     // Public for threading
     void GdbServerLoop();
 
+    void AlertLibrariesChanged() {
+      LibraryMapChanged = true;
+    }
+
 private:
     void Break(int signal);
 
+    void OpenListenSocket();
     std::unique_ptr<std::iostream> OpenSocket();
     void StartThread();
     std::string ReadPacket(std::iostream &stream);
@@ -37,6 +44,9 @@ private:
 
     void SendACK(std::ostream &stream, bool NACK);
 
+    Event ThreadBreakEvent{};
+    void WaitForThreadWakeup();
+
     struct HandledPacketType {
       std::string Response{};
       enum ResponseType {
@@ -60,6 +70,8 @@ private:
     HandledPacketType handleBreakpoint(const std::string &packet);
     HandledPacketType handleProgramOffsets();
 
+    HandledPacketType ThreadAction(char action, uint32_t tid);
+
     std::string readRegs();
     HandledPacketType readReg(const std::string& packet);
 
@@ -69,8 +81,17 @@ private:
     std::mutex sendMutex;
     bool SettingNoAckMode{false};
     bool NoAckMode{false};
+    bool NonStopMode{false};
     std::string ThreadString{};
+    std::string OSDataString{};
+    void buildLibraryMap();
+    std::atomic<bool> LibraryMapChanged = true;
+    std::string LibraryMapString{};
+
+    // Used to keep track of which signals to pass to the guest
+    std::array<bool, SignalDelegator::MAX_SIGNALS + 1> PassSignals{};
     uint32_t CurrentDebuggingThread{};
+    int ListenSocket{};
     FEX_CONFIG_OPT(Filename, APP_FILENAME);
 };
 

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

@@ -1,6 +1,7 @@
-#include "Interface/Core/HostFeatures.h"
+#include "Interface/Core/CPUID.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"
@@ -13,14 +14,149 @@
 
 namespace FEXCore {
 
-HostFeatures::HostFeatures() {
+// Data Zero Prohibited flag
+// 0b0 = ZVA/GVA/GZVA permitted
+// 0b1 = ZVA/GVA/GZVA prohibited
+constexpr uint32_t DCZID_DZP_MASK = 0b1'0000;
+// Log2 of the blocksize in 32-bit words
+constexpr uint32_t DCZID_BS_MASK = 0b0'1111;
+
 #ifdef _M_ARM_64
-  auto Features = vixl::CPUFeatures::InferFromOS();
-  SupportsAES = Features.Has(vixl::CPUFeatures::Feature::kAES);
+static uint32_t GetDCZID() {
+  uint64_t Result{};
+  __asm("mrs %[Res], DCZID_EL0"
+      : [Res] "=r" (Result));
+  return Result;
+}
+
+static uint32_t GetFPCR() {
+  uint64_t Result{};
+  __asm ("mrs %[Res], FPCR"
+    : [Res] "=r" (Result));
+  return Result;
+}
+
+static void SetFPCR(uint64_t Value) {
+  __asm ("msr FPCR, %[Value]"
+    :: [Value] "r" (Value));
+}
+
+#else
+static uint32_t GetDCZID() {
+  // Return unsupported
+  return DCZID_DZP_MASK;
+}
 #endif
-#ifdef _M_X86_64
+
+
+HostFeatures::HostFeatures() {
+#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);
+  SupportsRAND = Features.Has(vixl::CPUFeatures::Feature::kRNG);
+
+  // Only supported when FEAT_AFP is supported
+  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;
+  __asm volatile ("mrs %[ctr], ctr_el0"
+    : [ctr] "=r"(CTR));
+
+  DCacheLineSize = 4 << ((CTR >> 16) & 0xF);
+  ICacheLineSize = 4 << (CTR & 0xF);
+
+  // 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 =
+    (1U << 8) |  // Invalid Operation float exception trap enable
+    (1U << 9) |  // Divide by zero float exception trap enable
+    (1U << 10) | // Overflow float exception trap enable
+    (1U << 11) | // Underflow float exception trap enable
+    (1U << 12) | // Inexact float exception trap enable
+    (1U << 15);  // Input Denormal float exception trap enable
+
+  uint32_t OriginalFPCR = GetFPCR();
+  uint32_t FPCR = OriginalFPCR | ExceptionEnableTraps;
+  SetFPCR(FPCR);
+  FPCR = GetFPCR();
+  SupportsFloatExceptions = (FPCR & ExceptionEnableTraps) == ExceptionEnableTraps;
+
+  // Set FPCR back to original just in case anything changed
+  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) {
+    uint32_t DCZID_Log2 = DCZID & DCZID_BS_MASK;
+    uint32_t DCZID_Bytes = (1 << DCZID_Log2) * sizeof(uint32_t);
+    // If the DC ZVA size matches the emulated cache line size
+    // This means we can use the instruction
+    SupportsCLZERO = DCZID_Bytes == CPUIDEmu::CACHELINE_SIZE;
+  }
 #endif
 }
 }

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

@@ -1,9 +0,0 @@
-#pragma once
-
-namespace FEXCore {
-class HostFeatures final {
-  public:
-    HostFeatures();
-    bool SupportsAES{};
-};
-}

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

@@ -0,0 +1,777 @@
+/*
+$info$
+tags: backend|interpreter
+$end_info$
+*/
+
+#include "Interface/Core/Interpreter/InterpreterClass.h"
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/Interpreter/InterpreterDefines.h"
+
+#include <FEXCore/Utils/BitUtils.h>
+
+#include <cstdint>
+
+namespace FEXCore::CPU {
+
+#ifdef _M_X86_64
+uint8_t AtomicFetchNeg(uint8_t *Addr) {
+  using Type = uint8_t;
+  std::atomic<Type> *MemData = reinterpret_cast<std::atomic<Type>*>(Addr);
+  Type Expected = MemData->load();
+  Type Desired = -Expected;
+  do {
+    Desired = -Expected;
+  } while (!MemData->compare_exchange_strong(Expected, Desired, std::memory_order_seq_cst));
+
+  return Expected;
+}
+
+uint16_t AtomicFetchNeg(uint16_t *Addr) {
+  using Type = uint16_t;
+  std::atomic<Type> *MemData = reinterpret_cast<std::atomic<Type>*>(Addr);
+  Type Expected = MemData->load();
+  Type Desired = -Expected;
+  do {
+    Desired = -Expected;
+  } while (!MemData->compare_exchange_strong(Expected, Desired, std::memory_order_seq_cst));
+
+  return Expected;
+}
+
+uint32_t AtomicFetchNeg(uint32_t *Addr) {
+  using Type = uint32_t;
+  std::atomic<Type> *MemData = reinterpret_cast<std::atomic<Type>*>(Addr);
+  Type Expected = MemData->load();
+  Type Desired = -Expected;
+  do {
+    Desired = -Expected;
+  } while (!MemData->compare_exchange_strong(Expected, Desired, std::memory_order_seq_cst));
+
+  return Expected;
+}
+
+uint64_t AtomicFetchNeg(uint64_t *Addr) {
+  using Type = uint64_t;
+  std::atomic<Type> *MemData = reinterpret_cast<std::atomic<Type>*>(Addr);
+  Type Expected = MemData->load();
+  Type Desired = -Expected;
+  do {
+    Desired = -Expected;
+  } while (!MemData->compare_exchange_strong(Expected, Desired, std::memory_order_seq_cst));
+
+  return Expected;
+}
+
+template<typename T>
+T AtomicCompareAndSwap(T expected, T desired, T *addr)
+{
+  std::atomic<T> *MemData = reinterpret_cast<std::atomic<T>*>(addr);
+
+  T Src1 = expected;
+  T Src2 = desired;
+
+  T Expected = Src1;
+  bool Result = MemData->compare_exchange_strong(Expected, Src2);
+
+  return Result ? Src1 : Expected;
+}
+
+template uint8_t AtomicCompareAndSwap<uint8_t>(uint8_t expected, uint8_t desired, uint8_t *addr);
+template uint16_t AtomicCompareAndSwap<uint16_t>(uint16_t expected, uint16_t desired, uint16_t *addr);
+template uint32_t AtomicCompareAndSwap<uint32_t>(uint32_t expected, uint32_t desired, uint32_t *addr);
+template uint64_t AtomicCompareAndSwap<uint64_t>(uint64_t expected, uint64_t desired, uint64_t *addr);
+
+#else
+// Needs to match what the AArch64 JIT and unaligned signal handler expects
+uint8_t AtomicFetchNeg(uint8_t *Addr) {
+  using Type = uint8_t;
+  Type Result{};
+  Type Tmp{};
+  Type TmpStatus{};
+
+  __asm__ volatile(
+  R"(
+  1:
+    ldaxrb %w[Result], [%[Memory]];
+    neg %w[Tmp], %w[Result];
+    stlxrb %w[TmpStatus], %w[Tmp], [%[Memory]];
+    cbnz %w[TmpStatus], 1b;
+  )"
+  : [Result] "=r" (Result)
+  , [Tmp] "=r" (Tmp)
+  , [TmpStatus] "=r" (TmpStatus)
+  , [Memory] "+r" (Addr)
+  :: "memory"
+  );
+  return Result;
+}
+
+uint16_t AtomicFetchNeg(uint16_t *Addr) {
+  using Type = uint16_t;
+  Type Result{};
+  Type Tmp{};
+  Type TmpStatus{};
+
+  __asm__ volatile(
+  R"(
+  1:
+    ldaxrh %w[Result], [%[Memory]];
+    neg %w[Tmp], %w[Result];
+    stlxrh %w[TmpStatus], %w[Tmp], [%[Memory]];
+    cbnz %w[TmpStatus], 1b;
+  )"
+  : [Result] "=r" (Result)
+  , [Tmp] "=r" (Tmp)
+  , [TmpStatus] "=r" (TmpStatus)
+  , [Memory] "+r" (Addr)
+  :: "memory"
+  );
+  return Result;
+}
+
+uint32_t AtomicFetchNeg(uint32_t *Addr) {
+  using Type = uint32_t;
+  Type Result{};
+  Type Tmp{};
+  Type TmpStatus{};
+
+  __asm__ volatile(
+  R"(
+  1:
+    ldaxr %w[Result], [%[Memory]];
+    neg %w[Tmp], %w[Result];
+    stlxr %w[TmpStatus], %w[Tmp], [%[Memory]];
+    cbnz %w[TmpStatus], 1b;
+  )"
+  : [Result] "=r" (Result)
+  , [Tmp] "=r" (Tmp)
+  , [TmpStatus] "=r" (TmpStatus)
+  , [Memory] "+r" (Addr)
+  :: "memory"
+  );
+  return Result;
+}
+
+uint64_t AtomicFetchNeg(uint64_t *Addr) {
+  using Type = uint64_t;
+  Type Result{};
+  Type Tmp{};
+  Type TmpStatus{};
+
+  __asm__ volatile(
+  R"(
+  1:
+    ldaxr %[Result], [%[Memory]];
+    neg %[Tmp], %[Result];
+    stlxr %w[TmpStatus], %[Tmp], [%[Memory]];
+    cbnz %w[TmpStatus], 1b;
+  )"
+  : [Result] "=r" (Result)
+  , [Tmp] "=r" (Tmp)
+  , [TmpStatus] "=r" (TmpStatus)
+  , [Memory] "+r" (Addr)
+  :: "memory"
+  );
+  return Result;
+}
+
+template<>
+uint8_t AtomicCompareAndSwap(uint8_t expected, uint8_t desired, uint8_t *addr) {
+  using Type = uint8_t;
+  //force Result to r9 (scratch register) or clang spills to stack
+  register Type Result asm("r9"){};
+  Type Tmp{};
+  Type Tmp2{};
+  __asm__ volatile(
+  R"(
+  1:
+    ldaxrb %w[Tmp], [%[Memory]];
+    cmp %w[Tmp], %w[Expected], uxtb;
+    b.ne 2f;
+    stlxrb %w[Tmp2], %w[Desired], [%[Memory]];
+    cbnz %w[Tmp2], 1b;
+    mov %w[Result], %w[Expected];
+    b 3f;
+  2:
+    mov %w[Result], %w[Tmp];
+    clrex;
+  3:
+  )"
+  : [Tmp] "=r" (Tmp)
+  , [Tmp2] "=r" (Tmp2)
+  , [Desired] "+r" (desired)
+  , [Expected] "+r" (expected)
+  , [Result] "=r" (Result)
+  , [Memory] "+r" (addr)
+  :: "memory"
+  );
+  return Result;
+}
+
+template<>
+uint16_t AtomicCompareAndSwap(uint16_t expected, uint16_t desired, uint16_t *addr) {
+  using Type = uint16_t;
+  //force Result to r9 (scratch register) or clang spills to stack
+  register Type Result asm("r9"){};
+  Type Tmp{};
+  Type Tmp2{};
+  __asm__ volatile(
+  R"(
+  1:
+    ldaxrh %w[Tmp], [%[Memory]];
+    cmp %w[Tmp], %w[Expected], uxth;
+    b.ne 2f;
+    stlxrh %w[Tmp2], %w[Desired], [%[Memory]];
+    cbnz %w[Tmp2], 1b;
+    mov %w[Result], %w[Expected];
+    b 3f;
+  2:
+    mov %w[Result], %w[Tmp];
+    clrex;
+  3:
+  )"
+  : [Tmp] "=r" (Tmp)
+  , [Tmp2] "=r" (Tmp2)
+  , [Desired] "+r" (desired)
+  , [Expected] "+r" (expected)
+  , [Result] "=r" (Result)
+  , [Memory] "+r" (addr)
+  :: "memory"
+  );
+  return Result;
+}
+
+template<>
+uint32_t AtomicCompareAndSwap(uint32_t expected, uint32_t desired, uint32_t *addr) {
+  using Type = uint32_t;
+  //force Result to r9 (scratch register) or clang spills to stack
+  register Type Result asm("r9"){};
+  Type Tmp{};
+  Type Tmp2{};
+  __asm__ volatile(
+  R"(
+  1:
+    ldaxr %w[Tmp], [%[Memory]];
+    cmp %w[Tmp], %w[Expected];
+    b.ne 2f;
+    stlxr %w[Tmp2], %w[Desired], [%[Memory]];
+    cbnz %w[Tmp2], 1b;
+    mov %w[Result], %w[Expected];
+    b 3f;
+  2:
+    mov %w[Result], %w[Tmp];
+    clrex;
+  3:
+  )"
+  : [Tmp] "=r" (Tmp)
+  , [Tmp2] "=r" (Tmp2)
+  , [Desired] "+r" (desired)
+  , [Expected] "+r" (expected)
+  , [Result] "=r" (Result)
+  , [Memory] "+r" (addr)
+  :: "memory"
+  );
+  return Result;
+}
+
+template<>
+uint64_t AtomicCompareAndSwap(uint64_t expected, uint64_t desired, uint64_t *addr) {
+  using Type = uint64_t;
+  //force Result to r9 (scratch register) or clang spills to stack
+  register Type Result asm("r9"){};
+  Type Tmp{};
+  Type Tmp2{};
+  __asm__ volatile(
+  R"(
+  1:
+    ldaxr %[Tmp], [%[Memory]];
+    cmp %[Tmp], %[Expected];
+    b.ne 2f;
+    stlxr %w[Tmp2], %[Desired], [%[Memory]];
+    cbnz %w[Tmp2], 1b;
+    mov %[Result], %[Expected];
+    b 3f;
+  2:
+    mov %[Result], %[Tmp];
+    clrex;
+  3:
+  )"
+  : [Tmp] "=r" (Tmp)
+  , [Tmp2] "=r" (Tmp2)
+  , [Desired] "+r" (desired)
+  , [Expected] "+r" (expected)
+  , [Result] "=r" (Result)
+  , [Memory] "+r" (addr)
+  :: "memory"
+  );
+  return Result;
+}
+
+#endif
+
+#define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
+DEF_OP(CASPair) {
+  auto Op = IROp->C<IR::IROp_CASPair>();
+
+  // Size is the size of each pair element
+  switch (IROp->ElementSize) {
+    case 4: {
+      GD = AtomicCompareAndSwap(
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint64_t**>(Data->SSAData, Op->Addr)
+      );
+      break;
+    }
+    case 8: {
+      std::atomic<__uint128_t> *MemData = *GetSrc<std::atomic<__uint128_t> **>(Data->SSAData, Op->Addr);
+
+      __uint128_t Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Expected);
+      __uint128_t Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Desired);
+
+      __uint128_t Expected = Src1;
+      bool Result = MemData->compare_exchange_strong(Expected, Src2);
+      memcpy(GDP, Result ? &Src1 : &Expected, 16);
+      break;
+    }
+    default: LOGMAN_MSG_A_FMT("Unknown CAS size: {}", IROp->ElementSize); break;
+  }
+}
+
+DEF_OP(CAS) {
+  auto Op = IROp->C<IR::IROp_CAS>();
+  uint8_t OpSize = IROp->Size;
+
+  switch (OpSize) {
+    case 1: {
+      GD = AtomicCompareAndSwap(
+        *GetSrc<uint8_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint8_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint8_t**>(Data->SSAData, Op->Addr)
+      );
+      break;
+    }
+    case 2: {
+      GD = AtomicCompareAndSwap(
+        *GetSrc<uint16_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint16_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint16_t**>(Data->SSAData, Op->Addr)
+      );
+      break;
+    }
+    case 4: {
+      GD = AtomicCompareAndSwap(
+        *GetSrc<uint32_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint32_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint32_t**>(Data->SSAData, Op->Addr)
+      );
+      break;
+    }
+    case 8: {
+      GD = AtomicCompareAndSwap(
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Expected),
+        *GetSrc<uint64_t*>(Data->SSAData, Op->Desired),
+        *GetSrc<uint64_t**>(Data->SSAData, Op->Addr)
+      );
+      break;
+    }
+    default: LOGMAN_MSG_A_FMT("Unknown CAS size: {}", OpSize); break;
+  }
+}
+
+DEF_OP(AtomicAdd) {
+  auto Op = IROp->C<IR::IROp_AtomicAdd>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      *MemData += Src;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      *MemData += Src;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      *MemData += Src;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      *MemData += Src;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicSub) {
+  auto Op = IROp->C<IR::IROp_AtomicSub>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      *MemData -= Src;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      *MemData -= Src;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      *MemData -= Src;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      *MemData -= Src;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicAnd) {
+  auto Op = IROp->C<IR::IROp_AtomicAnd>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      *MemData &= Src;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      *MemData &= Src;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      *MemData &= Src;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      *MemData &= Src;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicOr) {
+  auto Op = IROp->C<IR::IROp_AtomicOr>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      *MemData |= Src;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      *MemData |= Src;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      *MemData |= Src;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      *MemData |= Src;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicXor) {
+  auto Op = IROp->C<IR::IROp_AtomicXor>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      *MemData ^= Src;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      *MemData ^= Src;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      *MemData ^= Src;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      *MemData ^= Src;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicSwap) {
+  auto Op = IROp->C<IR::IROp_AtomicSwap>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      uint8_t Previous = MemData->exchange(Src);
+      GD = Previous;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      uint16_t Previous = MemData->exchange(Src);
+      GD = Previous;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      uint32_t Previous = MemData->exchange(Src);
+      GD = Previous;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      uint64_t Previous = MemData->exchange(Src);
+      GD = Previous;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicFetchAdd) {
+  auto Op = IROp->C<IR::IROp_AtomicFetchAdd>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      uint8_t Previous = MemData->fetch_add(Src);
+      GD = Previous;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      uint16_t Previous = MemData->fetch_add(Src);
+      GD = Previous;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      uint32_t Previous = MemData->fetch_add(Src);
+      GD = Previous;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      uint64_t Previous = MemData->fetch_add(Src);
+      GD = Previous;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicFetchSub) {
+  auto Op = IROp->C<IR::IROp_AtomicFetchSub>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      uint8_t Previous = MemData->fetch_sub(Src);
+      GD = Previous;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      uint16_t Previous = MemData->fetch_sub(Src);
+      GD = Previous;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      uint32_t Previous = MemData->fetch_sub(Src);
+      GD = Previous;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      uint64_t Previous = MemData->fetch_sub(Src);
+      GD = Previous;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicFetchAnd) {
+  auto Op = IROp->C<IR::IROp_AtomicFetchAnd>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      uint8_t Previous = MemData->fetch_and(Src);
+      GD = Previous;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      uint16_t Previous = MemData->fetch_and(Src);
+      GD = Previous;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      uint32_t Previous = MemData->fetch_and(Src);
+      GD = Previous;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      uint64_t Previous = MemData->fetch_and(Src);
+      GD = Previous;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicFetchOr) {
+  auto Op = IROp->C<IR::IROp_AtomicFetchOr>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      uint8_t Previous = MemData->fetch_or(Src);
+      GD = Previous;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      uint16_t Previous = MemData->fetch_or(Src);
+      GD = Previous;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      uint32_t Previous = MemData->fetch_or(Src);
+      GD = Previous;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      uint64_t Previous = MemData->fetch_or(Src);
+      GD = Previous;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicFetchXor) {
+  auto Op = IROp->C<IR::IROp_AtomicFetchXor>();
+  switch (IROp->Size) {
+    case 1: {
+      std::atomic<uint8_t> *MemData = *GetSrc<std::atomic<uint8_t> **>(Data->SSAData, Op->Addr);
+      uint8_t Src = *GetSrc<uint8_t*>(Data->SSAData, Op->Value);
+      uint8_t Previous = MemData->fetch_xor(Src);
+      GD = Previous;
+      break;
+    }
+    case 2: {
+      std::atomic<uint16_t> *MemData = *GetSrc<std::atomic<uint16_t> **>(Data->SSAData, Op->Addr);
+      uint16_t Src = *GetSrc<uint16_t*>(Data->SSAData, Op->Value);
+      uint16_t Previous = MemData->fetch_xor(Src);
+      GD = Previous;
+      break;
+    }
+    case 4: {
+      std::atomic<uint32_t> *MemData = *GetSrc<std::atomic<uint32_t> **>(Data->SSAData, Op->Addr);
+      uint32_t Src = *GetSrc<uint32_t*>(Data->SSAData, Op->Value);
+      uint32_t Previous = MemData->fetch_xor(Src);
+      GD = Previous;
+      break;
+    }
+    case 8: {
+      std::atomic<uint64_t> *MemData = *GetSrc<std::atomic<uint64_t> **>(Data->SSAData, Op->Addr);
+      uint64_t Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+      uint64_t Previous = MemData->fetch_xor(Src);
+      GD = Previous;
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+DEF_OP(AtomicFetchNeg) {
+  auto Op = IROp->C<IR::IROp_AtomicFetchNeg>();
+  switch (IROp->Size) {
+    case 1: {
+      using Type = uint8_t;
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
+      break;
+    }
+    case 2: {
+      using Type = uint16_t;
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
+      break;
+    }
+    case 4: {
+      using Type = uint32_t;
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
+      break;
+    }
+    case 8: {
+      using Type = uint64_t;
+      GD = AtomicFetchNeg(*GetSrc<Type**>(Data->SSAData, Op->Addr));
+      break;
+    }
+    default:  LOGMAN_MSG_A_FMT("Unhandled Atomic size: {}", IROp->Size);
+  }
+}
+
+#undef DEF_OP
+
+} // namespace FEXCore::CPU

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

@@ -0,0 +1,160 @@
+/*
+$info$
+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"
+#include "Interface/HLE/Thunks/Thunks.h"
+
+#include <FEXCore/Utils/BitUtils.h>
+#include <FEXCore/HLE/SyscallHandler.h>
+
+#include <cstdint>
+#include <unistd.h>
+
+namespace FEXCore::CPU {
+[[noreturn]]
+static void SignalReturn(FEXCore::Core::InternalThreadState *Thread) {
+  Thread->CTX->SignalThread(Thread, FEXCore::Core::SignalEvent::Return);
+
+  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(SignalReturn) {
+  SignalReturn(Data->State);
+}
+
+DEF_OP(CallbackReturn) {
+  Data->State->CurrentFrame->Pointers.Interpreter.CallbackReturn(Data->State, Data->StackEntry);
+}
+
+DEF_OP(ExitFunction) {
+  auto Op = IROp->C<IR::IROp_ExitFunction>();
+  uint8_t OpSize = IROp->Size;
+
+  uintptr_t* ContextPtr = reinterpret_cast<uintptr_t*>(Data->State->CurrentFrame);
+
+  void *ContextData = reinterpret_cast<void*>(ContextPtr);
+  void *Src = GetSrc<void*>(Data->SSAData, Op->NewRIP);
+
+  memcpy(ContextData, Src, OpSize);
+
+  Data->BlockResults.Quit = true;
+}
+
+DEF_OP(Jump) {
+  auto Op = IROp->C<IR::IROp_Jump>();
+  const uintptr_t ListBegin = Data->CurrentIR->GetListData();
+  const uintptr_t DataBegin = Data->CurrentIR->GetData();
+
+  Data->BlockIterator = IR::NodeIterator(ListBegin, DataBegin, Op->TargetBlock);
+  Data->BlockResults.Redo = true;
+}
+
+DEF_OP(CondJump) {
+  auto Op = IROp->C<IR::IROp_CondJump>();
+  const uintptr_t ListBegin = Data->CurrentIR->GetListData();
+  const uintptr_t DataBegin = Data->CurrentIR->GetData();
+
+  bool CompResult;
+
+  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);
+  else
+    CompResult = IsConditionTrue<uint64_t, int64_t, double>(Op->Cond.Val, Src1, Src2);
+
+  if (CompResult) {
+    Data->BlockIterator = IR::NodeIterator(ListBegin, DataBegin, Op->TrueBlock);
+  }
+  else  {
+    Data->BlockIterator = IR::NodeIterator(ListBegin, DataBegin, Op->FalseBlock);
+  }
+  Data->BlockResults.Redo = true;
+}
+
+DEF_OP(Syscall) {
+  auto Op = IROp->C<IR::IROp_Syscall>();
+
+  FEXCore::HLE::SyscallArguments Args;
+  for (size_t j = 0; j < FEXCore::HLE::SyscallArguments::MAX_ARGS; ++j) {
+    if (Op->Header.Args[j].IsInvalid()) break;
+    Args.Argument[j] = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[j]);
+  }
+
+  uint64_t Res = FEXCore::Context::HandleSyscall(Data->State->CTX->SyscallHandler, Data->State->CurrentFrame, &Args);
+  GD = Res;
+}
+
+DEF_OP(InlineSyscall) {
+  auto Op = IROp->C<IR::IROp_InlineSyscall>();
+
+  FEXCore::HLE::SyscallArguments Args;
+  for (size_t j = 0; j < FEXCore::HLE::SyscallArguments::MAX_ARGS; ++j) {
+    if (Op->Header.Args[j].IsInvalid()) break;
+    Args.Argument[j] = *GetSrc<uint64_t*>(Data->SSAData, Op->Header.Args[j]);
+  }
+
+  // We don't want the errno handling but I also don't want to write inline ASM atm
+  uint64_t Res = syscall(
+    Op->HostSyscallNumber,
+    Args.Argument[0],
+    Args.Argument[1],
+    Args.Argument[2],
+    Args.Argument[3],
+    Args.Argument[4],
+    Args.Argument[5],
+    Args.Argument[6]
+  );
+
+  if (Res == -1) {
+    Res = -errno;
+  }
+
+  GD = Res;
+}
+
+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->ArgPtr));
+}
+
+DEF_OP(ValidateCode) {
+  auto Op = IROp->C<IR::IROp_ValidateCode>();
+
+  auto CodePtr = Data->CurrentEntry + Op->Offset;
+  if (memcmp((void*)CodePtr, &Op->CodeOriginalLow, Op->CodeLength) != 0) {
+    GD = 1;
+  } else {
+    GD = 0;
+  }
+}
+
+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);
+  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);
+}
+
+#undef DEF_OP
+
+} // namespace FEXCore::CPU

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

@@ -0,0 +1,261 @@
+/*
+$info$
+tags: backend|interpreter
+$end_info$
+*/
+
+#include "Interface/Core/Interpreter/InterpreterClass.h"
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/Interpreter/InterpreterDefines.h"
+
+#include <cstdint>
+
+namespace FEXCore::CPU {
+#define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
+DEF_OP(VInsGPR) {
+  const auto Op = IROp->C<IR::IROp_VInsGPR>();
+  const auto OpSize = IROp->Size;
+
+  const auto ElementSize = Op->Header.ElementSize;
+  const auto ElementSizeBits = ElementSize * 8;
+  constexpr auto SSEBitSize = Core::CPUState::XMM_SSE_REG_SIZE * 8;
+
+  const uint64_t Offset = Op->DestIdx * ElementSizeBits;
+  const auto InUpperLane = Offset >= SSEBitSize;
+
+  __uint128_t Mask = (1ULL << ElementSizeBits) - 1;
+  if (ElementSize == 8) {
+    Mask = ~0ULL;
+  }
+
+  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;
+
+  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->Src), Op->Header.ElementSize);
+}
+
+DEF_OP(Float_FromGPR_S) {
+  auto Op = IROp->C<IR::IROp_Float_FromGPR_S>();
+
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  switch (Conv) {
+    case 0x0404: { // Float <- int32_t
+      const float Dst = (float)*GetSrc<int32_t*>(Data->SSAData, Op->Src);
+      memcpy(GDP, &Dst, Op->Header.ElementSize);
+      break;
+    }
+    case 0x0408: { // Float <- int64_t
+      const float Dst = (float)*GetSrc<int64_t*>(Data->SSAData, Op->Src);
+      memcpy(GDP, &Dst, Op->Header.ElementSize);
+      break;
+    }
+    case 0x0804: { // Double <- int32_t
+      const double Dst = (double)*GetSrc<int32_t*>(Data->SSAData, Op->Src);
+      memcpy(GDP, &Dst, Op->Header.ElementSize);
+      break;
+    }
+    case 0x0808: { // Double <- int64_t
+      const double Dst = (double)*GetSrc<int64_t*>(Data->SSAData, Op->Src);
+      memcpy(GDP, &Dst, Op->Header.ElementSize);
+      break;
+    }
+  }
+}
+
+DEF_OP(Float_FToF) {
+  auto Op = IROp->C<IR::IROp_Float_FToF>();
+  const uint16_t Conv = (Op->Header.ElementSize << 8) | Op->SrcElementSize;
+  switch (Conv) {
+    case 0x0804: { // Double <- Float
+      const double Dst = (double)*GetSrc<float*>(Data->SSAData, Op->Scalar);
+      memcpy(GDP, &Dst, 8);
+      break;
+    }
+    case 0x0408: { // Float <- Double
+      const float Dst = (float)*GetSrc<double*>(Data->SSAData, Op->Scalar);
+      memcpy(GDP, &Dst, 4);
+      break;
+    }
+    default: LOGMAN_MSG_A_FMT("Unknown FCVT sizes: 0x{:x}", Conv);
+  }
+}
+
+DEF_OP(Vector_SToF) {
+  auto Op = IROp->C<IR::IROp_Vector_SToF>();
+  const uint8_t OpSize = IROp->Size;
+
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
+
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
+
+  const auto Func = [](auto a, auto min, auto max) { return a; };
+  switch (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: {}", ElementSize);
+      break;
+  }
+  memcpy(GDP, Tmp, OpSize);
+}
+
+DEF_OP(Vector_FToZS) {
+  const auto Op = IROp->C<IR::IROp_Vector_FToZS>();
+  const uint8_t OpSize = IROp->Size;
+
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
+
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
+
+  const auto Func = [](auto a, auto min, auto max) { return std::trunc(a); };
+  switch (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: {}", ElementSize);
+      break;
+  }
+  memcpy(GDP, Tmp, OpSize);
+}
+
+DEF_OP(Vector_FToS) {
+  const auto Op = IROp->C<IR::IROp_Vector_FToS>();
+  const uint8_t OpSize = IROp->Size;
+
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
+
+  const uint8_t ElementSize = Op->Header.ElementSize;
+  const uint8_t Elements = OpSize / ElementSize;
+
+  const auto Func = [](auto a, auto min, auto max) { return std::nearbyint(a); };
+  switch (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: {}", ElementSize);
+      break;
+  }
+  memcpy(GDP, Tmp, OpSize);
+}
+
+DEF_OP(Vector_FToF) {
+  const auto Op = IROp->C<IR::IROp_Vector_FToF>();
+  const uint8_t OpSize = IROp->Size;
+
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
+
+  const uint16_t ElementSize = Op->Header.ElementSize;
+  const uint16_t Conv = (ElementSize << 8) | Op->SrcElementSize;
+
+  const auto Func = [](auto a, auto min, auto max) { return a; };
+  switch (Conv) {
+    case 0x0804: { // Double <- float
+      // Only the lower elements from the source
+      // This uses half the source elements
+      uint8_t Elements = OpSize / 8;
+      DO_VECTOR_1SRC_2TYPE_OP_NOSIZE(double, float, Func, 0, 0)
+      break;
+    }
+    case 0x0408: { // Float <- Double
+      // Little bit tricky here
+      // 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 == 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;
+  }
+  memcpy(GDP, Tmp, OpSize);
+}
+
+DEF_OP(Vector_FToI) {
+  const auto Op = IROp->C<IR::IROp_Vector_FToI>();
+  const uint8_t OpSize = IROp->Size;
+
+  void *Src = GetSrc<void*>(Data->SSAData, Op->Vector);
+  uint8_t Tmp[Core::CPUState::XMM_AVX_REG_SIZE]{};
+
+  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 (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 (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 (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 (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 (ElementSize) {
+        DO_VECTOR_1SRC_OP(4, float, Func_Host)
+        DO_VECTOR_1SRC_OP(8, double, Func_Host)
+      }
+    break;
+  }
+  memcpy(GDP, Tmp, OpSize);
+}
+
+#undef DEF_OP
+
+} // namespace FEXCore::CPU

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

@@ -0,0 +1,556 @@
+/*
+$info$
+tags: backend|interpreter
+$end_info$
+*/
+
+#include "Interface/Core/Interpreter/InterpreterClass.h"
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/Interpreter/InterpreterDefines.h"
+
+#include <cstdint>
+
+namespace AES {
+  static __uint128_t InvShiftRows(uint8_t *State) {
+    uint8_t Shifted[16] = {
+      State[0], State[13], State[10], State[7],
+      State[4], State[1], State[14], State[11],
+      State[8], State[5], State[2], State[15],
+      State[12], State[9], State[6], State[3],
+    };
+    __uint128_t Res{};
+    memcpy(&Res, Shifted, 16);
+    return Res;
+  }
+
+  static __uint128_t InvSubBytes(uint8_t *State) {
+    // 16x16 matrix table
+    static const uint8_t InvSubstitutionTable[256] = {
+      0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38, 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb,
+      0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87, 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb,
+      0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d, 0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e,
+      0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2, 0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25,
+      0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16, 0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92,
+      0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda, 0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84,
+      0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a, 0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06,
+      0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02, 0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b,
+      0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea, 0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73,
+      0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85, 0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e,
+      0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89, 0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b,
+      0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20, 0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4,
+      0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31, 0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f,
+      0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d, 0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef,
+      0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0, 0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61,
+      0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26, 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d,
+    };
+
+    // Uses a byte substitution table with a constant set of values
+    // Needs to do a table look up
+    uint8_t Substituted[16];
+    for (size_t i = 0; i < 16; ++i) {
+      Substituted[i] = InvSubstitutionTable[State[i]];
+    }
+
+    __uint128_t Res{};
+    memcpy(&Res, Substituted, 16);
+    return Res;
+  }
+
+  static __uint128_t ShiftRows(uint8_t *State) {
+    uint8_t Shifted[16] = {
+      State[0], State[5], State[10], State[15],
+      State[4], State[9], State[14], State[3],
+      State[8], State[13], State[2], State[7],
+      State[12], State[1], State[6], State[11],
+    };
+    __uint128_t Res{};
+    memcpy(&Res, Shifted, 16);
+    return Res;
+  }
+
+  static __uint128_t SubBytes(uint8_t *State, size_t Bytes) {
+    // 16x16 matrix table
+    static const uint8_t SubstitutionTable[256] = {
+      0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5, 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76,
+      0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0, 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0,
+      0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc, 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15,
+      0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a, 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75,
+      0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0, 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84,
+      0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b, 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf,
+      0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85, 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8,
+      0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5, 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2,
+      0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17, 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73,
+      0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88, 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb,
+      0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c, 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79,
+      0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9, 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08,
+      0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6, 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a,
+      0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e, 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e,
+      0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94, 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf,
+      0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68, 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16,
+    };
+    // Uses a byte substitution table with a constant set of values
+    // Needs to do a table look up
+    uint8_t Substituted[16];
+    Bytes = std::min(Bytes, (size_t)16);
+    for (size_t i = 0; i < Bytes; ++i) {
+      Substituted[i] = SubstitutionTable[State[i]];
+    }
+
+    __uint128_t Res{};
+    memcpy(&Res, Substituted, Bytes);
+    return Res;
+  }
+
+  static uint8_t FFMul02(uint8_t in) {
+    static const uint8_t FFMul02[256] = {
+      0x00, 0x02, 0x04, 0x06, 0x08, 0x0a, 0x0c, 0x0e, 0x10, 0x12, 0x14, 0x16, 0x18, 0x1a, 0x1c, 0x1e,
+      0x20, 0x22, 0x24, 0x26, 0x28, 0x2a, 0x2c, 0x2e, 0x30, 0x32, 0x34, 0x36, 0x38, 0x3a, 0x3c, 0x3e,
+      0x40, 0x42, 0x44, 0x46, 0x48, 0x4a, 0x4c, 0x4e, 0x50, 0x52, 0x54, 0x56, 0x58, 0x5a, 0x5c, 0x5e,
+      0x60, 0x62, 0x64, 0x66, 0x68, 0x6a, 0x6c, 0x6e, 0x70, 0x72, 0x74, 0x76, 0x78, 0x7a, 0x7c, 0x7e,
+      0x80, 0x82, 0x84, 0x86, 0x88, 0x8a, 0x8c, 0x8e, 0x90, 0x92, 0x94, 0x96, 0x98, 0x9a, 0x9c, 0x9e,
+      0xa0, 0xa2, 0xa4, 0xa6, 0xa8, 0xaa, 0xac, 0xae, 0xb0, 0xb2, 0xb4, 0xb6, 0xb8, 0xba, 0xbc, 0xbe,
+      0xc0, 0xc2, 0xc4, 0xc6, 0xc8, 0xca, 0xcc, 0xce, 0xd0, 0xd2, 0xd4, 0xd6, 0xd8, 0xda, 0xdc, 0xde,
+      0xe0, 0xe2, 0xe4, 0xe6, 0xe8, 0xea, 0xec, 0xee, 0xf0, 0xf2, 0xf4, 0xf6, 0xf8, 0xfa, 0xfc, 0xfe,
+      0x1b, 0x19, 0x1f, 0x1d, 0x13, 0x11, 0x17, 0x15, 0x0b, 0x09, 0x0f, 0x0d, 0x03, 0x01, 0x07, 0x05,
+      0x3b, 0x39, 0x3f, 0x3d, 0x33, 0x31, 0x37, 0x35, 0x2b, 0x29, 0x2f, 0x2d, 0x23, 0x21, 0x27, 0x25,
+      0x5b, 0x59, 0x5f, 0x5d, 0x53, 0x51, 0x57, 0x55, 0x4b, 0x49, 0x4f, 0x4d, 0x43, 0x41, 0x47, 0x45,
+      0x7b, 0x79, 0x7f, 0x7d, 0x73, 0x71, 0x77, 0x75, 0x6b, 0x69, 0x6f, 0x6d, 0x63, 0x61, 0x67, 0x65,
+      0x9b, 0x99, 0x9f, 0x9d, 0x93, 0x91, 0x97, 0x95, 0x8b, 0x89, 0x8f, 0x8d, 0x83, 0x81, 0x87, 0x85,
+      0xbb, 0xb9, 0xbf, 0xbd, 0xb3, 0xb1, 0xb7, 0xb5, 0xab, 0xa9, 0xaf, 0xad, 0xa3, 0xa1, 0xa7, 0xa5,
+      0xdb, 0xd9, 0xdf, 0xdd, 0xd3, 0xd1, 0xd7, 0xd5, 0xcb, 0xc9, 0xcf, 0xcd, 0xc3, 0xc1, 0xc7, 0xc5,
+      0xfb, 0xf9, 0xff, 0xfd, 0xf3, 0xf1, 0xf7, 0xf5, 0xeb, 0xe9, 0xef, 0xed, 0xe3, 0xe1, 0xe7, 0xe5,
+    };
+    return FFMul02[in];
+  }
+
+  static uint8_t FFMul03(uint8_t in) {
+    static const uint8_t FFMul03[256] = {
+      0x00, 0x03, 0x06, 0x05, 0x0c, 0x0f, 0x0a, 0x09, 0x18, 0x1b, 0x1e, 0x1d, 0x14, 0x17, 0x12, 0x11,
+      0x30, 0x33, 0x36, 0x35, 0x3c, 0x3f, 0x3a, 0x39, 0x28, 0x2b, 0x2e, 0x2d, 0x24, 0x27, 0x22, 0x21,
+      0x60, 0x63, 0x66, 0x65, 0x6c, 0x6f, 0x6a, 0x69, 0x78, 0x7b, 0x7e, 0x7d, 0x74, 0x77, 0x72, 0x71,
+      0x50, 0x53, 0x56, 0x55, 0x5c, 0x5f, 0x5a, 0x59, 0x48, 0x4b, 0x4e, 0x4d, 0x44, 0x47, 0x42, 0x41,
+      0xc0, 0xc3, 0xc6, 0xc5, 0xcc, 0xcf, 0xca, 0xc9, 0xd8, 0xdb, 0xde, 0xdd, 0xd4, 0xd7, 0xd2, 0xd1,
+      0xf0, 0xf3, 0xf6, 0xf5, 0xfc, 0xff, 0xfa, 0xf9, 0xe8, 0xeb, 0xee, 0xed, 0xe4, 0xe7, 0xe2, 0xe1,
+      0xa0, 0xa3, 0xa6, 0xa5, 0xac, 0xaf, 0xaa, 0xa9, 0xb8, 0xbb, 0xbe, 0xbd, 0xb4, 0xb7, 0xb2, 0xb1,
+      0x90, 0x93, 0x96, 0x95, 0x9c, 0x9f, 0x9a, 0x99, 0x88, 0x8b, 0x8e, 0x8d, 0x84, 0x87, 0x82, 0x81,
+      0x9b, 0x98, 0x9d, 0x9e, 0x97, 0x94, 0x91, 0x92, 0x83, 0x80, 0x85, 0x86, 0x8f, 0x8c, 0x89, 0x8a,
+      0xab, 0xa8, 0xad, 0xae, 0xa7, 0xa4, 0xa1, 0xa2, 0xb3, 0xb0, 0xb5, 0xb6, 0xbf, 0xbc, 0xb9, 0xba,
+      0xfb, 0xf8, 0xfd, 0xfe, 0xf7, 0xf4, 0xf1, 0xf2, 0xe3, 0xe0, 0xe5, 0xe6, 0xef, 0xec, 0xe9, 0xea,
+      0xcb, 0xc8, 0xcd, 0xce, 0xc7, 0xc4, 0xc1, 0xc2, 0xd3, 0xd0, 0xd5, 0xd6, 0xdf, 0xdc, 0xd9, 0xda,
+      0x5b, 0x58, 0x5d, 0x5e, 0x57, 0x54, 0x51, 0x52, 0x43, 0x40, 0x45, 0x46, 0x4f, 0x4c, 0x49, 0x4a,
+      0x6b, 0x68, 0x6d, 0x6e, 0x67, 0x64, 0x61, 0x62, 0x73, 0x70, 0x75, 0x76, 0x7f, 0x7c, 0x79, 0x7a,
+      0x3b, 0x38, 0x3d, 0x3e, 0x37, 0x34, 0x31, 0x32, 0x23, 0x20, 0x25, 0x26, 0x2f, 0x2c, 0x29, 0x2a,
+      0x0b, 0x08, 0x0d, 0x0e, 0x07, 0x04, 0x01, 0x02, 0x13, 0x10, 0x15, 0x16, 0x1f, 0x1c, 0x19, 0x1a,
+    };
+    return FFMul03[in];
+  }
+
+  static __uint128_t MixColumns(uint8_t *State) {
+    uint8_t In0[16] = {
+      State[0], State[4], State[8], State[12],
+      State[1], State[5], State[9], State[13],
+      State[2], State[6], State[10], State[14],
+      State[3], State[7], State[11], State[15],
+    };
+
+    uint8_t Out0[4]{};
+    uint8_t Out1[4]{};
+    uint8_t Out2[4]{};
+    uint8_t Out3[4]{};
+
+    for (size_t i = 0; i < 4; ++i) {
+      Out0[i] = FFMul02(In0[0 + i]) ^ FFMul03(In0[4 + i]) ^ In0[8 + i] ^ In0[12 + i];
+      Out1[i] = In0[0 + i] ^ FFMul02(In0[4 + i]) ^ FFMul03(In0[8 + i]) ^ In0[12 + i];
+      Out2[i] = In0[0 + i] ^ In0[4 + i] ^ FFMul02(In0[8 + i]) ^ FFMul03(In0[12 + i]);
+      Out3[i] = FFMul03(In0[0 + i]) ^ In0[4 + i] ^ In0[8 + i] ^ FFMul02(In0[12 + i]);
+    }
+
+    uint8_t OutArray[16] = {
+      Out0[0], Out1[0], Out2[0], Out3[0],
+      Out0[1], Out1[1], Out2[1], Out3[1],
+      Out0[2], Out1[2], Out2[2], Out3[2],
+      Out0[3], Out1[3], Out2[3], Out3[3],
+    };
+    __uint128_t Res{};
+    memcpy(&Res, OutArray, 16);
+    return Res;
+  }
+
+  static uint8_t FFMul09(uint8_t in) {
+    static const uint8_t FFMul09[256] = {
+      0x00, 0x09, 0x12, 0x1b, 0x24, 0x2d, 0x36, 0x3f, 0x48, 0x41, 0x5a, 0x53, 0x6c, 0x65, 0x7e, 0x77,
+      0x90, 0x99, 0x82, 0x8b, 0xb4, 0xbd, 0xa6, 0xaf, 0xd8, 0xd1, 0xca, 0xc3, 0xfc, 0xf5, 0xee, 0xe7,
+      0x3b, 0x32, 0x29, 0x20, 0x1f, 0x16, 0x0d, 0x04, 0x73, 0x7a, 0x61, 0x68, 0x57, 0x5e, 0x45, 0x4c,
+      0xab, 0xa2, 0xb9, 0xb0, 0x8f, 0x86, 0x9d, 0x94, 0xe3, 0xea, 0xf1, 0xf8, 0xc7, 0xce, 0xd5, 0xdc,
+      0x76, 0x7f, 0x64, 0x6d, 0x52, 0x5b, 0x40, 0x49, 0x3e, 0x37, 0x2c, 0x25, 0x1a, 0x13, 0x08, 0x01,
+      0xe6, 0xef, 0xf4, 0xfd, 0xc2, 0xcb, 0xd0, 0xd9, 0xae, 0xa7, 0xbc, 0xb5, 0x8a, 0x83, 0x98, 0x91,
+      0x4d, 0x44, 0x5f, 0x56, 0x69, 0x60, 0x7b, 0x72, 0x05, 0x0c, 0x17, 0x1e, 0x21, 0x28, 0x33, 0x3a,
+      0xdd, 0xd4, 0xcf, 0xc6, 0xf9, 0xf0, 0xeb, 0xe2, 0x95, 0x9c, 0x87, 0x8e, 0xb1, 0xb8, 0xa3, 0xaa,
+      0xec, 0xe5, 0xfe, 0xf7, 0xc8, 0xc1, 0xda, 0xd3, 0xa4, 0xad, 0xb6, 0xbf, 0x80, 0x89, 0x92, 0x9b,
+      0x7c, 0x75, 0x6e, 0x67, 0x58, 0x51, 0x4a, 0x43, 0x34, 0x3d, 0x26, 0x2f, 0x10, 0x19, 0x02, 0x0b,
+      0xd7, 0xde, 0xc5, 0xcc, 0xf3, 0xfa, 0xe1, 0xe8, 0x9f, 0x96, 0x8d, 0x84, 0xbb, 0xb2, 0xa9, 0xa0,
+      0x47, 0x4e, 0x55, 0x5c, 0x63, 0x6a, 0x71, 0x78, 0x0f, 0x06, 0x1d, 0x14, 0x2b, 0x22, 0x39, 0x30,
+      0x9a, 0x93, 0x88, 0x81, 0xbe, 0xb7, 0xac, 0xa5, 0xd2, 0xdb, 0xc0, 0xc9, 0xf6, 0xff, 0xe4, 0xed,
+      0x0a, 0x03, 0x18, 0x11, 0x2e, 0x27, 0x3c, 0x35, 0x42, 0x4b, 0x50, 0x59, 0x66, 0x6f, 0x74, 0x7d,
+      0xa1, 0xa8, 0xb3, 0xba, 0x85, 0x8c, 0x97, 0x9e, 0xe9, 0xe0, 0xfb, 0xf2, 0xcd, 0xc4, 0xdf, 0xd6,
+      0x31, 0x38, 0x23, 0x2a, 0x15, 0x1c, 0x07, 0x0e, 0x79, 0x70, 0x6b, 0x62, 0x5d, 0x54, 0x4f, 0x46,
+    };
+    return FFMul09[in];
+  }
+
+  static uint8_t FFMul0B(uint8_t in) {
+    static const uint8_t FFMul0B[256] = {
+      0x00, 0x0b, 0x16, 0x1d, 0x2c, 0x27, 0x3a, 0x31, 0x58, 0x53, 0x4e, 0x45, 0x74, 0x7f, 0x62, 0x69,
+      0xb0, 0xbb, 0xa6, 0xad, 0x9c, 0x97, 0x8a, 0x81, 0xe8, 0xe3, 0xfe, 0xf5, 0xc4, 0xcf, 0xd2, 0xd9,
+      0x7b, 0x70, 0x6d, 0x66, 0x57, 0x5c, 0x41, 0x4a, 0x23, 0x28, 0x35, 0x3e, 0x0f, 0x04, 0x19, 0x12,
+      0xcb, 0xc0, 0xdd, 0xd6, 0xe7, 0xec, 0xf1, 0xfa, 0x93, 0x98, 0x85, 0x8e, 0xbf, 0xb4, 0xa9, 0xa2,
+      0xf6, 0xfd, 0xe0, 0xeb, 0xda, 0xd1, 0xcc, 0xc7, 0xae, 0xa5, 0xb8, 0xb3, 0x82, 0x89, 0x94, 0x9f,
+      0x46, 0x4d, 0x50, 0x5b, 0x6a, 0x61, 0x7c, 0x77, 0x1e, 0x15, 0x08, 0x03, 0x32, 0x39, 0x24, 0x2f,
+      0x8d, 0x86, 0x9b, 0x90, 0xa1, 0xaa, 0xb7, 0xbc, 0xd5, 0xde, 0xc3, 0xc8, 0xf9, 0xf2, 0xef, 0xe4,
+      0x3d, 0x36, 0x2b, 0x20, 0x11, 0x1a, 0x07, 0x0c, 0x65, 0x6e, 0x73, 0x78, 0x49, 0x42, 0x5f, 0x54,
+      0xf7, 0xfc, 0xe1, 0xea, 0xdb, 0xd0, 0xcd, 0xc6, 0xaf, 0xa4, 0xb9, 0xb2, 0x83, 0x88, 0x95, 0x9e,
+      0x47, 0x4c, 0x51, 0x5a, 0x6b, 0x60, 0x7d, 0x76, 0x1f, 0x14, 0x09, 0x02, 0x33, 0x38, 0x25, 0x2e,
+      0x8c, 0x87, 0x9a, 0x91, 0xa0, 0xab, 0xb6, 0xbd, 0xd4, 0xdf, 0xc2, 0xc9, 0xf8, 0xf3, 0xee, 0xe5,
+      0x3c, 0x37, 0x2a, 0x21, 0x10, 0x1b, 0x06, 0x0d, 0x64, 0x6f, 0x72, 0x79, 0x48, 0x43, 0x5e, 0x55,
+      0x01, 0x0a, 0x17, 0x1c, 0x2d, 0x26, 0x3b, 0x30, 0x59, 0x52, 0x4f, 0x44, 0x75, 0x7e, 0x63, 0x68,
+      0xb1, 0xba, 0xa7, 0xac, 0x9d, 0x96, 0x8b, 0x80, 0xe9, 0xe2, 0xff, 0xf4, 0xc5, 0xce, 0xd3, 0xd8,
+      0x7a, 0x71, 0x6c, 0x67, 0x56, 0x5d, 0x40, 0x4b, 0x22, 0x29, 0x34, 0x3f, 0x0e, 0x05, 0x18, 0x13,
+      0xca, 0xc1, 0xdc, 0xd7, 0xe6, 0xed, 0xf0, 0xfb, 0x92, 0x99, 0x84, 0x8f, 0xbe, 0xb5, 0xa8, 0xa3,
+    };
+    return FFMul0B[in];
+  }
+
+  static uint8_t FFMul0D(uint8_t in) {
+    static const uint8_t FFMul0D[256] = {
+      0x00, 0x0d, 0x1a, 0x17, 0x34, 0x39, 0x2e, 0x23, 0x68, 0x65, 0x72, 0x7f, 0x5c, 0x51, 0x46, 0x4b,
+      0xd0, 0xdd, 0xca, 0xc7, 0xe4, 0xe9, 0xfe, 0xf3, 0xb8, 0xb5, 0xa2, 0xaf, 0x8c, 0x81, 0x96, 0x9b,
+      0xbb, 0xb6, 0xa1, 0xac, 0x8f, 0x82, 0x95, 0x98, 0xd3, 0xde, 0xc9, 0xc4, 0xe7, 0xea, 0xfd, 0xf0,
+      0x6b, 0x66, 0x71, 0x7c, 0x5f, 0x52, 0x45, 0x48, 0x03, 0x0e, 0x19, 0x14, 0x37, 0x3a, 0x2d, 0x20,
+      0x6d, 0x60, 0x77, 0x7a, 0x59, 0x54, 0x43, 0x4e, 0x05, 0x08, 0x1f, 0x12, 0x31, 0x3c, 0x2b, 0x26,
+      0xbd, 0xb0, 0xa7, 0xaa, 0x89, 0x84, 0x93, 0x9e, 0xd5, 0xd8, 0xcf, 0xc2, 0xe1, 0xec, 0xfb, 0xf6,
+      0xd6, 0xdb, 0xcc, 0xc1, 0xe2, 0xef, 0xf8, 0xf5, 0xbe, 0xb3, 0xa4, 0xa9, 0x8a, 0x87, 0x90, 0x9d,
+      0x06, 0x0b, 0x1c, 0x11, 0x32, 0x3f, 0x28, 0x25, 0x6e, 0x63, 0x74, 0x79, 0x5a, 0x57, 0x40, 0x4d,
+      0xda, 0xd7, 0xc0, 0xcd, 0xee, 0xe3, 0xf4, 0xf9, 0xb2, 0xbf, 0xa8, 0xa5, 0x86, 0x8b, 0x9c, 0x91,
+      0x0a, 0x07, 0x10, 0x1d, 0x3e, 0x33, 0x24, 0x29, 0x62, 0x6f, 0x78, 0x75, 0x56, 0x5b, 0x4c, 0x41,
+      0x61, 0x6c, 0x7b, 0x76, 0x55, 0x58, 0x4f, 0x42, 0x09, 0x04, 0x13, 0x1e, 0x3d, 0x30, 0x27, 0x2a,
+      0xb1, 0xbc, 0xab, 0xa6, 0x85, 0x88, 0x9f, 0x92, 0xd9, 0xd4, 0xc3, 0xce, 0xed, 0xe0, 0xf7, 0xfa,
+      0xb7, 0xba, 0xad, 0xa0, 0x83, 0x8e, 0x99, 0x94, 0xdf, 0xd2, 0xc5, 0xc8, 0xeb, 0xe6, 0xf1, 0xfc,
+      0x67, 0x6a, 0x7d, 0x70, 0x53, 0x5e, 0x49, 0x44, 0x0f, 0x02, 0x15, 0x18, 0x3b, 0x36, 0x21, 0x2c,
+      0x0c, 0x01, 0x16, 0x1b, 0x38, 0x35, 0x22, 0x2f, 0x64, 0x69, 0x7e, 0x73, 0x50, 0x5d, 0x4a, 0x47,
+      0xdc, 0xd1, 0xc6, 0xcb, 0xe8, 0xe5, 0xf2, 0xff, 0xb4, 0xb9, 0xae, 0xa3, 0x80, 0x8d, 0x9a, 0x97,
+    };
+
+    return FFMul0D[in];
+  }
+
+  static uint8_t FFMul0E(uint8_t in) {
+    static const uint8_t FFMul0E[256] = {
+      0x00, 0x0e, 0x1c, 0x12, 0x38, 0x36, 0x24, 0x2a, 0x70, 0x7e, 0x6c, 0x62, 0x48, 0x46, 0x54, 0x5a,
+      0xe0, 0xee, 0xfc, 0xf2, 0xd8, 0xd6, 0xc4, 0xca, 0x90, 0x9e, 0x8c, 0x82, 0xa8, 0xa6, 0xb4, 0xba,
+      0xdb, 0xd5, 0xc7, 0xc9, 0xe3, 0xed, 0xff, 0xf1, 0xab, 0xa5, 0xb7, 0xb9, 0x93, 0x9d, 0x8f, 0x81,
+      0x3b, 0x35, 0x27, 0x29, 0x03, 0x0d, 0x1f, 0x11, 0x4b, 0x45, 0x57, 0x59, 0x73, 0x7d, 0x6f, 0x61,
+      0xad, 0xa3, 0xb1, 0xbf, 0x95, 0x9b, 0x89, 0x87, 0xdd, 0xd3, 0xc1, 0xcf, 0xe5, 0xeb, 0xf9, 0xf7,
+      0x4d, 0x43, 0x51, 0x5f, 0x75, 0x7b, 0x69, 0x67, 0x3d, 0x33, 0x21, 0x2f, 0x05, 0x0b, 0x19, 0x17,
+      0x76, 0x78, 0x6a, 0x64, 0x4e, 0x40, 0x52, 0x5c, 0x06, 0x08, 0x1a, 0x14, 0x3e, 0x30, 0x22, 0x2c,
+      0x96, 0x98, 0x8a, 0x84, 0xae, 0xa0, 0xb2, 0xbc, 0xe6, 0xe8, 0xfa, 0xf4, 0xde, 0xd0, 0xc2, 0xcc,
+      0x41, 0x4f, 0x5d, 0x53, 0x79, 0x77, 0x65, 0x6b, 0x31, 0x3f, 0x2d, 0x23, 0x09, 0x07, 0x15, 0x1b,
+      0xa1, 0xaf, 0xbd, 0xb3, 0x99, 0x97, 0x85, 0x8b, 0xd1, 0xdf, 0xcd, 0xc3, 0xe9, 0xe7, 0xf5, 0xfb,
+      0x9a, 0x94, 0x86, 0x88, 0xa2, 0xac, 0xbe, 0xb0, 0xea, 0xe4, 0xf6, 0xf8, 0xd2, 0xdc, 0xce, 0xc0,
+      0x7a, 0x74, 0x66, 0x68, 0x42, 0x4c, 0x5e, 0x50, 0x0a, 0x04, 0x16, 0x18, 0x32, 0x3c, 0x2e, 0x20,
+      0xec, 0xe2, 0xf0, 0xfe, 0xd4, 0xda, 0xc8, 0xc6, 0x9c, 0x92, 0x80, 0x8e, 0xa4, 0xaa, 0xb8, 0xb6,
+      0x0c, 0x02, 0x10, 0x1e, 0x34, 0x3a, 0x28, 0x26, 0x7c, 0x72, 0x60, 0x6e, 0x44, 0x4a, 0x58, 0x56,
+      0x37, 0x39, 0x2b, 0x25, 0x0f, 0x01, 0x13, 0x1d, 0x47, 0x49, 0x5b, 0x55, 0x7f, 0x71, 0x63, 0x6d,
+      0xd7, 0xd9, 0xcb, 0xc5, 0xef, 0xe1, 0xf3, 0xfd, 0xa7, 0xa9, 0xbb, 0xb5, 0x9f, 0x91, 0x83, 0x8d,
+    };
+
+    return FFMul0E[in];
+  }
+
+  static __uint128_t InvMixColumns(uint8_t *State) {
+    uint8_t In0[16] = {
+      State[0], State[4], State[8], State[12],
+      State[1], State[5], State[9], State[13],
+      State[2], State[6], State[10], State[14],
+      State[3], State[7], State[11], State[15],
+    };
+
+    uint8_t Out0[4]{};
+    uint8_t Out1[4]{};
+    uint8_t Out2[4]{};
+    uint8_t Out3[4]{};
+
+    for (size_t i = 0; i < 4; ++i) {
+      Out0[i] = FFMul0E(In0[0 + i]) ^ FFMul0B(In0[4 + i]) ^ FFMul0D(In0[8 + i]) ^ FFMul09(In0[12 + i]);
+      Out1[i] = FFMul09(In0[0 + i]) ^ FFMul0E(In0[4 + i]) ^ FFMul0B(In0[8 + i]) ^ FFMul0D(In0[12 + i]);
+      Out2[i] = FFMul0D(In0[0 + i]) ^ FFMul09(In0[4 + i]) ^ FFMul0E(In0[8 + i]) ^ FFMul0B(In0[12 + i]);
+      Out3[i] = FFMul0B(In0[0 + i]) ^ FFMul0D(In0[4 + i]) ^ FFMul09(In0[8 + i]) ^ FFMul0E(In0[12 + i]);
+    }
+
+    uint8_t OutArray[16] = {
+      Out0[0], Out1[0], Out2[0], Out3[0],
+      Out0[1], Out1[1], Out2[1], Out3[1],
+      Out0[2], Out1[2], Out2[2], Out3[2],
+      Out0[3], Out1[3], Out2[3], Out3[3],
+    };
+    __uint128_t Res{};
+    memcpy(&Res, OutArray, 16);
+    return Res;
+  }
+}
+
+namespace CRC32 {
+  // CRC32 per byte lookup table.
+  constexpr std::array<uint32_t, 256> CRC32CTable = []() consteval {
+    std::array<uint32_t, 256> Table{};
+
+    // Clang 11.x doesn't support bitreverse as a consteval
+    // constexpr uint32_t Polynomial = 0x1EDC6F41;
+    constexpr uint32_t PolynomialRev = 0x82F63B78; //__builtin_bitreverse32(Polynomial);
+
+    for (size_t Char = 0; Char < std::size(Table); ++Char) {
+      uint32_t CurrentChar = Char;
+      for (size_t i = 0; i < 8; ++i) {
+        if (CurrentChar & 1) {
+          CurrentChar = (CurrentChar >> 1) ^ PolynomialRev;
+        }
+        else {
+          CurrentChar >>= 1;
+        }
+      }
+      Table[Char] = CurrentChar;
+    }
+
+    return Table;
+  }();
+
+  uint32_t crc32cb(uint32_t Accumulator, uint8_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ data] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+
+  uint32_t crc32ch(uint32_t Accumulator, uint16_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 0) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 8) & 0xFF)] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+
+  uint32_t crc32cw(uint32_t Accumulator, uint32_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  0) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  8) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 16) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 24) & 0xFF)] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+
+  uint32_t crc32cx(uint32_t Accumulator, uint64_t data) {
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  0) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >>  8) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 16) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 24) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 32) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 40) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 48) & 0xFF)] ^ Accumulator >> 8;
+    Accumulator = CRC32CTable[(uint8_t)Accumulator ^ ((data >> 56) & 0xFF)] ^ Accumulator >> 8;
+    return Accumulator;
+  }
+}
+
+namespace FEXCore::CPU {
+#define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
+
+DEF_OP(AESImc) {
+  auto Op = IROp->C<IR::IROp_VAESImc>();
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Vector);
+
+  // Pseudo-code
+  // Dst = InvMixColumns(STATE)
+  __uint128_t Tmp{};
+  Tmp = AES::InvMixColumns(reinterpret_cast<uint8_t*>(&Src1));
+  memcpy(GDP, &Tmp, sizeof(Tmp));
+}
+
+DEF_OP(AESEnc) {
+  auto Op = IROp->C<IR::IROp_VAESEnc>();
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);
+
+  // Pseudo-code
+  // STATE = Src1
+  // RoundKey = Src2
+  // STATE = ShiftRows(STATE)
+  // STATE = SubBytes(STATE)
+  // STATE = MixColumns(STATE)
+  // Dst = STATE XOR RoundKey
+  __uint128_t Tmp{};
+  Tmp = AES::ShiftRows(reinterpret_cast<uint8_t*>(&Src1));
+  Tmp = AES::SubBytes(reinterpret_cast<uint8_t*>(&Tmp), 16);
+  Tmp = AES::MixColumns(reinterpret_cast<uint8_t*>(&Tmp));
+  Tmp = Tmp ^ Src2;
+  memcpy(GDP, &Tmp, sizeof(Tmp));
+}
+
+DEF_OP(AESEncLast) {
+  auto Op = IROp->C<IR::IROp_VAESEncLast>();
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);
+
+  // Pseudo-code
+  // STATE = Src1
+  // RoundKey = Src2
+  // STATE = ShiftRows(STATE)
+  // STATE = SubBytes(STATE)
+  // Dst = STATE XOR RoundKey
+  __uint128_t Tmp{};
+  Tmp = AES::ShiftRows(reinterpret_cast<uint8_t*>(&Src1));
+  Tmp = AES::SubBytes(reinterpret_cast<uint8_t*>(&Tmp), 16);
+  Tmp = Tmp ^ Src2;
+  memcpy(GDP, &Tmp, sizeof(Tmp));
+}
+
+DEF_OP(AESDec) {
+  auto Op = IROp->C<IR::IROp_VAESDec>();
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);
+
+  // Pseudo-code
+  // STATE = Src1
+  // RoundKey = Src2
+  // STATE = InvShiftRows(STATE)
+  // STATE = InvSubBytes(STATE)
+  // STATE = InvMixColumns(STATE)
+  // Dst = STATE XOR RoundKey
+  __uint128_t Tmp{};
+  Tmp = AES::InvShiftRows(reinterpret_cast<uint8_t*>(&Src1));
+  Tmp = AES::InvSubBytes(reinterpret_cast<uint8_t*>(&Tmp));
+  Tmp = AES::InvMixColumns(reinterpret_cast<uint8_t*>(&Tmp));
+  Tmp = Tmp ^ Src2;
+  memcpy(GDP, &Tmp, sizeof(Tmp));
+}
+
+DEF_OP(AESDecLast) {
+  auto Op = IROp->C<IR::IROp_VAESDecLast>();
+  auto Src1 = *GetSrc<__uint128_t*>(Data->SSAData, Op->State);
+  auto Src2 = *GetSrc<__uint128_t*>(Data->SSAData, Op->Key);
+
+  // Pseudo-code
+  // STATE = Src1
+  // RoundKey = Src2
+  // STATE = InvShiftRows(STATE)
+  // STATE = InvSubBytes(STATE)
+  // Dst = STATE XOR RoundKey
+  __uint128_t Tmp{};
+  Tmp = AES::InvShiftRows(reinterpret_cast<uint8_t*>(&Src1));
+  Tmp = AES::InvSubBytes(reinterpret_cast<uint8_t*>(&Tmp));
+  Tmp = Tmp ^ Src2;
+  memcpy(GDP, &Tmp, sizeof(Tmp));
+}
+
+DEF_OP(AESKeyGenAssist) {
+  auto Op = IROp->C<IR::IROp_VAESKeyGenAssist>();
+  const uint8_t *Src1 = GetSrc<uint8_t*>(Data->SSAData, Op->Src);
+
+  // Pseudo-code
+  // X3 = Src1[127:96]
+  // X2 = Src1[95:64]
+  // X1 = Src1[63:32]
+  // X0 = Src1[31:30]
+  // RCON = (Zext)rcon
+  // Dest[31:0] = SubWord(X1)
+  // Dest[63:32] = RotWord(SubWord(X1)) XOR RCON
+  // Dest[95:64] = SubWord(X3)
+  // Dest[127:96] = RotWord(SubWord(X3)) XOR RCON
+  __uint128_t Tmp{};
+  uint32_t X1{};
+  uint32_t X3{};
+  memcpy(&X1, &Src1[4], 4);
+  memcpy(&X3, &Src1[12], 4);
+  uint32_t SubWord_X1 = AES::SubBytes(reinterpret_cast<uint8_t*>(&X1), 4);
+  uint32_t SubWord_X3 = AES::SubBytes(reinterpret_cast<uint8_t*>(&X3), 4);
+
+  auto Ror = [] (auto In, auto R) {
+    auto RotateMask = sizeof(In) * 8 - 1;
+    R &= RotateMask;
+    return (In >> R) | (In << (sizeof(In) * 8 - R));
+  };
+
+  uint32_t Rot_X1 = Ror(SubWord_X1, 8);
+  uint32_t Rot_X3 = Ror(SubWord_X3, 8);
+
+  Tmp = Rot_X3 ^ Op->RCON;
+  Tmp <<= 32;
+  Tmp |= SubWord_X3;
+  Tmp <<= 32;
+  Tmp |= Rot_X1 ^ Op->RCON;
+  Tmp <<= 32;
+  Tmp |= SubWord_X1;
+  memcpy(GDP, &Tmp, sizeof(Tmp));
+}
+
+DEF_OP(CRC32) {
+  auto Op = IROp->C<IR::IROp_CRC32>();
+  uint32_t Src1 = *GetSrc<uint32_t*>(Data->SSAData, Op->Src1);
+  uint8_t *Src2 = GetSrc<uint8_t*>(Data->SSAData, Op->Src2);
+  uint32_t Tmp{};
+
+  switch (Op->SrcSize) {
+    case 1:
+      Tmp = CRC32::crc32cb(Src1, *(uint8_t*)Src2);
+      break;
+    case 2:
+      Tmp = CRC32::crc32ch(Src1, *(uint16_t*)Src2);
+      break;
+    case 4:
+      Tmp = CRC32::crc32cw(Src1, *(uint32_t*)Src2);
+      break;
+    case 8:
+      Tmp = CRC32::crc32cx(Src1, *(uint64_t*)Src2);
+      break;
+    default:
+      LOGMAN_MSG_A_FMT("Unknown CRC32C size: {}", Op->SrcSize);
+      break;
+
+  }
+  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

@@ -0,0 +1,423 @@
+/*
+$info$
+tags: backend|interpreter
+$end_info$
+*/
+
+#include "Interface/Core/Interpreter/InterpreterClass.h"
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/Interpreter/InterpreterDefines.h"
+
+#include "F80Ops.h"
+
+#include <cstdint>
+
+namespace FEXCore::CPU {
+#define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
+DEF_OP(F80LOADFCW) {
+  FEXCore::CPU::OpHandlers<IR::OP_F80LOADFCW>::handle(*GetSrc<uint16_t*>(Data->SSAData, IROp->Args[0]));
+}
+
+DEF_OP(F80ADD) {
+  auto Op = IROp->C<IR::IROp_F80Add>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  const uint8_t OpSize = IROp->Size;
+
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+
+  switch (OpSize) {
+    case 4: {
+      float Tmp = Src;
+      memcpy(GDP, &Tmp, OpSize);
+      break;
+    }
+    case 8: {
+      double Tmp = Src;
+      memcpy(GDP, &Tmp, OpSize);
+      break;
+    }
+    default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+  }
+}
+
+DEF_OP(F80CVTINT) {
+  auto Op = IROp->C<IR::IROp_F80CVTInt>();
+  const uint8_t OpSize = IROp->Size;
+
+  const auto Src = *GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src);
+
+  switch (OpSize) {
+    case 2: {
+      int16_t Tmp = (Op->Truncate? FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2t : FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2)(Src);
+      memcpy(GDP, &Tmp, sizeof(Tmp));
+      break;
+    }
+    case 4: {
+      int32_t Tmp = (Op->Truncate? FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4t : FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4)(Src);
+      memcpy(GDP, &Tmp, sizeof(Tmp));
+      break;
+    }
+    case 8: {
+      int64_t Tmp = (Op->Truncate? FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8t : FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8)(Src);
+      memcpy(GDP, &Tmp, sizeof(Tmp));
+      break;
+    }
+    default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+  }
+}
+
+DEF_OP(F80CVTTO) {
+  auto Op = IROp->C<IR::IROp_F80CVTTo>();
+
+  switch (Op->SrcSize) {
+    case 4: {
+      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->X80Src);
+      X80SoftFloat Tmp = Src;
+      memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
+      break;
+    }
+    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->SrcSize);
+  }
+}
+
+DEF_OP(F80CVTTOINT) {
+  auto Op = IROp->C<IR::IROp_F80CVTToInt>();
+
+  switch (Op->SrcSize) {
+    case 2: {
+      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->Src);
+      X80SoftFloat Tmp = Src;
+      memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
+      break;
+    }
+    default: LogMan::Msg::DFmt("Unhandled size: {}", Op->SrcSize);
+  }
+}
+
+DEF_OP(F80ROUND) {
+  auto Op = IROp->C<IR::IROp_F80Round>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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{};
+  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) &&
+      lt) {
+    ResultFlags |= (1 << IR::FCMP_FLAG_LT);
+  }
+  if (Op->Flags & (1 << IR::FCMP_FLAG_UNORDERED) &&
+      nan) {
+    ResultFlags |= (1 << IR::FCMP_FLAG_UNORDERED);
+  }
+  if (Op->Flags & (1 << IR::FCMP_FLAG_EQ) &&
+      eq) {
+    ResultFlags |= (1 << IR::FCMP_FLAG_EQ);
+  }
+
+  GD = ResultFlags;
+}
+
+DEF_OP(F80BCDLOAD) {
+  auto Op = IROp->C<IR::IROp_F80BCDLoad>();
+  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
+  // Only 0-9 is supported, A-F results in undefined data
+  // | 4 bit     | 4 bit    |
+  // | 10s place | 1s place |
+  // EG 0x48 = 48
+  // EG 0x4847 = 4847
+  // This gives us an 18digit value encoded in BCD
+  // The last byte lets us know if it negative or not
+  for (size_t i = 0; i < 9; ++i) {
+    uint8_t Digit = Src1[8 - i];
+    // First shift our last value over
+    BCD *= 100;
+
+    // Add the tens place digit
+    BCD += (Digit >> 4) * 10;
+
+    // Add the ones place digit
+    BCD += Digit & 0xF;
+  }
+
+  // Set negative flag once converted to x87
+  bool Negative = Src1[9] & 0x80;
+  X80SoftFloat Tmp;
+
+  Tmp = BCD;
+  Tmp.Sign = Negative;
+
+  memcpy(GDP, &Tmp, sizeof(X80SoftFloat));
+}
+
+DEF_OP(F80BCDSTORE) {
+  auto Op = IROp->C<IR::IROp_F80BCDStore>();
+  X80SoftFloat Src1 = X80SoftFloat::FRNDINT(*GetSrc<X80SoftFloat*>(Data->SSAData, Op->X80Src));
+  bool Negative = Src1.Sign;
+
+  // Clear the Sign bit
+  Src1.Sign = 0;
+
+  uint64_t Tmp = Src1;
+  uint8_t BCD[10]{};
+
+  for (size_t i = 0; i < 9; ++i) {
+    if (Tmp == 0) {
+      // Nothing left? Just leave
+      break;
+    }
+    // Extract the lower 100 values
+    uint8_t Digit = Tmp % 100;
+
+    // Now divide it for the next iteration
+    Tmp /= 100;
+
+    uint8_t UpperNibble = Digit / 10;
+    uint8_t LowerNibble = Digit % 10;
+
+    // Now store the BCD
+    BCD[i] = (UpperNibble << 4) | LowerNibble;
+  }
+
+  // Set negative flag once converted to x87
+  BCD[9] = Negative ? 0x80 : 0;
+
+  memcpy(GDP, BCD, 10);
+}
+
+DEF_OP(F64SIN) {
+  auto Op = IROp->C<IR::IROp_F64SIN>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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>();
+  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));
+}
+
+
+#undef DEF_OP
+
+} // namespace FEXCore::CPU

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

@@ -0,0 +1,399 @@
+#pragma once
+#include "Common/SoftFloat.h"
+#include "Common/SoftFloat-3e/softfloat.h"
+
+#include <FEXCore/IR/IR.h>
+
+namespace FEXCore::CPU {
+template<IR::IROps Op>
+struct OpHandlers {
+
+};
+
+template<>
+struct OpHandlers<IR::OP_F80CVTTO> {
+  static X80SoftFloat handle4(float src) {
+    return src;
+  }
+
+  static X80SoftFloat handle8(double src) {
+    return src;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80CMP> {
+  template<uint32_t Flags>
+  static uint64_t handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    bool eq, lt, nan;
+    uint64_t ResultFlags = 0;
+
+    X80SoftFloat::FCMP(Src1, Src2, &eq, &lt, &nan);
+    if (Flags & (1 << IR::FCMP_FLAG_LT) &&
+        lt) {
+      ResultFlags |= (1 << IR::FCMP_FLAG_LT);
+    }
+    if (Flags & (1 << IR::FCMP_FLAG_UNORDERED) &&
+        nan) {
+      ResultFlags |= (1 << IR::FCMP_FLAG_UNORDERED);
+    }
+    if (Flags & (1 << IR::FCMP_FLAG_EQ) &&
+        eq) {
+      ResultFlags |= (1 << IR::FCMP_FLAG_EQ);
+    }
+    return ResultFlags;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80CVT> {
+  static float handle4(X80SoftFloat src) {
+    return src;
+  }
+
+  static double handle8(X80SoftFloat src) {
+    return src;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80CVTINT> {
+  static  int16_t handle2(X80SoftFloat src) {
+    return src;
+  }
+
+  static int32_t handle4(X80SoftFloat src) {
+    return src;
+  }
+
+  static int64_t handle8(X80SoftFloat src) {
+    return src;
+  }
+
+  static  int16_t handle2t(X80SoftFloat src) {
+    auto rv = extF80_to_i32(src, softfloat_round_minMag, false);
+
+    if (rv > INT16_MAX) {
+      return INT16_MAX;
+    } else if (rv < INT16_MIN) {
+      return INT16_MIN;
+    } else {
+      return rv;
+    }
+  }
+
+  static int32_t handle4t(X80SoftFloat src) {
+    return extF80_to_i32(src, softfloat_round_minMag, false);
+  }
+
+  static int64_t handle8t(X80SoftFloat src) {
+    return extF80_to_i64(src, softfloat_round_minMag, false);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80CVTTOINT> {
+  static X80SoftFloat handle2(int16_t src) {
+    return src;
+  }
+
+  static X80SoftFloat handle4(int32_t src) {
+    return src;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80ROUND> {
+  static X80SoftFloat handle(X80SoftFloat Src1) {
+    return X80SoftFloat::FRNDINT(Src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80F2XM1> {
+  static X80SoftFloat handle(X80SoftFloat Src1) {
+    return X80SoftFloat::F2XM1(Src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80TAN> {
+  static X80SoftFloat handle(X80SoftFloat Src1) {
+    return X80SoftFloat::FTAN(Src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80SQRT> {
+  static X80SoftFloat handle(X80SoftFloat Src1) {
+    return X80SoftFloat::FSQRT(Src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80SIN> {
+  static X80SoftFloat handle(X80SoftFloat Src1) {
+    return X80SoftFloat::FSIN(Src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80COS> {
+  static X80SoftFloat handle(X80SoftFloat Src1) {
+    return X80SoftFloat::FCOS(Src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80XTRACT_EXP> {
+  static X80SoftFloat handle(X80SoftFloat Src1) {
+    return X80SoftFloat::FXTRACT_EXP(Src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80XTRACT_SIG> {
+  static X80SoftFloat handle(X80SoftFloat Src1) {
+    return X80SoftFloat::FXTRACT_SIG(Src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80ADD> {
+  static X80SoftFloat handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    return X80SoftFloat::FADD(Src1, Src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80SUB> {
+  static X80SoftFloat handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    return X80SoftFloat::FSUB(Src1, Src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80MUL> {
+  static X80SoftFloat handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    return X80SoftFloat::FMUL(Src1, Src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80DIV> {
+  static X80SoftFloat handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    return X80SoftFloat::FDIV(Src1, Src2);
+  }
+};
+
+
+template<>
+struct OpHandlers<IR::OP_F80FYL2X> {
+  static X80SoftFloat handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    return X80SoftFloat::FYL2X(Src1, Src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80ATAN> {
+  static X80SoftFloat handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    return X80SoftFloat::FATAN(Src1, Src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80FPREM1> {
+  static X80SoftFloat handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    return X80SoftFloat::FREM1(Src1, Src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80FPREM> {
+  static X80SoftFloat handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    return X80SoftFloat::FREM(Src1, Src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80SCALE> {
+  static X80SoftFloat handle(X80SoftFloat Src1, X80SoftFloat Src2) {
+    return X80SoftFloat::FSCALE(Src1, Src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64SIN> {
+  static double handle(double src) {
+    return sin(src);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64COS> {
+  static double handle(double src) {
+    return cos(src);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64TAN> {
+  static double handle(double src) {
+    return tan(src);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64F2XM1> {
+  static double handle(double src) {
+    return exp2(src) - 1.0;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64ATAN> {
+  static double handle(double src1, double src2) {
+    return atan2(src1, src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64FPREM> {
+  static double handle(double src1, double src2) {
+    return fmod(src1, src2);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64FPREM1> {
+  static double handle(double src1, double src2) {
+    return remainder(src1, src2);
+  }
+};
+
+
+template<>
+struct OpHandlers<IR::OP_F64FYL2X> {
+  static double handle(double src1, double src2) {
+    return src2 * log2(src1);
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F64SCALE> {
+  static double handle(double src1, double src2) {
+    double trunc = (double)(int64_t)(src2); //truncate
+    return src1 * exp2(trunc);
+  }
+};
+
+
+
+template<>
+struct OpHandlers<IR::OP_F80BCDSTORE> {
+  static X80SoftFloat handle(X80SoftFloat Src1) {
+    bool Negative = Src1.Sign;
+
+    Src1 = X80SoftFloat::FRNDINT(Src1);
+
+    // Clear the Sign bit
+    Src1.Sign = 0;
+
+    uint64_t Tmp = Src1;
+    X80SoftFloat Rv;
+    uint8_t *BCD = reinterpret_cast<uint8_t*>(&Rv);
+    memset(BCD, 0, 10);
+
+    for (size_t i = 0; i < 9; ++i) {
+      if (Tmp == 0) {
+        // Nothing left? Just leave
+        break;
+      }
+      // Extract the lower 100 values
+      uint8_t Digit = Tmp % 100;
+
+      // Now divide it for the next iteration
+      Tmp /= 100;
+
+      uint8_t UpperNibble = Digit / 10;
+      uint8_t LowerNibble = Digit % 10;
+
+      // Now store the BCD
+      BCD[i] = (UpperNibble << 4) | LowerNibble;
+    }
+
+    // Set negative flag once converted to x87
+    BCD[9] = Negative ? 0x80 : 0;
+
+    return Rv;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80BCDLOAD> {
+  static X80SoftFloat handle(X80SoftFloat Src) {
+    uint8_t *Src1 = reinterpret_cast<uint8_t *>(&Src);
+    uint64_t BCD{};
+    // We walk through each uint8_t and pull out the BCD encoding
+    // Each 4bit split is a digit
+    // Only 0-9 is supported, A-F results in undefined data
+    // | 4 bit     | 4 bit    |
+    // | 10s place | 1s place |
+    // EG 0x48 = 48
+    // EG 0x4847 = 4847
+    // This gives us an 18digit value encoded in BCD
+    // The last byte lets us know if it negative or not
+    for (size_t i = 0; i < 9; ++i) {
+      uint8_t Digit = Src1[8 - i];
+      // First shift our last value over
+      BCD *= 100;
+
+      // Add the tens place digit
+      BCD += (Digit >> 4) * 10;
+
+      // Add the ones place digit
+      BCD += Digit & 0xF;
+    }
+
+    // Set negative flag once converted to x87
+    bool Negative = Src1[9] & 0x80;
+    X80SoftFloat Tmp;
+
+    Tmp = BCD;
+    Tmp.Sign = Negative;
+    return Tmp;
+  }
+};
+
+template<>
+struct OpHandlers<IR::OP_F80LOADFCW> {
+  static void handle(uint16_t NewFCW) {
+
+    auto PC = (NewFCW >> 8) & 3;
+    switch(PC) {
+      case 0: extF80_roundingPrecision = 32; break;
+      case 2: extF80_roundingPrecision = 64; break;
+      case 3: extF80_roundingPrecision = 80; break;
+      case 1: LOGMAN_MSG_A_FMT("Invalid x87 precision mode, {}", PC);
+    }
+
+    auto RC = (NewFCW >> 10) & 3;
+    switch(RC) {
+      case 0:
+        softfloat_roundingMode = softfloat_round_near_even;
+        break;
+      case 1:
+        softfloat_roundingMode = softfloat_round_min;
+        break;
+      case 2:
+        softfloat_roundingMode = softfloat_round_max;
+        break;
+      case 3:
+        softfloat_roundingMode = softfloat_round_minMag;
+      break;
+    }
+  }
+};
+
+
+}

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

@@ -0,0 +1,21 @@
+/*
+$info$
+tags: backend|interpreter
+$end_info$
+*/
+
+#include "Interface/Core/Interpreter/InterpreterClass.h"
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/Interpreter/InterpreterDefines.h"
+
+#include <cstdint>
+
+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->Value) >> Op->Flag) & 1;
+}
+#undef DEF_OP
+
+} // namespace FEXCore::CPU

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;
 
@@ -20,31 +21,35 @@ using DestMapType = std::vector<uint32_t>;
 
 class InterpreterCore final : public CPUBackend {
 public:
-  explicit InterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread);
-  std::string GetName() override { return "Interpreter"; }
-  void *CompileCode(uint64_t Entry, FEXCore::IR::IRListView const *IR, FEXCore::Core::DebugData *DebugData, FEXCore::IR::RegisterAllocationData *RAData) override;
+  explicit InterpreterCore(Dispatcher *Dispatch,
+                           FEXCore::Core::InternalThreadState *Thread);
 
-  void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }
+  [[nodiscard]] std::string GetName() override { return "Interpreter"; }
 
-  bool NeedsOpDispatch() override { return true; }
+  [[nodiscard]] void *CompileCode(uint64_t Entry,
+                                  FEXCore::IR::IRListView const *IR,
+                                  FEXCore::Core::DebugData *DebugData,
+                                  FEXCore::IR::RegisterAllocationData *RAData, bool GDBEnabled) override;
 
-  void CreateAsmDispatch(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread);
+  [[nodiscard]] void *MapRegion(void* HostPtr, uint64_t, uint64_t) override { return HostPtr; }
 
-  bool HandleSIGBUS(int Signal, void *info, void *ucontext);
+  [[nodiscard]] bool NeedsOpDispatch() override { return true; }
+
+  static void InitializeSignalHandlers(FEXCore::Context::Context *CTX);
+  
+  void ClearCache() override;
 
 private:
-  FEXCore::Context::Context *CTX;
-  FEXCore::Core::InternalThreadState *State;
-
-  uint32_t AllocateTmpSpace(size_t Size);
-
-  template<typename Res>
-  Res GetDest(void* SSAData, IR::OrderedNodeWrapper Op);
-
-  template<typename Res>
-  Res GetSrc(void* SSAData, IR::OrderedNodeWrapper Src);
-
-  std::unique_ptr<Dispatcher> Dispatcher{};
+  size_t BufferUsed;
+  Dispatcher *Dispatch;
 };
 
-}
+template<typename T>
+T AtomicCompareAndSwap(T expected, T desired, T *addr);
+
+uint8_t AtomicFetchNeg(uint8_t *Addr);
+uint16_t AtomicFetchNeg(uint16_t *Addr);
+uint32_t AtomicFetchNeg(uint32_t *Addr);
+uint64_t AtomicFetchNeg(uint64_t *Addr);
+
+} // namespace FEXCore::CPU

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

@@ -9,127 +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 <bits/types/stack_t.h>
 #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);
+
+  ClearCache();
 }
 
-bool InterpreterCore::HandleSIGBUS(int Signal, void *info, void *ucontext) {
+void InterpreterCore::InitializeSignalHandlers(FEXCore::Context::Context *CTX) {
+
 #ifdef _M_ARM_64
-  constexpr bool is_arm64 = true;
-#else
-  constexpr bool is_arm64 = false;
+  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
+}
 
-  if constexpr (is_arm64) {
-    uint32_t *PC = reinterpret_cast<uint32_t*>(ArchHelpers::Context::GetPc(ucontext));
-    uint32_t Instr = PC[0];
-    if ((Instr & FEXCore::ArchHelpers::Arm64::CASPAL_MASK) == FEXCore::ArchHelpers::Arm64::CASPAL_INST) { // CASPAL
-      if (FEXCore::ArchHelpers::Arm64::HandleCASPAL(ucontext, info, Instr)) {
-        // Skip this instruction now
-        ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) + 4);
-        return true;
-      }
-      else {
-        LogMan::Msg::E("Unhandled JIT SIGBUS CASPAL: PC: %p Instruction: 0x%08x\n", PC, PC[0]);
-        return false;
-      }
-    }
-    else if ((Instr & FEXCore::ArchHelpers::Arm64::CASAL_MASK) == FEXCore::ArchHelpers::Arm64::CASAL_INST) { // CASAL
-      if (FEXCore::ArchHelpers::Arm64::HandleCASAL(ucontext, info, Instr)) {
-        // Skip this instruction now
-        ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) + 4);
-        return true;
-      }
-      else {
-        LogMan::Msg::E("Unhandled JIT SIGBUS CASAL: PC: %p Instruction: 0x%08x\n", PC, PC[0]);
-        return false;
-      }
-    }
-    else if ((Instr & FEXCore::ArchHelpers::Arm64::ATOMIC_MEM_MASK) == FEXCore::ArchHelpers::Arm64::ATOMIC_MEM_INST) { // Atomic memory op
-      if (FEXCore::ArchHelpers::Arm64::HandleAtomicMemOp(ucontext, info, Instr)) {
-        // Skip this instruction now
-        ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) + 4);
-        return true;
-      }
-      else {
-        uint8_t Op = (PC[0] >> 12) & 0xF;
-        LogMan::Msg::E("Unhandled JIT SIGBUS Atomic mem op 0x%02x: PC: %p Instruction: 0x%08x\n", Op, PC, PC[0]);
-        return false;
-      }
-    }
-    else if ((Instr & FEXCore::ArchHelpers::Arm64::LDAXR_MASK) == FEXCore::ArchHelpers::Arm64::LDAXR_INST) { // LDAXR*
-      uint64_t BytesToSkip = FEXCore::ArchHelpers::Arm64::HandleAtomicLoadstoreExclusive(ucontext, info);
-      if (BytesToSkip) {
-        // Skip this instruction now
-        ArchHelpers::Context::SetPc(ucontext, ArchHelpers::Context::GetPc(ucontext) + BytesToSkip);
-        return true;
-      }
-      else {
-        LogMan::Msg::E("Unhandled JIT SIGBUS LDAXR: PC: %p Instruction: 0x%08x\n", PC, PC[0]);
-        return false;
-      }
-    }
+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);
   }
-  return false;
-}
 
-InterpreterCore::InterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread)
-  : CTX {ctx}
-  , State {Thread} {
-  // Grab our space for temporary data
+  const auto BufferStart = CurrentCodeBuffer->Ptr + BufferUsed;
 
-  if (!CompileThread &&
-      CTX->Config.Core == FEXCore::Config::CONFIG_INTERPRETER) {
-    CreateAsmDispatch(ctx, Thread);
-    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);
+  auto DestBuffer = BufferStart;
 
-    CTX->SignalDelegation->RegisterHostSignalHandler(SIGBUS, [](FEXCore::Core::InternalThreadState *Thread, int Signal, void *info, void *ucontext) -> bool {
-      InterpreterCore *Core = reinterpret_cast<InterpreterCore*>(Thread->CPUBackend.get());
-      return Core->HandleSIGBUS(Signal, info, ucontext);
-    }, true);
-
-    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);
-    }
+  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::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::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, bool CompileThread) {
-  return std::make_unique<InterpreterCore>(ctx, Thread, CompileThread);
+std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *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

@@ -12,7 +12,11 @@ namespace FEXCore::Core {
 
 namespace FEXCore::CPU {
 class CPUBackend;
+struct DispatcherConfig;
 
-std::unique_ptr<CPUBackend> CreateInterpreterCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalThreadState *Thread, bool CompileThread);
+[[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

@@ -0,0 +1,184 @@
+#pragma once
+
+#include <FEXCore/IR/IR.h>
+
+#define GD *GetDest<uint64_t*>(Data->SSAData, Node)
+#define GDP GetDest<void*>(Data->SSAData, Node)
+
+#define DO_OP(size, type, func)              \
+  case size: {                                      \
+               auto *Dst_d  = reinterpret_cast<type*>(GDP);  \
+               auto *Src1_d = reinterpret_cast<type*>(Src1); \
+               auto *Src2_d = reinterpret_cast<type*>(Src2); \
+               *Dst_d = func(*Src1_d, *Src2_d);          \
+               break;                                            \
+             }
+#define DO_SCALAR_COMPARE_OP(size, type, type2, func)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type2*>(Tmp);  \
+  auto *Src1_d = reinterpret_cast<type*>(Src1); \
+  auto *Src2_d = reinterpret_cast<type*>(Src2); \
+  Dst_d[0] = func(Src1_d[0], Src2_d[0]);          \
+  break;                                            \
+  }
+
+#define DO_VECTOR_COMPARE_OP(size, type, type2, func)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type2*>(Tmp);  \
+  auto *Src1_d = reinterpret_cast<type*>(Src1); \
+  auto *Src2_d = reinterpret_cast<type*>(Src2); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = func(Src1_d[i], Src2_d[i]);          \
+  }                                                 \
+  break;                                            \
+  }
+#define DO_VECTOR_OP(size, type, func)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src1_d = reinterpret_cast<type*>(Src1); \
+  auto *Src2_d = reinterpret_cast<type*>(Src2); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = func(Src1_d[i], Src2_d[i]);          \
+  }                                                 \
+  break;                                            \
+  }
+#define DO_VECTOR_PAIR_OP(size, type, func)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src1_d = reinterpret_cast<type*>(Src1); \
+  auto *Src2_d = reinterpret_cast<type*>(Src2); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = func(Src1_d[i*2], Src1_d[i*2 + 1]);          \
+    Dst_d[i+Elements] = func(Src2_d[i*2], Src2_d[i*2 + 1]);          \
+  }                                                 \
+  break;                                            \
+  }
+#define DO_VECTOR_SCALAR_OP(size, type, func)\
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src1_d = reinterpret_cast<type*>(Src1); \
+  auto *Src2_d = reinterpret_cast<type*>(Src2); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = func(Src1_d[i], *Src2_d);          \
+  }                                                 \
+  break;                                            \
+  }
+#define DO_VECTOR_0SRC_OP(size, type, func)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = func();          \
+  }                                                 \
+  break;                                            \
+  }
+#define DO_VECTOR_1SRC_OP(size, type, func)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src_d = reinterpret_cast<type*>(Src); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = func(Src_d[i]);          \
+  }                                                 \
+  break;                                            \
+  }
+#define DO_VECTOR_REDUCE_1SRC_OP(size, type, func, start_val)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src_d = reinterpret_cast<type*>(Src); \
+  type begin = start_val;                           \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    begin = func(begin, Src_d[i]);          \
+  }                                                 \
+  Dst_d[0] = begin;                                 \
+  break;                                            \
+  }
+#define DO_VECTOR_SAT_OP(size, type, func, min, max)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src1_d = reinterpret_cast<type*>(Src1); \
+  auto *Src2_d = reinterpret_cast<type*>(Src2); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = func(Src1_d[i], Src2_d[i], min, max);          \
+  }                                                 \
+  break;                                            \
+  }
+
+#define DO_VECTOR_1SRC_2TYPE_OP(size, type, type2, func, min, max)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src_d = reinterpret_cast<type2*>(Src); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = (type)func(Src_d[i], min, max);          \
+  }                                                 \
+  break;                                            \
+  }
+
+#define DO_VECTOR_1SRC_2TYPE_OP_NOSIZE(type, type2, func, min, max)              \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src_d = reinterpret_cast<type2*>(Src); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = (type)func(Src_d[i], min, max);          \
+  }
+#define DO_VECTOR_1SRC_2TYPE_OP_TOP(size, type, type2, func, min, max)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src_d = reinterpret_cast<type2*>(Src2); \
+  memcpy(Dst_d, Src1, Elements * sizeof(type2));\
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i+Elements] = (type)func(Src_d[i], min, max);          \
+  }                                                 \
+  break;                                            \
+  }
+
+#define DO_VECTOR_1SRC_2TYPE_OP_TOP_SRC(size, type, type2, func, min, max)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src_d = reinterpret_cast<type2*>(Src); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = (type)func(Src_d[i+Elements], min, max);          \
+  }                                                 \
+  break;                                            \
+  }
+#define DO_VECTOR_2SRC_2TYPE_OP(size, type, type2, func)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src1_d = reinterpret_cast<type2*>(Src1); \
+  auto *Src2_d = reinterpret_cast<type2*>(Src2); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = (type)func((type)Src1_d[i], (type)Src2_d[i]);          \
+  }                                                 \
+  break;                                            \
+  }
+#define DO_VECTOR_2SRC_2TYPE_OP_TOP_SRC(size, type, type2, func)              \
+  case size: {                                      \
+  auto *Dst_d  = reinterpret_cast<type*>(Tmp);  \
+  auto *Src1_d = reinterpret_cast<type2*>(Src1); \
+  auto *Src2_d = reinterpret_cast<type2*>(Src2); \
+  for (uint8_t i = 0; i < Elements; ++i) {          \
+    Dst_d[i] = (type)func((type)Src1_d[i+Elements], (type)Src2_d[i+Elements]);          \
+  }                                                 \
+  break;                                            \
+  }
+
+struct InterpVector256 {
+  __uint128_t Lower;
+  __uint128_t Upper;
+};
+
+template<typename Res>
+Res GetDest(void* SSAData, FEXCore::IR::OrderedNodeWrapper Op) {
+  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<InterpVector256*>(SSAData)[Op.Value];
+  return reinterpret_cast<Res>(DstPtr);
+}
+
+
+template<typename Res>
+Res GetSrc(void* SSAData, FEXCore::IR::OrderedNodeWrapper Src) {
+  auto DstPtr = &reinterpret_cast<InterpVector256*>(SSAData)[Src.ID().Value];
+  return reinterpret_cast<Res>(DstPtr);
+}

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

@@ -0,0 +1,313 @@
+#include "FEXCore/Core/CoreState.h"
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/Interpreter/F80Ops.h"
+
+#include <cstddef>
+#include <cstdint>
+
+namespace FEXCore::CPU {
+
+template<typename R, typename... Args>
+static FallbackInfo GetFallbackInfo(R(*fn)(Args...), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_UNKNOWN, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(float), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F32, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(double), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F64, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int16_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_I16, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(void(*fn)(uint16_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_VOID_U16, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(int32_t), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_I32, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(float(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F32_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(double(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F64_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(double(*fn)(double), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F64_F64, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(double(*fn)(double,double), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F64_F64_F64, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(int16_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I16_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(int32_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I32_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(int64_t(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I64_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(uint64_t(*fn)(X80SoftFloat, X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_I64_F80_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F80, (void*)fn, HandlerIndex};
+}
+
+template<>
+FallbackInfo GetFallbackInfo(X80SoftFloat(*fn)(X80SoftFloat, X80SoftFloat), FEXCore::Core::FallbackHandlerIndex HandlerIndex) {
+  return {FABI_F80_F80_F80, (void*)fn, HandlerIndex};
+}
+
+void InterpreterOps::FillFallbackIndexPointers(uint64_t *Info) {
+  Info[Core::OPINDEX_F80LOADFCW] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80LOADFCW>::handle, Core::OPINDEX_F80LOADFCW).fn);
+  Info[Core::OPINDEX_F80CVTTO_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle4, Core::OPINDEX_F80CVTTO_4).fn);
+  Info[Core::OPINDEX_F80CVTTO_8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle8, Core::OPINDEX_F80CVTTO_8).fn);
+  Info[Core::OPINDEX_F80CVT_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle4, Core::OPINDEX_F80CVT_4).fn);
+  Info[Core::OPINDEX_F80CVT_8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle8, Core::OPINDEX_F80CVT_8).fn);
+  Info[Core::OPINDEX_F80CVTINT_2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2, Core::OPINDEX_F80CVTINT_2).fn);
+  Info[Core::OPINDEX_F80CVTINT_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4, Core::OPINDEX_F80CVTINT_4).fn);
+  Info[Core::OPINDEX_F80CVTINT_8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8, Core::OPINDEX_F80CVTINT_8).fn);
+  Info[Core::OPINDEX_F80CVTINT_TRUNC2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2t, Core::OPINDEX_F80CVTINT_TRUNC2).fn);
+  Info[Core::OPINDEX_F80CVTINT_TRUNC4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4t, Core::OPINDEX_F80CVTINT_TRUNC4).fn);
+  Info[Core::OPINDEX_F80CVTINT_TRUNC8] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8t, Core::OPINDEX_F80CVTINT_TRUNC8).fn);
+  Info[Core::OPINDEX_F80CMP_0] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<0>, Core::OPINDEX_F80CMP_0).fn);
+  Info[Core::OPINDEX_F80CMP_1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<1>, Core::OPINDEX_F80CMP_1).fn);
+  Info[Core::OPINDEX_F80CMP_2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<2>, Core::OPINDEX_F80CMP_2).fn);
+  Info[Core::OPINDEX_F80CMP_3] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<3>, Core::OPINDEX_F80CMP_3).fn);
+  Info[Core::OPINDEX_F80CMP_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<4>, Core::OPINDEX_F80CMP_4).fn);
+  Info[Core::OPINDEX_F80CMP_5] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<5>, Core::OPINDEX_F80CMP_5).fn);
+  Info[Core::OPINDEX_F80CMP_6] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<6>, Core::OPINDEX_F80CMP_6).fn);
+  Info[Core::OPINDEX_F80CMP_7] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<7>, Core::OPINDEX_F80CMP_7).fn);
+  Info[Core::OPINDEX_F80CVTTOINT_2] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle2, Core::OPINDEX_F80CVTTOINT_2).fn);
+  Info[Core::OPINDEX_F80CVTTOINT_4] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle4, Core::OPINDEX_F80CVTTOINT_4).fn);
+
+  // Unary
+  Info[Core::OPINDEX_F80ROUND] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80ROUND>::handle, Core::OPINDEX_F80ROUND).fn);
+  Info[Core::OPINDEX_F80F2XM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80F2XM1>::handle, Core::OPINDEX_F80F2XM1).fn);
+  Info[Core::OPINDEX_F80TAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80TAN>::handle, Core::OPINDEX_F80TAN).fn);
+  Info[Core::OPINDEX_F80SQRT] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SQRT>::handle, Core::OPINDEX_F80SQRT).fn);
+  Info[Core::OPINDEX_F80SIN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SIN>::handle, Core::OPINDEX_F80SIN).fn);
+  Info[Core::OPINDEX_F80COS] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80COS>::handle, Core::OPINDEX_F80COS).fn);
+  Info[Core::OPINDEX_F80XTRACT_EXP] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80XTRACT_EXP>::handle, Core::OPINDEX_F80XTRACT_EXP).fn);
+  Info[Core::OPINDEX_F80XTRACT_SIG] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80XTRACT_SIG>::handle, Core::OPINDEX_F80XTRACT_SIG).fn);
+  Info[Core::OPINDEX_F80BCDSTORE] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80BCDSTORE>::handle, Core::OPINDEX_F80BCDSTORE).fn);
+  Info[Core::OPINDEX_F80BCDLOAD] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80BCDLOAD>::handle, Core::OPINDEX_F80BCDLOAD).fn);
+
+  // Binary
+  Info[Core::OPINDEX_F80ADD] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80ADD>::handle, Core::OPINDEX_F80ADD).fn);
+  Info[Core::OPINDEX_F80SUB] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SUB>::handle, Core::OPINDEX_F80SUB).fn);
+  Info[Core::OPINDEX_F80MUL] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80MUL>::handle, Core::OPINDEX_F80MUL).fn);
+  Info[Core::OPINDEX_F80DIV] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80DIV>::handle, Core::OPINDEX_F80DIV).fn);
+  Info[Core::OPINDEX_F80FYL2X] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80FYL2X>::handle, Core::OPINDEX_F80FYL2X).fn);
+  Info[Core::OPINDEX_F80ATAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80ATAN>::handle, Core::OPINDEX_F80ATAN).fn);
+  Info[Core::OPINDEX_F80FPREM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80FPREM1>::handle, Core::OPINDEX_F80FPREM1).fn);
+  Info[Core::OPINDEX_F80FPREM] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80FPREM>::handle, Core::OPINDEX_F80FPREM).fn);
+  Info[Core::OPINDEX_F80SCALE] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80SCALE>::handle, Core::OPINDEX_F80SCALE).fn);
+
+  // Double Precision
+  Info[Core::OPINDEX_F64SIN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64SIN>::handle, Core::OPINDEX_F64SIN).fn);
+  Info[Core::OPINDEX_F64COS] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64COS>::handle, Core::OPINDEX_F64COS).fn);
+  Info[Core::OPINDEX_F64TAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64TAN>::handle, Core::OPINDEX_F64TAN).fn);
+  Info[Core::OPINDEX_F64ATAN] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64ATAN>::handle, Core::OPINDEX_F64ATAN).fn);
+  Info[Core::OPINDEX_F64F2XM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64F2XM1>::handle, Core::OPINDEX_F64F2XM1).fn);
+  Info[Core::OPINDEX_F64FYL2X] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64FYL2X>::handle, Core::OPINDEX_F64FYL2X).fn);
+  Info[Core::OPINDEX_F64FPREM] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64FPREM>::handle, Core::OPINDEX_F64FPREM).fn);
+  Info[Core::OPINDEX_F64FPREM1] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64FPREM1>::handle, Core::OPINDEX_F64FPREM1).fn);
+  Info[Core::OPINDEX_F64SCALE] = reinterpret_cast<uint64_t>(GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64SCALE>::handle, Core::OPINDEX_F64SCALE).fn);
+
+}
+
+bool InterpreterOps::GetFallbackHandler(IR::IROp_Header const *IROp, FallbackInfo *Info) {
+  uint8_t OpSize = IROp->Size;
+  switch(IROp->Op) {
+    case IR::OP_F80LOADFCW: {
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80LOADFCW>::handle, Core::OPINDEX_F80LOADFCW);
+      return true;
+    }
+
+    case IR::OP_F80CVTTO: {
+      auto Op = IROp->C<IR::IROp_F80CVTTo>();
+
+      switch (Op->SrcSize) {
+        case 4: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle4, Core::OPINDEX_F80CVTTO_4);
+          return true;
+        }
+        case 8: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTO>::handle8, Core::OPINDEX_F80CVTTO_8);
+          return true;
+        }
+      default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+    case IR::OP_F80CVT: {
+      switch (OpSize) {
+        case 4: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle4, Core::OPINDEX_F80CVT_4);
+          return true;
+        }
+        case 8: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVT>::handle8, Core::OPINDEX_F80CVT_8);
+          return true;
+        }
+        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+    case IR::OP_F80CVTINT: {
+      auto Op = IROp->C<IR::IROp_F80CVTInt>();
+
+      switch (OpSize) {
+        case 2: {
+          if (Op->Truncate) {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2t, Core::OPINDEX_F80CVTINT_TRUNC2);
+          }
+          else {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle2, Core::OPINDEX_F80CVTINT_2);
+          }
+          return true;
+        }
+        case 4: {
+          if (Op->Truncate) {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4t, Core::OPINDEX_F80CVTINT_TRUNC4);
+          }
+          else {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle4, Core::OPINDEX_F80CVTINT_4);
+          }
+          return true;
+        }
+        case 8: {
+          if (Op->Truncate) {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8t, Core::OPINDEX_F80CVTINT_TRUNC8);
+          }
+          else {
+            *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTINT>::handle8, Core::OPINDEX_F80CVTINT_8);
+          }
+          return true;
+        }
+        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+    case IR::OP_F80CMP: {
+      auto Op = IROp->C<IR::IROp_F80Cmp>();
+
+      static constexpr std::array handlers{
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<0>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<1>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<2>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<3>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<4>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<5>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<6>,
+        &FEXCore::CPU::OpHandlers<IR::OP_F80CMP>::handle<7>,
+      };
+
+      *Info = GetFallbackInfo(handlers[Op->Flags], (Core::FallbackHandlerIndex)(Core::OPINDEX_F80CMP_0 + Op->Flags));
+      return true;
+    }
+
+    case IR::OP_F80CVTTOINT: {
+      auto Op = IROp->C<IR::IROp_F80CVTToInt>();
+
+      switch (Op->SrcSize) {
+        case 2: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle2, Core::OPINDEX_F80CVTTOINT_2);
+          return true;
+        }
+        case 4: {
+          *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80CVTTOINT>::handle4, Core::OPINDEX_F80CVTTOINT_4);
+          return true;
+        }
+        default: LogMan::Msg::DFmt("Unhandled size: {}", OpSize);
+      }
+      break;
+    }
+
+#define COMMON_X87_OP(OP) \
+    case IR::OP_F80##OP: { \
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F80##OP>::handle, Core::OPINDEX_F80##OP); \
+      return true; \
+    }
+
+#define COMMON_F64_OP(OP) \
+    case IR::OP_F64##OP: { \
+      *Info = GetFallbackInfo(&FEXCore::CPU::OpHandlers<IR::OP_F64##OP>::handle, Core::OPINDEX_F64##OP); \
+      return true; \
+    }
+
+    // Unary
+    COMMON_X87_OP(ROUND)
+    COMMON_X87_OP(F2XM1)
+    COMMON_X87_OP(TAN)
+    COMMON_X87_OP(SQRT)
+    COMMON_X87_OP(SIN)
+    COMMON_X87_OP(COS)
+    COMMON_X87_OP(XTRACT_EXP)
+    COMMON_X87_OP(XTRACT_SIG)
+    COMMON_X87_OP(BCDSTORE)
+    COMMON_X87_OP(BCDLOAD)
+
+    // Binary
+    COMMON_X87_OP(ADD)
+    COMMON_X87_OP(SUB)
+    COMMON_X87_OP(MUL)
+    COMMON_X87_OP(DIV)
+    COMMON_X87_OP(FYL2X)
+    COMMON_X87_OP(ATAN)
+    COMMON_X87_OP(FPREM1)
+    COMMON_X87_OP(FPREM)
+    COMMON_X87_OP(SCALE)
+
+    // Double Precision Unary
+    COMMON_F64_OP(F2XM1)
+    COMMON_F64_OP(TAN)
+    COMMON_F64_OP(SIN)
+    COMMON_F64_OP(COS)
+
+    // Double Precision Binary
+    COMMON_F64_OP(FYL2X)
+    COMMON_F64_OP(ATAN)
+    COMMON_F64_OP(FPREM1)
+    COMMON_F64_OP(FPREM)
+    COMMON_F64_OP(SCALE)
+
+    default:
+      break;
+  }
+
+  return false;
+}
+
+
+}

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

@@ -1,6 +1,10 @@
 #pragma once
 #include <stdint.h>
 
+#include <FEXCore/Core/CoreState.h>
+#include <FEXCore/IR/IR.h>
+#include <FEXCore/IR/IntrusiveIRList.h>
+
 namespace FEXCore::Core {
   struct InternalThreadState;
 }
@@ -24,6 +28,8 @@ namespace FEXCore::CPU {
     FABI_F80_I32,
     FABI_F32_F80,
     FABI_F64_F80,
+    FABI_F64_F64,
+    FABI_F64_F64_F64,
     FABI_I16_F80,
     FABI_I32_F80,
     FABI_I64_F80,
@@ -35,12 +41,368 @@ namespace FEXCore::CPU {
   struct FallbackInfo {
     FallbackABI ABI;
     void *fn;
+    FEXCore::Core::FallbackHandlerIndex HandlerIndex;
   };
 
   class InterpreterOps {
 
     public:
-      static void InterpretIR(FEXCore::Core::InternalThreadState *Thread, uint64_t Entry, FEXCore::IR::IRListView *CurrentIR, FEXCore::Core::DebugData *DebugData);
-      static bool GetFallbackHandler(IR::IROp_Header *IROp, FallbackInfo *Info);
+      static void InterpretIR(FEXCore::Core::CpuStateFrame *Frame, FEXCore::IR::IRListView const *IR);
+      static void FillFallbackIndexPointers(uint64_t *Info);
+      static bool GetFallbackHandler(IR::IROp_Header const *IROp, FallbackInfo *Info);
+
+      struct IROpData {
+        FEXCore::Core::InternalThreadState *State{};
+        uint64_t CurrentEntry{};
+        FEXCore::IR::IRListView const *CurrentIR{};
+        volatile void *StackEntry{};
+        void *SSAData{};
+        struct {
+          bool Quit;
+          bool Redo;
+        } BlockResults{};
+
+        IR::NodeIterator BlockIterator{0, 0};
+      };
+
+#define DEF_OP(x) static void Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
+
+  ///< Unhandled handler
+  DEF_OP(Unhandled);
+
+  ///< No-op Handler
+  DEF_OP(NoOp);
+
+  ///< ALU Ops
+  DEF_OP(TruncElementPair);
+  DEF_OP(Constant);
+  DEF_OP(EntrypointOffset);
+  DEF_OP(InlineConstant);
+  DEF_OP(InlineEntrypointOffset);
+  DEF_OP(CycleCounter);
+  DEF_OP(Add);
+  DEF_OP(Sub);
+  DEF_OP(Neg);
+  DEF_OP(Mul);
+  DEF_OP(UMul);
+  DEF_OP(Div);
+  DEF_OP(UDiv);
+  DEF_OP(Rem);
+  DEF_OP(URem);
+  DEF_OP(MulH);
+  DEF_OP(UMulH);
+  DEF_OP(Or);
+  DEF_OP(And);
+  DEF_OP(Andn);
+  DEF_OP(Xor);
+  DEF_OP(Lshl);
+  DEF_OP(Lshr);
+  DEF_OP(Ashr);
+  DEF_OP(Rol);
+  DEF_OP(Ror);
+  DEF_OP(Extr);
+  DEF_OP(PDep);
+  DEF_OP(PExt);
+  DEF_OP(LDiv);
+  DEF_OP(LUDiv);
+  DEF_OP(LRem);
+  DEF_OP(LURem);
+  DEF_OP(Zext);
+  DEF_OP(Not);
+  DEF_OP(Popcount);
+  DEF_OP(FindLSB);
+  DEF_OP(FindMSB);
+  DEF_OP(FindTrailingZeros);
+  DEF_OP(CountLeadingZeroes);
+  DEF_OP(Rev);
+  DEF_OP(Bfi);
+  DEF_OP(Bfe);
+  DEF_OP(Sbfe);
+  DEF_OP(Select);
+  DEF_OP(VExtractToGPR);
+  DEF_OP(Float_ToGPR_ZU);
+  DEF_OP(Float_ToGPR_ZS);
+  DEF_OP(Float_ToGPR_S);
+  DEF_OP(FCmp);
+
+  ///< Atomic ops
+  DEF_OP(CASPair);
+  DEF_OP(CAS);
+  DEF_OP(AtomicAdd);
+  DEF_OP(AtomicSub);
+  DEF_OP(AtomicAnd);
+  DEF_OP(AtomicOr);
+  DEF_OP(AtomicXor);
+  DEF_OP(AtomicSwap);
+  DEF_OP(AtomicFetchAdd);
+  DEF_OP(AtomicFetchSub);
+  DEF_OP(AtomicFetchAnd);
+  DEF_OP(AtomicFetchOr);
+  DEF_OP(AtomicFetchXor);
+  DEF_OP(AtomicFetchNeg);
+
+  ///< Branch ops
+  DEF_OP(SignalReturn);
+  DEF_OP(CallbackReturn);
+  DEF_OP(ExitFunction);
+  DEF_OP(Jump);
+  DEF_OP(CondJump);
+  DEF_OP(Syscall);
+  DEF_OP(InlineSyscall);
+  DEF_OP(Thunk);
+  DEF_OP(ValidateCode);
+  DEF_OP(ThreadRemoveCodeEntry);
+  DEF_OP(CPUID);
+
+  ///< Conversion ops
+  DEF_OP(VInsGPR);
+  DEF_OP(VCastFromGPR);
+  DEF_OP(Float_FromGPR_S);
+  DEF_OP(Float_FToF);
+  DEF_OP(Vector_SToF);
+  DEF_OP(Vector_FToZS);
+  DEF_OP(Vector_FToS);
+  DEF_OP(Vector_FToF);
+  DEF_OP(Vector_FToI);
+
+  ///< Flag ops
+  DEF_OP(GetHostFlag);
+
+  ///< Memory ops
+  DEF_OP(LoadContext);
+  DEF_OP(StoreContext);
+  DEF_OP(LoadRegister);
+  DEF_OP(StoreRegister);
+  DEF_OP(LoadContextIndexed);
+  DEF_OP(StoreContextIndexed);
+  DEF_OP(SpillRegister);
+  DEF_OP(FillRegister);
+  DEF_OP(LoadFlag);
+  DEF_OP(StoreFlag);
+  DEF_OP(LoadMem);
+  DEF_OP(StoreMem);
+  DEF_OP(CacheLineClear);
+  DEF_OP(CacheLineZero);
+
+  ///< Misc ops
+  DEF_OP(EndBlock);
+  DEF_OP(Fence);
+  DEF_OP(Break);
+  DEF_OP(Phi);
+  DEF_OP(PhiValue);
+  DEF_OP(Print);
+  DEF_OP(GetRoundingMode);
+  DEF_OP(SetRoundingMode);
+  DEF_OP(ProcessorID);
+  DEF_OP(RDRAND);
+  DEF_OP(Yield);
+
+  ///< Move ops
+  DEF_OP(ExtractElementPair);
+  DEF_OP(CreateElementPair);
+  DEF_OP(Mov);
+
+  ///< Vector ops
+  DEF_OP(VectorZero);
+  DEF_OP(VectorImm);
+  DEF_OP(VMov);
+  DEF_OP(VAnd);
+  DEF_OP(VBic);
+  DEF_OP(VOr);
+  DEF_OP(VXor);
+  DEF_OP(VAdd);
+  DEF_OP(VSub);
+  DEF_OP(VUQAdd);
+  DEF_OP(VUQSub);
+  DEF_OP(VSQAdd);
+  DEF_OP(VSQSub);
+  DEF_OP(VAddP);
+  DEF_OP(VAddV);
+  DEF_OP(VUMinV);
+  DEF_OP(VURAvg);
+  DEF_OP(VAbs);
+  DEF_OP(VPopcount);
+  DEF_OP(VFAdd);
+  DEF_OP(VFAddP);
+  DEF_OP(VFSub);
+  DEF_OP(VFMul);
+  DEF_OP(VFDiv);
+  DEF_OP(VFMin);
+  DEF_OP(VFMax);
+  DEF_OP(VFRecp);
+  DEF_OP(VFSqrt);
+  DEF_OP(VFRSqrt);
+  DEF_OP(VNeg);
+  DEF_OP(VFNeg);
+  DEF_OP(VNot);
+  DEF_OP(VUMin);
+  DEF_OP(VSMin);
+  DEF_OP(VUMax);
+  DEF_OP(VSMax);
+  DEF_OP(VZip);
+  DEF_OP(VUnZip);
+  DEF_OP(VBSL);
+  DEF_OP(VCMPEQ);
+  DEF_OP(VCMPEQZ);
+  DEF_OP(VCMPGT);
+  DEF_OP(VCMPGTZ);
+  DEF_OP(VCMPLTZ);
+  DEF_OP(VFCMPEQ);
+  DEF_OP(VFCMPNEQ);
+  DEF_OP(VFCMPLT);
+  DEF_OP(VFCMPGT);
+  DEF_OP(VFCMPLE);
+  DEF_OP(VFCMPORD);
+  DEF_OP(VFCMPUNO);
+  DEF_OP(VUShl);
+  DEF_OP(VUShr);
+  DEF_OP(VSShr);
+  DEF_OP(VUShlS);
+  DEF_OP(VUShrS);
+  DEF_OP(VSShrS);
+  DEF_OP(VInsElement);
+  DEF_OP(VDupElement);
+  DEF_OP(VExtr);
+  DEF_OP(VUShrI);
+  DEF_OP(VSShrI);
+  DEF_OP(VShlI);
+  DEF_OP(VUShrNI);
+  DEF_OP(VUShrNI2);
+  DEF_OP(VSXTL);
+  DEF_OP(VSXTL2);
+  DEF_OP(VUXTL);
+  DEF_OP(VUXTL2);
+  DEF_OP(VSQXTN);
+  DEF_OP(VSQXTN2);
+  DEF_OP(VSQXTUN);
+  DEF_OP(VSQXTUN2);
+  DEF_OP(VUMul);
+  DEF_OP(VUMull);
+  DEF_OP(VSMul);
+  DEF_OP(VSMull);
+  DEF_OP(VUMull2);
+  DEF_OP(VSMull2);
+  DEF_OP(VUABDL);
+  DEF_OP(VTBL1);
+  DEF_OP(VRev64);
+
+  ///< Encryption ops
+  DEF_OP(AESImc);
+  DEF_OP(AESEnc);
+  DEF_OP(AESEncLast);
+  DEF_OP(AESDec);
+  DEF_OP(AESDecLast);
+  DEF_OP(AESKeyGenAssist);
+  DEF_OP(CRC32);
+  DEF_OP(PCLMUL);
+
+  ///< F80 ops
+  DEF_OP(F80LOADFCW);
+  DEF_OP(F80ADD);
+  DEF_OP(F80SUB);
+  DEF_OP(F80MUL);
+  DEF_OP(F80DIV);
+  DEF_OP(F80FYL2X);
+  DEF_OP(F80ATAN);
+  DEF_OP(F80FPREM1);
+  DEF_OP(F80FPREM);
+  DEF_OP(F80SCALE);
+  DEF_OP(F80CVT);
+  DEF_OP(F80CVTINT);
+  DEF_OP(F80CVTTO);
+  DEF_OP(F80CVTTOINT);
+  DEF_OP(F80ROUND);
+  DEF_OP(F80F2XM1);
+  DEF_OP(F80TAN);
+  DEF_OP(F80SQRT);
+  DEF_OP(F80SIN);
+  DEF_OP(F80COS);
+  DEF_OP(F80XTRACT_EXP);
+  DEF_OP(F80XTRACT_SIG);
+  DEF_OP(F80CMP);
+  DEF_OP(F80BCDLOAD);
+  DEF_OP(F80BCDSTORE);
+
+  //< F64 ops
+  DEF_OP(F64SIN);
+  DEF_OP(F64COS);
+  DEF_OP(F64TAN);
+  DEF_OP(F64F2XM1);
+  DEF_OP(F64ATAN);
+  DEF_OP(F64FPREM);
+  DEF_OP(F64FPREM1);
+  DEF_OP(F64FYL2X);
+  DEF_OP(F64SCALE);
+#undef DEF_OP
+  template<typename unsigned_type, typename signed_type, typename float_type>
+  [[nodiscard]] static bool IsConditionTrue(uint8_t Cond, uint64_t Src1, uint64_t Src2) {
+    bool CompResult = false;
+    switch (Cond) {
+      case FEXCore::IR::COND_EQ:
+        CompResult = static_cast<unsigned_type>(Src1) == static_cast<unsigned_type>(Src2);
+        break;
+      case FEXCore::IR::COND_NEQ:
+        CompResult = static_cast<unsigned_type>(Src1) != static_cast<unsigned_type>(Src2);
+        break;
+      case FEXCore::IR::COND_SGE:
+        CompResult = static_cast<signed_type>(Src1) >= static_cast<signed_type>(Src2);
+        break;
+      case FEXCore::IR::COND_SLT:
+        CompResult = static_cast<signed_type>(Src1) < static_cast<signed_type>(Src2);
+        break;
+      case FEXCore::IR::COND_SGT:
+        CompResult = static_cast<signed_type>(Src1) > static_cast<signed_type>(Src2);
+        break;
+      case FEXCore::IR::COND_SLE:
+        CompResult = static_cast<signed_type>(Src1) <= static_cast<signed_type>(Src2);
+        break;
+      case FEXCore::IR::COND_UGE:
+        CompResult = static_cast<unsigned_type>(Src1) >= static_cast<unsigned_type>(Src2);
+        break;
+      case FEXCore::IR::COND_ULT:
+        CompResult = static_cast<unsigned_type>(Src1) < static_cast<unsigned_type>(Src2);
+        break;
+      case FEXCore::IR::COND_UGT:
+        CompResult = static_cast<unsigned_type>(Src1) > static_cast<unsigned_type>(Src2);
+        break;
+      case FEXCore::IR::COND_ULE:
+        CompResult = static_cast<unsigned_type>(Src1) <= static_cast<unsigned_type>(Src2);
+        break;
+
+      case FEXCore::IR::COND_FLU:
+        CompResult = reinterpret_cast<float_type&>(Src1) < reinterpret_cast<float_type&>(Src2) || (std::isnan(reinterpret_cast<float_type&>(Src1)) || std::isnan(reinterpret_cast<float_type&>(Src2)));
+        break;
+      case FEXCore::IR::COND_FGE:
+        CompResult = reinterpret_cast<float_type&>(Src1) >= reinterpret_cast<float_type&>(Src2) && !(std::isnan(reinterpret_cast<float_type&>(Src1)) || std::isnan(reinterpret_cast<float_type&>(Src2)));
+        break;
+      case FEXCore::IR::COND_FLEU:
+        CompResult = reinterpret_cast<float_type&>(Src1) <= reinterpret_cast<float_type&>(Src2) || (std::isnan(reinterpret_cast<float_type&>(Src1)) || std::isnan(reinterpret_cast<float_type&>(Src2)));
+        break;
+      case FEXCore::IR::COND_FGT:
+        CompResult = reinterpret_cast<float_type&>(Src1) > reinterpret_cast<float_type&>(Src2) && !(std::isnan(reinterpret_cast<float_type&>(Src1)) || std::isnan(reinterpret_cast<float_type&>(Src2)));
+        break;
+      case FEXCore::IR::COND_FU:
+        CompResult = (std::isnan(reinterpret_cast<float_type&>(Src1)) || std::isnan(reinterpret_cast<float_type&>(Src2)));
+        break;
+      case FEXCore::IR::COND_FNU:
+        CompResult = !(std::isnan(reinterpret_cast<float_type&>(Src1)) || std::isnan(reinterpret_cast<float_type&>(Src2)));
+        break;
+      case FEXCore::IR::COND_MI:
+      case FEXCore::IR::COND_PL:
+      case FEXCore::IR::COND_VS:
+      case FEXCore::IR::COND_VC:
+      default:
+        LOGMAN_MSG_A_FMT("Unsupported compare type");
+        break;
+    }
+
+    return CompResult;
+  }
+
+  static uint8_t GetOpSize(FEXCore::IR::IRListView const *CurrentIR, IR::OrderedNodeWrapper Node) {
+    auto IROp = CurrentIR->GetOp<FEXCore::IR::IROp_Header>(Node);
+    return IROp->Size;
+  }
+
   };
-};
+} // namespace FEXCore::CPU

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

@@ -0,0 +1,302 @@
+/*
+$info$
+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"
+
+#include <cstdint>
+
+namespace FEXCore::CPU {
+static inline void CacheLineFlush(char *Addr) {
+#ifdef _M_X86_64
+  __asm volatile (
+    "clflush (%[Addr]);"
+    :: [Addr] "r" (Addr)
+    : "memory");
+#else
+  __builtin___clear_cache(Addr, Addr+64);
+#endif
+}
+
+#define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
+DEF_OP(LoadContext) {
+  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;
+
+  #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 LoadContext size: {}", OpSize);
+      break;
+  }
+  #undef LOAD_CTX
+}
+
+DEF_OP(StoreContext) {
+  const auto Op = IROp->C<IR::IROp_StoreContext>();
+  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(LoadRegister) {
+  const auto Op = IROp->C<IR::IROp_LoadRegister>();
+  const auto OpSize = IROp->Size;
+
+  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*>(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 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) {
+  const auto Op = IROp->C<IR::IROp_StoreContextIndexed>();
+  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 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) {
+  LOGMAN_MSG_A_FMT("Unimplemented");
+}
+
+DEF_OP(FillRegister) {
+  LOGMAN_MSG_A_FMT("Unimplemented");
+}
+
+DEF_OP(LoadFlag) {
+  auto Op = IROp->C<IR::IROp_LoadFlag>();
+
+  uintptr_t ContextPtr = reinterpret_cast<uintptr_t>(Data->State->CurrentFrame);
+  ContextPtr += offsetof(FEXCore::Core::CPUState, flags[0]);
+  ContextPtr += Op->Flag;
+  uint8_t const *MemData = reinterpret_cast<uint8_t const*>(ContextPtr);
+  GD = *MemData;
+}
+
+DEF_OP(StoreFlag) {
+  auto Op = IROp->C<IR::IROp_StoreFlag>();
+  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]);
+  ContextPtr += Op->Flag;
+  uint8_t *MemData = reinterpret_cast<uint8_t*>(ContextPtr);
+  *MemData = Arg;
+}
+
+DEF_OP(LoadMem) {
+  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);
+
+  if (!Op->Offset.IsInvalid()) {
+    auto Offset = *GetSrc<uintptr_t const*>(Data->SSAData, Op->Offset) * Op->OffsetScale;
+
+    switch(Op->OffsetType.Val) {
+      case IR::MEM_OFFSET_SXTX.Val: MemData +=  Offset; break;
+      case IR::MEM_OFFSET_UXTW.Val: MemData += (uint32_t)Offset; break;
+      case IR::MEM_OFFSET_SXTW.Val: MemData += (int32_t)Offset; break;
+    }
+  }
+
+  memset(GDP, 0, Core::CPUState::XMM_AVX_REG_SIZE);
+  switch (OpSize) {
+    case 1: {
+      auto D = reinterpret_cast<const std::atomic<uint8_t>*>(MemData);
+      GD = D->load();
+      break;
+    }
+    case 2: {
+      auto D = reinterpret_cast<const std::atomic<uint16_t>*>(MemData);
+      GD = D->load();
+      break;
+    }
+    case 4: {
+      auto D = reinterpret_cast<const std::atomic<uint32_t>*>(MemData);
+      GD = D->load();
+      break;
+    }
+    case 8: {
+      auto D = reinterpret_cast<const std::atomic<uint64_t>*>(MemData);
+      GD = D->load();
+      break;
+    }
+    default:
+      memcpy(GDP, MemData, OpSize);
+      break;
+  }
+}
+
+DEF_OP(StoreMem) {
+  const auto Op = IROp->C<IR::IROp_StoreMem>();
+  const auto OpSize = IROp->Size;
+
+  uint8_t *MemData = *GetSrc<uint8_t **>(Data->SSAData, Op->Addr);
+
+  if (!Op->Offset.IsInvalid()) {
+    auto Offset = *GetSrc<uintptr_t const*>(Data->SSAData, Op->Offset) * Op->OffsetScale;
+
+    switch(Op->OffsetType.Val) {
+      case IR::MEM_OFFSET_SXTX.Val: MemData +=  Offset; break;
+      case IR::MEM_OFFSET_UXTW.Val: MemData += (uint32_t)Offset; break;
+      case IR::MEM_OFFSET_SXTW.Val: MemData += (int32_t)Offset; break;
+    }
+  }
+  switch (OpSize) {
+    case 1: {
+      reinterpret_cast<std::atomic<uint8_t>*>(MemData)->store(*GetSrc<uint8_t*>(Data->SSAData, Op->Value));
+      break;
+    }
+    case 2: {
+      reinterpret_cast<std::atomic<uint16_t>*>(MemData)->store(*GetSrc<uint16_t*>(Data->SSAData, Op->Value));
+      break;
+    }
+    case 4: {
+      reinterpret_cast<std::atomic<uint32_t>*>(MemData)->store(*GetSrc<uint32_t*>(Data->SSAData, Op->Value));
+      break;
+    }
+    case 8: {
+      reinterpret_cast<std::atomic<uint64_t>*>(MemData)->store(*GetSrc<uint64_t*>(Data->SSAData, Op->Value));
+      break;
+    }
+
+    default:
+      memcpy(MemData, GetSrc<void*>(Data->SSAData, Op->Value), OpSize);
+      break;
+  }
+}
+
+DEF_OP(CacheLineClear) {
+  auto Op = IROp->C<IR::IROp_CacheLineClear>();
+
+  char *MemData = *GetSrc<char **>(Data->SSAData, Op->Addr);
+
+  // 64-byte cache line clear
+  CacheLineFlush(MemData);
+}
+
+DEF_OP(CacheLineZero) {
+  auto Op = IROp->C<IR::IROp_CacheLineZero>();
+
+  uintptr_t MemData = *GetSrc<uintptr_t*>(Data->SSAData, Op->Addr);
+
+  // Force cacheline alignment
+  MemData = MemData & ~(CPUIDEmu::CACHELINE_SIZE - 1);
+
+  using DataType = uint64_t;
+  DataType *MemData64 = reinterpret_cast<DataType*>(MemData);
+
+  // 64-byte cache line zero
+  for (size_t i = 0; i < (CPUIDEmu::CACHELINE_SIZE / sizeof(DataType)); ++i) {
+    MemData64[i] = 0;
+  }
+}
+
+#undef DEF_OP
+} // namespace FEXCore::CPU

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

@@ -0,0 +1,174 @@
+/*
+$info$
+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"
+
+#include <FEXHeaderUtils/Syscalls.h>
+
+#include <cstdint>
+#ifdef _M_X86_64
+#include <xmmintrin.h>
+#endif
+#include <sys/random.h>
+
+namespace FEXCore::CPU {
+
+#define DEF_OP(x) void InterpreterOps::Op_##x(IR::IROp_Header *IROp, IROpData *Data, IR::NodeID Node)
+DEF_OP(Fence) {
+  auto Op = IROp->C<IR::IROp_Fence>();
+  switch (Op->Fence) {
+    case IR::Fence_Load.Val:
+      std::atomic_thread_fence(std::memory_order_acquire);
+      break;
+    case IR::Fence_LoadStore.Val:
+      std::atomic_thread_fence(std::memory_order_seq_cst);
+      break;
+    case IR::Fence_Store.Val:
+      std::atomic_thread_fence(std::memory_order_release);
+      break;
+    default: LOGMAN_MSG_A_FMT("Unknown Fence: {}", Op->Fence); break;
+  }
+}
+
+DEF_OP(Break) {
+  auto Op = IROp->C<IR::IROp_Break>();
+
+  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 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;
+  }
+}
+
+DEF_OP(GetRoundingMode) {
+  uint32_t GuestRounding{};
+#ifdef _M_ARM_64
+  uint64_t Tmp{};
+  __asm(R"(
+    mrs %[Tmp], FPCR;
+  )"
+  : [Tmp] "=r" (Tmp));
+  // Extract the rounding
+  // On ARM the ordering is different than on x86
+  GuestRounding |= ((Tmp >> 24) & 1) ? IR::ROUND_MODE_FLUSH_TO_ZERO : 0;
+  uint8_t RoundingMode = (Tmp >> 22) & 0b11;
+  if (RoundingMode == 0)
+    GuestRounding |= IR::ROUND_MODE_NEAREST;
+  else if (RoundingMode == 1)
+    GuestRounding |= IR::ROUND_MODE_POSITIVE_INFINITY;
+  else if (RoundingMode == 2)
+    GuestRounding |= IR::ROUND_MODE_NEGATIVE_INFINITY;
+  else if (RoundingMode == 3)
+    GuestRounding |= IR::ROUND_MODE_TOWARDS_ZERO;
+#else
+  GuestRounding = _mm_getcsr();
+
+  // Extract the rounding
+  GuestRounding = (GuestRounding >> 13) & 0b111;
+#endif
+  memcpy(GDP, &GuestRounding, sizeof(GuestRounding));
+}
+
+DEF_OP(SetRoundingMode) {
+  auto Op = IROp->C<IR::IROp_SetRoundingMode>();
+  const auto GuestRounding = *GetSrc<uint8_t*>(Data->SSAData, Op->RoundMode);
+#ifdef _M_ARM_64
+  uint64_t HostRounding{};
+  __asm volatile(R"(
+    mrs %[Tmp], FPCR;
+  )"
+  : [Tmp] "=r" (HostRounding));
+  // Mask out the rounding
+  HostRounding &= ~(0b111 << 22);
+
+  HostRounding |= (GuestRounding & IR::ROUND_MODE_FLUSH_TO_ZERO) ? (1U << 24) : 0;
+
+  uint8_t RoundingMode = GuestRounding & 0b11;
+  if (RoundingMode == IR::ROUND_MODE_NEAREST)
+    HostRounding |= (0b00U << 22);
+  else if (RoundingMode == IR::ROUND_MODE_POSITIVE_INFINITY)
+    HostRounding |= (0b01U << 22);
+  else if (RoundingMode == IR::ROUND_MODE_NEGATIVE_INFINITY)
+    HostRounding |= (0b10U << 22);
+  else if (RoundingMode == IR::ROUND_MODE_TOWARDS_ZERO)
+    HostRounding |= (0b11U << 22);
+
+  __asm volatile(R"(
+    msr FPCR, %[Tmp];
+  )"
+  :: [Tmp] "r" (HostRounding));
+#else
+  uint32_t HostRounding = _mm_getcsr();
+
+  // Cut out the host rounding mode
+  HostRounding &= ~(0b111 << 13);
+
+  // Insert our new rounding mode
+  HostRounding |= GuestRounding << 13;
+  _mm_setcsr(HostRounding);
+#endif
+}
+
+DEF_OP(Print) {
+  auto Op = IROp->C<IR::IROp_Print>();
+  const uint8_t OpSize = IROp->Size;
+
+  if (OpSize <= 8) {
+    const auto Src = *GetSrc<uint64_t*>(Data->SSAData, Op->Value);
+    LogMan::Msg::IFmt(">>>> Value in Arg: 0x{:x}, {}", Src, Src);
+  }
+  else if (OpSize == 16) {
+    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);
+  }
+  else
+    LOGMAN_MSG_A_FMT("Unknown value size: {}", OpSize);
+}
+
+DEF_OP(ProcessorID) {
+  uint32_t CPU, CPUNode;
+  FHU::Syscalls::getcpu(&CPU, &CPUNode);
+  GD = (CPUNode << 12) | CPU;
+}
+
+DEF_OP(RDRAND) {
+  // We are ignoring Op->GetReseeded in the interpreter
+  uint64_t *DstPtr = GetDest<uint64_t*>(Data->SSAData, Node);
+  ssize_t Result = ::getrandom(&DstPtr[0], 8, 0);
+
+  // Second result is if we managed to read a valid random number or not
+  DstPtr[1] = Result == 8 ? 1 : 0;
+}
+
+DEF_OP(Yield) {
+  // Nop implementation
+}
+
+#undef DEF_OP
+
+} // namespace FEXCore::CPU

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

@@ -0,0 +1,35 @@
+/*
+$info$
+tags: backend|interpreter
+$end_info$
+*/
+
+#include "Interface/Core/Interpreter/InterpreterClass.h"
+#include "Interface/Core/Interpreter/InterpreterOps.h"
+#include "Interface/Core/Interpreter/InterpreterDefines.h"
+
+#include <cstdint>
+
+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>();
+  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>();
+  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);
+
+  memcpy(Dst, Src_Lower, IROp->ElementSize);
+  memcpy(Dst + IROp->ElementSize, Src_Upper, IROp->ElementSize);
+}
+
+#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

@@ -4,6 +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"
@@ -11,23 +14,11 @@ $end_info$
 
 #include <FEXCore/Core/X86Enums.h>
 #include <FEXCore/HLE/SyscallHandler.h>
+#include <FEXCore/Utils/MathUtils.h>
 #include <Interface/HLE/Thunks/Thunks.h>
 
 namespace FEXCore::CPU {
-using namespace vixl;
-using namespace vixl::aarch64;
-#define DEF_OP(x) void Arm64JITCore::Op_##x(FEXCore::IR::IROp_Header *IROp, uint32_t Node)
-DEF_OP(GuestCallDirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestCallIndirect) {
-  LogMan::Msg::DFmt("Unimplemented");
-}
-
-DEF_OP(GuestReturn) {
-  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
@@ -35,12 +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
-  LoadConstant(x0, ThreadSharedData.SignalReturnInstruction);
-  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();
 
@@ -48,15 +38,15 @@ DEF_OP(CallbackReturn) {
   ResetStack();
 
   // We can now lower the ref counter again
-  LoadConstant(x0, reinterpret_cast<uint64_t>(ThreadSharedData.SignalHandlerRefCounterPtr));
-  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();
 
@@ -67,137 +57,114 @@ 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{ThreadSharedData.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
-    LoadConstant(x0, ThreadState->LookupCache->GetL1Pointer());
+    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);
-    LoadConstant(TMP1, ThreadSharedData.Dispatcher->AbsoluteLoopTopAddress);
-    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) {
-  auto Op = IROp->C<IR::IROp_Jump>();
+  const auto Op = IROp->C<IR::IROp_Jump>();
+  const auto Target = Op->TargetBlock.ID();
 
-  Label *TargetLabel;
-  auto IsTarget = JumpTargets.find(Op->Header.Args[0].ID());
-  if (IsTarget == JumpTargets.end()) {
-    TargetLabel = &JumpTargets.try_emplace(Op->Header.Args[0].ID()).first->second;
-  }
-  else {
-    TargetLabel = &IsTarget->second;
-  }
-  PendingTargetLabel = TargetLabel;
+  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())
-
-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::hi;
-  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;
-  Label *FalseTargetLabel;
-
-  auto TrueIter = JumpTargets.find(Op->TrueBlock.ID());
-  auto FalseIter = JumpTargets.find(Op->FalseBlock.ID());
-
-  if (TrueIter == JumpTargets.end()) {
-    TrueTargetLabel = &JumpTargets.try_emplace(Op->TrueBlock.ID()).first->second;
-  }
-  else {
-    TrueTargetLabel = &TrueIter->second;
-  }
-
+  auto TrueTargetLabel = &JumpTargets.try_emplace(Op->TrueBlock.ID()).first->second;
 
   uint64_t Const;
-  bool isConst = IsInlineConstant(Op->Cmp2, &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);
-      else
-        cmp(GRCMP(Op->Cmp1.ID()), GRCMP(Op->Cmp2.ID()));
+      if (isConst) {
+        cmp(Size, GetReg(Op->Cmp1.ID()), Const);
+      } else {
+        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);
   }
 
-  if (FalseIter == JumpTargets.end()) {
-    FalseTargetLabel = &JumpTargets.try_emplace(Op->FalseBlock.ID()).first->second;
-  }
-  else {
-    FalseTargetLabel = &FalseIter->second;
-  }
-  PendingTargetLabel = FalseTargetLabel;
+  PendingTargetLabel = &JumpTargets.try_emplace(Op->FalseBlock.ID()).first->second;
 }
 
 DEF_OP(Syscall) {
@@ -207,32 +174,152 @@ DEF_OP(Syscall) {
   // X1: ThreadState
   // X2: Pointer to SyscallArguments
 
-  PushDynamicRegsAndLR();
-  SpillStaticRegs();
+  FEXCore::IR::SyscallFlags Flags = Op->Flags;
+  PushDynamicRegsAndLR(TMP1);
 
-  uint64_t SPOffset = AlignUp(FEXCore::HLE::SyscallArguments::MAX_ARGS * 8, 16);
-  sub(sp, sp, 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));
+  if ((Flags & FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) != FEXCore::IR::SyscallFlags::NOSYNCSTATEONENTRY) {
+    SpillStaticRegs();
+  }
+  else {
+    // Need to spill all caller saved registers still
+    SpillStaticRegs(true, CALLER_GPR_MASK, CALLER_FPR_MASK);
   }
 
-  LoadConstant(x0, reinterpret_cast<uint64_t>(CTX->SyscallHandler));
-  mov(x1, STATE);
-  mov(x2, sp);
+  uint64_t SPOffset = AlignUp(FEXCore::HLE::SyscallArguments::MAX_ARGS * 8, 16);
+  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(Op->Header.Args[i].ID()).X(), ARMEmitter::Reg::rsp, i * 8);
+  }
 
-  LoadConstant(x3, reinterpret_cast<uint64_t>(FEXCore::Context::HandleSyscall));
-  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) {
+    FillStaticRegs();
+  }
+  else {
+    // Result is now in x0
+    // Fix the stack and any values that were stepped on
+    FillStaticRegs(true, CALLER_GPR_MASK, CALLER_FPR_MASK);
+  }
 
-  // Result is now in x0
-  // Fix the stack and any values that were stepped on
-  FillStaticRegs();
   PopDynamicRegsAndLR();
 
-  // Move result to its destination register
-  mov(GetReg<RA_64>(Node), x0);
+  if ((Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
+    // Move result to its destination register
+    mov(ARMEmitter::Size::i64Bit, GetReg(Node), ARMEmitter::Reg::r0);
+  }
+}
+
+DEF_OP(InlineSyscall) {
+  auto Op = IROp->C<IR::IROp_InlineSyscall>();
+  // Arguments are passed as follows:
+  // X8: SyscallNumber - RA INTERSECT
+  // X0: Arg0 & Return
+  // X1: Arg1
+  // X2: Arg2
+  // X3: Arg3
+  // X4: Arg4 - RA INTERSECT
+  // X5: Arg5 - RA INTERSECT
+  // 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<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;
+  for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
+    if (Op->Header.Args[i].IsInvalid()) break;
+
+    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.Idx());
+      Intersects = true;
+    }
+  }
+
+  // Ordering is incredibly important here
+  // We must spill any overlapping registers first THEN claim we are in a syscall without invalidating state at all
+  // Only spill the registers that intersect with our usage
+  SpillStaticRegs(false, SpillMask);
+
+  // Now that we are spilled, store in the state that we are in a syscall
+  // Still without overwriting registers that matter
+  // 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(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;
+
+      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]));
+      }
+      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 {
+        mov(EmitSize, RegArgs[i].R(), Reg);
+      }
+    }
+  }
+  else {
+    for (uint32_t i = 0; i < FEXCore::HLE::SyscallArguments::MAX_ARGS-1; ++i) {
+      if (Op->Header.Args[i].IsInvalid()) break;
+
+      mov(EmitSize, RegArgs[i].R(), GetReg(Op->Header.Args[i].ID()));
+    }
+  }
+
+  LoadConstant(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r8, Op->HostSyscallNumber);
+  svc(0);
+  // On updated signal mask we can receive a signal RIGHT HERE
+
+  if ((Op->Flags & FEXCore::IR::SyscallFlags::NORETURN) != FEXCore::IR::SyscallFlags::NORETURN) {
+    // Now that we are done in the syscall we need to carefully peel back the state
+    // First unspill the registers from before
+    FillStaticRegs(false, SpillMask);
+
+    // 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(ARMEmitter::XReg::zr, STATE, offsetof(FEXCore::Core::CpuStateFrame, InSyscallInfo));
+
+    // Result is now in x0
+    // Move result to its destination register
+    mov(EmitSize, GetReg(Node), ARMEmitter::Reg::r0);
+  }
 }
 
 DEF_OP(Thunk) {
@@ -243,81 +330,90 @@ 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();
 
   FillStaticRegs(); // load from ctx after ra64 refill
 }
 
-
 DEF_OP(ValidateCode) {
   auto Op = IROp->C<IR::IROp_ValidateCode>();
   const auto *OldCode = (const uint8_t *)&Op->CodeOriginalLow;
   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(RemoveCodeEntry) {
+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);
 
-  LoadConstant(x2, reinterpret_cast<uintptr_t>(&Context::Context::RemoveCodeEntryFromJit));
+  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
@@ -327,52 +423,46 @@ DEF_OP(RemoveCodeEntry) {
 DEF_OP(CPUID) {
   auto Op = IROp->C<IR::IROp_CPUID>();
 
-  PushDynamicRegsAndLR();
+  PushDynamicRegsAndLR(TMP1);
+  SpillStaticRegs();
 
   // x0 = CPUID Handler
   // x1 = CPUID Function
   // x2 = CPUID Leaf
-  LoadConstant(x0, reinterpret_cast<uint64_t>(&CTX->CPUID));
-  mov(x1, GetReg<RA_64>(Op->Header.Args[0].ID()));
-  mov(x2, GetReg<RA_64>(Op->Header.Args[1].ID()));
+  ldr(ARMEmitter::XReg::x0, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDObj));
+  ldr(ARMEmitter::XReg::x3, STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.Common.CPUIDFunction));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r1, GetReg(Op->Function.ID()));
+  mov(ARMEmitter::Size::i64Bit, ARMEmitter::Reg::r2, GetReg(Op->Leaf.ID()));
+#ifdef VIXL_SIMULATOR
+  GenerateIndirectRuntimeCall<__uint128_t, void*, uint64_t, uint64_t>(ARMEmitter::Reg::r3);
+#else
+  blr(ARMEmitter::Reg::r3);
+#endif
 
-  using ClassPtrType = FEXCore::CPUID::FunctionResults (FEXCore::CPUIDEmu::*)(uint32_t, uint32_t);
-  union PtrCast {
-    ClassPtrType ClassPtr;
-    uintptr_t Data;
-  };
-
-  PtrCast Ptr;
-  Ptr.ClassPtr = &FEXCore::CPUIDEmu::RunFunction;
-  LoadConstant(x3, Ptr.Data);
-  SpillStaticRegs();
-  blr(x3);
   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(GUESTRETURN,       GuestReturn);
   REGISTER_OP(SIGNALRETURN,      SignalReturn);
   REGISTER_OP(CALLBACKRETURN,    CallbackReturn);
   REGISTER_OP(EXITFUNCTION,      ExitFunction);
   REGISTER_OP(JUMP,              Jump);
   REGISTER_OP(CONDJUMP,          CondJump);
   REGISTER_OP(SYSCALL,           Syscall);
+  REGISTER_OP(INLINESYSCALL,     InlineSyscall);
   REGISTER_OP(THUNK,             Thunk);
   REGISTER_OP(VALIDATECODE,      ValidateCode);
-  REGISTER_OP(REMOVECODEENTRY,   RemoveCodeEntry);
+  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(FEXCore::IR::IROp_Header *IROp, uint32_t 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->Index, 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->Index, 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->Index, GetReg<RA_32>(Op->Header.Args[1].ID()));
-    break;
-    }
-    case 8: {
-      ins(GetDst(Node).V2D(), Op->Index, 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()).W());
-      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()).W());
-      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

@@ -4,100 +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(FEXCore::IR::IROp_Header *IROp, uint32_t 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::Label Constant;
-  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
-  adr(TMP1.X(), &Constant);
-  ldr(VTMP3, MemOperand(TMP1.X()));
+  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);
-  bind(&Constant);
-  dc32(0x0B0E0104);
-  dc32(0x040B0E01);
-  dc32(0x0306090C);
-  dc32(0x0C030609);
-  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(Dst.W(), Src1.W(), Src2.W());
+      break;
+    case 2:
+      crc32ch(Dst.W(), Src1.W(), Src2.W());
+      break;
+    case 4:
+      crc32cw(Dst.W(), Src1.W(), Src2.W());
+      break;
+    case 8:
+      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(FEXCore::IR::IROp_Header *IROp, uint32_t 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